From: Craig Topper Date: Sat, 24 Dec 2011 07:55:14 +0000 (+0000) Subject: Add AVX2 permute intrinsics. Also add parentheses on some macro arguments in other... X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=34a1da4354959522cd1721ce9ca099cc5c743f01;p=clang Add AVX2 permute intrinsics. Also add parentheses on some macro arguments in other intrinsic headers. git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@147241 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Headers/avx2intrin.h b/lib/Headers/avx2intrin.h index dc293f33f3..283227fdcf 100644 --- a/lib/Headers/avx2intrin.h +++ b/lib/Headers/avx2intrin.h @@ -758,9 +758,83 @@ _mm_broadcastsi128_si256(__m128i const *a) #define _mm_blend_epi32(V1, V2, M) __extension__ ({ \ __m128i __V1 = (V1); \ __m128i __V2 = (V2); \ - (__m128i)__builtin_ia32_pblendd128((__v4si)__V1, (__v4si)__V2, M); }) + (__m128i)__builtin_ia32_pblendd128((__v4si)__V1, (__v4si)__V2, (M)); }) #define _mm256_blend_epi32(V1, V2, M) __extension__ ({ \ __m256i __V1 = (V1); \ __m256i __V2 = (V2); \ - (__m256i)__builtin_ia32_pblendd256((__v8si)__V1, (__v8si)__V2, M); }) + (__m256i)__builtin_ia32_pblendd256((__v8si)__V1, (__v8si)__V2, (M)); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastb_epi8(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastb256((__v16qi)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastw_epi16(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastw256((__v8hi)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastd_epi32(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastd256((__v4si)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastq_epi64(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastq256(__X); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastb_epi8(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastb128((__v16qi)__X); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastw_epi16(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastw128((__v8hi)__X); +} + + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastd_epi32(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastd128((__v4si)__X); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastq_epi64(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastq128(__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_permutevar8x32_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_permvarsi256((__v8si)a, (__v8si)b); +} + +#define _mm256_permute4x64_pd(V, M) __extension__ ({ \ + __m256d __V = (V); \ + (__m256d)__builtin_ia32_permdf256((__v4df)__V, (M)); }) + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_permutevar8x32_ps(__m256 a, __m256 b) +{ + return (__m256)__builtin_ia32_permvarsf256((__v8sf)a, (__v8sf)b); +} + +#define _mm256_permute4x64_epi64(V, M) __extension__ ({ \ + __m256i __V = (V); \ + (__m256i)__builtin_ia32_permdi256((__v4di)__V, (M)); }) + +#define _mm256_permute2x128_si256(V1, V2, M) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m256i __V2 = (V2); \ + (__m256i)__builtin_ia32_permti256((__v4di)__V1, (__v4di)__V2, (M)); }) diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h index d33a076392..620ee0fbb0 100644 --- a/lib/Headers/avxintrin.h +++ b/lib/Headers/avxintrin.h @@ -147,11 +147,11 @@ _mm256_rcp_ps(__m256 a) #define _mm256_round_pd(V, M) __extension__ ({ \ __m256d __V = (V); \ - (__m256d)__builtin_ia32_roundpd256((__v4df)__V, M); }) + (__m256d)__builtin_ia32_roundpd256((__v4df)__V, (M)); }) #define _mm256_round_ps(V, M) __extension__ ({ \ __m256 __V = (V); \ - (__m256)__builtin_ia32_roundps256((__v8sf)__V, M); }) + (__m256)__builtin_ia32_roundps256((__v8sf)__V, (M)); }) #define _mm256_ceil_pd(V) _mm256_round_pd((V), _MM_FROUND_CEIL) #define _mm256_floor_pd(V) _mm256_round_pd((V), _MM_FROUND_FLOOR) @@ -260,45 +260,45 @@ _mm256_permutevar_ps(__m256 a, __m256i c) #define _mm_permute_pd(A, C) __extension__ ({ \ __m128d __A = (A); \ - (__m128d)__builtin_ia32_vpermilpd((__v2df)__A, C); }) + (__m128d)__builtin_ia32_vpermilpd((__v2df)__A, (C)); }) #define _mm256_permute_pd(A, C) __extension__ ({ \ __m256d __A = (A); \ - (__m256d)__builtin_ia32_vpermilpd256((__v4df)__A, C); }) + (__m256d)__builtin_ia32_vpermilpd256((__v4df)__A, (C)); }) #define _mm_permute_ps(A, C) __extension__ ({ \ __m128 __A = (A); \ - (__m128)__builtin_ia32_vpermilps((__v4sf)__A, C); }) + (__m128)__builtin_ia32_vpermilps((__v4sf)__A, (C)); }) #define _mm256_permute_ps(A, C) __extension__ ({ \ __m256 __A = (A); \ - (__m256)__builtin_ia32_vpermilps256((__v8sf)__A, C); }) + (__m256)__builtin_ia32_vpermilps256((__v8sf)__A, (C)); }) #define _mm256_permute2f128_pd(V1, V2, M) __extension__ ({ \ __m256d __V1 = (V1); \ __m256d __V2 = (V2); \ - (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)__V1, (__v4df)__V2, M); }) + (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)__V1, (__v4df)__V2, (M)); }) #define _mm256_permute2f128_ps(V1, V2, M) __extension__ ({ \ __m256 __V1 = (V1); \ __m256 __V2 = (V2); \ - (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)__V1, (__v8sf)__V2, M); }) + (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)__V1, (__v8sf)__V2, (M)); }) #define _mm256_permute2f128_si256(V1, V2, M) __extension__ ({ \ __m256i __V1 = (V1); \ __m256i __V2 = (V2); \ - (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)__V1, (__v8si)__V2, M); }) + (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)__V1, (__v8si)__V2, (M)); }) /* Vector Blend */ #define _mm256_blend_pd(V1, V2, M) __extension__ ({ \ __m256d __V1 = (V1); \ __m256d __V2 = (V2); \ - (__m256d)__builtin_ia32_blendpd256((__v4df)__V1, (__v4df)__V2, M); }) + (__m256d)__builtin_ia32_blendpd256((__v4df)__V1, (__v4df)__V2, (M)); }) #define _mm256_blend_ps(V1, V2, M) __extension__ ({ \ __m256 __V1 = (V1); \ __m256 __V2 = (V2); \ - (__m256)__builtin_ia32_blendps256((__v8sf)__V1, (__v8sf)__V2, M); }) + (__m256)__builtin_ia32_blendps256((__v8sf)__V1, (__v8sf)__V2, (M)); }) static __inline __m256d __attribute__((__always_inline__, __nodebug__)) _mm256_blendv_pd(__m256d a, __m256d b, __m256d c) @@ -316,7 +316,7 @@ _mm256_blendv_ps(__m256 a, __m256 b, __m256 c) #define _mm256_dp_ps(V1, V2, M) __extension__ ({ \ __m256 __V1 = (V1); \ __m256 __V2 = (V2); \ - (__m256)__builtin_ia32_dpps256((__v8sf)__V1, (__v8sf)__V2, M); }) + (__m256)__builtin_ia32_dpps256((__v8sf)__V1, (__v8sf)__V2, (M)); }) /* Vector shuffle */ #define _mm256_shuffle_ps(a, b, mask) __extension__ ({ \ @@ -404,15 +404,15 @@ _mm256_blendv_ps(__m256 a, __m256 b, __m256 c) /* Vector extract */ #define _mm256_extractf128_pd(A, O) __extension__ ({ \ __m256d __A = (A); \ - (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)__A, O); }) + (__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); }) + (__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); }) + (__m128i)__builtin_ia32_vextractf128_si256((__v8si)__A, (O)); }) static __inline int __attribute__((__always_inline__, __nodebug__)) _mm256_extract_epi32(__m256i a, int const imm) @@ -448,17 +448,17 @@ _mm256_extract_epi64(__m256i a, const int imm) #define _mm256_insertf128_pd(V1, V2, O) __extension__ ({ \ __m256d __V1 = (V1); \ __m128d __V2 = (V2); \ - (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)__V1, (__v2df)__V2, O); }) + (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)__V1, (__v2df)__V2, (O)); }) #define _mm256_insertf128_ps(V1, V2, O) __extension__ ({ \ __m256 __V1 = (V1); \ __m128 __V2 = (V2); \ - (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)__V1, (__v4sf)__V2, O); }) + (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)__V1, (__v4sf)__V2, (O)); }) #define _mm256_insertf128_si256(V1, V2, O) __extension__ ({ \ __m256i __V1 = (V1); \ __m128i __V2 = (V2); \ - (__m256i)__builtin_ia32_vinsertf128_si256((__v8si)__V1, (__v4si)__V2, O); }) + (__m256i)__builtin_ia32_vinsertf128_si256((__v8si)__V1, (__v4si)__V2, (O)); }) static __inline __m256i __attribute__((__always_inline__, __nodebug__)) _mm256_insert_epi32(__m256i a, int b, int const imm) diff --git a/lib/Headers/xmmintrin.h b/lib/Headers/xmmintrin.h index e7957cb2cd..51ce5a135c 100644 --- a/lib/Headers/xmmintrin.h +++ b/lib/Headers/xmmintrin.h @@ -664,7 +664,7 @@ _mm_storer_ps(float *p, __m128 a) /* FIXME: We have to #define this because "sel" must be a constant integer, and Sema doesn't do any form of constant propagation yet. */ -#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, sel)) +#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, (sel))) static __inline__ void __attribute__((__always_inline__, __nodebug__)) _mm_stream_pi(__m64 *p, __m64 a)