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){
54 int av_get_cpu_flags(void)
60 flags = ff_get_cpu_flags_aarch64();
62 flags = ff_get_cpu_flags_arm();
64 flags = ff_get_cpu_flags_ppc();
66 flags = ff_get_cpu_flags_x86();
72 void av_set_cpu_flags_mask(int mask)
75 flags = av_get_cpu_flags() & mask;
79 int av_parse_cpu_flags(const char *s)
81 #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
82 #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
83 #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
84 #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
85 #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
86 #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
87 #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
88 #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
89 #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
90 #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
91 #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
92 #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
93 #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
94 #define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
95 #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
96 #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
97 #define CPUFLAG_BMI1 (AV_CPU_FLAG_BMI1)
98 #define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | CPUFLAG_BMI1)
99 static const AVOption cpuflags_opts[] = {
100 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
102 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
104 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
105 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
106 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
107 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
108 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
109 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
110 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
111 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
112 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
113 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
114 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
115 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
116 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
117 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
118 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
119 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
120 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI1 }, .unit = "flags" },
121 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
122 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
123 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
124 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
126 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
127 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
128 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
129 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
130 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
131 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
133 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
134 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
138 static const AVClass class = {
139 .class_name = "cpuflags",
140 .item_name = av_default_item_name,
141 .option = cpuflags_opts,
142 .version = LIBAVUTIL_VERSION_INT,
146 const AVClass *pclass = &class;
148 if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
151 return flags & INT_MAX;
154 int av_parse_cpu_caps(unsigned *flags, const char *s)
156 static const AVOption cpuflags_opts[] = {
157 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
159 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
161 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
162 { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
163 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
164 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
165 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
166 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
167 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
168 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
169 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
170 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
171 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
172 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
173 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
174 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
175 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
176 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
177 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
178 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
179 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
180 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
181 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
182 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
184 #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
185 #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
186 #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
187 { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
188 { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
189 { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
191 #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
192 #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
193 #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
194 #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
195 { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
196 { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
197 { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
198 { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
199 { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
201 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
202 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
203 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
204 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
205 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
206 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
210 static const AVClass class = {
211 .class_name = "cpuflags",
212 .item_name = av_default_item_name,
213 .option = cpuflags_opts,
214 .version = LIBAVUTIL_VERSION_INT,
216 const AVClass *pclass = &class;
218 return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
221 int av_cpu_count(void)
223 static volatile int printed;
226 #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
231 if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
232 nb_cpus = CPU_COUNT(&cpuset);
233 #elif HAVE_GETPROCESSAFFINITYMASK
234 DWORD_PTR proc_aff, sys_aff;
235 if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
236 nb_cpus = av_popcount64(proc_aff);
237 #elif HAVE_SYSCTL && defined(HW_NCPU)
238 int mib[2] = { CTL_HW, HW_NCPU };
239 size_t len = sizeof(nb_cpus);
241 if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
243 #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
244 nb_cpus = sysconf(_SC_NPROC_ONLN);
245 #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
246 nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
250 av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
261 static const struct {
266 { AV_CPU_FLAG_NEON, "neon" },
267 { AV_CPU_FLAG_VFP, "vfp" },
269 { AV_CPU_FLAG_ARMV5TE, "armv5te" },
270 { AV_CPU_FLAG_ARMV6, "armv6" },
271 { AV_CPU_FLAG_ARMV6T2, "armv6t2" },
272 { AV_CPU_FLAG_VFP, "vfp" },
273 { AV_CPU_FLAG_VFPV3, "vfpv3" },
274 { AV_CPU_FLAG_NEON, "neon" },
276 { AV_CPU_FLAG_ALTIVEC, "altivec" },
278 { AV_CPU_FLAG_MMX, "mmx" },
279 { AV_CPU_FLAG_MMXEXT, "mmxext" },
280 { AV_CPU_FLAG_SSE, "sse" },
281 { AV_CPU_FLAG_SSE2, "sse2" },
282 { AV_CPU_FLAG_SSE2SLOW, "sse2(slow)" },
283 { AV_CPU_FLAG_SSE3, "sse3" },
284 { AV_CPU_FLAG_SSE3SLOW, "sse3(slow)" },
285 { AV_CPU_FLAG_SSSE3, "ssse3" },
286 { AV_CPU_FLAG_ATOM, "atom" },
287 { AV_CPU_FLAG_SSE4, "sse4.1" },
288 { AV_CPU_FLAG_SSE42, "sse4.2" },
289 { AV_CPU_FLAG_AVX, "avx" },
290 { AV_CPU_FLAG_XOP, "xop" },
291 { AV_CPU_FLAG_FMA3, "fma3" },
292 { AV_CPU_FLAG_FMA4, "fma4" },
293 { AV_CPU_FLAG_3DNOW, "3dnow" },
294 { AV_CPU_FLAG_3DNOWEXT, "3dnowext" },
295 { AV_CPU_FLAG_CMOV, "cmov" },
296 { AV_CPU_FLAG_AVX2, "avx2" },
297 { AV_CPU_FLAG_BMI1, "bmi1" },
298 { AV_CPU_FLAG_BMI2, "bmi2" },
305 int cpu_flags = av_get_cpu_flags();
308 printf("cpu_flags = 0x%08X\n", cpu_flags);
309 printf("cpu_flags =");
310 for (i = 0; cpu_flag_tab[i].flag; i++)
311 if (cpu_flags & cpu_flag_tab[i].flag)
312 printf(" %s", cpu_flag_tab[i].name);