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
19 #include <stddef.h>
20 #include <stdint.h>
21 #include <stdatomic.h>
22
23 #include "attributes.h"
24 #include "cpu.h"
25 #include "cpu_internal.h"
26 #include "config.h"
27 #include "opt.h"
28 #include "common.h"
29
30 #if HAVE_SCHED_GETAFFINITY
31 #ifndef _GNU_SOURCE
32 # define _GNU_SOURCE
33 #endif
34 #include <sched.h>
35 #endif
36 #if HAVE_GETPROCESSAFFINITYMASK || HAVE_WINRT
37 #include <windows.h>
38 #endif
39 #if HAVE_SYSCTL
40 #if HAVE_SYS_PARAM_H
41 #include <sys/param.h>
42 #endif
43 #include <sys/types.h>
44 #include <sys/sysctl.h>
45 #endif
46 #if HAVE_UNISTD_H
47 #include <unistd.h>
48 #endif
49
50 static atomic_int cpu_flags = ATOMIC_VAR_INIT(-1);
51
get_cpu_flags(void)52 static int get_cpu_flags(void)
53 {
54 if (ARCH_AARCH64)
55 return ff_get_cpu_flags_aarch64();
56 if (ARCH_ARM)
57 return ff_get_cpu_flags_arm();
58 if (ARCH_PPC)
59 return ff_get_cpu_flags_ppc();
60 if (ARCH_X86)
61 return ff_get_cpu_flags_x86();
62 return 0;
63 }
64
av_force_cpu_flags(int arg)65 void av_force_cpu_flags(int arg){
66 if (ARCH_X86 &&
67 (arg & ( AV_CPU_FLAG_3DNOW |
68 AV_CPU_FLAG_3DNOWEXT |
69 AV_CPU_FLAG_MMXEXT |
70 AV_CPU_FLAG_SSE |
71 AV_CPU_FLAG_SSE2 |
72 AV_CPU_FLAG_SSE2SLOW |
73 AV_CPU_FLAG_SSE3 |
74 AV_CPU_FLAG_SSE3SLOW |
75 AV_CPU_FLAG_SSSE3 |
76 AV_CPU_FLAG_SSE4 |
77 AV_CPU_FLAG_SSE42 |
78 AV_CPU_FLAG_AVX |
79 AV_CPU_FLAG_AVXSLOW |
80 AV_CPU_FLAG_XOP |
81 AV_CPU_FLAG_FMA3 |
82 AV_CPU_FLAG_FMA4 |
83 AV_CPU_FLAG_AVX2 |
84 AV_CPU_FLAG_AVX512 ))
85 && !(arg & AV_CPU_FLAG_MMX)) {
86 av_log(NULL, AV_LOG_WARNING, "MMX implied by specified flags\n");
87 arg |= AV_CPU_FLAG_MMX;
88 }
89
90 atomic_store_explicit(&cpu_flags, arg, memory_order_relaxed);
91 }
92
av_get_cpu_flags(void)93 int av_get_cpu_flags(void)
94 {
95 int flags = atomic_load_explicit(&cpu_flags, memory_order_relaxed);
96 if (flags == -1) {
97 flags = get_cpu_flags();
98 atomic_store_explicit(&cpu_flags, flags, memory_order_relaxed);
99 }
100 return flags;
101 }
102
av_set_cpu_flags_mask(int mask)103 void av_set_cpu_flags_mask(int mask)
104 {
105 atomic_store_explicit(&cpu_flags, get_cpu_flags() & mask,
106 memory_order_relaxed);
107 }
108
av_parse_cpu_flags(const char * s)109 int av_parse_cpu_flags(const char *s)
110 {
111 #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
112 #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX)
113 #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
114 #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT)
115 #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE)
116 #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
117 #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2)
118 #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
119 #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3)
120 #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3)
121 #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4)
122 #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42)
123 #define CPUFLAG_AVXSLOW (AV_CPU_FLAG_AVXSLOW | CPUFLAG_AVX)
124 #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX)
125 #define CPUFLAG_FMA3 (AV_CPU_FLAG_FMA3 | CPUFLAG_AVX)
126 #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX)
127 #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX)
128 #define CPUFLAG_BMI2 (AV_CPU_FLAG_BMI2 | AV_CPU_FLAG_BMI1)
129 #define CPUFLAG_AESNI (AV_CPU_FLAG_AESNI | CPUFLAG_SSE42)
130 #define CPUFLAG_AVX512 (AV_CPU_FLAG_AVX512 | CPUFLAG_AVX2)
131 static const AVOption cpuflags_opts[] = {
132 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
133 #if ARCH_PPC
134 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
135 #elif ARCH_X86
136 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
137 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT }, .unit = "flags" },
138 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE }, .unit = "flags" },
139 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2 }, .unit = "flags" },
140 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW }, .unit = "flags" },
141 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3 }, .unit = "flags" },
142 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW }, .unit = "flags" },
143 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3 }, .unit = "flags" },
144 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
145 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4 }, .unit = "flags" },
146 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42 }, .unit = "flags" },
147 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX }, .unit = "flags" },
148 { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVXSLOW }, .unit = "flags" },
149 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP }, .unit = "flags" },
150 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3 }, .unit = "flags" },
151 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" },
152 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" },
153 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
154 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2 }, .unit = "flags" },
155 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" },
156 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" },
157 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
158 { "aesni" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AESNI }, .unit = "flags" },
159 { "avx512" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX512 }, .unit = "flags" },
160 #elif ARCH_ARM
161 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
162 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
163 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
164 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
165 { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
166 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
167 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
168 #elif ARCH_AARCH64
169 { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
170 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
