Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 1 | /*===------------- avx512ifmavlintrin.h - IFMA intrinsics ------------------=== |
| 2 | * |
| 3 | * |
| 4 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 5 | * of this software and associated documentation files (the "Software"), to deal |
| 6 | * in the Software without restriction, including without limitation the rights |
| 7 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 8 | * copies of the Software, and to permit persons to whom the Software is |
| 9 | * furnished to do so, subject to the following conditions: |
| 10 | * |
| 11 | * The above copyright notice and this permission notice shall be included in |
| 12 | * all copies or substantial portions of the Software. |
| 13 | * |
| 14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 17 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 18 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 19 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 20 | * THE SOFTWARE. |
| 21 | * |
| 22 | *===-----------------------------------------------------------------------=== |
| 23 | */ |
| 24 | #ifndef __IMMINTRIN_H |
| 25 | #error "Never use <avx512ifmavlintrin.h> directly; include <immintrin.h> instead." |
| 26 | #endif |
| 27 | |
| 28 | #ifndef __IFMAVLINTRIN_H |
| 29 | #define __IFMAVLINTRIN_H |
| 30 | |
| 31 | /* Define the default attributes for the functions in this file. */ |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 32 | #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx512ifma,avx512vl"), __min_vector_width__(128))) |
| 33 | #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx512ifma,avx512vl"), __min_vector_width__(256))) |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 34 | |
| 35 | |
| 36 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 37 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 38 | _mm_madd52hi_epu64 (__m128i __X, __m128i __Y, __m128i __Z) |
| 39 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 40 | return (__m128i)__builtin_ia32_vpmadd52huq128((__v2di) __X, (__v2di) __Y, |
| 41 | (__v2di) __Z); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 42 | } |
| 43 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 44 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 45 | _mm_mask_madd52hi_epu64 (__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y) |
| 46 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 47 | return (__m128i)__builtin_ia32_selectq_128(__M, |
| 48 | (__v2di)_mm_madd52hi_epu64(__W, __X, __Y), |
| 49 | (__v2di)__W); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 50 | } |
| 51 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 52 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 53 | _mm_maskz_madd52hi_epu64 (__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z) |
| 54 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 55 | return (__m128i)__builtin_ia32_selectq_128(__M, |
| 56 | (__v2di)_mm_madd52hi_epu64(__X, __Y, __Z), |
| 57 | (__v2di)_mm_setzero_si128()); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 58 | } |
| 59 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 60 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 61 | _mm256_madd52hi_epu64 (__m256i __X, __m256i __Y, __m256i __Z) |
| 62 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 63 | return (__m256i)__builtin_ia32_vpmadd52huq256((__v4di)__X, (__v4di)__Y, |
| 64 | (__v4di)__Z); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 65 | } |
| 66 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 67 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 68 | _mm256_mask_madd52hi_epu64 (__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 69 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 70 | return (__m256i)__builtin_ia32_selectq_256(__M, |
| 71 | (__v4di)_mm256_madd52hi_epu64(__W, __X, __Y), |
| 72 | (__v4di)__W); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 73 | } |
| 74 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 75 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 76 | _mm256_maskz_madd52hi_epu64 (__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z) |
| 77 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 78 | return (__m256i)__builtin_ia32_selectq_256(__M, |
| 79 | (__v4di)_mm256_madd52hi_epu64(__X, __Y, __Z), |
| 80 | (__v4di)_mm256_setzero_si256()); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 81 | } |
| 82 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 83 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 84 | _mm_madd52lo_epu64 (__m128i __X, __m128i __Y, __m128i __Z) |
| 85 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 86 | return (__m128i)__builtin_ia32_vpmadd52luq128((__v2di)__X, (__v2di)__Y, |
| 87 | (__v2di)__Z); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 88 | } |
| 89 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 90 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 91 | _mm_mask_madd52lo_epu64 (__m128i __W, __mmask8 __M, __m128i __X, __m128i __Y) |
| 92 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 93 | return (__m128i)__builtin_ia32_selectq_128(__M, |
| 94 | (__v2di)_mm_madd52lo_epu64(__W, __X, __Y), |
| 95 | (__v2di)__W); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 96 | } |
| 97 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 98 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 99 | _mm_maskz_madd52lo_epu64 (__mmask8 __M, __m128i __X, __m128i __Y, __m128i __Z) |
| 100 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 101 | return (__m128i)__builtin_ia32_selectq_128(__M, |
| 102 | (__v2di)_mm_madd52lo_epu64(__X, __Y, __Z), |
| 103 | (__v2di)_mm_setzero_si128()); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 104 | } |
| 105 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 106 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 107 | _mm256_madd52lo_epu64 (__m256i __X, __m256i __Y, __m256i __Z) |
| 108 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 109 | return (__m256i)__builtin_ia32_vpmadd52luq256((__v4di)__X, (__v4di)__Y, |
| 110 | (__v4di)__Z); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 111 | } |
| 112 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 113 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 114 | _mm256_mask_madd52lo_epu64 (__m256i __W, __mmask8 __M, __m256i __X, __m256i __Y) |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 115 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 116 | return (__m256i)__builtin_ia32_selectq_256(__M, |
| 117 | (__v4di)_mm256_madd52lo_epu64(__W, __X, __Y), |
| 118 | (__v4di)__W); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 119 | } |
| 120 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 121 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 122 | _mm256_maskz_madd52lo_epu64 (__mmask8 __M, __m256i __X, __m256i __Y, __m256i __Z) |
| 123 | { |
Craig Topper | 387b142 | 2018-05-26 18:55:26 +0000 | [diff] [blame] | 124 | return (__m256i)__builtin_ia32_selectq_256(__M, |
| 125 | (__v4di)_mm256_madd52lo_epu64(__X, __Y, __Z), |
| 126 | (__v4di)_mm256_setzero_si256()); |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 127 | } |
| 128 | |
| 129 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 130 | #undef __DEFAULT_FN_ATTRS128 |
| 131 | #undef __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 0190c65 | 2016-03-07 09:55:55 +0000 | [diff] [blame] | 132 | |
| 133 | #endif |