Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4349 | Serge | 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 "cpu.h" |
||
20 | #include "cpu_internal.h" |
||
21 | #include "config.h" |
||
22 | #include "opt.h" |
||
23 | #include "common.h" |
||
24 | |||
25 | #if HAVE_SCHED_GETAFFINITY |
||
26 | #ifndef _GNU_SOURCE |
||
27 | # define _GNU_SOURCE |
||
28 | #endif |
||
29 | #include |
||
30 | #endif |
||
31 | #if HAVE_GETPROCESSAFFINITYMASK |
||
32 | #include |
||
33 | #endif |
||
34 | #if HAVE_SYSCTL |
||
35 | #if HAVE_SYS_PARAM_H |
||
36 | #include |
||
37 | #endif |
||
38 | #include |
||
39 | #include |
||
40 | #endif |
||
41 | #if HAVE_SYSCONF |
||
42 | #include |
||
43 | #endif |
||
44 | |||
45 | static int flags, checked; |
||
46 | |||
47 | void av_force_cpu_flags(int arg){ |
||
48 | flags = arg; |
||
49 | checked = arg != -1; |
||
50 | } |
||
51 | |||
52 | int av_get_cpu_flags(void) |
||
53 | { |
||
54 | if (checked) |
||
55 | return flags; |
||
56 | |||
57 | if (ARCH_ARM) flags = ff_get_cpu_flags_arm(); |
||
58 | if (ARCH_PPC) flags = ff_get_cpu_flags_ppc(); |
||
59 | if (ARCH_X86) flags = ff_get_cpu_flags_x86(); |
||
60 | |||
61 | checked = 1; |
||
62 | return flags; |
||
63 | } |
||
64 | |||
65 | void av_set_cpu_flags_mask(int mask) |
||
66 | { |
||
67 | checked = 0; |
||
68 | flags = av_get_cpu_flags() & mask; |
||
69 | checked = 1; |
||
70 | } |
||
71 | |||
72 | int av_parse_cpu_flags(const char *s) |
||
73 | { |
||
74 | #define CPUFLAG_MMXEXT (AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV) |
||
75 | #define CPUFLAG_3DNOW (AV_CPU_FLAG_3DNOW | AV_CPU_FLAG_MMX) |
||
76 | #define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW) |
||
77 | #define CPUFLAG_SSE (AV_CPU_FLAG_SSE | CPUFLAG_MMXEXT) |
||
78 | #define CPUFLAG_SSE2 (AV_CPU_FLAG_SSE2 | CPUFLAG_SSE) |
||
79 | #define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2) |
||
80 | #define CPUFLAG_SSE3 (AV_CPU_FLAG_SSE3 | CPUFLAG_SSE2) |
||
81 | #define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3) |
||
82 | #define CPUFLAG_SSSE3 (AV_CPU_FLAG_SSSE3 | CPUFLAG_SSE3) |
||
83 | #define CPUFLAG_SSE4 (AV_CPU_FLAG_SSE4 | CPUFLAG_SSSE3) |
||
84 | #define CPUFLAG_SSE42 (AV_CPU_FLAG_SSE42 | CPUFLAG_SSE4) |
||
85 | #define CPUFLAG_AVX (AV_CPU_FLAG_AVX | CPUFLAG_SSE42) |
||
86 | #define CPUFLAG_XOP (AV_CPU_FLAG_XOP | CPUFLAG_AVX) |
||
87 | #define CPUFLAG_FMA4 (AV_CPU_FLAG_FMA4 | CPUFLAG_AVX) |
||
88 | #define CPUFLAG_AVX2 (AV_CPU_FLAG_AVX2 | CPUFLAG_AVX) |
||
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 | { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4 }, .unit = "flags" }, |
||
108 | { "avx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2 }, .unit = "flags" }, |
||
109 | { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW }, .unit = "flags" }, |
||
110 | { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT }, .unit = "flags" }, |
||
111 | { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" }, |
||
112 | #elif ARCH_ARM |
||
113 | { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" }, |
||
114 | { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" }, |
||
115 | { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" }, |
||
116 | { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" }, |
||
117 | { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" }, |
||
118 | { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" }, |
||
119 | #endif |
||
120 | { NULL }, |
||
121 | }; |
||
122 | static const AVClass class = { |
||
123 | .class_name = "cpuflags", |
||
124 | .item_name = av_default_item_name, |
||
125 | .option = cpuflags_opts, |
||
126 | .version = LIBAVUTIL_VERSION_INT, |
||
127 | }; |
||
128 | |||
129 | int flags = 0, ret; |
||
130 | const AVClass *pclass = &class; |
||
131 | |||
