diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU')
18 files changed, 4788 insertions, 1638 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/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-no-agpr.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-min-agpr-alloc.ll index 2ad6e68..f730199 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-no-agpr.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-min-agpr-alloc.ll @@ -70,7 +70,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_def() { define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR2:[0-9]+]] { ; CHECK-NEXT: [[DEF:%.*]] = call i64 asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -118,7 +118,7 @@ define amdgpu_kernel void @kernel_uses_asm_physreg() { define amdgpu_kernel void @kernel_uses_asm_physreg_tuple() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_tuple( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR2]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -154,7 +154,7 @@ define void @func_uses_asm_physreg_agpr() { define void @func_uses_asm_physreg_agpr_tuple() { ; CHECK-LABEL: define void @func_uses_asm_physreg_agpr_tuple( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR2]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -168,7 +168,7 @@ declare void @unknown() define amdgpu_kernel void @kernel_calls_extern() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR3:[0-9]+]] { ; CHECK-NEXT: call void @unknown() ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -180,8 +180,8 @@ define amdgpu_kernel void @kernel_calls_extern() { define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern_marked_callsite( -; CHECK-SAME: ) #[[ATTR1]] { -; CHECK-NEXT: call void @unknown() #[[ATTR10:[0-9]+]] +; CHECK-SAME: ) #[[ATTR3]] { +; CHECK-NEXT: call void @unknown() #[[ATTR29:[0-9]+]] ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; @@ -192,7 +192,7 @@ define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { define amdgpu_kernel void @kernel_calls_indirect(ptr %indirect) { ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect( -; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR1]] { +; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR3]] { ; CHECK-NEXT: call void [[INDIRECT]]() ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -204,8 +204,8 @@ define amdgpu_kernel void @kernel_calls_indirect(ptr %indirect) { 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:%.*]]) #[[ATTR1]] { -; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR10]] +; CHECK-SAME: ptr [[INDIRECT:%.*]]) #[[ATTR3]] { +; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR29]] ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void ; @@ -316,7 +316,7 @@ define amdgpu_kernel void @kernel_calls_workitem_id_x(ptr addrspace(1) %out) { define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { ; CHECK-LABEL: define amdgpu_kernel void @indirect_calls_none_agpr( -; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR1]] { +; 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:%.*]] @@ -342,7 +342,7 @@ define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR2]] { ; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -354,7 +354,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0() { 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: ) #[[ATTR1]] { +; 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 @@ -378,7 +378,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2() { define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR2]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -390,7 +390,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty() { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR2]] { ; CHECK-NEXT: [[DEF:%.*]] = call ptr asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -402,7 +402,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty() { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR5]] { ; CHECK-NEXT: [[DEF:%.*]] = call <2 x ptr> asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -414,7 +414,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty() { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR6:[0-9]+]] { ; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -426,7 +426,7 @@ define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0() { define amdgpu_kernel void @kernel_uses_asm_clobber() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR7:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -438,7 +438,7 @@ define amdgpu_kernel void @kernel_uses_asm_clobber() { define amdgpu_kernel void @kernel_uses_asm_clobber_tuple() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_tuple( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR8:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -450,7 +450,7 @@ define amdgpu_kernel void @kernel_uses_asm_clobber_tuple() { define amdgpu_kernel void @kernel_uses_asm_clobber_oob() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_oob( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR9:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -462,7 +462,7 @@ define amdgpu_kernel void @kernel_uses_asm_clobber_oob() { define amdgpu_kernel void @kernel_uses_asm_clobber_max() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_max( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR9]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -474,7 +474,7 @@ define amdgpu_kernel void @kernel_uses_asm_clobber_max() { define amdgpu_kernel void @kernel_uses_asm_physreg_oob() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_oob( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR9]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -486,7 +486,7 @@ define amdgpu_kernel void @kernel_uses_asm_physreg_oob() { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR10:[0-9]+]] { ; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -498,7 +498,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty() { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR10]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -510,7 +510,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty() { 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: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR10]] { ; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -522,7 +522,7 @@ define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty() { define amdgpu_kernel void @vreg_use_exceeds_register_file() { ; CHECK-LABEL: define amdgpu_kernel void @vreg_use_exceeds_register_file( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR9]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -534,7 +534,7 @@ define amdgpu_kernel void @vreg_use_exceeds_register_file() { define amdgpu_kernel void @vreg_def_exceeds_register_file() { ; CHECK-LABEL: define amdgpu_kernel void @vreg_def_exceeds_register_file( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR9]] { ; CHECK-NEXT: [[DEF:%.