src/misc/cpu.c

   1 /*****************************************************************************
   2  * cpu.c: CPU detection code
   3  *****************************************************************************
   4  * Copyright (C) 1998-2004 the VideoLAN team
   5  * $Id$
   6  *
   7  * Authors: Samuel Hocevar <sam@zoy.org>
   8  *          Christophe Massiot <massiot@via.ecp.fr>
   9  *          Eugenio Jarosiewicz <ej0@cise.ufl.eduEujenio>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License as published by
  13  * the Free Software Foundation; either version 2 of the License, or
  14  * (at your option) any later version.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  * GNU General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public License
  22  * along with this program; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
  24  *****************************************************************************/
  25
  26 /*****************************************************************************
  27  * Preamble
  28  *****************************************************************************/
  29 #ifdef HAVE_CONFIG_H
  30 # include "config.h"
  31 #endif
  32
  33 #include <vlc_common.h>
  34 #include <vlc_cpu.h>
  35
  36 #include <sys/types.h>
  37 #ifndef WIN32
  38 #include <unistd.h>
  39 #include <sys/wait.h>
  40 #include <signal.h>
  41 #else
  42 #include <errno.h>
  43 #endif
  44 #include <assert.h>
  45
  46 #include "libvlc.h"
  47
  48 #if defined(__APPLE__)
  49 #include <sys/sysctl.h>
  50 #endif
  51
  52 #if defined(__SunOS)
  53 #include <unistd.h>
  54 #include <sys/types.h>
  55 #include <sys/processor.h>
  56 #include <sys/pset.h>
  57 #endif
  58
  59 #if defined( __i386__ ) || defined( __x86_64__ ) || defined( __powerpc__ ) \
  60  || defined( __ppc__ ) || defined( __ppc64__ ) || defined( __powerpc64__ )
  61 # ifndef WIN32
  62 static bool check_OS_capability( const char *psz_capability, pid_t pid )
  63 {
  64     int status;
  65
  66     if( pid == -1 )
  67         return false; /* fail safe :-/ */
  68
  69     while( waitpid( pid, &status, 0 ) == -1 );
  70
  71     if( WIFEXITED( status ) && WEXITSTATUS( status ) == 0 )
  72         return true;
  73
  74     fprintf( stderr, "warning: your CPU has %s instructions, but not your "
  75                      "operating system.\n", psz_capability );
  76     fprintf( stderr, "         some optimizations will be disabled unless "
  77                      "you upgrade your OS\n" );
  78     return false;
  79 }
  80
  81 #  define check_capability(name, flag, code)   \
  82      do {                                      \
  83         pid_t pid = fork();                    \
  84         if( pid == 0 )                         \
  85         {                                      \
  86             signal(SIGILL, SIG_DFL);           \
  87             __asm__ __volatile__ ( code : : ); \
  88             _exit(0);                          \
  89         }                                      \
  90         if( check_OS_capability((name), pid )) \
  91             i_capabilities |= (flag);          \
  92      } while(0)
  93
  94 # else /* WIN32 */
  95 #  define check_capability(name, flag, code)   \
  96         i_capabilities |= (flag);
  97 # endif
  98 #endif
  99
 100 /*****************************************************************************
 101  * CPUCapabilities: get the CPU capabilities
 102  *****************************************************************************
 103  * This function is called to list extensions the CPU may have.
