Mercurial > mplayer.hg
annotate libvo/aclib.c @ 12404:5ca00576a26c
more wishes
| author | diego |
|---|---|
| date | Mon, 03 May 2004 01:27:48 +0000 |
| parents | e7153e62a7f4 |
| children | 4b8417674f1c |
| 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" |
|
8123
9fc45fe0d444
*HUGE* set of compiler warning fixes, unused variables removal
arpi
parents:
7072
diff
changeset
|
12 #include "fastmemcpy.h" |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
13 |
|
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
|
14 #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
|
15 #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
|
16 //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
|
17 |
|
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
|
18 //#define STATISTICS |
| 3393 | 19 #ifdef ARCH_X86 |
| 20 #define CAN_COMPILE_X86_ASM | |
| 21 #endif | |
| 22 | |
| 23 //Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one | |
| 24 //Plain C versions | |
| 25 //#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) | |
| 26 //#define COMPILE_C | |
| 27 //#endif | |
| 28 | |
| 29 #ifdef CAN_COMPILE_X86_ASM | |
| 30 | |
| 31 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
| 32 #define COMPILE_MMX | |
| 33 #endif | |
| 34 | |
| 5208 | 35 #if (defined (HAVE_MMX2) && !defined (HAVE_SSE2)) || defined (RUNTIME_CPUDETECT) |
| 3393 | 36 #define COMPILE_MMX2 |
| 37 #endif | |
| 38 | |
| 39 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
| 40 #define COMPILE_3DNOW | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
41 #endif |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
42 |
| 5208 | 43 #if defined (HAVE_SSE2) || defined (RUNTIME_CPUDETECT) |
| 44 #define COMPILE_SSE | |
| 45 #endif | |
| 46 | |
| 3393 | 47 #undef HAVE_MMX |
| 48 #undef HAVE_MMX2 | |
| 49 #undef HAVE_3DNOW | |
| 5208 | 50 #undef HAVE_SSE |
| 51 #undef HAVE_SSE2 | |
| 3393 | 52 #undef ARCH_X86 |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
53 /* |
| 3393 | 54 #ifdef COMPILE_C |
| 55 #undef HAVE_MMX | |
| 56 #undef HAVE_MMX2 | |
| 57 #undef HAVE_3DNOW | |
| 58 #undef ARCH_X86 | |
| 59 #define RENAME(a) a ## _C | |
| 60 #include "aclib_template.c" | |
| 61 #endif | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
62 */ |
| 3393 | 63 //MMX versions |
| 64 #ifdef COMPILE_MMX | |
| 65 #undef RENAME | |
| 66 #define HAVE_MMX | |
| 67 #undef HAVE_MMX2 | |
| 68 #undef HAVE_3DNOW | |
| 5208 | 69 #undef HAVE_SSE |
| 70 #undef HAVE_SSE2 | |
| 3393 | 71 #define ARCH_X86 |
| 72 #define RENAME(a) a ## _MMX | |
| 73 #include "aclib_template.c" | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
74 #endif |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
75 |
| 3393 | 76 //MMX2 versions |
| 77 #ifdef COMPILE_MMX2 | |
| 78 #undef RENAME | |
| 79 #define HAVE_MMX | |
| 80 #define HAVE_MMX2 | |
| 81 #undef HAVE_3DNOW | |
| 5208 | 82 #undef HAVE_SSE |
| 83 #undef HAVE_SSE2 | |
| 3393 | 84 #define ARCH_X86 |
| 85 #define RENAME(a) a ## _MMX2 | |
| 86 #include "aclib_template.c" | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
87 #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
|
88 |
| 3393 | 89 //3DNOW versions |
| 90 #ifdef COMPILE_3DNOW | |
| 91 #undef RENAME | |
| 92 #define HAVE_MMX | |
| 93 #undef HAVE_MMX2 | |
| 94 #define HAVE_3DNOW | |
| 5208 | 95 #undef HAVE_SSE |
| 96 #undef HAVE_SSE2 | |
| 3393 | 97 #define ARCH_X86 |
| 98 #define RENAME(a) a ## _3DNow | |
| 99 #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
