cpu.c 13.1 KB
Newer Older
Laurent Aimar's avatar
Laurent Aimar committed
1
/*****************************************************************************
Fiona Glaser's avatar
Fiona Glaser committed
2
 * cpu.c: cpu detection
Laurent Aimar's avatar
Laurent Aimar committed
3
 *****************************************************************************
Loren Merritt's avatar
Loren Merritt committed
4
 * Copyright (C) 2003-2013 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.
Fiona Glaser's avatar
Fiona Glaser committed
23 24 25
 *
 * This program is also available under a commercial proprietary license.
 * For more information, contact us at licensing@x264.com.
Laurent Aimar's avatar
Laurent Aimar committed
26 27
 *****************************************************************************/

28 29 30 31
#define _GNU_SOURCE // for sched_getaffinity
#include "common.h"
#include "cpu.h"

32
#if HAVE_POSIXTHREAD && SYS_LINUX
33 34
#include <sched.h>
#endif
Steven Walters's avatar
Steven Walters committed
35
#if SYS_BEOS
36 37
#include <kernel/OS.h>
#endif
Steven Walters's avatar
Steven Walters committed
38
#if SYS_MACOSX || SYS_FREEBSD
39 40 41
#include <sys/types.h>
#include <sys/sysctl.h>
#endif
Steven Walters's avatar
Steven Walters committed
42
#if SYS_OPENBSD
43 44 45 46
#include <sys/param.h>
#include <sys/sysctl.h>
#include <machine/cpu.h>
#endif
47

48 49
const x264_cpu_name_t x264_cpu_names[] =
{
Loren Merritt's avatar
Loren Merritt committed
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64
    {"Altivec",     X264_CPU_ALTIVEC},
//  {"MMX",         X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
    {"MMX2",        X264_CPU_MMX|X264_CPU_MMX2},
    {"MMXEXT",      X264_CPU_MMX|X264_CPU_MMX2},
//  {"SSE",         X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE}, // there are no sse1 functions in x264
#define SSE2 X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE|X264_CPU_SSE2
    {"SSE2Slow",    SSE2|X264_CPU_SSE2_IS_SLOW},
    {"SSE2",        SSE2},
    {"SSE2Fast",    SSE2|X264_CPU_SSE2_IS_FAST},
    {"SSE3",        SSE2|X264_CPU_SSE3},
    {"SSSE3",       SSE2|X264_CPU_SSE3|X264_CPU_SSSE3},
    {"FastShuffle", SSE2|X264_CPU_SHUFFLE_IS_FAST},
    {"SSE4.1",      SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
    {"SSE4",        SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
    {"SSE4.2",      SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4|X264_CPU_SSE42},
65 66 67 68 69 70 71
#define AVX SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4|X264_CPU_SSE42|X264_CPU_AVX
    {"AVX",         AVX},
    {"XOP",         AVX|X264_CPU_XOP},
    {"FMA4",        AVX|X264_CPU_FMA4},
    {"AVX2",        AVX|X264_CPU_AVX2},
    {"FMA3",        AVX|X264_CPU_FMA3},
#undef AVX
Loren Merritt's avatar
Loren Merritt committed
72 73 74 75 76
#undef SSE2
    {"Cache32",         X264_CPU_CACHELINE_32},
    {"Cache64",         X264_CPU_CACHELINE_64},
    {"SSEMisalign",     X264_CPU_SSE_MISALIGN},
    {"LZCNT",           X264_CPU_LZCNT},
77 78 79
    {"BMI1",            X264_CPU_BMI1},
    {"BMI2",            X264_CPU_BMI1|X264_CPU_BMI2},
    {"TBM",             X264_CPU_TBM},
80
    {"Slow_mod4_stack", X264_CPU_STACK_MOD4},
Loren Merritt's avatar
Loren Merritt committed
81 82 83 84 85
    {"ARMv6",           X264_CPU_ARMV6},
    {"NEON",            X264_CPU_NEON},
    {"Fast_NEON_MRC",   X264_CPU_FAST_NEON_MRC},
    {"SlowCTZ",         X264_CPU_SLOW_CTZ},
    {"SlowAtom",        X264_CPU_SLOW_ATOM},
86 87 88
    {"", 0},
};

