This is an archive of the discontinued LLVM Phabricator instance.

Differential D70403

AMDGPU/GlobalISel: Legalize FDIV64
ClosedPublic

Authored by kerbowa on Nov 18 2019, 9:27 AM.

Details

Reviewers
arsenm
Commits
rGf3225f2abe78: AMDGPU/GlobalISel: Legalize FDIV64

Diff Detail

Event Timeline

kerbowa created this revision.Nov 18 2019, 9:27 AM
Herald added a project: Restricted Project. · View Herald TranscriptNov 18 2019, 9:27 AM
Herald added subscribers: llvm-commits, Petar.Avramovic, hiraditya and 9 others. · View Herald Transcript
Harbormaster completed remote builds in B41125: Diff 229870.Nov 18 2019, 9:29 AM
arsenm added inline comments.Nov 18 2019, 7:44 PM
llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
1890

All these auto substitutions are unrelated

2125–2133

The ugliness of casting to a vector and extracting the elements isn't necessary. This isn't the canonical way to handle this in GlobalISel. You can use G_UNMERGE_VALUES to 2 S32 pieces more naturally

kerbowa updated this revision to Diff 230126.Nov 19 2019, 12:28 PM

Use G_UNMERGE_VALUES.

kerbowa updated this revision to Diff 230127.Nov 19 2019, 12:33 PM

Remove unrelated change to Flags type.

Harbormaster completed remote builds in B41194: Diff 230126.Nov 19 2019, 12:34 PM
Harbormaster completed remote builds in B41195: Diff 230127.
kerbowa marked 2 inline comments as done.Nov 19 2019, 12:34 PM
arsenm accepted this revision.Nov 19 2019, 6:59 PM

LGTM, but it looks like none of these tests have the preserved flags?

This revision is now accepted and ready to land.Nov 19 2019, 6:59 PM
Closed by commit rGf3225f2abe78: AMDGPU/GlobalISel: Legalize FDIV64 (authored by kerbowa). · Explain WhyNov 19 2019, 9:04 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
llvm/
lib/
Target/
AMDGPU/
2 lines
85 lines
test/
CodeGen/
AMDGPU/
GlobalISel/
509 lines

Diff 230127

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.h

Show First 20 Lines • Show All 84 Lines • ▼ Show 20 Lines bool legalizePreloadedArgIntrin(
AMDGPUFunctionArgInfo::PreloadedValue ArgType) const; AMDGPUFunctionArgInfo::PreloadedValue ArgType) const;
bool legalizeFDIV(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeFDIV(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const; MachineIRBuilder &B) const;
bool legalizeFDIV16(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeFDIV16(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const; MachineIRBuilder &B) const;
bool legalizeFDIV32(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeFDIV32(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const; MachineIRBuilder &B) const;
bool legalizeFDIV64(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const;
bool legalizeFastUnsafeFDIV(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeFastUnsafeFDIV(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const; MachineIRBuilder &B) const;
bool legalizeFDIVFastIntrin(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeFDIVFastIntrin(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const; MachineIRBuilder &B) const;
bool legalizeImplicitArgPtr(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeImplicitArgPtr(MachineInstr &MI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const; MachineIRBuilder &B) const;
bool legalizeIsAddrSpace(MachineInstr &MI, MachineRegisterInfo &MRI, bool legalizeIsAddrSpace(MachineInstr &MI, MachineRegisterInfo &MRI,
Show All 12 Lines

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

Show First 20 Lines • Show All 1,859 Lines • ▼ Show 20 Lines
bool AMDGPULegalizerInfo::legalizeFDIV(MachineInstr &MI, bool AMDGPULegalizerInfo::legalizeFDIV(MachineInstr &MI,
MachineRegisterInfo &MRI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const { MachineIRBuilder &B) const {
B.setInstr(MI); B.setInstr(MI);
Register Dst = MI.getOperand(0).getReg(); Register Dst = MI.getOperand(0).getReg();
LLT DstTy = MRI.getType(Dst); LLT DstTy = MRI.getType(Dst);
LLT S16 = LLT::scalar(16); LLT S16 = LLT::scalar(16);
LLT S32 = LLT::scalar(32); LLT S32 = LLT::scalar(32);
LLT S64 = LLT::scalar(64);
if (legalizeFastUnsafeFDIV(MI, MRI, B)) if (legalizeFastUnsafeFDIV(MI, MRI, B))
return true; return true;
if (DstTy == S16) if (DstTy == S16)
return legalizeFDIV16(MI, MRI, B); return legalizeFDIV16(MI, MRI, B);
if (DstTy == S32) if (DstTy == S32)
return legalizeFDIV32(MI, MRI, B); return legalizeFDIV32(MI, MRI, B);
if (DstTy == S64)
return legalizeFDIV64(MI, MRI, B);
return false; return false;
} }
bool AMDGPULegalizerInfo::legalizeFastUnsafeFDIV(MachineInstr &MI, bool AMDGPULegalizerInfo::legalizeFastUnsafeFDIV(MachineInstr &MI,
MachineRegisterInfo &MRI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const { MachineIRBuilder &B) const {
Register Res = MI.getOperand(0).getReg(); Register Res = MI.getOperand(0).getReg();
Register LHS = MI.getOperand(1).getReg(); Register LHS = MI.getOperand(1).getReg();
Register RHS = MI.getOperand(2).getReg(); Register RHS = MI.getOperand(2).getReg();
uint16_t Flags = MI.getFlags(); uint16_t Flags = MI.getFlags();
arsenmUnsubmitted
Done
ReplyInline Actions

