062b510ab083381816442c8445f2f5f1f2b145e3
[libav.git] / libavcodec / dsputil.h
1 /*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19 #ifndef DSPUTIL_H
20 #define DSPUTIL_H
21
22 #include "common.h"
23 #include "avcodec.h"
24
25 //#define DEBUG
26 /* dct code */
27 typedef short DCTELEM;
28
29 void fdct_ifast (DCTELEM *data);
30
31 void j_rev_dct (DCTELEM *data);
32
33 void fdct_mmx(DCTELEM *block);
34
35 extern void (*av_fdct)(DCTELEM *block);
36
37 /* encoding scans */
38 extern UINT8 ff_alternate_horizontal_scan[64];
39 extern UINT8 ff_alternate_vertical_scan[64];
40 extern UINT8 zigzag_direct[64];
41
42 /* permutation table */
43 extern UINT8 permutation[64];
44
45 /* pixel operations */
46 #define MAX_NEG_CROP 384
47
48 /* temporary */
49 extern UINT32 squareTbl[512];
50 extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
51
52 void dsputil_init(void);
53
54 /* pixel ops : interface with DCT */
55
56 extern void (*ff_idct)(DCTELEM *block);
57 extern void (*get_pixels)(DCTELEM *block, const UINT8 *pixels, int line_size);
58 extern void (*diff_pixels)(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
59 extern void (*put_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);
60 extern void (*add_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);
61 extern void (*gmc1)(UINT8 *dst, UINT8 *src, int srcStride, int h, int x16, int y16, int rounder);
62 extern void (*clear_blocks)(DCTELEM *blocks);
63
64
65 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
66 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
67 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
68 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
69 void clear_blocks_c(DCTELEM *blocks);
70
71 /* add and put pixel (decoding) */
72 typedef void (*op_pixels_func)(UINT8 *block, const UINT8 *pixels, int line_size, int h);
73 typedef void (*qpel_mc_func)(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my);
74
75 extern op_pixels_func put_pixels_tab[4];
76 extern op_pixels_func avg_pixels_tab[4];
77 extern op_pixels_func put_no_rnd_pixels_tab[4];
78 extern op_pixels_func avg_no_rnd_pixels_tab[4];
79 extern qpel_mc_func qpel_mc_rnd_tab[16];
80 extern qpel_mc_func qpel_mc_no_rnd_tab[16];
81
82 /* motion estimation */
83
84 typedef int (*op_pixels_abs_func)(UINT8 *blk1, UINT8 *blk2, int line_size);
85
86 extern op_pixels_abs_func pix_abs16x16;
87 extern op_pixels_abs_func pix_abs16x16_x2;
88 extern op_pixels_abs_func pix_abs16x16_y2;
89 extern op_pixels_abs_func pix_abs16x16_xy2;
90 extern op_pixels_abs_func pix_abs8x8;
91 extern op_pixels_abs_func pix_abs8x8_x2;
92 extern op_pixels_abs_func pix_abs8x8_y2;
93 extern op_pixels_abs_func pix_abs8x8_xy2;
94
95 int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx);
96 int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx);
97 int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx);
98 int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
99
100 static inline int block_permute_op(int j)
101 {
102 return permutation[j];
103 }
104
105 void block_permute(INT16 *block);
106
107 #if defined(HAVE_MMX)
108
109 #define MM_MMX 0x0001 /* standard MMX */
110 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
111 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
112 #define MM_SSE 0x0008 /* SSE functions */
113 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
114
115 extern int mm_flags;
116
117 int mm_support(void);
118
119 static inline void emms(void)
120 {
121 __asm __volatile ("emms;":::"memory");
122 }
123
124 #define emms_c() \
125 {\
126 if (mm_flags & MM_MMX)\
127 emms();\
128 }
129
130 #define __align8 __attribute__ ((aligned (8)))
131
132 void dsputil_init_mmx(void);
133 void dsputil_set_bit_exact_mmx(void);
134
135 #elif defined(ARCH_ARMV4L)
136
137 #define emms_c()
138
139 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
140 line ptimizations */
141 #define __align8 __attribute__ ((aligned (4)))
142
143 void dsputil_init_armv4l(void);
144
145 #elif defined(HAVE_MLIB)
146
147 #define emms_c()
148
149 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
150 #define __align8 __attribute__ ((aligned (8)))
151
152 void dsputil_init_mlib(void);
153
154 #elif defined(ARCH_ALPHA)
155
156 #define emms_c()
157 #define __align8 __attribute__ ((aligned (8)))
158
159 void dsputil_init_alpha(void);
160
161 #else
162
163 #define emms_c()
164
165 #define __align8
166
167 #endif
168
169 /* PSNR */
170 void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
171 int orig_linesize[3], int coded_linesize,
172 AVCodecContext *avctx);
173
174 #endif