cpu.c 11.4 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1 2 3
/*****************************************************************************
 * cpu.c: h264 encoder library
 *****************************************************************************
4
 * Copyright (C) 2003-2008 x264 project
Laurent Aimar's avatar
Laurent Aimar committed
5
 *
6 7
 * Authors: Loren Merritt <lorenm@u.washington.edu>
 *          Laurent Aimar <fenrir@via.ecp.fr>
8
 *          Fiona Glaser <fiona@x264.com>
Laurent Aimar's avatar
Laurent Aimar committed
9 10 11 12 13 14 15 16 17 18 19 20 21
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
Laurent Aimar's avatar
Laurent Aimar committed
23 24
 *****************************************************************************/

25 26 27 28
#define _GNU_SOURCE // for sched_getaffinity
#include "common.h"
#include "cpu.h"

29 30 31
#if defined(HAVE_PTHREAD) && defined(SYS_LINUX)
#include <sched.h>
#endif
32 33 34
#ifdef SYS_BEOS
#include <kernel/OS.h>
#endif
35
#if defined(SYS_MACOSX) || defined(SYS_FREEBSD)
36 37 38
#include <sys/types.h>
#include <sys/sysctl.h>
#endif
39 40 41 42 43
#ifdef SYS_OPENBSD
#include <sys/param.h>
#include <sys/sysctl.h>
#include <machine/cpu.h>
#endif
44

45 46 47
const x264_cpu_name_t x264_cpu_names[] = {
    {"Altivec", X264_CPU_ALTIVEC},
//  {"MMX",     X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
48 49
    {"MMX2",    X264_CPU_MMX|X264_CPU_MMXEXT},
    {"MMXEXT",  X264_CPU_MMX|X264_CPU_MMXEXT},
50 51
//  {"SSE",     X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE}, // there are no sse1 functions in x264
    {"SSE2Slow",X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE2_IS_SLOW},
52
    {"SSE2",    X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2},
53
    {"SSE2Fast",X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE2_IS_FAST},
54 55
    {"SSE3",    X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3},
    {"SSSE3",   X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3},
Fiona Glaser's avatar
Fiona Glaser committed
56
    {"FastShuffle",   X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SHUFFLE_IS_FAST},
57 58
    {"SSE4.1",  X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
    {"SSE4.2",  X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4|X264_CPU_SSE42},
59 60
    {"Cache32", X264_CPU_CACHELINE_32},
    {"Cache64", X264_CPU_CACHELINE_64},
Fiona Glaser's avatar
Fiona Glaser committed
61
    {"SSEMisalign", X264_CPU_SSE_MISALIGN},
62
    {"LZCNT", X264_CPU_LZCNT},
63
    {"Slow_mod4_stack", X264_CPU_STACK_MOD4},
64 65 66
    {"ARMv6", X264_CPU_ARMV6},
    {"NEON",  X264_CPU_NEON},
    {"Fast_NEON_MRC",  X264_CPU_FAST_NEON_MRC},
67 68
    {"SlowCTZ", X264_CPU_SLOW_CTZ},
    {"SlowAtom", X264_CPU_SLOW_ATOM},
69 70 71
    {"", 0},
};

72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89
#if (defined(ARCH_PPC) && defined(SYS_LINUX)) || (defined(ARCH_ARM) && !defined(HAVE_NEON))
#include <signal.h>
#include <setjmp.h>
static sigjmp_buf jmpbuf;
static volatile sig_atomic_t canjump = 0;

static void sigill_handler( int sig )
{
    if( !canjump )
    {
        signal( sig, SIG_DFL );
        raise( sig );
    }

    canjump = 0;
    siglongjmp( jmpbuf, 1 );
}
#endif
Fiona Glaser's avatar
Fiona Glaser committed
90

Loren Merritt's avatar
Loren Merritt committed
91
#ifdef HAVE_MMX
Anton Mitrofanov's avatar
Anton Mitrofanov committed
92 93
int x264_cpu_cpuid_test( void );
uint32_t x264_cpu_cpuid( uint32_t op, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx );
Laurent Aimar's avatar
Laurent Aimar committed
94 95 96 97 98

uint32_t x264_cpu_detect( void )
{
    uint32_t cpu = 0;
    uint32_t eax, ebx, ecx, edx;
99 100 101
    uint32_t vendor[4] = {0};
    int max_extended_cap;
    int cache;
Laurent Aimar's avatar
Laurent Aimar committed
102

