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_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
95 #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
96 static const AVOption cpuflags_opts[] = {
97 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
99 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
101 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
102 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
103 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
104 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
105 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
106 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
107 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
108 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
109 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
110 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
111 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
112 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
113 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
114 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
115 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
116 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
117 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
118 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
120 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
121 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
122 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
123 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
124 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
125 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
127 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
128 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
132 static const AVClass class = {
133 .class_name = "cpuflags",
134 .item_name = av_default_item_name,
135 .option = cpuflags_opts,
136 .version = LIBAVUTIL_VERSION_INT,
140 const AVClass *pclass = &class;
142 if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
145 return flags & INT_MAX;
148 int av_parse_cpu_caps(unsigned *flags, const char *s)
150 static const AVOption cpuflags_opts[] = {
151 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
153 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
155 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
156 { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
157 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
158 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
159 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
160 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
161 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
162 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
163 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
164 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
165 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
166 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
167 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
168 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
169 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
170 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
171 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
172 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
174 #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
175 #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
176 #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
177 { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
178 { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
179 { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
181 #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
182 #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
183 #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
184 #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
185 { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
186 { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
187 { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
188 { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
189 { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
191 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
192 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
193 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
194 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
195 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
196 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
200 static const AVClass class = {
201 .class_name = "cpuflags",
202 .item_name = av_default_item_name,
203 .option = cpuflags_opts,
204 .version = LIBAVUTIL_VERSION_INT,
206 const AVClass *pclass = &class;
208 return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
211 int av_cpu_count(void)
213 static volatile int printed;
216 #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
221 if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
222 nb_cpus = CPU_COUNT(&cpuset);
223 #elif HAVE_GETPROCESSAFFINITYMASK
224 DWORD_PTR proc_aff, sys_aff;
225 if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
226 nb_cpus = av_popcount64(proc_aff);
227 #elif HAVE_SYSCTL && defined(HW_NCPU)
228 int mib[2] = { CTL_HW, HW_NCPU };
229 size_t len = sizeof(nb_cpus);
231 if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
233 #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
234 nb_cpus = sysconf(_SC_NPROC_ONLN);
235 #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
236 nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
240 av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
251 static const struct {
256 { AV_CPU_FLAG_NEON, "neon" },
257 { AV_CPU_FLAG_VFP, "vfp" },
259 { AV_CPU_FLAG_ARMV5TE, "armv5te" },
260 { AV_CPU_FLAG_ARMV6, "armv6" },
261 { AV_CPU_FLAG_ARMV6T2, "armv6t2" },
262 { AV_CPU_FLAG_VFP, "vfp" },
263 { AV_CPU_FLAG_VFPV3, "vfpv3" },
264 { AV_CPU_FLAG_NEON, "neon" },
266 { AV_CPU_FLAG_ALTIVEC, "altivec" },
268 { AV_CPU_FLAG_MMX, "mmx" },
269 { AV_CPU_FLAG_MMXEXT, "mmxext" },
270 { AV_CPU_FLAG_SSE, "sse" },
271 { AV_CPU_FLAG_SSE2, "sse2" },
272 { AV_CPU_FLAG_SSE2SLOW, "sse2(slow)" },
273 { AV_CPU_FLAG_SSE3, "sse3" },
274 { AV_CPU_FLAG_SSE3SLOW, "sse3(slow)" },
275 { AV_CPU_FLAG_SSSE3, "ssse3" },
276 { AV_CPU_FLAG_ATOM, "atom" },
277 { AV_CPU_FLAG_SSE4, "sse4.1" },
278 { AV_CPU_FLAG_SSE42, "sse4.2" },
279 { AV_CPU_FLAG_AVX, "avx" },
280 { AV_CPU_FLAG_XOP, "xop" },
281 { AV_CPU_FLAG_FMA4, "fma4" },
282 { AV_CPU_FLAG_3DNOW, "3dnow" },
283 { AV_CPU_FLAG_3DNOWEXT, "3dnowext" },
284 { AV_CPU_FLAG_CMOV, "cmov" },
285 { AV_CPU_FLAG_AVX2, "avx2" },
292 int cpu_flags = av_get_cpu_flags();
295 printf("cpu_flags = 0x%08X\n", cpu_flags);
296 printf("cpu_flags =");
297 for (i = 0; cpu_flag_tab[i].flag; i++)
298 if (cpu_flags & cpu_flag_tab[i].flag)
299 printf(" %s", cpu_flag_tab[i].name);