All these auto substitutions are unrelated

arsenm: All these auto substitutions are unrelated
LLT ResTy = MRI.getType(Res); LLT ResTy = MRI.getType(Res);
LLT S32 = LLT::scalar(32); LLT S32 = LLT::scalar(32);
LLT S64 = LLT::scalar(64); LLT S64 = LLT::scalar(64);
const MachineFunction &MF = B.getMF(); const MachineFunction &MF = B.getMF();
bool Unsafe = bool Unsafe =
MF.getTarget().Options.UnsafeFPMath || MI.getFlag(MachineInstr::FmArcp); MF.getTarget().Options.UnsafeFPMath || MI.getFlag(MachineInstr::FmArcp);
▲ Show 20 Lines • Show All 165 Lines • ▼ Show 20 Lines B.buildIntrinsic(Intrinsic::amdgcn_div_fixup, Res, false)
.addUse(RHS) .addUse(RHS)
.addUse(LHS) .addUse(LHS)
.setMIFlags(Flags); .setMIFlags(Flags);
MI.eraseFromParent(); MI.eraseFromParent();
return true; return true;
} }
bool AMDGPULegalizerInfo::legalizeFDIV64(MachineInstr &MI,
MachineRegisterInfo &MRI,
MachineIRBuilder &B) const {
B.setInstr(MI);
Register Res = MI.getOperand(0).getReg();
Register LHS = MI.getOperand(1).getReg();
Register RHS = MI.getOperand(2).getReg();
uint16_t Flags = MI.getFlags();
LLT S64 = LLT::scalar(64);
LLT S1 = LLT::scalar(1);
auto One = B.buildFConstant(S64, 1.0);
auto DivScale0 = B.buildIntrinsic(Intrinsic::amdgcn_div_scale, {S64, S1}, false)
.addUse(RHS)
.addUse(RHS)
.addUse(LHS)
.setMIFlags(Flags);
auto NegDivScale0 = B.buildFNeg(S64, DivScale0.getReg(0), Flags);
auto Rcp = B.buildIntrinsic(Intrinsic::amdgcn_rcp, {S64}, false)
.addUse(DivScale0.getReg(0))
.setMIFlags(Flags);
auto Fma0 = B.buildFMA(S64, NegDivScale0, Rcp, One, Flags);
auto Fma1 = B.buildFMA(S64, Rcp, Fma0, Rcp, Flags);
auto Fma2 = B.buildFMA(S64, NegDivScale0, Fma1, One, Flags);
auto DivScale1 = B.buildIntrinsic(Intrinsic::amdgcn_div_scale, {S64, S1}, false)
.addUse(LHS)
.addUse(RHS)
.addUse(LHS)
.setMIFlags(Flags);
auto Fma3 = B.buildFMA(S64, Fma1, Fma2, Fma1, Flags);
auto Mul = B.buildMul(S64, DivScale1.getReg(0), Fma3, Flags);
auto Fma4 = B.buildFMA(S64, NegDivScale0, Mul, DivScale1.getReg(0), Flags);
Register Scale;
if (!ST.hasUsableDivScaleConditionOutput()) {
// Workaround a hardware bug on SI where the condition output from div_scale
// is not usable.
Scale = MRI.createGenericVirtualRegister(S1);
LLT S32 = LLT::scalar(32);
auto NumUnmerge = B.buildUnmerge(S32, LHS);
auto DenUnmerge = B.buildUnmerge(S32, RHS);
auto Scale0Unmerge = B.buildUnmerge(S32, DivScale0);
auto Scale1Unmerge = B.buildUnmerge(S32, DivScale1);
auto CmpNum = B.buildICmp(ICmpInst::ICMP_EQ, S1, NumUnmerge.getReg(1),
Scale1Unmerge.getReg(1));
auto CmpDen = B.buildICmp(ICmpInst::ICMP_EQ, S1, DenUnmerge.getReg(1),
Scale0Unmerge.getReg(1));
B.buildXor(Scale, CmpNum, CmpDen);
} else {
Scale = DivScale1.getReg(1);
arsenmUnsubmitted
Done
ReplyInline Actions