103
#ifndef ARCH_X86_64
Laurent Aimar's avatar
Laurent Aimar committed
104 105
    if( !x264_cpu_cpuid_test() )
        return 0;
106
#endif
Laurent Aimar's avatar
Laurent Aimar committed
107

108
    x264_cpu_cpuid( 0, &eax, vendor+0, vendor+2, vendor+1 );
Laurent Aimar's avatar
Laurent Aimar committed
109 110 111 112
    if( eax == 0 )
        return 0;

    x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
113 114 115
    if( edx&0x00800000 )
        cpu |= X264_CPU_MMX;
    else
Laurent Aimar's avatar
Laurent Aimar committed
116
        return 0;
117
    if( edx&0x02000000 )
Laurent Aimar's avatar
Laurent Aimar committed
118
        cpu |= X264_CPU_MMXEXT|X264_CPU_SSE;
119
    if( edx&0x04000000 )
Laurent Aimar's avatar
Laurent Aimar committed
120
        cpu |= X264_CPU_SSE2;
121
    if( ecx&0x00000001 )
122
        cpu |= X264_CPU_SSE3;
123
    if( ecx&0x00000200 )
124
        cpu |= X264_CPU_SSSE3;
125 126
    if( ecx&0x00080000 )
        cpu |= X264_CPU_SSE4;
127 128
    if( ecx&0x00100000 )
        cpu |= X264_CPU_SSE42;
Laurent Aimar's avatar
Laurent Aimar committed
129

130 131 132
    if( cpu & X264_CPU_SSSE3 )
        cpu |= X264_CPU_SSE2_IS_FAST;
    if( cpu & X264_CPU_SSE4 )
Fiona Glaser's avatar
Fiona Glaser committed
133
        cpu |= X264_CPU_SHUFFLE_IS_FAST;
134

Laurent Aimar's avatar
Laurent Aimar committed
135
    x264_cpu_cpuid( 0x80000000, &eax, &ebx, &ecx, &edx );
136 137 138
    max_extended_cap = eax;

    if( !strcmp((char*)vendor, "AuthenticAMD") && max_extended_cap >= 0x80000001 )
Laurent Aimar's avatar
Laurent Aimar committed
139
    {
140
        cpu |= X264_CPU_SLOW_CTZ;
141 142 143
        x264_cpu_cpuid( 0x80000001, &eax, &ebx, &ecx, &edx );
        if( edx&0x00400000 )
            cpu |= X264_CPU_MMXEXT;
144 145 146
        if( cpu & X264_CPU_SSE2 )
        {
            if( ecx&0x00000040 ) /* SSE4a */
Fiona Glaser's avatar
Fiona Glaser committed
147
            {
148
                cpu |= X264_CPU_SSE2_IS_FAST;
149
                cpu |= X264_CPU_LZCNT;
Fiona Glaser's avatar
Fiona Glaser committed
150
                cpu |= X264_CPU_SHUFFLE_IS_FAST;
151
                cpu &= ~X264_CPU_SLOW_CTZ;
Fiona Glaser's avatar
Fiona Glaser committed
152
            }
153 154
            else
                cpu |= X264_CPU_SSE2_IS_SLOW;
155 156 157 158 159 160

            if( ecx&0x00000080 ) /* Misalign SSE */
            {
                cpu |= X264_CPU_SSE_MISALIGN;
                x264_cpu_mask_misalign_sse();
            }
161
        }
Laurent Aimar's avatar
Laurent Aimar committed
162 163
    }