Steven Walters's avatar
Steven Walters committed
89
#if (ARCH_PPC && SYS_LINUX) || (ARCH_ARM && !HAVE_NEON)
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106
#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
107

Steven Walters's avatar
Steven Walters committed
108
#if HAVE_MMX
Anton Mitrofanov's avatar
Anton Mitrofanov committed
109
int x264_cpu_cpuid_test( void );
110
void x264_cpu_cpuid( uint32_t op, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx );
Fiona Glaser's avatar
Fiona Glaser committed
111
void x264_cpu_xgetbv( uint32_t op, uint32_t *eax, uint32_t *edx );
Laurent Aimar's avatar
Laurent Aimar committed
112 113 114 115 116

uint32_t x264_cpu_detect( void )
{
    uint32_t cpu = 0;
    uint32_t eax, ebx, ecx, edx;
117
    uint32_t vendor[4] = {0};
118
    uint32_t max_extended_cap;
119
    int cache;
Laurent Aimar's avatar
Laurent Aimar committed
120

Steven Walters's avatar
Steven Walters committed
121
#if !ARCH_X86_64
Laurent Aimar's avatar
Laurent Aimar committed
122 123
    if( !x264_cpu_cpuid_test() )
        return 0;
124
#endif
Laurent Aimar's avatar
Laurent Aimar committed
125

126
    x264_cpu_cpuid( 0, &eax, vendor+0, vendor+2, vendor+1 );
Laurent Aimar's avatar
Laurent Aimar committed
127 128 129 130
    if( eax == 0 )
        return 0;

    x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
131 132 133
    if( edx&0x00800000 )
        cpu |= X264_CPU_MMX;
    else
Laurent Aimar's avatar
Laurent Aimar committed
134
        return 0;
135
    if( edx&0x02000000 )
Loren Merritt's avatar
Loren Merritt committed
136
        cpu |= X264_CPU_MMX2|X264_CPU_SSE;
137
    if( edx&0x04000000 )
Laurent Aimar's avatar
Laurent Aimar committed
138
        cpu |= X264_CPU_SSE2;
139
    if( ecx&0x00000001 )
140
        cpu |= X264_CPU_SSE3;
141
    if( ecx&0x00000200 )
142
        cpu |= X264_CPU_SSSE3;
143 144
    if( ecx&0x00080000 )
        cpu |= X264_CPU_SSE4;
145 146
    if( ecx&0x00100000 )
        cpu |= X264_CPU_SSE42;
147 148 149 150 151 152
    /* Check OXSAVE and AVX bits */
    if( (ecx&0x18000000) == 0x18000000 )
    {
        /* Check for OS support */
        x264_cpu_xgetbv( 0, &eax, &edx );
        if( (eax&0x6) == 0x6 )
153
        {
154
            cpu |= X264_CPU_AVX;
155 156 157 158 159 160 161 162 163 164 165 166 167 168
            if( ecx&0x00001000 )
                cpu |= X264_CPU_FMA3;
        }
    }

    x264_cpu_cpuid( 7, &eax, &ebx, &ecx, &edx );
    /* AVX2 requires OS support, but BMI1/2 don't. */
    if( (cpu&X264_CPU_AVX) && (ebx&0x00000020) )
        cpu |= X264_CPU_AVX2;
    if( ebx&0x00000008 )
    {
        cpu |= X264_CPU_BMI1;
        if( ebx&0x00000100 )
            cpu |= X264_CPU_BMI2;
169
    }
Laurent Aimar's avatar
Laurent Aimar committed
170

171 172 173
    if( cpu & X264_CPU_SSSE3 )
        cpu |= X264_CPU_SSE2_IS_FAST;
    if( cpu & X264_CPU_SSE4 )
Fiona Glaser's avatar
Fiona Glaser committed
174
        cpu |= X264_CPU_SHUFFLE_IS_FAST;
175

