blob: 34c555e3339a20e79d30713a173f3d425a453169 [file] [log] [blame]
Coby Tayreef4811eb2017-12-27 08:37:47 +00001/*===----------------- gfniintrin.h - GFNI 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 <gfniintrin.h> directly; include <immintrin.h> instead."
26#endif
27
28#ifndef __GFNIINTRIN_H
29#define __GFNIINTRIN_H
30
31
32#define _mm_gf2p8affineinv_epi64_epi8(A, B, I) __extension__ ({ \
33 (__m128i)__builtin_ia32_vgf2p8affineinvqb_v16qi((__v16qi)(__m128i)(A), \
34 (__v16qi)(__m128i)(B), \
35 (char)(I)); })
36
37#define _mm_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, I) __extension__ ({ \
38 (__m128i)__builtin_ia32_selectb_128((__mmask16)(U), \
39 (__v16qi)_mm_gf2p8affineinv_epi64_epi8(A, B, I), \
40 (__v16qi)(__m128i)(S)); })
41
42
43#define _mm_maskz_gf2p8affineinv_epi64_epi8(U, A, B, I) __extension__ ({ \
44 (__m128i)_mm_mask_gf2p8affineinv_epi64_epi8((__m128i)_mm_setzero_si128(), \
45 U, A, B, I); })
46
47
48#define _mm256_gf2p8affineinv_epi64_epi8(A, B, I) __extension__ ({ \
49 (__m256i)__builtin_ia32_vgf2p8affineinvqb_v32qi((__v32qi)(__m256i)(A), \
50 (__v32qi)(__m256i)(B), \
51 (char)(I)); })
52
53#define _mm256_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, I) __extension__ ({ \
54 (__m256i)__builtin_ia32_selectb_256((__mmask32)(U), \
55 (__v32qi)_mm256_gf2p8affineinv_epi64_epi8(A, B, I), \
56 (__v32qi)(__m256i)(S)); })
57
58#define _mm256_maskz_gf2p8affineinv_epi64_epi8(U, A, B, I) __extension__ ({ \
59 (__m256i)_mm256_mask_gf2p8affineinv_epi64_epi8((__m256i)_mm256_setzero_si256(), \
60 U, A, B, I); })
61
62
63#define _mm512_gf2p8affineinv_epi64_epi8(A, B, I) __extension__ ({ \
64 (__m512i)__builtin_ia32_vgf2p8affineinvqb_v64qi((__v64qi)(__m512i)(A), \
65 (__v64qi)(__m512i)(B), \
66 (char)(I)); })
67
68#define _mm512_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, I) __extension__ ({ \
69 (__m512i)__builtin_ia32_selectb_512((__mmask64)(U), \
70 (__v64qi)_mm512_gf2p8affineinv_epi64_epi8(A, B, I), \
71 (__v64qi)(__m512i)(S)); })
72
73#define _mm512_maskz_gf2p8affineinv_epi64_epi8(U, A, B, I) __extension__ ({ \
Craig Topperdff5b312018-05-30 18:02:11 +000074 (__m512i)_mm512_mask_gf2p8affineinv_epi64_epi8((__m512i)_mm512_setzero_si512(), \
Coby Tayreef4811eb2017-12-27 08:37:47 +000075 U, A, B, I); })
76
77#define _mm_gf2p8affine_epi64_epi8(A, B, I) __extension__ ({ \
78 (__m128i)__builtin_ia32_vgf2p8affineqb_v16qi((__v16qi)(__m128i)(A), \
79 (__v16qi)(__m128i)(B), \
80 (char)(I)); })
81
82#define _mm_mask_gf2p8affine_epi64_epi8(S, U, A, B, I) __extension__ ({ \
83 (__m128i)__builtin_ia32_selectb_128((__mmask16)(U), \
84 (__v16qi)_mm_gf2p8affine_epi64_epi8(A, B, I), \
85 (__v16qi)(__m128i)(S)); })
86
87
88#define _mm_maskz_gf2p8affine_epi64_epi8(U, A, B, I) __extension__ ({ \
89 (__m128i)_mm_mask_gf2p8affine_epi64_epi8((__m128i)_mm_setzero_si128(), \
90 U, A, B, I); })
91
92
93#define _mm256_gf2p8affine_epi64_epi8(A, B, I) __extension__ ({ \
94 (__m256i)__builtin_ia32_vgf2p8affineqb_v32qi((__v32qi)(__m256i)(A), \
95 (__v32qi)(__m256i)(B), \
96 (char)(I)); })
97
98#define _mm256_mask_gf2p8affine_epi64_epi8(S, U, A, B, I) __extension__ ({ \
99 (__m256i)__builtin_ia32_selectb_256((__mmask32)(U), \
100 (__v32qi)_mm256_gf2p8affine_epi64_epi8(A, B, I), \
101 (__v32qi)(__m256i)(S)); })
102
103#define _mm256_maskz_gf2p8affine_epi64_epi8(U, A, B, I) __extension__ ({ \
104 (__m256i)_mm256_mask_gf2p8affine_epi64_epi8((__m256i)_mm256_setzero_si256(), \
105 U, A, B, I); })
106
107
108#define _mm512_gf2p8affine_epi64_epi8(A, B, I) __extension__ ({ \
109 (__m512i)__builtin_ia32_vgf2p8affineqb_v64qi((__v64qi)(__m512i)(A), \
110 (__v64qi)(__m512i)(B), \
111 (char)(I)); })
112
113#define _mm512_mask_gf2p8affine_epi64_epi8(S, U, A, B, I) __extension__ ({ \
114 (__m512i)__builtin_ia32_selectb_512((__mmask64)(U), \
115 (__v64qi)_mm512_gf2p8affine_epi64_epi8(A, B, I), \
116 (__v64qi)(__m512i)(S)); })
