Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ cbcd6c8c

History | View | Annotate | Download (28.4 KB)

1
/*
2
 * DSP utils
3
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22

    
23
/**
24
 * @file dsputil.h
25
 * DSP utils.
26
 * note, many functions in here may use MMX which trashes the FPU state, it is
27
 * absolutely necessary to call emms_c() between dsp & float/double code
28
 */
29

    
30
#ifndef FFMPEG_DSPUTIL_H
31
#define FFMPEG_DSPUTIL_H
32

    
33
#include "avcodec.h"
34

    
35

    
36
//#define DEBUG
37
/* dct code */
38
typedef short DCTELEM;
39
typedef int DWTELEM;
40
typedef short IDWTELEM;
41

    
42
void fdct_ifast (DCTELEM *data);
43
void fdct_ifast248 (DCTELEM *data);
44
void ff_jpeg_fdct_islow (DCTELEM *data);
45
void ff_fdct248_islow (DCTELEM *data);
46

    
47
void j_rev_dct (DCTELEM *data);
48
void j_rev_dct4 (DCTELEM *data);
49
void j_rev_dct2 (DCTELEM *data);
50
void j_rev_dct1 (DCTELEM *data);
51
void ff_wmv2_idct_c(DCTELEM *data);
52

    
53
void ff_fdct_mmx(DCTELEM *block);
54
void ff_fdct_mmx2(DCTELEM *block);
55
void ff_fdct_sse2(DCTELEM *block);
56

    
57
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
58
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
59
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
60
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
61
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
62
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
63

    
64
void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
65
                              const float *src2, int src3, int blocksize, int step);
66
void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
67

    
68
/* encoding scans */
69
extern const uint8_t ff_alternate_horizontal_scan[64];
70
extern const uint8_t ff_alternate_vertical_scan[64];
71
extern const uint8_t ff_zigzag_direct[64];
72
extern const uint8_t ff_zigzag248_direct[64];
73

    
74
/* pixel operations */
75
#define MAX_NEG_CROP 1024
76

    
77
/* temporary */
78
extern uint32_t ff_squareTbl[512];
79
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
80

    
81
/* VP3 DSP functions */
82
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
83
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
84
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
85

    
86
/* 1/2^n downscaling functions from imgconvert.c */
87
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
88
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
89
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
90
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
91

    
92
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
93
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
94

    
95
/* minimum alignment rules ;)
96
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
97
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
98

99
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
100
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
101

102
!future video codecs might need functions with less strict alignment
103
*/
104

    
105
/*
106
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
107
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
108
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
109
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
110
void clear_blocks_c(DCTELEM *blocks);
111
*/
112

    
113
/* add and put pixel (decoding) */
114
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
115
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
116
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
117
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);
118
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
119
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
120
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
121
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
122

    
123
#define DEF_OLD_QPEL(name)\
124
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
125
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
126
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
127

    
128
DEF_OLD_QPEL(qpel16_mc11_old_c)
129
DEF_OLD_QPEL(qpel16_mc31_old_c)
130
DEF_OLD_QPEL(qpel16_mc12_old_c)
131
DEF_OLD_QPEL(qpel16_mc32_old_c)
132
DEF_OLD_QPEL(qpel16_mc13_old_c)
133
DEF_OLD_QPEL(qpel16_mc33_old_c)
134
DEF_OLD_QPEL(qpel8_mc11_old_c)
135
DEF_OLD_QPEL(qpel8_mc31_old_c)
136
DEF_OLD_QPEL(qpel8_mc12_old_c)
137
DEF_OLD_QPEL(qpel8_mc32_old_c)
138
DEF_OLD_QPEL(qpel8_mc13_old_c)
139
DEF_OLD_QPEL(qpel8_mc33_old_c)
140

    
141
#define CALL_2X_PIXELS(a, b, n)\
142
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
143
    b(block  , pixels  , line_size, h);\
144
    b(block+n, pixels+n, line_size, h);\
145
}
146

    
147
/* motion estimation */
148
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
149
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
150
typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
151

