cpu.c 10.9 KB
Newer Older
1
/*
2
 * This file is part of Libav.
3
 *
4
 * Libav is free software; you can redistribute it and/or
5 6 7 8
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
9
 * Libav is distributed in the hope that it will be useful,
10 11 12 13 14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
15
 * License along with Libav; if not, write to the Free Software
16 17 18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

19 20
#include <stdint.h>

21
#include "cpu.h"
22
#include "cpu_internal.h"
23
#include "config.h"
24
#include "opt.h"
25
#include "common.h"
26

27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
#if HAVE_SCHED_GETAFFINITY
#define _GNU_SOURCE
#include <sched.h>
#endif
#if HAVE_GETPROCESSAFFINITYMASK
#include <windows.h>
#endif
#if HAVE_SYSCTL
#if HAVE_SYS_PARAM_H
#include <sys/param.h>
#endif
#include <sys/types.h>
#include <sys/sysctl.h>
#endif
#if HAVE_SYSCONF
#include <unistd.h>
#endif

45
static int cpuflags_mask = -1, checked;
46

47 48
int av_get_cpu_flags(void)
{
49
    static int flags;
50 51 52 53

    if (checked)
        return flags;

54 55 56 57 58 59 60 61
    if (ARCH_AARCH64)
        flags = ff_get_cpu_flags_aarch64();
    if (ARCH_ARM)
        flags = ff_get_cpu_flags_arm();
    if (ARCH_PPC)
        flags = ff_get_cpu_flags_ppc();
    if (ARCH_X86)
        flags = ff_get_cpu_flags_x86();
62

63
    flags  &= cpuflags_mask;
64
    checked = 1;
65

66
    return flags;
67 68
}

69 70 71 72 73 74
void av_set_cpu_flags_mask(int mask)
{
    cpuflags_mask = mask;
    checked       = 0;
}

75 76
int av_parse_cpu_flags(const char *s)
{
77
#define CPUFLAG_MMXEXT   (AV_CPU_FLAG_MMX      | AV_CPU_FLAG_MMXEXT | AV_CPU_FLAG_CMOV)
78 79
#define CPUFLAG_3DNOW    (AV_CPU_FLAG_3DNOW    | AV_CPU_FLAG_MMX)
#define CPUFLAG_3DNOWEXT (AV_CPU_FLAG_3DNOWEXT | CPUFLAG_3DNOW)
80
#define CPUFLAG_SSE      (AV_CPU_FLAG_SSE      | CPUFLAG_MMXEXT)
81 82 83 84 85 86 87 88
#define CPUFLAG_SSE2     (AV_CPU_FLAG_SSE2     | CPUFLAG_SSE)
#define CPUFLAG_SSE2SLOW (AV_CPU_FLAG_SSE2SLOW | CPUFLAG_SSE2)
#define CPUFLAG_SSE3     (AV_CPU_FLAG_SSE3     | CPUFLAG_SSE2)
#define CPUFLAG_SSE3SLOW (AV_CPU_FLAG_SSE3SLOW | CPUFLAG_SSE3)
#define CPUFLAG_SSSE3    (AV_CPU_FLAG_SSSE3    | CPUFLAG_SSE3)
#define CPUFLAG_SSE4     (AV_CPU_FLAG_SSE4     | CPUFLAG_SSSE3)
#define CPUFLAG_SSE42    (AV_CPU_FLAG_SSE42    | CPUFLAG_SSE4)
#define CPUFLAG_AVX      (AV_CPU_FLAG_AVX      | CPUFLAG_SSE42)
89
#define CPUFLAG_AVXSLOW  (AV_CPU_FLAG_AVXSLOW  | CPUFLAG_AVX)
90
#define CPUFLAG_XOP      (AV_CPU_FLAG_XOP      | CPUFLAG_AVX)
91
#define CPUFLAG_FMA3     (AV_CPU_FLAG_FMA3     | CPUFLAG_AVX)
92
#define CPUFLAG_FMA4     (AV_CPU_FLAG_FMA4     | CPUFLAG_AVX)
93
#define CPUFLAG_AVX2     (AV_CPU_FLAG_AVX2     | CPUFLAG_AVX)
94
#define CPUFLAG_BMI2     (AV_CPU_FLAG_BMI2     | AV_CPU_FLAG_BMI1)
95
    static const AVOption cpuflags_opts[] = {
96
        { "flags"   , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
97
#if   ARCH_PPC
98
        { "altivec" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ALTIVEC  },    .unit = "flags" },
99
#elif ARCH_X86
100 101 102 103 104 105 106 107 108 109 110 111
        { "mmx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_MMX      },    .unit = "flags" },
        { "mmxext"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_MMXEXT       },    .unit = "flags" },
        { "sse"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE          },    .unit = "flags" },
        { "sse2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2         },    .unit = "flags" },
        { "sse2slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE2SLOW     },    .unit = "flags" },
        { "sse3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3         },    .unit = "flags" },
        { "sse3slow", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE3SLOW     },    .unit = "flags" },
        { "ssse3"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSSE3        },    .unit = "flags" },
        { "atom"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ATOM     },    .unit = "flags" },
        { "sse4.1"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE4         },    .unit = "flags" },
        { "sse4.2"  , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_SSE42        },    .unit = "flags" },
        { "avx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX          },    .unit = "flags" },
112
        { "avxslow" , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVXSLOW      },    .unit = "flags" },
113
        { "xop"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP          },    .unit = "flags" },
114
        { "fma3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3         },    .unit = "flags" },
115
        { "fma4"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4         },    .unit = "flags" },
116
        { "avx2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2         },    .unit = "flags" },
117
        { "bmi1"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1     },    .unit = "flags" },
118
        { "bmi2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2         },    .unit = "flags" },
119 120 121
        { "3dnow"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW        },    .unit = "flags" },
        { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT     },    .unit = "flags" },
        { "cmov",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV     },    .unit = "flags" },
122
#elif ARCH_ARM
123 124 125 126
        { "armv5te",  NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV5TE  },    .unit = "flags" },
        { "armv6",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6    },    .unit = "flags" },
        { "armv6t2",  NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV6T2  },    .unit = "flags" },
        { "vfp",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP      },    .unit = "flags" },
127
        { "vfp_vm",   NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP_VM   },    .unit = "flags" },
128 129
        { "vfpv3",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFPV3    },    .unit = "flags" },
        { "neon",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON     },    .unit = "flags" },
130
#elif ARCH_AARCH64
131
        { "armv8",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_ARMV8    },    .unit = "flags" },
132 133
        { "neon",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_NEON     },    .unit = "flags" },
        { "vfp",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_VFP      },    .unit = "flags" },
134
#endif
135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152
        { NULL },
    };
    static const AVClass class = {
        .class_name = "cpuflags",
        .item_name  = av_default_item_name,
        .option     = cpuflags_opts,
        .version    = LIBAVUTIL_VERSION_INT,
    };