*]] = call <257 x i32> asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -546,7 +546,7 @@ define amdgpu_kernel void @vreg_def_exceeds_register_file() { define amdgpu_kernel void @multiple() { ; CHECK-LABEL: define amdgpu_kernel void @multiple( -; CHECK-SAME: ) #[[ATTR1]] { +; 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 @@ -558,7 +558,7 @@ define amdgpu_kernel void @multiple() { define amdgpu_kernel void @earlyclobber_0() { ; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_0( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR11:[0-9]+]] { ; CHECK-NEXT: [[DEF:%.*]] = call <8 x i32> asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -570,7 +570,7 @@ define amdgpu_kernel void @earlyclobber_0() { define amdgpu_kernel void @earlyclobber_1() { ; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_1( -; CHECK-SAME: ) #[[ATTR1]] { +; 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 @@ -582,7 +582,7 @@ define amdgpu_kernel void @earlyclobber_1() { define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512() { ; CHECK-LABEL: define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR13:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -594,7 +594,7 @@ define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512() { 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: ) #[[ATTR1]] { +; 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 @@ -606,7 +606,7 @@ define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512() { 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: ) #[[ATTR1]] { +; 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 @@ -618,7 +618,7 @@ define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256() { define amdgpu_kernel void @mixed_physreg_vreg_tuples_0() { ; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_0( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR11]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -630,7 +630,7 @@ define amdgpu_kernel void @mixed_physreg_vreg_tuples_0() { define amdgpu_kernel void @mixed_physreg_vreg_tuples_1() { ; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_1( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR15:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -642,7 +642,7 @@ define amdgpu_kernel void @mixed_physreg_vreg_tuples_1() { define amdgpu_kernel void @physreg_raises_limit() { ; CHECK-LABEL: define amdgpu_kernel void @physreg_raises_limit( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR16:[0-9]+]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -652,10 +652,9 @@ define amdgpu_kernel void @physreg_raises_limit() { ret void } -; FIXME: This should require 9. We cannot allocate an a128 at a0. define amdgpu_kernel void @physreg_tuple_alignment_raises_limit() { ; CHECK-LABEL: define amdgpu_kernel void @physreg_tuple_alignment_raises_limit( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR11]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -667,7 +666,7 @@ define amdgpu_kernel void @physreg_tuple_alignment_raises_limit() { define amdgpu_kernel void @align3_virtreg() { ; CHECK-LABEL: define amdgpu_kernel void @align3_virtreg( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR6]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -679,7 +678,7 @@ define amdgpu_kernel void @align3_virtreg() { define amdgpu_kernel void @align3_align4_virtreg() { ; CHECK-LABEL: define amdgpu_kernel void @align3_align4_virtreg( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR15]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -691,7 +690,7 @@ define amdgpu_kernel void @align3_align4_virtreg() { define amdgpu_kernel void @align2_align4_virtreg() { ; CHECK-LABEL: define amdgpu_kernel void @align2_align4_virtreg( -; CHECK-SAME: ) #[[ATTR1]] { +; CHECK-SAME: ) #[[ATTR15]] { ; CHECK-NEXT: call void asm sideeffect " ; CHECK-NEXT: call void @use_most() ; CHECK-NEXT: ret void @@ -703,7 +702,7 @@ define amdgpu_kernel void @align2_align4_virtreg() { define amdgpu_kernel void @kernel_uses_write_register_a55() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55( -; CHECK-SAME: ) #[[ATTR3:[0-9]+]] { +; CHECK-SAME: ) #[[ATTR17:[0-9]+]] { ; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META0:![0-9]+]], i32 0) ; CHECK-NEXT: ret void ; @@ -713,71 +712,313 @@ define amdgpu_kernel void @kernel_uses_write_register_a55() { define amdgpu_kernel void @kernel_uses_write_register_v55() { ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_v55( -; CHECK-SAME: ) #[[ATTR4:[0-9]+]] { +; 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: ) #[[ATTR3]] { +; 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:%.*]]) #[[ATTR3]] { +; 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:%.*]]) #[[ATTR3]] { +; 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:%.*]]) #[[ATTR3]] { +; 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: ) #[[ATTR3]] { +; 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"} @@ -787,16 +1028,35 @@ attributes #0 = { "amdgpu-agpr-alloc"="0" } ;. ; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR1]] = { "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR2:[0-9]+]] = { convergent nocallback nofree nosync nounwind willreturn memory(none) "target-cpu"="gfx90a" } -; 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" "target-cpu"="gfx90a" "uniform-work-group-size"="false" } -; CHECK: attributes #[[ATTR4]] = { "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 #[[ATTR5:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR6:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR7:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(read) "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR8:[0-9]+]] = { nounwind "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR9:[0-9]+]] = { nocallback nounwind "target-cpu"="gfx90a" } -; CHECK: attributes #[[ATTR10]] = { "amdgpu-agpr-alloc"="0" } +; 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"} diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll new file mode 100644 index 0000000..6c4f504 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-uniform-waterfall.ll @@ -0,0 +1,452 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6 +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -amdgpu-enable-uniform-intrinsic-combine=0 -O3 -S < %s | FileCheck %s -check-prefix=CURRENT-CHECK +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -O3 -S < %s | FileCheck %s -check-prefix=O3-CHECK + +define protected amdgpu_kernel void @trivial_waterfall_eq_zero(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_PEEL:%.*]] = icmp eq i32 [[TMP0]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_PEEL]], label %[[EXIT:.*]], label %[[IF_PEEL:.*]] +; CURRENT-CHECK: [[IF_PEEL]]: +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[EXIT]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0:[0-9]+]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ true, %[[IF:.*]] ] +; PASS-CHECK-NEXT: [[NOT_DONE:%.*]] = xor i1 [[DONE]], true +; PASS-CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[NOT_DONE]], true +; PASS-CHECK-NEXT: br i1 [[TMP0]], label %[[EXIT:.