    
152

    
153
// for snow slices
154
typedef struct slice_buffer_s slice_buffer;
155

    
156
/**
157
 * Scantable.
158
 */
159
typedef struct ScanTable{
160
    const uint8_t *scantable;
161
    uint8_t permutated[64];
162
    uint8_t raster_end[64];
163
#ifdef ARCH_POWERPC
164
                /** Used by dct_quantize_altivec to find last-non-zero */
165
    DECLARE_ALIGNED(16, uint8_t, inverse[64]);
166
#endif
167
} ScanTable;
168

    
169
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
170

    
171
void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
172
                         int block_w, int block_h,
173
                         int src_x, int src_y, int w, int h);
174

    
175
/**
176
 * DSPContext.
177
 */
178
typedef struct DSPContext {
179
    /* pixel ops : interface with DCT */
180
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
181
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
182
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
183
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
184
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
185
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
186
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
187
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
188
    /**
189
     * translational global motion compensation.
190
     */
191
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
192
    /**
193
     * global motion compensation.
194
     */
195
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
196
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
197
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
198
    int (*pix_sum)(uint8_t * pix, int line_size);
199
    int (*pix_norm1)(uint8_t * pix, int line_size);
200
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
201

    
202
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
203
    me_cmp_func sse[5];
204
    me_cmp_func hadamard8_diff[5];
205
    me_cmp_func dct_sad[5];
206
    me_cmp_func quant_psnr[5];
207
    me_cmp_func bit[5];
208
    me_cmp_func rd[5];
209
    me_cmp_func vsad[5];
210
    me_cmp_func vsse[5];
211
    me_cmp_func nsse[5];
212
    me_cmp_func w53[5];
213
    me_cmp_func w97[5];
214
    me_cmp_func dct_max[5];
215
    me_cmp_func dct264_sad[5];
216

    
217
    me_cmp_func me_pre_cmp[5];
218
    me_cmp_func me_cmp[5];
219
    me_cmp_func me_sub_cmp[5];
220
    me_cmp_func mb_cmp[5];
221
    me_cmp_func ildct_cmp[5]; //only width 16 used
222
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
223

    
224
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
225
                             int size);
226

    
227
    /**
228
     * Halfpel motion compensation with rounding (a+b+1)>>1.
229
     * this is an array[4][4] of motion compensation functions for 4
230
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
231
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
232
     * @param block destination where the result is stored
233
     * @param pixels source
234
     * @param line_size number of bytes in a horizontal line of block
235
     * @param h height
236
     */
237
    op_pixels_func put_pixels_tab[4][4];
238

    
239
    /**
240
     * Halfpel motion compensation with rounding (a+b+1)>>1.
241
     * This is an array[4][4] of motion compensation functions for 4
242
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
243
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
244
     * @param block destination into which the result is averaged (a+b+1)>>1
245
     * @param pixels source
246
     * @param line_size number of bytes in a horizontal line of block
247
     * @param h height
248
     */
249
    op_pixels_func avg_pixels_tab[4][4];
250

    
251
    /**
252
     * Halfpel motion compensation with no rounding (a+b)>>1.
253
     * this is an array[2][4] of motion compensation functions for 2
254
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
255
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
256
     * @param block destination where the result is stored
257
     * @param pixels source
258
     * @param line_size number of bytes in a horizontal line of block
259
     * @param h height
260
     */
261
    op_pixels_func put_no_rnd_pixels_tab[4][4];
262

    
263
    /**
264
     * Halfpel motion compensation with no rounding (a+b)>>1.
265
     * this is an array[2][4] of motion compensation functions for 2
266
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
267
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
268
     * @param block destination into which the result is averaged (a+b)>>1
269
     * @param pixels source
270
     * @param line_size number of bytes in a horizontal line of block
271
     * @param h height
272
     */
273
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
274

    
275
    void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h);
276

    
277
    /**
278
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
279
     * this is an array[12] of motion compensation functions for the 9 thirdpe
280
     * positions<br>
281
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
282
     * @param block destination where the result is stored
283
     * @param pixels source
284
     * @param line_size number of bytes in a horizontal line of block
285
     * @param h height
286
     */
287
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
288
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
289

