FFmpeg
cpu.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #include "libavutil/cpu.h"
20 #include "libavutil/cpu_internal.h"
21 #include "config.h"
22 
23 #if HAVE_GETAUXVAL || HAVE_ELF_AUX_INFO
24 #include <stdint.h>
25 #include <sys/auxv.h>
26 
27 #define HWCAP_AARCH64_PMULL (1 << 4)
28 #define HWCAP_AARCH64_CRC32 (1 << 7)
29 #define HWCAP_AARCH64_SHA3 (1 << 17)
30 #define HWCAP_AARCH64_ASIMDDP (1 << 20)
31 #define HWCAP_AARCH64_SVE (1 << 22)
32 #define HWCAP2_AARCH64_SVE2 (1 << 1)
33 #define HWCAP2_AARCH64_I8MM (1 << 13)
34 #define HWCAP2_AARCH64_SME (1 << 23)
35 #define HWCAP2_AARCH64_SME_I16I64 (1 << 24)
36 #define HWCAP2_AARCH64_SME2 (1ULL << 37)
37 
38 static int detect_flags(void)
39 {
40  int flags = 0;
41 
42  unsigned long hwcap = ff_getauxval(AT_HWCAP);
43  unsigned long hwcap2 = ff_getauxval(AT_HWCAP2);
44 
45  if (hwcap & HWCAP_AARCH64_PMULL)
46  flags |= AV_CPU_FLAG_PMULL;
47  if (hwcap & HWCAP_AARCH64_SHA3)
48  flags |= AV_CPU_FLAG_EOR3;
49  if (hwcap & HWCAP_AARCH64_CRC32)
50  flags |= AV_CPU_FLAG_ARM_CRC;
51  if (hwcap & HWCAP_AARCH64_ASIMDDP)
52  flags |= AV_CPU_FLAG_DOTPROD;
53  if (hwcap & HWCAP_AARCH64_SVE)
54  flags |= AV_CPU_FLAG_SVE;
55  if (hwcap2 & HWCAP2_AARCH64_SVE2)
56  flags |= AV_CPU_FLAG_SVE2;
57  if (hwcap2 & HWCAP2_AARCH64_I8MM)
58  flags |= AV_CPU_FLAG_I8MM;
59  if (hwcap2 & HWCAP2_AARCH64_SME)
60  flags |= AV_CPU_FLAG_SME;
61  if (hwcap2 & HWCAP2_AARCH64_SME_I16I64)
62  flags |= AV_CPU_FLAG_SME_I16I64;
63  if (hwcap2 & HWCAP2_AARCH64_SME2)
64  flags |= AV_CPU_FLAG_SME2;
65 
66  return flags;
67 }
68 
69 #elif defined(__APPLE__) && HAVE_SYSCTLBYNAME
70 #include <sys/sysctl.h>
71 
72 static int have_feature(const char *feature) {
73  uint32_t value = 0;
74  size_t size = sizeof(value);
75  if (!sysctlbyname(feature, &value, &size, NULL, 0))
76  return value;
77  return 0;
78 }
79 
80 static int detect_flags(void)
81 {
82  int flags = 0;
83 
84  if (have_feature("hw.optional.arm.FEAT_DotProd"))
85  flags |= AV_CPU_FLAG_DOTPROD;
86  if (have_feature("hw.optional.arm.FEAT_I8MM"))
87  flags |= AV_CPU_FLAG_I8MM;
88  if (have_feature("hw.optional.arm.FEAT_SME"))
89  flags |= AV_CPU_FLAG_SME;
90  if (have_feature("hw.optional.arm.FEAT_SME_I16I64"))
91  flags |= AV_CPU_FLAG_SME_I16I64;
92  if (have_feature("hw.optional.armv8_crc32"))
93  flags |= AV_CPU_FLAG_ARM_CRC;
94  if (have_feature("hw.optional.arm.FEAT_PMULL"))
95  flags |= AV_CPU_FLAG_PMULL;
96  if (have_feature("hw.optional.armv8_2_sha3"))
97  flags |= AV_CPU_FLAG_EOR3;
98  if (have_feature("hw.optional.arm.FEAT_SME2"))
99  flags |= AV_CPU_FLAG_SME2;
100 
101  return flags;
102 }
103 
104 #elif defined(__OpenBSD__)
105 #include <machine/armreg.h>
106 #include <machine/cpu.h>
107 #include <sys/types.h>
108 #include <sys/sysctl.h>
109 
110 static int detect_flags(void)
111 {
112  int flags = 0;
113 
