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 (defined(__linux__) || defined(__ANDROID__)) && HAVE_GETAUXVAL
24 #include <stdint.h>
25 #include <sys/auxv.h>
26 
27 #define get_cpu_feature_reg(reg, val) \
28  __asm__("mrs %0, " #reg : "=r" (val))
29 
30 static int detect_flags(void)
31 {
32  int flags = 0;
33 
34 #if defined(HWCAP_CPUID) && HAVE_INLINE_ASM
35  unsigned long hwcap = getauxval(AT_HWCAP);
36  // We can check for DOTPROD and I8MM using HWCAP_ASIMDDP and
37  // HWCAP2_I8MM too, avoiding to read the CPUID registers (which triggers
38  // a trap, handled by the kernel). However the HWCAP_* defines for these
39  // extensions are added much later than HWCAP_CPUID, so the userland
40  // headers might lack support for them even if the binary later is run
41  // on hardware that does support it (and where the kernel might support
42  // HWCAP_CPUID).
43  // See https://www.kernel.org/doc/html/latest/arm64/cpu-feature-registers.html
44  if (hwcap & HWCAP_CPUID) {
45  uint64_t tmp;
46 
47  get_cpu_feature_reg(ID_AA64ISAR0_EL1, tmp);
48  if (((tmp >> 44) & 0xf) == 0x1)
50  get_cpu_feature_reg(ID_AA64ISAR1_EL1, tmp);
51  if (((tmp >> 52) & 0xf) == 0x1)
53  }
54 #endif
55 
56  return flags;
57 }
58 
59 #elif defined(__APPLE__) && HAVE_SYSCTLBYNAME
60 #include <sys/sysctl.h>
61 
62 static int detect_flags(void)
63 {
64  uint32_t value = 0;
65  size_t size;
66  int flags = 0;
67 
68  size = sizeof(value);
69  if (!sysctlbyname("hw.optional.arm.FEAT_DotProd", &value, &size, NULL, 0)) {
70  if (value)
72  }
73  size = sizeof(value);
74  if (!sysctlbyname("hw.optional.arm.FEAT_I8MM", &value, &size, NULL, 0)) {
75  if (value)
77  }
78  return flags;
79 }
80 
81 #elif defined(_WIN32)
82 #include <windows.h>
83 
84 static int detect_flags(void)
85 {
86  int flags = 0;
87 #ifdef PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE
88  if (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE))
90 #endif
91  return flags;
92 }
93 #else
94 
95 static int detect_flags(void)
96 {
97  return 0;
98 }
99 
100 #endif
101 
103 {
104  int flags = AV_CPU_FLAG_ARMV8 * HAVE_ARMV8 |
105  AV_CPU_FLAG_NEON * HAVE_NEON;
106 
107 #ifdef __ARM_FEATURE_DOTPROD
109 #endif
110 #ifdef __ARM_FEATURE_MATMUL_INT8
112 #endif
113 
114  flags |= detect_flags();
115 
116  return flags;
117 }
118 
120 {
121  int flags = av_get_cpu_flags();
122 
123  if (flags & AV_CPU_FLAG_NEON)
124  return 16;
125 
126  return 8;
127 }
tmp
static uint8_t tmp[11]
Definition: aes_ctr.c:28
AV_CPU_FLAG_DOTPROD
#define AV_CPU_FLAG_DOTPROD
Definition: cpu.h:72
av_get_cpu_flags
int av_get_cpu_flags(void)
Return the flags which specify extensions supported by the CPU.
Definition: cpu.c:103
cpu_internal.h
ff_get_cpu_max_align_aarch64
size_t ff_get_cpu_max_align_aarch64(void)
Definition: cpu.c:119
NULL
#define NULL
Definition: coverity.c:32
ff_get_cpu_flags_aarch64
int ff_get_cpu_flags_aarch64(void)
Definition: cpu.c:102
detect_flags
static int detect_flags(void)
Definition: cpu.c:95
AV_CPU_FLAG_I8MM
#define AV_CPU_FLAG_I8MM
Definition: cpu.h:73
cpu.h
AV_CPU_FLAG_NEON
#define AV_CPU_FLAG_NEON
Definition: cpu.h:69
size
int size
Definition: twinvq_data.h:10344
xf
#define xf(width, name, var, range_min, range_max, subs,...)
Definition: cbs_av1.c:590
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_ARMV8
#define AV_CPU_FLAG_ARMV8
Definition: cpu.h:70
flags
#define flags(name, subs,...)
Definition: cbs_av1.c:474