Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 05493021

History | View | Annotate | Download (15.2 KB)

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

    
20
/**
21
 * @file dsputil.h
22
 * DSP utils.
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
25
 */
26

    
27
#ifndef DSPUTIL_H
28
#define DSPUTIL_H
29

    
30
#include "common.h"
31
#include "avcodec.h"
32

    
33

    
34
//#define DEBUG
35
/* dct code */
36
typedef short DCTELEM;
37
//typedef int DCTELEM;
38

    
39
void fdct_ifast (DCTELEM *data);
40
void ff_jpeg_fdct_islow (DCTELEM *data);
41

    
42
void j_rev_dct (DCTELEM *data);
43

    
44
void ff_fdct_mmx(DCTELEM *block);
45

    
46
/* encoding scans */
47
extern const uint8_t ff_alternate_horizontal_scan[64];
48
extern const uint8_t ff_alternate_vertical_scan[64];
49
extern const uint8_t ff_zigzag_direct[64];
50

    
51
/* pixel operations */
52
#define MAX_NEG_CROP 384
53

    
54
/* temporary */
55
extern uint32_t squareTbl[512];
56
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
57

    
58

    
59
/* minimum alignment rules ;)
60
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
61
or 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)
64
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
65

66
!future video codecs might need functions with less strict alignment
67
*/
68

    
69
/*
70
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
71
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
72
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
73
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
74
void clear_blocks_c(DCTELEM *blocks);
75
*/
76

    
77
/* add and put pixel (decoding) */
78
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
79
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
80
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
81
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
82

    
83
#define DEF_OLD_QPEL(name)\
84
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
85
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
86
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
87

    
88
DEF_OLD_QPEL(qpel16_mc11_old_c)
89
DEF_OLD_QPEL(qpel16_mc31_old_c)
90
DEF_OLD_QPEL(qpel16_mc12_old_c)
91
DEF_OLD_QPEL(qpel16_mc32_old_c)
92
DEF_OLD_QPEL(qpel16_mc13_old_c)
93
DEF_OLD_QPEL(qpel16_mc33_old_c)
94
DEF_OLD_QPEL(qpel8_mc11_old_c)
95
DEF_OLD_QPEL(qpel8_mc31_old_c)
96
DEF_OLD_QPEL(qpel8_mc12_old_c)
97
DEF_OLD_QPEL(qpel8_mc32_old_c)
98
DEF_OLD_QPEL(qpel8_mc13_old_c)
99
DEF_OLD_QPEL(qpel8_mc33_old_c)
100

    
101
#define CALL_2X_PIXELS(a, b, n)\
102
static 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
}
106

    
107
/* motion estimation */
108

    
109
typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
110

    
111
typedef 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))*/;
112

    
113

    
114
/**
115
 * DSPContext.
116
 */
117
typedef struct DSPContext {
118
    /* pixel ops : interface with DCT */
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);
123
    /**
124
     * translational global motion compensation.
125
     */
126
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
127
    /**
128
     * global motion compensation.
129
     */
130
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
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*/);
133
    int (*pix_sum)(uint8_t * pix, int line_size);
134
    int (*pix_norm1)(uint8_t * pix, int line_size);
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];
140
    me_cmp_func bit[2];
141
    me_cmp_func rd[2];
142
    int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
143

    
144
    me_cmp_func me_pre_cmp[11];
145
    me_cmp_func me_cmp[11];
146
    me_cmp_func me_sub_cmp[11];
147
    me_cmp_func mb_cmp[11];
148

    
149
    /* maybe create an array for 16/8 functions */
150
    /**
151
     * Halfpel motion compensation with rounding (a+b+1)>>1.
152
     * this is an array[2][4] of motion compensation funcions for 2 
153
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
154
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
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
     */
160
    op_pixels_func put_pixels_tab[2][4];
161

    
162
    /**
163
     * Halfpel motion compensation with rounding (a+b+1)>>1.
164
     * this is an array[2][4] of motion compensation funcions for 2 
165
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
166
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
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
     */
172
    op_pixels_func avg_pixels_tab[2][4];
173

    
174
    /**
175
     * Halfpel motion compensation with no rounding (a+b)>>1.
176
     * this is an array[2][4] of motion compensation funcions for 2 
177
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
178
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
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
     */
184
    op_pixels_func put_no_rnd_pixels_tab[2][4];
185

    
186
    /**
187
     * Halfpel motion compensation with no rounding (a+b)>>1.
188
     * this is an array[2][4] of motion compensation funcions for 2 
189
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
190
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
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
     */
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];
201
    qpel_mc_func put_mspel_pixels_tab[8];
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];
208

    
209
    qpel_mc_func put_h264_qpel_pixels_tab[3][16];
210
    qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
211
    
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;
220
    
221
    /* huffyuv specific */
222
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
223
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
224
    
225
    /* (I)DCT */
226
    void (*fdct)(DCTELEM *block/* align 16*/);
227
    
228
    /**
229
     * block -> idct -> clip to unsigned 8 bit -> dest.
230
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
231
     * @param line_size size in bytes of a horizotal line of dest
232
     */
233
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
234
    
235
    /**
236
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
237
     * @param line_size size in bytes of a horizotal line of dest
238
     */
239
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
240
    
241
    /**
242
     * idct input permutation.
243
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
244
     * IDCT)
245
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
246
     * with the zigzag/alternate scan<br>
247
     * an example to avoid confusion:
248
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
249
     * - (x -> referece dct -> reference idct -> x)
250
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
251
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
252
     */
253
    uint8_t idct_permutation[64];
254
    int idct_permutation_type;
255
#define FF_NO_IDCT_PERM 1
256
#define FF_LIBMPEG2_IDCT_PERM 2
257
#define FF_SIMPLE_IDCT_PERM 3
258
#define FF_TRANSPOSE_IDCT_PERM 4
259