132 | if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0) |
||
133 | return ret; |
||
134 | |||
135 | return flags & INT_MAX; |
||
136 | } |
||
137 | |||
138 | int av_parse_cpu_caps(unsigned *flags, const char *s) |
||
139 | { |
||
140 | static const AVOption cpuflags_opts[] = { |
||
141 | { "flags" , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" }, |
||
142 | #if ARCH_PPC |
||
143 | { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC }, .unit = "flags" }, |
||
144 | #elif ARCH_X86 |
||
145 | { "mmx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" }, |
||
146 | { "mmx2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" }, |
||
147 | { "mmxext" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2 }, .unit = "flags" }, |
||
148 | { "sse" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE }, .unit = "flags" }, |
||
149 | { "sse2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2 }, .unit = "flags" }, |
||
150 | { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW }, .unit = "flags" }, |
||
151 | { "sse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3 }, .unit = "flags" }, |
||
152 | { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW }, .unit = "flags" }, |
||
153 | { "ssse3" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3 }, .unit = "flags" }, |
||
154 | { "atom" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM }, .unit = "flags" }, |
||
155 | { "sse4.1" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4 }, .unit = "flags" }, |
||
156 | { "sse4.2" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42 }, .unit = "flags" }, |
||
157 | { "avx" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX }, .unit = "flags" }, |
||
158 | { "xop" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP }, .unit = "flags" }, |
||
159 | { "fma4" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4 }, .unit = "flags" }, |
||
160 | { "3dnow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW }, .unit = "flags" }, |
||
161 | { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT }, .unit = "flags" }, |
||
162 | { "cmov", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV }, .unit = "flags" }, |
||
163 | |||
164 | #define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX |
||
165 | #define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE |
||
166 | #define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2 |
||
167 | { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2 }, .unit = "flags" }, |
||
168 | { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3 }, .unit = "flags" }, |
||
169 | { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4 }, .unit = "flags" }, |
||
170 | |||
171 | #define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW |
||
172 | #define CPU_FLAG_ATHLON CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2 |
||
173 | #define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE |
||
174 | #define CPU_FLAG_K8 CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2 |
||
175 | { "k6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX }, .unit = "flags" }, |
||
176 | { "k62", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62 }, .unit = "flags" }, |
||
177 | { "athlon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON }, .unit = "flags" }, |
||
178 | { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP }, .unit = "flags" }, |
||
179 | { "k8", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8 }, .unit = "flags" }, |
||
180 | #elif ARCH_ARM |
||
181 | { "armv5te", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE }, .unit = "flags" }, |
||
182 | { "armv6", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6 }, .unit = "flags" }, |
||
183 | { "armv6t2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2 }, .unit = "flags" }, |
||
