Index: lib/Target/AMDGPU/AMDGPUGISel.td =================================================================== --- lib/Target/AMDGPU/AMDGPUGISel.td +++ lib/Target/AMDGPU/AMDGPUGISel.td @@ -63,3 +63,7 @@ def gi_vop3mods0 : GIComplexOperandMatcher, GIComplexPatternEquiv; + +def gi_vop3mods : + GIComplexOperandMatcher, + GIComplexPatternEquiv; Index: lib/Target/AMDGPU/AMDGPUInstructionSelector.h =================================================================== --- lib/Target/AMDGPU/AMDGPUInstructionSelector.h +++ lib/Target/AMDGPU/AMDGPUInstructionSelector.h @@ -76,6 +76,8 @@ InstructionSelector::ComplexRendererFns selectVOP3Mods0(MachineOperand &Root) const; + InstructionSelector::ComplexRendererFns + selectVOP3Mods(MachineOperand &Root) const; const SIInstrInfo &TII; const SIRegisterInfo &TRI; Index: lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp =================================================================== --- lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp +++ lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp @@ -594,6 +594,8 @@ switch (I.getOpcode()) { default: break; + case TargetOpcode::G_FMUL: + case TargetOpcode::G_FADD: case TargetOpcode::G_FPTOUI: case TargetOpcode::G_OR: return selectImpl(I, CoverageInfo); @@ -629,3 +631,11 @@ [=](MachineInstrBuilder &MIB) { MIB.addImm(0); } // omod }}; } + +InstructionSelector::ComplexRendererFns +AMDGPUInstructionSelector::selectVOP3Mods(MachineOperand &Root) const { + return {{ + [=](MachineInstrBuilder &MIB) { MIB.add(Root); }, + [=](MachineInstrBuilder &MIB) { MIB.addImm(0); } // src_mods + }}; +} Index: test/CodeGen/AMDGPU/GlobalISel/inst-select-fadd.mir =================================================================== --- /dev/null +++ test/CodeGen/AMDGPU/GlobalISel/inst-select-fadd.mir @@ -0,0 +1,37 @@ +# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s -check-prefixes=GCN + +--- | + define amdgpu_kernel void @fadd(i32 addrspace(1)* %global0) {ret void} +... +--- + +name: fadd +legalized: true +regBankSelected: true + +# GCN-LABEL: name: fadd +body: | + bb.0: + liveins: $sgpr0, $vgpr0, $vgpr1, $vgpr3_vgpr4 + %0:sgpr(s32) = COPY $sgpr0 + %1:vgpr(s32) = COPY $vgpr0 + %2:vgpr(s32) = COPY $vgpr1 + %3:vgpr(s64) = COPY $vgpr3_vgpr4 + + ; fadd vs + ; GCN: V_ADD_F32_e64 + %4:vgpr(s32) = G_FADD %1, %0 + + ; fadd sv + ; GCN: V_ADD_F32_e64 + %5:vgpr(s32) = G_FADD %0, %1 + + ; fadd vv + ; GCN: V_ADD_F32_e64 + %6:vgpr(s32) = G_FADD %1, %2 + + G_STORE %4, %3 :: (store 4 into %ir.global0) + G_STORE %5, %3 :: (store 4 into %ir.global0) + G_STORE %6, %3 :: (store 4 into %ir.global0) +... +--- Index: test/CodeGen/AMDGPU/GlobalISel/inst-select-fmul.mir =================================================================== --- /dev/null +++ test/CodeGen/AMDGPU/GlobalISel/inst-select-fmul.mir @@ -0,0 +1,37 @@ +# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s -check-prefixes=GCN + +--- | + define amdgpu_kernel void @fmul(i32 addrspace(1)* %global0) {ret void} +... +--- + +name: fmul +legalized: true +regBankSelected: true + +# GCN-LABEL: name: fmul +body: | + bb.0: + liveins: $sgpr0, $vgpr0, $vgpr1, $vgpr3_vgpr4 + %0:sgpr(s32) = COPY $sgpr0 + %1:vgpr(s32) = COPY $vgpr0 + %2:vgpr(s32) = COPY $vgpr1 + %3:vgpr(s64) = COPY $vgpr3_vgpr4 + + ; fmul vs + ; GCN: V_MUL_F32_e64 + %4:vgpr(s32) = G_FMUL %1, %0 + + ; fmul sv + ; GCN: V_MUL_F32_e64 + %5:vgpr(s32) = G_FMUL %0, %1 + + ; fmul vv + ; GCN: V_MUL_F32_e64 + %6:vgpr(s32) = G_FMUL %1, %2 + + G_STORE %4, %3 :: (store 4 into %ir.global0) + G_STORE %5, %3 :: (store 4 into %ir.global0) + G_STORE %6, %3 :: (store 4 into %ir.global0) +... +---