libvo/aclib.c

   1 /*
   2  * aclib - advanced C library ;)
   3  * Functions which improve and expand the standard C library, see aclib_template.c.
   4  * This file only contains runtime CPU detection and config option stuff.
   5  * runtime CPU detection by Michael Niedermayer (michaelni@gmx.at)
   6  *
   7  * This file is part of MPlayer.
   8  *
   9  * MPlayer is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * MPlayer is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with MPlayer; if not, write to the Free Software Foundation, Inc.,
  21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22  */
  23
  24 #include "config.h"
  25 #include <stddef.h>
  26 #include <string.h>
  27 #include "cpudetect.h"
  28 #include "fastmemcpy.h"
  29 #undef memcpy
  30
  31 #define BLOCK_SIZE 4096
  32 #define CONFUSION_FACTOR 0
  33 //Feel free to fine-tune the above 2, it might be possible to get some speedup with them :)
  34
  35 //#define STATISTICS
  36 #if ARCH_X86
  37 #define CAN_COMPILE_X86_ASM
  38 #endif
  39
  40 //Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
  41 //Plain C versions
  42 //#if !HAVE_MMX || defined (RUNTIME_CPUDETECT)
  43 //#define COMPILE_C
  44 //#endif
  45
  46 #ifdef CAN_COMPILE_X86_ASM
  47
  48 #if (HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
  49 #define COMPILE_MMX
  50 #endif
  51
  52 #if (HAVE_MMX2 && !HAVE_SSE2) || defined (RUNTIME_CPUDETECT)
  53 #define COMPILE_MMX2
  54 #endif
  55
  56 #if (HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
  57 #define COMPILE_3DNOW
  58 #endif
  59
  60 #if HAVE_SSE2 || defined (RUNTIME_CPUDETECT)
  61 #define COMPILE_SSE
  62 #endif
  63
  64 #undef HAVE_MMX
  65 #undef HAVE_MMX2
  66 #undef HAVE_AMD3DNOW
  67 #undef HAVE_SSE
  68 #undef HAVE_SSE2
  69 #define HAVE_MMX 0
  70 #define HAVE_MMX2 0
  71 #define HAVE_AMD3DNOW 0
  72 #define HAVE_SSE 0
  73 #define HAVE_SSE2 0
  74 /*
  75 #ifdef COMPILE_C
  76 #undef HAVE_MMX
  77 #undef HAVE_MMX2
  78 #undef HAVE_AMD3DNOW
  79 #undef HAVE_SSE
  80 #undef HAVE_SSE2
  81 #define HAVE_MMX 0
  82 #define HAVE_MMX2 0
  83 #define HAVE_AMD3DNOW 0
  84 #define HAVE_SSE 0
  85 #define HAVE_SSE2 0
  86 #define RENAME(a) a ## _C
  87 #include "aclib_template.c"
  88 #endif
  89 */
  90 //MMX versions
  91 #ifdef COMPILE_MMX
  92 #undef RENAME
  93 #undef HAVE_MMX
  94 #undef HAVE_MMX2
  95 #undef HAVE_AMD3DNOW
  96 #undef HAVE_SSE
  97 #undef HAVE_SSE2
  98 #define HAVE_MMX 1
  99 #define HAVE_MMX2 0
 100 #define HAVE_AMD3DNOW 0
 101 #define HAVE_SSE 0
 102 #define HAVE_SSE2 0
 103 #define RENAME(a) a ## _MMX
 104 #include "aclib_template.c"
 105 #endif
 106
 107 //MMX2 versions
 108 #ifdef COMPILE_MMX2
 109 #undef RENAME
 110 #undef HAVE_MMX
 111 #undef HAVE_MMX2
 112 #undef HAVE_AMD3DNOW
 113 #undef HAVE_SSE
 114 #undef HAVE_SSE2
 115 #define HAVE_MMX 1
 116 #define HAVE_MMX2 1
 117 #define HAVE_AMD3DNOW 0
 118 #define HAVE_SSE 0
 119 #define HAVE_SSE2 0
 120 #define RENAME(a) a ## _MMX2
 121 #include "aclib_template.c"
 122 #endif
 123
 124 //3DNOW versions
 125 #ifdef COMPILE_3DNOW
 126 #undef RENAME