    int flags = 0, ret;
    const AVClass *pclass = &class;

    if ((ret = av_opt_eval_flags(&pclass, &cpuflags_opts[0], s, &flags)) < 0)
        return ret;

    return flags & INT_MAX;
}

153 154
int av_cpu_count(void)
{
155
    int nb_cpus = 1;
156 157 158 159 160
#if HAVE_SCHED_GETAFFINITY && defined(CPU_COUNT)
    cpu_set_t cpuset;

    CPU_ZERO(&cpuset);

161
    if (!sched_getaffinity(0, sizeof(cpuset), &cpuset))
162 163 164
        nb_cpus = CPU_COUNT(&cpuset);
#elif HAVE_GETPROCESSAFFINITYMASK
    DWORD_PTR proc_aff, sys_aff;
165
    if (GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff))
166 167 168 169 170
        nb_cpus = av_popcount64(proc_aff);
#elif HAVE_SYSCTL && defined(HW_NCPU)
    int mib[2] = { CTL_HW, HW_NCPU };
    size_t len = sizeof(nb_cpus);

171
    if (sysctl(mib, 2, &nb_cpus, &len, NULL, 0) == -1)
172 173 174 175 176 177 178 179 180 181
        nb_cpus = 0;
#elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
    nb_cpus = sysconf(_SC_NPROC_ONLN);
#elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
    nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
#endif

    return nb_cpus;
}

182 183
#ifdef TEST

184
#include <stdio.h>
185 186 187 188
#if HAVE_UNISTD_H
#include <unistd.h>
#endif

