Mercurial > mplayer.hg
annotate libvo/aclib.c @ 35618:2cf9f80c4023
Fix a rendering bug in vf_ass.
Reported at http://thread.gmane.org/gmane.comp.video.mplayer.user/69250.
author | upsuper |
---|---|
date | Mon, 31 Dec 2012 04:35:03 +0000 |
parents | 4e2f4bd081ce |
children |
rev | line source |
---|---|
28446
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
1 /* |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
2 * aclib - advanced C library ;) |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
3 * Functions which improve and expand the standard C library, see aclib_template.c. |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
4 * This file only contains runtime CPU detection and config option stuff. |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
5 * runtime CPU detection by Michael Niedermayer (michaelni@gmx.at) |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
6 * |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
7 * This file is part of MPlayer. |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
8 * |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
9 * MPlayer is free software; you can redistribute it and/or modify |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
10 * it under the terms of the GNU General Public License as published by |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
11 * the Free Software Foundation; either version 2 of the License, or |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
12 * (at your option) any later version. |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
13 * |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
14 * MPlayer is distributed in the hope that it will be useful, |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
17 * GNU General Public License for more details. |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
18 * |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
19 * You should have received a copy of the GNU General Public License along |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
20 * with MPlayer; if not, write to the Free Software Foundation, Inc., |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
21 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
22 */ |
7681eab10aea
Add standard license headers, unify header formatting.
diego
parents:
28335
diff
changeset
|
23 |
12650
ac3fd2ff2561
Unify the config.h #include, use "config.h" instead of "../config.h"
diego
parents:
12492
diff
changeset
|
24 #include "config.h" |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
25 #include <stddef.h> |
30135
807fce7a4bb3
Do not assume that "long" is the size of a register.
reimar
parents:
29114
diff
changeset
|
26 #include <stdint.h> |
21982
fa66a03e8920
Include string.h to make sure memcpy is not used without prototype
reimar
parents:
20577
diff
changeset
|
27 #include <string.h> |
13787
e047e70a9767
Handle "xxx.h" vs "../xxx.h" include paths in a consistent way.
diego
parents:
13720
diff
changeset
|
28 #include "cpudetect.h" |
8123
9fc45fe0d444
*HUGE* set of compiler warning fixes, unused variables removal
arpi
parents:
7072
diff
changeset
|
29 #include "fastmemcpy.h" |
12492
4b8417674f1c
fix crash due to fast_memcpy calling itself instead of libc memcpy
reimar
parents:
8127
diff
changeset
|
30 #undef memcpy |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
31 |
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
|
32 #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
|
33 #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
|
34 //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
|
35 |
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
|
36 //#define STATISTICS |
3393 | 37 |
38 //Note: we have MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one | |
39 //Plain C versions | |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
40 //#if !HAVE_MMX || CONFIG_RUNTIME_CPUDETECT |
3393 | 41 //#define COMPILE_C |
42 //#endif | |
