release/18.x: [X86][EVEX512] Check hasEVEX512 for canExtendTo512DQ (#90390)

[llvmbot]

Backport 35b89dd

Requested by: @phoebewang

[llvmbot]

@RKSimon What do you think about merging this PR to the release branch?

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: None (llvmbot)

Changes

Backport 35b89dd

Requested by: @phoebewang

---

Full diff: https://github.com/llvm/llvm-project/pull/90422.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86Subtarget.h (+2-1)
• (modified) llvm/test/CodeGen/X86/avx512bwvl-arith.ll (+33-2)

diff --git a/llvm/lib/Target/X86/X86Subtarget.h b/llvm/lib/Target/X86/X86Subtarget.h
index a458b5f9ec8fbb..4d55a084b730e4 100644
--- a/llvm/lib/Target/X86/X86Subtarget.h
+++ b/llvm/lib/Target/X86/X86Subtarget.h
@@ -244,7 +244,8 @@ class X86Subtarget final : public X86GenSubtargetInfo {
   // TODO: Currently we're always allowing widening on CPUs without VLX,
   // because for many cases we don't have a better option.
   bool canExtendTo512DQ() const {
-    return hasAVX512() && (!hasVLX() || getPreferVectorWidth() >= 512);
+    return hasAVX512() && hasEVEX512() &&
+           (!hasVLX() || getPreferVectorWidth() >= 512);
   }
   bool canExtendTo512BW() const  {
     return hasBWI() && canExtendTo512DQ();
diff --git a/llvm/test/CodeGen/X86/avx512bwvl-arith.ll b/llvm/test/CodeGen/X86/avx512bwvl-arith.ll
index 4988fc35b10eef..fdc25f44b156a7 100644
--- a/llvm/test/CodeGen/X86/avx512bwvl-arith.ll
+++ b/llvm/test/CodeGen/X86/avx512bwvl-arith.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,-evex512 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefixes=CHECK,EVEX256
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,-evex512 | FileCheck %s --check-prefixes=CHECK,EVEX512
 
 ; 256-bit
 
@@ -236,3 +236,34 @@ define <8 x i16> @vpmullw128_test(<8 x i16> %i, <8 x i16> %j) {
   ret <8 x i16> %x
 }
 
+define i16 @PR90356(<16 x i1> %a) {
+; EVEX256-LABEL: PR90356:
+; EVEX256:       # %bb.0:
+; EVEX256-NEXT:    vpsllw $7, %xmm0, %xmm0
+; EVEX256-NEXT:    vpmovb2m %xmm0, %k1
+; EVEX256-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; EVEX256-NEXT:    movb $63, %al
+; EVEX256-NEXT:    kmovd %eax, %k1
+; EVEX256-NEXT:    vpexpandq %zmm0, %zmm0 {%k1} {z}
+; EVEX256-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; EVEX256-NEXT:    kmovd %k0, %eax
+; EVEX256-NEXT:    # kill: def $ax killed $ax killed $eax
+; EVEX256-NEXT:    vzeroupper
+; EVEX256-NEXT:    retq
+;
+; EVEX512-LABEL: PR90356:
+; EVEX512:       # %bb.0:
+; EVEX512-NEXT:    vpsllw $7, %xmm0, %xmm0
+; EVEX512-NEXT:    vpmovb2m %xmm0, %k0
+; EVEX512-NEXT:    vpmovm2w %k0, %ymm0
+; EVEX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; EVEX512-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; EVEX512-NEXT:    vpmovw2m %ymm0, %k0
+; EVEX512-NEXT:    kmovd %k0, %eax
+; EVEX512-NEXT:    # kill: def $ax killed $ax killed $eax
+; EVEX512-NEXT:    vzeroupper
+; EVEX512-NEXT:    retq
+  %1 = shufflevector <16 x i1> %a, <16 x i1> zeroinitializer, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
+  %2 = bitcast <16 x i1> %1 to i16
+  ret i16 %2
+}

[RKSimon]

LGTM

[tstellar]

Hi @phoebewang (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.

[phoebewang]

Hi @phoebewang (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.

This patch fixes a X86 bug introduced during LLVM18, which crashes when compiling some bit vector with AVX512.