diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU')
70 files changed, 6620 insertions, 5727 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-used-outside-loop.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-used-outside-loop.ll index dd01112..c1e6b4f 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-used-outside-loop.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-divergent-i1-used-outside-loop.ll @@ -21,14 +21,14 @@ define void @divergent_i1_phi_used_outside_loop(float %val, float %pre.cond.val, ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: v_cvt_f32_u32_e32 v1, s6 ; GFX10-NEXT: s_mov_b32 s8, exec_lo +; GFX10-NEXT: s_mov_b32 s9, s5 ; GFX10-NEXT: s_add_i32 s6, s6, 1 -; GFX10-NEXT: s_xor_b32 s8, s5, s8 +; GFX10-NEXT: s_xor_b32 s5, s5, s8 ; GFX10-NEXT: v_cmp_gt_f32_e32 vcc_lo, v1, v0 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 s7, s7, exec_lo -; GFX10-NEXT: s_and_b32 s9, exec_lo, s5 -; GFX10-NEXT: s_mov_b32 s5, s8 -; GFX10-NEXT: s_or_b32 s7, s7, s9 +; GFX10-NEXT: s_and_b32 s8, exec_lo, s9 +; GFX10-NEXT: s_or_b32 s7, s7, s8 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB0_1 ; GFX10-NEXT: ; %bb.2: ; %exit @@ -240,11 +240,11 @@ define void @divergent_i1_xor_used_outside_loop_larger_loop_body(i32 %num.elts, ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0 ; GFX10-NEXT: s_mov_b32 s6, exec_lo -; GFX10-NEXT: s_mov_b32 s8, 0 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_and_saveexec_b32 s7, vcc_lo ; GFX10-NEXT: s_cbranch_execz .LBB4_6 ; GFX10-NEXT: ; %bb.1: ; %loop.start.preheader -; GFX10-NEXT: s_mov_b32 s4, 0 +; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: ; implicit-def: $sgpr10 ; GFX10-NEXT: ; implicit-def: $sgpr11 ; GFX10-NEXT: ; implicit-def: $sgpr9 @@ -345,8 +345,8 @@ define void @divergent_i1_icmp_used_outside_loop(i32 %v0, i32 %v1, ptr addrspace ; GFX10-LABEL: divergent_i1_icmp_used_outside_loop: ; GFX10: ; %bb.0: ; %entry ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: s_mov_b32 s6, 0 ; GFX10-NEXT: s_mov_b32 s4, 0 +; GFX10-NEXT: s_mov_b32 s6, 0 ; GFX10-NEXT: ; implicit-def: $sgpr7 ; GFX10-NEXT: s_branch .LBB5_2 ; GFX10-NEXT: .LBB5_1: ; %Flow @@ -457,8 +457,8 @@ define amdgpu_ps void @divergent_i1_freeze_used_outside_loop(i32 %n, ptr addrspa ; GFX10-LABEL: divergent_i1_freeze_used_outside_loop: ; GFX10: ; %bb.0: ; %entry ; GFX10-NEXT: s_mov_b32 s1, exec_lo -; GFX10-NEXT: s_mov_b32 s2, 0 ; GFX10-NEXT: s_mov_b32 s0, 0 +; GFX10-NEXT: s_mov_b32 s2, 0 ; GFX10-NEXT: ; implicit-def: $sgpr4 ; GFX10-NEXT: ; implicit-def: $sgpr3 ; GFX10-NEXT: s_branch .LBB6_2 @@ -534,8 +534,8 @@ exit: define amdgpu_cs void @loop_with_1break(ptr addrspace(1) %x, ptr addrspace(1) %a, ptr addrspace(1) %a.break) { ; GFX10-LABEL: loop_with_1break: ; GFX10: ; %bb.0: ; %entry -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_mov_b32 s0, 0 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: ; implicit-def: $sgpr6 ; GFX10-NEXT: ; implicit-def: $sgpr7 ; GFX10-NEXT: ; implicit-def: $sgpr5 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-structurizer.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-structurizer.ll index fd08ab8..484536b 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-structurizer.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-structurizer.ll @@ -106,8 +106,8 @@ exit: define amdgpu_cs void @loop_with_1break(ptr addrspace(1) %x, ptr addrspace(1) %a) { ; GFX10-LABEL: loop_with_1break: ; GFX10: ; %bb.0: ; %entry -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_mov_b32 s0, 0 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: ; implicit-def: $sgpr5 ; GFX10-NEXT: s_branch .LBB2_2 ; GFX10-NEXT: .LBB2_1: ; %Flow @@ -180,8 +180,8 @@ exit: define amdgpu_cs void @loop_with_2breaks(ptr addrspace(1) %x, ptr addrspace(1) %a, ptr addrspace(1) %b) { ; GFX10-LABEL: loop_with_2breaks: ; GFX10: ; %bb.0: ; %entry -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_mov_b32 s0, 0 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: ; implicit-def: $sgpr5 ; GFX10-NEXT: s_branch .LBB3_3 ; GFX10-NEXT: .LBB3_1: ; %Flow3 @@ -278,8 +278,8 @@ exit: define amdgpu_cs void @loop_with_3breaks(ptr addrspace(1) %x, ptr addrspace(1) %a, ptr addrspace(1) %b, ptr addrspace(1) %c) { ; GFX10-LABEL: loop_with_3breaks: ; GFX10: ; %bb.0: ; %entry -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_mov_b32 s0, 0 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: ; implicit-def: $sgpr5 ; GFX10-NEXT: s_branch .LBB4_4 ; GFX10-NEXT: .LBB4_1: ; %Flow5 @@ -404,8 +404,8 @@ exit: define amdgpu_cs void @loop_with_div_break_with_body(ptr addrspace(1) %x, ptr addrspace(1) %a, ptr addrspace(1) %a.break) { ; GFX10-LABEL: loop_with_div_break_with_body: ; GFX10: ; %bb.0: ; %entry -; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: s_mov_b32 s0, 0 +; GFX10-NEXT: s_mov_b32 s4, 0 ; GFX10-NEXT: ; implicit-def: $sgpr6 ; GFX10-NEXT: ; implicit-def: $sgpr7 ; GFX10-NEXT: ; implicit-def: $sgpr5 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-temporal-divergent-i1.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-temporal-divergent-i1.ll index d13d6a1..69baf61 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-temporal-divergent-i1.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/divergence-temporal-divergent-i1.ll @@ -101,8 +101,8 @@ define amdgpu_cs void @loop_with_1break(ptr addrspace(1) %x, i32 %x.size, ptr ad ; GFX10-LABEL: loop_with_1break: ; GFX10: ; %bb.0: ; %entry ; GFX10-NEXT: v_mov_b32_e32 v3, 0 -; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: s_mov_b32 s4, 0 +; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: ; implicit-def: $sgpr10 ; GFX10-NEXT: ; implicit-def: $sgpr9 ; GFX10-NEXT: s_branch .LBB2_3 @@ -197,14 +197,14 @@ define void @nested_loops_temporal_divergence_inner(float %pre.cond.val, i32 %n. ; GFX10-LABEL: nested_loops_temporal_divergence_inner: ; GFX10: ; %bb.0: ; %entry ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_cmp_lt_f32_e64 s8, 1.0, v0 -; GFX10-NEXT: s_mov_b32 s5, 0 +; GFX10-NEXT: v_cmp_lt_f32_e64 s5, 1.0, v0 ; GFX10-NEXT: s_mov_b32 s6, 0 +; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: .LBB3_1: ; %OuterHeader ; GFX10-NEXT: ; =>This Loop Header: Depth=1 ; GFX10-NEXT: ; Child Loop BB3_2 Depth 2 ; GFX10-NEXT: s_ashr_i32 s7, s6, 31 -; GFX10-NEXT: s_mov_b32 s4, s8 +; GFX10-NEXT: s_mov_b32 s4, s5 ; GFX10-NEXT: s_lshl_b64 s[10:11], s[6:7], 2 ; GFX10-NEXT: ; implicit-def: $sgpr9 ; GFX10-NEXT: v_mov_b32_e32 v6, s10 @@ -239,13 +239,13 @@ define void @nested_loops_temporal_divergence_inner(float %pre.cond.val, i32 %n. ; GFX10-NEXT: s_add_i32 s6, s6, 1 ; GFX10-NEXT: v_add_co_u32 v6, s4, v4, v6 ; GFX10-NEXT: v_add_co_ci_u32_e64 v7, s4, v5, v7, s4 -; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 +; GFX10-NEXT: s_or_b32 s8, vcc_lo, s8 ; GFX10-NEXT: flat_store_byte v[6:7], v0 ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 +; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s8 ; GFX10-NEXT: s_cbranch_execnz .LBB3_1 ; GFX10-NEXT: ; %bb.4: ; %exit -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s5 +; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s8 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_setpc_b64 s[30:31] entry: @@ -288,14 +288,14 @@ define void @nested_loops_temporal_divergence_outer(float %pre.cond.val, i32 %n. ; GFX10-LABEL: nested_loops_temporal_divergence_outer: ; GFX10: ; %bb.0: ; %entry ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_cmp_lt_f32_e64 s8, 1.0, v0 -; GFX10-NEXT: s_mov_b32 s5, 0 +; GFX10-NEXT: v_cmp_lt_f32_e64 s5, 1.0, v0 ; GFX10-NEXT: s_mov_b32 s6, 0 +; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: .LBB4_1: ; %OuterHeader ; GFX10-NEXT: ; =>This Loop Header: Depth=1 ; GFX10-NEXT: ; Child Loop BB4_2 Depth 2 ; GFX10-NEXT: s_ashr_i32 s7, s6, 31 -; GFX10-NEXT: s_mov_b32 s4, s8 +; GFX10-NEXT: s_mov_b32 s4, s5 ; GFX10-NEXT: s_lshl_b64 s[10:11], s[6:7], 2 ; GFX10-NEXT: ; implicit-def: $sgpr9 ; GFX10-NEXT: v_mov_b32_e32 v6, s10 @@ -330,13 +330,13 @@ define void @nested_loops_temporal_divergence_outer(float %pre.cond.val, i32 %n. ; GFX10-NEXT: s_add_i32 s6, s6, 1 ; GFX10-NEXT: v_add_co_u32 v6, s4, v4, v6 ; GFX10-NEXT: v_add_co_ci_u32_e64 v7, s4, v5, v7, s4 -; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 +; GFX10-NEXT: s_or_b32 s8, vcc_lo, s8 ; GFX10-NEXT: flat_store_byte v[6:7], v0 ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 +; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s8 ; GFX10-NEXT: s_cbranch_execnz .LBB4_1 ; GFX10-NEXT: ; %bb.4: ; %exit -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s5 +; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s8 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_setpc_b64 s[30:31] entry: @@ -379,15 +379,15 @@ define void @nested_loops_temporal_divergence_both(float %pre.cond.val, i32 %n.i ; GFX10-LABEL: nested_loops_temporal_divergence_both: ; GFX10: ; %bb.0: ; %entry ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-NEXT: v_cmp_lt_f32_e64 s8, 1.0, v0 -; GFX10-NEXT: s_mov_b32 s5, 0 +; GFX10-NEXT: v_cmp_lt_f32_e64 s5, 1.0, v0 ; GFX10-NEXT: s_mov_b32 s6, 0 +; GFX10-NEXT: s_mov_b32 s8, 0 ; GFX10-NEXT: ; implicit-def: $sgpr9 ; GFX10-NEXT: .LBB5_1: ; %OuterHeader ; GFX10-NEXT: ; =>This Loop Header: Depth=1 ; GFX10-NEXT: ; Child Loop BB5_2 Depth 2 ; GFX10-NEXT: s_ashr_i32 s7, s6, 31 -; GFX10-NEXT: s_mov_b32 s4, s8 +; GFX10-NEXT: s_mov_b32 s4, s5 ; GFX10-NEXT: s_lshl_b64 s[10:11], s[6:7], 2 ; GFX10-NEXT: v_mov_b32_e32 v8, s10 ; GFX10-NEXT: v_mov_b32_e32 v9, s11 @@ -421,13 +421,13 @@ define void @nested_loops_temporal_divergence_both(float %pre.cond.val, i32 %n.i ; GFX10-NEXT: s_add_i32 s6, s6, 1 ; GFX10-NEXT: v_add_co_u32 v8, s4, v4, v8 ; GFX10-NEXT: v_add_co_ci_u32_e64 v9, s4, v5, v9, s4 -; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 +; GFX10-NEXT: s_or_b32 s8, vcc_lo, s8 ; GFX10-NEXT: flat_store_byte v[8:9], v0 ; GFX10-NEXT: s_waitcnt_depctr 0xffe3 -; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 +; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s8 ; GFX10-NEXT: s_cbranch_execnz .LBB5_1 ; GFX10-NEXT: ; %bb.4: ; %exit -; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s5 +; GFX10-NEXT: s_or_b32 exec_lo, exec_lo, s8 ; GFX10-NEXT: flat_store_byte v[6:7], v0 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_setpc_b64 s[30:31] diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-mui.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-mui.ll index 5240bf4..9aaa9635 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-mui.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-mui.ll @@ -547,8 +547,8 @@ define amdgpu_cs void @loop_with_2breaks(ptr addrspace(1) %x, ptr addrspace(1) % ; ; NEW_RBS-LABEL: loop_with_2breaks: ; NEW_RBS: ; %bb.0: ; %entry -; NEW_RBS-NEXT: s_mov_b32 s4, 0 ; NEW_RBS-NEXT: s_mov_b32 s0, 0 +; NEW_RBS-NEXT: s_mov_b32 s4, 0 ; NEW_RBS-NEXT: ; implicit-def: $sgpr5 ; NEW_RBS-NEXT: s_branch .LBB16_3 ; NEW_RBS-NEXT: .LBB16_1: ; %Flow3 diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll index 54b1554..df77e7d 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll @@ -40,34 +40,33 @@ define amdgpu_kernel void @udiv_i32(ptr addrspace(1) %out, i32 %x, i32 %y) { ; GFX6-LABEL: udiv_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s3 -; GFX6-NEXT: s_sub_i32 s4, 0, s3 -; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s5 +; GFX6-NEXT: s_sub_i32 s2, 0, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 ; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, v0 ; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_lo_u32 v1, s4, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: v_mul_lo_u32 v1, s2, v0 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GFX6-NEXT: v_mul_hi_u32 v0, s2, v0 -; GFX6-NEXT: v_readfirstlane_b32 s0, v0 -; GFX6-NEXT: s_mul_i32 s0, s0, s3 -; GFX6-NEXT: s_sub_i32 s0, s2, s0 -; GFX6-NEXT: s_sub_i32 s1, s0, s3 +; GFX6-NEXT: v_mul_hi_u32 v0, s4, v0 +; GFX6-NEXT: v_readfirstlane_b32 s6, v0 +; GFX6-NEXT: s_mul_i32 s6, s6, s5 +; GFX6-NEXT: s_sub_i32 s4, s4, s6 +; GFX6-NEXT: s_sub_i32 s6, s4, s5 ; GFX6-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; GFX6-NEXT: s_cmp_ge_u32 s0, s3 +; GFX6-NEXT: s_cmp_ge_u32 s4, s5 ; GFX6-NEXT: s_cselect_b64 vcc, -1, 0 ; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; GFX6-NEXT: s_cselect_b32 s0, s1, s0 +; GFX6-NEXT: s_cselect_b32 s4, s6, s4 ; GFX6-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; GFX6-NEXT: s_cmp_ge_u32 s0, s3 +; GFX6-NEXT: s_cmp_ge_u32 s4, s5 ; GFX6-NEXT: s_cselect_b64 vcc, -1, 0 ; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: udiv_i32: @@ -138,31 +137,30 @@ define amdgpu_kernel void @urem_i32(ptr addrspace(1) %out, i32 %x, i32 %y) { ; GFX6-LABEL: urem_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s3 -; GFX6-NEXT: s_sub_i32 s4, 0, s3 -; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s5 +; GFX6-NEXT: s_sub_i32 s2, 0, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 ; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, v0 ; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_lo_u32 v1, s4, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: v_mul_lo_u32 v1, s2, v0 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GFX6-NEXT: v_mul_hi_u32 v0, s2, v0 -; GFX6-NEXT: v_readfirstlane_b32 s0, v0 -; GFX6-NEXT: s_mul_i32 s0, s0, s3 -; GFX6-NEXT: s_sub_i32 s0, s2, s0 -; GFX6-NEXT: s_sub_i32 s1, s0, s3 -; GFX6-NEXT: s_cmp_ge_u32 s0, s3 -; GFX6-NEXT: s_cselect_b32 s0, s1, s0 -; GFX6-NEXT: s_sub_i32 s1, s0, s3 -; GFX6-NEXT: s_cmp_ge_u32 s0, s3 -; GFX6-NEXT: s_cselect_b32 s0, s1, s0 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: v_mul_hi_u32 v0, s4, v0 +; GFX6-NEXT: v_readfirstlane_b32 s6, v0 +; GFX6-NEXT: s_mul_i32 s6, s6, s5 +; GFX6-NEXT: s_sub_i32 s4, s4, s6 +; GFX6-NEXT: s_sub_i32 s6, s4, s5 +; GFX6-NEXT: s_cmp_ge_u32 s4, s5 +; GFX6-NEXT: s_cselect_b32 s4, s6, s4 +; GFX6-NEXT: s_sub_i32 s6, s4, s5 +; GFX6-NEXT: s_cmp_ge_u32 s4, s5 +; GFX6-NEXT: s_cselect_b32 s4, s6, s4 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: urem_i32: @@ -242,40 +240,39 @@ define amdgpu_kernel void @sdiv_i32(ptr addrspace(1) %out, i32 %x, i32 %y) { ; GFX6-LABEL: sdiv_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_abs_i32 s8, s3 -; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s8 -; GFX6-NEXT: s_sub_i32 s4, 0, s8 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: s_xor_b32 s1, s2, s3 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_abs_i32 s6, s5 +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s6 +; GFX6-NEXT: s_sub_i32 s2, 0, s6 +; GFX6-NEXT: s_abs_i32 s7, s4 +; GFX6-NEXT: s_xor_b32 s4, s4, s5 ; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GFX6-NEXT: s_ashr_i32 s1, s1, 31 +; GFX6-NEXT: s_ashr_i32 s4, s4, 31 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 ; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_lo_u32 v1, s4, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_abs_i32 s0, s2 +; GFX6-NEXT: v_mul_lo_u32 v1, s2, v0 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GFX6-NEXT: v_mul_hi_u32 v0, s0, v0 -; GFX6-NEXT: v_readfirstlane_b32 s2, v0 -; GFX6-NEXT: s_mul_i32 s2, s2, s8 -; GFX6-NEXT: s_sub_i32 s0, s0, s2 -; GFX6-NEXT: s_sub_i32 s2, s0, s8 +; GFX6-NEXT: v_mul_hi_u32 v0, s7, v0 +; GFX6-NEXT: v_readfirstlane_b32 s5, v0 +; GFX6-NEXT: s_mul_i32 s5, s5, s6 +; GFX6-NEXT: s_sub_i32 s5, s7, s5 +; GFX6-NEXT: s_sub_i32 s7, s5, s6 ; GFX6-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; GFX6-NEXT: s_cmp_ge_u32 s0, s8 +; GFX6-NEXT: s_cmp_ge_u32 s5, s6 ; GFX6-NEXT: s_cselect_b64 vcc, -1, 0 ; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; GFX6-NEXT: s_cselect_b32 s0, s2, s0 +; GFX6-NEXT: s_cselect_b32 s5, s7, s5 ; GFX6-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; GFX6-NEXT: s_cmp_ge_u32 s0, s8 +; GFX6-NEXT: s_cmp_ge_u32 s5, s6 ; GFX6-NEXT: s_cselect_b64 vcc, -1, 0 ; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; GFX6-NEXT: v_xor_b32_e32 v0, s1, v0 -; GFX6-NEXT: v_subrev_i32_e32 v0, vcc, s1, v0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: v_xor_b32_e32 v0, s4, v0 +; GFX6-NEXT: v_subrev_i32_e32 v0, vcc, s4, v0 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: sdiv_i32: @@ -360,36 +357,35 @@ define amdgpu_kernel void @srem_i32(ptr addrspace(1) %out, i32 %x, i32 %y) { ; GFX6-LABEL: srem_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_abs_i32 s3, s3 -; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s3 -; GFX6-NEXT: s_sub_i32 s4, 0, s3 -; GFX6-NEXT: s_abs_i32 s8, s2 -; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_abs_i32 s5, s5 +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s5 +; GFX6-NEXT: s_sub_i32 s2, 0, s5 +; GFX6-NEXT: s_abs_i32 s6, s4 +; GFX6-NEXT: s_ashr_i32 s4, s4, 31 ; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, v0 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 ; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_lo_u32 v1, s4, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_ashr_i32 s0, s2, 31 +; GFX6-NEXT: v_mul_lo_u32 v1, s2, v0 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GFX6-NEXT: v_mul_hi_u32 v0, s8, v0 -; GFX6-NEXT: v_readfirstlane_b32 s1, v0 -; GFX6-NEXT: s_mul_i32 s1, s1, s3 -; GFX6-NEXT: s_sub_i32 s1, s8, s1 -; GFX6-NEXT: s_sub_i32 s2, s1, s3 -; GFX6-NEXT: s_cmp_ge_u32 s1, s3 -; GFX6-NEXT: s_cselect_b32 s1, s2, s1 -; GFX6-NEXT: s_sub_i32 s2, s1, s3 -; GFX6-NEXT: s_cmp_ge_u32 s1, s3 -; GFX6-NEXT: s_cselect_b32 s1, s2, s1 -; GFX6-NEXT: s_xor_b32 s1, s1, s0 -; GFX6-NEXT: s_sub_i32 s0, s1, s0 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: v_mul_hi_u32 v0, s6, v0 +; GFX6-NEXT: v_readfirstlane_b32 s7, v0 +; GFX6-NEXT: s_mul_i32 s7, s7, s5 +; GFX6-NEXT: s_sub_i32 s6, s6, s7 +; GFX6-NEXT: s_sub_i32 s7, s6, s5 +; GFX6-NEXT: s_cmp_ge_u32 s6, s5 +; GFX6-NEXT: s_cselect_b32 s6, s7, s6 +; GFX6-NEXT: s_sub_i32 s7, s6, s5 +; GFX6-NEXT: s_cmp_ge_u32 s6, s5 +; GFX6-NEXT: s_cselect_b32 s5, s7, s6 +; GFX6-NEXT: s_xor_b32 s5, s5, s4 +; GFX6-NEXT: s_sub_i32 s4, s5, s4 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: srem_i32: @@ -5462,15 +5458,14 @@ define amdgpu_kernel void @udiv_i32_pow2_shl_denom(ptr addrspace(1) %out, i32 %x ; GFX6-LABEL: udiv_i32_pow2_shl_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_add_i32 s0, s3, 12 -; GFX6-NEXT: s_lshr_b32 s0, s2, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_add_i32 s5, s5, 12 +; GFX6-NEXT: s_lshr_b32 s4, s4, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: udiv_i32_pow2_shl_denom: @@ -5503,16 +5498,15 @@ define amdgpu_kernel void @udiv_v2i32_pow2k_denom(ptr addrspace(1) %out, <2 x i3 ; GFX6-LABEL: udiv_v2i32_pow2k_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: s_lshr_b32 s0, s2, 12 -; GFX6-NEXT: s_lshr_b32 s1, s3, 12 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: v_mov_b32_e32 v1, s1 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_lshr_b32 s4, s4, 12 +; GFX6-NEXT: s_lshr_b32 s5, s5, 12 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: udiv_v2i32_pow2k_denom: @@ -5546,19 +5540,18 @@ define amdgpu_kernel void @udiv_v2i32_mixed_pow2k_denom(ptr addrspace(1) %out, < ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GFX6-NEXT: v_mov_b32_e32 v0, 0x100101 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_mul_hi_u32 v0, s3, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_lshr_b32 s0, s2, 12 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_sub_i32_e32 v1, vcc, s3, v0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: v_mul_hi_u32 v0, s5, v0 +; GFX6-NEXT: s_lshr_b32 s4, s4, 12 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_sub_i32_e32 v1, vcc, s5, v0 ; GFX6-NEXT: v_lshrrev_b32_e32 v1, 1, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v1, v0 ; GFX6-NEXT: v_lshrrev_b32_e32 v1, 11, v0 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: udiv_v2i32_mixed_pow2k_denom: @@ -5855,16 +5848,15 @@ define amdgpu_kernel void @urem_i32_pow2_shl_denom(ptr addrspace(1) %out, i32 %x ; GFX6-LABEL: urem_i32_pow2_shl_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_lshl_b32 s0, 0x1000, s3 -; GFX6-NEXT: s_add_i32 s0, s0, -1 -; GFX6-NEXT: s_and_b32 s0, s2, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_lshl_b32 s5, 0x1000, s5 +; GFX6-NEXT: s_add_i32 s5, s5, -1 +; GFX6-NEXT: s_and_b32 s4, s4, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: urem_i32_pow2_shl_denom: @@ -5898,16 +5890,15 @@ define amdgpu_kernel void @urem_v2i32_pow2k_denom(ptr addrspace(1) %out, <2 x i3 ; GFX6-LABEL: urem_v2i32_pow2k_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: s_and_b32 s0, s2, 0xfff -; GFX6-NEXT: s_and_b32 s1, s3, 0xfff -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: v_mov_b32_e32 v1, s1 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_and_b32 s4, s4, 0xfff +; GFX6-NEXT: s_and_b32 s5, s5, 0xfff +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: urem_v2i32_pow2k_denom: @@ -6187,41 +6178,40 @@ define amdgpu_kernel void @sdiv_i32_pow2_shl_denom(ptr addrspace(1) %out, i32 %x ; GFX6-LABEL: sdiv_i32_pow2_shl_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_lshl_b32 s3, 0x1000, s3 -; GFX6-NEXT: s_abs_i32 s8, s3 -; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s8 -; GFX6-NEXT: s_sub_i32 s4, 0, s8 -; GFX6-NEXT: s_abs_i32 s9, s2 -; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_lshl_b32 s5, 0x1000, s5 +; GFX6-NEXT: s_abs_i32 s6, s5 +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s6 +; GFX6-NEXT: s_sub_i32 s2, 0, s6 +; GFX6-NEXT: s_abs_i32 s7, s4 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 ; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, v0 ; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_lo_u32 v1, s4, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: v_mul_lo_u32 v1, s2, v0 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GFX6-NEXT: v_mul_hi_u32 v0, s9, v0 -; GFX6-NEXT: v_readfirstlane_b32 s0, v0 -; GFX6-NEXT: s_mul_i32 s0, s0, s8 -; GFX6-NEXT: s_sub_i32 s0, s9, s0 -; GFX6-NEXT: s_sub_i32 s1, s0, s8 +; GFX6-NEXT: v_mul_hi_u32 v0, s7, v0 +; GFX6-NEXT: v_readfirstlane_b32 s8, v0 +; GFX6-NEXT: s_mul_i32 s8, s8, s6 +; GFX6-NEXT: s_sub_i32 s7, s7, s8 +; GFX6-NEXT: s_sub_i32 s8, s7, s6 ; GFX6-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; GFX6-NEXT: s_cmp_ge_u32 s0, s8 +; GFX6-NEXT: s_cmp_ge_u32 s7, s6 ; GFX6-NEXT: s_cselect_b64 vcc, -1, 0 -; GFX6-NEXT: s_cselect_b32 s0, s1, s0 +; GFX6-NEXT: s_cselect_b32 s7, s8, s7 ; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc ; GFX6-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; GFX6-NEXT: s_cmp_ge_u32 s0, s8 +; GFX6-NEXT: s_cmp_ge_u32 s7, s6 ; GFX6-NEXT: s_cselect_b64 vcc, -1, 0 -; GFX6-NEXT: s_xor_b32 s0, s2, s3 +; GFX6-NEXT: s_xor_b32 s4, s4, s5 ; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; GFX6-NEXT: s_ashr_i32 s0, s0, 31 -; GFX6-NEXT: v_xor_b32_e32 v0, s0, v0 -; GFX6-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_ashr_i32 s4, s4, 31 +; GFX6-NEXT: v_xor_b32_e32 v0, s4, v0 +; GFX6-NEXT: v_subrev_i32_e32 v0, vcc, s4, v0 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: sdiv_i32_pow2_shl_denom: @@ -6279,22 +6269,21 @@ define amdgpu_kernel void @sdiv_v2i32_pow2k_denom(ptr addrspace(1) %out, <2 x i3 ; GFX6-LABEL: sdiv_v2i32_pow2k_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: s_ashr_i32 s0, s2, 31 -; GFX6-NEXT: s_ashr_i32 s1, s3, 31 -; GFX6-NEXT: s_lshr_b32 s0, s0, 20 -; GFX6-NEXT: s_lshr_b32 s1, s1, 20 -; GFX6-NEXT: s_add_i32 s0, s2, s0 -; GFX6-NEXT: s_add_i32 s1, s3, s1 -; GFX6-NEXT: s_ashr_i32 s0, s0, 12 -; GFX6-NEXT: s_ashr_i32 s1, s1, 12 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: v_mov_b32_e32 v1, s1 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_ashr_i32 s6, s4, 31 +; GFX6-NEXT: s_ashr_i32 s7, s5, 31 +; GFX6-NEXT: s_lshr_b32 s6, s6, 20 +; GFX6-NEXT: s_lshr_b32 s7, s7, 20 +; GFX6-NEXT: s_add_i32 s4, s4, s6 +; GFX6-NEXT: s_add_i32 s5, s5, s7 +; GFX6-NEXT: s_ashr_i32 s4, s4, 12 +; GFX6-NEXT: s_ashr_i32 s5, s5, 12 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: sdiv_v2i32_pow2k_denom: @@ -6334,22 +6323,21 @@ define amdgpu_kernel void @ssdiv_v2i32_mixed_pow2k_denom(ptr addrspace(1) %out, ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GFX6-NEXT: v_mov_b32_e32 v0, 0x80080081 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_mul_hi_i32 v0, s3, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_ashr_i32 s0, s2, 31 -; GFX6-NEXT: s_lshr_b32 s0, s0, 20 -; GFX6-NEXT: s_add_i32 s0, s2, s0 -; GFX6-NEXT: v_add_i32_e32 v0, vcc, s3, v0 -; GFX6-NEXT: s_ashr_i32 s0, s0, 12 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: v_mul_hi_i32 v0, s5, v0 +; GFX6-NEXT: s_ashr_i32 s6, s4, 31 +; GFX6-NEXT: s_lshr_b32 s6, s6, 20 +; GFX6-NEXT: s_add_i32 s4, s4, s6 +; GFX6-NEXT: v_add_i32_e32 v0, vcc, s5, v0 +; GFX6-NEXT: s_ashr_i32 s4, s4, 12 ; GFX6-NEXT: v_lshrrev_b32_e32 v1, 31, v0 ; GFX6-NEXT: v_ashrrev_i32_e32 v0, 11, v0 -; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_add_i32_e32 v1, vcc, v0, v1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: ssdiv_v2i32_mixed_pow2k_denom: @@ -6700,37 +6688,36 @@ define amdgpu_kernel void @srem_i32_pow2_shl_denom(ptr addrspace(1) %out, i32 %x ; GFX6-LABEL: srem_i32_pow2_shl_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_lshl_b32 s3, 0x1000, s3 -; GFX6-NEXT: s_abs_i32 s3, s3 -; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s3 -; GFX6-NEXT: s_sub_i32 s4, 0, s3 -; GFX6-NEXT: s_abs_i32 s8, s2 -; GFX6-NEXT: s_mov_b32 s5, s1 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_lshl_b32 s2, 0x1000, s5 +; GFX6-NEXT: s_abs_i32 s5, s2 +; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s5 +; GFX6-NEXT: s_sub_i32 s2, 0, s5 +; GFX6-NEXT: s_abs_i32 s6, s4 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 ; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, v0 ; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_lo_u32 v1, s4, v0 -; GFX6-NEXT: s_mov_b32 s4, s0 +; GFX6-NEXT: v_mul_lo_u32 v1, s2, v0 +; GFX6-NEXT: s_mov_b32 s2, -1 ; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1 ; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GFX6-NEXT: v_mul_hi_u32 v0, s8, v0 -; GFX6-NEXT: v_readfirstlane_b32 s0, v0 -; GFX6-NEXT: s_mul_i32 s0, s0, s3 -; GFX6-NEXT: s_sub_i32 s0, s8, s0 -; GFX6-NEXT: s_sub_i32 s1, s0, s3 -; GFX6-NEXT: s_cmp_ge_u32 s0, s3 -; GFX6-NEXT: s_cselect_b32 s0, s1, s0 -; GFX6-NEXT: s_sub_i32 s1, s0, s3 -; GFX6-NEXT: s_cmp_ge_u32 s0, s3 -; GFX6-NEXT: s_cselect_b32 s0, s1, s0 -; GFX6-NEXT: s_ashr_i32 s1, s2, 31 -; GFX6-NEXT: s_xor_b32 s0, s0, s1 -; GFX6-NEXT: s_sub_i32 s0, s0, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: v_mul_hi_u32 v0, s6, v0 +; GFX6-NEXT: v_readfirstlane_b32 s7, v0 +; GFX6-NEXT: s_mul_i32 s7, s7, s5 +; GFX6-NEXT: s_sub_i32 s6, s6, s7 +; GFX6-NEXT: s_sub_i32 s7, s6, s5 +; GFX6-NEXT: s_cmp_ge_u32 s6, s5 +; GFX6-NEXT: s_cselect_b32 s6, s7, s6 +; GFX6-NEXT: s_sub_i32 s7, s6, s5 +; GFX6-NEXT: s_cmp_ge_u32 s6, s5 +; GFX6-NEXT: s_cselect_b32 s5, s7, s6 +; GFX6-NEXT: s_ashr_i32 s4, s4, 31 +; GFX6-NEXT: s_xor_b32 s5, s5, s4 +; GFX6-NEXT: s_sub_i32 s4, s5, s4 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: srem_i32_pow2_shl_denom: @@ -6785,24 +6772,23 @@ define amdgpu_kernel void @srem_v2i32_pow2k_denom(ptr addrspace(1) %out, <2 x i3 ; GFX6-LABEL: srem_v2i32_pow2k_denom: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: s_ashr_i32 s0, s2, 31 -; GFX6-NEXT: s_ashr_i32 s1, s3, 31 -; GFX6-NEXT: s_lshr_b32 s0, s0, 20 -; GFX6-NEXT: s_lshr_b32 s1, s1, 20 -; GFX6-NEXT: s_add_i32 s0, s2, s0 -; GFX6-NEXT: s_add_i32 s1, s3, s1 -; GFX6-NEXT: s_and_b32 s0, s0, 0xfffff000 -; GFX6-NEXT: s_and_b32 s1, s1, 0xfffff000 -; GFX6-NEXT: s_sub_i32 s0, s2, s0 -; GFX6-NEXT: s_sub_i32 s1, s3, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: v_mov_b32_e32 v1, s1 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_ashr_i32 s6, s4, 31 +; GFX6-NEXT: s_lshr_b32 s6, s6, 20 +; GFX6-NEXT: s_ashr_i32 s7, s5, 31 +; GFX6-NEXT: s_add_i32 s6, s4, s6 +; GFX6-NEXT: s_lshr_b32 s7, s7, 20 +; GFX6-NEXT: s_and_b32 s6, s6, 0xfffff000 +; GFX6-NEXT: s_sub_i32 s4, s4, s6 +; GFX6-NEXT: s_add_i32 s6, s5, s7 +; GFX6-NEXT: s_and_b32 s6, s6, 0xfffff000 +; GFX6-NEXT: s_sub_i32 s5, s5, s6 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX9-LABEL: srem_v2i32_pow2k_denom: diff --git a/llvm/test/CodeGen/AMDGPU/and.ll b/llvm/test/CodeGen/AMDGPU/and.ll index 29bfc25..fe9ec8e 100644 --- a/llvm/test/CodeGen/AMDGPU/and.ll +++ b/llvm/test/CodeGen/AMDGPU/and.ll @@ -123,27 +123,25 @@ define amdgpu_kernel void @s_and_i32(ptr addrspace(1) %out, i32 %a, i32 %b) { ; GFX6-LABEL: s_and_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_and_b32 s0, s2, s3 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_and_b32 s4, s4, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX8-LABEL: s_and_i32: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; GFX8-NEXT: s_and_b32 s0, s2, s3 -; 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_mov_b64 s[4:5], s[2:3] +; GFX8-NEXT: s_and_b32 s4, s4, s5 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX8-NEXT: s_endpgm %and = and i32 %a, %b store i32 %and, ptr addrspace(1) %out, align 4 @@ -189,36 +187,34 @@ define amdgpu_kernel void @s_and_multi_use_constant_i32_0(ptr addrspace(1) %out, ; GFX6-LABEL: s_and_multi_use_constant_i32_0: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 -; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_and_b32 s0, s2, 0x12d687 -; GFX6-NEXT: s_add_i32 s0, s0, s3 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 -; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) ; GFX6-NEXT: v_mov_b32_e32 v0, 0x12d687 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_waitcnt lgkmcnt(0) +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_and_b32 s4, s4, 0x12d687 +; GFX6-NEXT: s_add_i32 s4, s4, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v1, s4 +; GFX6-NEXT: buffer_store_dword v1, off, s[0:3], 0 +; GFX6-NEXT: s_waitcnt vmcnt(0) +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_waitcnt vmcnt(0) ; GFX6-NEXT: s_endpgm ; ; GFX8-LABEL: s_and_multi_use_constant_i32_0: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_load_dwordx4 s[0:3], 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, 0x12d687 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: s_mov_b32 s4, s0 -; GFX8-NEXT: s_and_b32 s0, s2, 0x12d687 -; GFX8-NEXT: s_add_i32 s0, s0, s3 -; 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_mov_b64 s[4:5], s[2:3] +; GFX8-NEXT: s_and_b32 s4, s4, 0x12d687 +; GFX8-NEXT: s_add_i32 s4, s4, s5 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: v_mov_b32_e32 v1, s4 +; GFX8-NEXT: buffer_store_dword v1, off, s[0:3], 0 ; GFX8-NEXT: s_waitcnt vmcnt(0) -; GFX8-NEXT: v_mov_b32_e32 v0, 0x12d687 -; GFX8-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: s_endpgm %and = and i32 %a, 1234567 @@ -236,32 +232,30 @@ define amdgpu_kernel void @s_and_multi_use_constant_i32_1(ptr addrspace(1) %out, ; GFX6-LABEL: s_and_multi_use_constant_i32_1: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_and_b32 s0, s2, 0x12d687 -; GFX6-NEXT: s_add_i32 s0, s0, s3 -; GFX6-NEXT: s_add_i32 s0, s0, 0x12d687 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_and_b32 s4, s4, 0x12d687 +; GFX6-NEXT: s_add_i32 s4, s4, s5 +; GFX6-NEXT: s_add_i32 s4, s4, 0x12d687 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_waitcnt vmcnt(0) ; GFX6-NEXT: s_endpgm ; ; GFX8-LABEL: s_and_multi_use_constant_i32_1: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; GFX8-NEXT: s_and_b32 s0, s2, 0x12d687 -; GFX8-NEXT: s_add_i32 s0, s0, s3 -; GFX8-NEXT: s_add_i32 s0, s0, 0x12d687 -; 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_mov_b64 s[4:5], s[2:3] +; GFX8-NEXT: s_and_b32 s4, s4, 0x12d687 +; GFX8-NEXT: s_add_i32 s4, s4, s5 +; GFX8-NEXT: s_add_i32 s4, s4, 0x12d687 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX8-NEXT: s_waitcnt vmcnt(0) ; GFX8-NEXT: s_endpgm %and = and i32 %a, 1234567 diff --git a/llvm/test/CodeGen/AMDGPU/bf16-math.ll b/llvm/test/CodeGen/AMDGPU/bf16-math.ll index 30a7864..39618b0 100644 --- a/llvm/test/CodeGen/AMDGPU/bf16-math.ll +++ b/llvm/test/CodeGen/AMDGPU/bf16-math.ll @@ -368,7 +368,10 @@ define amdgpu_ps float @test_clamp_v2bf16_s(<2 x bfloat> inreg %src) { define amdgpu_ps bfloat @test_clamp_bf16_folding(bfloat %src) { ; GCN-LABEL: test_clamp_bf16_folding: ; GCN: ; %bb.0: -; GCN-NEXT: v_exp_bf16_e64 v0, v0 clamp +; GCN-NEXT: v_exp_bf16_e32 v0, v0 +; GCN-NEXT: v_nop +; GCN-NEXT: s_delay_alu instid0(TRANS32_DEP_1) +; GCN-NEXT: v_pk_max_num_bf16 v0, v0, v0 clamp ; GCN-NEXT: ; return to shader part epilog %exp = call bfloat @llvm.exp2.bf16(bfloat %src) %max = call bfloat @llvm.maxnum.bf16(bfloat %exp, bfloat 0.0) diff --git a/llvm/test/CodeGen/AMDGPU/bfe-patterns.ll b/llvm/test/CodeGen/AMDGPU/bfe-patterns.ll index c14678c..c0d5f8a 100644 --- a/llvm/test/CodeGen/AMDGPU/bfe-patterns.ll +++ b/llvm/test/CodeGen/AMDGPU/bfe-patterns.ll @@ -120,17 +120,17 @@ define amdgpu_kernel void @s_ubfe_sub_i32(ptr addrspace(1) %out, i32 %src, i32 % ; SI-LABEL: s_ubfe_sub_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, 0 ; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0 -; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_sub_i32 s3, 32, s3 -; SI-NEXT: s_lshl_b32 s2, s2, s3 -; SI-NEXT: s_lshr_b32 s2, s2, s3 ; SI-NEXT: v_mov_b32_e32 v1, 0 -; SI-NEXT: s_mov_b64 s[4:5], s[0:1] -; SI-NEXT: v_mov_b32_e32 v2, s2 -; SI-NEXT: buffer_store_dword v2, v[0:1], s[4:7], 0 addr64 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_sub_i32 s2, 32, s5 +; SI-NEXT: s_lshl_b32 s4, s4, s2 +; SI-NEXT: s_lshr_b32 s4, s4, s2 +; SI-NEXT: s_mov_b32 s2, 0 +; SI-NEXT: v_mov_b32_e32 v2, s4 +; SI-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_ubfe_sub_i32: @@ -160,20 +160,20 @@ define amdgpu_kernel void @s_ubfe_sub_multi_use_shl_i32(ptr addrspace(1) %out, i ; SI-LABEL: s_ubfe_sub_multi_use_shl_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s6, 0 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0 -; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_sub_i32 s3, 32, s3 -; SI-NEXT: s_lshl_b32 s2, s2, s3 -; SI-NEXT: s_lshr_b32 s3, s2, s3 ; SI-NEXT: v_mov_b32_e32 v1, 0 -; SI-NEXT: s_mov_b64 s[4:5], s[0:1] -; SI-NEXT: v_mov_b32_e32 v2, s3 -; SI-NEXT: buffer_store_dword v2, v[0:1], s[4:7], 0 addr64 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s2, 0 +; SI-NEXT: s_sub_i32 s3, 32, s5 +; SI-NEXT: s_lshl_b32 s4, s4, s3 +; SI-NEXT: s_lshr_b32 s5, s4, s3 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: v_mov_b32_e32 v2, s5 +; SI-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_waitcnt vmcnt(0) ; SI-NEXT: s_endpgm ; @@ -322,17 +322,17 @@ define amdgpu_kernel void @s_sbfe_sub_i32(ptr addrspace(1) %out, i32 %src, i32 % ; SI-LABEL: s_sbfe_sub_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, 0 ; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0 -; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_sub_i32 s3, 32, s3 -; SI-NEXT: s_lshl_b32 s2, s2, s3 -; SI-NEXT: s_ashr_i32 s2, s2, s3 ; SI-NEXT: v_mov_b32_e32 v1, 0 -; SI-NEXT: s_mov_b64 s[4:5], s[0:1] -; SI-NEXT: v_mov_b32_e32 v2, s2 -; SI-NEXT: buffer_store_dword v2, v[0:1], s[4:7], 0 addr64 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_sub_i32 s2, 32, s5 +; SI-NEXT: s_lshl_b32 s4, s4, s2 +; SI-NEXT: s_ashr_i32 s4, s4, s2 +; SI-NEXT: s_mov_b32 s2, 0 +; SI-NEXT: v_mov_b32_e32 v2, s4 +; SI-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_sbfe_sub_i32: @@ -362,20 +362,20 @@ define amdgpu_kernel void @s_sbfe_sub_multi_use_shl_i32(ptr addrspace(1) %out, i ; SI-LABEL: s_sbfe_sub_multi_use_shl_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s6, 0 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: v_lshlrev_b32_e32 v0, 2, v0 -; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_sub_i32 s3, 32, s3 -; SI-NEXT: s_lshl_b32 s2, s2, s3 -; SI-NEXT: s_ashr_i32 s3, s2, s3 ; SI-NEXT: v_mov_b32_e32 v1, 0 -; SI-NEXT: s_mov_b64 s[4:5], s[0:1] -; SI-NEXT: v_mov_b32_e32 v2, s3 -; SI-NEXT: buffer_store_dword v2, v[0:1], s[4:7], 0 addr64 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s2, 0 +; SI-NEXT: s_sub_i32 s3, 32, s5 +; SI-NEXT: s_lshl_b32 s4, s4, s3 +; SI-NEXT: s_ashr_i32 s5, s4, s3 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: v_mov_b32_e32 v2, s5 +; SI-NEXT: buffer_store_dword v2, v[0:1], s[0:3], 0 addr64 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_waitcnt vmcnt(0) ; SI-NEXT: s_endpgm ; diff --git a/llvm/test/CodeGen/AMDGPU/bfi_nested.ll b/llvm/test/CodeGen/AMDGPU/bfi_nested.ll index bd76f34..7326adae 100644 --- a/llvm/test/CodeGen/AMDGPU/bfi_nested.ll +++ b/llvm/test/CodeGen/AMDGPU/bfi_nested.ll @@ -284,16 +284,15 @@ define amdgpu_kernel void @v_bfi_dont_applied_for_scalar_ops(ptr addrspace(1) %o ; GCN-LABEL: v_bfi_dont_applied_for_scalar_ops: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_and_b32 s3, s3, 0xffff0000 -; GCN-NEXT: s_and_b32 s2, s2, 0xffff -; GCN-NEXT: s_or_b32 s2, s2, s3 -; 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_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_and_b32 s2, s5, 0xffff0000 +; GCN-NEXT: s_and_b32 s4, s4, 0xffff +; GCN-NEXT: s_or_b32 s4, 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 %shift = lshr i32 %b, 16 %tr = trunc i32 %shift to i16 diff --git a/llvm/test/CodeGen/AMDGPU/bfm.ll b/llvm/test/CodeGen/AMDGPU/bfm.ll index a12b5ea..172e07f 100644 --- a/llvm/test/CodeGen/AMDGPU/bfm.ll +++ b/llvm/test/CodeGen/AMDGPU/bfm.ll @@ -6,14 +6,13 @@ define amdgpu_kernel void @s_bfm_pattern(ptr addrspace(1) %out, i32 %x, i32 %y) ; SI-LABEL: s_bfm_pattern: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_bfm_b32 s2, s2, s3 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_bfm_b32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_bfm_pattern: diff --git a/llvm/test/CodeGen/AMDGPU/bitreverse.ll b/llvm/test/CodeGen/AMDGPU/bitreverse.ll index d4f5617..e33b9ab 100644 --- a/llvm/test/CodeGen/AMDGPU/bitreverse.ll +++ b/llvm/test/CodeGen/AMDGPU/bitreverse.ll @@ -362,31 +362,29 @@ define amdgpu_kernel void @s_brev_v2i32(ptr addrspace(1) noalias %out, <2 x i32> ; SI-LABEL: s_brev_v2i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_brev_b32 s0, s3 -; SI-NEXT: s_brev_b32 s1, s2 -; SI-NEXT: v_mov_b32_e32 v0, s1 -; SI-NEXT: v_mov_b32_e32 v1, s0 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_brev_b32 s5, s5 +; SI-NEXT: s_brev_b32 s4, s4 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; FLAT-LABEL: s_brev_v2i32: ; FLAT: ; %bb.0: ; FLAT-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; FLAT-NEXT: s_mov_b32 s7, 0xf000 -; FLAT-NEXT: s_mov_b32 s6, -1 ; FLAT-NEXT: s_waitcnt lgkmcnt(0) -; FLAT-NEXT: s_mov_b32 s4, s0 -; FLAT-NEXT: s_mov_b32 s5, s1 -; FLAT-NEXT: s_brev_b32 s0, s3 -; FLAT-NEXT: s_brev_b32 s1, s2 -; FLAT-NEXT: v_mov_b32_e32 v0, s1 -; FLAT-NEXT: v_mov_b32_e32 v1, s0 -; FLAT-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; FLAT-NEXT: s_mov_b64 s[4:5], s[2:3] +; FLAT-NEXT: s_brev_b32 s5, s5 +; FLAT-NEXT: s_brev_b32 s4, s4 +; FLAT-NEXT: s_mov_b32 s3, 0xf000 +; FLAT-NEXT: s_mov_b32 s2, -1 +; FLAT-NEXT: v_mov_b32_e32 v0, s4 +; FLAT-NEXT: v_mov_b32_e32 v1, s5 +; FLAT-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; FLAT-NEXT: s_endpgm ; ; GISEL-LABEL: s_brev_v2i32: @@ -405,16 +403,14 @@ define amdgpu_kernel void @s_brev_v2i32(ptr addrspace(1) noalias %out, <2 x i32> ; GFX11-FLAT-LABEL: s_brev_v2i32: ; GFX11-FLAT: ; %bb.0: ; GFX11-FLAT-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GFX11-FLAT-NEXT: s_mov_b32 s7, 0x31016000 -; GFX11-FLAT-NEXT: s_mov_b32 s6, -1 ; GFX11-FLAT-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FLAT-NEXT: s_brev_b32 s2, s2 ; GFX11-FLAT-NEXT: s_brev_b32 s3, s3 ; GFX11-FLAT-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FLAT-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3 -; GFX11-FLAT-NEXT: s_mov_b32 s4, s0 -; GFX11-FLAT-NEXT: s_mov_b32 s5, s1 -; GFX11-FLAT-NEXT: buffer_store_b64 v[0:1], off, s[4:7], 0 +; GFX11-FLAT-NEXT: s_mov_b32 s3, 0x31016000 +; GFX11-FLAT-NEXT: s_mov_b32 s2, -1 +; GFX11-FLAT-NEXT: buffer_store_b64 v[0:1], off, s[0:3], 0 ; GFX11-FLAT-NEXT: s_endpgm ; ; GFX11-GISEL-LABEL: s_brev_v2i32: diff --git a/llvm/test/CodeGen/AMDGPU/build_vector.ll b/llvm/test/CodeGen/AMDGPU/build_vector.ll index 763f436..fbaaef0 100644 --- a/llvm/test/CodeGen/AMDGPU/build_vector.ll +++ b/llvm/test/CodeGen/AMDGPU/build_vector.ll @@ -255,16 +255,15 @@ define amdgpu_kernel void @build_v2i32_from_v4i16_shuffle(ptr addrspace(1) %out, ; GFX6-LABEL: build_v2i32_from_v4i16_shuffle: ; GFX6: ; %bb.0: ; %entry ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: s_lshl_b32 s0, s3, 16 -; GFX6-NEXT: s_lshl_b32 s1, s2, 16 -; GFX6-NEXT: v_mov_b32_e32 v0, s1 -; GFX6-NEXT: v_mov_b32_e32 v1, s0 -; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_lshl_b32 s5, s5, 16 +; GFX6-NEXT: s_lshl_b32 s4, s4, 16 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: v_mov_b32_e32 v1, s5 +; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX8-LABEL: build_v2i32_from_v4i16_shuffle: diff --git a/llvm/test/CodeGen/AMDGPU/cc-entry.ll b/llvm/test/CodeGen/AMDGPU/cc-entry.ll new file mode 100644 index 0000000..d807f32 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/cc-entry.ll @@ -0,0 +1,69 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 +; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 < %s | FileCheck %s + +define amdgpu_kernel void @entry_fn() { +; CHECK-LABEL: entry_fn: +; CHECK: ; %bb.0: ; %entry +; CHECK-NEXT: s_getpc_b64 s[4:5] +; CHECK-NEXT: s_sext_i32_i16 s5, s5 +; CHECK-NEXT: s_add_co_u32 s4, s4, entry_fn@gotpcrel32@lo+8 +; CHECK-NEXT: s_add_co_ci_u32 s5, s5, entry_fn@gotpcrel32@hi+16 +; CHECK-NEXT: s_mov_b32 s32, 0 +; CHECK-NEXT: s_load_b64 s[4:5], s[4:5], 0x0 +; CHECK-NEXT: s_wait_kmcnt 0x0 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[4:5] +; CHECK-NEXT: s_endpgm +entry: + call void @entry_fn() + ret void +} + +define void @caller() { +; CHECK-LABEL: caller: +; CHECK: ; %bb.0: ; %entry +; CHECK-NEXT: s_wait_loadcnt_dscnt 0x0 +; CHECK-NEXT: s_wait_expcnt 0x0 +; CHECK-NEXT: s_wait_samplecnt 0x0 +; CHECK-NEXT: s_wait_bvhcnt 0x0 +; CHECK-NEXT: s_wait_kmcnt 0x0 +; CHECK-NEXT: s_mov_b32 s0, s33 +; CHECK-NEXT: s_mov_b32 s33, s32 +; CHECK-NEXT: s_or_saveexec_b32 s1, -1 +; CHECK-NEXT: scratch_store_b32 off, v40, s33 ; 4-byte Folded Spill +; CHECK-NEXT: s_wait_alu 0xfffe +; CHECK-NEXT: s_mov_b32 exec_lo, s1 +; CHECK-NEXT: s_add_co_i32 s32, s32, 16 +; CHECK-NEXT: v_writelane_b32 v40, s0, 2 +; CHECK-NEXT: s_mov_b64 s[0:1], s[4:5] +; CHECK-NEXT: s_getpc_b64 s[4:5] +; CHECK-NEXT: s_wait_alu 0xfffe +; CHECK-NEXT: s_sext_i32_i16 s5, s5 +; CHECK-NEXT: s_add_co_u32 s4, s4, entry_fn@gotpcrel32@lo+12 +; CHECK-NEXT: s_wait_alu 0xfffe +; CHECK-NEXT: s_add_co_ci_u32 s5, s5, entry_fn@gotpcrel32@hi+24 +; CHECK-NEXT: v_mov_b32_e32 v0, v31 +; CHECK-NEXT: s_load_b64 s[4:5], s[4:5], 0x0 +; CHECK-NEXT: v_writelane_b32 v40, s30, 0 +; CHECK-NEXT: s_mov_b64 s[2:3], s[6:7] +; CHECK-NEXT: s_mov_b64 s[6:7], s[10:11] +; CHECK-NEXT: v_writelane_b32 v40, s31, 1 +; CHECK-NEXT: s_wait_kmcnt 0x0 +; CHECK-NEXT: s_wait_alu 0xfffe +; CHECK-NEXT: s_swappc_b64 s[30:31], s[4:5] +; CHECK-NEXT: s_delay_alu instid0(VALU_DEP_1) +; CHECK-NEXT: v_readlane_b32 s31, v40, 1 +; CHECK-NEXT: v_readlane_b32 s30, v40, 0 +; CHECK-NEXT: s_mov_b32 s32, s33 +; CHECK-NEXT: v_readlane_b32 s0, v40, 2 +; CHECK-NEXT: s_or_saveexec_b32 s1, -1 +; CHECK-NEXT: scratch_load_b32 v40, off, s33 ; 4-byte Folded Reload +; CHECK-NEXT: s_wait_alu 0xfffe +; CHECK-NEXT: s_mov_b32 exec_lo, s1 +; CHECK-NEXT: s_mov_b32 s33, s0 +; CHECK-NEXT: s_wait_loadcnt 0x0 +; CHECK-NEXT: s_wait_alu 0xfffe +; CHECK-NEXT: s_setpc_b64 s[30:31] +entry: + call void @entry_fn() + ret void +} diff --git a/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll b/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll index 3d315f8..4cbd41c 100644 --- a/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll +++ b/llvm/test/CodeGen/AMDGPU/combine-cond-add-sub.ll @@ -647,20 +647,20 @@ define amdgpu_kernel void @sub_zext_setcc_commute(ptr addrspace(1) nocapture %ar ; GCN-LABEL: sub_zext_setcc_commute: ; GCN: ; %bb.0: ; %bb ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, 0 ; GCN-NEXT: v_lshlrev_b32_e32 v2, 2, v0 -; GCN-NEXT: v_mov_b32_e32 v3, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b64 s[4:5], s[0:1] -; GCN-NEXT: buffer_load_dword v4, v[2:3], s[4:7], 0 addr64 +; GCN-NEXT: s_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, 0 +; GCN-NEXT: v_mov_b32_e32 v3, 0 +; GCN-NEXT: buffer_load_dword v4, v[2:3], s[0:3], 0 addr64 ; GCN-NEXT: v_cmp_gt_u32_e32 vcc, v0, v1 ; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 -; GCN-NEXT: v_add_i32_e32 v0, vcc, s2, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s3, v0 -; GCN-NEXT: buffer_store_dword v0, v[2:3], s[4:7], 0 addr64 +; GCN-NEXT: v_add_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s5, v0 +; GCN-NEXT: buffer_store_dword v0, v[2:3], s[0:3], 0 addr64 ; GCN-NEXT: s_endpgm ; ; GFX9-LABEL: sub_zext_setcc_commute: @@ -696,20 +696,20 @@ define amdgpu_kernel void @sub_sext_setcc_commute(ptr addrspace(1) nocapture %ar ; GCN-LABEL: sub_sext_setcc_commute: ; GCN: ; %bb.0: ; %bb ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, 0 ; GCN-NEXT: v_lshlrev_b32_e32 v2, 2, v0 -; GCN-NEXT: v_mov_b32_e32 v3, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b64 s[4:5], s[0:1] -; GCN-NEXT: buffer_load_dword v4, v[2:3], s[4:7], 0 addr64 +; GCN-NEXT: s_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, 0 +; GCN-NEXT: v_mov_b32_e32 v3, 0 +; GCN-NEXT: buffer_load_dword v4, v[2:3], s[0:3], 0 addr64 ; GCN-NEXT: v_cmp_gt_u32_e32 vcc, v0, v1 ; GCN-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 -; GCN-NEXT: v_add_i32_e32 v0, vcc, s2, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s3, v0 -; GCN-NEXT: buffer_store_dword v0, v[2:3], s[4:7], 0 addr64 +; GCN-NEXT: v_add_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s5, v0 +; GCN-NEXT: buffer_store_dword v0, v[2:3], s[0:3], 0 addr64 ; GCN-NEXT: s_endpgm ; ; GFX9-LABEL: sub_sext_setcc_commute: diff --git a/llvm/test/CodeGen/AMDGPU/compute-known-bits-nofpclass.ll b/llvm/test/CodeGen/AMDGPU/compute-known-bits-nofpclass.ll new file mode 100644 index 0000000..244c3f7c --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/compute-known-bits-nofpclass.ll @@ -0,0 +1,46 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s + +define i32 @known_positive(float nofpclass(nan ninf nzero nsub nnorm) %signbit.zero) #0 { +; CHECK-LABEL: known_positive: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %cast = bitcast float %signbit.zero to i32 + %and = and i32 %cast, 2147483647 + ret i32 %and +} + +define i32 @known_positive_maybe_nan(float nofpclass(ninf nzero nsub nnorm) %signbit.zero) #0 { +; CHECK-LABEL: known_positive_maybe_nan: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; CHECK-NEXT: s_setpc_b64 s[30:31] + %cast = bitcast float %signbit.zero to i32 + %and = and i32 %cast, 2147483647 + ret i32 %and +} + +define i32 @known_negative(float nofpclass(nan pinf pzero psub pnorm) %signbit.one) #0 { +; CHECK-LABEL: known_negative: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %cast = bitcast float %signbit.one to i32 + %or = or i32 %cast, -2147483648 + ret i32 %or +} + +define i32 @known_negative_maybe_nan(float nofpclass(pinf pzero psub pnorm) %signbit.one) #0 { +; CHECK-LABEL: known_negative_maybe_nan: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: v_or_b32_e32 v0, 0x80000000, v0 +; CHECK-NEXT: s_setpc_b64 s[30:31] + %cast = bitcast float %signbit.one to i32 + %or = or i32 %cast, -2147483648 + ret i32 %or +} + +attributes #0 = { nounwind } diff --git a/llvm/test/CodeGen/AMDGPU/copysign-simplify-demanded-bits.ll b/llvm/test/CodeGen/AMDGPU/copysign-simplify-demanded-bits.ll index f5227ee..ef676dd 100644 --- a/llvm/test/CodeGen/AMDGPU/copysign-simplify-demanded-bits.ll +++ b/llvm/test/CodeGen/AMDGPU/copysign-simplify-demanded-bits.ll @@ -345,15 +345,13 @@ define float @test_copysign_pow_fast_f32__integral_y(float %x, i32 %y.i) { ; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v3 ; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v4, vcc ; GFX9-NEXT: v_fma_f32 v2, v2, v1, v3 -; GFX9-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GFX9-NEXT: v_exp_f32_e32 v2, v2 +; GFX9-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GFX9-NEXT: v_not_b32_e32 v3, 63 ; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc -; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1 ; GFX9-NEXT: v_ldexp_f32 v2, v2, v3 -; GFX9-NEXT: v_and_b32_e32 v0, v1, v0 -; GFX9-NEXT: s_brev_b32 s4, -2 -; GFX9-NEXT: v_bfi_b32 v0, s4, v2, v0 +; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1 +; GFX9-NEXT: v_and_or_b32 v0, v1, v0, v2 ; GFX9-NEXT: s_setpc_b64 s[30:31] %y = sitofp i32 %y.i to float %y.fptosi = fptosi float %y to i32 @@ -370,4 +368,109 @@ define float @test_copysign_pow_fast_f32__integral_y(float %x, i32 %y.i) { ret float %pow_sign1 } +define double @test_pow_fast_f64integral_y(double %x, i32 %y.i) #0 { +; GFX9-LABEL: test_pow_fast_f64integral_y: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s16, s33 +; GFX9-NEXT: s_mov_b32 s33, s32 +; GFX9-NEXT: s_or_saveexec_b64 s[18:19], -1 +; GFX9-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill +; GFX9-NEXT: s_mov_b64 exec, s[18:19] +; GFX9-NEXT: v_writelane_b32 v43, s16, 14 +; GFX9-NEXT: v_writelane_b32 v43, s30, 0 +; GFX9-NEXT: v_writelane_b32 v43, s31, 1 +; GFX9-NEXT: v_writelane_b32 v43, s34, 2 +; GFX9-NEXT: v_writelane_b32 v43, s35, 3 +; GFX9-NEXT: v_writelane_b32 v43, s36, 4 +; GFX9-NEXT: v_writelane_b32 v43, s37, 5 +; GFX9-NEXT: v_writelane_b32 v43, s38, 6 +; GFX9-NEXT: v_writelane_b32 v43, s39, 7 +; GFX9-NEXT: v_writelane_b32 v43, s48, 8 +; GFX9-NEXT: v_writelane_b32 v43, s49, 9 +; GFX9-NEXT: v_writelane_b32 v43, s50, 10 +; GFX9-NEXT: s_addk_i32 s32, 0x800 +; GFX9-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill +; GFX9-NEXT: buffer_store_dword v42, off, s[0:3], s33 ; 4-byte Folded Spill +; GFX9-NEXT: v_writelane_b32 v43, s51, 11 +; GFX9-NEXT: v_mov_b32_e32 v42, v1 +; GFX9-NEXT: v_writelane_b32 v43, s52, 12 +; GFX9-NEXT: v_and_b32_e32 v1, 0x7fffffff, v42 +; GFX9-NEXT: s_getpc_b64 s[16:17] +; GFX9-NEXT: s_add_u32 s16, s16, _Z4log2d@rel32@lo+4 +; GFX9-NEXT: s_addc_u32 s17, s17, _Z4log2d@rel32@hi+12 +; GFX9-NEXT: v_writelane_b32 v43, s53, 13 +; GFX9-NEXT: v_mov_b32_e32 v40, v31 +; GFX9-NEXT: v_mov_b32_e32 v41, v2 +; GFX9-NEXT: s_mov_b32 s50, s15 +; GFX9-NEXT: s_mov_b32 s51, s14 +; GFX9-NEXT: s_mov_b32 s52, s13 +; GFX9-NEXT: s_mov_b32 s53, s12 +; GFX9-NEXT: s_mov_b64 s[34:35], s[10:11] +; GFX9-NEXT: s_mov_b64 s[36:37], s[8:9] +; GFX9-NEXT: s_mov_b64 s[38:39], s[6:7] +; GFX9-NEXT: s_mov_b64 s[48:49], s[4:5] +; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17] +; GFX9-NEXT: v_cvt_f64_i32_e32 v[2:3], v41 +; GFX9-NEXT: s_getpc_b64 s[16:17] +; GFX9-NEXT: s_add_u32 s16, s16, _Z4exp2d@rel32@lo+4 +; GFX9-NEXT: s_addc_u32 s17, s17, _Z4exp2d@rel32@hi+12 +; GFX9-NEXT: s_mov_b64 s[4:5], s[48:49] +; GFX9-NEXT: s_mov_b64 s[6:7], s[38:39] +; GFX9-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3] +; GFX9-NEXT: s_mov_b64 s[8:9], s[36:37] +; GFX9-NEXT: s_mov_b64 s[10:11], s[34:35] +; GFX9-NEXT: s_mov_b32 s12, s53 +; GFX9-NEXT: s_mov_b32 s13, s52 +; GFX9-NEXT: s_mov_b32 s14, s51 +; GFX9-NEXT: s_mov_b32 s15, s50 +; GFX9-NEXT: v_mov_b32_e32 v31, v40 +; GFX9-NEXT: s_swappc_b64 s[30:31], s[16:17] +; GFX9-NEXT: v_lshlrev_b32_e32 v2, 31, v41 +; GFX9-NEXT: v_and_b32_e32 v2, v2, v42 +; GFX9-NEXT: buffer_load_dword v42, off, s[0:3], s33 ; 4-byte Folded Reload +; GFX9-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload +; GFX9-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload +; GFX9-NEXT: v_or_b32_e32 v1, v1, v2 +; GFX9-NEXT: v_readlane_b32 s53, v43, 13 +; GFX9-NEXT: v_readlane_b32 s52, v43, 12 +; GFX9-NEXT: v_readlane_b32 s51, v43, 11 +; GFX9-NEXT: v_readlane_b32 s50, v43, 10 +; GFX9-NEXT: v_readlane_b32 s49, v43, 9 +; GFX9-NEXT: v_readlane_b32 s48, v43, 8 +; GFX9-NEXT: v_readlane_b32 s39, v43, 7 +; GFX9-NEXT: v_readlane_b32 s38, v43, 6 +; GFX9-NEXT: v_readlane_b32 s37, v43, 5 +; GFX9-NEXT: v_readlane_b32 s36, v43, 4 +; GFX9-NEXT: v_readlane_b32 s35, v43, 3 +; GFX9-NEXT: v_readlane_b32 s34, v43, 2 +; GFX9-NEXT: v_readlane_b32 s31, v43, 1 +; GFX9-NEXT: v_readlane_b32 s30, v43, 0 +; GFX9-NEXT: s_mov_b32 s32, s33 +; GFX9-NEXT: v_readlane_b32 s4, v43, 14 +; GFX9-NEXT: s_or_saveexec_b64 s[6:7], -1 +; GFX9-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload +; GFX9-NEXT: s_mov_b64 exec, s[6:7] +; GFX9-NEXT: s_mov_b32 s33, s4 +; GFX9-NEXT: s_waitcnt vmcnt(0) +; GFX9-NEXT: s_setpc_b64 s[30:31] + %fabs = call fast double @llvm.fabs.f64(double %x) + %log2 = call fast double @_Z4log2d(double %fabs) + %pownI2F = sitofp i32 %y.i to double + %ylogx = fmul fast double %log2, %pownI2F + %exp2 = call fast nofpclass(nan ninf nzero nsub nnorm) double @_Z4exp2d(double %ylogx) + %ytou = zext i32 %y.i to i64 + %yeven = shl i64 %ytou, 63 + %x.i64 = bitcast double %x to i64 + %pow_sign = and i64 %yeven, %x.i64 + %pow_sign.f64 = bitcast i64 %pow_sign to double + %pow_sign1 = call fast double @llvm.copysign.f64(double %exp2, double %pow_sign.f64) + ret double %pow_sign1 +} + +declare hidden double @_Z4exp2d(double) #1 +declare hidden double @_Z4log2d(double) #1 + attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #1 = { norecurse nounwind memory(read) } diff --git a/llvm/test/CodeGen/AMDGPU/copysign-to-disjoint-or-combine.ll b/llvm/test/CodeGen/AMDGPU/copysign-to-disjoint-or-combine.ll new file mode 100644 index 0000000..afd610f --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/copysign-to-disjoint-or-combine.ll @@ -0,0 +1,198 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s + +; Negative test, don't know %x is positive +define half @copysign_known_signmask_f16(half %x, i16 %sign) { +; GFX9-LABEL: copysign_known_signmask_f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b16_e32 v1, 15, v1 +; GFX9-NEXT: s_movk_i32 s4, 0x7fff +; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i16 %sign, 15 + %signmask.bitcast = bitcast i16 %signmask to half + %result = call half @llvm.copysign.f16(half %x, half %signmask.bitcast) + ret half %result +} + +; Negative test, don't know %x is positive +define float @copysign_known_signmask_f32(float %x, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1 +; GFX9-NEXT: s_brev_b32 s4, -2 +; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %x, float %signmask.bitcast) + ret float %result +} + +; Negative test, don't know %x is positive +define double @copysign_known_signmask_f64(double %x, i64 %sign) { +; GFX9-LABEL: copysign_known_signmask_f64: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v2, 31, v2 +; GFX9-NEXT: s_brev_b32 s4, -2 +; GFX9-NEXT: v_bfi_b32 v1, s4, v1, v2 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i64 %sign, 63 + %signmask.bitcast = bitcast i64 %signmask to double + %result = call double @llvm.copysign.f64(double %x, double %signmask.bitcast) + ret double %result +} + +; Negative test, don't know %x is positive +define float @copysign_known_signmask_f32_known_not_known_positive_mag_maybe_nan(float nofpclass(ninf nzero nsub nnorm) %sign.bit.known.zero, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_not_known_positive_mag_maybe_nan: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1 +; GFX9-NEXT: s_brev_b32 s4, -2 +; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %sign.bit.known.zero, float %signmask.bitcast) + ret float %result +} + +; Negative test, don't know %x is positive +define float @copysign_known_signmask_f32_known_not_known_positive_mag_maybe_negzero(float nofpclass(nan ninf nsub nnorm) %sign.bit.known.zero, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_not_known_positive_mag_maybe_negzero: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1 +; GFX9-NEXT: s_brev_b32 s4, -2 +; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %sign.bit.known.zero, float %signmask.bitcast) + ret float %result +} + +define half @copysign_known_signmask_f16_known_positive_mag(half nofpclass(nan ninf nzero nsub nnorm) %sign.bit.known.zero, i16 %sign) { +; GFX9-LABEL: copysign_known_signmask_f16_known_positive_mag: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b16_e32 v1, 15, v1 +; GFX9-NEXT: v_or_b32_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i16 %sign, 15 + %signmask.bitcast = bitcast i16 %signmask to half + %result = call half @llvm.copysign.f16(half %sign.bit.known.zero, half %signmask.bitcast) + ret half %result +} + +define float @copysign_known_signmask_f32_known_positive_mag(float nofpclass(nan ninf nzero nsub nnorm) %sign.bit.known.zero, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_positive_mag: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshl_or_b32 v0, v1, 31, v0 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %sign.bit.known.zero, float %signmask.bitcast) + ret float %result +} + +define double @copysign_known_signmask_f64_known_positive_mag(double nofpclass(nan ninf nzero nsub nnorm) %sign.bit.known.zero, i64 %sign) { +; GFX9-LABEL: copysign_known_signmask_f64_known_positive_mag: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v2, 31, v2 +; GFX9-NEXT: v_or_b32_e32 v1, v1, v2 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i64 %sign, 63 + %signmask.bitcast = bitcast i64 %signmask to double + %result = call double @llvm.copysign.f64(double %sign.bit.known.zero, double %signmask.bitcast) + ret double %result +} + +; exp always returns a positive result, excluding the unknown nan sign +; bit. +define float @copysign_known_signmask_f32_known_positive_mag__nnan_exp(float %x, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_positive_mag__nnan_exp: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s4, 0xc2aeac50 +; GFX9-NEXT: v_add_f32_e32 v2, 0x42800000, v0 +; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc +; GFX9-NEXT: v_mul_f32_e32 v0, 0x3fb8aa3b, v0 +; GFX9-NEXT: v_exp_f32_e32 v0, v0 +; GFX9-NEXT: v_mul_f32_e32 v2, 0x114b4ea4, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc +; GFX9-NEXT: v_lshl_or_b32 v0, v1, 31, v0 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signbit.known.zero = call nnan afn float @llvm.exp.f32(float %x) + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %signbit.known.zero, float %signmask.bitcast) + ret float %result +} + +define float @copysign_known_signmask_f32_known_positive_mag__nnan_exp2(float %x, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_positive_mag__nnan_exp2: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s4, 0xc2fc0000 +; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x42800000 +; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc +; GFX9-NEXT: v_add_f32_e32 v0, v0, v3 +; GFX9-NEXT: v_exp_f32_e32 v0, v0 +; GFX9-NEXT: v_not_b32_e32 v2, 63 +; GFX9-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc +; GFX9-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX9-NEXT: v_lshl_or_b32 v0, v1, 31, v0 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signbit.known.zero = call nnan afn float @llvm.exp2.f32(float %x) + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %signbit.known.zero, float %signmask.bitcast) + ret float %result +} + +define float @copysign_known_signmask_f32_known_positive_mag__nnan_exp10(float %x, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_positive_mag__nnan_exp10: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: s_mov_b32 s4, 0xc2fc0000 +; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0 +; GFX9-NEXT: v_mov_b32_e32 v3, 0x42800000 +; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc +; GFX9-NEXT: v_add_f32_e32 v0, v0, v3 +; GFX9-NEXT: v_exp_f32_e32 v0, v0 +; GFX9-NEXT: v_not_b32_e32 v2, 63 +; GFX9-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc +; GFX9-NEXT: v_ldexp_f32 v0, v0, v2 +; GFX9-NEXT: v_lshl_or_b32 v0, v1, 31, v0 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signbit.known.zero = call nnan afn float @llvm.exp2.f32(float %x) + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %result = call float @llvm.copysign.f32(float %signbit.known.zero, float %signmask.bitcast) + ret float %result +} + +define float @copysign_known_signmask_f32_known_positive_mag_through_fence(float nofpclass(nan ninf nzero nsub nnorm) %sign.bit.known.zero, i32 %sign) { +; GFX9-LABEL: copysign_known_signmask_f32_known_positive_mag_through_fence: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1 +; GFX9-NEXT: ;ARITH_FENCE +; GFX9-NEXT: s_brev_b32 s4, -2 +; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] + %signmask = shl i32 %sign, 31 + %signmask.bitcast = bitcast i32 %signmask to float + %fence = call float @llvm.arithmetic.fence.f32(float %sign.bit.known.zero) + %result = call float @llvm.copysign.f32(float %fence, float %signmask.bitcast) + ret float %result +} diff --git a/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll b/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll index ab96dcf..08545b9 100644 --- a/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll +++ b/llvm/test/CodeGen/AMDGPU/divergence-driven-buildvector.ll @@ -390,16 +390,15 @@ define amdgpu_kernel void @uniform_vec_i16_LH(ptr addrspace(1) %out, i16 %a, i32 ; GCN-LABEL: uniform_vec_i16_LH: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_and_b32 s3, s3, 0xffff0000 -; GCN-NEXT: s_and_b32 s2, s2, 0xffff -; GCN-NEXT: s_or_b32 s2, s2, s3 -; 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_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_and_b32 s2, s5, 0xffff0000 +; GCN-NEXT: s_and_b32 s4, s4, 0xffff +; GCN-NEXT: s_or_b32 s4, 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 ; ; GFX9-LABEL: uniform_vec_i16_LH: diff --git a/llvm/test/CodeGen/AMDGPU/divergence-driven-sext-inreg.ll b/llvm/test/CodeGen/AMDGPU/divergence-driven-sext-inreg.ll index 4c3fd40..d8f9bc1 100644 --- a/llvm/test/CodeGen/AMDGPU/divergence-driven-sext-inreg.ll +++ b/llvm/test/CodeGen/AMDGPU/divergence-driven-sext-inreg.ll @@ -5,15 +5,14 @@ define amdgpu_kernel void @uniform_sext_in_reg_i8_to_i32(ptr addrspace(1) %out, ; GCN-LABEL: uniform_sext_in_reg_i8_to_i32: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_add_i32 s2, s2, s3 -; GCN-NEXT: s_sext_i32_i8 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_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_add_i32 s2, s4, 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 %c = add i32 %a, %b ; add to prevent folding into extload %shl = shl i32 %c, 24 @@ -26,15 +25,14 @@ define amdgpu_kernel void @divergent_sext_in_reg_i8_to_i32(ptr addrspace(1) %out ; GCN-LABEL: divergent_sext_in_reg_i8_to_i32: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s4, s0 -; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: s_add_i32 s0, s2, s3 -; GCN-NEXT: v_add_i32_e32 v0, vcc, s0, v0 +; GCN-NEXT: s_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 +; GCN-NEXT: s_add_i32 s4, s4, s5 +; GCN-NEXT: v_add_i32_e32 v0, vcc, s4, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 8 -; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GCN-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %c = add i32 %a, %b ; add to prevent folding into extload @@ -49,15 +47,14 @@ define amdgpu_kernel void @uniform_sext_in_reg_i16_to_i32(ptr addrspace(1) %out, ; GCN-LABEL: uniform_sext_in_reg_i16_to_i32: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; 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_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_add_i32 s2, s4, s5 +; GCN-NEXT: s_sext_i32_i16 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 %c = add i32 %a, %b ; add to prevent folding into extload %shl = shl i32 %c, 16 @@ -70,15 +67,14 @@ define amdgpu_kernel void @divergent_sext_in_reg_i16_to_i32(ptr addrspace(1) %ou ; GCN-LABEL: divergent_sext_in_reg_i16_to_i32: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s4, s0 -; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: s_add_i32 s0, s2, s3 -; GCN-NEXT: v_add_i32_e32 v0, vcc, s0, v0 +; GCN-NEXT: s_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 +; GCN-NEXT: s_add_i32 s4, s4, s5 +; GCN-NEXT: v_add_i32_e32 v0, vcc, s4, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 16 -; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GCN-NEXT: s_endpgm %tid = call i32 @llvm.amdgcn.workitem.id.x() %c = add i32 %a, %b ; add to prevent folding into extload diff --git a/llvm/test/CodeGen/AMDGPU/ds-read2-write2-debug-info.ll b/llvm/test/CodeGen/AMDGPU/ds-read2-write2-debug-info.ll new file mode 100644 index 0000000..08730038 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/ds-read2-write2-debug-info.ll @@ -0,0 +1,89 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: opt -passes=debugify < %s | llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 | FileCheck %s + +@lds = addrspace(3) global [512 x float] poison, align 4 + +define amdgpu_kernel void @simple_write2_one_val_f32(ptr addrspace(1) %C, ptr addrspace(1) %in) #0 { +; CHECK-LABEL: simple_write2_one_val_f32: +; CHECK: .Lfunc_begin0: +; CHECK-NEXT: .cfi_sections .debug_frame +; CHECK-NEXT: .cfi_startproc +; CHECK-NEXT: ; %bb.0: +; CHECK-NEXT: .file 1 "/" "<stdin>" +; CHECK-NEXT: .loc 1 1 1 prologue_end ; <stdin>:1:1 +; CHECK-NEXT: s_load_dwordx2 s[0:1], s[8:9], 0x8 +; CHECK-NEXT: .Ltmp0: +; CHECK-NEXT: ;DEBUG_VALUE: simple_write2_one_val_f32:1 <- $vgpr0 +; CHECK-NEXT: ;DEBUG_VALUE: simple_write2_one_val_f32:5 <- [DW_OP_plus_uconst 8, DW_OP_stack_value] $vgpr0 +; CHECK-NEXT: ;DEBUG_VALUE: simple_write2_one_val_f32:3 <- undef +; CHECK-NEXT: .loc 1 2 1 ; <stdin>:2:1 +; CHECK-NEXT: v_lshlrev_b32_e32 v0, 2, v0 +; CHECK-NEXT: .Ltmp1: +; CHECK-NEXT: ;DEBUG_VALUE: simple_write2_one_val_f32:4 <- $vgpr0 +; CHECK-NEXT: ;DEBUG_VALUE: simple_write2_one_val_f32:2 <- undef +; CHECK-NEXT: .loc 1 3 1 ; <stdin>:3:1 +; CHECK-NEXT: s_waitcnt lgkmcnt(0) +; CHECK-NEXT: global_load_dword v1, v0, s[0:1] +; CHECK-NEXT: .Ltmp2: +; CHECK-NEXT: ;DEBUG_VALUE: simple_write2_one_val_f32:6 <- [DW_OP_plus_uconst 32, DW_OP_stack_value] $vgpr0 +; CHECK-NEXT: .loc 1 0 0 is_stmt 0 ; <stdin>:0 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: ds_write2_b32 v0, v1, v1 offset1:8 +; CHECK-NEXT: .loc 1 9 1 is_stmt 1 ; <stdin>:9:1 +; CHECK-NEXT: s_endpgm +; CHECK-NEXT: .Ltmp3: + %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1 + %in.gep = getelementptr float, ptr addrspace(1) %in, i32 %x.i + %val = load float, ptr addrspace(1) %in.gep, align 4 + %arrayidx0 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds, i32 0, i32 %x.i + store float %val, ptr addrspace(3) %arrayidx0, align 4 + %add.x = add nsw i32 %x.i, 8 + %arrayidx1 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds, i32 0, i32 %add.x + store float %val, ptr addrspace(3) %arrayidx1, align 4 + ret void +} + +define amdgpu_kernel void @simple_read2_f32(ptr addrspace(1) %out) #0 { +; CHECK-LABEL: simple_read2_f32: +; CHECK: .Lfunc_begin1: +; CHECK-NEXT: .cfi_startproc +; CHECK-NEXT: ; %bb.0: +; CHECK-NEXT: .loc 1 11 1 prologue_end ; <stdin>:11:1 +; CHECK-NEXT: v_lshlrev_b32_e32 v2, 2, v0 +; CHECK-NEXT: .Ltmp4: +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:8 <- $vgpr2 +; CHECK-NEXT: .loc 1 0 0 is_stmt 0 ; <stdin>:0 +; CHECK-NEXT: ds_read2_b32 v[0:1], v2 offset1:8 +; CHECK-NEXT: .Ltmp5: +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:9 <- undef +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:11 <- [DW_OP_plus_uconst 32, DW_OP_stack_value] $vgpr2 +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:12 <- undef +; CHECK-NEXT: .loc 1 10 1 is_stmt 1 ; <stdin>:10:1 +; CHECK-NEXT: s_load_dwordx2 s[0:1], s[8:9], 0x0 +; CHECK-NEXT: .Ltmp6: +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:7 <- undef +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:10 <- [DW_OP_plus_uconst 8, DW_OP_stack_value] undef +; CHECK-NEXT: .loc 1 16 1 ; <stdin>:16:1 +; CHECK-NEXT: s_waitcnt lgkmcnt(0) +; CHECK-NEXT: v_add_f32_e32 v0, v0, v1 +; CHECK-NEXT: .Ltmp7: +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:13 <- $vgpr0 +; CHECK-NEXT: ;DEBUG_VALUE: simple_read2_f32:14 <- undef +; CHECK-NEXT: .loc 1 18 1 ; <stdin>:18:1 +; CHECK-NEXT: global_store_dword v2, v0, s[0:1] +; CHECK-NEXT: .loc 1 19 1 ; <stdin>:19:1 +; CHECK-NEXT: s_endpgm +; CHECK-NEXT: .Ltmp8: + %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1 + %arrayidx0 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds, i32 0, i32 %x.i + %val0 = load float, ptr addrspace(3) %arrayidx0, align 4 + %add.x = add nsw i32 %x.i, 8 + %arrayidx1 = getelementptr inbounds [512 x float], ptr addrspace(3) @lds, i32 0, i32 %add.x + %val1 = load float, ptr addrspace(3) %arrayidx1, align 4 + %sum = fadd float %val0, %val1 + %out.gep = getelementptr inbounds float, ptr addrspace(1) %out, i32 %x.i + store float %sum, ptr addrspace(1) %out.gep, align 4 + ret void +} + +attributes #0 = { nounwind } diff --git a/llvm/test/CodeGen/AMDGPU/fabs.ll b/llvm/test/CodeGen/AMDGPU/fabs.ll index 6bcb086..97e23fc 100644 --- a/llvm/test/CodeGen/AMDGPU/fabs.ll +++ b/llvm/test/CodeGen/AMDGPU/fabs.ll @@ -99,16 +99,15 @@ define amdgpu_kernel void @fabs_v2f32(ptr addrspace(1) %out, <2 x float> %in) { ; SI-LABEL: fabs_v2f32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_and_b32 s0, s3, 0x7fffffff -; SI-NEXT: s_and_b32 s1, s2, 0x7fffffff -; SI-NEXT: v_mov_b32_e32 v0, s1 -; SI-NEXT: v_mov_b32_e32 v1, s0 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: s_bitset0_b32 s5, 31 +; SI-NEXT: s_bitset0_b32 s4, 31 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: fabs_v2f32: diff --git a/llvm/test/CodeGen/AMDGPU/fdiv.ll b/llvm/test/CodeGen/AMDGPU/fdiv.ll index b826e6c..4d448e6 100644 --- a/llvm/test/CodeGen/AMDGPU/fdiv.ll +++ b/llvm/test/CodeGen/AMDGPU/fdiv.ll @@ -333,18 +333,17 @@ define amdgpu_kernel void @s_fdiv_25ulp_f32(ptr addrspace(1) %out, float %a, flo ; GFX67-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GFX67-NEXT: v_mov_b32_e32 v0, 0x6f800000 ; GFX67-NEXT: v_mov_b32_e32 v1, 0x2f800000 -; GFX67-NEXT: s_mov_b32 s7, 0xf000 -; GFX67-NEXT: s_mov_b32 s6, -1 ; GFX67-NEXT: s_waitcnt lgkmcnt(0) -; GFX67-NEXT: v_cmp_gt_f32_e64 vcc, |s3|, v0 +; GFX67-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX67-NEXT: v_cmp_gt_f32_e64 vcc, |s5|, v0 ; GFX67-NEXT: v_cndmask_b32_e32 v0, 1.0, v1, vcc -; GFX67-NEXT: v_mul_f32_e32 v1, s3, v0 +; GFX67-NEXT: v_mul_f32_e32 v1, s5, v0 ; GFX67-NEXT: v_rcp_f32_e32 v1, v1 -; GFX67-NEXT: s_mov_b32 s4, s0 -; GFX67-NEXT: s_mov_b32 s5, s1 -; GFX67-NEXT: v_mul_f32_e32 v1, s2, v1 +; GFX67-NEXT: s_mov_b32 s3, 0xf000 +; GFX67-NEXT: s_mov_b32 s2, -1 +; GFX67-NEXT: v_mul_f32_e32 v1, s4, v1 ; GFX67-NEXT: v_mul_f32_e32 v0, v0, v1 -; GFX67-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX67-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX67-NEXT: s_endpgm ; ; GFX8-LABEL: s_fdiv_25ulp_f32: @@ -441,20 +440,19 @@ define amdgpu_kernel void @s_fdiv_25ulp_ieee_f32(ptr addrspace(1) %out, float %a ; GFX7-LABEL: s_fdiv_25ulp_ieee_f32: ; GFX7: ; %bb.0: ; %entry ; GFX7-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX7-NEXT: s_mov_b32 s7, 0xf000 -; GFX7-NEXT: s_mov_b32 s6, -1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_frexp_mant_f32_e32 v0, s3 +; GFX7-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX7-NEXT: v_frexp_mant_f32_e32 v0, s5 ; GFX7-NEXT: v_rcp_f32_e32 v0, v0 -; GFX7-NEXT: v_frexp_exp_i32_f32_e32 v1, s3 -; GFX7-NEXT: v_frexp_exp_i32_f32_e32 v2, s2 -; GFX7-NEXT: v_frexp_mant_f32_e32 v3, s2 +; GFX7-NEXT: v_frexp_exp_i32_f32_e32 v1, s5 +; GFX7-NEXT: v_frexp_exp_i32_f32_e32 v2, s4 +; GFX7-NEXT: v_frexp_mant_f32_e32 v3, s4 ; GFX7-NEXT: v_mul_f32_e32 v0, v3, v0 ; GFX7-NEXT: v_sub_i32_e32 v1, vcc, v2, v1 -; GFX7-NEXT: s_mov_b32 s4, s0 -; GFX7-NEXT: s_mov_b32 s5, s1 +; GFX7-NEXT: s_mov_b32 s3, 0xf000 +; GFX7-NEXT: s_mov_b32 s2, -1 ; GFX7-NEXT: v_ldexp_f32_e32 v0, v0, v1 -; GFX7-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX7-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX7-NEXT: s_endpgm ; ; GFX8-LABEL: s_fdiv_25ulp_ieee_f32: @@ -528,14 +526,13 @@ define amdgpu_kernel void @s_fdiv_fast_ieee_f32(ptr addrspace(1) %out, float %a, ; GFX67-LABEL: s_fdiv_fast_ieee_f32: ; GFX67: ; %bb.0: ; %entry ; GFX67-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX67-NEXT: s_mov_b32 s7, 0xf000 -; GFX67-NEXT: s_mov_b32 s6, -1 ; GFX67-NEXT: s_waitcnt lgkmcnt(0) -; GFX67-NEXT: v_rcp_f32_e32 v0, s3 -; GFX67-NEXT: s_mov_b32 s4, s0 -; GFX67-NEXT: s_mov_b32 s5, s1 -; GFX67-NEXT: v_mul_f32_e32 v0, s2, v0 -; GFX67-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX67-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX67-NEXT: v_rcp_f32_e32 v0, s5 +; GFX67-NEXT: s_mov_b32 s3, 0xf000 +; GFX67-NEXT: s_mov_b32 s2, -1 +; GFX67-NEXT: v_mul_f32_e32 v0, s4, v0 +; GFX67-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX67-NEXT: s_endpgm ; ; GFX8-LABEL: s_fdiv_fast_ieee_f32: @@ -590,14 +587,13 @@ define amdgpu_kernel void @s_fdiv_f32_fast_math(ptr addrspace(1) %out, float %a, ; GFX67-LABEL: s_fdiv_f32_fast_math: ; GFX67: ; %bb.0: ; %entry ; GFX67-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX67-NEXT: s_mov_b32 s7, 0xf000 -; GFX67-NEXT: s_mov_b32 s6, -1 ; GFX67-NEXT: s_waitcnt lgkmcnt(0) -; GFX67-NEXT: v_rcp_f32_e32 v0, s3 -; GFX67-NEXT: s_mov_b32 s4, s0 -; GFX67-NEXT: s_mov_b32 s5, s1 -; GFX67-NEXT: v_mul_f32_e32 v0, s2, v0 -; GFX67-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX67-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX67-NEXT: v_rcp_f32_e32 v0, s5 +; GFX67-NEXT: s_mov_b32 s3, 0xf000 +; GFX67-NEXT: s_mov_b32 s2, -1 +; GFX67-NEXT: v_mul_f32_e32 v0, s4, v0 +; GFX67-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX67-NEXT: s_endpgm ; ; GFX8-LABEL: s_fdiv_f32_fast_math: @@ -652,14 +648,13 @@ define amdgpu_kernel void @s_fdiv_ulp25_f32_fast_math(ptr addrspace(1) %out, flo ; GFX67-LABEL: s_fdiv_ulp25_f32_fast_math: ; GFX67: ; %bb.0: ; %entry ; GFX67-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX67-NEXT: s_mov_b32 s7, 0xf000 -; GFX67-NEXT: s_mov_b32 s6, -1 ; GFX67-NEXT: s_waitcnt lgkmcnt(0) -; GFX67-NEXT: v_rcp_f32_e32 v0, s3 -; GFX67-NEXT: s_mov_b32 s4, s0 -; GFX67-NEXT: s_mov_b32 s5, s1 -; GFX67-NEXT: v_mul_f32_e32 v0, s2, v0 -; GFX67-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX67-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX67-NEXT: v_rcp_f32_e32 v0, s5 +; GFX67-NEXT: s_mov_b32 s3, 0xf000 +; GFX67-NEXT: s_mov_b32 s2, -1 +; GFX67-NEXT: v_mul_f32_e32 v0, s4, v0 +; GFX67-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX67-NEXT: s_endpgm ; ; GFX8-LABEL: s_fdiv_ulp25_f32_fast_math: @@ -877,14 +872,13 @@ define amdgpu_kernel void @s_fdiv_f32_arcp_ninf(ptr addrspace(1) %out, float %a, ; GFX67-LABEL: s_fdiv_f32_arcp_ninf: ; GFX67: ; %bb.0: ; %entry ; GFX67-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX67-NEXT: s_mov_b32 s7, 0xf000 -; GFX67-NEXT: s_mov_b32 s6, -1 ; GFX67-NEXT: s_waitcnt lgkmcnt(0) -; GFX67-NEXT: v_rcp_f32_e32 v0, s3 -; GFX67-NEXT: s_mov_b32 s4, s0 -; GFX67-NEXT: s_mov_b32 s5, s1 -; GFX67-NEXT: v_mul_f32_e32 v0, s2, v0 -; GFX67-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX67-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX67-NEXT: v_rcp_f32_e32 v0, s5 +; GFX67-NEXT: s_mov_b32 s3, 0xf000 +; GFX67-NEXT: s_mov_b32 s2, -1 +; GFX67-NEXT: v_mul_f32_e32 v0, s4, v0 +; GFX67-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX67-NEXT: s_endpgm ; ; GFX8-LABEL: s_fdiv_f32_arcp_ninf: diff --git a/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll b/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll index defcffa..39eefa1 100644 --- a/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll +++ b/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll @@ -75,9 +75,12 @@ define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_fast(ptr addrspace(1) %out ; GCN-LABEL: {{^}}s_test_fmin_legacy_ule_f32_nnan_src: ; GCN: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}} -; GCN-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#LOAD + 2]], 1.0 -; GCN-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#LOAD + 3]], 2.0 +; SI: s_mov_b64 s[[[#COPY:]]:{{[0-9]+}}], s{{\[}}[[#LOAD + 2]]:[[#LOAD + 3]]{{\]}} +; SI-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#COPY]], 1.0 +; SI-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#COPY + 1]], 2.0 +; VI-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#LOAD + 2]], 1.0 +; VI-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#LOAD + 3]], 2.0 ; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]] ; VI: v_cmp_ngt_f32_e32 vcc, [[ADD_A]], [[ADD_B]] @@ -96,8 +99,12 @@ define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_nnan_src(ptr addrspace(1) ; GCN-LABEL: {{^}}s_test_fmin_legacy_ule_f32_nnan_src_fast: ; GCN: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}} -; GCN-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#LOAD + 2]], 1.0 -; GCN-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#LOAD + 3]], 2.0 +; SI: s_mov_b64 s[[[#COPY:]]:{{[0-9]+}}], s{{\[}}[[#LOAD + 2]]:[[#LOAD + 3]]{{\]}} +; SI-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#COPY]], 1.0 +; SI-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#COPY + 1]], 2.0 + +; VI-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#LOAD + 2]], 1.0 +; VI-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#LOAD + 3]], 2.0 ; GCN: v_min_f32_e32 {{v[0-9]+}}, [[ADD_A]], [[ADD_B]] define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_nnan_src_fast(ptr addrspace(1) %out, float %a, float %b) #0 { diff --git a/llvm/test/CodeGen/AMDGPU/fnearbyint.ll b/llvm/test/CodeGen/AMDGPU/fnearbyint.ll index a025c36..6c2ab5f 100644 --- a/llvm/test/CodeGen/AMDGPU/fnearbyint.ll +++ b/llvm/test/CodeGen/AMDGPU/fnearbyint.ll @@ -121,14 +121,13 @@ define amdgpu_kernel void @fnearbyint_v2f32(ptr addrspace(1) %out, <2 x float> % ; SICI-LABEL: fnearbyint_v2f32: ; SICI: ; %bb.0: ; %entry ; SICI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SICI-NEXT: s_mov_b32 s7, 0xf000 -; SICI-NEXT: s_mov_b32 s6, -1 ; SICI-NEXT: s_waitcnt lgkmcnt(0) -; SICI-NEXT: s_mov_b32 s4, s0 -; SICI-NEXT: s_mov_b32 s5, s1 -; SICI-NEXT: v_rndne_f32_e32 v1, s3 -; SICI-NEXT: v_rndne_f32_e32 v0, s2 -; SICI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SICI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SICI-NEXT: s_mov_b32 s3, 0xf000 +; SICI-NEXT: s_mov_b32 s2, -1 +; SICI-NEXT: v_rndne_f32_e32 v1, s5 +; SICI-NEXT: v_rndne_f32_e32 v0, s4 +; SICI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SICI-NEXT: s_endpgm ; ; VI-LABEL: fnearbyint_v2f32: diff --git a/llvm/test/CodeGen/AMDGPU/fneg-fabs.ll b/llvm/test/CodeGen/AMDGPU/fneg-fabs.ll index 1fa9bfa..214cced 100644 --- a/llvm/test/CodeGen/AMDGPU/fneg-fabs.ll +++ b/llvm/test/CodeGen/AMDGPU/fneg-fabs.ll @@ -199,16 +199,15 @@ define amdgpu_kernel void @fneg_fabsf_v2f32(ptr addrspace(1) %out, <2 x float> % ; SI-LABEL: fneg_fabsf_v2f32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_bitset1_b32 s3, 31 -; SI-NEXT: s_bitset1_b32 s2, 31 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: v_mov_b32_e32 v1, s3 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_bitset1_b32 s5, 31 +; SI-NEXT: s_bitset1_b32 s4, 31 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: fneg_fabsf_v2f32: diff --git a/llvm/test/CodeGen/AMDGPU/fneg.ll b/llvm/test/CodeGen/AMDGPU/fneg.ll index c3f4ebe3..0223515 100644 --- a/llvm/test/CodeGen/AMDGPU/fneg.ll +++ b/llvm/test/CodeGen/AMDGPU/fneg.ll @@ -52,16 +52,15 @@ define amdgpu_kernel void @s_fneg_v2f32(ptr addrspace(1) nocapture %out, <2 x fl ; SI-LABEL: s_fneg_v2f32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_xor_b32 s0, s3, 0x80000000 -; SI-NEXT: s_xor_b32 s1, s2, 0x80000000 -; SI-NEXT: v_mov_b32_e32 v0, s1 -; SI-NEXT: v_mov_b32_e32 v1, s0 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_xor_b32 s5, s5, 0x80000000 +; SI-NEXT: s_xor_b32 s4, s4, 0x80000000 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_fneg_v2f32: diff --git a/llvm/test/CodeGen/AMDGPU/fp_to_sint.ll b/llvm/test/CodeGen/AMDGPU/fp_to_sint.ll index 7ab8b30..0c5ed00 100644 --- a/llvm/test/CodeGen/AMDGPU/fp_to_sint.ll +++ b/llvm/test/CodeGen/AMDGPU/fp_to_sint.ll @@ -88,27 +88,24 @@ define amdgpu_kernel void @fp_to_sint_v2i32(ptr addrspace(1) %out, <2 x float> % ; SI-LABEL: fp_to_sint_v2i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_cvt_i32_f32_e32 v1, s3 -; SI-NEXT: v_cvt_i32_f32_e32 v0, s2 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_cvt_i32_f32_e32 v1, s5 +; SI-NEXT: v_cvt_i32_f32_e32 v0, s4 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: fp_to_sint_v2i32: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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: v_cvt_i32_f32_e32 v1, s3 ; VI-NEXT: v_cvt_i32_f32_e32 v0, s2 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; EG-LABEL: fp_to_sint_v2i32: @@ -294,26 +291,25 @@ entry: define amdgpu_kernel void @fp_to_sint_v2i64(ptr addrspace(1) %out, <2 x float> %x) { ; SI-LABEL: fp_to_sint_v2i64: ; SI: ; %bb.0: -; SI-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x9 +; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 +; SI-NEXT: s_mov_b32 s6, 0x2f800000 +; SI-NEXT: s_mov_b32 s7, 0xcf800000 +; SI-NEXT: s_waitcnt lgkmcnt(0) +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] ; SI-NEXT: s_mov_b32 s3, 0xf000 ; SI-NEXT: s_mov_b32 s2, -1 -; SI-NEXT: s_mov_b32 s8, 0x2f800000 -; SI-NEXT: s_mov_b32 s9, 0xcf800000 -; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s0, s4 -; SI-NEXT: s_mov_b32 s1, s5 -; SI-NEXT: v_trunc_f32_e32 v0, s7 -; SI-NEXT: v_trunc_f32_e32 v1, s6 -; SI-NEXT: v_mul_f32_e64 v2, |v0|, s8 +; SI-NEXT: v_trunc_f32_e32 v0, s5 +; SI-NEXT: v_trunc_f32_e32 v1, s4 +; SI-NEXT: v_mul_f32_e64 v2, |v0|, s6 ; SI-NEXT: v_ashrrev_i32_e32 v3, 31, v0 -; SI-NEXT: v_mul_f32_e64 v4, |v1|, s8 +; SI-NEXT: v_mul_f32_e64 v4, |v1|, s6 ; SI-NEXT: v_ashrrev_i32_e32 v5, 31, v1 ; SI-NEXT: v_floor_f32_e32 v2, v2 ; SI-NEXT: v_floor_f32_e32 v4, v4 ; SI-NEXT: v_cvt_u32_f32_e32 v6, v2 -; SI-NEXT: v_fma_f32 v0, v2, s9, |v0| +; SI-NEXT: v_fma_f32 v0, v2, s7, |v0| ; SI-NEXT: v_cvt_u32_f32_e32 v2, v4 -; SI-NEXT: v_fma_f32 v1, v4, s9, |v1| +; SI-NEXT: v_fma_f32 v1, v4, s7, |v1| ; SI-NEXT: v_cvt_u32_f32_e32 v0, v0 ; SI-NEXT: v_xor_b32_e32 v4, v6, v3 ; SI-NEXT: v_cvt_u32_f32_e32 v1, v1 @@ -330,36 +326,35 @@ define amdgpu_kernel void @fp_to_sint_v2i64(ptr addrspace(1) %out, <2 x float> % ; VI-LABEL: fp_to_sint_v2i64: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; VI-NEXT: s_mov_b32 s8, 0x2f800000 -; VI-NEXT: s_mov_b32 s7, 0xf000 -; VI-NEXT: s_mov_b32 s6, -1 +; VI-NEXT: s_mov_b32 s6, 0x2f800000 +; VI-NEXT: s_mov_b32 s7, 0xcf800000 ; VI-NEXT: s_waitcnt lgkmcnt(0) -; VI-NEXT: v_trunc_f32_e32 v0, s3 -; VI-NEXT: v_mul_f32_e64 v1, |v0|, s8 -; VI-NEXT: s_mov_b32 s4, s0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: v_trunc_f32_e32 v0, s5 +; VI-NEXT: v_mul_f32_e64 v1, |v0|, s6 ; VI-NEXT: v_floor_f32_e32 v1, v1 -; VI-NEXT: s_mov_b32 s0, 0xcf800000 -; VI-NEXT: v_fma_f32 v2, v1, s0, |v0| -; VI-NEXT: v_trunc_f32_e32 v4, s2 -; VI-NEXT: v_cvt_u32_f32_e32 v2, v2 -; VI-NEXT: v_mul_f32_e64 v3, |v4|, s8 -; VI-NEXT: v_cvt_u32_f32_e32 v1, v1 -; VI-NEXT: v_floor_f32_e32 v3, v3 -; VI-NEXT: v_cvt_u32_f32_e32 v5, v3 -; VI-NEXT: v_fma_f32 v3, v3, s0, |v4| +; VI-NEXT: v_cvt_u32_f32_e32 v2, v1 +; VI-NEXT: v_fma_f32 v1, v1, s7, |v0| ; VI-NEXT: v_ashrrev_i32_e32 v0, 31, v0 -; VI-NEXT: v_cvt_u32_f32_e32 v6, v3 -; VI-NEXT: v_xor_b32_e32 v2, v2, v0 +; VI-NEXT: v_trunc_f32_e32 v4, s4 +; VI-NEXT: v_xor_b32_e32 v3, v2, v0 +; VI-NEXT: v_mul_f32_e64 v2, |v4|, s6 +; VI-NEXT: v_cvt_u32_f32_e32 v1, v1 +; VI-NEXT: v_floor_f32_e32 v2, v2 +; VI-NEXT: v_cvt_u32_f32_e32 v5, v2 +; VI-NEXT: v_fma_f32 v2, v2, s7, |v4| +; VI-NEXT: v_cvt_u32_f32_e32 v6, v2 ; VI-NEXT: v_xor_b32_e32 v1, v1, v0 -; VI-NEXT: v_sub_u32_e32 v2, vcc, v2, v0 -; VI-NEXT: v_subb_u32_e32 v3, vcc, v1, v0, vcc +; VI-NEXT: v_sub_u32_e32 v2, vcc, v1, v0 ; VI-NEXT: v_ashrrev_i32_e32 v1, 31, v4 +; VI-NEXT: v_subb_u32_e32 v3, vcc, v3, v0, vcc ; VI-NEXT: v_xor_b32_e32 v0, v6, v1 ; VI-NEXT: v_xor_b32_e32 v4, v5, v1 ; VI-NEXT: v_sub_u32_e32 v0, vcc, v0, v1 -; VI-NEXT: s_mov_b32 s5, s1 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 ; VI-NEXT: v_subb_u32_e32 v1, vcc, v4, v1, vcc -; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; EG-LABEL: fp_to_sint_v2i64: diff --git a/llvm/test/CodeGen/AMDGPU/fp_to_uint.ll b/llvm/test/CodeGen/AMDGPU/fp_to_uint.ll index 5428ba8..c938475 100644 --- a/llvm/test/CodeGen/AMDGPU/fp_to_uint.ll +++ b/llvm/test/CodeGen/AMDGPU/fp_to_uint.ll @@ -48,27 +48,24 @@ define amdgpu_kernel void @fp_to_uint_v2f32_to_v2i32(ptr addrspace(1) %out, <2 x ; SI-LABEL: fp_to_uint_v2f32_to_v2i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_cvt_u32_f32_e32 v1, s3 -; SI-NEXT: v_cvt_u32_f32_e32 v0, s2 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_cvt_u32_f32_e32 v1, s5 +; SI-NEXT: v_cvt_u32_f32_e32 v0, s4 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: fp_to_uint_v2f32_to_v2i32: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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: v_cvt_u32_f32_e32 v1, s3 ; VI-NEXT: v_cvt_u32_f32_e32 v0, s2 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; EG-LABEL: fp_to_uint_v2f32_to_v2i32: @@ -241,32 +238,29 @@ define amdgpu_kernel void @fp_to_uint_v2f32_to_v2i64(ptr addrspace(1) %out, <2 x ; SI-LABEL: fp_to_uint_v2f32_to_v2i64: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s8, 0xcf800000 +; SI-NEXT: s_mov_b32 s6, 0xcf800000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_trunc_f32_e32 v0, s3 -; SI-NEXT: v_trunc_f32_e32 v2, s2 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_trunc_f32_e32 v0, s5 +; SI-NEXT: v_trunc_f32_e32 v2, s4 ; SI-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0 ; SI-NEXT: v_mul_f32_e32 v3, 0x2f800000, v2 ; SI-NEXT: v_floor_f32_e32 v4, v1 ; SI-NEXT: v_floor_f32_e32 v5, v3 ; SI-NEXT: v_cvt_u32_f32_e32 v3, v4 ; SI-NEXT: v_cvt_u32_f32_e32 v1, v5 -; SI-NEXT: v_fma_f32 v0, v4, s8, v0 -; SI-NEXT: v_fma_f32 v4, v5, s8, v2 +; SI-NEXT: v_fma_f32 v0, v4, s6, v0 +; SI-NEXT: v_fma_f32 v4, v5, s6, v2 ; SI-NEXT: v_cvt_u32_f32_e32 v2, v0 ; SI-NEXT: v_cvt_u32_f32_e32 v0, v4 -; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; SI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: fp_to_uint_v2f32_to_v2i64: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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: v_trunc_f32_e32 v0, s3 ; VI-NEXT: v_trunc_f32_e32 v4, s2 @@ -281,9 +275,9 @@ define amdgpu_kernel void @fp_to_uint_v2f32_to_v2i64(ptr addrspace(1) %out, <2 x ; VI-NEXT: v_cvt_u32_f32_e32 v3, v5 ; VI-NEXT: v_cvt_u32_f32_e32 v1, v6 ; VI-NEXT: v_cvt_u32_f32_e32 v0, v0 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; EG-LABEL: fp_to_uint_v2f32_to_v2i64: diff --git a/llvm/test/CodeGen/AMDGPU/fshl.ll b/llvm/test/CodeGen/AMDGPU/fshl.ll index ed1ee45..68b95cd 100644 --- a/llvm/test/CodeGen/AMDGPU/fshl.ll +++ b/llvm/test/CodeGen/AMDGPU/fshl.ll @@ -691,17 +691,16 @@ define amdgpu_kernel void @orxor2or1(ptr addrspace(1) %in, i32 %a, i32 %b) { ; SI-LABEL: orxor2or1: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_lshl_b32 s0, s2, 7 -; SI-NEXT: s_or_b32 s0, s3, s0 -; SI-NEXT: s_cmp_eq_u32 s0, 0 -; SI-NEXT: s_cselect_b32 s0, s2, s3 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_lshl_b32 s6, s4, 7 +; SI-NEXT: s_or_b32 s6, s5, s6 +; SI-NEXT: s_cmp_eq_u32 s6, 0 +; SI-NEXT: s_cselect_b32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: orxor2or1: diff --git a/llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll b/llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll index 44bd409..7cbf9ae 100644 --- a/llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll +++ b/llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll @@ -1508,35 +1508,33 @@ define amdgpu_kernel void @dynamic_insertelement_v2i16(ptr addrspace(1) %out, <2 ; SI-LABEL: dynamic_insertelement_v2i16: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; SI-NEXT: s_mov_b32 s7, 0x100f000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_lshl_b32 s1, s3, 4 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_xor_b32 s0, s2, 0x50005 -; SI-NEXT: s_lshl_b32 s1, 0xffff, s1 -; SI-NEXT: s_and_b32 s0, s0, s1 -; SI-NEXT: s_xor_b32 s0, s0, s2 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_lshl_b32 s5, s5, 4 +; SI-NEXT: s_xor_b32 s6, s4, 0x50005 +; SI-NEXT: s_lshl_b32 s5, 0xffff, s5 +; SI-NEXT: s_and_b32 s5, s6, s5 +; SI-NEXT: s_xor_b32 s4, s5, s4 +; SI-NEXT: s_mov_b32 s3, 0x100f000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: dynamic_insertelement_v2i16: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0 -; VI-NEXT: s_mov_b32 s7, 0x1100f000 -; VI-NEXT: s_mov_b32 s6, -1 ; VI-NEXT: s_waitcnt lgkmcnt(0) -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_lshl_b32 s1, s3, 4 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_xor_b32 s0, s2, 0x50005 -; VI-NEXT: s_lshl_b32 s1, 0xffff, s1 -; VI-NEXT: s_and_b32 s0, s0, s1 -; VI-NEXT: s_xor_b32 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_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_lshl_b32 s5, s5, 4 +; VI-NEXT: s_xor_b32 s6, s4, 0x50005 +; VI-NEXT: s_lshl_b32 s5, 0xffff, s5 +; VI-NEXT: s_and_b32 s5, s6, s5 +; VI-NEXT: s_xor_b32 s4, s5, s4 +; VI-NEXT: s_mov_b32 s3, 0x1100f000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm %vecins = insertelement <2 x i16> %a, i16 5, i32 %b store <2 x i16> %vecins, ptr addrspace(1) %out, align 8 diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ubfe.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ubfe.ll index d4aa2051..e421e2c 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ubfe.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ubfe.ll @@ -1612,29 +1612,27 @@ define amdgpu_kernel void @v_lshr_and(ptr addrspace(1) %out, i32 %a, i32 %b) #0 ; SI-LABEL: v_lshr_and: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_lshr_b32 s2, s2, s3 -; SI-NEXT: s_and_b32 s2, s2, 7 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_lshr_b32 s2, s4, s5 +; SI-NEXT: s_and_b32 s4, s2, 7 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: v_lshr_and: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; VI-NEXT: s_lshr_b32 s0, s2, s3 -; VI-NEXT: s_and_b32 s0, s0, 7 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_lshr_b32 s4, s4, s5 +; VI-NEXT: s_and_b32 s4, s4, 7 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm %c = lshr i32 %a, %b %d = and i32 %c, 7 diff --git a/llvm/test/CodeGen/AMDGPU/llvm.exp.ll b/llvm/test/CodeGen/AMDGPU/llvm.exp.ll index ac356fa..3897a0e 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.exp.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.exp.ll @@ -520,42 +520,41 @@ define amdgpu_kernel void @s_exp_v2f32(ptr addrspace(1) %out, <2 x float> %in) { ; ; SI-SDAG-LABEL: s_exp_v2f32: ; SI-SDAG: ; %bb.0: -; SI-SDAG-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x9 +; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; SI-SDAG-NEXT: v_mov_b32_e32 v0, 0x3fb8aa3b ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0x32a5705f -; SI-SDAG-NEXT: s_mov_b32 s3, 0xf000 -; SI-SDAG-NEXT: s_mov_b32 s2, -1 ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SI-SDAG-NEXT: v_mul_f32_e32 v2, s7, v0 +; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-SDAG-NEXT: v_mul_f32_e32 v2, s5, v0 ; SI-SDAG-NEXT: v_rndne_f32_e32 v3, v2 -; SI-SDAG-NEXT: v_fma_f32 v4, s7, v0, -v2 +; SI-SDAG-NEXT: v_fma_f32 v4, s5, v0, -v2 ; SI-SDAG-NEXT: v_sub_f32_e32 v2, v2, v3 -; SI-SDAG-NEXT: v_fma_f32 v4, s7, v1, v4 +; SI-SDAG-NEXT: v_fma_f32 v4, s5, v1, v4 ; SI-SDAG-NEXT: v_add_f32_e32 v2, v2, v4 -; SI-SDAG-NEXT: v_mul_f32_e32 v5, s6, v0 -; SI-SDAG-NEXT: v_exp_f32_e32 v2, v2 +; SI-SDAG-NEXT: v_mul_f32_e32 v5, s4, v0 ; SI-SDAG-NEXT: v_cvt_i32_f32_e32 v3, v3 +; SI-SDAG-NEXT: v_exp_f32_e32 v2, v2 ; SI-SDAG-NEXT: v_rndne_f32_e32 v6, v5 -; SI-SDAG-NEXT: v_fma_f32 v0, s6, v0, -v5 +; SI-SDAG-NEXT: v_fma_f32 v0, s4, v0, -v5 ; SI-SDAG-NEXT: v_sub_f32_e32 v7, v5, v6 -; SI-SDAG-NEXT: v_fma_f32 v0, s6, v1, v0 +; SI-SDAG-NEXT: v_fma_f32 v0, s4, v1, v0 ; SI-SDAG-NEXT: v_add_f32_e32 v0, v7, v0 ; SI-SDAG-NEXT: v_exp_f32_e32 v0, v0 ; SI-SDAG-NEXT: v_cvt_i32_f32_e32 v5, v6 ; SI-SDAG-NEXT: v_ldexp_f32_e32 v2, v2, v3 ; SI-SDAG-NEXT: v_mov_b32_e32 v3, 0xc2ce8ed0 -; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v3 +; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s5, v3 ; SI-SDAG-NEXT: v_mov_b32_e32 v4, 0x42b17218 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc ; SI-SDAG-NEXT: v_mov_b32_e32 v6, 0x7f800000 -; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s7, v4 +; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s5, v4 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v1, v6, v2, vcc ; SI-SDAG-NEXT: v_ldexp_f32_e32 v0, v0, v5 -; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s6, v3 +; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s4, v3 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc -; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v4 -; SI-SDAG-NEXT: s_mov_b32 s0, s4 -; SI-SDAG-NEXT: s_mov_b32 s1, s5 +; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s4, v4 +; SI-SDAG-NEXT: s_mov_b32 s3, 0xf000 +; SI-SDAG-NEXT: s_mov_b32 s2, -1 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc ; SI-SDAG-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-SDAG-NEXT: s_endpgm diff --git a/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll b/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll index d12ebe4..3928ec2 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll @@ -522,42 +522,41 @@ define amdgpu_kernel void @s_exp10_v2f32(ptr addrspace(1) %out, <2 x float> %in) ; ; SI-SDAG-LABEL: s_exp10_v2f32: ; SI-SDAG: ; %bb.0: -; SI-SDAG-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x9 +; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; SI-SDAG-NEXT: v_mov_b32_e32 v0, 0x40549a78 ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0x33979a37 -; SI-SDAG-NEXT: s_mov_b32 s3, 0xf000 -; SI-SDAG-NEXT: s_mov_b32 s2, -1 ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SI-SDAG-NEXT: v_mul_f32_e32 v2, s7, v0 +; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-SDAG-NEXT: v_mul_f32_e32 v2, s5, v0 ; SI-SDAG-NEXT: v_rndne_f32_e32 v3, v2 -; SI-SDAG-NEXT: v_fma_f32 v4, s7, v0, -v2 +; SI-SDAG-NEXT: v_fma_f32 v4, s5, v0, -v2 ; SI-SDAG-NEXT: v_sub_f32_e32 v2, v2, v3 -; SI-SDAG-NEXT: v_fma_f32 v4, s7, v1, v4 +; SI-SDAG-NEXT: v_fma_f32 v4, s5, v1, v4 ; SI-SDAG-NEXT: v_add_f32_e32 v2, v2, v4 -; SI-SDAG-NEXT: v_mul_f32_e32 v5, s6, v0 -; SI-SDAG-NEXT: v_exp_f32_e32 v2, v2 +; SI-SDAG-NEXT: v_mul_f32_e32 v5, s4, v0 ; SI-SDAG-NEXT: v_cvt_i32_f32_e32 v3, v3 +; SI-SDAG-NEXT: v_exp_f32_e32 v2, v2 ; SI-SDAG-NEXT: v_rndne_f32_e32 v6, v5 -; SI-SDAG-NEXT: v_fma_f32 v0, s6, v0, -v5 +; SI-SDAG-NEXT: v_fma_f32 v0, s4, v0, -v5 ; SI-SDAG-NEXT: v_sub_f32_e32 v7, v5, v6 -; SI-SDAG-NEXT: v_fma_f32 v0, s6, v1, v0 +; SI-SDAG-NEXT: v_fma_f32 v0, s4, v1, v0 ; SI-SDAG-NEXT: v_add_f32_e32 v0, v7, v0 ; SI-SDAG-NEXT: v_exp_f32_e32 v0, v0 ; SI-SDAG-NEXT: v_cvt_i32_f32_e32 v5, v6 ; SI-SDAG-NEXT: v_ldexp_f32_e32 v2, v2, v3 ; SI-SDAG-NEXT: v_mov_b32_e32 v3, 0xc23369f4 -; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s7, v3 +; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s5, v3 ; SI-SDAG-NEXT: v_mov_b32_e32 v4, 0x421a209b ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc ; SI-SDAG-NEXT: v_mov_b32_e32 v6, 0x7f800000 -; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s7, v4 +; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s5, v4 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v1, v6, v2, vcc ; SI-SDAG-NEXT: v_ldexp_f32_e32 v0, v0, v5 -; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s6, v3 +; SI-SDAG-NEXT: v_cmp_nlt_f32_e32 vcc, s4, v3 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc -; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s6, v4 -; SI-SDAG-NEXT: s_mov_b32 s0, s4 -; SI-SDAG-NEXT: s_mov_b32 s1, s5 +; SI-SDAG-NEXT: v_cmp_ngt_f32_e32 vcc, s4, v4 +; SI-SDAG-NEXT: s_mov_b32 s3, 0xf000 +; SI-SDAG-NEXT: s_mov_b32 s2, -1 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc ; SI-SDAG-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-SDAG-NEXT: s_endpgm diff --git a/llvm/test/CodeGen/AMDGPU/llvm.exp2.ll b/llvm/test/CodeGen/AMDGPU/llvm.exp2.ll index e30a586..dd44a1a 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.exp2.