Skip to content

[AArch64][GlobalISel] Add fp128 and i128 fptosi/fptoui handling. #95528

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

Merged
merged 2 commits into from
Jun 21, 2024
Merged

[AArch64][GlobalISel] Add fp128 and i128 fptosi/fptoui handling. #95528

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bc597b8
[AArch64][GlobalISel] Add fp128 and i128 fptosi/fptoui handling.
davemgreen Jun 14, 2024
c802007
Use getFloatTypeForLLT
davemgreen Jun 15, 2024
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
10 changes: 4 additions & 6 deletions llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1123,15 +1123,13 @@ LegalizerHelper::libcall(MachineInstr &MI, LostDebugLocObserver &LocObserver) {
case TargetOpcode::G_FPTOSI:
case TargetOpcode::G_FPTOUI: {
// FIXME: Support other types
unsigned FromSize = MRI.getType(MI.getOperand(1).getReg()).getSizeInBits();
Type *FromTy =
getFloatTypeForLLT(Ctx, MRI.getType(MI.getOperand(1).getReg()));
unsigned ToSize = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
if ((ToSize != 32 && ToSize != 64) || (FromSize != 32 && FromSize != 64))
if ((ToSize != 32 && ToSize != 64 && ToSize != 128) || !FromTy)
return UnableToLegalize;
LegalizeResult Status = conversionLibcall(
MI, MIRBuilder,
ToSize == 32 ? Type::getInt32Ty(Ctx) : Type::getInt64Ty(Ctx),
FromSize == 64 ? Type::getDoubleTy(Ctx) : Type::getFloatTy(Ctx),
LocObserver);
MI, MIRBuilder, Type::getIntNTy(Ctx, ToSize), FromTy, LocObserver);
if (Status != Legalized)
return Status;
break;
Expand Down
38 changes: 28 additions & 10 deletions llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -661,34 +661,52 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)

// Conversions
getActionDefinitionsBuilder({G_FPTOSI, G_FPTOUI})
.legalForCartesianProduct({s32, s64, v2s64, v4s32, v2s32})
.legalFor({{s32, s32},
{s64, s32},
{s32, s64},
{s64, s64},
{v2s64, v2s64},
{v4s32, v4s32},
{v2s32, v2s32}})
.legalIf([=](const LegalityQuery &Query) {
return HasFP16 &&
(Query.Types[1] == s16 || Query.Types[1] == v4s16 ||
Query.Types[1] == v8s16) &&
(Query.Types[0] == s32 || Query.Types[0] == s64 ||
Query.Types[0] == v4s16 || Query.Types[0] == v8s16);
})
.widenScalarToNextPow2(0)
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(1)
.clampScalarOrElt(1, MinFPScalar, s64)
.scalarizeIf(scalarOrEltWiderThan(0, 64), 0)
.scalarizeIf(scalarOrEltWiderThan(1, 64), 1)
// The range of a fp16 value fits into an i17, so we can lower the width
// to i64.
.narrowScalarIf(
[=](const LegalityQuery &Query) {
return Query.Types[1] == s16 && Query.Types[0].getSizeInBits() > 64;
},
changeTo(0, s64))
.moreElementsToNextPow2(0)
.widenScalarOrEltToNextPow2OrMinSize(0)
.minScalar(0, s32)
.widenScalarOrEltToNextPow2OrMinSize(1, /*MinSize=*/HasFP16 ? 16 : 32)
.widenScalarIf(
[=](const LegalityQuery &Query) {
return Query.Types[0].getScalarSizeInBits() >
Query.Types[1].getScalarSizeInBits();
return Query.Types[0].getScalarSizeInBits() <= 64 &&
Query.Types[0].getScalarSizeInBits() >
Query.Types[1].getScalarSizeInBits();
},
LegalizeMutations::changeElementSizeTo(1, 0))
.widenScalarIf(
[=](const LegalityQuery &Query) {
return Query.Types[0].getScalarSizeInBits() <
Query.Types[1].getScalarSizeInBits();
return Query.Types[1].getScalarSizeInBits() <= 64 &&
Query.Types[0].getScalarSizeInBits() <
Query.Types[1].getScalarSizeInBits();
},
LegalizeMutations::changeElementSizeTo(0, 1))
.clampNumElements(0, v4s16, v8s16)
.clampNumElements(0, v2s32, v4s32)
.clampMaxNumElements(0, s64, 2);
.clampMaxNumElements(0, s64, 2)
.libcallFor(
{{s32, s128}, {s64, s128}, {s128, s128}, {s128, s32}, {s128, s64}});

getActionDefinitionsBuilder({G_SITOFP, G_UITOFP})
.legalForCartesianProduct({s32, s64, v2s64, v4s32, v2s32})
Expand Down
8 changes: 0 additions & 8 deletions llvm/test/CodeGen/AArch64/GlobalISel/gisel-fail-intermediate-legalizer.ll
View file
Open in desktop

This file was deleted.

