maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[X86] Fold ANDNP(undef,x)/ANDNP(x,undef) -> 0 

Matches the fold in DAGCombiner::visitANDLike.
This commit is contained in:
Simon Pilgrim 2022-04-29 10:20:39 +01:00
parent d5ee20fcc9
commit 2d7f0b1c22
2 changed files with 19 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -50665,6 +50665,11 @@ static SDValue combineAndnp(SDNode *N, SelectionDAG &DAG,
SDValue N1 = N->getOperand(1); SDValue N1 = N->getOperand(1);
MVT VT = N->getSimpleValueType(0); MVT VT = N->getSimpleValueType(0);
// ANDNP(undef, x) -> 0
// ANDNP(x, undef) -> 0
if (N0.isUndef() || N1.isUndef())
return DAG.getConstant(0, SDLoc(N), VT);
// ANDNP(0, x) -> x // ANDNP(0, x) -> x
if (ISD::isBuildVectorAllZeros(N0.getNode())) if (ISD::isBuildVectorAllZeros(N0.getNode()))
return N1; return N1;

41
llvm/test/CodeGen/X86/pr55158.ll
View File

@ -1,33 +1,20 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-linux -opaque-pointers -mcpu=ivybridge -mattr=+avx2 | FileCheck %s --check-prefixes=IVB ; RUN: llc < %s -mtriple=x86_64-linux -opaque-pointers -mcpu=ivybridge -mattr=+avx2 | FileCheck %s
; RUN: llc < %s -mtriple=x86_64-linux -opaque-pointers -mcpu=haswell | FileCheck %s --check-prefixes=HSW ; RUN: llc < %s -mtriple=x86_64-linux -opaque-pointers -mcpu=haswell | FileCheck %s
define <2 x i64> @PR55158(ptr %0) { define <2 x i64> @PR55158(ptr %0) {
; IVB-LABEL: PR55158: ; CHECK-LABEL: PR55158:
; IVB: # %bb.0: ; CHECK: # %bb.0:
; IVB-NEXT: vmovdqa 64(%rdi), %xmm0 ; CHECK-NEXT: vmovdqa 64(%rdi), %xmm0
; IVB-NEXT: vmovdqa 128(%rdi), %xmm1 ; CHECK-NEXT: vmovdqa 128(%rdi), %xmm1
; IVB-NEXT: vpmovsxbd (%rdi), %xmm2 ; CHECK-NEXT: vpmovsxbd (%rdi), %xmm2
; IVB-NEXT: vpcmpgtd %xmm2, %xmm1, %xmm1 ; CHECK-NEXT: vpcmpgtd %xmm2, %xmm1, %xmm1
; IVB-NEXT: vphsubw %xmm0, %xmm0, %xmm0 ; CHECK-NEXT: vphsubw %xmm0, %xmm0, %xmm0
; IVB-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero ; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
; IVB-NEXT: vpxor %xmm2, %xmm2, %xmm2 ; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; IVB-NEXT: vpsrlvq %xmm1, %xmm2, %xmm1 ; CHECK-NEXT: vpsrlvq %xmm1, %xmm2, %xmm1
; IVB-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm0[1] ; CHECK-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm0[1]
; IVB-NEXT: retq ; CHECK-NEXT: retq
;
; HSW-LABEL: PR55158:
; HSW: # %bb.0:
; HSW-NEXT: vmovdqa 64(%rdi), %xmm0
; HSW-NEXT: vmovdqa 128(%rdi), %xmm1
; HSW-NEXT: vpmovsxbd (%rdi), %xmm2
; HSW-NEXT: vpcmpgtd %xmm2, %xmm1, %xmm1
; HSW-NEXT: vphsubw %xmm0, %xmm0, %xmm0
; HSW-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
; HSW-NEXT: vpandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm2
; HSW-NEXT: vpsrlvq %xmm1, %xmm2, %xmm1
; HSW-NEXT: vpunpckhqdq {{.*#+}} xmm0 = xmm1[1],xmm0[1]
; HSW-NEXT: retq
%2 = load <16 x i8>, ptr %0, align 16 %2 = load <16 x i8>, ptr %0, align 16
%3 = getelementptr inbounds i32, ptr %0, i64 16 %3 = getelementptr inbounds i32, ptr %0, i64 16
%4 = load <8 x i16>, ptr %3, align 16 %4 = load <8 x i16>, ptr %3, align 16