Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUGISel.td =================================================================== --- llvm/trunk/lib/Target/AMDGPU/AMDGPUGISel.td +++ llvm/trunk/lib/Target/AMDGPU/AMDGPUGISel.td @@ -18,6 +18,10 @@ GIComplexOperandMatcher, GIComplexPatternEquiv; +def gi_vop3mods0 : + GIComplexOperandMatcher, + GIComplexPatternEquiv; + class GISelSop2Pat < SDPatternOperator node, Instruction inst, Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.h =================================================================== --- llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.h +++ llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.h @@ -73,6 +73,9 @@ InstructionSelector::ComplexRendererFns selectVSRC0(MachineOperand &Root) const; + InstructionSelector::ComplexRendererFns + selectVOP3Mods0(MachineOperand &Root) const; + const SIInstrInfo &TII; const SIRegisterInfo &TRI; const AMDGPURegisterBankInfo &RBI; Index: llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp =================================================================== --- llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp +++ llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp @@ -455,6 +455,7 @@ switch (I.getOpcode()) { default: break; + case TargetOpcode::G_FPTOUI: case TargetOpcode::G_OR: return selectImpl(I, CoverageInfo); case TargetOpcode::G_ADD: @@ -482,3 +483,13 @@ [=](MachineInstrBuilder &MIB) { MIB.add(Root); } }}; } + +InstructionSelector::ComplexRendererFns +AMDGPUInstructionSelector::selectVOP3Mods0(MachineOperand &Root) const { + return {{ + [=](MachineInstrBuilder &MIB) { MIB.add(Root); }, + [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // src0_mods + [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // clamp + [=](MachineInstrBuilder &MIB) { MIB.addImm(0); } // omod + }}; +} Index: llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir =================================================================== --- llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir +++ llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir @@ -0,0 +1,36 @@ +# RUN: llc -march=amdgcn -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s -check-prefixes=GCN + +--- | + define amdgpu_kernel void @fptoui(i32 addrspace(1)* %global0) {ret void} +... +--- + +name: fptoui +legalized: true +regBankSelected: true + +# GCN-LABEL: name: fptoui +body: | + bb.0: + liveins: $sgpr0, $vgpr0, $vgpr3_vgpr4 + + ; GCN: [[SGPR:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0 + %0:sgpr(s32) = COPY $sgpr0 + + ; GCN: [[VGPR:%[0-9]+]]:vgpr_32 = COPY $vgpr0 + %1:vgpr(s32) = COPY $vgpr0 + + %2:vgpr(s64) = COPY $vgpr3_vgpr4 + + ; fptoui s + ; GCN: V_CVT_U32_F32_e64 0, [[SGPR]], 0, 0 + %3:vgpr(s32) = G_FPTOUI %0 + + ; fptoui v + ; GCN: V_CVT_U32_F32_e64 0, [[VGPR]], 0, 0 + %4:vgpr(s32) = G_FPTOUI %1 + + G_STORE %3, %2 :: (store 4 into %ir.global0) + G_STORE %4, %2 :: (store 4 into %ir.global0) +... +---