diff options
author | Michael Zuckerman <Michael.zuckerman@intel.com> | 2016-04-11 17:04:21 +0000 |
---|---|---|
committer | Michael Zuckerman <Michael.zuckerman@intel.com> | 2016-04-11 17:04:21 +0000 |
commit | 81f468c8599a0735d832ae7646128760c5820bc4 (patch) | |
tree | 8178cee7a6f73478ff8f1a6afe99a0c8a0780604 /clang/lib | |
parent | 88cbe5a1ccad5b8a937d5dd7488eabf4363317c8 (diff) | |
download | bcm5719-llvm-81f468c8599a0735d832ae7646128760c5820bc4.tar.gz bcm5719-llvm-81f468c8599a0735d832ae7646128760c5820bc4.zip |
[Clang][AVX512][BuiltIn] Adding avx512 ( psll{d|q}512,psllv{16si|8di},psra{d|q}512,psrav{16si|8di},pternlog{d|q}{128|256|512} ) builtin to clang
Differential Revision: http://reviews.llvm.org/D18926
llvm-svn: 265964
Diffstat (limited to 'clang/lib')
-rw-r--r-- | clang/lib/Headers/avx512fintrin.h | 390 | ||||
-rw-r--r-- | clang/lib/Headers/avx512vlintrin.h | 89 |
2 files changed, 479 insertions, 0 deletions
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h index e656e7cd168..620e096ec67 100644 --- a/clang/lib/Headers/avx512fintrin.h +++ b/clang/lib/Headers/avx512fintrin.h @@ -4005,6 +4005,396 @@ _mm512_unpackhi_epi32 (__m512i __A, __m512i __B) (__mmask16) -1); } +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_sll_epi32 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_sll_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) __W, + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_sll_epi32 (__mmask16 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_sll_epi64 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_sll_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) __W, + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_sll_epi64 (__mmask8 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_sllv_epi32 (__m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psllv16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_sllv_epi32 (__m512i __W, __mmask16 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psllv16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) __W, + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_sllv_epi32 (__mmask16 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psllv16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_sllv_epi64 (__m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psllv8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) + _mm512_undefined_pd (), + (__mmask8) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_sllv_epi64 (__m512i __W, __mmask8 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psllv8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) __W, + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_sllv_epi64 (__mmask8 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psllv8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_sra_epi32 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_sra_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) __W, + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_sra_epi32 (__mmask16 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_sra_epi64 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_sra_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) __W, + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_sra_epi64 (__mmask8 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srav_epi32 (__m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrav16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srav_epi32 (__m512i __W, __mmask16 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrav16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) __W, + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srav_epi32 (__mmask16 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrav16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srav_epi64 (__m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrav8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srav_epi64 (__m512i __W, __mmask8 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrav8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) __W, + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srav_epi64 (__mmask8 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrav8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srl_epi32 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srl_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) __W, + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srl_epi32 (__mmask16 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A, + (__v4si) __B, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srl_epi64 (__m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srl_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) __W, + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srl_epi64 (__mmask8 __U, __m512i __A, __m128i __B) +{ + return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A, + (__v2di) __B, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srlv_epi32 (__m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrlv16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srlv_epi32 (__m512i __W, __mmask16 