annotate clang/lib/Headers/gfniintrin.h @ 176:de4ac79aef9d

...
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Mon, 25 May 2020 17:13:11 +0900
parents 1d019706d866
children 2e18cbf3894f
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
150
anatofuz
parents:
diff changeset
1 /*===----------------- gfniintrin.h - GFNI intrinsics ----------------------===
anatofuz
parents:
diff changeset
2 *
anatofuz
parents:
diff changeset
3 *
anatofuz
parents:
diff changeset
4 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
anatofuz
parents:
diff changeset
5 * See https://llvm.org/LICENSE.txt for license information.
anatofuz
parents:
diff changeset
6 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
anatofuz
parents:
diff changeset
7 *
anatofuz
parents:
diff changeset
8 *===-----------------------------------------------------------------------===
anatofuz
parents:
diff changeset
9 */
anatofuz
parents:
diff changeset
10 #ifndef __IMMINTRIN_H
anatofuz
parents:
diff changeset
11 #error "Never use <gfniintrin.h> directly; include <immintrin.h> instead."
anatofuz
parents:
diff changeset
12 #endif
anatofuz
parents:
diff changeset
13
anatofuz
parents:
diff changeset
14 #ifndef __GFNIINTRIN_H
anatofuz
parents:
diff changeset
15 #define __GFNIINTRIN_H
anatofuz
parents:
diff changeset
16
anatofuz
parents:
diff changeset
17
anatofuz
parents:
diff changeset
18 #define _mm_gf2p8affineinv_epi64_epi8(A, B, I) \
anatofuz
parents:
diff changeset
19 (__m128i)__builtin_ia32_vgf2p8affineinvqb_v16qi((__v16qi)(__m128i)(A), \
anatofuz
parents:
diff changeset
20 (__v16qi)(__m128i)(B), \
anatofuz
parents:
diff changeset
21 (char)(I))
anatofuz
parents:
diff changeset
22
anatofuz
parents:
diff changeset
23 #define _mm_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, I) \
anatofuz
parents:
diff changeset
24 (__m128i)__builtin_ia32_selectb_128((__mmask16)(U), \
anatofuz
parents:
diff changeset
25 (__v16qi)_mm_gf2p8affineinv_epi64_epi8(A, B, I), \
anatofuz
parents:
diff changeset
26 (__v16qi)(__m128i)(S))
anatofuz
parents:
diff changeset
27
anatofuz
parents:
diff changeset
28
anatofuz
parents:
diff changeset
29 #define _mm_maskz_gf2p8affineinv_epi64_epi8(U, A, B, I) \
anatofuz
parents:
diff changeset
30 (__m128i)_mm_mask_gf2p8affineinv_epi64_epi8((__m128i)_mm_setzero_si128(), \
anatofuz
parents:
diff changeset
31 U, A, B, I)
anatofuz
parents:
diff changeset
32
anatofuz
parents:
diff changeset
33
anatofuz
parents:
diff changeset
34 #define _mm256_gf2p8affineinv_epi64_epi8(A, B, I) \
anatofuz
parents:
diff changeset
35 (__m256i)__builtin_ia32_vgf2p8affineinvqb_v32qi((__v32qi)(__m256i)(A), \
anatofuz
parents:
diff changeset
36 (__v32qi)(__m256i)(B), \
anatofuz
parents:
diff changeset
37 (char)(I))
anatofuz
parents:
diff changeset
38
anatofuz
parents:
diff changeset
39 #define _mm256_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, I) \
anatofuz
parents:
diff changeset
40 (__m256i)__builtin_ia32_selectb_256((__mmask32)(U), \
anatofuz
parents:
diff changeset
41 (__v32qi)_mm256_gf2p8affineinv_epi64_epi8(A, B, I), \
anatofuz
parents:
diff changeset
42 (__v32qi)(__m256i)(S))
anatofuz
parents:
diff changeset
43
anatofuz
parents:
diff changeset
44 #define _mm256_maskz_gf2p8affineinv_epi64_epi8(U, A, B, I) \
anatofuz
parents:
diff changeset
45 (__m256i)_mm256_mask_gf2p8affineinv_epi64_epi8((__m256i)_mm256_setzero_si256(), \
anatofuz
parents:
diff changeset
46 U, A, B, I)
