[X86, AVX] replace vextractf128 intrinsics with generic shuffles
This is very much like D8088 (checked in at r231792).
Now that we've replaced the vinsertf128 intrinsics,
do the same for their extract twins.
Differential Revision: http://reviews.llvm.org/D8275
llvm-svn: 232052
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index 2d1735e..f7bda6b 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -429,19 +429,6 @@
__m128 __b = (b); \
(__m128)__builtin_ia32_cmpss((__v4sf)__a, (__v4sf)__b, (c)); })
-/* Vector extract */
-#define _mm256_extractf128_pd(A, O) __extension__ ({ \
- __m256d __A = (A); \
- (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)__A, (O)); })
-
-#define _mm256_extractf128_ps(A, O) __extension__ ({ \
- __m256 __A = (A); \
- (__m128)__builtin_ia32_vextractf128_ps256((__v8sf)__A, (O)); })
-
-#define _mm256_extractf128_si256(A, O) __extension__ ({ \
- __m256i __A = (A); \
- (__m128i)__builtin_ia32_vextractf128_si256((__v8si)__A, (O)); })
-
static __inline int __attribute__((__always_inline__, __nodebug__))
_mm256_extract_epi32(__m256i __a, const int __imm)
{
@@ -1186,6 +1173,34 @@
(((M) & 1) ? 4 : 2), \
(((M) & 1) ? 5 : 3) );})
+/*
+ Vector extract.
+ We use macros rather than inlines because we only want to accept
+ invocations where the immediate M is a constant expression.
+*/
+#define _mm256_extractf128_ps(V, M) __extension__ ({ \
+ (__m128)__builtin_shufflevector( \
+ (__v8sf)(V), \
+ (__v8sf)(V), \
+ (((M) & 1) ? 4 : 0), \
+ (((M) & 1) ? 5 : 1), \
+ (((M) & 1) ? 6 : 2), \
+ (((M) & 1) ? 7 : 3) );})
+
+#define _mm256_extractf128_pd(V, M) __extension__ ({ \
+ (__m128d)__builtin_shufflevector( \
+ (__v4df)(V), \
+ (__v4df)(V), \
+ (((M) & 1) ? 2 : 0), \
+ (((M) & 1) ? 3 : 1) );})
+
+#define _mm256_extractf128_si256(V, M) __extension__ ({ \
+ (__m128i)__builtin_shufflevector( \
+ (__v4di)(V), \
+ (__v4di)(V), \
+ (((M) & 1) ? 2 : 0), \
+ (((M) & 1) ? 3 : 1) );})
+
/* SIMD load ops (unaligned) */
static __inline __m256 __attribute__((__always_inline__, __nodebug__))
_mm256_loadu2_m128(float const *__addr_hi, float const *__addr_lo)