Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 14 | pmbaty | 1 | /*===---- avx512vlbitalgintrin.h - BITALG 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 "Never use <avx512vlbitalgintrin.h> directly; include <immintrin.h> instead." |
||
| 12 | #endif |
||
| 13 | |||
| 14 | #ifndef __AVX512VLBITALGINTRIN_H |
||
| 15 | #define __AVX512VLBITALGINTRIN_H |
||
| 16 | |||
| 17 | /* Define the default attributes for the functions in this file. */ |
||
| 18 | #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512bitalg"), __min_vector_width__(128))) |
||
| 19 | #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512bitalg"), __min_vector_width__(256))) |
||
| 20 | |||
| 21 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 22 | _mm256_popcnt_epi16(__m256i __A) |
||
| 23 | { |
||
| 24 | return (__m256i) __builtin_ia32_vpopcntw_256((__v16hi) __A); |
||
| 25 | } |
||
| 26 | |||
| 27 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 28 | _mm256_mask_popcnt_epi16(__m256i __A, __mmask16 __U, __m256i __B) |
||
| 29 | { |
||
| 30 | return (__m256i) __builtin_ia32_selectw_256((__mmask16) __U, |
||
| 31 | (__v16hi) _mm256_popcnt_epi16(__B), |
||
| 32 | (__v16hi) __A); |
||
| 33 | } |
||
| 34 | |||
| 35 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 36 | _mm256_maskz_popcnt_epi16(__mmask16 __U, __m256i __B) |
||
| 37 | { |
||
| 38 | return _mm256_mask_popcnt_epi16((__m256i) _mm256_setzero_si256(), |
||
| 39 | __U, |
||
| 40 | __B); |
||
| 41 | } |
||
| 42 | |||
| 43 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 44 | _mm_popcnt_epi16(__m128i __A) |
||
| 45 | { |
||
| 46 | return (__m128i) __builtin_ia32_vpopcntw_128((__v8hi) __A); |
||
| 47 | } |
||
| 48 | |||
| 49 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 50 | _mm_mask_popcnt_epi16(__m128i __A, __mmask8 __U, __m128i __B) |
||
| 51 | { |
||
| 52 | return (__m128i) __builtin_ia32_selectw_128((__mmask8) __U, |
||
| 53 | (__v8hi) _mm_popcnt_epi16(__B), |
||
| 54 | (__v8hi) __A); |
||
| 55 | } |
||
| 56 | |||
| 57 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 58 | _mm_maskz_popcnt_epi16(__mmask8 __U, __m128i __B) |
||
| 59 | { |
||
| 60 | return _mm_mask_popcnt_epi16((__m128i) _mm_setzero_si128(), |
||
| 61 | __U, |
||
| 62 | __B); |
||
| 63 | } |
||
| 64 | |||
| 65 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 66 | _mm256_popcnt_epi8(__m256i __A) |
||
| 67 | { |
||
| 68 | return (__m256i) __builtin_ia32_vpopcntb_256((__v32qi) __A); |
||
| 69 | } |
||
| 70 | |||
| 71 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 72 | _mm256_mask_popcnt_epi8(__m256i __A, __mmask32 __U, __m256i __B) |
||
| 73 | { |
||
| 74 | return (__m256i) __builtin_ia32_selectb_256((__mmask32) __U, |
||
| 75 | (__v32qi) _mm256_popcnt_epi8(__B), |
||
| 76 | (__v32qi) __A); |
||
| 77 | } |
||
| 78 | |||
| 79 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
||
| 80 | _mm256_maskz_popcnt_epi8(__mmask32 __U, __m256i __B) |
||
| 81 | { |
||
| 82 | return _mm256_mask_popcnt_epi8((__m256i) _mm256_setzero_si256(), |
||
| 83 | __U, |
||
| 84 | __B); |
||
| 85 | } |
||
| 86 | |||
| 87 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 88 | _mm_popcnt_epi8(__m128i __A) |
||
| 89 | { |
||
| 90 | return (__m128i) __builtin_ia32_vpopcntb_128((__v16qi) __A); |
||
| 91 | } |
||
| 92 | |||
| 93 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 94 | _mm_mask_popcnt_epi8(__m128i __A, __mmask16 __U, __m128i __B) |
||
| 95 | { |
||
| 96 | return (__m128i) __builtin_ia32_selectb_128((__mmask16) __U, |
||
| 97 | (__v16qi) _mm_popcnt_epi8(__B), |
||
| 98 | (__v16qi) __A); |
||
| 99 | } |
||
| 100 | |||
| 101 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
||
| 102 | _mm_maskz_popcnt_epi8(__mmask16 __U, __m128i __B) |
||
| 103 | { |
||
| 104 | return _mm_mask_popcnt_epi8((__m128i) _mm_setzero_si128(), |
||
| 105 | __U, |
||
| 106 | __B); |
||
| 107 | } |
||
| 108 | |||
| 109 | static __inline__ __mmask32 __DEFAULT_FN_ATTRS256 |
||
| 110 | _mm256_mask_bitshuffle_epi64_mask(__mmask32 __U, __m256i __A, __m256i __B) |
||
| 111 | { |
||
| 112 | return (__mmask32) __builtin_ia32_vpshufbitqmb256_mask((__v32qi) __A, |
||
| 113 | (__v32qi) __B, |
||
| 114 | __U); |
||
| 115 | } |
||
| 116 | |||
| 117 | static __inline__ __mmask32 __DEFAULT_FN_ATTRS256 |
||
| 118 | _mm256_bitshuffle_epi64_mask(__m256i __A, __m256i __B) |
||
| 119 | { |
||
| 120 | return _mm256_mask_bitshuffle_epi64_mask((__mmask32) -1, |
||
| 121 | __A, |
||
| 122 | __B); |
||
| 123 | } |
||
| 124 | |||
| 125 | static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 |
||
| 126 | _mm_mask_bitshuffle_epi64_mask(__mmask16 __U, __m128i __A, __m128i __B) |
||
| 127 | { |
||
| 128 | return (__mmask16) __builtin_ia32_vpshufbitqmb128_mask((__v16qi) __A, |
||
| 129 | (__v16qi) __B, |
||
| 130 | __U); |
||
| 131 | } |
||
| 132 | |||
| 133 | static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 |
||
| 134 | _mm_bitshuffle_epi64_mask(__m128i __A, __m128i __B) |
||
| 135 | { |
||
| 136 | return _mm_mask_bitshuffle_epi64_mask((__mmask16) -1, |
||
| 137 | __A, |
||
| 138 | __B); |
||
| 139 | } |
||
| 140 | |||
| 141 | |||
| 142 | #undef __DEFAULT_FN_ATTRS128 |
||
| 143 | #undef __DEFAULT_FN_ATTRS256 |
||
| 144 | |||
| 145 | #endif |