 104  *****************************************************************************/
 105 uint32_t CPUCapabilities( void )
 106 {
 107     uint32_t i_capabilities = 0;
 108
 109 #if defined( __i386__ ) || defined( __x86_64__ )
 110      unsigned int i_eax, i_ebx, i_ecx, i_edx;
 111      bool b_amd;
 112
 113     /* Needed for x86 CPU capabilities detection */
 114 #   if defined( __x86_64__ )
 115 #       define cpuid( reg )                    \
 116             asm volatile ( "cpuid\n\t"         \
 117                            "movl %%ebx,%1\n\t" \
 118                          : "=a" ( i_eax ),     \
 119                            "=b" ( i_ebx ),     \
 120                            "=c" ( i_ecx ),     \
 121                            "=d" ( i_edx )      \
 122                          : "a"  ( reg )        \
 123                          : "cc" );
 124 #   else
 125 #       define cpuid( reg )                    \
 126             asm volatile ( "push %%ebx\n\t"    \
 127                            "cpuid\n\t"         \
 128                            "movl %%ebx,%1\n\t" \
 129                            "pop %%ebx\n\t"     \
 130                          : "=a" ( i_eax ),     \
 131                            "=r" ( i_ebx ),     \
 132                            "=c" ( i_ecx ),     \
 133                            "=d" ( i_edx )      \
 134                          : "a"  ( reg )        \
 135                          : "cc" );
 136 #   endif
 137      /* Check if the OS really supports the requested instructions */
 138 # if defined (__i386__) && !defined (__i486__) && !defined (__i586__) \
 139   && !defined (__i686__) && !defined (__pentium4__) \
 140   && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
 141     /* check if cpuid instruction is supported */
 142     asm volatile ( "push %%ebx\n\t"
 143                    "pushf\n\t"
 144                    "pop %%eax\n\t"
 145                    "movl %%eax, %%ebx\n\t"
 146                    "xorl $0x200000, %%eax\n\t"
 147                    "push %%eax\n\t"
 148                    "popf\n\t"
 149                    "pushf\n\t"
 150                    "pop %%eax\n\t"
 151                    "movl %%ebx,%1\n\t"
 152                    "pop %%ebx\n\t"
 153                  : "=a" ( i_eax ),
 154                    "=r" ( i_ebx )
 155                  :
 156                  : "cc" );
 157
 158     if( i_eax == i_ebx )
 159         goto out;
 160 # endif
 161
 162     /* the CPU supports the CPUID instruction - get its level */
 163     cpuid( 0x00000000 );
 164
 165 # if defined (__i386__) && !defined (__i586__) \
 166   && !defined (__i686__) && !defined (__pentium4__) \
 167   && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
 168     if( !i_eax )
 169         goto out;
 170 #endif
 171
 172     /* borrowed from mpeg2dec */
 173     b_amd = ( i_ebx == 0x68747541 ) && ( i_ecx == 0x444d4163 )
 174                     && ( i_edx == 0x69746e65 );
 175
 176     /* test for the MMX flag */
 177     cpuid( 0x00000001 );
 178 # if !defined (__MMX__)
 179     if( ! (i_edx & 0x00800000) )
 180         goto out;
 181 # endif
 182     i_capabilities |= CPU_CAPABILITY_MMX;
 183
 184 # if defined (__SSE__)
 185     i_capabilities |= CPU_CAPABILITY_MMXEXT | CPU_CAPABILITY_SSE;
 186 # else
 187     if( i_edx & 0x02000000 )
 188     {
 189         i_capabilities |= CPU_CAPABILITY_MMXEXT;
 190
 191 #   ifdef CAN_COMPILE_SSE
 192         check_capability( "SSE", CPU_CAPABILITY_SSE,
 193                           "xorps %%xmm0,%%xmm0\n" );
 194 #   endif
 195     }
 196 # endif
 197
 198 # if defined (__SSE2__)
 199     i_capabilities |= CPU_CAPABILITY_SSE2;
 200 # elif defined (CAN_COMPILE_SSE2)
 201     if( i_edx & 0x04000000 )
 202         check_capability( "SSE2", CPU_CAPABILITY_SSE2,
 203                           "movupd %%xmm0, %%xmm0\n" );
 204 # endif
 205
 206 # if defined (__SSE3__)
