shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[X86][TargetLowering] Avoid deleting temporary nodes in getNegatedExpression (#139029)

Browse Source 
In the original case, the third call to `getCheaperNegatedExpression`
deletes the SDNode returned by the first call.
Similar to 74e6030bcbcc8e628f9a99a424342a0c656456f9, this patch uses
`HandleSDNodes` to prevent nodes from being deleted by subsequent calls.
Closes https://github.com/llvm/llvm-project/issues/138944.

(cherry picked from commit 143cce72b1f50bc37363315793b80ae92d2b0ae3)
This commit is contained in:
Yingwei Zheng 2025-05-10 13:14:01 +08:00 committed by Tom Stellard
parent f233430d97
commit 2cacf46f35
2 changed files with 31 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -54147,12 +54147,19 @@ SDValue X86TargetLowering::getNegatedExpression(SDValue Op, SelectionDAG &DAG,
if (!Flags.hasNoSignedZeros())
break;
// Because getCheaperNegatedExpression can delete nodes we need a handle to
// keep temporary nodes alive.
std::list<HandleSDNode> Handles;
// This is always negatible for free but we might be able to remove some
// extra operand negations as well.
SmallVector<SDValue, 4> NewOps(Op.getNumOperands(), SDValue());
for (int i = 0; i != 3; ++i)
for (int i = 0; i != 3; ++i) {
NewOps[i] = getCheaperNegatedExpression(
Op.getOperand(i), DAG, LegalOperations, ForCodeSize, Depth + 1);
if (!!NewOps[i])
Handles.emplace_back(NewOps[i]);
}
bool NegA = !!NewOps[0];
bool NegB = !!NewOps[1];

23
llvm/test/CodeGen/X86/pr138982.ll Normal file
View File

@ -0,0 +1,23 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc < %s -mtriple=x86_64 -mattr=+fma | FileCheck %s
define <4 x float> @pr138982(<4 x float> %in_vec) {
; CHECK-LABEL: pr138982:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
; CHECK-NEXT: vrcpps %xmm0, %xmm2
; CHECK-NEXT: vrcpps %xmm1, %xmm1
; CHECK-NEXT: vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vcmpneqps %xmm0, %xmm3, %xmm0
; CHECK-NEXT: vbroadcastss {{.*#+}} xmm4 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; CHECK-NEXT: vblendvps %xmm0, %xmm1, %xmm4, %xmm0
; CHECK-NEXT: vfnmadd231ps {{.*#+}} xmm0 = -(xmm3 * xmm2) + xmm0
; CHECK-NEXT: retq
entry:
%fneg = fneg <4 x float> %in_vec
%rcp = tail call <4 x float> @llvm.x86.sse.rcp.ps(<4 x float> %fneg)
%cmp = fcmp une <4 x float> zeroinitializer, %in_vec
%sel = select <4 x i1> %cmp, <4 x float> %rcp, <4 x float> splat (float 1.000000e+00)
%fma = call nsz <4 x float> @llvm.fma.v4f32(<4 x float> %rcp, <4 x float> zeroinitializer, <4 x float> %sel)
ret <4 x float> %fma
}