annotate clang/lib/Headers/avx512vbmivlintrin.h @ 176:de4ac79aef9d

...
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Mon, 25 May 2020 17:13:11 +0900
parents 1d019706d866
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
150
anatofuz
parents:
diff changeset
1 /*===------------- avx512vbmivlintrin.h - VBMI 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 <avx512vbmivlintrin.h> directly; include <immintrin.h> instead."
anatofuz
parents:
diff changeset
12 #endif
anatofuz
parents:
diff changeset
13
anatofuz
parents:
diff changeset
14 #ifndef __VBMIVLINTRIN_H
anatofuz
parents:
diff changeset
15 #define __VBMIVLINTRIN_H
anatofuz
parents:
diff changeset
16
anatofuz
parents:
diff changeset
17 /* Define the default attributes for the functions in this file. */
anatofuz
parents:
diff changeset
18 #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi,avx512vl"), __min_vector_width__(128)))
anatofuz
parents:
diff changeset
19 #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi,avx512vl"), __min_vector_width__(256)))
anatofuz
parents:
diff changeset
20
anatofuz
parents:
diff changeset
21
anatofuz
parents:
diff changeset
22 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
23 _mm_permutex2var_epi8(__m128i __A, __m128i __I, __m128i __B)
anatofuz
parents:
diff changeset
24 {
anatofuz
parents:
diff changeset
25 return (__m128i)__builtin_ia32_vpermi2varqi128((__v16qi)__A,
anatofuz
parents:
diff changeset
26 (__v16qi)__I,
anatofuz
parents:
diff changeset
27 (__v16qi)__B);
anatofuz
parents:
diff changeset
28 }
anatofuz
parents:
diff changeset
29
anatofuz
parents:
diff changeset
30 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
31 _mm_mask_permutex2var_epi8(__m128i __A, __mmask16 __U, __m128i __I,
anatofuz
parents:
diff changeset
32 __m128i __B)
anatofuz
parents:
diff changeset
33 {
anatofuz
parents:
diff changeset
34 return (__m128i)__builtin_ia32_selectb_128(__U,
anatofuz
parents:
diff changeset
35 (__v16qi)_mm_permutex2var_epi8(__A, __I, __B),
anatofuz
parents:
diff changeset
36 (__v16qi)__A);
anatofuz
parents:
diff changeset
37 }
anatofuz
parents:
diff changeset
38
anatofuz
parents:
diff changeset
39 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
40 _mm_mask2_permutex2var_epi8(__m128i __A, __m128i __I, __mmask16 __U,
anatofuz
parents:
diff changeset
41 __m128i __B)
anatofuz
parents:
diff changeset
42 {
anatofuz
parents:
diff changeset
43 return (__m128i)__builtin_ia32_selectb_128(__U,
anatofuz
parents:
diff changeset
44 (__v16qi)_mm_permutex2var_epi8(__A, __I, __B),
anatofuz
parents:
diff changeset
45 (__v16qi)__I);
anatofuz
parents:
diff changeset
46 }
anatofuz
parents:
diff changeset
47
anatofuz
parents:
diff changeset
48 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
49 _mm_maskz_permutex2var_epi8(__mmask16 __U, __m128i __A, __m128i __I,
anatofuz
parents:
diff changeset
50 __m128i __B)
anatofuz
parents:
diff changeset
51 {
anatofuz
parents:
diff changeset
52 return (__m128i)__builtin_ia32_selectb_128(__U,
anatofuz
parents:
diff changeset
53 (__v16qi)_mm_permutex2var_epi8(__A, __I, __B),
anatofuz
parents:
diff changeset
54 (__v16qi)_mm_setzero_si128());
anatofuz
parents:
diff changeset
55 }
anatofuz
parents:
diff changeset
56
anatofuz
parents:
diff changeset
57 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
58 _mm256_permutex2var_epi8(__m256i __A, __m256i __I, __m256i __B)
anatofuz
parents:
diff changeset
59 {
anatofuz
parents:
diff changeset
60 return (__m256i)__builtin_ia32_vpermi2varqi256((__v32qi)__A, (__v32qi)__I,
anatofuz
parents:
diff changeset
61 (__v32qi)__B);
anatofuz
parents:
diff changeset
62 }
anatofuz
parents:
diff changeset
63
anatofuz
parents:
diff changeset
64 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
65 _mm256_mask_permutex2var_epi8(__m256i __A, __mmask32 __U, __m256i __I,
anatofuz
parents:
diff changeset
