diff options
author | Craig Topper <craig.topper@intel.com> | 2018-08-16 06:20:29 +0000 |
---|---|---|
committer | Craig Topper <craig.topper@intel.com> | 2018-08-16 06:20:29 +0000 |
commit | 347daedef2a65151b13d1e6c6feb813b6d9b0c93 (patch) | |
tree | 25a5169af576398cacdafff3935e49a2aa07ace5 /lib/Headers/avx512bwintrin.h | |
parent | 9327733b3c555b951f46da9dffaabeef5ebb783b (diff) | |
download | clang-347daedef2a65151b13d1e6c6feb813b6d9b0c93.tar.gz |
[X86] Remove masking from the 512-bit padds and psubs builtins. Use select builtin instead.
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@339843 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Headers/avx512bwintrin.h')
-rw-r--r-- | lib/Headers/avx512bwintrin.h | 88 |
1 files changed, 32 insertions, 56 deletions
diff --git a/lib/Headers/avx512bwintrin.h b/lib/Headers/avx512bwintrin.h index fc46323749..7eda8a6dde 100644 --- a/lib/Headers/avx512bwintrin.h +++ b/lib/Headers/avx512bwintrin.h @@ -422,57 +422,45 @@ _mm512_maskz_packus_epi16(__mmask64 __M, __m512i __A, __m512i __B) static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_adds_epi8 (__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_paddsb512_mask ((__v64qi) __A, - (__v64qi) __B, - (__v64qi) _mm512_setzero_si512(), - (__mmask64) -1); + return (__m512i)__builtin_ia32_paddsb512((__v64qi)__A, (__v64qi)__B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_adds_epi8 (__m512i __W, __mmask64 __U, __m512i __A, - __m512i __B) +_mm512_mask_adds_epi8 (__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_paddsb512_mask ((__v64qi) __A, - (__v64qi) __B, - (__v64qi) __W, - (__mmask64) __U); + return (__m512i)__builtin_ia32_selectb_512((__mmask64)__U, + (__v64qi)_mm512_adds_epi8(__A, __B), + (__v64qi)__W); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_adds_epi8 (__mmask64 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_paddsb512_mask ((__v64qi) __A, - (__v64qi) __B, - (__v64qi) _mm512_setzero_si512(), - (__mmask64) __U); + return (__m512i)__builtin_ia32_selectb_512((__mmask64)__U, + (__v64qi)_mm512_adds_epi8(__A, __B), + (__v64qi)_mm512_setzero_si512()); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_adds_epi16 (__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_paddsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_si512(), - (__mmask32) -1); + return (__m512i)__builtin_ia32_paddsw512((__v32hi)__A, (__v32hi)__B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_adds_epi16 (__m512i __W, __mmask32 __U, __m512i __A, - __m512i __B) +_mm512_mask_adds_epi16 (__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_paddsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) __W, - (__mmask32) __U); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_adds_epi16(__A, __B), + (__v32hi)__W); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_adds_epi16 (__mmask32 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_paddsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_si512(), - (__mmask32) __U); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_adds_epi16(__A, __B), + (__v32hi)_mm512_setzero_si512()); } static __inline__ __m512i __DEFAULT_FN_ATTRS @@ -787,57 +775,45 @@ _mm512_maskz_shuffle_epi8(__mmask64 __U, __m512i __A, __m512i __B) static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_subs_epi8 (__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_psubsb512_mask ((__v64qi) __A, - (__v64qi) __B, - (__v64qi) _mm512_setzero_si512(), - (__mmask64) -1); + return (__m512i)__builtin_ia32_psubsb512((__v64qi)__A, (__v64qi)__B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_subs_epi8 (__m512i __W, __mmask64 __U, __m512i __A, - __m512i __B) +_mm512_mask_subs_epi8 (__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_psubsb512_mask ((__v64qi) __A, - (__v64qi) __B, - (__v64qi) __W, - (__mmask64) __U); + return (__m512i)__builtin_ia32_selectb_512((__mmask64)__U, + (__v64qi)_mm512_subs_epi8(__A, __B), + (__v64qi)__W); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_subs_epi8 (__mmask64 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_psubsb512_mask ((__v64qi) __A, - (__v64qi) __B, - (__v64qi) _mm512_setzero_si512(), - (__mmask64) __U); + return (__m512i)__builtin_ia32_selectb_512((__mmask64)__U, + (__v64qi)_mm512_subs_epi8(__A, __B), + (__v64qi)_mm512_setzero_si512()); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_subs_epi16 (__m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_psubsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_si512(), - (__mmask32) -1); + return (__m512i)__builtin_ia32_psubsw512((__v32hi)__A, (__v32hi)__B); } static __inline__ __m512i __DEFAULT_FN_ATTRS -_mm512_mask_subs_epi16 (__m512i __W, __mmask32 __U, __m512i __A, - __m512i __B) +_mm512_mask_subs_epi16 (__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_psubsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) __W, - (__mmask32) __U); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_subs_epi16(__A, __B), + (__v32hi)__W); } static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_subs_epi16 (__mmask32 __U, __m512i __A, __m512i __B) { - return (__m512i) __builtin_ia32_psubsw512_mask ((__v32hi) __A, - (__v32hi) __B, - (__v32hi) _mm512_setzero_si512(), - (__mmask32) __U); + return (__m512i)__builtin_ia32_selectw_512((__mmask32)__U, + (__v32hi)_mm512_subs_epi16(__A, __B), + (__v32hi)_mm512_setzero_si512()); } static __inline__ __m512i __DEFAULT_FN_ATTRS |