You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

319 lines
14KB

  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
  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_SYSCONF
  41. #include <unistd.h>
  42. #endif
  43. static int flags, checked;
  44. void av_force_cpu_flags(int arg){
  45. flags = arg;
  46. checked = arg != -1;
  47. }
  48. int av_get_cpu_flags(void)
  49. {
  50. if (checked)
  51. return flags;
  52. if (ARCH_AARCH64)
  53. flags = ff_get_cpu_flags_aarch64();
  54. if (ARCH_ARM)
  55. flags = ff_get_cpu_flags_arm();
  56. if (ARCH_PPC)
  57. flags = ff_get_cpu_flags_ppc();
  58. if (ARCH_X86)
  59. flags = ff_get_cpu_flags_x86();
  60. checked = 1;
  61. return flags;
  62. }
  63. void av_set_cpu_flags_mask(int mask)
  64. {
  65. checked = 0;
  66. flags = av_get_cpu_flags() & mask;
  67. checked = 1;
  68. }
  69. int av_parse_cpu_flags(const char *s)
  70. {
  71. #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
  72. #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
  73. #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
  74. #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
  75. #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
  76. #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
  77. #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
  78. #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
  79. #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
  80. #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
  81. #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
  82. #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
  83. #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
  84. #define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
  85. #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
  86. #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
  87. #define CPUFLAG_BMI1 (AV_CPU_FLAG_BMI1)
  88. #define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | CPUFLAG_BMI1)
  89. static const AVOption cpuflags_opts[] = {
  90. { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
  91. #if ARCH_PPC
  92. { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
  93. #elif ARCH_X86
  94. { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  95. { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
  96. { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
  97. { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
  98. { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
  99. { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
  100. { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
  101. { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
  102. { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
  103. { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
  104. { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
  105. { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
  106. { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
  107. { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
  108. { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
  109. { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
  110. { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI1 }, .unit = "flags" },
  111. { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
  112. { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
  113. { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
  114. { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
  115. #elif ARCH_ARM
  116. { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
  117. { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
  118. { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
  119. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  120. { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
  121. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  122. #elif ARCH_AARCH64
  123. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  124. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  125. #endif
  126. { NULL },
  127. };
  128. static const AVClass class = {
  129. .class_name = "cpuflags",
  130. .item_name = av_default_item_name,
  131. .option = cpuflags_opts,
  132. .version = LIBAVUTIL_VERSION_INT,
  133. };
  134. int flags = 0, ret;
  135. const AVClass *pclass = &class;
  136. if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
  137. return ret;
  138. return flags & INT_MAX;
  139. }
  140. int av_parse_cpu_caps(unsigned *flags, const char *s)
  141. {
  142. static const AVOption cpuflags_opts[] = {
  143. { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
  144. #if ARCH_PPC
  145. { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
  146. #elif ARCH_X86
  147. { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  148. { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
  149. { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
  150. { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
  151. { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
  152. { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
  153. { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
  154. { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
  155. { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
