From 41e74e400dc17cc25a3d69accae3a793e669e2af Mon Sep 17 00:00:00 2001 From: Konstantin Pyzhov Date: Thu, 5 Nov 2020 11:15:50 -0500 Subject: [PATCH] [AMDGPU] Corrected declaration of VOPC instructions with SDWA addressing mode. Removed "implicit def VCC" from declarations of AMDGPU VOPC instructions since they do not implicitly write to VCC in SDWA mode. Differential Revision: https://reviews.llvm.org/D89168 --- llvm/lib/Target/AMDGPU/VOPCInstructions.td | 2 +- llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir | 42 ++++++++++++------------ llvm/test/CodeGen/AMDGPU/wave32.ll | 8 ++--- 3 files changed, 26 insertions(+), 26 deletions(-) diff --git a/llvm/lib/Target/AMDGPU/VOPCInstructions.td b/llvm/lib/Target/AMDGPU/VOPCInstructions.td index aa2fa26..99599c5 100644 --- a/llvm/lib/Target/AMDGPU/VOPCInstructions.td +++ b/llvm/lib/Target/AMDGPU/VOPCInstructions.td @@ -229,7 +229,7 @@ multiclass VOPC_Pseudos .ret in def _sdwa : VOPC_SDWA_Pseudo { - let Defs = !if(DefExec, [VCC, EXEC], [VCC]); + let Defs = !if(DefExec, [EXEC], []); let SchedRW = P.Schedule; let isConvergent = DefExec; let isCompare = 1; diff --git a/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir b/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir index fa55e1b..89b4dbf 100644 --- a/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir +++ b/llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir @@ -325,39 +325,39 @@ body: | # GCN-LABEL: {{^}}name: vopc_instructions # GFX89: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_e32 123, implicit $exec -# GFX89: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec -# GFX89: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# GFX89: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $exec -# GFX89: $vcc = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $exec +# GFX89: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $mode, implicit $exec +# GFX89: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# GFX89: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $exec +# GFX89: $vcc = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $exec -# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec +# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $mode, implicit $exec # VI: %{{[0-9]+}}:sreg_64 = V_CMPX_GT_F32_e64 0, 23, 0, killed %{{[0-9]+}}, 0, implicit-def $exec, implicit $mode, implicit $exec -# VI: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %3, 0, 6, 4, implicit-def $vcc, implicit $exec +# VI: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %3, 0, 6, 4, implicit $exec # VI: %{{[0-9]+}}:sreg_64 = V_CMPX_EQ_I32_e64 23, killed %{{[0-9]+}}, implicit-def $exec, implicit $exec -# GFX9: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec +# GFX9: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $mode, implicit $exec # GFX9: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_e32 23, implicit $exec -# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# GFX9: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit $exec +# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# GFX9: $vcc = V_CMP_LT_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit $exec # GFX9: %{{[0-9]+}}:vgpr_32 = V_MOV_B32_e32 23, implicit $exec -# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $exec +# GFX9: %{{[0-9]+}}:sreg_64 = V_CMPX_EQ_I32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $exec -# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec -# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit-def $vcc, implicit $mode, implicit $exec -# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 1, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec +# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit $mode, implicit $exec +# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# VI: $vcc = V_CMP_EQ_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 1, 6, 4, implicit $mode, implicit $exec +# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# VI: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# VI: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 1, 6, 4, implicit-def $exec, implicit $mode, implicit $exec # GFX9: $vcc = V_CMP_EQ_F32_e64 0, %{{[0-9]+}}, 0, killed %{{[0-9]+}}, 1, implicit $mode, implicit $exec -# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec +# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec # GFX9: $vcc = V_CMP_EQ_F32_e64 0, %{{[0-9]+}}, 0, killed %{{[0-9]+}}, 1, implicit $mode, implicit $exec -# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec -# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $vcc, implicit-def $exec, implicit $mode, implicit $exec +# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 0, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# GFX9: $vcc = V_CMPX_GT_F32_sdwa 0, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec +# GFX9: $vcc = V_CMPX_GT_F32_sdwa 1, %{{[0-9]+}}, 1, %{{[0-9]+}}, 0, 6, 4, implicit-def $exec, implicit $mode, implicit $exec # GFX9: $vcc = V_CMPX_GT_F32_e64 1, 23, 1, killed %{{[0-9]+}}, 1, implicit-def $exec, implicit $mode, implicit $exec diff --git a/llvm/test/CodeGen/AMDGPU/wave32.ll b/llvm/test/CodeGen/AMDGPU/wave32.ll index 896b974..58737e6 100644 --- a/llvm/test/CodeGen/AMDGPU/wave32.ll +++ b/llvm/test/CodeGen/AMDGPU/wave32.ll @@ -44,10 +44,10 @@ define amdgpu_ps void @test_vopc_vcmpx(float %x) { } ; GCN-LABEL: {{^}}test_vopc_2xf16: -; GFX1032: v_cmp_le_f16_sdwa [[SC:s[0-9]+]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD -; GFX1032: v_cndmask_b32_e64 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]] -; GFX1064: v_cmp_le_f16_sdwa [[SC:s\[[0-9:]+\]]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD -; GFX1064: v_cndmask_b32_e64 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]] +; GFX1032: v_cmp_le_f16_sdwa [[SC:vcc_lo|s[0-9]+]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD +; GFX1032: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]] +; GFX1064: v_cmp_le_f16_sdwa [[SC:vcc|s\[[0-9:]+\]]], {{[vs][0-9]+}}, v{{[0-9]+}} src0_sel:WORD_1 src1_sel:DWORD +; GFX1064: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3c003c00, v{{[0-9]+}}, [[SC]] define amdgpu_kernel void @test_vopc_2xf16(<2 x half> addrspace(1)* %arg) { %lid = tail call i32 @llvm.amdgcn.workitem.id.x() %gep = getelementptr inbounds <2 x half>, <2 x half> addrspace(1)* %arg, i32 %lid -- 2.7.4