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrlv16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) __W, + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srlv_epi32 (__mmask16 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrlv16si_mask ((__v16si) __X, + (__v16si) __Y, + (__v16si) + _mm512_setzero_si512 (), + (__mmask16) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_srlv_epi64 (__m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrlv8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) -1); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_mask_srlv_epi64 (__m512i __W, __mmask8 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrlv8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) __W, + (__mmask8) __U); +} + +static __inline__ __m512i __DEFAULT_FN_ATTRS +_mm512_maskz_srlv_epi64 (__mmask8 __U, __m512i __X, __m512i __Y) +{ + return (__m512i) __builtin_ia32_psrlv8di_mask ((__v8di) __X, + (__v8di) __Y, + (__v8di) + _mm512_setzero_si512 (), + (__mmask8) __U); +} + +#define _mm512_ternarylogic_epi32( __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd512_mask ((__v16si)( __A),\ + (__v16si)( __B),\ + (__v16si)( __C),\ + ( imm), (__mmask16) -1);\ +}) + +#define _mm512_mask_ternarylogic_epi32( __A, __U, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd512_mask ((__v16si)( __A),\ + (__v16si)( __B),\ + (__v16si)( __C),\ + ( imm), (__mmask16)( __U));\ +}) + +#define _mm512_maskz_ternarylogic_epi32( __U, __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd512_maskz ((__v16si)( __A),\ + (__v16si)( __B),\ + (__v16si)( __C),\ + ( imm), (__mmask16)( __U));\ +}) + +#define _mm512_ternarylogic_epi64( __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq512_mask ((__v8di)( __A),\ + (__v8di)( __B),\ + (__v8di)( __C),( imm),\ + (__mmask8) -1);\ +}) + +#define _mm512_mask_ternarylogic_epi64( __A, __U, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq512_mask ((__v8di)( __A),\ + (__v8di)( __B),\ + (__v8di)( __C),( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm512_maskz_ternarylogic_epi64( __U, __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq512_maskz ((__v8di)( __A),\ + (__v8di)( __B),\ + (__v8di)( __C),\ + ( imm), (__mmask8)( __U));\ +}) + static __inline__ __m512d __DEFAULT_FN_ATTRS _mm512_maskz_unpackhi_pd (__mmask8 __U, __m512d __A, __m512d __B) { diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h index fc947d563e7..acaaae58549 100644 --- a/clang/lib/Headers/avx512vlintrin.h +++ b/clang/lib/Headers/avx512vlintrin.h @@ -7407,6 +7407,95 @@ __builtin_ia32_psraqi256_mask ((__v4di)( __A),( __imm),\ (__mmask8)( __U));\ }) +#define _mm_ternarylogic_epi32( __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd128_mask ((__v4si)( __A),\ + (__v4si)( __B),\ + (__v4si)( __C),( imm),\ + (__mmask8) -1);\ +}) + +#define _mm_mask_ternarylogic_epi32( __A, __U, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd128_mask ((__v4si)( __A),\ + (__v4si)( __B),\ + (__v4si)( __C),( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm_maskz_ternarylogic_epi32( __U, __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd128_maskz ((__v4si)( __A),\ + (__v4si)( __B),\ + (__v4si)( __C),\ + ( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm256_ternarylogic_epi32( __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd256_mask ((__v8si)( __A),\ + (__v8si)( __B),\ + (__v8si)( __C),( imm),\ + (__mmask8) -1);\ +}) + +#define _mm256_mask_ternarylogic_epi32( __A, __U, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd256_mask ((__v8si)( __A),\ + (__v8si)( __B),\ + (__v8si)( __C),( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm256_maskz_ternarylogic_epi32( __U, __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogd256_maskz ((__v8si)( __A),\ + (__v8si)( __B),\ + (__v8si)( __C),\ + ( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm_ternarylogic_epi64( __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq128_mask ((__v2di)( __A),\ + (__v2di)( __B),\ + (__v2di)( __C),( imm),\ + (__mmask8) -1);\ +}) + +#define _mm_mask_ternarylogic_epi64( __A, __U, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq128_mask ((__v2di)( __A),\ + (__v2di)( __B),\ + (__v2di)( __C),( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm_maskz_ternarylogic_epi64( __U, __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq128_maskz ((__v2di)( __A),\ + (__v2di)( __B),\ + (__v2di)( __C),\ + ( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm256_ternarylogic_epi64( __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq256_mask ((__v4di)( __A),\ + (__v4di)( __B),\ + (__v4di)( __C),( imm),\ + (__mmask8) -1);\ +}) + +#define _mm256_mask_ternarylogic_epi64( __A, __U, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq256_mask ((__v4di)( __A),\ + (__v4di)( __B),\ + (__v4di)( __C),( imm),\ + (__mmask8)( __U));\ +}) + +#define _mm256_maskz_ternarylogic_epi64( __U, __A, __B, __C, imm) __extension__ ({ \ +__builtin_ia32_pternlogq256_maskz ((__v4di)( __A),\ + (__v4di)( __B),\ + (__v4di)( __C),\ + ( imm),\ + (__mmask8)( __U));\ +}) + + #undef __DEFAULT_FN_ATTRS #undef __DEFAULT_FN_ATTRS_BOTH |