another missed SSE optimization

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@39772 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/README-SSE.txt b/lib/Target/X86/README-SSE.txt
index 20e6a53267ed26b15ee5682b3dd367fe9388f1e6..f4b54c4dbbb006b9a1dfc6813080270f41462ced 100644 (file)
--- a/lib/Target/X86/README-SSE.txt
+++ b/lib/Target/X86/README-SSE.txt
@@ -598,3 +598,32 @@ entry:
 }
 
 //===---------------------------------------------------------------------===//
+
+These functions should produce the same code:
+
+#include <emmintrin.h>
+
+typedef long long __m128i __attribute__ ((__vector_size__ (16)));
+
+int foo(__m128i* val) {
+  return __builtin_ia32_vec_ext_v4si(*val, 1);
+}
+int bar(__m128i* val) {
+  union vs {
+    __m128i *_v;
+    int* _s;
+  } v = {val};
+  return v._s[1];
+}
+
+We currently produce (with -m64):
+
+_foo:
+        pshufd $1, (%rdi), %xmm0
+        movd %xmm0, %eax
+        ret
+_bar:
+        movl 4(%rdi), %eax
+        ret
+
+//===---------------------------------------------------------------------===//