fix motion vector decoding bug and reinstate interframes
[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 */
24641185
MN
19
20/**
21 * @file dsputil.h
983e3246 22 * DSP utils.
5755c27f
MN
23 * note, many functions in here may use MMX which trashes the FPU state, it is
24 * absolutely necessary to call emms_c() between dsp & float/double code
24641185
MN
25 */
26
de6d9b64
FB
27#ifndef DSPUTIL_H
28#define DSPUTIL_H
29
30#include "common.h"
43f1708f 31#include "avcodec.h"
de6d9b64 32
24641185 33
44eb4951 34//#define DEBUG
de6d9b64
FB
35/* dct code */
36typedef short DCTELEM;
0e15384d 37//typedef int DCTELEM;
de6d9b64 38
03c94ede 39void fdct_ifast (DCTELEM *data);
28db7fce 40void ff_jpeg_fdct_islow (DCTELEM *data);
de6d9b64
FB
41
42void j_rev_dct (DCTELEM *data);
43
3f09f52a 44void ff_fdct_mmx(DCTELEM *block);
de6d9b64 45
e0eac44e 46/* encoding scans */
0c1a9eda
ZK
47extern const uint8_t ff_alternate_horizontal_scan[64];
48extern const uint8_t ff_alternate_vertical_scan[64];
49extern const uint8_t ff_zigzag_direct[64];
5a240838 50
de6d9b64
FB
51/* pixel operations */
52#define MAX_NEG_CROP 384
53
54/* temporary */
0c1a9eda
ZK
55extern uint32_t squareTbl[512];
56extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 57
de6d9b64 58
b7c27ee6 59/* minimum alignment rules ;)
eb4b3dd3 60if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
61or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
62
63!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
64i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 65
b7c27ee6
MN
66!future video codecs might need functions with less strict alignment
67*/
68
eb4b3dd3 69/*
0c1a9eda
ZK
70void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
71void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
72void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
73void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 74void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 75*/
de6d9b64
FB
76
77/* add and put pixel (decoding) */
b7c27ee6 78// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
0c1a9eda
ZK
79typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
80typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 81typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
b3184779 82
db794953 83#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
84void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
85void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
86void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
87
88DEF_OLD_QPEL(qpel16_mc11_old_c)
89DEF_OLD_QPEL(qpel16_mc31_old_c)
90DEF_OLD_QPEL(qpel16_mc12_old_c)
91DEF_OLD_QPEL(qpel16_mc32_old_c)
92DEF_OLD_QPEL(qpel16_mc13_old_c)
93DEF_OLD_QPEL(qpel16_mc33_old_c)
94DEF_OLD_QPEL(qpel8_mc11_old_c)
95DEF_OLD_QPEL(qpel8_mc31_old_c)
96DEF_OLD_QPEL(qpel8_mc12_old_c)
97DEF_OLD_QPEL(qpel8_mc32_old_c)
98DEF_OLD_QPEL(qpel8_mc13_old_c)
99DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
100
101#define CALL_2X_PIXELS(a, b, n)\
102static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
103 b(block , pixels , line_size, h);\
104 b(block+n, pixels+n, line_size, h);\
105}
44eb4951 106
de6d9b64
FB
107/* motion estimation */
108
0c1a9eda 109typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
1457ab52 110
0c1a9eda 111typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
1457ab52 112
0da71265 113
24641185
MN
114/**
115 * DSPContext.
116 */
eb4b3dd3
ZK
117typedef struct DSPContext {
118 /* pixel ops : interface with DCT */
0c1a9eda
ZK
119 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
120 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
121 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
122 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
d518aebd
MN
123 /**
124 * translational global motion compensation.
125 */
0c1a9eda 126 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
127 /**
128 * global motion compensation.
129 */
0c1a9eda 130 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
eb4b3dd3
ZK
131 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
132 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
133 int (*pix_sum)(uint8_t * pix, int line_size);
134 int (*pix_norm1)(uint8_t * pix, int line_size);
1457ab52
MN
135 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
136 me_cmp_func sse[2];
137 me_cmp_func hadamard8_diff[2];
138 me_cmp_func dct_sad[2];
139 me_cmp_func quant_psnr[2];
3a87ac94
MN
140 me_cmp_func bit[2];
141 me_cmp_func rd[2];
1457ab52
MN
142 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
143
826f429a 144 me_cmp_func me_pre_cmp[11];
1457ab52
MN
145 me_cmp_func me_cmp[11];
146 me_cmp_func me_sub_cmp[11];
147 me_cmp_func mb_cmp[11];
eb4b3dd3
ZK
148
149 /* maybe create an array for 16/8 functions */
d518aebd
MN
150 /**
151 * Halfpel motion compensation with rounding (a+b+1)>>1.
