common/cpu.c

   1 /*****************************************************************************
   2  * cpu.c: h264 encoder library
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Loren Merritt <lorenm@u.washington.edu>
   7  *          Laurent Aimar <fenrir@via.ecp.fr>
   8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *****************************************************************************/
  24
  25 #if defined(HAVE_PTHREAD) && defined(SYS_LINUX)
  26 #define _GNU_SOURCE
  27 #include <sched.h>
  28 #endif
  29 #ifdef SYS_BEOS
  30 #include <kernel/OS.h>
  31 #endif
  32 #if defined(SYS_MACOSX) || defined(SYS_FREEBSD)
  33 #include <sys/types.h>
  34 #include <sys/sysctl.h>
  35 #endif
  36
  37 #include "common.h"
  38 #include "cpu.h"
  39
  40 const x264_cpu_name_t x264_cpu_names[] = {
  41     {"Altivec", X264_CPU_ALTIVEC},
  42 //  {"MMX",     X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
  43     {"MMX2",    X264_CPU_MMX|X264_CPU_MMXEXT},
  44     {"MMXEXT",  X264_CPU_MMX|X264_CPU_MMXEXT},
  45 //  {"SSE",     X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE}, // there are no sse1 functions in x264
  46     {"SSE2Slow",X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE2_IS_SLOW},
  47     {"SSE2",    X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2},
  48     {"SSE2Fast",X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE2_IS_FAST},
  49     {"SSE3",    X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3},
  50     {"SSSE3",   X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3},
  51     {"PHADD",   X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_PHADD_IS_FAST},
  52     {"SSE4.1",  X264_CPU_MMX|X264_CPU_MMXEXT|X264_CPU_SSE|X264_CPU_SSE2|X264_CPU_SSE3|X264_CPU_SSSE3|X264_CPU_SSE4},
  53     {"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},
  54     {"Cache32", X264_CPU_CACHELINE_32},
  55     {"Cache64", X264_CPU_CACHELINE_64},
  56     {"SSEMisalign", X264_CPU_SSE_MISALIGN},
  57     {"Slow_mod4_stack", X264_CPU_STACK_MOD4},
  58     {"", 0},
  59 };
  60
  61
  62 #ifdef HAVE_MMX
  63 extern int  x264_cpu_cpuid_test( void );
  64 extern uint32_t  x264_cpu_cpuid( uint32_t op, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx );
  65
  66 uint32_t x264_cpu_detect( void )
  67 {
  68     uint32_t cpu = 0;
  69     uint32_t eax, ebx, ecx, edx;
  70     uint32_t vendor[4] = {0};
  71     int max_extended_cap;
  72     int cache;
  73
  74 #ifndef ARCH_X86_64
  75     if( !x264_cpu_cpuid_test() )
  76         return 0;
  77 #endif
  78
  79     x264_cpu_cpuid( 0, &eax, vendor+0, vendor+2, vendor+1 );
  80     if( eax == 0 )
  81         return 0;
  82
  83     x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
  84     if( edx&0x00800000 )
  85         cpu |= X264_CPU_MMX;
  86     else
  87         return 0;
  88     if( edx&0x02000000 )
  89         cpu |= X264_CPU_MMXEXT|X264_CPU_SSE;
  90     if( edx&0x04000000 )
  91         cpu |= X264_CPU_SSE2;
  92     if( ecx&0x00000001 )
  93         cpu |= X264_CPU_SSE3;
  94     if( ecx&0x00000200 )
  95         cpu |= X264_CPU_SSSE3;
  96     if( ecx&0x00080000 )
  97         cpu |= X264_CPU_SSE4;
  98     if( ecx&0x00100000 )
  99         cpu |= X264_CPU_SSE42;
 100
 101     if( cpu & X264_CPU_SSSE3 )
 102         cpu |= X264_CPU_SSE2_IS_FAST;
 103     if( cpu & X264_CPU_SSE4 )
 104         cpu |= X264_CPU_PHADD_IS_FAST;
 105
 106     x264_cpu_cpuid( 0x80000000, &eax, &ebx, &ecx, &edx );
 107     max_extended_cap = eax;
 108
 109     if( !strcmp((char*)vendor, "AuthenticAMD") && max_extended_cap >= 0x80000001 )
 110     {
 111         x264_cpu_cpuid( 0x80000001, &eax, &ebx, &ecx, &edx );
 112         if( edx&0x00400000 )
 113             cpu |= X264_CPU_MMXEXT;
 114         if( cpu & X264_CPU_SSE2 )
 115         {
 116             if( ecx&0x00000040 ) /* SSE4a */
 117             {
 118                 cpu |= X264_CPU_SSE2_IS_FAST;
 119                 cpu |= X264_CPU_SSE_MISALIGN;
 120                 x264_cpu_mask_misalign_sse();
 121             }
 122             else
 123                 cpu |= X264_CPU_SSE2_IS_SLOW;
 124         }
 125     }
 126
 127     if( !strcmp((char*)vendor, "GenuineIntel") )
 128     {
 129         int family, model, stepping;
 130         x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
 131         family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
 132         model  = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
 133         stepping = eax&0xf;
 134         /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and 6/14 (core1 "yonah")
 135          * theoretically support sse2, but it's significantly slower than mmx for
 136          * almost all of x264's functions, so let's just pretend they don't. */
 137         if( family==6 && (model==9 || model==13 || model==14) )
 138         {
