summaryrefslogtreecommitdiff
path: root/libavutil/cpu.c (plain)
blob: 16e0c9278fd04b1446f4bc2ae0ade951cf38e116
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 <stdint.h>
20#include <stdatomic.h>
21
22#include "cpu.h"
23#include "cpu_internal.h"
24#include "config.h"
25#include "opt.h"
26#include "common.h"
27
28#if HAVE_SCHED_GETAFFINITY
29#ifndef _GNU_SOURCE
30# define _GNU_SOURCE
31#endif
32#include <sched.h>
33#endif
34#if HAVE_GETPROCESSAFFINITYMASK || HAVE_WINRT
35#include <windows.h>
36#endif
37#if HAVE_SYSCTL
38#if HAVE_SYS_PARAM_H
39#include <sys/param.h>
40#endif
41#include <sys/types.h>
42#include <sys/sysctl.h>
43#endif
44#if HAVE_UNISTD_H
45#include <unistd.h>
46#endif
47
48static atomic_int cpu_flags = ATOMIC_VAR_INIT(-1);
49
50static int get_cpu_flags(void)
51{
52 if (ARCH_AARCH64)
53 return ff_get_cpu_flags_aarch64();
54 if (ARCH_ARM)
55 return ff_get_cpu_flags_arm();
56 if (ARCH_PPC)
57 return ff_get_cpu_flags_ppc();
58 if (ARCH_X86)
59 return ff_get_cpu_flags_x86();
60 return 0;
61}
62
63void av_force_cpu_flags(int arg){
64 if ( (arg & ( AV_CPU_FLAG_3DNOW |
65 AV_CPU_FLAG_3DNOWEXT |
66 AV_CPU_FLAG_MMXEXT |
67 AV_CPU_FLAG_SSE |
68 AV_CPU_FLAG_SSE2 |
69 AV_CPU_FLAG_SSE2SLOW |
70 AV_CPU_FLAG_SSE3 |
71 AV_CPU_FLAG_SSE3SLOW |
72 AV_CPU_FLAG_SSSE3 |
73 AV_CPU_FLAG_SSE4 |
74 AV_CPU_FLAG_SSE42 |
75 AV_CPU_FLAG_AVX |
76 AV_CPU_FLAG_AVXSLOW |
77 AV_CPU_FLAG_XOP |
78 AV_CPU_FLAG_FMA3 |
79 AV_CPU_FLAG_FMA4 |
80 AV_CPU_FLAG_AVX2 ))
81 && !(arg & AV_CPU_FLAG_MMX)) {
82 av_log(NULL, AV_LOG_WARNING, "MMX implied by specified flags\n");
83 arg |= AV_CPU_FLAG_MMX;
84 }
85
86 atomic_store_explicit(&cpu_flags, arg, memory_order_relaxed);
87}
88
89int av_get_cpu_flags(void)
90{
91 int flags = atomic_load_explicit(&cpu_flags, memory_order_relaxed);
92 if (flags == -1) {
93 flags = get_cpu_flags();
94 atomic_store_explicit(&cpu_flags, flags, memory_order_relaxed);
95 }
96 return flags;
97}
98
99void av_set_cpu_flags_mask(int mask)
100{
101 atomic_store_explicit(&cpu_flags, get_cpu_flags() & mask,
102 memory_order_relaxed);
103}
104
105int av_parse_cpu_flags(const char *s)
106{
107#define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
108#define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
109#define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
110#define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
111#define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
112#define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
113#define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
114#define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
115#define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
116#define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
117#define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
118#define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
119#define CPUFLAG_AVXSLOW (AV_CPU_FLAG_AVXSLOW | CPUFLAG_AVX)
120#define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
121#define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
122#define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
123#define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
124#define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | AV_CPU_FLAG_BMI1)
125#define CPUFLAG_AESNI (AV_CPU_FLAG_AESNI | CPUFLAG_SSE42)
126 static const AVOption cpuflags_opts[] = {
127 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
128#if ARCH_PPC
129 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
130#elif ARCH_X86
131 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
132 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
133 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
134 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
135 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
136 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
137 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
138 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
139 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
140 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
141 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
142 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
143 { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVXSLOW }, .unit = "flags" },
144 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
145 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
146 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
147 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
148 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
149 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
150 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
151 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
152 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
153 { "aesni" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AESNI }, .unit = "flags" },
154#elif ARCH_ARM
155 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
156 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
157 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
158 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
159 { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
160 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
161 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
162#elif ARCH_AARCH64
163 { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
164 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
165 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
166#endif
167 { NULL },
168 };
169 static const AVClass class = {
170 .class_name = "cpuflags",
171 .item_name = av_default_item_name,
172 .option = cpuflags_opts,
173 .version = LIBAVUTIL_VERSION_INT,
174 };
175
176 int flags = 0, ret;
177 const AVClass *pclass = &class;
178
179 if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
180 return ret;
181
182 return flags & INT_MAX;
183}
184
185int av_parse_cpu_caps(unsigned *flags, const char *s)
186{
187 static const AVOption cpuflags_opts[] = {
188 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
189#if ARCH_PPC
190 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
191#elif ARCH_X86
192 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
193 { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
194 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
195 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
196 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
197 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
198 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
199 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
200 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
201 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
202 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
203 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
204 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
205 { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVXSLOW }, .unit = "flags" },
206 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
207 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
208 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
209 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
210 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
211 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
212 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
213 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
214 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
215 { "aesni", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AESNI }, .unit = "flags" },
216
217#define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
218#define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
219#define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
220 { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
221 { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
222 { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
223
224#define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
225#define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
226#define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
227#define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
228 { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
229 { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
230 { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
231 { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
232 { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
233#elif ARCH_ARM
234 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
235 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
236 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
237 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
238 { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
239 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
240 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
241 { "setend", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SETEND }, .unit = "flags" },
242#elif ARCH_AARCH64
243 { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
244 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
245 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
246#endif
247 { NULL },
248 };
249 static const AVClass class = {
250 .class_name = "cpuflags",
251 .item_name = av_default_item_name,
252 .option = cpuflags_opts,
253 .version = LIBAVUTIL_VERSION_INT,
254 };
255 const AVClass *pclass = &class;
256
257 return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
258}
259
260int av_cpu_count(void)
261{
262 static volatile int printed;
263
264 int nb_cpus = 1;
265#if HAVE_WINRT
266 SYSTEM_INFO sysinfo;
267#endif
268#if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
269 cpu_set_t cpuset;
270
271 CPU_ZERO(&cpuset);
272
273 if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
274 nb_cpus = CPU_COUNT(&cpuset);
275#elif HAVE_GETPROCESSAFFINITYMASK
276 DWORD_PTR proc_aff, sys_aff;
277 if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
278 nb_cpus = av_popcount64(proc_aff);
279#elif HAVE_SYSCTL && defined(HW_NCPU)
280 int mib[2] = { CTL_HW, HW_NCPU };
281 size_t len = sizeof(nb_cpus);
282
283 if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
284 nb_cpus = 0;
285#elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
286 nb_cpus = sysconf(_SC_NPROC_ONLN);
287#elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
288 nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
289#elif HAVE_WINRT
290 GetNativeSystemInfo(&sysinfo);
291 nb_cpus = sysinfo.dwNumberOfProcessors;
292#endif
293
294 if (!printed) {
295 av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
296 printed = 1;
297 }
298
299 return nb_cpus;
300}
301