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.

302 lines
15KB

  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 <stdatomic.h>
  20. #include "cpu.h"
  21. #include "cpu_internal.h"
  22. #include "config.h"
  23. #include "opt.h"
  24. #include "common.h"
  25. #if HAVE_SCHED_GETAFFINITY
  26. #ifndef _GNU_SOURCE
  27. # define _GNU_SOURCE
  28. #endif
  29. #include <sched.h>
  30. #endif
  31. #if HAVE_GETPROCESSAFFINITYMASK || HAVE_WINRT
  32. #include <windows.h>
  33. #endif
  34. #if HAVE_SYSCTL
  35. #if HAVE_SYS_PARAM_H
  36. #include <sys/param.h>
  37. #endif
  38. #include <sys/types.h>
  39. #include <sys/sysctl.h>
  40. #endif
  41. #if HAVE_UNISTD_H
  42. #include <unistd.h>
  43. #endif
  44. static atomic_int cpu_flags = ATOMIC_VAR_INIT(-1);
  45. static int get_cpu_flags(void)
  46. {
  47. if (ARCH_AARCH64)
  48. return ff_get_cpu_flags_aarch64();
  49. if (ARCH_ARM)
  50. return ff_get_cpu_flags_arm();
  51. if (ARCH_PPC)
  52. return ff_get_cpu_flags_ppc();
  53. if (ARCH_X86)
  54. return ff_get_cpu_flags_x86();
  55. return 0;
  56. }
  57. void av_force_cpu_flags(int arg){
  58. if (ARCH_X86 &&
  59. (arg & ( AV_CPU_FLAG_3DNOW |
  60. AV_CPU_FLAG_3DNOWEXT |
  61. AV_CPU_FLAG_MMXEXT |
  62. AV_CPU_FLAG_SSE |
  63. AV_CPU_FLAG_SSE2 |
  64. AV_CPU_FLAG_SSE2SLOW |
  65. AV_CPU_FLAG_SSE3 |
  66. AV_CPU_FLAG_SSE3SLOW |
  67. AV_CPU_FLAG_SSSE3 |
  68. AV_CPU_FLAG_SSE4 |
  69. AV_CPU_FLAG_SSE42 |
  70. AV_CPU_FLAG_AVX |
  71. AV_CPU_FLAG_AVXSLOW |
  72. AV_CPU_FLAG_XOP |
  73. AV_CPU_FLAG_FMA3 |
  74. AV_CPU_FLAG_FMA4 |
  75. AV_CPU_FLAG_AVX2 ))
  76. && !(arg & AV_CPU_FLAG_MMX)) {
  77. av_log(NULL, AV_LOG_WARNING, "MMX implied by specified flags\n");
  78. arg |= AV_CPU_FLAG_MMX;
  79. }
  80. atomic_store_explicit(&cpu_flags, arg, memory_order_relaxed);
  81. }
  82. int av_get_cpu_flags(void)
  83. {
  84. int flags = atomic_load_explicit(&cpu_flags, memory_order_relaxed);
  85. if (flags == -1) {
  86. flags = get_cpu_flags();
  87. atomic_store_explicit(&cpu_flags, flags, memory_order_relaxed);
  88. }
  89. return flags;
  90. }
  91. void av_set_cpu_flags_mask(int mask)
  92. {
  93. atomic_store_explicit(&cpu_flags, get_cpu_flags() & mask,
  94. memory_order_relaxed);
  95. }
  96. int av_parse_cpu_flags(const char *s)
  97. {
  98. #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
  99. #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
  100. #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
  101. #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
  102. #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
  103. #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
  104. #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
  105. #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
  106. #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
  107. #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
  108. #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
  109. #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
  110. #define CPUFLAG_AVXSLOW (AV_CPU_FLAG_AVXSLOW | CPUFLAG_AVX)
  111. #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
  112. #define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
  113. #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
  114. #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
  115. #define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | AV_CPU_FLAG_BMI1)
  116. #define CPUFLAG_AESNI (AV_CPU_FLAG_AESNI | CPUFLAG_SSE42)
  117. static const AVOption cpuflags_opts[] = {
  118. { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
  119. #if ARCH_PPC
  120. { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
  121. #elif ARCH_X86
  122. { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  123. { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
  124. { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
  125. { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
  126. { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
  127. { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
  128. { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
  129. { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
  130. { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
  131. { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
  132. { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
  133. { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
  134. { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVXSLOW }, .unit = "flags" },
  135. { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
  136. { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
  137. { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
  138. { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
  139. { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
  140. { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
  141. { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
  142. { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
  143. { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
  144. { "aesni" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AESNI }, .unit = "flags" },
  145. #elif ARCH_ARM
  146. { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
  147. { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
  148. { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
  149. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  150. { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .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" },