 207     i_capabilities |= CPU_CAPABILITY_SSE3;
 208 # elif defined (CAN_COMPILE_SSE3)
 209     if( i_ecx & 0x00000001 )
 210         check_capability( "SSE3", CPU_CAPABILITY_SSE3,
 211                           "movsldup %%xmm1, %%xmm0\n" );
 212 # endif
 213
 214 # if defined (__SSSE3__)
 215     i_capabilities |= CPU_CAPABILITY_SSSE3;
 216 # elif defined (CAN_COMPILE_SSSE3)
 217     if( i_ecx & 0x00000200 )
 218         check_capability( "SSSE3", CPU_CAPABILITY_SSSE3,
 219                           "pabsw %%xmm1, %%xmm0\n" );
 220 # endif
 221
 222 # if defined (__SSE4_1__)
 223     i_capabilities |= CPU_CAPABILITY_SSE4_1;
 224 # elif defined (CAN_COMPILE_SSE4_1)
 225     if( i_ecx & 0x00080000 )
 226         check_capability( "SSE4.1", CPU_CAPABILITY_SSE4_1,
 227                           "pmaxsb %%xmm1, %%xmm0\n" );
 228 # endif
 229
 230 # if defined (__SSE4_2__)
 231     i_capabilities |= CPU_CAPABILITY_SSE4_2;
 232 # elif defined (CAN_COMPILE_SSE4_2)
 233     if( i_ecx & 0x00100000 )
 234         check_capability( "SSE4.2", CPU_CAPABILITY_SSE4_2,
 235                           "pcmpgtq %%xmm1, %%xmm0\n" );
 236 # endif
 237
 238     /* test for additional capabilities */
 239     cpuid( 0x80000000 );
 240
 241     if( i_eax < 0x80000001 )
 242         goto out;
 243
 244     /* list these additional capabilities */
 245     cpuid( 0x80000001 );
 246
 247 # if defined (__3dNOW__)
 248     i_capabilities |= CPU_CAPABILITY_3DNOW;
 249 # elif defined (CAN_COMPILE_3DNOW)
 250     if( i_edx & 0x80000000 )
 251         check_capability( "3D Now!", CPU_CAPABILITY_3DNOW,
 252                           "pfadd %%mm0,%%mm0\n" "femms\n" );
 253 # endif
 254
 255     if( b_amd && ( i_edx & 0x00400000 ) )
 256     {
 257         i_capabilities |= CPU_CAPABILITY_MMXEXT;
 258     }
 259 out:
 260
 261 #elif defined( __arm__ )
 262 #   if defined( __ARM_NEON__ )
 263     i_capabilities |= CPU_CAPABILITY_NEON;
 264 #   endif
 265
 266 #elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __powerpc64__ ) \
 267     || defined( __ppc64__ )
 268
 269 #   if defined(__APPLE__)
 270     int selectors[2] = { CTL_HW, HW_VECTORUNIT };
 271     int i_has_altivec = 0;
 272     size_t i_length = sizeof( i_has_altivec );
 273     int i_error = sysctl( selectors, 2, &i_has_altivec, &i_length, NULL, 0);
 274
 275     if( i_error == 0 && i_has_altivec != 0 )
 276         i_capabilities |= CPU_CAPABILITY_ALTIVEC;
 277
 278 #   elif defined( CAN_COMPILE_ALTIVEC )
 279     pid_t pid = fork();
 280     if( pid == 0 )
 281     {
 282         signal(SIGILL, SIG_DFL);
 283         asm volatile ("mtspr 256, %0\n\t"
 284                       "vand %%v0, %%v0, %%v0"
 285                       :
 286                       : "r" (-1));
 287         _exit(0);
 288     }
 289
 290     if( check_OS_capability( "Altivec", pid ) )
 291         i_capabilities |= CPU_CAPABILITY_ALTIVEC;
 292
 293 #   endif
 294
 295 #endif
 296     return i_capabilities;
 297 }
 298
 299 uint32_t cpu_flags = 0;
 300
 301
 302 /*****************************************************************************
 303  * vlc_CPU: get pre-computed CPU capability flags
 304  ****************************************************************************/
 305 unsigned vlc_CPU (void)
 306 {
 307     return cpu_flags;
 308 }
 309
 310 const struct
 311 {
 312     uint32_t value;
 313     char name[12];
 314 } cap_dirs[] = {
 315 #if defined ( __i386__ ) || defined ( __x86_64__ )
 316     { CPU_CAPABILITY_MMX,     "mmx" },
 317     { CPU_CAPABILITY_MMXEXT,  "mmxext" },
 318     { CPU_CAPABILITY_3DNOW,   "3dnow" },
 319     { CPU_CAPABILITY_SSE,     "sse" },
 320 #endif
 321 #if defined (__ppc__) || defined (__ppc64__) || defined (__powerpc__)
 322     { CPU_CAPABILITY_ALTIVEC, "altivec" },
 323 #endif
 324 #if defined (__arm__)
 325     { CPU_CAPABILITY_NEON,    "arm_neon" },
 326 #endif
 327 };
 328
 329 /**
 330  * Return the number of available logical CPU.
