Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 1 | /*===------------- avx512ifmaintrin.h - IFMA intrinsics ------------------=== |
| 2 | * |
| 3 | * |
Logan Chien | df4f766 | 2019-09-04 16:45:23 -0700 | [diff] [blame] | 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 |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 7 | * |
| 8 | *===-----------------------------------------------------------------------=== |
| 9 | */ |
| 10 | #ifndef __IMMINTRIN_H |
| 11 | #error "Never use <avx512ifmaintrin.h> directly; include <immintrin.h> instead." |
| 12 | #endif |
| 13 | |
| 14 | #ifndef __IFMAINTRIN_H |
| 15 | #define __IFMAINTRIN_H |
| 16 | |
| 17 | /* Define the default attributes for the functions in this file. */ |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 18 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512ifma"), __min_vector_width__(512))) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 19 | |
| 20 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 21 | _mm512_madd52hi_epu64 (__m512i __X, __m512i __Y, __m512i __Z) |
| 22 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 23 | return (__m512i)__builtin_ia32_vpmadd52huq512((__v8di) __X, (__v8di) __Y, |
| 24 | (__v8di) __Z); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 25 | } |
| 26 | |
| 27 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 28 | _mm512_mask_madd52hi_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 29 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 30 | return (__m512i)__builtin_ia32_selectq_512(__M, |
| 31 | (__v8di)_mm512_madd52hi_epu64(__W, __X, __Y), |
| 32 | (__v8di)__W); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 33 | } |
| 34 | |
| 35 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 36 | _mm512_maskz_madd52hi_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z) |
| 37 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 38 | return (__m512i)__builtin_ia32_selectq_512(__M, |
| 39 | (__v8di)_mm512_madd52hi_epu64(__X, __Y, __Z), |
| 40 | (__v8di)_mm512_setzero_si512()); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 41 | } |
| 42 | |
| 43 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 44 | _mm512_madd52lo_epu64 (__m512i __X, __m512i __Y, __m512i __Z) |
| 45 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 46 | return (__m512i)__builtin_ia32_vpmadd52luq512((__v8di) __X, (__v8di) __Y, |
| 47 | (__v8di) __Z); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 48 | } |
| 49 | |
| 50 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 51 | _mm512_mask_madd52lo_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 52 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 53 | return (__m512i)__builtin_ia32_selectq_512(__M, |
| 54 | (__v8di)_mm512_madd52lo_epu64(__W, __X, __Y), |
| 55 | (__v8di)__W); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 56 | } |
| 57 | |
| 58 | static __inline__ __m512i __DEFAULT_FN_ATTRS |
| 59 | _mm512_maskz_madd52lo_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z) |
| 60 | { |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 61 | return (__m512i)__builtin_ia32_selectq_512(__M, |
| 62 | (__v8di)_mm512_madd52lo_epu64(__X, __Y, __Z), |
| 63 | (__v8di)_mm512_setzero_si512()); |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 64 | } |
| 65 | |
| 66 | #undef __DEFAULT_FN_ATTRS |
| 67 | |
| 68 | #endif |