Skip to content

[X86] canCreateUndefOrPoisonForTargetNode - SSE PINSR/PEXTR vector element insert/extract are never out of bounds #149822

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

Jump to bottom
Merged
merged 1 commit into from
Jul 22, 2025
Merged

[X86] canCreateUndefOrPoisonForTargetNode - SSE PINSR/PEXTR vector element insert/extract are never out of bounds #149822

merged 1 commit into from
Jul 22, 2025

Conversation

RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented Jul 21, 2025

The immediate index is guaranteed to be treated as modulo

@RKSimon
[X86] canCreateUndefOrPoisonForTargetNode - SSE PINSR/PEXTR vector el…
0fc6e33 
…ement insert/extract are never out of bounds

The immediate index is guaranteed to be treated as modulo
@RKSimon RKSimon requested a review from phoebewang July 21, 2025 14:17
@llvmbot
Copy link
Member

llvmbot commented Jul 21, 2025

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

The immediate index is guaranteed to be treated as modulo

Full diff: https://github.com/llvm/llvm-project/pull/149822.diff

2 Files Affected:

  • (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+11-4)
  • (modified) llvm/test/CodeGen/X86/avg.ll (+39-42)
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 62811244dcfee..2b6c461d20a98 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -45098,27 +45098,34 @@ bool X86TargetLowering::canCreateUndefOrPoisonForTargetNode(
     bool PoisonOnly, bool ConsiderFlags, unsigned Depth) const {
 