76 changes: 0 additions & 76 deletions llvm/test/CodeGen/AArch64/GlobalISel/select-fp-casts.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -711,79 +711,3 @@ body: |
%1:fpr(<2 x s32>) = G_UITOFP %0
$d0 = COPY %1(<2 x s32>)
...

Copy link
Contributor

Choose a reason for hiding this comment

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

Why remove these tests?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

They were testing a v2s32->v2s64 G_FPTOSI, which are nowadays no longer legal. We split into two operations earlier during legalization. It shouldn't alter the end result, but feels a bit cleaner.

---
name: fptosi_v2s64_v2s32
legalized: true
regBankSelected: true

body: |
bb.0:
liveins: $d0

; CHECK-LABEL: name: fptosi_v2s64_v2s32
; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY $d0
; CHECK: [[FCVTLv2i32_:%[0-9]+]]:fpr128 = nofpexcept FCVTLv2i32 [[COPY]]
; CHECK: [[FCVTZSv2f64_:%[0-9]+]]:fpr128 = nofpexcept FCVTZSv2f64 [[FCVTLv2i32_]]
; CHECK: $q0 = COPY [[FCVTZSv2f64_]]
%0:fpr(<2 x s32>) = COPY $d0
%1:fpr(<2 x s64>) = G_FPTOSI %0
$q0 = COPY %1(<2 x s64>)
...

---
name: fptoui_v2s64_v2s32
legalized: true
regBankSelected: true

body: |
bb.0:
liveins: $d0

; CHECK-LABEL: name: fptoui_v2s64_v2s32
; CHECK: [[COPY:%[0-9]+]]:fpr64 = COPY $d0
; CHECK: [[FCVTLv2i32_:%[0-9]+]]:fpr128 = nofpexcept FCVTLv2i32 [[COPY]]
; CHECK: [[FCVTZUv2f64_:%[0-9]+]]:fpr128 = nofpexcept FCVTZUv2f64 [[FCVTLv2i32_]]
; CHECK: $q0 = COPY [[FCVTZUv2f64_]]
%0:fpr(<2 x s32>) = COPY $d0
%1:fpr(<2 x s64>) = G_FPTOUI %0
$q0 = COPY %1(<2 x s64>)
...

---
name: fptosi_v2s32_v2s64
legalized: true
regBankSelected: true

body: |
bb.0:
liveins: $q0

; CHECK-LABEL: name: fptosi_v2s32_v2s64
; CHECK: [[COPY:%[0-9]+]]:fpr128 = COPY $q0
; CHECK: [[FCVTZSv2f64_:%[0-9]+]]:fpr128 = nofpexcept FCVTZSv2f64 [[COPY]]
; CHECK: [[XTNv2i32_:%[0-9]+]]:fpr64 = XTNv2i32 [[FCVTZSv2f64_]]
; CHECK: $d0 = COPY [[XTNv2i32_]]
%0:fpr(<2 x s64>) = COPY $q0
%1:fpr(<2 x s32>) = G_FPTOSI %0
$d0 = COPY %1(<2 x s32>)
...

---
name: fptoui_v2s32_v2s64
legalized: true
regBankSelected: true

body: |
bb.0:
liveins: $q0

; CHECK-LABEL: name: fptoui_v2s32_v2s64
; CHECK: [[COPY:%[0-9]+]]:fpr128 = COPY $q0
; CHECK: [[FCVTZUv2f64_:%[0-9]+]]:fpr128 = nofpexcept FCVTZUv2f64 [[COPY]]
; CHECK: [[XTNv2i32_:%[0-9]+]]:fpr64 = XTNv2i32 [[FCVTZUv2f64_]]
; CHECK: $d0 = COPY [[XTNv2i32_]]
%0:fpr(<2 x s64>) = COPY $q0
%1:fpr(<2 x s32>) = G_FPTOUI %0
$d0 = COPY %1(<2 x s32>)
...
Loading
Loading