Index: lib/CodeGen/GlobalISel/LegalizerHelper.cpp =================================================================== --- lib/CodeGen/GlobalISel/LegalizerHelper.cpp +++ lib/CodeGen/GlobalISel/LegalizerHelper.cpp @@ -3111,6 +3111,26 @@ moreElementsVectorDst(MI, MoreTy, 0); Observer.changedInstr(MI); return Legalized; + case TargetOpcode::G_UNMERGE_VALUES: { + if (TypeIdx != 1) + return UnableToLegalize; + + LLT DstTy = MRI.getType(MI.getOperand(0).getReg()); + int NumDst = MI.getNumOperands() - 1; + moreElementsVectorSrc(MI, MoreTy, NumDst); + + auto MIB = MIRBuilder.buildInstr(TargetOpcode::G_UNMERGE_VALUES); + for (int I = 0; I != NumDst; ++I) + MIB.addDef(MI.getOperand(I).getReg()); + + int NewNumDst = MoreTy.getSizeInBits() / DstTy.getSizeInBits(); + for (int I = NumDst; I != NewNumDst; ++I) + MIB.addDef(MRI.createGenericVirtualRegister(DstTy)); + + MIB.addUse(MI.getOperand(NumDst).getReg()); + MI.eraseFromParent(); + return Legalized; + } case TargetOpcode::G_PHI: return moreElementsVectorPhi(MI, TypeIdx, MoreTy); default: Index: lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp =================================================================== --- lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp +++ lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp @@ -737,7 +737,7 @@ // valid. .clampScalar(LitTyIdx, S16, S256) .widenScalarToNextPow2(LitTyIdx, /*Min*/ 32) - .legalIf(all(typeIs(0, S16), typeIs(1, LLT::vector(3, 16)))) // FIXME: Testing hack + .moreElementsIf(isSmallOddVector(BigTyIdx), oneMoreElement(BigTyIdx)) .fewerElementsIf(all(typeIs(0, S16), vectorWiderThan(1, 32), elementTypeIs(1, S16)), changeTo(1, V2S16)) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-anyext.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-anyext.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-anyext.mir @@ -103,11 +103,18 @@ ; CHECK-LABEL: name: test_anyext_v3s16_to_v3s32 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) - ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) - ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) - ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) + ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>) + ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>) + ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) + ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) + ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) + ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY1]](s32), [[COPY2]](s32), [[COPY3]](s32) ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>) %0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %1:_(<3 x s16>) = G_EXTRACT %0, 0 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-ashr.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-ashr.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-ashr.mir @@ -607,43 +607,59 @@ ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) - ; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) + ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV8]](s16) + ; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) ; SI: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SEXT]], [[ZEXT]](s32) - ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) - ; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) + ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV9]](s16) + ; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) ; SI: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SEXT1]], [[ZEXT1]](s32) - ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) - ; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) + ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV10]](s16) + ; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) ; SI: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SEXT2]], [[ZEXT2]](s32) ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ASHR]](s32) ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ASHR1]](s32) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[ASHR2]](s32) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; SI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; SI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; VI-LABEL: name: test_ashr_v3s16_v3s16 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[UV]], [[UV3]](s16) - ; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[UV1]], [[UV4]](s16) - ; VI: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[UV2]], [[UV5]](s16) + ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[UV2]], [[UV8]](s16) + ; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[UV3]], [[UV9]](s16) + ; VI: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[UV4]], [[UV10]](s16) ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR2]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; VI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; VI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; GFX9-LABEL: name: test_ashr_v3s16_v3s16 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fabs.mir @@ -230,14 +230,18 @@ ; SI-LABEL: name: test_fabs_v3s16 ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FABS:%[0-9]+]]:_(s32) = G_FABS [[FPEXT]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS]](s32) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FABS1:%[0-9]+]]:_(s32) = G_FABS [[FPEXT1]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS1]](s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[FABS2:%[0-9]+]]:_(s32) = G_FABS [[FPEXT2]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FABS2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -249,10 +253,14 @@ ; VI-LABEL: name: test_fabs_v3s16 ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[FABS:%[0-9]+]]:_(s16) = G_FABS [[UV]] - ; VI: [[FABS1:%[0-9]+]]:_(s16) = G_FABS [[UV1]] - ; VI: [[FABS2:%[0-9]+]]:_(s16) = G_FABS [[UV2]] + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[FABS:%[0-9]+]]:_(s16) = G_FABS [[UV2]] + ; VI: [[FABS1:%[0-9]+]]:_(s16) = G_FABS [[UV3]] + ; VI: [[FABS2:%[0-9]+]]:_(s16) = G_FABS [[UV4]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FABS]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FABS1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FABS2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fadd.mir @@ -325,18 +325,26 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT]], [[FPEXT1]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) - ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT2]], [[FPEXT3]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32) - ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) - ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) + ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -350,11 +358,19 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV]], [[UV3]] - ; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV1]], [[UV4]] - ; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[UV8]] + ; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV3]], [[UV9]] + ; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV4]], [[UV10]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fcanonicalize.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fcanonicalize.