Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 983e3246

History | View | Annotate | Download (12.4 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

    
80
#define DEF_OLD_QPEL(name)\
81
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
82
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
83
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
84

    
85
DEF_OLD_QPEL(qpel16_mc11_old_c)
86
DEF_OLD_QPEL(qpel16_mc31_old_c)
87
DEF_OLD_QPEL(qpel16_mc12_old_c)
88
DEF_OLD_QPEL(qpel16_mc32_old_c)
89
DEF_OLD_QPEL(qpel16_mc13_old_c)
90
DEF_OLD_QPEL(qpel16_mc33_old_c)
91
DEF_OLD_QPEL(qpel8_mc11_old_c)
92
DEF_OLD_QPEL(qpel8_mc31_old_c)
93
DEF_OLD_QPEL(qpel8_mc12_old_c)
94
DEF_OLD_QPEL(qpel8_mc32_old_c)
95
DEF_OLD_QPEL(qpel8_mc13_old_c)
96
DEF_OLD_QPEL(qpel8_mc33_old_c)
97

    
98
#define CALL_2X_PIXELS(a, b, n)\
99
static 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
}
103

    
104
/* motion estimation */
105

    
106
typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
107

    
108
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))*/;
109

    
110
/**
111
 * DSPContext.
112
 */
113
typedef struct DSPContext {
114
    /* pixel ops : interface with DCT */
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);
119
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
120
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
121
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
122
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
123
    int (*pix_sum)(uint8_t * pix, int line_size);
124
    int (*pix_norm1)(uint8_t * pix, int line_size);
125
    me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
126
    me_cmp_func sse[2];
127
    me_cmp_func hadamard8_diff[2];
128
    me_cmp_func dct_sad[2];
129
    me_cmp_func quant_psnr[2];
130
    me_cmp_func bit[2];
131
    me_cmp_func rd[2];
132
    int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
133

    
134
    me_cmp_func me_pre_cmp[11];
135
    me_cmp_func me_cmp[11];
136
    me_cmp_func me_sub_cmp[11];
137
    me_cmp_func mb_cmp[11];
138

    
139
    /* maybe create an array for 16/8 functions */
140
    op_pixels_func put_pixels_tab[2][4];
141
    op_pixels_func avg_pixels_tab[2][4];
142
    op_pixels_func put_no_rnd_pixels_tab[2][4];
143
    op_pixels_func avg_no_rnd_pixels_tab[2][4];
144
    qpel_mc_func put_qpel_pixels_tab[2][16];
145
    qpel_mc_func avg_qpel_pixels_tab[2][16];
146
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
147
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
148
    qpel_mc_func put_mspel_pixels_tab[8];
149

    
150
    op_pixels_abs_func pix_abs16x16;
151
    op_pixels_abs_func pix_abs16x16_x2;
152
    op_pixels_abs_func pix_abs16x16_y2;
153
    op_pixels_abs_func pix_abs16x16_xy2;
154
    op_pixels_abs_func pix_abs8x8;
155
    op_pixels_abs_func pix_abs8x8_x2;
156
    op_pixels_abs_func pix_abs8x8_y2;
157
    op_pixels_abs_func pix_abs8x8_xy2;
158
    
159
    /* huffyuv specific */
160
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
161
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
162
    
163
    /* (I)DCT */
164
    void (*fdct)(DCTELEM *block/* align 16*/);
165
    
166
    /**
167
     * block -> idct -> clip to unsigned 8 bit -> dest.
168
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
169
     * @param line_size size in pixels of a horizotal line of dest
170
     */
171
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
172
    
173
    /**
174
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
175
     * @param line_size size in pixels of a horizotal line of dest
176
     */
177
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
178
    
179
    /**
180
     * idct input permutation.
181
     * an example to avoid confusion:
182
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
183
     * - (x -> referece dct -> reference idct -> x)
184
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
185
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
186
     */
187
    uint8_t idct_permutation[64];
188
    int idct_permutation_type;
189
#define FF_NO_IDCT_PERM 1
190
#define FF_LIBMPEG2_IDCT_PERM 2
191
#define FF_SIMPLE_IDCT_PERM 3
192
#define FF_TRANSPOSE_IDCT_PERM 4
193

    
194
} DSPContext;
195

    
196
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
197

    
198
/**
199
 * permute block according to permuatation.
200
 * @param last last non zero element in scantable order
201
 */
202
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
203

    
204
/**
205
 * Empty mmx state.
206
 * this must be called between any dsp function and float/double code.
207
 * for example sin(); dsp->idct_put(); emms_c(); cos()
208
 */
209
#define emms_c()
210

    
211
/* should be defined by architectures supporting
212
   one or more MultiMedia extension */
213
int mm_support(void);
214

