Mercurial > libavcodec.hg

annotate x86/h264_idct_sse2.asm @ 9005:e5c9a3a813ea libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Remove (incorrect) filenames from x264 asm files, add descriptions.
author darkshikari
date Sun, 22 Feb 2009 11:33:09 +0000
parents cea216e44ee3
children 37ac731fe32c
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
8510
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
1 ;*****************************************************************************
9005
e5c9a3a813ea Remove (incorrect) filenames from x264 asm files, add descriptions.
darkshikari
parents: 8510
diff changeset
2 ;* SSE2-optimized H.264 iDCT
8510
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
3 ;*****************************************************************************
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
4 ;* Copyright (C) 2003-2008 x264 project
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
5 ;*
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
6 ;* Authors: Laurent Aimar <fenrir@via.ecp.fr>
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
7 ;* Loren Merritt <lorenm@u.washington.edu>
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
8 ;* Holger Lubitz <hal@duncan.ol.sub.de>
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
9 ;* Min Chen <chenm001.163.com>
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
10 ;*
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
11 ;* This program is free software; you can redistribute it and/or modify
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
12 ;* it under the terms of the GNU General Public License as published by
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
13 ;* the Free Software Foundation; either version 2 of the License, or
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
14 ;* (at your option) any later version.
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
15 ;*
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
16 ;* This program is distributed in the hope that it will be useful,
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
17 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
18 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
19 ;* GNU General Public License for more details.
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
20 ;*
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
21 ;* You should have received a copy of the GNU General Public License
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
22 ;* along with this program; if not, write to the Free Software
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
23 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
24 ;*****************************************************************************
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
25
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
26 %include "x86inc.asm"
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
27 %include "x86util.asm"
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
28
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
29 SECTION_RODATA
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
30 pw_32: times 8 dw 32
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
31
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
32 SECTION .text
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
33
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
34 %macro IDCT4_1D 6
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
35 SUMSUB_BA m%3, m%1
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
36 SUMSUBD2_AB m%2, m%4, m%6, m%5
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
37 SUMSUB_BADC m%2, m%3, m%5, m%1
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
38 SWAP %1, %2, %5, %4, %3
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
39 %endmacro
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
40
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
41 INIT_XMM
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
42 cglobal x264_add8x4_idct_sse2, 3,3
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
43 movq m0, [r1+ 0]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
44 movq m1, [r1+ 8]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
45 movq m2, [r1+16]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
46 movq m3, [r1+24]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
47 movhps m0, [r1+32]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
48 movhps m1, [r1+40]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
49 movhps m2, [r1+48]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
50 movhps m3, [r1+56]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
51 IDCT4_1D 0,1,2,3,4,5
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
52 TRANSPOSE2x4x4W 0,1,2,3,4
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
53 paddw m0, [pw_32 GLOBAL]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
54 IDCT4_1D 0,1,2,3,4,5
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
55 pxor m7, m7
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
56 STORE_DIFF m0, m4, m7, [r0]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
57 STORE_DIFF m1, m4, m7, [r0+r2]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
58 lea r0, [r0+r2*2]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
59 STORE_DIFF m2, m4, m7, [r0]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
60 STORE_DIFF m3, m4, m7, [r0+r2]
cea216e44ee3 Add x264 SSE2 iDCT functions to H.264 decoder.
darkshikari
parents:
diff changeset
61 RET