Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 0da71265

History | View | Annotate | Download (14.2 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
 */
24

    
25
#ifndef DSPUTIL_H
26
#define DSPUTIL_H
27

    
28
#include "common.h"
29
#include "avcodec.h"
30

    
31

    
32
//#define DEBUG
33
/* dct code */
34
typedef short DCTELEM;
35
//typedef int DCTELEM;
36

    
37
void fdct_ifast (DCTELEM *data);
38
void ff_jpeg_fdct_islow (DCTELEM *data);
39

    
40
void j_rev_dct (DCTELEM *data);
41

    
42
void ff_fdct_mmx(DCTELEM *block);
43

    
44
/* encoding scans */
45
extern const uint8_t ff_alternate_horizontal_scan[64];
46
extern const uint8_t ff_alternate_vertical_scan[64];
47
extern const uint8_t ff_zigzag_direct[64];
48

    
49
/* pixel operations */
50
#define MAX_NEG_CROP 384
51

    
52
/* temporary */
53
extern uint32_t squareTbl[512];
54
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
55

    
56

    
57
/* minimum alignment rules ;)
58
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
59
or 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)
62
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
63

64
!future video codecs might need functions with less strict alignment
65
*/
66

    
67
/*
68
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
69
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
70
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
71
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
72
void clear_blocks_c(DCTELEM *blocks);
73
*/
74

    
75
/* add and put pixel (decoding) */
76
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
77
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
78
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
79
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
80

    
81
#define DEF_OLD_QPEL(name)\
82
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
83
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
84
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
85

    
86
DEF_OLD_QPEL(qpel16_mc11_old_c)
87
DEF_OLD_QPEL(qpel16_mc31_old_c)
88
DEF_OLD_QPEL(qpel16_mc12_old_c)
89
DEF_OLD_QPEL(qpel16_mc32_old_c)
90
DEF_OLD_QPEL(qpel16_mc13_old_c)
91
DEF_OLD_QPEL(qpel16_mc33_old_c)
92
DEF_OLD_QPEL(qpel8_mc11_old_c)
93
DEF_OLD_QPEL(qpel8_mc31_old_c)
94
DEF_OLD_QPEL(qpel8_mc12_old_c)
95
DEF_OLD_QPEL(qpel8_mc32_old_c)
96
DEF_OLD_QPEL(qpel8_mc13_old_c)
97
DEF_OLD_QPEL(qpel8_mc33_old_c)
98

    
99
#define CALL_2X_PIXELS(a, b, n)\
100
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
101
    b(block  , pixels  , line_size, h);\
102
    b(block+n, pixels+n, line_size, h);\
103
}
104

    
105
/* motion estimation */
106

    
107
typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
108

    
109
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))*/;
110

    
111

    
112
/**
113
 * DSPContext.
114
 */
115
typedef struct DSPContext {
116
    /* pixel ops : interface with DCT */
117
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
118
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
119
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
120
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
121
    /**
122
     * translational global motion compensation.
123
     */
124
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
125
    /**
126
     * global motion compensation.
127
     */
128
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
129
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
130
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
131
    int (*pix_sum)(uint8_t * pix, int line_size);
132
    int (*pix_norm1)(uint8_t * pix, int line_size);
133
    me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
134
    me_cmp_func sse[2];
135
    me_cmp_func hadamard8_diff[2];
136
    me_cmp_func dct_sad[2];
137
    me_cmp_func quant_psnr[2];
138
    me_cmp_func bit[2];
139
    me_cmp_func rd[2];
140
    int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
141

    
142
    me_cmp_func me_pre_cmp[11];
143
    me_cmp_func me_cmp[11];
144
    me_cmp_func me_sub_cmp[11];
145
    me_cmp_func mb_cmp[11];
146

    
147
    /* maybe create an array for 16/8 functions */
148
    /**
149
     * Halfpel motion compensation with rounding (a+b+1)>>1.
150
     * *pixels_tab[ 0->16x16 1->8x8 ][ xhalfpel + 2*yhalfpel ]
151
     * @param block destination where the result is stored
152
     * @param pixels source
153
     * @param line_size number of bytes in a horizontal line of block
154
     * @param h height
155
     */
156
    op_pixels_func put_pixels_tab[2][4];
157