The ugliness of casting to a vector and extracting the elements isn't necessary. This isn't the canonical way to handle this in GlobalISel. You can use G_UNMERGE_VALUES to 2 S32 pieces more naturally

arsenm: The ugliness of casting to a vector and extracting the elements isn't necessary. This isn't the…
}
auto Fmas = B.buildIntrinsic(Intrinsic::amdgcn_div_fmas, {S64}, false)
.addUse(Fma4.getReg(0))
.addUse(Fma3.getReg(0))
.addUse(Mul.getReg(0))
.addUse(Scale)
.setMIFlags(Flags);
B.buildIntrinsic(Intrinsic::amdgcn_div_fixup, {S64}, false)
.addDef(Res)
.addUse(Fmas.getReg(0))
.addUse(RHS)
.addUse(LHS)
.setMIFlags(Flags);
MI.eraseFromParent();
return true;
}
bool AMDGPULegalizerInfo::legalizeFDIVFastIntrin(MachineInstr &MI, bool AMDGPULegalizerInfo::legalizeFDIVFastIntrin(MachineInstr &MI,
MachineRegisterInfo &MRI, MachineRegisterInfo &MRI,
MachineIRBuilder &B) const { MachineIRBuilder &B) const {
B.setInstr(MI); B.setInstr(MI);
Register Res = MI.getOperand(0).getReg(); Register Res = MI.getOperand(0).getReg();
Register LHS = MI.getOperand(2).getReg(); Register LHS = MI.getOperand(2).getReg();
Register RHS = MI.getOperand(3).getReg(); Register RHS = MI.getOperand(3).getReg();
uint16_t Flags = MI.getFlags(); uint16_t Flags = MI.getFlags();
▲ Show 20 Lines • Show All 232 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fdiv.mir