114 #ifdef CPU_ID_AA64ISAR0
115  int mib[2];
116  uint64_t isar0;
117  uint64_t isar1;
118  size_t len;
119 
120  mib[0] = CTL_MACHDEP;
121  mib[1] = CPU_ID_AA64ISAR0;
122  len = sizeof(isar0);
123  if (sysctl(mib, 2, &isar0, &len, NULL, 0) != -1) {
124  if (ID_AA64ISAR0_DP(isar0) >= ID_AA64ISAR0_DP_IMPL)
125  flags |= AV_CPU_FLAG_DOTPROD;
126  if (ID_AA64ISAR0_CRC32(isar0) >= ID_AA64ISAR0_CRC32_BASE)
127  flags |= AV_CPU_FLAG_ARM_CRC;
128  if (ID_AA64ISAR0_AES(isar0) >= ID_AA64ISAR0_AES_PMULL)
129  flags |= AV_CPU_FLAG_PMULL;
130  if (ID_AA64ISAR0_SHA3(isar0) >= ID_AA64ISAR0_SHA3_IMPL)
131  flags |= AV_CPU_FLAG_EOR3;
132  }
133 
134  mib[0] = CTL_MACHDEP;
135  mib[1] = CPU_ID_AA64ISAR1;
136  len = sizeof(isar1);
137  if (sysctl(mib, 2, &isar1, &len, NULL, 0) != -1) {
138 #ifdef ID_AA64ISAR1_I8MM_IMPL
139  if (ID_AA64ISAR1_I8MM(isar1) >= ID_AA64ISAR1_I8MM_IMPL)
140  flags |= AV_CPU_FLAG_I8MM;
141 #endif
142  }
143 #endif
144 
145  return flags;
146 }
147 
148 #elif defined(_WIN32)
149 #include <windows.h>
150 
151 static int detect_flags(void)
152 {
153  int flags = 0;
154 #ifdef PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE
155  if (IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE))
156  flags |= AV_CPU_FLAG_ARM_CRC;
157 #endif
158 #ifdef PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE
159  if (IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE))
160  flags |= AV_CPU_FLAG_PMULL;
161 #endif
162 #ifdef PF_ARM_SHA3_INSTRUCTIONS_AVAILABLE
163  if (IsProcessorFeaturePresent(PF_ARM_SHA3_INSTRUCTIONS_AVAILABLE))
164  flags |= AV_CPU_FLAG_EOR3;
165 #endif
166 #ifdef PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE
167  if (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE))
168  flags |= AV_CPU_FLAG_DOTPROD;
169 #endif
170 #ifdef PF_ARM_V82_I8MM_INSTRUCTIONS_AVAILABLE
171  if (IsProcessorFeaturePresent(PF_ARM_V82_I8MM_INSTRUCTIONS_AVAILABLE))
172  flags |= AV_CPU_FLAG_I8MM;
173 #endif
174 #ifdef PF_ARM_SVE_INSTRUCTIONS_AVAILABLE
175  if (IsProcessorFeaturePresent(PF_ARM_SVE_INSTRUCTIONS_AVAILABLE))
176  flags |= AV_CPU_FLAG_SVE;
177 #endif
178 #ifdef PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE
179  if (IsProcessorFeaturePresent(PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE))
180  flags |= AV_CPU_FLAG_SVE2;
181 #endif
182 #ifdef PF_ARM_SME_INSTRUCTIONS_AVAILABLE
183  if (IsProcessorFeaturePresent(PF_ARM_SME_INSTRUCTIONS_AVAILABLE))
184  flags |= AV_CPU_FLAG_SME;
185 #endif
186 #ifdef PF_ARM_SME_I16I64_INSTRUCTIONS_AVAILABLE
187  if (IsProcessorFeaturePresent(PF_ARM_SME_I16I64_INSTRUCTIONS_AVAILABLE))
188  flags |= AV_CPU_FLAG_SME_I16I64;
189 #endif
190 #ifdef PF_ARM_SME2_INSTRUCTIONS_AVAILABLE
191  if (IsProcessorFeaturePresent(PF_ARM_SME2_INSTRUCTIONS_AVAILABLE))
192  flags |= AV_CPU_FLAG_SME2;
193 #endif
194  return flags;
195 }
196 #else
197 
