src/misc/cpu.c

   1 /*****************************************************************************
   2  * cpu.c: CPU detection code
   3  *****************************************************************************
   4  * Copyright (C) 1998-2004 VLC authors and VideoLAN
   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 it
  12  * under the terms of the GNU Lesser General Public License as published by
  13  * the Free Software Foundation; either version 2.1 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 Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public License
  22  * along with this program; if not, write to the Free Software Foundation,
  23  * 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 #include <vlc_memstream.h>
  36 #include "libvlc.h"
  37
  38 #include <assert.h>
  39
  40 #ifndef __linux__
  41 #include <sys/types.h>
  42 #ifndef _WIN32
  43 #include <unistd.h>
  44 #include <sys/wait.h>
  45 #include <signal.h>
  46 #else
  47 #include <errno.h>
  48 #endif
  49
  50 #ifdef __APPLE__
  51 #include <sys/sysctl.h>
  52 #endif
  53 #ifdef __ANDROID__
  54 #include <cpu-features.h>
  55 #endif
  56
  57 #if defined(__OpenBSD__) && defined(__powerpc__)
  58 #include <sys/param.h>
  59 #include <sys/sysctl.h>
  60 #include <machine/cpu.h>
  61 #endif
  62
  63 static uint32_t cpu_flags;
  64
  65 #if defined (__i386__) || defined (__x86_64__) || defined (__powerpc__) \
  66  || defined (__ppc__) || defined (__ppc64__) || defined (__powerpc64__)
  67 # if defined (HAVE_FORK)
  68 static bool vlc_CPU_check (const char *name, void (*func) (void))
  69 {
  70     pid_t pid = fork();
  71
  72     switch (pid)
  73     {
  74         case 0:
  75             signal (SIGILL, SIG_DFL);
  76             func ();
  77             _exit (0);
  78         case -1:
  79             return false;
  80     }
  81
  82     int status;
  83     while( waitpid( pid, &status, 0 ) == -1 );
  84
  85     if( WIFEXITED( status ) && WEXITSTATUS( status ) == 0 )
  86         return true;
  87
  88     fprintf (stderr, "Warning: your CPU has %s instructions, but not your "
  89                      "operating system.\n", name);
  90     fprintf( stderr, "         some optimizations will be disabled unless "
  91                      "you upgrade your OS\n" );
  92     return false;
  93 }
  94
  95 #if defined (CAN_COMPILE_SSE) && !defined (__SSE__)
  96 VLC_SSE static void SSE_test (void)
  97 {
  98     asm volatile ("xorps %%xmm0,%%xmm0\n" : : : "xmm0", "xmm1");
  99 }
 100 #endif
 101 #if defined (CAN_COMPILE_3DNOW)
 102 VLC_MMX static void ThreeD_Now_test (void)
 103 {
 104     asm volatile ("pfadd %%mm0,%%mm0\n" "femms\n" : : : "mm0");
 105 }
 106 #endif
 107
 108 #if defined (CAN_COMPILE_ALTIVEC)
 109 static void Altivec_test (void)
 110 {
 111     asm volatile ("mtspr 256, %0\n" "vand %%v0, %%v0, %%v0\n" : : "r" (-1));
 112 }
 113 #endif
 114
 115 #else /* _WIN32 || __OS2__ */
 116 # define vlc_CPU_check(name, func) (1)
 117 #endif
 118 #endif
 119
 120 /**
 121  * Determines the CPU capabilities and stores them in cpu_flags.
 122  * The result can be retrieved with vlc_CPU().