 139             cpu &= ~(X264_CPU_SSE2|X264_CPU_SSE3);
 140             assert(!(cpu&(X264_CPU_SSSE3|X264_CPU_SSE4)));
 141         }
 142     }
 143
 144     if( (!strcmp((char*)vendor, "GenuineIntel") || !strcmp((char*)vendor, "CyrixInstead")) && !(cpu&X264_CPU_SSE42))
 145     {
 146         /* cacheline size is specified in 3 places, any of which may be missing */
 147         x264_cpu_cpuid( 1, &eax, &ebx, &ecx, &edx );
 148         cache = (ebx&0xff00)>>5; // cflush size
 149         if( !cache && max_extended_cap >= 0x80000006 )
 150         {
 151             x264_cpu_cpuid( 0x80000006, &eax, &ebx, &ecx, &edx );
 152             cache = ecx&0xff; // cacheline size
 153         }
 154         if( !cache )
 155         {
 156             // Cache and TLB Information
 157             static const char cache32_ids[] = { 0x0a, 0x0c, 0x41, 0x42, 0x43, 0x44, 0x45, 0x82, 0x83, 0x84, 0x85, 0 };
 158             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 };
 159             uint32_t buf[4];
 160             int max, i=0, j;
 161             do {
 162                 x264_cpu_cpuid( 2, buf+0, buf+1, buf+2, buf+3 );
 163                 max = buf[0]&0xff;
 164                 buf[0] &= ~0xff;
 165                 for(j=0; j<4; j++)
 166                     if( !(buf[j]>>31) )
 167                         while( buf[j] )
 168                         {
 169                             if( strchr( cache32_ids, buf[j]&0xff ) )
 170                                 cache = 32;
 171                             if( strchr( cache64_ids, buf[j]&0xff ) )
 172                                 cache = 64;
 173                             buf[j] >>= 8;
 174                         }
 175             } while( ++i < max );
 176         }
 177
 178         if( cache == 32 )
 179             cpu |= X264_CPU_CACHELINE_32;
 180         else if( cache == 64 )
 181             cpu |= X264_CPU_CACHELINE_64;
 182         else
 183             fprintf( stderr, "x264 [warning]: unable to determine cacheline size\n" );
 184     }
 185
 186 #ifdef BROKEN_STACK_ALIGNMENT
 187     cpu |= X264_CPU_STACK_MOD4;
 188 #endif
 189
 190     return cpu;
 191 }
 192
 193 #elif defined( ARCH_PPC )
 194
 195 #ifdef SYS_MACOSX
 196 #include <sys/sysctl.h>
 197 uint32_t x264_cpu_detect( void )
 198 {
 199     /* Thank you VLC */
 200     uint32_t cpu = 0;
 201     int      selectors[2] = { CTL_HW, HW_VECTORUNIT };
 202     int      has_altivec = 0;
 203     size_t   length = sizeof( has_altivec );
 204     int      error = sysctl( selectors, 2, &has_altivec, &length, NULL, 0 );
 205
 206     if( error == 0 && has_altivec != 0 )
 207     {
 208         cpu |= X264_CPU_ALTIVEC;
 209     }
 210
 211     return cpu;
 212 }
 213
 214 #elif defined( SYS_LINUX )
 215 #include <signal.h>
 216 #include <setjmp.h>
 217 static sigjmp_buf jmpbuf;
 218 static volatile sig_atomic_t canjump = 0;
 219
 220 static void sigill_handler( int sig )
 221 {
 222     if( !canjump )
 223     {
 224         signal( sig, SIG_DFL );
 225         raise( sig );
 226     }
 227
 228     canjump = 0;
 229     siglongjmp( jmpbuf, 1 );
 230 }
 231
 232 uint32_t x264_cpu_detect( void )
 233 {
 234     static void (* oldsig)( int );
 235
 236     oldsig = signal( SIGILL, sigill_handler );
 237     if( sigsetjmp( jmpbuf, 1 ) )
 238     {
 239         signal( SIGILL, oldsig );
 240         return 0;
 241     }
 242
 243     canjump = 1;
 244     asm volatile( "mtspr 256, %0\n\t"
 245                   "vand 0, 0, 0\n\t"
 246                   :
 247                   : "r"(-1) );
 248     canjump = 0;
 249
 250     signal( SIGILL, oldsig );
 251
 252     return X264_CPU_ALTIVEC;
 253 }
 254 #endif
 255
 256 #else
 257
 258 uint32_t x264_cpu_detect( void )
 259 {
 260     return 0;
 261 }
 262
 263 #endif
 264
 265 #ifndef HAVE_MMX
 266 void x264_emms( void )
 267 {
 268 }
 269 #endif
 270
 271
 272 int x264_cpu_num_processors( void )
 273 {
 274 #if !defined(HAVE_PTHREAD)
 275     return 1;
 276
 277 #elif defined(_WIN32)
 278     return pthread_num_processors_np();
 279
 280 #elif defined(SYS_LINUX)
 281     unsigned int bit;
 282     int np;
 283     cpu_set_t p_aff;
 284     memset( &p_aff, 0, sizeof(p_aff) );
 285     sched_getaffinity( 0, sizeof(p_aff), &p_aff );
 286     for( np = 0, bit = 0; bit < sizeof(p_aff); bit++ )
 287         np += (((uint8_t *)&p_aff)[bit / 8] >> (bit % 8)) & 1;
 288     return np;
 289
 290 #elif defined(SYS_BEOS)
 291     system_info info;
 292     get_system_info( &info );
 293     return info.cpu_count;
 294
 295 #elif defined(SYS_MACOSX) || defined(SYS_FREEBSD)
 296     int numberOfCPUs;
 297     size_t length = sizeof( numberOfCPUs );
 298     if( sysctlbyname("hw.ncpu", &numberOfCPUs, &length, NULL, 0) )
 299     {
 300         numberOfCPUs = 1;
 301     }
 302     return numberOfCPUs;
 303
 304 #else
 305     return 1;
 306 #endif
 307 }