diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU')
55 files changed, 9455 insertions, 5957 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/a-v-flat-atomicrmw.ll b/llvm/test/CodeGen/AMDGPU/a-v-flat-atomicrmw.ll index 7cc5051..003aa04 100644 --- a/llvm/test/CodeGen/AMDGPU/a-v-flat-atomicrmw.ll +++ b/llvm/test/CodeGen/AMDGPU/a-v-flat-atomicrmw.ll @@ -8759,9 +8759,8 @@ define void @flat_atomic_usub_sat_i64_ret_a_a(ptr %ptr) #0 { ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v6 ; GFX90A-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v7, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] -; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[0:3] glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3] @@ -8780,20 +8779,19 @@ define void @flat_atomic_usub_sat_i64_ret_a_a(ptr %ptr) #0 { ; GFX90A-NEXT: s_cbranch_execz .LBB113_6 ; GFX90A-NEXT: ; %bb.5: ; %atomicrmw.private ; GFX90A-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5] -; GFX90A-NEXT: v_cndmask_b32_e32 v4, -1, v4, vcc -; GFX90A-NEXT: buffer_load_dword v0, v4, s[0:3], 0 offen -; GFX90A-NEXT: buffer_load_dword v1, v4, s[0:3], 0 offen offset:4 +; GFX90A-NEXT: v_cndmask_b32_e32 v0, -1, v4, vcc +; GFX90A-NEXT: buffer_load_dword v1, v0, s[0:3], 0 offen +; GFX90A-NEXT: buffer_load_dword v2, v0, s[0:3], 0 offen offset:4 ; GFX90A-NEXT: s_waitcnt vmcnt(1) -; GFX90A-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v6 +; GFX90A-NEXT: v_sub_co_u32_e32 v3, vcc, v1, v6 ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v7, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; GFX90A-NEXT: v_accvgpr_write_b32 a0, v0 -; GFX90A-NEXT: v_cndmask_b32_e64 v0, v3, 0, vcc -; GFX90A-NEXT: v_accvgpr_write_b32 a1, v1 -; GFX90A-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc -; GFX90A-NEXT: buffer_store_dword v0, v4, s[0:3], 0 offen offset:4 -; GFX90A-NEXT: buffer_store_dword v2, v4, s[0:3], 0 offen +; GFX90A-NEXT: v_subb_co_u32_e32 v4, vcc, v2, v7, vcc +; GFX90A-NEXT: v_accvgpr_write_b32 a0, v1 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc +; GFX90A-NEXT: v_accvgpr_write_b32 a1, v2 +; GFX90A-NEXT: v_cndmask_b32_e64 v1, v4, 0, vcc +; GFX90A-NEXT: buffer_store_dword v3, v0, s[0:3], 0 offen +; GFX90A-NEXT: buffer_store_dword v1, v0, s[0:3], 0 offen offset:4 ; GFX90A-NEXT: .LBB113_6: ; %atomicrmw.phi ; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX90A-NEXT: ;;#ASMSTART @@ -8827,10 +8825,9 @@ define void @flat_atomic_usub_sat_i64_ret_a_a(ptr %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v6 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v7, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[4:5], v[0:3] sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3] @@ -8856,11 +8853,11 @@ define void @flat_atomic_usub_sat_i64_ret_a_a(ptr %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v6 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v7, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] ; GFX950-NEXT: v_accvgpr_write_b32 a0, v0 -; GFX950-NEXT: v_accvgpr_write_b32 a1, v1 +; GFX950-NEXT: s_nop 0 ; GFX950-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc +; GFX950-NEXT: v_accvgpr_write_b32 a1, v1 ; GFX950-NEXT: scratch_store_dwordx2 v4, v[2:3], off ; GFX950-NEXT: .LBB113_6: ; %atomicrmw.phi ; GFX950-NEXT: s_or_b64 exec, exec, s[0:1] @@ -8900,9 +8897,8 @@ define void @flat_atomic_usub_sat_i64_ret_av_av(ptr %ptr) #0 { ; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-NEXT: v_sub_co_u32_e32 v4, vcc, v6, v2 ; GFX90A-NEXT: v_subb_co_u32_e32 v5, vcc, v7, v3, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[4:5], v[6:7] -; GFX90A-NEXT: v_cndmask_b32_e64 v5, v5, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v4, v4, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v5, v5, 0, vcc ; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7] glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] @@ -8918,18 +8914,17 @@ define void @flat_atomic_usub_sat_i64_ret_av_av(ptr %ptr) #0 { ; GFX90A-NEXT: s_cbranch_execz .LBB114_6 ; GFX90A-NEXT: ; %bb.5: ; %atomicrmw.private ; GFX90A-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1] -; GFX90A-NEXT: v_cndmask_b32_e32 v6, -1, v0, vcc -; GFX90A-NEXT: buffer_load_dword v4, v6, s[0:3], 0 offen -; GFX90A-NEXT: buffer_load_dword v5, v6, s[0:3], 0 offen offset:4 +; GFX90A-NEXT: v_cndmask_b32_e32 v0, -1, v0, vcc +; GFX90A-NEXT: buffer_load_dword v4, v0, s[0:3], 0 offen +; GFX90A-NEXT: buffer_load_dword v5, v0, s[0:3], 0 offen offset:4 ; GFX90A-NEXT: s_waitcnt vmcnt(1) -; GFX90A-NEXT: v_sub_co_u32_e32 v0, vcc, v4, v2 +; GFX90A-NEXT: v_sub_co_u32_e32 v1, vcc, v4, v2 ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_subb_co_u32_e32 v1, vcc, v5, v3, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[4:5] -; GFX90A-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX90A-NEXT: v_subb_co_u32_e32 v2, vcc, v5, v3, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc -; GFX90A-NEXT: buffer_store_dword v0, v6, s[0:3], 0 offen -; GFX90A-NEXT: buffer_store_dword v1, v6, s[0:3], 0 offen offset:4 +; GFX90A-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc +; GFX90A-NEXT: buffer_store_dword v1, v0, s[0:3], 0 offen +; GFX90A-NEXT: buffer_store_dword v2, v0, s[0:3], 0 offen offset:4 ; GFX90A-NEXT: .LBB114_6: ; %atomicrmw.phi ; GFX90A-NEXT: s_or_b64 exec, exec, s[4:5] ; GFX90A-NEXT: ;;#ASMSTART @@ -8962,10 +8957,9 @@ define void @flat_atomic_usub_sat_i64_ret_av_av(ptr %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v2, vcc, v8, v0 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v3, vcc, v9, v1, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[8:9] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v6, v2, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX950-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[4:5], v[6:9] sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[8:9] @@ -8988,7 +8982,6 @@ define void @flat_atomic_usub_sat_i64_ret_av_av(ptr %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v0 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v1, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc @@ -17064,9 +17057,8 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_a_a(ptr inreg %ptr) #0 { ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v4 ; GFX90A-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v5, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] -; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[6:7], v[0:3] glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3] @@ -17085,20 +17077,19 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_a_a(ptr inreg %ptr) #0 { ; GFX90A-NEXT: ; %bb.5: ; %atomicrmw.private ; GFX90A-NEXT: s_cmp_lg_u64 s[4:5], 0 ; GFX90A-NEXT: s_cselect_b32 s4, s4, -1 -; GFX90A-NEXT: v_mov_b32_e32 v6, s4 -; GFX90A-NEXT: buffer_load_dword v0, v6, s[0:3], 0 offen -; GFX90A-NEXT: buffer_load_dword v1, v6, s[0:3], 0 offen offset:4 +; GFX90A-NEXT: v_mov_b32_e32 v0, s4 +; GFX90A-NEXT: buffer_load_dword v1, v0, s[0:3], 0 offen +; GFX90A-NEXT: buffer_load_dword v2, v0, s[0:3], 0 offen offset:4 ; GFX90A-NEXT: s_waitcnt vmcnt(1) -; GFX90A-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v4 +; GFX90A-NEXT: v_sub_co_u32_e32 v3, vcc, v1, v4 ; GFX90A-NEXT: s_waitcnt vmcnt(0) -; GFX90A-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v5, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; GFX90A-NEXT: v_accvgpr_write_b32 a0, v0 -; GFX90A-NEXT: v_cndmask_b32_e64 v0, v3, 0, vcc -; GFX90A-NEXT: v_accvgpr_write_b32 a1, v1 -; GFX90A-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc -; GFX90A-NEXT: buffer_store_dword v0, v6, s[0:3], 0 offen offset:4 -; GFX90A-NEXT: buffer_store_dword v2, v6, s[0:3], 0 offen +; GFX90A-NEXT: v_subb_co_u32_e32 v4, vcc, v2, v5, vcc +; GFX90A-NEXT: v_accvgpr_write_b32 a0, v1 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc +; GFX90A-NEXT: v_accvgpr_write_b32 a1, v2 +; GFX90A-NEXT: v_cndmask_b32_e64 v1, v4, 0, vcc +; GFX90A-NEXT: buffer_store_dword v3, v0, s[0:3], 0 offen +; GFX90A-NEXT: buffer_store_dword v1, v0, s[0:3], 0 offen offset:4 ; GFX90A-NEXT: .LBB221_6: ; %atomicrmw.phi ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; use a[0:1] @@ -17131,10 +17122,9 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_a_a(ptr inreg %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v4 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v5, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: flat_atomic_cmpswap_x2 v[0:1], v[6:7], v[0:3] sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3] @@ -17158,11 +17148,11 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_a_a(ptr inreg %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v4 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v5, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] ; GFX950-NEXT: v_accvgpr_write_b32 a0, v0 -; GFX950-NEXT: v_accvgpr_write_b32 a1, v1 +; GFX950-NEXT: s_nop 0 ; GFX950-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc +; GFX950-NEXT: v_accvgpr_write_b32 a1, v1 ; GFX950-NEXT: scratch_store_dwordx2 off, v[2:3], s0 ; GFX950-NEXT: .LBB221_6: ; %atomicrmw.phi ; GFX950-NEXT: ;;#ASMSTART @@ -17201,9 +17191,8 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_av_av(ptr inreg %ptr) #0 { ; GFX90A-NEXT: v_pk_mov_b32 v[8:9], v[2:3], v[2:3] op_sel:[0,1] ; GFX90A-NEXT: v_sub_co_u32_e32 v2, vcc, v8, v0 ; GFX90A-NEXT: v_subb_co_u32_e32 v3, vcc, v9, v1, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[8:9] -; GFX90A-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v6, v2, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX90A-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[4:5], v[6:9] glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[8:9] @@ -17226,7 +17215,6 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_av_av(ptr inreg %ptr) #0 { ; GFX90A-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v0 ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v1, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] ; GFX90A-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: buffer_store_dword v0, v4, s[0:3], 0 offen @@ -17262,10 +17250,9 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_av_av(ptr inreg %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v2, vcc, v8, v0 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v3, vcc, v9, v1, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[8:9] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v6, v2, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX950-NEXT: flat_atomic_cmpswap_x2 v[2:3], v[4:5], v[6:9] sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[8:9] @@ -17286,7 +17273,6 @@ define void @flat_atomic_usub_sat_i64_saddr_ret_av_av(ptr inreg %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v0 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v1, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc diff --git a/llvm/test/CodeGen/AMDGPU/a-v-global-atomicrmw.ll b/llvm/test/CodeGen/AMDGPU/a-v-global-atomicrmw.ll index c98fff9..34a4899 100644 --- a/llvm/test/CodeGen/AMDGPU/a-v-global-atomicrmw.ll +++ b/llvm/test/CodeGen/AMDGPU/a-v-global-atomicrmw.ll @@ -5804,9 +5804,8 @@ define void @global_atomic_usub_sat_i64_ret_a_a(ptr addrspace(1) %ptr) #0 { ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_sub_co_u32_e32 v2, vcc, v4, v6 ; GFX90A-NEXT: v_subb_co_u32_e32 v3, vcc, v5, v7, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[4:5] -; GFX90A-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc ; GFX90A-NEXT: global_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5], off offset:80 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] @@ -5839,10 +5838,9 @@ define void @global_atomic_usub_sat_i64_ret_a_a(ptr addrspace(1) %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v2, vcc, v4, v6 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v3, vcc, v5, v7, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[4:5] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v2, v2, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v3, v3, 0, vcc ; GFX950-NEXT: global_atomic_cmpswap_x2 v[2:3], v[0:1], v[2:5], off offset:80 sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] @@ -5880,9 +5878,8 @@ define void @global_atomic_usub_sat_i64_ret_av_av(ptr addrspace(1) %ptr) #0 { ; GFX90A-NEXT: v_pk_mov_b32 v[6:7], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-NEXT: v_sub_co_u32_e32 v4, vcc, v6, v2 ; GFX90A-NEXT: v_subb_co_u32_e32 v5, vcc, v7, v3, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[4:5], v[6:7] -; GFX90A-NEXT: v_cndmask_b32_e64 v5, v5, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v4, v4, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v5, v5, 0, vcc ; GFX90A-NEXT: global_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7], off offset:80 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] @@ -5911,10 +5908,9 @@ define void @global_atomic_usub_sat_i64_ret_av_av(ptr addrspace(1) %ptr) #0 { ; GFX950-NEXT: v_sub_co_u32_e32 v4, vcc, v6, v2 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v5, vcc, v7, v3, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[4:5], v[6:7] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v5, v5, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v4, v4, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v5, v5, 0, vcc ; GFX950-NEXT: global_atomic_cmpswap_x2 v[4:5], v[0:1], v[4:7], off offset:80 sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[6:7] @@ -11573,9 +11569,8 @@ define void @global_atomic_usub_sat_i64_saddr_ret_a_a(ptr addrspace(1) inreg %pt ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v4 ; GFX90A-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v5, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] -; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX90A-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[16:17] offset:80 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3] @@ -11609,10 +11604,9 @@ define void @global_atomic_usub_sat_i64_saddr_ret_a_a(ptr addrspace(1) inreg %pt ; GFX950-NEXT: v_sub_co_u32_e32 v0, vcc, v2, v4 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v1, vcc, v3, v5, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[0:1], v[2:3] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX950-NEXT: global_atomic_cmpswap_x2 v[0:1], v6, v[0:3], s[0:1] offset:80 sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[2:3] @@ -11651,9 +11645,8 @@ define void @global_atomic_usub_sat_i64_saddr_ret_av_av(ptr addrspace(1) inreg % ; GFX90A-NEXT: v_pk_mov_b32 v[8:9], v[2:3], v[2:3] op_sel:[0,1] ; GFX90A-NEXT: v_sub_co_u32_e32 v2, vcc, v8, v0 ; GFX90A-NEXT: v_subb_co_u32_e32 v3, vcc, v9, v1, vcc -; GFX90A-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[8:9] -; GFX90A-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX90A-NEXT: v_cndmask_b32_e64 v6, v2, 0, vcc +; GFX90A-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX90A-NEXT: global_atomic_cmpswap_x2 v[2:3], v4, v[6:9], s[16:17] offset:80 glc ; GFX90A-NEXT: s_waitcnt vmcnt(0) ; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[8:9] @@ -11683,10 +11676,9 @@ define void @global_atomic_usub_sat_i64_saddr_ret_av_av(ptr addrspace(1) inreg % ; GFX950-NEXT: v_sub_co_u32_e32 v2, vcc, v8, v0 ; GFX950-NEXT: s_nop 1 ; GFX950-NEXT: v_subb_co_u32_e32 v3, vcc, v9, v1, vcc -; GFX950-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[8:9] ; GFX950-NEXT: s_nop 1 -; GFX950-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX950-NEXT: v_cndmask_b32_e64 v6, v2, 0, vcc +; GFX950-NEXT: v_cndmask_b32_e64 v7, v3, 0, vcc ; GFX950-NEXT: global_atomic_cmpswap_x2 v[2:3], v4, v[6:9], s[0:1] offset:80 sc0 ; GFX950-NEXT: s_waitcnt vmcnt(0) ; GFX950-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[8:9] diff --git a/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll b/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll index 2d7ef2c..98fbbe1 100644 --- a/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll +++ b/llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll @@ -169,6 +169,6 @@ attributes #1 = { nounwind } ;. ; HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) } -; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/addsub64_carry.ll b/llvm/test/CodeGen/AMDGPU/addsub64_carry.ll index d326966..b72eba8 100644 --- a/llvm/test/CodeGen/AMDGPU/addsub64_carry.ll +++ b/llvm/test/CodeGen/AMDGPU/addsub64_carry.ll @@ -17,12 +17,9 @@ define %struct.uint96 @v_add64_32(i64 %val64A, i64 %val64B, i32 %val32) { ; CHECK-LABEL: v_add64_32: ; CHECK: ; %bb.0: ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_add_co_u32_e32 v5, vcc, v0, v2 -; CHECK-NEXT: v_addc_co_u32_e32 v6, vcc, v1, v3, vcc -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, v[5:6], v[0:1] -; CHECK-NEXT: v_mov_b32_e32 v0, v5 +; CHECK-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2 +; CHECK-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc ; CHECK-NEXT: v_addc_co_u32_e32 v2, vcc, 0, v4, vcc -; CHECK-NEXT: v_mov_b32_e32 v1, v6 ; CHECK-NEXT: s_setpc_b64 s[30:31] %sum64 = add i64 %val64A, %val64B %obit = icmp ult i64 %sum64, %val64A @@ -38,16 +35,14 @@ define <2 x i64> @v_uadd_v2i64(<2 x i64> %val0, <2 x i64> %val1, ptr %ptrval) { ; CHECK: ; %bb.0: ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; CHECK-NEXT: v_add_co_u32_e32 v6, vcc, v2, v6 +; CHECK-NEXT: v_add_co_u32_e64 v4, s[4:5], v0, v4 ; CHECK-NEXT: v_addc_co_u32_e32 v7, vcc, v3, v7, vcc -; CHECK-NEXT: v_add_co_u32_e32 v4, vcc, v0, v4 -; CHECK-NEXT: v_addc_co_u32_e32 v5, vcc, v1, v5, vcc -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, v[4:5], v[0:1] -; CHECK-NEXT: flat_store_dwordx4 v[8:9], v[4:7] -; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, v[6:7], v[2:3] -; CHECK-NEXT: v_mov_b32_e32 v1, v0 +; CHECK-NEXT: v_addc_co_u32_e64 v5, s[4:5], v1, v5, s[4:5] +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] ; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc +; CHECK-NEXT: v_mov_b32_e32 v1, v0 ; CHECK-NEXT: v_mov_b32_e32 v3, v2 +; CHECK-NEXT: flat_store_dwordx4 v[8:9], v[4:7] ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: s_setpc_b64 s[30:31] %pair = call {<2 x i64>, <2 x i1>} @llvm.uadd.with.overflow.v2i64(<2 x i64> %val0, <2 x i64> %val1) @@ -63,16 +58,14 @@ define <2 x i64> @v_usub_v2i64(<2 x i64> %val0, <2 x i64> %val1, ptr %ptrval) { ; CHECK: ; %bb.0: ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; CHECK-NEXT: v_sub_co_u32_e32 v6, vcc, v2, v6 +; CHECK-NEXT: v_sub_co_u32_e64 v4, s[4:5], v0, v4 ; CHECK-NEXT: v_subb_co_u32_e32 v7, vcc, v3, v7, vcc -; CHECK-NEXT: v_sub_co_u32_e32 v4, vcc, v0, v4 -; CHECK-NEXT: v_subb_co_u32_e32 v5, vcc, v1, v5, vcc -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, v[4:5], v[0:1] -; CHECK-NEXT: flat_store_dwordx4 v[8:9], v[4:7] -; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, v[6:7], v[2:3] -; CHECK-NEXT: v_mov_b32_e32 v1, v0 +; CHECK-NEXT: v_subb_co_u32_e64 v5, s[4:5], v1, v5, s[4:5] +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] ; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc +; CHECK-NEXT: v_mov_b32_e32 v1, v0 ; CHECK-NEXT: v_mov_b32_e32 v3, v2 +; CHECK-NEXT: flat_store_dwordx4 v[8:9], v[4:7] ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: s_setpc_b64 s[30:31] %pair = call {<2 x i64>, <2 x i1>} @llvm.usub.with.overflow.v2i64(<2 x i64> %val0, <2 x i64> %val1) @@ -87,10 +80,9 @@ define i64 @v_uadd_i64(i64 %val0, i64 %val1, ptr %ptrval) { ; CHECK-LABEL: v_uadd_i64: ; CHECK: ; %bb.0: ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_add_co_u32_e32 v2, vcc, v0, v2 -; CHECK-NEXT: v_addc_co_u32_e32 v3, vcc, v1, v3, vcc -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[2:3] +; CHECK-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2 +; CHECK-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc +; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[0:1] ; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc ; CHECK-NEXT: v_mov_b32_e32 v1, v0 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) @@ -109,7 +101,6 @@ define i64 @v_uadd_p1(i64 %val0, i64 %val1, ptr %ptrval) { ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; CHECK-NEXT: v_add_co_u32_e32 v0, vcc, 1, v0 ; CHECK-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] ; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[0:1] ; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc ; CHECK-NEXT: v_mov_b32_e32 v1, v0 @@ -147,10 +138,9 @@ define i64 @v_usub_p1(i64 %val0, i64 %val1, ptr %ptrval) { ; CHECK-LABEL: v_usub_p1: ; CHECK: ; %bb.0: ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_add_co_u32_e32 v2, vcc, -1, v0 -; CHECK-NEXT: v_addc_co_u32_e32 v3, vcc, -1, v1, vcc -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[2:3] +; CHECK-NEXT: v_subrev_co_u32_e32 v0, vcc, 1, v0 +; CHECK-NEXT: v_subbrev_co_u32_e32 v1, vcc, 0, v1, vcc +; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[0:1] ; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc ; CHECK-NEXT: v_mov_b32_e32 v1, v0 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) @@ -167,10 +157,9 @@ define i64 @v_usub_n1(i64 %val0, i64 %val1, ptr %ptrval) { ; CHECK-LABEL: v_usub_n1: ; CHECK: ; %bb.0: ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_add_co_u32_e32 v2, vcc, 1, v0 -; CHECK-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v1, vcc -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[2:3] +; CHECK-NEXT: v_subrev_co_u32_e32 v0, vcc, -1, v0 +; CHECK-NEXT: v_subbrev_co_u32_e32 v1, vcc, -1, v1, vcc +; CHECK-NEXT: flat_store_dwordx2 v[4:5], v[0:1] ; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc ; CHECK-NEXT: v_mov_b32_e32 v1, v0 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) @@ -190,15 +179,13 @@ define i64 @v_usub_n1(i64 %val0, i64 %val1, ptr %ptrval) { define amdgpu_ps %struct.uint96 @s_add64_32(i64 inreg %val64A, i64 inreg %val64B, i32 inreg %val32) { ; CHECK-LABEL: s_add64_32: ; CHECK: ; %bb.0: -; CHECK-NEXT: s_add_u32 s6, s0, s2 -; CHECK-NEXT: v_mov_b32_e32 v0, s0 -; CHECK-NEXT: s_addc_u32 s7, s1, s3 -; CHECK-NEXT: v_mov_b32_e32 v1, s1 -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[0:1] -; CHECK-NEXT: s_mov_b32 s0, s6 -; CHECK-NEXT: s_cmp_lg_u64 vcc, 0 +; CHECK-NEXT: s_add_u32 s0, s0, s2 +; CHECK-NEXT: s_cselect_b64 s[6:7], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[6:7], 0 +; CHECK-NEXT: s_addc_u32 s1, s1, s3 +; CHECK-NEXT: s_cselect_b64 s[2:3], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[2:3], 0 ; CHECK-NEXT: s_addc_u32 s2, s4, 0 -; CHECK-NEXT: s_mov_b32 s1, s7 ; CHECK-NEXT: ; return to shader part epilog %sum64 = add i64 %val64A, %val64B %obit = icmp ult i64 %sum64, %val64A @@ -212,24 +199,24 @@ define amdgpu_ps %struct.uint96 @s_add64_32(i64 inreg %val64A, i64 inreg %val64B define amdgpu_ps <2 x i64> @s_uadd_v2i64(<2 x i64> inreg %val0, <2 x i64> inreg %val1, ptr %ptrval) { ; CHECK-LABEL: s_uadd_v2i64: ; CHECK: ; %bb.0: -; CHECK-NEXT: s_add_u32 s6, s2, s6 -; CHECK-NEXT: v_mov_b32_e32 v9, s3 -; CHECK-NEXT: s_addc_u32 s7, s3, s7 -; CHECK-NEXT: v_mov_b32_e32 v8, s2 -; CHECK-NEXT: s_add_u32 s4, s0, s4 -; CHECK-NEXT: v_mov_b32_e32 v7, s1 -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[8:9] -; CHECK-NEXT: s_addc_u32 s5, s1, s5 -; CHECK-NEXT: v_mov_b32_e32 v6, s0 -; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7] -; CHECK-NEXT: v_readfirstlane_b32 s2, v8 -; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc -; CHECK-NEXT: v_readfirstlane_b32 s0, v6 -; CHECK-NEXT: v_mov_b32_e32 v2, s4 -; CHECK-NEXT: v_mov_b32_e32 v3, s5 -; CHECK-NEXT: v_mov_b32_e32 v4, s6 -; CHECK-NEXT: v_mov_b32_e32 v5, s7 +; CHECK-NEXT: s_add_u32 s10, s2, s6 +; CHECK-NEXT: s_cselect_b64 s[8:9], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[8:9], 0 +; CHECK-NEXT: s_addc_u32 s8, s3, s7 +; CHECK-NEXT: s_cselect_b64 s[2:3], -1, 0 +; CHECK-NEXT: s_add_u32 s0, s0, s4 +; CHECK-NEXT: s_cselect_b64 s[6:7], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[6:7], 0 +; CHECK-NEXT: s_addc_u32 s1, s1, s5 +; CHECK-NEXT: v_mov_b32_e32 v2, s0 +; CHECK-NEXT: v_mov_b32_e32 v3, s1 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[2:3] +; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v7 +; CHECK-NEXT: v_readfirstlane_b32 s2, v6 +; CHECK-NEXT: v_mov_b32_e32 v4, s10 +; CHECK-NEXT: v_mov_b32_e32 v5, s8 ; CHECK-NEXT: s_mov_b32 s1, s0 ; CHECK-NEXT: s_mov_b32 s3, s2 ; CHECK-NEXT: flat_store_dwordx4 v[0:1], v[2:5] @@ -246,24 +233,24 @@ define amdgpu_ps <2 x i64> @s_uadd_v2i64(<2 x i64> inreg %val0, <2 x i64> inreg define amdgpu_ps <2 x i64> @s_usub_v2i64(<2 x i64> inreg %val0, <2 x i64> inreg %val1, ptr %ptrval) { ; CHECK-LABEL: s_usub_v2i64: ; CHECK: ; %bb.0: -; CHECK-NEXT: s_sub_u32 s6, s2, s6 -; CHECK-NEXT: v_mov_b32_e32 v9, s3 -; CHECK-NEXT: s_subb_u32 s7, s3, s7 -; CHECK-NEXT: v_mov_b32_e32 v8, s2 -; CHECK-NEXT: s_sub_u32 s4, s0, s4 -; CHECK-NEXT: v_mov_b32_e32 v7, s1 -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, s[6:7], v[8:9] -; CHECK-NEXT: s_subb_u32 s5, s1, s5 -; CHECK-NEXT: v_mov_b32_e32 v6, s0 -; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, s[4:5], v[6:7] -; CHECK-NEXT: v_readfirstlane_b32 s2, v8 -; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc -; CHECK-NEXT: v_readfirstlane_b32 s0, v6 -; CHECK-NEXT: v_mov_b32_e32 v2, s4 -; CHECK-NEXT: v_mov_b32_e32 v3, s5 -; CHECK-NEXT: v_mov_b32_e32 v4, s6 -; CHECK-NEXT: v_mov_b32_e32 v5, s7 +; CHECK-NEXT: s_sub_u32 s10, s2, s6 +; CHECK-NEXT: s_cselect_b64 s[8:9], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[8:9], 0 +; CHECK-NEXT: s_subb_u32 s8, s3, s7 +; CHECK-NEXT: s_cselect_b64 s[2:3], -1, 0 +; CHECK-NEXT: s_sub_u32 s0, s0, s4 +; CHECK-NEXT: s_cselect_b64 s[6:7], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[6:7], 0 +; CHECK-NEXT: s_subb_u32 s1, s1, s5 +; CHECK-NEXT: v_mov_b32_e32 v2, s0 +; CHECK-NEXT: v_mov_b32_e32 v3, s1 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[2:3] +; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[0:1] +; CHECK-NEXT: v_readfirstlane_b32 s0, v7 +; CHECK-NEXT: v_readfirstlane_b32 s2, v6 +; CHECK-NEXT: v_mov_b32_e32 v4, s10 +; CHECK-NEXT: v_mov_b32_e32 v5, s8 ; CHECK-NEXT: s_mov_b32 s1, s0 ; CHECK-NEXT: s_mov_b32 s3, s2 ; CHECK-NEXT: flat_store_dwordx4 v[0:1], v[2:5] @@ -280,15 +267,15 @@ define amdgpu_ps <2 x i64> @s_usub_v2i64(<2 x i64> inreg %val0, <2 x i64> inreg define amdgpu_ps i64 @s_uadd_i64(i64 inreg %val0, i64 inreg %val1, ptr %ptrval) { ; CHECK-LABEL: s_uadd_i64: ; CHECK: ; %bb.0: -; CHECK-NEXT: s_add_u32 s2, s0, s2 -; CHECK-NEXT: v_mov_b32_e32 v3, s1 -; CHECK-NEXT: s_addc_u32 s3, s1, s3 +; CHECK-NEXT: s_add_u32 s0, s0, s2 +; CHECK-NEXT: s_cselect_b64 s[4:5], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[4:5], 0 +; CHECK-NEXT: s_addc_u32 s1, s1, s3 ; CHECK-NEXT: v_mov_b32_e32 v2, s0 -; CHECK-NEXT: v_mov_b32_e32 v5, s3 -; CHECK-NEXT: v_cmp_lt_u64_e32 vcc, s[2:3], v[2:3] -; CHECK-NEXT: v_mov_b32_e32 v4, s2 -; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[4:5] -; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc +; CHECK-NEXT: v_mov_b32_e32 v3, s1 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] ; CHECK-NEXT: v_readfirstlane_b32 s0, v0 ; CHECK-NEXT: s_mov_b32 s1, s0 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) @@ -305,10 +292,11 @@ define amdgpu_ps i64 @s_uadd_p1(i64 inreg %val0, i64 inreg %val1, ptr %ptrval) { ; CHECK-LABEL: s_uadd_p1: ; CHECK: ; %bb.0: ; CHECK-NEXT: s_add_u32 s0, s0, 1 +; CHECK-NEXT: s_cselect_b64 s[2:3], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[2:3], 0 ; CHECK-NEXT: s_addc_u32 s1, s1, 0 -; CHECK-NEXT: s_cmp_eq_u64 s[0:1], 0 -; CHECK-NEXT: v_mov_b32_e32 v3, s1 ; CHECK-NEXT: v_mov_b32_e32 v2, s0 +; CHECK-NEXT: v_mov_b32_e32 v3, s1 ; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 ; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[2:3] ; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] @@ -350,15 +338,15 @@ define amdgpu_ps i64 @s_uadd_n1(i64 inreg %val0, i64 inreg %val1, ptr %ptrval) { define amdgpu_ps i64 @s_usub_p1(i64 inreg %val0, i64 inreg %val1, ptr %ptrval) { ; CHECK-LABEL: s_usub_p1: ; CHECK: ; %bb.0: -; CHECK-NEXT: s_add_u32 s2, s0, -1 -; CHECK-NEXT: v_mov_b32_e32 v3, s1 -; CHECK-NEXT: s_addc_u32 s3, s1, -1 +; CHECK-NEXT: s_sub_u32 s0, s0, 1 +; CHECK-NEXT: s_cselect_b64 s[2:3], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[2:3], 0 +; CHECK-NEXT: s_subb_u32 s1, s1, 0 ; CHECK-NEXT: v_mov_b32_e32 v2, s0 -; CHECK-NEXT: v_mov_b32_e32 v5, s3 -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, s[2:3], v[2:3] -; CHECK-NEXT: v_mov_b32_e32 v4, s2 -; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[4:5] -; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc +; CHECK-NEXT: v_mov_b32_e32 v3, s1 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] ; CHECK-NEXT: v_readfirstlane_b32 s0, v0 ; CHECK-NEXT: s_mov_b32 s1, s0 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) @@ -374,15 +362,15 @@ define amdgpu_ps i64 @s_usub_p1(i64 inreg %val0, i64 inreg %val1, ptr %ptrval) { define amdgpu_ps i64 @s_usub_n1(i64 inreg %val0, i64 inreg %val1, ptr %ptrval) { ; CHECK-LABEL: s_usub_n1: ; CHECK: ; %bb.0: -; CHECK-NEXT: s_add_u32 s2, s0, 1 -; CHECK-NEXT: v_mov_b32_e32 v3, s1 -; CHECK-NEXT: s_addc_u32 s3, s1, 0 +; CHECK-NEXT: s_sub_u32 s0, s0, -1 +; CHECK-NEXT: s_cselect_b64 s[2:3], -1, 0 +; CHECK-NEXT: s_cmp_lg_u64 s[2:3], 0 +; CHECK-NEXT: s_subb_u32 s1, s1, -1 ; CHECK-NEXT: v_mov_b32_e32 v2, s0 -; CHECK-NEXT: v_mov_b32_e32 v5, s3 -; CHECK-NEXT: v_cmp_gt_u64_e32 vcc, s[2:3], v[2:3] -; CHECK-NEXT: v_mov_b32_e32 v4, s2 -; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[4:5] -; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc +; CHECK-NEXT: v_mov_b32_e32 v3, s1 +; CHECK-NEXT: s_cselect_b64 s[0:1], -1, 0 +; CHECK-NEXT: flat_store_dwordx2 v[0:1], v[2:3] +; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] ; CHECK-NEXT: v_readfirstlane_b32 s0, v0 ; CHECK-NEXT: s_mov_b32 s1, s0 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-min-agpr-alloc.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-min-agpr-alloc.ll new file mode 100644 index 0000000..f730199 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-min-agpr-alloc.ll @@ -0,0 +1,1066 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --check-globals all --version 4 +; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx90a -passes=amdgpu-attributor %s | FileCheck %s + +; Shrink result attribute list by preventing use of most attributes. +define internal void @use_most() { +; CHECK-LABEL: define internal void @use_most( +; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { +; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [256 x i8], align 1, addrspace(5) +; CHECK-NEXT: [[ALLOCA_CAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA]] to ptr +; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.amdgcn.workitem.id.z() +; CHECK-NEXT: [[TMP4:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x() +; CHECK-NEXT: [[TMP5:%.*]] = call i32 @llvm.amdgcn.workgroup.id.y() +; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.amdgcn.workgroup.id.z() +; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.amdgcn.cluster.id.x() +; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.amdgcn.cluster.id.y() +; CHECK-NEXT: [[TMP12:%.*]] = call i32 @llvm.amdgcn.cluster.id.z() +; CHECK-NEXT: [[TMP7:%.*]] = call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() +; CHECK-NEXT: [[TMP8:%.*]] = call ptr addrspace(4) @llvm.amdgcn.queue.ptr() +; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.amdgcn.dispatch.id() +; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.amdgcn.lds.kernel.id() +; CHECK-NEXT: [[IMPLICIT_ARG_PTR:%.*]] = call ptr addrspace(4) @llvm.amdgcn.implicitarg.ptr() +; CHECK-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr [[ALLOCA_CAST]], ptr addrspace(4) [[IMPLICIT_ARG_PTR]], i64 256, i1 false) +; CHECK-NEXT: ret void +; + %alloca = alloca [256 x i8], addrspace(5) + %alloca.cast = addrspacecast ptr addrspace(5) %alloca to ptr + call i32 @llvm.amdgcn.workitem.id.x() + call i32 @llvm.amdgcn.workitem.id.y() + call i32 @llvm.amdgcn.workitem.id.z() + call i32 @llvm.amdgcn.workgroup.id.x() + call i32 @llvm.amdgcn.workgroup.id.y() + call i32 @llvm.amdgcn.workgroup.id.z() + call i32 @llvm.amdgcn.cluster.id.x() + call i32 @llvm.amdgcn.cluster.id.y() + call i32 @llvm.amdgcn.cluster.id.z() + call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() + call ptr addrspace(4) @llvm.amdgcn.queue.ptr() + call i64 @llvm.amdgcn.dispatch.id() + call i32 @llvm.amdgcn.lds.kernel.id() + %implicit.arg.ptr = call ptr addrspace(4) @llvm.amdgcn.implicitarg.ptr() + call void @llvm.memcpy.p0.p4(ptr %alloca.cast, ptr addrspace(4) %implicit.arg.ptr, i64 256, i1 false) + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg( +; CHECK-SAME: ) #[[ATTR1:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(i32 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call i32 asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call i32 asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple( +; CHECK-SAME: ) #[[ATTR2:[0-9]+]] { +; CHECK-NEXT: [[DEF:%.*]] = call i64 asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call i64 asm sideeffect "; def $0", "={a[0:1]}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "v,a"(i32 poison, i32 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_non_agpr_asm() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_non_agpr_asm( +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "v"(i32 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "{a0}"(i32 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg_tuple() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_tuple( +; CHECK-SAME: ) #[[ATTR2]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) + call void @use_most() + ret void +} + +define void @func_uses_asm_virtreg_agpr() { +; CHECK-LABEL: define void @func_uses_asm_virtreg_agpr( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(i32 poison) + call void @use_most() + ret void +} + +define void @func_uses_asm_physreg_agpr() { +; CHECK-LABEL: define void @func_uses_asm_physreg_agpr( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "{a0}"(i32 poison) + call void @use_most() + ret void +} + +define void @func_uses_asm_physreg_agpr_tuple() { +; CHECK-LABEL: define void @func_uses_asm_physreg_agpr_tuple( +; CHECK-SAME: ) #[[ATTR2]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) + call void @use_most() + ret void +} + +declare void @unknown() + +define amdgpu_kernel void @kernel_calls_extern() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern( +; CHECK-SAME: ) #[[ATTR3:[0-9]+]] { +; CHECK-NEXT: call void @unknown() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @unknown() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern_marked_callsite( +; CHECK-SAME: ) #[[ATTR3]] { +; CHECK-NEXT: call void @unknown() #[[ATTR29:[0-9]+]] +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @unknown() #0 + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_indirect(ptr %indirect) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect( +; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: call void [[INDIRECT]]() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void %indirect() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_indirect_marked_callsite(ptr %indirect) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect_marked_callsite( +; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR29]] +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void %indirect() #0 + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_transitively_uses_agpr_asm() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_transitively_uses_agpr_asm( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @func_uses_asm_physreg_agpr() + call void @use_most() + ret void +} + +define void @empty() { +; CHECK-LABEL: define void @empty( +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @use_most() + ret void +} + +define void @also_empty() { +; CHECK-LABEL: define void @also_empty( +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_empty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_empty( +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void @empty() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @empty() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: call void @empty() +; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @empty() + call void @func_uses_asm_physreg_agpr() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_generic_intrinsic(ptr %ptr0, ptr %ptr1, i64 %size) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_generic_intrinsic( +; CHECK-SAME: ptr [[PTR0:%.*]], ptr [[PTR1:%.*]], i64 [[SIZE:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[PTR0]], ptr [[PTR1]], i64 [[SIZE]], i1 false) +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @llvm.memcpy.p0.p0.i64(ptr %ptr0, ptr %ptr1, i64 %size, i1 false) + call void @use_most() + ret void +} + +declare <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float, float, <32 x float>, i32 immarg, i32 immarg, i32 immarg) + +define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32(ptr addrspace(1) %out, float %a, float %b, <32 x float> %c) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32( +; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]], <32 x float> [[C:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[RESULT:%.*]] = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float [[A]], float [[B]], <32 x float> [[C]], i32 0, i32 0, i32 0) +; CHECK-NEXT: store <32 x float> [[RESULT]], ptr addrspace(1) [[OUT]], align 128 +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %result = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float %a, float %b, <32 x float> %c, i32 0, i32 0, i32 0) + store <32 x float> %result, ptr addrspace(1) %out + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_calls_workitem_id_x(ptr addrspace(1) %out) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_workitem_id_x( +; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[RESULT:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; CHECK-NEXT: store i32 [[RESULT]], ptr addrspace(1) [[OUT]], align 4 +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %result = call i32 @llvm.amdgcn.workitem.id.x() + store i32 %result, ptr addrspace(1) %out + call void @use_most() + ret void +} + +define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { +; CHECK-LABEL: define amdgpu_kernel void @indirect_calls_none_agpr( +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty +; CHECK-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty +; CHECK-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]] +; CHECK: 2: +; CHECK-NEXT: call void @also_empty() +; CHECK-NEXT: br label [[TMP6:%.*]] +; CHECK: 3: +; CHECK-NEXT: br i1 true, label [[TMP4:%.*]], label [[TMP5:%.*]] +; CHECK: 4: +; CHECK-NEXT: call void @empty() +; CHECK-NEXT: br label [[TMP6]] +; CHECK: 5: +; CHECK-NEXT: unreachable +; CHECK: 6: +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %fptr = select i1 %cond, ptr @empty, ptr @also_empty + call void %fptr() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0( +; CHECK-SAME: ) #[[ATTR2]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, i32} asm sideeffect "; def $0", "=a,=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_1() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_1( +; CHECK-SAME: ) #[[ATTR5:[0-9]+]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, <2 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, <2 x i32>} asm sideeffect "; def $0", "=a,=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2( +; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, <2 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, <2 x i32>} asm sideeffect "; def $0", "=a,=v"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty( +; CHECK-SAME: ) #[[ATTR2]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(ptr poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty( +; CHECK-SAME: ) #[[ATTR2]] { +; CHECK-NEXT: [[DEF:%.*]] = call ptr asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call ptr asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty( +; CHECK-SAME: ) #[[ATTR5]] { +; CHECK-NEXT: [[DEF:%.*]] = call <2 x ptr> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <2 x ptr> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0( +; CHECK-SAME: ) #[[ATTR6:[0-9]+]] { +; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {i32, i32} asm sideeffect "; def $0", "={a0},={a[4:5]}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber( +; CHECK-SAME: ) #[[ATTR7:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a4}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber_tuple() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_tuple( +; CHECK-SAME: ) #[[ATTR8:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a[10:13]}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber_oob() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_oob( +; CHECK-SAME: ) #[[ATTR9:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a256}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_clobber_max() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_max( +; CHECK-SAME: ) #[[ATTR9]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; clobber $0", "~{a255}"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_physreg_oob() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_oob( +; CHECK-SAME: ) #[[ATTR9]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "{a256}"(i32 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty( +; CHECK-SAME: ) #[[ATTR10:[0-9]+]] { +; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <32 x i32> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty( +; CHECK-SAME: ) #[[ATTR10]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(<32 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty( +; CHECK-SAME: ) #[[ATTR10]] { +; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <32 x i32> asm sideeffect "; use $0", "=a,a"(<32 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @vreg_use_exceeds_register_file() { +; CHECK-LABEL: define amdgpu_kernel void @vreg_use_exceeds_register_file( +; CHECK-SAME: ) #[[ATTR9]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(<257 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @vreg_def_exceeds_register_file() { +; CHECK-LABEL: define amdgpu_kernel void @vreg_def_exceeds_register_file( +; CHECK-SAME: ) #[[ATTR9]] { +; CHECK-NEXT: [[DEF:%.*]] = call <257 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <257 x i32> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @multiple() { +; CHECK-LABEL: define amdgpu_kernel void @multiple( +; CHECK-SAME: ) #[[ATTR10]] { +; CHECK-NEXT: [[DEF:%.*]] = call { <16 x i32>, <8 x i32>, <8 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call {<16 x i32>, <8 x i32>, <8 x i32>} asm sideeffect "; def $0", "=a,=a,=a,a,a,a"(<4 x i32> splat (i32 0), <8 x i32> splat (i32 1), i64 999) + call void @use_most() + ret void +} + +define amdgpu_kernel void @earlyclobber_0() { +; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_0( +; CHECK-SAME: ) #[[ATTR11:[0-9]+]] { +; CHECK-NEXT: [[DEF:%.*]] = call <8 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call <8 x i32> asm sideeffect "; def $0", "=&a,a"(i32 0) + call void @use_most() + ret void +} + +define amdgpu_kernel void @earlyclobber_1() { +; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_1( +; CHECK-SAME: ) #[[ATTR12:[0-9]+]] { +; CHECK-NEXT: [[DEF:%.*]] = call { <8 x i32>, <16 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %def = call { <8 x i32>, <16 x i32 > } asm sideeffect "; def $0, $1", "=&a,=&a,a,a"(i32 0, <16 x i32> splat (i32 1)) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512( +; CHECK-SAME: ) #[[ATTR13:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1, $2", "{a16},a,a"(i32 poison, <8 x i32> poison, <16 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512( +; CHECK-SAME: ) #[[ATTR13]] { +; CHECK-NEXT: [[TMP1:%.*]] = call { i32, <8 x i32>, <16 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call {i32, <8 x i32>, <16 x i32>} asm sideeffect "; def $0, $1, $2", "={a16},=a,=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256( +; CHECK-SAME: ) #[[ATTR14:[0-9]+]] { +; CHECK-NEXT: [[TMP1:%.*]] = call { i32, <16 x i32> } asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call {i32, <16 x i32>} asm sideeffect "; def $0, $1, $2", "={a16},=a,a"(<8 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @mixed_physreg_vreg_tuples_0() { +; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_0( +; CHECK-SAME: ) #[[ATTR11]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "{a[1:4]},a"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @mixed_physreg_vreg_tuples_1() { +; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_1( +; CHECK-SAME: ) #[[ATTR15:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,{a[0:3]}"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_raises_limit() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_raises_limit( +; CHECK-SAME: ) #[[ATTR16:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,{a[5:8]}"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @physreg_tuple_alignment_raises_limit() { +; CHECK-LABEL: define amdgpu_kernel void @physreg_tuple_alignment_raises_limit( +; CHECK-SAME: ) #[[ATTR11]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,{a[1:4]}"(<4 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @align3_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @align3_virtreg( +; CHECK-SAME: ) #[[ATTR6]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,a"(<3 x i32> poison, <3 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @align3_align4_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @align3_align4_virtreg( +; CHECK-SAME: ) #[[ATTR15]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,a"(<3 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @align2_align4_virtreg() { +; CHECK-LABEL: define amdgpu_kernel void @align2_align4_virtreg( +; CHECK-SAME: ) #[[ATTR15]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0, $1", "a,a"(<2 x i32> poison, <4 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_a55() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55( +; CHECK-SAME: ) #[[ATTR17:[0-9]+]] { +; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META0:![0-9]+]], i32 0) +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !0, i32 0) + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_v55() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_v55( +; CHECK-SAME: ) #[[ATTR0]] { +; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META1:![0-9]+]], i32 0) +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !1, i32 0) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_a55_57() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55_57( +; CHECK-SAME: ) #[[ATTR18:[0-9]+]] { +; CHECK-NEXT: call void @llvm.write_register.i96(metadata [[META2:![0-9]+]], i96 0) +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !2, i96 0) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_read_register_a55(ptr addrspace(1) %ptr) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_register_a55( +; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR19:[0-9]+]] { +; CHECK-NEXT: [[REG:%.*]] = call i32 @llvm.read_register.i32(metadata [[META0]]) +; CHECK-NEXT: store i32 [[REG]], ptr addrspace(1) [[PTR]], align 4 +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %reg = call i32 @llvm.read_register.i64(metadata !0) + store i32 %reg, ptr addrspace(1) %ptr + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_read_volatile_register_a55(ptr addrspace(1) %ptr) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_volatile_register_a55( +; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR19]] { +; CHECK-NEXT: [[REG:%.*]] = call i32 @llvm.read_volatile_register.i32(metadata [[META0]]) +; CHECK-NEXT: store i32 [[REG]], ptr addrspace(1) [[PTR]], align 4 +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %reg = call i32 @llvm.read_volatile_register.i64(metadata !0) + store i32 %reg, ptr addrspace(1) %ptr + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_read_register_a56_59(ptr addrspace(1) %ptr) { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_register_a56_59( +; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR20:[0-9]+]] { +; CHECK-NEXT: [[REG:%.*]] = call i128 @llvm.read_register.i128(metadata [[META3:![0-9]+]]) +; CHECK-NEXT: store i128 [[REG]], ptr addrspace(1) [[PTR]], align 8 +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %reg = call i128 @llvm.read_register.i64(metadata !3) + store i128 %reg, ptr addrspace(1) %ptr + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_write_register_out_of_bounds_a256() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_out_of_bounds_a256( +; CHECK-SAME: ) #[[ATTR9]] { +; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META4:![0-9]+]], i32 0) +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void @llvm.write_register.i64(metadata !4, i32 0) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_multiple_uses() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_multiple_uses( +; CHECK-SAME: ) #[[ATTR5]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(i64 poison) + call void asm sideeffect "; use $0", "a"(i32 poison) + call void asm sideeffect "; use $0", "a"(i128 poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_multiple_defs() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_multiple_defs( +; CHECK-SAME: ) #[[ATTR5]] { +; CHECK-NEXT: [[TMP1:%.*]] = call i64 asm sideeffect " +; CHECK-NEXT: [[TMP2:%.*]] = call i32 asm sideeffect " +; CHECK-NEXT: [[TMP3:%.*]] = call i128 asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call i64 asm sideeffect "; def $0", "=a"() + call i32 asm sideeffect "; def $0", "=a"() + call i128 asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_multiple_use_defs() { +; CHECK-LABEL: define amdgpu_kernel void @kernel_multiple_use_defs( +; CHECK-SAME: ) #[[ATTR5]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: [[TMP1:%.*]] = call i128 asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(i32 poison) + call i128 asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define void @callgraph_b() { +; CHECK-LABEL: define void @callgraph_b( +; CHECK-SAME: ) #[[ATTR15]] { +; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> asm sideeffect " +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call <4 x i32> asm sideeffect "; def $0", "=a"() + call void asm sideeffect "; use $0", "a"(<8 x i32> poison) + call void @use_most() + ret void +} + +define void @callgraph_c() { +; CHECK-LABEL: define void @callgraph_c( +; CHECK-SAME: ) #[[ATTR2]] { +; CHECK-NEXT: [[TMP1:%.*]] = call i32 asm sideeffect " +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call i32 asm sideeffect "; def $0", "=a"() + call void asm sideeffect "; use $0", "a"(<2 x i32> poison) + call void @use_most() + ret void +} + +define void @callgraph_a(i1 %cond) { +; CHECK-LABEL: define void @callgraph_a( +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR15]] { +; CHECK-NEXT: br i1 [[COND]], label [[A:%.*]], label [[B:%.*]] +; CHECK: a: +; CHECK-NEXT: call void @callgraph_b() +; CHECK-NEXT: ret void +; CHECK: b: +; CHECK-NEXT: call void @callgraph_c() +; CHECK-NEXT: ret void +; + br i1 %cond, label %a, label %b + +a: + call void @callgraph_b() + ret void + +b: + call void @callgraph_c() + ret void +} + + +define void @kernel_max_callgraph(i1 %cond) { +; CHECK-LABEL: define void @kernel_max_callgraph( +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR15]] { +; CHECK-NEXT: call void @callgraph_a(i1 [[COND]]) +; CHECK-NEXT: ret void +; + call void @callgraph_a(i1 %cond) + ret void +} + +define amdgpu_kernel void @kernel_uses_all_virtregs() #1 { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_all_virtregs( +; CHECK-SAME: ) #[[ATTR21:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a,a,a,a,a,a,a,a"(<32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison) + call void @use_most() + ret void +} + +define amdgpu_kernel void @kernel_uses_all_virtregs_plus_1() #1 { +; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_all_virtregs_plus_1( +; CHECK-SAME: ) #[[ATTR21]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a,a,a,a,a,a,a,a,a"(<32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, i32 poison) + call void @use_most() + ret void +} + +define void @recursive() { +; CHECK-LABEL: define void @recursive( +; CHECK-SAME: ) #[[ATTR22:[0-9]+]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: call void @recursive() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(<7 x i32> poison) + call void @use_most() + call void @recursive() + ret void +} + +define void @indirect_0() { +; CHECK-LABEL: define void @indirect_0( +; CHECK-SAME: ) #[[ATTR22]] { +; CHECK-NEXT: call void asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void asm sideeffect "; use $0", "a"(<7 x i32> poison) + call void @use_most() + ret void +} + +define void @indirect_1() { +; CHECK-LABEL: define void @indirect_1( +; CHECK-SAME: ) #[[ATTR23:[0-9]+]] { +; CHECK-NEXT: [[TMP1:%.*]] = call <3 x i32> asm sideeffect " +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call <3 x i32> asm sideeffect "; def $0", "=a"() + call void @use_most() + ret void +} + +define amdgpu_kernel void @knowable_indirect_call(i1 %cond) { +; CHECK-LABEL: define amdgpu_kernel void @knowable_indirect_call( +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR22]] { +; CHECK-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @indirect_0, ptr @indirect_1 +; CHECK-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @indirect_1 +; CHECK-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]] +; CHECK: 2: +; CHECK-NEXT: call void @indirect_1() +; CHECK-NEXT: br label [[TMP6:%.*]] +; CHECK: 3: +; CHECK-NEXT: br i1 true, label [[TMP4:%.*]], label [[TMP5:%.*]] +; CHECK: 4: +; CHECK-NEXT: call void @indirect_0() +; CHECK-NEXT: br label [[TMP6]] +; CHECK: 5: +; CHECK-NEXT: unreachable +; CHECK: 6: +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + %fptr = select i1 %cond, ptr @indirect_0, ptr @indirect_1 + call void %fptr() + call void @use_most() + ret void +} + +define amdgpu_kernel void @calls_poison(i1 %cond) { +; CHECK-LABEL: define amdgpu_kernel void @calls_poison( +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: call void poison() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void poison() + call void @use_most() + ret void +} + +define amdgpu_kernel void @calls_null(i1 %cond) { +; CHECK-LABEL: define amdgpu_kernel void @calls_null( +; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: call void null() +; CHECK-NEXT: call void @use_most() +; CHECK-NEXT: ret void +; + call void null() + call void @use_most() + ret void +} + +define amdgpu_kernel void @indirect_unknown(ptr %fptr) { +; CHECK-LABEL: define amdgpu_kernel void @indirect_unknown( +; CHECK-SAME: ptr [[FPTR:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: call void [[FPTR]]() +; CHECK-NEXT: ret void +; + call void %fptr() + ret void +} + +attributes #0 = { "amdgpu-agpr-alloc"="0" } +attributes #1 = { "amdgpu-waves-per-eu"="1,1" } + +!0 = !{!"a55"} +!1 = !{!"v55"} +!2 = !{!"a[55:57]"} +!3 = !{!"a[56:59]"} +!4 = !{!"a256"} + +;. +; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="1" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="2" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4:[0-9]+]] = { convergent nocallback nofree nosync nounwind willreturn memory(none) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="4" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="6" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { "amdgpu-agpr-alloc"="5" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { "amdgpu-agpr-alloc"="14" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR9]] = { "amdgpu-agpr-alloc"="256" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR10]] = { "amdgpu-agpr-alloc"="32" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR11]] = { "amdgpu-agpr-alloc"="9" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR12]] = { "amdgpu-agpr-alloc"="64" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR13]] = { "amdgpu-agpr-alloc"="49" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR14]] = { "amdgpu-agpr-alloc"="33" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR15]] = { "amdgpu-agpr-alloc"="8" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR16]] = { "amdgpu-agpr-alloc"="13" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR17]] = { "amdgpu-agpr-alloc"="56" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR18]] = { "amdgpu-agpr-alloc"="58" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR19]] = { "amdgpu-agpr-alloc"="56" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR20]] = { "amdgpu-agpr-alloc"="60" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR21]] = { "amdgpu-agpr-alloc"="256" "amdgpu-waves-per-eu"="1,1" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR22]] = { "amdgpu-agpr-alloc"="7" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR23]] = { "amdgpu-agpr-alloc"="3" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR24:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR25:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR26:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(read) "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR27:[0-9]+]] = { nounwind "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR28:[0-9]+]] = { nocallback nounwind "target-cpu"="gfx90a" } +; CHECK: attributes #[[ATTR29]] = { "amdgpu-agpr-alloc"="0" } +;. +; CHECK: [[META0]] = !{!"a55"} +; CHECK: [[META1]] = !{!"v55"} +; CHECK: [[META2]] = !{!"a[55:57]"} +; CHECK: [[META3]] = !{!"a[56:59]"} +; CHECK: [[META4]] = !{!"a256"} +;. diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll deleted file mode 100644 index 664dfa2..0000000 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll +++ /dev/null @@ -1,264 +0,0 @@ -; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --check-globals all --version 4 -; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx90a -passes=amdgpu-attributor %s | FileCheck %s - -define amdgpu_kernel void @kernel_uses_asm_virtreg() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg( -; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "a"(i32 poison) - ret void -} - -define amdgpu_kernel void @kernel_uses_asm_virtreg_def() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: [[DEF:%.*]] = call i32 asm sideeffect " -; CHECK-NEXT: ret void -; - %def = call i32 asm sideeffect "; def $0", "=a"() - ret void -} - -define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: [[DEF:%.*]] = call i64 asm sideeffect " -; CHECK-NEXT: ret void -; - %def = call i64 asm sideeffect "; def $0", "={a[0:1]}"() - ret void -} - -define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "v,a"(i32 poison, i32 poison) - ret void -} - -define amdgpu_kernel void @kernel_uses_non_agpr_asm() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_non_agpr_asm( -; CHECK-SAME: ) #[[ATTR1:[0-9]+]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "v"(i32 poison) - ret void -} - -define amdgpu_kernel void @kernel_uses_asm_physreg() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "{a0}"(i32 poison) - ret void -} - -define amdgpu_kernel void @kernel_uses_asm_physreg_tuple() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_tuple( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) - ret void -} - -define void @func_uses_asm_virtreg_agpr() { -; CHECK-LABEL: define void @func_uses_asm_virtreg_agpr( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "a"(i32 poison) - ret void -} - -define void @func_uses_asm_physreg_agpr() { -; CHECK-LABEL: define void @func_uses_asm_physreg_agpr( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "{a0}"(i32 poison) - ret void -} - -define void @func_uses_asm_physreg_agpr_tuple() { -; CHECK-LABEL: define void @func_uses_asm_physreg_agpr_tuple( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void asm sideeffect " -; CHECK-NEXT: ret void -; - call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) - ret void -} - -declare void @unknown() - -define amdgpu_kernel void @kernel_calls_extern() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern( -; CHECK-SAME: ) #[[ATTR2:[0-9]+]] { -; CHECK-NEXT: call void @unknown() -; CHECK-NEXT: ret void -; - call void @unknown() - ret void -} - -define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern_marked_callsite( -; CHECK-SAME: ) #[[ATTR2]] { -; CHECK-NEXT: call void @unknown() #[[ATTR6:[0-9]+]] -; CHECK-NEXT: ret void -; - call void @unknown() #0 - ret void -} - -define amdgpu_kernel void @kernel_calls_indirect(ptr %indirect) { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect( -; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR2]] { -; CHECK-NEXT: call void [[INDIRECT]]() -; CHECK-NEXT: ret void -; - call void %indirect() - ret void -} - -define amdgpu_kernel void @kernel_calls_indirect_marked_callsite(ptr %indirect) { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect_marked_callsite( -; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR2]] { -; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR6]] -; CHECK-NEXT: ret void -; - call void %indirect() #0 - ret void -} - -define amdgpu_kernel void @kernel_transitively_uses_agpr_asm() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_transitively_uses_agpr_asm( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() -; CHECK-NEXT: ret void -; - call void @func_uses_asm_physreg_agpr() - ret void -} - -define void @empty() { -; CHECK-LABEL: define void @empty( -; CHECK-SAME: ) #[[ATTR1]] { -; CHECK-NEXT: ret void -; - ret void -} - -define void @also_empty() { -; CHECK-LABEL: define void @also_empty( -; CHECK-SAME: ) #[[ATTR1]] { -; CHECK-NEXT: ret void -; - ret void -} - -define amdgpu_kernel void @kernel_calls_empty() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_empty( -; CHECK-SAME: ) #[[ATTR1]] { -; CHECK-NEXT: call void @empty() -; CHECK-NEXT: ret void -; - call void @empty() - ret void -} - -define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr() { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr( -; CHECK-SAME: ) #[[ATTR0]] { -; CHECK-NEXT: call void @empty() -; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() -; CHECK-NEXT: ret void -; - call void @empty() - call void @func_uses_asm_physreg_agpr() - ret void -} - -define amdgpu_kernel void @kernel_calls_generic_intrinsic(ptr %ptr0, ptr %ptr1, i64 %size) { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_generic_intrinsic( -; CHECK-SAME: ptr [[PTR0:%.*]], ptr [[PTR1:%.*]], i64 [[SIZE:%.*]]) #[[ATTR1]] { -; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[PTR0]], ptr [[PTR1]], i64 [[SIZE]], i1 false) -; CHECK-NEXT: ret void -; - call void @llvm.memcpy.p0.p0.i64(ptr %ptr0, ptr %ptr1, i64 %size, i1 false) - ret void -} - -declare <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float, float, <32 x float>, i32 immarg, i32 immarg, i32 immarg) - -define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32(ptr addrspace(1) %out, float %a, float %b, <32 x float> %c) { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32( -; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]], <32 x float> [[C:%.*]]) #[[ATTR1]] { -; CHECK-NEXT: [[RESULT:%.*]] = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float [[A]], float [[B]], <32 x float> [[C]], i32 0, i32 0, i32 0) -; CHECK-NEXT: store <32 x float> [[RESULT]], ptr addrspace(1) [[OUT]], align 128 -; CHECK-NEXT: ret void -; - %result = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float %a, float %b, <32 x float> %c, i32 0, i32 0, i32 0) - store <32 x float> %result, ptr addrspace(1) %out - ret void -} - -define amdgpu_kernel void @kernel_calls_workitem_id_x(ptr addrspace(1) %out) { -; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_workitem_id_x( -; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR1]] { -; CHECK-NEXT: [[RESULT:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() -; CHECK-NEXT: store i32 [[RESULT]], ptr addrspace(1) [[OUT]], align 4 -; CHECK-NEXT: ret void -; - %result = call i32 @llvm.amdgcn.workitem.id.x() - store i32 %result, ptr addrspace(1) %out - ret void -} - -define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { -; CHECK-LABEL: define amdgpu_kernel void @indirect_calls_none_agpr( -; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty -; CHECK-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty -; CHECK-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]] -; CHECK: 2: -; CHECK-NEXT: call void @also_empty() -; CHECK-NEXT: br label [[TMP6:%.*]] -; CHECK: 3: -; CHECK-NEXT: br i1 true, label [[TMP4:%.*]], label [[TMP5:%.*]] -; CHECK: 4: -; CHECK-NEXT: call void @empty() -; CHECK-NEXT: br label [[TMP6]] -; CHECK: 5: -; CHECK-NEXT: unreachable -; CHECK: 6: -; CHECK-NEXT: ret void -; - %fptr = select i1 %cond, ptr @empty, ptr @also_empty - call void %fptr() - ret void -} - - -attributes #0 = { "amdgpu-agpr-alloc"="0" } -;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2:[0-9]+]] = { "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3:[0-9]+]] = { convergent nocallback nofree nosync nounwind willreturn memory(none) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR5:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="0" } -;. diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll index fb566e5..9283bd5 100644 --- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll +++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll @@ -691,29 +691,29 @@ attributes #6 = { "enqueued-block" } ;. ; ATTRIBUTOR_HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; ATTRIBUTOR_HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR14]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR15:[0-9]+]] = { nounwind "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR16]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR17]] = { nounwind sanitize_address "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR18]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="fiji" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR14]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR15]] = { nounwind "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR16]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR17]] = { nounwind sanitize_address "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR18]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR19:[0-9]+]] = { nounwind sanitize_address "amdgpu-no-implicitarg-ptr" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR20:[0-9]+]] = { "enqueued-block" "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR21]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "enqueued-block" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR21]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "enqueued-block" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR22]] = { "uniform-work-group-size"="false" } -; ATTRIBUTOR_HSA: attributes #[[ATTR23]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_HSA: attributes #[[ATTR23]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_HSA: attributes #[[ATTR24]] = { nounwind } ; ATTRIBUTOR_HSA: attributes #[[ATTR25]] = { "enqueued-block" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll index 484ff77..8554485 100644 --- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll +++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll @@ -474,19 +474,19 @@ attributes #1 = { nounwind } ; AKF_HSA: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 500} ;. ; HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR3]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR4]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR5]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR6]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR7]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR8]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR9]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR10]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR11]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR12]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; HSA: attributes #[[ATTR13]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. ; HSA: [[META0]] = !