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.exp2.ll @@ -176,26 +176,25 @@ define amdgpu_kernel void @s_exp2_v2f32(ptr addrspace(1) %out, <2 x float> %in) ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; SI-SDAG-NEXT: v_mov_b32_e32 v0, 0xc2fc0000 ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0x42800000 -; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 -; SI-SDAG-NEXT: s_mov_b32 s6, -1 ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s3, v0 +; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s5, v0 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, 0, v1, vcc -; SI-SDAG-NEXT: s_mov_b32 s4, s0 -; SI-SDAG-NEXT: s_mov_b32 s5, s1 -; SI-SDAG-NEXT: s_and_b64 s[0:1], vcc, exec -; SI-SDAG-NEXT: v_add_f32_e32 v2, s3, v2 -; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s2, v0 -; SI-SDAG-NEXT: v_exp_f32_e32 v2, v2 +; SI-SDAG-NEXT: s_and_b64 s[6:7], vcc, exec +; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s4, v0 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc -; SI-SDAG-NEXT: v_add_f32_e32 v0, s2, v0 +; SI-SDAG-NEXT: v_add_f32_e32 v2, s5, v2 +; SI-SDAG-NEXT: v_add_f32_e32 v0, s4, v0 +; SI-SDAG-NEXT: v_exp_f32_e32 v2, v2 ; SI-SDAG-NEXT: v_exp_f32_e32 v0, v0 -; SI-SDAG-NEXT: s_cselect_b32 s0, 0xffffffc0, 0 -; SI-SDAG-NEXT: v_ldexp_f32_e64 v1, v2, s0 -; SI-SDAG-NEXT: s_and_b64 s[0:1], vcc, exec -; SI-SDAG-NEXT: s_cselect_b32 s0, 0xffffffc0, 0 -; SI-SDAG-NEXT: v_ldexp_f32_e64 v0, v0, s0 -; SI-SDAG-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-SDAG-NEXT: s_cselect_b32 s6, 0xffffffc0, 0 +; SI-SDAG-NEXT: s_and_b64 s[4:5], vcc, exec +; SI-SDAG-NEXT: s_cselect_b32 s4, 0xffffffc0, 0 +; SI-SDAG-NEXT: s_mov_b32 s3, 0xf000 +; SI-SDAG-NEXT: s_mov_b32 s2, -1 +; SI-SDAG-NEXT: v_ldexp_f32_e64 v1, v2, s6 +; SI-SDAG-NEXT: v_ldexp_f32_e64 v0, v0, s4 +; SI-SDAG-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-SDAG-NEXT: s_endpgm ; ; SI-GISEL-LABEL: s_exp2_v2f32: diff --git a/llvm/test/CodeGen/AMDGPU/llvm.log.ll b/llvm/test/CodeGen/AMDGPU/llvm.log.ll index b5038c8..fc6b2d9 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.log.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.log.ll @@ -321,39 +321,38 @@ define amdgpu_kernel void @s_log_f32(ptr addrspace(1) %out, float %in) { define amdgpu_kernel void @s_log_v2f32(ptr addrspace(1) %out, <2 x float> %in) { ; SI-SDAG-LABEL: s_log_v2f32: ; SI-SDAG: ; %bb.0: -; SI-SDAG-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x9 +; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; SI-SDAG-NEXT: v_mov_b32_e32 v0, 0x800000 ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0x41b17218 -; SI-SDAG-NEXT: s_mov_b32 s8, 0x3377d1cf +; SI-SDAG-NEXT: s_mov_b32 s8, 0x3f317217 ; SI-SDAG-NEXT: s_mov_b32 s9, 0x7f800000 ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s7, v0 -; SI-SDAG-NEXT: s_and_b64 s[0:1], vcc, exec -; SI-SDAG-NEXT: s_cselect_b32 s0, 32, 0 -; SI-SDAG-NEXT: v_mov_b32_e32 v3, s0 -; SI-SDAG-NEXT: v_ldexp_f32_e32 v3, s7, v3 +; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s5, v0 +; SI-SDAG-NEXT: s_and_b64 s[2:3], vcc, exec +; SI-SDAG-NEXT: s_cselect_b32 s2, 32, 0 +; SI-SDAG-NEXT: v_mov_b32_e32 v3, s2 +; SI-SDAG-NEXT: v_ldexp_f32_e32 v3, s5, v3 ; SI-SDAG-NEXT: v_log_f32_e32 v3, v3 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, 0, v1, vcc -; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s6, v0 -; SI-SDAG-NEXT: s_mov_b32 s0, s4 -; SI-SDAG-NEXT: s_mov_b32 s1, s5 -; SI-SDAG-NEXT: s_and_b64 s[4:5], vcc, exec -; SI-SDAG-NEXT: s_mov_b32 s7, 0x3f317217 +; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s4, v0 +; SI-SDAG-NEXT: s_and_b64 s[6:7], vcc, exec ; SI-SDAG-NEXT: v_mul_f32_e32 v4, 0x3f317217, v3 -; SI-SDAG-NEXT: s_cselect_b32 s4, 32, 0 -; SI-SDAG-NEXT: v_fma_f32 v5, v3, s7, -v4 +; SI-SDAG-NEXT: s_cselect_b32 s6, 32, 0 +; SI-SDAG-NEXT: s_mov_b32 s5, 0x3377d1cf +; SI-SDAG-NEXT: v_fma_f32 v5, v3, s8, -v4 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc -; SI-SDAG-NEXT: v_mov_b32_e32 v1, s4 -; SI-SDAG-NEXT: v_fma_f32 v5, v3, s8, v5 -; SI-SDAG-NEXT: v_ldexp_f32_e32 v1, s6, v1 +; SI-SDAG-NEXT: v_mov_b32_e32 v1, s6 +; SI-SDAG-NEXT: v_fma_f32 v5, v3, s5, v5 +; SI-SDAG-NEXT: v_ldexp_f32_e32 v1, s4, v1 ; SI-SDAG-NEXT: v_add_f32_e32 v4, v4, v5 ; SI-SDAG-NEXT: v_log_f32_e32 v5, v1 ; SI-SDAG-NEXT: v_cmp_lt_f32_e64 vcc, |v3|, s9 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc ; SI-SDAG-NEXT: v_sub_f32_e32 v1, v1, v2 ; SI-SDAG-NEXT: v_mul_f32_e32 v2, 0x3f317217, v5 -; SI-SDAG-NEXT: v_fma_f32 v3, v5, s7, -v2 -; SI-SDAG-NEXT: v_fma_f32 v3, v5, s8, v3 +; SI-SDAG-NEXT: v_fma_f32 v3, v5, s8, -v2 +; SI-SDAG-NEXT: v_fma_f32 v3, v5, s5, v3 ; SI-SDAG-NEXT: v_add_f32_e32 v2, v2, v3 ; SI-SDAG-NEXT: v_cmp_lt_f32_e64 vcc, |v5|, s9 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, v5, v2, vcc diff --git a/llvm/test/CodeGen/AMDGPU/llvm.log10.ll b/llvm/test/CodeGen/AMDGPU/llvm.log10.ll index 7465b49..a141bce 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.log10.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.log10.ll @@ -321,39 +321,38 @@ define amdgpu_kernel void @s_log10_f32(ptr addrspace(1) %out, float %in) { define amdgpu_kernel void @s_log10_v2f32(ptr addrspace(1) %out, <2 x float> %in) { ; SI-SDAG-LABEL: s_log10_v2f32: ; SI-SDAG: ; %bb.0: -; SI-SDAG-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x9 +; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; SI-SDAG-NEXT: v_mov_b32_e32 v0, 0x800000 ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0x411a209b -; SI-SDAG-NEXT: s_mov_b32 s8, 0x3284fbcf +; SI-SDAG-NEXT: s_mov_b32 s8, 0x3e9a209a ; SI-SDAG-NEXT: s_mov_b32 s9, 0x7f800000 ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) -; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s7, v0 -; SI-SDAG-NEXT: s_and_b64 s[0:1], vcc, exec -; SI-SDAG-NEXT: s_cselect_b32 s0, 32, 0 -; SI-SDAG-NEXT: v_mov_b32_e32 v3, s0 -; SI-SDAG-NEXT: v_ldexp_f32_e32 v3, s7, v3 +; SI-SDAG-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s5, v0 +; SI-SDAG-NEXT: s_and_b64 s[2:3], vcc, exec +; SI-SDAG-NEXT: s_cselect_b32 s2, 32, 0 +; SI-SDAG-NEXT: v_mov_b32_e32 v3, s2 +; SI-SDAG-NEXT: v_ldexp_f32_e32 v3, s5, v3 ; SI-SDAG-NEXT: v_log_f32_e32 v3, v3 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, 0, v1, vcc -; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s6, v0 -; SI-SDAG-NEXT: s_mov_b32 s0, s4 -; SI-SDAG-NEXT: s_mov_b32 s1, s5 -; SI-SDAG-NEXT: s_and_b64 s[4:5], vcc, exec -; SI-SDAG-NEXT: s_mov_b32 s7, 0x3e9a209a +; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s4, v0 +; SI-SDAG-NEXT: s_and_b64 s[6:7], vcc, exec ; SI-SDAG-NEXT: v_mul_f32_e32 v4, 0x3e9a209a, v3 -; SI-SDAG-NEXT: s_cselect_b32 s4, 32, 0 -; SI-SDAG-NEXT: v_fma_f32 v5, v3, s7, -v4 +; SI-SDAG-NEXT: s_cselect_b32 s6, 32, 0 +; SI-SDAG-NEXT: s_mov_b32 s5, 0x3284fbcf +; SI-SDAG-NEXT: v_fma_f32 v5, v3, s8, -v4 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc -; SI-SDAG-NEXT: v_mov_b32_e32 v1, s4 -; SI-SDAG-NEXT: v_fma_f32 v5, v3, s8, v5 -; SI-SDAG-NEXT: v_ldexp_f32_e32 v1, s6, v1 +; SI-SDAG-NEXT: v_mov_b32_e32 v1, s6 +; SI-SDAG-NEXT: v_fma_f32 v5, v3, s5, v5 +; SI-SDAG-NEXT: v_ldexp_f32_e32 v1, s4, v1 ; SI-SDAG-NEXT: v_add_f32_e32 v4, v4, v5 ; SI-SDAG-NEXT: v_log_f32_e32 v5, v1 ; SI-SDAG-NEXT: v_cmp_lt_f32_e64 vcc, |v3|, s9 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc ; SI-SDAG-NEXT: v_sub_f32_e32 v1, v1, v2 ; SI-SDAG-NEXT: v_mul_f32_e32 v2, 0x3e9a209a, v5 -; SI-SDAG-NEXT: v_fma_f32 v3, v5, s7, -v2 -; SI-SDAG-NEXT: v_fma_f32 v3, v5, s8, v3 +; SI-SDAG-NEXT: v_fma_f32 v3, v5, s8, -v2 +; SI-SDAG-NEXT: v_fma_f32 v3, v5, s5, v3 ; SI-SDAG-NEXT: v_add_f32_e32 v2, v2, v3 ; SI-SDAG-NEXT: v_cmp_lt_f32_e64 vcc, |v5|, s9 ; SI-SDAG-NEXT: v_cndmask_b32_e32 v2, v5, v2, vcc diff --git a/llvm/test/CodeGen/AMDGPU/llvm.log2.ll b/llvm/test/CodeGen/AMDGPU/llvm.log2.ll index 61a777f..b1407d3 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.log2.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.log2.ll @@ -221,8 +221,6 @@ define amdgpu_kernel void @s_log2_v2f32(ptr addrspace(1) %out, <2 x float> %in) ; SI-SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; SI-SDAG-NEXT: v_mov_b32_e32 v0, 0x800000 ; SI-SDAG-NEXT: v_mov_b32_e32 v1, 0x42000000 -; SI-SDAG-NEXT: s_mov_b32 s7, 0xf000 -; SI-SDAG-NEXT: s_mov_b32 s6, -1 ; SI-SDAG-NEXT: s_waitcnt lgkmcnt(0) ; SI-SDAG-NEXT: v_cmp_lt_f32_e32 vcc, s3, v0 ; SI-SDAG-NEXT: s_and_b64 s[4:5], vcc, exec @@ -238,11 +236,11 @@ define amdgpu_kernel void @s_log2_v2f32(ptr addrspace(1) %out, <2 x float> %in) ; SI-SDAG-NEXT: v_ldexp_f32_e32 v1, s2, v1 ; SI-SDAG-NEXT: v_log_f32_e32 v3, v3 ; SI-SDAG-NEXT: v_log_f32_e32 v4, v1 -; SI-SDAG-NEXT: s_mov_b32 s4, s0 -; SI-SDAG-NEXT: s_mov_b32 s5, s1 +; SI-SDAG-NEXT: s_mov_b32 s3, 0xf000 +; SI-SDAG-NEXT: s_mov_b32 s2, -1 ; SI-SDAG-NEXT: v_sub_f32_e32 v1, v3, v2 ; SI-SDAG-NEXT: v_sub_f32_e32 v0, v4, v0 -; SI-SDAG-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-SDAG-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-SDAG-NEXT: s_endpgm ; ; SI-GISEL-LABEL: s_log2_v2f32: diff --git a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fadd.ll b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fadd.ll index 6dc9199..b6eaaf1 100644 --- a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fadd.ll +++ b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fadd.ll @@ -326,12 +326,12 @@ define void @local_atomic_fadd_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v2, 4.0, v1 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_add_f32_e32 v1, 4.0, v2 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB2_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -347,12 +347,12 @@ define void @local_atomic_fadd_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v2, 4.0, v1 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_add_f32_e32 v1, 4.0, v2 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB2_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -440,12 +440,12 @@ define void @local_atomic_fadd_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v2, 4.0, v1 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_add_f32_e32 v1, 4.0, v2 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB3_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -462,12 +462,12 @@ define void @local_atomic_fadd_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v2, 4.0, v1 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_add_f32_e32 v1, 4.0, v2 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB3_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -880,13 +880,14 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX12-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f64_e32 v[3:4], 4.0, v[1:2] +; GFX12-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f64_e32 v[1:2], 4.0, v[3:4] ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] +; GFX12-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX12-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -913,13 +914,14 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX11-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 +; GFX11-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] +; GFX11-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX11-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -936,14 +938,14 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 +; GFX10-NEXT: v_mov_b32_e32 v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v3, v1 +; GFX10-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX10-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX10-NEXT: v_mov_b32_e32 v1, v3 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB6_1 @@ -968,13 +970,13 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX908-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX908-NEXT: v_mov_b32_e32 v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v3, v1 +; GFX908-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX908-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX908-NEXT: v_mov_b32_e32 v1, v3 +; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB6_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -990,13 +992,13 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX8-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX8-NEXT: v_mov_b32_e32 v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v3, v1 +; GFX8-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX8-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX8-NEXT: v_mov_b32_e32 v1, v3 +; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB6_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1012,13 +1014,13 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX7-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX7-NEXT: v_mov_b32_e32 v4, v2 +; GFX7-NEXT: v_mov_b32_e32 v3, v1 +; GFX7-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX7-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX7-NEXT: v_mov_b32_e32 v1, v3 +; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v2, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB6_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1034,13 +1036,13 @@ define void @local_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX6-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX6-NEXT: v_mov_b32_e32 v4, v2 +; GFX6-NEXT: v_mov_b32_e32 v3, v1 +; GFX6-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX6-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX6-NEXT: v_mov_b32_e32 v1, v3 +; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v2, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB6_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1063,13 +1065,14 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f64_e32 v[3:4], 4.0, v[1:2] +; GFX12-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f64_e32 v[1:2], 4.0, v[3:4] ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] offset:65528 +; GFX12-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] offset:65528 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX12-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -1096,13 +1099,14 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 +; GFX11-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] offset:65528 +; GFX11-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] offset:65528 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX11-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1119,14 +1123,14 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 +; GFX10-NEXT: v_mov_b32_e32 v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v3, v1 +; GFX10-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX10-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX10-NEXT: v_mov_b32_e32 v1, v3 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB7_1 @@ -1151,13 +1155,13 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX908-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX908-NEXT: v_mov_b32_e32 v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v3, v1 +; GFX908-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX908-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX908-NEXT: v_mov_b32_e32 v1, v3 +; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB7_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1173,13 +1177,13 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX8-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX8-NEXT: v_mov_b32_e32 v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v3, v1 +; GFX8-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX8-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX8-NEXT: v_mov_b32_e32 v1, v3 +; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB7_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1195,13 +1199,13 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f64 v[3:4], v[1:2], 4.0 -; GFX7-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX7-NEXT: v_mov_b32_e32 v4, v2 +; GFX7-NEXT: v_mov_b32_e32 v3, v1 +; GFX7-NEXT: v_add_f64 v[1:2], v[3:4], 4.0 +; GFX7-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX7-NEXT: v_mov_b32_e32 v1, v3 +; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v2, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB7_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1218,13 +1222,13 @@ define void @local_atomic_fadd_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f64 v[3:4], v[0:1], 4.0 -; GFX6-NEXT: ds_cmpst_rtn_b64 v[3:4], v2, v[0:1], v[3:4] +; GFX6-NEXT: v_mov_b32_e32 v4, v1 +; GFX6-NEXT: v_mov_b32_e32 v3, v0 +; GFX6-NEXT: v_add_f64 v[0:1], v[3:4], 4.0 +; GFX6-NEXT: ds_cmpst_rtn_b64 v[0:1], v2, v[3:4], v[0:1] ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[0:1] -; GFX6-NEXT: v_mov_b32_e32 v0, v3 +; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[3:4] ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB7_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2032,27 +2036,27 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: v_add_f16_e32 v4.l, 4.0, v4.l +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_add_f16_e32 v3.l, 4.0, v3.l ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2073,28 +2077,28 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_add_f16_e32 v4, 4.0, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_add_f16_e32 v3, 4.0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2119,15 +2123,15 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX942-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX942-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB10_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2140,27 +2144,27 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: v_add_f16_e32 v4.l, 4.0, v4.l +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_add_f16_e32 v3.l, 4.0, v3.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2175,28 +2179,28 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_add_f16_e32 v4, 4.0, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2211,23 +2215,23 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX10-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX10-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB10_1 @@ -2249,15 +2253,15 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX90A-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX90A-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB10_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2278,15 +2282,15 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX908-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX908-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB10_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2308,16 +2312,16 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX8-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX8-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB10_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2338,18 +2342,18 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB10_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2370,18 +2374,18 @@ define void @local_atomic_fadd_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB10_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2412,19 +2416,19 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add_f16_e32 v4.l, 4.0, v4.l -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_add_f16_e32 v3.l, 4.0, v3.l +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2455,19 +2459,20 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_add_f16_e32 v3, 4.0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2493,15 +2498,15 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX942-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX942-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX942-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB11_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2524,19 +2529,19 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add_f16_e32 v4.l, 4.0, v4.l -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_add_f16_e32 v3.l, 4.0, v3.l +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2561,19 +2566,20 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_add_f16_e32 v3, 4.0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2596,16 +2602,16 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX10-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX10-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB11_1 @@ -2628,15 +2634,15 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX90A-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX90A-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX90A-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB11_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2658,15 +2664,15 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX908-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX908-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB11_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2689,16 +2695,16 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX8-NEXT: v_add_f16_e32 v4, 4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX8-NEXT: v_add_f16_e32 v3, 4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB11_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2720,18 +2726,18 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB11_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2753,18 +2759,18 @@ define void @local_atomic_fadd_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB11_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3086,16 +3092,16 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_add_f16_e32 v2.l, 4.0, v1.l -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add_f16_e32 v1.l, 4.0, v2.l +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -3118,16 +3124,17 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_add_f16_e32 v2, 4.0, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -3147,13 +3154,13 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f16_e32 v2, 4.0, v1 -; GFX942-NEXT: v_and_or_b32 v2, v1, s2, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX942-NEXT: v_and_or_b32 v1, v2, s2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB13_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3168,16 +3175,16 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_add_f16_e32 v2.l, 4.0, v1.l -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add_f16_e32 v1.l, 4.0, v2.l +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -3194,16 +3201,17 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_add_f16_e32 v2, 4.0, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -3220,15 +3228,15 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f16_e32 v2, 4.0, v1 -; GFX10-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX10-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB13_1 @@ -3245,13 +3253,13 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f16_e32 v2, 4.0, v1 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB13_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3267,13 +3275,13 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f16_e32 v2, 4.0, v1 -; GFX908-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX908-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB13_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3289,14 +3297,14 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f16_e32 v2, 4.0, v1 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_add_f16_e32 v1, 4.0, v2 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB13_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3312,16 +3320,16 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX7-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB13_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3338,16 +3346,16 @@ define void @local_atomic_fadd_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB13_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4297,38 +4305,38 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -4349,37 +4357,37 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4405,22 +4413,22 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB16_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4433,38 +4441,38 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4479,37 +4487,37 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .p2align 6 ; GFX11-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4524,28 +4532,28 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB16_1 @@ -4568,20 +4576,20 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB16_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4603,20 +4611,20 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB16_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4638,22 +4646,22 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB16_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4674,18 +4682,18 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB16_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4706,18 +4714,18 @@ define void @local_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB16_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4748,29 +4756,30 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-TRUE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -4801,28 +4810,29 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4849,22 +4859,22 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX942-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB17_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4888,28 +4898,29 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-TRUE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4935,27 +4946,28 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4978,21 +4990,21 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX10-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB17_1 @@ -5016,20 +5028,20 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX90A-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB17_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5052,20 +5064,20 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX908-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB17_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5088,22 +5100,22 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX8-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB17_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5125,18 +5137,18 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX7-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB17_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5158,18 +5170,18 @@ define void @local_atomic_fadd_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX6-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_add_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_add_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB17_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5569,26 +5581,27 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-TRUE16-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -5611,25 +5624,26 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-FAKE16-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -5650,21 +5664,21 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX942-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX942-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX942-NEXT: v_add3_u32 v3, v3, v2, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX942-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX942-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX942-NEXT: v_add3_u32 v3, v3, v1, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX942-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX942-NEXT: v_and_or_b32 v2, v1, s3, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX942-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX942-NEXT: v_and_or_b32 v1, v2, s3, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB19_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5680,25 +5694,26 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5716,24 +5731,25 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5750,21 +5766,21 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX10-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX10-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX10-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo -; GFX10-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX10-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX10-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX10-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX10-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB19_1 @@ -5782,20 +5798,20 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX90A-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX90A-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX90A-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX90A-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX90A-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX90A-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX90A-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX90A-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX90A-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX90A-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB19_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5812,20 +5828,20 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX908-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX908-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX908-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX908-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX908-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX908-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX908-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX908-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX908-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX908-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX908-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX908-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB19_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5841,21 +5857,21 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX8-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX8-NEXT: v_bfe_u32 v4, v2, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX8-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX8-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v1 ; GFX8-NEXT: v_add_u32_e32 v4, vcc, 0x7fff, v4 -; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v2 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc -; GFX8-NEXT: v_or_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v1 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v5, vcc +; GFX8-NEXT: v_or_b32_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB19_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5871,16 +5887,16 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX7-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX7-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB19_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5897,16 +5913,16 @@ define void @local_atomic_fadd_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX6-NEXT: v_add_f32_e32 v2, 4.0, v2 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX6-NEXT: v_add_f32_e32 v1, 4.0, v1 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB19_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6399,13 +6415,14 @@ define void @local_atomic_fadd_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX11-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_add_f16 v3, v2, v1 +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_pk_add_f16 v2, v3, v1 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6422,13 +6439,13 @@ define void @local_atomic_fadd_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX10-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_add_f16 v3, v2, v1 +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_add_f16 v2, v3, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB22_1 @@ -6444,12 +6461,12 @@ define void @local_atomic_fadd_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX90A-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_add_f16 v3, v2, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_add_f16 v2, v3, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB22_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6464,12 +6481,12 @@ define void @local_atomic_fadd_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX908-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_add_f16 v3, v2, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_add_f16 v2, v3, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB22_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6485,14 +6502,14 @@ define void @local_atomic_fadd_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX8-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v3, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v4, v2, v1 -; GFX8-NEXT: v_or_b32_e32 v3, v4, v3 -; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX8-NEXT: v_mov_b32_e32 v3, v2 +; GFX8-NEXT: v_add_f16_sdwa v2, v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_add_f16_e32 v4, v3, v1 +; GFX8-NEXT: v_or_b32_e32 v2, v4, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v3 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB22_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6614,13 +6631,14 @@ define void @local_atomic_fadd_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX11-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_add_f16 v3, v2, v1 +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_pk_add_f16 v2, v3, v1 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6637,13 +6655,13 @@ define void @local_atomic_fadd_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX10-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_add_f16 v3, v2, v1 +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_add_f16 v2, v3, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB23_1 @@ -6659,12 +6677,12 @@ define void @local_atomic_fadd_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX90A-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_add_f16 v3, v2, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_add_f16 v2, v3, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB23_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6679,12 +6697,12 @@ define void @local_atomic_fadd_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX908-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_add_f16 v3, v2, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_add_f16 v2, v3, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB23_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6700,14 +6718,14 @@ define void @local_atomic_fadd_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX8-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f16_sdwa v3, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_add_f16_e32 v4, v2, v1 -; GFX8-NEXT: v_or_b32_e32 v3, v4, v3 -; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX8-NEXT: v_mov_b32_e32 v3, v2 +; GFX8-NEXT: v_add_f16_sdwa v2, v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_add_f16_e32 v4, v3, v1 +; GFX8-NEXT: v_or_b32_e32 v2, v4, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v3 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB23_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7547,30 +7565,32 @@ define void @local_atomic_fadd_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_add_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_add_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -7591,30 +7611,32 @@ define void @local_atomic_fadd_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_add_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -7634,27 +7656,27 @@ define void @local_atomic_fadd_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX10-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB26_1 @@ -7674,26 +7696,26 @@ define void @local_atomic_fadd_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX90A-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB26_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7712,26 +7734,26 @@ define void @local_atomic_fadd_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX908-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB26_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7749,29 +7771,29 @@ define void @local_atomic_fadd_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX8-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB26_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7888,30 +7910,32 @@ define void @local_atomic_fadd_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_add_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_add_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -7932,30 +7956,32 @@ define void @local_atomic_fadd_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_add_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -7975,27 +8001,27 @@ define void @local_atomic_fadd_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX10-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB27_1 @@ -8015,26 +8041,26 @@ define void @local_atomic_fadd_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX90A-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB27_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8053,26 +8079,26 @@ define void @local_atomic_fadd_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX908-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB27_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8090,29 +8116,29 @@ define void @local_atomic_fadd_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX8-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_add_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_add_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_add_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB27_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8849,20 +8875,20 @@ define amdgpu_kernel void @local_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) ; GFX7-NEXT: ; %bb.5: ; GFX7-NEXT: s_lshl_b32 s0, s3, 4 ; GFX7-NEXT: v_mov_b32_e32 v1, s0 -; GFX7-NEXT: ds_read_b32 v3, v1 +; GFX7-NEXT: ds_read_b32 v2, v1 ; GFX7-NEXT: s_bcnt1_i32_b64 s0, s[8:9] -; GFX7-NEXT: v_cvt_f32_ubyte0_e32 v2, s0 -; GFX7-NEXT: v_mul_f32_e32 v2, 0x42280000, v2 +; GFX7-NEXT: v_cvt_f32_ubyte0_e32 v3, s0 +; GFX7-NEXT: v_mul_f32_e32 v3, 0x42280000, v3 ; GFX7-NEXT: s_mov_b64 s[8:9], 0 ; GFX7-NEXT: .LBB28_6: ; %atomicrmw.start2 ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v4, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v2 +; GFX7-NEXT: v_add_f32_e32 v2, v4, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v1, v4, v2 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e64 s[0:1], v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e64 s[0:1], v2, v4 ; GFX7-NEXT: s_or_b64 s[8:9], s[0:1], s[8:9] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GFX7-NEXT: s_cbranch_execnz .LBB28_6 ; GFX7-NEXT: .LBB28_7: ; %Flow21 @@ -8973,20 +8999,20 @@ define amdgpu_kernel void @local_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) ; GFX6-NEXT: ; %bb.5: ; GFX6-NEXT: s_lshl_b32 s0, s3, 4 ; GFX6-NEXT: v_mov_b32_e32 v1, s0 -; GFX6-NEXT: ds_read_b32 v3, v1 +; GFX6-NEXT: ds_read_b32 v2, v1 ; GFX6-NEXT: s_bcnt1_i32_b64 s0, s[8:9] -; GFX6-NEXT: v_cvt_f32_ubyte0_e32 v2, s0 -; GFX6-NEXT: v_mul_f32_e32 v2, 0x42280000, v2 +; GFX6-NEXT: v_cvt_f32_ubyte0_e32 v3, s0 +; GFX6-NEXT: v_mul_f32_e32 v3, 0x42280000, v3 ; GFX6-NEXT: s_mov_b64 s[8:9], 0 ; GFX6-NEXT: .LBB28_6: ; %atomicrmw.start2 ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v4, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v2 +; GFX6-NEXT: v_add_f32_e32 v2, v4, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v1, v4, v2 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e64 s[0:1], v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e64 s[0:1], v2, v4 ; GFX6-NEXT: s_or_b64 s[8:9], s[0:1], s[8:9] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GFX6-NEXT: s_cbranch_execnz .LBB28_6 ; GFX6-NEXT: .LBB28_7: ; %Flow19 @@ -9677,20 +9703,20 @@ define amdgpu_kernel void @local_ds_fadd_one_as(ptr addrspace(1) %out, ptr addrs ; GFX7-NEXT: ; %bb.5: ; GFX7-NEXT: s_lshl_b32 s0, s3, 4 ; GFX7-NEXT: v_mov_b32_e32 v1, s0 -; GFX7-NEXT: ds_read_b32 v3, v1 +; GFX7-NEXT: ds_read_b32 v2, v1 ; GFX7-NEXT: s_bcnt1_i32_b64 s0, s[8:9] -; GFX7-NEXT: v_cvt_f32_ubyte0_e32 v2, s0 -; GFX7-NEXT: v_mul_f32_e32 v2, 0x42280000, v2 +; GFX7-NEXT: v_cvt_f32_ubyte0_e32 v3, s0 +; GFX7-NEXT: v_mul_f32_e32 v3, 0x42280000, v3 ; GFX7-NEXT: s_mov_b64 s[8:9], 0 ; GFX7-NEXT: .LBB29_6: ; %atomicrmw.start2 ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v4, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v2 +; GFX7-NEXT: v_add_f32_e32 v2, v4, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v1, v4, v2 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e64 s[0:1], v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e64 s[0:1], v2, v4 ; GFX7-NEXT: s_or_b64 s[8:9], s[0:1], s[8:9] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GFX7-NEXT: s_cbranch_execnz .LBB29_6 ; GFX7-NEXT: .LBB29_7: ; %Flow21 @@ -9801,20 +9827,20 @@ define amdgpu_kernel void @local_ds_fadd_one_as(ptr addrspace(1) %out, ptr addrs ; GFX6-NEXT: ; %bb.5: ; GFX6-NEXT: s_lshl_b32 s0, s3, 4 ; GFX6-NEXT: v_mov_b32_e32 v1, s0 -; GFX6-NEXT: ds_read_b32 v3, v1 +; GFX6-NEXT: ds_read_b32 v2, v1 ; GFX6-NEXT: s_bcnt1_i32_b64 s0, s[8:9] -; GFX6-NEXT: v_cvt_f32_ubyte0_e32 v2, s0 -; GFX6-NEXT: v_mul_f32_e32 v2, 0x42280000, v2 +; GFX6-NEXT: v_cvt_f32_ubyte0_e32 v3, s0 +; GFX6-NEXT: v_mul_f32_e32 v3, 0x42280000, v3 ; GFX6-NEXT: s_mov_b64 s[8:9], 0 ; GFX6-NEXT: .LBB29_6: ; %atomicrmw.start2 ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v4, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v2 +; GFX6-NEXT: v_add_f32_e32 v2, v4, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v1, v4, v2 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e64 s[0:1], v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e64 s[0:1], v2, v4 ; GFX6-NEXT: s_or_b64 s[8:9], s[0:1], s[8:9] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[8:9] ; GFX6-NEXT: s_cbranch_execnz .LBB29_6 ; GFX6-NEXT: .LBB29_7: ; %Flow19 @@ -10084,12 +10110,12 @@ define void @local_atomic_fadd_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX7-NEXT: .LBB31_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v2, 4.0, v1 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_add_f32_e32 v1, 4.0, v2 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB31_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -10105,12 +10131,12 @@ define void @local_atomic_fadd_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX6-NEXT: .LBB31_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v2, 4.0, v1 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_add_f32_e32 v1, 4.0, v2 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB31_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end diff --git a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmax.ll b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmax.ll index d6b7d8f..8e094a7 100644 --- a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmax.ll +++ b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmax.ll @@ -1598,29 +1598,29 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v4.l, v4.l, v4.l +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v3.l, v3.l, v3.l ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v4.l, 4.0, v4.l -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v3.l, 4.0, v3.l +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -1641,29 +1641,29 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v4, v4, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v3, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -1688,16 +1688,16 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX942-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX942-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX942-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX942-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB10_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1710,29 +1710,29 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: v_max_f16_e32 v4.l, v4.l, v4.l +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v3.l, v3.l, v3.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_max_f16_e32 v4.l, 4.0, v4.l -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v3.l, 4.0, v3.l +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1747,29 +1747,29 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_max_f16_e32 v4, v4, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v3, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1784,24 +1784,24 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX10-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX10-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX10-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX10-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB10_1 @@ -1823,16 +1823,16 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX90A-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX90A-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX90A-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX90A-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB10_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1853,16 +1853,16 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX908-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX908-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX908-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX908-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB10_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1884,17 +1884,17 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX8-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX8-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX8-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX8-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB10_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1915,18 +1915,18 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB10_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1947,18 +1947,18 @@ define void @local_atomic_fmax_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB10_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1989,20 +1989,21 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v4.l, v4.l, v4.l -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v4.l, 4.0, v4.l +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v3.l, v3.l, v3.l ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v3.l, 4.0, v3.l +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2033,21 +2034,21 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v4, v4, v4 -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v4, 4.0, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v3, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2073,16 +2074,16 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX942-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX942-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX942-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX942-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX942-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB11_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2105,20 +2106,21 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_max_f16_e32 v4.l, v4.l, v4.l -; GFX11-TRUE16-NEXT: v_max_f16_e32 v4.l, 4.0, v4.l +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v3.l, v3.l, v3.