Skip to content

[SDAG] Add missing ppc_fp128 ExpandFloatRes for sincos[pi] #128514

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 3 commits into from
Feb 25, 2025
Merged

[SDAG] Add missing ppc_fp128 ExpandFloatRes for sincos[pi] #128514

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
cccc09d
[SDAG] Add missing ppc_fp128 ExpandFloatRes for sincos[pi]
MacDue Feb 24, 2025
093dd26
Add tail call test
MacDue Feb 24, 2025
652535f
Add other tail call form
MacDue Feb 24, 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
11 changes: 11 additions & 0 deletions llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1570,6 +1570,8 @@ void DAGTypeLegalizer::ExpandFloatResult(SDNode *N, unsigned ResNo) {
case ISD::STRICT_FREM:
case ISD::FREM: ExpandFloatRes_FREM(N, Lo, Hi); break;
case ISD::FMODF: ExpandFloatRes_FMODF(N); break;
case ISD::FSINCOS: ExpandFloatRes_FSINCOS(N); break;
case ISD::FSINCOSPI: ExpandFloatRes_FSINCOSPI(N); break;
// clang-format on
}

Expand Down Expand Up @@ -1625,6 +1627,15 @@ void DAGTypeLegalizer::ExpandFloatRes_FMODF(SDNode *N) {
/*CallRetResNo=*/0);
}

void DAGTypeLegalizer::ExpandFloatRes_FSINCOS(SDNode *N) {
ExpandFloatRes_UnaryWithTwoFPResults(N, RTLIB::getSINCOS(N->getValueType(0)));
}

void DAGTypeLegalizer::ExpandFloatRes_FSINCOSPI(SDNode *N) {
ExpandFloatRes_UnaryWithTwoFPResults(N,
RTLIB::getSINCOSPI(N->getValueType(0)));
}

void DAGTypeLegalizer::ExpandFloatRes_UnaryWithTwoFPResults(
SDNode *N, RTLIB::Libcall LC, std::optional<unsigned> CallRetResNo) {
assert(!N->isStrictFPOpcode() && "strictfp not implemented");
Expand Down
2 changes: 2 additions & 0 deletions llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -718,6 +718,8 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
void ExpandFloatRes_LOAD (SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_XINT_TO_FP(SDNode *N, SDValue &Lo, SDValue &Hi);
void ExpandFloatRes_FMODF(SDNode *N);
void ExpandFloatRes_FSINCOS(SDNode* N);
void ExpandFloatRes_FSINCOSPI(SDNode* N);
// clang-format on

// Float Operand Expansion.
Expand Down
97 changes: 97 additions & 0 deletions llvm/test/CodeGen/PowerPC/llvm.sincos.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,97 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc -mcpu=pwr9 -mtriple=powerpc64le-gnu-linux \
; RUN: -ppc-vsr-nums-as-vr -ppc-asm-full-reg-names < %s | FileCheck %s

define { ppc_fp128, ppc_fp128 } @test_sincos_ppcf128(ppc_fp128 %a) {
; CHECK-LABEL: test_sincos_ppcf128:
; CHECK: # %bb.0:
; CHECK-NEXT: mflr r0
; CHECK-NEXT: stdu r1, -64(r1)
; CHECK-NEXT: std r0, 80(r1)
; CHECK-NEXT: .cfi_def_cfa_offset 64
; CHECK-NEXT: .cfi_offset lr, 16
; CHECK-NEXT: addi r5, r1, 48
; CHECK-NEXT: addi r6, r1, 32
; CHECK-NEXT: bl sincosl
; CHECK-NEXT: nop
; CHECK-NEXT: lfd f1, 48(r1)
; CHECK-NEXT: lfd f2, 56(r1)
; CHECK-NEXT: lfd f3, 32(r1)
; CHECK-NEXT: lfd f4, 40(r1)
; CHECK-NEXT: addi r1, r1, 64
; CHECK-NEXT: ld r0, 16(r1)
; CHECK-NEXT: mtlr r0
; CHECK-NEXT: blr
%result = call { ppc_fp128, ppc_fp128 } @llvm.sincos.ppcf128(ppc_fp128 %a)
ret { ppc_fp128, ppc_fp128 } %result
}

define { ppc_fp128, ppc_fp128 } @test_sincospi_ppcf128(ppc_fp128 %a) {
; CHECK-LABEL: test_sincospi_ppcf128:
; CHECK: # %bb.0:
; CHECK-NEXT: mflr r0
; CHECK-NEXT: stdu r1, -64(r1)
; CHECK-NEXT: std r0, 80(r1)
; CHECK-NEXT: .cfi_def_cfa_offset 64
; CHECK-NEXT: .cfi_offset lr, 16
; CHECK-NEXT: addi r5, r1, 48
; CHECK-NEXT: addi r6, r1, 32
; CHECK-NEXT: bl sincospil
; CHECK-NEXT: nop
; CHECK-NEXT: lfd f1, 48(r1)
; CHECK-NEXT: lfd f2, 56(r1)
; CHECK-NEXT: lfd f3, 32(r1)
; CHECK-NEXT: lfd f4, 40(r1)
; CHECK-NEXT: addi r1, r1, 64
; CHECK-NEXT: ld r0, 16(r1)
; CHECK-NEXT: mtlr r0
; CHECK-NEXT: blr
%result = call { ppc_fp128, ppc_fp128 } @llvm.sincospi.ppcf128(ppc_fp128 %a)
ret { ppc_fp128, ppc_fp128 } %result
}