anatofuz
parents:
diff changeset
47
anatofuz
parents:
diff changeset
48
anatofuz
parents:
diff changeset
49 #define _mm512_gf2p8affineinv_epi64_epi8(A, B, I) \
anatofuz
parents:
diff changeset
50 (__m512i)__builtin_ia32_vgf2p8affineinvqb_v64qi((__v64qi)(__m512i)(A), \
anatofuz
parents:
diff changeset
51 (__v64qi)(__m512i)(B), \
anatofuz
parents:
diff changeset
52 (char)(I))
anatofuz
parents:
diff changeset
53
anatofuz
parents:
diff changeset
54 #define _mm512_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, I) \
anatofuz
parents:
diff changeset
55 (__m512i)__builtin_ia32_selectb_512((__mmask64)(U), \
anatofuz
parents:
diff changeset
56 (__v64qi)_mm512_gf2p8affineinv_epi64_epi8(A, B, I), \
anatofuz
parents:
diff changeset
57 (__v64qi)(__m512i)(S))
anatofuz
parents:
diff changeset
58
anatofuz
parents:
diff changeset
59 #define _mm512_maskz_gf2p8affineinv_epi64_epi8(U, A, B, I) \
anatofuz
parents:
diff changeset
60 (__m512i)_mm512_mask_gf2p8affineinv_epi64_epi8((__m512i)_mm512_setzero_si512(), \
anatofuz
parents:
diff changeset
61 U, A, B, I)
anatofuz
parents:
diff changeset
62
anatofuz
parents:
diff changeset
63 #define _mm_gf2p8affine_epi64_epi8(A, B, I) \
anatofuz
parents:
diff changeset
64 (__m128i)__builtin_ia32_vgf2p8affineqb_v16qi((__v16qi)(__m128i)(A), \
anatofuz
parents:
diff changeset
65 (__v16qi)(__m128i)(B), \
anatofuz
parents:
diff changeset
66 (char)(I))
anatofuz
parents:
diff changeset
67
anatofuz
parents:
diff changeset
68 #define _mm_mask_gf2p8affine_epi64_epi8(S, U, A, B, I) \
anatofuz
parents:
diff changeset
69 (__m128i)__builtin_ia32_selectb_128((__mmask16)(U), \
anatofuz
parents:
diff changeset
70 (__v16qi)_mm_gf2p8affine_epi64_epi8(A, B, I), \
anatofuz
parents:
diff changeset
71 (__v16qi)(__m128i)(S))
anatofuz
parents:
diff changeset
72
anatofuz
parents:
diff changeset
73
anatofuz
parents:
diff changeset
74 #define _mm_maskz_gf2p8affine_epi64_epi8(U, A, B, I) \
anatofuz
parents:
diff changeset
75 (__m128i)_mm_mask_gf2p8affine_epi64_epi8((__m128i)_mm_setzero_si128(), \
anatofuz
parents:
diff changeset
76 U, A, B, I)
anatofuz
parents:
diff changeset
77
anatofuz
parents:
diff changeset
78
anatofuz
parents:
diff changeset
79 #define _mm256_gf2p8affine_epi64_epi8(A, B, I) \
anatofuz
parents:
diff changeset
80 (__m256i)__builtin_ia32_vgf2p8affineqb_v32qi((__v32qi)(__m256i)(A), \
anatofuz
parents:
diff changeset
81 (__v32qi)(__m256i)(B), \
anatofuz
parents:
diff changeset
82 (char)(I))
anatofuz
parents:
diff changeset
83
anatofuz
parents:
diff changeset
84 #define _mm256_mask_gf2p8affine_epi64_epi8(S, U, A, B, I) \
anatofuz
parents:
diff changeset
85 (__m256i)__builtin_ia32_selectb_256((__mmask32)(U), \
anatofuz
parents:
diff changeset
86 (__v32qi)_mm256_gf2p8affine_epi64_epi8(A, B, I), \
anatofuz
parents:
diff changeset
87 (__v32qi)(__m256i)(S))
anatofuz
parents:
diff changeset
88
anatofuz
parents:
diff changeset
89 #define _mm256_maskz_gf2p8affine_epi64_epi8(U, A, B, I) \
anatofuz
parents:
diff changeset
90 (__m256i)_mm256_mask_gf2p8affine_epi64_epi8((__m256i)_mm256_setzero_si256(), \
anatofuz
parents:
diff changeset
91 U, A, B, I)
anatofuz
parents:
diff changeset
92
anatofuz
parents:
diff changeset
93
anatofuz
parents:
diff changeset
94 #define _mm512_gf2p8affine_epi64_epi8(A, B, I) \
anatofuz
parents:
diff changeset
95 (__m512i)__builtin_ia32_vgf2p8affineqb_v64qi((__v64qi)(__m512i)(A), \
anatofuz
parents:
diff changeset
