fix some vector extractions to return properly zero extended values

(instead of sign extending) to match ICC.  GCC is changing this in 
a series of their own PRs (e.g. 41323).


git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@111637 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Headers/emmintrin.h

@@ -1194,7 +1194,7 @@ static __inline__ int __attribute__((__always_inline__, __nodebug__))
 _mm_extract_epi16(__m128i a, int imm)
 {
   __v8hi b = (__v8hi)a;
-  return b[imm];
+  return (unsigned short)b[imm];
 }
 
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))

lib/Headers/smmintrin.h

@@ -213,11 +213,13 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2)
                                                     __a;}))
 #endif /* __x86_64__ */
 
-/* Extract int from packed integer array at index.  */
+/* Extract int from packed integer array at index.  This returns the element
+ * as a zero extended value, so it is unsigned.
+ */
 #define _mm_extract_epi8(X, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \
-                                                 __a[N];}))
+                                                 (unsigned char)__a[N];}))
 #define _mm_extract_epi32(X, N) (__extension__ ({ __v4si __a = (__v4si)(X); \
-                                                  __a[N];}))
+                                                  (unsigned)__a[N];}))
 #ifdef __x86_64__
 #define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
                                                   __a[N];}))

test/CodeGen/vector.c

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -triple i386-apple-darwin9 -target-cpu pentium4 -g -emit-llvm %s -o -
+// RUN: %clang_cc1 -triple i386-apple-darwin9 -O1 -target-cpu pentium4 -target-feature +sse4.1 -g -emit-llvm %s -o - | FileCheck %s
 typedef short __v4hi __attribute__ ((__vector_size__ (8)));
 
 void test1() {
@@ -40,3 +40,16 @@ int test4(int argc, char *argv[]) {
 
   return result;
 }
+
+#include <smmintrin.h>
+
+unsigned long test_epi8(__m128i x) { return _mm_extract_epi8(x, 4); }
+// CHECK: @test_epi8
+// CHECK: extractelement <16 x i8> {{.*}}, i32 4
+// CHECK: zext i8 {{.*}} to i32
+
+unsigned long test_epi16(__m128i x) { return _mm_extract_epi16(x, 3); }
+
+// CHECK: @test_epi16
+// CHECK: extractelement <8 x i16> {{.*}}, i32 3
+// CHECK: zext i16 {{.*}} to i32