diff options
author | Michael Zuckerman <Michael.zuckerman@intel.com> | 2016-04-11 15:46:39 +0000 |
---|---|---|
committer | Michael Zuckerman <Michael.zuckerman@intel.com> | 2016-04-11 15:46:39 +0000 |
commit | 6b5f4d8ad1929e7dc0ac338c5c536c19352f4cbe (patch) | |
tree | 141a4e28074c90e74b8e856f83a0970a7ef1386c /clang/lib | |
parent | 6eaff5cec66e6e8164b53823b3b113f96269a6fc (diff) | |
download | bcm5719-llvm-6b5f4d8ad1929e7dc0ac338c5c536c19352f4cbe.tar.gz bcm5719-llvm-6b5f4d8ad1929e7dc0ac338c5c536c19352f4cbe.zip |
[CLANG] [AVX512] [BUILTIN] Adding PSRA{Q|D|QI|DI}{128|256|512} builtin
Differential Revision: http://reviews.llvm.org/D17693
llvm-svn: 265952
Diffstat (limited to 'clang/lib')
-rw-r--r-- | clang/lib/Headers/avx512fintrin.h | 43 | ||||
-rw-r--r-- | clang/lib/Headers/avx512vlintrin.h | 166 |
2 files changed, 209 insertions, 0 deletions
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h index edcab8cd116..e656e7cd168 100644 --- a/clang/lib/Headers/avx512fintrin.h +++ b/clang/lib/Headers/avx512fintrin.h @@ -4874,6 +4874,49 @@ __builtin_ia32_scalefss_round_mask ((__v4sf)( __A),\ _MM_FROUND_CUR_DIRECTION);\ }) +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srai_epi32 (__m512i __A, unsigned int __B) +{ + return (__m512i) __builtin_ia32_psradi512_mask ((__v16si) __A, __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +#define _mm512_mask_srai_epi32( __W, __U, __A, __B) __extension__ ({ \ +__builtin_ia32_psradi512_mask ((__v16si)( __A),( __B),\ + (__v16si)( __W),\ + (__mmask16)( __U));\ +}) + +#define _mm512_maskz_srai_epi32( __U, __A, __B) __extension__ ({ \ +__builtin_ia32_psradi512_mask ((__v16si)( __A),( __B),\ + (__v16si)\ + _mm512_setzero_si512 (),\ + (__mmask16)( __U));\ +}) + +#define _mm512_srai_epi64( __A, __B) __extension__ ({ \ +__builtin_ia32_psraqi512_mask ((__v8di)( __A),( __B),\ + (__v8di)\ + _mm512_setzero_si512 (),\ + (__mmask8) -1);\ +}) + +#define _mm512_mask_srai_epi64( __W, __U, __A, __B) __extension__ ({ \ +__builtin_ia32_psraqi512_mask ((__v8di)( __A),( __B),\ + (__v8di)( __W),\ + (__mmask8)( __U));\ +}) + +#define _mm512_maskz_srai_epi64( __U, __A, __B) __extension__ ({ \ +__builtin_ia32_psraqi512_mask ((__v8di)( __A),( __B),\ + (__v8di)\ + _mm512_setzero_si512 (),\ + (__mmask8)( __U));\ +}) + + #undef __DEFAULT_FN_ATTRS #endif // __AVX512FINTRIN_H diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h index 8e6230d5a08..fc947d563e7 100644 --- a/clang/lib/Headers/avx512vlintrin.h +++ b/clang/lib/Headers/avx512vlintrin.h @@ -7241,6 +7241,172 @@ _mm256_maskz_unpacklo_epi64 (__mmask8 __U, __m256i __A, __m256i __B) (__mmask8) __U); } +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_sra_epi32 (__m128i __W, __mmask8 __U, __m128i __A, + __m128i __B) +{ + return (__m128i) __builtin_ia32_psrad128_mask ((__v4si) __A, + (__v4si) __B, + (__v4si) __W, + (__mmask8) __U); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_sra_epi32 (__mmask8 __U, __m128i __A, __m128i __B) +{ + return (__m128i) __builtin_ia32_psrad128_mask ((__v4si) __A, + (__v4si) __B, + (__v4si) + _mm_setzero_si128 (), + (__mmask8) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_mask_sra_epi32 (__m256i __W, __mmask8 __U, __m256i __A, + __m128i __B) +{ + return (__m256i) __builtin_ia32_psrad256_mask ((__v8si) __A, + (__v4si) __B, + (__v8si) __W, + (__mmask8) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_maskz_sra_epi32 (__mmask8 __U, __m256i __A, __m128i __B) +{ + return (__m256i) __builtin_ia32_psrad256_mask ((__v8si) __A, + (__v4si) __B, + (__v8si) + _mm256_setzero_si256 (), + (__mmask8) __U); +} + +#define _mm_mask_srai_epi32( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psradi128_mask ((__v4si)( __A),( __imm),\ + (__v4si)( __W),\ + (__mmask8)( __U));\ +}) + +#define _mm_maskz_srai_epi32( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psradi128_mask ((__v4si)( __A),( __imm),\ + (__v4si)\ + _mm_setzero_si128 (),\ + (__mmask8)( __U));\ +}) + +#define _mm256_mask_srai_epi32( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psradi256_mask ((__v8si)( __A),( __imm),\ + (__v8si)( __W),\ + (__mmask8)( __U));\ +}) + +#define _mm256_maskz_srai_epi32( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psradi256_mask ((__v8si)( __A),( __imm),\ + (__v8si)\ + _mm256_setzero_si256 (),\ + (__mmask8)( __U));\ +}) + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_sra_epi64 (__m128i __A, __m128i __B) +{ + return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A, + (__v2di) __B, + (__v2di) + _mm_setzero_di (), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_sra_epi64 (__m128i __W, __mmask8 __U, __m128i __A, + __m128i __B) +{ + return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A, + (__v2di) __B, + (__v2di) __W, + (__mmask8) __U); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_sra_epi64 (__mmask8 __U, __m128i __A, __m128i __B) +{ + return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A, + (__v2di) __B, + (__v2di) + _mm_setzero_di (), + (__mmask8) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_sra_epi64 (__m256i __A, __m128i __B) +{ + return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A, + (__v2di) __B, + (__v4di) + _mm256_setzero_si256 (), + (__mmask8) -1); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_mask_sra_epi64 (__m256i __W, __mmask8 __U, __m256i __A, + __m128i __B) +{ + return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A, + (__v2di) __B, + (__v4di) __W, + (__mmask8) __U); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm256_maskz_sra_epi64 (__mmask8 __U, __m256i __A, __m128i __B) +{ + return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A, + (__v2di) __B, + (__v4di) + _mm256_setzero_si256 (), + (__mmask8) __U); +} + +#define _mm_srai_epi64( __A, __imm) __extension__ ({ \ +__builtin_ia32_psraqi128_mask ((__v2di)( __A),( __imm),\ + (__v2di)\ + _mm_setzero_di (),\ + (__mmask8) -1);\ +}) + +#define _mm_mask_srai_epi64( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psraqi128_mask ((__v2di)( __A),( __imm),\ + (__v2di)( __W),\ + (__mmask8)( __U));\ +}) + +#define _mm_maskz_srai_epi64( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psraqi128_mask ((__v2di)( __A),( __imm),\ + (__v2di)\ + _mm_setzero_si128 (),\ + (__mmask8)( __U));\ +}) + +#define _mm256_srai_epi64( __A, __imm) __extension__ ({ \ +__builtin_ia32_psraqi256_mask ((__v4di)( __A),( __imm),\ + (__v4di)\ + _mm256_setzero_si256 (),\ + (__mmask8) -1);\ +}) + +#define _mm256_mask_srai_epi64( __W, __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psraqi256_mask ((__v4di)( __A),( __imm),\ + (__v4di)( __W),\ + (__mmask8)( __U));\ +}) + +#define _mm256_maskz_srai_epi64( __U, __A, __imm) __extension__ ({ \ +__builtin_ia32_psraqi256_mask ((__v4di)( __A),( __imm),\ + (__v4di)\ + _mm256_setzero_si256 (),\ + (__mmask8)( __U));\ +}) + #undef __DEFAULT_FN_ATTRS #undef __DEFAULT_FN_ATTRS_BOTH |