Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 1 | /*===---- mm3dnow.h - 3DNow! intrinsics ------------------------------------=== |
| 2 | * |
Logan Chien | df4f766 | 2019-09-04 16:45:23 -0700 | [diff] [blame] | 3 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | * See https://llvm.org/LICENSE.txt for license information. |
| 5 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 6 | * |
| 7 | *===-----------------------------------------------------------------------=== |
| 8 | */ |
| 9 | |
| 10 | #ifndef _MM3DNOW_H_INCLUDED |
| 11 | #define _MM3DNOW_H_INCLUDED |
| 12 | |
| 13 | #include <mmintrin.h> |
| 14 | #include <prfchwintrin.h> |
| 15 | |
| 16 | typedef float __v2sf __attribute__((__vector_size__(8))); |
| 17 | |
| 18 | /* Define the default attributes for the functions in this file. */ |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 19 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnow"), __min_vector_width__(64))) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 20 | |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 21 | static __inline__ void __attribute__((__always_inline__, __nodebug__, __target__("3dnow"))) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 22 | _m_femms(void) { |
| 23 | __builtin_ia32_femms(); |
| 24 | } |
| 25 | |
| 26 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 27 | _m_pavgusb(__m64 __m1, __m64 __m2) { |
| 28 | return (__m64)__builtin_ia32_pavgusb((__v8qi)__m1, (__v8qi)__m2); |
| 29 | } |
| 30 | |
| 31 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 32 | _m_pf2id(__m64 __m) { |
| 33 | return (__m64)__builtin_ia32_pf2id((__v2sf)__m); |
| 34 | } |
| 35 | |
| 36 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 37 | _m_pfacc(__m64 __m1, __m64 __m2) { |
| 38 | return (__m64)__builtin_ia32_pfacc((__v2sf)__m1, (__v2sf)__m2); |
| 39 | } |
| 40 | |
| 41 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 42 | _m_pfadd(__m64 __m1, __m64 __m2) { |
| 43 | return (__m64)__builtin_ia32_pfadd((__v2sf)__m1, (__v2sf)__m2); |
| 44 | } |
| 45 | |
| 46 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 47 | _m_pfcmpeq(__m64 __m1, __m64 __m2) { |
| 48 | return (__m64)__builtin_ia32_pfcmpeq((__v2sf)__m1, (__v2sf)__m2); |
| 49 | } |
| 50 | |
| 51 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 52 | _m_pfcmpge(__m64 __m1, __m64 __m2) { |
| 53 | return (__m64)__builtin_ia32_pfcmpge((__v2sf)__m1, (__v2sf)__m2); |
| 54 | } |
| 55 | |
| 56 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 57 | _m_pfcmpgt(__m64 __m1, __m64 __m2) { |
| 58 | return (__m64)__builtin_ia32_pfcmpgt((__v2sf)__m1, (__v2sf)__m2); |
| 59 | } |
| 60 | |
| 61 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 62 | _m_pfmax(__m64 __m1, __m64 __m2) { |
| 63 | return (__m64)__builtin_ia32_pfmax((__v2sf)__m1, (__v2sf)__m2); |
| 64 | } |
| 65 | |
| 66 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 67 | _m_pfmin(__m64 __m1, __m64 __m2) { |
| 68 | return (__m64)__builtin_ia32_pfmin((__v2sf)__m1, (__v2sf)__m2); |
| 69 | } |
| 70 | |
| 71 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 72 | _m_pfmul(__m64 __m1, __m64 __m2) { |
| 73 | return (__m64)__builtin_ia32_pfmul((__v2sf)__m1, (__v2sf)__m2); |
