Mercurial > mplayer.hg
annotate libvo/aclib.c @ 15500:41fd530bd7a9
mux extradata
| author | nicodvb |
|---|---|
| date | Tue, 17 May 2005 21:44:28 +0000 |
| parents | e047e70a9767 |
| children | 6289755ce7c7 |
| rev | line source |
|---|---|
|
12650
ac3fd2ff2561
Unify the config.h #include, use "config.h" instead of "../config.h"
diego
parents:
12492
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> |
|
13787
e047e70a9767
Handle "xxx.h" vs "../xxx.h" include paths in a consistent way.
diego
parents:
13720
diff
changeset
|
11 #include "cpudetect.h" |
|
8123
9fc45fe0d444
*HUGE* set of compiler warning fixes, unused variables removal
arpi
parents:
7072
diff
changeset
|
12 #include "fastmemcpy.h" |
|
12492
4b8417674f1c
fix crash due to fast_memcpy calling itself instead of libc memcpy
reimar
parents:
8127
diff
changeset
|
13 #undef memcpy |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
14 |
|
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
|
15 #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
|
16 #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
|
17 //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
|
18 |
|
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
|
19 //#define STATISTICS |
|
13720
821f464b4d90
adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64
aurel
parents:
12650
diff
changeset
|
20 #if defined(ARCH_X86) || defined(ARCH_X86_64) |
| 3393 | 21 #define CAN_COMPILE_X86_ASM |
| 22 #endif | |
| 23 | |
| 24 //Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one | |
| 25 //Plain C versions | |
| 26 //#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) | |
| 27 //#define COMPILE_C | |
| 28 //#endif | |
| 29 | |
| 30 #ifdef CAN_COMPILE_X86_ASM | |
| 31 | |
| 32 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
| 33 #define COMPILE_MMX | |
| 34 #endif | |
| 35 | |
| 5208 | 36 #if (defined (HAVE_MMX2) && !defined (HAVE_SSE2)) || defined (RUNTIME_CPUDETECT) |
| 3393 | 37 #define COMPILE_MMX2 |
| 38 #endif | |
| 39 | |
| 40 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT) | |
| 41 #define COMPILE_3DNOW | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
42 #endif |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
43 |
| 5208 | 44 #if defined (HAVE_SSE2) || defined (RUNTIME_CPUDETECT) |
| 45 #define COMPILE_SSE | |
| 46 #endif | |
| 47 | |
| 3393 | 48 #undef HAVE_MMX |
| 49 #undef HAVE_MMX2 | |
| 50 #undef HAVE_3DNOW | |
| 5208 | 51 #undef HAVE_SSE |
| 52 #undef HAVE_SSE2 | |
|
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 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 RENAME(a) a ## _MMX2 |
| 84 #include "aclib_template.c" | |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
85 #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
|
86 |
| 3393 | 87 //3DNOW versions |
| 88 #ifdef COMPILE_3DNOW | |
| 89 #undef RENAME | |
| 90 #define HAVE_MMX | |
| 91 #undef HAVE_MMX2 | |
| 92 #define HAVE_3DNOW | |
| 5208 | 93 #undef HAVE_SSE |
| 94 #undef HAVE_SSE2 | |
| 3393 | 95 #define RENAME(a) a ## _3DNow |
| 96 #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
|
97 #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
|
98 |
| 5208 | 99 //SSE versions (only used on SSE2 cpus) |
| 100 #ifdef COMPILE_SSE | |
| 101 #undef RENAME | |