117
118#define _mm512_maskz_gf2p8affine_epi64_epi8(U, A, B, I) __extension__ ({ \
Craig Topperdff5b312018-05-30 18:02:11 +0000119 (__m512i)_mm512_mask_gf2p8affine_epi64_epi8((__m512i)_mm512_setzero_si512(), \
Coby Tayreef4811eb2017-12-27 08:37:47 +0000120 U, A, B, I); })
121
122/* Default attributes for simple form (no masking). */
123#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("gfni")))
124
Craig Topper66ef4182018-05-07 21:47:11 +0000125/* Default attributes for YMM unmasked form. */
126#define __DEFAULT_FN_ATTRS_Y __attribute__((__always_inline__, __nodebug__, __target__("avx,gfni")))
127
Coby Tayreef4811eb2017-12-27 08:37:47 +0000128/* Default attributes for ZMM forms. */
Craig Topper66ef4182018-05-07 21:47:11 +0000129#define __DEFAULT_FN_ATTRS_Z __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,gfni")))
Coby Tayreef4811eb2017-12-27 08:37:47 +0000130
131/* Default attributes for VLX forms. */
132#define __DEFAULT_FN_ATTRS_VL __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,avx512vl,gfni")))
133
134static __inline__ __m128i __DEFAULT_FN_ATTRS
135_mm_gf2p8mul_epi8(__m128i __A, __m128i __B)
136{
137 return (__m128i) __builtin_ia32_vgf2p8mulb_v16qi((__v16qi) __A,
138 (__v16qi) __B);
139}
140
141static __inline__ __m128i __DEFAULT_FN_ATTRS_VL
142_mm_mask_gf2p8mul_epi8(__m128i __S, __mmask16 __U, __m128i __A, __m128i __B)
143{
144 return (__m128i) __builtin_ia32_selectb_128(__U,
145 (__v16qi) _mm_gf2p8mul_epi8(__A, __B),
146 (__v16qi) __S);
147}
148
149static __inline__ __m128i __DEFAULT_FN_ATTRS_VL
150_mm_maskz_gf2p8mul_epi8(__mmask16 __U, __m128i __A, __m128i __B)
151{
152 return _mm_mask_gf2p8mul_epi8((__m128i)_mm_setzero_si128(),
153 __U, __A, __B);
154}
155
Craig Topper66ef4182018-05-07 21:47:11 +0000156static __inline__ __m256i __DEFAULT_FN_ATTRS_Y
Coby Tayreef4811eb2017-12-27 08:37:47 +0000157_mm256_gf2p8mul_epi8(__m256i __A, __m256i __B)
158{
159 return (__m256i) __builtin_ia32_vgf2p8mulb_v32qi((__v32qi) __A,
160 (__v32qi) __B);
161}
162
163static __inline__ __m256i __DEFAULT_FN_ATTRS_VL
164_mm256_mask_gf2p8mul_epi8(__m256i __S, __mmask32 __U, __m256i __A, __m256i __B)
165{
166 return (__m256i) __builtin_ia32_selectb_256(__U,
167 (__v32qi) _mm256_gf2p8mul_epi8(__A, __B),
168 (__v32qi) __S);
169}
170
171static __inline__ __m256i __DEFAULT_FN_ATTRS_VL
172_mm256_maskz_gf2p8mul_epi8(__mmask32 __U, __m256i __A, __m256i __B)
173{
174 return _mm256_mask_gf2p8mul_epi8((__m256i)_mm256_setzero_si256(),
175 __U, __A, __B);
176}
177
Craig Topper66ef4182018-05-07 21:47:11 +0000178static __inline__ __m512i __DEFAULT_FN_ATTRS_Z
Coby Tayreef4811eb2017-12-27 08:37:47 +0000179_mm512_gf2p8mul_epi8(__m512i __A, __m512i __B)
180{
181 return (__m512i) __builtin_ia32_vgf2p8mulb_v64qi((__v64qi) __A,
182 (__v64qi) __B);
183}
184
Craig Topper66ef4182018-05-07 21:47:11 +0000185static __inline__ __m512i __DEFAULT_FN_ATTRS_Z
Coby Tayreef4811eb2017-12-27 08:37:47 +0000186_mm512_mask_gf2p8mul_epi8(__m512i __S, __mmask64 __U, __m512i __A, __m512i __B)
187{
188 return (__m512i) __builtin_ia32_selectb_512(__U,
189 (__v64qi) _mm512_gf2p8mul_epi8(__A, __B),
190 (__v64qi) __S);
191}
192
Craig Topper66ef4182018-05-07 21:47:11 +0000193static __inline__ __m512i __DEFAULT_FN_ATTRS_Z
Coby Tayreef4811eb2017-12-27 08:37:47 +0000194_mm512_maskz_gf2p8mul_epi8(__mmask64 __U, __m512i __A, __m512i __B)
195{
Craig Topperdff5b312018-05-30 18:02:11 +0000196 return _mm512_mask_gf2p8mul_epi8((__m512i)_mm512_setzero_si512(),
Coby Tayreef4811eb2017-12-27 08:37:47 +0000197 __U, __A, __B);
198}
199
200#undef __DEFAULT_FN_ATTRS
Craig Topper66ef4182018-05-07 21:47:11 +0000201#undef __DEFAULT_FN_ATTRS_Y
202#undef __DEFAULT_FN_ATTRS_Z
Coby Tayreef4811eb2017-12-27 08:37:47 +0000203#undef __DEFAULT_FN_ATTRS_VL
204
Craig Topper73d1d402018-05-30 22:33:21 +0000205#endif /* __GFNIINTRIN_H */
Coby Tayreef4811eb2017-12-27 08:37:47 +0000206