*]], label %[[IF]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ 0, %entry ], [ 1, %if ] + %not_done = xor i1 %done, true + %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %not_done) + %is_done = icmp eq i64 %ballot, 0 ; in this case is_done = !not_done + br i1 %is_done, label %exit, label %if + +if: + store i32 5, ptr addrspace(1) %out + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @trivial_waterfall_eq_zero_swap_op(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero_swap_op( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_PEEL:%.*]] = icmp eq i32 [[TMP0]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_PEEL]], label %[[EXIT:.*]], label %[[IF_PEEL:.*]] +; CURRENT-CHECK: [[IF_PEEL]]: +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[EXIT]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero_swap_op( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ true, %[[IF:.*]] ] +; PASS-CHECK-NEXT: [[NOT_DONE:%.*]] = xor i1 [[DONE]], true +; PASS-CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[NOT_DONE]], true +; PASS-CHECK-NEXT: br i1 [[TMP0]], label %[[EXIT:.*]], label %[[IF]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero_swap_op( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ 0, %entry ], [ 1, %if ] + %not_done = xor i1 %done, true + %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %not_done) + %is_done = icmp eq i64 0, %ballot ; in this case is_done = !not_done + br i1 %is_done, label %exit, label %if + +if: + store i32 5, ptr addrspace(1) %out + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @trivial_waterfall_ne_zero(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[WHILE:.*]] +; CURRENT-CHECK: [[WHILE]]: +; CURRENT-CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_NOT:%.*]] = icmp eq i32 [[TMP0]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_NOT]], label %[[WHILE]], label %[[EXIT:.*]], !llvm.loop [[LOOP0:![0-9]+]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ true, %[[IF:.*]] ] +; PASS-CHECK-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[IF]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ 0, %entry ], [ 1, %if ] + %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %done) + %is_done = icmp ne i64 0, %ballot ; in this case is_done = done + br i1 %is_done, label %exit, label %if + +if: + store i32 5, ptr addrspace(1) %out + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @trivial_waterfall_ne_zero_swap(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero_swap( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[WHILE:.*]] +; CURRENT-CHECK: [[WHILE]]: +; CURRENT-CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_NOT:%.*]] = icmp eq i32 [[TMP0]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_NOT]], label %[[WHILE]], label %[[EXIT:.*]], !llvm.loop [[LOOP2:![0-9]+]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero_swap( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ true, %[[IF:.*]] ] +; PASS-CHECK-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[IF]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero_swap( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ 0, %entry ], [ 1, %if ] + %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %done) + %is_done = icmp ne i64 %ballot, 0 ; in this case is_done = done + br i1 %is_done, label %exit, label %if + +if: + store i32 5, ptr addrspace(1) %out + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @trivial_uniform_waterfall(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_uniform_waterfall( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_PEEL:%.*]] = icmp eq i32 [[TMP0]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_PEEL]], label %[[EXIT:.*]], label %[[WORK_PEEL:.*]] +; CURRENT-CHECK: [[WORK_PEEL]]: +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[EXIT]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_uniform_waterfall( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ [[NEW_DONE:%.*]], %[[TAIL:.*]] ] +; PASS-CHECK-NEXT: [[NOT_DONE:%.*]] = xor i1 [[DONE]], true +; PASS-CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[NOT_DONE]], true +; PASS-CHECK-NEXT: br i1 [[TMP0]], label %[[EXIT:.*]], label %[[IF:.*]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: [[IS_FIRST_ACTIVE_ID:%.*]] = icmp eq i32 0, 0 +; PASS-CHECK-NEXT: br i1 [[IS_FIRST_ACTIVE_ID]], label %[[WORK:.*]], label %[[TAIL]] +; PASS-CHECK: [[WORK]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[TAIL]] +; PASS-CHECK: [[TAIL]]: +; PASS-CHECK-NEXT: [[NEW_DONE]] = phi i1 [ true, %[[WORK]] ], [ false, %[[IF]] ] +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_uniform_waterfall( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ false, %entry ], [ %new_done, %tail ] + %not_done = xor i1 %done, true + %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %not_done) + %is_done = icmp eq i64 %ballot, 0 + br i1 %is_done, label %exit, label %if + +if: + %first_active_id = tail call noundef i32 @llvm.amdgcn.readfirstlane.i32(i32 0) + %is_first_active_id = icmp eq i32 0, %first_active_id + br i1 %is_first_active_id, label %work, label %tail + +work: + store i32 5, ptr addrspace(1) %out + br label %tail + +tail: + %new_done = phi i1 [ true, %work ], [ false, %if ] + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @uniform_waterfall(ptr addrspace(1) %out, i32 %mymask) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @uniform_waterfall( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]], i32 [[MYMASK:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_PEEL:%.*]] = icmp eq i32 [[TMP0]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_PEEL]], label %[[EXIT:.*]], label %[[WORK_PEEL:.*]] +; CURRENT-CHECK: [[WORK_PEEL]]: +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[EXIT]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @uniform_waterfall( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[MYMASK:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ [[NEW_DONE:%.*]], %[[TAIL:.*]] ] +; PASS-CHECK-NEXT: [[NOT_DONE:%.*]] = xor i1 [[DONE]], true +; PASS-CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[NOT_DONE]], true +; PASS-CHECK-NEXT: br i1 [[TMP0]], label %[[EXIT:.*]], label %[[IF:.*]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: [[IS_FIRST_ACTIVE_ID:%.*]] = icmp eq i32 [[MYMASK]], [[MYMASK]] +; PASS-CHECK-NEXT: br i1 [[IS_FIRST_ACTIVE_ID]], label %[[WORK:.*]], label %[[TAIL]] +; PASS-CHECK: [[WORK]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[TAIL]] +; PASS-CHECK: [[TAIL]]: +; PASS-CHECK-NEXT: [[NEW_DONE]] = phi i1 [ true, %[[WORK]] ], [ false, %[[IF]] ] +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @uniform_waterfall( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]], i32 [[MYMASK:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ false, %entry ], [ %new_done, %tail ] + %not_done = xor i1 %done, true + %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %not_done) + %is_done = icmp eq i64 %ballot, 0 + br i1 %is_done, label %exit, label %if + +if: + %first_active_id = tail call noundef i32 @llvm.amdgcn.readfirstlane.i32(i32 %mymask) + %is_first_active_id = icmp eq i32 %mymask, %first_active_id + br i1 %is_first_active_id, label %work, label %tail + +work: + store i32 5, ptr addrspace(1) %out + br label %tail + +tail: + %new_done = phi i1 [ true, %work ], [ false, %if ] + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @trivial_waterfall_eq_zero_i32(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero_i32( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: [[BALLOT_PEEL:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_PEEL:%.