Mercurial > mplayer.hg
annotate libvo/aclib.c @ 5428:a43b00b28081
fixed 10l bug: using ds instead of dp, and adding some debug prints
| author | arpi |
|---|---|
| date | Sun, 31 Mar 2002 20:09:17 +0000 |
| parents | b08228af4098 |
| children | c75f75806af1 |
| rev | line source |
|---|---|
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
1 #include "../config.h" |
| 3393 | 2 #ifdef USE_FASTMEMCPY |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
3 |
|
3077
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
4 /* |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
5 aclib - advanced C library ;) |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
6 This file contains functions which improve and expand standard C-library |
| 3393 | 7 see aclib_template.c ... this file only contains runtime cpu detection and config options stuff |
| 8 runtime cpu detection by michael niedermayer (michaelni@gmx.at) is under GPL | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
9 */ |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
10 #include <stddef.h> |
| 3393 | 11 #include "../cpudetect.h" |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
12 |
|
3077
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
13 #define BLOCK_SIZE 4096 |
|
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
14 #define CONFUSION_FACTOR 0 |
|
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
15 //Feel free to fine-tune the above 2, it might be possible to get some speedup with them :) |
|
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
16 |
|
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
17 //#define STATISTICS |
| 3393 | 18 #ifdef ARCH_X86 |
| 19 #define CAN_COMPILE_X86_ASM | |
| 20 #endif | |
| 21 | |
| 22 //Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one | |
| 23 //Plain C versions | |
| 24 //#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) | |
| 25 //#define COMPILE_C | |
| 26 //#endif | |
| 27 | |
| 28 #ifdef CAN_COMPILE_X86_ASM | |
| 29 | |
| 30 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
| 31 #define COMPILE_MMX | |
| 32 #endif | |
| 33 | |
| 5208 | 34 #if (defined (HAVE_MMX2) && !defined (HAVE_SSE2)) || defined (RUNTIME_CPUDETECT) |
| 3393 | 35 #define COMPILE_MMX2 |
| 36 #endif | |
| 37 | |
| 38 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
| 39 #define COMPILE_3DNOW | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
40 #endif |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
41 |
| 5208 | 42 #if defined (HAVE_SSE2) || defined (RUNTIME_CPUDETECT) |
| 43 #define COMPILE_SSE | |
| 44 #endif | |
| 45 | |
| 3393 | 46 #undef HAVE_MMX |
| 47 #undef HAVE_MMX2 | |
| 48 #undef HAVE_3DNOW | |
| 5208 | 49 #undef HAVE_SSE |
| 50 #undef HAVE_SSE2 | |
| 3393 | 51 #undef ARCH_X86 |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
52 /* |
| 3393 | 53 #ifdef COMPILE_C |
| 54 #undef HAVE_MMX | |
| 55 #undef HAVE_MMX2 | |
| 56 #undef HAVE_3DNOW | |
| 57 #undef ARCH_X86 | |
| 58 #define RENAME(a) a ## _C | |
| 59 #include "aclib_template.c" | |
| 60 #endif | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
61 */ |
| 3393 | 62 //MMX versions |
| 63 #ifdef COMPILE_MMX | |
| 64 #undef RENAME | |
| 65 #define HAVE_MMX | |
| 66 #undef HAVE_MMX2 | |
| 67 #undef HAVE_3DNOW | |
| 5208 | 68 #undef HAVE_SSE |
| 69 #undef HAVE_SSE2 | |
| 3393 | 70 #define ARCH_X86 |
| 71 #define RENAME(a) a ## _MMX | |
| 72 #include "aclib_template.c" | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
73 #endif |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
74 |
| 3393 | 75 //MMX2 versions |
| 76 #ifdef COMPILE_MMX2 | |
| 77 #undef RENAME | |
| 78 #define HAVE_MMX | |
| 79 #define HAVE_MMX2 | |
| 80 #undef HAVE_3DNOW | |
| 5208 | 81 #undef HAVE_SSE |
| 82 #undef HAVE_SSE2 | |
| 3393 | 83 #define ARCH_X86 |
| 84 #define RENAME(a) a ## _MMX2 | |