    
158
    /**
159
     * Halfpel motion compensation with rounding (a+b+1)>>1.
160
     * *pixels_tab[ 0->16x16 1->8x8 ][ xhalfpel + 2*yhalfpel ]
161
     * @param block destination into which the result is averaged (a+b+1)>>1
162
     * @param pixels source
163
     * @param line_size number of bytes in a horizontal line of block
164
     * @param h height
165
     */
166
    op_pixels_func avg_pixels_tab[2][4];
167

    
168
    /**
169
     * Halfpel motion compensation with no rounding (a+b)>>1.
170
     * *pixels_tab[ 0->16x16 1->8x8 ][ xhalfpel + 2*yhalfpel ]
171
     * @param block destination where the result is stored
172
     * @param pixels source
173
     * @param line_size number of bytes in a horizontal line of block
174
     * @param h height
175
     */
176
    op_pixels_func put_no_rnd_pixels_tab[2][4];
177

    
178
    /**
179
     * Halfpel motion compensation with no rounding (a+b)>>1.
180
     * *pixels_tab[ 0->16x16 1->8x8 ][ xhalfpel + 2*yhalfpel ]
181
     * @param block destination into which the result is averaged (a+b)>>1
182
     * @param pixels source
183
     * @param line_size number of bytes in a horizontal line of block
184
     * @param h height
185
     */
186
    op_pixels_func avg_no_rnd_pixels_tab[2][4];
187
    qpel_mc_func put_qpel_pixels_tab[2][16];
188
    qpel_mc_func avg_qpel_pixels_tab[2][16];
189
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
190
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
191
    qpel_mc_func put_mspel_pixels_tab[8];
192
    
193
    /**
194
     * h264 Chram MC
195
     */
196
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
197
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
198

    
199
    qpel_mc_func put_h264_qpel_pixels_tab[3][16];
200
    qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
201
    
202
    op_pixels_abs_func pix_abs16x16;
203
    op_pixels_abs_func pix_abs16x16_x2;
204
    op_pixels_abs_func pix_abs16x16_y2;
205
    op_pixels_abs_func pix_abs16x16_xy2;
206
    op_pixels_abs_func pix_abs8x8;
207
    op_pixels_abs_func pix_abs8x8_x2;
208
    op_pixels_abs_func pix_abs8x8_y2;
209
    op_pixels_abs_func pix_abs8x8_xy2;
210
    
211
    /* huffyuv specific */
212
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
213
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
214
    
215
    /* (I)DCT */
216
    void (*fdct)(DCTELEM *block/* align 16*/);
217
    
218
    /**
219
     * block -> idct -> clip to unsigned 8 bit -> dest.
220
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
221
     * @param line_size size in bytes of a horizotal line of dest
222
     */
223
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
224
    
225
    /**
226
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
227
     * @param line_size size in bytes of a horizotal line of dest
228
     */
229
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
230
    
231
    /**
232
     * idct input permutation.
233
     * an example to avoid confusion:
234
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
235
     * - (x -> referece dct -> reference idct -> x)
236
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
237
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
238
     */
239
    uint8_t idct_permutation[64];
240
    int idct_permutation_type;
241
#define FF_NO_IDCT_PERM 1
242
#define FF_LIBMPEG2_IDCT_PERM 2
243
#define FF_SIMPLE_IDCT_PERM 3
244
#define FF_TRANSPOSE_IDCT_PERM 4
245

    
246
} DSPContext;
247

    
248
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
249

    
250
/**
251
 * permute block according to permuatation.
252
 * @param last last non zero element in scantable order
253
 */
254
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
255

    
256
/**
257
 * Empty mmx state.
258
 * this must be called between any dsp function and float/double code.
259
 * for example sin(); dsp->idct_put(); emms_c(); cos()
260
 */
261
#define emms_c()
262

    
263
/* should be defined by architectures supporting
264
   one or more MultiMedia extension */
265
int mm_support(void);
266

    
267
#if defined(HAVE_MMX)
268

    
269
#undef emms_c
270

    
271
#define MM_MMX    0x0001 /* standard MMX */
272
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
273
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
274
#define MM_SSE    0x0008 /* SSE functions */
275
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
276

    
277
extern int mm_flags;
278

    
279
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
280
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
281

    
282
static inline void emms(void)
283
{
284
    __asm __volatile ("emms;":::"memory");
285
}
286

    
287

    
288
#define emms_c() \
289
{\
290
    if (mm_flags & MM_MMX)\
291
        emms();\
292
}
293

