Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 3899eb2f

History | View | Annotate | Download (34.7 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 libavcodec/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 AVCODEC_DSPUTIL_H
31
#define AVCODEC_DSPUTIL_H
32

    
33
#include "libavutil/intreadwrite.h"
34
#include "avcodec.h"
35

    
36

    
37
//#define DEBUG
38
/* dct code */
39
typedef short DCTELEM;
40
typedef int DWTELEM;
41
typedef short IDWTELEM;
42

    
43
void fdct_ifast (DCTELEM *data);
44
void fdct_ifast248 (DCTELEM *data);
45
void ff_jpeg_fdct_islow (DCTELEM *data);
46
void ff_fdct248_islow (DCTELEM *data);
47

    
48
void j_rev_dct (DCTELEM *data);
49
void j_rev_dct4 (DCTELEM *data);
50
void j_rev_dct2 (DCTELEM *data);
51
void j_rev_dct1 (DCTELEM *data);
52
void ff_wmv2_idct_c(DCTELEM *data);
53

    
54
void ff_fdct_mmx(DCTELEM *block);
55
void ff_fdct_mmx2(DCTELEM *block);
56
void ff_fdct_sse2(DCTELEM *block);
57

    
58
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
59
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
60
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
61
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
62
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
63
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
64
void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
65
void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
66
void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
67
void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
68

    
69
void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
70
                              const float *src2, int src3, int blocksize, int step);
71
void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
72
                             const float *win, float add_bias, int len);
73
void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
74
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
75

    
76
/* encoding scans */
77
extern const uint8_t ff_alternate_horizontal_scan[64];
78
extern const uint8_t ff_alternate_vertical_scan[64];
79
extern const uint8_t ff_zigzag_direct[64];
80
extern const uint8_t ff_zigzag248_direct[64];
81

    
82
/* pixel operations */
83
#define MAX_NEG_CROP 1024
84

    
85
/* temporary */
86
extern uint32_t ff_squareTbl[512];
87
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
88

    
89
/* VP3 DSP functions */
90
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
91
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
92
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
93

    
94
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
95
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
96

    
97
/* VP6 DSP functions */
98
void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride,
99
                           const int16_t *h_weights, const int16_t *v_weights);
100

    
101
/* 1/2^n downscaling functions from imgconvert.c */
102
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
103
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
104
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
105
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
106

    
107
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
108
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
109

    
110
/* minimum alignment rules ;)
111
If you notice errors in the align stuff, need more alignment for some ASM code
112
for some CPU or need to use a function with less aligned data then send a mail
113
to the ffmpeg-devel mailing list, ...
114

115
!warning These alignments might not match reality, (missing attribute((align))
116
stuff somewhere possible).
117
I (Michael) did not check them, these are just the alignments which I think
118
could be reached easily ...
119

120
!future video codecs might need functions with less strict alignment
121
*/
122

    
123
/*
124
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
125
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
126
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
127
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
128
void clear_blocks_c(DCTELEM *blocks);
129
*/
130

    
131
/* add and put pixel (decoding) */
132
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
133
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
134
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
135
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);
136
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
137
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
138
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
139
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
140

    
141
#define DEF_OLD_QPEL(name)\
142
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
143
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
144
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
145

    
146
DEF_OLD_QPEL(qpel16_mc11_old_c)
147
DEF_OLD_QPEL(qpel16_mc31_old_c)
148
DEF_OLD_QPEL(qpel16_mc12_old_c)
149
DEF_OLD_QPEL(qpel16_mc32_old_c)
150
DEF_OLD_QPEL(qpel16_mc13_old_c)
151
DEF_OLD_QPEL(qpel16_mc33_old_c)
152
DEF_OLD_QPEL(qpel8_mc11_old_c)
153
DEF_OLD_QPEL(qpel8_mc31_old_c)
154
DEF_OLD_QPEL(qpel8_mc12_old_c)
155
DEF_OLD_QPEL(qpel8_mc32_old_c)
156
DEF_OLD_QPEL(qpel8_mc13_old_c)
157
DEF_OLD_QPEL(qpel8_mc33_old_c)
158