{i32 1, i32 3, i32 4, i32 10} ; HSA: [[META1]] = !{i32 1, i32 5, i32 6, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll index 2efe024..e2a2deb 100644 --- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll +++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features.ll @@ -294,13 +294,13 @@ attributes #1 = { nounwind } ;. ; CHECK: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; CHECK: attributes #[[ATTR1]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR6]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR7]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR8]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR9]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR9]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workitem-id-x" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll b/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll index aaedb85..e67d7fdb 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_load_local.ll @@ -3,6 +3,8 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-TRUE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-FAKE16 %s define i8 @atomic_load_monotonic_i8(ptr addrspace(3) %ptr) { ; CI-LABEL: atomic_load_monotonic_i8: @@ -33,6 +35,14 @@ define i8 @atomic_load_monotonic_i8(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u8 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i8: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u8 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i8, ptr addrspace(3) %ptr monotonic, align 1 ret i8 %load } @@ -66,6 +76,14 @@ define i8 @atomic_load_monotonic_i8_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u8 v0, v0 offset:16 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i8_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u8 v0, v0 offset:16 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 16 %load = load atomic i8, ptr addrspace(3) %gep monotonic, align 1 ret i8 %load @@ -100,6 +118,14 @@ define i16 @atomic_load_monotonic_i16(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i16: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i16, ptr addrspace(3) %ptr monotonic, align 2 ret i16 %load } @@ -133,6 +159,14 @@ define i16 @atomic_load_monotonic_i16_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i16_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 offset:32 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i16, ptr addrspace(3) %ptr, i16 16 %load = load atomic i16, ptr addrspace(3) %gep monotonic, align 2 ret i16 %load @@ -160,6 +194,14 @@ define i32 @atomic_load_monotonic_i32(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b32 v0, v0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i32, ptr addrspace(3) %ptr monotonic, align 4 ret i32 %load } @@ -186,6 +228,14 @@ define i32 @atomic_load_monotonic_i32_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b32 v0, v0 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i32_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i32, ptr addrspace(3) %ptr, i32 16 %load = load atomic i32, ptr addrspace(3) %gep monotonic, align 4 ret i32 %load @@ -213,6 +263,14 @@ define i64 @atomic_load_monotonic_i64(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i64: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic i64, ptr addrspace(3) %ptr monotonic, align 8 ret i64 %load } @@ -239,6 +297,14 @@ define i64 @atomic_load_monotonic_i64_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_i64_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i64, ptr addrspace(3) %ptr, i32 16 %load = load atomic i64, ptr addrspace(3) %gep monotonic, align 8 ret i64 %load @@ -266,6 +332,14 @@ define float @atomic_load_monotonic_f32_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b32 v0, v0 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f32_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds float, ptr addrspace(3) %ptr, i32 16 %load = load atomic float, ptr addrspace(3) %gep monotonic, align 4 ret float %load @@ -293,6 +367,14 @@ define double @atomic_load_monotonic_f64_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f64_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds double, ptr addrspace(3) %ptr, i32 16 %load = load atomic double, ptr addrspace(3) %gep monotonic, align 8 ret double %load @@ -320,6 +402,14 @@ define ptr @atomic_load_monotonic_p0i8_offset(ptr addrspace(3) %ptr) { ; GFX11-NEXT: ds_load_b64 v[0:1], v0 offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_p0i8_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b64 v[0:1], v0 offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds ptr, ptr addrspace(3) %ptr, i32 16 %load = load atomic ptr, ptr addrspace(3) %gep monotonic, align 8 ret ptr %load @@ -347,6 +437,14 @@ define ptr addrspace(3) @atomic_load_monotonic_p3i8_offset(ptr addrspace(3) %ptr ; GFX11-NEXT: ds_load_b32 v0, v0 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_p3i8_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_b32 v0, v0 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds ptr addrspace(3), ptr addrspace(3) %ptr, i32 16 %load = load atomic ptr addrspace(3), ptr addrspace(3) %gep monotonic, align 4 ret ptr addrspace(3) %load @@ -381,6 +479,14 @@ define i16 @atomic_load_monotonic_f16(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f16: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic half, ptr addrspace(3) %ptr monotonic, align 2 %ret = bitcast half %load to i16 ret i16 %ret @@ -415,6 +521,14 @@ define i16 @atomic_load_monotonic_f16_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_f16_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 offset:32 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds half, ptr addrspace(3) %ptr, i32 16 %load = load atomic half, ptr addrspace(3) %gep monotonic, align 2 %ret = bitcast half %load to i16 @@ -450,6 +564,14 @@ define i16 @atomic_load_monotonic_bf16(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_bf16: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %load = load atomic bfloat, ptr addrspace(3) %ptr monotonic, align 2 %ret = bitcast bfloat %load to i16 ret i16 %ret @@ -484,6 +606,14 @@ define i16 @atomic_load_monotonic_bf16_offset(ptr addrspace(3) %ptr) { ; GFX11-FAKE16-NEXT: ds_load_u16 v0, v0 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_load_monotonic_bf16_offset: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_load_u16 v0, v0 offset:32 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds bfloat, ptr addrspace(3) %ptr, i32 16 %load = load atomic bfloat, ptr addrspace(3) %gep monotonic, align 2 %ret = bitcast bfloat %load to i16 @@ -491,3 +621,5 @@ define i16 @atomic_load_monotonic_bf16_offset(ptr addrspace(3) %ptr) { } ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: ; GCN: {{.*}} +; GFX1250-FAKE16: {{.*}} +; GFX1250-TRUE16: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll b/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll index c2bb4f00..31065f2 100644 --- a/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll +++ b/llvm/test/CodeGen/AMDGPU/atomic_store_local.ll @@ -3,6 +3,8 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-TRUE16 %s +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX1250,GFX1250-FAKE16 %s define void @atomic_store_monotonic_i8(ptr addrspace(3) %ptr, i8 %val) { ; CI-LABEL: atomic_store_monotonic_i8: @@ -41,6 +43,26 @@ define void @atomic_store_monotonic_i8(ptr addrspace(3) %ptr, i8 %val) { ; GFX11-FAKE16-NEXT: ds_store_b8 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_i8: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b8 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b8_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_i8: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i8 %val, 2 store atomic i8 %val, ptr addrspace(3) %ptr monotonic, align 1 store atomic i8 %val1, ptr addrspace(3) %ptr monotonic, align 1 @@ -84,6 +106,26 @@ define void @atomic_store_monotonic_offset_i8(ptr addrspace(3) %ptr, i8 %val) { ; GFX11-FAKE16-NEXT: ds_store_b8 v0, v2 offset:16 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_i8: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b8 v0, v1 offset:8 +; GFX1250-TRUE16-NEXT: ds_store_b8_d16_hi v0, v1 offset:16 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_i8: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v1 offset:8 +; GFX1250-FAKE16-NEXT: ds_store_b8 v0, v2 offset:16 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i8 %val, 2 %gep_1 = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 8 %gep_2 = getelementptr inbounds i8, ptr addrspace(3) %ptr, i8 16 @@ -129,6 +171,26 @@ define void @atomic_store_monotonic_i16(ptr addrspace(3) %ptr, i16 %val) { ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_i16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_i16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i16 %val, 2 store atomic i16 %val, ptr addrspace(3) %ptr monotonic, align 2 store atomic i16 %val1, ptr addrspace(3) %ptr monotonic, align 2 @@ -172,6 +234,26 @@ define void @atomic_store_monotonic_offset_i16(ptr addrspace(3) %ptr, i16 %val) ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_i16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_i16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %val1 = add i16 %val, 2 %gep = getelementptr inbounds i16, ptr addrspace(3) %ptr, i16 16 store atomic i16 %val, ptr addrspace(3) %gep monotonic, align 2 @@ -201,6 +283,14 @@ define void @atomic_store_monotonic_i32(ptr addrspace(3) %ptr, i32 %val) { ; GFX11-NEXT: ds_store_b32 v0, v1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_i32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_store_b32 v0, v1 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] store atomic i32 %val, ptr addrspace(3) %ptr monotonic, align 4 ret void } @@ -227,6 +317,14 @@ define void @atomic_store_monotonic_offset_i32(ptr addrspace(3) %ptr, i32 %val) ; GFX11-NEXT: ds_store_b32 v0, v1 offset:64 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_offset_i32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: ds_store_b32 v0, v1 offset:64 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i32, ptr addrspace(3) %ptr, i32 16 store atomic i32 %val, ptr addrspace(3) %gep monotonic, align 4 ret void @@ -254,6 +352,15 @@ define void @atomic_store_monotonic_i64(ptr addrspace(3) %ptr, i64 %val) { ; GFX11-NEXT: ds_store_b64 v0, v[1:2] ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_i64: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 +; GFX1250-NEXT: ds_store_b64 v0, v[2:3] +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] store atomic i64 %val, ptr addrspace(3) %ptr monotonic, align 8 ret void } @@ -280,6 +387,15 @@ define void @atomic_store_monotonic_offset_i64(ptr addrspace(3) %ptr, i64 %val) ; GFX11-NEXT: ds_store_b64 v0, v[1:2] offset:128 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: atomic_store_monotonic_offset_i64: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 +; GFX1250-NEXT: ds_store_b64 v0, v[2:3] offset:128 +; GFX1250-NEXT: s_wait_dscnt 0x0 +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %gep = getelementptr inbounds i64, ptr addrspace(3) %ptr, i64 16 store atomic i64 %val, ptr addrspace(3) %gep monotonic, align 8 ret void @@ -322,6 +438,26 @@ define void @atomic_store_monotonic_f16(ptr addrspace(3) %ptr, i16 %arg.val) { ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_f16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_f16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val = bitcast i16 %arg.val to half %val1 = bitcast i16 %arg.val1 to half @@ -367,6 +503,26 @@ define void @atomic_store_monotonic_offset_f16(ptr addrspace(3) %ptr, i16 %arg.v ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_f16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_f16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val1 = bitcast i16 %arg.val1 to half %val = bitcast i16 %arg.val to half @@ -413,6 +569,26 @@ define void @atomic_store_monotonic_bf16(ptr addrspace(3) %ptr, i16 %arg.val) { ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_bf16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_bf16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val1 = bitcast i16 %arg.val1 to bfloat %val = bitcast i16 %arg.val to bfloat @@ -458,6 +634,26 @@ define void @atomic_store_monotonic_offset_bf16(ptr addrspace(3) %ptr, i16 %arg. ; GFX11-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-TRUE16-LABEL: atomic_store_monotonic_offset_bf16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: v_add_nc_u16 v1.h, v1.l, 2 +; GFX1250-TRUE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: ds_store_b16_d16_hi v0, v1 offset:32 +; GFX1250-TRUE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250-FAKE16-LABEL: atomic_store_monotonic_offset_bf16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: v_add_nc_u16 v2, v1, 2 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v1 offset:32 +; GFX1250-FAKE16-NEXT: ds_store_b16 v0, v2 offset:32 +; GFX1250-FAKE16-NEXT: s_wait_dscnt 0x0 +; GFX1250-FAKE16-NEXT: s_set_pc_i64 s[30:31] %arg.val1 = add i16 %arg.val, 2 %val1 = bitcast i16 %arg.val1 to bfloat %val = bitcast i16 %arg.val to bfloat diff --git a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll index f63dd6e..c90611f 100644 --- a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll +++ b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit-undefined-behavior.ll @@ -147,10 +147,10 @@ define amdgpu_kernel void @call_calls_intrin_ascast_cc_kernel(ptr addrspace(3) % attributes #0 = { "amdgpu-no-flat-scratch-init" } ;. -; GFX9: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } ; GFX9: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx900" } ;. -; GFX10: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } ; GFX10: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1010" } ;. ; GFX9: [[META0]] = !{i32 1, i32 5, i32 6, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll index 60cd252..c005695a 100644 --- a/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll +++ b/llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll @@ -723,7 +723,7 @@ define void @also_empty() { define amdgpu_kernel void @indirect_call_known_callees(i1 %cond) { ; GFX9-LABEL: define amdgpu_kernel void @indirect_call_known_callees( -; GFX9-SAME: i1 [[COND:%.*]]) #[[ATTR3:[0-9]+]] { +; GFX9-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { ; GFX9-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty ; GFX9-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty ; GFX9-NEXT: br i1 [[TMP1]], label %[[BB2:.*]], label %[[BB3:.*]] @@ -741,7 +741,7 @@ define amdgpu_kernel void @indirect_call_known_callees(i1 %cond) { ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define amdgpu_kernel void @indirect_call_known_callees( -; GFX10-SAME: i1 [[COND:%.*]]) #[[ATTR3:[0-9]+]] { +; GFX10-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { ; GFX10-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty ; GFX10-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty ; GFX10-NEXT: br i1 [[TMP1]], label %[[BB2:.*]], label %[[BB3:.*]] @@ -767,13 +767,13 @@ declare i32 @llvm.amdgcn.workgroup.id.x() define void @use_intrinsic_workitem_id_x() { ; GFX9-LABEL: define void @use_intrinsic_workitem_id_x( -; GFX9-SAME: ) #[[ATTR5:[0-9]+]] { +; GFX9-SAME: ) #[[ATTR4:[0-9]+]] { ; GFX9-NEXT: [[VAL:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() ; GFX9-NEXT: store volatile i32 [[VAL]], ptr addrspace(1) null, align 4 ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define void @use_intrinsic_workitem_id_x( -; GFX10-SAME: ) #[[ATTR5:[0-9]+]] { +; GFX10-SAME: ) #[[ATTR4:[0-9]+]] { ; GFX10-NEXT: [[VAL:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() ; GFX10-NEXT: store volatile i32 [[VAL]], ptr addrspace(1) null, align 4 ; GFX10-NEXT: ret void @@ -803,12 +803,12 @@ define amdgpu_kernel void @use_intrinsic_workitem_id_x_cc_kernel() { define void @call_use_intrinsic_workitem_id_x() { ; GFX9-LABEL: define void @call_use_intrinsic_workitem_id_x( -; GFX9-SAME: ) #[[ATTR5]] { +; GFX9-SAME: ) #[[ATTR4]] { ; GFX9-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define void @call_use_intrinsic_workitem_id_x( -; GFX10-SAME: ) #[[ATTR5]] { +; GFX10-SAME: ) #[[ATTR4]] { ; GFX10-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX10-NEXT: ret void ; @@ -818,12 +818,12 @@ define void @call_use_intrinsic_workitem_id_x() { define amdgpu_kernel void @call_use_intrinsic_workitem_id_x_cc_kernel() { ; GFX9-LABEL: define amdgpu_kernel void @call_use_intrinsic_workitem_id_x_cc_kernel( -; GFX9-SAME: ) #[[ATTR5]] { +; GFX9-SAME: ) #[[ATTR4]] { ; GFX9-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define amdgpu_kernel void @call_use_intrinsic_workitem_id_x_cc_kernel( -; GFX10-SAME: ) #[[ATTR5]] { +; GFX10-SAME: ) #[[ATTR4]] { ; GFX10-NEXT: call void @use_intrinsic_workitem_id_x() ; GFX10-NEXT: ret void ; @@ -851,12 +851,12 @@ define amdgpu_kernel void @calls_intrin_ascast_cc_kernel(ptr addrspace(3) %ptr) define amdgpu_kernel void @with_inline_asm() { ; GFX9-LABEL: define amdgpu_kernel void @with_inline_asm( -; GFX9-SAME: ) #[[ATTR3]] { +; GFX9-SAME: ) #[[ATTR0]] { ; GFX9-NEXT: call void asm sideeffect " ; GFX9-NEXT: ret void ; ; GFX10-LABEL: define amdgpu_kernel void @with_inline_asm( -; GFX10-SAME: ) #[[ATTR3]] { +; GFX10-SAME: ) #[[ATTR0]] { ; GFX10-NEXT: call void asm sideeffect " ; GFX10-NEXT: ret void ; @@ -865,19 +865,17 @@ define amdgpu_kernel void @with_inline_asm() { } ;. -; GFX9: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; GFX9: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } ; GFX9: attributes #[[ATTR2]] = { "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; GFX9: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } -; GFX9: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx900" } -; GFX9: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } +; GFX9: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx900" } +; GFX9: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx900" "uniform-work-group-size"="false" } ;. -; GFX10: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } -; GFX10: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } ; GFX10: attributes #[[ATTR2]] = { "target-cpu"="gfx1010" "uniform-work-group-size"="false" } -; GFX10: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } -; GFX10: attributes #[[ATTR4:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1010" } -; GFX10: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } +; GFX10: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx1010" } +; GFX10: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx1010" "uniform-work-group-size"="false" } ;. ; GFX9: [[META0]] = !{i32 2, i32 10} ; GFX9: [[META1]] = !{i32 1, i32 2, i32 3, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/bf16.ll b/llvm/test/CodeGen/AMDGPU/bf16.ll index 6b5647e..4b14dc6 100644 --- a/llvm/test/CodeGen/AMDGPU/bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/bf16.ll @@ -7,11 +7,9 @@ ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1010 | FileCheck %s -check-prefixes=GFX10 ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 | FileCheck %s -check-prefixes=GFX11,GFX11TRUE16 ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 | FileCheck %s -check-prefixes=GFX11,GFX11FAKE16 -; xUN: llc < %s -mtriple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 | FileCheck %s -check-prefixes=GFX1250,GFX1250TRUE16 +; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1250 -mattr=+real-true16 | FileCheck %s -check-prefixes=GFX1250,GFX1250TRUE16 ; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1250 -mattr=-real-true16 | FileCheck %s -check-prefixes=GFX1250,GFX1250FAKE16 -; FIXME: real-true16 version of gfx1250 test fails - define void @test_load_store(ptr addrspace(1) %in, ptr addrspace(1) %out) { ; GCN-LABEL: test_load_store: ; GCN: ; %bb.0: @@ -2393,15 +2391,25 @@ define void @test_store_fpimm(ptr addrspace(1) %ptr0, ptr addrspace(1) %ptr1) { ; GFX11FAKE16-NEXT: global_store_b16 v[2:3], v5, off ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_store_fpimm: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_mov_b32_e32 v4, 0x3f80 -; GFX1250-NEXT: v_mov_b32_e32 v5, 0x4228 -; GFX1250-NEXT: global_store_b16 v[0:1], v4, off -; GFX1250-NEXT: global_store_b16 v[2:3], v5, off -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_store_fpimm: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v4.l, 0x3f80 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v4.h, 0x4228 +; GFX1250TRUE16-NEXT: global_store_b16 v[0:1], v4, off +; GFX1250TRUE16-NEXT: global_store_d16_hi_b16 v[2:3], v4, off +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_store_fpimm: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v4, 0x3f80 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v5, 0x4228 +; GFX1250FAKE16-NEXT: global_store_b16 v[0:1], v4, off +; GFX1250FAKE16-NEXT: global_store_b16 v[2:3], v5, off +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] store bfloat 1.0, ptr addrspace(1) %ptr0 store bfloat 42.0, ptr addrspace(1) %ptr1 ret void @@ -3796,13 +3804,21 @@ define amdgpu_gfx void @test_inreg_arg_store(bfloat inreg %in, ptr addrspace(1) ; GFX11FAKE16-NEXT: global_store_b16 v[0:1], v2, off ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_inreg_arg_store: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_mov_b32_e32 v2, s4 -; GFX1250-NEXT: global_store_b16 v[0:1], v2, off -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_inreg_arg_store: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, s4 +; GFX1250TRUE16-NEXT: global_store_b16 v[0:1], v2, off +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_inreg_arg_store: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v2, s4 +; GFX1250FAKE16-NEXT: global_store_b16 v[0:1], v2, off +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] store bfloat %in, ptr addrspace(1) %out ret void } @@ -3866,12 +3882,20 @@ define bfloat @test_byval(ptr addrspace(5) byval(bfloat) %bv, bfloat %val) { ; GFX11FAKE16-NEXT: scratch_store_b16 off, v0, s32 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_byval: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: scratch_store_b16 off, v0, s32 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_byval: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l +; GFX1250TRUE16-NEXT: scratch_store_b16 off, v1, s32 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_byval: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: scratch_store_b16 off, v0, s32 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] store bfloat %val, ptr addrspace(5) %bv %retval = load bfloat, ptr addrspace(5) %bv ret bfloat %retval @@ -6708,27 +6732,50 @@ define { <32 x i32>, bfloat } @test_overflow_stack(bfloat %a, <32 x i32> %b) { ; GFX11FAKE16-NEXT: scratch_store_b16 v0, v1, off offset:128 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: test_overflow_stack: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x2 -; GFX1250-NEXT: scratch_load_b32 v33, off, s32 offset:8 -; GFX1250-NEXT: scratch_load_b32 v32, off, s32 offset:4 -; GFX1250-NEXT: scratch_load_b32 v31, off, s32 -; GFX1250-NEXT: s_clause 0x5 -; GFX1250-NEXT: scratch_store_b128 v0, v[22:25], off offset:80 -; GFX1250-NEXT: scratch_store_b128 v0, v[18:21], off offset:64 -; GFX1250-NEXT: scratch_store_b128 v0, v[14:17], off offset:48 -; GFX1250-NEXT: scratch_store_b128 v0, v[10:13], off offset:32 -; GFX1250-NEXT: scratch_store_b128 v0, v[6:9], off offset:16 -; GFX1250-NEXT: scratch_store_b128 v0, v[2:5], off -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: s_clause 0x2 -; GFX1250-NEXT: scratch_store_b128 v0, v[30:33], off offset:112 -; GFX1250-NEXT: scratch_store_b128 v0, v[26:29], off offset:96 -; GFX1250-NEXT: scratch_store_b16 v0, v1, off offset:128 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: test_overflow_stack: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: s_clause 0x2 +; GFX1250TRUE16-NEXT: scratch_load_b32 v33, off, s32 offset:8 +; GFX1250TRUE16-NEXT: scratch_load_b32 v32, off, s32 offset:4 +; GFX1250TRUE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250TRUE16-NEXT: s_clause 0x3 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[22:25], off offset:80 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[18:21], off offset:64 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[14:17], off offset:48 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[10:13], off offset:32 +; GFX1250TRUE16-NEXT: s_clause 0x1 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[6:9], off offset:16 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[2:5], off +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250TRUE16-NEXT: s_clause 0x2 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[30:33], off offset:112 +; GFX1250TRUE16-NEXT: scratch_store_b128 v0, v[26:29], off offset:96 +; GFX1250TRUE16-NEXT: scratch_store_b16 v0, v1, off offset:128 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: test_overflow_stack: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: s_clause 0x2 +; GFX1250FAKE16-NEXT: scratch_load_b32 v33, off, s32 offset:8 +; GFX1250FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:4 +; GFX1250FAKE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250FAKE16-NEXT: s_clause 0x5 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[22:25], off offset:80 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[18:21], off offset:64 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[14:17], off offset:48 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[10:13], off offset:32 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[6:9], off offset:16 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[2:5], off +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250FAKE16-NEXT: s_clause 0x2 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[30:33], off offset:112 +; GFX1250FAKE16-NEXT: scratch_store_b128 v0, v[26:29], off offset:96 +; GFX1250FAKE16-NEXT: scratch_store_b16 v0, v1, off offset:128 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %ins.0 = insertvalue { <32 x i32>, bfloat } poison, <32 x i32> %b, 0 %ins.1 = insertvalue { <32 x i32>, bfloat } %ins.0 ,bfloat %a, 1 ret { <32 x i32>, bfloat } %ins.1 @@ -10726,15 +10773,29 @@ define bfloat @v_fadd_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fadd_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fadd_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fadd_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fadd bfloat %a, %b ret bfloat %op } @@ -15268,15 +15329,26 @@ define bfloat @v_fadd_bf16_fpimm_0(bfloat %arg0) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fadd_bf16_fpimm_0: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v0, 1.0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fadd_bf16_fpimm_0: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, 1.0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fadd_bf16_fpimm_0: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, 1.0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %add = fadd bfloat %arg0, 1.0 ret bfloat %add } @@ -15382,15 +15454,26 @@ define bfloat @v_fadd_bf16_fpimm_1(bfloat %arg0) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fadd_bf16_fpimm_1: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v0, 0x42280000, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fadd_bf16_fpimm_1: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, 0x42280000, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fadd_bf16_fpimm_1: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, 0x42280000, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %add = fadd bfloat %arg0, 42.0 ret bfloat %add } @@ -15507,15 +15590,29 @@ define bfloat @v_fsub_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fsub_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fsub_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fsub_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fsub bfloat %a, %b ret bfloat %op } @@ -15931,21 +16028,37 @@ define <3 x bfloat> @v_fsub_v3bf16(<3 x bfloat> %a, <3 x bfloat> %b) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fsub_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v3, 16, v3 -; GFX1250-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 -; GFX1250-NEXT: v_and_b32_e32 v5, 0xffff0000, v0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_dual_sub_f32 v4, v5, v4 :: v_dual_lshlrev_b32 v1, 16, v1 -; GFX1250-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v4 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fsub_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v3, 16, v3 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v1, v1, v3 +; GFX1250TRUE16-NEXT: v_dual_sub_f32 v3, v5, v4 :: v_dual_sub_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v3 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fsub_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v4, 0xffff0000, v2 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_dual_sub_f32 v4, v5, v4 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250FAKE16-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v4 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fsub <3 x bfloat> %a, %b ret <3 x bfloat> %op } @@ -16371,12 +16484,26 @@ define bfloat @v_fmul_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fmul_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_fma_mixlo_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fmul_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fmul_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_fma_mixlo_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fmul bfloat %a, %b ret bfloat %op } @@ -21012,31 +21139,60 @@ define bfloat @v_fdiv_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fdiv_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_div_scale_f32 v2, null, v1, v1, v0 -; GFX1250-NEXT: v_rcp_f32_e32 v3, v2 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fma_f32 v4, -v2, v3, 1.0 -; GFX1250-NEXT: v_fmac_f32_e32 v3, v4, v3 -; GFX1250-NEXT: v_div_scale_f32 v4, vcc_lo, v0, v1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX1250-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fmac_f32_e32 v5, v6, v3 -; GFX1250-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX1250-NEXT: v_div_fixup_f32 v0, v2, v1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fdiv_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v0.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, v2.l +; GFX1250TRUE16-NEXT: v_div_scale_f32 v1, null, v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_rcp_f32_e32 v3, v1 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_fma_f32 v4, -v1, v3, 1.0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fmac_f32_e32 v3, v4, v3 +; GFX1250TRUE16-NEXT: v_div_scale_f32 v4, vcc_lo, v2, v0, v2 +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v6, -v1, v5, v4 +; GFX1250TRUE16-NEXT: v_fmac_f32_e32 v5, v6, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v1, -v1, v5, v4 +; GFX1250TRUE16-NEXT: v_div_fmas_f32 v1, v1, v3, v5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_div_fixup_f32 v0, v1, v0, v2 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fdiv_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_div_scale_f32 v2, null, v1, v1, v0 +; GFX1250FAKE16-NEXT: v_rcp_f32_e32 v3, v2 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fma_f32 v4, -v2, v3, 1.0 +; GFX1250FAKE16-NEXT: v_fmac_f32_e32 v3, v4, v3 +; GFX1250FAKE16-NEXT: v_div_scale_f32 v4, vcc_lo, v0, v1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX1250FAKE16-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fmac_f32_e32 v5, v6, v3 +; GFX1250FAKE16-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX1250FAKE16-NEXT: v_div_fixup_f32 v0, v2, v1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fdiv bfloat %a, %b ret bfloat %op } @@ -21092,12 +21248,19 @@ define bfloat @v_fabs_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fabs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0x7fff, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fabs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 0x7fff, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fabs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.fabs.bf16(bfloat %a) ret bfloat %op } @@ -21198,12 +21361,19 @@ define bfloat @v_fneg_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_xor_b32_e32 v0, 0x8000, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fneg_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_xor_b32_e32 v0, 0x8000, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fneg_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fneg_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v0, 0x8000, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fneg bfloat %a ret bfloat %op } @@ -21317,12 +21487,19 @@ define bfloat @v_fneg_fabs_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_or_b32_e32 v0, 0x8000, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fneg_fabs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_or_b32_e32 v0, 0x8000, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fneg_fabs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_or_b16 v0.l, 0x8000, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fneg_fabs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_or_b32_e32 v0, 0x8000, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %fabs = call bfloat @llvm.fabs.bf16(bfloat %a) %op = fneg bfloat %fabs ret bfloat %op @@ -21511,15 +21688,29 @@ define bfloat @v_minnum_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_minnum_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_min_num_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_minnum_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_min_num_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_minnum_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_min_num_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.minnum.bf16(bfloat %a, bfloat %b) ret bfloat %op } @@ -26073,15 +26264,29 @@ define bfloat @v_maxnum_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_maxnum_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_max_num_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_maxnum_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_max_num_f32_e32 v0, v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_maxnum_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_max_num_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.maxnum.bf16(bfloat %a, bfloat %b) ret bfloat %op } @@ -30764,12 +30969,19 @@ define bfloat @v_sqrt_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sqrt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_sqrt_bf16_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sqrt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_sqrt_bf16_e32 v0.l, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sqrt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_sqrt_bf16_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.sqrt.bf16(bfloat %a) ret bfloat %op } @@ -30877,15 +31089,26 @@ define bfloat @v_ldexp_bf16_i32(bfloat %a, i32 %b) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_ldexp_bf16_i32: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_ldexp_bf16_i32: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v2, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_ldexp_bf16_i32: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.ldexp.bf16.i32(bfloat %a, i32 %b) ret bfloat %op } @@ -31005,16 +31228,28 @@ define { bfloat, i16 } @v_frexp_bf16_i16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_frexp_bf16_i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_frexp_mant_f32_e32 v0, v1 -; GFX1250-NEXT: v_frexp_exp_i32_f32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_frexp_bf16_i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_frexp_mant_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_frexp_exp_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_frexp_bf16_i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_frexp_mant_f32_e32 v0, v1 +; GFX1250FAKE16-NEXT: v_frexp_exp_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call { bfloat, i16 } @llvm.frexp.bf16.i16(bfloat %a) ret { bfloat, i16 } %op } @@ -31254,31 +31489,58 @@ define bfloat @v_log_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_log_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) -; GFX1250-NEXT: v_log_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: v_mul_f32_e32 v1, 0x3f317217, v0 -; GFX1250-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fma_f32 v2, 0x3f317217, v0, -v1 -; GFX1250-NEXT: v_fmamk_f32 v2, v0, 0x3377d1cf, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v1, v1, v2 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 0x41b17218, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_log_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 32, vcc_lo +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v1, 0x3f317217, v0 +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v2, 0x3f317217, v0, -v1 +; GFX1250TRUE16-NEXT: v_fmamk_f32 v2, v0, 0x3377d1cf, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x41b17218, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_log_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250FAKE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v1, 0x3f317217, v0 +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fma_f32 v2, 0x3f317217, v0, -v1 +; GFX1250FAKE16-NEXT: v_fmamk_f32 v2, v0, 0x3377d1cf, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x41b17218, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.log.bf16(bfloat %a) ret bfloat %op } @@ -31439,12 +31701,19 @@ define bfloat @v_log2_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_log2_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_log_bf16_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_log2_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_log_bf16_e32 v0.l, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_log2_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_log_bf16_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.log2.bf16(bfloat %a) ret bfloat %op } @@ -31679,31 +31948,58 @@ define bfloat @v_log10_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_log10_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) -; GFX1250-NEXT: v_log_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: v_mul_f32_e32 v1, 0x3e9a209a, v0 -; GFX1250-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_fma_f32 v2, 0x3e9a209a, v0, -v1 -; GFX1250-NEXT: v_fmamk_f32 v2, v0, 0x3284fbcf, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_f32_e32 v1, v1, v2 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 -; GFX1250-NEXT: v_cndmask_b32_e64 v1, 0, 0x411a209b, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_f32_e32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_log10_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 32, vcc_lo +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v1, 0x3e9a209a, v0 +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_fma_f32 v2, 0x3e9a209a, v0, -v1 +; GFX1250TRUE16-NEXT: v_fmamk_f32 v2, v0, 0x3284fbcf, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x411a209b, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_log10_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, 0x800000, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc_lo +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250FAKE16-NEXT: v_log_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v1, 0x3e9a209a, v0 +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e64 s0, 0x7f800000, |v0| +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_fma_f32 v2, 0x3e9a209a, v0, -v1 +; GFX1250FAKE16-NEXT: v_fmamk_f32 v2, v0, 0x3284fbcf, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v1, v1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, v0, v1, s0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v1, 0, 0x411a209b, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.log10.bf16(bfloat %a) ret bfloat %op } @@ -31946,34 +32242,65 @@ define bfloat @v_exp_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_exp_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: s_mov_b32 s0, 0x3fb8aa3b -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v1 -; GFX1250-NEXT: v_rndne_f32_e32 v3, v2 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] -; GFX1250-NEXT: s_mov_b32 s0, 0x32a5705f -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_sub_f32_e32 v2, v2, v3 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] -; GFX1250-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc2ce8ed0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_add_f32_e32 v0, v2, v0 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v3 -; GFX1250-NEXT: v_exp_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo -; GFX1250-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x42b17218, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_exp_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x3fb8aa3b +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v1 +; GFX1250TRUE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc2ce8ed0, v1 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v3, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v4, v2 +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x32a5705f +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v3 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v4 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250TRUE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x42b17218, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_exp_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x3fb8aa3b +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v1 +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v3, v2 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x32a5705f +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v2, v2, v3 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc2ce8ed0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v3 +; GFX1250FAKE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x42b17218, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.exp.bf16(bfloat %a) ret bfloat %op } @@ -32138,12 +32465,19 @@ define bfloat @v_exp2_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_exp2_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_exp_bf16_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_exp2_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_exp_bf16_e32 v0.l, v0.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_exp2_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_exp_bf16_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.exp2.bf16(bfloat %a) ret bfloat %op } @@ -32382,34 +32716,65 @@ define bfloat @v_exp10_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_exp10_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: s_mov_b32 s0, 0x40549a78 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mul_f32_e32 v2, 0x40549a78, v1 -; GFX1250-NEXT: v_rndne_f32_e32 v3, v2 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] -; GFX1250-NEXT: s_mov_b32 s0, 0x33979a37 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_sub_f32_e32 v2, v2, v3 -; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] -; GFX1250-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc23369f4, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_add_f32_e32 v0, v2, v0 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v3 -; GFX1250-NEXT: v_exp_f32_e32 v0, v0 -; GFX1250-NEXT: v_nop -; GFX1250-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo -; GFX1250-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x421a209b, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_exp10_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x40549a78 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mul_f32_e32 v2, 0x40549a78, v1 +; GFX1250TRUE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc23369f4, v1 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v3, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v4, v2 +; GFX1250TRUE16-NEXT: s_mov_b32 s0, 0x33979a37 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v3 op_sel_hi:[1,0,0] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v4 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(TRANS32_DEP_1) +; GFX1250TRUE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_nop +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250TRUE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x421a209b, v1 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_exp10_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x40549a78 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_mul_f32_e32 v2, 0x40549a78, v1 +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v3, v2 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v0, s0, -v2 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: s_mov_b32 s0, 0x33979a37 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v2, v2, v3 +; GFX1250FAKE16-NEXT: v_fma_mix_f32_bf16 v0, v0, s0, v4 op_sel_hi:[1,0,0] +; GFX1250FAKE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, 0xc23369f4, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v2, v0 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v3 +; GFX1250FAKE16-NEXT: v_exp_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_nop +; GFX1250FAKE16-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0x421a209b, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x7f800000, v0, vcc_lo +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.exp10.bf16(bfloat %a) ret bfloat %op } @@ -32517,15 +32882,26 @@ define bfloat @v_ceil_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_ceil_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_ceil_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_ceil_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_ceil_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_ceil_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_ceil_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.ceil.bf16(bfloat %a) ret bfloat %op } @@ -32633,15 +33009,26 @@ define bfloat @v_trunc_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_trunc_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_trunc_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_trunc_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_trunc_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_trunc_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_trunc_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.trunc.bf16(bfloat %a) ret bfloat %op } @@ -32749,15 +33136,26 @@ define bfloat @v_rint_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_rint_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_rndne_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_rint_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_rint_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.rint.bf16(bfloat %a) ret bfloat %op } @@ -32865,15 +33263,26 @@ define bfloat @v_nearbyint_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_nearbyint_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_rndne_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_nearbyint_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_nearbyint_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.nearbyint.bf16(bfloat %a) ret bfloat %op } @@ -33031,23 +33440,42 @@ define bfloat @v_round_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_round_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_trunc_f32_e32 v1, v0 -; GFX1250-NEXT: v_sub_f32_e32 v2, v0, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_ge_f32_e64 s0, |v2|, 0.5 -; GFX1250-NEXT: v_cndmask_b32_e64 v2, 0, 1.0, s0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_bfi_b32 v0, 0x7fffffff, v2, v0 -; GFX1250-NEXT: v_add_f32_e32 v0, v1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_round_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_trunc_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_sub_f32_e32 v2, v1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_ge_f32_e64 s0, |v2|, 0.5 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v2, 0, 1.0, s0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_bfi_b32 v1, 0x7fffffff, v2, v1 +; GFX1250TRUE16-NEXT: v_add_f32_e32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_round_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_trunc_f32_e32 v1, v0 +; GFX1250FAKE16-NEXT: v_sub_f32_e32 v2, v0, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_ge_f32_e64 s0, |v2|, 0.5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v2, 0, 1.0, s0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_bfi_b32 v0, 0x7fffffff, v2, v0 +; GFX1250FAKE16-NEXT: v_add_f32_e32 v0, v1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.round.bf16(bfloat %a) ret bfloat %op } @@ -33155,15 +33583,26 @@ define bfloat @v_roundeven_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_roundeven_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_rndne_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_roundeven_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_rndne_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_roundeven_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_rndne_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.roundeven.bf16(bfloat %a) ret bfloat %op } @@ -33271,15 +33710,26 @@ define bfloat @v_floor_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_floor_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_floor_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_floor_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_floor_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_floor_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_floor_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.floor.bf16(bfloat %a) ret bfloat %op } @@ -33385,15 +33835,26 @@ define bfloat @v_canonicalize_bf16(bfloat %a) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_canonicalize_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_max_num_f32_e32 v0, v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_canonicalize_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_max_num_f32_e32 v0, v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_canonicalize_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_max_num_f32_e32 v0, v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.canonicalize.bf16(bfloat %a) ret bfloat %op } @@ -33535,15 +33996,28 @@ define i1 @v_fcmp_oeq_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_oeq_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_oeq_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_oeq_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp oeq bfloat %a, %b ret i1 %op } @@ -33630,15 +34104,28 @@ define i1 @v_fcmp_ogt_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ogt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_gt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ogt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ogt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_gt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ogt bfloat %a, %b ret i1 %op } @@ -33725,15 +34212,28 @@ define i1 @v_fcmp_oge_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_oge_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_oge_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_ge_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_oge_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_ge_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp oge bfloat %a, %b ret i1 %op } @@ -33820,15 +34320,28 @@ define i1 @v_fcmp_olt_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_olt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_olt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_lt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_olt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_lt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp olt bfloat %a, %b ret i1 %op } @@ -33915,15 +34428,28 @@ define i1 @v_fcmp_ole_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ole_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ole_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_le_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ole_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_le_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ole bfloat %a, %b ret i1 %op } @@ -34010,15 +34536,28 @@ define i1 @v_fcmp_one_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_one_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_lg_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_one_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_lg_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_one_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_lg_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp one bfloat %a, %b ret i1 %op } @@ -34105,15 +34644,28 @@ define i1 @v_fcmp_uno_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_uno_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_uno_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_uno_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp uno bfloat %a, %b ret i1 %op } @@ -34200,15 +34752,28 @@ define i1 @v_fcmp_ueq_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ueq_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nlg_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ueq_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nlg_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ueq_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nlg_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ueq bfloat %a, %b ret i1 %op } @@ -34295,15 +34860,28 @@ define i1 @v_fcmp_ugt_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ugt_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nle_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ugt_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nle_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ugt_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nle_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ugt bfloat %a, %b ret i1 %op } @@ -34390,15 +34968,28 @@ define i1 @v_fcmp_uge_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_uge_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_uge_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_uge_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nlt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp uge bfloat %a, %b ret i1 %op } @@ -34485,15 +35076,28 @@ define i1 @v_fcmp_ult_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ult_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ult_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_nge_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ult_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_nge_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ult bfloat %a, %b ret i1 %op } @@ -34580,15 +35184,28 @@ define i1 @v_fcmp_ule_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_ule_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_ule_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_ule_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_ngt_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp ule bfloat %a, %b ret i1 %op } @@ -34675,15 +35292,28 @@ define i1 @v_fcmp_une_bf16(bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fcmp_une_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_neq_f32_e32 vcc_lo, v0, v1 -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fcmp_une_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, v1.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: v_cmp_neq_f32_e32 vcc_lo, v1, v2 +; GFX1250TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fcmp_une_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_neq_f32_e32 vcc_lo, v0, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fcmp une bfloat %a, %b ret i1 %op } @@ -34790,14 +35420,24 @@ define i16 @v_fptosi_bf16_to_i16(bfloat %x) { ; GFX11FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_bf16_to_i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_bf16_to_i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_bf16_to_i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi bfloat %x to i16 ret i16 %op } @@ -34899,18 +35539,31 @@ define <2 x i16> @v_fptosi_v2bf16_to_v2i16(<2 x bfloat> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_v2bf16_to_v2i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_v2bf16_to_v2i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v1, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_v2bf16_to_v2i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi <2 x bfloat> %x to <2 x i16> ret <2 x i16> %op } @@ -35032,19 +35685,33 @@ define <3 x i16> @v_fptosi_v3bf16_to_v3i16(<3 x bfloat> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v0, v0, v2, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_v3bf16_to_v3i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: v_perm_b32 v0, v0, v2, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_v3bf16_to_v3i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v2.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_v3bf16_to_v3i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v2, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi <3 x bfloat> %x to <3 x i16> ret <3 x i16> %op } @@ -35198,23 +35865,41 @@ define <4 x i16> @v_fptosi_v4bf16_to_v4i16(<4 x bfloat> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v1, v1, v2, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_v4bf16_to_v4i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v1 :: v_dual_lshlrev_b32 v3, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GFX1250-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v3, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_i32_f32_e32 v1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_perm_b32 v0, v0, v3, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v1, v2, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_v4bf16_to_v4i16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff0000, v0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 +; GFX1250TRUE16-NEXT: v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, v2.l +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v3.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_v4bf16_to_v4i16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v1 :: v_dual_lshlrev_b32 v3, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 0xffff0000, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v3, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v1, v2, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi <4 x bfloat> %x to <4 x i16> ret <4 x i16> %op } @@ -35274,14 +35959,24 @@ define i32 @v_fptosi_bf16_to_i32(bfloat %x) { ; GFX11FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_bf16_to_i32: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_i32_f32_e32 v0, v0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_bf16_to_i32: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_i32_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_bf16_to_i32: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_i32_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi bfloat %x to i32 ret i32 %op } @@ -35729,26 +36424,48 @@ define i64 @v_fptosi_bf16_to_i64(bfloat %x) { ; GFX11FAKE16-NEXT: v_sub_co_ci_u32_e64 v1, null, v1, v3, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fptosi_bf16_to_i64: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshlrev_b32_e32 v0, 16, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_trunc_f32_e32 v0, v0 -; GFX1250-NEXT: v_mul_f32_e64 v1, 0x2f800000, |v0| -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_floor_f32_e32 v1, v1 -; GFX1250-NEXT: v_fma_f32 v2, 0xcf800000, v1, |v0| -; GFX1250-NEXT: v_ashrrev_i32_e32 v0, 31, v0 -; GFX1250-NEXT: v_cvt_u32_f32_e32 v3, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_u32_f32_e32 v2, v2 -; GFX1250-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_bitop2_b32 v3, v3, v0 bitop3:0x14 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_xor_b32_e32 v2, v2, v0 -; GFX1250-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fptosi_bf16_to_i64: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_trunc_f32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_mul_f32_e64 v1, 0x2f800000, |v0| +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_floor_f32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_fma_f32 v2, 0xcf800000, v1, |v0| +; GFX1250TRUE16-NEXT: v_ashrrev_i32_e32 v0, 31, v0 +; GFX1250TRUE16-NEXT: v_cvt_u32_f32_e32 v3, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_bitop2_b32 v3, v3, v0 bitop3:0x14 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v2, v2, v0 +; GFX1250TRUE16-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1] +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fptosi_bf16_to_i64: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshlrev_b32_e32 v0, 16, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_trunc_f32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_mul_f32_e64 v1, 0x2f800000, |v0| +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_floor_f32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_fma_f32 v2, 0xcf800000, v1, |v0| +; GFX1250FAKE16-NEXT: v_ashrrev_i32_e32 v0, 31, v0 +; GFX1250FAKE16-NEXT: v_cvt_u32_f32_e32 v3, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_bitop2_b32 v3, v3, v0 bitop3:0x14 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v2, v2, v0 +; GFX1250FAKE16-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = fptosi bfloat %x to i64 ret i64 %op } @@ -37293,22 +38010,39 @@ define <3 x bfloat> @v_sitofp_v3i16_to_v3bf16(<3 x i16> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sitofp_v3i16_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_ashrrev_i32_e32 v2, 16, v0 -; GFX1250-NEXT: v_bfe_i32 v0, v0, 0, 16 -; GFX1250-NEXT: v_bfe_i32 v1, v1, 0, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_i32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_i32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sitofp_v3i16_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_bfe_i32 v1, v1, 0, 16 +; GFX1250TRUE16-NEXT: v_ashrrev_i32_e32 v2, 16, v0 +; GFX1250TRUE16-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sitofp_v3i16_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_ashrrev_i32_e32 v2, 16, v0 +; GFX1250FAKE16-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GFX1250FAKE16-NEXT: v_bfe_i32 v1, v1, 0, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = sitofp <3 x i16> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -37972,17 +38706,31 @@ define <3 x bfloat> @v_sitofp_v3i32_to_v3bf16(<3 x i32> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v2, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sitofp_v3i32_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v2, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sitofp_v3i32_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sitofp_v3i32_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = sitofp <3 x i32> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -39232,52 +39980,101 @@ define <3 x bfloat> @v_sitofp_v3i64_to_v3bf16(<3 x i64> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_sitofp_v3i64_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_xor_b32_e32 v8, v4, v5 -; GFX1250-NEXT: v_xor_b32_e32 v6, v2, v3 -; GFX1250-NEXT: v_cls_i32_e32 v10, v3 -; GFX1250-NEXT: v_cls_i32_e32 v9, v5 -; GFX1250-NEXT: v_cls_i32_e32 v11, v1 -; GFX1250-NEXT: v_dual_ashrrev_i32 v8, 31, v8 :: v_dual_bitop2_b32 v7, v0, v1 bitop3:0x14 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_dual_ashrrev_i32 v6, 31, v6 :: v_dual_ashrrev_i32 v7, 31, v7 -; GFX1250-NEXT: v_dual_add_nc_u32 v6, 32, v6 :: v_dual_add_nc_u32 v7, 32, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_add_min_u32_e64 v6, v10, -1, v6 -; GFX1250-NEXT: v_add_min_u32_e64 v7, v11, -1, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] -; GFX1250-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v2, 1, v2 -; GFX1250-NEXT: v_add_nc_u32_e32 v8, 32, v8 -; GFX1250-NEXT: v_min_u32_e32 v0, 1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX1250-NEXT: v_add_min_u32_e64 v8, v9, -1, v8 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v2, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] -; GFX1250-NEXT: v_sub_nc_u32_e32 v8, 32, v8 -; GFX1250-NEXT: v_ldexp_f32 v2, v2, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_min_u32_e32 v4, 1, v4 -; GFX1250-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 -; GFX1250-NEXT: v_cvt_f32_i32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_i32_e32 v1, v1 -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v4 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_ldexp_f32 v1, v1, v8 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_sitofp_v3i64_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v7, v2, v3 +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v6, v4, v5 +; GFX1250TRUE16-NEXT: v_cls_i32_e32 v10, v3 +; GFX1250TRUE16-NEXT: v_cls_i32_e32 v9, v5 +; GFX1250TRUE16-NEXT: v_cls_i32_e32 v11, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_dual_ashrrev_i32 v7, 31, v7 :: v_dual_ashrrev_i32 v6, 31, v6 +; GFX1250TRUE16-NEXT: v_xor_b32_e32 v8, v0, v1 +; GFX1250TRUE16-NEXT: v_dual_add_nc_u32 v7, 32, v7 :: v_dual_add_nc_u32 v6, 32, v6 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_ashrrev_i32_e32 v8, 31, v8 +; GFX1250TRUE16-NEXT: v_add_min_u32_e64 v7, v10, -1, v7 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_add_min_u32_e64 v6, v9, -1, v6 +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[2:3], v7, v[2:3] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[4:5], v6, v[4:5] +; GFX1250TRUE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250TRUE16-NEXT: v_add_nc_u32_e32 v8, 32, v8 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250TRUE16-NEXT: v_or_b32_e32 v2, v3, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_add_min_u32_e64 v8, v11, -1, v8 +; GFX1250TRUE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v4, v5, v4 bitop3:0x54 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[0:1], v8, v[0:1] +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v5, 32, v8 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: v_or_b32_e32 v0, v1, v0 +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v1, v4 +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v4, 32, v7 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v1, v1, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_ldexp_f32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_sitofp_v3i64_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v8, v4, v5 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v6, v2, v3 +; GFX1250FAKE16-NEXT: v_cls_i32_e32 v10, v3 +; GFX1250FAKE16-NEXT: v_cls_i32_e32 v9, v5 +; GFX1250FAKE16-NEXT: v_cls_i32_e32 v11, v1 +; GFX1250FAKE16-NEXT: v_dual_ashrrev_i32 v8, 31, v8 :: v_dual_bitop2_b32 v7, v0, v1 bitop3:0x14 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_dual_ashrrev_i32 v6, 31, v6 :: v_dual_ashrrev_i32 v7, 31, v7 +; GFX1250FAKE16-NEXT: v_dual_add_nc_u32 v6, 32, v6 :: v_dual_add_nc_u32 v7, 32, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_add_min_u32_e64 v6, v10, -1, v6 +; GFX1250FAKE16-NEXT: v_add_min_u32_e64 v7, v11, -1, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250FAKE16-NEXT: v_add_nc_u32_e32 v8, 32, v8 +; GFX1250FAKE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_or_b32_e32 v2, v3, v2 +; GFX1250FAKE16-NEXT: v_add_min_u32_e64 v8, v9, -1, v8 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v2, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] +; GFX1250FAKE16-NEXT: v_sub_nc_u32_e32 v8, 32, v8 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v2, v2, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v4 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v1, v1, v8 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = sitofp <3 x i64> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -40015,15 +40812,26 @@ define bfloat @v_uitofp_i16_to_bf16(i16 %x) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_i16_to_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_i16_to_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_i16_to_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp i16 %x to bfloat ret bfloat %op } @@ -40167,18 +40975,32 @@ define <2 x bfloat> @v_uitofp_v2i16_to_v2bf16(<2 x i16> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x7060302 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v2i16_to_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v1, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v2i16_to_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v0.h +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v2i16_to_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <2 x i16> %x to <2 x bfloat> ret <2 x bfloat> %op } @@ -40373,22 +41195,41 @@ define <3 x bfloat> @v_uitofp_v3i16_to_v3bf16(<3 x i16> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v3i16_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: v_and_b32_e32 v1, 0xffff, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v3i16_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v2, 0xffff, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v3.l, v0.h +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v3 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v3, v0, s0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v1, v2 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v3.