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fcanonicalize.mir @@ -215,14 +215,18 @@ ; SI-LABEL: name: test_fcanonicalize_v3s16 ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FCANONICALIZE:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE]](s32) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FCANONICALIZE1:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT1]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE1]](s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[FCANONICALIZE2:%[0-9]+]]:_(s32) = G_FCANONICALIZE [[FPEXT2]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FCANONICALIZE2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -234,10 +238,14 @@ ; VI-LABEL: name: test_fcanonicalize_v3s16 ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV]] - ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV1]] - ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] + ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] + ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FCANONICALIZE]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FCANONICALIZE1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FCANONICALIZE2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir @@ -390,22 +390,34 @@ ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT2:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF2]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT2]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV6]](s16) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF4:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF4]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[DEF5:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF5]], [[EXTRACT2]](<3 x s16>), 0 + ; SI: [[UV12:%[0-9]+]]:_(<2 x s16>), [[UV13:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT2]](<4 x s16>) + ; SI: [[UV14:%[0-9]+]]:_(s16), [[UV15:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV12]](<2 x s16>) + ; SI: [[UV16:%[0-9]+]]:_(s16), [[UV17:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV13]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV14]](s16) ; SI: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32) - ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) - ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) - ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV7]](s16) + ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) + ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV15]](s16) ; SI: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32) - ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) - ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) - ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) + ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) + ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[UV16]](s16) ; SI: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -421,12 +433,24 @@ ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT2:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF2]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT2]](<3 x s16>) - ; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[UV]], [[UV3]], [[UV6]] - ; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[UV1]], [[UV4]], [[UV7]] - ; VI: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[UV2]], [[UV5]], [[UV8]] + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF4:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF4]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[DEF5:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF5]], [[EXTRACT2]](<3 x s16>), 0 + ; VI: [[UV12:%[0-9]+]]:_(<2 x s16>), [[UV13:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT2]](<4 x s16>) + ; VI: [[UV14:%[0-9]+]]:_(s16), [[UV15:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV12]](<2 x s16>) + ; VI: [[UV16:%[0-9]+]]:_(s16), [[UV17:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV13]](<2 x s16>) + ; VI: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[UV2]], [[UV8]], [[UV14]] + ; VI: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[UV3]], [[UV9]], [[UV15]] + ; VI: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[UV4]], [[UV10]], [[UV16]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fmaxnum.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fmaxnum.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fmaxnum.mir @@ -387,25 +387,33 @@ ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC %24(s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) - ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC %21(s32) - ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) - ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) + ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC %18(s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) ; SI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC1]](s16) ; SI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC2]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; SI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; SI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; SI: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT4]], [[FPEXT5]] ; SI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT2]], [[FPEXT3]] ; SI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT]], [[FPEXT1]] @@ -414,24 +422,32 @@ ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) + ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT %13(s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT %14(s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT %15(s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; VI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) - ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] - ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; VI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) + ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] + ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV10]] ; VI: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE]], [[FCANONICALIZE1]] - ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV1]] - ; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] + ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] + ; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV9]] ; VI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE2]], [[FCANONICALIZE3]] - ; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV]] - ; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] + ; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] + ; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV8]] ; VI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE4]], [[FCANONICALIZE5]] ; GFX9-LABEL: name: test_fminnum_v3s16 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fminnum.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fminnum.