 123  */
 124 void vlc_CPU_init (void)
 125 {
 126     uint32_t i_capabilities = 0;
 127
 128 #if defined( __i386__ ) || defined( __x86_64__ )
 129      unsigned int i_eax, i_ebx, i_ecx, i_edx;
 130      bool b_amd;
 131
 132     /* Needed for x86 CPU capabilities detection */
 133 # if defined (__i386__) && defined (__PIC__)
 134 #  define cpuid(reg) \
 135      asm volatile ("xchgl %%ebx,%1\n\t" \
 136                    "cpuid\n\t" \
 137                    "xchgl %%ebx,%1\n\t" \
 138                    : "=a" (i_eax), "=r" (i_ebx), "=c" (i_ecx), "=d" (i_edx) \
 139                    : "a" (reg) \
 140                    : "cc");
 141 # else
 142 #  define cpuid(reg) \
 143      asm volatile ("cpuid\n\t" \
 144                    : "=a" (i_eax), "=b" (i_ebx), "=c" (i_ecx), "=d" (i_edx) \
 145                    : "a" (reg) \
 146                    : "cc");
 147 # endif
 148      /* Check if the OS really supports the requested instructions */
 149 # if defined (__i386__) && !defined (__i486__) && !defined (__i586__) \
 150   && !defined (__i686__) && !defined (__pentium4__) \
 151   && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
 152     /* check if cpuid instruction is supported */
 153     asm volatile ( "push %%ebx\n\t"
 154                    "pushf\n\t"
 155                    "pop %%eax\n\t"
 156                    "movl %%eax, %%ebx\n\t"
 157                    "xorl $0x200000, %%eax\n\t"
 158                    "push %%eax\n\t"
 159                    "popf\n\t"
 160                    "pushf\n\t"
 161                    "pop %%eax\n\t"
 162                    "movl %%ebx,%1\n\t"
 163                    "pop %%ebx\n\t"
 164                  : "=a" ( i_eax ),
 165                    "=r" ( i_ebx )
 166                  :
 167                  : "cc" );
 168
 169     if( i_eax == i_ebx )
 170         goto out;
 171 # endif
 172
 173     /* the CPU supports the CPUID instruction - get its level */
 174     cpuid( 0x00000000 );
 175
 176 # if defined (__i386__) && !defined (__i586__) \
 177   && !defined (__i686__) && !defined (__pentium4__) \
 178   && !defined (__k6__) && !defined (__athlon__) && !defined (__k8__)
 179     if( !i_eax )
 180         goto out;
 181 #endif
 182
 183     /* borrowed from mpeg2dec */
 184     b_amd = ( i_ebx == 0x68747541 ) && ( i_ecx == 0x444d4163 )
 185                     && ( i_edx == 0x69746e65 );
 186
 187     /* test for the MMX flag */
 188     cpuid( 0x00000001 );
 189 # if !defined (__MMX__)
 190     if( ! (i_edx & 0x00800000) )
 191         goto out;
 192 # endif
 193     i_capabilities |= VLC_CPU_MMX;
 194
 195     if( i_edx & 0x02000000 )
 196         i_capabilities |= VLC_CPU_MMXEXT;
 197 # if defined (CAN_COMPILE_SSE) && !defined (__SSE__)
 198     if (( i_edx & 0x02000000 ) && vlc_CPU_check ("SSE", SSE_test))
 199 # endif
 200     {
 201         /*if( i_edx & 0x02000000 )*/
 202             i_capabilities |= VLC_CPU_SSE;
 203         if (i_edx & 0x04000000)
 204             i_capabilities |= VLC_CPU_SSE2;
 205         if (i_ecx & 0x00000001)
 206             i_capabilities |= VLC_CPU_SSE3;
 207         if (i_ecx & 0x00000200)
 208             i_capabilities |= VLC_CPU_SSSE3;
 209         if (i_ecx & 0x00080000)
 210             i_capabilities |= VLC_CPU_SSE4_1;
 211         if (i_ecx & 0x00100000)
 212             i_capabilities |= VLC_CPU_SSE4_2;
 213     }
 214
 215     /* test for additional capabilities */
 216     cpuid( 0x80000000 );
 217
 218     if( i_eax < 0x80000001 )
 219         goto out;
 220
 221     /* list these additional capabilities */
 222     cpuid( 0x80000001 );
 223
 224 # if defined (CAN_COMPILE_3DNOW) && !defined (__3dNOW__)
