Skip to content

[X86][GISEL] - Legalize G_FPTOUI & G_UITOFP for X87 #155562

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 3 commits into
base: main
Choose a base branch
from
+732 −74
Draft

[X86][GISEL] - Legalize G_FPTOUI & G_UITOFP for X87 #155562

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fbce4a4
[X86][GISEL] - Legalize G_FPTOUI for X87
pawan-nirpal-031 Aug 27, 2025
9bf22ad
[X86][GISEL] - Legalize G_UITOFP in X87 mode
pawan-nirpal-031 Sep 1, 2025
cccb00f
Merge branch 'llvm:main' into main
pawan-nirpal-031 Oct 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
98 changes: 86 additions & 12 deletions llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -485,10 +485,13 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
getActionDefinitionsBuilder(G_UITOFP)
.legalFor(HasAVX512, {{s32, s32}, {s32, s64}, {s64, s32}, {s64, s64}})
.customIf([=](const LegalityQuery &Query) {
return !HasAVX512 &&
((HasSSE1 && typeIs(0, s32)(Query)) ||
(HasSSE2 && typeIs(0, s64)(Query))) &&
scalarNarrowerThan(1, Is64Bit ? 64 : 32)(Query);
if (!HasAVX512 &&
((HasSSE1 && typeIs(0, s32)(Query)) ||
(HasSSE2 && typeIs(0, s64)(Query))) &&
scalarNarrowerThan(1, Is64Bit ? 64 : 32)(Query))
return true;
return typeInSet(0, {s32, s64, s80})(Query) &&
typeInSet(1, {s8, s16, s32, s64})(Query);
})
.lowerIf([=](const LegalityQuery &Query) {
// Lower conversions from s64
Expand All @@ -503,12 +506,17 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
.widenScalarToNextPow2(1);

getActionDefinitionsBuilder(G_FPTOUI)
.legalFor(HasAVX512, {{s32, s32}, {s32, s64}, {s64, s32}, {s64, s64}})
.legalIf([=](const LegalityQuery &Query) {
return HasAVX512 && typeInSet(0, {s32, s64})(Query) &&
typeInSet(1, {s32, s64})(Query);
})
Comment on lines +509 to +512
Copy link
Contributor

@e-kud e-kud Oct 26, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it is the part of code before refactoring legalizer to legalFor usage. We should be fine with existing legalFor.

.customIf([=](const LegalityQuery &Query) {
return !HasAVX512 &&
((HasSSE1 && typeIs(1, s32)(Query)) ||
(HasSSE2 && typeIs(1, s64)(Query))) &&
scalarNarrowerThan(0, Is64Bit ? 64 : 32)(Query);
if (!HasAVX512 && (((typeIs(1, s32)(Query) && HasSSE1) ||
(typeIs(1, s64)(Query) && HasSSE2)) &&
scalarNarrowerThan(0, Is64Bit ? 64 : 32)(Query)))
return true;
return typeInSet(0, {s16, s32, s64})(Query) &&
typeInSet(1, {s32, s64, s80})(Query);
})
// TODO: replace with customized legalization using
// specifics of cvttsd2si. The selection of this node requires
Expand All @@ -520,7 +528,7 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
(HasSSE2 && typeIs(1, s64)(Query))) &&
(Is64Bit && typeIs(0, s64)(Query));
})
.clampScalar(0, s32, sMaxScalar)
.clampScalar(0, s16, sMaxScalar)
Copy link
Contributor

@e-kud e-kud Oct 26, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe clamp to s16 only if no SSE available?

.widenScalarToNextPow2(0)
.clampScalar(1, s32, HasSSE2 ? s64 : s32)
.widenScalarToNextPow2(1);
Expand Down Expand Up @@ -751,7 +759,68 @@ bool X86LegalizerInfo::legalizeFPTOUI(MachineInstr &MI,
return true;
}

return false;
MachineFunction &MF = *MI.getMF();
TypeSize MemSize = DstTy.getSizeInBytes();
MachinePointerInfo PtrInfo;
Align Alignmt = Helper.getStackTemporaryAlignment(DstTy);
auto SlotPointer = Helper.createStackTemporary(MemSize, Alignmt, PtrInfo);

if (DstTy == s64) {
// As x87 cmov is unsupported at the momement, we need to use the following
// arithmatic algorithm to do fptoui computation. y_tmp = y - sint_max.
// y_res = fptosi(y - !sign(y_tmp)sint_max) + !sign(y_tmp)sint_max

const LLT s1 = LLT::scalar(1);
const LLT s80 = LLT::scalar(80);
const llvm::fltSemantics &Sem =
(SrcTy != s80 ? getFltSemanticForLLT(SrcTy)
: APFloat::x87DoubleExtended());