    
290
    qpel_mc_func put_qpel_pixels_tab[2][16];
291
    qpel_mc_func avg_qpel_pixels_tab[2][16];
292
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
293
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
294
    qpel_mc_func put_mspel_pixels_tab[8];
295

    
296
    /**
297
     * h264 Chroma MC
298
     */
299
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
300
    /* This is really one func used in VC-1 decoding */
301
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
302
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
303

    
304
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
305
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
306

    
307
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
308
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
309

    
310
    h264_weight_func weight_h264_pixels_tab[10];
311
    h264_biweight_func biweight_h264_pixels_tab[10];
312

    
313
    /* AVS specific */
314
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
315
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
316
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
317
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
318
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
319
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
320
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
321

    
322
    me_cmp_func pix_abs[2][4];
323

    
324
    /* huffyuv specific */
325
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
326
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
327
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
328
    /**
329
     * subtract huffyuv's variant of median prediction
330
     * note, this might read from src1[-1], src2[-1]
331
     */
332
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
333
    /* this might write to dst[w] */
334
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
335
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
336

    
337
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
338
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
339
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
340
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
341
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
342
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
343
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
344
    void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
345
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);
346

    
347
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
348
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
349

    
350
    void (*h261_loop_filter)(uint8_t *src, int stride);
351

    
352
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
353
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
354

    
355
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
356
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
357
    /* no alignment needed */
358
    void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
359
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
360
    void (*vector_fmul)(float *dst, const float *src, int len);
361
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
362
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
363
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
364

    
365
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
366
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
367
    void (*float_to_int16)(int16_t *dst, const float *src, int len);
368

    
369
    /* (I)DCT */
370
    void (*fdct)(DCTELEM *block/* align 16*/);
371
    void (*fdct248)(DCTELEM *block/* align 16*/);
372

    
373
    /* IDCT really*/
374
    void (*idct)(DCTELEM *block/* align 16*/);
375

    
376
    /**
377
     * block -> idct -> clip to unsigned 8 bit -> dest.
378
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
379
     * @param line_size size in bytes of a horizontal line of dest
380
     */
381
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
382

    
383
    /**
384
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
385
     * @param line_size size in bytes of a horizontal line of dest
386
     */
387
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
388

    
389
    /**
390
     * idct input permutation.
391
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
392
     * IDCT)
393
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
394
     * with the zigzag/alternate scan<br>
395
     * an example to avoid confusion:
396
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
397
     * - (x -> referece dct -> reference idct -> x)
398
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
399
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
400
     */
401
    uint8_t idct_permutation[64];
402
    int idct_permutation_type;
403
#define FF_NO_IDCT_PERM 1
404
#define FF_LIBMPEG2_IDCT_PERM 2
405
#define FF_SIMPLE_IDCT_PERM 3
406
#define FF_TRANSPOSE_IDCT_PERM 4
407
#define FF_PARTTRANS_IDCT_PERM 5
408

    
409
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
410
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
411
#define BASIS_SHIFT 16
412
#define RECON_SHIFT 6
413

    
414
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
415
#define EDGE_WIDTH 16
416

    
417
    /* h264 functions */
418
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
419
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
420
    void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
421
    void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
422
    void (*h264_dct)(DCTELEM block[4][4]);
423

    
424
    /* snow wavelet */
425
    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
426
    void (*horizontal_compose97i)(IDWTELEM *b, int width);
427
    void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
428

    
429
    void (*prefetch)(void *mem, int stride, int h);
430

    
431
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
432

    
433
    /* vc1 functions */
434
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
435
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
436
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
437
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
438
    void (*vc1_v_overlap)(uint8_t* src, int stride);
439
    void (*vc1_h_overlap)(uint8_t* src, int stride);
440
    /* put 8x8 block with bicubic interpolation and quarterpel precision
441
     * last argument is actually round value instead of height
442
     */
443
    op_pixels_func put_vc1_mspel_pixels_tab[16];
444

    
445
    /* intrax8 functions */
446
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
447
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
448
           int * range, int * sum,  int edges);
449

    
450
} DSPContext;
451

    
452
void dsputil_static_init(void);
453
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
454