*]] = icmp eq i32 [[BALLOT_PEEL]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_PEEL]], label %[[EXIT:.*]], label %[[IF_PEEL:.*]] +; CURRENT-CHECK: [[IF_PEEL]]: +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[EXIT]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero_i32( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ true, %[[IF:.*]] ] +; PASS-CHECK-NEXT: [[NOT_DONE:%.*]] = xor i1 [[DONE]], true +; PASS-CHECK-NEXT: [[TMP0:%.*]] = xor i1 [[NOT_DONE]], true +; PASS-CHECK-NEXT: br i1 [[TMP0]], label %[[EXIT:.*]], label %[[IF]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_eq_zero_i32( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ 0, %entry ], [ 1, %if ] + %not_done = xor i1 %done, true + %ballot = tail call i32 @llvm.amdgcn.ballot.i32(i1 %not_done) + %is_done = icmp eq i32 %ballot, 0 ; in this case is_done = !not_done + br i1 %is_done, label %exit, label %if + +if: + store i32 5, ptr addrspace(1) %out + br label %while + +exit: + ret void +} + +define protected amdgpu_kernel void @trivial_waterfall_ne_zero_i32(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero_i32( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[ENTRY:.*:]] +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: br label %[[WHILE:.*]] +; CURRENT-CHECK: [[WHILE]]: +; CURRENT-CHECK-NEXT: [[BALLOT:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 true) +; CURRENT-CHECK-NEXT: [[IS_DONE_NOT:%.*]] = icmp eq i32 [[BALLOT]], 0 +; CURRENT-CHECK-NEXT: br i1 [[IS_DONE_NOT]], label %[[WHILE]], label %[[EXIT:.*]], !llvm.loop [[LOOP3:![0-9]+]] +; CURRENT-CHECK: [[EXIT]]: +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero_i32( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: br label %[[WHILE:.*]] +; PASS-CHECK: [[WHILE]]: +; PASS-CHECK-NEXT: [[DONE:%.*]] = phi i1 [ false, %[[ENTRY]] ], [ true, %[[IF:.*]] ] +; PASS-CHECK-NEXT: br i1 [[DONE]], label %[[EXIT:.*]], label %[[IF]] +; PASS-CHECK: [[IF]]: +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: br label %[[WHILE]] +; PASS-CHECK: [[EXIT]]: +; PASS-CHECK-NEXT: ret void +; +; O3-CHECK-LABEL: define protected amdgpu_kernel void @trivial_waterfall_ne_zero_i32( +; O3-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; O3-CHECK-NEXT: [[ENTRY:.*:]] +; O3-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; O3-CHECK-NEXT: ret void +; +entry: + br label %while + +while: + %done = phi i1 [ 0, %entry ], [ 1, %if ] + %ballot = tail call i32 @llvm.amdgcn.ballot.i32(i1 %done) + %is_done = icmp ne i32 0, %ballot ; in this case is_done = done + br i1 %is_done, label %exit, label %if + +if: + store i32 5, ptr addrspace(1) %out + br label %while + +exit: + ret void +} + +declare i64 @llvm.amdgcn.ballot.i64(i1) #1 +!6 = !{i64 690} +!7 = distinct !{!7, !8} +!8 = !{!"llvm.loop.mustprogress"} +;. +; CURRENT-CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]]} +; CURRENT-CHECK: [[META1]] = !{!"llvm.loop.peeled.count", i32 1} +; CURRENT-CHECK: [[LOOP2]] = distinct !{[[LOOP2]], [[META1]]} +; CURRENT-CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]} +;. diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll new file mode 100644 index 0000000..aa11574 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-intrinsic-combine.ll @@ -0,0 +1,790 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6 +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -amdgpu-enable-uniform-intrinsic-combine=0 -O3 -S < %s | FileCheck %s -check-prefix=CURRENT-CHECK +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine,dce -S < %s | FileCheck %s -check-prefix=DCE-CHECK + +define amdgpu_kernel void @permlane64_constant(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @permlane64_constant( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +; CURRENT-CHECK-NEXT: store i32 77, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @permlane64_constant( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0:[0-9]+]] { +; PASS-CHECK-NEXT: store i32 77, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @permlane64_constant( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0:[0-9]+]] { +; DCE-CHECK-NEXT: store i32 77, ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v = call i32 @llvm.amdgcn.permlane64(i32 77) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @permlane64_uniform(ptr addrspace(1) %out, i32 %src) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @permlane64_uniform( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]], i32 [[SRC:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 [[SRC]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @permlane64_uniform( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[SRC:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 [[SRC]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @permlane64_uniform( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[SRC:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 [[SRC]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v = call i32 @llvm.amdgcn.permlane64(i32 %src) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @permlane64_nonuniform(i32 addrspace(1)* %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @permlane64_nonuniform( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] { +; CURRENT-CHECK-NEXT: [[TID:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.permlane64.i32(i32 [[TID]]) +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TID]] to i64 +; CURRENT-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i64 [[TMP1]] +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @permlane64_nonuniform( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.permlane64.i32(i32 [[TID]]) +; PASS-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID]] +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @permlane64_nonuniform( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.permlane64.i32(i32 [[TID]]) +; DCE-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID]] +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tid = call i32 @llvm.amdgcn.workitem.id.x() + %v = call i32 @llvm.amdgcn.permlane64(i32 %tid) + %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid + store i32 %v, i32 addrspace(1)* %out_ptr + ret void +} + +define amdgpu_kernel void @permlane64_nonuniform_expression(i32 addrspace(1)* %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @permlane64_nonuniform_expression( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[TID:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TID2:%.*]] = add nuw nsw i32 [[TID]], 1 +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.permlane64.i32(i32 [[TID2]]) +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TID]] to i64 +; CURRENT-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i64 [[TMP1]] +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @permlane64_nonuniform_expression( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TID2:%.