    
294
#define __align8 __attribute__ ((aligned (8)))
295

    
296
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
297
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
298

    
299
#elif defined(ARCH_ARMV4L)
300

    
301
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
302
   line ptimizations */
303
#define __align8 __attribute__ ((aligned (4)))
304

    
305
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
306

    
307
#elif defined(HAVE_MLIB)
308

    
309
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
310
#define __align8 __attribute__ ((aligned (8)))
311

    
312
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
313

    
314
#elif defined(ARCH_ALPHA)
315

    
316
#define __align8 __attribute__ ((aligned (8)))
317

    
318
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
319

    
320
#elif defined(ARCH_POWERPC)
321

    
322
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
323

    
324
extern int mm_flags;
325

    
326
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
327
#include <altivec.h>
328
#endif
329

    
330
#define __align8 __attribute__ ((aligned (16)))
331

    
332
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
333

    
334
#elif defined(HAVE_MMI)
335

    
336
#define __align8 __attribute__ ((aligned (16)))
337

    
338
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
339

    
340
#else
341

    
342
#define __align8
343

    
344
#endif
345

    
346
#ifdef __GNUC__
347

    
348
struct unaligned_64 { uint64_t l; } __attribute__((packed));
349
struct unaligned_32 { uint32_t l; } __attribute__((packed));
350

    
351
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
352
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
353

    
354
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
355

    
356
#else /* __GNUC__ */
357

    
358
#define LD32(a) (*((uint32_t*)(a)))
359
#define LD64(a) (*((uint64_t*)(a)))
360

    
361
#define ST32(a, b) *((uint32_t*)(a)) = (b)
362

    
363
#endif /* !__GNUC__ */
364

    
365
/* PSNR */
366
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
367
              int orig_linesize[3], int coded_linesize,
368
              AVCodecContext *avctx);
369

    
370
/* FFT computation */
371

    
372
/* NOTE: soon integer code will be added, so you must use the
373
   FFTSample type */
374
typedef float FFTSample;
375

    
376
typedef struct FFTComplex {
377
    FFTSample re, im;
378
} FFTComplex;
379

    
380
typedef struct FFTContext {
381
    int nbits;
382
    int inverse;
383
    uint16_t *revtab;
384
    FFTComplex *exptab;
385
    FFTComplex *exptab1; /* only used by SSE code */
386
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
387
} FFTContext;
388

    
389
int fft_init(FFTContext *s, int nbits, int inverse);
390
void fft_permute(FFTContext *s, FFTComplex *z);
391
void fft_calc_c(FFTContext *s, FFTComplex *z);
392
void fft_calc_sse(FFTContext *s, FFTComplex *z);
393
void fft_calc_altivec(FFTContext *s, FFTComplex *z);
394

    
395
static inline void fft_calc(FFTContext *s, FFTComplex *z)
396
{
397
    s->fft_calc(s, z);
398
}
399
void fft_end(FFTContext *s);
400

    
401
/* MDCT computation */
402

    
403
typedef struct MDCTContext {
404
    int n;  /* size of MDCT (i.e. number of input data * 2) */
405
    int nbits; /* n = 2^nbits */
406
    /* pre/post rotation tables */
407
    FFTSample *tcos;
408
    FFTSample *tsin;
409
    FFTContext fft;
410
} MDCTContext;
411

    
412
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
413
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
414
                const FFTSample *input, FFTSample *tmp);
415
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
416
               const FFTSample *input, FFTSample *tmp);
417
void ff_mdct_end(MDCTContext *s);
418

    
419
#define WARPER88_1616(name8, name16)\
420
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
421
    return name8(s, dst           , src           , stride)\
422
          +name8(s, dst+8         , src+8         , stride)\
423
          +name8(s, dst  +8*stride, src  +8*stride, stride)\
424
          +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
425
}
426

    
427
#ifndef HAVE_LRINTF
428
/* XXX: add ISOC specific test to avoid specific BSD testing. */
429
/* better than nothing implementation. */
430
/* btw, rintf() is existing on fbsd too -- alex */
431
static inline long int lrintf(float x)
432
{
433
#ifdef CONFIG_WIN32
434
    /* XXX: incorrect, but make it compile */
435
    return (int)(x);
436
#else
437
    return (int)(rint(x));
438
#endif
439
}
440
#endif
441

    
442
#endif