   switch (Op.getOpcode()) {
+  // SSE vector insert/extracts use modulo indices.
+  case X86ISD::PINSRB:
+  case X86ISD::PINSRW:
+  case X86ISD::PEXTRB:
+  case X86ISD::PEXTRW:
+    return false;
   // SSE vector multiplies are either inbounds or saturate.
   case X86ISD::VPMADDUBSW:
   case X86ISD::VPMADDWD:
+    return false;
   // SSE vector shifts handle out of bounds shift amounts.
   case X86ISD::VSHLI:
   case X86ISD::VSRLI:
   case X86ISD::VSRAI:
     return false;
-    // SSE blends.
+  // SSE blends.
   case X86ISD::BLENDI:
   case X86ISD::BLENDV:
     return false;
-    // SSE target shuffles.
+  // SSE target shuffles.
   case X86ISD::PSHUFD:
   case X86ISD::UNPCKL:
   case X86ISD::UNPCKH:
   case X86ISD::VPERMILPI:
   case X86ISD::VPERMV3:
     return false;
-    // SSE comparisons handle all icmp/fcmp cases.
-    // TODO: Add CMPM/MM with test coverage.
+  // SSE comparisons handle all icmp/fcmp cases.
+  // TODO: Add CMPM/MM with test coverage.
   case X86ISD::CMPP:
   case X86ISD::PCMPEQ:
   case X86ISD::PCMPGT:
diff --git a/llvm/test/CodeGen/X86/avg.ll b/llvm/test/CodeGen/X86/avg.ll
index 217ccebdfb77f..9be816655072c 100644
--- a/llvm/test/CodeGen/X86/avg.ll
+++ b/llvm/test/CodeGen/X86/avg.ll
@@ -1829,73 +1829,70 @@ define void @not_avg_v16i8_wide_constants(ptr %a, ptr %b) nounwind {
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX1-NEXT:    vpextrd $2, %xmm5, %ecx
-; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
-; AVX1-NEXT:    vpextrd $2, %xmm4, %eax
 ; AVX1-NEXT:    vpextrw $3, %xmm3, %edx
+; AVX1-NEXT:    vpextrw $2, %xmm3, %ecx
+; AVX1-NEXT:    vpextrw $1, %xmm3, %eax
 ; AVX1-NEXT:    decl %edx
 ; AVX1-NEXT:    vmovd %edx, %xmm4
-; AVX1-NEXT:    vpextrw $2, %xmm3, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm5
-; AVX1-NEXT:    vpextrw $1, %xmm3, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm6
 ; AVX1-NEXT:    vpextrw $0, %xmm3, %edx
+; AVX1-NEXT:    decl %ecx
+; AVX1-NEXT:    vmovd %ecx, %xmm5
+; AVX1-NEXT:    vpextrw $3, %xmm2, %ecx
+; AVX1-NEXT:    decl %eax
+; AVX1-NEXT:    vmovd %eax, %xmm6
+; AVX1-NEXT:    vpextrw $2, %xmm2, %eax
 ; AVX1-NEXT:    decl %edx
 ; AVX1-NEXT:    vmovd %edx, %xmm7
-; AVX1-NEXT:    vpextrw $3, %xmm2, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm8
-; AVX1-NEXT:    vpextrw $2, %xmm2, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm9
 ; AVX1-NEXT:    vpextrw $1, %xmm2, %edx
+; AVX1-NEXT:    decl %ecx
+; AVX1-NEXT:    vmovd %ecx, %xmm8
+; AVX1-NEXT:    vpextrw $0, %xmm2, %ecx
+; AVX1-NEXT:    decl %eax
+; AVX1-NEXT:    vmovd %eax, %xmm9
+; AVX1-NEXT:    vpextrw $7, %xmm3, %eax
 ; AVX1-NEXT:    decl %edx
 ; AVX1-NEXT:    vmovd %edx, %xmm10
-; AVX1-NEXT:    vpextrw $0, %xmm2, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm11
-; AVX1-NEXT:    vpextrw $5, %xmm3, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm12
-; AVX1-NEXT:    vpextrw $4, %xmm3, %edx
+; AVX1-NEXT:    vpextrw $6, %xmm3, %edx
+; AVX1-NEXT:    decl %ecx
+; AVX1-NEXT:    vmovd %ecx, %xmm11
+; AVX1-NEXT:    vpextrw $7, %xmm2, %ecx
+; AVX1-NEXT:    decl %eax
+; AVX1-NEXT:    vmovd %eax, %xmm12
+; AVX1-NEXT:    vpextrw $6, %xmm2, %eax
 ; AVX1-NEXT:    decl %edx
 ; AVX1-NEXT:    vmovd %edx, %xmm13
-; AVX1-NEXT:    vpextrw $5, %xmm2, %edx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm14
-; AVX1-NEXT:    vpextrw $4, %xmm2, %edx
+; AVX1-NEXT:    vpextrw $5, %xmm3, %edx
+; AVX1-NEXT:    decl %ecx
+; AVX1-NEXT:    vmovd %ecx, %xmm14
+; AVX1-NEXT:    vpextrw $4, %xmm3, %ecx
+; AVX1-NEXT:    decl %eax
+; AVX1-NEXT:    vmovd %eax, %xmm3
+; AVX1-NEXT:    vpextrw $5, %xmm2, %eax
 ; AVX1-NEXT:    decl %edx
 ; AVX1-NEXT:    vmovd %edx, %xmm15
-; AVX1-NEXT:    vpextrw $7, %xmm3, %edx
+; AVX1-NEXT:    vpextrw $4, %xmm2, %edx
 ; AVX1-NEXT:    decl %ecx
-; AVX1-NEXT:    vmovd %ecx, %xmm3
-; AVX1-NEXT:    vpextrw $7, %xmm2, %ecx
-; AVX1-NEXT:    decl %edx
-; AVX1-NEXT:    vmovd %edx, %xmm2
+; AVX1-NEXT:    vmovd %ecx, %xmm2
 ; AVX1-NEXT:    decl %eax
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
 ; AVX1-NEXT:    vmovd %eax, %xmm5
-; AVX1-NEXT:    decl %ecx
+; AVX1-NEXT:    decl %edx
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX1-NEXT:    vmovd %ecx, %xmm7
+; AVX1-NEXT:    vmovd %edx, %xmm7
 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm6[0],xmm4[0],xmm6[1],xmm4[1]
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm8[0],xmm6[0],xmm8[1],xmm6[1]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm6, %ymm4
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3]
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm8, %ymm6
-; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX1-NEXT:    vmovddup {{.*#+}} ymm3 = ymm6[0,0,2,2]
-; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm7[0],xmm5[1],xmm7[1],xmm5[2],xmm7[2],xmm5[3],xmm7[3]
+; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm14[0],xmm3[1],xmm14[1],xmm3[2],xmm14[2],xmm3[3],xmm14[3]
+; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm3, %ymm3
+; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm15[0],xmm2[1],xmm15[1],xmm2[2],xmm15[2],xmm2[3],xmm15[3]
+; AVX1-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[0,0,0,0,4,4,4,4]
+; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm7[0],xmm5[0],xmm7[1],xmm5[1],xmm7[2],xmm5[2],xmm7[3],xmm5[3]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm5, %ymm2
-; AVX1-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,0,0,0,4,4,4,4]
-; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
+; AVX1-NEXT:    vmovddup {{.*#+}} ymm2 = ymm2[0,0,2,2]
+; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5,6],ymm3[7]
 ; AVX1-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    vandps %ymm0, %ymm2, %ymm1

phoebewang
phoebewang approved these changes Jul 22, 2025
View reviewed changes
Copy link
Contributor

@phoebewang phoebewang left a comment

Choose a reason for hiding this comment

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

LGTM.

@RKSimon RKSimon merged commit 069f0fe into llvm:main Jul 22, 2025
11 checks passed
@RKSimon RKSimon deleted the x86-pextr-pinsr-nopoison branch July 22, 2025 06:50
RKSimon added a commit to RKSimon/llvm-project that referenced this pull request Jul 24, 2025
@RKSimon
[X86] canCreateUndefOrPoisonForTargetNode - SSE PINSR/PEXTR vector el…
4f07800 
…ement insert/extract are never out of bounds (llvm#149822)

The immediate index is guaranteed to be treated as modulo
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@phoebewang phoebewang phoebewang approved these changes

Assignees
No one assigned
Labels
backend:X86
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@RKSimon @llvmbot @phoebewang