*]] = add i32 [[TID]], 1 +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.permlane64.i32(i32 [[TID2]]) +; PASS-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID]] +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @permlane64_nonuniform_expression( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TID2:%.*]] = add i32 [[TID]], 1 +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.permlane64.i32(i32 [[TID2]]) +; DCE-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID]] +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tid = call i32 @llvm.amdgcn.workitem.id.x() + %tid2 = add i32 %tid, 1 + %v = call i32 @llvm.amdgcn.permlane64(i32 %tid2) + %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid + store i32 %v, i32 addrspace(1)* %out_ptr + ret void +} + +define amdgpu_kernel void @readlane_constant(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_constant( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 7, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_constant( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 7, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_constant( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 7, ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v = call i32 @llvm.amdgcn.readlane(i32 7, i32 5) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readlane_nonuniform_indices(ptr addrspace(1) %out, i32 %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_indices( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]], i32 [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 [[SRC0]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_indices( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 [[SRC0]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_indices( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 [[SRC0]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v = call i32 @llvm.amdgcn.readlane(i32 %src0, i32 %src1) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readlane_nonuniform_workitem(i32 addrspace(1)* %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_workitem( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] { +; CURRENT-CHECK-NEXT: [[TIDX:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TIDY:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.y() +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TIDX]] to i64 +; CURRENT-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i64 [[TMP1]] +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_workitem( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; PASS-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TIDX]] +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_workitem( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; DCE-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TIDX]] +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tidx = call i32 @llvm.amdgcn.workitem.id.x() + %tidy = call i32 @llvm.amdgcn.workitem.id.y() + %v = call i32 @llvm.amdgcn.readlane(i32 %tidx, i32 %tidy) + %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tidx + store i32 %v, i32 addrspace(1)* %out_ptr + ret void +} + +define amdgpu_kernel void @readlane_nonuniform_expression(i32 addrspace(1)* %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_expression( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2]] { +; CURRENT-CHECK-NEXT: [[TIDX:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TIDY:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.y() +; CURRENT-CHECK-NEXT: [[TIDX2:%.*]] = add nuw nsw i32 [[TIDX]], 1 +; CURRENT-CHECK-NEXT: [[TIDY2:%.*]] = add nuw nsw i32 [[TIDY]], 2 +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX2]], i32 [[TIDY2]]) +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TIDX]] to i64 +; CURRENT-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i64 [[TMP1]] +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_expression( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; PASS-CHECK-NEXT: [[TIDX2:%.*]] = add i32 [[TIDX]], 1 +; PASS-CHECK-NEXT: [[TIDY2:%.*]] = add i32 [[TIDY]], 2 +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX2]], i32 [[TIDY2]]) +; PASS-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TIDX]] +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_nonuniform_expression( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; DCE-CHECK-NEXT: [[TIDX2:%.*]] = add i32 [[TIDX]], 1 +; DCE-CHECK-NEXT: [[TIDY2:%.*]] = add i32 [[TIDY]], 2 +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX2]], i32 [[TIDY2]]) +; DCE-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TIDX]] +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tidx = call i32 @llvm.amdgcn.workitem.id.x() + %tidy = call i32 @llvm.amdgcn.workitem.id.y() + %tidx2 = add i32 %tidx, 1 + %tidy2 = add i32 %tidy, 2 + %v = call i32 @llvm.amdgcn.readlane(i32 %tidx2, i32 %tidy2) + %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tidx + store i32 %v, i32 addrspace(1)* %out_ptr + ret void +} + +define amdgpu_kernel void @readfirstlane_constant(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_constant( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 7, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_constant( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 7, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_constant( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 7, ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v = call i32 @llvm.amdgcn.readfirstlane(i32 7) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readfirstlane_with_argument(ptr addrspace(1) %out, i32 %src0) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_argument( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]], i32 [[SRC0:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 [[SRC0]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_argument( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[SRC0:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 [[SRC0]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_argument( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[SRC0:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 [[SRC0]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v = call i32 @llvm.amdgcn.readfirstlane(i32 %src0) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readfirstlane_with_workitem_id(i32 addrspace(1)* %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_workitem_id( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[TID:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TID]]) +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TID]] to i64 +; CURRENT-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i64 [[TMP1]] +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_workitem_id( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TID]]) +; PASS-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID]] +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_workitem_id( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TID]]) +; DCE-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID]] +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tid = call i32 @llvm.amdgcn.workitem.id.x() + %v = call i32 @llvm.amdgcn.readfirstlane(i32 %tid) + %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid + store i32 %v, i32 addrspace(1)* %out_ptr + ret void +} + +define amdgpu_kernel void @readfirstlane_expression(i32 addrspace(1)* %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_expression( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[TID:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TID2:%.