|
100 #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
|
101 |
| 5208 | 102 //SSE versions (only used on SSE2 cpus) |
| 103 #ifdef COMPILE_SSE | |
| 104 #undef RENAME | |
| 105 #define HAVE_MMX | |
| 106 #define HAVE_MMX2 | |
| 107 #undef HAVE_3DNOW | |
| 108 #define HAVE_SSE | |
| 109 #define HAVE_SSE2 | |
| 110 #define ARCH_X86 | |
| 111 #define RENAME(a) a ## _SSE | |
| 112 #include "aclib_template.c" | |
| 113 #endif | |
| 114 | |
| 3393 | 115 #endif // CAN_COMPILE_X86_ASM |
| 116 | |
|
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
|
117 |
| 7072 | 118 void * fast_memcpy(void * to, const void * from, size_t len) |
| 3393 | 119 { |
| 120 #ifdef RUNTIME_CPUDETECT | |
| 121 #ifdef CAN_COMPILE_X86_ASM | |
| 122 // ordered per speed fasterst first | |
| 5208 | 123 if(gCpuCaps.hasSSE2) |
| 124 fast_memcpy_SSE(to, from, len); | |
| 125 else if(gCpuCaps.hasMMX2) | |
| 3393 | 126 fast_memcpy_MMX2(to, from, len); |
| 127 else if(gCpuCaps.has3DNow) | |
| 128 fast_memcpy_3DNow(to, from, len); | |
| 129 else if(gCpuCaps.hasMMX) | |
| 130 fast_memcpy_MMX(to, from, len); | |
| 131 else | |
| 132 #endif //CAN_COMPILE_X86_ASM | |
| 133 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 134 #else | |
| 5208 | 135 #ifdef HAVE_SSE2 |
| 136 fast_memcpy_SSE(to, from, len); | |
| 137 #elif defined (HAVE_MMX2) | |
| 3393 | 138 fast_memcpy_MMX2(to, from, len); |
| 139 #elif defined (HAVE_3DNOW) | |
| 140 fast_memcpy_3DNow(to, from, len); | |
| 141 #elif defined (HAVE_MMX) | |
| 142 fast_memcpy_MMX(to, from, len); | |
| 143 #else | |
| 144 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
|
145 #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
|
146 |
| 3393 | 147 #endif //!RUNTIME_CPUDETECT |
|
5543
c75f75806af1
memcpy must return destination ptr patch by Adam <adam@cfar.umd.edu>
michael
parents:
5208
diff
changeset
|
148 return to; |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
149 } |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
150 |
|
8127
e7153e62a7f4
On non-x86 platforms, memcpy was re-implemented in mplayer and was called
jkeil
parents:
8123
diff
changeset
|
151 #undef mem2agpcpy |
| 7072 | 152 void * mem2agpcpy(void * to, const void * from, size_t len) |
| 4681 | 153 { |
| 154 #ifdef RUNTIME_CPUDETECT | |
| 155 #ifdef CAN_COMPILE_X86_ASM | |
| 156 // ordered per speed fasterst first | |
| 5208 | 157 if(gCpuCaps.hasSSE2) |
| 158 mem2agpcpy_SSE(to, from, len); | |
| 159 else if(gCpuCaps.hasMMX2) | |
| 4681 | 160 mem2agpcpy_MMX2(to, from, len); |
| 161 else if(gCpuCaps.has3DNow) | |
| 162 mem2agpcpy_3DNow(to, from, len); | |
| 163 else if(gCpuCaps.hasMMX) | |
| 164 mem2agpcpy_MMX(to, from, len); | |
| 165 else | |
| 166 #endif //CAN_COMPILE_X86_ASM | |
| 167 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 168 #else | |
| 5208 | 169 #ifdef HAVE_SSE2 |
| 170 mem2agpcpy_SSE(to, from, len); | |
| 171 #elif defined (HAVE_MMX2) | |
| 4681 | 172 mem2agpcpy_MMX2(to, from, len); |
| 173 #elif defined (HAVE_3DNOW) | |
| 174 mem2agpcpy_3DNow(to, from, len); | |
| 175 #elif defined (HAVE_MMX) | |
| 176 mem2agpcpy_MMX(to, from, len); | |
| 177 #else | |
| 178 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 179 #endif | |
| 180 | |
| 181 #endif //!RUNTIME_CPUDETECT | |
|
8123
9fc45fe0d444
*HUGE* set of compiler warning fixes, unused variables removal
arpi
parents:
7072
diff
changeset
|
182 return to; |
| 4681 | 183 } |
| 184 | |
| 185 #endif /* use fastmemcpy */ | |
| 186 |