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