diff options
| author | Michael Zuckerman <Michael.zuckerman@intel.com> | 2016-04-14 07:56:51 +0000 |
|---|---|---|
| committer | Michael Zuckerman <Michael.zuckerman@intel.com> | 2016-04-14 07:56:51 +0000 |
| commit | 0a3508a8d39746b070f9ad4e3aabe3e0c3fbeed8 (patch) | |
| tree | d78014e111e915d4df118e55afe1b001b0f25dd4 /clang/lib | |
| parent | 0f26b0aeb4a91418d4c273bb25ab22f3b416a960 (diff) | |
| download | bcm5719-llvm-0a3508a8d39746b070f9ad4e3aabe3e0c3fbeed8.tar.gz bcm5719-llvm-0a3508a8d39746b070f9ad4e3aabe3e0c3fbeed8.zip | |
[Clang][AVX512][BUILTIN] Adding support for intrinsics of vpmov{d|q}{b|w|d}{128|256|512} instruction set
Differential Revision: http://reviews.llvm.org/D19055
llvm-svn: 266280
Diffstat (limited to 'clang/lib')
| -rw-r--r-- | clang/lib/Headers/avx512fintrin.h | 145 | ||||
| -rw-r--r-- | clang/lib/Headers/avx512vlintrin.h | 292 |
2 files changed, 437 insertions, 0 deletions
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h index 2699127bc67..8d0d9819eba 100644 --- a/clang/lib/Headers/avx512fintrin.h +++ b/clang/lib/Headers/avx512fintrin.h @@ -5913,6 +5913,151 @@ _mm512_mask_cvtusepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A) __builtin_ia32_pmovusqw512mem_mask ((__v8hi*) __P, (__v8di) __A, __M); } +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_cvtepi32_epi8 (__m512i __A) +{ + return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A, + (__v16qi) _mm_undefined_si128 (), + (__mmask16) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi32_epi8 (__m128i __O, __mmask16 __M, __m512i __A) +{ + return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A, + (__v16qi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_maskz_cvtepi32_epi8 (__mmask16 __M, __m512i __A) +{ + return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A, + (__v16qi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi32_storeu_epi8 (void * __P, __mmask16 __M, __m512i __A) +{ + __builtin_ia32_pmovdb512mem_mask ((__v16qi *) __P, (__v16si) __A, __M); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm512_cvtepi32_epi16 (__m512i __A) +{ + return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A, + (__v16hi) _mm256_undefined_si256 (), + (__mmask16) -1); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi32_epi16 (__m256i __O, __mmask16 __M, __m512i __A) +{ + return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A, + (__v16hi) __O, __M); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm512_maskz_cvtepi32_epi16 (__mmask16 __M, __m512i __A) +{ + return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A, + (__v16hi) _mm256_setzero_si256 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi32_storeu_epi16 (void * __P, __mmask16 __M, __m512i __A) +{ + __builtin_ia32_pmovdw512mem_mask ((__v16hi *) __P, (__v16si) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_cvtepi64_epi8 (__m512i __A) +{ + return (__m128i) __builtin_ia32_pmovqb512_mask ((__v8di) __A, + (__v16qi) _mm_undefined_si128 (), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi64_epi8 (__m128i __O, __mmask8 __M, __m512i __A) +{ + return (__m128i) __builtin_ia32_pmovqb512_mask ((__v8di) __A, + (__v16qi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_maskz_cvtepi64_epi8 (__mmask8 __M, __m512i __A) +{ + return (__m128i) __builtin_ia32_pmovqb512_mask ((__v8di) __A, + (__v16qi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m512i __A) +{ + __builtin_ia32_pmovqb512mem_mask ((__v16qi *) __P, (__v8di) __A, __M); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm512_cvtepi64_epi32 (__m512i __A) +{ + return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A, + (__v8si) _mm256_undefined_si256 (), + (__mmask8) -1); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi64_epi32 (__m256i __O, __mmask8 __M, __m512i __A) +{ + return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A, + (__v8si) __O, __M); +} + +static __inline__ __m256i __DEFAULT_FN_ATTRS +_mm512_maskz_cvtepi64_epi32 (__mmask8 __M, __m512i __A) +{ + return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A, + (__v8si) _mm256_setzero_si256 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi64_storeu_epi32 (void* __P, __mmask8 __M, __m512i __A) +{ + __builtin_ia32_pmovqd512mem_mask ((__v8si *) __P, (__v8di) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_cvtepi64_epi16 (__m512i __A) +{ + return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A, + (__v8hi) _mm_undefined_si128 (), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m512i __A) +{ + return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A, + (__v8hi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm512_maskz_cvtepi64_epi16 (__mmask8 __M, __m512i __A) +{ + return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A, + (__v8hi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A) +{ + __builtin_ia32_pmovqw512mem_mask ((__v8hi *) __P, (__v8di) __A, __M); +} + #undef __DEFAULT_FN_ATTRS #endif // __AVX512FINTRIN_H diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h index df00884fec3..0f0d5c7fbe9 100644 --- a/clang/lib/Headers/avx512vlintrin.h +++ b/clang/lib/Headers/avx512vlintrin.h @@ -8512,6 +8512,298 @@ _mm256_mask_cvtusepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A) return __builtin_ia32_pmovusqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M); } +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_cvtepi32_epi8 (__m128i __A) +{ + return (__m128i) __builtin_ia32_pmovdb128_mask ((__v4si) __A, + (__v16qi)_mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_cvtepi32_epi8 (__m128i __O, __mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovdb128_mask ((__v4si) __A, + (__v16qi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_cvtepi32_epi8 (__mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovdb128_mask ((__v4si) __A, + (__v16qi) + _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm_mask_cvtepi32_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A) +{ + __builtin_ia32_pmovdb128mem_mask ((__v16qi *) __P, (__v4si) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_cvtepi32_epi8 (__m256i __A) +{ + return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A, + (__v16qi)_mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi32_epi8 (__m128i __O, __mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A, + (__v16qi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_maskz_cvtepi32_epi8 (__mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A, + (__v16qi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi32_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A) +{ + __builtin_ia32_pmovdb256mem_mask ((__v16qi *) __P, (__v8si) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_cvtepi32_epi16 (__m128i __A) +{ + return (__m128i) __builtin_ia32_pmovdw128_mask ((__v4si) __A, + (__v8hi) _mm_setzero_si128 (), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_cvtepi32_epi16 (__m128i __O, __mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovdw128_mask ((__v4si) __A, + (__v8hi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_cvtepi32_epi16 (__mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovdw128_mask ((__v4si) __A, + (__v8hi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A) +{ + __builtin_ia32_pmovdw128mem_mask ((__v8hi *) __P, (__v4si) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_cvtepi32_epi16 (__m256i __A) +{ + return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A, + (__v8hi)_mm_setzero_si128 (), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi32_epi16 (__m128i __O, __mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A, + (__v8hi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_maskz_cvtepi32_epi16 (__mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A, + (__v8hi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A) +{ + __builtin_ia32_pmovdw256mem_mask ((__v8hi *) __P, (__v8si) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_cvtepi64_epi8 (__m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqb128_mask ((__v2di) __A, + (__v16qi) _mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_cvtepi64_epi8 (__m128i __O, __mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqb128_mask ((__v2di) __A, + (__v16qi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_cvtepi64_epi8 (__mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqb128_mask ((__v2di) __A, + (__v16qi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A) +{ + __builtin_ia32_pmovqb128mem_mask ((__v16qi *) __P, (__v2di) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_cvtepi64_epi8 (__m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqb256_mask ((__v4di) __A, + (__v16qi) _mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi64_epi8 (__m128i __O, __mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqb256_mask ((__v4di) __A, + (__v16qi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_maskz_cvtepi64_epi8 (__mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqb256_mask ((__v4di) __A, + (__v16qi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A) +{ + __builtin_ia32_pmovqb256mem_mask ((__v16qi *) __P, (__v4di) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_cvtepi64_epi32 (__m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqd128_mask ((__v2di) __A, + (__v4si)_mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_cvtepi64_epi32 (__m128i __O, __mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqd128_mask ((__v2di) __A, + (__v4si) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_cvtepi64_epi32 (__mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqd128_mask ((__v2di) __A, + (__v4si) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m128i __A) +{ + __builtin_ia32_pmovqd128mem_mask ((__v4si *) __P, (__v2di) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_cvtepi64_epi32 (__m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A, + (__v4si) _mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi64_epi32 (__m128i __O, __mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A, + (__v4si) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_maskz_cvtepi64_epi32 (__mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A, + (__v4si) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m256i __A) +{ + __builtin_ia32_pmovqd256mem_mask ((__v4si *) __P, (__v4di) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_cvtepi64_epi16 (__m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqw128_mask ((__v2di) __A, + (__v8hi) _mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqw128_mask ((__v2di) __A, + (__v8hi)__O, + __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm_maskz_cvtepi64_epi16 (__mmask8 __M, __m128i __A) +{ + return (__m128i) __builtin_ia32_pmovqw128_mask ((__v2di) __A, + (__v8hi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A) +{ + __builtin_ia32_pmovqw128mem_mask ((__v8hi *) __P, (__v2di) __A, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_cvtepi64_epi16 (__m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqw256_mask ((__v4di) __A, + (__v8hi)_mm_undefined_si128(), + (__mmask8) -1); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqw256_mask ((__v4di) __A, + (__v8hi) __O, __M); +} + +static __inline__ __m128i __DEFAULT_FN_ATTRS +_mm256_maskz_cvtepi64_epi16 (__mmask8 __M, __m256i __A) +{ + return (__m128i) __builtin_ia32_pmovqw256_mask ((__v4di) __A, + (__v8hi) _mm_setzero_si128 (), + __M); +} + +static __inline__ void __DEFAULT_FN_ATTRS +_mm256_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A) +{ + __builtin_ia32_pmovqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M); +} + #undef __DEFAULT_FN_ATTRS #undef __DEFAULT_FN_ATTRS_BOTH |