164
    if( !strcmp((char*)vendor, "GenuineIntel") )
Laurent Aimar's avatar
Laurent Aimar committed
165
    {
166
        x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
167 168
        int family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
        int model  = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
169 170 171
        /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and 6/14 (core1 "yonah")
         * theoretically support sse2, but it's significantly slower than mmx for
         * almost all of x264's functions, so let's just pretend they don't. */
172
        if( family == 6 && (model == 9 || model == 13 || model == 14) )
173 174 175 176
        {
            cpu &= ~(X264_CPU_SSE2|X264_CPU_SSE3);
            assert(!(cpu&(X264_CPU_SSSE3|X264_CPU_SSE4)));
        }
177 178 179 180 181 182
        /* Detect Atom CPU */
        if( family == 6 && model == 28 )
        {
            cpu |= X264_CPU_SLOW_ATOM;
            cpu |= X264_CPU_SLOW_CTZ;
        }
Laurent Aimar's avatar
Laurent Aimar committed
183
    }
184

185
    if( (!strcmp((char*)vendor, "GenuineIntel") || !strcmp((char*)vendor, "CyrixInstead")) && !(cpu&X264_CPU_SSE42))
Laurent Aimar's avatar
Laurent Aimar committed
186
    {
187 188 189 190 191 192 193 194 195 196 197 198
        /* cacheline size is specified in 3 places, any of which may be missing */
        x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
        cache = (ebx&0xff00)>>5; // cflush size
        if( !cache && max_extended_cap >= 0x80000006 )
        {
            x264_cpu_cpuid( 0x80000006, &eax, &ebx, &ecx, &edx );
            cache = ecx&0xff; // cacheline size
        }
        if( !cache )
        {
            // Cache and TLB Information
            static const char cache32_ids[] = { 0x0a, 0x0c, 0x41, 0x42, 0x43, 0x44, 0x45, 0x82, 0x83, 0x84, 0x85, 0 };
199 200
            static const char cache64_ids[] = { 0x22, 0x23, 0x25, 0x29, 0x2c, 0x46, 0x47, 0x49, 0x60, 0x66, 0x67,
                                                0x68, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7c, 0x7f, 0x86, 0x87, 0 };
201
            uint32_t buf[4];
202
            int max, i = 0;
203 204 205 206
            do {
                x264_cpu_cpuid( 2, buf+0, buf+1, buf+2, buf+3 );
                max = buf[0]&0xff;
                buf[0] &= ~0xff;
207
                for( int j = 0; j < 4; j++ )
208 209 210 211 212 213 214 215 216 217 218
                    if( !(buf[j]>>31) )
                        while( buf[j] )
                        {
                            if( strchr( cache32_ids, buf[j]&0xff ) )
                                cache = 32;
                            if( strchr( cache64_ids, buf[j]&0xff ) )
                                cache = 64;
                            buf[j] >>= 8;
                        }
            } while( ++i < max );
        }
Laurent Aimar's avatar
Laurent Aimar committed
219

220 221 222 223 224
        if( cache == 32 )
            cpu |= X264_CPU_CACHELINE_32;
        else if( cache == 64 )
            cpu |= X264_CPU_CACHELINE_64;
        else
225
            x264_log( NULL, X264_LOG_WARNING, "unable to determine cacheline size\n" );
226
    }
227

228 229 230 231
#ifdef BROKEN_STACK_ALIGNMENT
    cpu |= X264_CPU_STACK_MOD4;
#endif

Laurent Aimar's avatar
Laurent Aimar committed
232 233 234
    return cpu;
}

235
#elif defined( ARCH_PPC )
Laurent Aimar's avatar
Laurent Aimar committed
236

237
#if defined(SYS_MACOSX) || defined(SYS_OPENBSD)
238
#include <sys/sysctl.h>
Laurent Aimar's avatar
Laurent Aimar committed
239 240
uint32_t x264_cpu_detect( void )
{
241
    /* Thank you VLC */
Laurent Aimar's avatar
Laurent Aimar committed
242
    uint32_t cpu = 0;
243 244 245
#ifdef SYS_OPENBSD
    int      selectors[2] = { CTL_MACHDEP, CPU_ALTIVEC };
#else
Laurent Aimar's avatar
Laurent Aimar committed
246
    int      selectors[2] = { CTL_HW, HW_VECTORUNIT };
247
#endif
Laurent Aimar's avatar
Laurent Aimar committed
248 249 250 251 252 253 254 255 256 257
    int      has_altivec = 0;
    size_t   length = sizeof( has_altivec );
    int      error = sysctl( selectors, 2, &has_altivec, &length, NULL, 0 );

