2 * This file is part of FFmpeg.
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.
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.
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
22 #include "cpu_internal.h"
27 #if HAVE_SCHED_GETAFFINITY
33 #if HAVE_GETPROCESSAFFINITYMASK
38 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/sysctl.h>
47 static int flags, checked;
49 void av_force_cpu_flags(int arg){
50 if ( (arg & ( AV_CPU_FLAG_3DNOW |
51 AV_CPU_FLAG_3DNOWEXT |
55 AV_CPU_FLAG_SSE2SLOW |
57 AV_CPU_FLAG_SSE3SLOW |
66 && !(arg & AV_CPU_FLAG_MMX)) {
67 av_log(NULL, AV_LOG_WARNING, "MMX implied by specified flags\n");
68 arg |= AV_CPU_FLAG_MMX;
75 int av_get_cpu_flags(void)
81 flags = ff_get_cpu_flags_aarch64();
83 flags = ff_get_cpu_flags_arm();
85 flags = ff_get_cpu_flags_ppc();
87 flags = ff_get_cpu_flags_x86();
93 void av_set_cpu_flags_mask(int mask)
96 flags = av_get_cpu_flags() & mask;
100 int av_parse_cpu_flags(const char *s)
102 #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
103 #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
104 #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
105 #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
106 #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
107 #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
108 #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
109 #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
110 #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
111 #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
112 #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
113 #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
114 #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
115 #define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
116 #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
117 #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
118 #define CPUFLAG_BMI1 (AV_CPU_FLAG_BMI1)
119 #define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | CPUFLAG_BMI1)
120 static const AVOption cpuflags_opts[] = {
121 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
123 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
125 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
126 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
127 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
128 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
129 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
130 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
131 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
132 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
133 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
134 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
135 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
136 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
137 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
138 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
139 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
140 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
141 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI1 }, .unit = "flags" },
142 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
143 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
144 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
145 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
147 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
148 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
149 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
150 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
151 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
152 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
154 { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
155 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
156 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
160 static const AVClass class = {
161 .class_name = "cpuflags",
162 .item_name = av_default_item_name,
163 .option = cpuflags_opts,
164 .version = LIBAVUTIL_VERSION_INT,
168 const AVClass *pclass = &class;
170 if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
173 return flags & INT_MAX;
176 int av_parse_cpu_caps(unsigned *flags, const char *s)
178 static const AVOption cpuflags_opts[] = {
179 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
181 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
183 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
184 { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
185 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
186 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
187 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
188 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
189 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
190 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
191 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
192 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
193 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
194 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
195 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
196 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
197 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