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI %s # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -enable-unsafe-fp-math -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9-UNSAFE %s # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -enable-unsafe-fp-math -o - %s | FileCheck -check-prefix=GFX9-UNSAFE %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -run-pass=legalizer -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -run-pass=legalizer -o - %s | FileCheck -check-prefix=GFX10 %s
--- ---
name: test_fdiv_s16 name: test_fdiv_s16
body: | body: |
bb.0: bb.0:
liveins: $vgpr0, $vgpr1 liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_fdiv_s16 ; SI-LABEL: name: test_fdiv_s16
▲ Show 20 Lines • Show All 218 Lines • ▼ Show 20 Lines
name: test_fdiv_s64 name: test_fdiv_s64
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3 liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_fdiv_s64 ; SI-LABEL: name: test_fdiv_s64
; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
; SI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[COPY]], [[COPY1]] ; SI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; SI: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; SI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY1]](s64), [[COPY1]](s64), [[COPY]](s64)
; SI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; SI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; SI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; SI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; SI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY1]](s64), [[COPY]](s64)
; SI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; SI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; SI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; SI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; SI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; SI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; SI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; SI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
; SI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](s64)
; SI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](s64)
; SI: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV5]](s32), [[UV11]]
; SI: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV7]](s32), [[UV9]]
; SI: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[ICMP]], [[ICMP1]]
; SI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[XOR]](s1)
; SI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY1]](s64), [[COPY]](s64)
; SI: $vgpr0_vgpr1 = COPY %2(s64)
; VI-LABEL: name: test_fdiv_s64 ; VI-LABEL: name: test_fdiv_s64
; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
; VI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[COPY]], [[COPY1]] ; VI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; VI: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; VI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY1]](s64), [[COPY1]](s64), [[COPY]](s64)
; VI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; VI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; VI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; VI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; VI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY1]](s64), [[COPY]](s64)
; VI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; VI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; VI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; VI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; VI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; VI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; VI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; VI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; VI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; VI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY1]](s64), [[COPY]](s64)
; VI: $vgpr0_vgpr1 = COPY %2(s64)
; GFX9-LABEL: name: test_fdiv_s64 ; GFX9-LABEL: name: test_fdiv_s64
; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
; GFX9: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[COPY]], [[COPY1]] ; GFX9: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX9: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; GFX9: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY1]](s64), [[COPY1]](s64), [[COPY]](s64)
; GFX9: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX9: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; GFX9: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX9: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; GFX9: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY1]](s64), [[COPY]](s64)
; GFX9: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; GFX9: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX9: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX9: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX9: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX9: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY1]](s64), [[COPY]](s64)
; GFX9: $vgpr0_vgpr1 = COPY %2(s64)
; GFX9-UNSAFE-LABEL: name: test_fdiv_s64 ; GFX9-UNSAFE-LABEL: name: test_fdiv_s64
; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[COPY1]](s64) ; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[COPY1]](s64)
; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[COPY]], [[INT]] ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[COPY]], [[INT]]
; GFX9-UNSAFE: $vgpr0_vgpr1 = COPY [[FMUL]](s64) ; GFX9-UNSAFE: $vgpr0_vgpr1 = COPY [[FMUL]](s64)
; GFX10-LABEL: name: test_fdiv_s64 ; GFX10-LABEL: name: test_fdiv_s64
; GFX10: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX10: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3 ; GFX10: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
; GFX10: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[COPY]], [[COPY1]] ; GFX10: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX10: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; GFX10: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY1]](s64), [[COPY1]](s64), [[COPY]](s64)
; GFX10: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX10: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX10: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; GFX10: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX10: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; GFX10: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY1]](s64), [[COPY]](s64)
; GFX10: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX10: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX10: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; GFX10: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; GFX10: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; GFX10: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; GFX10: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX10: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX10: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX10: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX10: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY1]](s64), [[COPY]](s64)
; GFX10: $vgpr0_vgpr1 = COPY %2(s64)
%0:_(s64) = COPY $vgpr0_vgpr1 %0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s64) = COPY $vgpr2_vgpr3 %1:_(s64) = COPY $vgpr2_vgpr3
%2:_(s64) = G_FDIV %0, %1 %2:_(s64) = G_FDIV %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...
--- ---
name: test_fdiv_v2s32 name: test_fdiv_v2s32
▲ Show 20 Lines • Show All 560 Lines • ▼ Show 20 Linesbody: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7 liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7
; SI-LABEL: name: test_fdiv_v2s64 ; SI-LABEL: name: test_fdiv_v2s64
; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 ; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7 ; SI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>) ; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>) ; SI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
