|
145
|
1 /* Copyright (C) 2017-2020 Free Software Foundation, Inc.
|
|
131
|
2
|
|
|
3 This file is part of GCC.
|
|
|
4
|
|
|
5 GCC is free software; you can redistribute it and/or modify
|
|
|
6 it under the terms of the GNU General Public License as published by
|
|
|
7 the Free Software Foundation; either version 3, or (at your option)
|
|
|
8 any later version.
|
|
|
9
|
|
|
10 GCC is distributed in the hope that it will be useful,
|
|
|
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
13 GNU General Public License for more details.
|
|
|
14
|
|
|
15 Under Section 7 of GPL version 3, you are granted additional
|
|
|
16 permissions described in the GCC Runtime Library Exception, version
|
|
|
17 3.1, as published by the Free Software Foundation.
|
|
|
18
|
|
|
19 You should have received a copy of the GNU General Public License and
|
|
|
20 a copy of the GCC Runtime Library Exception along with this program;
|
|
|
21 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
|
22 <http://www.gnu.org/licenses/>. */
|
|
|
23
|
|
|
24 #if !defined _IMMINTRIN_H_INCLUDED
|
|
|
25 # error "Never use <avx512vpopcntdqvlintrin.h> directly; include <immintrin.h> instead."
|
|
|
26 #endif
|
|
|
27
|
|
|
28 #ifndef _AVX512VPOPCNTDQVLINTRIN_H_INCLUDED
|
|
|
29 #define _AVX512VPOPCNTDQVLINTRIN_H_INCLUDED
|
|
|
30
|
|
|
31 #if !defined(__AVX512VPOPCNTDQ__) || !defined(__AVX512VL__)
|
|
|
32 #pragma GCC push_options
|
|
|
33 #pragma GCC target("avx512vpopcntdq,avx512vl")
|
|
|
34 #define __DISABLE_AVX512VPOPCNTDQVL__
|
|
|
35 #endif /* __AVX512VPOPCNTDQVL__ */
|
|
|
36
|
|
|
37 extern __inline __m128i
|
|
|
38 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
39 _mm_popcnt_epi32 (__m128i __A)
|
|
|
40 {
|
|
|
41 return (__m128i) __builtin_ia32_vpopcountd_v4si ((__v4si) __A);
|
|
|
42 }
|
|
|
43
|
|
|
44 extern __inline __m128i
|
|
|
45 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
46 _mm_mask_popcnt_epi32 (__m128i __A, __mmask16 __U, __m128i __B)
|
|
|
47 {
|
|
|
48 return (__m128i) __builtin_ia32_vpopcountd_v4si_mask ((__v4si) __A,
|
|
|
49 (__v4si) __B,
|
|
|
50 (__mmask16) __U);
|
|
|
51 }
|
|
|
52
|
|
|
53 extern __inline __m128i
|
|
|
54 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
55 _mm_maskz_popcnt_epi32 (__mmask16 __U, __m128i __A)
|
|
|
56 {
|
|
|
57 return (__m128i) __builtin_ia32_vpopcountd_v4si_mask ((__v4si) __A,
|
|
|
58 (__v4si)
|
|
|
59 _mm_setzero_si128 (),
|
|
|
60 (__mmask16) __U);
|
|
|
61 }
|
|
|
62
|
|
|
63 extern __inline __m256i
|
|
|
64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
65 _mm256_popcnt_epi32 (__m256i __A)
|
|
|
66 {
|
|
|
67 return (__m256i) __builtin_ia32_vpopcountd_v8si ((__v8si) __A);
|
|
|
68 }
|
|
|
69
|
|
|
70 extern __inline __m256i
|
|
|
71 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
72 _mm256_mask_popcnt_epi32 (__m256i __A, __mmask16 __U, __m256i __B)
|
|
|
73 {
|
|
|
74 return (__m256i) __builtin_ia32_vpopcountd_v8si_mask ((__v8si) __A,
|
|
|
75 (__v8si) __B,
|
|
|
76 (__mmask16) __U);
|
|
|
77 }
|
|
|
78
|
|
|
79 extern __inline __m256i
|
|
|
80 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
81 _mm256_maskz_popcnt_epi32 (__mmask16 __U, __m256i __A)
|
|
|
82 {
|
|
|
83 return (__m256i) __builtin_ia32_vpopcountd_v8si_mask ((__v8si) __A,
|
|
|
84 (__v8si)
|
|
|
85 _mm256_setzero_si256 (),
|
|
|
86 (__mmask16) __U);
|
|
|
87 }
|
|
|
88
|
|
|
89 extern __inline __m128i
|
|
|
90 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
91 _mm_popcnt_epi64 (__m128i __A)
|
|
|
92 {
|
|
|
93 return (__m128i) __builtin_ia32_vpopcountq_v2di ((__v2di) __A);
|
|
|
94 }
|
|
|
95
|
|
|
96 extern __inline __m128i
|
|
|
97 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
98 _mm_mask_popcnt_epi64 (__m128i __A, __mmask8 __U, __m128i __B)
|
|
|
99 {
|
|
|
100 return (__m128i) __builtin_ia32_vpopcountq_v2di_mask ((__v2di) __A,
|
|
|
101 (__v2di) __B,
|
|
|
102 (__mmask8) __U);
|
|
|
103 }
|
|
|
104
|
|
|
105 extern __inline __m128i
|
|
|
106 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
107 _mm_maskz_popcnt_epi64 (__mmask8 __U, __m128i __A)
|
|
|
108 {
|
|
|
109 return (__m128i) __builtin_ia32_vpopcountq_v2di_mask ((__v2di) __A,
|
|
|
110 (__v2di)
|
|
|
111 _mm_setzero_si128 (),
|
|
|
112 (__mmask8) __U);
|
|
|
113 }
|
|
|
114
|
|
|
115 extern __inline __m256i
|
|
|
116 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
117 _mm256_popcnt_epi64 (__m256i __A)
|
|
|
118 {
|
|
|
119 return (__m256i) __builtin_ia32_vpopcountq_v4di ((__v4di) __A);
|
|
|
120 }
|
|
|
121
|
|
|
122 extern __inline __m256i
|
|
|
123 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
124 _mm256_mask_popcnt_epi64 (__m256i __A, __mmask8 __U, __m256i __B)
|
|
|
125 {
|
|
|
126 return (__m256i) __builtin_ia32_vpopcountq_v4di_mask ((__v4di) __A,
|
|
|
127 (__v4di) __B,
|
|
|
128 (__mmask8) __U);
|
|
|
129 }
|
|
|
130
|
|
|
131 extern __inline __m256i
|
|
|
132 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
|
|
|
133 _mm256_maskz_popcnt_epi64 (__mmask8 __U, __m256i __A)
|
|
|
134 {
|
|
|
135 return (__m256i) __builtin_ia32_vpopcountq_v4di_mask ((__v4di) __A,
|
|
|
136 (__v4di)
|
|
|
137 _mm256_setzero_si256 (),
|
|
|
138 (__mmask8) __U);
|
|
|
139 }
|
|
|
140
|
|
|
141 #ifdef __DISABLE_AVX512VPOPCNTDQVL__
|
|
|
142 #undef __DISABLE_AVX512VPOPCNTDQVL__
|
|
|
143 #pragma GCC pop_options
|
|
|
144 #endif /* __DISABLE_AVX512VPOPCNTDQVL__ */
|
|
|
145
|
|
|
146 #endif /* _AVX512VPOPCNTDQVLINTRIN_H_INCLUDED */
|
|
|
147
|
