From 35b89dda2b9734917824b1457f149192669b314c Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.wang@intel.com>
Date: Mon, 29 Apr 2024 08:40:26 +0800
Subject: [PATCH] [X86][EVEX512] Check hasEVEX512 for canExtendTo512DQ (#90390)

Fixes #90356
---
 llvm/lib/Target/X86/X86Subtarget.h        |  3 +-
 llvm/test/CodeGen/X86/avx512bwvl-arith.ll | 35 +++++++++++++++++++++--
 2 files changed, 35 insertions(+), 3 deletions(-)

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
+}