[X86] Add missing typecasts in intrinsic macros. This should make them more robust against inputs that aren't already the right type.
llvm-svn: 252700
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index 292f10d..6d1ca54 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -1157,7 +1157,7 @@
*/
#define _mm256_insertf128_ps(V1, V2, M) __extension__ ({ \
(__m256)__builtin_shufflevector( \
- (__v8sf)(V1), \
+ (__v8sf)(__m256)(V1), \
(__v8sf)_mm256_castps128_ps256((__m128)(V2)), \
(((M) & 1) ? 0 : 8), \
(((M) & 1) ? 1 : 9), \
@@ -1170,7 +1170,7 @@
#define _mm256_insertf128_pd(V1, V2, M) __extension__ ({ \
(__m256d)__builtin_shufflevector( \
- (__v4df)(V1), \
+ (__v4df)(__m256d)(V1), \
(__v4df)_mm256_castpd128_pd256((__m128d)(V2)), \
(((M) & 1) ? 0 : 4), \
(((M) & 1) ? 1 : 5), \
@@ -1179,7 +1179,7 @@
#define _mm256_insertf128_si256(V1, V2, M) __extension__ ({ \
(__m256i)__builtin_shufflevector( \
- (__v4di)(V1), \
+ (__v4di)(__m256i)(V1), \
(__v4di)_mm256_castsi128_si256((__m128i)(V2)), \
(((M) & 1) ? 0 : 4), \
(((M) & 1) ? 1 : 5), \
@@ -1193,7 +1193,7 @@
*/
#define _mm256_extractf128_ps(V, M) __extension__ ({ \
(__m128)__builtin_shufflevector( \
- (__v8sf)(V), \
+ (__v8sf)(__m256)(V), \
(__v8sf)(_mm256_setzero_ps()), \
(((M) & 1) ? 4 : 0), \
(((M) & 1) ? 5 : 1), \
@@ -1202,14 +1202,14 @@
#define _mm256_extractf128_pd(V, M) __extension__ ({ \
(__m128d)__builtin_shufflevector( \
- (__v4df)(V), \
+ (__v4df)(__m256d)(V), \
(__v4df)(_mm256_setzero_pd()), \
(((M) & 1) ? 2 : 0), \
(((M) & 1) ? 3 : 1) );})
#define _mm256_extractf128_si256(V, M) __extension__ ({ \
(__m128i)__builtin_shufflevector( \
- (__v4di)(V), \
+ (__v4di)(__m256i)(V), \
(__v4di)(_mm256_setzero_si256()), \
(((M) & 1) ? 2 : 0), \
(((M) & 1) ? 3 : 1) );})