 127 #undef HAVE_MMX
 128 #undef HAVE_MMX2
 129 #undef HAVE_AMD3DNOW
 130 #undef HAVE_SSE
 131 #undef HAVE_SSE2
 132 #define HAVE_MMX 1
 133 #define HAVE_MMX2 0
 134 #define HAVE_AMD3DNOW 1
 135 #define HAVE_SSE 0
 136 #define HAVE_SSE2 0
 137 #define RENAME(a) a ## _3DNow
 138 #include "aclib_template.c"
 139 #endif
 140
 141 //SSE versions (only used on SSE2 cpus)
 142 #ifdef COMPILE_SSE
 143 #undef RENAME
 144 #undef HAVE_MMX
 145 #undef HAVE_MMX2
 146 #undef HAVE_AMD3DNOW
 147 #undef HAVE_SSE
 148 #undef HAVE_SSE2
 149 #define HAVE_MMX 1
 150 #define HAVE_MMX2 1
 151 #define HAVE_AMD3DNOW 0
 152 #define HAVE_SSE 1
 153 #define HAVE_SSE2 1
 154 #define RENAME(a) a ## _SSE
 155 #include "aclib_template.c"
 156 #endif
 157
 158 #endif // CAN_COMPILE_X86_ASM
 159
 160
 161 #undef fast_memcpy
 162 void * fast_memcpy(void * to, const void * from, size_t len)
 163 {
 164 #ifdef RUNTIME_CPUDETECT
 165 #ifdef CAN_COMPILE_X86_ASM
 166         // ordered per speed fasterst first
 167         if(gCpuCaps.hasSSE2)
 168                 fast_memcpy_SSE(to, from, len);
 169         else if(gCpuCaps.hasMMX2)
 170                 fast_memcpy_MMX2(to, from, len);
 171         else if(gCpuCaps.has3DNow)
 172                 fast_memcpy_3DNow(to, from, len);
 173         else if(gCpuCaps.hasMMX)
 174                 fast_memcpy_MMX(to, from, len);
 175         else
 176 #endif //CAN_COMPILE_X86_ASM
 177                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 178 #else
 179 #if HAVE_SSE2
 180                 fast_memcpy_SSE(to, from, len);
 181 #elif HAVE_MMX2
 182                 fast_memcpy_MMX2(to, from, len);
 183 #elif HAVE_AMD3DNOW
 184                 fast_memcpy_3DNow(to, from, len);
 185 #elif HAVE_MMX
 186                 fast_memcpy_MMX(to, from, len);
 187 #else
 188                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 189 #endif
 190
 191 #endif //!RUNTIME_CPUDETECT
 192         return to;
 193 }
 194
 195 #undef  mem2agpcpy
 196 void * mem2agpcpy(void * to, const void * from, size_t len)
 197 {
 198 #ifdef RUNTIME_CPUDETECT
 199 #ifdef CAN_COMPILE_X86_ASM
 200         // ordered per speed fasterst first
 201         if(gCpuCaps.hasSSE2)
 202                 mem2agpcpy_SSE(to, from, len);
 203         else if(gCpuCaps.hasMMX2)
 204                 mem2agpcpy_MMX2(to, from, len);
 205         else if(gCpuCaps.has3DNow)
 206                 mem2agpcpy_3DNow(to, from, len);
 207         else if(gCpuCaps.hasMMX)
 208                 mem2agpcpy_MMX(to, from, len);
 209         else
 210 #endif //CAN_COMPILE_X86_ASM
 211                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 212 #else
 213 #if HAVE_SSE2
 214                 mem2agpcpy_SSE(to, from, len);
 215 #elif HAVE_MMX2
 216                 mem2agpcpy_MMX2(to, from, len);
 217 #elif HAVE_AMD3DNOW
 218                 mem2agpcpy_3DNow(to, from, len);
 219 #elif HAVE_MMX
 220                 mem2agpcpy_MMX(to, from, len);
 221 #else
 222                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 223 #endif
 224
 225 #endif //!RUNTIME_CPUDETECT
 226         return to;
 227 }