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v3i16_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <3 x i16> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -40626,23 +41467,43 @@ define <4 x bfloat> @v_uitofp_v4i16_to_v4bf16(<4 x i16> %x) { ; GFX11FAKE16-NEXT: v_perm_b32 v1, v1, v2, 0x7060302 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v4i16_to_v4bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v2, 16, v1 :: v_dual_lshrrev_b32 v3, 16, v0 -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: v_and_b32_e32 v1, 0xffff, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v3, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v3 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, v2 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v4i16_to_v4bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, v1.h +; GFX1250TRUE16-NEXT: v_and_b32_e32 v3, 0xffff, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v4, v2 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.h +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 0xffff, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v3, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v1, v2 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v3, v4 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v4i16_to_v4bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v2, 16, v1 :: v_dual_lshrrev_b32 v3, 16, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v3, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v3 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, v2 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <4 x i16> %x to <4 x bfloat> ret <4 x bfloat> %op } @@ -41058,17 +41919,31 @@ define <3 x bfloat> @v_uitofp_v3i32_to_v3bf16(<3 x i32> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v2, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v3i32_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v3i32_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, v2.l +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v3i32_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v1 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v2, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <3 x i32> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -42105,44 +42980,84 @@ define <3 x bfloat> @v_uitofp_v3i64_to_v3bf16(<3 x i64> %x) { ; GFX11FAKE16-NEXT: v_alignbit_b32 v1, s0, v1, 16 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_uitofp_v3i64_to_v3bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_clz_i32_u32_e32 v6, v3 -; GFX1250-NEXT: v_clz_i32_u32_e32 v7, v1 -; GFX1250-NEXT: v_clz_i32_u32_e32 v8, v5 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v6, 32, v6 -; GFX1250-NEXT: v_min_u32_e32 v7, 32, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v8, 32, v8 -; GFX1250-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] -; GFX1250-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v2, 1, v2 -; GFX1250-NEXT: v_min_u32_e32 v0, 1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_min_u32_e32 v4, 1, v4 -; GFX1250-NEXT: v_dual_sub_nc_u32 v8, 32, v8 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 -; GFX1250-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v2, v2 -; GFX1250-NEXT: v_cvt_f32_u32_e32 v0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cvt_f32_u32_e32 v1, v1 -; GFX1250-NEXT: v_ldexp_f32 v2, v2, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_ldexp_f32 v0, v0, v4 -; GFX1250-NEXT: v_ldexp_f32 v1, v1, v8 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 -; GFX1250-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_uitofp_v3i64_to_v3bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_clz_i32_u32_e32 v6, v5 +; GFX1250TRUE16-NEXT: v_clz_i32_u32_e32 v7, v3 +; GFX1250TRUE16-NEXT: v_clz_i32_u32_e32 v8, v1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v6, 32, v6 +; GFX1250TRUE16-NEXT: v_min_u32_e32 v7, 32, v7 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v8, 32, v8 +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[4:5], v6, v[4:5] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[2:3], v7, v[2:3] +; GFX1250TRUE16-NEXT: v_lshlrev_b64_e32 v[0:1], v8, v[0:1] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250TRUE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: v_or_b32_e32 v4, v5, v4 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_or_b32_e32 v2, v3, v2 +; GFX1250TRUE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v5, 32, v8 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_3) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v1, v4 +; GFX1250TRUE16-NEXT: v_sub_nc_u32_e32 v4, 32, v7 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250TRUE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v1, v1, v3 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250TRUE16-NEXT: v_ldexp_f32 v2, v2, v4 +; GFX1250TRUE16-NEXT: v_ldexp_f32 v0, v0, v5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_uitofp_v3i64_to_v3bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_clz_i32_u32_e32 v6, v3 +; GFX1250FAKE16-NEXT: v_clz_i32_u32_e32 v7, v1 +; GFX1250FAKE16-NEXT: v_clz_i32_u32_e32 v8, v5 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v6, 32, v6 +; GFX1250FAKE16-NEXT: v_min_u32_e32 v7, 32, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v8, 32, v8 +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[2:3], v6, v[2:3] +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[0:1], v7, v[0:1] +; GFX1250FAKE16-NEXT: v_lshlrev_b64_e32 v[4:5], v8, v[4:5] +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v2, 1, v2 +; GFX1250FAKE16-NEXT: v_min_u32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_min_u32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v8, 32, v8 :: v_dual_bitop2_b32 v2, v3, v2 bitop3:0x54 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v3, 32, v6 :: v_dual_bitop2_b32 v0, v1, v0 bitop3:0x54 +; GFX1250FAKE16-NEXT: v_dual_sub_nc_u32 v4, 32, v7 :: v_dual_bitop2_b32 v1, v5, v4 bitop3:0x54 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v2, v2 +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v2, v2, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_ldexp_f32 v0, v0, v4 +; GFX1250FAKE16-NEXT: v_ldexp_f32 v1, v1, v8 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, v2 +; GFX1250FAKE16-NEXT: v_cvt_pk_bf16_f32 v1, v1, s0 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = uitofp <3 x i64> %x to <3 x bfloat> ret <3 x bfloat> %op } @@ -42717,15 +43632,25 @@ define bfloat @v_select_bf16(i1 %cond, bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select i1 %cond, bfloat %a, bfloat %b ret bfloat %op } @@ -42810,16 +43735,27 @@ define bfloat @v_select_fneg_lhs_bf16(i1 %cond, bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_fneg_lhs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0 -; GFX1250-NEXT: v_xor_b32_e32 v1, 0x8000, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_fneg_lhs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v1.l +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_fneg_lhs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v1, 0x8000, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %neg.a = fneg bfloat %a %op = select i1 %cond, bfloat %neg.a, bfloat %b ret bfloat %op @@ -42905,16 +43841,27 @@ define bfloat @v_select_fneg_rhs_bf16(i1 %cond, bfloat %a, bfloat %b) { ; GFX11FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_fneg_rhs_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0 -; GFX1250-NEXT: v_xor_b32_e32 v2, 0x8000, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_fneg_rhs_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v2.l +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_fneg_rhs_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250FAKE16-NEXT: v_xor_b32_e32 v2, 0x8000, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %neg.b = fneg bfloat %b %op = select i1 %cond, bfloat %a, bfloat %neg.b ret bfloat %op @@ -43025,18 +43972,29 @@ define <2 x bfloat> @v_select_v2bf16(i1 %cond, <2 x bfloat> %a, <2 x bfloat> %b) ; GFX11FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_select_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v3, 16, v1 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_cndmask_b32 v0, v2, v1, vcc_lo -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_select_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b32_e32 v0, 1, v0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v1.h, vcc_lo +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_select_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v3, 16, v1 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_cndmask_b32 v0, v2, v1, vcc_lo +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v4, v3, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select i1 %cond, <2 x bfloat> %a, <2 x bfloat> %b ret <2 x bfloat> %op } @@ -43155,20 +44113,34 @@ define <2 x bfloat> @v_vselect_v2bf16(<2 x i1> %cond, <2 x bfloat> %a, <2 x bflo ; GFX11FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v5, 16, v3 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v4, 16, v2 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v5, 16, v3 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <2 x i1> %cond, <2 x bfloat> %a, <2 x bfloat> %b ret <2 x bfloat> %op } @@ -43256,16 +44228,26 @@ define amdgpu_ps i32 @s_select_bf16(bfloat inreg %a, bfloat inreg %b, i32 %c) { ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_select_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: v_mov_b32_e32 v1, s0 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s1, v1, vcc_lo -; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_select_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, s0 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, 0 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, s1, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_select_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v1, s0 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s1, v1, vcc_lo +; GFX1250FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq i32 %c, 0 %op = select i1 %cond, bfloat %a, bfloat %b %cast = bitcast bfloat %op to i16 @@ -43402,20 +44384,34 @@ define amdgpu_ps i32 @s_select_v2bf16(<2 x bfloat> inreg %a, <2 x bfloat> inreg ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_select_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_lshr_b32 s2, s0, 16 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: v_dual_mov_b32 v1, s2 :: v_dual_mov_b32 v2, s0 -; GFX1250-NEXT: s_lshr_b32 s3, s1, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s3, v1, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, s1, v2, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_select_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_lshr_b32 s2, s0, 16 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, s2 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, s0 +; GFX1250TRUE16-NEXT: s_lshr_b32 s0, s1, 16 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, s0, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, s1, v0.l, vcc_lo +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_select_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_lshr_b32 s2, s0, 16 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v1, s2 :: v_dual_mov_b32 v2, s0 +; GFX1250FAKE16-NEXT: s_lshr_b32 s3, s1, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s3, v1, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, s1, v2, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v0, v1, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq i32 %c, 0 %op = select i1 %cond, <2 x bfloat> %a, <2 x bfloat> %b %cast = bitcast <2 x bfloat> %op to i32 @@ -43554,21 +44550,36 @@ define amdgpu_ps i32 @s_vselect_v2bf16(<2 x bfloat> inreg %a, <2 x bfloat> inreg ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_vselect_v2bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_lshr_b32 s2, s0, 16 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX1250-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s0 -; GFX1250-NEXT: s_lshr_b32 s0, s1, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, s0, v2, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s1, v3, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_vselect_v2bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_lshr_b32 s3, s0, 16 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s2, 0, v1 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, s3 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, s0 +; GFX1250TRUE16-NEXT: s_lshr_b32 s0, s1, 16 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, s0, v0.l, s2 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, s1, v0.h, vcc_lo +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v1 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_vselect_v2bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_lshr_b32 s2, s0, 16 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s0 +; GFX1250FAKE16-NEXT: s_lshr_b32 s0, s1, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, s0, v2, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s1, v3, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq <2 x i32> %c, zeroinitializer %op = select <2 x i1> %cond, <2 x bfloat> %a, <2 x bfloat> %b %cast = bitcast <2 x bfloat> %op to i32 @@ -45557,32 +46568,55 @@ define amdgpu_ps <2 x i32> @s_vselect_v4bf16(<4 x bfloat> inreg %a, <4 x bfloat> ; GFX11FAKE16-NEXT: v_readfirstlane_b32 s1, v1 ; GFX11FAKE16-NEXT: ; return to shader part epilog ; -; GFX1250-LABEL: s_vselect_v4bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_lshr_b32 s4, s1, 16 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v3 -; GFX1250-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s1 -; GFX1250-NEXT: s_lshr_b32 s4, s3, 16 -; GFX1250-NEXT: s_lshr_b32 s5, s0, 16 -; GFX1250-NEXT: v_mov_b32_e32 v6, s0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cndmask_b32_e32 v3, s4, v4, vcc_lo -; GFX1250-NEXT: v_mov_b32_e32 v4, s5 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX1250-NEXT: s_lshr_b32 s0, s2, 16 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) -; GFX1250-NEXT: v_cndmask_b32_e32 v1, s0, v4, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, s2, v6, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v2 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_cndmask_b32_e32 v2, s3, v5, vcc_lo -; GFX1250-NEXT: v_readfirstlane_b32 s0, v0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_readfirstlane_b32 s1, v1 -; GFX1250-NEXT: ; return to shader part epilog +; GFX1250TRUE16-LABEL: s_vselect_v4bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_lshr_b32 s7, s1, 16 +; GFX1250TRUE16-NEXT: s_lshr_b32 s9, s0, 16 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s4, 0, v1 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s5, 0, v2 +; GFX1250TRUE16-NEXT: v_cmp_eq_u32_e64 s6, 0, v3 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.l, s7 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v0.h, s9 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.l, s0 +; GFX1250TRUE16-NEXT: v_mov_b16_e32 v1.h, s1 +; GFX1250TRUE16-NEXT: s_lshr_b32 s8, s3, 16 +; GFX1250TRUE16-NEXT: s_lshr_b32 s0, s2, 16 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, s8, v0.l, s6 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, s0, v0.h, s4 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, s2, v1.l, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, s3, v1.h, s5 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250TRUE16-NEXT: v_readfirstlane_b32 s1, v2 +; GFX1250TRUE16-NEXT: ; return to shader part epilog +; +; GFX1250FAKE16-LABEL: s_vselect_v4bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_lshr_b32 s4, s1, 16 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v3 +; GFX1250FAKE16-NEXT: v_dual_mov_b32 v4, s4 :: v_dual_mov_b32 v5, s1 +; GFX1250FAKE16-NEXT: s_lshr_b32 s4, s3, 16 +; GFX1250FAKE16-NEXT: s_lshr_b32 s5, s0, 16 +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v6, s0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, s4, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_mov_b32_e32 v4, s5 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX1250FAKE16-NEXT: s_lshr_b32 s0, s2, 16 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1) +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, s0, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, s2, v6, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v2 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v2, s3, v5, vcc_lo +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s0, v0 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_readfirstlane_b32 s1, v1 +; GFX1250FAKE16-NEXT: ; return to shader part epilog %cond = icmp eq <4 x i32> %c, zeroinitializer %op = select <4 x i1> %cond, <4 x bfloat> %a, <4 x bfloat> %b %cast = bitcast <4 x bfloat> %op to <2 x i32> @@ -45787,27 +46821,49 @@ define <4 x bfloat> @v_vselect_v4bf16(<4 x i1> %cond, <4 x bfloat> %a, <4 x bflo ; GFX11FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v4bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v2, 1, v2 -; GFX1250-NEXT: v_dual_lshrrev_b32 v8, 16, v4 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v9, 16, v6 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_dual_cndmask_b32 v2, v7, v5, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v7, 16, v7 :: v_dual_lshrrev_b32 v5, 16, v5 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v6, v4, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v9, v8, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v4bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v3.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v1.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v7.l, v5.l, vcc_lo +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v6.l, v4.l, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v6.h, v4.h, s1 +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v7.h, v5.h, s2 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v4bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v2, 1, v2 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v8, 16, v4 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v9, 16, v6 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v2, v7, v5, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v7, 16, v7 :: v_dual_lshrrev_b32 v5, 16, v5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v6, v4, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v9, v8, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <4 x i1> %cond, <4 x bfloat> %a, <4 x bfloat> %b ret <4 x bfloat> %op } @@ -46161,45 +47217,77 @@ define <8 x bfloat> @v_vselect_v8bf16(<8 x i1> %cond, <8 x bfloat> %a, <8 x bflo ; GFX11FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v8bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_and_b32_e32 v6, 1, v6 -; GFX1250-NEXT: v_and_b32_e32 v4, 1, v4 -; GFX1250-NEXT: v_dual_lshrrev_b32 v17, 16, v14 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v16, 16, v10 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 -; GFX1250-NEXT: v_and_b32_e32 v1, 1, v1 -; GFX1250-NEXT: v_dual_cndmask_b32 v6, v15, v11, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 -; GFX1250-NEXT: v_and_b32_e32 v7, 1, v7 -; GFX1250-NEXT: v_lshrrev_b32_e32 v11, 16, v11 -; GFX1250-NEXT: v_dual_cndmask_b32 v4, v14, v10 :: v_dual_lshrrev_b32 v15, 16, v15 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 -; GFX1250-NEXT: v_dual_lshrrev_b32 v14, 16, v12 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 -; GFX1250-NEXT: v_lshrrev_b32_e32 v10, 16, v8 -; GFX1250-NEXT: v_cndmask_b32_e32 v5, v17, v16, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v2, v13, v9, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v9, 16, v9 -; GFX1250-NEXT: v_dual_cndmask_b32 v0, v12, v8 :: v_dual_lshrrev_b32 v13, 16, v13 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v14, v10, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v13, v9, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 -; GFX1250-NEXT: v_cndmask_b32_e32 v7, v15, v11, vcc_lo -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v8bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v3.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v5.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.l +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v6.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v4.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.l, 1, v7.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s5, 1, v1.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s3, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s4, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s6, 1, v2.l +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v12.l, v8.l, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.l, v15.l, v11.l, s2 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, v14.l, v10.l, s3 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v13.l, v9.l, s4 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v12.h, v8.h, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v13.h, v9.h, s1 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, v14.h, v10.h, s5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.h, v15.h, v11.h, s6 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v8bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v6, 1, v6 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v17, 16, v14 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v16, 16, v10 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 1, v1 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v6, v15, v11, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v7, 1, v7 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v11, 16, v11 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v4, v14, v10 :: v_dual_lshrrev_b32 v15, 16, v15 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v14, 16, v12 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v10, 16, v8 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v5, v17, v16, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v2, v13, v9, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v9, 16, v9 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v0, v12, v8 :: v_dual_lshrrev_b32 v13, 16, v13 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v14, v10, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_3) | instid1(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v13, v9, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v7, v15, v11, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <8 x i1> %cond, <8 x bfloat> %a, <8 x bfloat> %b ret <8 x bfloat> %op } @@ -46939,73 +48027,129 @@ define <16 x bfloat> @v_vselect_v16bf16(<16 x i1> %cond, <16 x bfloat> %a, <16 x ; GFX11FAKE16-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v16bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: scratch_load_b32 v31, off, s32 -; GFX1250-NEXT: v_dual_lshrrev_b32 v52, 16, v25 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v53, 16, v16 :: v_dual_bitop2_b32 v13, 1, v13 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v33, 16, v22 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 -; GFX1250-NEXT: v_dual_lshrrev_b32 v34, 16, v30 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v51, 16, v17 :: v_dual_bitop2_b32 v10, 1, v10 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v12, v30, v22, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 -; GFX1250-NEXT: v_dual_lshrrev_b32 v50, 16, v26 :: v_dual_bitop2_b32 v11, 1, v11 bitop3:0x40 -; GFX1250-NEXT: v_and_b32_e32 v14, 1, v14 -; GFX1250-NEXT: v_dual_lshrrev_b32 v35, 16, v21 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v13, v34, v33, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 -; GFX1250-NEXT: v_dual_lshrrev_b32 v36, 16, v29 :: v_dual_bitop2_b32 v4, 1, v4 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v49, 16, v18 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v10, v29, v21, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 -; GFX1250-NEXT: v_dual_lshrrev_b32 v37, 16, v20 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v38, 16, v28 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v48, 16, v27 :: v_dual_bitop2_b32 v9, 1, v9 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v11, v36, v35, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 -; GFX1250-NEXT: v_dual_lshrrev_b32 v39, 16, v19 :: v_dual_bitop2_b32 v6, 1, v6 bitop3:0x40 -; GFX1250-NEXT: v_dual_lshrrev_b32 v32, 16, v23 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v8, v28, v20, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 -; GFX1250-NEXT: v_dual_lshrrev_b32 v54, 16, v24 :: v_dual_bitop2_b32 v15, 1, v15 bitop3:0x40 -; GFX1250-NEXT: v_cndmask_b32_e32 v9, v38, v37, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 -; GFX1250-NEXT: v_cndmask_b32_e32 v6, v27, v19, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 -; GFX1250-NEXT: v_cndmask_b32_e32 v4, v26, v18, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_cndmask_b32_e32 v2, v25, v17, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v52, v51, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_cndmask_b32_e32 v0, v24, v16, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v54, v53, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 -; GFX1250-NEXT: v_cndmask_b32_e32 v5, v50, v49, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_lshrrev_b32_e32 v3, 16, v31 -; GFX1250-NEXT: v_cndmask_b32_e32 v7, v48, v39, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 -; GFX1250-NEXT: v_cndmask_b32_e32 v14, v31, v23, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 -; GFX1250-NEXT: v_cndmask_b32_e32 v15, v3, v32, vcc_lo -; GFX1250-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX1250-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v16bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v3.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.l, 1, v5.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.h, 1, v4.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.l, 1, v7.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.h, 1, v6.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.l, 1, v9.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.h, 1, v8.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.l, 1, v11.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.h, 1, v10.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.l, 1, v13.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.h, 1, v12.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.l, 1, v15.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.h, 1, v14.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v1.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s3, 1, v2.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s4, 1, v2.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s5, 1, v3.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s6, 1, v3.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s7, 1, v4.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s8, 1, v4.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s9, 1, v5.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s10, 1, v6.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s11, 1, v6.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s12, 1, v5.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s13, 1, v7.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s14, 1, v7.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.l, v30.l, v22.l, s10 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.h, v30.h, v22.h, s11 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.l, v29.l, v21.l, s12 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.h, v29.h, v21.h, s9 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.l, v28.l, v20.l, s8 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.h, v28.h, v20.h, s7 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.l, v27.l, v19.l, s6 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.h, v27.h, v19.h, s5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, v26.l, v18.l, s4 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v25.l, v17.l, s2 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v24.l, v16.l, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v24.h, v16.h, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v25.h, v17.h, s1 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, v26.h, v18.h, s3 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.l, v31.l, v23.l, s14 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.h, v31.h, v23.h, s13 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v16bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: scratch_load_b32 v31, off, s32 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v52, 16, v25 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v53, 16, v16 :: v_dual_bitop2_b32 v13, 1, v13 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v33, 16, v22 :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v34, 16, v30 :: v_dual_bitop2_b32 v3, 1, v3 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v51, 16, v17 :: v_dual_bitop2_b32 v10, 1, v10 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v12, v30, v22, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v50, 16, v26 :: v_dual_bitop2_b32 v11, 1, v11 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v14, 1, v14 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v35, 16, v21 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v13, v34, v33, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v36, 16, v29 :: v_dual_bitop2_b32 v4, 1, v4 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v49, 16, v18 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v10, v29, v21, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v37, 16, v20 :: v_dual_bitop2_b32 v5, 1, v5 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v38, 16, v28 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v48, 16, v27 :: v_dual_bitop2_b32 v9, 1, v9 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v11, v36, v35, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v39, 16, v19 :: v_dual_bitop2_b32 v6, 1, v6 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v32, 16, v23 :: v_dual_bitop2_b32 v1, 1, v1 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v8, v28, v20, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v54, 16, v24 :: v_dual_bitop2_b32 v15, 1, v15 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v9, v38, v37, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v6, v27, v19, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v4, v26, v18, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v2, v25, v17, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v52, v51, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v0, v24, v16, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v54, v53, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v5, v50, v49, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_4) +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v31 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v7, v48, v39, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v14, v31, v23, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v15, v3, v32, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX1250FAKE16-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <16 x i1> %cond, <16 x bfloat> %a, <16 x bfloat> %b ret <16 x bfloat> %op } @@ -48861,177 +50005,330 @@ define <32 x bfloat> @v_vselect_v32bf16(<32 x i1> %cond, <32 x bfloat> %a, <32 x ; GFX11FAKE16-NEXT: v_perm_b32 v15, v31, v30, 0x5040100 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_vselect_v32bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x1b -; GFX1250-NEXT: scratch_load_b32 v31, off, s32 offset:60 -; GFX1250-NEXT: scratch_load_b32 v32, off, s32 offset:124 -; GFX1250-NEXT: scratch_load_u16 v33, off, s32 -; GFX1250-NEXT: scratch_load_b32 v34, off, s32 offset:128 -; GFX1250-NEXT: scratch_load_b32 v35, off, s32 offset:64 -; GFX1250-NEXT: scratch_load_b32 v36, off, s32 offset:120 -; GFX1250-NEXT: scratch_load_b32 v37, off, s32 offset:56 -; GFX1250-NEXT: scratch_load_b32 v38, off, s32 offset:116 -; GFX1250-NEXT: scratch_load_b32 v39, off, s32 offset:52 -; GFX1250-NEXT: scratch_load_b32 v48, off, s32 offset:112 -; GFX1250-NEXT: scratch_load_b32 v49, off, s32 offset:48 -; GFX1250-NEXT: scratch_load_b32 v50, off, s32 offset:108 -; GFX1250-NEXT: scratch_load_b32 v51, off, s32 offset:44 -; GFX1250-NEXT: scratch_load_b32 v52, off, s32 offset:104 -; GFX1250-NEXT: scratch_load_b32 v53, off, s32 offset:40 -; GFX1250-NEXT: scratch_load_b32 v54, off, s32 offset:100 -; GFX1250-NEXT: scratch_load_b32 v55, off, s32 offset:36 -; GFX1250-NEXT: scratch_load_b32 v64, off, s32 offset:76 -; GFX1250-NEXT: scratch_load_b32 v65, off, s32 offset:12 -; GFX1250-NEXT: scratch_load_b32 v66, off, s32 offset:96 -; GFX1250-NEXT: scratch_load_b32 v67, off, s32 offset:32 -; GFX1250-NEXT: scratch_load_b32 v68, off, s32 offset:80 -; GFX1250-NEXT: scratch_load_b32 v69, off, s32 offset:84 -; GFX1250-NEXT: scratch_load_b32 v70, off, s32 offset:92 -; GFX1250-NEXT: scratch_load_b32 v71, off, s32 offset:28 -; GFX1250-NEXT: scratch_load_b32 v80, off, s32 offset:20 -; GFX1250-NEXT: scratch_load_b32 v81, off, s32 offset:88 -; GFX1250-NEXT: scratch_load_b32 v82, off, s32 offset:24 -; GFX1250-NEXT: v_and_b32_e32 v30, 1, v30 -; GFX1250-NEXT: v_and_b32_e32 v29, 1, v29 -; GFX1250-NEXT: v_and_b32_e32 v26, 1, v26 -; GFX1250-NEXT: v_and_b32_e32 v24, 1, v24 -; GFX1250-NEXT: v_and_b32_e32 v22, 1, v22 -; GFX1250-NEXT: v_and_b32_e32 v20, 1, v20 -; GFX1250-NEXT: v_and_b32_e32 v18, 1, v18 -; GFX1250-NEXT: v_and_b32_e32 v16, 1, v16 -; GFX1250-NEXT: v_and_b32_e32 v10, 1, v10 -; GFX1250-NEXT: v_and_b32_e32 v6, 1, v6 -; GFX1250-NEXT: v_and_b32_e32 v4, 1, v4 -; GFX1250-NEXT: v_and_b32_e32 v1, 1, v1 -; GFX1250-NEXT: v_and_b32_e32 v3, 1, v3 -; GFX1250-NEXT: v_and_b32_e32 v5, 1, v5 -; GFX1250-NEXT: v_and_b32_e32 v23, 1, v23 -; GFX1250-NEXT: v_and_b32_e32 v9, 1, v9 -; GFX1250-NEXT: v_and_b32_e32 v13, 1, v13 -; GFX1250-NEXT: v_and_b32_e32 v15, 1, v15 -; GFX1250-NEXT: v_and_b32_e32 v21, 1, v21 -; GFX1250-NEXT: v_and_b32_e32 v11, 1, v11 -; GFX1250-NEXT: v_and_b32_e32 v19, 1, v19 -; GFX1250-NEXT: s_wait_loadcnt 0x1a -; GFX1250-NEXT: v_dual_lshrrev_b32 v83, 16, v32 :: v_dual_bitop2_b32 v17, 1, v17 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e64 s1, 1, v30 -; GFX1250-NEXT: v_and_b32_e32 v28, 1, v28 -; GFX1250-NEXT: s_wait_loadcnt 0x17 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX1250-NEXT: v_dual_cndmask_b32 v30, v34, v35, s1 :: v_dual_bitop2_b32 v33, 1, v33 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v28 -; GFX1250-NEXT: v_lshrrev_b32_e32 v28, 16, v31 -; GFX1250-NEXT: v_cmp_eq_u32_e64 s0, 1, v29 -; GFX1250-NEXT: scratch_load_b32 v29, off, s32 offset:16 -; GFX1250-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_lshrrev_b32 v34, 16, v34 -; GFX1250-NEXT: v_cndmask_b32_e32 v31, v32, v31, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v33 -; GFX1250-NEXT: scratch_load_b32 v32, off, s32 offset:72 -; GFX1250-NEXT: v_cndmask_b32_e64 v28, v83, v28, s0 -; GFX1250-NEXT: scratch_load_b32 v83, off, s32 offset:4 -; GFX1250-NEXT: v_cndmask_b32_e32 v34, v34, v35, vcc_lo -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: scratch_load_b32 v35, off, s32 offset:68 -; GFX1250-NEXT: scratch_load_b32 v33, off, s32 offset:8 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v26 -; GFX1250-NEXT: s_wait_loadcnt 0x1a -; GFX1250-NEXT: v_dual_cndmask_b32 v26, v36, v37, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v24 -; GFX1250-NEXT: v_dual_lshrrev_b32 v37, 16, v37 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x18 -; GFX1250-NEXT: v_dual_lshrrev_b32 v36, 16, v36 :: v_dual_cndmask_b32 v24, v38, v39, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v22 -; GFX1250-NEXT: v_dual_lshrrev_b32 v38, 16, v38 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x16 -; GFX1250-NEXT: v_dual_cndmask_b32 v22, v48, v49 :: v_dual_lshrrev_b32 v39, 16, v39 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v20 -; GFX1250-NEXT: v_dual_lshrrev_b32 v49, 16, v49 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x14 -; GFX1250-NEXT: v_dual_lshrrev_b32 v48, 16, v48 :: v_dual_cndmask_b32 v20, v50, v51, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v18 -; GFX1250-NEXT: v_dual_lshrrev_b32 v51, 16, v51 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x12 -; GFX1250-NEXT: v_dual_lshrrev_b32 v50, 16, v50 :: v_dual_cndmask_b32 v18, v52, v53, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v16 -; GFX1250-NEXT: v_dual_lshrrev_b32 v53, 16, v53 :: v_dual_bitop2_b32 v14, 1, v14 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x10 -; GFX1250-NEXT: v_dual_lshrrev_b32 v52, 16, v52 :: v_dual_cndmask_b32 v16, v54, v55, vcc_lo -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 -; GFX1250-NEXT: v_dual_lshrrev_b32 v55, 16, v55 :: v_dual_lshrrev_b32 v54, 16, v54 -; GFX1250-NEXT: s_wait_loadcnt 0xc -; GFX1250-NEXT: v_cndmask_b32_e32 v14, v66, v67, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 -; GFX1250-NEXT: v_dual_lshrrev_b32 v67, 16, v67 :: v_dual_lshrrev_b32 v66, 16, v66 -; GFX1250-NEXT: s_wait_loadcnt 0x8 -; GFX1250-NEXT: v_cndmask_b32_e32 v12, v70, v71, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 -; GFX1250-NEXT: v_dual_lshrrev_b32 v70, 16, v70 :: v_dual_bitop2_b32 v25, 1, v25 bitop3:0x40 -; GFX1250-NEXT: s_wait_loadcnt 0x5 -; GFX1250-NEXT: v_dual_cndmask_b32 v10, v81, v82 :: v_dual_lshrrev_b32 v71, 16, v71 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 -; GFX1250-NEXT: v_dual_lshrrev_b32 v82, 16, v82 :: v_dual_bitop2_b32 v27, 1, v27 bitop3:0x40 -; GFX1250-NEXT: v_dual_cndmask_b32 v8, v69, v80 :: v_dual_lshrrev_b32 v81, 16, v81 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 -; GFX1250-NEXT: v_dual_lshrrev_b32 v80, 16, v80 :: v_dual_lshrrev_b32 v69, 16, v69 -; GFX1250-NEXT: s_wait_loadcnt 0x4 -; GFX1250-NEXT: v_dual_cndmask_b32 v6, v68, v29 :: v_dual_lshrrev_b32 v29, 16, v29 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 -; GFX1250-NEXT: v_dual_lshrrev_b32 v68, 16, v68 :: v_dual_cndmask_b32 v4, v64, v65, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 -; GFX1250-NEXT: v_dual_lshrrev_b32 v65, 16, v65 :: v_dual_lshrrev_b32 v64, 16, v64 -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_dual_cndmask_b32 v2, v32, v33 :: v_dual_lshrrev_b32 v33, 16, v33 -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 -; GFX1250-NEXT: v_dual_lshrrev_b32 v32, 16, v32 :: v_dual_cndmask_b32 v0, v35, v83, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v27 -; GFX1250-NEXT: v_dual_lshrrev_b32 v83, 16, v83 :: v_dual_cndmask_b32 v27, v36, v37, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v25 -; GFX1250-NEXT: v_cndmask_b32_e32 v25, v38, v39, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v23 -; GFX1250-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_cndmask_b32 v23, v48, v49, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v21 -; GFX1250-NEXT: v_cndmask_b32_e32 v21, v50, v51, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v19 -; GFX1250-NEXT: v_cndmask_b32_e32 v19, v52, v53, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v17 -; GFX1250-NEXT: v_cndmask_b32_e32 v17, v54, v55, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 -; GFX1250-NEXT: v_cndmask_b32_e32 v15, v66, v67, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 -; GFX1250-NEXT: v_cndmask_b32_e32 v13, v70, v71, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 -; GFX1250-NEXT: v_cndmask_b32_e32 v11, v81, v82, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 -; GFX1250-NEXT: v_cndmask_b32_e32 v7, v68, v29, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 -; GFX1250-NEXT: v_cndmask_b32_e32 v3, v32, v33, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 -; GFX1250-NEXT: v_cndmask_b32_e32 v1, v35, v83, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 -; GFX1250-NEXT: v_cndmask_b32_e32 v5, v64, v65, vcc_lo -; GFX1250-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 -; GFX1250-NEXT: v_cndmask_b32_e32 v9, v69, v80, vcc_lo -; GFX1250-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v8, v17, v16, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v9, v19, v18, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v10, v21, v20, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v11, v23, v22, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v12, v25, v24, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v13, v27, v26, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v14, v28, v31, 0x5040100 -; GFX1250-NEXT: v_perm_b32 v15, v34, v30, 0x5040100 -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_vselect_v32bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: s_clause 0x20 +; GFX1250TRUE16-NEXT: scratch_load_u16 v31, off, s32 +; GFX1250TRUE16-NEXT: scratch_load_b32 v32, off, s32 offset:68 +; GFX1250TRUE16-NEXT: scratch_load_b32 v33, off, s32 offset:72 +; GFX1250TRUE16-NEXT: scratch_load_b32 v34, off, s32 offset:76 +; GFX1250TRUE16-NEXT: scratch_load_b32 v35, off, s32 offset:124 +; GFX1250TRUE16-NEXT: scratch_load_b32 v36, off, s32 offset:128 +; GFX1250TRUE16-NEXT: scratch_load_b32 v37, off, s32 offset:64 +; GFX1250TRUE16-NEXT: scratch_load_b32 v38, off, s32 offset:60 +; GFX1250TRUE16-NEXT: scratch_load_b32 v39, off, s32 offset:120 +; GFX1250TRUE16-NEXT: scratch_load_b32 v48, off, s32 offset:56 +; GFX1250TRUE16-NEXT: scratch_load_b32 v49, off, s32 offset:116 +; GFX1250TRUE16-NEXT: scratch_load_b32 v50, off, s32 offset:52 +; GFX1250TRUE16-NEXT: scratch_load_b32 v51, off, s32 offset:112 +; GFX1250TRUE16-NEXT: scratch_load_b32 v52, off, s32 offset:48 +; GFX1250TRUE16-NEXT: scratch_load_b32 v53, off, s32 offset:108 +; GFX1250TRUE16-NEXT: scratch_load_b32 v54, off, s32 offset:44 +; GFX1250TRUE16-NEXT: scratch_load_b32 v55, off, s32 offset:104 +; GFX1250TRUE16-NEXT: scratch_load_b32 v64, off, s32 offset:40 +; GFX1250TRUE16-NEXT: scratch_load_b32 v65, off, s32 offset:100 +; GFX1250TRUE16-NEXT: scratch_load_b32 v66, off, s32 offset:36 +; GFX1250TRUE16-NEXT: scratch_load_b32 v67, off, s32 offset:96 +; GFX1250TRUE16-NEXT: scratch_load_b32 v68, off, s32 offset:32 +; GFX1250TRUE16-NEXT: scratch_load_b32 v69, off, s32 offset:92 +; GFX1250TRUE16-NEXT: scratch_load_b32 v70, off, s32 offset:28 +; GFX1250TRUE16-NEXT: scratch_load_b32 v71, off, s32 offset:88 +; GFX1250TRUE16-NEXT: scratch_load_b32 v80, off, s32 offset:24 +; GFX1250TRUE16-NEXT: scratch_load_b32 v81, off, s32 offset:84 +; GFX1250TRUE16-NEXT: scratch_load_b32 v82, off, s32 offset:20 +; GFX1250TRUE16-NEXT: scratch_load_b32 v83, off, s32 offset:80 +; GFX1250TRUE16-NEXT: scratch_load_b32 v84, off, s32 offset:16 +; GFX1250TRUE16-NEXT: scratch_load_b32 v85, off, s32 offset:12 +; GFX1250TRUE16-NEXT: scratch_load_b32 v86, off, s32 offset:8 +; GFX1250TRUE16-NEXT: scratch_load_b32 v87, off, s32 offset:4 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v1.l +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v0.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v3.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v2.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.l, 1, v9.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e32 vcc_lo, 1, v0.h +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v4.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s0, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s2, 1, v1.h +; GFX1250TRUE16-NEXT: v_and_b16 v0.l, 1, v5.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.l, 1, v7.l +; GFX1250TRUE16-NEXT: v_and_b16 v1.h, 1, v6.l +; GFX1250TRUE16-NEXT: v_and_b16 v2.h, 1, v8.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.l, 1, v11.l +; GFX1250TRUE16-NEXT: v_and_b16 v3.h, 1, v10.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.l, 1, v13.l +; GFX1250TRUE16-NEXT: v_and_b16 v4.h, 1, v12.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.l, 1, v15.l +; GFX1250TRUE16-NEXT: v_and_b16 v5.h, 1, v14.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.l, 1, v17.l +; GFX1250TRUE16-NEXT: v_and_b16 v6.h, 1, v16.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.l, 1, v19.l +; GFX1250TRUE16-NEXT: v_and_b16 v7.h, 1, v18.l +; GFX1250TRUE16-NEXT: v_and_b16 v8.l, 1, v21.l +; GFX1250TRUE16-NEXT: v_and_b16 v8.h, 1, v20.l +; GFX1250TRUE16-NEXT: v_and_b16 v9.l, 1, v23.l +; GFX1250TRUE16-NEXT: v_and_b16 v9.h, 1, v22.l +; GFX1250TRUE16-NEXT: v_and_b16 v10.l, 1, v25.l +; GFX1250TRUE16-NEXT: v_and_b16 v10.h, 1, v24.l +; GFX1250TRUE16-NEXT: v_and_b16 v11.l, 1, v27.l +; GFX1250TRUE16-NEXT: v_and_b16 v11.h, 1, v26.l +; GFX1250TRUE16-NEXT: v_and_b16 v12.l, 1, v29.l +; GFX1250TRUE16-NEXT: v_and_b16 v12.h, 1, v28.l +; GFX1250TRUE16-NEXT: v_and_b16 v13.l, 1, v30.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s4, 1, v0.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s3, 1, v0.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s5, 1, v1.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s6, 1, v1.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s7, 1, v2.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s8, 1, v2.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s9, 1, v3.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s10, 1, v3.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s11, 1, v4.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s12, 1, v4.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s13, 1, v5.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s14, 1, v5.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s15, 1, v6.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s16, 1, v6.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s17, 1, v7.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s18, 1, v7.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s19, 1, v8.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s20, 1, v8.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s21, 1, v9.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s22, 1, v9.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s23, 1, v10.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s24, 1, v10.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s25, 1, v11.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s26, 1, v13.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s27, 1, v12.h +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s28, 1, v12.l +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s29, 1, v11.h +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x20 +; GFX1250TRUE16-NEXT: v_and_b16 v0.h, 1, v31.l +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x1a +; GFX1250TRUE16-NEXT: v_cndmask_b16 v15.l, v36.l, v37.l, s26 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x19 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v14.l, v35.l, v38.l, s27 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v14.h, v35.h, v38.h, s28 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x17 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v13.l, v39.l, v48.l, s29 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v13.h, v39.h, v48.h, s25 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x15 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v12.l, v49.l, v50.l, s24 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v12.h, v49.h, v50.h, s23 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x13 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v11.l, v51.l, v52.l, s22 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v11.h, v51.h, v52.h, s21 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x11 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v10.l, v53.l, v54.l, s20 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v10.h, v53.h, v54.h, s19 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0xf +; GFX1250TRUE16-NEXT: v_cndmask_b16 v9.l, v55.l, v64.l, s18 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v9.h, v55.h, v64.h, s17 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0xd +; GFX1250TRUE16-NEXT: v_cndmask_b16 v8.l, v65.l, v66.l, s16 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v8.h, v65.h, v66.h, s15 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0xb +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.l, v67.l, v68.l, s14 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v7.h, v67.h, v68.h, s13 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x9 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.l, v69.l, v70.l, s12 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v6.h, v69.h, v70.h, s11 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x7 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.l, v71.l, v80.l, s10 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v5.h, v71.h, v80.h, s9 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.l, v81.l, v82.l, s8 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v4.h, v81.h, v82.h, s7 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x3 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.l, v83.l, v84.l, s6 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x2 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.l, v34.l, v85.l, s4 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x1 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.l, v33.l, v86.l, s2 +; GFX1250TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.l, v32.l, v87.l, s1 +; GFX1250TRUE16-NEXT: v_cmp_eq_u16_e64 s1, 1, v0.h +; GFX1250TRUE16-NEXT: v_cndmask_b16 v0.h, v32.h, v87.h, vcc_lo +; GFX1250TRUE16-NEXT: v_cndmask_b16 v1.h, v33.h, v86.h, s0 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v2.h, v34.h, v85.h, s3 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v3.h, v83.h, v84.h, s5 +; GFX1250TRUE16-NEXT: v_cndmask_b16 v15.h, v36.h, v37.h, s1 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_vselect_v32bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: s_clause 0x1b +; GFX1250FAKE16-NEXT: scratch_load_b32 v31, off, s32 offset:60 +; GFX1250FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:124 +; GFX1250FAKE16-NEXT: scratch_load_u16 v33, off, s32 +; GFX1250FAKE16-NEXT: scratch_load_b32 v34, off, s32 offset:128 +; GFX1250FAKE16-NEXT: scratch_load_b32 v35, off, s32 offset:64 +; GFX1250FAKE16-NEXT: scratch_load_b32 v36, off, s32 offset:120 +; GFX1250FAKE16-NEXT: scratch_load_b32 v37, off, s32 offset:56 +; GFX1250FAKE16-NEXT: scratch_load_b32 v38, off, s32 offset:116 +; GFX1250FAKE16-NEXT: scratch_load_b32 v39, off, s32 offset:52 +; GFX1250FAKE16-NEXT: scratch_load_b32 v48, off, s32 offset:112 +; GFX1250FAKE16-NEXT: scratch_load_b32 v49, off, s32 offset:48 +; GFX1250FAKE16-NEXT: scratch_load_b32 v50, off, s32 offset:108 +; GFX1250FAKE16-NEXT: scratch_load_b32 v51, off, s32 offset:44 +; GFX1250FAKE16-NEXT: scratch_load_b32 v52, off, s32 offset:104 +; GFX1250FAKE16-NEXT: scratch_load_b32 v53, off, s32 offset:40 +; GFX1250FAKE16-NEXT: scratch_load_b32 v54, off, s32 offset:100 +; GFX1250FAKE16-NEXT: scratch_load_b32 v55, off, s32 offset:36 +; GFX1250FAKE16-NEXT: scratch_load_b32 v64, off, s32 offset:76 +; GFX1250FAKE16-NEXT: scratch_load_b32 v65, off, s32 offset:12 +; GFX1250FAKE16-NEXT: scratch_load_b32 v66, off, s32 offset:96 +; GFX1250FAKE16-NEXT: scratch_load_b32 v67, off, s32 offset:32 +; GFX1250FAKE16-NEXT: scratch_load_b32 v68, off, s32 offset:80 +; GFX1250FAKE16-NEXT: scratch_load_b32 v69, off, s32 offset:84 +; GFX1250FAKE16-NEXT: scratch_load_b32 v70, off, s32 offset:92 +; GFX1250FAKE16-NEXT: scratch_load_b32 v71, off, s32 offset:28 +; GFX1250FAKE16-NEXT: scratch_load_b32 v80, off, s32 offset:20 +; GFX1250FAKE16-NEXT: scratch_load_b32 v81, off, s32 offset:88 +; GFX1250FAKE16-NEXT: scratch_load_b32 v82, off, s32 offset:24 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v30, 1, v30 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v29, 1, v29 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v26, 1, v26 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v24, 1, v24 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v22, 1, v22 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v20, 1, v20 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v18, 1, v18 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v16, 1, v16 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v10, 1, v10 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v6, 1, v6 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v4, 1, v4 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v1, 1, v1 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v3, 1, v3 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v5, 1, v5 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v23, 1, v23 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v9, 1, v9 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v13, 1, v13 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v15, 1, v15 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v21, 1, v21 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v11, 1, v11 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v19, 1, v19 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x1a +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v83, 16, v32 :: v_dual_bitop2_b32 v17, 1, v17 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e64 s1, 1, v30 +; GFX1250FAKE16-NEXT: v_and_b32_e32 v28, 1, v28 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x17 +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v30, v34, v35, s1 :: v_dual_bitop2_b32 v33, 1, v33 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v28 +; GFX1250FAKE16-NEXT: v_lshrrev_b32_e32 v28, 16, v31 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e64 s0, 1, v29 +; GFX1250FAKE16-NEXT: scratch_load_b32 v29, off, s32 offset:16 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_lshrrev_b32 v34, 16, v34 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v31, v32, v31, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v33 +; GFX1250FAKE16-NEXT: scratch_load_b32 v32, off, s32 offset:72 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e64 v28, v83, v28, s0 +; GFX1250FAKE16-NEXT: scratch_load_b32 v83, off, s32 offset:4 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v34, v34, v35, vcc_lo +; GFX1250FAKE16-NEXT: s_clause 0x1 +; GFX1250FAKE16-NEXT: scratch_load_b32 v35, off, s32 offset:68 +; GFX1250FAKE16-NEXT: scratch_load_b32 v33, off, s32 offset:8 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v26 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x1a +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v26, v36, v37, vcc_lo :: v_dual_bitop2_b32 v0, 1, v0 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v24 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v37, 16, v37 :: v_dual_bitop2_b32 v2, 1, v2 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x18 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v36, 16, v36 :: v_dual_cndmask_b32 v24, v38, v39, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v22 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v38, 16, v38 :: v_dual_bitop2_b32 v7, 1, v7 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x16 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v22, v48, v49 :: v_dual_lshrrev_b32 v39, 16, v39 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v20 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v49, 16, v49 :: v_dual_bitop2_b32 v8, 1, v8 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x14 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v48, 16, v48 :: v_dual_cndmask_b32 v20, v50, v51, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v18 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v51, 16, v51 :: v_dual_bitop2_b32 v12, 1, v12 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x12 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v50, 16, v50 :: v_dual_cndmask_b32 v18, v52, v53, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v16 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v53, 16, v53 :: v_dual_bitop2_b32 v14, 1, v14 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x10 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v52, 16, v52 :: v_dual_cndmask_b32 v16, v54, v55, vcc_lo +; GFX1250FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v14 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v55, 16, v55 :: v_dual_lshrrev_b32 v54, 16, v54 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0xc +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v14, v66, v67, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v12 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v67, 16, v67 :: v_dual_lshrrev_b32 v66, 16, v66 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x8 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v12, v70, v71, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v10 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v70, 16, v70 :: v_dual_bitop2_b32 v25, 1, v25 bitop3:0x40 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x5 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v10, v81, v82 :: v_dual_lshrrev_b32 v71, 16, v71 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v8 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v82, 16, v82 :: v_dual_bitop2_b32 v27, 1, v27 bitop3:0x40 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v8, v69, v80 :: v_dual_lshrrev_b32 v81, 16, v81 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v6 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v80, 16, v80 :: v_dual_lshrrev_b32 v69, 16, v69 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x4 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v6, v68, v29 :: v_dual_lshrrev_b32 v29, 16, v29 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v4 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v68, 16, v68 :: v_dual_cndmask_b32 v4, v64, v65, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v2 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v65, 16, v65 :: v_dual_lshrrev_b32 v64, 16, v64 +; GFX1250FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250FAKE16-NEXT: v_dual_cndmask_b32 v2, v32, v33 :: v_dual_lshrrev_b32 v33, 16, v33 +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v32, 16, v32 :: v_dual_cndmask_b32 v0, v35, v83, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v27 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v83, 16, v83 :: v_dual_cndmask_b32 v27, v36, v37, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v25 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v25, v38, v39, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v23 +; GFX1250FAKE16-NEXT: v_dual_lshrrev_b32 v35, 16, v35 :: v_dual_cndmask_b32 v23, v48, v49, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v21 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v21, v50, v51, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v19 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v19, v52, v53, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v17 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v17, v54, v55, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v15 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v15, v66, v67, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v13 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v13, v70, v71, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v11 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v11, v81, v82, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v7 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v7, v68, v29, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v3 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v3, v32, v33, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v1 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v1, v35, v83, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v5 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v5, v64, v65, vcc_lo +; GFX1250FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v9 +; GFX1250FAKE16-NEXT: v_cndmask_b32_e32 v9, v69, v80, vcc_lo +; GFX1250FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v1, v3, v2, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v2, v5, v4, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v3, v7, v6, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v4, v9, v8, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v5, v11, v10, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v6, v13, v12, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v7, v15, v14, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v8, v17, v16, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v9, v19, v18, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v10, v21, v20, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v11, v23, v22, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v12, v25, v24, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v13, v27, v26, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v14, v28, v31, 0x5040100 +; GFX1250FAKE16-NEXT: v_perm_b32 v15, v34, v30, 0x5040100 +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = select <32 x i1> %cond, <32 x bfloat> %a, <32 x bfloat> %b ret <32 x bfloat> %op } @@ -49167,12 +50464,21 @@ define bfloat @v_fma_bf16(bfloat %a, bfloat %b, bfloat %c) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fma_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fma_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fma_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.fma.bf16(bfloat %a, bfloat %b, bfloat %c) ret bfloat %op } @@ -54791,12 +56097,21 @@ define bfloat @v_fmuladd_bf16(bfloat %a, bfloat %b, bfloat %c) { ; GFX11FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0 ; GFX11FAKE16-NEXT: s_setpc_b64 s[30:31] ; -; GFX1250-LABEL: v_fmuladd_bf16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] -; GFX1250-NEXT: s_set_pc_i64 s[30:31] +; GFX1250TRUE16-LABEL: v_fmuladd_bf16: +; GFX1250TRUE16: ; %bb.0: +; GFX1250TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250TRUE16-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250TRUE16-NEXT: v_cvt_pk_bf16_f32 v0, v0, s0 +; GFX1250TRUE16-NEXT: s_set_pc_i64 s[30:31] +; +; GFX1250FAKE16-LABEL: v_fmuladd_bf16: +; GFX1250FAKE16: ; %bb.0: +; GFX1250FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250FAKE16-NEXT: v_fma_mixlo_bf16 v0, v0, v1, v2 op_sel_hi:[1,1,1] +; GFX1250FAKE16-NEXT: s_set_pc_i64 s[30:31] %op = call bfloat @llvm.fmuladd.bf16(bfloat %a, bfloat %b, bfloat %c) ret bfloat %op } @@ -55652,5 +56967,3 @@ define <4 x bfloat> @v_fmuladd_v4bf16(<4 x bfloat> %a, <4 x bfloat> %b, <4 x bfl %op = call <4 x bfloat> @llvm.fmuladd.v4bf16(<4 x bfloat> %a, <4 x bfloat> %b, <4 x bfloat> %c) ret <4 x bfloat> %op } -;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: -; GFX1250FAKE16: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll index 363a248..cbf6b66 100644 --- a/llvm/test/CodeGen/AMDGPU/calling-conventions.ll +++ b/llvm/test/CodeGen/AMDGPU/calling-conventions.ll @@ -1262,7 +1262,7 @@ define amdgpu_ps void @ps_mesa_i16(i16 %arg0) { ; GFX1250-TRUE16-LABEL: ps_mesa_i16: ; GFX1250-TRUE16: ; %bb.0: ; GFX1250-TRUE16-NEXT: v_add_nc_u16 v0.l, v0.l, v0.l -; GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v0 +; GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v0, off ; GFX1250-TRUE16-NEXT: s_endpgm ; ; GFX1250-FAKE16-LABEL: ps_mesa_i16: @@ -3013,7 +3013,7 @@ define amdgpu_cs void @amdgpu_cs_v8i1(<8 x i1> %arg0) { ; GFX1250-TRUE16-NEXT: v_lshlrev_b16 v0.h, 4, v0.h ; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX1250-TRUE16-NEXT: v_bitop3_b16 v0.l, v0.l, v0.h, 15 bitop3:0xec -; GFX1250-TRUE16-NEXT: flat_store_b8 v[0:1], v0 +; GFX1250-TRUE16-NEXT: global_store_b8 v[0:1], v0, off ; GFX1250-TRUE16-NEXT: s_endpgm ; ; GFX1250-FAKE16-LABEL: amdgpu_cs_v8i1: @@ -3297,7 +3297,7 @@ define amdgpu_cs void @amdgpu_cs_v16i1(<16 x i1> %arg0) { ; GFX1250-TRUE16-NEXT: v_or_b16 v0.h, v1.h, v1.l ; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX1250-TRUE16-NEXT: v_bitop3_b16 v0.l, v0.l, v0.h, 0xff bitop3:0xec -; GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v0 +; GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v0, off ; GFX1250-TRUE16-NEXT: s_endpgm ; ; GFX1250-FAKE16-LABEL: amdgpu_cs_v16i1: diff --git a/llvm/test/CodeGen/AMDGPU/carryout-selection.ll b/llvm/test/CodeGen/AMDGPU/carryout-selection.ll index 2ae6fc2..4a6fa4f 100644 --- a/llvm/test/CodeGen/AMDGPU/carryout-selection.ll +++ b/llvm/test/CodeGen/AMDGPU/carryout-selection.ll @@ -691,7 +691,8 @@ define amdgpu_kernel void @uaddo32_vcc_user(ptr addrspace(1) %out, ptr addrspace ; GCN-ISEL-LABEL: name: suaddo64 ; GCN-ISEL-LABEL: body: ; GCN-ISEL-LABEL: bb.0 -; GCN-ISEL: S_ADD_U64_PSEUDO +; GCN-ISEL: S_UADDO_PSEUDO +; GCN-ISEL: S_ADD_CO_PSEUDO define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i64 %a, i64 %b) #0 { ; CISI-LABEL: suaddo64: @@ -700,21 +701,23 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; CISI-NEXT: s_mov_b32 s11, 0xf000 ; CISI-NEXT: s_mov_b32 s10, -1 ; CISI-NEXT: s_waitcnt lgkmcnt(0) -; CISI-NEXT: s_add_u32 s6, s4, s6 -; CISI-NEXT: v_mov_b32_e32 v0, s4 -; CISI-NEXT: s_addc_u32 s7, s5, s7 -; CISI-NEXT: v_mov_b32_e32 v1, s5 -; CISI-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[0:1] -; CISI-NEXT: v_mov_b32_e32 v2, s6 +; CISI-NEXT: s_add_u32 s4, s4, s6 +; CISI-NEXT: s_cselect_b64 s[12:13], -1, 0 +; CISI-NEXT: s_or_b32 s6, s12, s13 +; CISI-NEXT: s_cmp_lg_u32 s6, 0 +; CISI-NEXT: s_addc_u32 s5, s5, s7 ; CISI-NEXT: s_mov_b32 s8, s0 ; CISI-NEXT: s_mov_b32 s9, s1 +; CISI-NEXT: v_mov_b32_e32 v0, s4 +; CISI-NEXT: v_mov_b32_e32 v1, s5 +; CISI-NEXT: s_cselect_b64 s[4:5], -1, 0 ; CISI-NEXT: s_mov_b32 s0, s2 ; CISI-NEXT: s_mov_b32 s1, s3 ; CISI-NEXT: s_mov_b32 s2, s10 ; CISI-NEXT: s_mov_b32 s3, s11 -; CISI-NEXT: v_mov_b32_e32 v3, s7 -; CISI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; CISI-NEXT: buffer_store_dwordx2 v[2:3], off, s[8:11], 0 +; CISI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; CISI-NEXT: s_waitcnt expcnt(0) +; CISI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] ; CISI-NEXT: buffer_store_byte v0, off, s[0:3], 0 ; CISI-NEXT: s_endpgm ; @@ -722,37 +725,37 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_add_u32 s2, s4, s6 ; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: s_add_u32 s0, s4, s6 -; VI-NEXT: v_mov_b32_e32 v4, s4 ; VI-NEXT: v_mov_b32_e32 v1, s1 -; VI-NEXT: s_addc_u32 s1, s5, s7 -; VI-NEXT: v_mov_b32_e32 v5, s5 -; VI-NEXT: v_mov_b32_e32 v7, s1 -; VI-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[4:5] -; VI-NEXT: v_mov_b32_e32 v6, s0 -; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: s_cmp_lg_u64 s[0:1], 0 +; VI-NEXT: s_addc_u32 s0, s5, s7 +; VI-NEXT: v_mov_b32_e32 v4, s2 +; VI-NEXT: v_mov_b32_e32 v5, s0 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 ; VI-NEXT: v_mov_b32_e32 v3, s3 -; VI-NEXT: flat_store_dwordx2 v[0:1], v[6:7] -; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc +; VI-NEXT: flat_store_dwordx2 v[0:1], v[4:5] +; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; VI-NEXT: flat_store_byte v[2:3], v0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: suaddo64: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v4, 0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_add_u32 s0, s12, s14 -; GFX9-NEXT: v_mov_b32_e32 v0, s12 -; GFX9-NEXT: v_mov_b32_e32 v1, s13 -; GFX9-NEXT: s_addc_u32 s1, s13, s15 -; GFX9-NEXT: v_mov_b32_e32 v3, s1 -; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[0:1] -; GFX9-NEXT: v_mov_b32_e32 v2, s0 -; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GFX9-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] -; GFX9-NEXT: global_store_byte v4, v0, s[10:11] +; GFX9-NEXT: s_add_u32 s2, s12, s14 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: s_cmp_lg_u64 s[0:1], 0 +; GFX9-NEXT: s_addc_u32 s0, s13, s15 +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_mov_b32_e32 v1, s0 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[0:1] +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] +; GFX9-NEXT: global_store_byte v2, v3, s[10:11] ; GFX9-NEXT: s_endpgm ; ; GFX1010-LABEL: suaddo64: @@ -761,10 +764,12 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1010-NEXT: v_mov_b32_e32 v2, 0 ; GFX1010-NEXT: s_waitcnt lgkmcnt(0) ; GFX1010-NEXT: s_add_u32 s0, s12, s14 -; GFX1010-NEXT: s_addc_u32 s1, s13, s15 +; GFX1010-NEXT: s_cselect_b32 s1, -1, 0 ; GFX1010-NEXT: v_mov_b32_e32 v0, s0 +; GFX1010-NEXT: s_cmp_lg_u32 s1, 0 +; GFX1010-NEXT: s_addc_u32 s1, s13, s15 +; GFX1010-NEXT: s_cselect_b32 s0, -1, 0 ; GFX1010-NEXT: v_mov_b32_e32 v1, s1 -; GFX1010-NEXT: v_cmp_lt_u64_e64 s0, s[0:1], s[12:13] ; GFX1010-NEXT: v_cndmask_b32_e64 v3, 0, 1, s0 ; GFX1010-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] ; GFX1010-NEXT: global_store_byte v2, v3, s[10:11] @@ -775,11 +780,13 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W32-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; GFX1030W32-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W32-NEXT: s_waitcnt lgkmcnt(0) -; GFX1030W32-NEXT: s_add_u32 s6, s4, s6 -; GFX1030W32-NEXT: s_addc_u32 s7, s5, s7 -; GFX1030W32-NEXT: v_mov_b32_e32 v0, s6 -; GFX1030W32-NEXT: v_cmp_lt_u64_e64 s4, s[6:7], s[4:5] -; GFX1030W32-NEXT: v_mov_b32_e32 v1, s7 +; GFX1030W32-NEXT: s_add_u32 s4, s4, s6 +; GFX1030W32-NEXT: s_cselect_b32 s6, -1, 0 +; GFX1030W32-NEXT: v_mov_b32_e32 v0, s4 +; GFX1030W32-NEXT: s_cmp_lg_u32 s6, 0 +; GFX1030W32-NEXT: s_addc_u32 s5, s5, s7 +; GFX1030W32-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1030W32-NEXT: v_mov_b32_e32 v1, s5 ; GFX1030W32-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1030W32-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W32-NEXT: global_store_byte v2, v3, s[2:3] @@ -790,11 +797,13 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W64-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; GFX1030W64-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W64-NEXT: s_waitcnt lgkmcnt(0) -; GFX1030W64-NEXT: s_add_u32 s6, s4, s6 -; GFX1030W64-NEXT: s_addc_u32 s7, s5, s7 -; GFX1030W64-NEXT: v_mov_b32_e32 v0, s6 -; GFX1030W64-NEXT: v_cmp_lt_u64_e64 s[4:5], s[6:7], s[4:5] -; GFX1030W64-NEXT: v_mov_b32_e32 v1, s7 +; GFX1030W64-NEXT: s_add_u32 s4, s4, s6 +; GFX1030W64-NEXT: s_cselect_b64 s[8:9], -1, 0 +; GFX1030W64-NEXT: v_mov_b32_e32 v0, s4 +; GFX1030W64-NEXT: s_cmp_lg_u64 s[8:9], 0 +; GFX1030W64-NEXT: s_addc_u32 s5, s5, s7 +; GFX1030W64-NEXT: v_mov_b32_e32 v1, s5 +; GFX1030W64-NEXT: s_cselect_b64 s[4:5], -1, 0 ; GFX1030W64-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[4:5] ; GFX1030W64-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W64-NEXT: global_store_byte v2, v3, s[2:3] @@ -804,12 +813,13 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX11: ; %bb.0: ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_add_u32 s6, s4, s6 -; GFX11-NEXT: s_addc_u32 s7, s5, s7 -; GFX11-NEXT: v_mov_b32_e32 v0, s6 -; GFX11-NEXT: v_cmp_lt_u64_e64 s4, s[6:7], s[4:5] -; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s7 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-NEXT: s_add_u32 s4, s4, s6 +; GFX11-NEXT: s_cselect_b32 s6, -1, 0 +; GFX11-NEXT: v_mov_b32_e32 v0, s4 +; GFX11-NEXT: s_cmp_lg_u32 s6, 0 +; GFX11-NEXT: s_addc_u32 s5, s5, s7 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s5 ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] @@ -819,12 +829,14 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1250-LABEL: suaddo64: ; GFX1250: ; %bb.0: ; GFX1250-NEXT: s_load_b256 s[8:15], s[4:5], 0x24 -; GFX1250-NEXT: v_mov_b32_e32 v2, 0 ; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_add_nc_u64 s[0:1], s[12:13], s[14:15] -; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mov_b64_e32 v[0:1], s[0:1] -; GFX1250-NEXT: v_cmp_lt_u64_e64 s0, s[0:1], s[12:13] +; GFX1250-NEXT: s_add_co_u32 s0, s12, s14 +; GFX1250-NEXT: s_cselect_b32 s1, -1, 0 +; GFX1250-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v0, s0 +; GFX1250-NEXT: s_cmp_lg_u32 s1, 0 +; GFX1250-NEXT: s_add_co_ci_u32 s1, s13, s15 +; GFX1250-NEXT: s_cselect_b32 s0, -1, 0 +; GFX1250-NEXT: v_mov_b32_e32 v1, s1 ; GFX1250-NEXT: v_cndmask_b32_e64 v3, 0, 1, s0 ; GFX1250-NEXT: s_clause 0x1 ; GFX1250-NEXT: global_store_b64 v2, v[0:1], s[8:9] @@ -841,7 +853,8 @@ define amdgpu_kernel void @suaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GCN-ISEL-LABEL: name: vuaddo64 ; GCN-ISEL-LABEL: body: ; GCN-ISEL-LABEL: bb.0 -; GCN-ISEL: V_ADD_U64_PSEUDO +; GCN-ISEL: V_ADD_CO_U32_e64 +; GCN-ISEL: V_ADDC_U32_e64 define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i64 %a) #0 { ; CISI-LABEL: vuaddo64: @@ -854,9 +867,8 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; CISI-NEXT: s_mov_b32 s4, s0 ; CISI-NEXT: v_mov_b32_e32 v1, s9 ; CISI-NEXT: v_add_i32_e32 v0, vcc, s8, v0 -; CISI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; CISI-NEXT: v_cmp_gt_u64_e32 vcc, s[8:9], v[0:1] ; CISI-NEXT: s_mov_b32 s5, s1 +; CISI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; CISI-NEXT: s_mov_b32 s0, s2 ; CISI-NEXT: s_mov_b32 s1, s3 ; CISI-NEXT: s_mov_b32 s2, s6 @@ -876,7 +888,6 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; VI-NEXT: v_mov_b32_e32 v6, s5 ; VI-NEXT: v_add_u32_e32 v5, vcc, s4, v0 ; VI-NEXT: v_addc_u32_e32 v6, vcc, 0, v6, vcc -; VI-NEXT: v_cmp_gt_u64_e32 vcc, s[4:5], v[5:6] ; VI-NEXT: v_mov_b32_e32 v2, s1 ; VI-NEXT: v_mov_b32_e32 v3, s2 ; VI-NEXT: v_mov_b32_e32 v4, s3 @@ -894,7 +905,6 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX9-NEXT: v_mov_b32_e32 v1, s7 ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v0 ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc -; GFX9-NEXT: v_cmp_gt_u64_e32 vcc, s[6:7], v[0:1] ; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; GFX9-NEXT: global_store_byte v2, v0, s[2:3] @@ -909,8 +919,7 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1010-NEXT: s_waitcnt lgkmcnt(0) ; GFX1010-NEXT: v_add_co_u32 v0, s4, s6, v0 ; GFX1010-NEXT: v_add_co_ci_u32_e64 v1, s4, s7, 0, s4 -; GFX1010-NEXT: v_cmp_gt_u64_e32 vcc_lo, s[6:7], v[0:1] -; GFX1010-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc_lo +; GFX1010-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1010-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1010-NEXT: global_store_byte v2, v3, s[2:3] ; GFX1010-NEXT: s_endpgm @@ -923,9 +932,8 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W32-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W32-NEXT: s_waitcnt lgkmcnt(0) ; GFX1030W32-NEXT: v_add_co_u32 v0, s4, s6, v0 -; GFX1030W32-NEXT: v_add_co_ci_u32_e64 v1, null, s7, 0, s4 -; GFX1030W32-NEXT: v_cmp_gt_u64_e32 vcc_lo, s[6:7], v[0:1] -; GFX1030W32-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc_lo +; GFX1030W32-NEXT: v_add_co_ci_u32_e64 v1, s4, s7, 0, s4 +; GFX1030W32-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1030W32-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W32-NEXT: global_store_byte v2, v3, s[2:3] ; GFX1030W32-NEXT: s_endpgm @@ -938,9 +946,8 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W64-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W64-NEXT: s_waitcnt lgkmcnt(0) ; GFX1030W64-NEXT: v_add_co_u32 v0, s[4:5], s6, v0 -; GFX1030W64-NEXT: v_add_co_ci_u32_e64 v1, null, s7, 0, s[4:5] -; GFX1030W64-NEXT: v_cmp_gt_u64_e32 vcc, s[6:7], v[0:1] -; GFX1030W64-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc +; GFX1030W64-NEXT: v_add_co_ci_u32_e64 v1, s[4:5], s7, 0, s[4:5] +; GFX1030W64-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[4:5] ; GFX1030W64-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W64-NEXT: global_store_byte v2, v3, s[2:3] ; GFX1030W64-NEXT: s_endpgm @@ -955,10 +962,9 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-NEXT: v_add_co_u32 v0, s4, s6, v0 -; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, s7, 0, s4 +; GFX11-NEXT: v_add_co_ci_u32_e64 v1, s4, s7, 0, s4 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_cmp_gt_u64_e32 vcc_lo, s[6:7], v[0:1] -; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc_lo +; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] ; GFX11-NEXT: global_store_b8 v2, v3, s[2:3] @@ -969,16 +975,17 @@ define amdgpu_kernel void @vuaddo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1250-NEXT: s_clause 0x1 ; GFX1250-NEXT: s_load_b64 s[6:7], s[4:5], 0x34 ; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GFX1250-NEXT: v_mov_b32_e32 v1, 0 ; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; GFX1250-NEXT: v_mov_b32_e32 v2, 0 ; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_add_nc_u64_e32 v[2:3], s[6:7], v[0:1] -; GFX1250-NEXT: v_cmp_gt_u64_e32 vcc_lo, s[6:7], v[2:3] -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250-NEXT: v_add_co_u32 v0, s4, s6, v0 +; GFX1250-NEXT: v_add_co_ci_u32_e64 v1, s4, s7, 0, s4 +; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: global_store_b64 v1, v[2:3], s[0:1] -; GFX1250-NEXT: global_store_b8 v1, v0, s[2:3] +; GFX1250-NEXT: global_store_b64 v2, v[0:1], s[0:1] +; GFX1250-NEXT: global_store_b8 v2, v3, s[2:3] ; GFX1250-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %tid.ext = sext i32 %tid to i64 @@ -1671,7 +1678,8 @@ define amdgpu_kernel void @usubo32_vcc_user(ptr addrspace(1) %out, ptr addrspace ; GCN-ISEL-LABEL: name: susubo64 ; GCN-ISEL-LABEL: body: ; GCN-ISEL-LABEL: bb.0 -; GCN-ISEL: S_SUB_U64_PSEUDO +; GCN-ISEL: S_USUBO_PSEUDO +; GCN-ISEL: S_SUB_CO_PSEUDO define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i64 %a, i64 %b) #0 { ; CISI-LABEL: susubo64: @@ -1680,21 +1688,23 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; CISI-NEXT: s_mov_b32 s11, 0xf000 ; CISI-NEXT: s_mov_b32 s10, -1 ; CISI-NEXT: s_waitcnt lgkmcnt(0) -; CISI-NEXT: s_sub_u32 s6, s4, s6 -; CISI-NEXT: v_mov_b32_e32 v0, s4 -; CISI-NEXT: s_subb_u32 s7, s5, s7 -; CISI-NEXT: v_mov_b32_e32 v1, s5 -; CISI-NEXT: v_cmp_gt_u64_e32 vcc, s[6:7], v[0:1] -; CISI-NEXT: v_mov_b32_e32 v2, s6 +; CISI-NEXT: s_sub_u32 s4, s4, s6 +; CISI-NEXT: s_cselect_b64 s[12:13], -1, 0 +; CISI-NEXT: s_or_b32 s6, s12, s13 +; CISI-NEXT: s_cmp_lg_u32 s6, 0 +; CISI-NEXT: s_subb_u32 s5, s5, s7 ; CISI-NEXT: s_mov_b32 s8, s0 ; CISI-NEXT: s_mov_b32 s9, s1 +; CISI-NEXT: v_mov_b32_e32 v0, s4 +; CISI-NEXT: v_mov_b32_e32 v1, s5 +; CISI-NEXT: s_cselect_b64 s[4:5], -1, 0 ; CISI-NEXT: s_mov_b32 s0, s2 ; CISI-NEXT: s_mov_b32 s1, s3 ; CISI-NEXT: s_mov_b32 s2, s10 ; CISI-NEXT: s_mov_b32 s3, s11 -; CISI-NEXT: v_mov_b32_e32 v3, s7 -; CISI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; CISI-NEXT: buffer_store_dwordx2 v[2:3], off, s[8:11], 0 +; CISI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; CISI-NEXT: s_waitcnt expcnt(0) +; CISI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] ; CISI-NEXT: buffer_store_byte v0, off, s[0:3], 0 ; CISI-NEXT: s_endpgm ; @@ -1702,37 +1712,37 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_sub_u32 s2, s4, s6 ; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: s_sub_u32 s0, s4, s6 -; VI-NEXT: v_mov_b32_e32 v4, s4 ; VI-NEXT: v_mov_b32_e32 v1, s1 -; VI-NEXT: s_subb_u32 s1, s5, s7 -; VI-NEXT: v_mov_b32_e32 v5, s5 -; VI-NEXT: v_mov_b32_e32 v7, s1 -; VI-NEXT: v_cmp_gt_u64_e32 vcc, s[0:1], v[4:5] -; VI-NEXT: v_mov_b32_e32 v6, s0 -; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: s_cmp_lg_u64 s[0:1], 0 +; VI-NEXT: s_subb_u32 s0, s5, s7 +; VI-NEXT: v_mov_b32_e32 v4, s2 +; VI-NEXT: v_mov_b32_e32 v5, s0 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 ; VI-NEXT: v_mov_b32_e32 v3, s3 -; VI-NEXT: flat_store_dwordx2 v[0:1], v[6:7] -; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc +; VI-NEXT: flat_store_dwordx2 v[0:1], v[4:5] +; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; VI-NEXT: flat_store_byte v[2:3], v0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: susubo64: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v4, 0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_sub_u32 s0, s12, s14 -; GFX9-NEXT: v_mov_b32_e32 v0, s12 -; GFX9-NEXT: v_mov_b32_e32 v1, s13 -; GFX9-NEXT: s_subb_u32 s1, s13, s15 -; GFX9-NEXT: v_mov_b32_e32 v3, s1 -; GFX9-NEXT: v_cmp_gt_u64_e32 vcc, s[0:1], v[0:1] -; GFX9-NEXT: v_mov_b32_e32 v2, s0 -; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GFX9-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] -; GFX9-NEXT: global_store_byte v4, v0, s[10:11] +; GFX9-NEXT: s_sub_u32 s2, s12, s14 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: s_cmp_lg_u64 s[0:1], 0 +; GFX9-NEXT: s_subb_u32 s0, s13, s15 +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_mov_b32_e32 v1, s0 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[0:1] +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] +; GFX9-NEXT: global_store_byte v2, v3, s[10:11] ; GFX9-NEXT: s_endpgm ; ; GFX1010-LABEL: susubo64: @@ -1741,10 +1751,12 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1010-NEXT: v_mov_b32_e32 v2, 0 ; GFX1010-NEXT: s_waitcnt lgkmcnt(0) ; GFX1010-NEXT: s_sub_u32 s0, s12, s14 -; GFX1010-NEXT: s_subb_u32 s1, s13, s15 +; GFX1010-NEXT: s_cselect_b32 s1, -1, 0 ; GFX1010-NEXT: v_mov_b32_e32 v0, s0 +; GFX1010-NEXT: s_cmp_lg_u32 s1, 0 +; GFX1010-NEXT: s_subb_u32 s1, s13, s15 +; GFX1010-NEXT: s_cselect_b32 s0, -1, 0 ; GFX1010-NEXT: v_mov_b32_e32 v1, s1 -; GFX1010-NEXT: v_cmp_gt_u64_e64 s0, s[0:1], s[12:13] ; GFX1010-NEXT: v_cndmask_b32_e64 v3, 0, 1, s0 ; GFX1010-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] ; GFX1010-NEXT: global_store_byte v2, v3, s[10:11] @@ -1755,11 +1767,13 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W32-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; GFX1030W32-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W32-NEXT: s_waitcnt lgkmcnt(0) -; GFX1030W32-NEXT: s_sub_u32 s6, s4, s6 -; GFX1030W32-NEXT: s_subb_u32 s7, s5, s7 -; GFX1030W32-NEXT: v_mov_b32_e32 v0, s6 -; GFX1030W32-NEXT: v_cmp_gt_u64_e64 s4, s[6:7], s[4:5] -; GFX1030W32-NEXT: v_mov_b32_e32 v1, s7 +; GFX1030W32-NEXT: s_sub_u32 s4, s4, s6 +; GFX1030W32-NEXT: s_cselect_b32 s6, -1, 0 +; GFX1030W32-NEXT: v_mov_b32_e32 v0, s4 +; GFX1030W32-NEXT: s_cmp_lg_u32 s6, 0 +; GFX1030W32-NEXT: s_subb_u32 s5, s5, s7 +; GFX1030W32-NEXT: s_cselect_b32 s4, -1, 0 +; GFX1030W32-NEXT: v_mov_b32_e32 v1, s5 ; GFX1030W32-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1030W32-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W32-NEXT: global_store_byte v2, v3, s[2:3] @@ -1770,11 +1784,13 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W64-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; GFX1030W64-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W64-NEXT: s_waitcnt lgkmcnt(0) -; GFX1030W64-NEXT: s_sub_u32 s6, s4, s6 -; GFX1030W64-NEXT: s_subb_u32 s7, s5, s7 -; GFX1030W64-NEXT: v_mov_b32_e32 v0, s6 -; GFX1030W64-NEXT: v_cmp_gt_u64_e64 s[4:5], s[6:7], s[4:5] -; GFX1030W64-NEXT: v_mov_b32_e32 v1, s7 +; GFX1030W64-NEXT: s_sub_u32 s4, s4, s6 +; GFX1030W64-NEXT: s_cselect_b64 s[8:9], -1, 0 +; GFX1030W64-NEXT: v_mov_b32_e32 v0, s4 +; GFX1030W64-NEXT: s_cmp_lg_u64 s[8:9], 0 +; GFX1030W64-NEXT: s_subb_u32 s5, s5, s7 +; GFX1030W64-NEXT: v_mov_b32_e32 v1, s5 +; GFX1030W64-NEXT: s_cselect_b64 s[4:5], -1, 0 ; GFX1030W64-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[4:5] ; GFX1030W64-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W64-NEXT: global_store_byte v2, v3, s[2:3] @@ -1784,12 +1800,13 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX11: ; %bb.0: ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_sub_u32 s6, s4, s6 -; GFX11-NEXT: s_subb_u32 s7, s5, s7 -; GFX11-NEXT: v_mov_b32_e32 v0, s6 -; GFX11-NEXT: v_cmp_gt_u64_e64 s4, s[6:7], s[4:5] -; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s7 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-NEXT: s_sub_u32 s4, s4, s6 +; GFX11-NEXT: s_cselect_b32 s6, -1, 0 +; GFX11-NEXT: v_mov_b32_e32 v0, s4 +; GFX11-NEXT: s_cmp_lg_u32 s6, 0 +; GFX11-NEXT: s_subb_u32 s5, s5, s7 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s5 ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] @@ -1799,12 +1816,14 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1250-LABEL: susubo64: ; GFX1250: ; %bb.0: ; GFX1250-NEXT: s_load_b256 s[8:15], s[4:5], 0x24 -; GFX1250-NEXT: v_mov_b32_e32 v2, 0 ; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_sub_nc_u64 s[0:1], s[12:13], s[14:15] -; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_mov_b64_e32 v[0:1], s[0:1] -; GFX1250-NEXT: v_cmp_gt_u64_e64 s0, s[0:1], s[12:13] +; GFX1250-NEXT: s_sub_co_u32 s0, s12, s14 +; GFX1250-NEXT: s_cselect_b32 s1, -1, 0 +; GFX1250-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v0, s0 +; GFX1250-NEXT: s_cmp_lg_u32 s1, 0 +; GFX1250-NEXT: s_sub_co_ci_u32 s1, s13, s15 +; GFX1250-NEXT: s_cselect_b32 s0, -1, 0 +; GFX1250-NEXT: v_mov_b32_e32 v1, s1 ; GFX1250-NEXT: v_cndmask_b32_e64 v3, 0, 1, s0 ; GFX1250-NEXT: s_clause 0x1 ; GFX1250-NEXT: global_store_b64 v2, v[0:1], s[8:9] @@ -1821,7 +1840,8 @@ define amdgpu_kernel void @susubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GCN-ISEL-LABEL: name: vusubo64 ; GCN-ISEL-LABEL: body: ; GCN-ISEL-LABEL: bb.0 -; GCN-ISEL: V_SUB_U64_PSEUDO +; GCN-ISEL: V_SUB_CO_U32_e64 +; GCN-ISEL: V_SUBB_U32_e64 define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i64 %a) #0 { ; CISI-LABEL: vusubo64: @@ -1834,9 +1854,8 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; CISI-NEXT: s_mov_b32 s4, s0 ; CISI-NEXT: v_mov_b32_e32 v1, s9 ; CISI-NEXT: v_sub_i32_e32 v0, vcc, s8, v0 -; CISI-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc -; CISI-NEXT: v_cmp_lt_u64_e32 vcc, s[8:9], v[0:1] ; CISI-NEXT: s_mov_b32 s5, s1 +; CISI-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc ; CISI-NEXT: s_mov_b32 s0, s2 ; CISI-NEXT: s_mov_b32 s1, s3 ; CISI-NEXT: s_mov_b32 s2, s6 @@ -1856,7 +1875,6 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; VI-NEXT: v_mov_b32_e32 v6, s5 ; VI-NEXT: v_sub_u32_e32 v5, vcc, s4, v0 ; VI-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v6, vcc -; VI-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[5:6] ; VI-NEXT: v_mov_b32_e32 v2, s1 ; VI-NEXT: v_mov_b32_e32 v3, s2 ; VI-NEXT: v_mov_b32_e32 v4, s3 @@ -1874,7 +1892,6 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX9-NEXT: v_mov_b32_e32 v1, s7 ; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, s6, v0 ; GFX9-NEXT: v_subbrev_co_u32_e32 v1, vcc, 0, v1, vcc -; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[0:1] ; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; GFX9-NEXT: global_store_byte v2, v0, s[2:3] @@ -1889,8 +1906,7 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1010-NEXT: s_waitcnt lgkmcnt(0) ; GFX1010-NEXT: v_sub_co_u32 v0, s4, s6, v0 ; GFX1010-NEXT: v_sub_co_ci_u32_e64 v1, s4, s7, 0, s4 -; GFX1010-NEXT: v_cmp_lt_u64_e32 vcc_lo, s[6:7], v[0:1] -; GFX1010-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc_lo +; GFX1010-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1010-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1010-NEXT: global_store_byte v2, v3, s[2:3] ; GFX1010-NEXT: s_endpgm @@ -1903,9 +1919,8 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W32-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W32-NEXT: s_waitcnt lgkmcnt(0) ; GFX1030W32-NEXT: v_sub_co_u32 v0, s4, s6, v0 -; GFX1030W32-NEXT: v_sub_co_ci_u32_e64 v1, null, s7, 0, s4 -; GFX1030W32-NEXT: v_cmp_lt_u64_e32 vcc_lo, s[6:7], v[0:1] -; GFX1030W32-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc_lo +; GFX1030W32-NEXT: v_sub_co_ci_u32_e64 v1, s4, s7, 0, s4 +; GFX1030W32-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1030W32-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W32-NEXT: global_store_byte v2, v3, s[2:3] ; GFX1030W32-NEXT: s_endpgm @@ -1918,9 +1933,8 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1030W64-NEXT: v_mov_b32_e32 v2, 0 ; GFX1030W64-NEXT: s_waitcnt lgkmcnt(0) ; GFX1030W64-NEXT: v_sub_co_u32 v0, s[4:5], s6, v0 -; GFX1030W64-NEXT: v_sub_co_ci_u32_e64 v1, null, s7, 0, s[4:5] -; GFX1030W64-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[0:1] -; GFX1030W64-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc +; GFX1030W64-NEXT: v_sub_co_ci_u32_e64 v1, s[4:5], s7, 0, s[4:5] +; GFX1030W64-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[4:5] ; GFX1030W64-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX1030W64-NEXT: global_store_byte v2, v3, s[2:3] ; GFX1030W64-NEXT: s_endpgm @@ -1935,10 +1949,9 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-NEXT: v_sub_co_u32 v0, s4, s6, v0 -; GFX11-NEXT: v_sub_co_ci_u32_e64 v1, null, s7, 0, s4 +; GFX11-NEXT: v_sub_co_ci_u32_e64 v1, s4, s7, 0, s4 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_cmp_lt_u64_e32 vcc_lo, s[6:7], v[0:1] -; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc_lo +; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] ; GFX11-NEXT: global_store_b8 v2, v3, s[2:3] @@ -1949,16 +1962,17 @@ define amdgpu_kernel void @vusubo64(ptr addrspace(1) %out, ptr addrspace(1) %car ; GFX1250-NEXT: s_clause 0x1 ; GFX1250-NEXT: s_load_b64 s[6:7], s[4:5], 0x34 ; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GFX1250-NEXT: v_mov_b32_e32 v1, 0 ; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; GFX1250-NEXT: v_mov_b32_e32 v2, 0 ; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX1250-NEXT: v_sub_nc_u64_e32 v[2:3], s[6:7], v[0:1] -; GFX1250-NEXT: v_cmp_lt_u64_e32 vcc_lo, s[6:7], v[2:3] -; GFX1250-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250-NEXT: v_sub_co_u32 v0, s4, s6, v0 +; GFX1250-NEXT: v_sub_co_ci_u32_e64 v1, s4, s7, 0, s4 +; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: global_store_b64 v1, v[2:3], s[0:1] -; GFX1250-NEXT: global_store_b8 v1, v0, s[2:3] +; GFX1250-NEXT: global_store_b64 v2, v[0:1], s[0:1] +; GFX1250-NEXT: global_store_b8 v2, v3, s[2:3] ; GFX1250-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %tid.ext = sext i32 %tid to i64 diff --git a/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll b/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll index f706f53..eb40e5c 100644 --- a/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll +++ b/llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll @@ -35,6 +35,6 @@ define amdgpu_kernel void @test_direct_indirect_call() { ret void } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll b/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll index 8da204b..c02ff28 100644 --- a/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll +++ b/llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll @@ -28,6 +28,6 @@ define amdgpu_kernel void @test_simple_indirect_call() #0 { attributes #0 = { "amdgpu-no-dispatch-id" } ;. -; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_GCN: attributes #[[ATTR1]] = { "amdgpu-no-dispatch-id" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll b/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll index ab51693..05d3e9c3 100644 --- a/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll +++ b/llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.ll @@ -497,12 +497,10 @@ define amdgpu_kernel void @test_fold_canonicalize_minnum_value_f32(ptr addrspace ret void } -; FIXME: Should there be more checks here? minnum with NaN operand is simplified away. +; FIXME: Should there be more checks here? minnum with sNaN operand is simplified to qNaN. ; GCN-LABEL: test_fold_canonicalize_sNaN_value_f32: -; GCN: {{flat|global}}_load_dword [[LOAD:v[0-9]+]] -; VI: v_mul_f32_e32 v{{[0-9]+}}, 1.0, [[LOAD]] -; GFX9: v_max_f32_e32 v{{[0-9]+}}, [[LOAD]], [[LOAD]] +; GCN: v_mov_b32_e32 v{{.+}}, 0x7fc00000 define amdgpu_kernel void @test_fold_canonicalize_sNaN_value_f32(ptr addrspace(1) %arg) { %id = tail call i32 @llvm.amdgcn.workitem.id.x() %gep = getelementptr inbounds float, ptr addrspace(1) %arg, i32 %id diff --git a/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll b/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll index 3de6df2..833be20 100644 --- a/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll +++ b/llvm/test/CodeGen/AMDGPU/fneg-combines.new.ll @@ -1949,8 +1949,7 @@ define float @v_fneg_self_minimumnum_f32_ieee(float %a) #0 { ; GCN-LABEL: v_fneg_self_minimumnum_f32_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_mul_f32_e32 v0, -1.0, v0 -; GCN-NEXT: v_max_f32_e32 v0, v0, v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %min = call float @llvm.minimumnum.f32(float %a, float %a) %min.fneg = fneg float %min @@ -1961,7 +1960,7 @@ define float @v_fneg_self_minimumnum_f32_no_ieee(float %a) #4 { ; GCN-LABEL: v_fneg_self_minimumnum_f32_no_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_max_f32_e64 v0, -v0, -v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %min = call float @llvm.minimumnum.f32(float %a, float %a) %min.fneg = fneg float %min @@ -2285,8 +2284,7 @@ define float @v_fneg_self_maximumnum_f32_ieee(float %a) #0 { ; GCN-LABEL: v_fneg_self_maximumnum_f32_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_mul_f32_e32 v0, -1.0, v0 -; GCN-NEXT: v_min_f32_e32 v0, v0, v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %max = call float @llvm.maximumnum.f32(float %a, float %a) %max.fneg = fneg float %max @@ -2297,7 +2295,7 @@ define float @v_fneg_self_maximumnum_f32_no_ieee(float %a) #4 { ; GCN-LABEL: v_fneg_self_maximumnum_f32_no_ieee: ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-NEXT: v_min_f32_e64 v0, -v0, -v0 +; GCN-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] %max = call float @llvm.maximumnum.f32(float %a, float %a) %max.fneg = fneg float %max diff --git a/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll b/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll index 40d2765..b0dd187 100644 --- a/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/fptrunc.f16.ll @@ -11,9 +11,9 @@ ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -global-isel=0 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-SDAG-FAKE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GISEL-TRUE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-GISEL-FAKE16 %s -; TODO: FIXME-TRUE16 llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=0 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-SDAG-TRUE16 %s +; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=0 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-SDAG-TRUE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=0 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-SDAG-FAKE16 %s -; TODO: FIXME-TRUE16 llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=1 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-GISEL-TRUE16 %s +; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=1 -mattr=-flat-for-global,+real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-GISEL-TRUE16 %s ; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1250 -global-isel=1 -mattr=-flat-for-global,-real-true16 -denormal-fp-math=preserve-sign < %s | FileCheck -enable-var-scope -check-prefixes=GFX1250-GISEL-FAKE16 %s define amdgpu_kernel void @fptrunc_f32_to_f16( @@ -197,6 +197,24 @@ define amdgpu_kernel void @fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -215,6 +233,21 @@ define amdgpu_kernel void @fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -419,6 +452,24 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_afn(ptr addrspace(1) %r, ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16_afn: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16_afn: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -437,6 +488,21 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_afn(ptr addrspace(1) %r, ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16_afn: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16_afn: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1160,6 +1226,73 @@ define amdgpu_kernel void @fptrunc_f64_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f64_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s2, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s3, s2, 0x1ff +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s2, 8 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v0, s3, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_bfe_u32 s3, s2, 0xb0014 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-SDAG-TRUE16-NEXT: s_sub_co_i32 s4, 0x3f1, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2) +; GFX1250-SDAG-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX1250-SDAG-TRUE16-NEXT: v_med3_i32 v1, s4, 0, 13 +; GFX1250-SDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s8, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s4, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s4, s5, s4 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s4, 0x1000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s9, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s9, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s8, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_addk_co_i32 s3, 0xfc10 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s9, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s3, 12 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s4, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s8, s5, 7 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_gt_i32 s8, 5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s9, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s8, 3 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s5, 2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s8, s9 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_add_co_i32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 31 +; GFX1250-SDAG-TRUE16-NEXT: s_movk_i32 s8, 0x7e00 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s4, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s4, s8, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s3, 0x40f +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s4, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s2, s2, 16 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s2, s2, 0x8000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s2, s2, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f64_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1227,6 +1360,63 @@ define amdgpu_kernel void @fptrunc_f64_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f64_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b64 s[2:3], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s6, s3, 0x1ff +; GFX1250-GISEL-TRUE16-NEXT: s_bfe_u32 s4, s3, 0xb0014 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s5, s3, 8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s6, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_addk_co_i32 s4, 0xfc10 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s2, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s5, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s2, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_sub_co_i32 s6, 1, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s8, s2, 0x1000 +; GFX1250-GISEL-TRUE16-NEXT: s_max_i32 s6, s6, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s7, s4, 12 +; GFX1250-GISEL-TRUE16-NEXT: s_min_i32 s6, s6, 13 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s5, s5, 9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s9, s8, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s2, s7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s6, s9, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s5, s5, 0x7c00 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s6, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s9, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lt_i32 s4, 1 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, s6, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s6, s2, 7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s2, s2, 2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s6, 3 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s7, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s6, 5 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s7, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_add_co_i32 s2, s2, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s4, 30 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, 0x7c00, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s4, 0x40f +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, s5, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s3, 16 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s3, s3, 0x8000 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s3, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f64_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1489,6 +1679,26 @@ define amdgpu_kernel void @fptrunc_f64_to_f16_afn( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f64_to_f16_afn: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, v[0:1] +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f64_to_f16_afn: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1509,6 +1719,20 @@ define amdgpu_kernel void @fptrunc_f64_to_f16_afn( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f64_to_f16_afn: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b64 s[2:3], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, s[2:3] +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f64_to_f16_afn: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1740,6 +1964,24 @@ define amdgpu_kernel void @fptrunc_v2f32_to_v2f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_v2f32_to_v2f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b64 v[0:1], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v1 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_v2f32_to_v2f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -1758,6 +2000,20 @@ define amdgpu_kernel void @fptrunc_v2f32_to_v2f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_v2f32_to_v2f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b64 s[2:3], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b64_e32 v[0:1], s[2:3] +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_v2f32_to_v2f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3017,6 +3273,122 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_v2f64_to_v2f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b128 v[0:3], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s2, v3 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s3, s2, 0x1ff +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s2, 8 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v2, s3, v2 +; GFX1250-SDAG-TRUE16-NEXT: s_bfe_u32 s3, s2, 0xb0014 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-SDAG-TRUE16-NEXT: s_sub_co_i32 s4, 0x3f1, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2) +; GFX1250-SDAG-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v2 +; GFX1250-SDAG-TRUE16-NEXT: v_med3_i32 v3, s4, 0, 13 +; GFX1250-SDAG-TRUE16-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s8, v3 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s4, v2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s4, s5, s4 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s4, 0x1000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s9, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s9, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s8, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_addk_co_i32 s3, 0xfc10 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s5, s9, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s8, s3, 12 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s4, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s8, s5, 7 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_gt_i32 s8, 5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s9, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s8, 3 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s5, s5, 2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s8, s8, s9 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_add_co_i32 s5, s5, s8 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s3, 31 +; GFX1250-SDAG-TRUE16-NEXT: s_movk_i32 s8, 0x7e00 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s5, s5, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s4, 0 +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s4, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s9, s8, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s3, 0x40f +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s9, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s5, s4, 0x1ff +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s10, s4, 8 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v0, s5, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_bfe_u32 s5, s4, 0xb0014 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s10, s10, 0xffe +; GFX1250-SDAG-TRUE16-NEXT: s_sub_co_i32 s9, 0x3f1, s5 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s2, s2, 16 +; GFX1250-SDAG-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX1250-SDAG-TRUE16-NEXT: v_med3_i32 v1, s9, 0, 13 +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s2, s2, 0x8000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s2, s2, s3 +; GFX1250-SDAG-TRUE16-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s11, v1 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_readfirstlane_b32 s9, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s9, s10, s9 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s10, s9, 0x1000 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s12, s10, s11 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s11, s12, s11 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s11, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_addk_co_i32 s5, 0xfc10 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s3, s12, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_lshl_b32 s10, s5, 12 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s10, s9, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s5, 1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s3, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s10, s3, 7 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_gt_i32 s10, 5 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s11, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s10, 3 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s10, 1, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s3, s3, 2 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s10, s10, s11 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_add_co_i32 s3, s3, s10 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lt_i32 s5, 31 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s3, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_lg_u32 s9, 0 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s8, s8, 0x7c00 +; GFX1250-SDAG-TRUE16-NEXT: s_cmp_eq_u32 s5, 0x40f +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_cselect_b32 s3, s8, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_lshr_b32 s4, s4, 16 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: s_and_b32 s4, s4, 0x8000 +; GFX1250-SDAG-TRUE16-NEXT: s_or_b32 s3, s4, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: s_pack_ll_b32_b16 s2, s3, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_v2f64_to_v2f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3133,6 +3505,109 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_v2f64_to_v2f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[4:7], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s8, s5, 0x1ff +; GFX1250-GISEL-TRUE16-NEXT: s_bfe_u32 s2, s5, 0xb0014 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s5, 8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s4, s8, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_addk_co_i32 s2, 0xfc10 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s3, s3, 0xffe +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s4, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s4, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s3, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s3, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s4, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_sub_co_i32 s8, 1, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s10, s3, 0x1000 +; GFX1250-GISEL-TRUE16-NEXT: s_max_i32 s8, s8, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s9, s2, 12 +; GFX1250-GISEL-TRUE16-NEXT: s_min_i32 s8, s8, 13 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s4, s4, 9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s11, s10, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s3, s9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s8, s11, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s4, s4, 0x7c00 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s8, s10 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s8, s11, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lt_i32 s2, 1 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, s8, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s8, s3, 7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s3, 2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s8, 3 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s9, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s8, 5 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s8, s9, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_add_co_i32 s3, s3, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s2, 30 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, 0x7c00, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s2, 0x40f +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s2, s4, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s5, 16 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s8, s7, 0x1ff +; GFX1250-GISEL-TRUE16-NEXT: s_bfe_u32 s4, s7, 0xb0014 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s5, s7, 8 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s3, s3, 0x8000 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s8, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_addk_co_i32 s4, 0xfc10 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s5, s5, 0xffe +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s2, s3, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s6, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s5, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s3, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s5, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_sub_co_i32 s6, 1, s4 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s9, s3, 0x1000 +; GFX1250-GISEL-TRUE16-NEXT: s_max_i32 s6, s6, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s8, s4, 12 +; GFX1250-GISEL-TRUE16-NEXT: s_min_i32 s6, s6, 13 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s5, s5, 9 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s10, s9, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s3, s8 +; GFX1250-GISEL-TRUE16-NEXT: s_lshl_b32 s6, s10, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s5, s5, 0x7c00 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lg_u32 s6, s9 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s10, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_lt_i32 s4, 1 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, s6, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s6, s3, 7 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s3, s3, 2 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s6, 3 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s8, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s6, 5 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s6, 1, 0 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s6, s8, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_add_co_i32 s3, s3, s6 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_gt_i32 s4, 30 +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, 0x7c00, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_cmp_eq_u32 s4, 0x40f +; GFX1250-GISEL-TRUE16-NEXT: s_cselect_b32 s3, s5, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_lshr_b32 s4, s7, 16 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s4, s4, 0x8000 +; GFX1250-GISEL-TRUE16-NEXT: s_or_b32 s3, s4, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_pack_ll_b32_b16 s2, s2, s3 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_v2f64_to_v2f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3481,6 +3956,27 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16_afn( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_v2f64_to_v2f16_afn: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b128 v[0:3], off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f32_f64_e32 v2, v[2:3] +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, v[0:1] +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_pk_f16_f32 v0, v0, v2 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_v2f64_to_v2f16_afn: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3502,6 +3998,25 @@ define amdgpu_kernel void @fptrunc_v2f64_to_v2f16_afn( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_v2f64_to_v2f16_afn: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[4:7], s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f32_f64_e32 v0, s[4:5] +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f32_f64_e32 v1, s[6:7] +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-GISEL-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v1 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-GISEL-TRUE16-NEXT: v_pack_b32_f16 v0, v0.l, v0.h +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_v2f64_to_v2f16_afn: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3710,6 +4225,26 @@ define amdgpu_kernel void @fneg_fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fneg_fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80000000, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fneg_fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3730,6 +4265,22 @@ define amdgpu_kernel void @fneg_fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fneg_fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_xor_b32 s2, s2, 0x80000000 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fneg_fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3936,6 +4487,26 @@ define amdgpu_kernel void @fabs_fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fabs_fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fabs_fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -3956,6 +4527,22 @@ define amdgpu_kernel void @fabs_fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fabs_fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_bitset0_b32 s2, 31 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fabs_fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4162,6 +4749,26 @@ define amdgpu_kernel void @fneg_fabs_fptrunc_f32_to_f16( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b16 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_or_b32_e32 v0, 0x80000000, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4182,6 +4789,22 @@ define amdgpu_kernel void @fneg_fabs_fptrunc_f32_to_f16( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_bitset1_b32 s2, 31 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b16 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fneg_fabs_fptrunc_f32_to_f16: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4396,6 +5019,26 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_zext_i32( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16_zext_i32: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16_zext_i32: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4416,6 +5059,22 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_zext_i32( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16_zext_i32: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s2, 0xffff, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16_zext_i32: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4630,6 +5289,27 @@ define amdgpu_kernel void @fptrunc_fabs_f32_to_f16_zext_i32( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: v_and_b32_e32 v0, 0xffff, v0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4651,6 +5331,24 @@ define amdgpu_kernel void @fptrunc_fabs_f32_to_f16_zext_i32( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_bitset0_b32 s2, 31 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_3) +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_and_b32 s2, 0xffff, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_fabs_f32_to_f16_zext_i32: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4877,6 +5575,26 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_sext_i32( ; GFX11-GISEL-FAKE16-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-GISEL-FAKE16-NEXT: s_endpgm ; +; GFX1250-SDAG-TRUE16-LABEL: fptrunc_f32_to_f16_sext_i32: +; GFX1250-SDAG-TRUE16: ; %bb.0: ; %entry +; GFX1250-SDAG-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s6, -1 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s7, 0x31016000 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s10, s6 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s11, s7 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s8, s2 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s9, s3 +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s4, s0 +; GFX1250-SDAG-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], null +; GFX1250-SDAG-TRUE16-NEXT: s_mov_b32 s5, s1 +; GFX1250-SDAG-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-SDAG-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v0 +; GFX1250-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX1250-SDAG-TRUE16-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GFX1250-SDAG-TRUE16-NEXT: buffer_store_b32 v0, off, s[4:7], null +; GFX1250-SDAG-TRUE16-NEXT: s_endpgm +; ; GFX1250-SDAG-FAKE16-LABEL: fptrunc_f32_to_f16_sext_i32: ; GFX1250-SDAG-FAKE16: ; %bb.0: ; %entry ; GFX1250-SDAG-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 @@ -4897,6 +5615,22 @@ define amdgpu_kernel void @fptrunc_f32_to_f16_sext_i32( ; GFX1250-SDAG-FAKE16-NEXT: buffer_store_b32 v0, off, s[4:7], null ; GFX1250-SDAG-FAKE16-NEXT: s_endpgm ; +; GFX1250-GISEL-TRUE16-LABEL: fptrunc_f32_to_f16_sext_i32: +; GFX1250-GISEL-TRUE16: ; %bb.0: ; %entry +; GFX1250-GISEL-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_load_b32 s2, s[2:3], 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_xcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s3, 0x31016000 +; GFX1250-GISEL-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-GISEL-TRUE16-NEXT: s_cvt_f16_f32 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3) | instskip(NEXT) | instid1(SALU_CYCLE_1) +; GFX1250-GISEL-TRUE16-NEXT: s_sext_i32_i16 s2, s2 +; GFX1250-GISEL-TRUE16-NEXT: v_mov_b32_e32 v0, s2 +; GFX1250-GISEL-TRUE16-NEXT: s_mov_b32 s2, -1 +; GFX1250-GISEL-TRUE16-NEXT: buffer_store_b32 v0, off, s[0:3], null +; GFX1250-GISEL-TRUE16-NEXT: s_endpgm +; ; GFX1250-GISEL-FAKE16-LABEL: fptrunc_f32_to_f16_sext_i32: ; GFX1250-GISEL-FAKE16: ; %bb.0: ; %entry ; GFX1250-GISEL-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 diff --git a/llvm/test/CodeGen/AMDGPU/fsub.ll b/llvm/test/CodeGen/AMDGPU/fsub.ll index 743431c..d6a9cb1 100644 --- a/llvm/test/CodeGen/AMDGPU/fsub.ll +++ b/llvm/test/CodeGen/AMDGPU/fsub.ll @@ -92,43 +92,11 @@ define amdgpu_kernel void @v_fneg_fsub_nsz_f32(ptr addrspace(1) %out, ptr addrsp ret void } -; FUNC-LABEL: {{^}}v_fneg_fsub_nsz_attribute_f32: -; SI: v_sub_f32_e32 [[SUB:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}} -; SI-NOT: xor -define amdgpu_kernel void @v_fneg_fsub_nsz_attribute_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { - %b_ptr = getelementptr float, ptr addrspace(1) %in, i32 1 - %a = load float, ptr addrspace(1) %in, align 4 - %b = load float, ptr addrspace(1) %b_ptr, align 4 - %result = fsub float %a, %b - %neg.result = fsub float -0.0, %result - store float %neg.result, ptr addrspace(1) %out, align 4 - ret void -} - -; For some reason the attribute has a string "true" or "false", so -; make sure it is disabled and the fneg is not folded if it is not -; "true". -; FUNC-LABEL: {{^}}v_fneg_fsub_nsz_false_attribute_f32: -; SI: v_sub_f32_e32 [[SUB:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}} -; SI: v_xor_b32_e32 v{{[0-9]+}}, 0x80000000, [[SUB]] -define amdgpu_kernel void @v_fneg_fsub_nsz_false_attribute_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #1 { - %b_ptr = getelementptr float, ptr addrspace(1) %in, i32 1 - %a = load float, ptr addrspace(1) %in, align 4 - %b = load float, ptr addrspace(1) %b_ptr, align 4 - %result = fsub float %a, %b - %neg.result = fsub float -0.0, %result - store float %neg.result, ptr addrspace(1) %out, align 4 - ret void -} - -; FUNC-LABEL: {{^}}v_fsub_0_nsz_attribute_f32: +; FUNC-LABEL: {{^}}v_fsub_0_nsz_flag_f32: ; SI-NOT: v_sub -define amdgpu_kernel void @v_fsub_0_nsz_attribute_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { +define amdgpu_kernel void @v_fsub_0_nsz_flag_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) { %a = load float, ptr addrspace(1) %in, align 4 - %result = fsub float %a, 0.0 + %result = fsub nsz float %a, 0.0 store float %result, ptr addrspace(1) %out, align 4 ret void } - -attributes #0 = { nounwind "no-signed-zeros-fp-math"="true" } -attributes #1 = { nounwind "no-signed-zeros-fp-math"="false" } diff --git a/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll b/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll index 3089054..32f7d6b 100644 --- a/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll +++ b/llvm/test/CodeGen/AMDGPU/implicitarg-offset-attributes.ll @@ -276,23 +276,23 @@ attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memo ;. ; V4: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; V4: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V4: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V4: attributes #[[ATTR5]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } ;. ; V5: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; V5: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V5: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V5: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V5: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V5: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } ;. ; V6: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } -; V6: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V6: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V6: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } -; V6: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-default-queue" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } +; V6: attributes #[[ATTR4]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-multigrid-sync-arg" "uniform-work-group-size"="false" } ;. ; V4: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. diff --git a/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll b/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll index d3ef1b7..a0f5d2f 100644 --- a/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll +++ b/llvm/test/CodeGen/AMDGPU/indirect-call-set-from-other-function.ll @@ -68,6 +68,6 @@ if.end: ret void } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll b/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll index 71a330e..4e952b6 100644 --- a/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll +++ b/llvm/test/CodeGen/AMDGPU/issue120256-annotate-constexpr-addrspacecast.ll @@ -55,8 +55,8 @@ define amdgpu_kernel void @issue120256_private(ptr addrspace(1) %out) { ; FIXME: Inference of amdgpu-no-queue-ptr should not depend on code object version. !0 = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "target-cpu"="gfx803" "uniform-work-group-size"="false" } ;. ; CHECK: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll index 6ccfad7..ff47563 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.f16.ll @@ -14,7 +14,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_bf8_f16 v0.l, v0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[2:3], v0 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[2:3], v0, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_bf8_f16_v: @@ -28,7 +28,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_dual_mov_b32 v4, v1 :: v_dual_mov_b32 v5, v2 ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_bf8_f16 v0.l, v0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[4:5], v0 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[4:5], v0, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_bf8_f16_v: @@ -46,7 +46,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_bf8_f16_s: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, s0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_bf8_f16_s: @@ -58,7 +58,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_bf8_f16_s: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, s0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_bf8_f16_s: @@ -75,7 +75,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_l(ptr addrspace(1) %out) { ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_bf8_f16_l: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, 0x56400000 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_bf8_f16_l: @@ -87,7 +87,7 @@ define amdgpu_ps void @test_cvt_pk_bf8_f16_l(ptr addrspace(1) %out) { ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_bf8_f16_l: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_bf8_f16 v2.l, 0x56400000 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_bf8_f16_l: @@ -105,7 +105,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_mov_b32 v2, v1 ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_fp8_f16 v0.l, v0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[2:3], v0 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[2:3], v0, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_fp8_f16_v: @@ -119,7 +119,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_v(<2 x half> %a, ptr addrspace(1) %ou ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_dual_mov_b32 v4, v1 :: v_dual_mov_b32 v5, v2 ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_fp8_f16 v0.l, v0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[4:5], v0 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[4:5], v0, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_fp8_f16_v: @@ -137,7 +137,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_fp8_f16_s: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, s0 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_fp8_f16_s: @@ -149,7 +149,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_s(<2 x half> inreg %a, ptr addrspace( ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_fp8_f16_s: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, s0 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_fp8_f16_s: @@ -166,7 +166,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_l(ptr addrspace(1) %out) { ; GFX1250-SDAG-REAL16-LABEL: test_cvt_pk_fp8_f16_l: ; GFX1250-SDAG-REAL16: ; %bb.0: ; GFX1250-SDAG-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, 0x56400000 -; GFX1250-SDAG-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-SDAG-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-SDAG-REAL16-NEXT: s_endpgm ; ; GFX1250-SDAG-FAKE16-LABEL: test_cvt_pk_fp8_f16_l: @@ -178,7 +178,7 @@ define amdgpu_ps void @test_cvt_pk_fp8_f16_l(ptr addrspace(1) %out) { ; GFX1250-GISEL-REAL16-LABEL: test_cvt_pk_fp8_f16_l: ; GFX1250-GISEL-REAL16: ; %bb.0: ; GFX1250-GISEL-REAL16-NEXT: v_cvt_pk_fp8_f16 v2.l, 0x56400000 -; GFX1250-GISEL-REAL16-NEXT: flat_store_b16 v[0:1], v2 +; GFX1250-GISEL-REAL16-NEXT: global_store_b16 v[0:1], v2, off ; GFX1250-GISEL-REAL16-NEXT: s_endpgm ; ; GFX1250-GISEL-FAKE16-LABEL: test_cvt_pk_fp8_f16_l: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll index 1e44a09..dbea832 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rcp.bf16.ll @@ -15,7 +15,7 @@ define amdgpu_kernel void @rcp_bf16(ptr addrspace(1) %out, bfloat %src) #1 { ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 ; SDAG-TRUE16-NEXT: v_rcp_bf16_e32 v0.l, s2 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_bf16: @@ -35,10 +35,10 @@ define amdgpu_kernel void @rcp_bf16_constant_4(ptr addrspace(1) %out) #1 { ; SDAG-TRUE16-LABEL: rcp_bf16_constant_4: ; SDAG-TRUE16: ; %bb.0: ; SDAG-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 -; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3e80 ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 +; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3e80 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_bf16_constant_4: @@ -57,10 +57,10 @@ define amdgpu_kernel void @rcp_bf16_constant_100(ptr addrspace(1) %out) #1 { ; SDAG-TRUE16-LABEL: rcp_bf16_constant_100: ; SDAG-TRUE16: ; %bb.0: ; SDAG-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 -; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3c24 ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 +; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x3c24 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_bf16_constant_100: @@ -79,10 +79,10 @@ define amdgpu_kernel void @rcp_undef_bf16(ptr addrspace(1) %out) #1 { ; SDAG-TRUE16-LABEL: rcp_undef_bf16: ; SDAG-TRUE16: ; %bb.0: ; SDAG-TRUE16-NEXT: s_load_b64 s[0:1], s[4:5], 0x0 -; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7fc0 ; SDAG-TRUE16-NEXT: v_mov_b32_e32 v1, 0 +; SDAG-TRUE16-NEXT: v_mov_b16_e32 v0.l, 0x7fc0 ; SDAG-TRUE16-NEXT: s_wait_kmcnt 0x0 -; SDAG-TRUE16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-TRUE16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rcp_undef_bf16: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll index 42d12fd..662dc613 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.rsq.bf16.ll @@ -15,7 +15,7 @@ define amdgpu_kernel void @rsq_bf16(ptr addrspace(1) %out, bfloat %src) #1 { ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 ; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, s2 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rsq_bf16: @@ -38,7 +38,7 @@ define amdgpu_kernel void @rsq_bf16_constant_4(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, 4.0 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rsq_bf16_constant_4: @@ -61,7 +61,7 @@ define amdgpu_kernel void @rsq_bf16_constant_100(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_rsq_bf16_e32 v0.l, 0x42c8 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: rsq_bf16_constant_100: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll index dd89f80..ba769ef 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.tanh.ll @@ -100,7 +100,7 @@ define amdgpu_kernel void @tanh_f16(ptr addrspace(1) %out, half %src) #1 { ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 ; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, s2 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_f16: @@ -123,7 +123,7 @@ define amdgpu_kernel void @tanh_f16_constant_4.0(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, 4.0 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_f16_constant_4.0: @@ -146,7 +146,7 @@ define amdgpu_kernel void @tanh_f16_constant_100.0(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_f16_e32 v0.l, 0x5640 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_f16_constant_100.0: @@ -182,7 +182,7 @@ define amdgpu_kernel void @tanh_bf16(ptr addrspace(1) %out, bfloat %src) #1 { ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 ; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, s2 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_bf16: @@ -205,7 +205,7 @@ define amdgpu_kernel void @tanh_bf16_constant_4(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, 4.0 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_bf16_constant_4: @@ -228,7 +228,7 @@ define amdgpu_kernel void @tanh_bf16_constant_100(ptr addrspace(1) %out) #1 { ; SDAG-REAL16-NEXT: v_tanh_bf16_e32 v0.l, 0x42c8 ; SDAG-REAL16-NEXT: v_mov_b32_e32 v1, 0 ; SDAG-REAL16-NEXT: s_wait_kmcnt 0x0 -; SDAG-REAL16-NEXT: flat_store_b16 v1, v0, s[0:1] +; SDAG-REAL16-NEXT: global_store_b16 v1, v0, s[0:1] ; SDAG-REAL16-NEXT: s_endpgm ; ; SDAG-FAKE16-LABEL: tanh_bf16_constant_100: diff --git a/llvm/test/CodeGen/AMDGPU/mad_int24.ll b/llvm/test/CodeGen/AMDGPU/mad_int24.ll index 93fda94..dd88310 100644 --- a/llvm/test/CodeGen/AMDGPU/mad_int24.ll +++ b/llvm/test/CodeGen/AMDGPU/mad_int24.ll @@ -1,17 +1,79 @@ -; RUN: llc < %s -mtriple=amdgcn | FileCheck %s --check-prefix=GCN --check-prefix=FUNC -; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global | FileCheck %s --check-prefix=GCN --check-prefix=FUNC -; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG --check-prefix=FUNC -; RUN: llc < %s -mtriple=r600 -mcpu=cayman | FileCheck %s --check-prefix=CM --check-prefix=FUNC +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc < %s -mtriple=amdgcn| FileCheck %s --check-prefixes=GCN +; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global | FileCheck %s --check-prefixes=VI +; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s --check-prefixes=EG,R600,RW +; RUN: llc < %s -mtriple=r600 -mcpu=cayman | FileCheck %s --check-prefixes=EG,R600,CM -; FUNC-LABEL: {{^}}i32_mad24: ; Signed 24-bit multiply is not supported on pre-Cayman GPUs. -; EG: MULLO_INT -; CM: MULLO_INT -; GCN: s_bfe_i32 -; GCN: s_bfe_i32 -; GCN: s_mul_i32 -; GCN: s_add_i32 define amdgpu_kernel void @i32_mad24(ptr addrspace(1) %out, i32 %a, i32 %b, i32 %c) { +; GCN-LABEL: i32_mad24: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0xb +; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_bfe_i32 s0, s0, 0x180000 +; GCN-NEXT: s_bfe_i32 s1, s1, 0x180000 +; GCN-NEXT: s_mul_i32 s0, s0, s1 +; GCN-NEXT: s_add_i32 s0, s0, s2 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: v_mov_b32_e32 v0, s0 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; VI-LABEL: i32_mad24: +; VI: ; %bb.0: ; %entry +; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2c +; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x24 +; VI-NEXT: s_mov_b32 s7, 0xf000 +; VI-NEXT: s_mov_b32 s6, -1 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: s_bfe_i32 s0, s0, 0x180000 +; VI-NEXT: s_bfe_i32 s1, s1, 0x180000 +; VI-NEXT: s_mul_i32 s0, s0, s1 +; VI-NEXT: s_add_i32 s0, s0, s2 +; VI-NEXT: v_mov_b32_e32 v0, s0 +; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_endpgm +; +; RW-LABEL: i32_mad24: +; RW: ; %bb.0: ; %entry +; RW-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[] +; RW-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; RW-NEXT: CF_END +; RW-NEXT: PAD +; RW-NEXT: ALU clause starting at 4: +; RW-NEXT: LSHL T0.W, KC0[2].Z, literal.x, +; RW-NEXT: LSHL * T1.W, KC0[2].W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: ASHR T1.W, PS, literal.x, +; RW-NEXT: ASHR * T0.