    if( error == 0 && has_altivec != 0 )
        cpu |= X264_CPU_ALTIVEC;

    return cpu;
}

258
#elif defined( SYS_LINUX )
259

260 261
uint32_t x264_cpu_detect( void )
{
262
    static void (*oldsig)( int );
263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279

    oldsig = signal( SIGILL, sigill_handler );
    if( sigsetjmp( jmpbuf, 1 ) )
    {
        signal( SIGILL, oldsig );
        return 0;
    }

    canjump = 1;
    asm volatile( "mtspr 256, %0\n\t"
                  "vand 0, 0, 0\n\t"
                  :
                  : "r"(-1) );
    canjump = 0;

    signal( SIGILL, oldsig );

280 281 282 283
    return X264_CPU_ALTIVEC;
}
#endif

284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
#elif defined( ARCH_ARM )

void x264_cpu_neon_test();
int x264_cpu_fast_neon_mrc_test();

uint32_t x264_cpu_detect( void )
{
    int flags = 0;
#ifdef HAVE_ARMV6
    flags |= X264_CPU_ARMV6;

    // don't do this hack if compiled with -mfpu=neon
#ifndef HAVE_NEON
    static void (* oldsig)( int );
    oldsig = signal( SIGILL, sigill_handler );
    if( sigsetjmp( jmpbuf, 1 ) )
    {
        signal( SIGILL, oldsig );
        return flags;
    }

    canjump = 1;
    x264_cpu_neon_test();
    canjump = 0;
    signal( SIGILL, oldsig );
#endif

    flags |= X264_CPU_NEON;

    // fast neon -> arm (Cortex-A9) detection relies on user access to the
    // cycle counter; this assumes ARMv7 performance counters.
    // NEON requires at least ARMv7, ARMv8 may require changes here, but
    // hopefully this hacky detection method will have been replaced by then.
    // Note that there is potential for a race condition if another program or
    // x264 instance disables or reinits the counters while x264 is using them,
    // which may result in incorrect detection and the counters stuck enabled.
    flags |= x264_cpu_fast_neon_mrc_test() ? X264_CPU_FAST_NEON_MRC : 0;
    // TODO: write dual issue test? currently it's A8 (dual issue) vs. A9 (fast mrc)
#endif
    return flags;
}

Laurent Aimar's avatar
Laurent Aimar committed
326 327 328 329 330 331 332
#else

uint32_t x264_cpu_detect( void )
{
    return 0;
}

333 334
#endif

335 336
int x264_cpu_num_processors( void )
{
337 338 339
#if !defined(HAVE_PTHREAD)
    return 1;

340
#elif defined(_WIN32)
341 342 343
    return pthread_num_processors_np();

#elif defined(SYS_LINUX)
344
    unsigned int bit;
345
    int np;
346 347 348 349 350
    cpu_set_t p_aff;
    memset( &p_aff, 0, sizeof(p_aff) );
    sched_getaffinity( 0, sizeof(p_aff), &p_aff );
    for( np = 0, bit = 0; bit < sizeof(p_aff); bit++ )
        np += (((uint8_t *)&p_aff)[bit / 8] >> (bit % 8)) & 1;
351 352
    return np;

353 354 355 356 357
#elif defined(SYS_BEOS)
    system_info info;
    get_system_info( &info );
    return info.cpu_count;

358
#elif defined(SYS_MACOSX) || defined(SYS_FREEBSD) || defined(SYS_OPENBSD)
359 360
    int numberOfCPUs;
    size_t length = sizeof( numberOfCPUs );
361 362 363 364
#ifdef SYS_OPENBSD
    int mib[2] = { CTL_HW, HW_NCPU };
    if( sysctl(mib, 2, &numberOfCPUs, &length, NULL, 0) )
#else
365
    if( sysctlbyname("hw.ncpu", &numberOfCPUs, &length, NULL, 0) )
366
#endif
367 368
    {
        numberOfCPUs = 1;
369
    }
370
    return numberOfCPUs;
371

372
#else
373 374
    return 1;
#endif
375
}