; FIXME: Recognise this as a tail call and omit the stack frame:
Copy link
Contributor

Choose a reason for hiding this comment

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

I don't think this can be recognized as a tail call, the return needs to be the immediate next instruction after the call

Copy link
Member Author

Choose a reason for hiding this comment

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

the return needs to be the immediate next instruction after the call

That is what codegen emits, the stores and extractvalues fold away to nothing, though the tail call logic does not recognise this.

Copy link
Contributor

Choose a reason for hiding this comment

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

What do you mean by "what the codegen emits"? The issue is what is in the IR. This function should return the original { ppc_fp128, ppc_fp128 } with no extracts or stores

Copy link
Member Author

@MacDue MacDue Feb 24, 2025 •
edited
Loading

Choose a reason for hiding this comment

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

Right, but returning a struct would never result in a tail call for the standard expansion.
This IR does result in something that codegen could recognise as a tail call (and there's logic in SDAG that could do this), but right now it fails to do so. Look at the current codegen:

; CHECK-LABEL: test_sincos_ppcf128_tail_call:
; CHECK:       # %bb.0:
<frame setup>
; CHECK-NEXT:    mflr r0
; CHECK-NEXT:    stdu r1, -32(r1)
; CHECK-NEXT:    std r0, 48(r1)
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset lr, 16

<call to sincos>
; CHECK-NEXT:    bl sincosl

<frame destruction> 
; CHECK-NEXT:    nop
; CHECK-NEXT:    addi r1, r1, 32
; CHECK-NEXT:    ld r0, 16(r1)
; CHECK-NEXT:    mtlr r0
; CHECK-NEXT:    blr

This could simply be a jump to sincosl.

define void @test_sincos_ppcf128_tail_call(ppc_fp128 %a, ptr noalias %out_sin, ptr noalias %out_cos) {
Copy link
Contributor

Choose a reason for hiding this comment

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

This should return the raw structure type

Copy link
Member Author

Choose a reason for hiding this comment

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

That won't result in something that could be a tail call on any target that uses the (semi)-standard GNU sincos function (since it'll need to emit loads after the call).

; CHECK-LABEL: test_sincos_ppcf128_tail_call:
; CHECK: # %bb.0:
; CHECK-NEXT: mflr r0
; CHECK-NEXT: stdu r1, -32(r1)
; CHECK-NEXT: std r0, 48(r1)
; CHECK-NEXT: .cfi_def_cfa_offset 32
; CHECK-NEXT: .cfi_offset lr, 16
; CHECK-NEXT: bl sincosl
; CHECK-NEXT: nop
; CHECK-NEXT: addi r1, r1, 32
; CHECK-NEXT: ld r0, 16(r1)
; CHECK-NEXT: mtlr r0
; CHECK-NEXT: blr
%result = tail call { ppc_fp128, ppc_fp128 } @llvm.sincos.ppcf128(ppc_fp128 %a)
%result.0 = extractvalue { ppc_fp128, ppc_fp128 } %result, 0
%result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 1
store ppc_fp128 %result.0, ptr %out_sin, align 16
store ppc_fp128 %result.1, ptr %out_cos, align 16
ret void
}

; FIXME: Recognise this as a tail call and omit the stack frame:
define void @test_sincospi_ppcf128_tail_call(ppc_fp128 %a, ptr noalias %out_sin, ptr noalias %out_cos) {
; CHECK-LABEL: test_sincospi_ppcf128_tail_call:
; CHECK: # %bb.0:
; CHECK-NEXT: mflr r0
; CHECK-NEXT: stdu r1, -32(r1)
; CHECK-NEXT: std r0, 48(r1)
; CHECK-NEXT: .cfi_def_cfa_offset 32
; CHECK-NEXT: .cfi_offset lr, 16
; CHECK-NEXT: bl sincospil
; CHECK-NEXT: nop
; CHECK-NEXT: addi r1, r1, 32
; CHECK-NEXT: ld r0, 16(r1)
; CHECK-NEXT: mtlr r0
; CHECK-NEXT: blr
%result = tail call { ppc_fp128, ppc_fp128 } @llvm.sincospi.ppcf128(ppc_fp128 %a)
%result.0 = extractvalue { ppc_fp128, ppc_fp128 } %result, 0
%result.1 = extractvalue { ppc_fp128, ppc_fp128 } %result, 1
store ppc_fp128 %result.0, ptr %out_sin, align 16
store ppc_fp128 %result.1, ptr %out_cos, align 16
ret void
}
Loading