; SI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[UV]], [[UV2]] ; SI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; SI: [[FDIV1:%[0-9]+]]:_(s64) = G_FDIV [[UV1]], [[UV3]] ; SI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV2]](s64), [[UV2]](s64), [[UV]](s64)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FDIV]](s64), [[FDIV1]](s64) ; SI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; SI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; SI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; SI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; SI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV]](s64), [[UV2]](s64), [[UV]](s64)
; SI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; SI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV6]]
; SI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV5]], [[UV6]]
; SI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV7]]
; SI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV4]], [[UV6]]
; SI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; SI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; SI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; SI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV]](s64)
; SI: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV2]](s64)
; SI: [[UV12:%[0-9]+]]:_(s32), [[UV13:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](s64)
; SI: [[UV14:%[0-9]+]]:_(s32), [[UV15:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV9]](s32), [[UV15]]
; SI: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV11]](s32), [[UV13]]
; SI: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[ICMP]], [[ICMP1]]
; SI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[XOR]](s1)
; SI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %7(s64), [[INT5]](s64), [[UV2]](s64), [[UV]](s64)
; SI: [[INT7:%[0-9]+]]:_(s64), [[INT8:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV3]](s64), [[UV3]](s64), [[UV1]](s64)
; SI: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[INT7]]
; SI: [[INT9:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT7]](s64)
; SI: [[FMA5:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[INT9]], [[C]]
; SI: [[FMA6:%[0-9]+]]:_(s64) = G_FMA [[INT9]], [[FMA5]], [[INT9]]
; SI: [[FMA7:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[FMA6]], [[C]]
; SI: [[INT10:%[0-9]+]]:_(s64), [[INT11:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV1]](s64), [[UV3]](s64), [[UV1]](s64)
; SI: [[FMA8:%[0-9]+]]:_(s64) = G_FMA [[FMA6]], [[FMA7]], [[FMA6]]
; SI: [[UV16:%[0-9]+]]:_(s32), [[UV17:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT10]](s64)
; SI: [[UV18:%[0-9]+]]:_(s32), [[UV19:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA8]](s64)
; SI: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[UV16]], [[UV18]]
; SI: [[MUL4:%[0-9]+]]:_(s32) = G_MUL [[UV17]], [[UV18]]
; SI: [[MUL5:%[0-9]+]]:_(s32) = G_MUL [[UV16]], [[UV19]]
; SI: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV16]], [[UV18]]
; SI: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[MUL4]], [[MUL5]]
; SI: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[ADD2]], [[UMULH1]]
; SI: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL3]](s32), [[ADD3]](s32)
; SI: [[FMA9:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[MV1]], [[INT10]]
; SI: [[UV20:%[0-9]+]]:_(s32), [[UV21:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV1]](s64)
; SI: [[UV22:%[0-9]+]]:_(s32), [[UV23:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[UV3]](s64)
; SI: [[UV24:%[0-9]+]]:_(s32), [[UV25:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT7]](s64)
; SI: [[UV26:%[0-9]+]]:_(s32), [[UV27:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT10]](s64)
; SI: [[ICMP2:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV21]](s32), [[UV27]]
; SI: [[ICMP3:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV23]](s32), [[UV25]]
; SI: [[XOR1:%[0-9]+]]:_(s1) = G_XOR [[ICMP2]], [[ICMP3]]
; SI: [[INT12:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA9]](s64), [[FMA8]](s64), [[MV1]](s64), [[XOR1]](s1)
; SI: [[INT13:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %8(s64), [[INT12]](s64), [[UV3]](s64), [[UV1]](s64)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR %7(s64), %8(s64)
; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>) ; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
; VI-LABEL: name: test_fdiv_v2s64 ; VI-LABEL: name: test_fdiv_v2s64
; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 ; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7 ; VI: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>) ; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>) ; VI: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
; VI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[UV]], [[UV2]] ; VI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; VI: [[FDIV1:%[0-9]+]]:_(s64) = G_FDIV [[UV1]], [[UV3]] ; VI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV2]](s64), [[UV2]](s64), [[UV]](s64)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FDIV]](s64), [[FDIV1]](s64) ; VI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; VI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; VI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; VI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; VI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV]](s64), [[UV2]](s64), [[UV]](s64)
; VI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; VI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; VI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; VI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV6]]
; VI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV5]], [[UV6]]
; VI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV7]]
; VI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV4]], [[UV6]]
; VI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; VI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; VI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; VI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; VI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %7(s64), [[INT5]](s64), [[UV2]](s64), [[UV]](s64)
; VI: [[INT7:%[0-9]+]]:_(s64), [[INT8:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV3]](s64), [[UV3]](s64), [[UV1]](s64)
; VI: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[INT7]]
; VI: [[INT9:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT7]](s64)
; VI: [[FMA5:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[INT9]], [[C]]
; VI: [[FMA6:%[0-9]+]]:_(s64) = G_FMA [[INT9]], [[FMA5]], [[INT9]]
; VI: [[FMA7:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[FMA6]], [[C]]
; VI: [[INT10:%[0-9]+]]:_(s64), [[INT11:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV1]](s64), [[UV3]](s64), [[UV1]](s64)
; VI: [[FMA8:%[0-9]+]]:_(s64) = G_FMA [[FMA6]], [[FMA7]], [[FMA6]]
; VI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT10]](s64)
; VI: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA8]](s64)
; VI: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[UV8]], [[UV10]]
; VI: [[MUL4:%[0-9]+]]:_(s32) = G_MUL [[UV9]], [[UV10]]
; VI: [[MUL5:%[0-9]+]]:_(s32) = G_MUL [[UV8]], [[UV11]]
; VI: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV8]], [[UV10]]
; VI: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[MUL4]], [[MUL5]]
; VI: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[ADD2]], [[UMULH1]]
; VI: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL3]](s32), [[ADD3]](s32)
; VI: [[FMA9:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[MV1]], [[INT10]]
; VI: [[INT12:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA9]](s64), [[FMA8]](s64), [[MV1]](s64), [[INT11]](s1)
; VI: [[INT13:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %8(s64), [[INT12]](s64), [[UV3]](s64), [[UV1]](s64)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR %7(s64), %8(s64)
; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>) ; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
; GFX9-LABEL: name: test_fdiv_v2s64 ; GFX9-LABEL: name: test_fdiv_v2s64
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 ; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7 ; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>) ; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>) ; GFX9: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
; GFX9: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[UV]], [[UV2]] ; GFX9: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX9: [[FDIV1:%[0-9]+]]:_(s64) = G_FDIV [[UV1]], [[UV3]] ; GFX9: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV2]](s64), [[UV2]](s64), [[UV]](s64)
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FDIV]](s64), [[FDIV1]](s64) ; GFX9: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX9: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; GFX9: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX9: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; GFX9: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV]](s64), [[UV2]](s64), [[UV]](s64)
; GFX9: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX9: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX9: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV6]]
; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV5]], [[UV6]]
; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV7]]
; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV4]], [[UV6]]
; GFX9: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX9: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX9: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX9: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX9: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %7(s64), [[INT5]](s64), [[UV2]](s64), [[UV]](s64)
; GFX9: [[INT7:%[0-9]+]]:_(s64), [[INT8:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV3]](s64), [[UV3]](s64), [[UV1]](s64)
; GFX9: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[INT7]]
; GFX9: [[INT9:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT7]](s64)
; GFX9: [[FMA5:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[INT9]], [[C]]
; GFX9: [[FMA6:%[0-9]+]]:_(s64) = G_FMA [[INT9]], [[FMA5]], [[INT9]]
; GFX9: [[FMA7:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[FMA6]], [[C]]
; GFX9: [[INT10:%[0-9]+]]:_(s64), [[INT11:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV1]](s64), [[UV3]](s64), [[UV1]](s64)
; GFX9: [[FMA8:%[0-9]+]]:_(s64) = G_FMA [[FMA6]], [[FMA7]], [[FMA6]]
; GFX9: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT10]](s64)
; GFX9: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA8]](s64)
; GFX9: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[UV8]], [[UV10]]
; GFX9: [[MUL4:%[0-9]+]]:_(s32) = G_MUL [[UV9]], [[UV10]]
; GFX9: [[MUL5:%[0-9]+]]:_(s32) = G_MUL [[UV8]], [[UV11]]
; GFX9: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV8]], [[UV10]]
; GFX9: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[MUL4]], [[MUL5]]
; GFX9: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[ADD2]], [[UMULH1]]
; GFX9: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL3]](s32), [[ADD3]](s32)
; GFX9: [[FMA9:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[MV1]], [[INT10]]
; GFX9: [[INT12:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA9]](s64), [[FMA8]](s64), [[MV1]](s64), [[INT11]](s1)
; GFX9: [[INT13:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %8(s64), [[INT12]](s64), [[UV3]](s64), [[UV1]](s64)
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR %7(s64), %8(s64)
; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>) ; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
; GFX9-UNSAFE-LABEL: name: test_fdiv_v2s64 ; GFX9-UNSAFE-LABEL: name: test_fdiv_v2s64
; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7 ; GFX9-UNSAFE: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>) ; GFX9-UNSAFE: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; GFX9-UNSAFE: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>) ; GFX9-UNSAFE: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[UV2]](s64) ; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[UV2]](s64)
; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[UV]], [[INT]] ; GFX9-UNSAFE: [[FMUL:%[0-9]+]]:_(s64) = G_FMUL [[UV]], [[INT]]
; GFX9-UNSAFE: [[INT1:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[UV3]](s64) ; GFX9-UNSAFE: [[INT1:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[UV3]](s64)
; GFX9-UNSAFE: [[FMUL1:%[0-9]+]]:_(s64) = G_FMUL [[UV1]], [[INT1]] ; GFX9-UNSAFE: [[FMUL1:%[0-9]+]]:_(s64) = G_FMUL [[UV1]], [[INT1]]
; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMUL]](s64), [[FMUL1]](s64) ; GFX9-UNSAFE: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMUL]](s64), [[FMUL1]](s64)
; GFX9-UNSAFE: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>) ; GFX9-UNSAFE: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
; GFX10-LABEL: name: test_fdiv_v2s64 ; GFX10-LABEL: name: test_fdiv_v2s64
; GFX10: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 ; GFX10: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; GFX10: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7 ; GFX10: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
; GFX10: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>) ; GFX10: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; GFX10: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>) ; GFX10: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
; GFX10: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[UV]], [[UV2]] ; GFX10: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX10: [[FDIV1:%[0-9]+]]:_(s64) = G_FDIV [[UV1]], [[UV3]] ; GFX10: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV2]](s64), [[UV2]](s64), [[UV]](s64)
; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FDIV]](s64), [[FDIV1]](s64) ; GFX10: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX10: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX10: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; GFX10: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX10: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; GFX10: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV]](s64), [[UV2]](s64), [[UV]](s64)
; GFX10: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX10: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX10: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX10: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV6]]
; GFX10: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV5]], [[UV6]]
; GFX10: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV4]], [[UV7]]
; GFX10: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV4]], [[UV6]]
; GFX10: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX10: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX10: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX10: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX10: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %7(s64), [[INT5]](s64), [[UV2]](s64), [[UV]](s64)
; GFX10: [[INT7:%[0-9]+]]:_(s64), [[INT8:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV3]](s64), [[UV3]](s64), [[UV1]](s64)
; GFX10: [[FNEG1:%[0-9]+]]:_(s64) = G_FNEG [[INT7]]
; GFX10: [[INT9:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT7]](s64)
; GFX10: [[FMA5:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[INT9]], [[C]]
; GFX10: [[FMA6:%[0-9]+]]:_(s64) = G_FMA [[INT9]], [[FMA5]], [[INT9]]
; GFX10: [[FMA7:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[FMA6]], [[C]]
; GFX10: [[INT10:%[0-9]+]]:_(s64), [[INT11:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[UV1]](s64), [[UV3]](s64), [[UV1]](s64)
; GFX10: [[FMA8:%[0-9]+]]:_(s64) = G_FMA [[FMA6]], [[FMA7]], [[FMA6]]
; GFX10: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT10]](s64)
; GFX10: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA8]](s64)
; GFX10: [[MUL3:%[0-9]+]]:_(s32) = G_MUL [[UV8]], [[UV10]]