*]] = add nuw nsw i32 [[TID]], 1 +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TID2]]) +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TID2]] to i64 +; CURRENT-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i64 [[TMP1]] +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_expression( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TID2:%.*]] = add i32 [[TID]], 1 +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TID2]]) +; PASS-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID2]] +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_expression( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TID2:%.*]] = add i32 [[TID]], 1 +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TID2]]) +; DCE-CHECK-NEXT: [[OUT_PTR:%.*]] = getelementptr i32, ptr addrspace(1) [[OUT]], i32 [[TID2]] +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT_PTR]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tid = call i32 @llvm.amdgcn.workitem.id.x() + %tid2 = add i32 %tid, 1 + %v = call i32 @llvm.amdgcn.readfirstlane(i32 %tid2) + %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid2 + store i32 %v, i32 addrspace(1)* %out_ptr + ret void +} + +define amdgpu_kernel void @readfirstlane_with_readfirstlane(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_readfirstlane( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_readfirstlane( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_readfirstlane( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 5, ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %v1 = call i32 @llvm.amdgcn.readfirstlane(i32 5) + %v2 = call i32 @llvm.amdgcn.readfirstlane(i32 %v1) + store i32 %v2, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readfirstlane_with_readlane(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_readlane( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2]] { +; CURRENT-CHECK-NEXT: [[TIDX:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TIDY:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.y() +; CURRENT-CHECK-NEXT: [[V1:%.*]] = tail call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; CURRENT-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_readlane( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; PASS-CHECK-NEXT: [[V1:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; PASS-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_with_readlane( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; DCE-CHECK-NEXT: [[V1:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; DCE-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tidx = call i32 @llvm.amdgcn.workitem.id.x() + %tidy = call i32 @llvm.amdgcn.workitem.id.y() + %v1 = call i32 @llvm.amdgcn.readlane(i32 %tidx, i32 %tidy) + %v2 = call i32 @llvm.amdgcn.readfirstlane(i32 %v1) + store i32 %v2, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readlane_with_firstlane(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_with_firstlane( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[TIDX:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[V1:%.*]] = tail call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TIDX]]) +; CURRENT-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_with_firstlane( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[V1:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TIDX]]) +; PASS-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_with_firstlane( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[V1:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[TIDX]]) +; DCE-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tidx = call i32 @llvm.amdgcn.workitem.id.x() + %v1 = call i32 @llvm.amdgcn.readfirstlane(i32 %tidx) + %v2 = call i32 @llvm.amdgcn.readlane(i32 %v1, i32 3) + store i32 %v2, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readlane_readlane(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_readlane( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2]] { +; CURRENT-CHECK-NEXT: [[TIDX:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TIDY:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.y() +; CURRENT-CHECK-NEXT: [[V1:%.*]] = tail call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; CURRENT-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_readlane( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; PASS-CHECK-NEXT: [[V1:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; PASS-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_readlane( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TIDY:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() +; DCE-CHECK-NEXT: [[V1:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; DCE-CHECK-NEXT: store i32 [[V1]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tidx = call i32 @llvm.amdgcn.workitem.id.x() + %tidy = call i32 @llvm.amdgcn.workitem.id.y() + %v1 = call i32 @llvm.amdgcn.readlane(i32 %tidx, i32 %tidy) + %v2 = call i32 @llvm.amdgcn.readlane(i32 %v1, i32 2) + store i32 %v2, ptr addrspace(1) %out + ret void +} + + +define amdgpu_kernel void @permlane64_boundary(ptr addrspace(1) %out_min, ptr addrspace(1) %out_max) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @permlane64_boundary( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT_MIN:%.*]], ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT_MAX:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 -2147483648, ptr addrspace(1) [[OUT_MIN]], align 4 +; CURRENT-CHECK-NEXT: store i32 2147483647, ptr addrspace(1) [[OUT_MAX]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @permlane64_boundary( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT_MIN:%.*]], ptr addrspace(1) [[OUT_MAX:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: store i32 -2147483648, ptr addrspace(1) [[OUT_MIN]], align 4 +; PASS-CHECK-NEXT: store i32 2147483647, ptr addrspace(1) [[OUT_MAX]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @permlane64_boundary( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT_MIN:%.*]], ptr addrspace(1) [[OUT_MAX:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: store i32 -2147483648, ptr addrspace(1) [[OUT_MIN]], align 4 +; DCE-CHECK-NEXT: store i32 2147483647, ptr addrspace(1) [[OUT_MAX]], align 4 +; DCE-CHECK-NEXT: ret void +; + %min_v = call i32 @llvm.amdgcn.permlane64(i32 -2147483648) + store i32 %min_v, ptr addrspace(1) %out_min + %max_v = call i32 @llvm.amdgcn.permlane64(i32 2147483647) + store i32 %max_v, ptr addrspace(1) %out_max + ret void +} + +define amdgpu_kernel void @readlane_cross_lane(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_cross_lane( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[TIDX:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[TIDY:%.