cpu.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296
  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. #include <stdint.h>
  19. #include "cpu.h"
  20. #include "cpu_internal.h"
  21. #include "config.h"
  22. #include "opt.h"
  23. #include "common.h"
  24. #if HAVE_SCHED_GETAFFINITY
  25. #ifndef _GNU_SOURCE
  26. # define _GNU_SOURCE
  27. #endif
  28. #include <sched.h>
  29. #endif
  30. #if HAVE_GETPROCESSAFFINITYMASK || HAVE_WINRT
  31. #include <windows.h>
  32. #endif
  33. #if HAVE_SYSCTL
  34. #if HAVE_SYS_PARAM_H
  35. #include <sys/param.h>
  36. #endif
  37. #include <sys/types.h>
  38. #include <sys/sysctl.h>
  39. #endif
  40. #if HAVE_UNISTD_H
  41. #include <unistd.h>
  42. #endif
  43. static int flags, checked;
  44. void av_force_cpu_flags(int arg){
  45. if ( (arg & ( AV_CPU_FLAG_3DNOW |
  46. AV_CPU_FLAG_3DNOWEXT |
  47. AV_CPU_FLAG_MMXEXT |
  48. AV_CPU_FLAG_SSE |
  49. AV_CPU_FLAG_SSE2 |
  50. AV_CPU_FLAG_SSE2SLOW |
  51. AV_CPU_FLAG_SSE3 |
  52. AV_CPU_FLAG_SSE3SLOW |
  53. AV_CPU_FLAG_SSSE3 |
  54. AV_CPU_FLAG_SSE4 |
  55. AV_CPU_FLAG_SSE42 |
  56. AV_CPU_FLAG_AVX |
  57. AV_CPU_FLAG_AVXSLOW |
  58. AV_CPU_FLAG_XOP |
  59. AV_CPU_FLAG_FMA3 |
  60. AV_CPU_FLAG_FMA4 |
  61. AV_CPU_FLAG_AVX2 ))
  62. && !(arg & AV_CPU_FLAG_MMX)) {
  63. av_log(NULL, AV_LOG_WARNING, "MMX implied by specified flags\n");
  64. arg |= AV_CPU_FLAG_MMX;
  65. }
  66. flags = arg;
  67. checked = arg != -1;
  68. }
  69. int av_get_cpu_flags(void)
  70. {
  71. if (checked)
  72. return flags;
  73. if (ARCH_AARCH64)
  74. flags = ff_get_cpu_flags_aarch64();
  75. if (ARCH_ARM)
  76. flags = ff_get_cpu_flags_arm();
  77. if (ARCH_PPC)
  78. flags = ff_get_cpu_flags_ppc();
  79. if (ARCH_X86)
  80. flags = ff_get_cpu_flags_x86();
  81. checked = 1;
  82. return flags;
  83. }
  84. void av_set_cpu_flags_mask(int mask)
  85. {
  86. checked = 0;
  87. flags = av_get_cpu_flags() & mask;
  88. checked = 1;
  89. }
  90. int av_parse_cpu_flags(const char *s)
  91. {
  92. #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
  93. #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
  94. #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
  95. #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
  96. #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
  97. #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
  98. #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
  99. #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
  100. #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
  101. #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
  102. #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
  103. #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
  104. #define CPUFLAG_AVXSLOW (AV_CPU_FLAG_AVXSLOW | CPUFLAG_AVX)
  105. #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
  106. #define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
  107. #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
  108. #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
  109. #define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | AV_CPU_FLAG_BMI1)
  110. #define CPUFLAG_AESNI (AV_CPU_FLAG_AESNI | CPUFLAG_SSE42)
  111. static const AVOption cpuflags_opts[] = {
  112. { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
  113. #if ARCH_PPC
  114. { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
  115. #elif ARCH_X86
  116. { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  117. { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
  118. { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
  119. { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
  120. { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
  121. { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
  122. { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
  123. { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
  124. { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
  125. { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
  126. { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
  127. { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
  128. { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVXSLOW }, .unit = "flags" },
  129. { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
  130. { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
  131. { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
  132. { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
  133. { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
  134. { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
  135. { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
  136. { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
  137. { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
  138. { "aesni" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AESNI }, .unit = "flags" },
  139. #elif ARCH_ARM
  140. { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
  141. { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
  142. { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
  143. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  144. { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
  145. { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
  146. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  147. #elif ARCH_AARCH64
  148. { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
  149. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  150. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  151. #endif