APFloat SintMaxFP = APFloat::getZero(Sem);
SintMaxFP.convertFromAPInt(APInt::getOneBitSet(64, 63), false,
APFloat::rmNearestTiesToEven);
Register SintMax = MIRBuilder.buildFConstant(SrcTy, SintMaxFP).getReg(0);

Register YTmp = MIRBuilder.buildFSub(SrcTy, Src, SintMax).getReg(0);

APFloat ZeroFP = APFloat::getZero(Sem);
Register Zero = MIRBuilder.buildFConstant(SrcTy, ZeroFP).getReg(0);
Register Sign =
MIRBuilder.buildFCmp(CmpInst::FCMP_OLT, s1, YTmp, Zero).getReg(0);

Register One = MIRBuilder.buildConstant(s1, 1).getReg(0);
Register NSign = MIRBuilder.buildXor(s1, Sign, One).getReg(0);

Register NSign64 = MIRBuilder.buildZExt(s64, NSign).getReg(0);

Register NSignF = MIRBuilder.buildSITOFP(SrcTy, NSign64).getReg(0);
Register Offset = MIRBuilder.buildFMul(SrcTy, NSignF, SintMax).getReg(0);

Register YAdj = MIRBuilder.buildFSub(SrcTy, Src, Offset).getReg(0);

Register YAdjI = MIRBuilder.buildFPTOSI(s64, YAdj).getReg(0);

Register OffsetI = MIRBuilder.buildFPTOSI(s64, Offset).getReg(0);

Register Result = MIRBuilder.buildAdd(s64, YAdjI, OffsetI).getReg(0);

MIRBuilder.buildCopy(Dst, Result);
MI.eraseFromParent();
return true;
}

MachineMemOperand *StoreMMO = MF.getMachineMemOperand(
PtrInfo, MachineMemOperand::MOStore, MemSize, Align(MemSize));

MIRBuilder.buildInstr(X86::G_FIST)
.addUse(Src)
.addUse(SlotPointer.getReg(0))
.addMemOperand(StoreMMO);

MIRBuilder.buildLoad(Dst, SlotPointer, PtrInfo, Align(MemSize));
MI.eraseFromParent();

return true;
}

bool X86LegalizerInfo::legalizeUITOFP(MachineInstr &MI,
Expand All @@ -764,7 +833,12 @@ bool X86LegalizerInfo::legalizeUITOFP(MachineInstr &MI,