Laurent Aimar's avatar
Laurent Aimar committed
176
    x264_cpu_cpuid( 0x80000000, &eax, &ebx, &ecx, &edx );
177 178 179
    max_extended_cap = eax;

    if( !strcmp((char*)vendor, "AuthenticAMD") && max_extended_cap >= 0x80000001 )
Laurent Aimar's avatar
Laurent Aimar committed
180
    {
181
        cpu |= X264_CPU_SLOW_CTZ;
182 183
        x264_cpu_cpuid( 0x80000001, &eax, &ebx, &ecx, &edx );
        if( edx&0x00400000 )
Loren Merritt's avatar
Loren Merritt committed
184
            cpu |= X264_CPU_MMX2;
185 186 187
        if( cpu & X264_CPU_SSE2 )
        {
            if( ecx&0x00000040 ) /* SSE4a */
Fiona Glaser's avatar
Fiona Glaser committed
188
            {
189
                cpu |= X264_CPU_SSE2_IS_FAST;
190
                cpu |= X264_CPU_LZCNT;
Fiona Glaser's avatar
Fiona Glaser committed
191
                cpu |= X264_CPU_SHUFFLE_IS_FAST;
192
                cpu &= ~X264_CPU_SLOW_CTZ;
Fiona Glaser's avatar
Fiona Glaser committed
193
            }
194 195
            else
                cpu |= X264_CPU_SSE2_IS_SLOW;
196 197 198 199 200 201

            if( ecx&0x00000080 ) /* Misalign SSE */
            {
                cpu |= X264_CPU_SSE_MISALIGN;
                x264_cpu_mask_misalign_sse();
            }
202 203 204 205 206 207 208 209

            if( cpu & X264_CPU_AVX )
            {
                if( ecx&0x00000800 ) /* XOP */
                    cpu |= X264_CPU_XOP;
                if( ecx&0x00010000 ) /* FMA4 */
                    cpu |= X264_CPU_FMA4;
            }
210 211 212

            if( ecx&0x00200000 )
                cpu |= X264_CPU_TBM;
213
        }
Laurent Aimar's avatar
Laurent Aimar committed
214 215
    }

216
    if( !strcmp((char*)vendor, "GenuineIntel") )
Laurent Aimar's avatar
Laurent Aimar committed
217
    {
218
        x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
219 220
        int family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
        int model  = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
221
        if( family == 6 )
222
        {
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
            /* 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. */
            if( model == 9 || model == 13 || model == 14 )
            {
                cpu &= ~(X264_CPU_SSE2|X264_CPU_SSE3);
                assert(!(cpu&(X264_CPU_SSSE3|X264_CPU_SSE4)));
            }
            /* Detect Atom CPU */
            else if( model == 28 )
            {
                cpu |= X264_CPU_SLOW_ATOM;
                cpu |= X264_CPU_SLOW_CTZ;
            }
            /* Some Penryns and Nehalems are pointlessly crippled (SSE4 disabled), so
             * detect them here. */
            else if( model >= 23 )
                cpu |= X264_CPU_SHUFFLE_IS_FAST;
241
        }
Laurent Aimar's avatar
Laurent Aimar committed
242
    }
243

244
    if( (!strcmp((char*)vendor, "GenuineIntel") || !strcmp((char*)vendor, "CyrixInstead")) && !(cpu&X264_CPU_SSE42))
Laurent Aimar's avatar
Laurent Aimar committed
245
    {
246 247 248 249 250 251 252 253 254 255 256 257
        /* 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 };
258 259
            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 };
260
            uint32_t buf[4];
261
            int max, i = 0;
262 263 264 265
            do {
                x264_cpu_cpuid( 2, buf+0, buf+1, buf+2, buf+3 );
                max = buf[0]&0xff;
                buf[0] &= ~0xff;
266
                for( int j = 0; j < 4; j++ )
267 268 269 270 271 272 273 274 275 276 277
                    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
278

279 280 281 282 283
        if( cache == 32 )
            cpu |= X264_CPU_CACHELINE_32;
        else if( cache == 64 )
            cpu |= X264_CPU_CACHELINE_64;
        else
284
            x264_log( NULL, X264_LOG_WARNING, "unable to determine cacheline size\n" );
285
    }
286

Steven Walters's avatar
Steven Walters committed
287
#if BROKEN_STACK_ALIGNMENT
288 289 290
    cpu |= X264_CPU_STACK_MOD4;
#endif

Laurent Aimar's avatar
Laurent Aimar committed
291 292 293
    return cpu;
}

Steven Walters's avatar
Steven Walters committed
294
#elif ARCH_PPC
Laurent Aimar's avatar
Laurent Aimar committed
295

