diff options
| author | Craig Topper <craig.topper@gmail.com> | 2016-12-11 01:26:52 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@gmail.com> | 2016-12-11 01:26:52 +0000 |
| commit | 5074e7e38b126b06700c3aa396c58daa9c99652f (patch) | |
| tree | 1634286e5b1ecbd8f7309a4bc2b1ff7b3e46f8d3 /lib/Headers/avx512fintrin.h | |
| parent | f2ba17e2d09858e3be53511c94d970cfd166729c (diff) | |
[AVX-512] Remove masking from 512-bit vpermil builtins. The backend now has versions without masking so wrap it with select.
This will allow the backend to constant fold these to generic shuffle vectors like 128-bit and 256-bit without having to working about handling masking.
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@289351 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Headers/avx512fintrin.h')
| -rw-r--r-- | lib/Headers/avx512fintrin.h | 54 |
1 files changed, 20 insertions, 34 deletions
diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h index 86999574c8..71b645230d 100644 --- a/lib/Headers/avx512fintrin.h +++ b/lib/Headers/avx512fintrin.h @@ -6588,61 +6588,47 @@ _mm512_mask2_permutex2var_epi64 (__m512i __A, __m512i __I, (__v16sf)_mm512_setzero_ps()); }) static __inline__ __m512d __DEFAULT_FN_ATTRS -_mm512_permutevar_pd (__m512d __A, __m512i __C) +_mm512_permutevar_pd(__m512d __A, __m512i __C) { - return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A, - (__v8di) __C, - (__v8df) - _mm512_undefined_pd (), - (__mmask8) -1); + return (__m512d)__builtin_ia32_vpermilvarpd512((__v8df)__A, (__v8di)__C); } static __inline__ __m512d __DEFAULT_FN_ATTRS -_mm512_mask_permutevar_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512i __C) +_mm512_mask_permutevar_pd(__m512d __W, __mmask8 __U, __m512d __A, __m512i __C) { - return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A, - (__v8di) __C, - (__v8df) __W, - (__mmask8) __U); + return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, + (__v8df)_mm512_permutevar_pd(__A, __C), + (__v8df)__W); } static __inline__ __m512d __DEFAULT_FN_ATTRS -_mm512_maskz_permutevar_pd (__mmask8 __U, __m512d __A, __m512i __C) +_mm512_maskz_permutevar_pd(__mmask8 __U, __m512d __A, __m512i __C) { - return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A, - (__v8di) __C, - (__v8df) - _mm512_setzero_pd (), - (__mmask8) __U); + return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__U, + (__v8df)_mm512_permutevar_pd(__A, __C), + (__v8df)_mm512_setzero_pd()); } static __inline__ __m512 __DEFAULT_FN_ATTRS -_mm512_permutevar_ps (__m512 __A, __m512i __C) +_mm512_permutevar_ps(__m512 __A, __m512i __C) { - return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A, - (__v16si) __C, - (__v16sf) - _mm512_undefined_ps (), - (__mmask16) -1); + return (__m512)__builtin_ia32_vpermilvarps512((__v16sf)__A, (__v16si)__C); } static __inline__ __m512 __DEFAULT_FN_ATTRS -_mm512_mask_permutevar_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512i __C) +_mm512_mask_permutevar_ps(__m512 __W, __mmask16 __U, __m512 __A, __m512i __C) { - return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A, - (__v16si) __C, - (__v16sf) __W, - (__mmask16) __U); + return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, + (__v16sf)_mm512_permutevar_ps(__A, __C), + (__v16sf)__W); } static __inline__ __m512 __DEFAULT_FN_ATTRS -_mm512_maskz_permutevar_ps (__mmask16 __U, __m512 __A, __m512i __C) +_mm512_maskz_permutevar_ps(__mmask16 __U, __m512 __A, __m512i __C) { - return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A, - (__v16si) __C, - (__v16sf) - _mm512_setzero_ps (), - (__mmask16) __U); + return (__m512)__builtin_ia32_selectps_512((__mmask16)__U, + (__v16sf)_mm512_permutevar_ps(__A, __C), + (__v16sf)_mm512_setzero_ps()); } static __inline __m512d __DEFAULT_FN_ATTRS |