43 | |
28921
62f0032e736a
Get rid of pointless preprocessor condition indirection and use ARCH_X86
diego
parents:
28448
diff
changeset
|
44 #if ARCH_X86 |
3393 | 45 |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
46 #if (HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT |
3393 | 47 #define COMPILE_MMX |
48 #endif | |
49 | |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
50 #if (HAVE_MMX2 && !HAVE_SSE2) || CONFIG_RUNTIME_CPUDETECT |
3393 | 51 #define COMPILE_MMX2 |
52 #endif | |
53 | |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
54 #if (HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT |
3393 | 55 #define COMPILE_3DNOW |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
56 #endif |
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
57 |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
58 #if HAVE_SSE2 || CONFIG_RUNTIME_CPUDETECT |
5208 | 59 #define COMPILE_SSE |
60 #endif | |
61 | |
3393 | 62 #undef HAVE_MMX |
63 #undef HAVE_MMX2 | |
28335 | 64 #undef HAVE_AMD3DNOW |
5208 | 65 #undef HAVE_SSE |
66 #undef HAVE_SSE2 | |
28290 | 67 #define HAVE_MMX 0 |
68 #define HAVE_MMX2 0 | |
28335 | 69 #define HAVE_AMD3DNOW 0 |
28290 | 70 #define HAVE_SSE 0 |
71 #define HAVE_SSE2 0 | |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
72 /* |
3393 | 73 #ifdef COMPILE_C |
74 #undef HAVE_MMX | |
75 #undef HAVE_MMX2 | |
28335 | 76 #undef HAVE_AMD3DNOW |
28290 | 77 #undef HAVE_SSE |
78 #undef HAVE_SSE2 | |
79 #define HAVE_MMX 0 | |
80 #define HAVE_MMX2 0 | |
28335 | 81 #define HAVE_AMD3DNOW 0 |
28290 | 82 #define HAVE_SSE 0 |
83 #define HAVE_SSE2 0 | |
3393 | 84 #define RENAME(a) a ## _C |
85 #include "aclib_template.c" | |
86 #endif | |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
87 */ |
3393 | 88 //MMX versions |
89 #ifdef COMPILE_MMX | |
90 #undef RENAME | |
28290 | 91 #undef HAVE_MMX |
3393 | 92 #undef HAVE_MMX2 |
28335 | 93 #undef HAVE_AMD3DNOW |
5208 | 94 #undef HAVE_SSE |
95 #undef HAVE_SSE2 | |
28290 | 96 #define HAVE_MMX 1 |
97 #define HAVE_MMX2 0 | |
28335 | 98 #define HAVE_AMD3DNOW 0 |
28290 | 99 #define HAVE_SSE 0 |
100 #define HAVE_SSE2 0 | |
3393 | 101 #define RENAME(a) a ## _MMX |
102 #include "aclib_template.c" | |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
103 #endif |
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
104 |
3393 | 105 //MMX2 versions |
106 #ifdef COMPILE_MMX2 | |
107 #undef RENAME | |
28290 | 108 #undef HAVE_MMX |
109 #undef HAVE_MMX2 | |
28335 | 110 #undef HAVE_AMD3DNOW |
5208 | 111 #undef HAVE_SSE |
112 #undef HAVE_SSE2 | |
28290 | 113 #define HAVE_MMX 1 |
114 #define HAVE_MMX2 1 | |
28335 | 115 #define HAVE_AMD3DNOW 0 |
28290 | 116 #define HAVE_SSE 0 |
117 #define HAVE_SSE2 0 | |
3393 | 118 #define RENAME(a) a ## _MMX2 |
119 #include "aclib_template.c" | |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
120 #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
|
121 |
3393 | 122 //3DNOW versions |
123 #ifdef COMPILE_3DNOW | |
124 #undef RENAME | |
28290 | 125 #undef HAVE_MMX |
3393 | 126 #undef HAVE_MMX2 |
28335 | 127 #undef HAVE_AMD3DNOW |
5208 | 128 #undef HAVE_SSE |
129 #undef HAVE_SSE2 | |
28290 | 130 #define HAVE_MMX 1 |
131 #define HAVE_MMX2 0 | |
28335 | 132 #define HAVE_AMD3DNOW 1 |
28290 | 133 #define HAVE_SSE 0 |
134 #define HAVE_SSE2 0 | |
3393 | 135 #define RENAME(a) a ## _3DNow |
136 #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
|
137 #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
|
138 |
5208 | 139 //SSE versions (only used on SSE2 cpus) |
140 #ifdef COMPILE_SSE | |
141 #undef RENAME | |
28290 | 142 #undef HAVE_MMX |
143 #undef HAVE_MMX2 | |
28335 | 144 #undef HAVE_AMD3DNOW |
28290 | 145 #undef HAVE_SSE |
146 #undef HAVE_SSE2 | |
147 #define HAVE_MMX 1 | |
148 #define HAVE_MMX2 1 | |
28335 | 149 #define HAVE_AMD3DNOW 0 |
28290 | 150 #define HAVE_SSE 1 |
151 #define HAVE_SSE2 1 | |
5208 | 152 #define RENAME(a) a ## _SSE |
153 #include "aclib_template.c" | |
154 #endif | |