// Simply reuse SITOFP when it is possible to widen the type
if (SrcTy.getSizeInBits() <= 32) {
auto Ext = MIRBuilder.buildZExt(SrcTy == s32 ? s64 : s32, Src);
const LLT s8 = LLT::scalar(8);
const LLT s16 = LLT::scalar(16);
auto Ext = MIRBuilder.buildZExt((SrcTy == s8 ? s16
: SrcTy == s32 ? s64
: s32),
Src);
MIRBuilder.buildSITOFP(Dst, Ext);
MI.eraseFromParent();
return true;
Expand Down
111 changes: 49 additions & 62 deletions llvm/test/CodeGen/X86/isel-fp-to-int.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,18 +18,22 @@ define i64 @test_double_to_ui64(double %x) {
;
; GISEL-X64-LABEL: test_double_to_ui64:
; GISEL-X64: # %bb.0: # %entry
; GISEL-X64-NEXT: cvttsd2si %xmm0, %rcx
; GISEL-X64-NEXT: movsd {{.*#+}} xmm1 = [9.2233720368547758E+18,0.0E+0]
; GISEL-X64-NEXT: movapd %xmm0, %xmm2
; GISEL-X64-NEXT: subsd %xmm1, %xmm2
; GISEL-X64-NEXT: cvttsd2si %xmm2, %rdx
; GISEL-X64-NEXT: movabsq $-9223372036854775808, %rax # imm = 0x8000000000000000
; GISEL-X64-NEXT: xorq %rdx, %rax
; GISEL-X64-NEXT: xorl %edx, %edx
; GISEL-X64-NEXT: ucomisd %xmm1, %xmm0
; GISEL-X64-NEXT: setb %dl
; GISEL-X64-NEXT: andl $1, %edx
; GISEL-X64-NEXT: cmovneq %rcx, %rax
; GISEL-X64-NEXT: movsd {{.*#+}} xmm3 = [0.0E+0,0.0E+0]
; GISEL-X64-NEXT: ucomisd %xmm2, %xmm3
; GISEL-X64-NEXT: seta %al
; GISEL-X64-NEXT: xorb $1, %al
; GISEL-X64-NEXT: movzbl %al, %eax
; GISEL-X64-NEXT: andq $1, %rax
; GISEL-X64-NEXT: xorps %xmm2, %xmm2
; GISEL-X64-NEXT: cvtsi2sd %rax, %xmm2
; GISEL-X64-NEXT: mulsd %xmm1, %xmm2
; GISEL-X64-NEXT: subsd %xmm2, %xmm0
; GISEL-X64-NEXT: cvttsd2si %xmm0, %rcx
; GISEL-X64-NEXT: cvttsd2si %xmm2, %rax
; GISEL-X64-NEXT: addq %rcx, %rax
; GISEL-X64-NEXT: retq
;
; AVX512-LABEL: test_double_to_ui64:
Expand Down Expand Up @@ -64,17 +68,11 @@ define zeroext i16 @test_double_to_ui16(double %x) {
; X64-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NEXT: retq
;
; SDAG-AVX512-LABEL: test_double_to_ui16:
; SDAG-AVX512: # %bb.0: # %entry
; SDAG-AVX512-NEXT: vcvttsd2si %xmm0, %eax
; SDAG-AVX512-NEXT: # kill: def $ax killed $ax killed $eax
; SDAG-AVX512-NEXT: retq
;
; GISEL-AVX512-LABEL: test_double_to_ui16:
; GISEL-AVX512: # %bb.0: # %entry
; GISEL-AVX512-NEXT: vcvttsd2usi %xmm0, %eax
; GISEL-AVX512-NEXT: # kill: def $ax killed $ax killed $eax
; GISEL-AVX512-NEXT: retq
; AVX512-LABEL: test_double_to_ui16:
; AVX512: # %bb.0: # %entry
; AVX512-NEXT: vcvttsd2si %xmm0, %eax
; AVX512-NEXT: # kill: def $ax killed $ax killed $eax
; AVX512-NEXT: retq
entry:
%conv = fptoui double %x to i16
ret i16 %conv
Expand All @@ -87,17 +85,11 @@ define zeroext i8 @test_double_to_ui8(double %x) {
; X64-NEXT: # kill: def $al killed $al killed $eax
; X64-NEXT: retq
;
; SDAG-AVX512-LABEL: test_double_to_ui8:
; SDAG-AVX512: # %bb.0: # %entry
; SDAG-AVX512-NEXT: vcvttsd2si %xmm0, %eax
; SDAG-AVX512-NEXT: # kill: def $al killed $al killed $eax
; SDAG-AVX512-NEXT: retq
;
; GISEL-AVX512-LABEL: test_double_to_ui8:
; GISEL-AVX512: # %bb.0: # %entry
; GISEL-AVX512-NEXT: vcvttsd2usi %xmm0, %eax
; GISEL-AVX512-NEXT: # kill: def $al killed $al killed $eax
; GISEL-AVX512-NEXT: retq
; AVX512-LABEL: test_double_to_ui8:
; AVX512: # %bb.0: # %entry
; AVX512-NEXT: vcvttsd2si %xmm0, %eax
; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: retq
entry:
%conv = fptoui double %x to i8
ret i8 %conv
Expand All @@ -117,18 +109,22 @@ define i64 @test_float_to_ui64(float %x) {
;
; GISEL-X64-LABEL: test_float_to_ui64:
; GISEL-X64: # %bb.0: # %entry
; GISEL-X64-NEXT: cvttss2si %xmm0, %rcx
; GISEL-X64-NEXT: movss {{.*#+}} xmm1 = [9.22337203E+18,0.0E+0,0.0E+0,0.0E+0]
; GISEL-X64-NEXT: movaps %xmm0, %xmm2
; GISEL-X64-NEXT: subss %xmm1, %xmm2
; GISEL-X64-NEXT: cvttss2si %xmm2, %rdx
; GISEL-X64-NEXT: movabsq $-9223372036854775808, %rax # imm = 0x8000000000000000
; GISEL-X64-NEXT: xorq %rdx, %rax
; GISEL-X64-NEXT: xorl %edx, %edx
; GISEL-X64-NEXT: ucomiss %xmm1, %xmm0
; GISEL-X64-NEXT: setb %dl
; GISEL-X64-NEXT: andl $1, %edx
; GISEL-X64-NEXT: cmovneq %rcx, %rax
; GISEL-X64-NEXT: movss {{.*#+}} xmm3 = [0.0E+0,0.0E+0,0.0E+0,0.0E+0]
; GISEL-X64-NEXT: ucomiss %xmm2, %xmm3
; GISEL-X64-NEXT: seta %al
; GISEL-X64-NEXT: xorb $1, %al
; GISEL-X64-NEXT: movzbl %al, %eax
; GISEL-X64-NEXT: andq $1, %rax
; GISEL-X64-NEXT: xorps %xmm2, %xmm2
; GISEL-X64-NEXT: cvtsi2ss %rax, %xmm2
; GISEL-X64-NEXT: mulss %xmm1, %xmm2
Copy link
Contributor Author