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v3.l, 4.0, v3.l +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2143,21 +2145,21 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX11-FAKE16-NEXT: v_max_f16_e32 v4, 4.0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v3, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2180,17 +2182,17 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX10-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX10-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX10-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX10-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB11_1 @@ -2213,16 +2215,16 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX90A-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX90A-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX90A-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX90A-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX90A-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB11_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2244,16 +2246,16 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX908-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX908-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX908-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX908-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB11_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2276,17 +2278,17 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX8-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX8-NEXT: v_max_f16_e32 v4, 4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX8-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX8-NEXT: v_max_f16_e32 v3, 4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB11_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2308,18 +2310,18 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB11_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2341,18 +2343,18 @@ define void @local_atomic_fmax_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB11_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2685,17 +2687,18 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v2.l, v1.l, v1.l -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v2.l, 4.0, v2.l -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v1.l, v2.l, v2.l +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v1.l, 4.0, v1.l +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2718,18 +2721,18 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v2, v1, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v2, 4.0, v2 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v1, v2, v2 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v1, 4.0, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2749,14 +2752,14 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX942-NEXT: v_max_f16_e32 v2, 4.0, v2 -; GFX942-NEXT: v_and_or_b32 v2, v1, s2, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX942-NEXT: v_max_f16_e32 v1, 4.0, v1 +; GFX942-NEXT: v_and_or_b32 v1, v2, s2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB13_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2771,17 +2774,18 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_max_f16_e32 v2.l, v1.l, v1.l -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_max_f16_e32 v2.l, 4.0, v2.l -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v1.l, v2.l, v2.l +; GFX11-TRUE16-NEXT: v_max_f16_e32 v1.l, 4.0, v1.l +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2798,18 +2802,18 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_max_f16_e32 v2, v1, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_max_f16_e32 v2, 4.0, v2 -; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v1, 4.0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2826,16 +2830,16 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX10-NEXT: v_max_f16_e32 v2, 4.0, v2 -; GFX10-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX10-NEXT: v_max_f16_e32 v1, 4.0, v1 +; GFX10-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB13_1 @@ -2852,14 +2856,14 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX90A-NEXT: v_max_f16_e32 v2, 4.0, v2 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX90A-NEXT: v_max_f16_e32 v1, 4.0, v1 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB13_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2875,14 +2879,14 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX908-NEXT: v_max_f16_e32 v2, 4.0, v2 -; GFX908-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX908-NEXT: v_max_f16_e32 v1, 4.0, v1 +; GFX908-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB13_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2898,15 +2902,15 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_max_f16_e32 v2, 4.0, v2 -; GFX8-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_max_f16_e32 v1, 4.0, v1 +; GFX8-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB13_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2922,16 +2926,16 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX7-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB13_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2948,16 +2952,16 @@ define void @local_atomic_fmax_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB13_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3911,38 +3915,38 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v4, 4.0, v4 -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v3, 4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -3963,37 +3967,37 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4019,22 +4023,22 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB16_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4047,38 +4051,38 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4093,37 +4097,37 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .p2align 6 ; GFX11-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4138,28 +4142,28 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB16_1 @@ -4182,20 +4186,20 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB16_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4217,20 +4221,20 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB16_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4252,22 +4256,22 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB16_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4288,19 +4292,19 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX7-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX7-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB16_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4321,19 +4325,19 @@ define void @local_atomic_fmax_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX6-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX6-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB16_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4364,29 +4368,30 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v4, 4.0, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v3, 4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -4417,28 +4422,29 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4465,22 +4471,22 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX942-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB17_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4504,28 +4510,29 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-TRUE16-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4551,27 +4558,28 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4594,21 +4602,21 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX10-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB17_1 @@ -4632,20 +4640,20 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX90A-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB17_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4668,20 +4676,20 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX908-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB17_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4704,22 +4712,22 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX8-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB17_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4741,19 +4749,19 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX7-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX7-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX7-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB17_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4775,19 +4783,19 @@ define void @local_atomic_fmax_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX6-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX6-NEXT: v_max_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX6-NEXT: v_max_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB17_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5189,26 +5197,27 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v2, 4.0, v2 -; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v1, 4.0, v1 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -5231,25 +5240,26 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v2, 4.0, v2 -; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v1, 4.0, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -5270,21 +5280,21 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX942-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX942-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX942-NEXT: v_add3_u32 v3, v3, v2, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX942-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX942-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX942-NEXT: v_add3_u32 v3, v3, v1, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX942-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX942-NEXT: v_and_or_b32 v2, v1, s3, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX942-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX942-NEXT: v_and_or_b32 v1, v2, s3, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB19_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5300,25 +5310,26 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5336,24 +5347,25 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-FAKE16-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5370,21 +5382,21 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX10-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX10-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX10-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo -; GFX10-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX10-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX10-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX10-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX10-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB19_1 @@ -5402,20 +5414,20 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX90A-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX90A-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX90A-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX90A-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX90A-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX90A-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX90A-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX90A-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX90A-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX90A-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB19_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5432,20 +5444,20 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX908-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX908-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX908-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX908-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX908-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX908-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX908-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX908-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX908-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX908-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX908-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX908-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB19_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5461,21 +5473,21 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX8-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX8-NEXT: v_bfe_u32 v4, v2, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX8-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX8-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v1 ; GFX8-NEXT: v_add_u32_e32 v4, vcc, 0x7fff, v4 -; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v2 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc -; GFX8-NEXT: v_or_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v1 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v5, vcc +; GFX8-NEXT: v_or_b32_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB19_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5491,17 +5503,17 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v2 -; GFX7-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v1 +; GFX7-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB19_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5518,17 +5530,17 @@ define void @local_atomic_fmax_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX6-NEXT: v_mul_f32_e32 v2, 1.0, v2 -; GFX6-NEXT: v_max_f32_e32 v2, 4.0, v2 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX6-NEXT: v_mul_f32_e32 v1, 1.0, v1 +; GFX6-NEXT: v_max_f32_e32 v1, 4.0, v1 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB19_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6101,15 +6113,15 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX12-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_pk_max_num_f16 v3, v2, v2 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_pk_max_num_f16 v3, v3, v1 +; GFX12-NEXT: v_mov_b32_e32 v3, v2 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_pk_max_num_f16 v2, v3, v3 +; GFX12-NEXT: v_pk_max_num_f16 v2, v2, v1 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX12-NEXT: v_mov_b32_e32 v2, v3 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -6129,14 +6141,14 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX942-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_pk_max_f16 v3, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v3, v2 +; GFX942-NEXT: v_pk_max_f16 v2, v3, v3 ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_pk_max_f16 v3, v3, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX942-NEXT: v_pk_max_f16 v2, v2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v2, v3 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB22_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6152,15 +6164,15 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX11-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_pk_max_f16 v3, v3, v1 +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX11-NEXT: v_pk_max_f16 v2, v2, v1 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6178,14 +6190,14 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX10-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX10-NEXT: v_pk_max_f16 v3, v3, v1 +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX10-NEXT: v_pk_max_f16 v2, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB22_1 @@ -6202,13 +6214,13 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX90A-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX90A-NEXT: v_pk_max_f16 v3, v3, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX90A-NEXT: v_pk_max_f16 v2, v2, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB22_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6224,13 +6236,13 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX908-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX908-NEXT: v_pk_max_f16 v3, v3, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX908-NEXT: v_pk_max_f16 v2, v2, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB22_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6248,16 +6260,16 @@ define void @local_atomic_fmax_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX8-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_max_f16_sdwa v4, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_max_f16_e32 v5, v3, v3 -; GFX8-NEXT: v_max_f16_sdwa v4, v4, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_max_f16_sdwa v3, v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_max_f16_e32 v5, v4, v4 +; GFX8-NEXT: v_max_f16_sdwa v3, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX8-NEXT: v_max_f16_e32 v5, v5, v1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB22_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6363,15 +6375,15 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX12-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_pk_max_num_f16 v3, v2, v2 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_pk_max_num_f16 v3, v3, v1 +; GFX12-NEXT: v_mov_b32_e32 v3, v2 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_pk_max_num_f16 v2, v3, v3 +; GFX12-NEXT: v_pk_max_num_f16 v2, v2, v1 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX12-NEXT: v_mov_b32_e32 v2, v3 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -6391,14 +6403,14 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX942-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_pk_max_f16 v3, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v3, v2 +; GFX942-NEXT: v_pk_max_f16 v2, v3, v3 ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_pk_max_f16 v3, v3, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX942-NEXT: v_pk_max_f16 v2, v2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v2, v3 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB23_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6414,15 +6426,15 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX11-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_pk_max_f16 v3, v3, v1 +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX11-NEXT: v_pk_max_f16 v2, v2, v1 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6440,14 +6452,14 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX10-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX10-NEXT: v_pk_max_f16 v3, v3, v1 +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX10-NEXT: v_pk_max_f16 v2, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB23_1 @@ -6464,13 +6476,13 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX90A-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX90A-NEXT: v_pk_max_f16 v3, v3, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX90A-NEXT: v_pk_max_f16 v2, v2, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB23_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6486,13 +6498,13 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX908-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX908-NEXT: v_pk_max_f16 v3, v3, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX908-NEXT: v_pk_max_f16 v2, v2, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB23_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6510,16 +6522,16 @@ define void @local_atomic_fmax_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX8-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_max_f16_sdwa v4, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_max_f16_e32 v5, v3, v3 -; GFX8-NEXT: v_max_f16_sdwa v4, v4, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_max_f16_sdwa v3, v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_max_f16_e32 v5, v4, v4 +; GFX8-NEXT: v_max_f16_sdwa v3, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX8-NEXT: v_max_f16_e32 v5, v5, v1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB23_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7589,31 +7601,34 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX12-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_max_num_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v4, v4, v1 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v5, v5, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v3, v3, v1 ; GFX12-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX12-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -7638,32 +7653,33 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX12-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_max_num_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v4, v4, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v5, v5, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v3, v3, v2 ; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX12-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 +; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo +; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xf1ff -; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX12-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -7686,27 +7702,27 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX942-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX942-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX942-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX942-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX942-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX942-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX942-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX942-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX942-NEXT: v_add3_u32 v6, v6, v4, s4 +; GFX942-NEXT: v_add3_u32 v6, v6, v3, s4 ; GFX942-NEXT: v_add3_u32 v8, v8, v5, s4 ; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v4, v4 +; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v3, v3 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX942-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[0:1] -; GFX942-NEXT: v_perm_b32 v4, v5, v4, s5 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] +; GFX942-NEXT: v_perm_b32 v3, v5, v3, s5 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB26_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7724,30 +7740,32 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_max_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_max_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -7768,30 +7786,32 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_max_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_max_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -7811,27 +7831,27 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX10-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB26_1 @@ -7851,26 +7871,26 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX90A-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB26_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7889,26 +7909,26 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX908-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB26_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7926,29 +7946,29 @@ define void @local_atomic_fmax_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX8-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB26_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8047,31 +8067,34 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX12-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_max_num_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v4, v4, v1 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v5, v5, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_max_num_f32_e32 v3, v3, v1 ; GFX12-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX12-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -8096,32 +8119,33 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX12-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_max_num_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v4, v4, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v5, v5, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX12-FAKE16-NEXT: v_max_num_f32_e32 v3, v3, v2 ; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX12-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 +; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo +; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xf1ff -; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX12-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -8144,27 +8168,27 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX942-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX942-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX942-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX942-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX942-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX942-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX942-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX942-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX942-NEXT: v_add3_u32 v6, v6, v4, s4 +; GFX942-NEXT: v_add3_u32 v6, v6, v3, s4 ; GFX942-NEXT: v_add3_u32 v8, v8, v5, s4 ; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v4, v4 +; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v3, v3 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX942-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[0:1] -; GFX942-NEXT: v_perm_b32 v4, v5, v4, s5 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX942-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] +; GFX942-NEXT: v_perm_b32 v3, v5, v3, s5 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB27_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8182,30 +8206,32 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_max_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_max_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_max_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -8226,30 +8252,32 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_max_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_max_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -8269,27 +8297,27 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX10-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB27_1 @@ -8309,26 +8337,26 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX90A-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB27_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8347,26 +8375,26 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX908-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB27_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8384,29 +8412,29 @@ define void @local_atomic_fmax_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX8-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_max_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_max_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_max_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB27_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end diff --git a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmin.ll b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmin.ll index 11ed43d..0aa8d33 100644 --- a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmin.ll +++ b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fmin.ll @@ -1598,29 +1598,29 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v4.l, v4.l, v4.l +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v3.l, v3.l, v3.l ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_min_num_f16_e32 v4.l, 4.0, v4.l -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-TRUE16-NEXT: v_min_num_f16_e32 v3.l, 4.0, v3.l +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -1641,29 +1641,29 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v4, v4, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v3, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_min_num_f16_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX12-FAKE16-NEXT: v_min_num_f16_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -1688,16 +1688,16 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX942-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX942-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX942-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX942-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB10_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1710,29 +1710,29 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: v_max_f16_e32 v4.l, v4.l, v4.l +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v3.l, v3.l, v3.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_min_f16_e32 v4.l, 4.0, v4.l -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-TRUE16-NEXT: v_min_f16_e32 v3.l, 4.0, v3.l +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1747,29 +1747,29 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_max_f16_e32 v4, v4, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v3, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX11-FAKE16-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1784,24 +1784,24 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX10-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX10-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX10-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX10-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB10_1 @@ -1823,16 +1823,16 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX90A-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX90A-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX90A-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX90A-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB10_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1853,16 +1853,16 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX908-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX908-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX908-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX908-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB10_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1884,17 +1884,17 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX8-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX8-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX8-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX8-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB10_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1915,18 +1915,18 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB10_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1947,18 +1947,18 @@ define void @local_atomic_fmin_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB10_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1989,20 +1989,21 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v4.l, v4.l, v4.l -; GFX12-TRUE16-NEXT: v_min_num_f16_e32 v4.l, 4.0, v4.l +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v3.l, v3.l, v3.l ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: v_min_num_f16_e32 v3.l, 4.0, v3.l +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2033,21 +2034,21 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v4, v4, v4 -; GFX12-FAKE16-NEXT: v_min_num_f16_e32 v4, 4.0, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v3, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_min_num_f16_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2073,16 +2074,16 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX942-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX942-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX942-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX942-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX942-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB11_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2105,20 +2106,21 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_max_f16_e32 v4.l, v4.l, v4.l -; GFX11-TRUE16-NEXT: v_min_f16_e32 v4.l, 4.0, v4.l +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v3.l, v3.l, v3.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_min_f16_e32 v3.l, 4.0, v3.l +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2143,21 +2145,21 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX11-FAKE16-NEXT: v_min_f16_e32 v4, 4.0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v3, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2180,17 +2182,17 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX10-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX10-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX10-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX10-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB11_1 @@ -2213,16 +2215,16 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX90A-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX90A-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX90A-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX90A-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX90A-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB11_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2244,16 +2246,16 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX908-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX908-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX908-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX908-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB11_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2276,17 +2278,17 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX8-NEXT: v_max_f16_e32 v4, v4, v4 -; GFX8-NEXT: v_min_f16_e32 v4, 4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX8-NEXT: v_max_f16_e32 v3, v3, v3 +; GFX8-NEXT: v_min_f16_e32 v3, 4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB11_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2308,18 +2310,18 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB11_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2341,18 +2343,18 @@ define void @local_atomic_fmin_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB11_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2685,17 +2687,18 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v2.l, v1.l, v1.l -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_min_num_f16_e32 v2.l, 4.0, v2.l -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-TRUE16-NEXT: v_max_num_f16_e32 v1.l, v2.l, v2.l +; GFX12-TRUE16-NEXT: v_min_num_f16_e32 v1.l, 4.0, v1.l +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2718,18 +2721,18 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v2, v1, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_min_num_f16_e32 v2, 4.0, v2 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_max_num_f16_e32 v1, v2, v2 +; GFX12-FAKE16-NEXT: v_min_num_f16_e32 v1, 4.0, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2749,14 +2752,14 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX942-NEXT: v_min_f16_e32 v2, 4.0, v2 -; GFX942-NEXT: v_and_or_b32 v2, v1, s2, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX942-NEXT: v_min_f16_e32 v1, 4.0, v1 +; GFX942-NEXT: v_and_or_b32 v1, v2, s2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB13_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2771,17 +2774,18 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_max_f16_e32 v2.l, v1.l, v1.l -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_min_f16_e32 v2.l, 4.0, v2.l -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-TRUE16-NEXT: v_max_f16_e32 v1.l, v2.l, v2.l +; GFX11-TRUE16-NEXT: v_min_f16_e32 v1.l, 4.0, v1.l +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2798,18 +2802,18 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_max_f16_e32 v2, v1, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_min_f16_e32 v2, 4.0, v2 -; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX11-FAKE16-NEXT: v_min_f16_e32 v1, 4.0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2826,16 +2830,16 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX10-NEXT: v_min_f16_e32 v2, 4.0, v2 -; GFX10-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX10-NEXT: v_min_f16_e32 v1, 4.0, v1 +; GFX10-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB13_1 @@ -2852,14 +2856,14 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX90A-NEXT: v_min_f16_e32 v2, 4.0, v2 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX90A-NEXT: v_min_f16_e32 v1, 4.0, v1 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB13_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2875,14 +2879,14 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX908-NEXT: v_min_f16_e32 v2, 4.0, v2 -; GFX908-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX908-NEXT: v_min_f16_e32 v1, 4.0, v1 +; GFX908-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB13_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2898,15 +2902,15 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_max_f16_e32 v2, v1, v1 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_min_f16_e32 v2, 4.0, v2 -; GFX8-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_max_f16_e32 v1, v2, v2 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_min_f16_e32 v1, 4.0, v1 +; GFX8-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB13_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2922,16 +2926,16 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX7-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB13_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2948,16 +2952,16 @@ define void @local_atomic_fmin_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB13_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3911,38 +3915,38 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v4, 4.0, v4 -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v3, 4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -3963,37 +3967,37 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4019,22 +4023,22 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB16_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4047,38 +4051,38 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4093,37 +4097,37 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .p2align 6 ; GFX11-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4138,28 +4142,28 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB16_1 @@ -4182,20 +4186,20 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB16_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4217,20 +4221,20 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB16_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4252,22 +4256,22 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB16_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4288,19 +4292,19 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX7-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX7-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB16_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4321,19 +4325,19 @@ define void @local_atomic_fmin_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX6-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX6-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB16_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4364,29 +4368,30 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v4, 4.0, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v3, 4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -4417,28 +4422,29 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v4, 4.0, v4 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v3, 4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4465,22 +4471,22 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX942-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB17_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4504,28 +4510,29 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-TRUE16-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4551,27 +4558,28 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4594,21 +4602,21 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX10-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB17_1 @@ -4632,20 +4640,20 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX90A-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB17_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4668,20 +4676,20 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX908-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB17_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4704,22 +4712,22 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX8-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB17_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4741,19 +4749,19 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX7-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX7-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX7-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB17_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4775,19 +4783,19 @@ define void @local_atomic_fmin_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX6-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_mul_f32_e32 v4, 1.0, v4 -; GFX6-NEXT: v_min_f32_e32 v4, 4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_mul_f32_e32 v3, 1.0, v3 +; GFX6-NEXT: v_min_f32_e32 v3, 4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB17_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5189,26 +5197,27 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v2, 4.0, v2 -; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v1, 4.0, v1 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -5231,25 +5240,26 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v2, 4.0, v2 -; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v1, 4.0, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -5270,21 +5280,21 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX942-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX942-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX942-NEXT: v_add3_u32 v3, v3, v2, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX942-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX942-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX942-NEXT: v_add3_u32 v3, v3, v1, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX942-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX942-NEXT: v_and_or_b32 v2, v1, s3, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX942-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX942-NEXT: v_and_or_b32 v1, v2, s3, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB19_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5300,25 +5310,26 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5336,24 +5347,25 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-FAKE16-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5370,21 +5382,21 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX10-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX10-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX10-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo -; GFX10-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX10-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX10-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX10-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX10-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB19_1 @@ -5402,20 +5414,20 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX90A-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX90A-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX90A-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX90A-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX90A-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX90A-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX90A-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX90A-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX90A-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX90A-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB19_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5432,20 +5444,20 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX908-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX908-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX908-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX908-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX908-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX908-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX908-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX908-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX908-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX908-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX908-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX908-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB19_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5461,21 +5473,21 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX8-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX8-NEXT: v_bfe_u32 v4, v2, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX8-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX8-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v1 ; GFX8-NEXT: v_add_u32_e32 v4, vcc, 0x7fff, v4 -; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v2 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc -; GFX8-NEXT: v_or_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v1 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v5, vcc +; GFX8-NEXT: v_or_b32_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB19_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5491,17 +5503,17 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX7-NEXT: v_mul_f32_e32 v2, 1.0, v2 -; GFX7-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX7-NEXT: v_mul_f32_e32 v1, 1.0, v1 +; GFX7-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB19_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5518,17 +5530,17 @@ define void @local_atomic_fmin_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX6-NEXT: v_mul_f32_e32 v2, 1.0, v2 -; GFX6-NEXT: v_min_f32_e32 v2, 4.0, v2 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX6-NEXT: v_mul_f32_e32 v1, 1.0, v1 +; GFX6-NEXT: v_min_f32_e32 v1, 4.0, v1 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB19_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6101,15 +6113,15 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX12-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_pk_max_num_f16 v3, v2, v2 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_pk_min_num_f16 v3, v3, v1 +; GFX12-NEXT: v_mov_b32_e32 v3, v2 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_pk_max_num_f16 v2, v3, v3 +; GFX12-NEXT: v_pk_min_num_f16 v2, v2, v1 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX12-NEXT: v_mov_b32_e32 v2, v3 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -6129,14 +6141,14 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX942-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_pk_max_f16 v3, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v3, v2 +; GFX942-NEXT: v_pk_max_f16 v2, v3, v3 ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_pk_min_f16 v3, v3, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX942-NEXT: v_pk_min_f16 v2, v2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v2, v3 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB22_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6152,15 +6164,15 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX11-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_pk_min_f16 v3, v3, v1 +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX11-NEXT: v_pk_min_f16 v2, v2, v1 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6178,14 +6190,14 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX10-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX10-NEXT: v_pk_min_f16 v3, v3, v1 +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX10-NEXT: v_pk_min_f16 v2, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB22_1 @@ -6202,13 +6214,13 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX90A-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX90A-NEXT: v_pk_min_f16 v3, v3, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX90A-NEXT: v_pk_min_f16 v2, v2, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB22_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6224,13 +6236,13 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX908-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX908-NEXT: v_pk_min_f16 v3, v3, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX908-NEXT: v_pk_min_f16 v2, v2, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB22_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6248,16 +6260,16 @@ define void @local_atomic_fmin_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX8-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_max_f16_sdwa v4, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_max_f16_e32 v5, v3, v3 -; GFX8-NEXT: v_min_f16_sdwa v4, v4, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_max_f16_sdwa v3, v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_max_f16_e32 v5, v4, v4 +; GFX8-NEXT: v_min_f16_sdwa v3, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX8-NEXT: v_min_f16_e32 v5, v5, v1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB22_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6363,15 +6375,15 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX12-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_pk_max_num_f16 v3, v2, v2 -; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-NEXT: v_pk_min_num_f16 v3, v3, v1 +; GFX12-NEXT: v_mov_b32_e32 v3, v2 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_pk_max_num_f16 v2, v3, v3 +; GFX12-NEXT: v_pk_min_num_f16 v2, v2, v1 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX12-NEXT: v_mov_b32_e32 v2, v3 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -6391,14 +6403,14 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX942-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_pk_max_f16 v3, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v3, v2 +; GFX942-NEXT: v_pk_max_f16 v2, v3, v3 ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_pk_min_f16 v3, v3, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX942-NEXT: v_pk_min_f16 v2, v2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v2, v3 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB23_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6414,15 +6426,15 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX11-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-NEXT: v_pk_min_f16 v3, v3, v1 +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX11-NEXT: v_pk_min_f16 v2, v2, v1 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6440,14 +6452,14 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX10-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX10-NEXT: v_pk_min_f16 v3, v3, v1 +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX10-NEXT: v_pk_min_f16 v2, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB23_1 @@ -6464,13 +6476,13 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX90A-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX90A-NEXT: v_pk_min_f16 v3, v3, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX90A-NEXT: v_pk_min_f16 v2, v2, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB23_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6486,13 +6498,13 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX908-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_max_f16 v3, v2, v2 -; GFX908-NEXT: v_pk_min_f16 v3, v3, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_max_f16 v2, v3, v3 +; GFX908-NEXT: v_pk_min_f16 v2, v2, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB23_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6510,16 +6522,16 @@ define void @local_atomic_fmin_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX8-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_max_f16_sdwa v4, v3, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_max_f16_e32 v5, v3, v3 -; GFX8-NEXT: v_min_f16_sdwa v4, v4, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_max_f16_sdwa v3, v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_max_f16_e32 v5, v4, v4 +; GFX8-NEXT: v_min_f16_sdwa v3, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX8-NEXT: v_min_f16_e32 v5, v5, v1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB23_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7589,31 +7601,34 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX12-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_min_num_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v4, v4, v1 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v5, v5, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v3, v3, v1 ; GFX12-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX12-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -7638,32 +7653,33 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX12-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_min_num_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v4, v4, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v5, v5, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX12-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 +; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo +; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xf1ff -; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX12-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -7686,27 +7702,27 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX942-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX942-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX942-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX942-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX942-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX942-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX942-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX942-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX942-NEXT: v_add3_u32 v6, v6, v4, s4 +; GFX942-NEXT: v_add3_u32 v6, v6, v3, s4 ; GFX942-NEXT: v_add3_u32 v8, v8, v5, s4 ; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v4, v4 +; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v3, v3 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX942-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[0:1] -; GFX942-NEXT: v_perm_b32 v4, v5, v4, s5 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] +; GFX942-NEXT: v_perm_b32 v3, v5, v3, s5 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB26_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7724,30 +7740,32 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_min_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_min_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -7768,30 +7786,32 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_min_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_min_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -7811,27 +7831,27 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX10-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB26_1 @@ -7851,26 +7871,26 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX90A-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB26_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7889,26 +7909,26 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX908-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB26_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7926,29 +7946,29 @@ define void @local_atomic_fmin_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX8-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB26_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8047,31 +8067,34 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX12-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_min_num_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v4, v4, v1 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v5, v5, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_min_num_f32_e32 v3, v3, v1 ; GFX12-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX12-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -8096,32 +8119,33 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX12-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_min_num_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v4, v4, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v5, v5, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX12-FAKE16-NEXT: v_min_num_f32_e32 v3, v3, v2 ; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX12-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 +; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo +; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xf1ff -; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX12-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -8144,27 +8168,27 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX942-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX942-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX942-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX942-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX942-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX942-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX942-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX942-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX942-NEXT: v_add3_u32 v6, v6, v4, s4 +; GFX942-NEXT: v_add3_u32 v6, v6, v3, s4 ; GFX942-NEXT: v_add3_u32 v8, v8, v5, s4 ; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v4, v4 +; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v3, v3 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX942-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[0:1] -; GFX942-NEXT: v_perm_b32 v4, v5, v4, s5 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX942-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] +; GFX942-NEXT: v_perm_b32 v3, v5, v3, s5 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB27_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8182,30 +8206,32 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_min_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_min_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_min_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -8226,30 +8252,32 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_min_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_min_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -8269,27 +8297,27 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX10-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB27_1 @@ -8309,26 +8337,26 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX90A-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB27_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8347,26 +8375,26 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX908-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB27_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8384,29 +8412,29 @@ define void @local_atomic_fmin_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX8-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_min_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_min_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_min_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB27_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end diff --git a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fsub.ll b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fsub.ll index d74338c..929bb61 100644 --- a/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fsub.ll +++ b/llvm/test/CodeGen/AMDGPU/local-atomicrmw-fsub.ll @@ -453,13 +453,14 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX12-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX12-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -478,12 +479,12 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB2_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -498,13 +499,14 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX11-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX11-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -521,13 +523,13 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB2_1 @@ -543,12 +545,12 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB2_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -563,12 +565,12 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB2_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -584,12 +586,12 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB2_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -605,12 +607,12 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB2_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -626,12 +628,12 @@ define void @local_atomic_fsub_noret_f32(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB2_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB2_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -654,13 +656,14 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX12-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65532 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65532 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -679,12 +682,12 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX942-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB3_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -699,13 +702,14 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX11-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65532 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65532 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -722,13 +726,13 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB3_1 @@ -744,12 +748,12 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX90A-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB3_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -764,12 +768,12 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB3_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -785,12 +789,12 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB3_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -806,12 +810,12 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65532 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65532 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB3_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -828,12 +832,12 @@ define void @local_atomic_fsub_noret_f32__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB3_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB3_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1296,13 +1300,14 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX12-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f64_e32 v[3:4], -4.0, v[1:2] +; GFX12-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f64_e32 v[1:2], -4.0, v[3:4] ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] +; GFX12-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX12-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -1321,12 +1326,12 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f64 v[4:5], v[2:3], -4.0 -; GFX942-NEXT: ds_cmpst_rtn_b64 v[4:5], v0, v[2:3], v[4:5] +; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[2:3] +; GFX942-NEXT: v_add_f64 v[2:3], v[4:5], -4.0 +; GFX942-NEXT: ds_cmpst_rtn_b64 v[2:3], v0, v[4:5], v[2:3] ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[2:3] +; GFX942-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b64_e32 v[2:3], v[4:5] ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB6_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1341,13 +1346,14 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX11-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 +; GFX11-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] +; GFX11-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX11-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1364,14 +1370,14 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 +; GFX10-NEXT: v_mov_b32_e32 v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v3, v1 +; GFX10-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX10-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX10-NEXT: v_mov_b32_e32 v1, v3 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB6_1 @@ -1387,12 +1393,12 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f64 v[4:5], v[2:3], -4.0 -; GFX90A-NEXT: ds_cmpst_rtn_b64 v[4:5], v0, v[2:3], v[4:5] +; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] +; GFX90A-NEXT: v_add_f64 v[2:3], v[4:5], -4.0 +; GFX90A-NEXT: ds_cmpst_rtn_b64 v[2:3], v0, v[4:5], v[2:3] ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[2:3] +; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB6_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1407,13 +1413,13 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX908-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX908-NEXT: v_mov_b32_e32 v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v3, v1 +; GFX908-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX908-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX908-NEXT: v_mov_b32_e32 v1, v3 +; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB6_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1429,13 +1435,13 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX8-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX8-NEXT: v_mov_b32_e32 v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v3, v1 +; GFX8-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX8-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX8-NEXT: v_mov_b32_e32 v1, v3 +; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB6_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1451,13 +1457,13 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX7-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX7-NEXT: v_mov_b32_e32 v4, v2 +; GFX7-NEXT: v_mov_b32_e32 v3, v1 +; GFX7-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX7-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX7-NEXT: v_mov_b32_e32 v1, v3 +; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v2, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB6_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1473,13 +1479,13 @@ define void @local_atomic_fsub_noret_f64(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB6_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX6-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] +; GFX6-NEXT: v_mov_b32_e32 v4, v2 +; GFX6-NEXT: v_mov_b32_e32 v3, v1 +; GFX6-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX6-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX6-NEXT: v_mov_b32_e32 v1, v3 +; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v2, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB6_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1502,13 +1508,14 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f64_e32 v[3:4], -4.0, v[1:2] +; GFX12-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f64_e32 v[1:2], -4.0, v[3:4] ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] offset:65528 +; GFX12-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] offset:65528 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX12-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -1527,12 +1534,12 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX942-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f64 v[4:5], v[2:3], -4.0 -; GFX942-NEXT: ds_cmpst_rtn_b64 v[4:5], v0, v[2:3], v[4:5] offset:65528 +; GFX942-NEXT: v_mov_b64_e32 v[4:5], v[2:3] +; GFX942-NEXT: v_add_f64 v[2:3], v[4:5], -4.0 +; GFX942-NEXT: ds_cmpst_rtn_b64 v[2:3], v0, v[4:5], v[2:3] offset:65528 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[2:3] +; GFX942-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b64_e32 v[2:3], v[4:5] ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB7_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1547,13 +1554,14 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 +; GFX11-NEXT: v_dual_mov_b32 v4, v2 :: v_dual_mov_b32 v3, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b64 v[3:4], v0, v[3:4], v[1:2] offset:65528 +; GFX11-NEXT: ds_cmpstore_rtn_b64 v[1:2], v0, v[1:2], v[3:4] offset:65528 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX11-NEXT: v_dual_mov_b32 v1, v3 :: v_dual_mov_b32 v2, v4 +; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -1570,14 +1578,14 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 +; GFX10-NEXT: v_mov_b32_e32 v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v3, v1 +; GFX10-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX10-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[3:4], v[1:2] -; GFX10-NEXT: v_mov_b32_e32 v1, v3 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, v[1:2], v[3:4] ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB7_1 @@ -1593,12 +1601,12 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX90A-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f64 v[4:5], v[2:3], -4.0 -; GFX90A-NEXT: ds_cmpst_rtn_b64 v[4:5], v0, v[2:3], v[4:5] offset:65528 +; GFX90A-NEXT: v_pk_mov_b32 v[4:5], v[2:3], v[2:3] op_sel:[0,1] +; GFX90A-NEXT: v_add_f64 v[2:3], v[4:5], -4.0 +; GFX90A-NEXT: ds_cmpst_rtn_b64 v[2:3], v0, v[4:5], v[2:3] offset:65528 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[4:5], v[2:3] +; GFX90A-NEXT: v_cmp_eq_u64_e32 vcc, v[2:3], v[4:5] ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_pk_mov_b32 v[2:3], v[4:5], v[4:5] op_sel:[0,1] ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB7_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1613,13 +1621,13 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX908-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX908-NEXT: v_mov_b32_e32 v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v3, v1 +; GFX908-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX908-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX908-NEXT: v_mov_b32_e32 v1, v3 +; GFX908-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB7_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1635,13 +1643,13 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX8-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX8-NEXT: v_mov_b32_e32 v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v3, v1 +; GFX8-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX8-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX8-NEXT: v_mov_b32_e32 v1, v3 +; GFX8-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB7_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1657,13 +1665,13 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f64 v[3:4], v[1:2], -4.0 -; GFX7-NEXT: ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4] offset:65528 +; GFX7-NEXT: v_mov_b32_e32 v4, v2 +; GFX7-NEXT: v_mov_b32_e32 v3, v1 +; GFX7-NEXT: v_add_f64 v[1:2], v[3:4], -4.0 +; GFX7-NEXT: ds_cmpst_rtn_b64 v[1:2], v0, v[3:4], v[1:2] offset:65528 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2] -; GFX7-NEXT: v_mov_b32_e32 v1, v3 +; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, v[1:2], v[3:4] ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v2, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB7_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -1680,13 +1688,13 @@ define void @local_atomic_fsub_noret_f64__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB7_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f64 v[3:4], v[0:1], -4.0 -; GFX6-NEXT: ds_cmpst_rtn_b64 v[3:4], v2, v[0:1], v[3:4] +; GFX6-NEXT: v_mov_b32_e32 v4, v1 +; GFX6-NEXT: v_mov_b32_e32 v3, v0 +; GFX6-NEXT: v_add_f64 v[0:1], v[3:4], -4.0 +; GFX6-NEXT: ds_cmpst_rtn_b64 v[0:1], v2, v[3:4], v[0:1] ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[3:4], v[0:1] -; GFX6-NEXT: v_mov_b32_e32 v0, v3 +; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, v[0:1], v[3:4] ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB7_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2494,27 +2502,27 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: v_add_f16_e32 v4.l, -4.0, v4.l +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_add_f16_e32 v3.l, -4.0, v3.l ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2535,28 +2543,28 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_add_f16_e32 v4, -4.0, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_add_f16_e32 v3, -4.0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2581,15 +2589,15 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX942-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX942-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB10_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2602,27 +2610,27 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: v_add_f16_e32 v4.l, -4.0, v4.l +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_add_f16_e32 v3.l, -4.0, v3.l ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2637,28 +2645,28 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_add_f16_e32 v4, -4.0, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_add_f16_e32 v3, -4.0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -2673,23 +2681,23 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX10-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX10-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB10_1 @@ -2711,15 +2719,15 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX90A-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX90A-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB10_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2740,15 +2748,15 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX908-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX908-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB10_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2770,16 +2778,16 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX8-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX8-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB10_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2800,18 +2808,18 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB10_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2832,18 +2840,18 @@ define void @local_atomic_fsub_noret_f16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB10_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB10_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2874,19 +2882,19 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add_f16_e32 v4.l, -4.0, v4.l -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX12-TRUE16-NEXT: v_add_f16_e32 v3.l, -4.0, v3.l +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -2917,19 +2925,20 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX12-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_add_f16_e32 v3, -4.0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -2955,15 +2964,15 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX942-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX942-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX942-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX942-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB11_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -2986,19 +2995,19 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-TRUE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add_f16_e32 v4.l, -4.0, v4.l -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 +; GFX11-TRUE16-NEXT: v_add_f16_e32 v3.l, -4.0, v3.l +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -3023,19 +3032,20 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX11-FAKE16-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_add_f16_e32 v3, -4.0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -3058,16 +3068,16 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX10-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX10-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX10-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB11_1 @@ -3090,15 +3100,15 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX90A-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX90A-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX90A-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB11_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3120,15 +3130,15 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX908-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX908-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX908-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX908-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB11_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3151,16 +3161,16 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX8-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX8-NEXT: v_add_f16_e32 v4, -4.0, v4 -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX8-NEXT: v_add_f16_e32 v3, -4.0, v3 +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB11_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3182,18 +3192,18 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX7-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX7-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX7-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB11_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3215,18 +3225,18 @@ define void @local_atomic_fsub_noret_f16__offset(ptr addrspace(3) %ptr) nounwind ; GFX6-NEXT: .LBB11_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX6-NEXT: v_cvt_f16_f32_e32 v4, v4 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_cvt_f32_f16_e32 v3, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX6-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB11_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3548,16 +3558,16 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_add_f16_e32 v2.l, -4.0, v1.l -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add_f16_e32 v1.l, -4.0, v2.l +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -3580,16 +3590,17 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_add_f16_e32 v2, -4.0, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -3609,13 +3620,13 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f16_e32 v2, -4.0, v1 -; GFX942-NEXT: v_and_or_b32 v2, v1, s2, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX942-NEXT: v_and_or_b32 v1, v2, s2, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB13_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3630,16 +3641,16 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_add_f16_e32 v2.l, -4.0, v1.l -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v2.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v1.h, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add_f16_e32 v1.l, -4.0, v2.l +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -3656,16 +3667,17 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_add_f16_e32 v2, -4.0, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -3682,15 +3694,15 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f16_e32 v2, -4.0, v1 -; GFX10-NEXT: v_and_b32_e32 v2, 0xffff, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX10-NEXT: v_and_b32_e32 v1, 0xffff, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB13_1 @@ -3707,13 +3719,13 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f16_e32 v2, -4.0, v1 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB13_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3729,13 +3741,13 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f16_e32 v2, -4.0, v1 -; GFX908-NEXT: v_and_or_b32 v2, v1, s6, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX908-NEXT: v_and_or_b32 v1, v2, s6, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB13_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3751,14 +3763,14 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f16_e32 v2, -4.0, v1 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_add_f16_e32 v1, -4.0, v2 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB13_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3774,16 +3786,16 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX7-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX7-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB13_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -3800,16 +3812,16 @@ define void @local_atomic_fsub_noret_f16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB13_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cvt_f32_f16_e32 v2, v1 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX6-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v2 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB13_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4759,38 +4771,38 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX12-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -4811,37 +4823,37 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX12-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX12-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX12-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX12-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX12-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX12-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX12-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX12-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -4867,22 +4879,22 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX942-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB16_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -4895,38 +4907,38 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-TRUE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-TRUE16-NEXT: s_mov_b32 s0, 0 -; GFX11-TRUE16-NEXT: ds_load_b32 v2, v1 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-TRUE16-NEXT: ds_load_b32 v3, v1 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-TRUE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-TRUE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-TRUE16-NEXT: .p2align 6 ; GFX11-TRUE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v0, v5 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v0, v5 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4941,37 +4953,37 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX11-FAKE16-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX11-FAKE16-NEXT: s_mov_b32 s0, 0 -; GFX11-FAKE16-NEXT: ds_load_b32 v2, v1 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX11-FAKE16-NEXT: ds_load_b32 v3, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 24, v0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-FAKE16-NEXT: v_not_b32_e32 v3, v3 +; GFX11-FAKE16-NEXT: v_not_b32_e32 v2, v2 ; GFX11-FAKE16-NEXT: .p2align 6 ; GFX11-FAKE16-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v0, v2 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v0, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v0, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v1, v4, v2 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v1, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -4986,28 +4998,28 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX10-NEXT: v_and_b32_e32 v1, -4, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 3, v0 ; GFX10-NEXT: s_mov_b32 s4, 0 -; GFX10-NEXT: ds_read_b32 v2, v1 -; GFX10-NEXT: v_lshlrev_b32_e64 v3, v0, 0xffff +; GFX10-NEXT: ds_read_b32 v3, v1 +; GFX10-NEXT: v_lshlrev_b32_e64 v2, v0, 0xffff ; GFX10-NEXT: v_and_b32_e32 v0, 24, v0 -; GFX10-NEXT: v_not_b32_e32 v3, v3 +; GFX10-NEXT: v_not_b32_e32 v2, v2 ; GFX10-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v2, v3, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v1, v2, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB16_1 @@ -5030,20 +5042,20 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX90A-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB16_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5065,20 +5077,20 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX908-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB16_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5100,22 +5112,22 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX8-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB16_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5136,18 +5148,18 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX7-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB16_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5168,18 +5180,18 @@ define void @local_atomic_fsub_noret_bf16(ptr addrspace(3) %ptr) nounwind { ; GFX6-NEXT: .LBB16_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v0, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v0, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v1, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v0, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v0, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v1, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB16_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5210,29 +5222,30 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-TRUE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX12-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX12-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -5263,28 +5276,29 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX12-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX12-FAKE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX12-FAKE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX12-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX12-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -5311,22 +5325,22 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX942-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX942-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX942-NEXT: v_add3_u32 v5, v5, v4, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 +; GFX942-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX942-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX942-NEXT: v_add3_u32 v5, v5, v3, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX942-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX942-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX942-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX942-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB17_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5350,28 +5364,29 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-TRUE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-TRUE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v1, v5 -; GFX11-TRUE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX11-TRUE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.h, 0 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, v1, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5397,27 +5412,28 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX11-FAKE16-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v1, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX11-FAKE16-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) -; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v4, 0x7fff +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v4 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX11-FAKE16-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX11-FAKE16-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -5440,21 +5456,21 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX10-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX10-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX10-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 -; GFX10-NEXT: v_add3_u32 v5, v5, v4, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc_lo -; GFX10-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX10-NEXT: v_and_or_b32 v4, v3, v2, v4 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX10-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX10-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX10-NEXT: v_add3_u32 v5, v5, v3, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc_lo +; GFX10-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX10-NEXT: v_and_or_b32 v3, v4, v2, v3 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB17_1 @@ -5478,20 +5494,20 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX90A-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX90A-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX90A-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX90A-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX90A-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX90A-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX90A-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX90A-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX90A-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX90A-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX90A-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB17_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5514,20 +5530,20 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX908-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX908-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX908-NEXT: v_bfe_u32 v5, v4, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v4 -; GFX908-NEXT: v_add3_u32 v5, v5, v4, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX908-NEXT: v_cndmask_b32_e32 v4, v5, v6, vcc -; GFX908-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX908-NEXT: v_and_or_b32 v4, v3, v2, v4 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX908-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX908-NEXT: v_bfe_u32 v5, v3, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v6, 0x400000, v3 +; GFX908-NEXT: v_add3_u32 v5, v5, v3, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX908-NEXT: v_cndmask_b32_e32 v3, v5, v6, vcc +; GFX908-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX908-NEXT: v_and_or_b32 v3, v4, v2, v3 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB17_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5550,22 +5566,22 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX8-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshrrev_b32_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; GFX8-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshrrev_b32_sdwa v3, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; GFX8-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v4, v4 -; GFX8-NEXT: v_cndmask_b32_e32 v4, v6, v7, vcc -; GFX8-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX8-NEXT: v_lshlrev_b32_sdwa v4, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v3, v3 +; GFX8-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc +; GFX8-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX8-NEXT: v_lshlrev_b32_sdwa v3, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB17_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5587,18 +5603,18 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX7-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX7-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX7-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX7-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX7-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX7-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX7-NEXT: v_mov_b32_e32 v4, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX7-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX7-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX7-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX7-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX7-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v3, v4 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB17_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -5620,18 +5636,18 @@ define void @local_atomic_fsub_noret_bf16__offset(ptr addrspace(3) %ptr) nounwin ; GFX6-NEXT: .LBB17_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshrrev_b32_e32 v4, v1, v3 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_add_f32_e32 v4, -4.0, v4 -; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v4 -; GFX6-NEXT: v_and_b32_e32 v5, v3, v2 -; GFX6-NEXT: v_lshlrev_b32_e32 v4, v1, v4 -; GFX6-NEXT: v_or_b32_e32 v4, v5, v4 -; GFX6-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX6-NEXT: v_mov_b32_e32 v4, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, v1, v4 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_add_f32_e32 v3, -4.0, v3 +; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v3 +; GFX6-NEXT: v_and_b32_e32 v5, v4, v2 +; GFX6-NEXT: v_lshlrev_b32_e32 v3, v1, v3 +; GFX6-NEXT: v_or_b32_e32 v3, v5, v3 +; GFX6-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v3, v4 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB17_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6031,26 +6047,27 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-TRUE16-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX12-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -6073,25 +6090,26 