189
#include "avstring.h"
190

191 192 193 194
#if !HAVE_GETOPT
#include "compat/getopt.c"
#endif

195 196 197 198
static const struct {
    int flag;
    const char *name;
} cpu_flag_tab[] = {
199
#if   ARCH_AARCH64
200
    { AV_CPU_FLAG_ARMV8,     "armv8"      },
201 202 203
    { AV_CPU_FLAG_NEON,      "neon"       },
    { AV_CPU_FLAG_VFP,       "vfp"        },
#elif ARCH_ARM
204 205 206 207
    { AV_CPU_FLAG_ARMV5TE,   "armv5te"    },
    { AV_CPU_FLAG_ARMV6,     "armv6"      },
    { AV_CPU_FLAG_ARMV6T2,   "armv6t2"    },
    { AV_CPU_FLAG_VFP,       "vfp"        },
208
    { AV_CPU_FLAG_VFP_VM,    "vfp_vm"     },
209 210 211
    { AV_CPU_FLAG_VFPV3,     "vfpv3"      },
    { AV_CPU_FLAG_NEON,      "neon"       },
#elif ARCH_PPC
212 213 214
    { AV_CPU_FLAG_ALTIVEC,   "altivec"    },
#elif ARCH_X86
    { AV_CPU_FLAG_MMX,       "mmx"        },
215
    { AV_CPU_FLAG_MMXEXT,    "mmxext"     },
216 217 218 219 220 221 222 223 224 225
    { AV_CPU_FLAG_SSE,       "sse"        },
    { AV_CPU_FLAG_SSE2,      "sse2"       },
    { AV_CPU_FLAG_SSE2SLOW,  "sse2(slow)" },
    { AV_CPU_FLAG_SSE3,      "sse3"       },
    { AV_CPU_FLAG_SSE3SLOW,  "sse3(slow)" },
    { AV_CPU_FLAG_SSSE3,     "ssse3"      },
    { AV_CPU_FLAG_ATOM,      "atom"       },
    { AV_CPU_FLAG_SSE4,      "sse4.1"     },
    { AV_CPU_FLAG_SSE42,     "sse4.2"     },
    { AV_CPU_FLAG_AVX,       "avx"        },
226
    { AV_CPU_FLAG_AVXSLOW,   "avxslow"    },
227
    { AV_CPU_FLAG_XOP,       "xop"        },
228
    { AV_CPU_FLAG_FMA3,      "fma3"       },
229
    { AV_CPU_FLAG_FMA4,      "fma4"       },
230 231
    { AV_CPU_FLAG_3DNOW,     "3dnow"      },
    { AV_CPU_FLAG_3DNOWEXT,  "3dnowext"   },
232
    { AV_CPU_FLAG_CMOV,      "cmov"       },
233
    { AV_CPU_FLAG_AVX2,      "avx2"       },
234 235
    { AV_CPU_FLAG_BMI1,      "bmi1"       },
    { AV_CPU_FLAG_BMI2,      "bmi2"       },
236 237 238 239
#endif
    { 0 }
};

240
static void print_cpu_flags(int cpu_flags, const char *type)
241
{
242
    int i;
243

244 245
    fprintf(stderr, "cpu_flags(%s) = 0x%08X\n", type, cpu_flags);
    fprintf(stderr, "cpu_flags_str(%s) =", type);
246 247
    for (i = 0; cpu_flag_tab[i].flag; i++)
        if (cpu_flags & cpu_flag_tab[i].flag)
248 249 250 251 252 253 254 255 256
            fprintf(stderr, " %s", cpu_flag_tab[i].name);
    fprintf(stderr, "\n");
}


int main(int argc, char **argv)
{
    int cpu_flags_raw = av_get_cpu_flags();
    int cpu_flags_eff;
257 258
    int cpu_count = av_cpu_count();
    char threads[5] = "auto";
259 260 261 262 263

    if (cpu_flags_raw < 0)
        return 1;

    for (;;) {
264
        int c = getopt(argc, argv, "c:t:");
265 266 267 268 269 270 271 272 273 274 275
        if (c == -1)
            break;
        switch (c) {
        case 'c':
        {
            int cpuflags = av_parse_cpu_flags(optarg);
            if (cpuflags < 0)
                return 2;
            av_set_cpu_flags_mask(cpuflags);
            break;
        }
276 277 278 279 280 281 282 283
        case 't':
        {
            int len = av_strlcpy(threads, optarg, sizeof(threads));
            if (len >= sizeof(threads)) {
                fprintf(stderr, "Invalid thread count '%s'\n", optarg);
                return 2;
            }
        }
284 285 286 287 288 289 290 291 292 293
        }
    }

    cpu_flags_eff = av_get_cpu_flags();

    if (cpu_flags_eff < 0)
        return 3;

    print_cpu_flags(cpu_flags_raw, "raw");
    print_cpu_flags(cpu_flags_eff, "effective");
294
    fprintf(stderr, "threads = %s (cpu_count = %d)\n", threads, cpu_count);
295

296 297 298 299
    return 0;
}

#endif