    
215
#if defined(HAVE_MMX)
216

    
217
#undef emms_c
218

    
219
#define MM_MMX    0x0001 /* standard MMX */
220
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
221
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
222
#define MM_SSE    0x0008 /* SSE functions */
223
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
224

    
225
extern int mm_flags;
226

    
227
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
228
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
229

    
230
static inline void emms(void)
231
{
232
    __asm __volatile ("emms;":::"memory");
233
}
234

    
235

    
236
#define emms_c() \
237
{\
238
    if (mm_flags & MM_MMX)\
239
        emms();\
240
}
241

    
242
#define __align8 __attribute__ ((aligned (8)))
243

    
244
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
245
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
246

    
247
#elif defined(ARCH_ARMV4L)
248

    
249
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
250
   line ptimizations */
251
#define __align8 __attribute__ ((aligned (4)))
252

    
253
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
254

    
255
#elif defined(HAVE_MLIB)
256

    
257
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
258
#define __align8 __attribute__ ((aligned (8)))
259

    
260
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
261

    
262
#elif defined(ARCH_ALPHA)
263

    
264
#define __align8 __attribute__ ((aligned (8)))
265

    
266
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
267

    
268
#elif defined(ARCH_POWERPC)
269

    
270
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
271

    
272
extern int mm_flags;
273

    
274
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
275
#include <altivec.h>
276
#endif
277

    
278
#define __align8 __attribute__ ((aligned (16)))
279

    
280
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
281

    
282
#elif defined(HAVE_MMI)
283

    
284
#define __align8 __attribute__ ((aligned (16)))
285

    
286
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
287

    
288
#else
289

    
290
#define __align8
291

    
292
#endif
293

    
294
#ifdef __GNUC__
295

    
296
struct unaligned_64 { uint64_t l; } __attribute__((packed));
297
struct unaligned_32 { uint32_t l; } __attribute__((packed));
298

    
299
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
300
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
301

    
302
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
303

    
304
#else /* __GNUC__ */
305

    
306
#define LD32(a) (*((uint32_t*)(a)))
307
#define LD64(a) (*((uint64_t*)(a)))
308

    
309
#define ST32(a, b) *((uint32_t*)(a)) = (b)
310

    
311
#endif /* !__GNUC__ */
312

    
313
/* PSNR */
314
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
315
              int orig_linesize[3], int coded_linesize,
316
              AVCodecContext *avctx);
317

    
318
/* FFT computation */
319

    
320
/* NOTE: soon integer code will be added, so you must use the
321
   FFTSample type */
322
typedef float FFTSample;
323

    
324
typedef struct FFTComplex {
325
    FFTSample re, im;
326
} FFTComplex;
327

    
328
typedef struct FFTContext {
329
    int nbits;
330
    int inverse;
331
    uint16_t *revtab;
332
    FFTComplex *exptab;
333
    FFTComplex *exptab1; /* only used by SSE code */
334
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
335
} FFTContext;
336

    
337
int fft_init(FFTContext *s, int nbits, int inverse);
338
void fft_permute(FFTContext *s, FFTComplex *z);
339
void fft_calc_c(FFTContext *s, FFTComplex *z);
340
void fft_calc_sse(FFTContext *s, FFTComplex *z);
341
void fft_calc_altivec(FFTContext *s, FFTComplex *z);
342

    
343
static inline void fft_calc(FFTContext *s, FFTComplex *z)
344
{
345
    s->fft_calc(s, z);
346
}
347
void fft_end(FFTContext *s);
348

    
349
/* MDCT computation */
350

    
351
typedef struct MDCTContext {
352
    int n;  /* size of MDCT (i.e. number of input data * 2) */
353
    int nbits; /* n = 2^nbits */
354
    /* pre/post rotation tables */
355
    FFTSample *tcos;
356
    FFTSample *tsin;
357
    FFTContext fft;
358
} MDCTContext;
359

    
360
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
361
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
362
                const FFTSample *input, FFTSample *tmp);
363
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
364
               const FFTSample *input, FFTSample *tmp);
365
void ff_mdct_end(MDCTContext *s);
366

    
367
#define WARPER88_1616(name8, name16)\
368
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
369
    return name8(s, dst           , src           , stride)\
370
          +name8(s, dst+8         , src+8         , stride)\
371
          +name8(s, dst  +8*stride, src  +8*stride, stride)\
372
          +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
373
}
374

    
375
#ifndef HAVE_LRINTF
376
/* XXX: add ISOC specific test to avoid specific BSD testing. */
377
/* better than nothing implementation. */
378
/* btw, rintf() is existing on fbsd too -- alex */
379
static inline long int lrintf(float x)
380
{
381
#ifdef CONFIG_WIN32
382
    /* XXX: incorrect, but make it compile */
383
    return (int)(x);
384
#else
385
    return (int)(rint(x));
386
#endif
387
}
388
#endif
389

    
390
#endif