66 __m256i __B)
anatofuz
parents:
diff changeset
67 {
anatofuz
parents:
diff changeset
68 return (__m256i)__builtin_ia32_selectb_256(__U,
anatofuz
parents:
diff changeset
69 (__v32qi)_mm256_permutex2var_epi8(__A, __I, __B),
anatofuz
parents:
diff changeset
70 (__v32qi)__A);
anatofuz
parents:
diff changeset
71 }
anatofuz
parents:
diff changeset
72
anatofuz
parents:
diff changeset
73 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
74 _mm256_mask2_permutex2var_epi8(__m256i __A, __m256i __I, __mmask32 __U,
anatofuz
parents:
diff changeset
75 __m256i __B)
anatofuz
parents:
diff changeset
76 {
anatofuz
parents:
diff changeset
77 return (__m256i)__builtin_ia32_selectb_256(__U,
anatofuz
parents:
diff changeset
78 (__v32qi)_mm256_permutex2var_epi8(__A, __I, __B),
anatofuz
parents:
diff changeset
79 (__v32qi)__I);
anatofuz
parents:
diff changeset
80 }
anatofuz
parents:
diff changeset
81
anatofuz
parents:
diff changeset
82 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
83 _mm256_maskz_permutex2var_epi8(__mmask32 __U, __m256i __A, __m256i __I,
anatofuz
parents:
diff changeset
84 __m256i __B)
anatofuz
parents:
diff changeset
85 {
anatofuz
parents:
diff changeset
86 return (__m256i)__builtin_ia32_selectb_256(__U,
anatofuz
parents:
diff changeset
87 (__v32qi)_mm256_permutex2var_epi8(__A, __I, __B),
anatofuz
parents:
diff changeset
88 (__v32qi)_mm256_setzero_si256());
anatofuz
parents:
diff changeset
89 }
anatofuz
parents:
diff changeset
90
anatofuz
parents:
diff changeset
91 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
92 _mm_permutexvar_epi8 (__m128i __A, __m128i __B)
anatofuz
parents:
diff changeset
93 {
anatofuz
parents:
diff changeset
94 return (__m128i)__builtin_ia32_permvarqi128((__v16qi)__B, (__v16qi)__A);
anatofuz
parents:
diff changeset
95 }
anatofuz
parents:
diff changeset
96
anatofuz
parents:
diff changeset
97 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
98 _mm_maskz_permutexvar_epi8 (__mmask16 __M, __m128i __A, __m128i __B)
anatofuz
parents:
diff changeset
99 {
anatofuz
parents:
diff changeset
100 return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
anatofuz
parents:
diff changeset
101 (__v16qi)_mm_permutexvar_epi8(__A, __B),
anatofuz
parents:
diff changeset
102 (__v16qi)_mm_setzero_si128());
anatofuz
parents:
diff changeset
103 }
anatofuz
parents:
diff changeset
104
anatofuz
parents:
diff changeset
105 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
106 _mm_mask_permutexvar_epi8 (__m128i __W, __mmask16 __M, __m128i __A,
anatofuz
parents:
diff changeset
107 __m128i __B)
anatofuz
parents:
diff changeset
108 {
anatofuz
parents:
diff changeset
109 return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
anatofuz
parents:
diff changeset
110 (__v16qi)_mm_permutexvar_epi8(__A, __B),
anatofuz
parents:
diff changeset
111 (__v16qi)__W);
anatofuz
parents:
diff changeset
112 }
anatofuz
parents:
diff changeset
113
anatofuz
parents:
diff changeset
114 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
115 _mm256_permutexvar_epi8 (__m256i __A, __m256i __B)
anatofuz
parents:
diff changeset
116 {
anatofuz
parents:
diff changeset
117 return (__m256i)__builtin_ia32_permvarqi256((__v32qi) __B, (__v32qi) __A);
anatofuz
parents:
diff changeset
118 }
anatofuz
parents:
diff changeset
119
anatofuz
parents:
diff changeset
120 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
121 _mm256_maskz_permutexvar_epi8 (__mmask32 __M, __m256i __A,
anatofuz
parents:
diff changeset
122 __m256i __B)
anatofuz
parents:
diff changeset
123 {
anatofuz
parents:
diff changeset
124 return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
anatofuz
parents:
diff changeset
125 (__v32qi)_mm256_permutexvar_epi8(__A, __B),
anatofuz
parents:
diff changeset
126 (__v32qi)_mm256_setzero_si256());
anatofuz
parents:
diff changeset
127 }
anatofuz
parents:
diff changeset