| 74 | } |
| 75 | |
| 76 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 77 | _m_pfrcp(__m64 __m) { |
| 78 | return (__m64)__builtin_ia32_pfrcp((__v2sf)__m); |
| 79 | } |
| 80 | |
| 81 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 82 | _m_pfrcpit1(__m64 __m1, __m64 __m2) { |
| 83 | return (__m64)__builtin_ia32_pfrcpit1((__v2sf)__m1, (__v2sf)__m2); |
| 84 | } |
| 85 | |
| 86 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 87 | _m_pfrcpit2(__m64 __m1, __m64 __m2) { |
| 88 | return (__m64)__builtin_ia32_pfrcpit2((__v2sf)__m1, (__v2sf)__m2); |
| 89 | } |
| 90 | |
| 91 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 92 | _m_pfrsqrt(__m64 __m) { |
| 93 | return (__m64)__builtin_ia32_pfrsqrt((__v2sf)__m); |
| 94 | } |
| 95 | |
| 96 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 97 | _m_pfrsqrtit1(__m64 __m1, __m64 __m2) { |
| 98 | return (__m64)__builtin_ia32_pfrsqit1((__v2sf)__m1, (__v2sf)__m2); |
| 99 | } |
| 100 | |
| 101 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 102 | _m_pfsub(__m64 __m1, __m64 __m2) { |
| 103 | return (__m64)__builtin_ia32_pfsub((__v2sf)__m1, (__v2sf)__m2); |
| 104 | } |
| 105 | |
| 106 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 107 | _m_pfsubr(__m64 __m1, __m64 __m2) { |
| 108 | return (__m64)__builtin_ia32_pfsubr((__v2sf)__m1, (__v2sf)__m2); |
| 109 | } |
| 110 | |
| 111 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 112 | _m_pi2fd(__m64 __m) { |
| 113 | return (__m64)__builtin_ia32_pi2fd((__v2si)__m); |
| 114 | } |
| 115 | |
| 116 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 117 | _m_pmulhrw(__m64 __m1, __m64 __m2) { |
| 118 | return (__m64)__builtin_ia32_pmulhrw((__v4hi)__m1, (__v4hi)__m2); |
| 119 | } |
| 120 | |
| 121 | /* Handle the 3dnowa instructions here. */ |
| 122 | #undef __DEFAULT_FN_ATTRS |
Logan Chien | 55afb0a | 2018-10-15 10:42:14 +0800 | [diff] [blame] | 123 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnowa"), __min_vector_width__(64))) |
Logan Chien | 2833ffb | 2018-10-09 10:03:24 +0800 | [diff] [blame] | 124 | |
| 125 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 126 | _m_pf2iw(__m64 __m) { |
| 127 | return (__m64)__builtin_ia32_pf2iw((__v2sf)__m); |
| 128 | } |
| 129 | |
| 130 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 131 | _m_pfnacc(__m64 __m1, __m64 __m2) { |
| 132 | return (__m64)__builtin_ia32_pfnacc((__v2sf)__m1, (__v2sf)__m2); |
| 133 | } |
| 134 | |
| 135 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 136 | _m_pfpnacc(__m64 __m1, __m64 __m2) { |
| 137 | return (__m64)__builtin_ia32_pfpnacc((__v2sf)__m1, (__v2sf)__m2); |
| 138 | } |
| 139 | |
| 140 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 141 | _m_pi2fw(__m64 __m) { |
| 142 | return (__m64)__builtin_ia32_pi2fw((__v2si)__m); |
| 143 | } |
| 144 | |
| 145 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 146 | _m_pswapdsf(__m64 __m) { |
| 147 | return (__m64)__builtin_ia32_pswapdsf((__v2sf)__m); |
| 148 | } |
| 149 | |
| 150 | static __inline__ __m64 __DEFAULT_FN_ATTRS |
| 151 | _m_pswapdsi(__m64 __m) { |
| 152 | return (__m64)__builtin_ia32_pswapdsi((__v2si)__m); |
| 153 | } |
| 154 | |
| 155 | #undef __DEFAULT_FN_ATTRS |
| 156 | |
| 157 | #endif |