  156. { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
  157. { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
  158. { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
  159. { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
  160. { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
  161. { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
  162. { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
  163. { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
  164. { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
  165. { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
  166. { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
  167. { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
  168. { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
  169. #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
  170. #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
  171. #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
  172. { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
  173. { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
  174. { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
  175. #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
  176. #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
  177. #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
  178. #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
  179. { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  180. { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
  181. { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
  182. { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
  183. { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
  184. #elif ARCH_ARM
  185. { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
  186. { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
  187. { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
  188. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  189. { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
  190. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  191. #endif
  192. { NULL },
  193. };
  194. static const AVClass class = {
  195. .class_name = "cpuflags",
  196. .item_name = av_default_item_name,
  197. .option = cpuflags_opts,
  198. .version = LIBAVUTIL_VERSION_INT,
  199. };
  200. const AVClass *pclass = &class;
  201. return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
  202. }
  203. int av_cpu_count(void)
  204. {
  205. static volatile int printed;
  206. int nb_cpus = 1;
  207. #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
  208. cpu_set_t cpuset;
  209. CPU_ZERO(&cpuset);
  210. if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
  211. nb_cpus = CPU_COUNT(&cpuset);
  212. #elif HAVE_GETPROCESSAFFINITYMASK
  213. DWORD_PTR proc_aff, sys_aff;
  214. if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
  215. nb_cpus = av_popcount64(proc_aff);
  216. #elif HAVE_SYSCTL && defined(HW_NCPU)
  217. int mib[2] = { CTL_HW, HW_NCPU };
  218. size_t len = sizeof(nb_cpus);
  219. if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
  220. nb_cpus = 0;
  221. #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
  222. nb_cpus = sysconf(_SC_NPROC_ONLN);
  223. #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
  224. nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
  225. #endif
  226. if (!printed) {
  227. av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
  228. printed = 1;
  229. }
  230. return nb_cpus;
  231. }
  232. #ifdef TEST
  233. #include <stdio.h>
  234. static const struct {
  235. int flag;
  236. const char *name;
  237. } cpu_flag_tab[] = {
  238. #if ARCH_AARCH64
  239. { AV_CPU_FLAG_NEON, "neon" },
  240. { AV_CPU_FLAG_VFP, "vfp" },
  241. #elif ARCH_ARM
  242. { AV_CPU_FLAG_ARMV5TE, "armv5te" },
  243. { AV_CPU_FLAG_ARMV6, "armv6" },
  244. { AV_CPU_FLAG_ARMV6T2, "armv6t2" },
  245. { AV_CPU_FLAG_VFP, "vfp" },
  246. { AV_CPU_FLAG_VFPV3, "vfpv3" },
  247. { AV_CPU_FLAG_NEON, "neon" },
  248. #elif ARCH_PPC
  249. { AV_CPU_FLAG_ALTIVEC, "altivec" },
  250. #elif ARCH_X86
  251. { AV_CPU_FLAG_MMX, "mmx" },
  252. { AV_CPU_FLAG_MMXEXT, "mmxext" },
  253. { AV_CPU_FLAG_SSE, "sse" },
  254. { AV_CPU_FLAG_SSE2, "sse2" },
  255. { AV_CPU_FLAG_SSE2SLOW, "sse2(slow)" },
  256. { AV_CPU_FLAG_SSE3, "sse3" },
  257. { AV_CPU_FLAG_SSE3SLOW, "sse3(slow)" },
  258. { AV_CPU_FLAG_SSSE3, "ssse3" },
  259. { AV_CPU_FLAG_ATOM, "atom" },
  260. { AV_CPU_FLAG_SSE4, "sse4.1" },
  261. { AV_CPU_FLAG_SSE42, "sse4.2" },
  262. { AV_CPU_FLAG_AVX, "avx" },
  263. { AV_CPU_FLAG_XOP, "xop" },
  264. { AV_CPU_FLAG_FMA3, "fma3" },
  265. { AV_CPU_FLAG_FMA4, "fma4" },
  266. { AV_CPU_FLAG_3DNOW, "3dnow" },
  267. { AV_CPU_FLAG_3DNOWEXT, "3dnowext" },
  268. { AV_CPU_FLAG_CMOV, "cmov" },
  269. { AV_CPU_FLAG_AVX2, "avx2" },
  270. { AV_CPU_FLAG_BMI1, "bmi1" },
  271. { AV_CPU_FLAG_BMI2, "bmi2" },
  272. #endif
  273. { 0 }
  274. };
  275. int main(void)
  276. {
  277. int cpu_flags = av_get_cpu_flags();
  278. int i;
  279. printf("cpu_flags = 0x%08X\n", cpu_flags);
  280. printf("cpu_flags =");
  281. for (i = 0; cpu_flag_tab[i].flag; i++)
  282. if (cpu_flags & cpu_flag_tab[i].flag)
  283. printf(" %s", cpu_flag_tab[i].name);
  284. printf("\n");
  285. return 0;
  286. }
  287. #endif