    
455
int ff_check_alignment(void);
456

    
457
/**
458
 * permute block according to permuatation.
459
 * @param last last non zero element in scantable order
460
 */
461
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
462

    
463
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
464

    
465
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
466

    
467
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
468
{
469
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
470
}
471

    
472
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
473
{
474
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
475
}
476

    
477
static inline int get_penalty_factor(int lambda, int lambda2, int type){
478
    switch(type&0xFF){
479
    default:
480
    case FF_CMP_SAD:
481
        return lambda>>FF_LAMBDA_SHIFT;
482
    case FF_CMP_DCT:
483
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
484
    case FF_CMP_W53:
485
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
486
    case FF_CMP_W97:
487
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
488
    case FF_CMP_SATD:
489
    case FF_CMP_DCT264:
490
        return (2*lambda)>>FF_LAMBDA_SHIFT;
491
    case FF_CMP_RD:
492
    case FF_CMP_PSNR:
493
    case FF_CMP_SSE:
494
    case FF_CMP_NSSE:
495
        return lambda2>>FF_LAMBDA_SHIFT;
496
    case FF_CMP_BIT:
497
        return 1;
498
    }
499
}
500

    
501
/**
502
 * Empty mmx state.
503
 * this must be called between any dsp function and float/double code.
504
 * for example sin(); dsp->idct_put(); emms_c(); cos()
505
 */
506
#define emms_c()
507

    
508
/* should be defined by architectures supporting
509
   one or more MultiMedia extension */
510
int mm_support(void);
511

    
512
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
513
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
514
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
515
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
516
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
517
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
518
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
519
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
520
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
521

    
522
#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
523

    
524
#if defined(HAVE_MMX)
525

    
526
#undef emms_c
527

    
528
#define MM_MMX    0x0001 /* standard MMX */
529
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
530
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
531
#define MM_SSE    0x0008 /* SSE functions */
532
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
533
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
534
#define MM_SSE3   0x0040 /* Prescott SSE3 functions */
535
#define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
536

    
537
extern int mm_flags;
538

    
539
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
540
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
541
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
542

    
543
static inline void emms(void)
544
{
545
    asm volatile ("emms;":::"memory");
546
}
547

    
548

    
549
#define emms_c() \
550
{\
551
    if (mm_flags & MM_MMX)\
552
        emms();\
553
}
554

    
555
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
556

    
557
#elif defined(ARCH_ARMV4L)
558

    
559
#define MM_IWMMXT    0x0100 /* XScale IWMMXT */
560

    
561
extern int mm_flags;
562

    
563
#elif defined(ARCH_POWERPC)
564

    
565
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
566

    
567
extern int mm_flags;
568

    
569
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
570
#define STRIDE_ALIGN 16
571

    
572
#elif defined(HAVE_MMI)
573

    
574
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
575
#define STRIDE_ALIGN 16
576

    
577
#endif
578

    
579
#ifndef DECLARE_ALIGNED_8
580
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
581
#endif
582

    
583
#ifndef STRIDE_ALIGN
584
#   define STRIDE_ALIGN 8
585
#endif
586

    
587
/* PSNR */
588
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
589
              int orig_linesize[3], int coded_linesize,
590
              AVCodecContext *avctx);
591

    
592
/* FFT computation */
593

    
594
/* NOTE: soon integer code will be added, so you must use the
595
   FFTSample type */
596
typedef float FFTSample;
597

    
598
struct MDCTContext;
599

    
600
typedef struct FFTComplex {
601
    FFTSample re, im;
602
} FFTComplex;
603

    
604
typedef struct FFTContext {
605
    int nbits;
606
    int inverse;
607
    uint16_t *revtab;
608
    FFTComplex *exptab;
609
    FFTComplex *exptab1; /* only used by SSE code */
610
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
611
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
612
                       const FFTSample *input, FFTSample *tmp);
613
} FFTContext;
614

    
615
int ff_fft_init(FFTContext *s, int nbits, int inverse);
616
void ff_fft_permute(FFTContext *s, FFTComplex *z);
617
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
618
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
619
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
620
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
621
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
622