eb14c713
MN
152 * this is an array[2][4] of motion compensation funcions for 2
153 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 154 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
155 * @param block destination where the result is stored
156 * @param pixels source
157 * @param line_size number of bytes in a horizontal line of block
158 * @param h height
159 */
eb4b3dd3 160 op_pixels_func put_pixels_tab[2][4];
d518aebd
MN
161
162 /**
163 * Halfpel motion compensation with rounding (a+b+1)>>1.
eb14c713
MN
164 * this is an array[2][4] of motion compensation funcions for 2
165 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 166 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
167 * @param block destination into which the result is averaged (a+b+1)>>1
168 * @param pixels source
169 * @param line_size number of bytes in a horizontal line of block
170 * @param h height
171 */
eb4b3dd3 172 op_pixels_func avg_pixels_tab[2][4];
d518aebd
MN
173
174 /**
175 * Halfpel motion compensation with no rounding (a+b)>>1.
eb14c713
MN
176 * this is an array[2][4] of motion compensation funcions for 2
177 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 178 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
179 * @param block destination where the result is stored
180 * @param pixels source
181 * @param line_size number of bytes in a horizontal line of block
182 * @param h height
183 */
eb4b3dd3 184 op_pixels_func put_no_rnd_pixels_tab[2][4];
d518aebd
MN
185
186 /**
187 * Halfpel motion compensation with no rounding (a+b)>>1.
eb14c713
MN
188 * this is an array[2][4] of motion compensation funcions for 2
189 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 190 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
191 * @param block destination into which the result is averaged (a+b)>>1
192 * @param pixels source
193 * @param line_size number of bytes in a horizontal line of block
194 * @param h height
195 */
eb4b3dd3
ZK
196 op_pixels_func avg_no_rnd_pixels_tab[2][4];
197 qpel_mc_func put_qpel_pixels_tab[2][16];
198 qpel_mc_func avg_qpel_pixels_tab[2][16];
199 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
200 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 201 qpel_mc_func put_mspel_pixels_tab[8];
0da71265
MN
202
203 /**
204 * h264 Chram MC
205 */
206 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
207 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 208
0da71265
MN
209 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
210 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
211
eb4b3dd3
ZK
212 op_pixels_abs_func pix_abs16x16;
213 op_pixels_abs_func pix_abs16x16_x2;
214 op_pixels_abs_func pix_abs16x16_y2;
215 op_pixels_abs_func pix_abs16x16_xy2;
216 op_pixels_abs_func pix_abs8x8;
217 op_pixels_abs_func pix_abs8x8_x2;
218 op_pixels_abs_func pix_abs8x8_y2;
219 op_pixels_abs_func pix_abs8x8_xy2;
11f18faf
MN
220
221 /* huffyuv specific */
11f18faf 222 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 223 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
b0368839
MN
224
225 /* (I)DCT */
226 void (*fdct)(DCTELEM *block/* align 16*/);
24641185
MN
227
228 /**
77c92c2d 229 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 230 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 231 * @param line_size size in bytes of a horizotal line of dest
24641185 232 */
b0368839 233 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
234
235 /**
236 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 237 * @param line_size size in bytes of a horizotal line of dest
24641185 238 */
b0368839 239 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
240
241 /**
77c92c2d 242 * idct input permutation.
24641185
MN
243 * an example to avoid confusion:
244 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
245 * - (x -> referece dct -> reference idct -> x)
246 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
247 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
248 */
b0368839
MN
249 uint8_t idct_permutation[64];
250 int idct_permutation_type;
251#define FF_NO_IDCT_PERM 1
252#define FF_LIBMPEG2_IDCT_PERM 2
253#define FF_SIMPLE_IDCT_PERM 3
254#define FF_TRANSPOSE_IDCT_PERM 4
255
eb4b3dd3
ZK
256} DSPContext;
257
59cf08ce 258void dsputil_static_init(void);
b0368839 259void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 260
7801d21d
MN
261/**
262 * permute block according to permuatation.
263 * @param last last non zero element in scantable order
264 */
0c1a9eda 265void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 266
24641185 267/**
77c92c2d 268 * Empty mmx state.
24641185
MN
269 * this must be called between any dsp function and float/double code.