Steven Walters's avatar
Steven Walters committed
296
#if SYS_MACOSX || SYS_OPENBSD
297
#include <sys/sysctl.h>
Laurent Aimar's avatar
Laurent Aimar committed
298 299
uint32_t x264_cpu_detect( void )
{
300
    /* Thank you VLC */
Laurent Aimar's avatar
Laurent Aimar committed
301
    uint32_t cpu = 0;
Steven Walters's avatar
Steven Walters committed
302
#if SYS_OPENBSD
303 304
    int      selectors[2] = { CTL_MACHDEP, CPU_ALTIVEC };
#else
Laurent Aimar's avatar
Laurent Aimar committed
305
    int      selectors[2] = { CTL_HW, HW_VECTORUNIT };
306
#endif
Laurent Aimar's avatar
Laurent Aimar committed
307 308 309 310 311 312 313 314 315 316
    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;
}

Steven Walters's avatar
Steven Walters committed
317
#elif SYS_LINUX
318

319 320
uint32_t x264_cpu_detect( void )
{
321
    static void (*oldsig)( int );
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338

    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 );

339 340 341 342
    return X264_CPU_ALTIVEC;
}
#endif

Steven Walters's avatar
Steven Walters committed
343
#elif ARCH_ARM
344

345 346
void x264_cpu_neon_test( void );
int x264_cpu_fast_neon_mrc_test( void );
347 348 349 350

uint32_t x264_cpu_detect( void )
{
    int flags = 0;
Steven Walters's avatar
Steven Walters committed
351
#if HAVE_ARMV6
352 353 354
    flags |= X264_CPU_ARMV6;

    // don't do this hack if compiled with -mfpu=neon
Steven Walters's avatar
Steven Walters committed
355
#if !HAVE_NEON
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
    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.
379 380
    // right now Apple does not seem to support performance counters for this test
#ifndef __MACH__
381
    flags |= x264_cpu_fast_neon_mrc_test() ? X264_CPU_FAST_NEON_MRC : 0;
382
#endif
383 384 385 386 387
    // 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
388 389 390 391 392 393 394
#else

uint32_t x264_cpu_detect( void )
{
    return 0;
}

395 396
#endif

397 398
int x264_cpu_num_processors( void )
{
399
#if !HAVE_THREAD
400 401
    return 1;

402
#elif SYS_WINDOWS
403
    return x264_pthread_num_processors_np();
404

405
#elif SYS_CYGWIN || SYS_SunOS
406 407
    return sysconf( _SC_NPROCESSORS_ONLN );

Steven Walters's avatar
Steven Walters committed
408
#elif SYS_LINUX
409
    cpu_set_t p_aff;
410 411 412 413 414 415 416 417 418 419 420
    memset( &p_aff, 0, sizeof(p_aff) );
    if( sched_getaffinity( 0, sizeof(p_aff), &p_aff ) )
        return 1;
#if HAVE_CPU_COUNT
    return CPU_COUNT(&p_aff);
#else
    int np = 0;
    for( unsigned int bit = 0; bit < 8 * sizeof(p_aff); bit++ )
        np += (((uint8_t *)&p_aff)[bit / 8] >> (bit % 8)) & 1;
    return np;
#endif
421

Steven Walters's avatar
Steven Walters committed
422
#elif SYS_BEOS
423 424 425 426
    system_info info;
    get_system_info( &info );
    return info.cpu_count;

Steven Walters's avatar
Steven Walters committed
427
#elif SYS_MACOSX || SYS_FREEBSD || SYS_OPENBSD
Anton Mitrofanov's avatar
Anton Mitrofanov committed
428 429
    int ncpu;
    size_t length = sizeof( ncpu );
Steven Walters's avatar
Steven Walters committed
430
#if SYS_OPENBSD
431
    int mib[2] = { CTL_HW, HW_NCPU };
Anton Mitrofanov's avatar
Anton Mitrofanov committed
432
    if( sysctl(mib, 2, &ncpu, &length, NULL, 0) )
433
#else
Anton Mitrofanov's avatar
Anton Mitrofanov committed
434
    if( sysctlbyname("hw.ncpu", &ncpu, &length, NULL, 0) )
435
#endif
436
    {
Anton Mitrofanov's avatar
Anton Mitrofanov committed
437
        ncpu = 1;
438
    }
Anton Mitrofanov's avatar
Anton Mitrofanov committed
439
    return ncpu;
440

441
#else
442 443
    return 1;
#endif
444
}