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: MULLO_INT * T0.X, PS, PV.W, +; RW-NEXT: ADD_INT T0.X, PS, KC0[3].X, +; RW-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; RW-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: i32_mad24: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 12, @4, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: ALU clause starting at 4: +; CM-NEXT: LSHL T0.Z, KC0[2].Z, literal.x, +; CM-NEXT: LSHL * T0.W, KC0[2].W, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: ASHR T1.Z, PV.W, literal.x, +; CM-NEXT: ASHR * T0.W, PV.Z, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, T1.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T1.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T1.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T1.Z, +; CM-NEXT: ADD_INT * T0.X, PV.X, KC0[3].X, +; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) entry: %0 = shl i32 %a, 8 %a_24 = ashr i32 %0, 8 @@ -23,13 +85,25 @@ entry: ret void } -; GCN-LABEL: {{^}}mad24_known_bits_destroyed: -; GCN: s_waitcnt -; GCN-NEXT: v_mad_i32_i24 -; GCN-NEXT: v_mul_i32_i24 -; GCN-NEXT: s_setpc_b64 define i32 @mad24_known_bits_destroyed(i32 %a, i32 %b, i32 %c) { - +; GCN-LABEL: mad24_known_bits_destroyed: +; GCN: ; %bb.0: +; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-NEXT: v_mad_i32_i24 v1, v0, v1, v2 +; GCN-NEXT: v_mul_i32_i24_e32 v0, v1, v0 +; GCN-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: mad24_known_bits_destroyed: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_mad_i32_i24 v1, v0, v1, v2 +; VI-NEXT: v_mul_i32_i24_e32 v0, v1, v0 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; EG-LABEL: mad24_known_bits_destroyed: +; EG: ; %bb.0: +; EG-NEXT: CF_END +; EG-NEXT: PAD %shl.0 = shl i32 %a, 8 %sra.0 = ashr i32 %shl.0, 8 %shl.1 = shl i32 %b, 8 @@ -48,12 +122,25 @@ define i32 @mad24_known_bits_destroyed(i32 %a, i32 %b, i32 %c) { ret i32 %mul1 } -; GCN-LABEL: {{^}}mad24_intrin_known_bits_destroyed: -; GCN: s_waitcnt -; GCN-NEXT: v_mad_i32_i24 -; GCN-NEXT: v_mul_i32_i24 -; GCN-NEXT: s_setpc_b64 define i32 @mad24_intrin_known_bits_destroyed(i32 %a, i32 %b, i32 %c) { +; GCN-LABEL: mad24_intrin_known_bits_destroyed: +; GCN: ; %bb.0: +; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-NEXT: v_mad_i32_i24 v1, v0, v1, v2 +; GCN-NEXT: v_mul_i32_i24_e32 v0, v1, v0 +; GCN-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: mad24_intrin_known_bits_destroyed: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_mad_i32_i24 v1, v0, v1, v2 +; VI-NEXT: v_mul_i32_i24_e32 v0, v1, v0 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; EG-LABEL: mad24_intrin_known_bits_destroyed: +; EG: ; %bb.0: +; EG-NEXT: CF_END +; EG-NEXT: PAD %shl.0 = shl i32 %a, 8 %sra.0 = ashr i32 %shl.0, 8 %shl.1 = shl i32 %b, 8 @@ -73,17 +160,177 @@ define i32 @mad24_intrin_known_bits_destroyed(i32 %a, i32 %b, i32 %c) { } ; Make sure no unnecessary BFEs are emitted in the loop. -; GCN-LABEL: {{^}}mad24_destroyed_knownbits_2: -; GCN-NOT: v_bfe -; GCN: v_mad_i32_i24 -; GCN-NOT: v_bfe -; GCN: v_mad_i32_i24 -; GCN-NOT: v_bfe -; GCN: v_mad_i32_i24 -; GCN-NOT: v_bfe -; GCN: v_mad_i32_i24 -; GCN-NOT: v_bfe define void @mad24_destroyed_knownbits_2(i32 %arg, i32 %arg1, i32 %arg2, ptr addrspace(1) %arg3) { +; GCN-LABEL: mad24_destroyed_knownbits_2: +; GCN: ; %bb.0: ; %bb +; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-NEXT: v_mov_b32_e32 v5, 1 +; GCN-NEXT: s_mov_b64 s[4:5], 0 +; GCN-NEXT: .LBB3_1: ; %bb6 +; GCN-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-NEXT: v_mad_i32_i24 v0, v0, v5, v5 +; GCN-NEXT: v_add_i32_e32 v1, vcc, -1, v1 +; GCN-NEXT: v_mad_i32_i24 v5, v0, v5, v0 +; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-NEXT: v_mad_i32_i24 v0, v5, v0, v5 +; GCN-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GCN-NEXT: v_mad_i32_i24 v0, v0, v5, v0 +; GCN-NEXT: v_mov_b32_e32 v5, v2 +; GCN-NEXT: s_andn2_b64 exec, exec, s[4:5] +; GCN-NEXT: s_cbranch_execnz .LBB3_1 +; GCN-NEXT: ; %bb.2: ; %bb5 +; GCN-NEXT: s_or_b64 exec, exec, s[4:5] +; GCN-NEXT: s_mov_b32 s6, 0 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s4, s6 +; GCN-NEXT: s_mov_b32 s5, s6 +; GCN-NEXT: buffer_store_dword v0, v[3:4], s[4:7], 0 addr64 +; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) +; GCN-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: mad24_destroyed_knownbits_2: +; VI: ; %bb.0: ; %bb +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v5, 1 +; VI-NEXT: s_mov_b64 s[4:5], 0 +; VI-NEXT: .LBB3_1: ; %bb6 +; VI-NEXT: ; =>This Inner Loop Header: Depth=1 +; VI-NEXT: v_mad_i32_i24 v0, v0, v5, v5 +; VI-NEXT: v_mad_i32_i24 v5, v0, v5, v0 +; VI-NEXT: v_add_u32_e32 v1, vcc, -1, v1 +; VI-NEXT: v_mad_i32_i24 v0, v5, v0, v5 +; VI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; VI-NEXT: v_mad_i32_i24 v0, v0, v5, v0 +; VI-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; VI-NEXT: v_mov_b32_e32 v5, v2 +; VI-NEXT: s_andn2_b64 exec, exec, s[4:5] +; VI-NEXT: s_cbranch_execnz .LBB3_1 +; VI-NEXT: ; %bb.2: ; %bb5 +; VI-NEXT: s_or_b64 exec, exec, s[4:5] +; VI-NEXT: flat_store_dword v[3:4], v0 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: s_setpc_b64 s[30:31] +; +; RW-LABEL: mad24_destroyed_knownbits_2: +; RW: ; %bb.0: ; %bb +; RW-NEXT: ALU 5, @10, KC0[CB0:0-32], KC1[] +; RW-NEXT: LOOP_START_DX10 @7 +; RW-NEXT: ALU_PUSH_BEFORE 30, @16, KC0[], KC1[] +; RW-NEXT: JUMP @6 POP:1 +; RW-NEXT: LOOP_BREAK @6 +; RW-NEXT: POP @6 POP:1 +; RW-NEXT: END_LOOP @2 +; RW-NEXT: ALU 1, @47, KC0[], KC1[] +; RW-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; RW-NEXT: CF_END +; RW-NEXT: ALU clause starting at 10: +; RW-NEXT: MOV T0.X, KC0[2].Y, +; RW-NEXT: MOV T0.Y, KC0[2].Z, +; RW-NEXT: MOV * T0.Z, KC0[2].W, +; RW-NEXT: MOV T0.W, KC0[3].X, +; RW-NEXT: MOV * T1.W, literal.x, +; RW-NEXT: 1(1.401298e-45), 0(0.000000e+00) +; RW-NEXT: ALU clause starting at 16: +; RW-NEXT: LSHL T2.W, T1.W, literal.x, +; RW-NEXT: LSHL * T3.W, T0.X, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: ASHR T3.W, PS, literal.x, +; RW-NEXT: ASHR * T2.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: MULLO_INT * T0.X, PV.W, PS, +; RW-NEXT: ADD_INT * T1.W, PS, T1.W, +; RW-NEXT: LSHL * T3.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: ASHR * T3.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: MULLO_INT * T0.X, PV.W, T2.W, +; RW-NEXT: ADD_INT * T1.W, PS, T1.W, +; RW-NEXT: LSHL * T2.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: ASHR * T2.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: MULLO_INT * T0.X, PV.W, T3.W, +; RW-NEXT: ADD_INT * T1.W, PS, T1.W, +; RW-NEXT: LSHL * T3.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: ASHR * T3.W, PV.W, literal.x, +; RW-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; RW-NEXT: ADD_INT T0.Y, T0.Y, literal.x, +; RW-NEXT: MULLO_INT * T0.X, PV.W, T2.W, +; RW-NEXT: -1(nan), 0(0.000000e+00) +; RW-NEXT: ADD_INT T0.X, PS, T1.W, +; RW-NEXT: SETE_INT T2.W, PV.Y, 0.0, +; RW-NEXT: MOV * T1.W, T0.Z, +; RW-NEXT: PRED_SETNE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; RW-NEXT: ALU clause starting at 47: +; RW-NEXT: LSHR * T1.X, T0.W, literal.x, +; RW-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: mad24_destroyed_knownbits_2: +; CM: ; %bb.0: ; %bb +; CM-NEXT: ALU 5, @10, KC0[CB0:0-32], KC1[] +; CM-NEXT: LOOP_START_DX10 @7 +; CM-NEXT: ALU_PUSH_BEFORE 41, @16, KC0[], KC1[] +; CM-NEXT: JUMP @6 POP:1 +; CM-NEXT: LOOP_BREAK @6 +; CM-NEXT: POP @6 POP:1 +; CM-NEXT: END_LOOP @2 +; CM-NEXT: ALU 1, @58, KC0[], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T1.X, T0.X +; CM-NEXT: CF_END +; CM-NEXT: ALU clause starting at 10: +; CM-NEXT: MOV * T1.X, KC0[2].Y, +; CM-NEXT: MOV T0.X, KC0[2].Z, +; CM-NEXT: MOV T0.Y, KC0[2].W, +; CM-NEXT: MOV T0.Z, KC0[3].X, +; CM-NEXT: MOV * T0.W, literal.x, +; CM-NEXT: 1(1.401298e-45), 0(0.000000e+00) +; CM-NEXT: ALU clause starting at 16: +; CM-NEXT: LSHL T1.Z, T0.W, literal.x, +; CM-NEXT: LSHL * T1.W, T1.X, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: ASHR T2.Z, PV.W, literal.x, +; CM-NEXT: ASHR * T1.W, PV.Z, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T1.X, T2.Z, T1.W, +; CM-NEXT: MULLO_INT T1.Y (MASKED), T2.Z, T1.W, +; CM-NEXT: MULLO_INT T1.Z (MASKED), T2.Z, T1.W, +; CM-NEXT: MULLO_INT * T1.W (MASKED), T2.Z, T1.W, +; CM-NEXT: ADD_INT * T0.W, PV.X, T0.W, +; CM-NEXT: LSHL * T2.W, PV.W, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: ASHR * T2.W, PV.W, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T1.X, T2.W, T1.W, +; CM-NEXT: MULLO_INT T1.Y (MASKED), T2.W, T1.W, +; CM-NEXT: MULLO_INT T1.Z (MASKED), T2.W, T1.W, +; CM-NEXT: MULLO_INT * T1.W (MASKED), T2.W, T1.W, +; CM-NEXT: ADD_INT * T0.W, PV.X, T0.W, +; CM-NEXT: LSHL * T1.W, PV.W, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: ASHR * T1.W, PV.W, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T1.X, T1.W, T2.W, +; CM-NEXT: MULLO_INT T1.Y (MASKED), T1.W, T2.W, +; CM-NEXT: MULLO_INT T1.Z (MASKED), T1.W, T2.W, +; CM-NEXT: MULLO_INT * T1.W (MASKED), T1.W, T2.W, +; CM-NEXT: ADD_INT * T0.W, PV.X, T0.W, +; CM-NEXT: LSHL * T2.W, PV.W, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: ADD_INT T0.X, T0.X, literal.x, +; CM-NEXT: ASHR * T2.W, PV.W, literal.y, +; CM-NEXT: -1(nan), 8(1.121039e-44) +; CM-NEXT: MULLO_INT T1.X, T2.W, T1.W, +; CM-NEXT: MULLO_INT T1.Y (MASKED), T2.W, T1.W, +; CM-NEXT: MULLO_INT T1.Z (MASKED), T2.W, T1.W, +; CM-NEXT: MULLO_INT * T1.W (MASKED), T2.W, T1.W, +; CM-NEXT: ADD_INT T1.X, PV.X, T0.W, +; CM-NEXT: SETE_INT T1.Z, T0.X, 0.0, +; CM-NEXT: MOV * T0.W, T0.Y, +; CM-NEXT: PRED_SETNE_INT * ExecMask,PredicateBit (MASKED), PV.Z, 0.0, +; CM-NEXT: ALU clause starting at 58: +; CM-NEXT: LSHR * T0.X, T0.Z, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) bb: br label %bb6 @@ -119,3 +366,5 @@ bb6: ; preds = %bb6, %bb } declare i32 @llvm.amdgcn.mul.i24(i32, i32) +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; R600: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/mad_uint24.ll b/llvm/test/CodeGen/AMDGPU/mad_uint24.ll index a6d458e..46b8df4 100644 --- a/llvm/test/CodeGen/AMDGPU/mad_uint24.ll +++ b/llvm/test/CodeGen/AMDGPU/mad_uint24.ll @@ -1,19 +1,75 @@ -; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG --check-prefix=FUNC -; RUN: llc < %s -mtriple=r600 -mcpu=cayman | FileCheck %s --check-prefix=EG --check-prefix=FUNC -; RUN: llc < %s -mtriple=amdgcn | FileCheck %s --check-prefix=SI --check-prefix=FUNC --check-prefix=GCN -; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global | FileCheck %s --check-prefix=VI --check-prefix=FUNC --check-prefix=GCN --check-prefix=GCN2 -; RUN: llc < %s -mtriple=amdgcn -mcpu=fiji -mattr=-flat-for-global | FileCheck %s --check-prefix=VI --check-prefix=FUNC --check-prefix=GCN --check-prefix=GCN2 +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s --check-prefixes=EG +; RUN: llc < %s -mtriple=r600 -mcpu=cayman | FileCheck %s --check-prefixes=CM +; RUN: llc < %s -mtriple=amdgcn | FileCheck %s --check-prefixes=GCN +; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global | FileCheck %s --check-prefixes=GFX8,SI +; RUN: llc < %s -mtriple=amdgcn -mcpu=fiji -mattr=-flat-for-global | FileCheck %s --check-prefixes=GFX8,VI declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone -; FUNC-LABEL: {{^}}u32_mad24: -; EG: MULLO_INT -; SI: s_mul_i32 -; SI: s_add_i32 -; VI: s_mul_{{[iu]}}32 -; VI: s_add_{{[iu]}}32 - define amdgpu_kernel void @u32_mad24(ptr addrspace(1) %out, i32 %a, i32 %b, i32 %c) { +; EG-LABEL: u32_mad24: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 6, @4, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 4: +; EG-NEXT: AND_INT T0.W, KC0[2].W, literal.x, +; EG-NEXT: AND_INT * T1.W, KC0[2].Z, literal.x, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: MULLO_INT * T0.X, PS, PV.W, +; EG-NEXT: ADD_INT T0.X, PS, KC0[3].X, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: u32_mad24: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 9, @4, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: ALU clause starting at 4: +; CM-NEXT: AND_INT T0.Z, KC0[2].W, literal.x, +; CM-NEXT: AND_INT * T0.W, KC0[2].Z, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, T0.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T0.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T0.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T0.Z, +; CM-NEXT: ADD_INT * T0.X, PV.X, KC0[3].X, +; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: u32_mad24: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0xb +; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_and_b32 s0, s0, 0xffffff +; GCN-NEXT: s_and_b32 s1, s1, 0xffffff +; GCN-NEXT: s_mul_i32 s0, s0, s1 +; GCN-NEXT: s_add_i32 s0, s0, s2 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: v_mov_b32_e32 v0, s0 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; GFX8-LABEL: u32_mad24: +; GFX8: ; %bb.0: ; %entry +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2c +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_and_b32 s0, s0, 0xffffff +; GFX8-NEXT: s_and_b32 s1, s1, 0xffffff +; GFX8-NEXT: s_mul_i32 s0, s0, s1 +; GFX8-NEXT: s_add_i32 s0, s0, s2 +; GFX8-NEXT: v_mov_b32_e32 v0, s0 +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm entry: %0 = shl i32 %a, 8 %a_24 = lshr i32 %0, 8 @@ -25,18 +81,88 @@ entry: ret void } -; FUNC-LABEL: {{^}}i16_mad24: ; The order of A and B does not matter. -; EG: MULLO_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; EG: ADD_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] ; The result must be sign-extended -; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x -; EG: 16 -; GCN: s_mul_i32 [[MUL:s[0-9]]], {{[s][0-9], [s][0-9]}} -; GCN: s_add_i32 [[MAD:s[0-9]]], [[MUL]], s{{[0-9]}} -; GCN: s_sext_i32_i16 [[EXT:s[0-9]]], [[MAD]] -; GCN: v_mov_b32_e32 v0, [[EXT]] define amdgpu_kernel void @i16_mad24(ptr addrspace(1) %out, i16 %a, i16 %b, i16 %c) { +; EG-LABEL: i16_mad24: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 0, @12, KC0[], KC1[] +; EG-NEXT: TEX 2 @6 +; EG-NEXT: ALU 4, @13, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_16 T1.X, T0.X, 40, #3 +; EG-NEXT: VTX_READ_16 T2.X, T0.X, 42, #3 +; EG-NEXT: VTX_READ_16 T0.X, T0.X, 44, #3 +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: MOV * T0.X, 0.0, +; EG-NEXT: ALU clause starting at 13: +; EG-NEXT: MULLO_INT * T0.Y, T1.X, T2.X, +; EG-NEXT: ADD_INT * T0.W, PS, T0.X, +; EG-NEXT: BFE_INT T0.X, PV.W, 0.0, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 16(2.242078e-44), 2(2.802597e-45) +; +; CM-LABEL: i16_mad24: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 0, @12, KC0[], KC1[] +; CM-NEXT: TEX 2 @6 +; CM-NEXT: ALU 8, @13, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: Fetch clause starting at 6: +; CM-NEXT: VTX_READ_16 T1.X, T0.X, 40, #3 +; CM-NEXT: VTX_READ_16 T2.X, T0.X, 42, #3 +; CM-NEXT: VTX_READ_16 T0.X, T0.X, 44, #3 +; CM-NEXT: ALU clause starting at 12: +; CM-NEXT: MOV * T0.X, 0.0, +; CM-NEXT: ALU clause starting at 13: +; CM-NEXT: MULLO_INT T0.X (MASKED), T1.X, T2.X, +; CM-NEXT: MULLO_INT T0.Y, T1.X, T2.X, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T1.X, T2.X, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T1.X, T2.X, +; CM-NEXT: ADD_INT * T0.W, PV.Y, T0.X, +; CM-NEXT: BFE_INT * T0.X, PV.W, 0.0, literal.x, +; CM-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: i16_mad24: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 +; GCN-NEXT: s_load_dword s4, s[4:5], 0xb +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_lshr_b32 s2, s2, 16 +; GCN-NEXT: s_mul_i32 s2, s4, s2 +; GCN-NEXT: s_add_i32 s2, s2, s3 +; GCN-NEXT: s_sext_i32_i16 s2, s2 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s4, s0 +; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: v_mov_b32_e32 v0, s2 +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; GFX8-LABEL: i16_mad24: +; GFX8: ; %bb.0: ; %entry +; GFX8-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 +; GFX8-NEXT: s_load_dword s8, s[4:5], 0x2c +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_mov_b32 s4, s0 +; GFX8-NEXT: s_lshr_b32 s0, s2, 16 +; GFX8-NEXT: s_mul_i32 s0, s8, s0 +; GFX8-NEXT: s_add_i32 s0, s0, s3 +; GFX8-NEXT: s_sext_i32_i16 s0, s0 +; GFX8-NEXT: s_mov_b32 s5, s1 +; GFX8-NEXT: v_mov_b32_e32 v0, s0 +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm entry: %0 = mul i16 %a, %b %1 = add i16 %0, %c @@ -46,17 +172,85 @@ entry: } ; FIXME: Need to handle non-uniform case for function below (load without gep). -; FUNC-LABEL: {{^}}i8_mad24: -; EG: MULLO_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; EG: ADD_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] ; The result must be sign-extended -; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x -; EG: 8 -; GCN: s_mul_i32 [[MUL:s[0-9]]], {{[s][0-9], [s][0-9]}} -; GCN: s_add_i32 [[MAD:s[0-9]]], [[MUL]], s{{[0-9]}} -; GCN: s_sext_i32_i8 [[EXT:s[0-9]]], [[MAD]] -; GCN: v_mov_b32_e32 v0, [[EXT]] define amdgpu_kernel void @i8_mad24(ptr addrspace(1) %out, i8 %a, i8 %b, i8 %c) { +; EG-LABEL: i8_mad24: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 0, @12, KC0[], KC1[] +; EG-NEXT: TEX 2 @6 +; EG-NEXT: ALU 4, @13, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 6: +; EG-NEXT: VTX_READ_8 T1.X, T0.X, 40, #3 +; EG-NEXT: VTX_READ_8 T2.X, T0.X, 41, #3 +; EG-NEXT: VTX_READ_8 T0.X, T0.X, 42, #3 +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: MOV * T0.X, 0.0, +; EG-NEXT: ALU clause starting at 13: +; EG-NEXT: MULLO_INT * T0.Y, T1.X, T2.X, +; EG-NEXT: ADD_INT * T0.W, PS, T0.X, +; EG-NEXT: BFE_INT T0.X, PV.W, 0.0, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 8(1.121039e-44), 2(2.802597e-45) +; +; CM-LABEL: i8_mad24: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 0, @12, KC0[], KC1[] +; CM-NEXT: TEX 2 @6 +; CM-NEXT: ALU 8, @13, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: Fetch clause starting at 6: +; CM-NEXT: VTX_READ_8 T1.X, T0.X, 40, #3 +; CM-NEXT: VTX_READ_8 T2.X, T0.X, 41, #3 +; CM-NEXT: VTX_READ_8 T0.X, T0.X, 42, #3 +; CM-NEXT: ALU clause starting at 12: +; CM-NEXT: MOV * T0.X, 0.0, +; CM-NEXT: ALU clause starting at 13: +; CM-NEXT: MULLO_INT T0.X (MASKED), T1.X, T2.X, +; CM-NEXT: MULLO_INT T0.Y, T1.X, T2.X, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T1.X, T2.X, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T1.X, T2.X, +; CM-NEXT: ADD_INT * T0.W, PV.Y, T0.X, +; CM-NEXT: BFE_INT * T0.X, PV.W, 0.0, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: i8_mad24: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_load_dword s2, s[4:5], 0xb +; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_lshr_b32 s4, s2, 8 +; GCN-NEXT: s_lshr_b32 s5, s2, 16 +; GCN-NEXT: s_mul_i32 s2, s2, s4 +; GCN-NEXT: s_add_i32 s2, s2, s5 +; GCN-NEXT: s_sext_i32_i8 s4, s2 +; GCN-NEXT: s_mov_b32 s2, -1 +; GCN-NEXT: v_mov_b32_e32 v0, s4 +; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GCN-NEXT: s_endpgm +; +; GFX8-LABEL: i8_mad24: +; GFX8: ; %bb.0: ; %entry +; GFX8-NEXT: s_load_dword s6, s[4:5], 0x2c +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_lshr_b32 s4, s6, 8 +; GFX8-NEXT: s_lshr_b32 s5, s6, 16 +; GFX8-NEXT: s_mul_i32 s4, s6, s4 +; GFX8-NEXT: s_add_i32 s4, s4, s5 +; GFX8-NEXT: s_sext_i32_i8 s4, s4 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX8-NEXT: s_endpgm entry: %0 = mul i8 %a, %b %1 = add i8 %0, %c @@ -72,11 +266,75 @@ entry: ; 24-bit mad pattern wasn't being matched. ; Check that the select instruction is not deleted. -; FUNC-LABEL: {{^}}i24_i32_i32_mad: -; EG: CNDE_INT -; SI: s_cselect -; GCN2: s_cselect define amdgpu_kernel void @i24_i32_i32_mad(ptr addrspace(1) %out, i32 %a, i32 %b, i32 %c, i32 %d) { +; EG-LABEL: i24_i32_i32_mad: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 7, @4, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: ALU clause starting at 4: +; EG-NEXT: ASHR * T0.W, KC0[2].Z, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: CNDE_INT * T0.W, KC0[3].X, literal.x, PV.W, +; EG-NEXT: 34(4.764415e-44), 0(0.000000e+00) +; EG-NEXT: MULLO_INT * T0.X, PV.W, KC0[3].X, +; EG-NEXT: ADD_INT T0.X, PS, KC0[3].Y, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: i24_i32_i32_mad: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 10, @4, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: ALU clause starting at 4: +; CM-NEXT: ASHR * T0.W, KC0[2].Z, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: CNDE_INT * T0.W, KC0[3].X, literal.x, PV.W, +; CM-NEXT: 34(4.764415e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, KC0[3].X, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, KC0[3].X, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, KC0[3].X, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, KC0[3].X, +; CM-NEXT: ADD_INT * T0.X, PV.X, KC0[3].Y, +; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: i24_i32_i32_mad: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_load_dword s2, s[4:5], 0xb +; GCN-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0xd +; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_ashr_i32 s2, s2, 8 +; GCN-NEXT: s_cmp_lg_u32 s6, 0 +; GCN-NEXT: s_cselect_b32 s2, s2, 34 +; GCN-NEXT: s_mul_i32 s2, s2, s6 +; GCN-NEXT: s_add_i32 s4, s2, s7 +; GCN-NEXT: s_mov_b32 s2, -1 +; GCN-NEXT: v_mov_b32_e32 v0, s4 +; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GCN-NEXT: s_endpgm +; +; GFX8-LABEL: i24_i32_i32_mad: +; GFX8: ; %bb.0: ; %entry +; GFX8-NEXT: s_load_dword s8, s[4:5], 0x2c +; GFX8-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_ashr_i32 s4, s8, 8 +; GFX8-NEXT: s_cmp_lg_u32 s6, 0 +; GFX8-NEXT: s_cselect_b32 s4, s4, 34 +; GFX8-NEXT: s_mul_i32 s4, s4, s6 +; GFX8-NEXT: s_add_i32 s4, s4, s7 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; GFX8-NEXT: s_endpgm entry: %0 = ashr i32 %a, 8 %1 = icmp ne i32 %c, 0 @@ -87,13 +345,139 @@ entry: ret void } -; FUNC-LABEL: {{^}}extra_and: -; SI-NOT: v_and -; SI: s_mul_i32 -; SI: s_mul_i32 -; SI: s_add_i32 -; SI: s_add_i32 define amdgpu_kernel void @extra_and(ptr addrspace(1) %arg, i32 %arg2, i32 %arg3) { +; EG-LABEL: extra_and: +; EG: ; %bb.0: ; %bb +; EG-NEXT: ALU 5, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: LOOP_START_DX10 @7 +; EG-NEXT: ALU_PUSH_BEFORE 12, @16, KC0[], KC1[] +; EG-NEXT: JUMP @6 POP:1 +; EG-NEXT: LOOP_BREAK @6 +; EG-NEXT: POP @6 POP:1 +; EG-NEXT: END_LOOP @2 +; EG-NEXT: ALU 1, @29, KC0[], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: ALU clause starting at 10: +; EG-NEXT: MOV * T1.W, literal.x, +; EG-NEXT: 0(0.000000e+00), 0(0.000000e+00) +; EG-NEXT: MOV * T3.W, PV.W, +; EG-NEXT: MOV T0.Z, KC0[2].Y, +; EG-NEXT: MOV T0.W, KC0[2].Z, +; EG-NEXT: MOV * T2.W, KC0[2].W, +; EG-NEXT: ALU clause starting at 16: +; EG-NEXT: AND_INT T1.W, T1.W, literal.x, +; EG-NEXT: AND_INT * T4.W, T3.W, literal.x, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: AND_INT T3.W, T3.W, literal.x, +; EG-NEXT: MULLO_INT * T0.X, PS, PV.W, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: MULLO_INT * T0.Y, PV.W, T1.W, +; EG-NEXT: ADD_INT T3.W, T2.W, PS, +; EG-NEXT: ADD_INT * T1.W, T0.W, T0.X, +; EG-NEXT: ADD_INT * T0.X, PS, PV.W, +; EG-NEXT: SETNE_INT * T4.W, PV.X, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: PRED_SETE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; EG-NEXT: ALU clause starting at 29: +; EG-NEXT: LSHR * T1.X, T0.Z, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: extra_and: +; CM: ; %bb.0: ; %bb +; CM-NEXT: ALU 5, @10, KC0[CB0:0-32], KC1[] +; CM-NEXT: LOOP_START_DX10 @7 +; CM-NEXT: ALU_PUSH_BEFORE 17, @16, KC0[], KC1[] +; CM-NEXT: JUMP @6 POP:1 +; CM-NEXT: LOOP_BREAK @6 +; CM-NEXT: POP @6 POP:1 +; CM-NEXT: END_LOOP @2 +; CM-NEXT: ALU 1, @34, KC0[], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: ALU clause starting at 10: +; CM-NEXT: MOV * T0.W, literal.x, +; CM-NEXT: 0(0.000000e+00), 0(0.000000e+00) +; CM-NEXT: MOV * T1.Z, PV.W, +; CM-NEXT: MOV T0.Y, KC0[2].Y, +; CM-NEXT: MOV T0.Z, KC0[2].Z, +; CM-NEXT: MOV * T1.W, KC0[2].W, +; CM-NEXT: ALU clause starting at 16: +; CM-NEXT: AND_INT T1.Y, T1.Z, literal.x, +; CM-NEXT: AND_INT T2.Z, T0.W, literal.x, +; CM-NEXT: AND_INT * T0.W, T1.Z, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, T2.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T2.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T2.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T2.Z, +; CM-NEXT: MULLO_INT T0.X (MASKED), T1.Y, T2.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T1.Y, T2.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T1.Y, T2.Z, +; CM-NEXT: MULLO_INT * T0.W, T1.Y, T2.Z, +; CM-NEXT: ADD_INT T1.Z, T1.W, PV.W, +; CM-NEXT: ADD_INT * T0.W, T0.Z, T0.X, +; CM-NEXT: ADD_INT * T0.X, PV.W, PV.Z, +; CM-NEXT: SETNE_INT * T2.W, PV.X, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: PRED_SETE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; CM-NEXT: ALU clause starting at 34: +; CM-NEXT: LSHR * T1.X, T0.Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: extra_and: +; GCN: ; %bb.0: ; %bb +; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xb +; GCN-NEXT: s_mov_b32 s2, 0 +; GCN-NEXT: s_mov_b32 s6, 0 +; GCN-NEXT: .LBB4_1: ; %bb4 +; GCN-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-NEXT: s_and_b32 s3, s6, 0xffffff +; GCN-NEXT: s_and_b32 s6, s6, 0xffffff +; GCN-NEXT: s_and_b32 s2, s2, 0xffffff +; GCN-NEXT: s_mul_i32 s3, s3, s2 +; GCN-NEXT: s_mul_i32 s6, s6, s2 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_add_i32 s2, s0, s3 +; GCN-NEXT: s_add_i32 s6, s1, s6 +; GCN-NEXT: s_add_i32 s3, s2, s6 +; GCN-NEXT: s_cmp_lg_u32 s3, 8 +; GCN-NEXT: s_cbranch_scc1 .LBB4_1 +; GCN-NEXT: ; %bb.2: ; %bb18 +; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: v_mov_b32_e32 v0, s3 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; GFX8-LABEL: extra_and: +; GFX8: ; %bb.0: ; %bb +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x2c +; GFX8-NEXT: s_mov_b32 s2, 0 +; GFX8-NEXT: s_mov_b32 s6, 0 +; GFX8-NEXT: .LBB4_1: ; %bb4 +; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 +; GFX8-NEXT: s_and_b32 s3, s6, 0xffffff +; GFX8-NEXT: s_and_b32 s6, s6, 0xffffff +; GFX8-NEXT: s_and_b32 s2, s2, 0xffffff +; GFX8-NEXT: s_mul_i32 s3, s3, s2 +; GFX8-NEXT: s_mul_i32 s6, s6, s2 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_add_i32 s2, s0, s3 +; GFX8-NEXT: s_add_i32 s6, s1, s6 +; GFX8-NEXT: s_add_i32 s3, s2, s6 +; GFX8-NEXT: s_cmp_lg_u32 s3, 8 +; GFX8-NEXT: s_cbranch_scc1 .LBB4_1 +; GFX8-NEXT: ; %bb.2: ; %bb18 +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: v_mov_b32_e32 v0, s3 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm bb: br label %bb4 @@ -119,13 +503,139 @@ bb18: ; preds = %bb4 ret void } -; FUNC-LABEL: {{^}}dont_remove_shift -; SI: s_lshr -; SI: s_mul_i32 -; SI: s_mul_i32 -; SI: s_add_i32 -; SI: s_add_i32 define amdgpu_kernel void @dont_remove_shift(ptr addrspace(1) %arg, i32 %arg2, i32 %arg3) { +; EG-LABEL: dont_remove_shift: +; EG: ; %bb.0: ; %bb +; EG-NEXT: ALU 5, @10, KC0[CB0:0-32], KC1[] +; EG-NEXT: LOOP_START_DX10 @7 +; EG-NEXT: ALU_PUSH_BEFORE 12, @16, KC0[], KC1[] +; EG-NEXT: JUMP @6 POP:1 +; EG-NEXT: LOOP_BREAK @6 +; EG-NEXT: POP @6 POP:1 +; EG-NEXT: END_LOOP @2 +; EG-NEXT: ALU 1, @29, KC0[], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: ALU clause starting at 10: +; EG-NEXT: MOV * T1.W, literal.x, +; EG-NEXT: 0(0.000000e+00), 0(0.000000e+00) +; EG-NEXT: MOV * T3.W, PV.W, +; EG-NEXT: MOV T0.Z, KC0[2].Y, +; EG-NEXT: MOV T0.W, KC0[2].Z, +; EG-NEXT: MOV * T2.W, KC0[2].W, +; EG-NEXT: ALU clause starting at 16: +; EG-NEXT: LSHR T1.W, T1.W, literal.x, +; EG-NEXT: LSHR * T4.W, T3.W, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: LSHR T3.W, T3.W, literal.x, +; EG-NEXT: MULLO_INT * T0.X, PS, PV.W, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: MULLO_INT * T0.Y, PV.W, T1.W, +; EG-NEXT: ADD_INT T3.W, T2.W, PS, +; EG-NEXT: ADD_INT * T1.W, T0.W, T0.X, +; EG-NEXT: ADD_INT * T0.X, PS, PV.W, +; EG-NEXT: SETNE_INT * T4.W, PV.X, literal.x, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: PRED_SETE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; EG-NEXT: ALU clause starting at 29: +; EG-NEXT: LSHR * T1.X, T0.Z, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: dont_remove_shift: +; CM: ; %bb.0: ; %bb +; CM-NEXT: ALU 5, @10, KC0[CB0:0-32], KC1[] +; CM-NEXT: LOOP_START_DX10 @7 +; CM-NEXT: ALU_PUSH_BEFORE 17, @16, KC0[], KC1[] +; CM-NEXT: JUMP @6 POP:1 +; CM-NEXT: LOOP_BREAK @6 +; CM-NEXT: POP @6 POP:1 +; CM-NEXT: END_LOOP @2 +; CM-NEXT: ALU 1, @34, KC0[], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: ALU clause starting at 10: +; CM-NEXT: MOV * T0.W, literal.x, +; CM-NEXT: 0(0.000000e+00), 0(0.000000e+00) +; CM-NEXT: MOV * T1.Z, PV.W, +; CM-NEXT: MOV T0.Y, KC0[2].Y, +; CM-NEXT: MOV T0.Z, KC0[2].Z, +; CM-NEXT: MOV * T1.W, KC0[2].W, +; CM-NEXT: ALU clause starting at 16: +; CM-NEXT: LSHR T1.Y, T1.Z, literal.x, +; CM-NEXT: LSHR T2.Z, T0.W, literal.x, +; CM-NEXT: LSHR * T0.W, T1.Z, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, T2.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T2.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T2.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T2.Z, +; CM-NEXT: MULLO_INT T0.X (MASKED), T1.Y, T2.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T1.Y, T2.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T1.Y, T2.Z, +; CM-NEXT: MULLO_INT * T0.W, T1.Y, T2.Z, +; CM-NEXT: ADD_INT T1.Z, T1.W, PV.W, +; CM-NEXT: ADD_INT * T0.W, T0.Z, T0.X, +; CM-NEXT: ADD_INT * T0.X, PV.W, PV.Z, +; CM-NEXT: SETNE_INT * T2.W, PV.X, literal.x, +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: PRED_SETE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; CM-NEXT: ALU clause starting at 34: +; CM-NEXT: LSHR * T1.X, T0.Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: dont_remove_shift: +; GCN: ; %bb.0: ; %bb +; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0xb +; GCN-NEXT: s_mov_b32 s2, 0 +; GCN-NEXT: s_mov_b32 s6, 0 +; GCN-NEXT: .LBB5_1: ; %bb4 +; GCN-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-NEXT: s_lshr_b32 s3, s6, 8 +; GCN-NEXT: s_lshr_b32 s6, s6, 8 +; GCN-NEXT: s_lshr_b32 s2, s2, 8 +; GCN-NEXT: s_mul_i32 s3, s3, s2 +; GCN-NEXT: s_mul_i32 s6, s6, s2 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_add_i32 s2, s0, s3 +; GCN-NEXT: s_add_i32 s6, s1, s6 +; GCN-NEXT: s_add_i32 s3, s2, s6 +; GCN-NEXT: s_cmp_lg_u32 s3, 8 +; GCN-NEXT: s_cbranch_scc1 .LBB5_1 +; GCN-NEXT: ; %bb.2: ; %bb18 +; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x9 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: v_mov_b32_e32 v0, s3 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_endpgm +; +; GFX8-LABEL: dont_remove_shift: +; GFX8: ; %bb.0: ; %bb +; GFX8-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x2c +; GFX8-NEXT: s_mov_b32 s2, 0 +; GFX8-NEXT: s_mov_b32 s6, 0 +; GFX8-NEXT: .LBB5_1: ; %bb4 +; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 +; GFX8-NEXT: s_lshr_b32 s3, s6, 8 +; GFX8-NEXT: s_lshr_b32 s6, s6, 8 +; GFX8-NEXT: s_lshr_b32 s2, s2, 8 +; GFX8-NEXT: s_mul_i32 s3, s3, s2 +; GFX8-NEXT: s_mul_i32 s6, s6, s2 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: s_add_i32 s2, s0, s3 +; GFX8-NEXT: s_add_i32 s6, s1, s6 +; GFX8-NEXT: s_add_i32 s3, s2, s6 +; GFX8-NEXT: s_cmp_lg_u32 s3, 8 +; GFX8-NEXT: s_cbranch_scc1 .LBB5_1 +; GFX8-NEXT: ; %bb.2: ; %bb18 +; GFX8-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x24 +; GFX8-NEXT: s_mov_b32 s7, 0xf000 +; GFX8-NEXT: s_mov_b32 s6, -1 +; GFX8-NEXT: v_mov_b32_e32 v0, s3 +; GFX8-NEXT: s_waitcnt lgkmcnt(0) +; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: s_endpgm bb: br label %bb4 @@ -151,19 +661,234 @@ bb18: ; preds = %bb4 ret void } -; FUNC-LABEL: {{^}}i8_mad_sat_16: -; EG: MULLO_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; EG: ADD_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; The result must be sign-extended -; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x -; EG: 8 -; SI: v_mad_u32_u24 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}} -; SI: v_bfe_i32 [[EXT:v[0-9]]], [[MAD]], 0, 16 -; SI: v_med3_i32 v{{[0-9]}}, [[EXT]], -; VI: v_mad_u16 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}} -; VI: v_max_i16_e32 [[MAX:v[0-9]]], 0xff80, [[MAD]] -; VI: v_min_i16_e32 {{v[0-9]}}, 0x7f, [[MAX]] define amdgpu_kernel void @i8_mad_sat_16(ptr addrspace(1) %out, ptr addrspace(1) %in0, ptr addrspace(1) %in1, ptr addrspace(1) %in2, ptr addrspace(5) %idx) { +; EG-LABEL: i8_mad_sat_16: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 4, @14, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @8 +; EG-NEXT: ALU 1, @19, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @10 +; EG-NEXT: ALU 24, @21, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT MSKOR T0.XW, T1.X +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 8: +; EG-NEXT: VTX_READ_8 T1.X, T1.X, 0, #1 +; EG-NEXT: Fetch clause starting at 10: +; EG-NEXT: VTX_READ_8 T3.X, T3.X, 0, #1 +; EG-NEXT: VTX_READ_8 T2.X, T2.X, 0, #1 +; EG-NEXT: ALU clause starting at 14: +; EG-NEXT: LSHR * T0.W, KC0[3].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: MOVA_INT * AR.x (MASKED), PV.W, +; EG-NEXT: MOV * T0.X, T(0 + AR.x).X+, +; EG-NEXT: ADD_INT * T1.X, KC0[2].W, PV.X, +; EG-NEXT: ALU clause starting at 19: +; EG-NEXT: ADD_INT T2.X, KC0[2].Z, T0.X, +; EG-NEXT: ADD_INT * T3.X, KC0[3].X, T0.X, +; EG-NEXT: ALU clause starting at 21: +; EG-NEXT: BFE_INT T0.Z, T1.X, 0.0, literal.x, +; EG-NEXT: BFE_INT * T0.W, T2.X, 0.0, literal.x, BS:VEC_120/SCL_212 +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: BFE_INT T1.W, T3.X, 0.0, literal.x, +; EG-NEXT: MULLO_INT * T0.Y, PV.Z, PV.W, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: ADD_INT * T0.W, PS, PV.W, +; EG-NEXT: BFE_INT * T0.W, PV.W, 0.0, literal.x, +; EG-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; EG-NEXT: MAX_INT T0.W, PV.W, literal.x, +; EG-NEXT: ADD_INT * T1.W, KC0[2].Y, T0.X, +; EG-NEXT: -128(nan), 0(0.000000e+00) +; EG-NEXT: AND_INT T2.W, PS, literal.x, +; EG-NEXT: MIN_INT * T0.W, PV.W, literal.y, +; EG-NEXT: 3(4.203895e-45), 127(1.779649e-43) +; EG-NEXT: AND_INT T0.W, PS, literal.x, +; EG-NEXT: LSHL * T2.W, PV.W, literal.y, +; EG-NEXT: 255(3.573311e-43), 3(4.203895e-45) +; EG-NEXT: LSHL T0.X, PV.W, PS, +; EG-NEXT: LSHL * T0.W, literal.x, PS, +; EG-NEXT: 255(3.573311e-43), 0(0.000000e+00) +; EG-NEXT: MOV T0.Y, 0.0, +; EG-NEXT: MOV * T0.Z, 0.0, +; EG-NEXT: LSHR * T1.X, T1.W, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; CM-LABEL: i8_mad_sat_16: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 4, @14, KC0[CB0:0-32], KC1[] +; CM-NEXT: TEX 0 @8 +; CM-NEXT: ALU 1, @19, KC0[CB0:0-32], KC1[] +; CM-NEXT: TEX 1 @10 +; CM-NEXT: ALU 26, @21, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT MSKOR T1.XW, T0.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: Fetch clause starting at 8: +; CM-NEXT: VTX_READ_8 T1.X, T1.X, 0, #1 +; CM-NEXT: Fetch clause starting at 10: +; CM-NEXT: VTX_READ_8 T3.X, T3.X, 0, #1 +; CM-NEXT: VTX_READ_8 T2.X, T2.X, 0, #1 +; CM-NEXT: ALU clause starting at 14: +; CM-NEXT: LSHR * T0.W, KC0[3].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; CM-NEXT: MOVA_INT * AR.x (MASKED), PV.W, +; CM-NEXT: MOV * T0.X, T(0 + AR.x).X+, +; CM-NEXT: ADD_INT * T1.X, KC0[3].X, PV.X, +; CM-NEXT: ALU clause starting at 19: +; CM-NEXT: ADD_INT * T2.X, KC0[2].W, T0.X, +; CM-NEXT: ADD_INT * T3.X, KC0[2].Z, T0.X, +; CM-NEXT: ALU clause starting at 21: +; CM-NEXT: BFE_INT T0.Y, T1.X, 0.0, literal.x, +; CM-NEXT: BFE_INT T0.Z, T2.X, 0.0, literal.x, BS:VEC_120/SCL_212 +; CM-NEXT: BFE_INT * T0.W, T3.X, 0.0, literal.x, BS:VEC_201 +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X (MASKED), T0.Z, T0.W, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.Z, T0.W, +; CM-NEXT: MULLO_INT T0.Z, T0.Z, T0.W, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.Z, T0.W, +; CM-NEXT: ADD_INT * T0.W, PV.Z, T0.Y, +; CM-NEXT: BFE_INT * T0.W, PV.W, 0.0, literal.x, +; CM-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; CM-NEXT: MAX_INT T0.Z, PV.W, literal.x, +; CM-NEXT: ADD_INT * T0.W, KC0[2].Y, T0.X, +; CM-NEXT: -128(nan), 0(0.000000e+00) +; CM-NEXT: AND_INT T1.Z, PV.W, literal.x, +; CM-NEXT: MIN_INT * T1.W, PV.Z, literal.y, +; CM-NEXT: 3(4.203895e-45), 127(1.779649e-43) +; CM-NEXT: AND_INT T0.Z, PV.W, literal.x, +; CM-NEXT: LSHL * T1.W, PV.Z, literal.y, +; CM-NEXT: 255(3.573311e-43), 3(4.203895e-45) +; CM-NEXT: LSHL T1.X, PV.Z, PV.W, +; CM-NEXT: LSHL * T1.W, literal.x, PV.W, +; CM-NEXT: 255(3.573311e-43), 0(0.000000e+00) +; CM-NEXT: MOV T1.Y, 0.0, +; CM-NEXT: MOV * T1.Z, 0.0, +; CM-NEXT: LSHR * T0.X, T0.W, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: i8_mad_sat_16: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_mov_b32 s20, SCRATCH_RSRC_DWORD0 +; GCN-NEXT: s_mov_b32 s21, SCRATCH_RSRC_DWORD1 +; GCN-NEXT: s_mov_b32 s22, -1 +; GCN-NEXT: s_mov_b32 s23, 0xe8f000 +; GCN-NEXT: s_add_u32 s20, s20, s11 +; GCN-NEXT: s_addc_u32 s21, s21, 0 +; GCN-NEXT: s_load_dword s8, s[4:5], 0x11 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_add_i32 s9, s8, 4 +; GCN-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 +; GCN-NEXT: v_mov_b32_e32 v0, s8 +; GCN-NEXT: v_mov_b32_e32 v1, s9 +; GCN-NEXT: buffer_load_dword v1, v1, s[20:23], 0 offen +; GCN-NEXT: buffer_load_dword v0, v0, s[20:23], 0 offen +; GCN-NEXT: s_mov_b32 s11, 0xf000 +; GCN-NEXT: s_mov_b32 s10, 0 +; GCN-NEXT: s_mov_b64 s[14:15], s[10:11] +; GCN-NEXT: s_mov_b64 s[18:19], s[10:11] +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b64 s[8:9], s[2:3] +; GCN-NEXT: s_mov_b64 s[12:13], s[4:5] +; GCN-NEXT: s_mov_b64 s[16:17], s[6:7] +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: buffer_load_sbyte v2, v[0:1], s[12:15], 0 addr64 +; GCN-NEXT: buffer_load_sbyte v3, v[0:1], s[8:11], 0 addr64 +; GCN-NEXT: buffer_load_sbyte v4, v[0:1], s[16:19], 0 addr64 +; GCN-NEXT: s_movk_i32 s2, 0xff80 +; GCN-NEXT: s_waitcnt vmcnt(2) +; GCN-NEXT: v_and_b32_e32 v2, 0xffff, v2 +; GCN-NEXT: s_waitcnt vmcnt(1) +; GCN-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_mad_u32_u24 v2, v2, v3, v4 +; GCN-NEXT: v_bfe_i32 v2, v2, 0, 16 +; GCN-NEXT: v_mov_b32_e32 v3, 0x7f +; GCN-NEXT: v_med3_i32 v2, v2, s2, v3 +; GCN-NEXT: s_mov_b64 s[2:3], s[10:11] +; GCN-NEXT: buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 +; GCN-NEXT: s_endpgm +; +; SI-LABEL: i8_mad_sat_16: +; SI: ; %bb.0: ; %entry +; SI-NEXT: s_mov_b32 s88, SCRATCH_RSRC_DWORD0 +; SI-NEXT: s_load_dword s0, s[4:5], 0x44 +; SI-NEXT: s_mov_b32 s89, SCRATCH_RSRC_DWORD1 +; SI-NEXT: s_mov_b32 s90, -1 +; SI-NEXT: s_mov_b32 s91, 0xe80000 +; SI-NEXT: s_add_u32 s88, s88, s11 +; SI-NEXT: s_addc_u32 s89, s89, 0 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_add_i32 s1, s0, 4 +; SI-NEXT: v_mov_b32_e32 v0, s0 +; SI-NEXT: buffer_load_dword v6, v0, s[88:91], 0 offen +; SI-NEXT: v_mov_b32_e32 v0, s1 +; SI-NEXT: buffer_load_dword v7, v0, s[88:91], 0 offen +; SI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: v_mov_b32_e32 v1, s3 +; SI-NEXT: v_mov_b32_e32 v3, s5 +; SI-NEXT: v_mov_b32_e32 v5, s7 +; SI-NEXT: s_waitcnt vmcnt(1) +; SI-NEXT: v_add_u32_e32 v0, vcc, s2, v6 +; SI-NEXT: s_waitcnt vmcnt(0) +; SI-NEXT: v_addc_u32_e32 v1, vcc, v1, v7, vcc +; SI-NEXT: v_add_u32_e32 v2, vcc, s4, v6 +; SI-NEXT: v_addc_u32_e32 v3, vcc, v3, v7, vcc +; SI-NEXT: v_add_u32_e32 v4, vcc, s6, v6 +; SI-NEXT: v_addc_u32_e32 v5, vcc, v5, v7, vcc +; SI-NEXT: flat_load_sbyte v0, v[0:1] +; SI-NEXT: flat_load_sbyte v1, v[2:3] +; SI-NEXT: flat_load_sbyte v2, v[4:5] +; SI-NEXT: v_mov_b32_e32 v3, s1 +; SI-NEXT: s_waitcnt vmcnt(0) +; SI-NEXT: v_mad_u16 v0, v1, v0, v2 +; SI-NEXT: v_max_i16_e32 v0, 0xff80, v0 +; SI-NEXT: v_min_i16_e32 v2, 0x7f, v0 +; SI-NEXT: v_add_u32_e32 v0, vcc, s0, v6 +; SI-NEXT: v_addc_u32_e32 v1, vcc, v3, v7, vcc +; SI-NEXT: flat_store_byte v[0:1], v2 +; SI-NEXT: s_endpgm +; +; VI-LABEL: i8_mad_sat_16: +; VI: ; %bb.0: ; %entry +; VI-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0 +; VI-NEXT: s_load_dword s0, s[4:5], 0x44 +; VI-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1 +; VI-NEXT: s_mov_b32 s14, -1 +; VI-NEXT: s_mov_b32 s15, 0xe80000 +; VI-NEXT: s_add_u32 s12, s12, s11 +; VI-NEXT: s_addc_u32 s13, s13, 0 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: s_add_i32 s1, s0, 4 +; VI-NEXT: v_mov_b32_e32 v0, s0 +; VI-NEXT: buffer_load_dword v6, v0, s[12:15], 0 offen +; VI-NEXT: v_mov_b32_e32 v0, s1 +; VI-NEXT: buffer_load_dword v7, v0, s[12:15], 0 offen +; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v1, s3 +; VI-NEXT: v_mov_b32_e32 v3, s5 +; VI-NEXT: v_mov_b32_e32 v5, s7 +; VI-NEXT: s_waitcnt vmcnt(1) +; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v6 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: v_addc_u32_e32 v1, vcc, v1, v7, vcc +; VI-NEXT: v_add_u32_e32 v2, vcc, s4, v6 +; VI-NEXT: v_addc_u32_e32 v3, vcc, v3, v7, vcc +; VI-NEXT: v_add_u32_e32 v4, vcc, s6, v6 +; VI-NEXT: v_addc_u32_e32 v5, vcc, v5, v7, vcc +; VI-NEXT: flat_load_sbyte v0, v[0:1] +; VI-NEXT: flat_load_sbyte v1, v[2:3] +; VI-NEXT: flat_load_sbyte v2, v[4:5] +; VI-NEXT: v_mov_b32_e32 v3, s1 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: v_mad_u16 v0, v1, v0, v2 +; VI-NEXT: v_max_i16_e32 v0, 0xff80, v0 +; VI-NEXT: v_min_i16_e32 v2, 0x7f, v0 +; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v6 +; VI-NEXT: v_addc_u32_e32 v1, vcc, v3, v7, vcc +; VI-NEXT: flat_store_byte v[0:1], v2 +; VI-NEXT: s_endpgm entry: %retval.0.i = load i64, ptr addrspace(5) %idx %arrayidx = getelementptr inbounds i8, ptr addrspace(1) %in0, i64 %retval.0.i @@ -187,16 +912,201 @@ entry: ret void } -; FUNC-LABEL: {{^}}i8_mad_32: -; EG: MULLO_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; EG: ADD_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; The result must be sign-extended -; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x -; EG: 8 -; SI: v_mad_u32_u24 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}} -; VI: v_mad_u16 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}} -; GCN: v_bfe_i32 [[EXT:v[0-9]]], [[MAD]], 0, 16 define amdgpu_kernel void @i8_mad_32(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b, ptr addrspace(1) %c, ptr addrspace(5) %idx) { +; EG-LABEL: i8_mad_32: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 4, @14, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @8 +; EG-NEXT: ALU 1, @19, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @10 +; EG-NEXT: ALU 9, @21, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 8: +; EG-NEXT: VTX_READ_8 T1.X, T1.X, 0, #1 +; EG-NEXT: Fetch clause starting at 10: +; EG-NEXT: VTX_READ_8 T0.X, T0.X, 0, #1 +; EG-NEXT: VTX_READ_8 T2.X, T2.X, 0, #1 +; EG-NEXT: ALU clause starting at 14: +; EG-NEXT: LSHR * T0.W, KC0[3].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: MOVA_INT * AR.x (MASKED), PV.W, +; EG-NEXT: MOV * T0.X, T(0 + AR.x).X+, +; EG-NEXT: ADD_INT * T1.X, KC0[2].W, PV.X, +; EG-NEXT: ALU clause starting at 19: +; EG-NEXT: ADD_INT T2.X, KC0[2].Z, T0.X, +; EG-NEXT: ADD_INT * T0.X, KC0[3].X, T0.X, +; EG-NEXT: ALU clause starting at 21: +; EG-NEXT: BFE_INT T0.Z, T1.X, 0.0, literal.x, +; EG-NEXT: BFE_INT * T0.W, T2.X, 0.0, literal.x, BS:VEC_120/SCL_212 +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: BFE_INT T1.W, T0.X, 0.0, literal.x, +; EG-NEXT: MULLO_INT * T0.X, PV.W, PV.Z, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: ADD_INT * T0.W, PS, PV.W, +; EG-NEXT: BFE_INT T0.X, PV.W, 0.0, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 16(2.242078e-44), 2(2.802597e-45) +; +; CM-LABEL: i8_mad_32: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 4, @14, KC0[CB0:0-32], KC1[] +; CM-NEXT: TEX 0 @8 +; CM-NEXT: ALU 1, @19, KC0[CB0:0-32], KC1[] +; CM-NEXT: TEX 1 @10 +; CM-NEXT: ALU 12, @21, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: Fetch clause starting at 8: +; CM-NEXT: VTX_READ_8 T1.X, T1.X, 0, #1 +; CM-NEXT: Fetch clause starting at 10: +; CM-NEXT: VTX_READ_8 T0.X, T0.X, 0, #1 +; CM-NEXT: VTX_READ_8 T2.X, T2.X, 0, #1 +; CM-NEXT: ALU clause starting at 14: +; CM-NEXT: LSHR * T0.W, KC0[3].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; CM-NEXT: MOVA_INT * AR.x (MASKED), PV.W, +; CM-NEXT: MOV * T0.X, T(0 + AR.x).X+, +; CM-NEXT: ADD_INT * T1.X, KC0[3].X, PV.X, +; CM-NEXT: ALU clause starting at 19: +; CM-NEXT: ADD_INT * T2.X, KC0[2].W, T0.X, +; CM-NEXT: ADD_INT * T0.X, KC0[2].Z, T0.X, +; CM-NEXT: ALU clause starting at 21: +; CM-NEXT: BFE_INT T0.Y, T1.X, 0.0, literal.x, +; CM-NEXT: BFE_INT T0.Z, T2.X, 0.0, literal.x, BS:VEC_120/SCL_212 +; CM-NEXT: BFE_INT * T0.W, T0.X, 0.0, literal.x, BS:VEC_201 +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, T0.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T0.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T0.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T0.Z, +; CM-NEXT: ADD_INT * T0.W, PV.X, T0.Y, +; CM-NEXT: BFE_INT * T0.X, PV.W, 0.0, literal.x, +; CM-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; CM-NEXT: LSHR * T1.X, KC0[2].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; +; GCN-LABEL: i8_mad_32: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_mov_b32 s24, SCRATCH_RSRC_DWORD0 +; GCN-NEXT: s_mov_b32 s25, SCRATCH_RSRC_DWORD1 +; GCN-NEXT: s_mov_b32 s26, -1 +; GCN-NEXT: s_mov_b32 s27, 0xe8f000 +; GCN-NEXT: s_add_u32 s24, s24, s11 +; GCN-NEXT: s_addc_u32 s25, s25, 0 +; GCN-NEXT: s_load_dword s8, s[4:5], 0x11 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_add_i32 s9, s8, 4 +; GCN-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 +; GCN-NEXT: v_mov_b32_e32 v0, s8 +; GCN-NEXT: v_mov_b32_e32 v1, s9 +; GCN-NEXT: buffer_load_dword v1, v1, s[24:27], 0 offen +; GCN-NEXT: buffer_load_dword v0, v0, s[24:27], 0 offen +; GCN-NEXT: s_mov_b32 s11, 0xf000 +; GCN-NEXT: s_mov_b32 s14, 0 +; GCN-NEXT: s_mov_b32 s15, s11 +; GCN-NEXT: s_mov_b64 s[18:19], s[14:15] +; GCN-NEXT: s_mov_b64 s[22:23], s[14:15] +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b64 s[12:13], s[2:3] +; GCN-NEXT: s_mov_b64 s[16:17], s[4:5] +; GCN-NEXT: s_mov_b64 s[20:21], s[6:7] +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: buffer_load_sbyte v2, v[0:1], s[12:15], 0 addr64 +; GCN-NEXT: buffer_load_sbyte v3, v[0:1], s[16:19], 0 addr64 +; GCN-NEXT: buffer_load_sbyte v0, v[0:1], s[20:23], 0 addr64 +; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_mov_b32 s8, s0 +; GCN-NEXT: s_mov_b32 s9, s1 +; GCN-NEXT: s_waitcnt vmcnt(2) +; GCN-NEXT: v_and_b32_e32 v1, 0xffff, v2 +; GCN-NEXT: s_waitcnt vmcnt(1) +; GCN-NEXT: v_and_b32_e32 v2, 0xffff, v3 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_mad_u32_u24 v0, v1, v2, v0 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GCN-NEXT: buffer_store_dword v0, off, s[8:11], 0 +; GCN-NEXT: s_endpgm +; +; SI-LABEL: i8_mad_32: +; SI: ; %bb.0: ; %entry +; SI-NEXT: s_mov_b32 s88, SCRATCH_RSRC_DWORD0 +; SI-NEXT: s_load_dword s0, s[4:5], 0x44 +; SI-NEXT: s_mov_b32 s89, SCRATCH_RSRC_DWORD1 +; SI-NEXT: s_mov_b32 s90, -1 +; SI-NEXT: s_mov_b32 s91, 0xe80000 +; SI-NEXT: s_add_u32 s88, s88, s11 +; SI-NEXT: s_addc_u32 s89, s89, 0 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_add_i32 s1, s0, 4 +; SI-NEXT: v_mov_b32_e32 v0, s0 +; SI-NEXT: buffer_load_dword v4, v0, s[88:91], 0 offen +; SI-NEXT: v_mov_b32_e32 v0, s1 +; SI-NEXT: buffer_load_dword v5, v0, s[88:91], 0 offen +; SI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: v_mov_b32_e32 v1, s3 +; SI-NEXT: v_mov_b32_e32 v3, s5 +; SI-NEXT: v_mov_b32_e32 v6, s7 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_waitcnt vmcnt(1) +; SI-NEXT: v_add_u32_e32 v0, vcc, s2, v4 +; SI-NEXT: s_waitcnt vmcnt(0) +; SI-NEXT: v_addc_u32_e32 v1, vcc, v1, v5, vcc +; SI-NEXT: v_add_u32_e32 v2, vcc, s4, v4 +; SI-NEXT: v_addc_u32_e32 v3, vcc, v3, v5, vcc +; SI-NEXT: v_add_u32_e32 v4, vcc, s6, v4 +; SI-NEXT: v_addc_u32_e32 v5, vcc, v6, v5, vcc +; SI-NEXT: flat_load_sbyte v0, v[0:1] +; SI-NEXT: flat_load_sbyte v1, v[2:3] +; SI-NEXT: flat_load_sbyte v2, v[4:5] +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: s_waitcnt vmcnt(0) +; SI-NEXT: v_mad_u16 v0, v0, v1, v2 +; SI-NEXT: v_bfe_i32 v0, v0, 0, 16 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; SI-NEXT: s_endpgm +; +; VI-LABEL: i8_mad_32: +; VI: ; %bb.0: ; %entry +; VI-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0 +; VI-NEXT: s_load_dword s0, s[4:5], 0x44 +; VI-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1 +; VI-NEXT: s_mov_b32 s14, -1 +; VI-NEXT: s_mov_b32 s15, 0xe80000 +; VI-NEXT: s_add_u32 s12, s12, s11 +; VI-NEXT: s_addc_u32 s13, s13, 0 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: s_add_i32 s1, s0, 4 +; VI-NEXT: v_mov_b32_e32 v0, s0 +; VI-NEXT: buffer_load_dword v4, v0, s[12:15], 0 offen +; VI-NEXT: v_mov_b32_e32 v0, s1 +; VI-NEXT: buffer_load_dword v5, v0, s[12:15], 0 offen +; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v1, s3 +; VI-NEXT: v_mov_b32_e32 v3, s5 +; VI-NEXT: v_mov_b32_e32 v6, s7 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_waitcnt vmcnt(1) +; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v4 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: v_addc_u32_e32 v1, vcc, v1, v5, vcc +; VI-NEXT: v_add_u32_e32 v2, vcc, s4, v4 +; VI-NEXT: v_addc_u32_e32 v3, vcc, v3, v5, vcc +; VI-NEXT: v_add_u32_e32 v4, vcc, s6, v4 +; VI-NEXT: v_addc_u32_e32 v5, vcc, v6, v5, vcc +; VI-NEXT: flat_load_sbyte v0, v[0:1] +; VI-NEXT: flat_load_sbyte v1, v[2:3] +; VI-NEXT: flat_load_sbyte v2, v[4:5] +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: v_mad_u16 v0, v0, v1, v2 +; VI-NEXT: v_bfe_i32 v0, v0, 0, 16 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 +; VI-NEXT: s_endpgm entry: %retval.0.i = load i64, ptr addrspace(5) %idx %arrayidx = getelementptr inbounds i8, ptr addrspace(1) %a, i64 %retval.0.i @@ -215,16 +1125,207 @@ entry: ret void } -; FUNC-LABEL: {{^}}i8_mad_64: -; EG: MULLO_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; EG: ADD_INT {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]] -; The result must be sign-extended -; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x -; EG: 8 -; SI: v_mad_u32_u24 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}} -; VI: v_mad_u16 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}} -; GCN: v_bfe_i32 [[EXT:v[0-9]]], [[MAD]], 0, 16 define amdgpu_kernel void @i8_mad_64(ptr addrspace(1) %out, ptr addrspace(1) %a, ptr addrspace(1) %b, ptr addrspace(1) %c, ptr addrspace(5) %idx) { +; EG-LABEL: i8_mad_64: +; EG: ; %bb.0: ; %entry +; EG-NEXT: ALU 4, @14, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 0 @8 +; EG-NEXT: ALU 1, @19, KC0[CB0:0-32], KC1[] +; EG-NEXT: TEX 1 @10 +; EG-NEXT: ALU 11, @21, KC0[CB0:0-32], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1 +; EG-NEXT: CF_END +; EG-NEXT: PAD +; EG-NEXT: Fetch clause starting at 8: +; EG-NEXT: VTX_READ_8 T1.X, T1.X, 0, #1 +; EG-NEXT: Fetch clause starting at 10: +; EG-NEXT: VTX_READ_8 T0.X, T0.X, 0, #1 +; EG-NEXT: VTX_READ_8 T2.X, T2.X, 0, #1 +; EG-NEXT: ALU clause starting at 14: +; EG-NEXT: LSHR * T0.W, KC0[3].Y, literal.x, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: MOVA_INT * AR.x (MASKED), PV.W, +; EG-NEXT: MOV * T0.X, T(0 + AR.x).X+, +; EG-NEXT: ADD_INT * T1.X, KC0[2].W, PV.X, +; EG-NEXT: ALU clause starting at 19: +; EG-NEXT: ADD_INT T2.X, KC0[2].Z, T0.X, +; EG-NEXT: ADD_INT * T0.X, KC0[3].X, T0.X, +; EG-NEXT: ALU clause starting at 21: +; EG-NEXT: BFE_INT T0.Z, T1.X, 0.0, literal.x, +; EG-NEXT: BFE_INT * T0.W, T2.X, 0.0, literal.x, BS:VEC_120/SCL_212 +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: BFE_INT T1.W, T0.X, 0.0, literal.x, +; EG-NEXT: MULLO_INT * T0.X, PV.W, PV.Z, +; EG-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; EG-NEXT: ADD_INT * T0.W, PS, PV.W, +; EG-NEXT: BFE_INT T0.X, PV.W, 0.0, literal.x, +; EG-NEXT: LSHR * T1.X, KC0[2].Y, literal.y, +; EG-NEXT: 16(2.242078e-44), 2(2.802597e-45) +; EG-NEXT: ASHR * T0.Y, PV.X, literal.x, +; EG-NEXT: 31(4.344025e-44), 0(0.000000e+00) +; +; CM-LABEL: i8_mad_64: +; CM: ; %bb.0: ; %entry +; CM-NEXT: ALU 4, @14, KC0[CB0:0-32], KC1[] +; CM-NEXT: TEX 0 @8 +; CM-NEXT: ALU 1, @19, KC0[CB0:0-32], KC1[] +; CM-NEXT: TEX 1 @10 +; CM-NEXT: ALU 13, @21, KC0[CB0:0-32], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0, T1.X +; CM-NEXT: CF_END +; CM-NEXT: PAD +; CM-NEXT: Fetch clause starting at 8: +; CM-NEXT: VTX_READ_8 T1.X, T1.X, 0, #1 +; CM-NEXT: Fetch clause starting at 10: +; CM-NEXT: VTX_READ_8 T0.X, T0.X, 0, #1 +; CM-NEXT: VTX_READ_8 T2.X, T2.X, 0, #1 +; CM-NEXT: ALU clause starting at 14: +; CM-NEXT: LSHR * T0.W, KC0[3].Y, literal.x, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; CM-NEXT: MOVA_INT * AR.x (MASKED), PV.W, +; CM-NEXT: MOV * T0.X, T(0 + AR.x).X+, +; CM-NEXT: ADD_INT * T1.X, KC0[3].X, PV.X, +; CM-NEXT: ALU clause starting at 19: +; CM-NEXT: ADD_INT * T2.X, KC0[2].W, T0.X, +; CM-NEXT: ADD_INT * T0.X, KC0[2].Z, T0.X, +; CM-NEXT: ALU clause starting at 21: +; CM-NEXT: BFE_INT T0.Y, T1.X, 0.0, literal.x, +; CM-NEXT: BFE_INT T0.Z, T2.X, 0.0, literal.x, BS:VEC_120/SCL_212 +; CM-NEXT: BFE_INT * T0.W, T0.X, 0.0, literal.x, BS:VEC_201 +; CM-NEXT: 8(1.121039e-44), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T0.W, T0.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T0.Z, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T0.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T0.Z, +; CM-NEXT: ADD_INT * T0.W, PV.X, T0.Y, +; CM-NEXT: BFE_INT * T0.X, PV.W, 0.0, literal.x, +; CM-NEXT: 16(2.242078e-44), 0(0.000000e+00) +; CM-NEXT: LSHR T1.X, KC0[2].Y, literal.x, +; CM-NEXT: ASHR * T0.Y, PV.X, literal.y, +; CM-NEXT: 2(2.802597e-45), 31(4.344025e-44) +; +; GCN-LABEL: i8_mad_64: +; GCN: ; %bb.0: ; %entry +; GCN-NEXT: s_mov_b32 s24, SCRATCH_RSRC_DWORD0 +; GCN-NEXT: s_mov_b32 s25, SCRATCH_RSRC_DWORD1 +; GCN-NEXT: s_mov_b32 s26, -1 +; GCN-NEXT: s_mov_b32 s27, 0xe8f000 +; GCN-NEXT: s_add_u32 s24, s24, s11 +; GCN-NEXT: s_addc_u32 s25, s25, 0 +; GCN-NEXT: s_load_dword s8, s[4:5], 0x11 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_add_i32 s9, s8, 4 +; GCN-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x9 +; GCN-NEXT: v_mov_b32_e32 v0, s8 +; GCN-NEXT: v_mov_b32_e32 v1, s9 +; GCN-NEXT: buffer_load_dword v1, v1, s[24:27], 0 offen +; GCN-NEXT: buffer_load_dword v0, v0, s[24:27], 0 offen +; GCN-NEXT: s_mov_b32 s11, 0xf000 +; GCN-NEXT: s_mov_b32 s14, 0 +; GCN-NEXT: s_mov_b32 s15, s11 +; GCN-NEXT: s_mov_b64 s[18:19], s[14:15] +; GCN-NEXT: s_mov_b64 s[22:23], s[14:15] +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b64 s[12:13], s[2:3] +; GCN-NEXT: s_mov_b64 s[16:17], s[4:5] +; GCN-NEXT: s_mov_b64 s[20:21], s[6:7] +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: buffer_load_sbyte v2, v[0:1], s[12:15], 0 addr64 +; GCN-NEXT: buffer_load_sbyte v3, v[0:1], s[16:19], 0 addr64 +; GCN-NEXT: buffer_load_sbyte v0, v[0:1], s[20:23], 0 addr64 +; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_mov_b32 s8, s0 +; GCN-NEXT: s_mov_b32 s9, s1 +; GCN-NEXT: s_waitcnt vmcnt(2) +; GCN-NEXT: v_and_b32_e32 v1, 0xffff, v2 +; GCN-NEXT: s_waitcnt vmcnt(1) +; GCN-NEXT: v_and_b32_e32 v2, 0xffff, v3 +; GCN-NEXT: s_waitcnt vmcnt(0) +; GCN-NEXT: v_mad_u32_u24 v0, v1, v2, v0 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 16 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: s_endpgm +; +; SI-LABEL: i8_mad_64: +; SI: ; %bb.0: ; %entry +; SI-NEXT: s_mov_b32 s88, SCRATCH_RSRC_DWORD0 +; SI-NEXT: s_load_dword s0, s[4:5], 0x44 +; SI-NEXT: s_mov_b32 s89, SCRATCH_RSRC_DWORD1 +; SI-NEXT: s_mov_b32 s90, -1 +; SI-NEXT: s_mov_b32 s91, 0xe80000 +; SI-NEXT: s_add_u32 s88, s88, s11 +; SI-NEXT: s_addc_u32 s89, s89, 0 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_add_i32 s1, s0, 4 +; SI-NEXT: v_mov_b32_e32 v0, s0 +; SI-NEXT: buffer_load_dword v4, v0, s[88:91], 0 offen +; SI-NEXT: v_mov_b32_e32 v0, s1 +; SI-NEXT: buffer_load_dword v5, v0, s[88:91], 0 offen +; SI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: v_mov_b32_e32 v1, s3 +; SI-NEXT: v_mov_b32_e32 v3, s5 +; SI-NEXT: v_mov_b32_e32 v6, s7 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_waitcnt vmcnt(1) +; SI-NEXT: v_add_u32_e32 v0, vcc, s2, v4 +; SI-NEXT: s_waitcnt vmcnt(0) +; SI-NEXT: v_addc_u32_e32 v1, vcc, v1, v5, vcc +; SI-NEXT: v_add_u32_e32 v2, vcc, s4, v4 +; SI-NEXT: v_addc_u32_e32 v3, vcc, v3, v5, vcc +; SI-NEXT: v_add_u32_e32 v4, vcc, s6, v4 +; SI-NEXT: v_addc_u32_e32 v5, vcc, v6, v5, vcc +; SI-NEXT: flat_load_sbyte v0, v[0:1] +; SI-NEXT: flat_load_sbyte v1, v[2:3] +; SI-NEXT: flat_load_sbyte v2, v[4:5] +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: s_waitcnt vmcnt(0) +; SI-NEXT: v_mad_u16 v0, v0, v1, v2 +; SI-NEXT: v_bfe_i32 v0, v0, 0, 16 +; SI-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; SI-NEXT: s_endpgm +; +; VI-LABEL: i8_mad_64: +; VI: ; %bb.0: ; %entry +; VI-NEXT: s_mov_b32 s12, SCRATCH_RSRC_DWORD0 +; VI-NEXT: s_load_dword s0, s[4:5], 0x44 +; VI-NEXT: s_mov_b32 s13, SCRATCH_RSRC_DWORD1 +; VI-NEXT: s_mov_b32 s14, -1 +; VI-NEXT: s_mov_b32 s15, 0xe80000 +; VI-NEXT: s_add_u32 s12, s12, s11 +; VI-NEXT: s_addc_u32 s13, s13, 0 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: s_add_i32 s1, s0, 4 +; VI-NEXT: v_mov_b32_e32 v0, s0 +; VI-NEXT: buffer_load_dword v4, v0, s[12:15], 0 offen +; VI-NEXT: v_mov_b32_e32 v0, s1 +; VI-NEXT: buffer_load_dword v5, v0, s[12:15], 0 offen +; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 +; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v1, s3 +; VI-NEXT: v_mov_b32_e32 v3, s5 +; VI-NEXT: v_mov_b32_e32 v6, s7 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_waitcnt vmcnt(1) +; VI-NEXT: v_add_u32_e32 v0, vcc, s2, v4 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: v_addc_u32_e32 v1, vcc, v1, v5, vcc +; VI-NEXT: v_add_u32_e32 v2, vcc, s4, v4 +; VI-NEXT: v_addc_u32_e32 v3, vcc, v3, v5, vcc +; VI-NEXT: v_add_u32_e32 v4, vcc, s6, v4 +; VI-NEXT: v_addc_u32_e32 v5, vcc, v6, v5, vcc +; VI-NEXT: flat_load_sbyte v0, v[0:1] +; VI-NEXT: flat_load_sbyte v1, v[2:3] +; VI-NEXT: flat_load_sbyte v2, v[4:5] +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: s_waitcnt vmcnt(0) +; VI-NEXT: v_mad_u16 v0, v0, v1, v2 +; VI-NEXT: v_bfe_i32 v0, v0, 0, 16 +; VI-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; VI-NEXT: s_endpgm entry: %retval.0.i = load i64, ptr addrspace(5) %idx %arrayidx = getelementptr inbounds i8, ptr addrspace(1) %a, i64 %retval.0.i @@ -248,17 +1349,236 @@ entry: ; had a chance to form mul24. The mul combine would then see ; extractelement with no known bits and fail. All of the mul/add ; combos in this loop should form v_mad_u32_u24. - -; FUNC-LABEL: {{^}}mad24_known_bits_destroyed: -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 -; GCN: v_mad_u32_u24 define void @mad24_known_bits_destroyed(i32 %arg, <4 x i32> %arg1, <4 x i32> %arg2, <4 x i32> %arg3, i32 %arg4, i32 %arg5, i32 %arg6, ptr addrspace(1) %arg7, ptr addrspace(1) %arg8) #0 { +; EG-LABEL: mad24_known_bits_destroyed: +; EG: ; %bb.0: ; %bb +; EG-NEXT: ALU 21, @12, KC0[CB0:0-32], KC1[] +; EG-NEXT: LOOP_START_DX10 @11 +; EG-NEXT: ALU 8, @34, KC0[], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.X, T2.X, 0 +; EG-NEXT: ALU 14, @43, KC0[], KC1[] +; EG-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 0 +; EG-NEXT: ALU_PUSH_BEFORE 3, @58, KC0[], KC1[] +; EG-NEXT: JUMP @10 POP:1 +; EG-NEXT: LOOP_BREAK @10 +; EG-NEXT: POP @10 POP:1 +; EG-NEXT: END_LOOP @2 +; EG-NEXT: CF_END +; EG-NEXT: ALU clause starting at 12: +; EG-NEXT: MOV * T0.W, KC0[5].X, +; EG-NEXT: MOV * T0.Z, KC0[4].W, +; EG-NEXT: MOV * T0.Y, KC0[4].Z, +; EG-NEXT: MOV T0.X, KC0[2].Y, +; EG-NEXT: AND_INT * T1.Y, KC0[4].X, literal.x, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: AND_INT T1.Z, KC0[3].W, literal.x, +; EG-NEXT: AND_INT T1.W, KC0[3].Z, literal.x, +; EG-NEXT: MOV * T2.W, KC0[7].Y, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: LSHR T1.X, PS, literal.x, +; EG-NEXT: AND_INT T2.Y, KC0[6].Y, literal.y, +; EG-NEXT: MOV T2.Z, KC0[6].X, +; EG-NEXT: MOV * T2.W, KC0[5].W, +; EG-NEXT: 2(2.802597e-45), 16777215(2.350989e-38) +; EG-NEXT: MOV * T3.W, KC0[7].X, +; EG-NEXT: LSHR T2.X, PV.W, literal.x, +; EG-NEXT: MOV T3.Y, KC0[5].Z, +; EG-NEXT: MOV T3.Z, KC0[6].Z, +; EG-NEXT: MOV * T3.W, KC0[6].W, +; EG-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; EG-NEXT: MOV * T4.W, KC0[4].Y, +; EG-NEXT: ALU clause starting at 34: +; EG-NEXT: MULLO_INT * T0.X, T0.X, T2.Y, +; EG-NEXT: ADD_INT * T4.W, PS, T3.Z, +; EG-NEXT: AND_INT * T4.W, PV.W, literal.x, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: MULLO_INT * T0.X, PV.W, T2.Y, +; EG-NEXT: MULLO_INT * T0.W, T0.W, T1.Y, +; EG-NEXT: MULLO_INT * T0.Z, T0.Z, T1.Z, +; EG-NEXT: MULLO_INT * T0.Y, T0.Y, T1.W, +; EG-NEXT: ADD_INT * T0.X, T0.X, T3.Z, +; EG-NEXT: ALU clause starting at 43: +; EG-NEXT: ADD_INT * T4.W, T0.Y, T3.Y, +; EG-NEXT: AND_INT T4.W, PV.W, literal.x, +; EG-NEXT: ADD_INT * T5.W, T0.Z, T2.W, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: AND_INT T0.Z, PS, literal.x, +; EG-NEXT: ADD_INT T0.W, T0.W, T2.Z, +; EG-NEXT: MULLO_INT * T0.Y, PV.W, T1.W, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: ADD_INT T0.Y, PS, T3.Y, +; EG-NEXT: AND_INT T0.W, PV.W, literal.x, +; EG-NEXT: MULLO_INT * T0.Z, PV.Z, T1.Z, +; EG-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; EG-NEXT: ADD_INT T0.Z, PS, T2.W, +; EG-NEXT: MULLO_INT * T0.W, PV.W, T1.Y, +; EG-NEXT: ADD_INT * T0.W, PS, T2.Z, +; EG-NEXT: ALU clause starting at 58: +; EG-NEXT: ADD_INT * T3.W, T3.W, literal.x, +; EG-NEXT: -1(nan), 0(0.000000e+00) +; EG-NEXT: SETE_INT * T4.W, PV.W, 0.0, +; EG-NEXT: PRED_SETNE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; +; CM-LABEL: mad24_known_bits_destroyed: +; CM: ; %bb.0: ; %bb +; CM-NEXT: ALU 22, @12, KC0[CB0:0-32], KC1[] +; CM-NEXT: LOOP_START_DX10 @11 +; CM-NEXT: ALU 23, @35, KC0[], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0.X, T2.X +; CM-NEXT: ALU 23, @59, KC0[], KC1[] +; CM-NEXT: MEM_RAT_CACHELESS STORE_DWORD T0, T1.X +; CM-NEXT: ALU_PUSH_BEFORE 3, @83, KC0[], KC1[] +; CM-NEXT: JUMP @10 POP:1 +; CM-NEXT: LOOP_BREAK @10 +; CM-NEXT: POP @10 POP:1 +; CM-NEXT: END_LOOP @2 +; CM-NEXT: CF_END +; CM-NEXT: ALU clause starting at 12: +; CM-NEXT: MOV * T0.W, KC0[5].X, +; CM-NEXT: MOV * T0.Z, KC0[4].W, +; CM-NEXT: MOV * T0.Y, KC0[4].Z, +; CM-NEXT: MOV T0.X, KC0[2].Y, +; CM-NEXT: AND_INT * T1.Y, KC0[4].X, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: AND_INT T1.Z, KC0[3].W, literal.x, +; CM-NEXT: AND_INT * T1.W, KC0[3].Z, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: AND_INT T2.Y, KC0[6].Y, literal.x, +; CM-NEXT: MOV T2.Z, KC0[6].X, +; CM-NEXT: MOV * T2.W, KC0[7].Y, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: LSHR T1.X, PV.W, literal.x, +; CM-NEXT: MOV T3.Y, KC0[5].W, +; CM-NEXT: MOV T3.Z, KC0[5].Z, +; CM-NEXT: MOV * T2.W, KC0[7].X, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; CM-NEXT: LSHR T2.X, PV.W, literal.x, +; CM-NEXT: MOV T4.Y, KC0[6].Z, +; CM-NEXT: MOV T4.Z, KC0[6].W, +; CM-NEXT: MOV * T2.W, KC0[4].Y, +; CM-NEXT: 2(2.802597e-45), 0(0.000000e+00) +; CM-NEXT: ALU clause starting at 35: +; CM-NEXT: MULLO_INT T0.X, T0.X, T2.Y, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.X, T2.Y, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.X, T2.Y, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.X, T2.Y, +; CM-NEXT: ADD_INT * T2.W, PV.X, T4.Y, +; CM-NEXT: AND_INT * T2.W, PV.W, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X, T2.W, T2.Y, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T2.W, T2.Y, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T2.W, T2.Y, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T2.W, T2.Y, +; CM-NEXT: MULLO_INT T0.X (MASKED), T0.W, T1.Y, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T1.Y, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T1.Y, +; CM-NEXT: MULLO_INT * T0.W, T0.W, T1.Y, +; CM-NEXT: MULLO_INT T0.X (MASKED), T0.Z, T1.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.Z, T1.Z, +; CM-NEXT: MULLO_INT T0.Z, T0.Z, T1.