[X86][EVEX512] Do not allow 512-bit memcpy without EVEX512 (#70420)

Solves crash mentioned in #65920.

Author: phoebewang

llvm/lib/Target/X86/X86ISelLoweringCall.cpp

@@ -281,7 +281,7 @@ EVT X86TargetLowering::getOptimalMemOpType(
     if (Op.size() >= 16 &&
         (!Subtarget.isUnalignedMem16Slow() || Op.isAligned(Align(16)))) {
       // FIXME: Check if unaligned 64-byte accesses are slow.
-      if (Op.size() >= 64 && Subtarget.hasAVX512() &&
+      if (Op.size() >= 64 && Subtarget.hasAVX512() && Subtarget.hasEVEX512() &&
           (Subtarget.getPreferVectorWidth() >= 512)) {
         return Subtarget.hasBWI() ? MVT::v64i8 : MVT::v16i32;
       }
@@ -395,7 +395,7 @@ bool X86TargetLowering::allowsMemoryAccess(LLVMContext &Context,
         return true;
       return false;
     case 512:
-      if (Subtarget.hasAVX512())
+      if (Subtarget.hasAVX512() && Subtarget.hasEVEX512())
         return true;
       return false;
     default:

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

@@ -180,7 +180,7 @@ X86TTIImpl::getRegisterBitWidth(TargetTransformInfo::RegisterKind K) const {
   case TargetTransformInfo::RGK_Scalar:
     return TypeSize::getFixed(ST->is64Bit() ? 64 : 32);
   case TargetTransformInfo::RGK_FixedWidthVector:
-    if (ST->hasAVX512() && PreferVectorWidth >= 512)
+    if (ST->hasAVX512() && ST->hasEVEX512() && PreferVectorWidth >= 512)
       return TypeSize::getFixed(512);
     if (ST->hasAVX() && PreferVectorWidth >= 256)
       return TypeSize::getFixed(256);
@@ -6131,7 +6131,8 @@ X86TTIImpl::enableMemCmpExpansion(bool OptSize, bool IsZeroCmp) const {
     // Only enable vector loads for equality comparison. Right now the vector
     // version is not as fast for three way compare (see #33329).
     const unsigned PreferredWidth = ST->getPreferVectorWidth();
-    if (PreferredWidth >= 512 && ST->hasAVX512()) Options.LoadSizes.push_back(64);
+    if (PreferredWidth >= 512 && ST->hasAVX512() && ST->hasEVEX512())
+      Options.LoadSizes.push_back(64);
     if (PreferredWidth >= 256 && ST->hasAVX()) Options.LoadSizes.push_back(32);
     if (PreferredWidth >= 128 && ST->hasSSE2()) Options.LoadSizes.push_back(16);
   }

llvm/test/CodeGen/X86/evex512-mem.ll

@@ -0,0 +1,29 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=avx512f,avx512bw,avx512vl < %s | FileCheck %s --check-prefix=AVX512
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=avx512f,avx512bw,avx512vl,-evex512 < %s | FileCheck %s --check-prefix=AVX256
+
+define void @test1() {
+; AVX512-LABEL: test1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    movq 64, %rax
+; AVX512-NEXT:    movq %rax, (%rax)
+; AVX512-NEXT:    vmovups 0, %zmm0
+; AVX512-NEXT:    vmovups %zmm0, (%rax)
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
+;
+; AVX256-LABEL: test1:
+; AVX256:       # %bb.0:
+; AVX256-NEXT:    movq 64, %rax
+; AVX256-NEXT:    movq %rax, (%rax)
+; AVX256-NEXT:    vmovups 0, %ymm0
+; AVX256-NEXT:    vmovups 32, %ymm1
+; AVX256-NEXT:    vmovups %ymm1, (%rax)
+; AVX256-NEXT:    vmovups %ymm0, (%rax)
+; AVX256-NEXT:    vzeroupper
+; AVX256-NEXT:    retq
+  call void @llvm.memcpy.p0.p0.i64(ptr align 8 poison, ptr align 8 null, i64 72, i1 false)
+  ret void
+}
+
+declare void @llvm.memcpy.p0.p0.i64(ptr, ptr, i64, i1)