Mercurial > libavcodec.hg
annotate x86/rv40dsp_mmx.c @ 10822:9ccc13d1144d libavcodec
Move AVCodecs from h263dec.c to msmpeg4.c and disentangle init decode init.
author | michael |
---|---|
date | Fri, 08 Jan 2010 19:40:10 +0000 |
parents | cc64e1343397 |
children | 34a65026fa06 |
rev | line source |
---|---|
8519
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
1 /* |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
2 * Copyright (c) 2008 Konstantin Shishkov, Mathieu Velten |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
3 * |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
4 * MMX-optimized DSP functions for RV40, based on H.264 optimizations by |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
5 * Michael Niedermayer and Loren Merritt |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
6 * |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
7 * This file is part of FFmpeg. |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
8 * |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
9 * FFmpeg is free software; you can redistribute it and/or |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
10 * modify it under the terms of the GNU Lesser General Public |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
11 * License as published by the Free Software Foundation; either |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
12 * version 2.1 of the License, or (at your option) any later version. |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
13 * |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
14 * FFmpeg is distributed in the hope that it will be useful, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
17 * Lesser General Public License for more details. |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
18 * |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
19 * You should have received a copy of the GNU Lesser General Public |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
20 * License along with FFmpeg; if not, write to the Free Software |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
22 */ |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
23 |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
24 #include "dsputil_mmx.h" |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
25 |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
26 /* bias interleaved with bias div 8, use p+1 to access bias div 8 */ |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
27 DECLARE_ALIGNED_8(static const uint64_t, rv40_bias_reg[4][8]) = { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
28 { 0x0000000000000000ULL, 0x0000000000000000ULL, 0x0010001000100010ULL, 0x0002000200020002ULL, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
29 0x0020002000200020ULL, 0x0004000400040004ULL, 0x0010001000100010ULL, 0x0002000200020002ULL }, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
30 { 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
31 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL }, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
32 { 0x0000000000000000ULL, 0x0000000000000000ULL, 0x0020002000200020ULL, 0x0004000400040004ULL, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
33 0x0010001000100010ULL, 0x0002000200020002ULL, 0x0020002000200020ULL, 0x0004000400040004ULL }, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
34 { 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL, |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
35 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL } |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
36 }; |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
37 |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
38 static void put_rv40_chroma_mc8_mmx(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
39 { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
40 put_h264_chroma_generic_mc8_mmx(dst, src, stride, h, x, y, &rv40_bias_reg[y>>1][x&(~1)]); |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
41 } |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
42 static void put_rv40_chroma_mc4_mmx(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
43 { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
44 put_h264_chroma_generic_mc4_mmx(dst, src, stride, h, x, y, &rv40_bias_reg[y>>1][x&(~1)]); |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
45 } |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
46 static void avg_rv40_chroma_mc8_mmx2(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
47 { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
48 avg_h264_chroma_generic_mc8_mmx2(dst, src, stride, h, x, y, &rv40_bias_reg[y>>1][x&(~1)]); |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
49 } |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
50 static void avg_rv40_chroma_mc4_mmx2(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
51 { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
52 avg_h264_chroma_generic_mc4_mmx2(dst, src, stride, h, x, y, &rv40_bias_reg[y>>1][x&(~1)]); |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
53 } |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
54 static void avg_rv40_chroma_mc8_3dnow(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
55 { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
56 avg_h264_chroma_generic_mc8_3dnow(dst, src, stride, h, x, y, &rv40_bias_reg[y>>1][x&(~1)]); |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
57 } |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
58 static void avg_rv40_chroma_mc4_3dnow(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
59 { |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
60 avg_h264_chroma_generic_mc4_3dnow(dst, src, stride, h, x, y, &rv40_bias_reg[y>>1][x&(~1)]); |
cc64e1343397
Use H264 MMX chroma functions to accelerate RV40 decoding.
cehoyos
parents:
diff
changeset
|
61 } |