155 | |
28921
62f0032e736a
Get rid of pointless preprocessor condition indirection and use ARCH_X86
diego
parents:
28448
diff
changeset
|
156 #endif /* ARCH_X86 */ |
3393 | 157 |
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
|
158 |
23523
273aa6124f66
avoid utter breakage on non-x86, patch from Chris Roccati <roccati@at@pobox.dot.com>
lu_zero
parents:
21982
diff
changeset
|
159 #undef fast_memcpy |
7072 | 160 void * fast_memcpy(void * to, const void * from, size_t len) |
3393 | 161 { |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
162 #if CONFIG_RUNTIME_CPUDETECT |
28921
62f0032e736a
Get rid of pointless preprocessor condition indirection and use ARCH_X86
diego
parents:
28448
diff
changeset
|
163 #if ARCH_X86 |
3393 | 164 // ordered per speed fasterst first |
5208 | 165 if(gCpuCaps.hasSSE2) |
166 fast_memcpy_SSE(to, from, len); | |
167 else if(gCpuCaps.hasMMX2) | |
3393 | 168 fast_memcpy_MMX2(to, from, len); |
169 else if(gCpuCaps.has3DNow) | |
170 fast_memcpy_3DNow(to, from, len); | |
171 else if(gCpuCaps.hasMMX) | |
172 fast_memcpy_MMX(to, from, len); | |
173 else | |
28921
62f0032e736a
Get rid of pointless preprocessor condition indirection and use ARCH_X86
diego
parents:
28448
diff
changeset
|
174 #endif |
3393 | 175 memcpy(to, from, len); // prior to mmx we use the standart memcpy |
176 #else | |
28290 | 177 #if HAVE_SSE2 |
5208 | 178 fast_memcpy_SSE(to, from, len); |
28290 | 179 #elif HAVE_MMX2 |
3393 | 180 fast_memcpy_MMX2(to, from, len); |
28335 | 181 #elif HAVE_AMD3DNOW |
3393 | 182 fast_memcpy_3DNow(to, from, len); |
28290 | 183 #elif HAVE_MMX |
3393 | 184 fast_memcpy_MMX(to, from, len); |
185 #else | |
186 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
|
187 #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
|
188 |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
189 #endif //!CONFIG_RUNTIME_CPUDETECT |
5543
c75f75806af1
memcpy must return destination ptr patch by Adam <adam@cfar.umd.edu>
michael
parents:
5208
diff
changeset
|
190 return to; |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
191 } |
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
diff
changeset
|
192 |
8127
e7153e62a7f4
On non-x86 platforms, memcpy was re-implemented in mplayer and was called
jkeil
parents:
8123
diff
changeset
|
193 #undef mem2agpcpy |
7072 | 194 void * mem2agpcpy(void * to, const void * from, size_t len) |
4681 | 195 { |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
196 #if CONFIG_RUNTIME_CPUDETECT |
28921
62f0032e736a
Get rid of pointless preprocessor condition indirection and use ARCH_X86
diego
parents:
28448
diff
changeset
|
197 #if ARCH_X86 |
4681 | 198 // ordered per speed fasterst first |
5208 | 199 if(gCpuCaps.hasSSE2) |
200 mem2agpcpy_SSE(to, from, len); | |
201 else if(gCpuCaps.hasMMX2) | |
4681 | 202 mem2agpcpy_MMX2(to, from, len); |
203 else if(gCpuCaps.has3DNow) | |
204 mem2agpcpy_3DNow(to, from, len); | |
205 else if(gCpuCaps.hasMMX) | |
206 mem2agpcpy_MMX(to, from, len); | |
207 else | |
28921
62f0032e736a
Get rid of pointless preprocessor condition indirection and use ARCH_X86
diego
parents:
28448
diff
changeset
|
208 #endif |
4681 | 209 memcpy(to, from, len); // prior to mmx we use the standart memcpy |
210 #else | |
28290 | 211 #if HAVE_SSE2 |
5208 | 212 mem2agpcpy_SSE(to, from, len); |
28290 | 213 #elif HAVE_MMX2 |
4681 | 214 mem2agpcpy_MMX2(to, from, len); |
28335 | 215 #elif HAVE_AMD3DNOW |
4681 | 216 mem2agpcpy_3DNow(to, from, len); |
28290 | 217 #elif HAVE_MMX |
4681 | 218 mem2agpcpy_MMX(to, from, len); |
219 #else | |
220 memcpy(to, from, len); // prior to mmx we use the standart memcpy | |
221 #endif | |
222 | |
29114
06540eb5ef6a
Rename RUNTIME_CPUDETECT to CONFIG_RUNTIME_CPUDETECT and always define it.
ramiro
parents:
28921
diff
changeset
|
223 #endif //!CONFIG_RUNTIME_CPUDETECT |
8123
9fc45fe0d444
*HUGE* set of compiler warning fixes, unused variables removal
arpi
parents:
7072
diff
changeset
|
224 return to; |
4681 | 225 } |