270 * for example sin(); dsp->idct_put(); emms_c(); cos()
271 */
eb4b3dd3
ZK
272#define emms_c()
273
e629ab68
RD
274/* should be defined by architectures supporting
275 one or more MultiMedia extension */
276int mm_support(void);
277
3d03c0a2 278#if defined(HAVE_MMX)
de6d9b64 279
18f77016 280#undef emms_c
eb4b3dd3 281
de6d9b64
FB
282#define MM_MMX 0x0001 /* standard MMX */
283#define MM_3DNOW 0x0004 /* AMD 3DNOW */
284#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
285#define MM_SSE 0x0008 /* SSE functions */
286#define MM_SSE2 0x0010 /* PIV SSE2 functions */
287
288extern int mm_flags;
289
0c1a9eda
ZK
290void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
291void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
292
293static inline void emms(void)
294{
fb16b7e7
FB
295 __asm __volatile ("emms;":::"memory");
296}
297
1457ab52 298
fb16b7e7
FB
299#define emms_c() \
300{\
301 if (mm_flags & MM_MMX)\
302 emms();\
de6d9b64
FB
303}
304
305#define __align8 __attribute__ ((aligned (8)))
306
b0368839
MN
307void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
308void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 309
3d03c0a2
FB
310#elif defined(ARCH_ARMV4L)
311
3d03c0a2
FB
312/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
313 line ptimizations */
314#define __align8 __attribute__ ((aligned (4)))
315
b0368839 316void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 317
c34270f5 318#elif defined(HAVE_MLIB)
c34270f5
FB
319
320/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
321#define __align8 __attribute__ ((aligned (8)))
322
b0368839 323void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 324
1e98dffb
NK
325#elif defined(ARCH_ALPHA)
326
1e98dffb
NK
327#define __align8 __attribute__ ((aligned (8)))
328
b0368839 329void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 330
59925ef2
BF
331#elif defined(ARCH_POWERPC)
332
404d2241
BF
333#define MM_ALTIVEC 0x0001 /* standard AltiVec */
334
335extern int mm_flags;
336
3b991c54
RD
337#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
338#include <altivec.h>
339#endif
340
59925ef2
BF
341#define __align8 __attribute__ ((aligned (16)))
342
b0368839 343void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 344
d46aba26
LS
345#elif defined(HAVE_MMI)
346
d46aba26
LS
347#define __align8 __attribute__ ((aligned (16)))
348
b0368839 349void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 350
de6d9b64
FB
351#else
352
353#define __align8
354
355#endif
356
6d4985bb
FB
357#ifdef __GNUC__
358
359struct unaligned_64 { uint64_t l; } __attribute__((packed));
360struct unaligned_32 { uint32_t l; } __attribute__((packed));
361
362#define LD32(a) (((const struct unaligned_32 *) (a))->l)
363#define LD64(a) (((const struct unaligned_64 *) (a))->l)
364
365#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
366
367#else /* __GNUC__ */
368
369#define LD32(a) (*((uint32_t*)(a)))
370#define LD64(a) (*((uint64_t*)(a)))
371
372#define ST32(a, b) *((uint32_t*)(a)) = (b)
373
374#endif /* !__GNUC__ */
375
43f1708f 376/* PSNR */
0c1a9eda 377void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
378 int orig_linesize[3], int coded_linesize,
379 AVCodecContext *avctx);
bb6f5690
FB
380
381/* FFT computation */
382
383/* NOTE: soon integer code will be added, so you must use the
384 FFTSample type */
385typedef float FFTSample;
386
387typedef struct FFTComplex {
388 FFTSample re, im;
389} FFTComplex;
390
391typedef struct FFTContext {
392 int nbits;
393 int inverse;
394 uint16_t *revtab;
395 FFTComplex *exptab;
396 FFTComplex *exptab1; /* only used by SSE code */
397 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
398} FFTContext;
399
400int fft_init(FFTContext *s, int nbits, int inverse);
401void fft_permute(FFTContext *s, FFTComplex *z);
402void fft_calc_c(FFTContext *s, FFTComplex *z);
403void fft_calc_sse(FFTContext *s, FFTComplex *z);
8d268a7d
FB
404void fft_calc_altivec(FFTContext *s, FFTComplex *z);
405
bb6f5690
FB
406static inline void fft_calc(FFTContext *s, FFTComplex *z)
407{
408 s->fft_calc(s, z);
409}
410void fft_end(FFTContext *s);
411
412/* MDCT computation */
413
414typedef struct MDCTContext {
415 int n; /* size of MDCT (i.e. number of input data * 2) */
416 int nbits; /* n = 2^nbits */
417 /* pre/post rotation tables */
418 FFTSample *tcos;
419 FFTSample *tsin;
420 FFTContext fft;
421} MDCTContext;
422
82696bee 423int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 424void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 425 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 426void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 427 const FFTSample *input, FFTSample *tmp);
82696bee 428void ff_mdct_end(MDCTContext *s);
bb6f5690 429
1457ab52
MN
430#define WARPER88_1616(name8, name16)\
431static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
432 return name8(s, dst , src , stride)\
433 +name8(s, dst+8 , src+8 , stride)\
434 +name8(s, dst +8*stride, src +8*stride, stride)\
435 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
436}
437
95e2ce4a 438#ifndef HAVE_LRINTF
9d85cbd9
FB
439/* XXX: add ISOC specific test to avoid specific BSD testing. */
440/* better than nothing implementation. */
6234d753 441/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
442static inline long int lrintf(float x)
443{
ea937d01
FB
444#ifdef CONFIG_WIN32
445 /* XXX: incorrect, but make it compile */
446 return (int)(x);
447#else
9d85cbd9 448 return (int)(rint(x));
ea937d01 449#endif
9d85cbd9
FB
450}
451#endif
452
de6d9b64 453#endif