    
159
#define CALL_2X_PIXELS(a, b, n)\
160
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
161
    b(block  , pixels  , line_size, h);\
162
    b(block+n, pixels+n, line_size, h);\
163
}
164

    
165
/* motion estimation */
166
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
167
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
168
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))*/;
169

    
170

    
171
// for snow slices
172
typedef struct slice_buffer_s slice_buffer;
173

    
174
/**
175
 * Scantable.
176
 */
177
typedef struct ScanTable{
178
    const uint8_t *scantable;
179
    uint8_t permutated[64];
180
    uint8_t raster_end[64];
181
#if ARCH_PPC
182
                /** Used by dct_quantize_altivec to find last-non-zero */
183
    DECLARE_ALIGNED(16, uint8_t, inverse[64]);
184
#endif
185
} ScanTable;
186

    
187
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
188

    
189
void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
190
                         int block_w, int block_h,
191
                         int src_x, int src_y, int w, int h);
192

    
193
/**
194
 * DSPContext.
195
 */
196
typedef struct DSPContext {
197
    /* pixel ops : interface with DCT */
198
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
199
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
200
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
201
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
202
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
203
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
204
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
205
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
206
    /**
207
     * translational global motion compensation.
208
     */
209
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
210
    /**
211
     * global motion compensation.
212
     */
213
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
214
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
215
    void (*clear_block)(DCTELEM *block/*align 16*/);
216
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
217
    int (*pix_sum)(uint8_t * pix, int line_size);
218
    int (*pix_norm1)(uint8_t * pix, int line_size);
219
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
220

    
221
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
222
    me_cmp_func sse[6];
223
    me_cmp_func hadamard8_diff[6];
224
    me_cmp_func dct_sad[6];
225
    me_cmp_func quant_psnr[6];
226
    me_cmp_func bit[6];
227
    me_cmp_func rd[6];
228
    me_cmp_func vsad[6];
229
    me_cmp_func vsse[6];
230
    me_cmp_func nsse[6];
231
    me_cmp_func w53[6];
232
    me_cmp_func w97[6];
233
    me_cmp_func dct_max[6];
234
    me_cmp_func dct264_sad[6];
235

    
236
    me_cmp_func me_pre_cmp[6];
237
    me_cmp_func me_cmp[6];
238
    me_cmp_func me_sub_cmp[6];
239
    me_cmp_func mb_cmp[6];
240
    me_cmp_func ildct_cmp[6]; //only width 16 used
241
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
242

    
243
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
244
                             int size);
245

    
246
    /**
247
     * Halfpel motion compensation with rounding (a+b+1)>>1.
248
     * this is an array[4][4] of motion compensation functions for 4
249
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
250
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
251
     * @param block destination where the result is stored
252
     * @param pixels source
253
     * @param line_size number of bytes in a horizontal line of block
254
     * @param h height
255
     */
256
    op_pixels_func put_pixels_tab[4][4];
257

    
258
    /**
259
     * Halfpel motion compensation with rounding (a+b+1)>>1.
260
     * This is an array[4][4] of motion compensation functions for 4
261
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
262
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
263
     * @param block destination into which the result is averaged (a+b+1)>>1
264
     * @param pixels source
265
     * @param line_size number of bytes in a horizontal line of block
266
     * @param h height
267
     */
268
    op_pixels_func avg_pixels_tab[4][4];
269

    
270
    /**
271
     * Halfpel motion compensation with no rounding (a+b)>>1.
272
     * this is an array[2][4] of motion compensation functions for 2
273
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
274
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
275
     * @param block destination where the result is stored
276
     * @param pixels source
277
     * @param line_size number of bytes in a horizontal line of block
278
     * @param h height
279
     */
280
    op_pixels_func put_no_rnd_pixels_tab[4][4];
281