    
260
} DSPContext;
261

    
262
void dsputil_static_init(void);
263
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
264

    
265
/**
266
 * permute block according to permuatation.
267
 * @param last last non zero element in scantable order
268
 */
269
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
270

    
271
/**
272
 * Empty mmx state.
273
 * this must be called between any dsp function and float/double code.
274
 * for example sin(); dsp->idct_put(); emms_c(); cos()
275
 */
276
#define emms_c()
277

    
278
/* should be defined by architectures supporting
279
   one or more MultiMedia extension */
280
int mm_support(void);
281

    
282
#if defined(HAVE_MMX)
283

    
284
#undef emms_c
285

    
286
#define MM_MMX    0x0001 /* standard MMX */
287
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
288
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
289
#define MM_SSE    0x0008 /* SSE functions */
290
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
291

    
292
extern int mm_flags;
293

    
294
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
295
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
296

    
297
static inline void emms(void)
298
{
299
    __asm __volatile ("emms;":::"memory");
300
}
301

    
302

    
303
#define emms_c() \
304
{\
305
    if (mm_flags & MM_MMX)\
306
        emms();\
307
}
308

    
309
#define __align8 __attribute__ ((aligned (8)))
310

    
311
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
312
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
313

    
314
#elif defined(ARCH_ARMV4L)
315

    
316
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
317
   line ptimizations */
318
#define __align8 __attribute__ ((aligned (4)))
319

    
320
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
321

    
322
#elif defined(HAVE_MLIB)
323

    
324
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
325
#define __align8 __attribute__ ((aligned (8)))
326

    
327
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
328

    
329
#elif defined(ARCH_ALPHA)
330

    
331
#define __align8 __attribute__ ((aligned (8)))
332

    
333
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
334

    
335
#elif defined(ARCH_POWERPC)
336

    
337
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
338

    
339
extern int mm_flags;
340

    
341
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
342
#include <altivec.h>
343
#endif
344

    
345
#define __align8 __attribute__ ((aligned (16)))
346

    
347
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
348

    
349
#elif defined(HAVE_MMI)
350

    
351
#define __align8 __attribute__ ((aligned (16)))
352

    
353
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
354

    
355
#else
356

    
357
#define __align8
358

    
359
#endif
360

    
361
#ifdef __GNUC__
362

    
363
struct unaligned_64 { uint64_t l; } __attribute__((packed));
364
struct unaligned_32 { uint32_t l; } __attribute__((packed));
365

    
366
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
367
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
368

    
369
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
370

    
371
#else /* __GNUC__ */
372

    
373
#define LD32(a) (*((uint32_t*)(a)))
374
#define LD64(a) (*((uint64_t*)(a)))
375

    
376
#define ST32(a, b) *((uint32_t*)(a)) = (b)
377

    
378
#endif /* !__GNUC__ */
379

    
380
/* PSNR */
381
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
382
              int orig_linesize[3], int coded_linesize,
383
              AVCodecContext *avctx);
384

    
385
/* FFT computation */
386

    
387
/* NOTE: soon integer code will be added, so you must use the
388
   FFTSample type */
389
typedef float FFTSample;
390

    
391
typedef struct FFTComplex {
392
    FFTSample re, im;
393
} FFTComplex;
394

    
395
typedef struct FFTContext {
396
    int nbits;
397
    int inverse;
398
    uint16_t *revtab;
399
    FFTComplex *exptab;
400
    FFTComplex *exptab1; /* only used by SSE code */
401
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
402
} FFTContext;
403

    
404
int fft_init(FFTContext *s, int nbits, int inverse);
405
void fft_permute(FFTContext *s, FFTComplex *z);
406
void fft_calc_c(FFTContext *s, FFTComplex *z);
407
void fft_calc_sse(FFTContext *s, FFTComplex *z);
408
void fft_calc_altivec(FFTContext *s, FFTComplex *z);
409

    
410
static inline void fft_calc(FFTContext *s, FFTComplex *z)
411
{
412
    s->fft_calc(s, z);
413
}
414
void fft_end(FFTContext *s);
415

    
416
/* MDCT computation */
417

    
418
typedef struct MDCTContext {
419
    int n;  /* size of MDCT (i.e. number of input data * 2) */
420
    int nbits; /* n = 2^nbits */
421
    /* pre/post rotation tables */
422
    FFTSample *tcos;
423
    FFTSample *tsin;
424
    FFTContext fft;
425
} MDCTContext;
426

    
427
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
428
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
429
                const FFTSample *input, FFTSample *tmp);
430
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
431
               const FFTSample *input, FFTSample *tmp);
432
void ff_mdct_end(MDCTContext *s);
433

    
434
#define WARPER88_1616(name8, name16)\
435
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
436
    return name8(s, dst           , src           , stride)\
437
          +name8(s, dst+8         , src+8         , stride)\
438
          +name8(s, dst  +8*stride, src  +8*stride, stride)\
439
          +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
440
}
441

    
442
#ifndef HAVE_LRINTF
443
/* XXX: add ISOC specific test to avoid specific BSD testing. */
444
/* better than nothing implementation. */
445
/* btw, rintf() is existing on fbsd too -- alex */
446
static inline long int lrintf(float x)
447
{
448
#ifdef CONFIG_WIN32
449
    /* XXX: incorrect, but make it compile */
450
    return (int)(x);
451
#else
452
    return (int)(rint(x));
453
#endif
454
}
455
#endif
456

    
457
#endif