*]] = add nuw nsw i32 [[TIDX]], 5 +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_cross_lane( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[TIDY:%.*]] = add i32 [[TIDX]], 5 +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_cross_lane( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[TIDX:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[TIDY:%.*]] = add i32 [[TIDX]], 5 +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[TIDX]], i32 [[TIDY]]) +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %tidx = call i32 @llvm.amdgcn.workitem.id.x() + %tidy = add i32 %tidx, 5 + %v = call i32 @llvm.amdgcn.readlane(i32 %tidx, i32 %tidy) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readfirstlane_random(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_random( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR0]] { +; CURRENT-CHECK-NEXT: store i32 435, ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_random( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[RANDOM:%.*]] = xor i32 123, 456 +; PASS-CHECK-NEXT: store i32 [[RANDOM]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readfirstlane_random( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[RANDOM:%.*]] = xor i32 123, 456 +; DCE-CHECK-NEXT: store i32 [[RANDOM]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %random = xor i32 123, 456 + %v = call i32 @llvm.amdgcn.readfirstlane(i32 %random) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @readlane_expression(ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @readlane_expression( +; CURRENT-CHECK-SAME: ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[IDX1:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x() +; CURRENT-CHECK-NEXT: [[IDX2:%.*]] = shl nuw nsw i32 [[IDX1]], 1 +; CURRENT-CHECK-NEXT: [[V:%.*]] = tail call i32 @llvm.amdgcn.readlane.i32(i32 [[IDX1]], i32 [[IDX2]]) +; CURRENT-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT]], align 4 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @readlane_expression( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[IDX1:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: [[IDX2:%.*]] = mul i32 [[IDX1]], 2 +; PASS-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[IDX1]], i32 [[IDX2]]) +; PASS-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @readlane_expression( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[IDX1:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; DCE-CHECK-NEXT: [[IDX2:%.*]] = mul i32 [[IDX1]], 2 +; DCE-CHECK-NEXT: [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[IDX1]], i32 [[IDX2]]) +; DCE-CHECK-NEXT: store i32 [[V]], ptr addrspace(1) [[OUT]], align 4 +; DCE-CHECK-NEXT: ret void +; + %idx1 = call i32 @llvm.amdgcn.workitem.id.x() + %idx2 = mul i32 %idx1, 2 + %v = call i32 @llvm.amdgcn.readlane(i32 %idx1, i32 %idx2) + store i32 %v, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @ballot_i32(i32 %v, ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @ballot_i32( +; CURRENT-CHECK-SAME: i32 [[V:%.*]], ptr addrspace(1) writeonly captures(none) initializes((0, 1)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[C:%.*]] = trunc i32 [[V]] to i1 +; CURRENT-CHECK-NEXT: [[BALLOT:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 [[C]]) +; CURRENT-CHECK-NEXT: [[BALLOT_NE_ZERO:%.*]] = icmp ne i32 [[BALLOT]], 0 +; CURRENT-CHECK-NEXT: store i1 [[BALLOT_NE_ZERO]], ptr addrspace(1) [[OUT]], align 1 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @ballot_i32( +; PASS-CHECK-SAME: i32 [[V:%.*]], ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[C:%.*]] = trunc i32 [[V]] to i1 +; PASS-CHECK-NEXT: store i1 [[C]], ptr addrspace(1) [[OUT]], align 1 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @ballot_i32( +; DCE-CHECK-SAME: i32 [[V:%.*]], ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[C:%.*]] = trunc i32 [[V]] to i1 +; DCE-CHECK-NEXT: store i1 [[C]], ptr addrspace(1) [[OUT]], align 1 +; DCE-CHECK-NEXT: ret void +; + %c = trunc i32 %v to i1 + %ballot = call i32 @llvm.amdgcn.ballot.i32(i1 %c) + %ballot_ne_zero = icmp ne i32 %ballot, 0 + store i1 %ballot_ne_zero, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @ballot_i64(i32 %v, ptr addrspace(1) %out) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @ballot_i64( +; CURRENT-CHECK-SAME: i32 [[V:%.*]], ptr addrspace(1) writeonly captures(none) initializes((0, 1)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR1]] { +; CURRENT-CHECK-NEXT: [[C:%.*]] = trunc i32 [[V]] to i1 +; CURRENT-CHECK-NEXT: [[TMP1:%.*]] = tail call i32 @llvm.amdgcn.ballot.i32(i1 [[C]]) +; CURRENT-CHECK-NEXT: [[BALLOT_NE_ZERO:%.*]] = icmp ne i32 [[TMP1]], 0 +; CURRENT-CHECK-NEXT: store i1 [[BALLOT_NE_ZERO]], ptr addrspace(1) [[OUT]], align 1 +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @ballot_i64( +; PASS-CHECK-SAME: i32 [[V:%.*]], ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[C:%.*]] = trunc i32 [[V]] to i1 +; PASS-CHECK-NEXT: store i1 [[C]], ptr addrspace(1) [[OUT]], align 1 +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @ballot_i64( +; DCE-CHECK-SAME: i32 [[V:%.*]], ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[C:%.*]] = trunc i32 [[V]] to i1 +; DCE-CHECK-NEXT: store i1 [[C]], ptr addrspace(1) [[OUT]], align 1 +; DCE-CHECK-NEXT: ret void +; + %c = trunc i32 %v to i1 + %ballot = call i64 @llvm.amdgcn.ballot.i64(i1 %c) + %ballot_ne_zero = icmp ne i64 %ballot, 0 + store i1 %ballot_ne_zero, ptr addrspace(1) %out + ret void +} + +define amdgpu_kernel void @test_readlane_i16(i16 %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @test_readlane_i16( +; CURRENT-CHECK-SAME: i16 [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3:[0-9]+]] { +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @test_readlane_i16( +; PASS-CHECK-SAME: i16 [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @test_readlane_i16( +; DCE-CHECK-SAME: i16 [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %readlane = call i16 @llvm.amdgcn.readlane.i16(i16 %src0, i32 %src1) + call void asm sideeffect "; use $0", "s"(i16 %readlane) + ret void +} + +define amdgpu_kernel void @test_readlane_i64(i64 %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @test_readlane_i64( +; CURRENT-CHECK-SAME: i64 [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3]] { +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @test_readlane_i64( +; PASS-CHECK-SAME: i64 [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @test_readlane_i64( +; DCE-CHECK-SAME: i64 [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %readlane = call i64 @llvm.amdgcn.readlane.i64(i64 %src0, i32 %src1) + call void asm sideeffect "; use $0", "s"(i64 %readlane) + ret void +} + +define amdgpu_kernel void @test_readlane_bf16(bfloat %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @test_readlane_bf16( +; CURRENT-CHECK-SAME: bfloat [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3]] { +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @test_readlane_bf16( +; PASS-CHECK-SAME: bfloat [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @test_readlane_bf16( +; DCE-CHECK-SAME: bfloat [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %readlane = call bfloat @llvm.amdgcn.readlane.bf16(bfloat %src0, i32 %src1) + call void asm sideeffect "; use $0", "s"(bfloat %readlane) + ret void +} + +define amdgpu_kernel void @test_readlane_f16(half %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f16( +; CURRENT-CHECK-SAME: half [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3]] { +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f16( +; PASS-CHECK-SAME: half [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f16( +; DCE-CHECK-SAME: half [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %readlane = call half @llvm.amdgcn.readlane.f16(half %src0, i32 %src1) + call void asm sideeffect "; use $0", "s"(half %readlane) + ret void +} + +define amdgpu_kernel void @test_readlane_f32(float %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f32( +; CURRENT-CHECK-SAME: float [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3]] { +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f32( +; PASS-CHECK-SAME: float [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f32( +; DCE-CHECK-SAME: float [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %readlane = call float @llvm.amdgcn.readlane.f32(float %src0, i32 %src1) + call void asm sideeffect "; use $0", "s"(float %readlane) + ret void +} + +define amdgpu_kernel void @test_readlane_f64(double %src0, i32 %src1) { +; CURRENT-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f64( +; CURRENT-CHECK-SAME: double [[SRC0:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3]] { +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f64( +; PASS-CHECK-SAME: double [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define amdgpu_kernel void @test_readlane_f64( +; DCE-CHECK-SAME: double [[SRC0:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %readlane = call double @llvm.amdgcn.readlane.f64(double %src0, i32 %src1) + call void asm sideeffect "; use $0", "s"(double %readlane) + ret void +} +; All such cases can be optimised, given generic way to query getDeclarationIfExists() +define void @test_readlane_v8i16(ptr addrspace(1) %out, <8 x i16> %src, i32 %src1) { +; CURRENT-CHECK-LABEL: define void @test_readlane_v8i16( +; CURRENT-CHECK-SAME: ptr addrspace(1) readnone captures(none) [[OUT:%.*]], <8 x i16> [[SRC:%.*]], i32 [[SRC1:%.*]]) local_unnamed_addr #[[ATTR3]] { +; CURRENT-CHECK-NEXT: [[X:%.*]] = tail call <8 x i16> @llvm.amdgcn.readlane.v8i16(<8 x i16> [[SRC]], i32 [[SRC1]]) +; CURRENT-CHECK-NEXT: tail call void asm sideeffect " +; CURRENT-CHECK-NEXT: ret void +; +; PASS-CHECK-LABEL: define void @test_readlane_v8i16( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], <8 x i16> [[SRC:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; PASS-CHECK-NEXT: [[X:%.*]] = call <8 x i16> @llvm.amdgcn.readlane.v8i16(<8 x i16> [[SRC]], i32 [[SRC1]]) +; PASS-CHECK-NEXT: call void asm sideeffect " +; PASS-CHECK-NEXT: ret void +; +; DCE-CHECK-LABEL: define void @test_readlane_v8i16( +; DCE-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], <8 x i16> [[SRC:%.*]], i32 [[SRC1:%.*]]) #[[ATTR0]] { +; DCE-CHECK-NEXT: [[X:%.*]] = call <8 x i16> @llvm.amdgcn.readlane.v8i16(<8 x i16> [[SRC]], i32 [[SRC1]]) +; DCE-CHECK-NEXT: call void asm sideeffect " +; DCE-CHECK-NEXT: ret void +; + %x = call <8 x i16> @llvm.amdgcn.readlane.v8i16(<8 x i16> %src, i32 %src1) + call void asm sideeffect "; use $0", "s"(<8 x i16> %x) + ret void +} diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-temporal-divergence.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-temporal-divergence.ll new file mode 100644 index 0000000..2fde3e3 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-uniform-temporal-divergence.ll @@ -0,0 +1,57 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s -check-prefix=PASS-CHECK +; RUN: opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=amdgpu-uniform-intrinsic-combine,instcombine,early-cse,simplifycfg -S < %s | FileCheck %s -check-prefix=COMB-CHECK + +; This should not be optimized +define amdgpu_cs void @temporal_divergence(ptr addrspace(1) %out, i32 %n) { +; PASS-CHECK-LABEL: define amdgpu_cs void @temporal_divergence( +; PASS-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[N:%.*]]) #[[ATTR0:[0-9]+]] { +; PASS-CHECK-NEXT: [[ENTRY:.*]]: +; PASS-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; PASS-CHECK-NEXT: br label %[[H:.*]] +; PASS-CHECK: [[H]]: +; PASS-CHECK-NEXT: [[UNI_MERGE_H:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[UNI_INC:%.*]], %[[H]] ] +; PASS-CHECK-NEXT: [[UNI_INC]] = add i32 [[UNI_MERGE_H]], 1 +; PASS-CHECK-NEXT: [[DIV_EXITX:%.*]] = icmp eq i32 [[TID]], 0 +; PASS-CHECK-NEXT: br i1 [[DIV_EXITX]], label %[[X:.*]], label %[[H]] +; PASS-CHECK: [[X]]: +; PASS-CHECK-NEXT: [[UNI_JOIN:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[UNI_INC]]) +; PASS-CHECK-NEXT: [[JOIN_USER:%.*]] = add i32 [[UNI_JOIN]], 5 +; PASS-CHECK-NEXT: store i32 [[JOIN_USER]], ptr addrspace(1) [[OUT]], align 4 +; PASS-CHECK-NEXT: ret void +; +; COMB-CHECK-LABEL: define amdgpu_cs void @temporal_divergence( +; COMB-CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i32 [[N:%.*]]) #[[ATTR0:[0-9]+]] { +; COMB-CHECK-NEXT: [[ENTRY:.*]]: +; COMB-CHECK-NEXT: [[TID:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() +; COMB-CHECK-NEXT: br label %[[H:.*]] +; COMB-CHECK: [[H]]: +; COMB-CHECK-NEXT: [[UNI_MERGE_H:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[UNI_INC:%.*]], %[[H]] ] +; COMB-CHECK-NEXT: [[UNI_INC]] = add i32 [[UNI_MERGE_H]], 1 +; COMB-CHECK-NEXT: [[DIV_EXITX:%.*]] = icmp eq i32 [[TID]], 0 +; COMB-CHECK-NEXT: br i1 [[DIV_EXITX]], label %[[X:.*]], label %[[H]] +; COMB-CHECK: [[X]]: +; COMB-CHECK-NEXT: [[UNI_JOIN:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[UNI_INC]]) +; COMB-CHECK-NEXT: [[JOIN_USER:%.*]] = add i32 [[UNI_JOIN]], 5 +; COMB-CHECK-NEXT: store i32 [[JOIN_USER]], ptr addrspace(1) [[OUT]], align 4 +; COMB-CHECK-NEXT: ret void +; +entry: + %tid = call i32 @llvm.amdgcn.workitem.id.x() + br label %H + +H: + %uni.merge.h = phi i32 [ 0, %entry ], [ %uni.inc, %H ] + %uni.inc = add i32 %uni.merge.h, 1 + %div.exitx = icmp eq i32 %tid, 0 + br i1 %div.exitx, label %X, label %H ; divergent branch + +X: + %uni.join = call i32 @llvm.amdgcn.readfirstlane.i32(i32 %uni.inc) + %join.user = add i32 %uni.join, 5 + store i32 %join.user, ptr addrspace(1) %out + ret void +} + +declare i32 @llvm.amdgcn.workitem.id.x() +declare i32 @llvm.amdgcn.readfirstlane.i32(i32) 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/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/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/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/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 } |