    
282
    /**
283
     * Halfpel motion compensation with no rounding (a+b)>>1.
284
     * this is an array[2][4] of motion compensation functions for 2
285
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
286
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
287
     * @param block destination into which the result is averaged (a+b)>>1
288
     * @param pixels source
289
     * @param line_size number of bytes in a horizontal line of block
290
     * @param h height
291
     */
292
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
293

    
294
    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);
295

    
296
    /**
297
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
298
     * this is an array[12] of motion compensation functions for the 9 thirdpe
299
     * positions<br>
300
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
301
     * @param block destination where the result is stored
302
     * @param pixels source
303
     * @param line_size number of bytes in a horizontal line of block
304
     * @param h height
305
     */
306
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
307
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
308

    
309
    qpel_mc_func put_qpel_pixels_tab[2][16];
310
    qpel_mc_func avg_qpel_pixels_tab[2][16];
311
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
312
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
313
    qpel_mc_func put_mspel_pixels_tab[8];
314

    
315
    /**
316
     * h264 Chroma MC
317
     */
318
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
319
    /* This is really one func used in VC-1 decoding */
320
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
321
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
322

    
323
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
324
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
325

    
326
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
327
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
328

    
329
    h264_weight_func weight_h264_pixels_tab[10];
330
    h264_biweight_func biweight_h264_pixels_tab[10];
331

    
332
    /* AVS specific */
333
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
334
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
335
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
336
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
337
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
338
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
339
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
340

    
341
    me_cmp_func pix_abs[2][4];
342

    
343
    /* huffyuv specific */
344
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
345
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
346
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
347
    /**
348
     * subtract huffyuv's variant of median prediction
349
     * note, this might read from src1[-1], src2[-1]
350
     */
351
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
352
    void (*add_hfyu_median_prediction)(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top);
353
    /* this might write to dst[w] */
354
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
355
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
356

    
357
    void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
358
    void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
359
    /* v/h_loop_filter_luma_intra: align 16 */
360
    void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
361
    void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
362
    void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
363
    void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
364
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
365
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
366
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
367
    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],
368
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
369

    
370
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
371
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
372

    
373
    void (*h261_loop_filter)(uint8_t *src, int stride);
374

    
375
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
376
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
377

    
378
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
379
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
380

    
381
    void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride,
382
                             const int16_t *h_weights,const int16_t *v_weights);
383

    
384
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
385
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
386
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
387
    /* no alignment needed */
388
    void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
389
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
390
    void (*vector_fmul)(float *dst, const float *src, int len);
391
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
392
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
393
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
394
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
395
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
396
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
397
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
398

    
399
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
400
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
401
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
402
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
403

    
404
    /* (I)DCT */
405
    void (*fdct)(DCTELEM *block/* align 16*/);
406
    void (*fdct248)(DCTELEM *block/* align 16*/);
407

    
408
    /* IDCT really*/
409
    void (*idct)(DCTELEM *block/* align 16*/);
410

    
411
    /**
412
     * block -> idct -> clip to unsigned 8 bit -> dest.
413
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
414
     * @param line_size size in bytes of a horizontal line of dest
415
     */
416
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
417

    
418
    /**
419
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
420
     * @param line_size size in bytes of a horizontal line of dest
421
     */
422
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
423

    
424
    /**
425
     * idct input permutation.
426
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
427
     * IDCT)
428
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
429
     * with the zigzag/alternate scan<br>
430
     * an example to avoid confusion:
431
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
432
     * - (x -> referece dct -> reference idct -> x)
433
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
434
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
435
     */
436
    uint8_t idct_permutation[64];
437
    int idct_permutation_type;
438
#define FF_NO_IDCT_PERM 1
439
#define FF_LIBMPEG2_IDCT_PERM 2
440
#define FF_SIMPLE_IDCT_PERM 3
441
#define FF_TRANSPOSE_IDCT_PERM 4
442
#define FF_PARTTRANS_IDCT_PERM 5
443
#define FF_SSE2_IDCT_PERM 6
444

    
445
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
446
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
447
#define BASIS_SHIFT 16
448
#define RECON_SHIFT 6
449

    
450
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
451
#define EDGE_WIDTH 16
452

