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