198 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
199 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
200 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
201 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
202 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
203 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
204 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
206 #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
207 #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
208 #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
209 { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
210 { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
211 { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
213 #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
214 #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
215 #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
216 #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
217 { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
218 { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
219 { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
220 { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
221 { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
223 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
224 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
225 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
226 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
227 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
228 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
229 { "setend", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SETEND }, .unit = "flags" },
231 { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
232 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
233 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
237 static const AVClass class = {
238 .class_name = "cpuflags",
239 .item_name = av_default_item_name,
240 .option = cpuflags_opts,
241 .version = LIBAVUTIL_VERSION_INT,
243 const AVClass *pclass = &class;
245 return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
248 int av_cpu_count(void)
250 static volatile int printed;
253 #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
258 if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
259 nb_cpus = CPU_COUNT(&cpuset);
260 #elif HAVE_GETPROCESSAFFINITYMASK
261 DWORD_PTR proc_aff, sys_aff;
262 if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
263 nb_cpus = av_popcount64(proc_aff);
264 #elif HAVE_SYSCTL && defined(HW_NCPU)
265 int mib[2] = { CTL_HW, HW_NCPU };
266 size_t len = sizeof(nb_cpus);
268 if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
270 #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
271 nb_cpus = sysconf(_SC_NPROC_ONLN);
272 #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
273 nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
277 av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
287 #include "avstring.h"
290 #include "compat/getopt.c"
293 static const struct {
298 { AV_CPU_FLAG_ARMV8, "armv8" },
299 { AV_CPU_FLAG_NEON, "neon" },
300 { AV_CPU_FLAG_VFP, "vfp" },
302 { AV_CPU_FLAG_ARMV5TE, "armv5te" },
303 { AV_CPU_FLAG_ARMV6, "armv6" },
304 { AV_CPU_FLAG_ARMV6T2, "armv6t2" },
305 { AV_CPU_FLAG_VFP, "vfp" },
306 { AV_CPU_FLAG_VFPV3, "vfpv3" },
307 { AV_CPU_FLAG_NEON, "neon" },
308 { AV_CPU_FLAG_SETEND, "setend" },
310 { AV_CPU_FLAG_ALTIVEC, "altivec" },
312 { AV_CPU_FLAG_MMX, "mmx" },
313 { AV_CPU_FLAG_MMXEXT, "mmxext" },
314 { AV_CPU_FLAG_SSE, "sse" },
315 { AV_CPU_FLAG_SSE2, "sse2" },
316 { AV_CPU_FLAG_SSE2SLOW, "sse2slow" },
317 { AV_CPU_FLAG_SSE3, "sse3" },
318 { AV_CPU_FLAG_SSE3SLOW, "sse3slow" },
319 { AV_CPU_FLAG_SSSE3, "ssse3" },
320 { AV_CPU_FLAG_ATOM, "atom" },
321 { AV_CPU_FLAG_SSE4, "sse4.1" },
322 { AV_CPU_FLAG_SSE42, "sse4.2" },
323 { AV_CPU_FLAG_AVX, "avx" },
324 { AV_CPU_FLAG_XOP, "xop" },
325 { AV_CPU_FLAG_FMA3, "fma3" },
326 { AV_CPU_FLAG_FMA4, "fma4" },
327 { AV_CPU_FLAG_3DNOW, "3dnow" },
328 { AV_CPU_FLAG_3DNOWEXT, "3dnowext" },
329 { AV_CPU_FLAG_CMOV, "cmov" },
330 { AV_CPU_FLAG_AVX2, "avx2" },
331 { AV_CPU_FLAG_BMI1, "bmi1" },
332 { AV_CPU_FLAG_BMI2, "bmi2" },
337 static void print_cpu_flags(int cpu_flags, const char *type)
341 printf("cpu_flags(%s) = 0x%08X\n", type, cpu_flags);
342 printf("cpu_flags_str(%s) =", type);
343 for (i = 0; cpu_flag_tab[i].flag; i++)
344 if (cpu_flags & cpu_flag_tab[i].flag)
345 printf(" %s", cpu_flag_tab[i].name);
350 int main(int argc, char **argv)
352 int cpu_flags_raw = av_get_cpu_flags();
354 int cpu_count = av_cpu_count();
355 char threads[5] = "auto";
358 for(i = 0; cpu_flag_tab[i].flag; i++) {
360 if (av_parse_cpu_caps(&tmp, cpu_flag_tab[i].name) < 0) {
361 fprintf(stderr, "Table missing %s\n", cpu_flag_tab[i].name);
366 if (cpu_flags_raw < 0)
370 int c = getopt(argc, argv, "c:t:");
376 unsigned flags = av_get_cpu_flags();
377 if (av_parse_cpu_caps(&flags, optarg) < 0)
380 av_force_cpu_flags(flags);
385 int len = av_strlcpy(threads, optarg, sizeof(threads));
386 if (len >= sizeof(threads)) {
387 fprintf(stderr, "Invalid thread count '%s'\n", optarg);
394 cpu_flags_eff = av_get_cpu_flags();
396 if (cpu_flags_eff < 0)
399 print_cpu_flags(cpu_flags_raw, "raw");
400 print_cpu_flags(cpu_flags_eff, "effective");
401 printf("threads = %s (cpu_count = %d)\n", threads, cpu_count);