    
453
    /* h264 functions */
454
    /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
455
       NOTE!!! if you implement any of h264_idct_add, h264_idct_add16, h264_idct_add16intra, h264_idct_add8 then you must implement all of them
456
        The reason for above, is that no 2 out of one list may use a different permutation.
457
    */
458
    void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
459
    void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
460
    void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
461
    void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
462
    void (*h264_dct)(DCTELEM block[4][4]);
463
    void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
464
    void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
465
    void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
466
    void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
467

    
468
    /* snow wavelet */
469
    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
470
    void (*horizontal_compose97i)(IDWTELEM *b, int width);
471
    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);
472

    
473
    void (*prefetch)(void *mem, int stride, int h);
474

    
475
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
476

    
477
    /* vc1 functions */
478
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
479
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
480
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
481
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
482
    void (*vc1_v_overlap)(uint8_t* src, int stride);
483
    void (*vc1_h_overlap)(uint8_t* src, int stride);
484
    /* put 8x8 block with bicubic interpolation and quarterpel precision
485
     * last argument is actually round value instead of height
486
     */
487
    op_pixels_func put_vc1_mspel_pixels_tab[16];
488

    
489
    /* intrax8 functions */
490
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
491
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
492
           int * range, int * sum,  int edges);
493

    
494
    /* ape functions */
495
    /**
496
     * Add contents of the second vector to the first one.
497
     * @param len length of vectors, should be multiple of 16
498
     */
499
    void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
500
    /**
501
     * Add contents of the second vector to the first one.
502
     * @param len length of vectors, should be multiple of 16
503
     */
504
    void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
505
    /**
506
     * Calculate scalar product of two vectors.
507
     * @param len length of vectors, should be multiple of 16
508
     * @param shift number of bits to discard from product
509
     */
510
    int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
511

    
512
    /* rv30 functions */
513
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
514
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
515

    
516
    /* rv40 functions */
517
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
518
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
519
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
520
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
521
} DSPContext;
522

    
523
void dsputil_static_init(void);
524
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
525

    
526
int ff_check_alignment(void);
527

    
528
/**
529
 * permute block according to permuatation.
530
 * @param last last non zero element in scantable order
531
 */
532
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
533

    
534
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
535

    
536
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
537

    
538
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
539
{
540
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
541
}
542

    
543
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
544
{
545
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
546
}
547

    
548
static inline int get_penalty_factor(int lambda, int lambda2, int type){
549
    switch(type&0xFF){
550
    default:
551
    case FF_CMP_SAD:
552
        return lambda>>FF_LAMBDA_SHIFT;
553
    case FF_CMP_DCT:
554
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
555
    case FF_CMP_W53:
556
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
557
    case FF_CMP_W97:
558
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
559
    case FF_CMP_SATD:
560
    case FF_CMP_DCT264:
561
        return (2*lambda)>>FF_LAMBDA_SHIFT;
562
    case FF_CMP_RD:
563
    case FF_CMP_PSNR:
564
    case FF_CMP_SSE:
565
    case FF_CMP_NSSE:
566
        return lambda2>>FF_LAMBDA_SHIFT;
567
    case FF_CMP_BIT:
568
        return 1;
569
    }
570
}
571

    
572
/**
573
 * Empty mmx state.
574
 * this must be called between any dsp function and float/double code.
575
 * for example sin(); dsp->idct_put(); emms_c(); cos()
576
 */
577
#define emms_c()
578

    
579
/* should be defined by architectures supporting
580
   one or more MultiMedia extension */
581
int mm_support(void);
582

    
583
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
584
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
585
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
586
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
587
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
588
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
589
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
590
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
591
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
592

    
593
#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
594

    
595
#if HAVE_MMX
596

    
597
#undef emms_c
598

    
599
extern int mm_flags;
600

    
601
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
602
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
603
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
604

    
605
static inline void emms(void)
606
{
607
    __asm__ volatile ("emms;":::"memory");
608
}
609