  153. #elif ARCH_AARCH64
  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" },
  157. #endif
  158. { NULL },
  159. };
  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,
  165. };
  166. int flags = 0, ret;
  167. const AVClass *pclass = &class;
  168. if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
  169. return ret;
  170. return flags & INT_MAX;
  171. }
  172. int av_parse_cpu_caps(unsigned *flags, const char *s)
  173. {
  174. static const AVOption cpuflags_opts[] = {
  175. { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
  176. #if ARCH_PPC
  177. { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
  178. #elif ARCH_X86
  179. { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  180. { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
  181. { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
  182. { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
  183. { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
  184. { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
  185. { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
  186. { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
  187. { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
  188. { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
  189. { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
  190. { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
  191. { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
  192. { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVXSLOW }, .unit = "flags" },
  193. { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
  194. { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
  195. { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
  196. { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
  197. { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
  198. { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
  199. { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
  200. { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
  201. { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
  202. { "aesni", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AESNI }, .unit = "flags" },
  203. #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
  204. #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
  205. #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
  206. { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
  207. { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
  208. { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
  209. #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
  210. #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
  211. #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
  212. #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
  213. { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
  214. { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
  215. { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
  216. { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
  217. { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
  218. #elif ARCH_ARM
  219. { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
  220. { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
  221. { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
  222. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  223. { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
  224. { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
  225. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  226. { "setend", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SETEND }, .unit = "flags" },
  227. #elif ARCH_AARCH64
  228. { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
  229. { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
  230. { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
  231. #endif
  232. { NULL },
  233. };
  234. static const AVClass class = {
  235. .class_name = "cpuflags",
  236. .item_name = av_default_item_name,
  237. .option = cpuflags_opts,
  238. .version = LIBAVUTIL_VERSION_INT,
  239. };
  240. const AVClass *pclass = &class;
  241. return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
  242. }
  243. int av_cpu_count(void)
  244. {
  245. static volatile int printed;
  246. int nb_cpus = 1;
  247. #if HAVE_WINRT
  248. SYSTEM_INFO sysinfo;
  249. #endif
  250. #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
  251. cpu_set_t cpuset;
  252. CPU_ZERO(&cpuset);
  253. if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
  254. nb_cpus = CPU_COUNT(&cpuset);
  255. #elif HAVE_GETPROCESSAFFINITYMASK
  256. DWORD_PTR proc_aff, sys_aff;
  257. if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
  258. nb_cpus = av_popcount64(proc_aff);
  259. #elif HAVE_SYSCTL && defined(HW_NCPU)
  260. int mib[2] = { CTL_HW, HW_NCPU };
  261. size_t len = sizeof(nb_cpus);
  262. if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
  263. nb_cpus = 0;
  264. #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
  265. nb_cpus = sysconf(_SC_NPROC_ONLN);
  266. #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
  267. nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
  268. #elif HAVE_WINRT
  269. GetNativeSystemInfo(&sysinfo);
  270. nb_cpus = sysinfo.dwNumberOfProcessors;
  271. #endif
  272. if (!printed) {
  273. av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
  274. printed = 1;
  275. }
  276. return nb_cpus;
  277. }