Mercurial > libavcodec.hg
annotate x86/h264_idct_sse2.asm @ 9626:bd3e11b60ccd libavcodec
Add a chroma_sample_location field to define positioning of chroma samples
author | conrad |
---|---|
date | Mon, 11 May 2009 04:34:23 +0000 |
parents | 37ac731fe32c |
children | c08ca946c80a |
rev | line source |
---|---|
9006
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
1 ;***************************************************************************** |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
2 ;* SSE2-optimized H.264 iDCT |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
3 ;***************************************************************************** |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
4 ;* Copyright (C) 2003-2008 x264 project |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
5 ;* |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
6 ;* Authors: Laurent Aimar <fenrir@via.ecp.fr> |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
7 ;* Loren Merritt <lorenm@u.washington.edu> |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
8 ;* Holger Lubitz <hal@duncan.ol.sub.de> |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
9 ;* Min Chen <chenm001.163.com> |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
10 ;* |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
11 ;* This program is free software; you can redistribute it and/or modify |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
12 ;* it under the terms of the GNU General Public License as published by |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
13 ;* the Free Software Foundation; either version 2 of the License, or |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
14 ;* (at your option) any later version. |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
15 ;* |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
16 ;* This program is distributed in the hope that it will be useful, |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
17 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
18 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
19 ;* GNU General Public License for more details. |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
20 ;* |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
21 ;* You should have received a copy of the GNU General Public License |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
22 ;* along with this program; if not, write to the Free Software |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
23 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
24 ;***************************************************************************** |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
25 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
26 %include "x86inc.asm" |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
27 %include "x86util.asm" |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
28 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
29 SECTION_RODATA |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
30 pw_32: times 8 dw 32 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
31 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
32 SECTION .text |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
33 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
34 %macro IDCT4_1D 6 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
35 SUMSUB_BA m%3, m%1 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
36 SUMSUBD2_AB m%2, m%4, m%6, m%5 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
37 SUMSUB_BADC m%2, m%3, m%5, m%1 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
38 SWAP %1, %2, %5, %4, %3 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
39 %endmacro |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
40 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
41 INIT_XMM |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
42 cglobal x264_add8x4_idct_sse2, 3,3 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
43 movq m0, [r1+ 0] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
44 movq m1, [r1+ 8] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
45 movq m2, [r1+16] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
46 movq m3, [r1+24] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
47 movhps m0, [r1+32] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
48 movhps m1, [r1+40] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
49 movhps m2, [r1+48] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
50 movhps m3, [r1+56] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
51 IDCT4_1D 0,1,2,3,4,5 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
52 TRANSPOSE2x4x4W 0,1,2,3,4 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
53 paddw m0, [pw_32 GLOBAL] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
54 IDCT4_1D 0,1,2,3,4,5 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
55 pxor m7, m7 |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
56 STORE_DIFF m0, m4, m7, [r0] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
57 STORE_DIFF m1, m4, m7, [r0+r2] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
58 lea r0, [r0+r2*2] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
59 STORE_DIFF m2, m4, m7, [r0] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
60 STORE_DIFF m3, m4, m7, [r0+r2] |
37ac731fe32c
Convert x264 asm files to proper unix line breaks
darkshikari
parents:
9005
diff
changeset
|
61 RET |