Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | /*===---- avx512vpopcntdqintrin.h - AVX512VPOPCNTDQ intrinsics -------------=== |
| 2 | * |
||
| 3 | * |
||
| 4 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
||
| 5 | * See https://llvm.org/LICENSE.txt for license information. |
||
| 6 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
||
| 7 | * |
||
| 8 | *===-----------------------------------------------------------------------=== |
||
| 9 | */ |
||
| 10 | #ifndef __IMMINTRIN_H |
||
| 11 | #error \ |
||
| 12 | "Never use <avx512vpopcntdqvlintrin.h> directly; include <immintrin.h> instead." |
||
| 13 | #endif |
||
| 14 | |||
| 15 | #ifndef __AVX512VPOPCNTDQVLINTRIN_H |
||
| 16 | #define __AVX512VPOPCNTDQVLINTRIN_H |
||
| 17 | |||
| 18 | /* Define the default attributes for the functions in this file. */ |
||
| 19 | #define __DEFAULT_FN_ATTRS128 \ |
||
| 20 | __attribute__((__always_inline__, __nodebug__, __target__("avx512vpopcntdq,avx512vl"), __min_vector_width__(128))) |
||
| 21 | #define __DEFAULT_FN_ATTRS256 \ |
||
| 22 | __attribute__((__always_inline__, __nodebug__, __target__("avx512vpopcntdq,avx512vl"), __min_vector_width__(256))) |
||
| 23 | |||
| 24 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 25 | _mm_popcnt_epi64(__m128i __A) { |
||
| 26 | return (__m128i)__builtin_ia32_vpopcntq_128((__v2di)__A); |
||
| 27 | } |
||
| 28 | |||
| 29 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 30 | _mm_mask_popcnt_epi64(__m128i __W, __mmask8 __U, __m128i __A) { |
||
| 31 | return (__m128i)__builtin_ia32_selectq_128( |
||
| 32 | (__mmask8)__U, (__v2di)_mm_popcnt_epi64(__A), (__v2di)__W); |
||
| 33 | } |
||
| 34 | |||
| 35 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 36 | _mm_maskz_popcnt_epi64(__mmask8 __U, __m128i __A) { |
||
| 37 | return _mm_mask_popcnt_epi64((__m128i)_mm_setzero_si128(), __U, __A); |
||
| 38 | } |
||
| 39 | |||
| 40 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 41 | _mm_popcnt_epi32(__m128i __A) { |
||
| 42 | return (__m128i)__builtin_ia32_vpopcntd_128((__v4si)__A); |
||
| 43 | } |
||
| 44 | |||
| 45 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 46 | _mm_mask_popcnt_epi32(__m128i __W, __mmask8 __U, __m128i __A) { |
||
| 47 | return (__m128i)__builtin_ia32_selectd_128( |
||
| 48 | (__mmask8)__U, (__v4si)_mm_popcnt_epi32(__A), (__v4si)__W); |
||
| 49 | } |
||
| 50 | |||
| 51 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 52 | _mm_maskz_popcnt_epi32(__mmask8 __U, __m128i __A) { |
||
| 53 | return _mm_mask_popcnt_epi32((__m128i)_mm_setzero_si128(), __U, __A); |
||
| 54 | } |
||
| 55 | |||
| 56 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 57 | _mm256_popcnt_epi64(__m256i __A) { |
||
| 58 | return (__m256i)__builtin_ia32_vpopcntq_256((__v4di)__A); |
||
| 59 | } |
||
| 60 | |||
| 61 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 62 | _mm256_mask_popcnt_epi64(__m256i __W, __mmask8 __U, __m256i __A) { |
||
| 63 | return (__m256i)__builtin_ia32_selectq_256( |
||
| 64 | (__mmask8)__U, (__v4di)_mm256_popcnt_epi64(__A), (__v4di)__W); |
||
| 65 | } |
||
| 66 | |||
| 67 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 68 | _mm256_maskz_popcnt_epi64(__mmask8 __U, __m256i __A) { |
||
| 69 | return _mm256_mask_popcnt_epi64((__m256i)_mm256_setzero_si256(), __U, __A); |
||
| 70 | } |
||
| 71 | |||
| 72 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 73 | _mm256_popcnt_epi32(__m256i __A) { |
||
| 74 | return (__m256i)__builtin_ia32_vpopcntd_256((__v8si)__A); |
||
| 75 | } |
||
| 76 | |||
| 77 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 78 | _mm256_mask_popcnt_epi32(__m256i __W, __mmask8 __U, __m256i __A) { |
||
| 79 | return (__m256i)__builtin_ia32_selectd_256( |
||
| 80 | (__mmask8)__U, (__v8si)_mm256_popcnt_epi32(__A), (__v8si)__W); |
||
| 81 | } |
||
| 82 | |||
| 83 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 84 | _mm256_maskz_popcnt_epi32(__mmask8 __U, __m256i __A) { |
||
| 85 | return _mm256_mask_popcnt_epi32((__m256i)_mm256_setzero_si256(), __U, __A); |
||
| 86 | } |
||
| 87 | |||
| 88 | #undef __DEFAULT_FN_ATTRS128 |
||
| 89 | #undef __DEFAULT_FN_ATTRS256 |
||
| 90 | |||
| 91 | #endif |