| 102 #define HAVE_MMX | |
| 103 #define HAVE_MMX2 | |
| 104 #undef HAVE_3DNOW | |
| 105 #define HAVE_SSE | |
| 106 #define HAVE_SSE2 | |
| 107 #define RENAME(a) a ## _SSE | |
| 108 #include "aclib_template.c" | |
| 109 #endif | |
| 110 | |
| 3393 | 111 #endif // CAN_COMPILE_X86_ASM |
| 112 | |
|
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
|
113 |
| 7072 | 114 void * fast_memcpy(void * to, const void * from, size_t len) |
| 3393 | 115 { |
| 116 #ifdef RUNTIME_CPUDETECT | |
| 117 #ifdef CAN_COMPILE_X86_ASM | |
| 118 // ordered per speed fasterst first | |
| 5208 | 119 if(gCpuCaps.hasSSE2) |
| 120 fast_memcpy_SSE(to, from, len); | |
| 121 else if(gCpuCaps.hasMMX2) | |
| 3393 | 122 fast_memcpy_MMX2(to, from, len); |
| 123 else if(gCpuCaps.has3DNow) | |
| 124 fast_memcpy_3DNow(to, from, len); | |
| 125 else if(gCpuCaps.hasMMX) | |
| 126 fast_memcpy_MMX(to, from, len); | |
| 127 else | |
| 128 #endif //CAN_COMPILE_X86_ASM | |
| 129 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 130 #else | |
| 5208 | 131 #ifdef HAVE_SSE2 |
| 132 fast_memcpy_SSE(to, from, len); | |
| 133 #elif defined (HAVE_MMX2) | |
| 3393 | 134 fast_memcpy_MMX2(to, from, len); |
| 135 #elif defined (HAVE_3DNOW) | |
| 136 fast_memcpy_3DNow(to, from, len); | |
| 137 #elif defined (HAVE_MMX) | |
| 138 fast_memcpy_MMX(to, from, len); | |
| 139 #else | |
| 140 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
|
141 #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
|
142 |
| 3393 | 143 #endif //!RUNTIME_CPUDETECT |
|
5543
c75f75806af1
memcpy must return destination ptr patch by Adam <adam@cfar.umd.edu>
michael
parents:
5208
diff
changeset
|
144 return to; |
|
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
145 } |
|
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
146 |
|
8127
e7153e62a7f4
On non-x86 platforms, memcpy was re-implemented in mplayer and was called
jkeil
parents:
8123
diff
changeset
|
147 #undef mem2agpcpy |
| 7072 | 148 void * mem2agpcpy(void * to, const void * from, size_t len) |
| 4681 | 149 { |
| 150 #ifdef RUNTIME_CPUDETECT | |
| 151 #ifdef CAN_COMPILE_X86_ASM | |
| 152 // ordered per speed fasterst first | |
| 5208 | 153 if(gCpuCaps.hasSSE2) |
| 154 mem2agpcpy_SSE(to, from, len); | |
| 155 else if(gCpuCaps.hasMMX2) | |
| 4681 | 156 mem2agpcpy_MMX2(to, from, len); |
| 157 else if(gCpuCaps.has3DNow) | |
| 158 mem2agpcpy_3DNow(to, from, len); | |
| 159 else if(gCpuCaps.hasMMX) | |
| 160 mem2agpcpy_MMX(to, from, len); | |
| 161 else | |
| 162 #endif //CAN_COMPILE_X86_ASM | |
| 163 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 164 #else | |
| 5208 | 165 #ifdef HAVE_SSE2 |
| 166 mem2agpcpy_SSE(to, from, len); | |
| 167 #elif defined (HAVE_MMX2) | |
| 4681 | 168 mem2agpcpy_MMX2(to, from, len); |
| 169 #elif defined (HAVE_3DNOW) | |
| 170 mem2agpcpy_3DNow(to, from, len); | |
| 171 #elif defined (HAVE_MMX) | |
| 172 mem2agpcpy_MMX(to, from, len); | |
| 173 #else | |
| 174 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
| 175 #endif | |
| 176 | |
| 177 #endif //!RUNTIME_CPUDETECT | |
|
8123
9fc45fe0d444
*HUGE* set of compiler warning fixes, unused variables removal
arpi
parents:
7072
diff
changeset
|
178 return to; |
| 4681 | 179 } |
| 180 | |
| 181 #endif /* use fastmemcpy */ | |
| 182 |