@pawan-nirpal-031 pawan-nirpal-031 Aug 27, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Legalizer rules need fixing.

; GISEL-X64-NEXT: subss %xmm2, %xmm0
; GISEL-X64-NEXT: cvttss2si %xmm0, %rcx
; GISEL-X64-NEXT: cvttss2si %xmm2, %rax
; GISEL-X64-NEXT: addq %rcx, %rax
; GISEL-X64-NEXT: retq
;
; AVX512-LABEL: test_float_to_ui64:
Expand Down Expand Up @@ -163,17 +159,11 @@ define zeroext i16 @test_float_to_ui16(float %x) {
; X64-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NEXT: retq
;
; SDAG-AVX512-LABEL: test_float_to_ui16:
; SDAG-AVX512: # %bb.0: # %entry
; SDAG-AVX512-NEXT: vcvttss2si %xmm0, %eax
; SDAG-AVX512-NEXT: # kill: def $ax killed $ax killed $eax
; SDAG-AVX512-NEXT: retq
;
; GISEL-AVX512-LABEL: test_float_to_ui16:
; GISEL-AVX512: # %bb.0: # %entry
; GISEL-AVX512-NEXT: vcvttss2usi %xmm0, %eax
; GISEL-AVX512-NEXT: # kill: def $ax killed $ax killed $eax
; GISEL-AVX512-NEXT: retq
; AVX512-LABEL: test_float_to_ui16:
; AVX512: # %bb.0: # %entry
; AVX512-NEXT: vcvttss2si %xmm0, %eax
; AVX512-NEXT: # kill: def $ax killed $ax killed $eax
; AVX512-NEXT: retq
entry:
%conv = fptoui float %x to i16
ret i16 %conv
Expand All @@ -186,17 +176,11 @@ define zeroext i8 @test_float_to_ui8(float %x) {
; X64-NEXT: # kill: def $al killed $al killed $eax
; X64-NEXT: retq
;
; SDAG-AVX512-LABEL: test_float_to_ui8:
; SDAG-AVX512: # %bb.0: # %entry
; SDAG-AVX512-NEXT: vcvttss2si %xmm0, %eax
; SDAG-AVX512-NEXT: # kill: def $al killed $al killed $eax
; SDAG-AVX512-NEXT: retq
;
; GISEL-AVX512-LABEL: test_float_to_ui8:
; GISEL-AVX512: # %bb.0: # %entry
; GISEL-AVX512-NEXT: vcvttss2usi %xmm0, %eax
; GISEL-AVX512-NEXT: # kill: def $al killed $al killed $eax
; GISEL-AVX512-NEXT: retq
; AVX512-LABEL: test_float_to_ui8:
; AVX512: # %bb.0: # %entry
; AVX512-NEXT: vcvttss2si %xmm0, %eax
; AVX512-NEXT: # kill: def $al killed $al killed $eax
; AVX512-NEXT: retq
entry:
%conv = fptoui float %x to i8
ret i8 %conv
Expand Down Expand Up @@ -389,3 +373,6 @@ entry:
%conv = fptosi float %x to i33
ret i33 %conv
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GISEL-AVX512: {{.*}}
; SDAG-AVX512: {{.*}}
Loading
Loading