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fminnum.mir @@ -387,25 +387,33 @@ ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC %24(s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) - ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC %21(s32) - ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) - ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) + ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC %18(s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) ; SI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC1]](s16) ; SI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC2]](s16) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; SI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; SI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; SI: [[FMINNUM_IEEE:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT4]], [[FPEXT5]] ; SI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT2]], [[FPEXT3]] ; SI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s32) = G_FMINNUM_IEEE [[FPEXT]], [[FPEXT1]] @@ -414,24 +422,32 @@ ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) + ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT %13(s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT %14(s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT %15(s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; VI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) - ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] - ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; VI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) + ; VI: [[FCANONICALIZE:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] + ; VI: [[FCANONICALIZE1:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV10]] ; VI: [[FMINNUM_IEEE:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE]], [[FCANONICALIZE1]] - ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV1]] - ; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV4]] + ; VI: [[FCANONICALIZE2:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] + ; VI: [[FCANONICALIZE3:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV9]] ; VI: [[FMINNUM_IEEE1:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE2]], [[FCANONICALIZE3]] - ; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV]] - ; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV3]] + ; VI: [[FCANONICALIZE4:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV2]] + ; VI: [[FCANONICALIZE5:%[0-9]+]]:_(s16) = G_FCANONICALIZE [[UV8]] ; VI: [[FMINNUM_IEEE2:%[0-9]+]]:_(s16) = G_FMINNUM_IEEE [[FCANONICALIZE4]], [[FCANONICALIZE5]] ; GFX9-LABEL: name: test_fminnum_v3s16 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fmul.mir @@ -324,18 +324,26 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) ; SI: [[FMUL:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT]], [[FPEXT1]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL]](s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) - ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) ; SI: [[FMUL1:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT2]], [[FPEXT3]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL1]](s32) - ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) - ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) + ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) ; SI: [[FMUL2:%[0-9]+]]:_(s32) = G_FMUL [[FPEXT4]], [[FPEXT5]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMUL2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -349,11 +357,19 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[UV]], [[UV3]] - ; VI: [[FMUL1:%[0-9]+]]:_(s16) = G_FMUL [[UV1]], [[UV4]] - ; VI: [[FMUL2:%[0-9]+]]:_(s16) = G_FMUL [[UV2]], [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[FMUL:%[0-9]+]]:_(s16) = G_FMUL [[UV2]], [[UV8]] + ; VI: [[FMUL1:%[0-9]+]]:_(s16) = G_FMUL [[UV3]], [[UV9]] + ; VI: [[FMUL2:%[0-9]+]]:_(s16) = G_FMUL [[UV4]], [[UV10]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMUL]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FMUL1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FMUL2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fneg.mir @@ -228,14 +228,18 @@ ; SI-LABEL: name: test_fneg_v3s16 ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT1]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT2]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -247,10 +251,14 @@ ; VI-LABEL: name: test_fneg_v3s16 ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV]] - ; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV1]] - ; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV2]] + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV2]] + ; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV3]] + ; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV4]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FNEG]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FNEG1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FNEG2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fpext.mir @@ -71,10 +71,20 @@ ; CHECK-LABEL: name: test_fpext_v3f16_to_v3f32 ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) - ; CHECK: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) - ; CHECK: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; CHECK: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>) + ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32) + ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32) + ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>) + ; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32) + ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; CHECK: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16) + ; CHECK: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16) + ; CHECK: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16) ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FPEXT]](s32), [[FPEXT1]](s32), [[FPEXT2]](s32) ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>) %0:_(<3 x s16>) = G_IMPLICIT_DEF Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fsqrt.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fsqrt.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fsqrt.mir @@ -237,14 +237,18 @@ ; SI-LABEL: name: test_fsqrt_v3s16 ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32) - ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32) ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) @@ -256,10 +260,14 @@ ; VI-LABEL: name: test_fsqrt_v3s16 ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[UV]] - ; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[UV1]] - ; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[UV2]] + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[UV2]] + ; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[UV3]] + ; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[UV4]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT2]](s16) @@ -269,10 +277,14 @@ ; GFX9-LABEL: name: test_fsqrt_v3s16 ; GFX9: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[UV]] - ; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[UV1]] - ; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[UV2]] + ; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; GFX9: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[UV2]] + ; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[UV3]] + ; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[UV4]] ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16) ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT1]](s16) ; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-fsub.mir @@ -377,26 +377,34 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[UV8]](s16) ; SI: [[FNEG:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT]] ; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG]](s32) - ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV]](s16) + ; SI: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) ; SI: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC]](s16) ; SI: [[FADD:%[0-9]+]]:_(s32) = G_FADD [[FPEXT1]], [[FPEXT2]] ; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD]](s32) - ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) + ; SI: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[UV9]](s16) ; SI: [[FNEG1:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT3]] ; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG1]](s32) - ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV1]](s16) + ; SI: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[UV3]](s16) ; SI: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC2]](s16) ; SI: [[FADD1:%[0-9]+]]:_(s32) = G_FADD [[FPEXT4]], [[FPEXT5]] ; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD1]](s32) - ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV5]](s16) + ; SI: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[UV10]](s16) ; SI: [[FNEG2:%[0-9]+]]:_(s32) = G_FNEG [[FPEXT6]] ; SI: [[FPTRUNC4:%[0-9]+]]:_(s16) = G_FPTRUNC [[FNEG2]](s32) - ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV2]](s16) + ; SI: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[UV4]](s16) ; SI: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[FPTRUNC4]](s16) ; SI: [[FADD2:%[0-9]+]]:_(s32) = G_FADD [[FPEXT7]], [[FPEXT8]] ; SI: [[FPTRUNC5:%[0-9]+]]:_(s16) = G_FPTRUNC [[FADD2]](s32) @@ -411,14 +419,22 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV3]] - ; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV]], [[FNEG]] - ; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV4]] - ; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV1]], [[FNEG1]] - ; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV5]] - ; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[FNEG2]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV8]] + ; VI: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[FNEG]] + ; VI: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV9]] + ; VI: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV3]], [[FNEG1]] + ; VI: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV10]] + ; VI: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV4]], [[FNEG2]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD2]](s16) @@ -430,14 +446,22 @@ ; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; GFX9: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; GFX9: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; GFX9: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; GFX9: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV3]] - ; GFX9: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV]], [[FNEG]] - ; GFX9: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV4]] - ; GFX9: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV1]], [[FNEG1]] - ; GFX9: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV5]] - ; GFX9: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[FNEG2]] + ; GFX9: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; GFX9: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; GFX9: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; GFX9: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; GFX9: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; GFX9: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; GFX9: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; GFX9: [[FNEG:%[0-9]+]]:_(s16) = G_FNEG [[UV8]] + ; GFX9: [[FADD:%[0-9]+]]:_(s16) = G_FADD [[UV2]], [[FNEG]] + ; GFX9: [[FNEG1:%[0-9]+]]:_(s16) = G_FNEG [[UV9]] + ; GFX9: [[FADD1:%[0-9]+]]:_(s16) = G_FADD [[UV3]], [[FNEG1]] + ; GFX9: [[FNEG2:%[0-9]+]]:_(s16) = G_FNEG [[UV10]] + ; GFX9: [[FADD2:%[0-9]+]]:_(s16) = G_FADD [[UV4]], [[FNEG2]] ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD]](s16) ; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD1]](s16) ; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FADD2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-lshr.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-lshr.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-lshr.mir @@ -592,43 +592,59 @@ ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) - ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) + ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV8]](s16) + ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT1]], [[ZEXT]](s32) - ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) - ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) + ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV9]](s16) + ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT3]], [[ZEXT2]](s32) - ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) - ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) + ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV10]](s16) + ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT5]], [[ZEXT4]](s32) ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; SI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; SI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; VI-LABEL: name: test_lshr_v3s16_v3s16 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[LSHR:%[0-9]+]]:_(s16) = G_LSHR [[UV]], [[UV3]](s16) - ; VI: [[LSHR1:%[0-9]+]]:_(s16) = G_LSHR [[UV1]], [[UV4]](s16) - ; VI: [[LSHR2:%[0-9]+]]:_(s16) = G_LSHR [[UV2]], [[UV5]](s16) + ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[LSHR:%[0-9]+]]:_(s16) = G_LSHR [[UV2]], [[UV8]](s16) + ; VI: [[LSHR1:%[0-9]+]]:_(s16) = G_LSHR [[UV3]], [[UV9]](s16) + ; VI: [[LSHR2:%[0-9]+]]:_(s16) = G_LSHR [[UV4]], [[UV10]](s16) ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[LSHR]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[LSHR1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[LSHR2]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; VI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; VI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; GFX9-LABEL: name: test_lshr_v3s16_v3s16 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-phi.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-phi.