| 85 #include "aclib_template.c" | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
86 #endif |
|
3077
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
87 |
| 3393 | 88 //3DNOW versions |
| 89 #ifdef COMPILE_3DNOW | |
| 90 #undef RENAME | |
| 91 #define HAVE_MMX | |
| 92 #undef HAVE_MMX2 | |
| 93 #define HAVE_3DNOW | |
| 5208 | 94 #undef HAVE_SSE |
| 95 #undef HAVE_SSE2 | |
| 3393 | 96 #define ARCH_X86 |
| 97 #define RENAME(a) a ## _3DNow | |
| 98 #include "aclib_template.c" | |
|
3077
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
99 #endif |
|
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
100 |
| 5208 | 101 //SSE versions (only used on SSE2 cpus) |
| 102 #ifdef COMPILE_SSE | |
| 103 #undef RENAME | |
| 104 #define HAVE_MMX | |
| 105 #define HAVE_MMX2 | |
| 106 #undef HAVE_3DNOW | |
| 107 #define HAVE_SSE | |
| 108 #define HAVE_SSE2 | |
| 109 #define ARCH_X86 | |
| 110 #define RENAME(a) a ## _SSE | |
| 111 #include "aclib_template.c" | |
| 112 #endif | |
| 113 | |
| 3393 | 114 #endif // CAN_COMPILE_X86_ASM |
| 115 | |
|
3077
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
116 |
| 3393 | 117 inline void * fast_memcpy(void * to, const void * from, size_t len) |
| 118 { | |
| 119 #ifdef RUNTIME_CPUDETECT | |
| 120 #ifdef CAN_COMPILE_X86_ASM | |
| 121 // ordered per speed fasterst first | |
| 5208 | 122 if(gCpuCaps.hasSSE2) |
| 123 fast_memcpy_SSE(to, from, len); | |
| 124 else if(gCpuCaps.hasMMX2) | |
| 3393 | 125 fast_memcpy_MMX2(to, from, len); |
| 126 else if(gCpuCaps.has3DNow) | |
| 127 fast_memcpy_3DNow(to, from, len); | |
| 128 else if(gCpuCaps.hasMMX) | |
| 129 fast_memcpy_MMX(to, from, len); | |
| 130 else | |
| 131 #endif //CAN_COMPILE_X86_ASM | |
| 132 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 133 #else | |
| 5208 | 134 #ifdef HAVE_SSE2 |
| 135 fast_memcpy_SSE(to, from, len); | |
| 136 #elif defined (HAVE_MMX2) | |
| 3393 | 137 fast_memcpy_MMX2(to, from, len); |
| 138 #elif defined (HAVE_3DNOW) | |
| 139 fast_memcpy_3DNow(to, from, len); | |
| 140 #elif defined (HAVE_MMX) | |
| 141 fast_memcpy_MMX(to, from, len); | |
| 142 #else | |
| 143 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
144 #endif |
|
3077
99f6db3255aa
10-20% faster fastmemcpy :) on my p3 at least but the algo is mostly from "amd athlon processor x86 code optimization guide" so it should be faster for amd chips too, but i fear it might be slower for mem->vram copies (someone should check that, i cant) ... there are 2 #defines to finetune it (BLOCK_SIZE & CONFUSION_FACTOR)
michael
parents:
1123
diff
changeset
|
145 |
| 3393 | 146 #endif //!RUNTIME_CPUDETECT |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
147 } |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
148 |
| 4681 | 149 inline void * mem2agpcpy(void * to, const void * from, size_t len) |
| 150 { | |
| 151 #ifdef RUNTIME_CPUDETECT | |
| 152 #ifdef CAN_COMPILE_X86_ASM | |
| 153 // ordered per speed fasterst first | |
| 5208 | 154 if(gCpuCaps.hasSSE2) |
| 155 mem2agpcpy_SSE(to, from, len); | |
| 156 else if(gCpuCaps.hasMMX2) | |
| 4681 | 157 mem2agpcpy_MMX2(to, from, len); |
| 158 else if(gCpuCaps.has3DNow) | |
| 159 mem2agpcpy_3DNow(to, from, len); | |
| 160 else if(gCpuCaps.hasMMX) | |
| 161 mem2agpcpy_MMX(to, from, len); | |
| 162 else | |
| 163 #endif //CAN_COMPILE_X86_ASM | |
| 164 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 165 #else | |
| 5208 | 166 #ifdef HAVE_SSE2 |
| 167 mem2agpcpy_SSE(to, from, len); | |
| 168 #elif defined (HAVE_MMX2) | |
| 4681 | 169 mem2agpcpy_MMX2(to, from, len); |
| 170 #elif defined (HAVE_3DNOW) | |
| 171 mem2agpcpy_3DNow(to, from, len); | |
| 172 #elif defined (HAVE_MMX) | |
| 173 mem2agpcpy_MMX(to, from, len); | |
| 174 #else | |
| 175 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 176 #endif | |
| 177 | |
| 178 #endif //!RUNTIME_CPUDETECT | |
| 179 } | |
| 180 | |
| 181 #endif /* use fastmemcpy */ | |
| 182 |