184 | { "vfp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP }, .unit = "flags" }, |
||
185 | { "vfpv3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3 }, .unit = "flags" }, |
||
186 | { "neon", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON }, .unit = "flags" }, |
||
187 | #endif |
||
188 | { NULL }, |
||
189 | }; |
||
190 | static const AVClass class = { |
||
191 | .class_name = "cpuflags", |
||
192 | .item_name = av_default_item_name, |
||
193 | .option = cpuflags_opts, |
||
194 | .version = LIBAVUTIL_VERSION_INT, |
||
195 | }; |
||
196 | const AVClass *pclass = &class; |
||
197 | |||
198 | return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags); |
||
199 | } |
||
200 | |||
201 | int av_cpu_count(void) |
||
202 | { |
||
203 | static volatile int printed; |
||
204 | |||
205 | int nb_cpus = 1; |
||
206 | #if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT) |
||
207 | cpu_set_t cpuset; |
||
208 | |||
209 | CPU_ZERO(&cpuset); |
||
210 | |||
211 | if (!sched_getaffinity(0, sizeof(cpuset), &cpuset)) |
||
212 | nb_cpus = CPU_COUNT(&cpuset); |
||
213 | #elif HAVE_GETPROCESSAFFINITYMASK |
||
214 | DWORD_PTR proc_aff, sys_aff; |
||
215 | if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff)) |
||
216 | nb_cpus = av_popcount64(proc_aff); |
||
217 | #elif HAVE_SYSCTL && defined(HW_NCPU) |
||
218 | int mib[2] = { CTL_HW, HW_NCPU }; |
||
219 | size_t len = sizeof(nb_cpus); |
||
220 | |||
221 | if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1) |
||
222 | nb_cpus = 0; |
||
223 | #elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN) |
||
224 | nb_cpus = sysconf(_SC_NPROC_ONLN); |
||
225 | #elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN) |
||
226 | nb_cpus = sysconf(_SC_NPROCESSORS_ONLN); |
||
227 | #endif |
||
228 | |||
229 | if (!printed) { |
||
230 | av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus); |
||
231 | printed = 1; |
||
232 | } |
||
233 | |||
234 | return nb_cpus; |
||
235 | } |
||
236 | |||
237 | #ifdef TEST |
||
238 | |||
239 | #include |
||
240 | |||
241 | static const struct { |
||
242 | int flag; |
||
243 | const char *name; |
||
244 | } cpu_flag_tab[] = { |
||
245 | #if ARCH_ARM |
||
246 | { AV_CPU_FLAG_ARMV5TE, "armv5te" }, |
||
247 | { AV_CPU_FLAG_ARMV6, "armv6" }, |
||
248 | { AV_CPU_FLAG_ARMV6T2, "armv6t2" }, |
||
249 | { AV_CPU_FLAG_VFP, "vfp" }, |
||
250 | { AV_CPU_FLAG_VFPV3, "vfpv3" }, |
||
251 | { AV_CPU_FLAG_NEON, "neon" }, |
||
252 | #elif ARCH_PPC |
||
253 | { AV_CPU_FLAG_ALTIVEC, "altivec" }, |
||
254 | #elif ARCH_X86 |
||
255 | { AV_CPU_FLAG_MMX, "mmx" }, |
||
256 | { AV_CPU_FLAG_MMXEXT, "mmxext" }, |
||
257 | { AV_CPU_FLAG_SSE, "sse" }, |
||
258 | { AV_CPU_FLAG_SSE2, "sse2" }, |
||
259 | { AV_CPU_FLAG_SSE2SLOW, "sse2(slow)" }, |
||
260 | { AV_CPU_FLAG_SSE3, "sse3" }, |
||
261 | { AV_CPU_FLAG_SSE3SLOW, "sse3(slow)" }, |
||
262 | { AV_CPU_FLAG_SSSE3, "ssse3" }, |
||
263 | { AV_CPU_FLAG_ATOM, "atom" }, |
||
264 | { AV_CPU_FLAG_SSE4, "sse4.1" }, |
||
265 | { AV_CPU_FLAG_SSE42, "sse4.2" }, |
||
266 | { AV_CPU_FLAG_AVX, "avx" }, |
||
267 | { AV_CPU_FLAG_XOP, "xop" }, |
||
268 | { AV_CPU_FLAG_FMA4, "fma4" }, |
||
269 | { AV_CPU_FLAG_3DNOW, "3dnow" }, |
||
270 | { AV_CPU_FLAG_3DNOWEXT, "3dnowext" }, |
||
271 | { AV_CPU_FLAG_CMOV, "cmov" }, |
||
272 | { AV_CPU_FLAG_AVX2, "avx2" }, |
||
273 | #endif |
||
274 | { 0 } |
||
275 | }; |
||
276 | |||
277 | int main(void) |
||
278 | { |
||
279 | int cpu_flags = av_get_cpu_flags(); |
||
280 | int i; |
||
281 | |||
282 | printf("cpu_flags = 0x%08X\n", cpu_flags); |
||
283 | printf("cpu_flags ="); |
||
284 | for (i = 0; cpu_flag_tab[i].flag; i++) |
||
285 | if (cpu_flags & cpu_flag_tab[i].flag) |
||
286 | printf(" %s", cpu_flag_tab[i].name); |
||
287 | printf("\n"); |
||
288 | |||
289 | return 0; |
||
290 | } |
||
291 | |||
292 | #endif> |