Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 84705403

History | View | Annotate | Download (16.9 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

    
38
void fdct_ifast (DCTELEM *data);
39
void ff_jpeg_fdct_islow (DCTELEM *data);
40

    
41
void j_rev_dct (DCTELEM *data);
42

    
43
void ff_fdct_mmx(DCTELEM *block);
44

    
45
/* encoding scans */
46
extern const uint8_t ff_alternate_horizontal_scan[64];
47
extern const uint8_t ff_alternate_vertical_scan[64];
48
extern const uint8_t ff_zigzag_direct[64];
49

    
50
/* pixel operations */
51
#define MAX_NEG_CROP 384
52

    
53
/* temporary */
54
extern uint32_t squareTbl[512];
55
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
56

    
57

    
58
/* minimum alignment rules ;)
59
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
60
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
61

62
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
63
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
64

65
!future video codecs might need functions with less strict alignment
66
*/
67

    
68
/*
69
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
70
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
71
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
72
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
73
void clear_blocks_c(DCTELEM *blocks);
74
*/
75

    
76
/* add and put pixel (decoding) */
77
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
78
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
79
typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, 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 16/8/4/2 functions */
150
    /**
151
     * Halfpel motion compensation with rounding (a+b+1)>>1.
152
     * this is an array[4][4] of motion compensation funcions for 4 
153
     * horizontal blocksizes (2,4,8,16) 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[4][4];
161

    
162
    /**
163
     * Halfpel motion compensation with rounding (a+b+1)>>1.
164
     * This is an array[4][4] of motion compensation functions for 4 
165
     * horizontal blocksizes (2,4,8,16) 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[4][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
    
198
    /**
199
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
200
     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
201
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
202
     * @param block destination where the result is stored
203
     * @param pixels source
204
     * @param line_size number of bytes in a horizontal line of block
205
     * @param h height
206
     */
207
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
208
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
209

    
210
    qpel_mc_func put_qpel_pixels_tab[2][16];
211
    qpel_mc_func avg_qpel_pixels_tab[2][16];
212
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
213
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
214
    qpel_mc_func put_mspel_pixels_tab[8];
215
    
216
    /**
217
     * h264 Chram MC
218
     */
219
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
220
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
221

    
222
    qpel_mc_func put_h264_qpel_pixels_tab[3][16];
223
    qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
224
    
225
    op_pixels_abs_func pix_abs16x16;
226
    op_pixels_abs_func pix_abs16x16_x2;
227
    op_pixels_abs_func pix_abs16x16_y2;
228
    op_pixels_abs_func pix_abs16x16_xy2;
229
    op_pixels_abs_func pix_abs8x8;
230
    op_pixels_abs_func pix_abs8x8_x2;
231
    op_pixels_abs_func pix_abs8x8_y2;
232
    op_pixels_abs_func pix_abs8x8_xy2;
233
    
234
    /* huffyuv specific */
235
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
236
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
237
    /**
238
     * subtract huffyuv's variant of median prediction
239
     * note, this might read from src1[-1], src2[-1]
240
     */
241
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
242
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
243
    
244
    /* (I)DCT */
245
    void (*fdct)(DCTELEM *block/* align 16*/);
246
    
247
    /* IDCT really*/
248
    void (*idct)(DCTELEM *block/* align 16*/);
249
    
250
    /**
251
     * block -> idct -> clip to unsigned 8 bit -> dest.
252
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
253
     * @param line_size size in bytes of a horizotal line of dest
254
     */
255
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
256
    
257
    /**
258
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
259
     * @param line_size size in bytes of a horizotal line of dest
260
     */
261
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
262
    
263
    /**
264
     * idct input permutation.
265
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
266
     * IDCT)
267
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
268
     * with the zigzag/alternate scan<br>
269
     * an example to avoid confusion:
270
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
271
     * - (x -> referece dct -> reference idct -> x)
272
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
273
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
274
     */
275
    uint8_t idct_permutation[64];
276
    int idct_permutation_type;
277
#define FF_NO_IDCT_PERM 1
278
#define FF_LIBMPEG2_IDCT_PERM 2
279
#define FF_SIMPLE_IDCT_PERM 3
280
#define FF_TRANSPOSE_IDCT_PERM 4
281

    
282
} DSPContext;
283

    
284
void dsputil_static_init(void);
285
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
286

    
287
/**
288
 * permute block according to permuatation.
289
 * @param last last non zero element in scantable order
290
 */
291
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
292

    
293
#define        BYTE_VEC32(c)        ((c)*0x01010101UL)
294

    
295
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
296
{
297
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
298
}
299

    
300
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
301
{
302
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
303
}
304

    
305
/**
306
 * Empty mmx state.
307
 * this must be called between any dsp function and float/double code.
308
 * for example sin(); dsp->idct_put(); emms_c(); cos()
309
 */
310
#define emms_c()
311

    
312
/* should be defined by architectures supporting
313
   one or more MultiMedia extension */
314
int mm_support(void);
315

    
316
#if defined(HAVE_MMX)
317

    
318
#undef emms_c
319

    
320
#define MM_MMX    0x0001 /* standard MMX */
321
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
322
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
323
#define MM_SSE    0x0008 /* SSE functions */
324
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
325