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.Z, T1.Z, +; CM-NEXT: MULLO_INT T0.X (MASKED), T0.Y, T1.W, +; CM-NEXT: MULLO_INT T0.Y, T0.Y, T1.W, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.Y, T1.W, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.Y, T1.W, +; CM-NEXT: ADD_INT * T0.X, T0.X, T4.Y, +; CM-NEXT: ALU clause starting at 59: +; CM-NEXT: ADD_INT * T2.W, T0.Y, T3.Z, +; CM-NEXT: ADD_INT T0.Z, T0.Z, T3.Y, +; CM-NEXT: AND_INT * T2.W, PV.W, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X (MASKED), T2.W, T1.W, +; CM-NEXT: MULLO_INT T0.Y, T2.W, T1.W, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T2.W, T1.W, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T2.W, T1.W, +; CM-NEXT: ADD_INT T0.Y, PV.Y, T3.Z, +; CM-NEXT: ADD_INT T5.Z, T0.W, T2.Z, BS:VEC_021/SCL_122 +; CM-NEXT: AND_INT * T0.W, T0.Z, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X (MASKED), T0.W, T1.Z, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T1.Z, +; CM-NEXT: MULLO_INT T0.Z, T0.W, T1.Z, +; CM-NEXT: MULLO_INT * T0.W (MASKED), T0.W, T1.Z, +; CM-NEXT: ADD_INT T0.Z, PV.Z, T3.Y, +; CM-NEXT: AND_INT * T0.W, T5.Z, literal.x, +; CM-NEXT: 16777215(2.350989e-38), 0(0.000000e+00) +; CM-NEXT: MULLO_INT T0.X (MASKED), T0.W, T1.Y, +; CM-NEXT: MULLO_INT T0.Y (MASKED), T0.W, T1.Y, +; CM-NEXT: MULLO_INT T0.Z (MASKED), T0.W, T1.Y, +; CM-NEXT: MULLO_INT * T0.W, T0.W, T1.Y, +; CM-NEXT: ADD_INT * T0.W, PV.W, T2.Z, +; CM-NEXT: ALU clause starting at 83: +; CM-NEXT: ADD_INT * T4.Z, T4.Z, literal.x, +; CM-NEXT: -1(nan), 0(0.000000e+00) +; CM-NEXT: SETE_INT * T2.W, PV.Z, 0.0, +; CM-NEXT: PRED_SETNE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0, +; +; GCN-LABEL: mad24_known_bits_destroyed: +; GCN: ; %bb.0: ; %bb +; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-NEXT: v_mov_b32_e32 v5, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v13 +; GCN-NEXT: v_and_b32_e32 v1, 0xffffff, v2 +; GCN-NEXT: v_and_b32_e32 v2, 0xffffff, v3 +; GCN-NEXT: v_and_b32_e32 v3, 0xffffff, v4 +; GCN-NEXT: s_mov_b64 s[8:9], 0 +; GCN-NEXT: s_mov_b32 s6, 0 +; GCN-NEXT: s_mov_b32 s7, 0xf000 +; GCN-NEXT: s_mov_b32 s4, s6 +; GCN-NEXT: s_mov_b32 s5, s6 +; GCN-NEXT: .LBB9_1: ; %bb19 +; GCN-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-NEXT: v_mad_u32_u24 v4, v5, v0, v14 +; GCN-NEXT: s_waitcnt expcnt(0) +; GCN-NEXT: v_mad_u32_u24 v6, v6, v1, v10 +; GCN-NEXT: v_mad_u32_u24 v7, v7, v2, v11 +; GCN-NEXT: v_mad_u32_u24 v8, v8, v3, v12 +; GCN-NEXT: v_add_i32_e32 v15, vcc, -1, v15 +; GCN-NEXT: v_mad_u32_u24 v5, v4, v0, v14 +; GCN-NEXT: v_mad_u32_u24 v6, v6, v1, v10 +; GCN-NEXT: v_mad_u32_u24 v7, v7, v2, v11 +; GCN-NEXT: v_mad_u32_u24 v8, v8, v3, v12 +; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v15 +; GCN-NEXT: buffer_store_dword v5, v[16:17], s[4:7], 0 addr64 +; GCN-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-NEXT: buffer_store_dwordx4 v[5:8], v[18:19], s[4:7], 0 addr64 +; GCN-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-NEXT: s_cbranch_execnz .LBB9_1 +; GCN-NEXT: ; %bb.2: ; %bb18 +; GCN-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) +; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GFX8-LABEL: mad24_known_bits_destroyed: +; GFX8: ; %bb.0: ; %bb +; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX8-NEXT: v_mov_b32_e32 v5, v0 +; GFX8-NEXT: v_and_b32_e32 v0, 0xffffff, v13 +; GFX8-NEXT: v_and_b32_e32 v1, 0xffffff, v2 +; GFX8-NEXT: v_and_b32_e32 v2, 0xffffff, v3 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffffff, v4 +; GFX8-NEXT: s_mov_b64 s[4:5], 0 +; GFX8-NEXT: .LBB9_1: ; %bb19 +; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 +; GFX8-NEXT: v_add_u32_e32 v15, vcc, -1, v15 +; GFX8-NEXT: v_mad_u32_u24 v4, v5, v0, v14 +; GFX8-NEXT: v_mad_u32_u24 v6, v6, v1, v10 +; GFX8-NEXT: v_mad_u32_u24 v7, v7, v2, v11 +; GFX8-NEXT: v_mad_u32_u24 v8, v8, v3, v12 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v15 +; GFX8-NEXT: v_mad_u32_u24 v5, v4, v0, v14 +; GFX8-NEXT: v_mad_u32_u24 v6, v6, v1, v10 +; GFX8-NEXT: v_mad_u32_u24 v7, v7, v2, v11 +; GFX8-NEXT: v_mad_u32_u24 v8, v8, v3, v12 +; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GFX8-NEXT: flat_store_dword v[16:17], v5 +; GFX8-NEXT: flat_store_dwordx4 v[18:19], v[5:8] +; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] +; GFX8-NEXT: s_cbranch_execnz .LBB9_1 +; GFX8-NEXT: ; %bb.2: ; %bb18 +; GFX8-NEXT: s_or_b64 exec, exec, s[4:5] +; GFX8-NEXT: s_waitcnt vmcnt(0) +; GFX8-NEXT: s_setpc_b64 s[30:31] bb: %tmp = and i32 %arg4, 16777215 %tmp9 = extractelement <4 x i32> %arg1, i64 1 diff --git a/llvm/test/CodeGen/AMDGPU/min.ll b/llvm/test/CodeGen/AMDGPU/min.ll index 6a3d31f..0458a64 100644 --- a/llvm/test/CodeGen/AMDGPU/min.ll +++ b/llvm/test/CodeGen/AMDGPU/min.ll @@ -6,9 +6,7 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck --check-prefix=GFX10 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX11,GFX11-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX11,GFX11-FAKE16 %s -; TODO: FIXME-TRUE16 - Enable this llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-TRUE16 %s -; Crashing on v_test_imin_slt_i16 -; LLVM ERROR: Cannot select: 0x5f895f65b050: i16,ch = load<(load (s16) from %ir.b.gep, addrspace 1)> +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-TRUE16 %s ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck --check-prefixes=GFX1250,GFX1250-FAKE16 %s define amdgpu_kernel void @v_test_imin_sle_i32(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 { @@ -1482,20 +1480,35 @@ define amdgpu_kernel void @v_test_imin_slt_i16(ptr addrspace(1) %out, ptr addrsp ; GFX11-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] ; GFX11-FAKE16-NEXT: s_endpgm ; -; GFX1250-LABEL: v_test_imin_slt_i16: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX1250-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 -; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: global_load_u16 v1, v0, s[2:3] scale_offset -; GFX1250-NEXT: global_load_u16 v2, v0, s[6:7] scale_offset -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_min_i16 v1, v1, v2 -; GFX1250-NEXT: global_store_b16 v0, v1, s[0:1] scale_offset -; GFX1250-NEXT: s_endpgm +; GFX1250-TRUE16-LABEL: v_test_imin_slt_i16: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-TRUE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-TRUE16-NEXT: v_and_b32_e32 v1, 0x3ff, v0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3] scale_offset +; GFX1250-TRUE16-NEXT: global_load_u16 v2, v1, s[6:7] scale_offset +; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-TRUE16-NEXT: v_min_i16 v0.l, v0.l, v2.l +; GFX1250-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] scale_offset +; GFX1250-TRUE16-NEXT: s_endpgm +; +; GFX1250-FAKE16-LABEL: v_test_imin_slt_i16: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-FAKE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3] scale_offset +; GFX1250-FAKE16-NEXT: global_load_u16 v2, v0, s[6:7] scale_offset +; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-FAKE16-NEXT: v_min_i16 v1, v1, v2 +; GFX1250-FAKE16-NEXT: global_store_b16 v0, v1, s[0:1] scale_offset +; GFX1250-FAKE16-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %a.gep = getelementptr inbounds i16, ptr addrspace(1) %aptr, i32 %tid %b.gep = getelementptr inbounds i16, ptr addrspace(1) %bptr, i32 %tid @@ -2769,20 +2782,35 @@ define amdgpu_kernel void @v_test_umin_ult_i8(ptr addrspace(1) %out, ptr addrspa ; GFX11-FAKE16-NEXT: global_store_b8 v0, v1, s[0:1] ; GFX11-FAKE16-NEXT: s_endpgm ; -; GFX1250-LABEL: v_test_umin_ult_i8: -; GFX1250: ; %bb.0: -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 -; GFX1250-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 -; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0 -; GFX1250-NEXT: s_wait_kmcnt 0x0 -; GFX1250-NEXT: s_clause 0x1 -; GFX1250-NEXT: global_load_u8 v1, v0, s[2:3] -; GFX1250-NEXT: global_load_u8 v2, v0, s[6:7] -; GFX1250-NEXT: s_wait_loadcnt 0x0 -; GFX1250-NEXT: v_min_u16 v1, v1, v2 -; GFX1250-NEXT: global_store_b8 v0, v1, s[0:1] -; GFX1250-NEXT: s_endpgm +; GFX1250-TRUE16-LABEL: v_test_umin_ult_i8: +; GFX1250-TRUE16: ; %bb.0: +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-TRUE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-TRUE16-NEXT: v_and_b32_e32 v1, 0x3ff, v0 +; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: global_load_u8 v0, v1, s[2:3] +; GFX1250-TRUE16-NEXT: global_load_u8 v2, v1, s[6:7] +; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-TRUE16-NEXT: v_min_u16 v0.l, v0.l, v2.l +; GFX1250-TRUE16-NEXT: global_store_b8 v1, v0, s[0:1] +; GFX1250-TRUE16-NEXT: s_endpgm +; +; GFX1250-FAKE16-LABEL: v_test_umin_ult_i8: +; GFX1250-FAKE16: ; %bb.0: +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x0 +; GFX1250-FAKE16-NEXT: s_load_b64 s[6:7], s[4:5], 0x10 +; GFX1250-FAKE16-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 +; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: global_load_u8 v1, v0, s[2:3] +; GFX1250-FAKE16-NEXT: global_load_u8 v2, v0, s[6:7] +; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0 +; GFX1250-FAKE16-NEXT: v_min_u16 v1, v1, v2 +; GFX1250-FAKE16-NEXT: global_store_b8 v0, v1, s[0:1] +; GFX1250-FAKE16-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %a.gep = getelementptr inbounds i8, ptr addrspace(1) %a.ptr, i32 %tid %b.gep = getelementptr inbounds i8, ptr addrspace(1) %b.ptr, i32 %tid @@ -5069,5 +5097,3 @@ declare i32 @llvm.amdgcn.workitem.id.x() #1 attributes #0 = { nounwind } attributes #1 = { nounwind readnone } -;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: -; GFX1250-FAKE16: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/minmax.ll b/llvm/test/CodeGen/AMDGPU/minmax.ll index 57e6943..56f9c5d 100644 --- a/llvm/test/CodeGen/AMDGPU/minmax.ll +++ b/llvm/test/CodeGen/AMDGPU/minmax.ll @@ -638,6 +638,14 @@ define void @test_med3_minimumnum_maximumnum_f32(ptr addrspace(1) %arg, float %x ; GFX12-NEXT: v_med3_num_f32 v2, v2, v3, v4 ; GFX12-NEXT: global_store_b32 v[0:1], v2, off ; GFX12-NEXT: s_setpc_b64 s[30:31] +; +; GFX1250-LABEL: test_med3_minimumnum_maximumnum_f32: +; GFX1250: ; %bb.0: +; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX1250-NEXT: s_wait_kmcnt 0x0 +; GFX1250-NEXT: v_med3_num_f32 v2, v2, v3, v4 +; GFX1250-NEXT: global_store_b32 v[0:1], v2, off +; GFX1250-NEXT: s_set_pc_i64 s[30:31] %tmp0 = call float @llvm.minimumnum.f32(float %x, float %y) %tmp1 = call float @llvm.maximumnum.f32(float %x, float %y) %tmp2 = call float @llvm.minimumnum.f32(float %tmp1, float %z) @@ -798,7 +806,7 @@ define amdgpu_ps void @s_test_minmax_f16_ieee_false(half inreg %a, half inreg %b ; SDAG-GFX1250-TRUE16-NEXT: s_mov_b32 s5, s4 ; SDAG-GFX1250-TRUE16-NEXT: s_mov_b32 s4, s3 ; SDAG-GFX1250-TRUE16-NEXT: v_maxmin_num_f16 v0.l, s0, s1, v0.l -; SDAG-GFX1250-TRUE16-NEXT: flat_store_b16 v1, v0, s[4:5] +; SDAG-GFX1250-TRUE16-NEXT: global_store_b16 v1, v0, s[4:5] ; SDAG-GFX1250-TRUE16-NEXT: s_endpgm ; ; SDAG-GFX1250-FAKE16-LABEL: s_test_minmax_f16_ieee_false: @@ -813,12 +821,12 @@ define amdgpu_ps void @s_test_minmax_f16_ieee_false(half inreg %a, half inreg %b ; GISEL-GFX1250-TRUE16-LABEL: s_test_minmax_f16_ieee_false: ; GISEL-GFX1250-TRUE16: ; %bb.0: ; GISEL-GFX1250-TRUE16-NEXT: s_max_num_f16 s0, s0, s1 +; GISEL-GFX1250-TRUE16-NEXT: v_mov_b32_e32 v1, 0 ; GISEL-GFX1250-TRUE16-NEXT: s_mov_b32 s6, s3 ; GISEL-GFX1250-TRUE16-NEXT: s_mov_b32 s7, s4 -; GISEL-GFX1250-TRUE16-NEXT: v_mov_b32_e32 v1, 0 ; GISEL-GFX1250-TRUE16-NEXT: s_min_num_f16 s0, s0, s2 -; GISEL-GFX1250-TRUE16-NEXT: v_mov_b32_e32 v0, s0 -; GISEL-GFX1250-TRUE16-NEXT: flat_store_b16 v1, v0, s[6:7] +; GISEL-GFX1250-TRUE16-NEXT: v_mov_b16_e32 v0.l, s0 +; GISEL-GFX1250-TRUE16-NEXT: global_store_b16 v1, v0, s[6:7] ; GISEL-GFX1250-TRUE16-NEXT: s_endpgm ; ; GISEL-GFX1250-FAKE16-LABEL: s_test_minmax_f16_ieee_false: @@ -1246,7 +1254,7 @@ define void @test_med3_f16(ptr addrspace(1) %arg, half %x, half %y, half %z) #0 ; SDAG-GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 ; SDAG-GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 ; SDAG-GFX1250-TRUE16-NEXT: v_med3_num_f16 v2.l, v2.l, v3.l, v4.l -; SDAG-GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v2 +; SDAG-GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v2, off ; SDAG-GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] ; ; SDAG-GFX1250-FAKE16-LABEL: test_med3_f16: @@ -1262,7 +1270,7 @@ define void @test_med3_f16(ptr addrspace(1) %arg, half %x, half %y, half %z) #0 ; GISEL-GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 ; GISEL-GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GISEL-GFX1250-TRUE16-NEXT: v_med3_num_f16 v2.l, v2.l, v3.l, v4.l -; GISEL-GFX1250-TRUE16-NEXT: flat_store_b16 v[0:1], v2 +; GISEL-GFX1250-TRUE16-NEXT: global_store_b16 v[0:1], v2, off ; GISEL-GFX1250-TRUE16-NEXT: s_set_pc_i64 s[30:31] ; ; GISEL-GFX1250-FAKE16-LABEL: test_med3_f16: diff --git a/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll b/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll index 42469c8..23e90b3 100644 --- a/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll +++ b/llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll @@ -202,13 +202,13 @@ attributes #5 = { "amdgpu-flat-work-group-size"="128,512" } attributes #6 = { "amdgpu-flat-work-group-size"="512,512" } attributes #7 = { "amdgpu-flat-work-group-size"="64,256" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="64,128" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="64,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="128,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="512,1024" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR7]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="64,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR8]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-flat-work-group-size"="1,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-flat-work-group-size"="64,128" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="128,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-flat-work-group-size"="64,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-flat-work-group-size"="128,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-flat-work-group-size"="512,1024" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="512,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { "amdgpu-flat-work-group-size"="64,256" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll b/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll index 06533b4..0be3147 100644 --- a/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll +++ b/llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll @@ -399,25 +399,25 @@ attributes #17 = { "amdgpu-waves-per-eu"="5,8" } attributes #18 = { "amdgpu-waves-per-eu"="9,10" } attributes #19 = { "amdgpu-waves-per-eu"="8,9" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,2" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,4" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,9" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,1" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR7]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,9" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR8]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR9]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR10]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR11]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,123" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR12]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR13]] = { "amdgpu-agpr-alloc"="0" "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR14]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR15]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR16]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="6,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR17]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,5" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR18]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,8" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR19]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR20]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="8,9" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,2" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,4" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,9" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,1" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR6]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR7]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="2,9" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR8]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR9]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR10]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR11]] = { "amdgpu-flat-work-group-size"="1,64" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="1,123" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR12]] = { "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR13]] = { "amdgpu-flat-work-group-size"="1,512" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR14]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="3,6" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR15]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="4,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR16]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="6,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR17]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,5" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR18]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="5,8" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR19]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="9,10" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR20]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-waves-per-eu"="8,9" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll b/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll index 8930626..33da671 100644 --- a/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll +++ b/llvm/test/CodeGen/AMDGPU/recursive_global_initializer.ll @@ -19,5 +19,5 @@ define void @hoge() { ret void } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll b/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll index 3dfb0e1..f847d66 100644 --- a/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll +++ b/llvm/test/CodeGen/AMDGPU/remove-no-kernel-id-attribute.ll @@ -191,12 +191,12 @@ define amdgpu_kernel void @kernel_lds_recursion() { !1 = !{i32 1, !"amdhsa_code_object_version", i32 400} ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR3]] = { "amdgpu-lds-size"="4" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "amdgpu-agpr-alloc"="0" "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="0" "amdgpu-lds-size"="4" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR4]] = { "amdgpu-lds-size"="2" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR5]] = { "amdgpu-lds-size"="4" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR6:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) } ; CHECK: attributes #[[ATTR7:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } ;. diff --git a/llvm/test/CodeGen/AMDGPU/sdiv64.ll b/llvm/test/CodeGen/AMDGPU/sdiv64.ll index 697bcc3..5f6d622 100644 --- a/llvm/test/CodeGen/AMDGPU/sdiv64.ll +++ b/llvm/test/CodeGen/AMDGPU/sdiv64.ll @@ -206,8 +206,11 @@ define amdgpu_kernel void @s_test_sdiv(ptr addrspace(1) %out, i64 %x, i64 %y) { ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s18, s16, 1 -; GCN-IR-NEXT: s_addc_u32 s19, s17, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[10:11], s[18:19], 0 +; GCN-IR-NEXT: s_cselect_b64 s[10:11], -1, 0 +; GCN-IR-NEXT: s_or_b32 s10, s10, s11 +; GCN-IR-NEXT: s_cmp_lg_u32 s10, 0 +; GCN-IR-NEXT: s_addc_u32 s10, s17, 0 +; GCN-IR-NEXT: s_cselect_b64 s[10:11], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s16, 63, s16 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[10:11] ; GCN-IR-NEXT: s_lshl_b64 s[10:11], s[12:13], s16 @@ -217,9 +220,9 @@ define amdgpu_kernel void @s_test_sdiv(ptr addrspace(1) %out, i64 %x, i64 %y) { ; GCN-IR-NEXT: s_add_u32 s18, s2, -1 ; GCN-IR-NEXT: s_addc_u32 s19, s3, -1 ; GCN-IR-NEXT: s_not_b64 s[8:9], s[14:15] -; GCN-IR-NEXT: s_add_u32 s12, s8, s20 -; GCN-IR-NEXT: s_addc_u32 s13, s9, 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], 0 +; GCN-IR-NEXT: s_add_u32 s14, s8, s20 +; GCN-IR-NEXT: s_addc_u32 s15, s9, 0 +; GCN-IR-NEXT: s_mov_b64 s[12:13], 0 ; GCN-IR-NEXT: s_mov_b32 s9, 0 ; GCN-IR-NEXT: .LBB0_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -227,19 +230,22 @@ define amdgpu_kernel void @s_test_sdiv(ptr addrspace(1) %out, i64 %x, i64 %y) { ; GCN-IR-NEXT: s_lshr_b32 s8, s11, 31 ; GCN-IR-NEXT: s_lshl_b64 s[10:11], s[10:11], 1 ; GCN-IR-NEXT: s_or_b64 s[16:17], s[16:17], s[8:9] -; GCN-IR-NEXT: s_or_b64 s[10:11], s[14:15], s[10:11] +; GCN-IR-NEXT: s_or_b64 s[10:11], s[12:13], s[10:11] ; GCN-IR-NEXT: s_sub_u32 s8, s18, s16 ; GCN-IR-NEXT: s_subb_u32 s8, s19, s17 -; GCN-IR-NEXT: s_ashr_i32 s14, s8, 31 -; GCN-IR-NEXT: s_mov_b32 s15, s14 -; GCN-IR-NEXT: s_and_b32 s8, s14, 1 -; GCN-IR-NEXT: s_and_b64 s[14:15], s[14:15], s[2:3] -; GCN-IR-NEXT: s_sub_u32 s16, s16, s14 -; GCN-IR-NEXT: s_subb_u32 s17, s17, s15 -; GCN-IR-NEXT: s_add_u32 s12, s12, 1 -; GCN-IR-NEXT: s_addc_u32 s13, s13, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[20:21], s[12:13], 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], s[8:9] +; GCN-IR-NEXT: s_ashr_i32 s12, s8, 31 +; GCN-IR-NEXT: s_mov_b32 s13, s12 +; GCN-IR-NEXT: s_and_b32 s8, s12, 1 +; GCN-IR-NEXT: s_and_b64 s[20:21], s[12:13], s[2:3] +; GCN-IR-NEXT: s_sub_u32 s16, s16, s20 +; GCN-IR-NEXT: s_subb_u32 s17, s17, s21 +; GCN-IR-NEXT: s_add_u32 s14, s14, 1 +; GCN-IR-NEXT: s_cselect_b64 s[20:21], -1, 0 +; GCN-IR-NEXT: s_or_b32 s20, s20, s21 +; GCN-IR-NEXT: s_cmp_lg_u32 s20, 0 +; GCN-IR-NEXT: s_addc_u32 s15, s15, 0 +; GCN-IR-NEXT: s_cselect_b64 s[20:21], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[12:13], s[8:9] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[20:21] ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_3 ; GCN-IR-NEXT: .LBB0_4: ; %Flow7 @@ -389,25 +395,25 @@ define i64 @v_test_sdiv(i64 %x, i64 %y) { ; GCN-IR-LABEL: v_test_sdiv: ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v1 -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v13, 31, v3 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v12 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v1, v12, vcc -; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v13 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v3, v13 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v13 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v13, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v10 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v11, 31, v3 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v10 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v1, v10, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v11 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v3, v11 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v11 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v11, vcc ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e64 v2, s[6:7], 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v6 ; GCN-IR-NEXT: v_add_i32_e64 v2, s[6:7], 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v7 -; GCN-IR-NEXT: v_min_u32_e32 v11, v2, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v2, s[6:7], v10, v11 +; GCN-IR-NEXT: v_min_u32_e32 v9, v2, v3 +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[6:7], v8, v9 ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[6:7] ; GCN-IR-NEXT: v_subb_u32_e64 v3, s[6:7], 0, 0, s[6:7] @@ -416,70 +422,69 @@ define i64 @v_test_sdiv(i64 %x, i64 %y) { ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] ; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[2:3] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 -; GCN-IR-NEXT: v_mov_b32_e32 v14, v12 -; GCN-IR-NEXT: v_mov_b32_e32 v15, v13 +; GCN-IR-NEXT: v_mov_b32_e32 v12, v10 +; GCN-IR-NEXT: v_mov_b32_e32 v13, v11 ; GCN-IR-NEXT: v_cndmask_b32_e64 v5, v7, 0, s[4:5] ; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v6, 0, s[4:5] ; GCN-IR-NEXT: s_and_b64 s[4:5], s[6:7], vcc ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB1_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[6:7], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB1_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v16, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v17, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_not_b32_e32 v4, v10 -; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[6:7], v8 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, v4, v11 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v7, s[4:5], -1, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], v[6:7], v14 +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_not_b32_e32 v4, v8 +; GCN-IR-NEXT: v_add_i32_e32 v16, vcc, v4, v9 +; GCN-IR-NEXT: v_addc_u32_e64 v17, s[8:9], -1, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB1_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v16, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v17, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v14, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v15, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v16, vcc, 1, v16 +; GCN-IR-NEXT: v_addc_u32_e32 v17, vcc, 0, v17, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB1_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB1_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB1_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v1 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v0 ; GCN-IR-NEXT: .LBB1_6: ; %Flow5 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] -; GCN-IR-NEXT: v_xor_b32_e32 v0, v13, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v15, v14 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v11, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v13, v12 ; GCN-IR-NEXT: v_xor_b32_e32 v3, v4, v0 ; GCN-IR-NEXT: v_xor_b32_e32 v2, v5, v1 ; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v3, v0 @@ -1293,34 +1298,37 @@ define amdgpu_kernel void @s_test_sdiv_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_xor_b64 s[2:3], s[2:3], s[4:5] ; GCN-IR-NEXT: s_sub_u32 s2, s2, s4 ; GCN-IR-NEXT: s_subb_u32 s3, s3, s4 -; GCN-IR-NEXT: s_flbit_i32_b64 s14, s[2:3] -; GCN-IR-NEXT: s_add_u32 s10, s14, 0xffffffc5 +; GCN-IR-NEXT: s_flbit_i32_b64 s16, s[2:3] +; GCN-IR-NEXT: s_add_u32 s10, s16, 0xffffffc5 ; GCN-IR-NEXT: s_addc_u32 s11, 0, -1 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[2:3], 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[12:13], s[10:11], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[10:11], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[14:15], s[10:11], 63 ; GCN-IR-NEXT: s_or_b64 s[12:13], s[8:9], s[12:13] ; GCN-IR-NEXT: s_and_b64 s[8:9], s[12:13], exec ; GCN-IR-NEXT: s_cselect_b32 s8, 0, 24 -; GCN-IR-NEXT: s_or_b64 s[12:13], s[12:13], s[16:17] +; GCN-IR-NEXT: s_or_b64 s[12:13], s[12:13], s[14:15] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[12:13] ; GCN-IR-NEXT: s_mov_b32 s9, 0 ; GCN-IR-NEXT: s_cbranch_vccz .LBB10_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s12, s10, 1 -; GCN-IR-NEXT: s_addc_u32 s13, s11, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[12:13], 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 +; GCN-IR-NEXT: s_or_b32 s8, s8, s9 +; GCN-IR-NEXT: s_cmp_lg_u32 s8, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s11, 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s10, 63, s10 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[8:9] ; GCN-IR-NEXT: s_lshl_b64 s[8:9], 24, s10 ; GCN-IR-NEXT: s_cbranch_vccz .LBB10_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[12:13], 24, s12 -; GCN-IR-NEXT: s_add_u32 s16, s2, -1 -; GCN-IR-NEXT: s_addc_u32 s17, s3, -1 -; GCN-IR-NEXT: s_sub_u32 s10, 58, s14 -; GCN-IR-NEXT: s_subb_u32 s11, 0, 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], 0 +; GCN-IR-NEXT: s_add_u32 s14, s2, -1 +; GCN-IR-NEXT: s_addc_u32 s15, s3, -1 +; GCN-IR-NEXT: s_sub_u32 s16, 58, s16 +; GCN-IR-NEXT: s_subb_u32 s17, 0, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 ; GCN-IR-NEXT: s_mov_b32 s7, 0 ; GCN-IR-NEXT: .LBB10_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -1328,19 +1336,22 @@ define amdgpu_kernel void @s_test_sdiv_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_lshr_b32 s6, s9, 31 ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[8:9], 1 ; GCN-IR-NEXT: s_or_b64 s[12:13], s[12:13], s[6:7] -; GCN-IR-NEXT: s_or_b64 s[8:9], s[14:15], s[8:9] -; GCN-IR-NEXT: s_sub_u32 s6, s16, s12 -; GCN-IR-NEXT: s_subb_u32 s6, s17, s13 -; GCN-IR-NEXT: s_ashr_i32 s14, s6, 31 -; GCN-IR-NEXT: s_mov_b32 s15, s14 -; GCN-IR-NEXT: s_and_b32 s6, s14, 1 -; GCN-IR-NEXT: s_and_b64 s[14:15], s[14:15], s[2:3] -; GCN-IR-NEXT: s_sub_u32 s12, s12, s14 -; GCN-IR-NEXT: s_subb_u32 s13, s13, s15 -; GCN-IR-NEXT: s_add_u32 s10, s10, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s11, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[18:19], s[10:11], 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], s[6:7] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[10:11], s[8:9] +; GCN-IR-NEXT: s_sub_u32 s6, s14, s12 +; GCN-IR-NEXT: s_subb_u32 s6, s15, s13 +; GCN-IR-NEXT: s_ashr_i32 s10, s6, 31 +; GCN-IR-NEXT: s_mov_b32 s11, s10 +; GCN-IR-NEXT: s_and_b32 s6, s10, 1 +; GCN-IR-NEXT: s_and_b64 s[18:19], s[10:11], s[2:3] +; GCN-IR-NEXT: s_sub_u32 s12, s12, s18 +; GCN-IR-NEXT: s_subb_u32 s13, s13, s19 +; GCN-IR-NEXT: s_add_u32 s16, s16, 1 +; GCN-IR-NEXT: s_cselect_b64 s[18:19], -1, 0 +; GCN-IR-NEXT: s_or_b32 s18, s18, s19 +; GCN-IR-NEXT: s_cmp_lg_u32 s18, 0 +; GCN-IR-NEXT: s_addc_u32 s17, s17, 0 +; GCN-IR-NEXT: s_cselect_b64 s[18:19], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], s[6:7] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[18:19] ; GCN-IR-NEXT: s_cbranch_vccz .LBB10_3 ; GCN-IR-NEXT: .LBB10_4: ; %Flow6 @@ -1472,17 +1483,17 @@ define i64 @v_test_sdiv_k_num_i64(i64 %x) { ; GCN-IR-LABEL: v_test_sdiv_k_num_i64: ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v1 -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v12 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v12, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v10 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v10, vcc ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 ; GCN-IR-NEXT: s_movk_i32 s6, 0xffc5 -; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v10 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v8 ; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] @@ -1490,69 +1501,68 @@ define i64 @v_test_sdiv_k_num_i64(i64 %x) { ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc ; GCN-IR-NEXT: v_cndmask_b32_e64 v4, 24, 0, s[4:5] ; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 -; GCN-IR-NEXT: v_mov_b32_e32 v13, v12 +; GCN-IR-NEXT: v_mov_b32_e32 v11, v10 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB11_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], 24, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB11_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_lshr_b64 v[8:9], 24, v6 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 58, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v14, vcc, 58, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], 24, v6 +; GCN-IR-NEXT: v_subb_u32_e64 v15, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB11_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v14, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v15, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v12, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v13, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, 1, v14 +; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, 0, v15, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB11_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB11_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB11_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v1 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v0 ; GCN-IR-NEXT: .LBB11_6: ; %Flow5 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] -; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v5, v13 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v13, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v5, v11 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v11, vcc ; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 24, %x ret i64 %result @@ -1665,17 +1675,17 @@ define i64 @v_test_sdiv_pow2_k_num_i64(i64 %x) { ; GCN-IR-LABEL: v_test_sdiv_pow2_k_num_i64: ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v1 -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v12 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v12, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v10 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v10, vcc ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 ; GCN-IR-NEXT: s_movk_i32 s6, 0xffd0 -; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v10 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v8 ; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] @@ -1684,70 +1694,69 @@ define i64 @v_test_sdiv_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc ; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v4, 0, s[4:5] ; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 -; GCN-IR-NEXT: v_mov_b32_e32 v13, v12 +; GCN-IR-NEXT: v_mov_b32_e32 v11, v10 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB12_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc -; GCN-IR-NEXT: s_mov_b64 s[4:5], 0x8000 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0x8000 +; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[8:9], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[4:5], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[8:9] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[10:11], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[10:11] ; GCN-IR-NEXT: s_cbranch_execz .LBB12_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[4:5], v6 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 47, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v14, vcc, 47, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], s[8:9], v6 +; GCN-IR-NEXT: v_subb_u32_e64 v15, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB12_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v14, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v15, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v12, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v13, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, 1, v14 +; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, 0, v15, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB12_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB12_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB12_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v1 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v0 ; GCN-IR-NEXT: .LBB12_6: ; %Flow5 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] -; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v5, v13 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v13, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v5, v11 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v11, vcc ; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 32768, %x ret i64 %result @@ -1767,20 +1776,20 @@ define i64 @v_test_sdiv_pow2_k_den_i64(i64 %x) { ; GCN-IR-LABEL: v_test_sdiv_pow2_k_den_i64: ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v1 -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v10 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v10 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v0, v10 -; GCN-IR-NEXT: v_subb_u32_e32 v5, vcc, v1, v10, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v8 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v8 +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v0, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v5, vcc, v1, v8, vcc ; GCN-IR-NEXT: v_ffbh_u32_e32 v0, v4 ; GCN-IR-NEXT: v_add_i32_e64 v0, s[4:5], 32, v0 ; GCN-IR-NEXT: v_ffbh_u32_e32 v1, v5 -; GCN-IR-NEXT: v_min_u32_e32 v8, v0, v1 -; GCN-IR-NEXT: v_sub_i32_e64 v0, s[4:5], 48, v8 +; GCN-IR-NEXT: v_min_u32_e32 v6, v0, v1 +; GCN-IR-NEXT: v_sub_i32_e64 v0, s[4:5], 48, v6 ; GCN-IR-NEXT: v_subb_u32_e64 v1, s[4:5], 0, 0, s[4:5] ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5] ; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[0:1] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v10 +; GCN-IR-NEXT: v_mov_b32_e32 v9, v8 ; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] ; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[0:1] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 @@ -1790,61 +1799,60 @@ define i64 @v_test_sdiv_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB13_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v7, vcc, 1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v0, s[4:5], 63, v0 -; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], v0 +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB13_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_lshr_b64 v[6:7], v[4:5], v6 -; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 0xffffffcf, v8 -; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v5, s[4:5], 0, -1, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, 0xffffffcf, v6 +; GCN-IR-NEXT: v_lshr_b64 v[4:5], v[4:5], v7 +; GCN-IR-NEXT: v_addc_u32_e64 v11, s[8:9], 0, -1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 -; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: s_movk_i32 s10, 0x7fff ; GCN-IR-NEXT: .LBB13_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v1 -; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v2 -; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, s12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v2 ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[0:1], 1 -; GCN-IR-NEXT: v_subb_u32_e32 v2, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 1, v4 -; GCN-IR-NEXT: v_or_b32_e32 v0, v8, v0 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc -; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v8 -; GCN-IR-NEXT: v_and_b32_e32 v8, 0x8000, v8 -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5] -; GCN-IR-NEXT: v_or_b32_e32 v1, v9, v1 -; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v6, v8 -; GCN-IR-NEXT: v_mov_b32_e32 v9, v3 -; GCN-IR-NEXT: v_subbrev_u32_e64 v7, s[4:5], 0, v7, s[4:5] -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v8, v2 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, s10, v4 +; GCN-IR-NEXT: v_subb_u32_e32 v2, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_or_b32_e32 v0, v6, v0 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 0x8000, v6 +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v4, v6 +; GCN-IR-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, 1, v10 +; GCN-IR-NEXT: v_or_b32_e32 v1, v7, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB13_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB13_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB13_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[0:1], 1 ; GCN-IR-NEXT: v_or_b32_e32 v3, v3, v1 ; GCN-IR-NEXT: v_or_b32_e32 v2, v2, v0 ; GCN-IR-NEXT: .LBB13_6: ; %Flow5 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] -; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v10 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v3, v11 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v11, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v8 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v3, v9 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v9, vcc ; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 %x, 32768 ret i64 %result diff --git a/llvm/test/CodeGen/AMDGPU/select-fabs-fneg-extract.v2f16.ll b/llvm/test/CodeGen/AMDGPU/select-fabs-fneg-extract.v2f16.ll index bb22144..9814ed8 100644 --- a/llvm/test/CodeGen/AMDGPU/select-fabs-fneg-extract.v2f16.ll +++ b/llvm/test/CodeGen/AMDGPU/select-fabs-fneg-extract.v2f16.ll @@ -1,15 +1,9 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn -mcpu=hawaii < %s | FileCheck -check-prefixes=CI,CI-SAFE %s -; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -check-prefixes=VI,VI-SAFE %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx906 < %s | FileCheck -check-prefixes=GFX9,GFX9-SAFE %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SAFE,GFX11-SAFE-TRUE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SAFE,GFX11-SAFE-FAKE16 %s - -; RUN: llc -mtriple=amdgcn -mcpu=hawaii -enable-no-signed-zeros-fp-math < %s | FileCheck -check-prefixes=CI,CI-NSZ %s -; RUN: llc -mtriple=amdgcn -mcpu=fiji -enable-no-signed-zeros-fp-math < %s | FileCheck -check-prefixes=VI,VI-NSZ %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -enable-no-signed-zeros-fp-math < %s | FileCheck -check-prefixes=GFX9,GFX9-NSZ %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -enable-no-signed-zeros-fp-math < %s | FileCheck -check-prefixes=GFX11,GFX11-NSZ,GFX11-NSZ-TRUE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -enable-no-signed-zeros-fp-math < %s | FileCheck -check-prefixes=GFX11,GFX11-NSZ,GFX11-NSZ-FAKE16 %s +; RUN: llc -mtriple=amdgcn -mcpu=hawaii < %s | FileCheck -check-prefixes=CI %s +; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -check-prefixes=VI %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx906 < %s | FileCheck -check-prefixes=GFX9 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s define <2 x half> @add_select_fabs_fabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 x half> %y, <2 x half> %z) { ; CI-LABEL: add_select_fabs_fabs_v2f16: @@ -63,69 +57,37 @@ define <2 x half> @add_select_fabs_fabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -198,73 +160,39 @@ define { <2 x half>, <2 x half> } @add_select_multi_use_lhs_fabs_fabs_v2f16(<2 x ; GFX9-NEXT: v_pk_add_f16 v1, v1, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_multi_use_lhs_fabs_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v1, v2, v4 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v0.h, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v5 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_multi_use_lhs_fabs_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v1, v2, v4 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v5 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_multi_use_lhs_fabs_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v1, v2, v4 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v0.h, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v5 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_multi_use_lhs_fabs_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v1, v2, v4 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v5 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_multi_use_lhs_fabs_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v1, v2, v4 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v0.h, v2.h, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v5 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_multi_use_lhs_fabs_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v1, v2, v4 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v5 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -328,73 +256,39 @@ define { <2 x half>, <2 x half> } @add_select_multi_store_use_lhs_fabs_fabs_v2f1 ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_multi_store_use_lhs_fabs_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-TRUE16-NEXT: v_mov_b32_e32 v1, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v0.h, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_multi_store_use_lhs_fabs_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_mov_b32_e32 v1, v2 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_multi_store_use_lhs_fabs_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NSZ-TRUE16-NEXT: v_mov_b32_e32 v1, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v0.h, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_multi_store_use_lhs_fabs_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_mov_b32_e32 v1, v2 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_multi_store_use_lhs_fabs_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v0.l, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v0.h, v2.h, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_multi_store_use_lhs_fabs_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -469,73 +363,39 @@ define { <2 x half>, <2 x half> } @add_select_multi_use_rhs_fabs_fabs_v2f16(<2 x ; GFX9-NEXT: v_pk_add_f16 v1, v2, v5 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_multi_use_rhs_fabs_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v1, v2, v5 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_multi_use_rhs_fabs_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v1, v3, v5 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_multi_use_rhs_fabs_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v1, v2, v5 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_multi_use_rhs_fabs_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v1, v3, v5 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_multi_use_rhs_fabs_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v1, v2, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_multi_use_rhs_fabs_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v1, v3, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -597,63 +457,34 @@ define <2 x half> @add_select_fabs_var_v2f16(<2 x i32> %c, <2 x half> %x, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_var_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v1.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v1.l, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_var_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_var_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v1.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v1.l, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_var_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_var_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v1.h, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v1.l, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_var_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %select = select <2 x i1> %cmp, <2 x half> %fabs.x, <2 x half> %y @@ -709,61 +540,33 @@ define <2 x half> @add_select_fabs_negk_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_negk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_negk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_negk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_negk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_negk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_negk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %select = select <2 x i1> %cmp, <2 x half> %fabs, <2 x half> <half -1.0, half -1.0> @@ -815,61 +618,33 @@ define <2 x half> @add_select_fabs_negk_negk_v2f16(<2 x i32> %c, <2 x half> %x) ; GFX9-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_negk_negk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_negk_negk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_negk_negk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_negk_negk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_negk_negk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_negk_negk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %select = select <2 x i1> %cmp, <2 x half> <half -2.0, half -2.0>, <2 x half> <half -1.0, half -1.0> %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %select) @@ -920,61 +695,33 @@ define <2 x half> @add_select_posk_posk_v2f16(<2 x i32> %c, <2 x half> %x) { ; GFX9-NEXT: v_pk_add_f16 v0, v0, v2 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_posk_posk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0x3c00 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0x4000, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0x4000, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_posk_posk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_mov_b32_e32 v3, 0x4000 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_posk_posk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0x3c00 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0x4000, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0x4000, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_posk_posk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_mov_b32_e32 v3, 0x4000 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_posk_posk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0x3c00 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0x4000, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0x4000, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_posk_posk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, 0x4000 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %select = select <2 x i1> %cmp, <2 x half> <half 2.0, half 2.0>, <2 x half> <half 1.0, half 1.0> %add = fadd <2 x half> %select, %x @@ -1029,61 +776,33 @@ define <2 x half> @add_select_negk_fabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negk_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negk_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negk_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negk_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negk_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negk_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %select = select <2 x i1> %cmp, <2 x half> <half -1.0, half -1.0>, <2 x half> %fabs @@ -1140,61 +859,33 @@ define <2 x half> @add_select_negliteralk_fabs_v2f16(<2 x i32> %c, <2 x half> %x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negliteralk_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xe400, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xe400, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negliteralk_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xe400, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xe400, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negliteralk_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xe400, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xe400, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negliteralk_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xe400, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xe400, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negliteralk_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xe400, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xe400, v2.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negliteralk_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xe400, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xe400, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %select = select <2 x i1> %cmp, <2 x half> <half -1024.0, half -1024.0>, <2 x half> %fabs @@ -1250,61 +941,33 @@ define <2 x half> @add_select_fabs_posk_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_posk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_posk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_posk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_posk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_posk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_posk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %select = select <2 x i1> %cmp, <2 x half> %fabs, <2 x half> <half 1.0, half 1.0> @@ -1360,61 +1023,33 @@ define <2 x half> @add_select_posk_fabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_posk_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_posk_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_posk_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_posk_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_posk_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_posk_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %select = select <2 x i1> %cmp, <2 x half> <half 1.0, half 1.0>, <2 x half> %fabs @@ -1470,57 +1105,31 @@ define <2 x half> @add_select_fneg_fneg_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_fneg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_fneg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_fneg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_fneg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_fneg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_fneg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %fneg.y = fneg <2 x half> %y @@ -1587,61 +1196,33 @@ define { <2 x half>, <2 x half> } @add_select_multi_use_lhs_fneg_fneg_v2f16(<2 x ; GFX9-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_multi_use_lhs_fneg_fneg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_multi_use_lhs_fneg_fneg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_multi_use_lhs_fneg_fneg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_multi_use_lhs_fneg_fneg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_multi_use_lhs_fneg_fneg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_multi_use_lhs_fneg_fneg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v1, v5, v2 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %fneg.y = fneg <2 x half> %y @@ -1705,61 +1286,33 @@ define { <2 x half>, <2 x half> } @add_select_multi_store_use_lhs_fneg_fneg_v2f1 ; GFX9-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_multi_store_use_lhs_fneg_fneg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_multi_store_use_lhs_fneg_fneg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_multi_store_use_lhs_fneg_fneg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_multi_store_use_lhs_fneg_fneg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_multi_store_use_lhs_fneg_fneg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_multi_store_use_lhs_fneg_fneg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %fneg.y = fneg <2 x half> %y @@ -1828,61 +1381,33 @@ define { <2 x half>, <2 x half> } @add_select_multi_use_rhs_fneg_fneg_v2f16(<2 x ; GFX9-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_multi_use_rhs_fneg_fneg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_multi_use_rhs_fneg_fneg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_multi_use_rhs_fneg_fneg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_multi_use_rhs_fneg_fneg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_multi_use_rhs_fneg_fneg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_multi_use_rhs_fneg_fneg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v7, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v1, v5, v3 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %fneg.y = fneg <2 x half> %y @@ -1948,63 +1473,34 @@ define <2 x half> @add_select_fneg_var_v2f16(<2 x i32> %c, <2 x half> %x, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_var_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v1.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v1.l, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_var_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_var_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v1.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v1.l, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_var_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_var_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v1, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.h, v1.h, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, v1.l, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_var_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> %fneg.x, <2 x half> %y @@ -2058,55 +1554,30 @@ define <2 x half> @add_select_fneg_negk_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_negk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_negk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_negk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_negk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_negk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_negk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> %fneg.x, <2 x half> <half -1.0, half -1.0> @@ -2161,55 +1632,30 @@ define <2 x half> @add_select_fneg_inv2pi_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_inv2pi_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xb118, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xb118, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_inv2pi_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xb118, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xb118, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_inv2pi_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xb118, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xb118, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_inv2pi_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xb118, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xb118, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_inv2pi_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xb118, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xb118, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_inv2pi_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xb118, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xb118, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> %fneg.x, <2 x half> <half 0xH3118, half 0xH3118> @@ -2264,55 +1710,30 @@ define <2 x half> @add_select_fneg_neginv2pi_v2f16(<2 x i32> %c, <2 x half> %x, ; GFX9-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_neginv2pi_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3118, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3118, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_neginv2pi_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3118, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3118, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_neginv2pi_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3118, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3118, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_neginv2pi_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3118, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3118, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_neginv2pi_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3118, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3118, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_neginv2pi_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3118, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3118, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> %fneg.x, <2 x half> <half 0xHB118, half 0xHB118> @@ -2363,61 +1784,33 @@ define <2 x half> @add_select_negk_negk_v2f16(<2 x i32> %c, <2 x half> %x) { ; GFX9-NEXT: v_pk_add_f16 v0, v0, v2 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negk_negk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negk_negk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negk_negk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negk_negk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negk_negk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negk_negk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %select = select <2 x i1> %cmp, <2 x half> <half -2.0, half -2.0>, <2 x half> <half -1.0, half -1.0> %add = fadd <2 x half> %select, %x @@ -2469,61 +1862,33 @@ define <2 x half> @add_select_negliteralk_negliteralk_v2f16(<2 x i32> %c, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v2 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negliteralk_negliteralk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xec00 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xe800, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xe800, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negliteralk_negliteralk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_mov_b32_e32 v3, 0xe800 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xec00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xec00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negliteralk_negliteralk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xec00 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xe800, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xe800, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negliteralk_negliteralk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_mov_b32_e32 v3, 0xe800 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xec00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xec00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negliteralk_negliteralk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xec00 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xe800, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xe800, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negliteralk_negliteralk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, 0xe800 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xec00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xec00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %select = select <2 x i1> %cmp, <2 x half> <half -2048.0, half -2048.0>, <2 x half> <half -4096.0, half -4096.0> %add = fadd <2 x half> %select, %x @@ -2573,61 +1938,33 @@ define <2 x half> @add_select_fneg_negk_negk_v2f16(<2 x i32> %c, <2 x half> %x) ; GFX9-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_negk_negk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_negk_negk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_negk_negk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_negk_negk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_negk_negk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0xbc00 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v3.l, 0xc000, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v3.l, 0xc000, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pack_b32_f16 v0, v0.h, v0.l +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_negk_negk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, 0xc000 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_pack_b32_f16 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v2, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %select = select <2 x i1> %cmp, <2 x half> <half -2.0, half -2.0>, <2 x half> <half -1.0, half -1.0> %fneg.x = fneg <2 x half> %select @@ -2681,55 +2018,30 @@ define <2 x half> @add_select_negk_fneg_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negk_fneg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negk_fneg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negk_fneg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negk_fneg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negk_fneg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x3c00, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x3c00, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negk_fneg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x3c00, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x3c00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> <half -1.0, half -1.0>, <2 x half> %fneg.x @@ -2783,55 +2095,30 @@ define <2 x half> @add_select_fneg_posk_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fneg_posk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fneg_posk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fneg_posk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fneg_posk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fneg_posk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fneg_posk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> %fneg.x, <2 x half> <half 1.0, half 1.0> @@ -2885,55 +2172,30 @@ define <2 x half> @add_select_posk_fneg_v2f16(<2 x i32> %c, <2 x half> %x, <2 x ; GFX9-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_posk_fneg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_posk_fneg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_posk_fneg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_posk_fneg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_posk_fneg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xbc00, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xbc00, v2.l, s0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_posk_fneg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xbc00, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xbc00, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v3, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %select = select <2 x i1> %cmp, <2 x half> <half 1.0, half 1.0>, <2 x half> %fneg.x @@ -2997,69 +2259,37 @@ define <2 x half> @add_select_negfabs_fabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negfabs_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negfabs_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negfabs_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negfabs_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negfabs_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negfabs_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.fabs.x = fneg <2 x half> %fabs.x @@ -3125,69 +2355,37 @@ define <2 x half> @add_select_fabs_negfabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_negfabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_negfabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_or_b32_e32 v3, 0x80008000, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_negfabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_negfabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_or_b32_e32 v3, 0x80008000, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_negfabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_negfabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_or_b32_e32 v3, 0x80008000, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -3253,69 +2451,37 @@ define <2 x half> @add_select_neg_fabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_neg_fabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_neg_fabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_neg_fabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_neg_fabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_neg_fabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_neg_fabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -3380,69 +2546,37 @@ define <2 x half> @add_select_fabs_neg_v2f16(<2 x i32> %c, <2 x half> %x, <2 x h ; GFX9-NEXT: v_pk_add_f16 v0, v0, v4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_fabs_neg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_fabs_neg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v3, 0x80008000, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_fabs_neg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_fabs_neg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_xor_b32_e32 v3, 0x80008000, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_fabs_neg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_fabs_neg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v3, 0x80008000, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v0, v4 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.y = fneg <2 x half> %y @@ -3501,63 +2635,34 @@ define <2 x half> @add_select_neg_negfabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_neg_negfabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v3 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v2.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v2.l, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_neg_negfabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_neg_negfabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v3 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v2.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v2.l, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_neg_negfabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_neg_negfabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v3 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v2.h, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v2.l, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_neg_negfabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v3 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v3, v2, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fneg.x = fneg <2 x half> %x %fabs.y = call <2 x half> @llvm.fabs.v2f16(<2 x half> %y) @@ -3617,63 +2722,34 @@ define <2 x half> @add_select_negfabs_neg_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: add_select_negfabs_neg_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v3.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v3.l, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: add_select_negfabs_neg_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: add_select_negfabs_neg_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v3.h, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v3.l, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: add_select_negfabs_neg_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: add_select_negfabs_neg_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v3.h, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v3.l, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: add_select_negfabs_neg_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v3 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0x7fff7fff, v2 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v6, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_add_f16 v0, v4, v0 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.fabs.x = fneg <2 x half> %fabs.x @@ -3735,61 +2811,33 @@ define <2 x half> @mul_select_negfabs_posk_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_mul_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: mul_select_negfabs_posk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4400, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4400, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: mul_select_negfabs_posk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4400, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4400, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: mul_select_negfabs_posk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4400, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4400, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: mul_select_negfabs_posk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4400, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4400, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: mul_select_negfabs_posk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4400, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4400, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: mul_select_negfabs_posk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4400, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4400, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.fabs.x = fneg <2 x half> %fabs.x @@ -3850,61 +2898,33 @@ define <2 x half> @mul_select_posk_negfabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_mul_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: mul_select_posk_negfabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4400, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4400, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: mul_select_posk_negfabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4400, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4400, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: mul_select_posk_negfabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4400, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4400, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: mul_select_posk_negfabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4400, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4400, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: mul_select_posk_negfabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4400, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4400, v2.