128
anatofuz
parents:
diff changeset
129 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
130 _mm256_mask_permutexvar_epi8 (__m256i __W, __mmask32 __M, __m256i __A,
anatofuz
parents:
diff changeset
131 __m256i __B)
anatofuz
parents:
diff changeset
132 {
anatofuz
parents:
diff changeset
133 return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
anatofuz
parents:
diff changeset
134 (__v32qi)_mm256_permutexvar_epi8(__A, __B),
anatofuz
parents:
diff changeset
135 (__v32qi)__W);
anatofuz
parents:
diff changeset
136 }
anatofuz
parents:
diff changeset
137
anatofuz
parents:
diff changeset
138 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
139 _mm_multishift_epi64_epi8(__m128i __X, __m128i __Y)
anatofuz
parents:
diff changeset
140 {
anatofuz
parents:
diff changeset
141 return (__m128i)__builtin_ia32_vpmultishiftqb128((__v16qi)__X, (__v16qi)__Y);
anatofuz
parents:
diff changeset
142 }
anatofuz
parents:
diff changeset
143
anatofuz
parents:
diff changeset
144 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
145 _mm_mask_multishift_epi64_epi8(__m128i __W, __mmask16 __M, __m128i __X,
anatofuz
parents:
diff changeset
146 __m128i __Y)
anatofuz
parents:
diff changeset
147 {
anatofuz
parents:
diff changeset
148 return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
anatofuz
parents:
diff changeset
149 (__v16qi)_mm_multishift_epi64_epi8(__X, __Y),
anatofuz
parents:
diff changeset
150 (__v16qi)__W);
anatofuz
parents:
diff changeset
151 }
anatofuz
parents:
diff changeset
152
anatofuz
parents:
diff changeset
153 static __inline__ __m128i __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
154 _mm_maskz_multishift_epi64_epi8(__mmask16 __M, __m128i __X, __m128i __Y)
anatofuz
parents:
diff changeset
155 {
anatofuz
parents:
diff changeset
156 return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
anatofuz
parents:
diff changeset
157 (__v16qi)_mm_multishift_epi64_epi8(__X, __Y),
anatofuz
parents:
diff changeset
158 (__v16qi)_mm_setzero_si128());
anatofuz
parents:
diff changeset
159 }
anatofuz
parents:
diff changeset
160
anatofuz
parents:
diff changeset
161 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
162 _mm256_multishift_epi64_epi8(__m256i __X, __m256i __Y)
anatofuz
parents:
diff changeset
163 {
anatofuz
parents:
diff changeset
164 return (__m256i)__builtin_ia32_vpmultishiftqb256((__v32qi)__X, (__v32qi)__Y);
anatofuz
parents:
diff changeset
165 }
anatofuz
parents:
diff changeset
166
anatofuz
parents:
diff changeset
167 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
168 _mm256_mask_multishift_epi64_epi8(__m256i __W, __mmask32 __M, __m256i __X,
anatofuz
parents:
diff changeset
169 __m256i __Y)
anatofuz
parents:
diff changeset
170 {
anatofuz
parents:
diff changeset
171 return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
anatofuz
parents:
diff changeset
172 (__v32qi)_mm256_multishift_epi64_epi8(__X, __Y),
anatofuz
parents:
diff changeset
173 (__v32qi)__W);
anatofuz
parents:
diff changeset
174 }
anatofuz
parents:
diff changeset
175
anatofuz
parents:
diff changeset
176 static __inline__ __m256i __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
177 _mm256_maskz_multishift_epi64_epi8(__mmask32 __M, __m256i __X, __m256i __Y)
anatofuz
parents:
diff changeset
178 {
anatofuz
parents:
diff changeset
179 return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
anatofuz
parents:
diff changeset
180 (__v32qi)_mm256_multishift_epi64_epi8(__X, __Y),
anatofuz
parents:
diff changeset
181 (__v32qi)_mm256_setzero_si256());
anatofuz
parents:
diff changeset
182 }
anatofuz
parents:
diff changeset
183
anatofuz
parents:
diff changeset
184
anatofuz
parents:
diff changeset
185 #undef __DEFAULT_FN_ATTRS128
anatofuz
parents:
diff changeset
186 #undef __DEFAULT_FN_ATTRS256
anatofuz
parents:
diff changeset
187
anatofuz
parents:
diff changeset
188 #endif