    
623
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
624
{
625
    s->fft_calc(s, z);
626
}
627
void ff_fft_end(FFTContext *s);
628

    
629
/* MDCT computation */
630

    
631
typedef struct MDCTContext {
632
    int n;  /* size of MDCT (i.e. number of input data * 2) */
633
    int nbits; /* n = 2^nbits */
634
    /* pre/post rotation tables */
635
    FFTSample *tcos;
636
    FFTSample *tsin;
637
    FFTContext fft;
638
} MDCTContext;
639

    
640
/**
641
 * Generate a Kaiser-Bessel Derived Window.
642
 * @param   window  pointer to half window
643
 * @param   alpha   determines window shape
644
 * @param   n       size of half window
645
 */
646
void ff_kbd_window_init(float *window, float alpha, int n);
647

    
648
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
649
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
650
                const FFTSample *input, FFTSample *tmp);
651
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
652
                        const FFTSample *input, FFTSample *tmp);
653
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
654
                       const FFTSample *input, FFTSample *tmp);
655
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
656
               const FFTSample *input, FFTSample *tmp);
657
void ff_mdct_end(MDCTContext *s);
658

    
659
#define WRAPPER8_16(name8, name16)\
660
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
661
    return name8(s, dst           , src           , stride, h)\
662
          +name8(s, dst+8         , src+8         , stride, h);\
663
}
664

    
665
#define WRAPPER8_16_SQ(name8, name16)\
666
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
667
    int score=0;\
668
    score +=name8(s, dst           , src           , stride, 8);\
669
    score +=name8(s, dst+8         , src+8         , stride, 8);\
670
    if(h==16){\
671
        dst += 8*stride;\
672
        src += 8*stride;\
673
        score +=name8(s, dst           , src           , stride, 8);\
674
        score +=name8(s, dst+8         , src+8         , stride, 8);\
675
    }\
676
    return score;\
677
}
678

    
679

    
680
static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
681
{
682
    int i;
683
    for(i=0; i<h; i++)
684
    {
685
        AV_WN16(dst   , AV_RN16(src   ));
686
        dst+=dstStride;
687
        src+=srcStride;
688
    }
689
}
690

    
691
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
692
{
693
    int i;
694
    for(i=0; i<h; i++)
695
    {
696
        AV_WN32(dst   , AV_RN32(src   ));
697
        dst+=dstStride;
698
        src+=srcStride;
699
    }
700
}
701

    
702
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
703
{
704
    int i;
705
    for(i=0; i<h; i++)
706
    {
707
        AV_WN32(dst   , AV_RN32(src   ));
708
        AV_WN32(dst+4 , AV_RN32(src+4 ));
709
        dst+=dstStride;
710
        src+=srcStride;
711
    }
712
}
713

    
714
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
715
{
716
    int i;
717
    for(i=0; i<h; i++)
718
    {
719
        AV_WN32(dst   , AV_RN32(src   ));
720
        AV_WN32(dst+4 , AV_RN32(src+4 ));
721
        dst[8]= src[8];
722
        dst+=dstStride;
723
        src+=srcStride;
724
    }
725
}
726

    
727
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
728
{
729
    int i;
730
    for(i=0; i<h; i++)
731
    {
732
        AV_WN32(dst   , AV_RN32(src   ));
733
        AV_WN32(dst+4 , AV_RN32(src+4 ));
734
        AV_WN32(dst+8 , AV_RN32(src+8 ));
735
        AV_WN32(dst+12, AV_RN32(src+12));
736
        dst+=dstStride;
737
        src+=srcStride;
738
    }
739
}
740

    
741
static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
742
{
743
    int i;
744
    for(i=0; i<h; i++)
745
    {
746
        AV_WN32(dst   , AV_RN32(src   ));
747
        AV_WN32(dst+4 , AV_RN32(src+4 ));
748
        AV_WN32(dst+8 , AV_RN32(src+8 ));
749
        AV_WN32(dst+12, AV_RN32(src+12));
750
        dst[16]= src[16];
751
        dst+=dstStride;
752
        src+=srcStride;
753
    }
754
}
755

    
756
#endif /* FFMPEG_DSPUTIL_H */