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-phi.mir @@ -126,31 +126,39 @@ ; CHECK: G_BR %bb.2 ; CHECK: bb.1: ; CHECK: successors: %bb.2(0x80000000) - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) - ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) + ; CHECK: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; CHECK: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; CHECK: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; CHECK: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT2]](<4 x s16>) + ; CHECK: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; CHECK: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) + ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV8]](s16) ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT]], [[ANYEXT1]] - ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) - ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16) + ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) + ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV9]](s16) ; CHECK: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT2]], [[ANYEXT3]] - ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) - ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s16) + ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16) + ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV10]](s16) ; CHECK: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[ANYEXT4]], [[ANYEXT5]] ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ADD]](s32) ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ADD1]](s32) ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[ADD2]](s32) ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32) ; CHECK: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; CHECK: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; CHECK: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[TRUNC]](<3 x s16>), 0 + ; CHECK: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[TRUNC]](<3 x s16>), 0 ; CHECK: G_BR %bb.2 ; CHECK: bb.2: - ; CHECK: [[PHI:%[0-9]+]]:_(<4 x s16>) = G_PHI [[INSERT]](<4 x s16>), %bb.0, [[INSERT1]](<4 x s16>), %bb.1 + ; CHECK: [[PHI:%[0-9]+]]:_(<4 x s16>) = G_PHI [[INSERT]](<4 x s16>), %bb.0, [[INSERT3]](<4 x s16>), %bb.1 ; CHECK: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[PHI]](<4 x s16>), 0 - ; CHECK: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; CHECK: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT1]](<3 x s16>), 0 - ; CHECK: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) + ; CHECK: [[DEF4:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT4:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF4]], [[EXTRACT1]](<3 x s16>), 0 + ; CHECK: $vgpr0_vgpr1 = COPY [[INSERT4]](<4 x s16>) ; CHECK: S_SETPC_B64 undef $sgpr30_sgpr31 bb.0: successors: %bb.1, %bb.2 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-sext.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-sext.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-sext.mir @@ -108,11 +108,24 @@ ; CHECK-LABEL: name: test_sext_v3s16_to_v3s32 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) - ; CHECK: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) - ; CHECK: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) - ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT]](s32), [[SEXT1]](s32), [[SEXT2]](s32) + ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>) + ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>) + ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) + ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY1]], [[C]](s32) + ; CHECK: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32) + ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) + ; CHECK: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C]](s32) + ; CHECK: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C]](s32) + ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) + ; CHECK: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[C]](s32) + ; CHECK: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C]](s32) + ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ASHR]](s32), [[ASHR1]](s32), [[ASHR2]](s32) ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>) %0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %1:_(<3 x s16>) = G_EXTRACT %0, 0 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-shl.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-shl.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-shl.mir @@ -580,43 +580,59 @@ ; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) - ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) + ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV8]](s16) + ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT]], [[ZEXT]](s32) - ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) - ; SI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) + ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV9]](s16) + ; SI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16) ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT1]], [[ZEXT1]](s32) - ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) - ; SI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) + ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV10]](s16) + ; SI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16) ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[ANYEXT2]], [[ZEXT2]](s32) ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SHL]](s32) ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[SHL1]](s32) ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SHL2]](s32) ; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY3]](s32), [[COPY4]](s32) ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; SI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; SI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; VI-LABEL: name: test_shl_v3s16_v3s16 ; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[UV]], [[UV3]](s16) - ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[UV1]], [[UV4]](s16) - ; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[UV2]], [[UV5]](s16) + ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[UV2]], [[UV8]](s16) + ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[UV3]], [[UV9]](s16) + ; VI: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[UV4]], [[UV10]](s16) ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SHL]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[SHL1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[SHL2]](s16) ; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32) ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s16>) = G_TRUNC [[BUILD_VECTOR]](<3 x s32>) - ; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF - ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[TRUNC]](<3 x s16>), 0 - ; VI: $vgpr0_vgpr1 = COPY [[INSERT]](<4 x s16>) + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[TRUNC]](<3 x s16>), 0 + ; VI: $vgpr0_vgpr1 = COPY [[INSERT2]](<4 x s16>) ; GFX9-LABEL: name: test_shl_v3s16_v3s16 ; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-shuffle-vector.mir @@ -264,32 +264,81 @@ ; CHECK: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3 ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 ; CHECK: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY1]](<4 x s16>), 0 - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; CHECK: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) - ; CHECK: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) - ; CHECK: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) - ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT]](s32), [[SEXT1]](s32), [[SEXT2]](s32) + ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT1]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>) + ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>) + ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) + ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY2]], [[C]](s32) + ; CHECK: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SHL]], [[C]](s32) + ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) + ; CHECK: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[COPY3]], [[C]](s32) + ; CHECK: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[SHL1]], [[C]](s32) + ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) + ; CHECK: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[COPY4]], [[C]](s32) + ; CHECK: [[ASHR2:%[0-9]+]]:_(s32) = G_ASHR [[SHL2]], [[C]](s32) + ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ASHR]](s32), [[ASHR1]](s32), [[ASHR2]](s32) ; CHECK: [[EXTRACT2:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR]](<3 x s32>), 64 ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT2]](s32) - ; CHECK: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) - ; CHECK: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) - ; CHECK: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV5]](s16) - ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT3]](s32), [[SEXT4]](s32), [[SEXT5]](s32) + ; CHECK: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; CHECK: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>) + ; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32) + ; CHECK: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>) + ; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32) + ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32) + ; CHECK: [[SHL3:%[0-9]+]]:_(s32) = G_SHL [[COPY5]], [[C]](s32) + ; CHECK: [[ASHR3:%[0-9]+]]:_(s32) = G_ASHR [[SHL3]], [[C]](s32) + ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32) + ; CHECK: [[SHL4:%[0-9]+]]:_(s32) = G_SHL [[COPY6]], [[C]](s32) + ; CHECK: [[ASHR4:%[0-9]+]]:_(s32) = G_ASHR [[SHL4]], [[C]](s32) + ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32) + ; CHECK: [[SHL5:%[0-9]+]]:_(s32) = G_SHL [[COPY7]], [[C]](s32) + ; CHECK: [[ASHR5:%[0-9]+]]:_(s32) = G_ASHR [[SHL5]], [[C]](s32) + ; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ASHR3]](s32), [[ASHR4]](s32), [[ASHR5]](s32) ; CHECK: [[EXTRACT3:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR1]](<3 x s32>), 32 ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT3]](s32) - ; CHECK: [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16), [[UV8:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; CHECK: [[SEXT6:%[0-9]+]]:_(s32) = G_SEXT [[UV6]](s16) - ; CHECK: [[SEXT7:%[0-9]+]]:_(s32) = G_SEXT [[UV7]](s16) - ; CHECK: [[SEXT8:%[0-9]+]]:_(s32) = G_SEXT [[UV8]](s16) - ; CHECK: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT6]](s32), [[SEXT7]](s32), [[SEXT8]](s32) + ; CHECK: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT2:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT1]](<3 x s16>), 0 + ; CHECK: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT2]](<4 x s16>) + ; CHECK: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>) + ; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32) + ; CHECK: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>) + ; CHECK: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32) + ; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[BITCAST4]](s32) + ; CHECK: [[SHL6:%[0-9]+]]:_(s32) = G_SHL [[COPY8]], [[C]](s32) + ; CHECK: [[ASHR6:%[0-9]+]]:_(s32) = G_ASHR [[SHL6]], [[C]](s32) + ; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32) + ; CHECK: [[SHL7:%[0-9]+]]:_(s32) = G_SHL [[COPY9]], [[C]](s32) + ; CHECK: [[ASHR7:%[0-9]+]]:_(s32) = G_ASHR [[SHL7]], [[C]](s32) + ; CHECK: [[COPY10:%[0-9]+]]:_(s32) = COPY [[BITCAST5]](s32) + ; CHECK: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[COPY10]], [[C]](s32) + ; CHECK: [[ASHR8:%[0-9]+]]:_(s32) = G_ASHR [[SHL8]], [[C]](s32) + ; CHECK: [[BUILD_VECTOR2:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ASHR6]](s32), [[ASHR7]](s32), [[ASHR8]](s32) ; CHECK: [[EXTRACT4:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR2]](<3 x s32>), 0 ; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT4]](s32) - ; CHECK: [[UV9:%[0-9]+]]:_(s16), [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[SEXT9:%[0-9]+]]:_(s32) = G_SEXT [[UV9]](s16) - ; CHECK: [[SEXT10:%[0-9]+]]:_(s32) = G_SEXT [[UV10]](s16) - ; CHECK: [[SEXT11:%[0-9]+]]:_(s32) = G_SEXT [[UV11]](s16) - ; CHECK: [[BUILD_VECTOR3:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SEXT9]](s32), [[SEXT10]](s32), [[SEXT11]](s32) + ; CHECK: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT3:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT3]](<4 x s16>) + ; CHECK: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>) + ; CHECK: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST6]], [[C]](s32) + ; CHECK: [[BITCAST7:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>) + ; CHECK: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST7]], [[C]](s32) + ; CHECK: [[COPY11:%[0-9]+]]:_(s32) = COPY [[BITCAST6]](s32) + ; CHECK: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[COPY11]], [[C]](s32) + ; CHECK: [[ASHR9:%[0-9]+]]:_(s32) = G_ASHR [[SHL9]], [[C]](s32) + ; CHECK: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32) + ; CHECK: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[COPY12]], [[C]](s32) + ; CHECK: [[ASHR10:%[0-9]+]]:_(s32) = G_ASHR [[SHL10]], [[C]](s32) + ; CHECK: [[COPY13:%[0-9]+]]:_(s32) = COPY [[BITCAST7]](s32) + ; CHECK: [[SHL11:%[0-9]+]]:_(s32) = G_SHL [[COPY13]], [[C]](s32) + ; CHECK: [[ASHR11:%[0-9]+]]:_(s32) = G_ASHR [[SHL11]], [[C]](s32) + ; CHECK: [[BUILD_VECTOR3:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ASHR9]](s32), [[ASHR10]](s32), [[ASHR11]](s32) ; CHECK: [[EXTRACT5:%[0-9]+]]:_(s32) = G_EXTRACT [[BUILD_VECTOR3]](<3 x s32>), 0 ; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[EXTRACT5]](s32) ; CHECK: [[BUILD_VECTOR4:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[TRUNC3]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir @@ -96,32 +96,32 @@ ; VI-LABEL: name: test_smax_s8 ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 - ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) - ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[TRUNC]](s16) - ; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC]](s16) + ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) + ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[TRUNC1]](s16) + ; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC1]](s16) ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) - ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C2]](s32) - ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC]](s16) - ; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC]](s16) + ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 + ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[C1]](s32) + ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC3]](s16) + ; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC3]](s16) ; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[ASHR]], [[ASHR1]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16) ; VI: $vgpr0 = COPY [[ANYEXT]](s32) ; GFX9-LABEL: name: test_smax_s8 ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 - ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) - ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[TRUNC]](s16) - ; GFX9: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC]](s16) + ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) + ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[TRUNC1]](s16) + ; GFX9: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC1]](s16) ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) - ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) - ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC]](s16) - ; GFX9: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC]](s16) + ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 + ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[C1]](s32) + ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC3]](s16) + ; GFX9: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC3]](s16) ; GFX9: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[ASHR]], [[ASHR1]] ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16) ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32) @@ -339,16 +339,24 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) - ; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) + ; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV8]](s16) ; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT]], [[SEXT1]] - ; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) - ; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) + ; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) + ; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV9]](s16) ; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[SEXT2]], [[SEXT3]] - ; SI: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) - ; SI: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV5]](s16) + ; SI: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) + ; SI: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV10]](s16) ; SI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[SEXT4]], [[SEXT5]] ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32) ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[SMAX1]](s32) @@ -361,11 +369,19 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[UV]], [[UV3]] - ; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[UV1]], [[UV4]] - ; VI: [[SMAX2:%[0-9]+]]:_(s16) = G_SMAX [[UV2]], [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[UV2]], [[UV8]] + ; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[UV3]], [[UV9]] + ; VI: [[SMAX2:%[0-9]+]]:_(s16) = G_SMAX [[UV4]], [[UV10]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir @@ -96,32 +96,32 @@ ; VI-LABEL: name: test_smin_s8 ; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 - ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) - ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[TRUNC]](s16) - ; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC]](s16) + ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) + ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[TRUNC1]](s16) + ; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC1]](s16) ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) - ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C2]](s32) - ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC]](s16) - ; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC]](s16) + ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 + ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[C1]](s32) + ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC3]](s16) + ; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC3]](s16) ; VI: [[SMIN:%[0-9]+]]:_(s16) = G_SMIN [[ASHR]], [[ASHR1]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMIN]](s16) ; VI: $vgpr0 = COPY [[ANYEXT]](s32) ; GFX9-LABEL: name: test_smin_s8 ; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0 ; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1 - ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) + ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32) ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) - ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[TRUNC]](s16) - ; GFX9: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC]](s16) + ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) + ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[TRUNC1]](s16) + ; GFX9: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[TRUNC1]](s16) ; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32) - ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 - ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[C]](s32) - ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC]](s16) - ; GFX9: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC]](s16) + ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8 + ; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[C1]](s32) + ; GFX9: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[TRUNC3]](s16) + ; GFX9: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[TRUNC3]](s16) ; GFX9: [[SMIN:%[0-9]+]]:_(s16) = G_SMIN [[ASHR]], [[ASHR1]] ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMIN]](s16) ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32) @@ -339,16 +339,24 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s16) - ; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) + ; SI: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV8]](s16) ; SI: [[SMIN:%[0-9]+]]:_(s32) = G_SMIN [[SEXT]], [[SEXT1]] - ; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s16) - ; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) + ; SI: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s16) + ; SI: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV9]](s16) ; SI: [[SMIN1:%[0-9]+]]:_(s32) = G_SMIN [[SEXT2]], [[SEXT3]] - ; SI: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s16) - ; SI: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV5]](s16) + ; SI: [[SEXT4:%[0-9]+]]:_(s32) = G_SEXT [[UV4]](s16) + ; SI: [[SEXT5:%[0-9]+]]:_(s32) = G_SEXT [[UV10]](s16) ; SI: [[SMIN2:%[0-9]+]]:_(s32) = G_SMIN [[SEXT4]], [[SEXT5]] ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY [[SMIN]](s32) ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[SMIN1]](s32) @@ -361,11 +369,19 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[SMIN:%[0-9]+]]:_(s16) = G_SMIN [[UV]], [[UV3]] - ; VI: [[SMIN1:%[0-9]+]]:_(s16) = G_SMIN [[UV1]], [[UV4]] - ; VI: [[SMIN2:%[0-9]+]]:_(s16) = G_SMIN [[UV2]], [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[SMIN:%[0-9]+]]:_(s16) = G_SMIN [[UV2]], [[UV8]] + ; VI: [[SMIN1:%[0-9]+]]:_(s16) = G_SMIN [[UV3]], [[UV9]] + ; VI: [[SMIN2:%[0-9]+]]:_(s16) = G_SMIN [[UV4]], [[UV10]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMIN]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[SMIN1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[SMIN2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir @@ -318,16 +318,24 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) - ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) + ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV8]](s16) ; SI: [[UMAX:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT]], [[ZEXT1]] - ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) - ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) + ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) + ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV9]](s16) ; SI: [[UMAX1:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT2]], [[ZEXT3]] - ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) - ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) + ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) + ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV10]](s16) ; SI: [[UMAX2:%[0-9]+]]:_(s32) = G_UMAX [[ZEXT4]], [[ZEXT5]] ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY [[UMAX]](s32) ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UMAX1]](s32) @@ -340,11 +348,19 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[UMAX:%[0-9]+]]:_(s16) = G_UMAX [[UV]], [[UV3]] - ; VI: [[UMAX1:%[0-9]+]]:_(s16) = G_UMAX [[UV1]], [[UV4]] - ; VI: [[UMAX2:%[0-9]+]]:_(s16) = G_UMAX [[UV2]], [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[UMAX:%[0-9]+]]:_(s16) = G_UMAX [[UV2]], [[UV8]] + ; VI: [[UMAX1:%[0-9]+]]:_(s16) = G_UMAX [[UV3]], [[UV9]] + ; VI: [[UMAX2:%[0-9]+]]:_(s16) = G_UMAX [[UV4]], [[UV10]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UMAX]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UMAX1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UMAX2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir @@ -318,16 +318,24 @@ ; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; SI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; SI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; SI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) - ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) + ; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; SI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; SI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; SI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; SI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; SI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; SI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) + ; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV8]](s16) ; SI: [[UMIN:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT]], [[ZEXT1]] - ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) - ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) + ; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV3]](s16) + ; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[UV9]](s16) ; SI: [[UMIN1:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT2]], [[ZEXT3]] - ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) - ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV5]](s16) + ; SI: [[ZEXT4:%[0-9]+]]:_(s32) = G_ZEXT [[UV4]](s16) + ; SI: [[ZEXT5:%[0-9]+]]:_(s32) = G_ZEXT [[UV10]](s16) ; SI: [[UMIN2:%[0-9]+]]:_(s32) = G_UMIN [[ZEXT4]], [[ZEXT5]] ; SI: [[COPY:%[0-9]+]]:_(s32) = COPY [[UMIN]](s32) ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UMIN1]](s32) @@ -340,11 +348,19 @@ ; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 ; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; VI: [[EXTRACT1:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF1]](<4 x s16>), 0 - ; VI: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; VI: [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT1]](<3 x s16>) - ; VI: [[UMIN:%[0-9]+]]:_(s16) = G_UMIN [[UV]], [[UV3]] - ; VI: [[UMIN1:%[0-9]+]]:_(s16) = G_UMIN [[UV1]], [[UV4]] - ; VI: [[UMIN2:%[0-9]+]]:_(s16) = G_UMIN [[UV2]], [[UV5]] + ; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF2]], [[EXTRACT]](<3 x s16>), 0 + ; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; VI: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV]](<2 x s16>) + ; VI: [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV1]](<2 x s16>) + ; VI: [[DEF3:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; VI: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF3]], [[EXTRACT1]](<3 x s16>), 0 + ; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>) + ; VI: [[UV8:%[0-9]+]]:_(s16), [[UV9:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV6]](<2 x s16>) + ; VI: [[UV10:%[0-9]+]]:_(s16), [[UV11:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[UV7]](<2 x s16>) + ; VI: [[UMIN:%[0-9]+]]:_(s16) = G_UMIN [[UV2]], [[UV8]] + ; VI: [[UMIN1:%[0-9]+]]:_(s16) = G_UMIN [[UV3]], [[UV9]] + ; VI: [[UMIN2:%[0-9]+]]:_(s16) = G_UMIN [[UV4]], [[UV10]] ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UMIN]](s16) ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UMIN1]](s16) ; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UMIN2]](s16) Index: test/CodeGen/AMDGPU/GlobalISel/legalize-unmerge-values.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-unmerge-values.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-unmerge-values.mir @@ -61,13 +61,20 @@ ; CHECK-LABEL: name: test_unmerge_s16_v3s16 ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[DEF]](<4 x s16>), 0 - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16) - ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16) - ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16) - ; CHECK: $vgpr0 = COPY [[ANYEXT]](s32) - ; CHECK: $vgpr1 = COPY [[ANYEXT1]](s32) - ; CHECK: $vgpr2 = COPY [[ANYEXT2]](s32) + ; CHECK: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>) + ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>) + ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) + ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) + ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) + ; CHECK: $vgpr0 = COPY [[COPY]](s32) + ; CHECK: $vgpr1 = COPY [[COPY1]](s32) + ; CHECK: $vgpr2 = COPY [[COPY2]](s32) %0:_(<3 x s16>) = G_IMPLICIT_DEF %1:_(s16), %2:_(s16), %3:_(s16) = G_UNMERGE_VALUES %0 %4:_(s32) = G_ANYEXT %1 Index: test/CodeGen/AMDGPU/GlobalISel/legalize-zext.mir =================================================================== --- test/CodeGen/AMDGPU/GlobalISel/legalize-zext.mir +++ test/CodeGen/AMDGPU/GlobalISel/legalize-zext.mir @@ -110,11 +110,22 @@ ; CHECK-LABEL: name: test_zext_v3s16_to_v3s32 ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1 ; CHECK: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[COPY]](<4 x s16>), 0 - ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[EXTRACT]](<3 x s16>) - ; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16) - ; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16) - ; CHECK: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[UV2]](s16) - ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ZEXT]](s32), [[ZEXT1]](s32), [[ZEXT2]](s32) + ; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF + ; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF]], [[EXTRACT]](<3 x s16>), 0 + ; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>) + ; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>) + ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16 + ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32) + ; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>) + ; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32) + ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535 + ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32) + ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C1]] + ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32) + ; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]] + ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32) + ; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]] + ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[AND]](s32), [[AND1]](s32), [[AND2]](s32) ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>) %0:_(<4 x s16>) = COPY $vgpr0_vgpr1 %1:_(<3 x s16>) = G_EXTRACT %0, 0