annotate binkidct.c @ 12413:e6e4059ea421 libavcodec

Mark xmm registers as clobbered in simple loopfilter. Should fix the last two VP8-related fate failures on Win64.
author rbultje
date Tue, 24 Aug 2010 16:52:27 +0000
parents 7dd2a45249a9
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
11231
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
1 /*
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
2 * Bink IDCT algorithm
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
3 * Copyright (c) 2009 Kostya Shishkov
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
4 *
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
5 * This file is part of FFmpeg.
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
6 *
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
7 * FFmpeg is free software; you can redistribute it and/or
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
8 * modify it under the terms of the GNU Lesser General Public
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
9 * License as published by the Free Software Foundation; either
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
10 * version 2.1 of the License, or (at your option) any later version.
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
11 *
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
12 * FFmpeg is distributed in the hope that it will be useful,
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
15 * Lesser General Public License for more details.
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
16 *
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
17 * You should have received a copy of the GNU Lesser General Public
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
18 * License along with FFmpeg; if not, write to the Free Software
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
20 */
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
21
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
22 /**
11644
7dd2a45249a9 Remove explicit filename from Doxygen @file commands.
diego
parents: 11231
diff changeset
23 * @file
11231
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
24 * Bink IDCT algorithm
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
25 */
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
26
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
27 #include "dsputil.h"
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
28
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
29 #define A1 2896 /* (1/sqrt(2))<<12 */
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
30 #define A2 2217
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
31 #define A3 3784
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
32 #define A4 -5352
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
33
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
34 #define IDCT_TRANSFORM(dest,s0,s1,s2,s3,s4,s5,s6,s7,d0,d1,d2,d3,d4,d5,d6,d7,munge,src) {\
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
35 const int a0 = (src)[s0] + (src)[s4]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
36 const int a1 = (src)[s0] - (src)[s4]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
37 const int a2 = (src)[s2] + (src)[s6]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
38 const int a3 = (A1*((src)[s2] - (src)[s6])) >> 11; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
39 const int a4 = (src)[s5] + (src)[s3]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
40 const int a5 = (src)[s5] - (src)[s3]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
41 const int a6 = (src)[s1] + (src)[s7]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
42 const int a7 = (src)[s1] - (src)[s7]; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
43 const int b0 = a4 + a6; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
44 const int b1 = (A3*(a5 + a7)) >> 11; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
45 const int b2 = ((A4*a5) >> 11) - b0 + b1; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
46 const int b3 = (A1*(a6 - a4) >> 11) - b2; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
47 const int b4 = ((A2*a7) >> 11) + b3 - b1; \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
48 (dest)[d0] = munge(a0+a2 +b0); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
49 (dest)[d1] = munge(a1+a3-a2+b2); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
50 (dest)[d2] = munge(a1-a3+a2+b3); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
51 (dest)[d3] = munge(a0-a2 -b4); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
52 (dest)[d4] = munge(a0-a2 +b4); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
53 (dest)[d5] = munge(a1-a3+a2-b3); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
54 (dest)[d6] = munge(a1+a3-a2-b2); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
55 (dest)[d7] = munge(a0+a2 -b0); \
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
56 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
57 /* end IDCT_TRANSFORM macro */
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
58
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
59 #define MUNGE_NONE(x) (x)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
60 #define IDCT_COL(dest,src) IDCT_TRANSFORM(dest,0,8,16,24,32,40,48,56,0,8,16,24,32,40,48,56,MUNGE_NONE,src)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
61
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
62 #define MUNGE_ROW(x) (((x) + 0x7F)>>8)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
63 #define IDCT_ROW(dest,src) IDCT_TRANSFORM(dest,0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7,MUNGE_ROW,src)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
64
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
65 static inline void bink_idct_col(DCTELEM *dest, const DCTELEM *src)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
66 {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
67 if ((src[8]|src[16]|src[24]|src[32]|src[40]|src[48]|src[56])==0) {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
68 dest[0] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
69 dest[8] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
70 dest[16] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
71 dest[24] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
72 dest[32] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
73 dest[40] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
74 dest[48] =
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
75 dest[56] = src[0];
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
76 } else {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
77 IDCT_COL(dest, src);
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
78 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
79 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
80
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
81 void ff_bink_idct_c(DCTELEM *block)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
82 {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
83 int i;
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
84 DCTELEM temp[64];
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
85
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
86 for (i = 0; i < 8; i++)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
87 bink_idct_col(&temp[i], &block[i]);
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
88 for (i = 0; i < 8; i++) {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
89 IDCT_ROW( (&block[8*i]), (&temp[8*i]) );
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
90 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
91 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
92
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
93 void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
94 {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
95 int i, j;
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
96
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
97 ff_bink_idct_c(block);
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
98 for (i = 0; i < 8; i++, dest += linesize, block += 8)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
99 for (j = 0; j < 8; j++)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
100 dest[j] += block[j];
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
101 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
102
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
103 void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
104 {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
105 int i;
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
106 DCTELEM temp[64];
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
107 for (i = 0; i < 8; i++)
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
108 bink_idct_col(&temp[i], &block[i]);
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
109 for (i = 0; i < 8; i++) {
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
110 IDCT_ROW( (&dest[i*linesize]), (&temp[8*i]) );
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
111 }
0fc1cdd984b7 Bink video decoder
kostya
parents:
diff changeset
112 }