    
326
extern int mm_flags;
327

    
328
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
329
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
330

    
331
static inline void emms(void)
332
{
333
    __asm __volatile ("emms;":::"memory");
334
}
335

    
336

    
337
#define emms_c() \
338
{\
339
    if (mm_flags & MM_MMX)\
340
        emms();\
341
}
342

    
343
#define __align8 __attribute__ ((aligned (8)))
344

    
345
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
346
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
347

    
348
#elif defined(ARCH_ARMV4L)
349

    
350
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
351
   line ptimizations */
352
#define __align8 __attribute__ ((aligned (4)))
353

    
354
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
355

    
356
#elif defined(HAVE_MLIB)
357

    
358
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
359
#define __align8 __attribute__ ((aligned (8)))
360

    
361
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
362

    
363
#elif defined(ARCH_ALPHA)
364

    
365
#define __align8 __attribute__ ((aligned (8)))
366

    
367
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
368

    
369
#elif defined(ARCH_POWERPC)
370

    
371
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
372

    
373
extern int mm_flags;
374

    
375
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
376
#include <altivec.h>
377
#endif
378

    
379
#define __align8 __attribute__ ((aligned (16)))
380

    
381
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
382

    
383
#elif defined(HAVE_MMI)
384

    
385
#define __align8 __attribute__ ((aligned (16)))
386

    
387
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
388

    
389
#elif defined(ARCH_SH4)
390

    
391
#define __align8 __attribute__ ((aligned (8)))
392

    
393
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
394

    
395
#else
396

    
397
#define __align8
398

    
399
#endif
400

    
401
#ifdef __GNUC__
402

    
403
struct unaligned_64 { uint64_t l; } __attribute__((packed));
404
struct unaligned_32 { uint32_t l; } __attribute__((packed));
405
struct unaligned_16 { uint16_t l; } __attribute__((packed));
406

    
407
#define LD16(a) (((const struct unaligned_16 *) (a))->l)
408
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
409
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
410

    
411
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
412

    
413
#else /* __GNUC__ */
414

    
415
#define LD16(a) (*((uint16_t*)(a)))
416
#define LD32(a) (*((uint32_t*)(a)))
417
#define LD64(a) (*((uint64_t*)(a)))
418

    
419
#define ST32(a, b) *((uint32_t*)(a)) = (b)
420

    
421
#endif /* !__GNUC__ */
422

    
423
/* PSNR */
424
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
425
              int orig_linesize[3], int coded_linesize,
426
              AVCodecContext *avctx);
427

    
428
/* FFT computation */
429

    
430
/* NOTE: soon integer code will be added, so you must use the
431
   FFTSample type */
432
typedef float FFTSample;
433

    
434
typedef struct FFTComplex {
435
    FFTSample re, im;
436
} FFTComplex;
437

    
438
typedef struct FFTContext {
439
    int nbits;
440
    int inverse;
441
    uint16_t *revtab;
442
    FFTComplex *exptab;
443
    FFTComplex *exptab1; /* only used by SSE code */
444
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
445
} FFTContext;
446

    
447
int fft_init(FFTContext *s, int nbits, int inverse);
448
void fft_permute(FFTContext *s, FFTComplex *z);
449
void fft_calc_c(FFTContext *s, FFTComplex *z);
450
void fft_calc_sse(FFTContext *s, FFTComplex *z);
451
void fft_calc_altivec(FFTContext *s, FFTComplex *z);
452

    
453
static inline void fft_calc(FFTContext *s, FFTComplex *z)
454
{
455
    s->fft_calc(s, z);
456
}
457
void fft_end(FFTContext *s);
458

    
459
/* MDCT computation */
460

    
461
typedef struct MDCTContext {
462
    int n;  /* size of MDCT (i.e. number of input data * 2) */
463
    int nbits; /* n = 2^nbits */
464
    /* pre/post rotation tables */
465
    FFTSample *tcos;
466
    FFTSample *tsin;
467
    FFTContext fft;
468
} MDCTContext;
469

    
470
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
471
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
472
                const FFTSample *input, FFTSample *tmp);
473
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
474
               const FFTSample *input, FFTSample *tmp);
475
void ff_mdct_end(MDCTContext *s);
476

    
477
#define WARPER88_1616(name8, name16)\
478
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
479
    return name8(s, dst           , src           , stride)\
480
          +name8(s, dst+8         , src+8         , stride)\
481
          +name8(s, dst  +8*stride, src  +8*stride, stride)\
482
          +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
483
}
484

    
485
#ifndef HAVE_LRINTF
486
/* XXX: add ISOC specific test to avoid specific BSD testing. */
487
/* better than nothing implementation. */
488
/* btw, rintf() is existing on fbsd too -- alex */
489
static inline long int lrintf(float x)
490
{
491
#ifdef CONFIG_WIN32
492
    /* XXX: incorrect, but make it compile */
493
    return (int)(x);
494
#else
495
    return (int)(rint(x));
496
#endif
497
}
498
#endif
499

    
500
#if defined(CONFIG_OS2) || defined(CONFIG_SUNOS)
501
static inline float floorf(float f) { 
502
    return floor(f); 
503
}
504
#endif
505

    
506
#endif