    
610

    
611
#define emms_c() \
612
{\
613
    if (mm_flags & FF_MM_MMX)\
614
        emms();\
615
}
616

    
617
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
618

    
619
#elif ARCH_ARM
620

    
621
extern int mm_flags;
622

    
623
#if HAVE_NEON
624
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
625
#   define STRIDE_ALIGN 16
626
#endif
627

    
628
#elif ARCH_PPC
629

    
630
extern int mm_flags;
631

    
632
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
633
#define STRIDE_ALIGN 16
634

    
635
#elif HAVE_MMI
636

    
637
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
638
#define STRIDE_ALIGN 16
639

    
640
#else
641

    
642
#define mm_flags 0
643
#define mm_support() 0
644

    
645
#endif
646

    
647
#ifndef DECLARE_ALIGNED_8
648
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
649
#endif
650

    
651
#ifndef STRIDE_ALIGN
652
#   define STRIDE_ALIGN 8
653
#endif
654

    
655
/* PSNR */
656
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
657
              int orig_linesize[3], int coded_linesize,
658
              AVCodecContext *avctx);
659

    
660
/* FFT computation */
661

    
662
/* NOTE: soon integer code will be added, so you must use the
663
   FFTSample type */
664
typedef float FFTSample;
665

    
666
struct MDCTContext;
667

    
668
typedef struct FFTComplex {
669
    FFTSample re, im;
670
} FFTComplex;
671

    
672
typedef struct FFTContext {
673
    int nbits;
674
    int inverse;
675
    uint16_t *revtab;
676
    FFTComplex *exptab;
677
    FFTComplex *exptab1; /* only used by SSE code */
678
    FFTComplex *tmp_buf;
679
    void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
680
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
681
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
682
    void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
683
} FFTContext;
684

    
685
extern FFTSample* ff_cos_tabs[13];
686

    
687
/**
688
 * Sets up a complex FFT.
689
 * @param nbits           log2 of the length of the input array
690
 * @param inverse         if 0 perform the forward transform, if 1 perform the inverse
691
 */
692
int ff_fft_init(FFTContext *s, int nbits, int inverse);
693
void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
694
void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
695
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
696
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
697
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
698
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
699
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
700

    
701
/**
702
 * Do the permutation needed BEFORE calling ff_fft_calc().
703
 */
704
static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
705
{
706
    s->fft_permute(s, z);
707
}
708
/**
709
 * Do a complex FFT with the parameters defined in ff_fft_init(). The
710
 * input data must be permuted before. No 1.0/sqrt(n) normalization is done.
711
 */
712
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
713
{
714
    s->fft_calc(s, z);
715
}
716
void ff_fft_end(FFTContext *s);
717

    
718
/* MDCT computation */
719

    
720
typedef struct MDCTContext {
721
    int n;  /* size of MDCT (i.e. number of input data * 2) */
722
    int nbits; /* n = 2^nbits */
723
    /* pre/post rotation tables */
724
    FFTSample *tcos;
725
    FFTSample *tsin;
726
    FFTContext fft;
727
} MDCTContext;
728

    
729
static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
730
{
731
    s->fft.imdct_calc(s, output, input);
732
}
733
static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
734
{
735
    s->fft.imdct_half(s, output, input);
736
}
737

    
738
/**
739
 * Generate a Kaiser-Bessel Derived Window.
740
 * @param   window  pointer to half window
741
 * @param   alpha   determines window shape
742
 * @param   n       size of half window
743
 */
744
void ff_kbd_window_init(float *window, float alpha, int n);
745

    
746
/**
747
 * Generate a sine window.
748
 * @param   window  pointer to half window
749
 * @param   n       size of half window
750
 */
751
void ff_sine_window_init(float *window, int n);
752
extern float ff_sine_128 [ 128];
753
extern float ff_sine_256 [ 256];
754
extern float ff_sine_512 [ 512];
755
extern float ff_sine_1024[1024];
756
extern float ff_sine_2048[2048];
757
extern float ff_sine_4096[4096];
758
extern float *ff_sine_windows[6];
759