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: mul_select_posk_negfabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4400, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4400, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.fabs.x = fneg <2 x half> %fabs.x @@ -3965,61 +2985,33 @@ define <2 x half> @mul_select_negfabs_negk_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_mul_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: mul_select_negfabs_negk_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xc400, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xc400, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: mul_select_negfabs_negk_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xc400, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc400, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: mul_select_negfabs_negk_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xc400, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xc400, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: mul_select_negfabs_negk_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xc400, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc400, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: mul_select_negfabs_negk_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xc400, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xc400, v0.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: mul_select_negfabs_negk_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xc400, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc400, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.fabs.x = fneg <2 x half> %fabs.x @@ -4080,61 +3072,33 @@ define <2 x half> @mul_select_negk_negfabs_v2f16(<2 x i32> %c, <2 x half> %x, <2 ; GFX9-NEXT: v_pk_mul_f16 v0, v0, v3 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: mul_select_negk_negfabs_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xc400, v2.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xc400, v2.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: mul_select_negk_negfabs_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xc400, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc400, v4, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: mul_select_negk_negfabs_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xc400, v2.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xc400, v2.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: mul_select_negk_negfabs_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xc400, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc400, v4, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: mul_select_negk_negfabs_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_ne_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0xc400, v2.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0xc400, v2.h, s0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: mul_select_negk_negfabs_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_or_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0xc400, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0xc400, v4, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_pk_mul_f16 v0, v0, v3 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fabs.x = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x) %fneg.fabs.x = fneg <2 x half> %fabs.x @@ -4171,115 +3135,63 @@ define <2 x half> @select_fneg_posk_src_add_v2f16(<2 x i32> %c, <2 x half> %x, < ; CI-NEXT: v_cndmask_b32_e32 v1, 2.0, v2, vcc ; CI-NEXT: s_setpc_b64 s[30:31] ; -; VI-SAFE-LABEL: select_fneg_posk_src_add_v2f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-SAFE-NEXT: v_mov_b32_e32 v1, 0x4400 -; VI-SAFE-NEXT: v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD -; VI-SAFE-NEXT: v_add_f16_e32 v2, 4.0, v2 -; VI-SAFE-NEXT: v_or_b32_e32 v1, v2, v1 -; VI-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; VI-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; VI-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-SAFE-LABEL: select_fneg_posk_src_add_v2f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-SAFE-NEXT: v_pk_add_f16 v1, v2, 4.0 op_sel_hi:[1,0] -; GFX9-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; GFX9-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_add_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v2, v2, 4.0 op_sel_hi:[1,0] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_add_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v2, v2, 4.0 op_sel_hi:[1,0] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; VI-NSZ-LABEL: select_fneg_posk_src_add_v2f16: -; VI-NSZ: ; %bb.0: -; VI-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-NSZ-NEXT: v_mov_b32_e32 v1, 0xc400 -; VI-NSZ-NEXT: v_sub_f16_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; VI-NSZ-NEXT: v_sub_f16_e32 v2, -4.0, v2 -; VI-NSZ-NEXT: v_mov_b32_e32 v3, 0x4000 -; VI-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-NSZ-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[4:5] -; VI-NSZ-NEXT: v_cndmask_b32_sdwa v1, v3, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-NSZ-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NSZ-LABEL: select_fneg_posk_src_add_v2f16: -; GFX9-NSZ: ; %bb.0: -; GFX9-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-NSZ-NEXT: v_pk_add_f16 v1, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX9-NSZ-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-NSZ-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-NSZ-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-NSZ-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-NSZ-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_add_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_add_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; VI-LABEL: select_fneg_posk_src_add_v2f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; VI-NEXT: v_mov_b32_e32 v1, 0x4400 +; VI-NEXT: v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD +; VI-NEXT: v_add_f16_e32 v2, 4.0, v2 +; VI-NEXT: v_or_b32_e32 v1, v2, v1 +; VI-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; VI-NEXT: v_mov_b32_e32 v2, 0x4000 +; VI-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; VI-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; VI-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-LABEL: select_fneg_posk_src_add_v2f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GFX9-NEXT: v_pk_add_f16 v1, v2, 4.0 op_sel_hi:[1,0] +; GFX9-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x4000 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; GFX9-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: select_fneg_posk_src_add_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v2, v2, 4.0 op_sel_hi:[1,0] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_add_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v2, v2, 4.0 op_sel_hi:[1,0] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %add = fadd <2 x half> %x, <half 4.0, half 4.0> %fneg = fneg <2 x half> %add @@ -4330,55 +3242,30 @@ define <2 x half> @select_fneg_posk_src_add_v2f16_nsz(<2 x i32> %c, <2 x half> % ; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_add_v2f16_nsz: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v0, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_add_v2f16_nsz: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_add_v2f16_nsz: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_add_v2f16_nsz: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: select_fneg_posk_src_add_v2f16_nsz: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_add_v2f16_nsz: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %add = fadd nsz <2 x half> %x, <half 4.0, half 4.0> %fneg = fneg <2 x half> %add @@ -4387,153 +3274,86 @@ define <2 x half> @select_fneg_posk_src_add_v2f16_nsz(<2 x i32> %c, <2 x half> % } define <2 x half> @select_fneg_posk_src_sub_v2f16(<2 x i32> %c, <2 x half> %x) { -; CI-SAFE-LABEL: select_fneg_posk_src_sub_v2f16: -; CI-SAFE: ; %bb.0: -; CI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; CI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; CI-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0 -; CI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; CI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; CI-SAFE-NEXT: v_add_f32_e32 v3, -4.0, v3 -; CI-SAFE-NEXT: v_add_f32_e32 v2, -4.0, v2 -; CI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; CI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; CI-SAFE-NEXT: v_lshlrev_b32_e32 v3, 16, v3 -; CI-SAFE-NEXT: v_or_b32_e32 v2, v2, v3 -; CI-SAFE-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; CI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; CI-SAFE-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; CI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; CI-SAFE-NEXT: v_cndmask_b32_e32 v0, 2.0, v3, vcc -; CI-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; CI-SAFE-NEXT: v_cndmask_b32_e32 v1, 2.0, v2, vcc -; CI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: select_fneg_posk_src_sub_v2f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-SAFE-NEXT: v_mov_b32_e32 v1, 0xc400 -; VI-SAFE-NEXT: v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD -; VI-SAFE-NEXT: v_add_f16_e32 v2, -4.0, v2 -; VI-SAFE-NEXT: v_or_b32_e32 v1, v2, v1 -; VI-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; VI-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; VI-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-SAFE-LABEL: select_fneg_posk_src_sub_v2f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-SAFE-NEXT: v_pk_add_f16 v1, v2, -4.0 op_sel_hi:[1,0] -; GFX9-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; GFX9-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_sub_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_sub_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; CI-NSZ-LABEL: select_fneg_posk_src_sub_v2f16: -; CI-NSZ: ; %bb.0: -; CI-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CI-NSZ-NEXT: v_cvt_f16_f32_e32 v2, v2 -; CI-NSZ-NEXT: v_cvt_f16_f32_e32 v3, v3 -; CI-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0 -; CI-NSZ-NEXT: v_cvt_f32_f16_e32 v2, v2 -; CI-NSZ-NEXT: v_cvt_f32_f16_e32 v3, v3 -; CI-NSZ-NEXT: v_sub_f32_e32 v2, 4.0, v2 -; CI-NSZ-NEXT: v_sub_f32_e32 v3, 4.0, v3 -; CI-NSZ-NEXT: v_cndmask_b32_e32 v0, 2.0, v2, vcc -; CI-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; CI-NSZ-NEXT: v_cndmask_b32_e32 v1, 2.0, v3, vcc -; CI-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; VI-NSZ-LABEL: select_fneg_posk_src_sub_v2f16: -; VI-NSZ: ; %bb.0: -; VI-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-NSZ-NEXT: v_mov_b32_e32 v1, 0x4400 -; VI-NSZ-NEXT: v_sub_f16_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; VI-NSZ-NEXT: v_sub_f16_e32 v2, 4.0, v2 -; VI-NSZ-NEXT: v_mov_b32_e32 v3, 0x4000 -; VI-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-NSZ-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[4:5] -; VI-NSZ-NEXT: v_cndmask_b32_sdwa v1, v3, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-NSZ-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NSZ-LABEL: select_fneg_posk_src_sub_v2f16: -; GFX9-NSZ: ; %bb.0: -; GFX9-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-NSZ-NEXT: v_pk_add_f16 v1, v2, 4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX9-NSZ-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-NSZ-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-NSZ-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-NSZ-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-NSZ-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_sub_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_add_f16 v0, v2, 4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_sub_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_add_f16 v2, v2, 4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; CI-LABEL: select_fneg_posk_src_sub_v2f16: +; CI: ; %bb.0: +; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; CI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; CI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0 +; CI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; CI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; CI-NEXT: v_add_f32_e32 v3, -4.0, v3 +; CI-NEXT: v_add_f32_e32 v2, -4.0, v2 +; CI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; CI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; CI-NEXT: v_or_b32_e32 v2, v2, v3 +; CI-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; CI-NEXT: v_cvt_f32_f16_e32 v3, v2 +; CI-NEXT: v_lshrrev_b32_e32 v2, 16, v2 +; CI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; CI-NEXT: v_cndmask_b32_e32 v0, 2.0, v3, vcc +; CI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; CI-NEXT: v_cndmask_b32_e32 v1, 2.0, v2, vcc +; CI-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: select_fneg_posk_src_sub_v2f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; VI-NEXT: v_mov_b32_e32 v1, 0xc400 +; VI-NEXT: v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD +; VI-NEXT: v_add_f16_e32 v2, -4.0, v2 +; VI-NEXT: v_or_b32_e32 v1, v2, v1 +; VI-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; VI-NEXT: v_mov_b32_e32 v2, 0x4000 +; VI-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; VI-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; VI-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-LABEL: select_fneg_posk_src_sub_v2f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GFX9-NEXT: v_pk_add_f16 v1, v2, -4.0 op_sel_hi:[1,0] +; GFX9-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x4000 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; GFX9-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: select_fneg_posk_src_sub_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_sub_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v2, v2, -4.0 op_sel_hi:[1,0] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %add = fsub <2 x half> %x, <half 4.0, half 4.0> %fneg = fneg <2 x half> %add @@ -4541,6 +3361,80 @@ define <2 x half> @select_fneg_posk_src_sub_v2f16(<2 x i32> %c, <2 x half> %x) { ret <2 x half> %select } +define <2 x half> @select_fneg_posk_src_sub_v2f16_nsz(<2 x i32> %c, <2 x half> %x) { +; CI-LABEL: select_fneg_posk_src_sub_v2f16_nsz: +; CI: ; %bb.0: +; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; CI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; CI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0 +; CI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; CI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; CI-NEXT: v_sub_f32_e32 v2, 4.0, v2 +; CI-NEXT: v_sub_f32_e32 v3, 4.0, v3 +; CI-NEXT: v_cndmask_b32_e32 v0, 2.0, v2, vcc +; CI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; CI-NEXT: v_cndmask_b32_e32 v1, 2.0, v3, vcc +; CI-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: select_fneg_posk_src_sub_v2f16_nsz: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; VI-NEXT: v_mov_b32_e32 v1, 0x4400 +; VI-NEXT: v_sub_f16_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; VI-NEXT: v_sub_f16_e32 v2, 4.0, v2 +; VI-NEXT: v_mov_b32_e32 v3, 0x4000 +; VI-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[4:5] +; VI-NEXT: v_cndmask_b32_sdwa v1, v3, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; VI-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-LABEL: select_fneg_posk_src_sub_v2f16_nsz: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GFX9-NEXT: v_pk_add_f16 v1, v2, 4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] +; GFX9-NEXT: v_mov_b32_e32 v2, 0x4000 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; GFX9-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: select_fneg_posk_src_sub_v2f16_nsz: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v2, 4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_sub_v2f16_nsz: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_add_f16 v2, v2, 4.0 op_sel_hi:[1,0] neg_lo:[1,0] neg_hi:[1,0] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] + %cmp = icmp eq <2 x i32> %c, zeroinitializer + %add = fsub <2 x half> %x, <half 4.0, half 4.0> + %fneg = fneg nsz <2 x half> %add + %select = select <2 x i1> %cmp, <2 x half> %fneg, <2 x half> <half 2.0, half 2.0> + ret <2 x half> %select +} + define <2 x half> @select_fneg_posk_src_mul_v2f16(<2 x i32> %c, <2 x half> %x) { ; CI-LABEL: select_fneg_posk_src_mul_v2f16: ; CI: ; %bb.0: @@ -4584,55 +3478,30 @@ define <2 x half> @select_fneg_posk_src_mul_v2f16(<2 x i32> %c, <2 x half> %x) { ; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_mul_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_mul_f16 v0, v2, -4.0 op_sel_hi:[1,0] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_mul_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_mul_f16 v2, v2, -4.0 op_sel_hi:[1,0] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_mul_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_mul_f16 v0, v2, -4.0 op_sel_hi:[1,0] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_mul_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_mul_f16 v2, v2, -4.0 op_sel_hi:[1,0] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: select_fneg_posk_src_mul_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_pk_mul_f16 v0, v2, -4.0 op_sel_hi:[1,0] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_mul_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_mul_f16 v2, v2, -4.0 op_sel_hi:[1,0] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %mul = fmul <2 x half> %x, <half 4.0, half 4.0> %fneg = fneg <2 x half> %mul @@ -4668,118 +3537,65 @@ define <2 x half> @select_fneg_posk_src_fma_v2f16(<2 x i32> %c, <2 x half> %x, < ; CI-NEXT: v_cndmask_b32_e32 v1, 2.0, v2, vcc ; CI-NEXT: s_setpc_b64 s[30:31] ; -; VI-SAFE-LABEL: select_fneg_posk_src_fma_v2f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v1, 16, v3 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-SAFE-NEXT: v_fma_f16 v1, v4, 4.0, v1 -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v1, 16, v1 -; VI-SAFE-NEXT: v_fma_f16 v2, v2, 4.0, v3 -; VI-SAFE-NEXT: v_or_b32_e32 v1, v2, v1 -; VI-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; VI-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; VI-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-SAFE-LABEL: select_fneg_posk_src_fma_v2f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-SAFE-NEXT: v_pk_fma_f16 v1, v2, 4.0, v3 op_sel_hi:[1,0,1] -; GFX9-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; GFX9-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_fma_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_fma_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; VI-NSZ-LABEL: select_fneg_posk_src_fma_v2f16: -; VI-NSZ: ; %bb.0: -; VI-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-NSZ-NEXT: v_lshrrev_b32_e32 v1, 16, v3 -; VI-NSZ-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-NSZ-NEXT: v_fma_f16 v1, v4, -4.0, -v1 -; VI-NSZ-NEXT: v_fma_f16 v2, v2, -4.0, -v3 -; VI-NSZ-NEXT: v_mov_b32_e32 v3, 0x4000 -; VI-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-NSZ-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[4:5] -; VI-NSZ-NEXT: v_cndmask_b32_sdwa v1, v3, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-NSZ-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NSZ-LABEL: select_fneg_posk_src_fma_v2f16: -; GFX9-NSZ: ; %bb.0: -; GFX9-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-NSZ-NEXT: v_pk_fma_f16 v1, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX9-NSZ-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-NSZ-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-NSZ-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-NSZ-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-NSZ-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_fma_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_fma_f16 v0, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_fma_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_fma_f16 v2, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; VI-LABEL: select_fneg_posk_src_fma_v2f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; VI-NEXT: v_lshrrev_b32_e32 v1, 16, v3 +; VI-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; VI-NEXT: v_fma_f16 v1, v4, 4.0, v1 +; VI-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; VI-NEXT: v_fma_f16 v2, v2, 4.0, v3 +; VI-NEXT: v_or_b32_e32 v1, v2, v1 +; VI-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; VI-NEXT: v_mov_b32_e32 v2, 0x4000 +; VI-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; VI-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; VI-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-LABEL: select_fneg_posk_src_fma_v2f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GFX9-NEXT: v_pk_fma_f16 v1, v2, 4.0, v3 op_sel_hi:[1,0,1] +; GFX9-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x4000 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; GFX9-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: select_fneg_posk_src_fma_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_fma_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fma = call <2 x half> @llvm.fma.v2f16(<2 x half> %x, <2 x half> <half 4.0, half 4.0>, <2 x half> %z) %fneg = fneg <2 x half> %fma @@ -4817,118 +3633,65 @@ define <2 x half> @select_fneg_posk_src_fmad_v2f16(<2 x i32> %c, <2 x half> %x, ; CI-NEXT: v_cndmask_b32_e32 v1, 2.0, v2, vcc ; CI-NEXT: s_setpc_b64 s[30:31] ; -; VI-SAFE-LABEL: select_fneg_posk_src_fmad_v2f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v1, 16, v3 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-SAFE-NEXT: v_fma_f16 v1, v4, 4.0, v1 -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v1, 16, v1 -; VI-SAFE-NEXT: v_fma_f16 v2, v2, 4.0, v3 -; VI-SAFE-NEXT: v_or_b32_e32 v1, v2, v1 -; VI-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; VI-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; VI-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-SAFE-LABEL: select_fneg_posk_src_fmad_v2f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-SAFE-NEXT: v_pk_fma_f16 v1, v2, 4.0, v3 op_sel_hi:[1,0,1] -; GFX9-SAFE-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 -; GFX9-SAFE-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-SAFE-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-SAFE-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-SAFE-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_fmad_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_fmad_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; VI-NSZ-LABEL: select_fneg_posk_src_fmad_v2f16: -; VI-NSZ: ; %bb.0: -; VI-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; VI-NSZ-NEXT: v_lshrrev_b32_e32 v1, 16, v3 -; VI-NSZ-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-NSZ-NEXT: v_fma_f16 v1, v4, -4.0, -v1 -; VI-NSZ-NEXT: v_fma_f16 v2, v2, -4.0, -v3 -; VI-NSZ-NEXT: v_mov_b32_e32 v3, 0x4000 -; VI-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; VI-NSZ-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[4:5] -; VI-NSZ-NEXT: v_cndmask_b32_sdwa v1, v3, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-NSZ-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NSZ-LABEL: select_fneg_posk_src_fmad_v2f16: -; GFX9-NSZ: ; %bb.0: -; GFX9-NSZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 -; GFX9-NSZ-NEXT: v_pk_fma_f16 v1, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX9-NSZ-NEXT: v_mov_b32_e32 v2, 0x4000 -; GFX9-NSZ-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 -; GFX9-NSZ-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] -; GFX9-NSZ-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX9-NSZ-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-NSZ-NEXT: v_perm_b32 v0, v1, v0, s4 -; GFX9-NSZ-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_fmad_v2f16: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_fma_f16 v0, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_fmad_v2f16: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_fma_f16 v2, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; VI-LABEL: select_fneg_posk_src_fmad_v2f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; VI-NEXT: v_lshrrev_b32_e32 v1, 16, v3 +; VI-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; VI-NEXT: v_fma_f16 v1, v4, 4.0, v1 +; VI-NEXT: v_lshlrev_b32_e32 v1, 16, v1 +; VI-NEXT: v_fma_f16 v2, v2, 4.0, v3 +; VI-NEXT: v_or_b32_e32 v1, v2, v1 +; VI-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; VI-NEXT: v_mov_b32_e32 v2, 0x4000 +; VI-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; VI-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; VI-NEXT: v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; GFX9-LABEL: select_fneg_posk_src_fmad_v2f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GFX9-NEXT: v_pk_fma_f16 v1, v2, 4.0, v3 op_sel_hi:[1,0,1] +; GFX9-NEXT: v_xor_b32_e32 v1, 0x80008000, v1 +; GFX9-NEXT: v_mov_b32_e32 v2, 0x4000 +; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, v1, s[4:5] +; GFX9-NEXT: v_cndmask_b32_sdwa v1, v2, v1, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: select_fneg_posk_src_fmad_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_xor_b32_e32 v0, 0x80008000, v2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_fmad_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_fma_f16 v2, v2, 4.0, v3 op_sel_hi:[1,0,1] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_xor_b32_e32 v2, 0x80008000, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fmad = call <2 x half> @llvm.fmuladd.v2f16(<2 x half> %x, <2 x half> <half 4.0, half 4.0>, <2 x half> %z) %fneg = fneg <2 x half> %fmad @@ -4986,55 +3749,30 @@ define <2 x half> @select_fneg_posk_src_fmad_v2f16_nsz(<2 x i32> %c, <2 x half> ; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4 ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX11-SAFE-TRUE16-LABEL: select_fneg_posk_src_fmad_v2f16_nsz: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-TRUE16-NEXT: v_pk_fma_f16 v0, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-SAFE-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: select_fneg_posk_src_fmad_v2f16_nsz: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_pk_fma_f16 v2, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-TRUE16-LABEL: select_fneg_posk_src_fmad_v2f16_nsz: -; GFX11-NSZ-TRUE16: ; %bb.0: -; GFX11-NSZ-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-TRUE16-NEXT: v_pk_fma_f16 v0, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-NSZ-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 -; GFX11-NSZ-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo -; GFX11-NSZ-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 -; GFX11-NSZ-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NSZ-FAKE16-LABEL: select_fneg_posk_src_fmad_v2f16_nsz: -; GFX11-NSZ-FAKE16: ; %bb.0: -; GFX11-NSZ-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NSZ-FAKE16-NEXT: v_pk_fma_f16 v2, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-NSZ-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 -; GFX11-NSZ-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo -; GFX11-NSZ-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NSZ-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 -; GFX11-NSZ-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX11-TRUE16-LABEL: select_fneg_posk_src_fmad_v2f16_nsz: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-TRUE16-NEXT: v_pk_fma_f16 v0, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e64 s0, 0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, 0x4000, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, 0x4000, v0.h, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: select_fneg_posk_src_fmad_v2f16_nsz: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_pk_fma_f16 v2, v2, -4.0, v3 op_sel_hi:[1,0,1] neg_lo:[0,0,1] neg_hi:[0,0,1] +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, 0x4000, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, 0x4000, v3, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v1, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = icmp eq <2 x i32> %c, zeroinitializer %fmad = call nsz <2 x half> @llvm.fmuladd.v2f16(<2 x half> %x, <2 x half> <half 4.0, half 4.0>, <2 x half> %z) %fneg = fneg <2 x half> %fmad @@ -5049,5 +3787,3 @@ declare <2 x half> @llvm.fmuladd.v2f16(<2 x half>, <2 x half>, <2 x half>) #0 attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: ; GFX11: {{.*}} -; GFX11-NSZ: {{.*}} -; GFX11-SAFE: {{.*}} diff --git a/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll b/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll index f1cadea..0868148 100644 --- a/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll +++ b/llvm/test/CodeGen/AMDGPU/simple-indirect-call-2.ll @@ -63,7 +63,7 @@ define amdgpu_kernel void @foo(ptr noundef %fp) { ; OW-NEXT: ret void ; ; CW-LABEL: define {{[^@]+}}@foo -; CW-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR1:[0-9]+]] { +; CW-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR0]] { ; CW-NEXT: entry: ; CW-NEXT: [[FP_ADDR:%.*]] = alloca ptr, align 8, addrspace(5) ; CW-NEXT: store ptr [[FP]], ptr addrspace(5) [[FP_ADDR]], align 8 @@ -84,7 +84,7 @@ define amdgpu_kernel void @foo(ptr noundef %fp) { ; CW-NEXT: ret void ; ; NO-LABEL: define {{[^@]+}}@foo -; NO-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR1:[0-9]+]] { +; NO-SAME: (ptr noundef [[FP:%.*]]) #[[ATTR0]] { ; NO-NEXT: entry: ; NO-NEXT: [[FP_ADDR:%.*]] = alloca ptr, align 8, addrspace(5) ; NO-NEXT: store ptr [[FP]], ptr addrspace(5) [[FP_ADDR]], align 8 @@ -101,14 +101,12 @@ entry: } ;. -; NO: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; NO: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; NO: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. -; OW: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; OW: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; OW: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. -; CW: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CW: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CW: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. ; NO: [[META0]] = !{ptr @bar1, ptr @bar2} ;. diff --git a/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll b/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll index 775d2f9..8fcaf5e 100644 --- a/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll +++ b/llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll @@ -58,7 +58,7 @@ define amdgpu_kernel void @test_simple_indirect_call() { ;. -; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; ATTRIBUTOR_GCN: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" } ;. ; ATTRIBUTOR_GCN: [[META0]] = !{i32 1, i32 5, i32 6, i32 10} diff --git a/llvm/test/CodeGen/AMDGPU/srem64.ll b/llvm/test/CodeGen/AMDGPU/srem64.ll index 465024a..33b0a5d 100644 --- a/llvm/test/CodeGen/AMDGPU/srem64.ll +++ b/llvm/test/CodeGen/AMDGPU/srem64.ll @@ -170,35 +170,38 @@ define amdgpu_kernel void @s_test_srem(ptr addrspace(1) %out, i64 %x, i64 %y) { ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[6:7], 0 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[2:3], 0 ; GCN-IR-NEXT: s_flbit_i32_b64 s10, s[6:7] -; GCN-IR-NEXT: s_flbit_i32_b64 s18, s[2:3] +; GCN-IR-NEXT: s_flbit_i32_b64 s16, s[2:3] ; GCN-IR-NEXT: s_or_b64 s[8:9], s[8:9], s[12:13] -; GCN-IR-NEXT: s_sub_u32 s12, s10, s18 +; GCN-IR-NEXT: s_sub_u32 s12, s10, s16 ; GCN-IR-NEXT: s_subb_u32 s13, 0, 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[14:15], s[12:13], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[12:13], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[18:19], s[12:13], 63 ; GCN-IR-NEXT: s_or_b64 s[14:15], s[8:9], s[14:15] ; GCN-IR-NEXT: s_and_b64 s[8:9], s[14:15], exec ; GCN-IR-NEXT: s_cselect_b32 s9, 0, s3 ; GCN-IR-NEXT: s_cselect_b32 s8, 0, s2 -; GCN-IR-NEXT: s_or_b64 s[14:15], s[14:15], s[16:17] +; GCN-IR-NEXT: s_or_b64 s[14:15], s[14:15], s[18:19] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[14:15] ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s14, s12, 1 -; GCN-IR-NEXT: s_addc_u32 s15, s13, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[14:15], 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 +; GCN-IR-NEXT: s_or_b32 s8, s8, s9 +; GCN-IR-NEXT: s_cmp_lg_u32 s8, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s13, 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s12, 63, s12 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[8:9] ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[2:3], s12 ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[12:13], s[2:3], s14 -; GCN-IR-NEXT: s_add_u32 s16, s6, -1 -; GCN-IR-NEXT: s_addc_u32 s17, s7, -1 +; GCN-IR-NEXT: s_add_u32 s14, s6, -1 +; GCN-IR-NEXT: s_addc_u32 s15, s7, -1 ; GCN-IR-NEXT: s_not_b64 s[4:5], s[10:11] -; GCN-IR-NEXT: s_add_u32 s10, s4, s18 -; GCN-IR-NEXT: s_addc_u32 s11, s5, 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], 0 +; GCN-IR-NEXT: s_add_u32 s16, s4, s16 +; GCN-IR-NEXT: s_addc_u32 s17, s5, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB0_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -206,19 +209,22 @@ define amdgpu_kernel void @s_test_srem(ptr addrspace(1) %out, i64 %x, i64 %y) { ; GCN-IR-NEXT: s_lshr_b32 s4, s9, 31 ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[8:9], 1 ; GCN-IR-NEXT: s_or_b64 s[12:13], s[12:13], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[8:9], s[14:15], s[8:9] -; GCN-IR-NEXT: s_sub_u32 s4, s16, s12 -; GCN-IR-NEXT: s_subb_u32 s4, s17, s13 -; GCN-IR-NEXT: s_ashr_i32 s14, s4, 31 -; GCN-IR-NEXT: s_mov_b32 s15, s14 -; GCN-IR-NEXT: s_and_b32 s4, s14, 1 -; GCN-IR-NEXT: s_and_b64 s[14:15], s[14:15], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s12, s12, s14 -; GCN-IR-NEXT: s_subb_u32 s13, s13, s15 -; GCN-IR-NEXT: s_add_u32 s10, s10, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s11, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[18:19], s[10:11], 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[10:11], s[8:9] +; GCN-IR-NEXT: s_sub_u32 s4, s14, s12 +; GCN-IR-NEXT: s_subb_u32 s4, s15, s13 +; GCN-IR-NEXT: s_ashr_i32 s10, s4, 31 +; GCN-IR-NEXT: s_mov_b32 s11, s10 +; GCN-IR-NEXT: s_and_b32 s4, s10, 1 +; GCN-IR-NEXT: s_and_b64 s[18:19], s[10:11], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s12, s12, s18 +; GCN-IR-NEXT: s_subb_u32 s13, s13, s19 +; GCN-IR-NEXT: s_add_u32 s16, s16, 1 +; GCN-IR-NEXT: s_cselect_b64 s[18:19], -1, 0 +; GCN-IR-NEXT: s_or_b32 s18, s18, s19 +; GCN-IR-NEXT: s_cmp_lg_u32 s18, 0 +; GCN-IR-NEXT: s_addc_u32 s17, s17, 0 +; GCN-IR-NEXT: s_cselect_b64 s[18:19], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[18:19] ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_3 ; GCN-IR-NEXT: .LBB0_4: ; %Flow7 @@ -373,12 +379,12 @@ define i64 @v_test_srem(i64 %x, i64 %y) { ; GCN-IR-LABEL: v_test_srem: ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-IR-NEXT: v_ashrrev_i32_e32 v14, 31, v1 -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v14 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v14 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v14 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v12 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 ; GCN-IR-NEXT: v_ashrrev_i32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v14, vcc +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v12, vcc ; GCN-IR-NEXT: v_xor_b32_e32 v2, v2, v4 ; GCN-IR-NEXT: v_xor_b32_e32 v3, v3, v4 ; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v2, v4 @@ -386,12 +392,12 @@ define i64 @v_test_srem(i64 %x, i64 %y) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v2 ; GCN-IR-NEXT: v_add_i32_e64 v4, s[6:7], 32, v4 ; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v3 -; GCN-IR-NEXT: v_min_u32_e32 v12, v4, v5 +; GCN-IR-NEXT: v_min_u32_e32 v10, v4, v5 ; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v0 ; GCN-IR-NEXT: v_add_i32_e64 v4, s[6:7], 32, v4 ; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v1 -; GCN-IR-NEXT: v_min_u32_e32 v13, v4, v5 -; GCN-IR-NEXT: v_sub_i32_e64 v4, s[6:7], v12, v13 +; GCN-IR-NEXT: v_min_u32_e32 v11, v4, v5 +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[6:7], v10, v11 ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_subb_u32_e64 v5, s[6:7], 0, 0, s[6:7] @@ -400,7 +406,7 @@ define i64 @v_test_srem(i64 %x, i64 %y) { ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] ; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[4:5] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 -; GCN-IR-NEXT: v_mov_b32_e32 v15, v14 +; GCN-IR-NEXT: v_mov_b32_e32 v13, v12 ; GCN-IR-NEXT: v_cndmask_b32_e64 v7, v1, 0, s[4:5] ; GCN-IR-NEXT: v_cndmask_b32_e64 v6, v0, 0, s[4:5] ; GCN-IR-NEXT: s_and_b64 s[4:5], s[6:7], vcc @@ -408,54 +414,53 @@ define i64 @v_test_srem(i64 %x, i64 %y) { ; GCN-IR-NEXT: s_cbranch_execz .LBB1_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v4 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 63, v4 -; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[0:1], v4 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB1_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v16, vcc, -1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v17, vcc, -1, v3, vcc -; GCN-IR-NEXT: v_not_b32_e32 v6, v12 -; GCN-IR-NEXT: v_lshr_b64 v[10:11], v[0:1], v8 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, v6, v13 -; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v9, s[4:5], -1, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v13, 0 +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_not_b32_e32 v6, v10 +; GCN-IR-NEXT: v_add_i32_e32 v16, vcc, v6, v11 +; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v8 +; GCN-IR-NEXT: v_addc_u32_e64 v17, s[8:9], -1, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 ; GCN-IR-NEXT: .LBB1_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[10:11], 1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v5 -; GCN-IR-NEXT: v_or_b32_e32 v10, v10, v6 +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v6 ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v16, v10 -; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, v17, v11, vcc -; GCN-IR-NEXT: v_or_b32_e32 v4, v12, v4 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v6 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v8 -; GCN-IR-NEXT: v_or_b32_e32 v5, v13, v5 -; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v12 -; GCN-IR-NEXT: v_and_b32_e32 v13, v12, v3 -; GCN-IR-NEXT: v_and_b32_e32 v12, v12, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[8:9] -; GCN-IR-NEXT: v_sub_i32_e64 v10, s[4:5], v10, v12 -; GCN-IR-NEXT: v_subb_u32_e64 v11, s[4:5], v11, v13, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v13, v7 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v12, v6 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v14, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, v15, v9, vcc +; GCN-IR-NEXT: v_or_b32_e32 v4, v10, v4 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_or_b32_e32 v5, v11, v5 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v3 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, v8, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v9, vcc, v9, v11, vcc +; GCN-IR-NEXT: v_add_i32_e32 v16, vcc, 1, v16 +; GCN-IR-NEXT: v_addc_u32_e32 v17, vcc, 0, v17, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB1_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB1_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB1_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 ; GCN-IR-NEXT: v_or_b32_e32 v7, v7, v5 ; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 @@ -469,10 +474,10 @@ define i64 @v_test_srem(i64 %x, i64 %y) { ; GCN-IR-NEXT: v_add_i32_e32 v3, vcc, v4, v3 ; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 ; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v14 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v15 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v14 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v15, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v13 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v13, vcc ; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = srem i64 %x, %y ret i64 %result @@ -1148,35 +1153,38 @@ define amdgpu_kernel void @s_test_srem33_64(ptr addrspace(1) %out, i64 %x, i64 % ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[2:3], s[8:9], 0 ; GCN-IR-NEXT: s_flbit_i32_b64 s12, s[8:9] ; GCN-IR-NEXT: s_or_b64 s[10:11], s[2:3], s[10:11] -; GCN-IR-NEXT: s_flbit_i32_b64 s20, s[6:7] -; GCN-IR-NEXT: s_sub_u32 s14, s12, s20 +; GCN-IR-NEXT: s_flbit_i32_b64 s18, s[6:7] +; GCN-IR-NEXT: s_sub_u32 s14, s12, s18 ; GCN-IR-NEXT: s_subb_u32 s15, 0, 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[16:17], s[14:15], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[18:19], s[14:15], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[20:21], s[14:15], 63 ; GCN-IR-NEXT: s_or_b64 s[16:17], s[10:11], s[16:17] ; GCN-IR-NEXT: s_and_b64 s[10:11], s[16:17], exec ; GCN-IR-NEXT: s_cselect_b32 s11, 0, s7 ; GCN-IR-NEXT: s_cselect_b32 s10, 0, s6 -; GCN-IR-NEXT: s_or_b64 s[16:17], s[16:17], s[18:19] +; GCN-IR-NEXT: s_or_b64 s[16:17], s[16:17], s[20:21] ; GCN-IR-NEXT: s_mov_b64 s[2:3], 0 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[16:17] ; GCN-IR-NEXT: s_cbranch_vccz .LBB8_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s16, s14, 1 -; GCN-IR-NEXT: s_addc_u32 s17, s15, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[10:11], s[16:17], 0 +; GCN-IR-NEXT: s_cselect_b64 s[10:11], -1, 0 +; GCN-IR-NEXT: s_or_b32 s10, s10, s11 +; GCN-IR-NEXT: s_cmp_lg_u32 s10, 0 +; GCN-IR-NEXT: s_addc_u32 s10, s15, 0 +; GCN-IR-NEXT: s_cselect_b64 s[10:11], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s14, 63, s14 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[10:11] ; GCN-IR-NEXT: s_lshl_b64 s[10:11], s[6:7], s14 ; GCN-IR-NEXT: s_cbranch_vccz .LBB8_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[14:15], s[6:7], s16 -; GCN-IR-NEXT: s_add_u32 s18, s8, -1 -; GCN-IR-NEXT: s_addc_u32 s19, s9, -1 +; GCN-IR-NEXT: s_add_u32 s16, s8, -1 +; GCN-IR-NEXT: s_addc_u32 s17, s9, -1 ; GCN-IR-NEXT: s_not_b64 s[2:3], s[12:13] -; GCN-IR-NEXT: s_add_u32 s12, s2, s20 -; GCN-IR-NEXT: s_addc_u32 s13, s3, 0 -; GCN-IR-NEXT: s_mov_b64 s[16:17], 0 +; GCN-IR-NEXT: s_add_u32 s18, s2, s18 +; GCN-IR-NEXT: s_addc_u32 s19, s3, 0 +; GCN-IR-NEXT: s_mov_b64 s[12:13], 0 ; GCN-IR-NEXT: s_mov_b32 s3, 0 ; GCN-IR-NEXT: .LBB8_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -1184,19 +1192,22 @@ define amdgpu_kernel void @s_test_srem33_64(ptr addrspace(1) %out, i64 %x, i64 % ; GCN-IR-NEXT: s_lshr_b32 s2, s11, 31 ; GCN-IR-NEXT: s_lshl_b64 s[10:11], s[10:11], 1 ; GCN-IR-NEXT: s_or_b64 s[14:15], s[14:15], s[2:3] -; GCN-IR-NEXT: s_or_b64 s[10:11], s[16:17], s[10:11] -; GCN-IR-NEXT: s_sub_u32 s2, s18, s14 -; GCN-IR-NEXT: s_subb_u32 s2, s19, s15 -; GCN-IR-NEXT: s_ashr_i32 s16, s2, 31 -; GCN-IR-NEXT: s_mov_b32 s17, s16 -; GCN-IR-NEXT: s_and_b32 s2, s16, 1 -; GCN-IR-NEXT: s_and_b64 s[16:17], s[16:17], s[8:9] -; GCN-IR-NEXT: s_sub_u32 s14, s14, s16 -; GCN-IR-NEXT: s_subb_u32 s15, s15, s17 -; GCN-IR-NEXT: s_add_u32 s12, s12, 1 -; GCN-IR-NEXT: s_addc_u32 s13, s13, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[20:21], s[12:13], 0 -; GCN-IR-NEXT: s_mov_b64 s[16:17], s[2:3] +; GCN-IR-NEXT: s_or_b64 s[10:11], s[12:13], s[10:11] +; GCN-IR-NEXT: s_sub_u32 s2, s16, s14 +; GCN-IR-NEXT: s_subb_u32 s2, s17, s15 +; GCN-IR-NEXT: s_ashr_i32 s12, s2, 31 +; GCN-IR-NEXT: s_mov_b32 s13, s12 +; GCN-IR-NEXT: s_and_b32 s2, s12, 1 +; GCN-IR-NEXT: s_and_b64 s[20:21], s[12:13], s[8:9] +; GCN-IR-NEXT: s_sub_u32 s14, s14, s20 +; GCN-IR-NEXT: s_subb_u32 s15, s15, s21 +; GCN-IR-NEXT: s_add_u32 s18, s18, 1 +; GCN-IR-NEXT: s_cselect_b64 s[20:21], -1, 0 +; GCN-IR-NEXT: s_or_b32 s20, s20, s21 +; GCN-IR-NEXT: s_cmp_lg_u32 s20, 0 +; GCN-IR-NEXT: s_addc_u32 s19, s19, 0 +; GCN-IR-NEXT: s_cselect_b64 s[20:21], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[12:13], s[2:3] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[20:21] ; GCN-IR-NEXT: s_cbranch_vccz .LBB8_3 ; GCN-IR-NEXT: .LBB8_4: ; %Flow7 @@ -1461,34 +1472,37 @@ define amdgpu_kernel void @s_test_srem_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_xor_b64 s[2:3], s[2:3], s[8:9] ; GCN-IR-NEXT: s_sub_u32 s4, s2, s8 ; GCN-IR-NEXT: s_subb_u32 s5, s3, s8 -; GCN-IR-NEXT: s_flbit_i32_b64 s12, s[4:5] -; GCN-IR-NEXT: s_add_u32 s2, s12, 0xffffffc5 +; GCN-IR-NEXT: s_flbit_i32_b64 s14, s[4:5] +; GCN-IR-NEXT: s_add_u32 s2, s14, 0xffffffc5 ; GCN-IR-NEXT: s_addc_u32 s3, 0, -1 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[4:5], 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[10:11], s[2:3], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[14:15], s[2:3], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[2:3], 63 ; GCN-IR-NEXT: s_or_b64 s[10:11], s[8:9], s[10:11] ; GCN-IR-NEXT: s_and_b64 s[8:9], s[10:11], exec ; GCN-IR-NEXT: s_cselect_b32 s8, 0, 24 -; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[14:15] +; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[12:13] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[10:11] ; GCN-IR-NEXT: s_mov_b32 s9, 0 ; GCN-IR-NEXT: s_cbranch_vccz .LBB10_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s8, s2, 1 -; GCN-IR-NEXT: s_addc_u32 s9, s3, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[10:11], s[8:9], 0 +; GCN-IR-NEXT: s_cselect_b64 s[10:11], -1, 0 +; GCN-IR-NEXT: s_or_b32 s9, s10, s11 +; GCN-IR-NEXT: s_cmp_lg_u32 s9, 0 +; GCN-IR-NEXT: s_addc_u32 s3, s3, 0 +; GCN-IR-NEXT: s_cselect_b64 s[10:11], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s2, 63, s2 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[10:11] ; GCN-IR-NEXT: s_lshl_b64 s[2:3], 24, s2 ; GCN-IR-NEXT: s_cbranch_vccz .LBB10_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[10:11], 24, s8 -; GCN-IR-NEXT: s_add_u32 s14, s4, -1 -; GCN-IR-NEXT: s_addc_u32 s15, s5, -1 -; GCN-IR-NEXT: s_sub_u32 s8, 58, s12 -; GCN-IR-NEXT: s_subb_u32 s9, 0, 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], 0 +; GCN-IR-NEXT: s_add_u32 s12, s4, -1 +; GCN-IR-NEXT: s_addc_u32 s13, s5, -1 +; GCN-IR-NEXT: s_sub_u32 s14, 58, s14 +; GCN-IR-NEXT: s_subb_u32 s15, 0, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 ; GCN-IR-NEXT: s_mov_b32 s7, 0 ; GCN-IR-NEXT: .LBB10_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -1496,19 +1510,22 @@ define amdgpu_kernel void @s_test_srem_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_lshr_b32 s6, s3, 31 ; GCN-IR-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[6:7] -; GCN-IR-NEXT: s_or_b64 s[2:3], s[12:13], s[2:3] -; GCN-IR-NEXT: s_sub_u32 s6, s14, s10 -; GCN-IR-NEXT: s_subb_u32 s6, s15, s11 -; GCN-IR-NEXT: s_ashr_i32 s12, s6, 31 -; GCN-IR-NEXT: s_mov_b32 s13, s12 -; GCN-IR-NEXT: s_and_b32 s6, s12, 1 -; GCN-IR-NEXT: s_and_b64 s[12:13], s[12:13], s[4:5] -; GCN-IR-NEXT: s_sub_u32 s10, s10, s12 -; GCN-IR-NEXT: s_subb_u32 s11, s11, s13 -; GCN-IR-NEXT: s_add_u32 s8, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s9, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[8:9], 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], s[6:7] +; GCN-IR-NEXT: s_or_b64 s[2:3], s[8:9], s[2:3] +; GCN-IR-NEXT: s_sub_u32 s6, s12, s10 +; GCN-IR-NEXT: s_subb_u32 s6, s13, s11 +; GCN-IR-NEXT: s_ashr_i32 s8, s6, 31 +; GCN-IR-NEXT: s_mov_b32 s9, s8 +; GCN-IR-NEXT: s_and_b32 s6, s8, 1 +; GCN-IR-NEXT: s_and_b64 s[16:17], s[8:9], s[4:5] +; GCN-IR-NEXT: s_sub_u32 s10, s10, s16 +; GCN-IR-NEXT: s_subb_u32 s11, s11, s17 +; GCN-IR-NEXT: s_add_u32 s14, s14, 1 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_or_b32 s16, s16, s17 +; GCN-IR-NEXT: s_cmp_lg_u32 s16, 0 +; GCN-IR-NEXT: s_addc_u32 s15, s15, 0 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], s[6:7] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[16:17] ; GCN-IR-NEXT: s_cbranch_vccz .LBB10_3 ; GCN-IR-NEXT: .LBB10_4: ; %Flow6 @@ -1647,9 +1664,9 @@ define i64 @v_test_srem_k_num_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 ; GCN-IR-NEXT: s_movk_i32 s6, 0xffc5 -; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v10 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v8 ; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] @@ -1663,53 +1680,52 @@ define i64 @v_test_srem_k_num_i64(i64 %x) { ; GCN-IR-NEXT: s_cbranch_execz .LBB11_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], 24, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB11_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_lshr_b64 v[8:9], 24, v6 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 58, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, 58, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], 24, v6 +; GCN-IR-NEXT: v_subb_u32_e64 v13, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB11_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v12, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v13, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v10, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v11, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 1, v12 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB11_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB11_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB11_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v3 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v2 @@ -1838,9 +1854,9 @@ define i64 @v_test_srem_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 ; GCN-IR-NEXT: s_movk_i32 s6, 0xffd0 -; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v10 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, s6, v8 ; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] @@ -1855,54 +1871,53 @@ define i64 @v_test_srem_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: s_cbranch_execz .LBB12_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc -; GCN-IR-NEXT: s_mov_b64 s[4:5], 0x8000 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0x8000 +; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[8:9], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[4:5], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[8:9] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[10:11], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[10:11] ; GCN-IR-NEXT: s_cbranch_execz .LBB12_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[4:5], v6 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 47, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, 47, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], s[8:9], v6 +; GCN-IR-NEXT: v_subb_u32_e64 v13, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB12_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v12, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v13, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v10, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v11, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 1, v12 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB12_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB12_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB12_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v3 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v2 @@ -1937,20 +1952,20 @@ define i64 @v_test_srem_pow2_k_den_i64(i64 %x) { ; GCN-IR-LABEL: v_test_srem_pow2_k_den_i64: ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v1 -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v12 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v12, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v10 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v10, vcc ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 48, v10 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 48, v8 ; GCN-IR-NEXT: v_subb_u32_e64 v3, s[4:5], 0, 0, s[4:5] ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[2:3] -; GCN-IR-NEXT: v_mov_b32_e32 v13, v12 +; GCN-IR-NEXT: v_mov_b32_e32 v11, v10 ; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] ; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[2:3] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 @@ -1961,51 +1976,50 @@ define i64 @v_test_srem_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: s_cbranch_execz .LBB13_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB13_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v6 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 0xffffffcf, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v7, s[4:5], 0, -1, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 0xffffffcf, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], v[0:1], v6 +; GCN-IR-NEXT: v_addc_u32_e64 v13, s[8:9], 0, -1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: s_movk_i32 s10, 0x7fff ; GCN-IR-NEXT: .LBB13_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s12, v8 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v9, vcc -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v10, 0x8000, v10 -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: v_subbrev_u32_e64 v9, s[4:5], 0, v9, s[4:5] -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s10, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 0x8000, v8 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subbrev_u32_e32 v7, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 1, v12 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB13_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB13_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB13_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v3 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v2 @@ -2014,10 +2028,10 @@ define i64 @v_test_srem_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[4:5], 15 ; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 ; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc -; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v12 -; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v13 -; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v12 -; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v13, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v10 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v11 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v11, vcc ; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = srem i64 %x, 32768 ret i64 %result diff --git a/llvm/test/CodeGen/AMDGPU/uaddo.ll b/llvm/test/CodeGen/AMDGPU/uaddo.ll index e1574dc..bb5918b2 100644 --- a/llvm/test/CodeGen/AMDGPU/uaddo.ll +++ b/llvm/test/CodeGen/AMDGPU/uaddo.ll @@ -14,15 +14,16 @@ define amdgpu_kernel void @s_uaddo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) ; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_add_u32 s0, s2, s8 -; SI-NEXT: v_mov_b32_e32 v0, s2 +; SI-NEXT: s_add_u32 s2, s2, s8 ; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_addc_u32 s1, s3, s9 +; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; SI-NEXT: s_or_b32 s0, s0, s1 +; SI-NEXT: s_cmp_lg_u32 s0, 0 +; SI-NEXT: s_addc_u32 s3, s3, s9 +; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; SI-NEXT: v_mov_b32_e32 v1, s3 -; SI-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[0:1] -; SI-NEXT: v_mov_b32_e32 v1, s1 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; SI-NEXT: v_add_i32_e32 v0, vcc, s0, v0 +; SI-NEXT: v_add_i32_e32 v0, vcc, s2, v0 ; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; SI-NEXT: s_endpgm @@ -33,15 +34,15 @@ define amdgpu_kernel void @s_uaddo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34 ; VI-NEXT: s_waitcnt lgkmcnt(0) ; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: s_add_u32 s0, s2, s4 -; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_add_u32 s2, s2, s4 ; VI-NEXT: v_mov_b32_e32 v1, s1 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: s_cmp_lg_u64 s[0:1], 0 +; VI-NEXT: s_addc_u32 s3, s3, s5 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[0:1] ; VI-NEXT: v_mov_b32_e32 v3, s3 -; VI-NEXT: s_addc_u32 s1, s3, s5 -; VI-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[2:3] -; VI-NEXT: v_mov_b32_e32 v3, s1 -; VI-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc -; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v2 +; VI-NEXT: v_add_u32_e32 v2, vcc, s2, v2 ; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] ; VI-NEXT: s_endpgm @@ -52,14 +53,14 @@ define amdgpu_kernel void @s_uaddo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 ; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mov_b32_e32 v0, s2 -; GFX9-NEXT: s_add_u32 s4, s2, s6 -; GFX9-NEXT: v_mov_b32_e32 v1, s3 -; GFX9-NEXT: s_addc_u32 s5, s3, s7 -; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[0:1] -; GFX9-NEXT: v_mov_b32_e32 v1, s5 -; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s4, v0 +; GFX9-NEXT: s_add_u32 s6, s2, s6 +; GFX9-NEXT: s_cselect_b64 s[4:5], -1, 0 +; GFX9-NEXT: s_cmp_lg_u64 s[4:5], 0 +; GFX9-NEXT: s_addc_u32 s4, s3, s7 +; GFX9-NEXT: s_cselect_b64 s[2:3], -1, 0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[2:3] +; GFX9-NEXT: v_mov_b32_e32 v1, s4 +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v0 ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc ; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX9-NEXT: s_endpgm @@ -71,12 +72,14 @@ define amdgpu_kernel void @s_uaddo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: s_add_u32 s4, s2, s6 -; GFX10-NEXT: s_addc_u32 s5, s3, s7 -; GFX10-NEXT: v_cmp_lt_u64_e64 s2, s[4:5], s[2:3] -; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s2 -; GFX10-NEXT: v_add_co_u32 v0, s2, s4, v0 -; GFX10-NEXT: v_add_co_ci_u32_e64 v1, s2, s5, 0, s2 +; GFX10-NEXT: s_add_u32 s2, s2, s6 +; GFX10-NEXT: s_cselect_b32 s4, -1, 0 +; GFX10-NEXT: s_cmp_lg_u32 s4, 0 +; GFX10-NEXT: s_addc_u32 s3, s3, s7 +; GFX10-NEXT: s_cselect_b32 s4, -1, 0 +; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s4 +; GFX10-NEXT: v_add_co_u32 v0, s2, s2, v0 +; GFX10-NEXT: v_add_co_ci_u32_e64 v1, s2, s3, 0, s2 ; GFX10-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX10-NEXT: s_endpgm ; @@ -87,14 +90,16 @@ define amdgpu_kernel void @s_uaddo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34 ; GFX11-NEXT: v_mov_b32_e32 v2, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_add_u32 s4, s2, s4 -; GFX11-NEXT: s_addc_u32 s5, s3, s5 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_cmp_lt_u64_e64 s2, s[4:5], s[2:3] -; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s2 +; GFX11-NEXT: s_add_u32 s2, s2, s4 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX11-NEXT: s_cmp_lg_u32 s4, 0 +; GFX11-NEXT: s_addc_u32 s3, s3, s5 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s4 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_add_co_u32 v0, s2, s4, v0 -; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, s5, 0, s2 +; GFX11-NEXT: v_add_co_u32 v0, s2, s2, v0 +; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, s3, 0, s2 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] ; GFX11-NEXT: s_endpgm %uadd = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b) @@ -436,21 +441,23 @@ define amdgpu_kernel void @s_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; SI-NEXT: s_mov_b32 s11, 0xf000 ; SI-NEXT: s_mov_b32 s10, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_add_u32 s6, s4, s6 -; SI-NEXT: v_mov_b32_e32 v0, s4 -; SI-NEXT: s_addc_u32 s7, s5, s7 -; SI-NEXT: v_mov_b32_e32 v1, s5 -; SI-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[0:1] -; SI-NEXT: v_mov_b32_e32 v2, s6 +; SI-NEXT: s_add_u32 s4, s4, s6 +; SI-NEXT: s_cselect_b64 s[12:13], -1, 0 +; SI-NEXT: s_or_b32 s6, s12, s13 +; SI-NEXT: s_cmp_lg_u32 s6, 0 +; SI-NEXT: s_addc_u32 s5, s5, s7 ; SI-NEXT: s_mov_b32 s8, s0 ; SI-NEXT: s_mov_b32 s9, s1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: s_cselect_b64 s[4:5], -1, 0 ; SI-NEXT: s_mov_b32 s0, s2 ; SI-NEXT: s_mov_b32 s1, s3 ; SI-NEXT: s_mov_b32 s2, s10 ; SI-NEXT: s_mov_b32 s3, s11 -; SI-NEXT: v_mov_b32_e32 v3, s7 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; SI-NEXT: buffer_store_dwordx2 v[2:3], off, s[8:11], 0 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; SI-NEXT: s_waitcnt expcnt(0) +; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] ; SI-NEXT: buffer_store_byte v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; @@ -458,37 +465,37 @@ define amdgpu_kernel void @s_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_add_u32 s2, s4, s6 ; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: s_add_u32 s0, s4, s6 -; VI-NEXT: v_mov_b32_e32 v4, s4 ; VI-NEXT: v_mov_b32_e32 v1, s1 -; VI-NEXT: s_addc_u32 s1, s5, s7 -; VI-NEXT: v_mov_b32_e32 v5, s5 -; VI-NEXT: v_mov_b32_e32 v7, s1 -; VI-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[4:5] -; VI-NEXT: v_mov_b32_e32 v6, s0 -; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: s_cmp_lg_u64 s[0:1], 0 +; VI-NEXT: s_addc_u32 s0, s5, s7 +; VI-NEXT: v_mov_b32_e32 v4, s2 +; VI-NEXT: v_mov_b32_e32 v5, s0 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 ; VI-NEXT: v_mov_b32_e32 v3, s3 -; VI-NEXT: flat_store_dwordx2 v[0:1], v[6:7] -; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc +; VI-NEXT: flat_store_dwordx2 v[0:1], v[4:5] +; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; VI-NEXT: flat_store_byte v[2:3], v0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: s_uaddo_i64: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v4, 0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_add_u32 s0, s12, s14 -; GFX9-NEXT: v_mov_b32_e32 v0, s12 -; GFX9-NEXT: v_mov_b32_e32 v1, s13 -; GFX9-NEXT: s_addc_u32 s1, s13, s15 -; GFX9-NEXT: v_mov_b32_e32 v3, s1 -; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[0:1], v[0:1] -; GFX9-NEXT: v_mov_b32_e32 v2, s0 -; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GFX9-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] -; GFX9-NEXT: global_store_byte v4, v0, s[10:11] +; GFX9-NEXT: s_add_u32 s2, s12, s14 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: s_cmp_lg_u64 s[0:1], 0 +; GFX9-NEXT: s_addc_u32 s0, s13, s15 +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_mov_b32_e32 v1, s0 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[0:1] +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] +; GFX9-NEXT: global_store_byte v2, v3, s[10:11] ; GFX9-NEXT: s_endpgm ; ; GFX10-LABEL: s_uaddo_i64: @@ -497,10 +504,12 @@ define amdgpu_kernel void @s_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_add_u32 s0, s12, s14 -; GFX10-NEXT: s_addc_u32 s1, s13, s15 +; GFX10-NEXT: s_cselect_b32 s1, -1, 0 ; GFX10-NEXT: v_mov_b32_e32 v0, s0 +; GFX10-NEXT: s_cmp_lg_u32 s1, 0 +; GFX10-NEXT: s_addc_u32 s1, s13, s15 +; GFX10-NEXT: s_cselect_b32 s0, -1, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, s1 -; GFX10-NEXT: v_cmp_lt_u64_e64 s0, s[0:1], s[12:13] ; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s0 ; GFX10-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] ; GFX10-NEXT: global_store_byte v2, v3, s[10:11] @@ -510,12 +519,13 @@ define amdgpu_kernel void @s_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX11: ; %bb.0: ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_add_u32 s6, s4, s6 -; GFX11-NEXT: s_addc_u32 s7, s5, s7 -; GFX11-NEXT: v_mov_b32_e32 v0, s6 -; GFX11-NEXT: v_cmp_lt_u64_e64 s4, s[6:7], s[4:5] -; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s7 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-NEXT: s_add_u32 s4, s4, s6 +; GFX11-NEXT: s_cselect_b32 s6, -1, 0 +; GFX11-NEXT: v_mov_b32_e32 v0, s4 +; GFX11-NEXT: s_cmp_lg_u32 s6, 0 +; GFX11-NEXT: s_addc_u32 s5, s5, s7 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s5 ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] @@ -551,10 +561,10 @@ define amdgpu_kernel void @v_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; SI-NEXT: s_mov_b32 s4, s2 ; SI-NEXT: s_mov_b32 s5, s3 ; SI-NEXT: s_waitcnt vmcnt(0) -; SI-NEXT: v_add_i32_e32 v2, vcc, v0, v2 -; SI-NEXT: v_addc_u32_e32 v3, vcc, v1, v3, vcc -; SI-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; SI-NEXT: buffer_store_dwordx2 v[2:3], off, s[8:11], 0 +; SI-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; SI-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; SI-NEXT: s_waitcnt expcnt(0) ; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; SI-NEXT: buffer_store_byte v0, off, s[4:7], 0 ; SI-NEXT: s_endpgm @@ -574,10 +584,9 @@ define amdgpu_kernel void @v_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; VI-NEXT: v_mov_b32_e32 v6, s2 ; VI-NEXT: v_mov_b32_e32 v7, s3 ; VI-NEXT: s_waitcnt vmcnt(0) -; VI-NEXT: v_add_u32_e32 v2, vcc, v0, v2 -; VI-NEXT: v_addc_u32_e32 v3, vcc, v1, v3, vcc -; VI-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; VI-NEXT: flat_store_dwordx2 v[4:5], v[2:3] +; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v2 +; VI-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc +; VI-NEXT: flat_store_dwordx2 v[4:5], v[0:1] ; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; VI-NEXT: flat_store_byte v[6:7], v0 ; VI-NEXT: s_endpgm @@ -590,10 +599,9 @@ define amdgpu_kernel void @v_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX9-NEXT: global_load_dwordx2 v[0:1], v4, s[12:13] ; GFX9-NEXT: global_load_dwordx2 v[2:3], v4, s[14:15] ; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v0, v2 -; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, v1, v3, vcc -; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; GFX9-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2 +; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc +; GFX9-NEXT: global_store_dwordx2 v4, v[0:1], s[8:9] ; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; GFX9-NEXT: global_store_byte v4, v0, s[10:11] ; GFX9-NEXT: s_endpgm @@ -607,12 +615,11 @@ define amdgpu_kernel void @v_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX10-NEXT: global_load_dwordx2 v[0:1], v4, s[12:13] ; GFX10-NEXT: global_load_dwordx2 v[2:3], v4, s[14:15] ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_add_co_u32 v2, vcc_lo, v0, v2 -; GFX10-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, v1, v3, vcc_lo -; GFX10-NEXT: v_cmp_lt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX10-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] -; GFX10-NEXT: global_store_byte v4, v0, s[10:11] +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo +; GFX10-NEXT: global_store_dwordx2 v4, v[0:1], s[8:9] +; GFX10-NEXT: global_store_byte v4, v2, s[10:11] ; GFX10-NEXT: s_endpgm ; ; GFX11-LABEL: v_uaddo_i64: @@ -624,14 +631,12 @@ define amdgpu_kernel void @v_uaddo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX11-NEXT: global_load_b64 v[0:1], v4, s[4:5] ; GFX11-NEXT: global_load_b64 v[2:3], v4, s[6:7] ; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v0, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, v1, v3, vcc_lo -; GFX11-NEXT: v_cmp_lt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2 +; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo ; GFX11-NEXT: s_clause 0x1 -; GFX11-NEXT: global_store_b64 v4, v[2:3], s[0:1] -; GFX11-NEXT: global_store_b8 v4, v0, s[2:3] +; GFX11-NEXT: global_store_b64 v4, v[0:1], s[0:1] +; GFX11-NEXT: global_store_b8 v4, v2, s[2:3] ; GFX11-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %tid.ext = sext i32 %tid to i64 diff --git a/llvm/test/CodeGen/AMDGPU/uaddsat.ll b/llvm/test/CodeGen/AMDGPU/uaddsat.ll index 9230174..7f89581 100644 --- a/llvm/test/CodeGen/AMDGPU/uaddsat.ll +++ b/llvm/test/CodeGen/AMDGPU/uaddsat.ll @@ -693,52 +693,47 @@ define i64 @v_uaddsat_i64(i64 %lhs, i64 %rhs) { ; GFX6-LABEL: v_uaddsat_i64: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-NEXT: v_add_i32_e32 v2, vcc, v0, v2 -; GFX6-NEXT: v_addc_u32_e32 v3, vcc, v1, v3, vcc -; GFX6-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; GFX6-NEXT: v_cndmask_b32_e64 v0, v2, -1, vcc -; GFX6-NEXT: v_cndmask_b32_e64 v1, v3, -1, vcc +; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GFX6-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, -1, vcc +; GFX6-NEXT: v_cndmask_b32_e64 v1, v1, -1, vcc ; GFX6-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: v_uaddsat_i64: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_add_u32_e32 v2, vcc, v0, v2 -; GFX8-NEXT: v_addc_u32_e32 v3, vcc, v1, v3, vcc -; GFX8-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; GFX8-NEXT: v_cndmask_b32_e64 v0, v2, -1, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v1, v3, -1, vcc +; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v2 +; GFX8-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc +; GFX8-NEXT: v_cndmask_b32_e64 v0, v0, -1, vcc +; GFX8-NEXT: v_cndmask_b32_e64 v1, v1, -1, vcc ; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX9-LABEL: v_uaddsat_i64: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v0, v2 -; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, v1, v3, vcc -; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, v[2:3], v[0:1] -; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, -1, vcc -; GFX9-NEXT: v_cndmask_b32_e64 v1, v3, -1, vcc +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2 +; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v0, v0, -1, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v1, v1, -1, vcc ; GFX9-NEXT: s_setpc_b64 s[30:31] ; ; GFX10-LABEL: v_uaddsat_i64: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_add_co_u32 v2, vcc_lo, v0, v2 -; GFX10-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, v1, v3, vcc_lo -; GFX10-NEXT: v_cmp_lt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX10-NEXT: v_cndmask_b32_e64 v0, v2, -1, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v1, v3, -1, vcc_lo +; GFX10-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2 +; GFX10-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX10-NEXT: v_cndmask_b32_e64 v0, v0, -1, vcc_lo +; GFX10-NEXT: v_cndmask_b32_e64 v1, v1, -1, vcc_lo ; GFX10-NEXT: s_setpc_b64 s[30:31] ; ; GFX11-LABEL: v_uaddsat_i64: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_add_co_u32 v2, vcc_lo, v0, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_add_co_ci_u32_e64 v3, null, v1, v3, vcc_lo -; GFX11-NEXT: v_cmp_lt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX11-NEXT: v_cndmask_b32_e64 v0, v2, -1, vcc_lo -; GFX11-NEXT: v_cndmask_b32_e64 v1, v3, -1, vcc_lo +; GFX11-NEXT: v_add_co_u32 v0, vcc_lo, v0, v2 +; GFX11-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-NEXT: v_cndmask_b32_e64 v0, v0, -1, vcc_lo +; GFX11-NEXT: v_cndmask_b32_e64 v1, v1, -1, vcc_lo ; GFX11-NEXT: s_setpc_b64 s[30:31] %result = call i64 @llvm.uadd.sat.i64(i64 %lhs, i64 %rhs) ret i64 %result diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll index 1ed04f8..41199b0 100644 --- a/llvm/test/CodeGen/AMDGPU/udiv64.ll +++ b/llvm/test/CodeGen/AMDGPU/udiv64.ll @@ -146,8 +146,11 @@ define amdgpu_kernel void @s_test_udiv_i64(ptr addrspace(1) %out, i64 %x, i64 %y ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s14, s12, 1 -; GCN-IR-NEXT: s_addc_u32 s15, s13, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[14:15], 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 +; GCN-IR-NEXT: s_or_b32 s8, s8, s9 +; GCN-IR-NEXT: s_cmp_lg_u32 s8, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s13, 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s12, 63, s12 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[8:9] ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[2:3], s12 @@ -157,9 +160,9 @@ define amdgpu_kernel void @s_test_udiv_i64(ptr addrspace(1) %out, i64 %x, i64 %y ; GCN-IR-NEXT: s_add_u32 s14, s6, -1 ; GCN-IR-NEXT: s_addc_u32 s15, s7, -1 ; GCN-IR-NEXT: s_not_b64 s[2:3], s[10:11] -; GCN-IR-NEXT: s_add_u32 s2, s2, s16 -; GCN-IR-NEXT: s_addc_u32 s3, s3, 0 -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 +; GCN-IR-NEXT: s_add_u32 s10, s2, s16 +; GCN-IR-NEXT: s_addc_u32 s11, s3, 0 +; GCN-IR-NEXT: s_mov_b64 s[2:3], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB0_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -167,19 +170,22 @@ define amdgpu_kernel void @s_test_udiv_i64(ptr addrspace(1) %out, i64 %x, i64 %y ; GCN-IR-NEXT: s_lshr_b32 s4, s9, 31 ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[8:9], 1 ; GCN-IR-NEXT: s_or_b64 s[12:13], s[12:13], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[8:9], s[10:11], s[8:9] -; GCN-IR-NEXT: s_sub_u32 s4, s14, s12 -; GCN-IR-NEXT: s_subb_u32 s4, s15, s13 -; GCN-IR-NEXT: s_ashr_i32 s10, s4, 31 -; GCN-IR-NEXT: s_mov_b32 s11, s10 -; GCN-IR-NEXT: s_and_b32 s4, s10, 1 -; GCN-IR-NEXT: s_and_b64 s[10:11], s[10:11], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s12, s12, s10 -; GCN-IR-NEXT: s_subb_u32 s13, s13, s11 -; GCN-IR-NEXT: s_add_u32 s2, s2, 1 -; GCN-IR-NEXT: s_addc_u32 s3, s3, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[2:3], 0 -; GCN-IR-NEXT: s_mov_b64 s[10:11], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[2:3], s[8:9] +; GCN-IR-NEXT: s_sub_u32 s2, s14, s12 +; GCN-IR-NEXT: s_subb_u32 s2, s15, s13 +; GCN-IR-NEXT: s_ashr_i32 s2, s2, 31 +; GCN-IR-NEXT: s_mov_b32 s3, s2 +; GCN-IR-NEXT: s_and_b32 s4, s2, 1 +; GCN-IR-NEXT: s_and_b64 s[16:17], s[2:3], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s12, s12, s16 +; GCN-IR-NEXT: s_subb_u32 s13, s13, s17 +; GCN-IR-NEXT: s_add_u32 s10, s10, 1 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_or_b32 s16, s16, s17 +; GCN-IR-NEXT: s_cmp_lg_u32 s16, 0 +; GCN-IR-NEXT: s_addc_u32 s11, s11, 0 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[2:3], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[16:17] ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_3 ; GCN-IR-NEXT: .LBB0_4: ; %Flow7 @@ -313,19 +319,19 @@ define i64 @v_test_udiv_i64(i64 %x, i64 %y) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v2 ; GCN-IR-NEXT: v_add_i32_e64 v4, s[6:7], 32, v4 ; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v3 -; GCN-IR-NEXT: v_min_u32_e32 v14, v4, v5 +; GCN-IR-NEXT: v_min_u32_e32 v8, v4, v5 ; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v0 ; GCN-IR-NEXT: v_add_i32_e64 v4, s[6:7], 32, v4 ; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v1 -; GCN-IR-NEXT: v_min_u32_e32 v15, v4, v5 -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[6:7], v14, v15 +; GCN-IR-NEXT: v_min_u32_e32 v9, v4, v5 +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[6:7], v8, v9 ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[6:7], 0, 0, s[6:7] -; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[6:7], 63, v[8:9] +; GCN-IR-NEXT: v_subb_u32_e64 v7, s[6:7], 0, 0, s[6:7] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[6:7], 63, v[6:7] ; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[8:9] +; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[6:7] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 ; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v1, 0, s[4:5] ; GCN-IR-NEXT: v_cndmask_b32_e64 v5, v0, 0, s[4:5] @@ -333,55 +339,54 @@ define i64 @v_test_udiv_i64(i64 %x, i64 %y) { ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB1_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, 1, v8 -; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, 0, v9, vcc -; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 63, v8 -; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[10:11] +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, 1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v4, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 63, v6 ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[0:1], v4 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB1_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v2 -; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v10 -; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v3, vcc -; GCN-IR-NEXT: v_not_b32_e32 v0, v14 -; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v15 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v1, s[4:5], -1, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_lshr_b64 v[0:1], v[0:1], v10 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_not_b32_e32 v6, v8 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, v6, v9 +; GCN-IR-NEXT: v_addc_u32_e64 v13, s[8:9], -1, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 ; GCN-IR-NEXT: .LBB1_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[0:1], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v5 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v6 +; GCN-IR-NEXT: v_or_b32_e32 v0, v0, v6 ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v12, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, v13, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v4, v10, v4 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 -; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v0 -; GCN-IR-NEXT: v_or_b32_e32 v5, v11, v5 -; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v3 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v10, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, v11, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v4, v8, v4 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v6 +; GCN-IR-NEXT: v_or_b32_e32 v5, v9, v5 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v3 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 1, v12 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v7 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB1_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB1_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB1_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 ; GCN-IR-NEXT: v_or_b32_e32 v4, v7, v1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v6, v0 @@ -923,34 +928,37 @@ define amdgpu_kernel void @s_test_udiv_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GCN-IR-NEXT: s_mov_b64 s[4:5], 0 ; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) -; GCN-IR-NEXT: s_flbit_i32_b64 s12, s[2:3] -; GCN-IR-NEXT: s_add_u32 s8, s12, 0xffffffc5 +; GCN-IR-NEXT: s_flbit_i32_b64 s14, s[2:3] +; GCN-IR-NEXT: s_add_u32 s8, s14, 0xffffffc5 ; GCN-IR-NEXT: s_addc_u32 s9, 0, -1 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[2:3], 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[10:11], s[8:9], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[14:15], s[8:9], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[8:9], 63 ; GCN-IR-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] ; GCN-IR-NEXT: s_and_b64 s[6:7], s[10:11], exec ; GCN-IR-NEXT: s_cselect_b32 s6, 0, 24 -; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[14:15] +; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[12:13] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[10:11] ; GCN-IR-NEXT: s_mov_b32 s7, 0 ; GCN-IR-NEXT: s_cbranch_vccz .LBB8_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s10, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[10:11], 0 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 +; GCN-IR-NEXT: s_or_b32 s6, s6, s7 +; GCN-IR-NEXT: s_cmp_lg_u32 s6, 0 +; GCN-IR-NEXT: s_addc_u32 s6, s9, 0 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s8, 63, s8 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[6:7] ; GCN-IR-NEXT: s_lshl_b64 s[6:7], 24, s8 ; GCN-IR-NEXT: s_cbranch_vccz .LBB8_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[10:11], 24, s10 -; GCN-IR-NEXT: s_add_u32 s14, s2, -1 -; GCN-IR-NEXT: s_addc_u32 s15, s3, -1 -; GCN-IR-NEXT: s_sub_u32 s8, 58, s12 -; GCN-IR-NEXT: s_subb_u32 s9, 0, 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], 0 +; GCN-IR-NEXT: s_add_u32 s12, s2, -1 +; GCN-IR-NEXT: s_addc_u32 s13, s3, -1 +; GCN-IR-NEXT: s_sub_u32 s14, 58, s14 +; GCN-IR-NEXT: s_subb_u32 s15, 0, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB8_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -958,19 +966,22 @@ define amdgpu_kernel void @s_test_udiv_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_lshr_b32 s4, s7, 31 ; GCN-IR-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 ; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[6:7], s[12:13], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s4, s14, s10 -; GCN-IR-NEXT: s_subb_u32 s4, s15, s11 -; GCN-IR-NEXT: s_ashr_i32 s12, s4, 31 -; GCN-IR-NEXT: s_mov_b32 s13, s12 -; GCN-IR-NEXT: s_and_b32 s4, s12, 1 -; GCN-IR-NEXT: s_and_b64 s[12:13], s[12:13], s[2:3] -; GCN-IR-NEXT: s_sub_u32 s10, s10, s12 -; GCN-IR-NEXT: s_subb_u32 s11, s11, s13 -; GCN-IR-NEXT: s_add_u32 s8, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s9, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[8:9], 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s4, s12, s10 +; GCN-IR-NEXT: s_subb_u32 s4, s13, s11 +; GCN-IR-NEXT: s_ashr_i32 s8, s4, 31 +; GCN-IR-NEXT: s_mov_b32 s9, s8 +; GCN-IR-NEXT: s_and_b32 s4, s8, 1 +; GCN-IR-NEXT: s_and_b64 s[16:17], s[8:9], s[2:3] +; GCN-IR-NEXT: s_sub_u32 s10, s10, s16 +; GCN-IR-NEXT: s_subb_u32 s11, s11, s17 +; GCN-IR-NEXT: s_add_u32 s14, s14, 1 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_or_b32 s16, s16, s17 +; GCN-IR-NEXT: s_cmp_lg_u32 s16, 0 +; GCN-IR-NEXT: s_addc_u32 s15, s15, 0 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[16:17] ; GCN-IR-NEXT: s_cbranch_vccz .LBB8_3 ; GCN-IR-NEXT: .LBB8_4: ; %Flow6 @@ -1094,12 +1105,12 @@ define i64 @v_test_udiv_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 0xffffffd0, v10 -; GCN-IR-NEXT: v_addc_u32_e64 v7, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 0xffffffd0, v8 +; GCN-IR-NEXT: v_addc_u32_e64 v5, s[6:7], 0, -1, vcc ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] -; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[6:7] -; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[6:7] +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] ; GCN-IR-NEXT: v_mov_b32_e32 v3, 0x8000 ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc ; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v3, 0, s[4:5] @@ -1109,55 +1120,54 @@ define i64 @v_test_udiv_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB9_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v6 -; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v6 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v7, vcc -; GCN-IR-NEXT: s_mov_b64 s[4:5], 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v2, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0x8000 +; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[8:9], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] -; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[4:5], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[8:9] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[10:11], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[10:11] ; GCN-IR-NEXT: s_cbranch_execz .LBB9_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[4:5], v8 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 47, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, 47, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], s[8:9], v6 +; GCN-IR-NEXT: v_subb_u32_e64 v13, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB9_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v12, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v13, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v10, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v11, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 1, v12 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB9_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB9_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB9_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v2, v5, v1 ; GCN-IR-NEXT: v_or_b32_e32 v3, v4, v0 @@ -1184,13 +1194,13 @@ define i64 @v_test_udiv_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], 48, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_min_u32_e32 v6, v2, v3 +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 48, v6 +; GCN-IR-NEXT: v_subb_u32_e64 v5, s[4:5], 0, 0, s[4:5] ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] -; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[6:7] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[4:5] ; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[4:5] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 ; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v1, 0, s[4:5] ; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v0, 0, s[4:5] @@ -1198,52 +1208,51 @@ define i64 @v_test_udiv_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB10_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v6 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v6 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] +; GCN-IR-NEXT: v_add_i32_e32 v7, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v2, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB10_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_lshr_b64 v[6:7], v[0:1], v8 -; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 0xffffffcf, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v1, s[4:5], 0, -1, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 0xffffffcf, v6 +; GCN-IR-NEXT: v_lshr_b64 v[0:1], v[0:1], v7 +; GCN-IR-NEXT: v_addc_u32_e64 v9, s[8:9], 0, -1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: s_movk_i32 s10, 0x7fff ; GCN-IR-NEXT: .LBB10_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[0:1], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v0, v0, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v0 -; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 -; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 -; GCN-IR-NEXT: v_and_b32_e32 v8, 0x8000, v8 -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] -; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v6, v8 -; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 -; GCN-IR-NEXT: v_subbrev_u32_e64 v7, s[4:5], 0, v7, s[4:5] -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s10, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v6, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 0x8000, v6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v6 +; GCN-IR-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v8 +; GCN-IR-NEXT: v_or_b32_e32 v3, v7, v3 +; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v7, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v6, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB10_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB10_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB10_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v2, v5, v1 ; GCN-IR-NEXT: v_or_b32_e32 v3, v4, v0 @@ -1290,52 +1299,58 @@ define amdgpu_kernel void @s_test_udiv_k_den_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) -; GCN-IR-NEXT: s_flbit_i32_b64 s12, s[2:3] -; GCN-IR-NEXT: s_sub_u32 s8, 59, s12 +; GCN-IR-NEXT: s_flbit_i32_b64 s10, s[2:3] +; GCN-IR-NEXT: s_sub_u32 s8, 59, s10 ; GCN-IR-NEXT: s_subb_u32 s9, 0, 0 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], s[2:3], 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[6:7], s[8:9], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[10:11], s[8:9], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[8:9], 63 ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] ; GCN-IR-NEXT: s_and_b64 s[6:7], s[4:5], exec ; GCN-IR-NEXT: s_cselect_b32 s7, 0, s3 ; GCN-IR-NEXT: s_cselect_b32 s6, 0, s2 -; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[10:11] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[12:13] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[4:5] ; GCN-IR-NEXT: s_mov_b64 s[4:5], 0 ; GCN-IR-NEXT: s_cbranch_vccz .LBB11_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: s_add_u32 s10, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[10:11], 0 +; GCN-IR-NEXT: s_add_u32 s11, s8, 1 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 +; GCN-IR-NEXT: s_or_b32 s6, s6, s7 +; GCN-IR-NEXT: s_cmp_lg_u32 s6, 0 +; GCN-IR-NEXT: s_addc_u32 s6, s9, 0 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s8, 63, s8 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[6:7] ; GCN-IR-NEXT: s_lshl_b64 s[6:7], s[2:3], s8 ; GCN-IR-NEXT: s_cbranch_vccz .LBB11_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: s_lshr_b64 s[8:9], s[2:3], s10 -; GCN-IR-NEXT: s_add_u32 s2, s12, 0xffffffc4 -; GCN-IR-NEXT: s_addc_u32 s3, 0, -1 -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 +; GCN-IR-NEXT: s_lshr_b64 s[2:3], s[2:3], s11 +; GCN-IR-NEXT: s_add_u32 s10, s10, 0xffffffc4 +; GCN-IR-NEXT: s_addc_u32 s11, 0, -1 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB11_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[8:9], 1 +; GCN-IR-NEXT: s_lshl_b64 s[2:3], s[2:3], 1 ; GCN-IR-NEXT: s_lshr_b32 s4, s7, 31 ; GCN-IR-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 -; GCN-IR-NEXT: s_or_b64 s[8:9], s[8:9], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s4, 23, s8 -; GCN-IR-NEXT: s_subb_u32 s4, 0, s9 -; GCN-IR-NEXT: s_ashr_i32 s10, s4, 31 -; GCN-IR-NEXT: s_and_b32 s4, s10, 1 -; GCN-IR-NEXT: s_and_b32 s10, s10, 24 -; GCN-IR-NEXT: s_sub_u32 s8, s8, s10 -; GCN-IR-NEXT: s_subb_u32 s9, s9, 0 -; GCN-IR-NEXT: s_add_u32 s2, s2, 1 -; GCN-IR-NEXT: s_addc_u32 s3, s3, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[2:3], 0 -; GCN-IR-NEXT: s_mov_b64 s[10:11], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[2:3], s[2:3], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s4, 23, s2 +; GCN-IR-NEXT: s_subb_u32 s4, 0, s3 +; GCN-IR-NEXT: s_ashr_i32 s8, s4, 31 +; GCN-IR-NEXT: s_and_b32 s4, s8, 1 +; GCN-IR-NEXT: s_and_b32 s8, s8, 24 +; GCN-IR-NEXT: s_sub_u32 s2, s2, s8 +; GCN-IR-NEXT: s_subb_u32 s3, s3, 0 +; GCN-IR-NEXT: s_add_u32 s10, s10, 1 +; GCN-IR-NEXT: s_cselect_b64 s[12:13], -1, 0 +; GCN-IR-NEXT: s_or_b32 s12, s12, s13 +; GCN-IR-NEXT: s_cmp_lg_u32 s12, 0 +; GCN-IR-NEXT: s_addc_u32 s11, s11, 0 +; GCN-IR-NEXT: s_cselect_b64 s[12:13], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[12:13] ; GCN-IR-NEXT: s_cbranch_vccz .LBB11_3 ; GCN-IR-NEXT: .LBB11_4: ; %Flow6 @@ -1384,13 +1399,13 @@ define i64 @v_test_udiv_k_den_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], 59, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_min_u32_e32 v6, v2, v3 +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 59, v6 +; GCN-IR-NEXT: v_subb_u32_e64 v5, s[4:5], 0, 0, s[4:5] ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] -; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[6:7] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[4:5] ; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 63, v[4:5] ; GCN-IR-NEXT: s_xor_b64 s[6:7], s[4:5], -1 ; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v1, 0, s[4:5] ; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v0, 0, s[4:5] @@ -1398,51 +1413,50 @@ define i64 @v_test_udiv_k_den_i64(i64 %x) { ; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] ; GCN-IR-NEXT: s_cbranch_execz .LBB12_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v6 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v6 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] +; GCN-IR-NEXT: v_add_i32_e32 v7, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v2, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB12_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_lshr_b64 v[6:7], v[0:1], v8 -; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 0xffffffc4, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v1, s[4:5], 0, -1, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 0xffffffc4, v6 +; GCN-IR-NEXT: v_lshr_b64 v[0:1], v[0:1], v7 +; GCN-IR-NEXT: v_addc_u32_e64 v9, s[8:9], 0, -1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB12_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[0:1], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, 23, v6 +; GCN-IR-NEXT: v_or_b32_e32 v0, v0, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v0 -; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 -; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 -; GCN-IR-NEXT: v_and_b32_e32 v8, 24, v8 -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] -; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v6, v8 -; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 -; GCN-IR-NEXT: v_subbrev_u32_e64 v7, s[4:5], 0, v7, s[4:5] -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, 23, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v6, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 24, v6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v6 +; GCN-IR-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v8 +; GCN-IR-NEXT: v_or_b32_e32 v3, v7, v3 +; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v7, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v6, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB12_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB12_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB12_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v2, v5, v1 ; GCN-IR-NEXT: v_or_b32_e32 v3, v4, v0 diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll index a1557418..8dfd3b7 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll @@ -31,5 +31,5 @@ define amdgpu_kernel void @kernel1() #1 { attributes #0 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll index fb225a9..fa01ee9 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll @@ -98,7 +98,7 @@ define amdgpu_kernel void @kernel2() #0 { attributes #0 = { "uniform-work-group-size"="true" } ;. ; CHECK: attributes #[[ATTR0]] = { "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR2]] = { "uniform-work-group-size"="true" } -; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR3]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll index cfede0c..09001ca 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll @@ -41,6 +41,6 @@ define amdgpu_kernel void @kernel3() #2 { attributes #2 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll index 854b724..4dede21 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll @@ -41,6 +41,6 @@ define amdgpu_kernel void @kernel2() #2 { attributes #1 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll index c4e0a60..08e1556 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-propagate-attribute.ll @@ -52,8 +52,8 @@ attributes #0 = { nounwind } attributes #1 = { "uniform-work-group-size"="false" } attributes #2 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { nounwind "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR2]] = { nounwind "uniform-work-group-size"="false" } ; CHECK: attributes #[[ATTR3]] = { "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll index 05af74d..9090d605 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll @@ -101,7 +101,7 @@ define amdgpu_kernel void @kernel(ptr addrspace(1) %m) #1 { attributes #0 = { nounwind readnone } attributes #1 = { "uniform-work-group-size"="true" } ;. -; CHECK: attributes #[[ATTR0]] = { nounwind memory(none) "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { nounwind memory(none) "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } -; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR0]] = { nounwind memory(none) "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR1]] = { nounwind memory(none) "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } +; CHECK: attributes #[[ATTR2]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="true" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll b/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll index cdbca7f..5e109f4 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll @@ -61,5 +61,5 @@ define amdgpu_kernel void @kernel3() #0 { attributes #0 = { "uniform-work-group-size"="false" } ;. -; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } +; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } ;. diff --git a/llvm/test/CodeGen/AMDGPU/urem64.ll b/llvm/test/CodeGen/AMDGPU/urem64.ll index b846ce7..cdcc914 100644 --- a/llvm/test/CodeGen/AMDGPU/urem64.ll +++ b/llvm/test/CodeGen/AMDGPU/urem64.ll @@ -170,35 +170,38 @@ define amdgpu_kernel void @s_test_urem_i64(ptr addrspace(1) %out, i64 %x, i64 %y ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[6:7], 0 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[2:3], 0 ; GCN-IR-NEXT: s_flbit_i32_b64 s10, s[6:7] -; GCN-IR-NEXT: s_flbit_i32_b64 s18, s[2:3] +; GCN-IR-NEXT: s_flbit_i32_b64 s16, s[2:3] ; GCN-IR-NEXT: s_or_b64 s[8:9], s[8:9], s[12:13] -; GCN-IR-NEXT: s_sub_u32 s12, s10, s18 +; GCN-IR-NEXT: s_sub_u32 s12, s10, s16 ; GCN-IR-NEXT: s_subb_u32 s13, 0, 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[14:15], s[12:13], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[12:13], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[18:19], s[12:13], 63 ; GCN-IR-NEXT: s_or_b64 s[14:15], s[8:9], s[14:15] ; GCN-IR-NEXT: s_and_b64 s[8:9], s[14:15], exec ; GCN-IR-NEXT: s_cselect_b32 s9, 0, s3 ; GCN-IR-NEXT: s_cselect_b32 s8, 0, s2 -; GCN-IR-NEXT: s_or_b64 s[14:15], s[14:15], s[16:17] +; GCN-IR-NEXT: s_or_b64 s[14:15], s[14:15], s[18:19] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[14:15] ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s14, s12, 1 -; GCN-IR-NEXT: s_addc_u32 s15, s13, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[14:15], 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 +; GCN-IR-NEXT: s_or_b32 s8, s8, s9 +; GCN-IR-NEXT: s_cmp_lg_u32 s8, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s13, 0 +; GCN-IR-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s12, 63, s12 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[8:9] ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[2:3], s12 ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[12:13], s[2:3], s14 -; GCN-IR-NEXT: s_add_u32 s16, s6, -1 -; GCN-IR-NEXT: s_addc_u32 s17, s7, -1 +; GCN-IR-NEXT: s_add_u32 s14, s6, -1 +; GCN-IR-NEXT: s_addc_u32 s15, s7, -1 ; GCN-IR-NEXT: s_not_b64 s[4:5], s[10:11] -; GCN-IR-NEXT: s_add_u32 s10, s4, s18 -; GCN-IR-NEXT: s_addc_u32 s11, s5, 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], 0 +; GCN-IR-NEXT: s_add_u32 s16, s4, s16 +; GCN-IR-NEXT: s_addc_u32 s17, s5, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB0_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -206,19 +209,22 @@ define amdgpu_kernel void @s_test_urem_i64(ptr addrspace(1) %out, i64 %x, i64 %y ; GCN-IR-NEXT: s_lshr_b32 s4, s9, 31 ; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[8:9], 1 ; GCN-IR-NEXT: s_or_b64 s[12:13], s[12:13], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[8:9], s[14:15], s[8:9] -; GCN-IR-NEXT: s_sub_u32 s4, s16, s12 -; GCN-IR-NEXT: s_subb_u32 s4, s17, s13 -; GCN-IR-NEXT: s_ashr_i32 s14, s4, 31 -; GCN-IR-NEXT: s_mov_b32 s15, s14 -; GCN-IR-NEXT: s_and_b32 s4, s14, 1 -; GCN-IR-NEXT: s_and_b64 s[14:15], s[14:15], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s12, s12, s14 -; GCN-IR-NEXT: s_subb_u32 s13, s13, s15 -; GCN-IR-NEXT: s_add_u32 s10, s10, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s11, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[18:19], s[10:11], 0 -; GCN-IR-NEXT: s_mov_b64 s[14:15], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[10:11], s[8:9] +; GCN-IR-NEXT: s_sub_u32 s4, s14, s12 +; GCN-IR-NEXT: s_subb_u32 s4, s15, s13 +; GCN-IR-NEXT: s_ashr_i32 s10, s4, 31 +; GCN-IR-NEXT: s_mov_b32 s11, s10 +; GCN-IR-NEXT: s_and_b32 s4, s10, 1 +; GCN-IR-NEXT: s_and_b64 s[18:19], s[10:11], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s12, s12, s18 +; GCN-IR-NEXT: s_subb_u32 s13, s13, s19 +; GCN-IR-NEXT: s_add_u32 s16, s16, 1 +; GCN-IR-NEXT: s_cselect_b64 s[18:19], -1, 0 +; GCN-IR-NEXT: s_or_b32 s18, s18, s19 +; GCN-IR-NEXT: s_cmp_lg_u32 s18, 0 +; GCN-IR-NEXT: s_addc_u32 s17, s17, 0 +; GCN-IR-NEXT: s_cselect_b64 s[18:19], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[18:19] ; GCN-IR-NEXT: s_cbranch_vccz .LBB0_3 ; GCN-IR-NEXT: .LBB0_4: ; %Flow7 @@ -362,12 +368,12 @@ define i64 @v_test_urem_i64(i64 %x, i64 %y) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v2 ; GCN-IR-NEXT: v_add_i32_e64 v4, s[6:7], 32, v4 ; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v3 -; GCN-IR-NEXT: v_min_u32_e32 v12, v4, v5 +; GCN-IR-NEXT: v_min_u32_e32 v10, v4, v5 ; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v0 ; GCN-IR-NEXT: v_add_i32_e64 v4, s[6:7], 32, v4 ; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v1 -; GCN-IR-NEXT: v_min_u32_e32 v13, v4, v5 -; GCN-IR-NEXT: v_sub_i32_e64 v4, s[6:7], v12, v13 +; GCN-IR-NEXT: v_min_u32_e32 v11, v4, v5 +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[6:7], v10, v11 ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_subb_u32_e64 v5, s[6:7], 0, 0, s[6:7] @@ -383,54 +389,53 @@ define i64 @v_test_urem_i64(i64 %x, i64 %y) { ; GCN-IR-NEXT: s_cbranch_execz .LBB1_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v4 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 63, v4 -; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9] ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[0:1], v4 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB1_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, -1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, -1, v3, vcc -; GCN-IR-NEXT: v_not_b32_e32 v6, v12 -; GCN-IR-NEXT: v_lshr_b64 v[10:11], v[0:1], v8 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, v6, v13 -; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v9, s[4:5], -1, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v13, 0 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_not_b32_e32 v6, v10 +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, v6, v11 +; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v8 +; GCN-IR-NEXT: v_addc_u32_e64 v15, s[8:9], -1, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 ; GCN-IR-NEXT: .LBB1_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[10:11], 1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v5 -; GCN-IR-NEXT: v_or_b32_e32 v10, v10, v6 +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v6 ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v14, v10 -; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, v15, v11, vcc -; GCN-IR-NEXT: v_or_b32_e32 v4, v12, v4 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v6 -; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 1, v8 -; GCN-IR-NEXT: v_or_b32_e32 v5, v13, v5 -; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v12 -; GCN-IR-NEXT: v_and_b32_e32 v13, v12, v3 -; GCN-IR-NEXT: v_and_b32_e32 v12, v12, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, 0, v9, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[8:9] -; GCN-IR-NEXT: v_sub_i32_e64 v10, s[4:5], v10, v12 -; GCN-IR-NEXT: v_subb_u32_e64 v11, s[4:5], v11, v13, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v13, v7 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v12, v6 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v12, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, v13, v9, vcc +; GCN-IR-NEXT: v_or_b32_e32 v4, v10, v4 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_or_b32_e32 v5, v11, v5 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v3 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, v8, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v9, vcc, v9, v11, vcc +; GCN-IR-NEXT: v_add_i32_e32 v14, vcc, 1, v14 +; GCN-IR-NEXT: v_addc_u32_e32 v15, vcc, 0, v15, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB1_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB1_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB1_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 ; GCN-IR-NEXT: v_or_b32_e32 v7, v7, v5 ; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 @@ -948,34 +953,37 @@ define amdgpu_kernel void @s_test_urem_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GCN-IR-NEXT: s_mov_b64 s[4:5], 0 ; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) -; GCN-IR-NEXT: s_flbit_i32_b64 s12, s[2:3] -; GCN-IR-NEXT: s_add_u32 s8, s12, 0xffffffc5 +; GCN-IR-NEXT: s_flbit_i32_b64 s14, s[2:3] +; GCN-IR-NEXT: s_add_u32 s8, s14, 0xffffffc5 ; GCN-IR-NEXT: s_addc_u32 s9, 0, -1 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[2:3], 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[10:11], s[8:9], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[14:15], s[8:9], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[8:9], 63 ; GCN-IR-NEXT: s_or_b64 s[10:11], s[6:7], s[10:11] ; GCN-IR-NEXT: s_and_b64 s[6:7], s[10:11], exec ; GCN-IR-NEXT: s_cselect_b32 s6, 0, 24 -; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[14:15] +; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[12:13] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[10:11] ; GCN-IR-NEXT: s_mov_b32 s7, 0 ; GCN-IR-NEXT: s_cbranch_vccz .LBB6_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: s_add_u32 s10, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[10:11], 0 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 +; GCN-IR-NEXT: s_or_b32 s6, s6, s7 +; GCN-IR-NEXT: s_cmp_lg_u32 s6, 0 +; GCN-IR-NEXT: s_addc_u32 s6, s9, 0 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s8, 63, s8 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[6:7] ; GCN-IR-NEXT: s_lshl_b64 s[6:7], 24, s8 ; GCN-IR-NEXT: s_cbranch_vccz .LBB6_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader ; GCN-IR-NEXT: s_lshr_b64 s[10:11], 24, s10 -; GCN-IR-NEXT: s_add_u32 s14, s2, -1 -; GCN-IR-NEXT: s_addc_u32 s15, s3, -1 -; GCN-IR-NEXT: s_sub_u32 s8, 58, s12 -; GCN-IR-NEXT: s_subb_u32 s9, 0, 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], 0 +; GCN-IR-NEXT: s_add_u32 s12, s2, -1 +; GCN-IR-NEXT: s_addc_u32 s13, s3, -1 +; GCN-IR-NEXT: s_sub_u32 s14, 58, s14 +; GCN-IR-NEXT: s_subb_u32 s15, 0, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB6_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 @@ -983,19 +991,22 @@ define amdgpu_kernel void @s_test_urem_k_num_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR-NEXT: s_lshr_b32 s4, s7, 31 ; GCN-IR-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 ; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[6:7], s[12:13], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s4, s14, s10 -; GCN-IR-NEXT: s_subb_u32 s4, s15, s11 -; GCN-IR-NEXT: s_ashr_i32 s12, s4, 31 -; GCN-IR-NEXT: s_mov_b32 s13, s12 -; GCN-IR-NEXT: s_and_b32 s4, s12, 1 -; GCN-IR-NEXT: s_and_b64 s[12:13], s[12:13], s[2:3] -; GCN-IR-NEXT: s_sub_u32 s10, s10, s12 -; GCN-IR-NEXT: s_subb_u32 s11, s11, s13 -; GCN-IR-NEXT: s_add_u32 s8, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s9, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[16:17], s[8:9], 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[8:9], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s4, s12, s10 +; GCN-IR-NEXT: s_subb_u32 s4, s13, s11 +; GCN-IR-NEXT: s_ashr_i32 s8, s4, 31 +; GCN-IR-NEXT: s_mov_b32 s9, s8 +; GCN-IR-NEXT: s_and_b32 s4, s8, 1 +; GCN-IR-NEXT: s_and_b64 s[16:17], s[8:9], s[2:3] +; GCN-IR-NEXT: s_sub_u32 s10, s10, s16 +; GCN-IR-NEXT: s_subb_u32 s11, s11, s17 +; GCN-IR-NEXT: s_add_u32 s14, s14, 1 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_or_b32 s16, s16, s17 +; GCN-IR-NEXT: s_cmp_lg_u32 s16, 0 +; GCN-IR-NEXT: s_addc_u32 s15, s15, 0 +; GCN-IR-NEXT: s_cselect_b64 s[16:17], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[16:17] ; GCN-IR-NEXT: s_cbranch_vccz .LBB6_3 ; GCN-IR-NEXT: .LBB6_4: ; %Flow6 @@ -1064,52 +1075,58 @@ define amdgpu_kernel void @s_test_urem_k_den_i64(ptr addrspace(1) %out, i64 %x) ; GCN-IR: ; %bb.0: ; %_udiv-special-cases ; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) -; GCN-IR-NEXT: s_flbit_i32_b64 s12, s[2:3] -; GCN-IR-NEXT: s_sub_u32 s8, 59, s12 +; GCN-IR-NEXT: s_flbit_i32_b64 s10, s[2:3] +; GCN-IR-NEXT: s_sub_u32 s8, 59, s10 ; GCN-IR-NEXT: s_subb_u32 s9, 0, 0 ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], s[2:3], 0 ; GCN-IR-NEXT: v_cmp_gt_u64_e64 s[6:7], s[8:9], 63 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[10:11], s[8:9], 63 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[8:9], 63 ; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7] ; GCN-IR-NEXT: s_and_b64 s[6:7], s[4:5], exec ; GCN-IR-NEXT: s_cselect_b32 s7, 0, s3 ; GCN-IR-NEXT: s_cselect_b32 s6, 0, s2 -; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[10:11] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], s[12:13] ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[4:5] ; GCN-IR-NEXT: s_mov_b64 s[4:5], 0 ; GCN-IR-NEXT: s_cbranch_vccz .LBB7_5 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 -; GCN-IR-NEXT: s_add_u32 s10, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s11, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[10:11], 0 +; GCN-IR-NEXT: s_add_u32 s11, s8, 1 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 +; GCN-IR-NEXT: s_or_b32 s6, s6, s7 +; GCN-IR-NEXT: s_cmp_lg_u32 s6, 0 +; GCN-IR-NEXT: s_addc_u32 s6, s9, 0 +; GCN-IR-NEXT: s_cselect_b64 s[6:7], -1, 0 ; GCN-IR-NEXT: s_sub_i32 s8, 63, s8 ; GCN-IR-NEXT: s_andn2_b64 vcc, exec, s[6:7] ; GCN-IR-NEXT: s_lshl_b64 s[6:7], s[2:3], s8 ; GCN-IR-NEXT: s_cbranch_vccz .LBB7_4 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: s_lshr_b64 s[10:11], s[2:3], s10 -; GCN-IR-NEXT: s_add_u32 s8, s12, 0xffffffc4 -; GCN-IR-NEXT: s_addc_u32 s9, 0, -1 -; GCN-IR-NEXT: s_mov_b64 s[12:13], 0 +; GCN-IR-NEXT: s_lshr_b64 s[8:9], s[2:3], s11 +; GCN-IR-NEXT: s_add_u32 s12, s10, 0xffffffc4 +; GCN-IR-NEXT: s_addc_u32 s13, 0, -1 +; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 ; GCN-IR-NEXT: s_mov_b32 s5, 0 ; GCN-IR-NEXT: .LBB7_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: s_lshl_b64 s[10:11], s[10:11], 1 +; GCN-IR-NEXT: s_lshl_b64 s[8:9], s[8:9], 1 ; GCN-IR-NEXT: s_lshr_b32 s4, s7, 31 ; GCN-IR-NEXT: s_lshl_b64 s[6:7], s[6:7], 1 -; GCN-IR-NEXT: s_or_b64 s[10:11], s[10:11], s[4:5] -; GCN-IR-NEXT: s_or_b64 s[6:7], s[12:13], s[6:7] -; GCN-IR-NEXT: s_sub_u32 s4, 23, s10 -; GCN-IR-NEXT: s_subb_u32 s4, 0, s11 -; GCN-IR-NEXT: s_ashr_i32 s12, s4, 31 -; GCN-IR-NEXT: s_and_b32 s4, s12, 1 -; GCN-IR-NEXT: s_and_b32 s12, s12, 24 -; GCN-IR-NEXT: s_sub_u32 s10, s10, s12 -; GCN-IR-NEXT: s_subb_u32 s11, s11, 0 -; GCN-IR-NEXT: s_add_u32 s8, s8, 1 -; GCN-IR-NEXT: s_addc_u32 s9, s9, 0 -; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[14:15], s[8:9], 0 -; GCN-IR-NEXT: s_mov_b64 s[12:13], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[8:9], s[4:5] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[10:11], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s4, 23, s8 +; GCN-IR-NEXT: s_subb_u32 s4, 0, s9 +; GCN-IR-NEXT: s_ashr_i32 s10, s4, 31 +; GCN-IR-NEXT: s_and_b32 s4, s10, 1 +; GCN-IR-NEXT: s_and_b32 s10, s10, 24 +; GCN-IR-NEXT: s_sub_u32 s8, s8, s10 +; GCN-IR-NEXT: s_subb_u32 s9, s9, 0 +; GCN-IR-NEXT: s_add_u32 s12, s12, 1 +; GCN-IR-NEXT: s_cselect_b64 s[14:15], -1, 0 +; GCN-IR-NEXT: s_or_b32 s14, s14, s15 +; GCN-IR-NEXT: s_cmp_lg_u32 s14, 0 +; GCN-IR-NEXT: s_addc_u32 s13, s13, 0 +; GCN-IR-NEXT: s_cselect_b64 s[14:15], -1, 0 +; GCN-IR-NEXT: s_mov_b64 s[10:11], s[4:5] ; GCN-IR-NEXT: s_and_b64 vcc, exec, s[14:15] ; GCN-IR-NEXT: s_cbranch_vccz .LBB7_3 ; GCN-IR-NEXT: .LBB7_4: ; %Flow6 @@ -1241,8 +1258,8 @@ define i64 @v_test_urem_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 -; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 0xffffffd0, v10 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 0xffffffd0, v8 ; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc ; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] @@ -1257,54 +1274,53 @@ define i64 @v_test_urem_pow2_k_num_i64(i64 %x) { ; GCN-IR-NEXT: s_cbranch_execz .LBB8_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc -; GCN-IR-NEXT: s_mov_b64 s[4:5], 0x8000 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0x8000 +; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[8:9], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_lshl_b64 v[2:3], s[4:5], v2 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[8:9] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[10:11], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[10:11] ; GCN-IR-NEXT: s_cbranch_execz .LBB8_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc -; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[4:5], v6 -; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 47, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, 47, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], s[8:9], v6 +; GCN-IR-NEXT: v_subb_u32_e64 v13, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 ; GCN-IR-NEXT: .LBB8_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v12, v8 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v13, v9, vcc -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 -; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], v9, v11, s[4:5] -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, v10, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, v11, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v9, v8, v1 +; GCN-IR-NEXT: v_and_b32_e32 v8, v8, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v7, vcc, v7, v9, vcc +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, 1, v12 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB8_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB8_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB8_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v3 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v2 @@ -1337,8 +1353,8 @@ define i64 @v_test_urem_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 ; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 ; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 -; GCN-IR-NEXT: v_min_u32_e32 v10, v2, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 48, v10 +; GCN-IR-NEXT: v_min_u32_e32 v8, v2, v3 +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 48, v8 ; GCN-IR-NEXT: v_subb_u32_e64 v3, s[4:5], 0, 0, s[4:5] ; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] ; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[2:3] @@ -1352,51 +1368,50 @@ define i64 @v_test_urem_pow2_k_den_i64(i64 %x) { ; GCN-IR-NEXT: s_cbranch_execz .LBB9_6 ; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 ; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v2 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; GCN-IR-NEXT: v_sub_i32_e64 v2, s[4:5], 63, v2 -; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 -; GCN-IR-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[6:7] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc -; GCN-IR-NEXT: s_xor_b64 s[8:9], exec, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], vcc, -1 +; GCN-IR-NEXT: s_and_saveexec_b64 s[8:9], s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execz .LBB9_5 ; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader -; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v6 -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 0xffffffcf, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 -; GCN-IR-NEXT: v_addc_u32_e64 v7, s[4:5], 0, -1, vcc -; GCN-IR-NEXT: s_mov_b64 s[10:11], 0 -; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, 0xffffffcf, v8 +; GCN-IR-NEXT: v_lshr_b64 v[6:7], v[0:1], v6 +; GCN-IR-NEXT: v_addc_u32_e64 v11, s[8:9], 0, -1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 ; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 -; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: s_movk_i32 s10, 0x7fff ; GCN-IR-NEXT: .LBB9_3: ; %udiv-do-while ; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 -; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 ; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v3 -; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 -; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s12, v8 +; GCN-IR-NEXT: v_or_b32_e32 v6, v6, v4 ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 -; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v9, vcc -; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v6 -; GCN-IR-NEXT: v_or_b32_e32 v2, v10, v2 -; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 -; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v7, vcc -; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 -; GCN-IR-NEXT: v_and_b32_e32 v10, 0x8000, v10 -; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] -; GCN-IR-NEXT: v_or_b32_e32 v3, v11, v3 -; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v10 -; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 -; GCN-IR-NEXT: v_subbrev_u32_e64 v9, s[4:5], 0, v9, s[4:5] -; GCN-IR-NEXT: s_or_b64 s[10:11], vcc, s[10:11] -; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 -; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s10, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v2, v8, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 0x8000, v8 +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v6, v8 +; GCN-IR-NEXT: v_subbrev_u32_e32 v7, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, 1, v10 +; GCN-IR-NEXT: v_or_b32_e32 v3, v9, v3 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, 0, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GCN-IR-NEXT: s_cbranch_execnz .LBB9_3 ; GCN-IR-NEXT: ; %bb.4: ; %Flow -; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] -; GCN-IR-NEXT: .LBB9_5: ; %Flow4 ; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: .LBB9_5: ; %Flow4 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[4:5] ; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 1 ; GCN-IR-NEXT: v_or_b32_e32 v5, v5, v3 ; GCN-IR-NEXT: v_or_b32_e32 v4, v4, v2 diff --git a/llvm/test/CodeGen/AMDGPU/usubo.ll b/llvm/test/CodeGen/AMDGPU/usubo.ll index 0289dab..d67a7b1 100644 --- a/llvm/test/CodeGen/AMDGPU/usubo.ll +++ b/llvm/test/CodeGen/AMDGPU/usubo.ll @@ -14,15 +14,16 @@ define amdgpu_kernel void @s_usubo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) ; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_sub_u32 s0, s2, s8 -; SI-NEXT: v_mov_b32_e32 v0, s2 +; SI-NEXT: s_sub_u32 s2, s2, s8 ; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_subb_u32 s1, s3, s9 +; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; SI-NEXT: s_or_b32 s0, s0, s1 +; SI-NEXT: s_cmp_lg_u32 s0, 0 +; SI-NEXT: s_subb_u32 s3, s3, s9 +; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; SI-NEXT: v_mov_b32_e32 v1, s3 -; SI-NEXT: v_cmp_gt_u64_e32 vcc, s[0:1], v[0:1] -; SI-NEXT: v_mov_b32_e32 v1, s1 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; SI-NEXT: v_add_i32_e32 v0, vcc, s0, v0 +; SI-NEXT: v_add_i32_e32 v0, vcc, s2, v0 ; SI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; SI-NEXT: s_endpgm @@ -33,15 +34,15 @@ define amdgpu_kernel void @s_usubo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34 ; VI-NEXT: s_waitcnt lgkmcnt(0) ; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: s_sub_u32 s0, s2, s4 -; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_sub_u32 s2, s2, s4 ; VI-NEXT: v_mov_b32_e32 v1, s1 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: s_cmp_lg_u64 s[0:1], 0 +; VI-NEXT: s_subb_u32 s3, s3, s5 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[0:1] ; VI-NEXT: v_mov_b32_e32 v3, s3 -; VI-NEXT: s_subb_u32 s1, s3, s5 -; VI-NEXT: v_cmp_gt_u64_e32 vcc, s[0:1], v[2:3] -; VI-NEXT: v_mov_b32_e32 v3, s1 -; VI-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc -; VI-NEXT: v_add_u32_e32 v2, vcc, s0, v2 +; VI-NEXT: v_add_u32_e32 v2, vcc, s2, v2 ; VI-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc ; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] ; VI-NEXT: s_endpgm @@ -52,14 +53,14 @@ define amdgpu_kernel void @s_usubo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 ; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: v_mov_b32_e32 v0, s2 -; GFX9-NEXT: s_sub_u32 s4, s2, s6 -; GFX9-NEXT: v_mov_b32_e32 v1, s3 -; GFX9-NEXT: s_subb_u32 s5, s3, s7 -; GFX9-NEXT: v_cmp_gt_u64_e32 vcc, s[4:5], v[0:1] -; GFX9-NEXT: v_mov_b32_e32 v1, s5 -; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s4, v0 +; GFX9-NEXT: s_sub_u32 s6, s2, s6 +; GFX9-NEXT: s_cselect_b64 s[4:5], -1, 0 +; GFX9-NEXT: s_cmp_lg_u64 s[4:5], 0 +; GFX9-NEXT: s_subb_u32 s4, s3, s7 +; GFX9-NEXT: s_cselect_b64 s[2:3], -1, 0 +; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[2:3] +; GFX9-NEXT: v_mov_b32_e32 v1, s4 +; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s6, v0 ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc ; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX9-NEXT: s_endpgm @@ -71,12 +72,14 @@ define amdgpu_kernel void @s_usubo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; GFX10-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34 ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: s_sub_u32 s4, s2, s6 -; GFX10-NEXT: s_subb_u32 s5, s3, s7 -; GFX10-NEXT: v_cmp_gt_u64_e64 s2, s[4:5], s[2:3] -; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s2 -; GFX10-NEXT: v_add_co_u32 v0, s2, s4, v0 -; GFX10-NEXT: v_add_co_ci_u32_e64 v1, s2, s5, 0, s2 +; GFX10-NEXT: s_sub_u32 s2, s2, s6 +; GFX10-NEXT: s_cselect_b32 s4, -1, 0 +; GFX10-NEXT: s_cmp_lg_u32 s4, 0 +; GFX10-NEXT: s_subb_u32 s3, s3, s7 +; GFX10-NEXT: s_cselect_b32 s4, -1, 0 +; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, s4 +; GFX10-NEXT: v_add_co_u32 v0, s2, s2, v0 +; GFX10-NEXT: v_add_co_ci_u32_e64 v1, s2, s3, 0, s2 ; GFX10-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] ; GFX10-NEXT: s_endpgm ; @@ -87,14 +90,16 @@ define amdgpu_kernel void @s_usubo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 % ; GFX11-NEXT: s_load_b64 s[4:5], s[4:5], 0x34 ; GFX11-NEXT: v_mov_b32_e32 v2, 0 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_sub_u32 s4, s2, s4 -; GFX11-NEXT: s_subb_u32 s5, s3, s5 -; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_cmp_gt_u64_e64 s2, s[4:5], s[2:3] -; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s2 +; GFX11-NEXT: s_sub_u32 s2, s2, s4 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1) +; GFX11-NEXT: s_cmp_lg_u32 s4, 0 +; GFX11-NEXT: s_subb_u32 s3, s3, s5 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, s4 ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_add_co_u32 v0, s2, s4, v0 -; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, s5, 0, s2 +; GFX11-NEXT: v_add_co_u32 v0, s2, s2, v0 +; GFX11-NEXT: v_add_co_ci_u32_e64 v1, null, s3, 0, s2 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] ; GFX11-NEXT: s_endpgm %usub = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %b) #0 @@ -435,21 +440,23 @@ define amdgpu_kernel void @s_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; SI-NEXT: s_mov_b32 s11, 0xf000 ; SI-NEXT: s_mov_b32 s10, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_sub_u32 s6, s4, s6 -; SI-NEXT: v_mov_b32_e32 v0, s4 -; SI-NEXT: s_subb_u32 s7, s5, s7 -; SI-NEXT: v_mov_b32_e32 v1, s5 -; SI-NEXT: v_cmp_gt_u64_e32 vcc, s[6:7], v[0:1] -; SI-NEXT: v_mov_b32_e32 v2, s6 +; SI-NEXT: s_sub_u32 s4, s4, s6 +; SI-NEXT: s_cselect_b64 s[12:13], -1, 0 +; SI-NEXT: s_or_b32 s6, s12, s13 +; SI-NEXT: s_cmp_lg_u32 s6, 0 +; SI-NEXT: s_subb_u32 s5, s5, s7 ; SI-NEXT: s_mov_b32 s8, s0 ; SI-NEXT: s_mov_b32 s9, s1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: s_cselect_b64 s[4:5], -1, 0 ; SI-NEXT: s_mov_b32 s0, s2 ; SI-NEXT: s_mov_b32 s1, s3 ; SI-NEXT: s_mov_b32 s2, s10 ; SI-NEXT: s_mov_b32 s3, s11 -; SI-NEXT: v_mov_b32_e32 v3, s7 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; SI-NEXT: buffer_store_dwordx2 v[2:3], off, s[8:11], 0 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; SI-NEXT: s_waitcnt expcnt(0) +; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] ; SI-NEXT: buffer_store_byte v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; @@ -457,37 +464,37 @@ define amdgpu_kernel void @s_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx8 s[0:7], s[4:5], 0x24 ; VI-NEXT: s_waitcnt lgkmcnt(0) +; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_sub_u32 s2, s4, s6 ; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: s_sub_u32 s0, s4, s6 -; VI-NEXT: v_mov_b32_e32 v4, s4 ; VI-NEXT: v_mov_b32_e32 v1, s1 -; VI-NEXT: s_subb_u32 s1, s5, s7 -; VI-NEXT: v_mov_b32_e32 v5, s5 -; VI-NEXT: v_mov_b32_e32 v7, s1 -; VI-NEXT: v_cmp_gt_u64_e32 vcc, s[0:1], v[4:5] -; VI-NEXT: v_mov_b32_e32 v6, s0 -; VI-NEXT: v_mov_b32_e32 v2, s2 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 +; VI-NEXT: s_cmp_lg_u64 s[0:1], 0 +; VI-NEXT: s_subb_u32 s0, s5, s7 +; VI-NEXT: v_mov_b32_e32 v4, s2 +; VI-NEXT: v_mov_b32_e32 v5, s0 +; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 ; VI-NEXT: v_mov_b32_e32 v3, s3 -; VI-NEXT: flat_store_dwordx2 v[0:1], v[6:7] -; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc +; VI-NEXT: flat_store_dwordx2 v[0:1], v[4:5] +; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1] ; VI-NEXT: flat_store_byte v[2:3], v0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: s_usubo_i64: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x24 -; GFX9-NEXT: v_mov_b32_e32 v4, 0 +; GFX9-NEXT: v_mov_b32_e32 v2, 0 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_sub_u32 s0, s12, s14 -; GFX9-NEXT: v_mov_b32_e32 v0, s12 -; GFX9-NEXT: v_mov_b32_e32 v1, s13 -; GFX9-NEXT: s_subb_u32 s1, s13, s15 -; GFX9-NEXT: v_mov_b32_e32 v3, s1 -; GFX9-NEXT: v_cmp_gt_u64_e32 vcc, s[0:1], v[0:1] -; GFX9-NEXT: v_mov_b32_e32 v2, s0 -; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GFX9-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] -; GFX9-NEXT: global_store_byte v4, v0, s[10:11] +; GFX9-NEXT: s_sub_u32 s2, s12, s14 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: s_cmp_lg_u64 s[0:1], 0 +; GFX9-NEXT: s_subb_u32 s0, s13, s15 +; GFX9-NEXT: v_mov_b32_e32 v0, s2 +; GFX9-NEXT: v_mov_b32_e32 v1, s0 +; GFX9-NEXT: s_cselect_b64 s[0:1], -1, 0 +; GFX9-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[0:1] +; GFX9-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] +; GFX9-NEXT: global_store_byte v2, v3, s[10:11] ; GFX9-NEXT: s_endpgm ; ; GFX10-LABEL: s_usubo_i64: @@ -496,10 +503,12 @@ define amdgpu_kernel void @s_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX10-NEXT: v_mov_b32_e32 v2, 0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_sub_u32 s0, s12, s14 -; GFX10-NEXT: s_subb_u32 s1, s13, s15 +; GFX10-NEXT: s_cselect_b32 s1, -1, 0 ; GFX10-NEXT: v_mov_b32_e32 v0, s0 +; GFX10-NEXT: s_cmp_lg_u32 s1, 0 +; GFX10-NEXT: s_subb_u32 s1, s13, s15 +; GFX10-NEXT: s_cselect_b32 s0, -1, 0 ; GFX10-NEXT: v_mov_b32_e32 v1, s1 -; GFX10-NEXT: v_cmp_gt_u64_e64 s0, s[0:1], s[12:13] ; GFX10-NEXT: v_cndmask_b32_e64 v3, 0, 1, s0 ; GFX10-NEXT: global_store_dwordx2 v2, v[0:1], s[8:9] ; GFX10-NEXT: global_store_byte v2, v3, s[10:11] @@ -509,12 +518,13 @@ define amdgpu_kernel void @s_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX11: ; %bb.0: ; GFX11-NEXT: s_load_b256 s[0:7], s[4:5], 0x24 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: s_sub_u32 s6, s4, s6 -; GFX11-NEXT: s_subb_u32 s7, s5, s7 -; GFX11-NEXT: v_mov_b32_e32 v0, s6 -; GFX11-NEXT: v_cmp_gt_u64_e64 s4, s[6:7], s[4:5] -; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s7 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-NEXT: s_sub_u32 s4, s4, s6 +; GFX11-NEXT: s_cselect_b32 s6, -1, 0 +; GFX11-NEXT: v_mov_b32_e32 v0, s4 +; GFX11-NEXT: s_cmp_lg_u32 s6, 0 +; GFX11-NEXT: s_subb_u32 s5, s5, s7 +; GFX11-NEXT: s_cselect_b32 s4, -1, 0 +; GFX11-NEXT: v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s5 ; GFX11-NEXT: v_cndmask_b32_e64 v3, 0, 1, s4 ; GFX11-NEXT: s_clause 0x1 ; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1] @@ -550,10 +560,10 @@ define amdgpu_kernel void @v_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; SI-NEXT: s_mov_b32 s4, s2 ; SI-NEXT: s_mov_b32 s5, s3 ; SI-NEXT: s_waitcnt vmcnt(0) -; SI-NEXT: v_sub_i32_e32 v2, vcc, v0, v2 -; SI-NEXT: v_subb_u32_e32 v3, vcc, v1, v3, vcc -; SI-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; SI-NEXT: buffer_store_dwordx2 v[2:3], off, s[8:11], 0 +; SI-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; SI-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; SI-NEXT: s_waitcnt expcnt(0) ; SI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; SI-NEXT: buffer_store_byte v0, off, s[4:7], 0 ; SI-NEXT: s_endpgm @@ -573,10 +583,9 @@ define amdgpu_kernel void @v_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; VI-NEXT: v_mov_b32_e32 v6, s2 ; VI-NEXT: v_mov_b32_e32 v7, s3 ; VI-NEXT: s_waitcnt vmcnt(0) -; VI-NEXT: v_sub_u32_e32 v2, vcc, v0, v2 -; VI-NEXT: v_subb_u32_e32 v3, vcc, v1, v3, vcc -; VI-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; VI-NEXT: flat_store_dwordx2 v[4:5], v[2:3] +; VI-NEXT: v_sub_u32_e32 v0, vcc, v0, v2 +; VI-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; VI-NEXT: flat_store_dwordx2 v[4:5], v[0:1] ; VI-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; VI-NEXT: flat_store_byte v[6:7], v0 ; VI-NEXT: s_endpgm @@ -589,10 +598,9 @@ define amdgpu_kernel void @v_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX9-NEXT: global_load_dwordx2 v[0:1], v4, s[12:13] ; GFX9-NEXT: global_load_dwordx2 v[2:3], v4, s[14:15] ; GFX9-NEXT: s_waitcnt vmcnt(0) -; GFX9-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v2 -; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v3, vcc -; GFX9-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; GFX9-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] +; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v2 +; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v3, vcc +; GFX9-NEXT: global_store_dwordx2 v4, v[0:1], s[8:9] ; GFX9-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; GFX9-NEXT: global_store_byte v4, v0, s[10:11] ; GFX9-NEXT: s_endpgm @@ -606,12 +614,11 @@ define amdgpu_kernel void @v_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX10-NEXT: global_load_dwordx2 v[0:1], v4, s[12:13] ; GFX10-NEXT: global_load_dwordx2 v[2:3], v4, s[14:15] ; GFX10-NEXT: s_waitcnt vmcnt(0) -; GFX10-NEXT: v_sub_co_u32 v2, vcc_lo, v0, v2 -; GFX10-NEXT: v_sub_co_ci_u32_e32 v3, vcc_lo, v1, v3, vcc_lo -; GFX10-NEXT: v_cmp_gt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX10-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo -; GFX10-NEXT: global_store_dwordx2 v4, v[2:3], s[8:9] -; GFX10-NEXT: global_store_byte v4, v0, s[10:11] +; GFX10-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2 +; GFX10-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX10-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo +; GFX10-NEXT: global_store_dwordx2 v4, v[0:1], s[8:9] +; GFX10-NEXT: global_store_byte v4, v2, s[10:11] ; GFX10-NEXT: s_endpgm ; ; GFX11-LABEL: v_usubo_i64: @@ -623,14 +630,12 @@ define amdgpu_kernel void @v_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) % ; GFX11-NEXT: global_load_b64 v[0:1], v4, s[4:5] ; GFX11-NEXT: global_load_b64 v[2:3], v4, s[6:7] ; GFX11-NEXT: s_waitcnt vmcnt(0) -; GFX11-NEXT: v_sub_co_u32 v2, vcc_lo, v0, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-NEXT: v_sub_co_ci_u32_e64 v3, null, v1, v3, vcc_lo -; GFX11-NEXT: v_cmp_gt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc_lo +; GFX11-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2 +; GFX11-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX11-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc_lo ; GFX11-NEXT: s_clause 0x1 -; GFX11-NEXT: global_store_b64 v4, v[2:3], s[0:1] -; GFX11-NEXT: global_store_b8 v4, v0, s[2:3] +; GFX11-NEXT: global_store_b64 v4, v[0:1], s[0:1] +; GFX11-NEXT: global_store_b8 v4, v2, s[2:3] ; GFX11-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %tid.ext = sext i32 %tid to i64 diff --git a/llvm/test/CodeGen/AMDGPU/usubsat.ll b/llvm/test/CodeGen/AMDGPU/usubsat.ll index 90491a0..3ddb2f0 100644 --- a/llvm/test/CodeGen/AMDGPU/usubsat.ll +++ b/llvm/test/CodeGen/AMDGPU/usubsat.ll @@ -730,52 +730,38 @@ define i64 @v_usubsat_i64(i64 %lhs, i64 %rhs) { ; GFX6-LABEL: v_usubsat_i64: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-NEXT: v_sub_i32_e32 v2, vcc, v0, v2 -; GFX6-NEXT: v_subb_u32_e32 v3, vcc, v1, v3, vcc -; GFX6-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; GFX6-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc -; GFX6-NEXT: v_cndmask_b32_e64 v1, v3, 0, vcc +; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GFX6-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GFX6-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX6-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX6-NEXT: s_setpc_b64 s[30:31] ; ; GFX8-LABEL: v_usubsat_i64: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-NEXT: v_sub_u32_e32 v2, vcc, v0, v2 -; GFX8-NEXT: v_subb_u32_e32 v3, vcc, v1, v3, vcc -; GFX8-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; GFX8-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v1, v3, 0, vcc +; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v2 +; GFX8-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GFX8-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX8-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX8-NEXT: s_setpc_b64 s[30:31] ; ; GFX9-LABEL: v_usubsat_i64: ; GFX9: ; %bb.0: ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v2 -; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v3, vcc -; GFX9-NEXT: v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1] -; GFX9-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc -; GFX9-NEXT: v_cndmask_b32_e64 v1, v3, 0, vcc +; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v2 +; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v3, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc +; GFX9-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc ; GFX9-NEXT: s_setpc_b64 s[30:31] ; -; GFX10-LABEL: v_usubsat_i64: -; GFX10: ; %bb.0: -; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_sub_co_u32 v2, vcc_lo, v0, v2 -; GFX10-NEXT: v_sub_co_ci_u32_e32 v3, vcc_lo, v1, v3, vcc_lo -; GFX10-NEXT: v_cmp_gt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX10-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v1, v3, 0, vcc_lo -; GFX10-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-LABEL: v_usubsat_i64: -; GFX11: ; %bb.0: -; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_sub_co_u32 v2, vcc_lo, v0, v2 -; GFX11-NEXT: v_sub_co_ci_u32_e64 v3, null, v1, v3, vcc_lo -; GFX11-NEXT: v_cmp_gt_u64_e32 vcc_lo, v[2:3], v[0:1] -; GFX11-NEXT: v_cndmask_b32_e64 v0, v2, 0, vcc_lo -; GFX11-NEXT: v_cndmask_b32_e64 v1, v3, 0, vcc_lo -; GFX11-NEXT: s_setpc_b64 s[30:31] +; GFX10PLUS-LABEL: v_usubsat_i64: +; GFX10PLUS: ; %bb.0: +; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX10PLUS-NEXT: v_sub_co_u32 v0, vcc_lo, v0, v2 +; GFX10PLUS-NEXT: v_sub_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo +; GFX10PLUS-NEXT: v_cndmask_b32_e64 v0, v0, 0, vcc_lo +; GFX10PLUS-NEXT: v_cndmask_b32_e64 v1, v1, 0, vcc_lo +; GFX10PLUS-NEXT: s_setpc_b64 s[30:31] %result = call i64 @llvm.usub.sat.i64(i64 %lhs, i64 %rhs) ret i64 %result } |