 225     if ((i_edx & 0x80000000) && vlc_CPU_check ("3D Now!", ThreeD_Now_test))
 226 # endif
 227         i_capabilities |= VLC_CPU_3dNOW;
 228
 229     if( b_amd && ( i_edx & 0x00400000 ) )
 230         i_capabilities |= VLC_CPU_MMXEXT;
 231 out:
 232
 233 #elif defined( __powerpc__ ) || defined( __ppc__ ) || defined( __powerpc64__ ) \
 234     || defined( __ppc64__ )
 235
 236 #   if defined(__APPLE__) || defined(__OpenBSD__)
 237 #   if defined(__OpenBSD__)
 238     int selectors[2] = { CTL_MACHDEP, CPU_ALTIVEC };
 239 #   else
 240     int selectors[2] = { CTL_HW, HW_VECTORUNIT };
 241 #   endif
 242     int i_has_altivec = 0;
 243     size_t i_length = sizeof( i_has_altivec );
 244     int i_error = sysctl( selectors, 2, &i_has_altivec, &i_length, NULL, 0);
 245
 246     if( i_error == 0 && i_has_altivec != 0 )
 247         i_capabilities |= VLC_CPU_ALTIVEC;
 248
 249 #   elif defined( CAN_COMPILE_ALTIVEC )
 250     if (vlc_CPU_check ("Altivec", Altivec_test))
 251         i_capabilities |= VLC_CPU_ALTIVEC;
 252
 253 #   endif
 254
 255 #elif defined ( __arm__)
 256 # ifdef __ANDROID__
 257     if (android_getCpuFeatures() & ANDROID_CPU_ARM_FEATURE_NEON)
 258         i_capabilities |= VLC_CPU_ARM_NEON;
 259 # endif
 260
 261 #endif
 262
 263     cpu_flags = i_capabilities;
 264 }
 265
 266 /**
 267  * Retrieves pre-computed CPU capability flags
 268  */
 269 unsigned vlc_CPU (void)
 270 {
 271 /* On Windows and OS/2,
 272  * initialized from DllMain() and _DLL_InitTerm() respectively, instead */
 273 #if !defined(_WIN32) && !defined(__OS2__)
 274     static pthread_once_t once = PTHREAD_ONCE_INIT;
 275     pthread_once (&once, vlc_CPU_init);
 276 #endif
 277     return cpu_flags;
 278 }
 279 #endif
 280
 281 void vlc_CPU_dump (vlc_object_t *obj)
 282 {
 283     struct vlc_memstream stream;
 284
 285     vlc_memstream_open(&stream);
 286
 287 #if defined (__i386__) || defined (__x86_64__)
 288     if (vlc_CPU_MMX())
 289         vlc_memstream_puts(&stream, "MMX ");
 290     if (vlc_CPU_MMXEXT())
 291         vlc_memstream_puts(&stream, "MMXEXT ");
 292     if (vlc_CPU_SSE())
 293         vlc_memstream_puts(&stream, "SSE ");
 294     if (vlc_CPU_SSE2())
 295         vlc_memstream_puts(&stream, "SSE2 ");
 296     if (vlc_CPU_SSE3())
 297         vlc_memstream_puts(&stream, "SSE3 ");
 298     if (vlc_CPU_SSSE3())
 299         vlc_memstream_puts(&stream, "SSSE3 ");
 300     if (vlc_CPU_SSE4_1())
 301         vlc_memstream_puts(&stream, "SSE4.1 ");
 302     if (vlc_CPU_SSE4_2())
 303         vlc_memstream_puts(&stream, "SSE4.2 ");
 304     if (vlc_CPU_SSE4A())
 305         vlc_memstream_puts(&stream, "SSE4A ");
 306     if (vlc_CPU_AVX())
 307         vlc_memstream_puts(&stream, "AVX ");
 308     if (vlc_CPU_AVX2())
 309         vlc_memstream_puts(&stream, "AVX2 ");
 310     if (vlc_CPU_3dNOW())
 311         vlc_memstream_puts(&stream, "3DNow! ");
 312     if (vlc_CPU_XOP())
 313         vlc_memstream_puts(&stream, "XOP ");
 314     if (vlc_CPU_FMA4())
 315         vlc_memstream_puts(&stream, "FMA4 ");
 316
 317 #elif defined (__powerpc__) || defined (__ppc__) || defined (__ppc64__)
 318     if (vlc_CPU_ALTIVEC())
 319         vlc_memstream_puts(&stream, "AltiVec");
 320
 321 #elif defined (__arm__)
 322     if (vlc_CPU_ARM_NEON())
 323         vlc_memstream_puts(&stream, "ARM_NEON ");
 324
 325 #endif
 326
 327 #if HAVE_FPU
 328     vlc_memstream_puts(&stream, "FPU ");
 329 #endif
 330
 331     if (vlc_memstream_close(&stream) == 0)
 332     {
 333         msg_Dbg (obj, "CPU has capabilities %s", stream.ptr);
 334         free(stream.ptr);
 335     }
 336 }