198 static int detect_flags(void)
199 {
200  return 0;
201 }
202 
203 #endif
204 
205 int ff_get_cpu_flags_aarch64(void)
206 {
207  int flags = AV_CPU_FLAG_ARMV8 * HAVE_ARMV8 |
208  AV_CPU_FLAG_NEON * HAVE_NEON;
209 
210 #ifdef __ARM_FEATURE_DOTPROD
211  flags |= AV_CPU_FLAG_DOTPROD;
212 #endif
213 #ifdef __ARM_FEATURE_MATMUL_INT8
214  flags |= AV_CPU_FLAG_I8MM;
215 #endif
216 #ifdef __ARM_FEATURE_SVE
217  flags |= AV_CPU_FLAG_SVE;
218 #endif
219 #ifdef __ARM_FEATURE_SVE2
220  flags |= AV_CPU_FLAG_SVE2;
221 #endif
222 #ifdef __ARM_FEATURE_SME
223  flags |= AV_CPU_FLAG_SME;
224 #endif
225 #ifdef __ARM_FEATURE_CRC32
226  flags |= AV_CPU_FLAG_ARM_CRC;
227 #endif
228 #ifdef __ARM_FEATURE_AES
229  flags |= AV_CPU_FLAG_PMULL;
230 #endif
231 #ifdef __ARM_FEATURE_SHA3
232  flags |= AV_CPU_FLAG_EOR3;
233 #endif
234 #ifdef __ARM_FEATURE_SME_I16I64
235  flags |= AV_CPU_FLAG_SME_I16I64;
236 #endif
237 #ifdef __ARM_FEATURE_SME2
238  flags |= AV_CPU_FLAG_SME2;
239 #endif
240 
241  flags |= detect_flags();
242 
243  return flags;
244 }
245 
246 size_t ff_get_cpu_max_align_aarch64(void)
247 {
248  int flags = av_get_cpu_flags();
249 
250  if (flags & AV_CPU_FLAG_NEON)
251  return 16;
252 
253  return 8;
254 }
flags
const SwsFlags flags[]
Definition: swscale.c:71
AT_HWCAP
#define AT_HWCAP
Definition: cpu.c:50
AV_CPU_FLAG_SVE2
#define AV_CPU_FLAG_SVE2
Definition: cpu.h:79
AV_CPU_FLAG_SVE
#define AV_CPU_FLAG_SVE
Definition: cpu.h:78
AV_CPU_FLAG_DOTPROD
#define AV_CPU_FLAG_DOTPROD
Definition: cpu.h:76
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:109
AV_CPU_FLAG_PMULL
#define AV_CPU_FLAG_PMULL
Definition: cpu.h:85
cpu_internal.h
ff_get_cpu_max_align_aarch64
size_t ff_get_cpu_max_align_aarch64(void)
Definition: cpu.c:246
AV_CPU_FLAG_SME2
#define AV_CPU_FLAG_SME2
Definition: cpu.h:82
ff_getauxval
unsigned long ff_getauxval(unsigned long type)
Definition: cpu.c:309
NULL
#define NULL
Definition: coverity.c:32
ff_get_cpu_flags_aarch64
int ff_get_cpu_flags_aarch64(void)
Definition: cpu.c:205
detect_flags
static int detect_flags(void)
Definition: cpu.c:198
AV_CPU_FLAG_I8MM
#define AV_CPU_FLAG_I8MM
Definition: cpu.h:77
cpu.h
AV_CPU_FLAG_ARM_CRC
#define AV_CPU_FLAG_ARM_CRC
Definition: cpu.h:81
AV_CPU_FLAG_NEON
#define AV_CPU_FLAG_NEON
Definition: cpu.h:73
size
int size
Definition: twinvq_data.h:10344
value
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default value
Definition: writing_filters.txt:86
AV_CPU_FLAG_SME_I16I64
#define AV_CPU_FLAG_SME_I16I64
Definition: cpu.h:83
AV_CPU_FLAG_ARMV8
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:74
len
int len
Definition: vorbis_enc_data.h:426
AV_CPU_FLAG_EOR3
#define AV_CPU_FLAG_EOR3
Definition: cpu.h:86
AV_CPU_FLAG_SME
#define AV_CPU_FLAG_SME
Definition: cpu.h:80
AT_HWCAP2
#define AT_HWCAP2
Definition: cpu.c:53
Generated on Sat Mar 14 2026 19:23:38 for FFmpeg by   doxygen 1.8.17