96 (__v64qi)(__m512i)(B), \
anatofuz
parents:
diff changeset
97 (char)(I))
anatofuz
parents:
diff changeset
98
anatofuz
parents:
diff changeset
99 #define _mm512_mask_gf2p8affine_epi64_epi8(S, U, A, B, I) \
anatofuz
parents:
diff changeset
100 (__m512i)__builtin_ia32_selectb_512((__mmask64)(U), \
anatofuz
parents:
diff changeset
101 (__v64qi)_mm512_gf2p8affine_epi64_epi8(A, B, I), \
anatofuz
parents:
diff changeset
102 (__v64qi)(__m512i)(S))
anatofuz
parents:
diff changeset
103
anatofuz
parents:
diff changeset
104 #define _mm512_maskz_gf2p8affine_epi64_epi8(U, A, B, I) \
anatofuz
parents:
diff changeset
105 (__m512i)_mm512_mask_gf2p8affine_epi64_epi8((__m512i)_mm512_setzero_si512(), \
anatofuz
parents:
diff changeset
106 U, A, B, I)
anatofuz
parents:
diff changeset
107
anatofuz
parents:
diff changeset
108 /* Default attributes for simple form (no masking). */
anatofuz
parents:
diff changeset
109 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("gfni"), __min_vector_width__(128)))
anatofuz
parents:
diff changeset
110
anatofuz
parents:
diff changeset
111 /* Default attributes for YMM unmasked form. */
anatofuz
parents:
diff changeset
112 #define __DEFAULT_FN_ATTRS_Y __attribute__((__always_inline__, __nodebug__, __target__("avx,gfni"), __min_vector_width__(256)))
anatofuz
parents:
diff changeset
113
anatofuz
parents:
diff changeset
114 /* Default attributes for ZMM forms. */
anatofuz
parents:
diff changeset
115 #define __DEFAULT_FN_ATTRS_Z __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,gfni"), __min_vector_width__(512)))
anatofuz
parents:
diff changeset
116
anatofuz
parents:
diff changeset
117 /* Default attributes for VLX forms. */
anatofuz
parents:
diff changeset
118 #define __DEFAULT_FN_ATTRS_VL128 __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,avx512vl,gfni"), __min_vector_width__(128)))
anatofuz
parents:
diff changeset
119 #define __DEFAULT_FN_ATTRS_VL256 __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,avx512vl,gfni"), __min_vector_width__(256)))
anatofuz
parents:
diff changeset
120
anatofuz
parents:
diff changeset
121 static __inline__ __m128i __DEFAULT_FN_ATTRS
anatofuz
parents:
diff changeset
122 _mm_gf2p8mul_epi8(__m128i __A, __m128i __B)
anatofuz
parents:
diff changeset
123 {
anatofuz
parents:
diff changeset
124 return (__m128i) __builtin_ia32_vgf2p8mulb_v16qi((__v16qi) __A,
anatofuz
parents:
diff changeset
125 (__v16qi) __B);
anatofuz
parents:
diff changeset
126 }
anatofuz
parents:
diff changeset
127
anatofuz
parents:
diff changeset
128 static __inline__ __m128i __DEFAULT_FN_ATTRS_VL128
anatofuz
parents:
diff changeset
129 _mm_mask_gf2p8mul_epi8(__m128i __S, __mmask16 __U, __m128i __A, __m128i __B)
anatofuz
parents:
diff changeset
130 {
anatofuz
parents:
diff changeset
131 return (__m128i) __builtin_ia32_selectb_128(__U,
anatofuz
parents:
diff changeset
132 (__v16qi) _mm_gf2p8mul_epi8(__A, __B),
anatofuz
parents:
diff changeset
133 (__v16qi) __S);
anatofuz
parents:
diff changeset
134 }
anatofuz
parents:
diff changeset
135
anatofuz
parents:
diff changeset
136 static __inline__ __m128i __DEFAULT_FN_ATTRS_VL128
anatofuz
parents:
diff changeset
137 _mm_maskz_gf2p8mul_epi8(__mmask16 __U, __m128i __A, __m128i __B)
anatofuz
parents:
diff changeset
138 {
anatofuz
parents:
diff changeset
139 return _mm_mask_gf2p8mul_epi8((__m128i)_mm_setzero_si128(),
anatofuz
parents:
diff changeset
140 __U, __A, __B);
anatofuz
