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 <stdint.h>
20
21#include "cpu.h"
22#include "cpu_internal.h"
23#include "config.h"
24#include "opt.h"
25#include "common.h"
26
27#if HAVE_SCHED_GETAFFINITY
28#ifndef _GNU_SOURCE
29# define _GNU_SOURCE
30#endif
31#include <sched.h>
32#endif
33#if HAVE_GETPROCESSAFFINITYMASK
34#include <windows.h>
35#endif
36#if HAVE_SYSCTL
37#if HAVE_SYS_PARAM_H
38#include <sys/param.h>
39#endif
40#include <sys/types.h>
41#include <sys/sysctl.h>
42#endif
43#if HAVE_UNISTD_H
44#include <unistd.h>
45#endif
46
47static int flags, checked;
48
49void av_force_cpu_flags(int arg){
50    if (   (arg & ( AV_CPU_FLAG_3DNOW    |
51                    AV_CPU_FLAG_3DNOWEXT |
52                    AV_CPU_FLAG_SSE      |
53                    AV_CPU_FLAG_SSE2     |
54                    AV_CPU_FLAG_SSE2SLOW |
55                    AV_CPU_FLAG_SSE3     |
56                    AV_CPU_FLAG_SSE3SLOW |
57                    AV_CPU_FLAG_SSSE3    |
58                    AV_CPU_FLAG_SSE4     |
59                    AV_CPU_FLAG_SSE42    |
60                    AV_CPU_FLAG_AVX      |
61                    AV_CPU_FLAG_XOP      |
62                    AV_CPU_FLAG_FMA3     |
63                    AV_CPU_FLAG_FMA4     |
64                    AV_CPU_FLAG_AVX2     ))
65        && !(arg & AV_CPU_FLAG_MMX)) {
66        av_log(NULL, AV_LOG_WARNING, "MMX implied by specified flags\n");
67        arg |= AV_CPU_FLAG_MMX;
68    }
69
70    flags   = arg;
71    checked = arg != -1;
72}
73
74int av_get_cpu_flags(void)
75{
76    if (checked)
77        return flags;
78
79    if (ARCH_AARCH64)
80        flags = ff_get_cpu_flags_aarch64();
81    if (ARCH_ARM)
82        flags = ff_get_cpu_flags_arm();
83    if (ARCH_PPC)
84        flags = ff_get_cpu_flags_ppc();
85    if (ARCH_X86)
86        flags = ff_get_cpu_flags_x86();
87
88    checked = 1;
89    return flags;
90}
91
92void av_set_cpu_flags_mask(int mask)
93{
94    checked       = 0;
95    flags         = av_get_cpu_flags() & mask;
96    checked       = 1;
97}
98
99int av_parse_cpu_flags(const char *s)
100{
101#define CPUFLAG_MMXEXT   (AV_CPU_FLAG_MMX      | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
102#define CPUFLAG_3DNOW    (AV_CPU_FLAG_3DNOW    | AV_CPU_FLAG_MMX)
103#define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
104#define CPUFLAG_SSE      (AV_CPU_FLAG_SSE      | CPUFLAG_MMXEXT)
105#define CPUFLAG_SSE2     (AV_CPU_FLAG_SSE2     | CPUFLAG_SSE)
106#define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
107#define CPUFLAG_SSE3     (AV_CPU_FLAG_SSE3     | CPUFLAG_SSE2)
108#define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
109#define CPUFLAG_SSSE3    (AV_CPU_FLAG_SSSE3    | CPUFLAG_SSE3)
110#define CPUFLAG_SSE4     (AV_CPU_FLAG_SSE4     | CPUFLAG_SSSE3)
111#define CPUFLAG_SSE42    (AV_CPU_FLAG_SSE42    | CPUFLAG_SSE4)
112#define CPUFLAG_AVX      (AV_CPU_FLAG_AVX      | CPUFLAG_SSE42)
113#define CPUFLAG_XOP      (AV_CPU_FLAG_XOP      | CPUFLAG_AVX)
114#define CPUFLAG_FMA3     (AV_CPU_FLAG_FMA3     | CPUFLAG_AVX)
115#define CPUFLAG_FMA4     (AV_CPU_FLAG_FMA4     | CPUFLAG_AVX)
116#define CPUFLAG_AVX2     (AV_CPU_FLAG_AVX2     | CPUFLAG_AVX)
117#define CPUFLAG_BMI1     (AV_CPU_FLAG_BMI1)
118#define CPUFLAG_BMI2     (AV_CPU_FLAG_BMI2     | CPUFLAG_BMI1)
119    static const AVOption cpuflags_opts[] = {
120        { "flags"   , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
121#if   ARCH_PPC
122        { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC  },    .unit = "flags" },
123#elif ARCH_X86
124        { "mmx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX      },    .unit = "flags" },
125        { "mmxext"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT       },    .unit = "flags" },
126        { "sse"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE          },    .unit = "flags" },
127        { "sse2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2         },    .unit = "flags" },
128        { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW     },    .unit = "flags" },
129        { "sse3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3         },    .unit = "flags" },
130        { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW     },    .unit = "flags" },
131        { "ssse3"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3        },    .unit = "flags" },
132        { "atom"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM     },    .unit = "flags" },
133        { "sse4.1"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4         },    .unit = "flags" },
134        { "sse4.2"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42        },    .unit = "flags" },
135        { "avx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX          },    .unit = "flags" },
136        { "xop"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP          },    .unit = "flags" },
137        { "fma3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3         },    .unit = "flags" },
138        { "fma4"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4         },    .unit = "flags" },
