updated - feel free to correct/improve
[libav.git] / libavcodec / dsputil.h
CommitLineData
ff4ec49e
FB
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 */
de6d9b64
FB
19#ifndef DSPUTIL_H
20#define DSPUTIL_H
21
22#include "common.h"
43f1708f 23#include "avcodec.h"
de6d9b64 24
44eb4951 25//#define DEBUG
de6d9b64
FB
26/* dct code */
27typedef short DCTELEM;
28
03c94ede 29void fdct_ifast (DCTELEM *data);
28db7fce 30void ff_jpeg_fdct_islow (DCTELEM *data);
de6d9b64
FB
31
32void j_rev_dct (DCTELEM *data);
33
3f09f52a 34void ff_fdct_mmx(DCTELEM *block);
de6d9b64 35
e0eac44e 36/* encoding scans */
2ad1516a
MN
37extern const UINT8 ff_alternate_horizontal_scan[64];
38extern const UINT8 ff_alternate_vertical_scan[64];
39extern const UINT8 ff_zigzag_direct[64];
5a240838 40
de6d9b64
FB
41/* pixel operations */
42#define MAX_NEG_CROP 384
43
44/* temporary */
45extern UINT32 squareTbl[512];
0cfa9713 46extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64
FB
47
48void dsputil_init(void);
49
b7c27ee6
MN
50/* minimum alignment rules ;)
51if u notice errors in the align stuff, need more alignment for some asm code for some cpu
52or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
53
54!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
55i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 56
b7c27ee6
MN
57!future video codecs might need functions with less strict alignment
58*/
59
60/* pixel ops : interface with DCT */
b7c27ee6
MN
61extern void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
62extern void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
63extern void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
64extern void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
073b013d
MN
65extern void (*ff_gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
66extern void (*ff_gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
67 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
b7c27ee6 68extern void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
3aa102be
MN
69extern int (*pix_sum)(UINT8 * pix, int line_size);
70extern int (*pix_norm1)(UINT8 * pix, int line_size);
71
44eb4951 72
de6d9b64
FB
73
74void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
9dbcbd92 75void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
de6d9b64
FB
76void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
77void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
649c00c9 78void clear_blocks_c(DCTELEM *blocks);
de6d9b64
FB
79
80/* add and put pixel (decoding) */
b7c27ee6
MN
81// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
82typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
83typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
b3184779
MN
84
85extern op_pixels_func put_pixels_tab[2][4];
86extern op_pixels_func avg_pixels_tab[2][4];
87extern op_pixels_func put_no_rnd_pixels_tab[2][4];
88extern op_pixels_func avg_no_rnd_pixels_tab[2][4];
89extern qpel_mc_func put_qpel_pixels_tab[2][16];
90extern qpel_mc_func avg_qpel_pixels_tab[2][16];
91extern qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
92extern qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
93
94#define CALL_2X_PIXELS(a, b, n)\
95static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
96 b(block , pixels , line_size, h);\
97 b(block+n, pixels+n, line_size, h);\
98}
44eb4951 99
de6d9b64
FB
100/* motion estimation */
101
b7c27ee6 102typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size);
de6d9b64
FB
103
104extern op_pixels_abs_func pix_abs16x16;
105extern op_pixels_abs_func pix_abs16x16_x2;
106extern op_pixels_abs_func pix_abs16x16_y2;
107extern op_pixels_abs_func pix_abs16x16_xy2;
ba6802de
MN
108extern op_pixels_abs_func pix_abs8x8;
109extern op_pixels_abs_func pix_abs8x8_x2;
110extern op_pixels_abs_func pix_abs8x8_y2;
111extern op_pixels_abs_func pix_abs8x8_xy2;
112
113int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx);
114int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx);
115int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx);
116int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
de6d9b64 117
7801d21d
MN
118/**
119 * permute block according to permuatation.
120 * @param last last non zero element in scantable order
121 */
122void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
e0eac44e 123
3d03c0a2 124#if defined(HAVE_MMX)
de6d9b64
FB
125
126#define MM_MMX 0x0001 /* standard MMX */
127#define MM_3DNOW 0x0004 /* AMD 3DNOW */
128#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
129#define MM_SSE 0x0008 /* SSE functions */
130#define MM_SSE2 0x0010 /* PIV SSE2 functions */
131
132extern int mm_flags;
133
134int mm_support(void);
135
136static inline void emms(void)
137{
fb16b7e7
FB
138 __asm __volatile ("emms;":::"memory");
139}
140
141#define emms_c() \
142{\
143 if (mm_flags & MM_MMX)\
144 emms();\
de6d9b64
FB
145}
146
147#define __align8 __attribute__ ((aligned (8)))
148
149void dsputil_init_mmx(void);
57060b1e 150void dsputil_set_bit_exact_mmx(void);
de6d9b64 151
3d03c0a2
FB
152#elif defined(ARCH_ARMV4L)
153
154#define emms_c()
155
156/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
157 line ptimizations */
158#define __align8 __attribute__ ((aligned (4)))
159
160void dsputil_init_armv4l(void);
161
c34270f5
FB
162#elif defined(HAVE_MLIB)
163
164#define emms_c()
165
166/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
167#define __align8 __attribute__ ((aligned (8)))
168
169void dsputil_init_mlib(void);
170
1e98dffb
NK
171#elif defined(ARCH_ALPHA)
172
173#define emms_c()
174#define __align8 __attribute__ ((aligned (8)))
175
176void dsputil_init_alpha(void);
177
59925ef2
BF
178#elif defined(ARCH_POWERPC)
179
180#define emms_c()
181#define __align8 __attribute__ ((aligned (16)))
182
ab6c65f6 183void dsputil_init_ppc(void);
59925ef2 184
d46aba26
LS
185#elif defined(HAVE_MMI)
186
187#define emms_c()
188
189#define __align8 __attribute__ ((aligned (16)))
190
191void dsputil_init_mmi(void);
192
de6d9b64
FB
193#else
194
fb16b7e7
FB
195#define emms_c()
196
de6d9b64
FB
197#define __align8
198
199#endif
200
6d4985bb
FB
201#ifdef __GNUC__
202
203struct unaligned_64 { uint64_t l; } __attribute__((packed));
204struct unaligned_32 { uint32_t l; } __attribute__((packed));
205
206#define LD32(a) (((const struct unaligned_32 *) (a))->l)
207#define LD64(a) (((const struct unaligned_64 *) (a))->l)
208
209#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
210
211#else /* __GNUC__ */
212
213#define LD32(a) (*((uint32_t*)(a)))
214#define LD64(a) (*((uint64_t*)(a)))
215
216#define ST32(a, b) *((uint32_t*)(a)) = (b)
217
218#endif /* !__GNUC__ */
219
43f1708f
J
220/* PSNR */
221void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
222 int orig_linesize[3], int coded_linesize,
223 AVCodecContext *avctx);
224
de6d9b64 225#endif