parents:
diff changeset
141 }
anatofuz
parents:
diff changeset
142
anatofuz
parents:
diff changeset
143 static __inline__ __m256i __DEFAULT_FN_ATTRS_Y
anatofuz
parents:
diff changeset
144 _mm256_gf2p8mul_epi8(__m256i __A, __m256i __B)
anatofuz
parents:
diff changeset
145 {
anatofuz
parents:
diff changeset
146 return (__m256i) __builtin_ia32_vgf2p8mulb_v32qi((__v32qi) __A,
anatofuz
parents:
diff changeset
147 (__v32qi) __B);
anatofuz
parents:
diff changeset
148 }
anatofuz
parents:
diff changeset
149
anatofuz
parents:
diff changeset
150 static __inline__ __m256i __DEFAULT_FN_ATTRS_VL256
anatofuz
parents:
diff changeset
151 _mm256_mask_gf2p8mul_epi8(__m256i __S, __mmask32 __U, __m256i __A, __m256i __B)
anatofuz
parents:
diff changeset
152 {
anatofuz
parents:
diff changeset
153 return (__m256i) __builtin_ia32_selectb_256(__U,
anatofuz
parents:
diff changeset
154 (__v32qi) _mm256_gf2p8mul_epi8(__A, __B),
anatofuz
parents:
diff changeset
155 (__v32qi) __S);
anatofuz
parents:
diff changeset
156 }
anatofuz
parents:
diff changeset
157
anatofuz
parents:
diff changeset
158 static __inline__ __m256i __DEFAULT_FN_ATTRS_VL256
anatofuz
parents:
diff changeset
159 _mm256_maskz_gf2p8mul_epi8(__mmask32 __U, __m256i __A, __m256i __B)
anatofuz
parents:
diff changeset
160 {
anatofuz
parents:
diff changeset
161 return _mm256_mask_gf2p8mul_epi8((__m256i)_mm256_setzero_si256(),
anatofuz
parents:
diff changeset
162 __U, __A, __B);
anatofuz
parents:
diff changeset
163 }
anatofuz
parents:
diff changeset
164
anatofuz
parents:
diff changeset
165 static __inline__ __m512i __DEFAULT_FN_ATTRS_Z
anatofuz
parents:
diff changeset
166 _mm512_gf2p8mul_epi8(__m512i __A, __m512i __B)
anatofuz
parents:
diff changeset
167 {
anatofuz
parents:
diff changeset
168 return (__m512i) __builtin_ia32_vgf2p8mulb_v64qi((__v64qi) __A,
anatofuz
parents:
diff changeset
169 (__v64qi) __B);
anatofuz
parents:
diff changeset
170 }
anatofuz
parents:
diff changeset
171
anatofuz
parents:
diff changeset
172 static __inline__ __m512i __DEFAULT_FN_ATTRS_Z
anatofuz
parents:
diff changeset
173 _mm512_mask_gf2p8mul_epi8(__m512i __S, __mmask64 __U, __m512i __A, __m512i __B)
anatofuz
parents:
diff changeset
174 {
anatofuz
parents:
diff changeset
175 return (__m512i) __builtin_ia32_selectb_512(__U,
anatofuz
parents:
diff changeset
176 (__v64qi) _mm512_gf2p8mul_epi8(__A, __B),
anatofuz
parents:
diff changeset
177 (__v64qi) __S);
anatofuz
parents:
diff changeset
178 }
anatofuz
parents:
diff changeset
179
anatofuz
parents:
diff changeset
180 static __inline__ __m512i __DEFAULT_FN_ATTRS_Z
anatofuz
parents:
diff changeset
181 _mm512_maskz_gf2p8mul_epi8(__mmask64 __U, __m512i __A, __m512i __B)
anatofuz
parents:
diff changeset
182 {
anatofuz
parents:
diff changeset
183 return _mm512_mask_gf2p8mul_epi8((__m512i)_mm512_setzero_si512(),
anatofuz
parents:
diff changeset
184 __U, __A, __B);
anatofuz
parents:
diff changeset
185 }
anatofuz
parents:
diff changeset
186
anatofuz
parents:
diff changeset
187 #undef __DEFAULT_FN_ATTRS
anatofuz
parents:
diff changeset
188 #undef __DEFAULT_FN_ATTRS_Y
anatofuz
parents:
diff changeset
189 #undef __DEFAULT_FN_ATTRS_Z
anatofuz
parents:
diff changeset
190 #undef __DEFAULT_FN_ATTRS_VL128
anatofuz
parents:
diff changeset
191 #undef __DEFAULT_FN_ATTRS_VL256
anatofuz
parents:
diff changeset
192
anatofuz
parents:
diff changeset
193 #endif /* __GFNIINTRIN_H */
anatofuz
parents:
diff changeset
194