171 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
172 #endif
173 { NULL },
174 };
175 static const AVClass class = {
176 .class_name = "cpuflags",
177 .item_name = av_default_item_name,
178 .option = cpuflags_opts,
179 .version = LIBAVUTIL_VERSION_INT,
180 };
181
182 int flags = 0, ret;
183 const AVClass *pclass = &class;
184
185 if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
186 return ret;
187
188 return flags & INT_MAX;
189 }
190
av_parse_cpu_caps(unsigned * flags,const char * s)191 int av_parse_cpu_caps(unsigned *flags, const char *s)
192 {
193 static const AVOption cpuflags_opts[] = {
194 { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
195 #if ARCH_PPC
196 { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" },
197 #elif ARCH_X86
198 { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
199 { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
200 { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" },
201 { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" },
202 { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" },
203 { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" },
204 { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" },
205 { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" },
206 { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" },
207 { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" },
208 { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" },
209 { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" },
210 { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" },
211 { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVXSLOW }, .unit = "flags" },
212 { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" },
213 { "fma3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3 }, .unit = "flags" },
214 { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" },
215 { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2 }, .unit = "flags" },
216 { "bmi1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1 }, .unit = "flags" },
217 { "bmi2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2 }, .unit = "flags" },
218 { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" },
219 { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" },
220 { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" },
221 { "aesni", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AESNI }, .unit = "flags" },
222 { "avx512" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX512 }, .unit = "flags" },
223
224 #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
225 #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
226 #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
227 { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" },
228 { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" },
229 { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" },
230
231 #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
232 #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
233 #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
234 #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
235 { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" },
236 { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" },
237 { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" },
238 { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" },
239 { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" },
240 #elif ARCH_ARM
241 { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" },
242 { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" },
243 { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" },
244 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
245 { "vfp_vm", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM }, .unit = "flags" },
246 { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" },
247 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
248 { "setend", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SETEND }, .unit = "flags" },
249 #elif ARCH_AARCH64
250 { "armv8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8 }, .unit = "flags" },
251 { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" },
252 { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" },
253 #endif
254 { NULL },
255 };
256 static const AVClass class = {
257 .class_name = "cpuflags",
258 .item_name = av_default_item_name,
259 .option = cpuflags_opts,
260 .version = LIBAVUTIL_VERSION_INT,
261 };
262 const AVClass *pclass = &class;
263
264 return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
265 }
266
av_cpu_count(void)267 int av_cpu_count(void)
268 {
269 static volatile int printed;
270
271 int nb_cpus = 1;
272 #if HAVE_WINRT
273 SYSTEM_INFO sysinfo;
274 #endif
275 #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
276 cpu_set_t cpuset;
277
278 CPU_ZERO(&cpuset);
279
280 if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
281 nb_cpus = CPU_COUNT(&cpuset);
282 #elif HAVE_GETPROCESSAFFINITYMASK
283 DWORD_PTR proc_aff, sys_aff;
284 if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
285 nb_cpus = av_popcount64(proc_aff);
286 #elif HAVE_SYSCTL && defined(HW_NCPU)
287 int mib[2] = { CTL_HW, HW_NCPU };
288 size_t len = sizeof(nb_cpus);
289
290 if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
291 nb_cpus = 0;
292 #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
293 nb_cpus = sysconf(_SC_NPROC_ONLN);
294 #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
295 nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
296 #elif HAVE_WINRT
297 GetNativeSystemInfo(&sysinfo);
298 nb_cpus = sysinfo.dwNumberOfProcessors;
299 #endif
300
301 if (!printed) {
302 av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
303 printed = 1;
304 }
305
306 return nb_cpus;
307 }
308
av_cpu_max_align(void)309 size_t av_cpu_max_align(void)
310 {
311 if (ARCH_AARCH64)
312 return ff_get_cpu_max_align_aarch64();
313 if (ARCH_ARM)
314 return ff_get_cpu_max_align_arm();
315 if (ARCH_PPC)
316 return ff_get_cpu_max_align_ppc();
317 if (ARCH_X86)
318 return ff_get_cpu_max_align_x86();
319
320 return 8;
321 }
322