    
760
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
761
void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
762
void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
763
void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
764
void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
765
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
766
void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
767
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
768
void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
769
void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
770
void ff_mdct_end(MDCTContext *s);
771

    
772
/* Real Discrete Fourier Transform */
773

    
774
enum RDFTransformType {
775
    RDFT,
776
    IRDFT,
777
    RIDFT,
778
    IRIDFT,
779
};
780

    
781
typedef struct {
782
    int nbits;
783
    int inverse;
784
    int sign_convention;
785

    
786
    /* pre/post rotation tables */
787
    FFTSample *tcos;
788
    FFTSample *tsin;
789
    FFTContext fft;
790
} RDFTContext;
791

    
792
/**
793
 * Sets up a real FFT.
794
 * @param nbits           log2 of the length of the input array
795
 * @param trans           the type of transform
796
 */
797
int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans);
798
void ff_rdft_calc(RDFTContext *s, FFTSample *data);
799
void ff_rdft_end(RDFTContext *s);
800

    
801
#define WRAPPER8_16(name8, name16)\
802
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
803
    return name8(s, dst           , src           , stride, h)\
804
          +name8(s, dst+8         , src+8         , stride, h);\
805
}
806

    
807
#define WRAPPER8_16_SQ(name8, name16)\
808
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
809
    int score=0;\
810
    score +=name8(s, dst           , src           , stride, 8);\
811
    score +=name8(s, dst+8         , src+8         , stride, 8);\
812
    if(h==16){\
813
        dst += 8*stride;\
814
        src += 8*stride;\
815
        score +=name8(s, dst           , src           , stride, 8);\
816
        score +=name8(s, dst+8         , src+8         , stride, 8);\
817
    }\
818
    return score;\
819
}
820

    
821

    
822
static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
823
{
824
    int i;
825
    for(i=0; i<h; i++)
826
    {
827
        AV_WN16(dst   , AV_RN16(src   ));
828
        dst+=dstStride;
829
        src+=srcStride;
830
    }
831
}
832

    
833
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
834
{
835
    int i;
836
    for(i=0; i<h; i++)
837
    {
838
        AV_WN32(dst   , AV_RN32(src   ));
839
        dst+=dstStride;
840
        src+=srcStride;
841
    }
842
}
843

    
844
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
845
{
846
    int i;
847
    for(i=0; i<h; i++)
848
    {
849
        AV_WN32(dst   , AV_RN32(src   ));
850
        AV_WN32(dst+4 , AV_RN32(src+4 ));
851
        dst+=dstStride;
852
        src+=srcStride;
853
    }
854
}
855

    
856
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
857
{
858
    int i;
859
    for(i=0; i<h; i++)
860
    {
861
        AV_WN32(dst   , AV_RN32(src   ));
862
        AV_WN32(dst+4 , AV_RN32(src+4 ));
863
        dst[8]= src[8];
864
        dst+=dstStride;
865
        src+=srcStride;
866
    }
867
}
868

    
869
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
870
{
871
    int i;
872
    for(i=0; i<h; i++)
873
    {
874
        AV_WN32(dst   , AV_RN32(src   ));
875
        AV_WN32(dst+4 , AV_RN32(src+4 ));
876
        AV_WN32(dst+8 , AV_RN32(src+8 ));
877
        AV_WN32(dst+12, AV_RN32(src+12));
878
        dst+=dstStride;
879
        src+=srcStride;
880
    }
881
}
882

    
883
static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
884
{
885
    int i;
886
    for(i=0; i<h; i++)
887
    {
888
        AV_WN32(dst   , AV_RN32(src   ));
889
        AV_WN32(dst+4 , AV_RN32(src+4 ));
890
        AV_WN32(dst+8 , AV_RN32(src+8 ));
891
        AV_WN32(dst+12, AV_RN32(src+12));
892
        dst[16]= src[16];
893
        dst+=dstStride;
894
        src+=srcStride;
895
    }
896
}
897

    
898
#endif /* AVCODEC_DSPUTIL_H */