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX12-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX12-FAKE16-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX12-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX12-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX12-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX12-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -6112,21 +6130,21 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX942-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX942-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX942-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX942-NEXT: v_add3_u32 v3, v3, v2, s2 -; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX942-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX942-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX942-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX942-NEXT: v_add3_u32 v3, v3, v1, s2 +; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 ; GFX942-NEXT: s_nop 1 -; GFX942-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX942-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX942-NEXT: v_and_or_b32 v2, v1, s3, v2 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX942-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX942-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX942-NEXT: v_and_or_b32 v1, v2, s3, v1 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB19_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6142,25 +6160,26 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-TRUE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-TRUE16-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-TRUE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo ; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.h, 0 -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v2.h -; GFX11-TRUE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v3 +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v3.l, v1.h +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v3 ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6178,24 +6197,25 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX11-FAKE16-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v1 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v2, v1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX11-FAKE16-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX11-FAKE16-NEXT: v_add3_u32 v3, v3, v1, 0x7fff ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX11-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 offset:65534 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 offset:65534 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6212,21 +6232,21 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX10-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX10-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX10-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v2, v2 -; GFX10-NEXT: v_add3_u32 v3, v3, v2, 0x7fff -; GFX10-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc_lo -; GFX10-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX10-NEXT: v_and_or_b32 v2, 0xffff0000, v1, v2 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX10-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX10-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX10-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v1, v1 +; GFX10-NEXT: v_add3_u32 v3, v3, v1, 0x7fff +; GFX10-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc_lo +; GFX10-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX10-NEXT: v_and_or_b32 v1, 0xffff0000, v2, v1 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB19_1 @@ -6244,20 +6264,20 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX90A-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX90A-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX90A-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX90A-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX90A-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX90A-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX90A-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX90A-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX90A-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX90A-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX90A-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX90A-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX90A-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX90A-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB19_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6274,20 +6294,20 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX908-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX908-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX908-NEXT: v_bfe_u32 v3, v2, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v2 -; GFX908-NEXT: v_add3_u32 v3, v3, v2, s6 -; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX908-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GFX908-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX908-NEXT: v_and_or_b32 v2, v1, s7, v2 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX908-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX908-NEXT: v_bfe_u32 v3, v1, 16, 1 +; GFX908-NEXT: v_or_b32_e32 v4, 0x400000, v1 +; GFX908-NEXT: v_add3_u32 v3, v3, v1, s6 +; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX908-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc +; GFX908-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX908-NEXT: v_and_or_b32 v1, v2, s7, v1 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB19_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6303,21 +6323,21 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX8-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX8-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX8-NEXT: v_bfe_u32 v4, v2, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX8-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX8-NEXT: v_bfe_u32 v4, v1, 16, 1 +; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v1 ; GFX8-NEXT: v_add_u32_e32 v4, vcc, 0x7fff, v4 -; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v2 -; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v2, v2 -; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX8-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc -; GFX8-NEXT: v_or_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX8-NEXT: v_or_b32_e32 v5, 0x400000, v1 +; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v1, v1 +; GFX8-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX8-NEXT: v_cndmask_b32_e32 v1, v4, v5, vcc +; GFX8-NEXT: v_or_b32_sdwa v1, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB19_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6333,16 +6353,16 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX7-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX7-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX7-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX7-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 offset:65534 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX7-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX7-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX7-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX7-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 offset:65534 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB19_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6359,16 +6379,16 @@ define void @local_atomic_fsub_noret_bf16__offset__align4(ptr addrspace(3) %ptr) ; GFX6-NEXT: .LBB19_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_lshlrev_b32_e32 v2, 16, v1 -; GFX6-NEXT: v_add_f32_e32 v2, -4.0, v2 -; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v1 -; GFX6-NEXT: v_lshrrev_b32_e32 v2, 16, v2 -; GFX6-NEXT: v_or_b32_e32 v2, v3, v2 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v2 +; GFX6-NEXT: v_add_f32_e32 v1, -4.0, v1 +; GFX6-NEXT: v_and_b32_e32 v3, 0xffff0000, v2 +; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v1 +; GFX6-NEXT: v_or_b32_e32 v1, v3, v1 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB19_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6906,13 +6926,14 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX12-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX12-NEXT: v_mov_b32_e32 v3, v2 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX12-NEXT: v_mov_b32_e32 v2, v3 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -6931,12 +6952,12 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX942-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] -; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX942-NEXT: v_mov_b32_e32 v3, v2 +; GFX942-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v2, v3 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB22_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -6951,13 +6972,14 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX11-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -6974,13 +6996,13 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX10-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB22_1 @@ -6996,12 +7018,12 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX90A-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB22_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7016,12 +7038,12 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX908-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB22_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7037,14 +7059,14 @@ define void @local_atomic_fsub_noret_v2f16(ptr addrspace(3) %ptr, <2 x half> %va ; GFX8-NEXT: .LBB22_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_sub_f16_sdwa v3, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_sub_f16_e32 v4, v2, v1 -; GFX8-NEXT: v_or_b32_e32 v3, v4, v3 -; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 +; GFX8-NEXT: v_mov_b32_e32 v3, v2 +; GFX8-NEXT: v_sub_f16_sdwa v2, v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_sub_f16_e32 v4, v3, v1 +; GFX8-NEXT: v_or_b32_e32 v2, v4, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v3 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB22_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7149,13 +7171,14 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX12-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX12-NEXT: v_mov_b32_e32 v3, v2 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX12-NEXT: v_mov_b32_e32 v2, v3 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -7174,12 +7197,12 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX942-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] -; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX942-NEXT: v_mov_b32_e32 v3, v2 +; GFX942-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v2, v3 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB23_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7194,13 +7217,14 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX11-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX11-NEXT: v_mov_b32_e32 v3, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v2 offset:65532 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v3 offset:65532 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX11-NEXT: v_mov_b32_e32 v2, v3 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -7217,13 +7241,13 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX10-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX10-NEXT: v_mov_b32_e32 v3, v2 +; GFX10-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v2 -; GFX10-NEXT: v_mov_b32_e32 v2, v3 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB23_1 @@ -7239,12 +7263,12 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX90A-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] -; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX90A-NEXT: v_mov_b32_e32 v3, v2 +; GFX90A-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v2, v3 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB23_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7259,12 +7283,12 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX908-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_pk_add_f16 v3, v2, v1 neg_lo:[0,1] neg_hi:[0,1] -; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX908-NEXT: v_mov_b32_e32 v3, v2 +; GFX908-NEXT: v_pk_add_f16 v2, v3, v1 neg_lo:[0,1] neg_hi:[0,1] +; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v2, v3 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB23_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -7280,14 +7304,14 @@ define void @local_atomic_fsub_noret_v2f16__offset(ptr addrspace(3) %ptr, <2 x h ; GFX8-NEXT: .LBB23_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_sub_f16_sdwa v3, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; GFX8-NEXT: v_sub_f16_e32 v4, v2, v1 -; GFX8-NEXT: v_or_b32_e32 v3, v4, v3 -; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v2, v3 offset:65532 +; GFX8-NEXT: v_mov_b32_e32 v3, v2 +; GFX8-NEXT: v_sub_f16_sdwa v2, v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; GFX8-NEXT: v_sub_f16_e32 v4, v3, v1 +; GFX8-NEXT: v_or_b32_e32 v2, v4, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v3, v2 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v2 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v3 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v2, v3 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB23_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8357,31 +8381,34 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX12-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_sub_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-TRUE16-NEXT: v_sub_f32_e32 v4, v4, v1 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-TRUE16-NEXT: v_sub_f32_e32 v5, v5, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_sub_f32_e32 v3, v3, v1 ; GFX12-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX12-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -8406,32 +8433,33 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX12-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_sub_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-FAKE16-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-FAKE16-NEXT: v_sub_f32_e32 v5, v5, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX12-FAKE16-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX12-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 +; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo +; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xf1ff -; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX12-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -8454,27 +8482,27 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX942-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX942-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX942-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX942-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX942-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX942-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX942-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX942-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX942-NEXT: v_add3_u32 v6, v6, v4, s4 +; GFX942-NEXT: v_add3_u32 v6, v6, v3, s4 ; GFX942-NEXT: v_add3_u32 v8, v8, v5, s4 ; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v4, v4 +; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v3, v3 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX942-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[0:1] -; GFX942-NEXT: v_perm_b32 v4, v5, v4, s5 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX942-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] +; GFX942-NEXT: v_perm_b32 v3, v5, v3, s5 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB26_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8492,30 +8520,32 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-TRUE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_sub_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_sub_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_sub_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_sub_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -8536,30 +8566,32 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX11-FAKE16-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_sub_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_sub_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -8579,27 +8611,27 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX10-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB26_1 @@ -8619,26 +8651,26 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX90A-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB26_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8657,26 +8689,26 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX908-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB26_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8694,29 +8726,29 @@ define void @local_atomic_fsub_noret_v2bf16(ptr addrspace(3) %ptr, <2 x bfloat> ; GFX8-NEXT: .LBB26_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB26_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8815,31 +8847,34 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX12-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX12-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 -; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-TRUE16-NEXT: v_dual_sub_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-TRUE16-NEXT: v_sub_f32_e32 v4, v4, v1 +; GFX12-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-TRUE16-NEXT: v_sub_f32_e32 v5, v5, v2 +; GFX12-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX12-TRUE16-NEXT: v_sub_f32_e32 v3, v3, v1 ; GFX12-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX12-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX12-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1) +; GFX12-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd -; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX12-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffd ; GFX12-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX12-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 -; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX12-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX12-TRUE16-NEXT: s_wait_dscnt 0x0 ; GFX12-TRUE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe ; GFX12-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-TRUE16-NEXT: s_wait_alu 0xfffe @@ -8864,32 +8899,33 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX12-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX12-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 -; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX12-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_dual_sub_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX12-FAKE16-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX12-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX12-FAKE16-NEXT: v_sub_f32_e32 v5, v5, v1 +; GFX12-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX12-FAKE16-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX12-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX12-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX12-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 +; GFX12-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX12-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX12-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffd -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2) ; GFX12-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo +; GFX12-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX12-FAKE16-NEXT: s_wait_alu 0xf1ff -; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX12-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX12-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX12-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX12-FAKE16-NEXT: s_wait_storecnt 0x0 -; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX12-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX12-FAKE16-NEXT: s_wait_dscnt 0x0 ; GFX12-FAKE16-NEXT: global_inv scope:SCOPE_SE -; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX12-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX12-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe ; GFX12-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX12-FAKE16-NEXT: s_wait_alu 0xfffe @@ -8912,27 +8948,27 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX942-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX942-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX942-NEXT: v_mov_b32_e32 v4, v3 +; GFX942-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX942-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX942-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX942-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX942-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX942-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX942-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX942-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX942-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX942-NEXT: v_add3_u32 v6, v6, v4, s4 +; GFX942-NEXT: v_add3_u32 v6, v6, v3, s4 ; GFX942-NEXT: v_add3_u32 v8, v8, v5, s4 ; GFX942-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v4, v4 +; GFX942-NEXT: v_cmp_u_f32_e64 s[0:1], v3, v3 ; GFX942-NEXT: s_nop 0 ; GFX942-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX942-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[0:1] -; GFX942-NEXT: v_perm_b32 v4, v5, v4, s5 -; GFX942-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX942-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] +; GFX942-NEXT: v_perm_b32 v3, v5, v3, s5 +; GFX942-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX942-NEXT: s_or_b64 s[2:3], vcc, s[2:3] -; GFX942-NEXT: v_mov_b32_e32 v3, v4 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[2:3] ; GFX942-NEXT: s_cbranch_execnz .LBB27_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -8950,30 +8986,32 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-TRUE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-TRUE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-TRUE16-NEXT: v_dual_sub_f32 v5, v5, v2 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-TRUE16-NEXT: v_sub_f32_e32 v4, v4, v1 +; GFX11-TRUE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-TRUE16-NEXT: v_sub_f32_e32 v5, v5, v2 +; GFX11-TRUE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-TRUE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 -; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v4, v4 +; GFX11-TRUE16-NEXT: v_sub_f32_e32 v3, v3, v1 ; GFX11-TRUE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) ; GFX11-TRUE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_4) -; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v4, v6, v8, vcc_lo +; GFX11-TRUE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-TRUE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v3, v3 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-TRUE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc_lo ; GFX11-TRUE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 ; GFX11-TRUE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo ; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) -; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v4.h +; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v3.h ; GFX11-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v5, v3 offset:65532 +; GFX11-TRUE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v5, v4 offset:65532 ; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-TRUE16-NEXT: buffer_gl0_inv -; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-TRUE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-TRUE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-TRUE16-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-TRUE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -8994,30 +9032,32 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX11-FAKE16-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX11-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 +; GFX11-FAKE16-NEXT: v_mov_b32_e32 v4, v3 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_dual_sub_f32 v5, v5, v1 :: v_dual_lshlrev_b32 v4, 16, v3 -; GFX11-FAKE16-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX11-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX11-FAKE16-NEXT: v_sub_f32_e32 v5, v5, v1 +; GFX11-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v4 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2) ; GFX11-FAKE16-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v4, 16, 1 -; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX11-FAKE16-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX11-FAKE16-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX11-FAKE16-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v4, 0x7fff -; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v4, v4 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_bfe_u32 v6, v3, 16, 1 +; GFX11-FAKE16-NEXT: v_or_b32_e32 v8, 0x400000, v3 +; GFX11-FAKE16-NEXT: v_cmp_u_f32_e64 s0, v3, v3 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) ; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v4, v6, v8, s0 -; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) -; GFX11-FAKE16-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX11-FAKE16-NEXT: v_add3_u32 v6, v6, v3, 0x7fff +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cndmask_b32_e64 v3, v6, v8, s0 +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX11-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v4, v0, v4, v3 offset:65532 +; GFX11-FAKE16-NEXT: ds_cmpstore_rtn_b32 v3, v0, v3, v4 offset:65532 ; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-FAKE16-NEXT: buffer_gl0_inv -; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX11-FAKE16-NEXT: v_mov_b32_e32 v3, v4 +; GFX11-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX11-FAKE16-NEXT: s_or_b32 s1, vcc_lo, s1 ; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s1 @@ -9037,27 +9077,27 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX10-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX10-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX10-NEXT: v_mov_b32_e32 v4, v3 +; GFX10-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX10-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX10-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX10-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX10-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX10-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX10-NEXT: v_bfe_u32 v7, v5, 16, 1 -; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v4 +; GFX10-NEXT: v_or_b32_e32 v8, 0x400000, v3 ; GFX10-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX10-NEXT: v_cmp_u_f32_e32 vcc_lo, v5, v5 -; GFX10-NEXT: v_add3_u32 v6, v6, v4, 0x7fff +; GFX10-NEXT: v_add3_u32 v6, v6, v3, 0x7fff ; GFX10-NEXT: v_add3_u32 v7, v7, v5, 0x7fff -; GFX10-NEXT: v_cmp_u_f32_e64 s4, v4, v4 +; GFX10-NEXT: v_cmp_u_f32_e64 s4, v3, v3 ; GFX10-NEXT: v_cndmask_b32_e32 v5, v7, v9, vcc_lo -; GFX10-NEXT: v_cndmask_b32_e64 v4, v6, v8, s4 -; GFX10-NEXT: v_perm_b32 v4, v5, v4, 0x7060302 +; GFX10-NEXT: v_cndmask_b32_e64 v3, v6, v8, s4 +; GFX10-NEXT: v_perm_b32 v3, v5, v3, 0x7060302 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX10-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v3 -; GFX10-NEXT: v_mov_b32_e32 v3, v4 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v4 ; GFX10-NEXT: s_or_b32 s5, vcc_lo, s5 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s5 ; GFX10-NEXT: s_cbranch_execnz .LBB27_1 @@ -9077,26 +9117,26 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX90A-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX90A-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX90A-NEXT: v_mov_b32_e32 v4, v3 +; GFX90A-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX90A-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX90A-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX90A-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX90A-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX90A-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX90A-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX90A-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX90A-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX90A-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX90A-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX90A-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX90A-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX90A-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX90A-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX90A-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX90A-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX90A-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX90A-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX90A-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX90A-NEXT: v_mov_b32_e32 v3, v4 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX90A-NEXT: s_cbranch_execnz .LBB27_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9115,26 +9155,26 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX908-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX908-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX908-NEXT: v_mov_b32_e32 v4, v3 +; GFX908-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX908-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX908-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX908-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX908-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX908-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX908-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v4 +; GFX908-NEXT: v_or_b32_e32 v7, 0x400000, v3 ; GFX908-NEXT: v_or_b32_e32 v9, 0x400000, v5 -; GFX908-NEXT: v_add3_u32 v6, v6, v4, s8 +; GFX908-NEXT: v_add3_u32 v6, v6, v3, s8 ; GFX908-NEXT: v_add3_u32 v8, v8, v5, s8 ; GFX908-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 -; GFX908-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX908-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 +; GFX908-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX908-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX908-NEXT: v_perm_b32 v4, v5, v4, s9 -; GFX908-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX908-NEXT: v_perm_b32 v3, v5, v3, s9 +; GFX908-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX908-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX908-NEXT: v_mov_b32_e32 v3, v4 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX908-NEXT: s_cbranch_execnz .LBB27_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9152,29 +9192,29 @@ define void @local_atomic_fsub_noret_v2bf16__ofset(ptr addrspace(3) %ptr, <2 x b ; GFX8-NEXT: .LBB27_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_lshlrev_b32_e32 v4, 16, v3 -; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v3 -; GFX8-NEXT: v_sub_f32_e32 v4, v4, v2 +; GFX8-NEXT: v_mov_b32_e32 v4, v3 +; GFX8-NEXT: v_lshlrev_b32_e32 v3, 16, v4 +; GFX8-NEXT: v_and_b32_e32 v5, 0xffff0000, v4 +; GFX8-NEXT: v_sub_f32_e32 v3, v3, v2 ; GFX8-NEXT: v_sub_f32_e32 v5, v5, v1 -; GFX8-NEXT: v_bfe_u32 v6, v4, 16, 1 +; GFX8-NEXT: v_bfe_u32 v6, v3, 16, 1 ; GFX8-NEXT: v_bfe_u32 v8, v5, 16, 1 -; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v4 +; GFX8-NEXT: v_add_u32_e32 v6, vcc, v6, v3 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, v8, v5 ; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x7fff, v6 ; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x7fff, v8 ; GFX8-NEXT: v_or_b32_e32 v9, 0x400000, v5 ; GFX8-NEXT: v_cmp_u_f32_e32 vcc, v5, v5 -; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v4 -; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v4, v4 +; GFX8-NEXT: v_or_b32_e32 v7, 0x400000, v3 +; GFX8-NEXT: v_cmp_u_f32_e64 s[4:5], v3, v3 ; GFX8-NEXT: v_cndmask_b32_e32 v5, v8, v9, vcc -; GFX8-NEXT: v_cndmask_b32_e64 v4, v6, v7, s[4:5] +; GFX8-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[4:5] ; GFX8-NEXT: v_lshrrev_b32_e32 v5, 16, v5 -; GFX8-NEXT: v_alignbit_b32 v4, v5, v4, 16 -; GFX8-NEXT: ds_cmpst_rtn_b32 v4, v0, v3, v4 offset:65532 +; GFX8-NEXT: v_alignbit_b32 v3, v5, v3, 16 +; GFX8-NEXT: ds_cmpst_rtn_b32 v3, v0, v4, v3 offset:65532 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v4, v3 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v3, v4 ; GFX8-NEXT: s_or_b64 s[6:7], vcc, s[6:7] -; GFX8-NEXT: v_mov_b32_e32 v3, v4 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[6:7] ; GFX8-NEXT: s_cbranch_execnz .LBB27_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9489,13 +9529,14 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX12-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX12-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX12-NEXT: s_wait_dscnt 0x0 -; GFX12-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX12-NEXT: v_mov_b32_e32 v2, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX12-NEXT: s_wait_storecnt 0x0 -; GFX12-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 +; GFX12-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 ; GFX12-NEXT: s_wait_dscnt 0x0 ; GFX12-NEXT: global_inv scope:SCOPE_SE -; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX12-NEXT: v_mov_b32_e32 v1, v2 +; GFX12-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX12-NEXT: s_wait_alu 0xfffe ; GFX12-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX12-NEXT: s_wait_alu 0xfffe @@ -9514,12 +9555,12 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX942-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX942-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX942-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX942-NEXT: v_mov_b32_e32 v2, v1 +; GFX942-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX942-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX942-NEXT: s_waitcnt lgkmcnt(0) -; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX942-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX942-NEXT: s_or_b64 s[0:1], vcc, s[0:1] -; GFX942-NEXT: v_mov_b32_e32 v1, v2 ; GFX942-NEXT: s_andn2_b64 exec, exec, s[0:1] ; GFX942-NEXT: s_cbranch_execnz .LBB29_1 ; GFX942-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9534,13 +9575,14 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX11-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX11-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) -; GFX11-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX11-NEXT: v_mov_b32_e32 v2, v1 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX11-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX11-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX11-NEXT: ds_cmpstore_rtn_b32 v2, v0, v2, v1 +; GFX11-NEXT: ds_cmpstore_rtn_b32 v1, v0, v1, v2 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: buffer_gl0_inv -; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX11-NEXT: v_mov_b32_e32 v1, v2 +; GFX11-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX11-NEXT: s_or_b32 s0, vcc_lo, s0 ; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX11-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0 @@ -9557,13 +9599,13 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX10-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX10-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) -; GFX10-NEXT: v_add_f32_e32 v2, -4.0, v1 +; GFX10-NEXT: v_mov_b32_e32 v2, v1 +; GFX10-NEXT: v_add_f32_e32 v1, -4.0, v2 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 -; GFX10-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX10-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: buffer_gl0_inv -; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v1 -; GFX10-NEXT: v_mov_b32_e32 v1, v2 +; GFX10-NEXT: v_cmp_eq_u32_e32 vcc_lo, v1, v2 ; GFX10-NEXT: s_or_b32 s4, vcc_lo, s4 ; GFX10-NEXT: s_andn2_b32 exec_lo, exec_lo, s4 ; GFX10-NEXT: s_cbranch_execnz .LBB29_1 @@ -9579,12 +9621,12 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX90A-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX90A-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX90A-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX90A-NEXT: v_mov_b32_e32 v2, v1 +; GFX90A-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX90A-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX90A-NEXT: s_waitcnt lgkmcnt(0) -; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX90A-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX90A-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX90A-NEXT: v_mov_b32_e32 v1, v2 ; GFX90A-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX90A-NEXT: s_cbranch_execnz .LBB29_1 ; GFX90A-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9599,12 +9641,12 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX908-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX908-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX908-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX908-NEXT: v_mov_b32_e32 v2, v1 +; GFX908-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX908-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX908-NEXT: s_waitcnt lgkmcnt(0) -; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX908-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX908-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX908-NEXT: v_mov_b32_e32 v1, v2 ; GFX908-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX908-NEXT: s_cbranch_execnz .LBB29_1 ; GFX908-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9620,12 +9662,12 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX8-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX8-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX8-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX8-NEXT: v_mov_b32_e32 v2, v1 +; GFX8-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX8-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX8-NEXT: s_waitcnt lgkmcnt(0) -; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX8-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX8-NEXT: v_mov_b32_e32 v1, v2 ; GFX8-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX8-NEXT: s_cbranch_execnz .LBB29_1 ; GFX8-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9641,12 +9683,12 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX7-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX7-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX7-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX7-NEXT: v_mov_b32_e32 v2, v1 +; GFX7-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX7-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX7-NEXT: s_waitcnt lgkmcnt(0) -; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX7-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX7-NEXT: v_mov_b32_e32 v1, v2 ; GFX7-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX7-NEXT: s_cbranch_execnz .LBB29_1 ; GFX7-NEXT: ; %bb.2: ; %atomicrmw.end @@ -9662,12 +9704,12 @@ define void @local_atomic_fsub_noret_f32__amdgpu_ignore_denormal_mode(ptr addrsp ; GFX6-NEXT: .LBB29_1: ; %atomicrmw.start ; GFX6-NEXT: ; =>This Inner Loop Header: Depth=1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_add_f32_e32 v2, -4.0, v1 -; GFX6-NEXT: ds_cmpst_rtn_b32 v2, v0, v1, v2 +; GFX6-NEXT: v_mov_b32_e32 v2, v1 +; GFX6-NEXT: v_add_f32_e32 v1, -4.0, v2 +; GFX6-NEXT: ds_cmpst_rtn_b32 v1, v0, v2, v1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v2, v1 +; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, v1, v2 ; GFX6-NEXT: s_or_b64 s[4:5], vcc, s[4:5] -; GFX6-NEXT: v_mov_b32_e32 v1, v2 ; GFX6-NEXT: s_andn2_b64 exec, exec, s[4:5] ; GFX6-NEXT: s_cbranch_execnz .LBB29_1 ; GFX6-NEXT: ; %bb.2: ; %atomicrmw.end diff --git a/llvm/test/CodeGen/AMDGPU/lshr.v2i16.ll b/llvm/test/CodeGen/AMDGPU/lshr.v2i16.ll index 68506ce..9056d40 100644 --- a/llvm/test/CodeGen/AMDGPU/lshr.v2i16.ll +++ b/llvm/test/CodeGen/AMDGPU/lshr.v2i16.ll @@ -36,20 +36,19 @@ define amdgpu_kernel void @s_lshr_v2i16(ptr addrspace(1) %out, <2 x i16> %lhs, < ; CI-LABEL: s_lshr_v2i16: ; CI: ; %bb.0: ; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; CI-NEXT: s_mov_b32 s7, 0xf000 -; CI-NEXT: s_mov_b32 s6, -1 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_mov_b32 s4, s0 -; CI-NEXT: s_mov_b32 s5, s1 -; CI-NEXT: s_and_b32 s0, s2, 0xffff -; CI-NEXT: s_lshr_b32 s1, s2, 16 -; CI-NEXT: s_lshr_b32 s2, s3, 16 -; CI-NEXT: s_lshr_b32 s1, s1, s2 -; CI-NEXT: s_lshl_b32 s1, s1, 16 -; CI-NEXT: s_lshr_b32 s0, s0, s3 -; CI-NEXT: s_or_b32 s0, s0, s1 -; CI-NEXT: v_mov_b32_e32 v0, s0 -; CI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; CI-NEXT: s_mov_b64 s[4:5], s[2:3] +; CI-NEXT: s_and_b32 s6, s4, 0xffff +; CI-NEXT: s_lshr_b32 s4, s4, 16 +; CI-NEXT: s_lshr_b32 s7, s5, 16 +; CI-NEXT: s_lshr_b32 s4, s4, s7 +; CI-NEXT: s_lshl_b32 s4, s4, 16 +; CI-NEXT: s_lshr_b32 s5, s6, s5 +; CI-NEXT: s_or_b32 s4, s5, s4 +; CI-NEXT: s_mov_b32 s3, 0xf000 +; CI-NEXT: s_mov_b32 s2, -1 +; CI-NEXT: v_mov_b32_e32 v0, s4 +; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; CI-NEXT: s_endpgm ; ; GFX10-LABEL: s_lshr_v2i16: diff --git a/llvm/test/CodeGen/AMDGPU/machine-sink-temporal-divergence-swdev407790.ll b/llvm/test/CodeGen/AMDGPU/machine-sink-temporal-divergence-swdev407790.ll index 680942fcb..9ecd35e 100644 --- a/llvm/test/CodeGen/AMDGPU/machine-sink-temporal-divergence-swdev407790.ll +++ b/llvm/test/CodeGen/AMDGPU/machine-sink-temporal-divergence-swdev407790.ll @@ -133,7 +133,7 @@ define protected amdgpu_kernel void @kernel_round1(ptr addrspace(1) nocapture no ; CHECK-NEXT: ; %bb.3: ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s4 ; CHECK-NEXT: v_add_nc_u32_e32 v45, -1, v42 -; CHECK-NEXT: s_mov_b32 s53, 0 +; CHECK-NEXT: s_mov_b32 s55, 0 ; CHECK-NEXT: v_cmp_ne_u32_e32 vcc_lo, 0, v45 ; CHECK-NEXT: s_and_b32 exec_lo, exec_lo, vcc_lo ; CHECK-NEXT: s_cbranch_execz .LBB0_25 @@ -141,7 +141,7 @@ define protected amdgpu_kernel void @kernel_round1(ptr addrspace(1) nocapture no ; CHECK-NEXT: v_lshlrev_b32_e32 v43, 10, v43 ; CHECK-NEXT: v_add_nc_u32_e32 v46, 0x3c05, v0 ; CHECK-NEXT: v_mov_b32_e32 v47, 0 -; CHECK-NEXT: s_mov_b32 s55, 0 +; CHECK-NEXT: s_mov_b32 s53, 0 ; CHECK-NEXT: .LBB0_5: ; =>This Loop Header: Depth=1 ; CHECK-NEXT: ; Child Loop BB0_8 Depth 2 ; CHECK-NEXT: ; Child Loop BB0_20 Depth 2 @@ -866,8 +866,8 @@ define protected amdgpu_kernel void @kernel_round1_short(ptr addrspace(1) nocapt ; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] ; CHECK-NEXT: v_mov_b32_e32 v41, v0 ; CHECK-NEXT: v_lshlrev_b32_e32 v42, 10, v42 -; CHECK-NEXT: s_mov_b32 s52, 0 ; CHECK-NEXT: s_mov_b32 s4, 0 +; CHECK-NEXT: s_mov_b32 s52, 0 ; CHECK-NEXT: ds_write_b8 v46, v43 offset:15364 ; CHECK-NEXT: v_add_nc_u32_e32 v45, -1, v41 ; CHECK-NEXT: .LBB1_1: ; %.37 diff --git a/llvm/test/CodeGen/AMDGPU/mad_uint24.ll b/llvm/test/CodeGen/AMDGPU/mad_uint24.ll index 46b8df4..9cc0e62 100644 --- a/llvm/test/CodeGen/AMDGPU/mad_uint24.ll +++ b/llvm/test/CodeGen/AMDGPU/mad_uint24.ll @@ -133,35 +133,33 @@ define amdgpu_kernel void @i16_mad24(ptr addrspace(1) %out, i16 %a, i16 %b, i16 ; 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_load_dword s6, s[4:5], 0xb ; 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_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_lshr_b32 s2, s4, 16 +; GCN-NEXT: s_mul_i32 s2, s6, s2 +; GCN-NEXT: s_add_i32 s2, s2, s5 +; GCN-NEXT: s_sext_i32_i16 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: 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_load_dword s6, s[4:5], 0x2c ; 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_mov_b64 s[4:5], s[2:3] +; GFX8-NEXT: s_lshr_b32 s4, s4, 16 +; GFX8-NEXT: s_mul_i32 s4, s6, s4 +; GFX8-NEXT: s_add_i32 s4, s4, s5 +; GFX8-NEXT: s_sext_i32_i16 s4, s4 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; 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 i16 %a, %b diff --git a/llvm/test/CodeGen/AMDGPU/max.ll b/llvm/test/CodeGen/AMDGPU/max.ll index ba53294..c48e25f3 100644 --- a/llvm/test/CodeGen/AMDGPU/max.ll +++ b/llvm/test/CodeGen/AMDGPU/max.ll @@ -155,14 +155,13 @@ define amdgpu_kernel void @s_test_imax_sge_i32(ptr addrspace(1) %out, i32 %a, i3 ; SI-LABEL: s_test_imax_sge_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_max_i32 s0, s2, s3 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_max_i32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; GFX1250-LABEL: s_test_imax_sge_i32: @@ -357,16 +356,15 @@ define amdgpu_kernel void @s_test_imax_sgt_imm_v2i32(ptr addrspace(1) %out, <2 x ; SI-LABEL: s_test_imax_sgt_imm_v2i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_max_i32 s0, s3, 9 -; SI-NEXT: s_max_i32 s1, s2, 9 -; SI-NEXT: v_mov_b32_e32 v0, s1 -; SI-NEXT: v_mov_b32_e32 v1, s0 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_max_i32 s5, s5, 9 +; SI-NEXT: s_max_i32 s4, s4, 9 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; GFX1250-LABEL: s_test_imax_sgt_imm_v2i32: @@ -472,14 +470,13 @@ define amdgpu_kernel void @s_test_imax_sgt_i32(ptr addrspace(1) %out, i32 %a, i3 ; SI-LABEL: s_test_imax_sgt_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_max_i32 s0, s2, s3 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_max_i32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; GFX1250-LABEL: s_test_imax_sgt_i32: @@ -582,14 +579,13 @@ define amdgpu_kernel void @s_test_umax_uge_i32(ptr addrspace(1) %out, i32 %a, i3 ; SI-LABEL: s_test_umax_uge_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_max_u32 s0, s2, s3 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_max_u32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; GFX1250-LABEL: s_test_umax_uge_i32: @@ -817,14 +813,13 @@ define amdgpu_kernel void @s_test_umax_ugt_i32(ptr addrspace(1) %out, i32 %a, i3 ; SI-LABEL: s_test_umax_ugt_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_max_u32 s0, s2, s3 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_max_u32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; GFX1250-LABEL: s_test_umax_ugt_i32: @@ -858,16 +853,15 @@ define amdgpu_kernel void @s_test_umax_ugt_imm_v2i32(ptr addrspace(1) %out, <2 x ; SI-LABEL: s_test_umax_ugt_imm_v2i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_max_u32 s0, s3, 23 -; SI-NEXT: s_max_u32 s1, s2, 15 -; SI-NEXT: v_mov_b32_e32 v0, s1 -; SI-NEXT: v_mov_b32_e32 v1, s0 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_max_u32 s5, s5, 23 +; SI-NEXT: s_max_u32 s4, s4, 15 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mov_b32_e32 v1, s5 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; GFX1250-LABEL: s_test_umax_ugt_imm_v2i32: diff --git a/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll b/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll index ca4f5d2..43752c2 100644 --- a/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll +++ b/llvm/test/CodeGen/AMDGPU/memcpy-crash-issue63986.ll @@ -90,18 +90,18 @@ define void @issue63986(i64 %0, i64 %idxprom, ptr inreg %ptr) { ; CHECK-NEXT: .LBB0_13: ; %loop-memcpy-expansion2 ; CHECK-NEXT: ; Parent Loop BB0_11 Depth=1 ; CHECK-NEXT: ; => This Inner Loop Header: Depth=2 -; CHECK-NEXT: v_mov_b32_e32 v6, s12 -; CHECK-NEXT: v_mov_b32_e32 v7, s13 +; CHECK-NEXT: v_mov_b32_e32 v6, s10 +; CHECK-NEXT: v_mov_b32_e32 v7, s11 ; CHECK-NEXT: flat_load_dwordx4 v[10:13], v[6:7] -; CHECK-NEXT: v_add_co_u32_e32 v6, vcc, s12, v8 -; CHECK-NEXT: s_add_u32 s12, s12, 16 +; CHECK-NEXT: v_add_co_u32_e32 v6, vcc, s10, v8 +; CHECK-NEXT: s_add_u32 s10, s10, 16 ; CHECK-NEXT: v_addc_co_u32_e32 v7, vcc, v9, v7, vcc -; CHECK-NEXT: s_addc_u32 s13, s13, 0 -; CHECK-NEXT: v_cmp_ge_u64_e32 vcc, s[12:13], v[0:1] -; CHECK-NEXT: s_or_b64 s[10:11], vcc, s[10:11] +; CHECK-NEXT: s_addc_u32 s11, s11, 0 +; CHECK-NEXT: v_cmp_ge_u64_e32 vcc, s[10:11], v[0:1] +; CHECK-NEXT: s_or_b64 s[12:13], vcc, s[12:13] ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: flat_store_dwordx4 v[6:7], v[10:13] -; CHECK-NEXT: s_andn2_b64 exec, exec, s[10:11] +; CHECK-NEXT: s_andn2_b64 exec, exec, s[12:13] ; CHECK-NEXT: s_cbranch_execnz .LBB0_13 ; CHECK-NEXT: .LBB0_14: ; %Flow15 ; CHECK-NEXT: ; in Loop: Header=BB0_11 Depth=1 @@ -115,8 +115,8 @@ define void @issue63986(i64 %0, i64 %idxprom, ptr inreg %ptr) { ; CHECK-NEXT: s_cbranch_execz .LBB0_9 ; CHECK-NEXT: ; %bb.16: ; %loop-memcpy-residual4.preheader ; CHECK-NEXT: ; in Loop: Header=BB0_11 Depth=1 -; CHECK-NEXT: s_mov_b64 s[12:13], 0 ; CHECK-NEXT: s_mov_b64 s[14:15], 0 +; CHECK-NEXT: s_mov_b64 s[12:13], 0 ; CHECK-NEXT: .LBB0_17: ; %loop-memcpy-residual4 ; CHECK-NEXT: ; Parent Loop BB0_11 Depth=1 ; CHECK-NEXT: ; => This Inner Loop Header: Depth=2 diff --git a/llvm/test/CodeGen/AMDGPU/memmove-var-size.ll b/llvm/test/CodeGen/AMDGPU/memmove-var-size.ll index 14b0729..953511d 100644 --- a/llvm/test/CodeGen/AMDGPU/memmove-var-size.ll +++ b/llvm/test/CodeGen/AMDGPU/memmove-var-size.ll @@ -10,13 +10,13 @@ define void @memmove_p0_p0(ptr addrspace(0) align 1 %dst, ptr addrspace(0) align ; CHECK-LABEL: memmove_p0_p0: ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_and_b32_e32 v8, 15, v4 -; CHECK-NEXT: v_mov_b32_e32 v9, 0 -; CHECK-NEXT: v_and_b32_e32 v6, -16, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v5 +; CHECK-NEXT: v_and_b32_e32 v6, 15, v4 +; CHECK-NEXT: v_mov_b32_e32 v7, 0 +; CHECK-NEXT: v_and_b32_e32 v8, -16, v4 +; CHECK-NEXT: v_mov_b32_e32 v9, v5 ; CHECK-NEXT: s_mov_b32 s6, exec_lo -; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[8:9] -; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9] ; CHECK-NEXT: v_cmpx_ge_u64_e64 v[2:3], v[0:1] ; CHECK-NEXT: s_xor_b32 s7, exec_lo, s6 ; CHECK-NEXT: s_cbranch_execnz .LBB0_3 @@ -33,10 +33,10 @@ define void @memmove_p0_p0(ptr addrspace(0) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: ; %bb.4: ; %memmove_fwd_main_loop.preheader ; CHECK-NEXT: v_mov_b32_e32 v5, v3 ; CHECK-NEXT: v_mov_b32_e32 v11, v1 -; CHECK-NEXT: v_mov_b32_e32 v13, v7 +; CHECK-NEXT: v_mov_b32_e32 v13, v9 ; CHECK-NEXT: v_mov_b32_e32 v4, v2 ; CHECK-NEXT: v_mov_b32_e32 v10, v0 -; CHECK-NEXT: v_mov_b32_e32 v12, v6 +; CHECK-NEXT: v_mov_b32_e32 v12, v8 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB0_5: ; %memmove_fwd_main_loop @@ -59,20 +59,20 @@ define void @memmove_p0_p0(ptr addrspace(0) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: s_and_saveexec_b32 s8, s4 ; CHECK-NEXT: s_cbranch_execz .LBB0_9 ; CHECK-NEXT: ; %bb.7: ; %memmove_fwd_residual_loop.preheader -; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v7, s5 -; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v7, s5 +; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v9, s5 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB0_8: ; %memmove_fwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: flat_load_ubyte v4, v[2:3] -; CHECK-NEXT: v_add_co_u32 v8, s5, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v6, s5, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s5 ; CHECK-NEXT: v_add_co_u32 v2, s5, v2, 1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, s5 -; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s9, s5, s9 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: flat_store_byte v[0:1], v4 @@ -82,10 +82,10 @@ define void @memmove_p0_p0(ptr addrspace(0) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: s_cbranch_execnz .LBB0_8 ; CHECK-NEXT: .LBB0_9: ; %Flow28 ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s8 -; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 +; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1 ; CHECK-NEXT: ; implicit-def: $vgpr2_vgpr3 -; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 +; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 ; CHECK-NEXT: ; implicit-def: $vgpr4_vgpr5 ; CHECK-NEXT: s_andn2_saveexec_b32 s6, s7 ; CHECK-NEXT: s_cbranch_execz .LBB0_2 @@ -104,11 +104,11 @@ define void @memmove_p0_p0(ptr addrspace(0) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: .LBB0_12: ; %memmove_bwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: flat_load_ubyte v12, v[10:11] -; CHECK-NEXT: v_add_co_u32 v8, s4, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s4 +; CHECK-NEXT: v_add_co_u32 v6, s4, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s4 ; CHECK-NEXT: v_add_co_u32 v10, s4, v10, -1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, -1, v11, s4 -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s8, s4, s8 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: flat_store_byte v[4:5], v12 @@ -129,19 +129,19 @@ define void @memmove_p0_p0(ptr addrspace(0) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB0_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v7, vcc_lo -; CHECK-NEXT: v_add_co_u32 v12, s4, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v13, null, v1, v7, s4 -; CHECK-NEXT: flat_load_dwordx4 v[8:11], v[4:5] -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v6, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v7, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[4:5] -; CHECK-NEXT: v_mov_b32_e32 v7, v5 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v11, v9 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v10 +; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v11, vcc_lo +; CHECK-NEXT: v_add_co_u32 v8, vcc_lo, v10, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v11, vcc_lo +; CHECK-NEXT: flat_load_dwordx4 v[4:7], v[4:5] +; CHECK-NEXT: v_add_co_u32 v10, s4, v0, v10 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[8:9] +; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, v1, v11, s4 ; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; CHECK-NEXT: flat_store_dwordx4 v[12:13], v[8:11] +; CHECK-NEXT: flat_store_dwordx4 v[10:11], v[4:7] ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB0_15 ; CHECK-NEXT: .LBB0_16: ; %Flow32 @@ -158,13 +158,13 @@ define void @memmove_p0_p1(ptr addrspace(0) align 1 %dst, ptr addrspace(1) align ; CHECK-LABEL: memmove_p0_p1: ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_and_b32_e32 v8, 15, v4 -; CHECK-NEXT: v_mov_b32_e32 v9, 0 -; CHECK-NEXT: v_and_b32_e32 v6, -16, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v5 +; CHECK-NEXT: v_and_b32_e32 v6, 15, v4 +; CHECK-NEXT: v_mov_b32_e32 v7, 0 +; CHECK-NEXT: v_and_b32_e32 v8, -16, v4 +; CHECK-NEXT: v_mov_b32_e32 v9, v5 ; CHECK-NEXT: s_mov_b32 s6, exec_lo -; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[8:9] -; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9] ; CHECK-NEXT: v_cmpx_ge_u64_e64 v[2:3], v[0:1] ; CHECK-NEXT: s_xor_b32 s7, exec_lo, s6 ; CHECK-NEXT: s_cbranch_execnz .LBB1_3 @@ -181,10 +181,10 @@ define void @memmove_p0_p1(ptr addrspace(0) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: ; %bb.4: ; %memmove_fwd_main_loop.preheader ; CHECK-NEXT: v_mov_b32_e32 v5, v3 ; CHECK-NEXT: v_mov_b32_e32 v11, v1 -; CHECK-NEXT: v_mov_b32_e32 v13, v7 +; CHECK-NEXT: v_mov_b32_e32 v13, v9 ; CHECK-NEXT: v_mov_b32_e32 v4, v2 ; CHECK-NEXT: v_mov_b32_e32 v10, v0 -; CHECK-NEXT: v_mov_b32_e32 v12, v6 +; CHECK-NEXT: v_mov_b32_e32 v12, v8 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB1_5: ; %memmove_fwd_main_loop @@ -207,20 +207,20 @@ define void @memmove_p0_p1(ptr addrspace(0) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: s_and_saveexec_b32 s8, s4 ; CHECK-NEXT: s_cbranch_execz .LBB1_9 ; CHECK-NEXT: ; %bb.7: ; %memmove_fwd_residual_loop.preheader -; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v7, s5 -; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v7, s5 +; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v9, s5 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB1_8: ; %memmove_fwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v4, v[2:3], off -; CHECK-NEXT: v_add_co_u32 v8, s5, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v6, s5, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s5 ; CHECK-NEXT: v_add_co_u32 v2, s5, v2, 1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, s5 -; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s9, s5, s9 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: flat_store_byte v[0:1], v4 @@ -230,10 +230,10 @@ define void @memmove_p0_p1(ptr addrspace(0) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: s_cbranch_execnz .LBB1_8 ; CHECK-NEXT: .LBB1_9: ; %Flow30 ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s8 -; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 +; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 ; CHECK-NEXT: ; implicit-def: $vgpr2_vgpr3 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1 -; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 +; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 ; CHECK-NEXT: ; implicit-def: $vgpr4_vgpr5 ; CHECK-NEXT: s_andn2_saveexec_b32 s6, s7 ; CHECK-NEXT: s_cbranch_execz .LBB1_2 @@ -252,11 +252,11 @@ define void @memmove_p0_p1(ptr addrspace(0) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: .LBB1_12: ; %memmove_bwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v12, v[4:5], off -; CHECK-NEXT: v_add_co_u32 v8, s4, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s4 +; CHECK-NEXT: v_add_co_u32 v6, s4, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s4 ; CHECK-NEXT: v_add_co_u32 v4, s4, v4, -1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v5, s4 -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s8, s4, s8 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: flat_store_byte v[10:11], v12 @@ -277,19 +277,19 @@ define void @memmove_p0_p1(ptr addrspace(0) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB1_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v7, vcc_lo -; CHECK-NEXT: v_add_co_u32 v12, s4, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v13, null, v1, v7, s4 -; CHECK-NEXT: global_load_dwordx4 v[8:11], v[4:5], off -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v6, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v7, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[4:5] -; CHECK-NEXT: v_mov_b32_e32 v7, v5 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v11, v9 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v10 +; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v11, vcc_lo +; CHECK-NEXT: v_add_co_u32 v8, vcc_lo, v10, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v11, vcc_lo +; CHECK-NEXT: global_load_dwordx4 v[4:7], v[4:5], off +; CHECK-NEXT: v_add_co_u32 v10, s4, v0, v10 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[8:9] +; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, v1, v11, s4 ; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) -; CHECK-NEXT: flat_store_dwordx4 v[12:13], v[8:11] +; CHECK-NEXT: flat_store_dwordx4 v[10:11], v[4:7] ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB1_15 ; CHECK-NEXT: .LBB1_16: ; %Flow34 @@ -423,17 +423,17 @@ define void @memmove_p0_p3(ptr addrspace(0) align 1 %dst, ptr addrspace(3) align ; CHECK-NEXT: .LBB2_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: ds_read_b128 v[7:10], v2 -; CHECK-NEXT: v_add_co_u32 v3, vcc_lo, v5, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v4, null, -1, v6, vcc_lo -; CHECK-NEXT: v_add_co_u32 v11, vcc_lo, v0, v5 -; CHECK-NEXT: v_add_co_ci_u32_e64 v12, null, v1, v6, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[3:4] -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v3, v5 +; CHECK-NEXT: v_mov_b32_e32 v4, v6 ; CHECK-NEXT: v_add_nc_u32_e32 v2, -16, v2 -; CHECK-NEXT: v_mov_b32_e32 v5, v3 -; CHECK-NEXT: s_or_b32 s7, s4, s7 +; CHECK-NEXT: v_add_co_u32 v5, vcc_lo, v3, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v6, null, -1, v4, vcc_lo +; CHECK-NEXT: v_add_co_u32 v3, s4, v0, v3 +; CHECK-NEXT: v_add_co_ci_u32_e64 v4, null, v1, v4, s4 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[5:6] +; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt lgkmcnt(0) -; CHECK-NEXT: flat_store_dwordx4 v[11:12], v[7:10] +; CHECK-NEXT: flat_store_dwordx4 v[3:4], v[7:10] ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB2_15 ; CHECK-NEXT: .LBB2_16: ; %Flow36 @@ -450,13 +450,13 @@ define void @memmove_p0_p4(ptr addrspace(0) align 1 %dst, ptr addrspace(4) align ; CHECK-LABEL: memmove_p0_p4: ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_and_b32_e32 v8, 15, v4 -; CHECK-NEXT: v_mov_b32_e32 v9, 0 -; CHECK-NEXT: v_and_b32_e32 v6, -16, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v5 +; CHECK-NEXT: v_and_b32_e32 v6, 15, v4 +; CHECK-NEXT: v_mov_b32_e32 v7, 0 +; CHECK-NEXT: v_and_b32_e32 v8, -16, v4 +; CHECK-NEXT: v_mov_b32_e32 v9, v5 ; CHECK-NEXT: s_mov_b32 s6, exec_lo -; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[8:9] -; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9] ; CHECK-NEXT: v_cmpx_ge_u64_e64 v[2:3], v[0:1] ; CHECK-NEXT: s_xor_b32 s7, exec_lo, s6 ; CHECK-NEXT: s_cbranch_execnz .LBB3_3 @@ -473,10 +473,10 @@ define void @memmove_p0_p4(ptr addrspace(0) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: ; %bb.4: ; %memmove_fwd_main_loop.preheader ; CHECK-NEXT: v_mov_b32_e32 v5, v3 ; CHECK-NEXT: v_mov_b32_e32 v11, v1 -; CHECK-NEXT: v_mov_b32_e32 v13, v7 +; CHECK-NEXT: v_mov_b32_e32 v13, v9 ; CHECK-NEXT: v_mov_b32_e32 v4, v2 ; CHECK-NEXT: v_mov_b32_e32 v10, v0 -; CHECK-NEXT: v_mov_b32_e32 v12, v6 +; CHECK-NEXT: v_mov_b32_e32 v12, v8 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB3_5: ; %memmove_fwd_main_loop @@ -499,20 +499,20 @@ define void @memmove_p0_p4(ptr addrspace(0) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: s_and_saveexec_b32 s8, s4 ; CHECK-NEXT: s_cbranch_execz .LBB3_9 ; CHECK-NEXT: ; %bb.7: ; %memmove_fwd_residual_loop.preheader -; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v7, s5 -; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v7, s5 +; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v9, s5 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB3_8: ; %memmove_fwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v4, v[2:3], off -; CHECK-NEXT: v_add_co_u32 v8, s5, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v6, s5, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s5 ; CHECK-NEXT: v_add_co_u32 v2, s5, v2, 1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, s5 -; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s9, s5, s9 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: flat_store_byte v[0:1], v4 @@ -522,10 +522,10 @@ define void @memmove_p0_p4(ptr addrspace(0) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: s_cbranch_execnz .LBB3_8 ; CHECK-NEXT: .LBB3_9: ; %Flow29 ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s8 -; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 +; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 ; CHECK-NEXT: ; implicit-def: $vgpr2_vgpr3 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1 -; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 +; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 ; CHECK-NEXT: ; implicit-def: $vgpr4_vgpr5 ; CHECK-NEXT: s_andn2_saveexec_b32 s6, s7 ; CHECK-NEXT: s_cbranch_execz .LBB3_2 @@ -544,11 +544,11 @@ define void @memmove_p0_p4(ptr addrspace(0) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: .LBB3_12: ; %memmove_bwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v12, v[10:11], off -; CHECK-NEXT: v_add_co_u32 v8, s4, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s4 +; CHECK-NEXT: v_add_co_u32 v6, s4, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s4 ; CHECK-NEXT: v_add_co_u32 v10, s4, v10, -1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, -1, v11, s4 -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s8, s4, s8 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: flat_store_byte v[4:5], v12 @@ -569,19 +569,19 @@ define void @memmove_p0_p4(ptr addrspace(0) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB3_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v7, vcc_lo -; CHECK-NEXT: v_add_co_u32 v12, s4, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v13, null, v1, v7, s4 -; CHECK-NEXT: global_load_dwordx4 v[8:11], v[4:5], off -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v6, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v7, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[4:5] -; CHECK-NEXT: v_mov_b32_e32 v7, v5 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v11, v9 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v10 +; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v11, vcc_lo +; CHECK-NEXT: v_add_co_u32 v8, vcc_lo, v10, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v11, vcc_lo +; CHECK-NEXT: global_load_dwordx4 v[4:7], v[4:5], off +; CHECK-NEXT: v_add_co_u32 v10, s4, v0, v10 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[8:9] +; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, v1, v11, s4 ; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) -; CHECK-NEXT: flat_store_dwordx4 v[12:13], v[8:11] +; CHECK-NEXT: flat_store_dwordx4 v[10:11], v[4:7] ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB3_15 ; CHECK-NEXT: .LBB3_16: ; %Flow33 @@ -723,17 +723,17 @@ define void @memmove_p0_p5(ptr addrspace(0) align 1 %dst, ptr addrspace(5) align ; CHECK-NEXT: buffer_load_dword v8, v2, s[0:3], 0 offen offset:4 ; CHECK-NEXT: buffer_load_dword v9, v2, s[0:3], 0 offen offset:8 ; CHECK-NEXT: buffer_load_dword v10, v2, s[0:3], 0 offen offset:12 -; CHECK-NEXT: v_add_co_u32 v3, vcc_lo, v5, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v4, null, -1, v6, vcc_lo -; CHECK-NEXT: v_add_co_u32 v11, vcc_lo, v0, v5 -; CHECK-NEXT: v_add_co_ci_u32_e64 v12, null, v1, v6, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[3:4] -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v3, v5 +; CHECK-NEXT: v_mov_b32_e32 v4, v6 ; CHECK-NEXT: v_add_nc_u32_e32 v2, -16, v2 -; CHECK-NEXT: v_mov_b32_e32 v5, v3 -; CHECK-NEXT: s_or_b32 s7, s4, s7 +; CHECK-NEXT: v_add_co_u32 v5, vcc_lo, v3, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v6, null, -1, v4, vcc_lo +; CHECK-NEXT: v_add_co_u32 v3, s4, v0, v3 +; CHECK-NEXT: v_add_co_ci_u32_e64 v4, null, v1, v4, s4 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[5:6] +; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) -; CHECK-NEXT: flat_store_dwordx4 v[11:12], v[7:10] +; CHECK-NEXT: flat_store_dwordx4 v[3:4], v[7:10] ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB4_15 ; CHECK-NEXT: .LBB4_16: ; %Flow36 @@ -751,13 +751,13 @@ define void @memmove_p1_p0(ptr addrspace(1) align 1 %dst, ptr addrspace(0) align ; CHECK-LABEL: memmove_p1_p0: ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_and_b32_e32 v8, 15, v4 -; CHECK-NEXT: v_mov_b32_e32 v9, 0 -; CHECK-NEXT: v_and_b32_e32 v6, -16, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v5 +; CHECK-NEXT: v_and_b32_e32 v6, 15, v4 +; CHECK-NEXT: v_mov_b32_e32 v7, 0 +; CHECK-NEXT: v_and_b32_e32 v8, -16, v4 +; CHECK-NEXT: v_mov_b32_e32 v9, v5 ; CHECK-NEXT: s_mov_b32 s6, exec_lo -; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[8:9] -; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9] ; CHECK-NEXT: v_cmpx_ge_u64_e64 v[2:3], v[0:1] ; CHECK-NEXT: s_xor_b32 s7, exec_lo, s6 ; CHECK-NEXT: s_cbranch_execnz .LBB5_3 @@ -773,10 +773,10 @@ define void @memmove_p1_p0(ptr addrspace(1) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: ; %bb.4: ; %memmove_fwd_main_loop.preheader ; CHECK-NEXT: v_mov_b32_e32 v5, v3 ; CHECK-NEXT: v_mov_b32_e32 v11, v1 -; CHECK-NEXT: v_mov_b32_e32 v13, v7 +; CHECK-NEXT: v_mov_b32_e32 v13, v9 ; CHECK-NEXT: v_mov_b32_e32 v4, v2 ; CHECK-NEXT: v_mov_b32_e32 v10, v0 -; CHECK-NEXT: v_mov_b32_e32 v12, v6 +; CHECK-NEXT: v_mov_b32_e32 v12, v8 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB5_5: ; %memmove_fwd_main_loop @@ -799,20 +799,20 @@ define void @memmove_p1_p0(ptr addrspace(1) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: s_and_saveexec_b32 s8, s4 ; CHECK-NEXT: s_cbranch_execz .LBB5_9 ; CHECK-NEXT: ; %bb.7: ; %memmove_fwd_residual_loop.preheader -; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v7, s5 -; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v7, s5 +; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v9, s5 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB5_8: ; %memmove_fwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: flat_load_ubyte v4, v[2:3] -; CHECK-NEXT: v_add_co_u32 v8, s5, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v6, s5, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s5 ; CHECK-NEXT: v_add_co_u32 v2, s5, v2, 1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, s5 -; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s9, s5, s9 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: global_store_byte v[0:1], v4, off @@ -822,10 +822,10 @@ define void @memmove_p1_p0(ptr addrspace(1) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: s_cbranch_execnz .LBB5_8 ; CHECK-NEXT: .LBB5_9: ; %Flow30 ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s8 -; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 +; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1 ; CHECK-NEXT: ; implicit-def: $vgpr2_vgpr3 -; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 +; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 ; CHECK-NEXT: ; implicit-def: $vgpr4_vgpr5 ; CHECK-NEXT: s_andn2_saveexec_b32 s6, s7 ; CHECK-NEXT: s_cbranch_execz .LBB5_2 @@ -844,11 +844,11 @@ define void @memmove_p1_p0(ptr addrspace(1) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: .LBB5_12: ; %memmove_bwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: flat_load_ubyte v12, v[10:11] -; CHECK-NEXT: v_add_co_u32 v8, s4, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s4 +; CHECK-NEXT: v_add_co_u32 v6, s4, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s4 ; CHECK-NEXT: v_add_co_u32 v10, s4, v10, -1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, -1, v11, s4 -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s8, s4, s8 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CHECK-NEXT: global_store_byte v[4:5], v12, off @@ -869,19 +869,19 @@ define void @memmove_p1_p0(ptr addrspace(1) align 1 %dst, ptr addrspace(0) align ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB5_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v7, vcc_lo -; CHECK-NEXT: v_add_co_u32 v12, s4, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v13, null, v1, v7, s4 -; CHECK-NEXT: flat_load_dwordx4 v[8:11], v[4:5] -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v6, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v7, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[4:5] -; CHECK-NEXT: v_mov_b32_e32 v7, v5 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v11, v9 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v10 +; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v11, vcc_lo +; CHECK-NEXT: v_add_co_u32 v8, vcc_lo, v10, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v11, vcc_lo +; CHECK-NEXT: flat_load_dwordx4 v[4:7], v[4:5] +; CHECK-NEXT: v_add_co_u32 v10, s4, v0, v10 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[8:9] +; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, v1, v11, s4 ; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) -; CHECK-NEXT: global_store_dwordx4 v[12:13], v[8:11], off +; CHECK-NEXT: global_store_dwordx4 v[10:11], v[4:7], off ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB5_15 ; CHECK-NEXT: .LBB5_16: ; %Flow34 @@ -897,13 +897,13 @@ define void @memmove_p1_p1(ptr addrspace(1) align 1 %dst, ptr addrspace(1) align ; CHECK-LABEL: memmove_p1_p1: ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_and_b32_e32 v8, 15, v4 -; CHECK-NEXT: v_mov_b32_e32 v9, 0 -; CHECK-NEXT: v_and_b32_e32 v6, -16, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v5 +; CHECK-NEXT: v_and_b32_e32 v6, 15, v4 +; CHECK-NEXT: v_mov_b32_e32 v7, 0 +; CHECK-NEXT: v_and_b32_e32 v8, -16, v4 +; CHECK-NEXT: v_mov_b32_e32 v9, v5 ; CHECK-NEXT: s_mov_b32 s6, exec_lo -; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[8:9] -; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9] ; CHECK-NEXT: v_cmpx_ge_u64_e64 v[2:3], v[0:1] ; CHECK-NEXT: s_xor_b32 s7, exec_lo, s6 ; CHECK-NEXT: s_cbranch_execnz .LBB6_3 @@ -919,10 +919,10 @@ define void @memmove_p1_p1(ptr addrspace(1) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: ; %bb.4: ; %memmove_fwd_main_loop.preheader ; CHECK-NEXT: v_mov_b32_e32 v5, v3 ; CHECK-NEXT: v_mov_b32_e32 v11, v1 -; CHECK-NEXT: v_mov_b32_e32 v13, v7 +; CHECK-NEXT: v_mov_b32_e32 v13, v9 ; CHECK-NEXT: v_mov_b32_e32 v4, v2 ; CHECK-NEXT: v_mov_b32_e32 v10, v0 -; CHECK-NEXT: v_mov_b32_e32 v12, v6 +; CHECK-NEXT: v_mov_b32_e32 v12, v8 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB6_5: ; %memmove_fwd_main_loop @@ -945,20 +945,20 @@ define void @memmove_p1_p1(ptr addrspace(1) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: s_and_saveexec_b32 s8, s4 ; CHECK-NEXT: s_cbranch_execz .LBB6_9 ; CHECK-NEXT: ; %bb.7: ; %memmove_fwd_residual_loop.preheader -; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v7, s5 -; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v7, s5 +; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v9, s5 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB6_8: ; %memmove_fwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v4, v[2:3], off -; CHECK-NEXT: v_add_co_u32 v8, s5, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v6, s5, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s5 ; CHECK-NEXT: v_add_co_u32 v2, s5, v2, 1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, s5 -; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s9, s5, s9 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: global_store_byte v[0:1], v4, off @@ -968,10 +968,10 @@ define void @memmove_p1_p1(ptr addrspace(1) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: s_cbranch_execnz .LBB6_8 ; CHECK-NEXT: .LBB6_9: ; %Flow32 ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s8 -; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 +; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1 ; CHECK-NEXT: ; implicit-def: $vgpr2_vgpr3 -; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 +; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 ; CHECK-NEXT: ; implicit-def: $vgpr4_vgpr5 ; CHECK-NEXT: s_andn2_saveexec_b32 s6, s7 ; CHECK-NEXT: s_cbranch_execz .LBB6_2 @@ -990,11 +990,11 @@ define void @memmove_p1_p1(ptr addrspace(1) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: .LBB6_12: ; %memmove_bwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v12, v[10:11], off -; CHECK-NEXT: v_add_co_u32 v8, s4, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s4 +; CHECK-NEXT: v_add_co_u32 v6, s4, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s4 ; CHECK-NEXT: v_add_co_u32 v10, s4, v10, -1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, -1, v11, s4 -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s8, s4, s8 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: global_store_byte v[4:5], v12, off @@ -1015,19 +1015,19 @@ define void @memmove_p1_p1(ptr addrspace(1) align 1 %dst, ptr addrspace(1) align ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB6_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v7, vcc_lo -; CHECK-NEXT: v_add_co_u32 v12, s4, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v13, null, v1, v7, s4 -; CHECK-NEXT: global_load_dwordx4 v[8:11], v[4:5], off -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v6, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v7, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[4:5] -; CHECK-NEXT: v_mov_b32_e32 v7, v5 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v11, v9 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v10 +; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v11, vcc_lo +; CHECK-NEXT: v_add_co_u32 v8, vcc_lo, v10, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v11, vcc_lo +; CHECK-NEXT: global_load_dwordx4 v[4:7], v[4:5], off +; CHECK-NEXT: v_add_co_u32 v10, s4, v0, v10 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[8:9] +; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, v1, v11, s4 ; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) -; CHECK-NEXT: global_store_dwordx4 v[12:13], v[8:11], off +; CHECK-NEXT: global_store_dwordx4 v[10:11], v[4:7], off ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB6_15 ; CHECK-NEXT: .LBB6_16: ; %Flow36 @@ -1109,13 +1109,13 @@ define void @memmove_p1_p4(ptr addrspace(1) align 1 %dst, ptr addrspace(4) align ; CHECK-LABEL: memmove_p1_p4: ; CHECK: ; %bb.0: ; %entry ; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; CHECK-NEXT: v_and_b32_e32 v8, 15, v4 -; CHECK-NEXT: v_mov_b32_e32 v9, 0 -; CHECK-NEXT: v_and_b32_e32 v6, -16, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v5 +; CHECK-NEXT: v_and_b32_e32 v6, 15, v4 +; CHECK-NEXT: v_mov_b32_e32 v7, 0 +; CHECK-NEXT: v_and_b32_e32 v8, -16, v4 +; CHECK-NEXT: v_mov_b32_e32 v9, v5 ; CHECK-NEXT: s_mov_b32 s6, exec_lo -; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[8:9] -; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e64 s4, 0, v[6:7] +; CHECK-NEXT: v_cmp_ne_u64_e32 vcc_lo, 0, v[8:9] ; CHECK-NEXT: v_cmpx_ge_u64_e64 v[2:3], v[0:1] ; CHECK-NEXT: s_xor_b32 s7, exec_lo, s6 ; CHECK-NEXT: s_cbranch_execnz .LBB8_3 @@ -1131,10 +1131,10 @@ define void @memmove_p1_p4(ptr addrspace(1) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: ; %bb.4: ; %memmove_fwd_main_loop.preheader ; CHECK-NEXT: v_mov_b32_e32 v5, v3 ; CHECK-NEXT: v_mov_b32_e32 v11, v1 -; CHECK-NEXT: v_mov_b32_e32 v13, v7 +; CHECK-NEXT: v_mov_b32_e32 v13, v9 ; CHECK-NEXT: v_mov_b32_e32 v4, v2 ; CHECK-NEXT: v_mov_b32_e32 v10, v0 -; CHECK-NEXT: v_mov_b32_e32 v12, v6 +; CHECK-NEXT: v_mov_b32_e32 v12, v8 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB8_5: ; %memmove_fwd_main_loop @@ -1157,20 +1157,20 @@ define void @memmove_p1_p4(ptr addrspace(1) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: s_and_saveexec_b32 s8, s4 ; CHECK-NEXT: s_cbranch_execz .LBB8_9 ; CHECK-NEXT: ; %bb.7: ; %memmove_fwd_residual_loop.preheader -; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v7, s5 -; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v7, s5 +; CHECK-NEXT: v_add_co_u32 v0, s5, v0, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v1, null, v1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v2, s5, v2, v8 +; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, v3, v9, s5 ; CHECK-NEXT: s_mov_b32 s9, 0 ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB8_8: ; %memmove_fwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v4, v[2:3], off -; CHECK-NEXT: v_add_co_u32 v8, s5, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s5 +; CHECK-NEXT: v_add_co_u32 v6, s5, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s5 ; CHECK-NEXT: v_add_co_u32 v2, s5, v2, 1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v3, null, 0, v3, s5 -; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s5, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s9, s5, s9 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: global_store_byte v[0:1], v4, off @@ -1180,10 +1180,10 @@ define void @memmove_p1_p4(ptr addrspace(1) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: s_cbranch_execnz .LBB8_8 ; CHECK-NEXT: .LBB8_9: ; %Flow31 ; CHECK-NEXT: s_or_b32 exec_lo, exec_lo, s8 -; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 +; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1 ; CHECK-NEXT: ; implicit-def: $vgpr2_vgpr3 -; CHECK-NEXT: ; implicit-def: $vgpr8_vgpr9 +; CHECK-NEXT: ; implicit-def: $vgpr6_vgpr7 ; CHECK-NEXT: ; implicit-def: $vgpr4_vgpr5 ; CHECK-NEXT: s_andn2_saveexec_b32 s6, s7 ; CHECK-NEXT: s_cbranch_execz .LBB8_2 @@ -1202,11 +1202,11 @@ define void @memmove_p1_p4(ptr addrspace(1) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: .LBB8_12: ; %memmove_bwd_residual_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 ; CHECK-NEXT: global_load_ubyte v12, v[10:11], off -; CHECK-NEXT: v_add_co_u32 v8, s4, v8, -1 -; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v9, s4 +; CHECK-NEXT: v_add_co_u32 v6, s4, v6, -1 +; CHECK-NEXT: v_add_co_ci_u32_e64 v7, null, -1, v7, s4 ; CHECK-NEXT: v_add_co_u32 v10, s4, v10, -1 ; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, -1, v11, s4 -; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[8:9] +; CHECK-NEXT: v_cmp_eq_u64_e64 s4, 0, v[6:7] ; CHECK-NEXT: s_or_b32 s8, s4, s8 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: global_store_byte v[4:5], v12, off @@ -1227,19 +1227,19 @@ define void @memmove_p1_p4(ptr addrspace(1) align 1 %dst, ptr addrspace(4) align ; CHECK-NEXT: .p2align 6 ; CHECK-NEXT: .LBB8_15: ; %memmove_bwd_main_loop ; CHECK-NEXT: ; =>This Inner Loop Header: Depth=1 -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v7, vcc_lo -; CHECK-NEXT: v_add_co_u32 v12, s4, v0, v6 -; CHECK-NEXT: v_add_co_ci_u32_e64 v13, null, v1, v7, s4 -; CHECK-NEXT: global_load_dwordx4 v[8:11], v[4:5], off -; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v6, -16 -; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, -1, v7, vcc_lo -; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[4:5] -; CHECK-NEXT: v_mov_b32_e32 v7, v5 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 +; CHECK-NEXT: v_mov_b32_e32 v11, v9 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_add_co_u32 v4, vcc_lo, v2, v10 +; CHECK-NEXT: v_add_co_ci_u32_e64 v5, null, v3, v11, vcc_lo +; CHECK-NEXT: v_add_co_u32 v8, vcc_lo, v10, -16 +; CHECK-NEXT: v_add_co_ci_u32_e64 v9, null, -1, v11, vcc_lo +; CHECK-NEXT: global_load_dwordx4 v[4:7], v[4:5], off +; CHECK-NEXT: v_add_co_u32 v10, s4, v0, v10 +; CHECK-NEXT: v_cmp_eq_u64_e32 vcc_lo, 0, v[8:9] +; CHECK-NEXT: v_add_co_ci_u32_e64 v11, null, v1, v11, s4 ; CHECK-NEXT: s_or_b32 s7, vcc_lo, s7 ; CHECK-NEXT: s_waitcnt vmcnt(0) -; CHECK-NEXT: global_store_dwordx4 v[12:13], v[8:11], off +; CHECK-NEXT: global_store_dwordx4 v[10:11], v[4:7], off ; CHECK-NEXT: s_andn2_b32 exec_lo, exec_lo, s7 ; CHECK-NEXT: s_cbranch_execnz .LBB8_15 ; CHECK-NEXT: .LBB8_16: ; %Flow35 diff --git a/llvm/test/CodeGen/AMDGPU/mul_int24.ll b/llvm/test/CodeGen/AMDGPU/mul_int24.ll index 3d9c2a2..10d4eb0 100644 --- a/llvm/test/CodeGen/AMDGPU/mul_int24.ll +++ b/llvm/test/CodeGen/AMDGPU/mul_int24.ll @@ -10,46 +10,43 @@ define amdgpu_kernel void @test_smul24_i32(ptr addrspace(1) %out, i32 %a, i32 %b ; SI-LABEL: test_smul24_i32: ; SI: ; %bb.0: ; %entry ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_bfe_i32 s2, s2, 0x180000 -; SI-NEXT: s_bfe_i32 s3, s3, 0x180000 -; SI-NEXT: s_mul_i32 s2, s2, s3 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_bfe_i32 s2, s4, 0x180000 +; SI-NEXT: s_bfe_i32 s4, s5, 0x180000 +; SI-NEXT: s_mul_i32 s4, s2, s4 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: test_smul24_i32: ; VI: ; %bb.0: ; %entry ; VI-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_bfe_i32 s0, s2, 0x180000 -; VI-NEXT: s_bfe_i32 s1, s3, 0x180000 -; VI-NEXT: s_mul_i32 s0, s0, s1 -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_bfe_i32 s4, s4, 0x180000 +; VI-NEXT: s_bfe_i32 s5, s5, 0x180000 +; VI-NEXT: s_mul_i32 s4, s4, s5 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: test_smul24_i32: ; GFX9: ; %bb.0: ; %entry ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; GFX9-NEXT: s_mov_b32 s7, 0xf000 -; GFX9-NEXT: s_mov_b32 s6, -1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_mov_b32 s4, s0 -; GFX9-NEXT: s_mov_b32 s5, s1 -; GFX9-NEXT: s_bfe_i32 s0, s2, 0x180000 -; GFX9-NEXT: s_bfe_i32 s1, s3, 0x180000 -; GFX9-NEXT: s_mul_i32 s0, s0, s1 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX9-NEXT: s_bfe_i32 s4, s4, 0x180000 +; GFX9-NEXT: s_bfe_i32 s5, s5, 0x180000 +; GFX9-NEXT: s_mul_i32 s4, s4, s5 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX9-NEXT: s_endpgm ; ; EG-LABEL: test_smul24_i32: @@ -127,16 +124,15 @@ define amdgpu_kernel void @test_smulhi24_i64(ptr addrspace(1) %out, i32 %a, i32 ; GFX9-LABEL: test_smulhi24_i64: ; GFX9: ; %bb.0: ; %entry ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; GFX9-NEXT: s_mov_b32 s7, 0xf000 -; GFX9-NEXT: s_mov_b32 s6, -1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_mov_b32 s4, s0 -; GFX9-NEXT: s_mov_b32 s5, s1 -; GFX9-NEXT: s_bfe_i32 s0, s2, 0x180000 -; GFX9-NEXT: s_bfe_i32 s1, s3, 0x180000 -; GFX9-NEXT: s_mul_hi_i32 s0, s0, s1 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX9-NEXT: s_bfe_i32 s4, s4, 0x180000 +; GFX9-NEXT: s_bfe_i32 s5, s5, 0x180000 +; GFX9-NEXT: s_mul_hi_i32 s4, s4, s5 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX9-NEXT: s_endpgm ; ; EG-LABEL: test_smulhi24_i64: @@ -464,29 +460,26 @@ define amdgpu_kernel void @test_smul24_i33(ptr addrspace(1) %out, i33 %a, i33 %b ; SI-LABEL: test_smul24_i33: ; SI: ; %bb.0: ; %entry ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0xd -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 +; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_bfe_i32 s0, s8, 0x180000 -; SI-NEXT: s_bfe_i32 s1, s2, 0x180000 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: s_mul_i32 s0, s1, s0 -; SI-NEXT: v_mul_hi_i32_i24_e32 v1, s1, v0 -; SI-NEXT: v_mov_b32_e32 v0, s0 +; SI-NEXT: s_mov_b64 s[6:7], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: s_bfe_i32 s4, s4, 0x180000 +; SI-NEXT: s_bfe_i32 s5, s6, 0x180000 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: s_mul_i32 s4, s5, s4 +; SI-NEXT: v_mul_hi_i32_i24_e32 v1, s5, v0 +; SI-NEXT: v_mov_b32_e32 v0, s4 ; SI-NEXT: v_lshl_b64 v[0:1], v[0:1], 31 ; SI-NEXT: v_ashr_i64 v[0:1], v[0:1], 31 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: test_smul24_i33: ; VI: ; %bb.0: ; %entry ; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 ; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34 -; 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 s2, s2, 0x180000 ; VI-NEXT: s_bfe_i32 s3, s4, 0x180000 @@ -494,10 +487,10 @@ define amdgpu_kernel void @test_smul24_i33(ptr addrspace(1) %out, i33 %a, i33 %b ; VI-NEXT: v_mul_hi_i32_i24_e32 v1, s2, v0 ; VI-NEXT: v_mul_i32_i24_e32 v0, s2, v0 ; VI-NEXT: v_lshlrev_b64 v[0:1], 31, v[0:1] -; VI-NEXT: s_mov_b32 s4, s0 +; VI-NEXT: s_mov_b32 s3, 0xf000 ; VI-NEXT: v_ashrrev_i64 v[0:1], 31, v[0:1] -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: test_smul24_i33: @@ -577,31 +570,29 @@ define amdgpu_kernel void @test_smulhi24_i33(ptr addrspace(1) %out, i33 %a, i33 ; SI-LABEL: test_smulhi24_i33: ; SI: ; %bb.0: ; %entry ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0xd -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 +; SI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s8 -; SI-NEXT: v_mul_hi_i32_i24_e32 v0, s2, v0 +; SI-NEXT: s_mov_b64 s[6:7], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: v_mul_hi_i32_i24_e32 v0, s6, v0 ; SI-NEXT: v_and_b32_e32 v0, 1, v0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: test_smulhi24_i33: ; VI: ; %bb.0: ; %entry ; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; VI-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0x34 -; VI-NEXT: s_mov_b32 s7, 0xf000 -; VI-NEXT: s_mov_b32 s6, -1 +; VI-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x34 ; VI-NEXT: s_waitcnt lgkmcnt(0) -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: v_mov_b32_e32 v0, s8 -; VI-NEXT: v_mul_hi_i32_i24_e32 v0, s2, v0 -; VI-NEXT: s_mov_b32 s5, s1 +; VI-NEXT: s_mov_b64 s[6:7], s[2:3] +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: v_mul_hi_i32_i24_e32 v0, s6, v0 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 ; VI-NEXT: v_and_b32_e32 v0, 1, v0 -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: test_smulhi24_i33: diff --git a/llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll b/llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll index e29da3a..1165401 100644 --- a/llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll +++ b/llvm/test/CodeGen/AMDGPU/mul_uint24-amdgcn.ll @@ -10,46 +10,43 @@ define amdgpu_kernel void @test_umul24_i32(ptr addrspace(1) %out, i32 %a, i32 %b ; SI-LABEL: test_umul24_i32: ; SI: ; %bb.0: ; %entry ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_and_b32 s2, s2, 0xffffff -; SI-NEXT: s_and_b32 s3, s3, 0xffffff -; SI-NEXT: s_mul_i32 s2, s2, s3 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_and_b32 s2, s4, 0xffffff +; SI-NEXT: s_and_b32 s4, s5, 0xffffff +; SI-NEXT: s_mul_i32 s4, s2, s4 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: test_umul24_i32: ; VI: ; %bb.0: ; %entry ; VI-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_and_b32 s0, s2, 0xffffff -; VI-NEXT: s_and_b32 s1, s3, 0xffffff -; VI-NEXT: s_mul_i32 s0, s0, s1 -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_and_b32 s4, s4, 0xffffff +; VI-NEXT: s_and_b32 s5, s5, 0xffffff +; VI-NEXT: s_mul_i32 s4, s4, s5 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: test_umul24_i32: ; GFX9: ; %bb.0: ; %entry ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; GFX9-NEXT: s_mov_b32 s7, 0xf000 -; GFX9-NEXT: s_mov_b32 s6, -1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_mov_b32 s4, s0 -; GFX9-NEXT: s_mov_b32 s5, s1 -; GFX9-NEXT: s_and_b32 s0, s2, 0xffffff -; GFX9-NEXT: s_and_b32 s1, s3, 0xffffff -; GFX9-NEXT: s_mul_i32 s0, s0, s1 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX9-NEXT: s_and_b32 s4, s4, 0xffffff +; GFX9-NEXT: s_and_b32 s5, s5, 0xffffff +; GFX9-NEXT: s_mul_i32 s4, s4, s5 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX9-NEXT: s_endpgm entry: %0 = shl i32 %a, 8 @@ -406,16 +403,15 @@ define amdgpu_kernel void @test_umulhi24_i32_i64(ptr addrspace(1) %out, i32 %a, ; GFX9-LABEL: test_umulhi24_i32_i64: ; GFX9: ; %bb.0: ; %entry ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; GFX9-NEXT: s_mov_b32 s7, 0xf000 -; GFX9-NEXT: s_mov_b32 s6, -1 ; GFX9-NEXT: s_waitcnt lgkmcnt(0) -; GFX9-NEXT: s_mov_b32 s4, s0 -; GFX9-NEXT: s_mov_b32 s5, s1 -; GFX9-NEXT: s_and_b32 s0, s2, 0xffffff -; GFX9-NEXT: s_and_b32 s1, s3, 0xffffff -; GFX9-NEXT: s_mul_hi_u32 s0, s0, s1 -; GFX9-NEXT: v_mov_b32_e32 v0, s0 -; GFX9-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX9-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX9-NEXT: s_and_b32 s4, s4, 0xffffff +; GFX9-NEXT: s_and_b32 s5, s5, 0xffffff +; GFX9-NEXT: s_mul_hi_u32 s4, s4, s5 +; GFX9-NEXT: s_mov_b32 s3, 0xf000 +; GFX9-NEXT: s_mov_b32 s2, -1 +; GFX9-NEXT: v_mov_b32_e32 v0, s4 +; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX9-NEXT: s_endpgm entry: %a.24 = and i32 %a, 16777215 @@ -632,33 +628,31 @@ define amdgpu_kernel void @test_umulhi16_i32(ptr addrspace(1) %out, i32 %a, i32 ; SI-LABEL: test_umulhi16_i32: ; SI: ; %bb.0: ; %entry ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_and_b32 s2, s2, 0xffff -; SI-NEXT: s_and_b32 s3, s3, 0xffff -; SI-NEXT: s_mul_i32 s2, s2, s3 -; SI-NEXT: s_lshr_b32 s2, s2, 16 -; SI-NEXT: s_mov_b32 s6, -1 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s2 -; SI-NEXT: buffer_store_short v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_and_b32 s2, s4, 0xffff +; SI-NEXT: s_and_b32 s4, s5, 0xffff +; SI-NEXT: s_mul_i32 s2, s2, s4 +; SI-NEXT: s_lshr_b32 s4, s2, 16 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_short v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: test_umulhi16_i32: ; VI: ; %bb.0: ; %entry ; VI-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_and_b32 s0, s2, 0xffff -; VI-NEXT: s_and_b32 s1, s3, 0xffff -; VI-NEXT: s_mul_i32 s0, s0, s1 -; VI-NEXT: s_lshr_b32 s0, s0, 16 -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: buffer_store_short v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_and_b32 s4, s4, 0xffff +; VI-NEXT: s_and_b32 s5, s5, 0xffff +; VI-NEXT: s_mul_i32 s4, s4, s5 +; VI-NEXT: s_lshr_b32 s4, s4, 16 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_short v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GFX9-LABEL: test_umulhi16_i32: diff --git a/llvm/test/CodeGen/AMDGPU/nofpclass-call.ll b/llvm/test/CodeGen/AMDGPU/nofpclass-call.ll new file mode 100644 index 0000000..5f303cc --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/nofpclass-call.ll @@ -0,0 +1,191 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s + +; Check that nofpclass attributes on call returns are used in +; selectiondag. + +define internal float @func_f32(ptr addrspace(1) %ptr) { +; CHECK-LABEL: func_f32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: global_load_dword v0, v[0:1], off glc +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %ld = load volatile float, ptr addrspace(1) %ptr + ret float %ld +} + +define float @call_nofpclass_funcs_f32(ptr addrspace(1) %ptr) { +; CHECK-LABEL: call_nofpclass_funcs_f32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: s_mov_b32 s18, s33 +; CHECK-NEXT: s_mov_b32 s33, s32 +; CHECK-NEXT: s_xor_saveexec_b64 s[16:17], -1 +; CHECK-NEXT: buffer_store_dword v4, off, s[0:3], s33 ; 4-byte Folded Spill +; CHECK-NEXT: s_mov_b64 exec, s[16:17] +; CHECK-NEXT: s_addk_i32 s32, 0x400 +; CHECK-NEXT: v_writelane_b32 v4, s30, 0 +; CHECK-NEXT: s_getpc_b64 s[16:17] +; CHECK-NEXT: s_add_u32 s16, s16, func_f32@rel32@lo+4 +; CHECK-NEXT: s_addc_u32 s17, s17, func_f32@rel32@hi+12 +; CHECK-NEXT: v_writelane_b32 v4, s31, 1 +; CHECK-NEXT: v_mov_b32_e32 v2, v0 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] +; CHECK-NEXT: v_mov_b32_e32 v3, v0 +; CHECK-NEXT: v_mov_b32_e32 v0, v2 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] +; CHECK-NEXT: v_min_f32_e32 v0, v3, v0 +; CHECK-NEXT: v_readlane_b32 s31, v4, 1 +; CHECK-NEXT: v_readlane_b32 s30, v4, 0 +; CHECK-NEXT: s_mov_b32 s32, s33 +; CHECK-NEXT: s_xor_saveexec_b64 s[4:5], -1 +; CHECK-NEXT: buffer_load_dword v4, off, s[0:3], s33 ; 4-byte Folded Reload +; CHECK-NEXT: s_mov_b64 exec, s[4:5] +; CHECK-NEXT: s_mov_b32 s33, s18 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %call0 = call nofpclass(nan) float @func_f32(ptr addrspace(1) %ptr) + %call1 = call nofpclass(nan) float @func_f32(ptr addrspace(1) %ptr) + %min = call float @llvm.minnum.f32(float %call0, float %call1) + ret float %min +} + +define internal <2 x float> @func_v2f32(ptr addrspace(1) %ptr) { +; CHECK-LABEL: func_v2f32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: global_load_dwordx2 v[0:1], v[0:1], off glc +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %ld = load volatile <2 x float>, ptr addrspace(1) %ptr + ret <2 x float> %ld +} + +define <2 x float> @call_nofpclass_funcs_v2f32(ptr addrspace(1) %ptr) { +; CHECK-LABEL: call_nofpclass_funcs_v2f32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: s_mov_b32 s18, s33 +; CHECK-NEXT: s_mov_b32 s33, s32 +; CHECK-NEXT: s_xor_saveexec_b64 s[16:17], -1 +; CHECK-NEXT: buffer_store_dword v6, off, s[0:3], s33 ; 4-byte Folded Spill +; CHECK-NEXT: s_mov_b64 exec, s[16:17] +; CHECK-NEXT: s_addk_i32 s32, 0x400 +; CHECK-NEXT: v_writelane_b32 v6, s30, 0 +; CHECK-NEXT: s_getpc_b64 s[16:17] +; CHECK-NEXT: s_add_u32 s16, s16, func_v2f32@rel32@lo+4 +; CHECK-NEXT: s_addc_u32 s17, s17, func_v2f32@rel32@hi+12 +; CHECK-NEXT: v_writelane_b32 v6, s31, 1 +; CHECK-NEXT: v_mov_b32_e32 v2, v1 +; CHECK-NEXT: v_mov_b32_e32 v3, v0 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] +; CHECK-NEXT: v_mov_b32_e32 v4, v0 +; CHECK-NEXT: v_mov_b32_e32 v5, v1 +; CHECK-NEXT: v_mov_b32_e32 v0, v3 +; CHECK-NEXT: v_mov_b32_e32 v1, v2 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] +; CHECK-NEXT: v_min_f32_e32 v0, v4, v0 +; CHECK-NEXT: v_min_f32_e32 v1, v5, v1 +; CHECK-NEXT: v_readlane_b32 s31, v6, 1 +; CHECK-NEXT: v_readlane_b32 s30, v6, 0 +; CHECK-NEXT: s_mov_b32 s32, s33 +; CHECK-NEXT: s_xor_saveexec_b64 s[4:5], -1 +; CHECK-NEXT: buffer_load_dword v6, off, s[0:3], s33 ; 4-byte Folded Reload +; CHECK-NEXT: s_mov_b64 exec, s[4:5] +; CHECK-NEXT: s_mov_b32 s33, s18 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %call0 = call nofpclass(nan) <2 x float> @func_v2f32(ptr addrspace(1) %ptr) + %call1 = call nofpclass(nan) <2 x float> @func_v2f32(ptr addrspace(1) %ptr) + %min = call <2 x float> @llvm.minnum.v2f32(<2 x float> %call0, <2 x float> %call1) + ret <2 x float> %min +} + +define internal double @func_f64(ptr addrspace(1) %ptr) { +; CHECK-LABEL: func_f64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: global_load_dwordx2 v[0:1], v[0:1], off glc +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %ld = load volatile double, ptr addrspace(1) %ptr + ret double %ld +} + +define double @call_nofpclass_funcs_f64(ptr addrspace(1) %ptr) { +; CHECK-LABEL: call_nofpclass_funcs_f64: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: s_mov_b32 s18, s33 +; CHECK-NEXT: s_mov_b32 s33, s32 +; CHECK-NEXT: s_xor_saveexec_b64 s[16:17], -1 +; CHECK-NEXT: buffer_store_dword v6, off, s[0:3], s33 ; 4-byte Folded Spill +; CHECK-NEXT: s_mov_b64 exec, s[16:17] +; CHECK-NEXT: s_addk_i32 s32, 0x400 +; CHECK-NEXT: v_writelane_b32 v6, s30, 0 +; CHECK-NEXT: s_getpc_b64 s[16:17] +; CHECK-NEXT: s_add_u32 s16, s16, func_f64@rel32@lo+4 +; CHECK-NEXT: s_addc_u32 s17, s17, func_f64@rel32@hi+12 +; CHECK-NEXT: v_writelane_b32 v6, s31, 1 +; CHECK-NEXT: v_mov_b32_e32 v4, v1 +; CHECK-NEXT: v_mov_b32_e32 v5, v0 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] +; CHECK-NEXT: v_mov_b32_e32 v2, v0 +; CHECK-NEXT: v_mov_b32_e32 v3, v1 +; CHECK-NEXT: v_mov_b32_e32 v0, v5 +; CHECK-NEXT: v_mov_b32_e32 v1, v4 +; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17] +; CHECK-NEXT: v_min_f64 v[0:1], v[2:3], v[0:1] +; CHECK-NEXT: v_readlane_b32 s31, v6, 1 +; CHECK-NEXT: v_readlane_b32 s30, v6, 0 +; CHECK-NEXT: s_mov_b32 s32, s33 +; CHECK-NEXT: s_xor_saveexec_b64 s[4:5], -1 +; CHECK-NEXT: buffer_load_dword v6, off, s[0:3], s33 ; 4-byte Folded Reload +; CHECK-NEXT: s_mov_b64 exec, s[4:5] +; CHECK-NEXT: s_mov_b32 s33, s18 +; CHECK-NEXT: s_waitcnt vmcnt(0) +; CHECK-NEXT: s_setpc_b64 s[30:31] + %call0 = call nofpclass(nan) double @func_f64(ptr addrspace(1) %ptr) + %call1 = call nofpclass(nan) double @func_f64(ptr addrspace(1) %ptr) + %min = call double @llvm.minnum.f64(double %call0, double %call1) + ret double %min +} + +define float @call_nofpclass_intrinsic_f32(float %x, float %y, float %z) { +; CHECK-LABEL: call_nofpclass_intrinsic_f32: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: v_sqrt_f32_e32 v0, v0 +; CHECK-NEXT: v_sqrt_f32_e32 v1, v1 +; CHECK-NEXT: v_cmp_lt_f32_e32 vcc, v0, v1 +; CHECK-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; CHECK-NEXT: s_setpc_b64 s[30:31] + %call0 = call nofpclass(nan) float @llvm.amdgcn.sqrt.f32(float %x) + %call1 = call nofpclass(nan) float @llvm.amdgcn.sqrt.f32(float %y) + %lt = fcmp olt float %call0, %call1 + %min = select nsz i1 %lt, float %call0, float %call1 + ret float %min +} + +define <2 x half> @call_nofpclass_intrinsic_v2f16(float %x, float %y, float %z, float %w) { +; CHECK-LABEL: call_nofpclass_intrinsic_v2f16: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: v_cvt_pkrtz_f16_f32 v0, v0, v1 +; CHECK-NEXT: v_cvt_pkrtz_f16_f32 v1, v2, v3 +; CHECK-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; CHECK-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; CHECK-NEXT: v_cmp_lt_f16_e32 vcc, v0, v1 +; CHECK-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; CHECK-NEXT: v_cmp_lt_f16_e32 vcc, v3, v2 +; CHECK-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc +; CHECK-NEXT: s_mov_b32 s4, 0x5040100 +; CHECK-NEXT: v_perm_b32 v0, v1, v0, s4 +; CHECK-NEXT: s_setpc_b64 s[30:31] + %call0 = call nofpclass(nan) <2 x half> @llvm.amdgcn.cvt.pkrtz(float %x, float %y) + %call1 = call nofpclass(nan) <2 x half> @llvm.amdgcn.cvt.pkrtz(float %z, float %w) + %lt = fcmp olt <2 x half> %call0, %call1 + %min = select nsz <2 x i1> %lt, <2 x half> %call0, <2 x half> %call1 + ret <2 x half> %min +} diff --git a/llvm/test/CodeGen/AMDGPU/or.ll b/llvm/test/CodeGen/AMDGPU/or.ll index 728067e..9afaab5 100644 --- a/llvm/test/CodeGen/AMDGPU/or.ll +++ b/llvm/test/CodeGen/AMDGPU/or.ll @@ -136,27 +136,25 @@ define amdgpu_kernel void @scalar_or_i32(ptr addrspace(1) %out, i32 %a, i32 %b) ; GFX6-LABEL: scalar_or_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_or_b32 s0, s2, s3 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_or_b32 s4, s4, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX8-LABEL: scalar_or_i32: ; GFX8: ; %bb.0: ; GFX8-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; GFX8-NEXT: s_or_b32 s0, s2, s3 -; 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_mov_b64 s[4:5], s[2:3] +; GFX8-NEXT: s_or_b32 s4, s4, s5 +; GFX8-NEXT: s_mov_b32 s3, 0xf000 +; GFX8-NEXT: s_mov_b32 s2, -1 +; GFX8-NEXT: v_mov_b32_e32 v0, s4 +; GFX8-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX8-NEXT: s_endpgm %or = or i32 %a, %b store i32 %or, ptr addrspace(1) %out diff --git a/llvm/test/CodeGen/AMDGPU/reg-coalescer-subreg-liveness.mir b/llvm/test/CodeGen/AMDGPU/reg-coalescer-subreg-liveness.mir index 381cb8c..f098618 100644 --- a/llvm/test/CodeGen/AMDGPU/reg-coalescer-subreg-liveness.mir +++ b/llvm/test/CodeGen/AMDGPU/reg-coalescer-subreg-liveness.mir @@ -16,11 +16,11 @@ body: | ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5 ; CHECK-NEXT: undef [[S_LOAD_DWORD_IMM:%[0-9]+]].sub1:sgpr_128 = S_LOAD_DWORD_IMM [[COPY]], 0, 0 :: (dereferenceable invariant load (s32), align 16, addrspace 4) ; CHECK-NEXT: undef [[S_MOV_B32_:%[0-9]+]].sub0:sgpr_128 = S_MOV_B32 1 + ; CHECK-NEXT: [[S_LOAD_DWORD_IMM:%[0-9]+]].sub2:sgpr_128 = S_MOV_B32 0 ; CHECK-NEXT: undef [[S_MOV_B32_1:%[0-9]+]].sub0:sgpr_256 = S_MOV_B32 0 ; CHECK-NEXT: TENSOR_LOAD_TO_LDS_D2 [[S_MOV_B32_]], [[S_MOV_B32_1]], 0, 0, implicit-def dead $tensorcnt, implicit $exec, implicit $tensorcnt ; CHECK-NEXT: [[S_LOAD_DWORD_IMM:%[0-9]+]].sub0:sgpr_128 = S_MOV_B32 1 ; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]].sub1:sgpr_128 = COPY [[S_MOV_B32_]].sub0 - ; CHECK-NEXT: [[S_LOAD_DWORD_IMM:%[0-9]+]].sub2:sgpr_128 = S_MOV_B32 0 ; CHECK-NEXT: {{ $}} ; CHECK-NEXT: bb.1: ; CHECK-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000) diff --git a/llvm/test/CodeGen/AMDGPU/remat-fp64-constants.ll b/llvm/test/CodeGen/AMDGPU/remat-fp64-constants.ll index c552f9d..88a51e9 100644 --- a/llvm/test/CodeGen/AMDGPU/remat-fp64-constants.ll +++ b/llvm/test/CodeGen/AMDGPU/remat-fp64-constants.ll @@ -1,10 +1,13 @@ ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 --stress-regalloc=10 < %s | FileCheck -check-prefix=GCN %s ; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 --stress-regalloc=10 < %s | FileCheck -check-prefix=GCN %s +; Rematerialization test for fp64 constants (w/ intentionally high register pressure). +; Check to make sure we have at least six constant MOVs, not necessarily consecutive, inside the loop. + ; GCN-LABEL: {{^}}test_remat_sgpr: ; GCN-NOT: v_writelane_b32 -; GCN-COUNT-4: s_mov_b32 s{{[0-9]+}}, 0x ; GCN: {{^}}[[LOOP:.LBB[0-9_]+]]: +; GCN-COUNT-6: {{s_mov_b32|v_mov_b32_e32}} {{[sv]}}{{[0-9]+}}, 0x ; GCN-NOT: v_writelane_b32 ; GCN: s_cbranch_{{[^ ]+}} [[LOOP]] ; GCN: .sgpr_spill_count: 0 diff --git a/llvm/test/CodeGen/AMDGPU/set-inactive-wwm-overwrite.ll b/llvm/test/CodeGen/AMDGPU/set-inactive-wwm-overwrite.ll index 5c90957..bcece19 100644 --- a/llvm/test/CodeGen/AMDGPU/set-inactive-wwm-overwrite.ll +++ b/llvm/test/CodeGen/AMDGPU/set-inactive-wwm-overwrite.ll @@ -16,11 +16,11 @@ define amdgpu_cs void @if_then(ptr addrspace(8) inreg %input, ptr addrspace(8) i ; GCN-NEXT: s_cbranch_execz .LBB0_4 ; GCN-NEXT: ; %bb.3: ; %.then ; GCN-NEXT: s_or_saveexec_b32 s1, -1 -; GCN-NEXT: v_cndmask_b32_e64 v1, 0, v3, s1 -; GCN-NEXT: v_mov_b32_e32 v2, 0 -; GCN-NEXT: v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf +; GCN-NEXT: v_mov_b32_e32 v1, 0 +; GCN-NEXT: v_cndmask_b32_e64 v2, 0, v3, s1 +; GCN-NEXT: v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf ; GCN-NEXT: s_mov_b32 exec_lo, s1 -; GCN-NEXT: v_mov_b32_e32 v0, v2 +; GCN-NEXT: v_mov_b32_e32 v0, v1 ; GCN-NEXT: v_mov_b32_e32 v4, -1 ; GCN-NEXT: v_mov_b32_e32 v3, 0 ; GCN-NEXT: buffer_store_dword v4, v0, s[4:7], 0 offen diff --git a/llvm/test/CodeGen/AMDGPU/sext-divergence-driven-isel.ll b/llvm/test/CodeGen/AMDGPU/sext-divergence-driven-isel.ll index a0bac53..e589a63 100644 --- a/llvm/test/CodeGen/AMDGPU/sext-divergence-driven-isel.ll +++ b/llvm/test/CodeGen/AMDGPU/sext-divergence-driven-isel.ll @@ -5,15 +5,14 @@ define amdgpu_kernel void @sext_i16_to_i32_uniform(ptr addrspace(1) %out, i16 %a ; GCN-LABEL: sext_i16_to_i32_uniform: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s4, s0 -; GCN-NEXT: s_sext_i32_i16 s0, s2 -; GCN-NEXT: s_add_i32 s0, s3, s0 -; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_sext_i32_i16 s4, s4 +; GCN-NEXT: s_add_i32 s4, s5, s4 +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; 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 %sext = sext i16 %a to i32 %res = add i32 %b, %sext diff --git a/llvm/test/CodeGen/AMDGPU/shl.v2i16.ll b/llvm/test/CodeGen/AMDGPU/shl.v2i16.ll index d8511c8..17db379 100644 --- a/llvm/test/CodeGen/AMDGPU/shl.v2i16.ll +++ b/llvm/test/CodeGen/AMDGPU/shl.v2i16.ll @@ -22,63 +22,57 @@ define amdgpu_kernel void @s_shl_v2i16(ptr addrspace(1) %out, <2 x i16> %lhs, <2 ; VI-LABEL: s_shl_v2i16: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_lshr_b32 s0, s3, 16 -; VI-NEXT: s_lshr_b32 s1, s2, 16 -; VI-NEXT: s_lshl_b32 s0, s1, s0 -; VI-NEXT: s_lshl_b32 s1, s2, s3 -; VI-NEXT: s_lshl_b32 s0, s0, 16 -; VI-NEXT: s_and_b32 s1, s1, 0xffff -; VI-NEXT: s_or_b32 s0, s1, s0 -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_lshr_b32 s6, s5, 16 +; VI-NEXT: s_lshr_b32 s7, s4, 16 +; VI-NEXT: s_lshl_b32 s4, s4, s5 +; VI-NEXT: s_lshl_b32 s5, s7, s6 +; VI-NEXT: s_lshl_b32 s5, s5, 16 +; VI-NEXT: s_and_b32 s4, s4, 0xffff +; VI-NEXT: s_or_b32 s4, s4, s5 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; CI-LABEL: s_shl_v2i16: ; CI: ; %bb.0: ; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; CI-NEXT: s_mov_b32 s7, 0xf000 -; CI-NEXT: s_mov_b32 s6, -1 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_mov_b32 s4, s0 -; CI-NEXT: s_mov_b32 s5, s1 -; CI-NEXT: s_lshr_b32 s0, s2, 16 -; CI-NEXT: s_lshr_b32 s1, s3, 16 -; CI-NEXT: s_lshl_b32 s0, s0, s1 -; CI-NEXT: s_lshl_b32 s1, s2, s3 -; CI-NEXT: s_lshl_b32 s0, s0, 16 -; CI-NEXT: s_and_b32 s1, s1, 0xffff -; CI-NEXT: s_or_b32 s0, s1, s0 -; CI-NEXT: v_mov_b32_e32 v0, s0 -; CI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; CI-NEXT: s_mov_b64 s[4:5], s[2:3] +; CI-NEXT: s_lshr_b32 s6, s4, 16 +; CI-NEXT: s_lshr_b32 s7, s5, 16 +; CI-NEXT: s_lshl_b32 s4, s4, s5 +; CI-NEXT: s_lshl_b32 s5, s6, s7 +; CI-NEXT: s_lshl_b32 s5, s5, 16 +; CI-NEXT: s_and_b32 s4, s4, 0xffff +; CI-NEXT: s_or_b32 s4, s4, s5 +; CI-NEXT: s_mov_b32 s3, 0xf000 +; CI-NEXT: s_mov_b32 s2, -1 +; CI-NEXT: v_mov_b32_e32 v0, s4 +; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; CI-NEXT: s_endpgm ; ; GFX10-LABEL: s_shl_v2i16: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; GFX10-NEXT: s_mov_b32 s7, 0x31016000 -; GFX10-NEXT: s_mov_b32 s6, -1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: v_pk_lshlrev_b16 v0, s3, s2 -; GFX10-NEXT: s_mov_b32 s4, s0 -; GFX10-NEXT: s_mov_b32 s5, s1 -; GFX10-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX10-NEXT: s_mov_b32 s3, 0x31016000 +; GFX10-NEXT: s_mov_b32 s2, -1 +; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX10-NEXT: s_endpgm ; ; GFX11-LABEL: s_shl_v2i16: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GFX11-NEXT: s_mov_b32 s7, 0x31016000 -; GFX11-NEXT: s_mov_b32 s6, -1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: v_pk_lshlrev_b16 v0, s3, s2 -; GFX11-NEXT: s_mov_b32 s4, s0 -; GFX11-NEXT: s_mov_b32 s5, s1 -; GFX11-NEXT: buffer_store_b32 v0, off, s[4:7], 0 +; GFX11-NEXT: s_mov_b32 s3, 0x31016000 +; GFX11-NEXT: s_mov_b32 s2, -1 +; GFX11-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-NEXT: s_endpgm %result = shl <2 x i16> %lhs, %rhs store <2 x i16> %result, ptr addrspace(1) %out diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v3f32.ll b/llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v3f32.ll index d4ee6fa..7c841783 100644 --- a/llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v3f32.ll +++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v3f32.ll @@ -3272,9 +3272,8 @@ define void @v_shuffle_v4f32_v3f32__1_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v4, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -3287,8 +3286,7 @@ define void @v_shuffle_v4f32_v3f32__1_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v4, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -3416,12 +3414,11 @@ define void @v_shuffle_v4f32_v3f32__4_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v7, 0 ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[4:6] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v7, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -3434,12 +3431,12 @@ define void @v_shuffle_v4f32_v3f32__4_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v7, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 ; GFX942-NEXT: ;;#ASMSTART ; GFX942-NEXT: ; def v[4:6] ; GFX942-NEXT: ;;#ASMEND +; GFX942-NEXT: s_nop 0 +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[0:1] op_sel:[1,0] ; GFX942-NEXT: global_store_dwordx4 v7, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: s_setpc_b64 s[30:31] @@ -6083,9 +6080,8 @@ define void @v_shuffle_v4f32_v3f32__1_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v5, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v5, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6102,8 +6098,7 @@ define void @v_shuffle_v4f32_v3f32__1_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -6241,9 +6236,8 @@ define void @v_shuffle_v4f32_v3f32__4_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v4, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6256,8 +6250,7 @@ define void @v_shuffle_v4f32_v3f32__4_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v4, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v3i32.ll b/llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v3i32.ll index 1a669ad..f714935 100644 --- a/llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v3i32.ll +++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v3i32.ll @@ -3272,9 +3272,8 @@ define void @v_shuffle_v4i32_v3i32__1_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v4, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -3287,8 +3286,7 @@ define void @v_shuffle_v4i32_v3i32__1_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v4, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -3416,12 +3414,11 @@ define void @v_shuffle_v4i32_v3i32__4_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v7, 0 ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[4:6] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v7, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -3434,12 +3431,12 @@ define void @v_shuffle_v4i32_v3i32__4_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v7, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 ; GFX942-NEXT: ;;#ASMSTART ; GFX942-NEXT: ; def v[4:6] ; GFX942-NEXT: ;;#ASMEND +; GFX942-NEXT: s_nop 0 +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[0:1] op_sel:[1,0] ; GFX942-NEXT: global_store_dwordx4 v7, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: s_setpc_b64 s[30:31] @@ -6083,9 +6080,8 @@ define void @v_shuffle_v4i32_v3i32__1_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v5, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v5, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6102,8 +6098,7 @@ define void @v_shuffle_v4i32_v3i32__1_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -6241,9 +6236,8 @@ define void @v_shuffle_v4i32_v3i32__4_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v4, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6256,8 +6250,7 @@ define void @v_shuffle_v4i32_v3i32__4_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v4, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) diff --git a/llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v3p3.ll b/llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v3p3.ll index 8039e12..aa9e23b 100644 --- a/llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v3p3.ll +++ b/llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v3p3.ll @@ -3272,9 +3272,8 @@ define void @v_shuffle_v4p3_v3p3__1_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v4, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -3287,8 +3286,7 @@ define void @v_shuffle_v4p3_v3p3__1_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v4, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -3416,12 +3414,11 @@ define void @v_shuffle_v4p3_v3p3__4_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v7, 0 ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[4:6] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v7, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -3434,12 +3431,12 @@ define void @v_shuffle_v4p3_v3p3__4_2_2_2(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v7, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 ; GFX942-NEXT: ;;#ASMSTART ; GFX942-NEXT: ; def v[4:6] ; GFX942-NEXT: ;;#ASMEND +; GFX942-NEXT: s_nop 0 +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[4:5], v[0:1] op_sel:[1,0] ; GFX942-NEXT: global_store_dwordx4 v7, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) ; GFX942-NEXT: s_setpc_b64 s[30:31] @@ -6083,9 +6080,8 @@ define void @v_shuffle_v4p3_v3p3__1_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v5, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v5, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6102,8 +6098,7 @@ define void @v_shuffle_v4p3_v3p3__1_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[2:3], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) @@ -6241,9 +6236,8 @@ define void @v_shuffle_v4p3_v3p3__4_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX90A-NEXT: ;;#ASMSTART ; GFX90A-NEXT: ; def v[0:2] ; GFX90A-NEXT: ;;#ASMEND -; GFX90A-NEXT: v_mov_b32_e32 v0, v2 ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 -; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX90A-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX90A-NEXT: v_mov_b32_e32 v3, v2 ; GFX90A-NEXT: global_store_dwordx4 v4, v[0:3], s[16:17] ; GFX90A-NEXT: s_waitcnt vmcnt(0) @@ -6256,8 +6250,7 @@ define void @v_shuffle_v4p3_v3p3__4_5_5_5(ptr addrspace(1) inreg %ptr) { ; GFX942-NEXT: ; def v[0:2] ; GFX942-NEXT: ;;#ASMEND ; GFX942-NEXT: v_mov_b32_e32 v4, 0 -; GFX942-NEXT: v_mov_b32_e32 v0, v2 -; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[0:1] op_sel:[1,0] +; GFX942-NEXT: v_pk_mov_b32 v[0:1], v[0:1], v[2:3] op_sel:[1,0] ; GFX942-NEXT: v_mov_b32_e32 v3, v2 ; GFX942-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] ; GFX942-NEXT: s_waitcnt vmcnt(0) diff --git a/llvm/test/CodeGen/AMDGPU/sign_extend.ll b/llvm/test/CodeGen/AMDGPU/sign_extend.ll index cb8bbde..ece46b5 100644 --- a/llvm/test/CodeGen/AMDGPU/sign_extend.ll +++ b/llvm/test/CodeGen/AMDGPU/sign_extend.ll @@ -6,29 +6,27 @@ define amdgpu_kernel void @s_sext_i1_to_i32(ptr addrspace(1) %out, i32 %a, i32 % ; SI-LABEL: s_sext_i1_to_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_cmp_eq_u32 s2, s3 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_cmp_eq_u32 s4, s5 +; SI-NEXT: s_cselect_b64 s[4:5], -1, 0 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_sext_i1_to_i32: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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_cmp_eq_u32 s2, s3 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 -; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_cmp_eq_u32 s4, s5 +; VI-NEXT: s_cselect_b64 s[4:5], -1, 0 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm %cmp = icmp eq i32 %a, %b %sext = sext i1 %cmp to i32 @@ -78,31 +76,29 @@ define amdgpu_kernel void @s_sext_i1_to_i64(ptr addrspace(1) %out, i32 %a, i32 % ; SI-LABEL: s_sext_i1_to_i64: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_cmp_eq_u32 s2, s3 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_cmp_eq_u32 s4, s5 +; SI-NEXT: s_cselect_b64 s[4:5], -1, 0 +; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 ; SI-NEXT: v_mov_b32_e32 v1, v0 -; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; SI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_sext_i1_to_i64: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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_cmp_eq_u32 s2, s3 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 -; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_cmp_eq_u32 s4, s5 +; VI-NEXT: s_cselect_b64 s[4:5], -1, 0 +; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 ; VI-NEXT: v_mov_b32_e32 v1, v0 -; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; VI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; VI-NEXT: s_endpgm %cmp = icmp eq i32 %a, %b %sext = sext i1 %cmp to i64 @@ -218,29 +214,27 @@ define amdgpu_kernel void @s_sext_i1_to_i16(ptr addrspace(1) %out, i32 %a, i32 % ; SI-LABEL: s_sext_i1_to_i16: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_cmp_eq_u32 s2, s3 -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: s_cselect_b64 s[0:1], -1, 0 -; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] -; SI-NEXT: buffer_store_short v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_cmp_eq_u32 s4, s5 +; SI-NEXT: s_cselect_b64 s[4:5], -1, 0 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] +; SI-NEXT: buffer_store_short v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_sext_i1_to_i16: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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_cmp_eq_u32 s2, s3 -; VI-NEXT: s_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_cselect_b64 s[0:1], -1, 0 -; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[0:1] -; VI-NEXT: buffer_store_short v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_cmp_eq_u32 s4, s5 +; VI-NEXT: s_cselect_b64 s[4:5], -1, 0 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] +; VI-NEXT: buffer_store_short v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm %cmp = icmp eq i32 %a, %b %sext = sext i1 %cmp to i16 diff --git a/llvm/test/CodeGen/AMDGPU/skip-if-dead.ll b/llvm/test/CodeGen/AMDGPU/skip-if-dead.ll index 5461532..e836366 100644 --- a/llvm/test/CodeGen/AMDGPU/skip-if-dead.ll +++ b/llvm/test/CodeGen/AMDGPU/skip-if-dead.ll @@ -1797,8 +1797,8 @@ define amdgpu_ps void @complex_loop(i32 inreg %cmpa, i32 %cmpb, i32 %cmpc) { ; GFX10-WAVE32-NEXT: s_cbranch_scc1 .LBB15_7 ; GFX10-WAVE32-NEXT: ; %bb.1: ; %.lr.ph ; GFX10-WAVE32-NEXT: s_mov_b32 s1, exec_lo -; GFX10-WAVE32-NEXT: s_mov_b32 s0, 0 ; GFX10-WAVE32-NEXT: s_mov_b32 s2, 0 +; GFX10-WAVE32-NEXT: s_mov_b32 s0, 0 ; GFX10-WAVE32-NEXT: s_branch .LBB15_3 ; GFX10-WAVE32-NEXT: .LBB15_2: ; %latch ; GFX10-WAVE32-NEXT: ; in Loop: Header=BB15_3 Depth=1 diff --git a/llvm/test/CodeGen/AMDGPU/sminmax.v2i16.ll b/llvm/test/CodeGen/AMDGPU/sminmax.v2i16.ll index 4799876..76f8f48 100644 --- a/llvm/test/CodeGen/AMDGPU/sminmax.v2i16.ll +++ b/llvm/test/CodeGen/AMDGPU/sminmax.v2i16.ll @@ -369,42 +369,41 @@ define amdgpu_kernel void @s_abs_v4i16(ptr addrspace(1) %out, <4 x i16> %val) #0 ; CI-LABEL: s_abs_v4i16: ; CI: ; %bb.0: ; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; CI-NEXT: s_mov_b32 s7, 0xf000 -; CI-NEXT: s_mov_b32 s6, -1 ; CI-NEXT: s_waitcnt lgkmcnt(0) -; CI-NEXT: s_mov_b32 s4, s0 -; CI-NEXT: s_mov_b32 s5, s1 -; CI-NEXT: s_ashr_i32 s0, s3, 16 -; CI-NEXT: s_ashr_i32 s1, s2, 16 -; CI-NEXT: s_lshr_b32 s8, s2, 16 -; CI-NEXT: s_lshr_b32 s9, s3, 16 -; CI-NEXT: s_sext_i32_i16 s10, s3 -; CI-NEXT: s_sext_i32_i16 s11, s2 -; CI-NEXT: s_sub_i32 s3, 0, s3 -; CI-NEXT: s_sub_i32 s2, 0, s2 -; CI-NEXT: s_sext_i32_i16 s3, s3 -; CI-NEXT: s_sext_i32_i16 s2, s2 +; CI-NEXT: s_mov_b64 s[4:5], s[2:3] +; CI-NEXT: s_ashr_i32 s6, s5, 16 +; CI-NEXT: s_lshr_b32 s9, s5, 16 +; CI-NEXT: s_sext_i32_i16 s10, s5 +; CI-NEXT: s_sub_i32 s5, 0, s5 +; CI-NEXT: s_ashr_i32 s7, s4, 16 +; CI-NEXT: s_lshr_b32 s8, s4, 16 +; CI-NEXT: s_sext_i32_i16 s11, s4 +; CI-NEXT: s_sext_i32_i16 s5, s5 +; CI-NEXT: s_sub_i32 s4, 0, s4 ; CI-NEXT: s_sub_i32 s9, 0, s9 -; CI-NEXT: s_sub_i32 s8, 0, s8 +; CI-NEXT: s_sext_i32_i16 s4, s4 ; CI-NEXT: s_sext_i32_i16 s9, s9 +; CI-NEXT: s_sub_i32 s8, 0, s8 +; CI-NEXT: s_max_i32 s5, s10, s5 ; CI-NEXT: s_sext_i32_i16 s8, s8 -; CI-NEXT: s_max_i32 s2, s11, s2 -; CI-NEXT: s_max_i32 s3, s10, s3 -; CI-NEXT: s_max_i32 s1, s1, s8 -; CI-NEXT: s_max_i32 s0, s0, s9 -; CI-NEXT: s_add_i32 s3, s3, 2 -; CI-NEXT: s_add_i32 s2, s2, 2 -; CI-NEXT: s_lshl_b32 s0, s0, 16 -; CI-NEXT: s_and_b32 s3, s3, 0xffff -; CI-NEXT: s_lshl_b32 s1, s1, 16 -; CI-NEXT: s_and_b32 s2, s2, 0xffff -; CI-NEXT: s_or_b32 s0, s0, s3 -; CI-NEXT: s_or_b32 s1, s1, s2 -; CI-NEXT: s_add_i32 s0, s0, 0x20000 -; CI-NEXT: s_add_i32 s1, s1, 0x20000 -; CI-NEXT: v_mov_b32_e32 v0, s1 -; CI-NEXT: v_mov_b32_e32 v1, s0 -; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; CI-NEXT: s_max_i32 s6, s6, s9 +; CI-NEXT: s_max_i32 s4, s11, s4 +; CI-NEXT: s_add_i32 s5, s5, 2 +; CI-NEXT: s_max_i32 s7, s7, s8 +; CI-NEXT: s_lshl_b32 s6, s6, 16 +; CI-NEXT: s_and_b32 s5, s5, 0xffff +; CI-NEXT: s_add_i32 s4, s4, 2 +; CI-NEXT: s_or_b32 s5, s6, s5 +; CI-NEXT: s_lshl_b32 s6, s7, 16 +; CI-NEXT: s_and_b32 s4, s4, 0xffff +; CI-NEXT: s_or_b32 s4, s6, s4 +; CI-NEXT: s_add_i32 s5, s5, 0x20000 +; CI-NEXT: s_add_i32 s4, s4, 0x20000 +; CI-NEXT: s_mov_b32 s3, 0xf000 +; CI-NEXT: s_mov_b32 s2, -1 +; CI-NEXT: v_mov_b32_e32 v0, s4 +; CI-NEXT: v_mov_b32_e32 v1, s5 +; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; CI-NEXT: s_endpgm %z0 = insertelement <4 x i16> poison, i16 0, i16 0 %z1 = insertelement <4 x i16> %z0, i16 0, i16 1 diff --git a/llvm/test/CodeGen/AMDGPU/spillv16.ll b/llvm/test/CodeGen/AMDGPU/spillv16.ll index 2d54ac8..9686c9d 100644 --- a/llvm/test/CodeGen/AMDGPU/spillv16.ll +++ b/llvm/test/CodeGen/AMDGPU/spillv16.ll @@ -1,6 +1,8 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck %s -check-prefixes=GCN,GCN-TRUE16 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck %s -check-prefixes=GCN,GCN-FAKE16 +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=+real-true16,+d16-write-vgpr32 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck %s -check-prefixes=GFX12-TRUE16,GFX12-TRUE16-D16W32 +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=+real-true16,-d16-write-vgpr32 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck %s -check-prefixes=GFX12-TRUE16,GFX12-TRUE16-D16W16 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=+real-true16 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck %s -check-prefixes=GFX1250,GFX1250-TRUE16 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 -mattr=-real-true16 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck %s -check-prefixes=GFX1250,GFX1250-FAKE16 @@ -35,6 +37,26 @@ define void @spill_i16_alu() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_i16_alu: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: v_add_nc_u16 v0.l, 0x7b, v0.l +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:2 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:2 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-TRUE16-LABEL: spill_i16_alu: ; GFX1250-TRUE16: ; %bb.0: ; %entry ; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 @@ -126,6 +148,56 @@ define void @spill_i16_alu_two_vals() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-D16W32-LABEL: spill_i16_alu_two_vals: +; GFX12-TRUE16-D16W32: ; %bb.0: ; %entry +; GFX12-TRUE16-D16W32-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-D16W32-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: v_add_nc_u16 v0.l, 0x7b, v0.l +; GFX12-TRUE16-D16W32-NEXT: scratch_store_b16 off, v0, s32 offset:6 ; 2-byte Folded Spill +; GFX12-TRUE16-D16W32-NEXT: ;;#ASMSTART +; GFX12-TRUE16-D16W32-NEXT: ;;#ASMEND +; GFX12-TRUE16-D16W32-NEXT: scratch_load_d16_b16 v0, off, s32 offset:4 scope:SCOPE_SYS +; GFX12-TRUE16-D16W32-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: scratch_load_d16_hi_b16 v0, off, s32 offset:6 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-D16W32-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: v_add_nc_u16 v0.l, 0x7b, v0.l +; GFX12-TRUE16-D16W32-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: scratch_store_d16_hi_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-D16W32-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: scratch_store_b16 off, v0, s32 offset:4 scope:SCOPE_SYS +; GFX12-TRUE16-D16W32-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-D16W32-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-TRUE16-D16W16-LABEL: spill_i16_alu_two_vals: +; GFX12-TRUE16-D16W16: ; %bb.0: ; %entry +; GFX12-TRUE16-D16W16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-D16W16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: v_add_nc_u16 v0.l, 0x7b, v0.l +; GFX12-TRUE16-D16W16-NEXT: scratch_store_b16 off, v0, s32 offset:6 ; 2-byte Folded Spill +; GFX12-TRUE16-D16W16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-D16W16-NEXT: ;;#ASMEND +; GFX12-TRUE16-D16W16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:4 scope:SCOPE_SYS +; GFX12-TRUE16-D16W16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: scratch_load_d16_hi_b16 v0, off, s32 offset:6 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-D16W16-NEXT: v_add_nc_u16 v0.l, 0x7b, v0.l +; GFX12-TRUE16-D16W16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: scratch_store_d16_hi_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-D16W16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: scratch_store_b16 off, v0, s32 offset:4 scope:SCOPE_SYS +; GFX12-TRUE16-D16W16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-D16W16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-TRUE16-LABEL: spill_i16_alu_two_vals: ; GFX1250-TRUE16: ; %bb.0: ; %entry ; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 @@ -223,6 +295,25 @@ define void @spill_i16() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_i16: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:2 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:2 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-LABEL: spill_i16: ; GFX1250: ; %bb.0: ; %entry ; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 @@ -282,6 +373,25 @@ define void @spill_half() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_half: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:2 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:2 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-LABEL: spill_half: ; GFX1250: ; %bb.0: ; %entry ; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 @@ -341,6 +451,25 @@ define void @spill_i16_from_v2i16() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_i16_from_v2i16: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:2 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:8 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:8 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:2 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-LABEL: spill_i16_from_v2i16: ; GFX1250: ; %bb.0: ; %entry ; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 @@ -414,13 +543,39 @@ define void @spill_2xi16_from_v2i16() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_2xi16_from_v2i16: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:2 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:8 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:10 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:8 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:2 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:10 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-TRUE16-LABEL: spill_2xi16_from_v2i16: ; GFX1250-TRUE16: ; %bb.0: ; %entry ; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 ; GFX1250-TRUE16-NEXT: s_wait_kmcnt 0x0 ; GFX1250-TRUE16-NEXT: scratch_load_u16 v0, off, s32 offset:2 scope:SCOPE_SYS ; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0 -; GFX1250-TRUE16-NEXT: s_clause 0x1 +; GFX1250-TRUE16-NEXT: s_clause 0x1 ; 4-byte Folded Spill ; GFX1250-TRUE16-NEXT: scratch_store_b32 off, v0, s32 offset:12 ; GFX1250-TRUE16-NEXT: scratch_load_u16 v0, off, s32 scope:SCOPE_SYS ; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0 @@ -444,7 +599,7 @@ define void @spill_2xi16_from_v2i16() { ; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0 ; GFX1250-FAKE16-NEXT: scratch_load_u16 v0, off, s32 offset:2 scope:SCOPE_SYS ; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0 -; GFX1250-FAKE16-NEXT: s_clause 0x1 +; GFX1250-FAKE16-NEXT: s_clause 0x1 ; 4-byte Folded Spill ; GFX1250-FAKE16-NEXT: scratch_store_b32 off, v0, s32 offset:8 ; GFX1250-FAKE16-NEXT: scratch_load_u16 v0, off, s32 scope:SCOPE_SYS ; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0 @@ -520,6 +675,32 @@ define void @spill_2xi16_from_v2i16_one_free_reg() { ; GCN-FAKE16-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-FAKE16-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_2xi16_from_v2i16_one_free_reg: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:2 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:8 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:10 ; 2-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:8 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 offset:2 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_d16_b16 v0, off, s32 offset:10 th:TH_LOAD_LU ; 2-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b16 off, v0, s32 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-TRUE16-LABEL: spill_2xi16_from_v2i16_one_free_reg: ; GFX1250-TRUE16: ; %bb.0: ; %entry ; GFX1250-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 @@ -595,6 +776,25 @@ define void @spill_v2i16() { ; GCN-NEXT: s_waitcnt_vscnt null, 0x0 ; GCN-NEXT: s_setpc_b64 s[30:31] ; +; GFX12-TRUE16-LABEL: spill_v2i16: +; GFX12-TRUE16: ; %bb.0: ; %entry +; GFX12-TRUE16-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_expcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_samplecnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_bvhcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_kmcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_load_b32 v0, off, s32 offset:4 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b32 off, v0, s32 offset:8 ; 4-byte Folded Spill +; GFX12-TRUE16-NEXT: ;;#ASMSTART +; GFX12-TRUE16-NEXT: ;;#ASMEND +; GFX12-TRUE16-NEXT: scratch_load_b32 v0, off, s32 offset:8 th:TH_LOAD_LU ; 4-byte Folded Reload +; GFX12-TRUE16-NEXT: s_wait_loadcnt 0x0 +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: scratch_store_b32 off, v0, s32 offset:4 scope:SCOPE_SYS +; GFX12-TRUE16-NEXT: s_wait_storecnt 0x0 +; GFX12-TRUE16-NEXT: s_setpc_b64 s[30:31] +; ; GFX1250-LABEL: spill_v2i16: ; GFX1250: ; %bb.0: ; %entry ; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0 diff --git a/llvm/test/CodeGen/AMDGPU/sub.ll b/llvm/test/CodeGen/AMDGPU/sub.ll index 5c113d8..0a51601 100644 --- a/llvm/test/CodeGen/AMDGPU/sub.ll +++ b/llvm/test/CodeGen/AMDGPU/sub.ll @@ -11,14 +11,13 @@ define amdgpu_kernel void @s_sub_i32(ptr addrspace(1) %out, i32 %a, i32 %b) { ; GFX6-LABEL: s_sub_i32: ; GFX6: ; %bb.0: ; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GFX6-NEXT: s_mov_b32 s7, 0xf000 -; GFX6-NEXT: s_mov_b32 s6, -1 ; GFX6-NEXT: s_waitcnt lgkmcnt(0) -; GFX6-NEXT: s_mov_b32 s4, s0 -; GFX6-NEXT: s_sub_i32 s0, s2, s3 -; GFX6-NEXT: s_mov_b32 s5, s1 -; GFX6-NEXT: v_mov_b32_e32 v0, s0 -; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX6-NEXT: s_mov_b64 s[4:5], s[2:3] +; GFX6-NEXT: s_sub_i32 s4, s4, s5 +; GFX6-NEXT: s_mov_b32 s3, 0xf000 +; GFX6-NEXT: s_mov_b32 s2, -1 +; GFX6-NEXT: v_mov_b32_e32 v0, s4 +; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX6-NEXT: s_endpgm ; ; GFX8-LABEL: s_sub_i32: diff --git a/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll b/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll index 6a273e5..82ef28f 100644 --- a/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll +++ b/llvm/test/CodeGen/AMDGPU/sub.v2i16.ll @@ -223,44 +223,39 @@ define amdgpu_kernel void @s_test_sub_v2i16_kernarg(ptr addrspace(1) %out, <2 x ; VI-LABEL: s_test_sub_v2i16_kernarg: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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_mov_b32 s4, s0 -; VI-NEXT: s_mov_b32 s5, s1 -; VI-NEXT: s_lshr_b32 s0, s3, 16 -; VI-NEXT: s_lshr_b32 s1, s2, 16 -; VI-NEXT: s_sub_i32 s0, s1, s0 -; VI-NEXT: s_sub_i32 s1, s2, s3 -; VI-NEXT: s_lshl_b32 s0, s0, 16 -; VI-NEXT: s_and_b32 s1, s1, 0xffff -; VI-NEXT: s_or_b32 s0, s1, s0 -; VI-NEXT: v_mov_b32_e32 v0, s0 -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: s_lshr_b32 s6, s5, 16 +; VI-NEXT: s_lshr_b32 s7, s4, 16 +; VI-NEXT: s_sub_i32 s4, s4, s5 +; VI-NEXT: s_sub_i32 s5, s7, s6 +; VI-NEXT: s_lshl_b32 s5, s5, 16 +; VI-NEXT: s_and_b32 s4, s4, 0xffff +; VI-NEXT: s_or_b32 s4, s4, s5 +; VI-NEXT: s_mov_b32 s3, 0xf000 +; VI-NEXT: s_mov_b32 s2, -1 +; VI-NEXT: v_mov_b32_e32 v0, s4 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GFX10-LABEL: s_test_sub_v2i16_kernarg: ; GFX10: ; %bb.0: ; GFX10-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24 -; GFX10-NEXT: s_mov_b32 s7, 0x31016000 -; GFX10-NEXT: s_mov_b32 s6, -1 ; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: v_pk_sub_i16 v0, s2, s3 -; GFX10-NEXT: s_mov_b32 s4, s0 -; GFX10-NEXT: s_mov_b32 s5, s1 -; GFX10-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GFX10-NEXT: s_mov_b32 s3, 0x31016000 +; GFX10-NEXT: s_mov_b32 s2, -1 +; GFX10-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GFX10-NEXT: s_endpgm ; ; GFX11-LABEL: s_test_sub_v2i16_kernarg: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_load_b128 s[0:3], s[4:5], 0x24 -; GFX11-NEXT: s_mov_b32 s7, 0x31016000 -; GFX11-NEXT: s_mov_b32 s6, -1 ; GFX11-NEXT: s_waitcnt lgkmcnt(0) ; GFX11-NEXT: v_pk_sub_i16 v0, s2, s3 -; GFX11-NEXT: s_mov_b32 s4, s0 -; GFX11-NEXT: s_mov_b32 s5, s1 -; GFX11-NEXT: buffer_store_b32 v0, off, s[4:7], 0 +; GFX11-NEXT: s_mov_b32 s3, 0x31016000 +; GFX11-NEXT: s_mov_b32 s2, -1 +; GFX11-NEXT: buffer_store_b32 v0, off, s[0:3], 0 ; GFX11-NEXT: s_endpgm %add = sub <2 x i16> %a, %b store <2 x i16> %add, ptr addrspace(1) %out diff --git a/llvm/test/CodeGen/AMDGPU/udiv.ll b/llvm/test/CodeGen/AMDGPU/udiv.ll index 063c56f..1f93bf7 100644 --- a/llvm/test/CodeGen/AMDGPU/udiv.ll +++ b/llvm/test/CodeGen/AMDGPU/udiv.ll @@ -189,67 +189,65 @@ define amdgpu_kernel void @s_udiv_i32(ptr addrspace(1) %out, i32 %a, i32 %b) { ; SI-LABEL: s_udiv_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: v_cvt_f32_u32_e32 v0, s3 -; SI-NEXT: s_sub_i32 s4, 0, s3 -; SI-NEXT: s_mov_b32 s5, s1 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: v_cvt_f32_u32_e32 v0, s5 +; SI-NEXT: s_sub_i32 s2, 0, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 ; SI-NEXT: v_rcp_iflag_f32_e32 v0, v0 ; SI-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; SI-NEXT: v_cvt_u32_f32_e32 v0, v0 -; SI-NEXT: v_mul_lo_u32 v1, s4, v0 -; SI-NEXT: s_mov_b32 s4, s0 +; SI-NEXT: v_mul_lo_u32 v1, s2, v0 +; SI-NEXT: s_mov_b32 s2, -1 ; SI-NEXT: v_mul_hi_u32 v1, v0, v1 ; SI-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; SI-NEXT: v_mul_hi_u32 v0, s2, v0 -; SI-NEXT: v_readfirstlane_b32 s0, v0 -; SI-NEXT: s_mul_i32 s0, s0, s3 -; SI-NEXT: s_sub_i32 s0, s2, s0 -; SI-NEXT: s_sub_i32 s1, s0, s3 +; SI-NEXT: v_mul_hi_u32 v0, s4, v0 +; SI-NEXT: v_readfirstlane_b32 s6, v0 +; SI-NEXT: s_mul_i32 s6, s6, s5 +; SI-NEXT: s_sub_i32 s4, s4, s6 +; SI-NEXT: s_sub_i32 s6, s4, s5 ; SI-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; SI-NEXT: s_cmp_ge_u32 s0, s3 +; SI-NEXT: s_cmp_ge_u32 s4, s5 ; SI-NEXT: s_cselect_b64 vcc, -1, 0 ; SI-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; SI-NEXT: s_cselect_b32 s0, s1, s0 +; SI-NEXT: s_cselect_b32 s4, s6, s4 ; SI-NEXT: v_add_i32_e32 v1, vcc, 1, v0 -; SI-NEXT: s_cmp_ge_u32 s0, s3 +; SI-NEXT: s_cmp_ge_u32 s4, s5 ; SI-NEXT: s_cselect_b64 vcc, -1, 0 ; SI-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: s_udiv_i32: ; VI: ; %bb.0: ; VI-NEXT: s_load_dwordx4 s[0:3], 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: v_cvt_f32_u32_e32 v0, s3 -; VI-NEXT: s_sub_i32 s4, 0, s3 -; VI-NEXT: s_mov_b32 s5, s1 +; VI-NEXT: s_mov_b64 s[4:5], s[2:3] +; VI-NEXT: v_cvt_f32_u32_e32 v0, s5 +; VI-NEXT: s_sub_i32 s2, 0, s5 +; VI-NEXT: s_mov_b32 s3, 0xf000 ; VI-NEXT: v_rcp_iflag_f32_e32 v0, v0 ; VI-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0 ; VI-NEXT: v_cvt_u32_f32_e32 v0, v0 -; VI-NEXT: v_mul_lo_u32 v1, s4, v0 -; VI-NEXT: s_mov_b32 s4, s0 +; VI-NEXT: v_mul_lo_u32 v1, s2, v0 +; VI-NEXT: s_mov_b32 s2, -1 ; VI-NEXT: v_mul_hi_u32 v1, v0, v1 ; VI-NEXT: v_add_u32_e32 v0, vcc, v0, v1 -; VI-NEXT: v_mul_hi_u32 v0, s2, v0 -; VI-NEXT: v_readfirstlane_b32 s0, v0 -; VI-NEXT: s_mul_i32 s0, s0, s3 -; VI-NEXT: s_sub_i32 s0, s2, s0 -; VI-NEXT: s_sub_i32 s1, s0, s3 +; VI-NEXT: v_mul_hi_u32 v0, s4, v0 +; VI-NEXT: v_readfirstlane_b32 s6, v0 +; VI-NEXT: s_mul_i32 s6, s6, s5 +; VI-NEXT: s_sub_i32 s4, s4, s6 +; VI-NEXT: s_sub_i32 s6, s4, s5 ; VI-NEXT: v_add_u32_e32 v1, vcc, 1, v0 -; VI-NEXT: s_cmp_ge_u32 s0, s3 +; VI-NEXT: s_cmp_ge_u32 s4, s5 ; VI-NEXT: s_cselect_b64 vcc, -1, 0 ; VI-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; VI-NEXT: s_cselect_b32 s0, s1, s0 +; VI-NEXT: s_cselect_b32 s4, s6, s4 ; VI-NEXT: v_add_u32_e32 v1, vcc, 1, v0 -; VI-NEXT: s_cmp_ge_u32 s0, s3 +; VI-NEXT: s_cmp_ge_u32 s4, s5 ; VI-NEXT: s_cselect_b64 vcc, -1, 0 ; VI-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; VI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; VI-NEXT: s_endpgm ; ; GCN-LABEL: s_udiv_i32: diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll index 775483c..1c50f93 100644 --- a/llvm/test/CodeGen/AMDGPU/udiv64.ll +++ b/llvm/test/CodeGen/AMDGPU/udiv64.ll @@ -716,8 +716,6 @@ define amdgpu_kernel void @s_test_udiv24_i48(ptr addrspace(1) %out, i48 %x, i48 ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: v_mov_b32_e32 v3, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: s_and_b32 s3, s3, 0xffff @@ -729,25 +727,23 @@ define amdgpu_kernel void @s_test_udiv24_i48(ptr addrspace(1) %out, i48 %x, i48 ; GCN-NEXT: s_lshr_b64 s[2:3], s[2:3], 24 ; GCN-NEXT: v_cvt_f32_u32_e32 v1, s2 ; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 -; GCN-NEXT: s_mov_b32 s4, s0 -; GCN-NEXT: s_mov_b32 s5, s1 +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: v_mul_f32_e32 v2, v1, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_cvt_u32_f32_e32 v4, v2 ; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1 -; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 ; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v0 -; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v4, vcc ; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 -; GCN-NEXT: buffer_store_short v3, off, s[4:7], 0 offset:4 -; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: buffer_store_short v3, off, s[0:3], 0 offset:4 +; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GCN-NEXT: s_endpgm ; ; GCN-IR-LABEL: s_test_udiv24_i48: ; GCN-IR: ; %bb.0: ; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 ; GCN-IR-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd -; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 -; GCN-IR-NEXT: s_mov_b32 s6, -1 ; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 ; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) ; GCN-IR-NEXT: s_and_b32 s3, s3, 0xffff @@ -759,17 +755,17 @@ define amdgpu_kernel void @s_test_udiv24_i48(ptr addrspace(1) %out, i48 %x, i48 ; GCN-IR-NEXT: s_lshr_b64 s[2:3], s[2:3], 24 ; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s2 ; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v0 -; GCN-IR-NEXT: s_mov_b32 s4, s0 -; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 ; GCN-IR-NEXT: v_mul_f32_e32 v2, v1, v2 ; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v4, v2 ; GCN-IR-NEXT: v_mad_f32 v1, -v2, v0, v1 -; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 ; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v0 -; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v4, vcc ; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 -; GCN-IR-NEXT: buffer_store_short v3, off, s[4:7], 0 offset:4 -; GCN-IR-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-IR-NEXT: buffer_store_short v3, off, s[0:3], 0 offset:4 +; GCN-IR-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; GCN-IR-NEXT: s_endpgm %1 = lshr i48 %x, 24 %2 = lshr i48 %y, 24 diff --git a/llvm/test/CodeGen/AMDGPU/while-break.ll b/llvm/test/CodeGen/AMDGPU/while-break.ll index 19c8e84..2b7e283 100644 --- a/llvm/test/CodeGen/AMDGPU/while-break.ll +++ b/llvm/test/CodeGen/AMDGPU/while-break.ll @@ -157,8 +157,8 @@ define amdgpu_ps < 2 x float> @while_break_two_chains_of_phi(float %v, i32 %x, i ; GCN-LABEL: while_break_two_chains_of_phi: ; GCN: ; %bb.0: ; %entry ; GCN-NEXT: v_mov_b32_e32 v6, 0 -; GCN-NEXT: s_mov_b32 s2, 0 ; GCN-NEXT: s_mov_b32 s0, 0 +; GCN-NEXT: s_mov_b32 s2, 0 ; GCN-NEXT: s_branch .LBB2_2 ; GCN-NEXT: .LBB2_1: ; %Flow1 ; GCN-NEXT: ; in Loop: Header=BB2_2 Depth=1 diff --git a/llvm/test/CodeGen/AMDGPU/xor.ll b/llvm/test/CodeGen/AMDGPU/xor.ll index feb6ecd..92280b9 100644 --- a/llvm/test/CodeGen/AMDGPU/xor.ll +++ b/llvm/test/CodeGen/AMDGPU/xor.ll @@ -298,14 +298,13 @@ define amdgpu_kernel void @scalar_xor_i32(ptr addrspace(1) %out, i32 %a, i32 %b) ; SI-LABEL: scalar_xor_i32: ; SI: ; %bb.0: ; SI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; SI-NEXT: s_mov_b32 s7, 0xf000 -; SI-NEXT: s_mov_b32 s6, -1 ; SI-NEXT: s_waitcnt lgkmcnt(0) -; SI-NEXT: s_mov_b32 s4, s0 -; SI-NEXT: s_xor_b32 s0, s2, s3 -; SI-NEXT: s_mov_b32 s5, s1 -; SI-NEXT: v_mov_b32_e32 v0, s0 -; SI-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; SI-NEXT: s_mov_b64 s[4:5], s[2:3] +; SI-NEXT: s_xor_b32 s4, s4, s5 +; SI-NEXT: s_mov_b32 s3, 0xf000 +; SI-NEXT: s_mov_b32 s2, -1 +; SI-NEXT: v_mov_b32_e32 v0, s4 +; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0 ; SI-NEXT: s_endpgm ; ; VI-LABEL: scalar_xor_i32: diff --git a/llvm/test/CodeGen/AMDGPU/zext-divergence-driven-isel.ll b/llvm/test/CodeGen/AMDGPU/zext-divergence-driven-isel.ll index c393582..d9f5ba9 100644 --- a/llvm/test/CodeGen/AMDGPU/zext-divergence-driven-isel.ll +++ b/llvm/test/CodeGen/AMDGPU/zext-divergence-driven-isel.ll @@ -5,15 +5,14 @@ define amdgpu_kernel void @zext_i16_to_i32_uniform(ptr addrspace(1) %out, i16 %a ; GCN-LABEL: zext_i16_to_i32_uniform: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s4, s0 -; GCN-NEXT: s_and_b32 s0, s2, 0xffff -; GCN-NEXT: s_add_i32 s0, s3, s0 -; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-NEXT: s_mov_b64 s[4:5], s[2:3] +; GCN-NEXT: s_and_b32 s4, s4, 0xffff +; GCN-NEXT: s_add_i32 s4, s5, s4 +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; 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 %zext = zext i16 %a to i32 %res = add i32 %b, %zext |