139        { "avx2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2         },    .unit = "flags" },
140        { "bmi1"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI1         },    .unit = "flags" },
141        { "bmi2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2         },    .unit = "flags" },
142        { "3dnow"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW        },    .unit = "flags" },
143        { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT     },    .unit = "flags" },
144        { "cmov",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV     },    .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        { "vfpv3",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3    },    .unit = "flags" },
151        { "neon",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON     },    .unit = "flags" },
152#elif ARCH_AARCH64
153        { "armv8",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8    },    .unit = "flags" },
154        { "neon",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON     },    .unit = "flags" },
155        { "vfp",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP      },    .unit = "flags" },
156#endif
157        { NULL },
158    };
159    static const AVClass class = {
160        .class_name = "cpuflags",
161        .item_name  = av_default_item_name,
162        .option     = cpuflags_opts,
163        .version    = LIBAVUTIL_VERSION_INT,
164    };
165
166    int flags = 0, ret;
167    const AVClass *pclass = &class;
168
169    if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
170        return ret;
171
172    return flags & INT_MAX;
173}
174
175int av_parse_cpu_caps(unsigned *flags, const char *s)
176{
177        static const AVOption cpuflags_opts[] = {
178        { "flags"   , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
179#if   ARCH_PPC
180        { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC  },    .unit = "flags" },
181#elif ARCH_X86
182        { "mmx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX      },    .unit = "flags" },
183        { "mmx2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2     },    .unit = "flags" },
184        { "mmxext"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX2     },    .unit = "flags" },
185        { "sse"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE      },    .unit = "flags" },
186        { "sse2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2     },    .unit = "flags" },
187        { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE2SLOW },    .unit = "flags" },
188        { "sse3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3     },    .unit = "flags" },
189        { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE3SLOW },    .unit = "flags" },
190        { "ssse3"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSSE3    },    .unit = "flags" },
191        { "atom"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM     },    .unit = "flags" },
192        { "sse4.1"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE4     },    .unit = "flags" },
193        { "sse4.2"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_SSE42    },    .unit = "flags" },
194        { "avx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX      },    .unit = "flags" },
195        { "xop"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_XOP      },    .unit = "flags" },
196        { "fma3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3     },    .unit = "flags" },
197        { "fma4"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4     },    .unit = "flags" },
198        { "avx2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2     },    .unit = "flags" },
199        { "bmi1"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1     },    .unit = "flags" },
200        { "bmi2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2     },    .unit = "flags" },
201        { "3dnow"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW    },    .unit = "flags" },
202        { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT },    .unit = "flags" },
203        { "cmov",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV     },    .unit = "flags" },
204
205#define CPU_FLAG_P2 AV_CPU_FLAG_CMOV | AV_CPU_FLAG_MMX
206#define CPU_FLAG_P3 CPU_FLAG_P2 | AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE
207#define CPU_FLAG_P4 CPU_FLAG_P3| AV_CPU_FLAG_SSE2
208        { "pentium2", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P2          },    .unit = "flags" },
209        { "pentium3", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P3          },    .unit = "flags" },
210        { "pentium4", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_P4          },    .unit = "flags" },
211
212#define CPU_FLAG_K62 AV_CPU_FLAG_MMX | AV_CPU_FLAG_3DNOW
213#define CPU_FLAG_ATHLON   CPU_FLAG_K62 | AV_CPU_FLAG_CMOV | AV_CPU_FLAG_3DNOWEXT | AV_CPU_FLAG_MMX2
214#define CPU_FLAG_ATHLONXP CPU_FLAG_ATHLON | AV_CPU_FLAG_SSE
215#define CPU_FLAG_K8  CPU_FLAG_ATHLONXP | AV_CPU_FLAG_SSE2