; GFX10: [[MUL4:%[0-9]+]]:_(s32) = G_MUL [[UV9]], [[UV10]]
; GFX10: [[MUL5:%[0-9]+]]:_(s32) = G_MUL [[UV8]], [[UV11]]
; GFX10: [[UMULH1:%[0-9]+]]:_(s32) = G_UMULH [[UV8]], [[UV10]]
; GFX10: [[ADD2:%[0-9]+]]:_(s32) = G_ADD [[MUL4]], [[MUL5]]
; GFX10: [[ADD3:%[0-9]+]]:_(s32) = G_ADD [[ADD2]], [[UMULH1]]
; GFX10: [[MV1:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL3]](s32), [[ADD3]](s32)
; GFX10: [[FMA9:%[0-9]+]]:_(s64) = G_FMA [[FNEG1]], [[MV1]], [[INT10]]
; GFX10: [[INT12:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA9]](s64), [[FMA8]](s64), [[MV1]](s64), [[INT11]](s1)
; GFX10: [[INT13:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %8(s64), [[INT12]](s64), [[UV3]](s64), [[UV1]](s64)
; GFX10: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR %7(s64), %8(s64)
; GFX10: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>) ; GFX10: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
%0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3 %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7 %1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
%2:_(<2 x s64>) = G_FDIV %0, %1 %2:_(<2 x s64>) = G_FDIV %0, %1
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %2 $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %2
... ...
--- ---
▲ Show 20 Lines • Show All 929 Lines • ▼ Show 20 Lines
name: test_fdiv_s64_constant_one_rcp name: test_fdiv_s64_constant_one_rcp
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fdiv_s64_constant_one_rcp ; SI-LABEL: name: test_fdiv_s64_constant_one_rcp
; SI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00 ; SI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; SI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; SI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; SI: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; SI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; SI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; SI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; SI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; SI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; SI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; SI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; SI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; SI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; SI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; SI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; SI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; SI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[C]](s64)
; SI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
; SI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](s64)
; SI: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV5]](s32), [[UV11]]
; SI: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV7]](s32), [[UV9]]
; SI: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[ICMP]], [[ICMP1]]
; SI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[XOR]](s1)
; SI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; SI: $vgpr0_vgpr1 = COPY %2(s64)
; VI-LABEL: name: test_fdiv_s64_constant_one_rcp ; VI-LABEL: name: test_fdiv_s64_constant_one_rcp
; VI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00 ; VI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; VI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; VI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; VI: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; VI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; VI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; VI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; VI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; VI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; VI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; VI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; VI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; VI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; VI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; VI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; VI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; VI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; VI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; VI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; VI: $vgpr0_vgpr1 = COPY %2(s64)
; GFX9-LABEL: name: test_fdiv_s64_constant_one_rcp ; GFX9-LABEL: name: test_fdiv_s64_constant_one_rcp
; GFX9: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00 ; GFX9: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; GFX9: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; GFX9: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; GFX9: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX9: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; GFX9: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX9: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; GFX9: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; GFX9: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; GFX9: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX9: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX9: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX9: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX9: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; GFX9: $vgpr0_vgpr1 = COPY %2(s64)
; GFX9-UNSAFE-LABEL: name: test_fdiv_s64_constant_one_rcp ; GFX9-UNSAFE-LABEL: name: test_fdiv_s64_constant_one_rcp
; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[COPY]](s64) ; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[COPY]](s64)
; GFX9-UNSAFE: $vgpr0_vgpr1 = COPY [[INT]](s64) ; GFX9-UNSAFE: $vgpr0_vgpr1 = COPY [[INT]](s64)
; GFX10-LABEL: name: test_fdiv_s64_constant_one_rcp ; GFX10-LABEL: name: test_fdiv_s64_constant_one_rcp
; GFX10: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00 ; GFX10: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX10: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX10: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX10: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; GFX10: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; GFX10: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; GFX10: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX10: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX10: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C]]
; GFX10: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX10: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C]]
; GFX10: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; GFX10: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX10: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX10: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; GFX10: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; GFX10: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; GFX10: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; GFX10: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX10: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX10: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX10: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX10: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; GFX10: $vgpr0_vgpr1 = COPY %2(s64)
%0:_(s64) = G_FCONSTANT double 1.0 %0:_(s64) = G_FCONSTANT double 1.0
%1:_(s64) = COPY $vgpr0_vgpr1 %1:_(s64) = COPY $vgpr0_vgpr1
%2:_(s64) = G_FDIV %0, %1 %2:_(s64) = G_FDIV %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...
--- ---
name: test_fdiv_s64_constant_negative_one_rcp name: test_fdiv_s64_constant_negative_one_rcp
body: | body: |
bb.0: bb.0:
liveins: $vgpr0_vgpr1 liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fdiv_s64_constant_negative_one_rcp ; SI-LABEL: name: test_fdiv_s64_constant_negative_one_rcp
; SI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00 ; SI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00
; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; SI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; SI: [[C1:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; SI: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; SI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; SI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; SI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; SI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C1]]
; SI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; SI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C1]]
; SI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; SI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; SI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; SI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; SI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; SI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; SI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; SI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; SI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; SI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; SI: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[C]](s64)
; SI: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
; SI: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT]](s64)
; SI: [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV5]](s32), [[UV11]]
; SI: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(eq), [[UV7]](s32), [[UV9]]
; SI: [[XOR:%[0-9]+]]:_(s1) = G_XOR [[ICMP]], [[ICMP1]]
; SI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[XOR]](s1)
; SI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; SI: $vgpr0_vgpr1 = COPY %2(s64)
; VI-LABEL: name: test_fdiv_s64_constant_negative_one_rcp ; VI-LABEL: name: test_fdiv_s64_constant_negative_one_rcp
; VI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00 ; VI: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00
; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; VI: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; VI: [[C1:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; VI: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; VI: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; VI: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; VI: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; VI: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C1]]
; VI: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; VI: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C1]]
; VI: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; VI: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; VI: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; VI: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; VI: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; VI: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; VI: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; VI: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; VI: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; VI: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; VI: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; VI: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; VI: $vgpr0_vgpr1 = COPY %2(s64)
; GFX9-LABEL: name: test_fdiv_s64_constant_negative_one_rcp ; GFX9-LABEL: name: test_fdiv_s64_constant_negative_one_rcp
; GFX9: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00 ; GFX9: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00
; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; GFX9: [[C1:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX9: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; GFX9: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; GFX9: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX9: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX9: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C1]]
; GFX9: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX9: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C1]]
; GFX9: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; GFX9: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX9: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; GFX9: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; GFX9: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; GFX9: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; GFX9: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX9: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX9: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX9: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX9: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX9: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; GFX9: $vgpr0_vgpr1 = COPY %2(s64)
; GFX9-UNSAFE-LABEL: name: test_fdiv_s64_constant_negative_one_rcp ; GFX9-UNSAFE-LABEL: name: test_fdiv_s64_constant_negative_one_rcp
; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX9-UNSAFE: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9-UNSAFE: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[COPY]] ; GFX9-UNSAFE: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[COPY]]
; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[FNEG]](s64) ; GFX9-UNSAFE: [[INT:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[FNEG]](s64)
; GFX9-UNSAFE: $vgpr0_vgpr1 = COPY [[INT]](s64) ; GFX9-UNSAFE: $vgpr0_vgpr1 = COPY [[INT]](s64)
; GFX10-LABEL: name: test_fdiv_s64_constant_negative_one_rcp ; GFX10-LABEL: name: test_fdiv_s64_constant_negative_one_rcp
; GFX10: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00 ; GFX10: [[C:%[0-9]+]]:_(s64) = G_FCONSTANT double -1.000000e+00
; GFX10: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1 ; GFX10: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX10: [[FDIV:%[0-9]+]]:_(s64) = G_FDIV [[C]], [[COPY]] ; GFX10: [[C1:%[0-9]+]]:_(s64) = G_FCONSTANT double 1.000000e+00
; GFX10: $vgpr0_vgpr1 = COPY [[FDIV]](s64) ; GFX10: [[INT:%[0-9]+]]:_(s64), [[INT1:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[COPY]](s64), [[COPY]](s64), [[C]](s64)
; GFX10: [[FNEG:%[0-9]+]]:_(s64) = G_FNEG [[INT]]
; GFX10: [[INT2:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.rcp), [[INT]](s64)
; GFX10: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[INT2]], [[C1]]
; GFX10: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[INT2]], [[FMA]], [[INT2]]
; GFX10: [[FMA2:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[FMA1]], [[C1]]
; GFX10: [[INT3:%[0-9]+]]:_(s64), [[INT4:%[0-9]+]]:_(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.scale), [[C]](s64), [[COPY]](s64), [[C]](s64)
; GFX10: [[FMA3:%[0-9]+]]:_(s64) = G_FMA [[FMA1]], [[FMA2]], [[FMA1]]
; GFX10: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[INT3]](s64)
; GFX10: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[FMA3]](s64)
; GFX10: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV2]]
; GFX10: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[UV1]], [[UV2]]
; GFX10: [[MUL2:%[0-9]+]]:_(s32) = G_MUL [[UV]], [[UV3]]
; GFX10: [[UMULH:%[0-9]+]]:_(s32) = G_UMULH [[UV]], [[UV2]]
; GFX10: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[MUL1]], [[MUL2]]
; GFX10: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ADD]], [[UMULH]]
; GFX10: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[MUL]](s32), [[ADD1]](s32)
; GFX10: [[FMA4:%[0-9]+]]:_(s64) = G_FMA [[FNEG]], [[MV]], [[INT3]]
; GFX10: [[INT5:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fmas), [[FMA4]](s64), [[FMA3]](s64), [[MV]](s64), [[INT4]](s1)
; GFX10: [[INT6:%[0-9]+]]:_(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.div.fixup), def %2(s64), [[INT5]](s64), [[COPY]](s64), [[C]](s64)
; GFX10: $vgpr0_vgpr1 = COPY %2(s64)
%0:_(s64) = G_FCONSTANT double -1.0 %0:_(s64) = G_FCONSTANT double -1.0
%1:_(s64) = COPY $vgpr0_vgpr1 %1:_(s64) = COPY $vgpr0_vgpr1
%2:_(s64) = G_FDIV %0, %1 %2:_(s64) = G_FDIV %0, %1
$vgpr0_vgpr1 = COPY %2 $vgpr0_vgpr1 = COPY %2
... ...