Mercurial > mplayer.hg
annotate libvo/fastmemcpy.h @ 23450:d739d60eb577
sync w/23467 (was just a typo in english doc, no incidence here)
author | gpoirier |
---|---|
date | Tue, 05 Jun 2007 07:35:07 +0000 |
parents | 499d246cd549 |
children | a124f3abc1ec |
rev | line source |
---|---|
23385 | 1 /* |
2 * This file is part of MPlayer. | |
3 * | |
4 * MPlayer is free software; you can redistribute it and/or | |
5 * modify it under the terms of the GNU Lesser General Public | |
6 * License as published by the Free Software Foundation; either | |
7 * version 2.1 of the License, or (at your option) any later version. | |
8 * | |
9 * MPlayer is distributed in the hope that it will be useful, | |
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 * Lesser General Public License for more details. | |
13 * | |
14 * You should have received a copy of the GNU Lesser General Public | |
15 * License along with MPlayer; if not, write to the Free Software | |
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
17 */ | |
18 | |
477 | 19 #ifndef __MPLAYER_MEMCPY |
644
88eb1a3f7bfb
Changed code, should be faster on Athlon/K6 but slower on PIII with SSE, more portable.
atmosfear
parents:
581
diff
changeset
|
20 #define __MPLAYER_MEMCPY 1 |
88eb1a3f7bfb
Changed code, should be faster on Athlon/K6 but slower on PIII with SSE, more portable.
atmosfear
parents:
581
diff
changeset
|
21 |
13787
e047e70a9767
Handle "xxx.h" vs "../xxx.h" include paths in a consistent way.
diego
parents:
12663
diff
changeset
|
22 #include "config.h" |
22003
6fd6bf6269f3
fastmemcpy.h always has to include inttypes.h on systems that need it.
diego
parents:
22000
diff
changeset
|
23 #include <inttypes.h> |
800 | 24 |
1131 | 25 #ifdef USE_FASTMEMCPY |
26 #if defined(HAVE_MMX) || defined(HAVE_MMX2) || defined(HAVE_3DNOW) \ | |
27 /* || defined(HAVE_SSE) || defined(HAVE_SSE2) */ | |
644
88eb1a3f7bfb
Changed code, should be faster on Athlon/K6 but slower on PIII with SSE, more portable.
atmosfear
parents:
581
diff
changeset
|
28 #include <stddef.h> |
477 | 29 |
698
f0fbf1a9bf31
Moving fast_memcpy to separate file (Size optimization)
nickols_k
parents:
685
diff
changeset
|
30 extern void * fast_memcpy(void * to, const void * from, size_t len); |
4681 | 31 extern void * mem2agpcpy(void * to, const void * from, size_t len); |
376 | 32 #define memcpy(a,b,c) fast_memcpy(a,b,c) |
513 | 33 |
4681 | 34 #else /* HAVE_MMX/MMX2/3DNOW/SSE/SSE2 */ |
35 #define mem2agpcpy(a,b,c) memcpy(a,b,c) | |
478 | 36 #endif |
4681 | 37 |
38 #else /* USE_FASTMEMCPY */ | |
39 #define mem2agpcpy(a,b,c) memcpy(a,b,c) | |
40 #endif | |
4708 | 41 |
17095 | 42 static inline void * mem2agpcpy_pic(void * dst, const void * src, int bytesPerLine, int height, int dstStride, int srcStride) |
4708 | 43 { |
44 int i; | |
45 void *retval=dst; | |
46 | |
15069
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
47 if(dstStride == srcStride) |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
48 { |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
49 if (srcStride < 0) { |
22000
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
50 src = (uint8_t*)src + (height-1)*srcStride; |
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
51 dst = (uint8_t*)dst + (height-1)*dstStride; |
15069
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
52 srcStride = -srcStride; |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
53 } |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
54 |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
55 mem2agpcpy(dst, src, srcStride*height); |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
56 } |
4708 | 57 else |
58 { | |
59 for(i=0; i<height; i++) | |
60 { | |
61 mem2agpcpy(dst, src, bytesPerLine); | |
22000
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
62 src = (uint8_t*)src + srcStride; |
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
63 dst = (uint8_t*)dst + dstStride; |
4708 | 64 } |
65 } | |
66 | |
67 return retval; | |
68 } | |
69 | |
17095 | 70 static inline void * memcpy_pic(void * dst, const void * src, int bytesPerLine, int height, int dstStride, int srcStride) |
5504
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
71 { |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
72 int i; |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
73 void *retval=dst; |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
74 |
15069
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
75 if(dstStride == srcStride) |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
76 { |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
77 if (srcStride < 0) { |
22000
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
78 src = (uint8_t*)src + (height-1)*srcStride; |
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
79 dst = (uint8_t*)dst + (height-1)*dstStride; |
15069
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
80 srcStride = -srcStride; |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
81 } |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
82 |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
83 memcpy(dst, src, srcStride*height); |
3f5daa60e049
support for negative strides (fixes -vf spp,flip crash)
henry
parents:
13787
diff
changeset
|
84 } |
5504
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
85 else |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
86 { |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
87 for(i=0; i<height; i++) |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
88 { |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
89 memcpy(dst, src, bytesPerLine); |
22000
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
90 src = (uint8_t*)src + srcStride; |
f7c209352770
get rid of void pointer arithmetic, as spotted by ICC
gpoirier
parents:
17095
diff
changeset
|
91 dst = (uint8_t*)dst + dstStride; |
5504
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
92 } |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
93 } |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
94 |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
95 return retval; |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
96 } |
23ba417cf64b
memcpy_pic() added (copy image plane with src/dst stride)
arpi
parents:
4708
diff
changeset
|
97 |
4681 | 98 #endif |