 331  */
 332 unsigned vlc_GetCPUCount(void)
 333 {
 334 #if defined(WIN32) && !defined(UNDER_CE)
 335     DWORD process_mask;
 336     DWORD system_mask;
 337     if (!GetProcessAffinityMask(GetCurrentProcess(), &process_mask, &system_mask))
 338         return 1;
 339
 340     unsigned count = 0;
 341     while (system_mask) {
 342         count++;
 343         system_mask >>= 1;
 344     }
 345     return count;
 346 #elif defined(HAVE_SCHED_GETAFFINITY)
 347     cpu_set_t cpu;
 348     CPU_ZERO(&cpu);
 349     if (sched_getaffinity(0, sizeof(cpu), &cpu) < 0)
 350         return 1;
 351     unsigned count = 0;
 352     for (unsigned i = 0; i < CPU_SETSIZE; i++)
 353         count += CPU_ISSET(i, &cpu) != 0;
 354     return count;
 355 #elif defined(__APPLE__)
 356     int count;
 357     size_t size = sizeof(count) ;
 358     if (sysctlbyname("hw.ncpu", &count, &size, NULL, 0))
 359         return 1; /* Failure */
 360     return count;
 361 #elif defined(__SunOS)
 362     unsigned count = 0;
 363     int type;
 364     u_int numcpus;
 365     processorid_t *cpulist;
 366     processor_info_t cpuinfo;
 367     cpulist = malloc(sizeof(processorid_t) * sysconf(_SC_NPROCESSORS_MAX));
 368     if (!cpulist) return 1;
 369     if (pset_info(PS_MYID, &type, &numcpus, cpulist)==0)
 370     {
 371         for (u_int i = 0; i < numcpus; i++)
 372         {
 373             if (!processor_info(cpulist[i], &cpuinfo))
 374                 count += (cpuinfo.pi_state == P_ONLINE)?1:0;
 375         }
 376     } else {
 377         count = sysconf(_SC_NPROCESSORS_ONLN);
 378     }
 379     free(cpulist);
 380     return (count>0)?count:1;
 381 #else
 382 #   warning "vlc_GetCPUCount is not implemented for your platform"
 383     return 1;
 384 #endif
 385 }
 386
 387 /**
 388  * Check if a directory name contains usable plugins w.r.t. the hardware
 389  * capabilities. Loading a plugin when the hardware has insufficient
 390  * capabilities may lead to illegal instructions (SIGILL) and must be avoided.
 391  *
 392  * @param name the name of the directory (<b>not</b> the path)
 393  *
 394  * @return true if the hardware has sufficient capabilities or the directory
 395  * does not require any special capability; false if the running hardware has
 396  * insufficient capabilities.
 397  */
 398 bool vlc_CPU_CheckPluginDir (const char *name)
 399 {
 400     const unsigned flags = vlc_CPU ();
 401     for (size_t i = 0; i < sizeof (cap_dirs) / sizeof (cap_dirs[0]); i++)
 402     {
 403         if (strcmp (name, cap_dirs[i].name))
 404             continue;
 405         return (flags & cap_dirs[i].value) != 0;
 406     }
 407     return true;
 408 }
 409
 410 static vlc_memcpy_t pf_vlc_memcpy = memcpy;
 411 static vlc_memset_t pf_vlc_memset = memset;
 412
 413 void vlc_fastmem_register (vlc_memcpy_t cpy, vlc_memset_t set)
 414 {
 415     if (cpy)
 416         pf_vlc_memcpy = cpy;
 417     if (set)
 418         pf_vlc_memset = set;
 419 }
 420
 421 /**
 422  * vlc_memcpy: fast CPU-dependent memcpy
 423  */
 424 void *vlc_memcpy (void *tgt, const void *src, size_t n)
 425 {
 426     return pf_vlc_memcpy (tgt, src, n);
 427 }
 428
 429 /**
 430  * vlc_memset: fast CPU-dependent memset
 431  */
 432 void *vlc_memset (void *tgt, int c, size_t n)
 433 {
 434     return pf_vlc_memset (tgt, c, n);
 435 }
 436
 437 /**
 438  * Returned an aligned pointer on newly allocated memory.
 439  * \param alignment must be a power of 2 and a multiple of sizeof(void*)
 440  * \param size is the size of the usable memory returned.
 441  *
 442  * It must not be freed directly, *base must.
 443  */
 444 void *vlc_memalign(void **base, size_t alignment, size_t size)
 445 {
 446     assert(alignment >= sizeof(void*));
 447     for (size_t t = alignment; t > 1; t >>= 1)
 448         assert((t&1) == 0);
 449 #if defined(HAVE_POSIX_MEMALIGN)
 450     if (posix_memalign(base, alignment, size)) {
 451         *base = NULL;
 452         return NULL;
 453     }
 454     return *base;
 455 #elif defined(HAVE_MEMALIGN)
 456     return *base = memalign(alignment, size);
 457 #else
 458     unsigned char *p = *base = malloc(size + alignment - 1);
 459     if (!p)
 460         return NULL;
 461     return (void*)((uintptr_t)(p + alignment - 1) & ~(alignment - 1));
 462 #endif
 463 }
 464