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 <stdint.h>
  27 #include <string.h>
  28 #include "cpudetect.h"
  29 #include "fastmemcpy.h"
  30 #include "libavutil/x86_cpu.h"
  31 #undef memcpy
  32
  33 #define BLOCK_SIZE 4096
  34 #define CONFUSION_FACTOR 0
  35 //Feel free to fine-tune the above 2, it might be possible to get some speedup with them :)
  36
  37 //#define STATISTICS
  38
  39 //Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
  40 //Plain C versions
  41 //#if !HAVE_MMX || CONFIG_RUNTIME_CPUDETECT
  42 //#define COMPILE_C
  43 //#endif
  44
  45 #if ARCH_X86
  46
  47 #if (HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT
  48 #define COMPILE_MMX
  49 #endif
  50
  51 #if (HAVE_MMX2 && !HAVE_SSE2) || CONFIG_RUNTIME_CPUDETECT
  52 #define COMPILE_MMX2
  53 #endif
  54
  55 #if (HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT
  56 #define COMPILE_3DNOW
  57 #endif
  58
  59 #if HAVE_SSE2 || CONFIG_RUNTIME_CPUDETECT
  60 #define COMPILE_SSE
  61 #endif
  62
  63 #undef HAVE_MMX
  64 #undef HAVE_MMX2
  65 #undef HAVE_AMD3DNOW
  66 #undef HAVE_SSE
  67 #undef HAVE_SSE2
  68 #define HAVE_MMX 0
  69 #define HAVE_MMX2 0
  70 #define HAVE_AMD3DNOW 0
  71 #define HAVE_SSE 0
  72 #define HAVE_SSE2 0
  73 /*
  74 #ifdef COMPILE_C
  75 #undef HAVE_MMX
  76 #undef HAVE_MMX2
  77 #undef HAVE_AMD3DNOW
  78 #undef HAVE_SSE
  79 #undef HAVE_SSE2
  80 #define HAVE_MMX 0
  81 #define HAVE_MMX2 0
  82 #define HAVE_AMD3DNOW 0
  83 #define HAVE_SSE 0
  84 #define HAVE_SSE2 0
  85 #define RENAME(a) a ## _C
  86 #include "aclib_template.c"
  87 #endif
  88 */
  89 //MMX versions
  90 #ifdef COMPILE_MMX
  91 #undef RENAME
  92 #undef HAVE_MMX
  93 #undef HAVE_MMX2
  94 #undef HAVE_AMD3DNOW
  95 #undef HAVE_SSE
  96 #undef HAVE_SSE2
  97 #define HAVE_MMX 1
  98 #define HAVE_MMX2 0
  99 #define HAVE_AMD3DNOW 0
 100 #define HAVE_SSE 0
 101 #define HAVE_SSE2 0
 102 #define RENAME(a) a ## _MMX
 103 #include "aclib_template.c"
 104 #endif
 105
 106 //MMX2 versions
 107 #ifdef COMPILE_MMX2
 108 #undef RENAME
 109 #undef HAVE_MMX
 110 #undef HAVE_MMX2
 111 #undef HAVE_AMD3DNOW
 112 #undef HAVE_SSE
 113 #undef HAVE_SSE2
 114 #define HAVE_MMX 1
 115 #define HAVE_MMX2 1
 116 #define HAVE_AMD3DNOW 0
 117 #define HAVE_SSE 0
 118 #define HAVE_SSE2 0
 119 #define RENAME(a) a ## _MMX2
 120 #include "aclib_template.c"
 121 #endif
 122
 123 //3DNOW versions
 124 #ifdef COMPILE_3DNOW
 125 #undef RENAME
 126 #undef HAVE_MMX
 127 #undef HAVE_MMX2
 128 #undef HAVE_AMD3DNOW
 129 #undef HAVE_SSE
 130 #undef HAVE_SSE2
 131 #define HAVE_MMX 1
 132 #define HAVE_MMX2 0
 133 #define HAVE_AMD3DNOW 1
 134 #define HAVE_SSE 0
 135 #define HAVE_SSE2 0
 136 #define RENAME(a) a ## _3DNow
 137 #include "aclib_template.c"
 138 #endif
 139
 140 //SSE versions (only used on SSE2 cpus)
 141 #ifdef COMPILE_SSE
 142 #undef RENAME
 143 #undef HAVE_MMX
 144 #undef HAVE_MMX2
 145 #undef HAVE_AMD3DNOW
 146 #undef HAVE_SSE
 147 #undef HAVE_SSE2
 148 #define HAVE_MMX 1
 149 #define HAVE_MMX2 1
 150 #define HAVE_AMD3DNOW 0
 151 #define HAVE_SSE 1
 152 #define HAVE_SSE2 1
 153 #define RENAME(a) a ## _SSE
 154 #include "aclib_template.c"
 155 #endif
 156
 157 #endif /* ARCH_X86 */
 158
 159
 160 #undef fast_memcpy
 161 void * fast_memcpy(void * to, const void * from, size_t len)
 162 {
 163 #if CONFIG_RUNTIME_CPUDETECT
 164 #if ARCH_X86
 165         // ordered per speed fasterst first
 166         if(gCpuCaps.hasSSE2)
 167                 fast_memcpy_SSE(to, from, len);
 168         else if(gCpuCaps.hasMMX2)
 169                 fast_memcpy_MMX2(to, from, len);
 170         else if(gCpuCaps.has3DNow)
 171                 fast_memcpy_3DNow(to, from, len);
 172         else if(gCpuCaps.hasMMX)
 173                 fast_memcpy_MMX(to, from, len);
 174         else
 175 #endif
 176                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 177 #else
 178 #if HAVE_SSE2
 179                 fast_memcpy_SSE(to, from, len);
 180 #elif HAVE_MMX2
 181                 fast_memcpy_MMX2(to, from, len);
 182 #elif HAVE_AMD3DNOW
 183                 fast_memcpy_3DNow(to, from, len);
 184 #elif HAVE_MMX
 185                 fast_memcpy_MMX(to, from, len);
 186 #else
 187                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 188 #endif
 189
 190 #endif //!CONFIG_RUNTIME_CPUDETECT
 191         return to;
 192 }
 193
 194 #undef  mem2agpcpy
 195 void * mem2agpcpy(void * to, const void * from, size_t len)
 196 {
 197 #if CONFIG_RUNTIME_CPUDETECT
 198 #if ARCH_X86
 199         // ordered per speed fasterst first
 200         if(gCpuCaps.hasSSE2)
 201                 mem2agpcpy_SSE(to, from, len);
 202         else if(gCpuCaps.hasMMX2)
 203                 mem2agpcpy_MMX2(to, from, len);
 204         else if(gCpuCaps.has3DNow)
 205                 mem2agpcpy_3DNow(to, from, len);
 206         else if(gCpuCaps.hasMMX)
 207                 mem2agpcpy_MMX(to, from, len);
 208         else
 209 #endif
 210                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 211 #else
 212 #if HAVE_SSE2
 213                 mem2agpcpy_SSE(to, from, len);
 214 #elif HAVE_MMX2
 215                 mem2agpcpy_MMX2(to, from, len);
 216 #elif HAVE_AMD3DNOW
 217                 mem2agpcpy_3DNow(to, from, len);
 218 #elif HAVE_MMX
 219                 mem2agpcpy_MMX(to, from, len);
 220 #else
 221                 memcpy(to, from, len); // prior to mmx we use the standart memcpy
 222 #endif
 223
 224 #endif //!CONFIG_RUNTIME_CPUDETECT
 225         return to;
 226 }