  152. { NULL },
  153. };
  154. static const AVClass class = {
  155. .class_name = "cpuflags",
  156. .item_name = av_default_item_name,
  157. .option = cpuflags_opts,
  158. .version = LIBAVUTIL_VERSION_INT,
  159. };
  160. int flags = 0, ret;
  161. const AVClass *pclass = &class;
  162. if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
  163. return ret;
  164. return flags & INT_MAX;
  165. }
  166. int av_parse_cpu_caps(unsigned *flags, const char *s)
  167. {
  168. static const AVOption cpuflags_opts[] = {
  169. { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
  170. #if ARCH_PPC
  171. { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
  172. #elif ARCH_X86
  173. { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  174. { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
  175. { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
  176. { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
  177. { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
  178. { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
  179. { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
  180. { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
  181. { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
  182. { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
  183. { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
  184. { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
  185. { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
  186. { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVXSLOW }, .unit = "flags" },
  187. { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
  188. { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
  189. { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
  190. { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
  191. { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
  192. { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
  193. { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
  194. { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
  195. { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
  196. { "aesni", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AESNI }, .unit = "flags" },
  197. #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
  198. #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
  199. #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
  200. { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
  201. { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
  202. { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
  203. #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
  204. #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
  205. #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
  206. #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
  207. { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  208. { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
  209. { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
  210. { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
  211. { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
  212. #elif ARCH_ARM
  213. { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
  214. { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
  215. { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
  216. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  217. { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
  218. { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
  219. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  220. { "setend", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SETEND }, .unit = "flags" },
  221. #elif ARCH_AARCH64
  222. { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
  223. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  224. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  225. #endif
  226. { NULL },
  227. };
  228. static const AVClass class = {
  229. .class_name = "cpuflags",
  230. .item_name = av_default_item_name,
  231. .option = cpuflags_opts,
  232. .version = LIBAVUTIL_VERSION_INT,
  233. };
  234. const AVClass *pclass = &class;
  235. return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
  236. }
  237. int av_cpu_count(void)
  238. {
  239. static volatile int printed;
  240. int nb_cpus = 1;
  241. #if HAVE_WINRT
  242. SYSTEM_INFO sysinfo;
  243. #endif
  244. #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
  245. cpu_set_t cpuset;
  246. CPU_ZERO(&cpuset);
  247. if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
  248. nb_cpus = CPU_COUNT(&cpuset);
  249. #elif HAVE_GETPROCESSAFFINITYMASK
  250. DWORD_PTR proc_aff, sys_aff;
  251. if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
  252. nb_cpus = av_popcount64(proc_aff);
  253. #elif HAVE_SYSCTL && defined(HW_NCPU)
  254. int mib[2] = { CTL_HW, HW_NCPU };
  255. size_t len = sizeof(nb_cpus);
  256. if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
  257. nb_cpus = 0;
  258. #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
  259. nb_cpus = sysconf(_SC_NPROC_ONLN);
  260. #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
  261. nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
  262. #elif HAVE_WINRT
  263. GetNativeSystemInfo(&sysinfo);
  264. nb_cpus = sysinfo.dwNumberOfProcessors;
  265. #endif
  266. if (!printed) {
  267. av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
  268. printed = 1;
  269. }
  270. return nb_cpus;
  271. }