[llvm] e838471 - [X86] Add regression test case from pr64593

[Philip Reames via llvm-commits]

Author: Philip Reames
Date: 2023-08-10T08:49:41-07:00
New Revision: e838471bc4936854ced32709f7f105af87aed798

URL: https://github.com/llvm/llvm-project/commit/e838471bc4936854ced32709f7f105af87aed798
DIFF: https://github.com/llvm/llvm-project/commit/e838471bc4936854ced32709f7f105af87aed798.diff

LOG: [X86] Add regression test case from pr64593

This is the case which triggered the revert of 660b740.  Note that the test is extremely fragile as it depends on getting a truncating store at the right moment rather than folding the constant to a narrower bitwidth.  This appears to happen on skylake, but not e.g. plain avx.

Added: 
    llvm/test/CodeGen/X86/pr64593.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/pr64593.ll b/llvm/test/CodeGen/X86/pr64593.ll
new file mode 100644
index 00000000000000..8228661b820247
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr64593.ll
@@ -0,0 +1,21 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=skx < %s | FileCheck %s
+
+; If we have a truncating store, we need to truncate the elements when
+; when combining the constants.
+define void @pr64593(ptr %p) {
+; CHECK-LABEL: pr64593:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vpmovqb %zmm0, (%rdi)
+; CHECK-NEXT:    movq $0, 8(%rdi)
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %v = insertelement <8 x i64> zeroinitializer, i64 0, i32 1
+  %trunc = trunc <8 x i64> %v to <8 x i8>
+  %p1 = getelementptr i8, ptr %p, i32 0
+  %p2 = getelementptr i8, ptr %p, i32 8
+  store <8 x i8> %trunc, ptr %p1
+  store <8 x i8> zeroinitializer, ptr %p2
+  ret void
+}