216        { "k6",       NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX      },    .unit = "flags" },
217        { "k62",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K62         },    .unit = "flags" },
218        { "athlon",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLON      },    .unit = "flags" },
219        { "athlonxp", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_ATHLONXP    },    .unit = "flags" },
220        { "k8",       NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPU_FLAG_K8          },    .unit = "flags" },
221#elif ARCH_ARM
222        { "armv5te",  NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE  },    .unit = "flags" },
223        { "armv6",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6    },    .unit = "flags" },
224        { "armv6t2",  NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2  },    .unit = "flags" },
225        { "vfp",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP      },    .unit = "flags" },
226        { "vfpv3",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3    },    .unit = "flags" },
227        { "neon",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON     },    .unit = "flags" },
228#elif ARCH_AARCH64
229        { "armv8",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8    },    .unit = "flags" },
230        { "neon",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON     },    .unit = "flags" },
231        { "vfp",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP      },    .unit = "flags" },
232#endif
233        { NULL },
234    };
235    static const AVClass class = {
236        .class_name = "cpuflags",
237        .item_name  = av_default_item_name,
238        .option     = cpuflags_opts,
239        .version    = LIBAVUTIL_VERSION_INT,
240    };
241    const AVClass *pclass = &class;
242
243    return av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, flags);
244}
245
246int av_cpu_count(void)
247{
248    static volatile int printed;
249
250    int nb_cpus = 1;
251#if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
252    cpu_set_t cpuset;
253
254    CPU_ZERO(&cpuset);
255
256    if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
257        nb_cpus = CPU_COUNT(&cpuset);
258#elif HAVE_GETPROCESSAFFINITYMASK
259    DWORD_PTR proc_aff, sys_aff;
260    if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
261        nb_cpus = av_popcount64(proc_aff);
262#elif HAVE_SYSCTL && defined(HW_NCPU)
263    int mib[2] = { CTL_HW, HW_NCPU };
264    size_t len = sizeof(nb_cpus);
265
266    if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
267        nb_cpus = 0;
268#elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
269    nb_cpus = sysconf(_SC_NPROC_ONLN);
270#elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
271    nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
272#endif
273
274    if (!printed) {
275        av_log(NULL, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
276        printed = 1;
277    }
278
279    return nb_cpus;
280}
281
282#ifdef TEST
283
284#include <stdio.h>
285
286static const struct {
287    int flag;
288    const char *name;
289} cpu_flag_tab[] = {
290#if   ARCH_AARCH64
291    { AV_CPU_FLAG_ARMV8,     "armv8"      },
292    { AV_CPU_FLAG_NEON,      "neon"       },
293    { AV_CPU_FLAG_VFP,       "vfp"        },
294#elif ARCH_ARM
295    { AV_CPU_FLAG_ARMV5TE,   "armv5te"    },
296    { AV_CPU_FLAG_ARMV6,     "armv6"      },
297    { AV_CPU_FLAG_ARMV6T2,   "armv6t2"    },
298    { AV_CPU_FLAG_VFP,       "vfp"        },
299    { AV_CPU_FLAG_VFPV3,     "vfpv3"      },
300    { AV_CPU_FLAG_NEON,      "neon"       },
301#elif ARCH_PPC
302    { AV_CPU_FLAG_ALTIVEC,   "altivec"    },
303#elif ARCH_X86
304    { AV_CPU_FLAG_MMX,       "mmx"        },
305    { AV_CPU_FLAG_MMXEXT,    "mmxext"     },
306    { AV_CPU_FLAG_SSE,       "sse"        },
307    { AV_CPU_FLAG_SSE2,      "sse2"       },
308    { AV_CPU_FLAG_SSE2SLOW,  "sse2(slow)" },
309    { AV_CPU_FLAG_SSE3,      "sse3"       },
310    { AV_CPU_FLAG_SSE3SLOW,  "sse3(slow)" },
311    { AV_CPU_FLAG_SSSE3,     "ssse3"      },
312    { AV_CPU_FLAG_ATOM,      "atom"       },
313    { AV_CPU_FLAG_SSE4,      "sse4.1"     },
314    { AV_CPU_FLAG_SSE42,     "sse4.2"     },
315    { AV_CPU_FLAG_AVX,       "avx"        },
316    { AV_CPU_FLAG_XOP,       "xop"        },
317    { AV_CPU_FLAG_FMA3,      "fma3"       },
318    { AV_CPU_FLAG_FMA4,      "fma4"       },
319    { AV_CPU_FLAG_3DNOW,     "3dnow"      },
320    { AV_CPU_FLAG_3DNOWEXT,  "3dnowext"   },
321    { AV_CPU_FLAG_CMOV,      "cmov"       },
322    { AV_CPU_FLAG_AVX2,      "avx2"       },
323    { AV_CPU_FLAG_BMI1,      "bmi1"       },
324    { AV_CPU_FLAG_BMI2,      "bmi2"       },
325#endif
326    { 0 }
327};
328
329int main(void)
330{
331    int cpu_flags = av_get_cpu_flags();
332    int i;
333
334    printf("cpu_flags = 0x%08X\n", cpu_flags);
335    printf("cpu_flags =");
336    for (i = 0; cpu_flag_tab[i].flag; i++)
337        if (cpu_flags & cpu_flag_tab[i].flag)
338            printf(" %s", cpu_flag_tab[i].name);
339    printf("\n");
340
341    return 0;
342}
343
344#endif
345