Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 84dc2d8a

History | View | Annotate | Download (39.9 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_window_c(float *dst, const float *src0, const float *src1,
70
                             const float *win, float add_bias, int len);
71
void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
72
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
73

    
74
/* encoding scans */
75
extern const uint8_t ff_alternate_horizontal_scan[64];
76
extern const uint8_t ff_alternate_vertical_scan[64];
77
extern const uint8_t ff_zigzag_direct[64];
78
extern const uint8_t ff_zigzag248_direct[64];
79

    
80
/* pixel operations */
81
#define MAX_NEG_CROP 1024
82

    
83
/* temporary */
84
extern uint32_t ff_squareTbl[512];
85
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
86

    
87
/* VP3 DSP functions */
88
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
89
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
90
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
91

    
92
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
93
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
94

    
95
/* VP6 DSP functions */
96
void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride,
97
                           const int16_t *h_weights, const int16_t *v_weights);
98

    
99
/* 1/2^n downscaling functions from imgconvert.c */
100
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
101
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
102
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
103
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
104

    
105
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
106
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
107

    
108
/* minimum alignment rules ;)
109
If you notice errors in the align stuff, need more alignment for some ASM code
110
for some CPU or need to use a function with less aligned data then send a mail
111
to the ffmpeg-devel mailing list, ...
112

113
!warning These alignments might not match reality, (missing attribute((align))
114
stuff somewhere possible).
115
I (Michael) did not check them, these are just the alignments which I think
116
could be reached easily ...
117

118
!future video codecs might need functions with less strict alignment
119
*/
120

    
121
/*
122
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
123
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
124
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
125
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
126
void clear_blocks_c(DCTELEM *blocks);
127
*/
128

    
129
/* add and put pixel (decoding) */
130
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
131
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
132
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
133
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);
134
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
135
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
136
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
137
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
138

    
139
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
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 (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
203
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
204
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
205
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
206
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
207
    /**
208
     * translational global motion compensation.
209
     */
210
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
211
    /**
212
     * global motion compensation.
213
     */
214
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
215
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
216
    void (*clear_block)(DCTELEM *block/*align 16*/);
217
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
218
    int (*pix_sum)(uint8_t * pix, int line_size);
219
    int (*pix_norm1)(uint8_t * pix, int line_size);
220
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
221

    
222
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
223
    me_cmp_func sse[6];
224
    me_cmp_func hadamard8_diff[6];
225
    me_cmp_func dct_sad[6];
226
    me_cmp_func quant_psnr[6];
227
    me_cmp_func bit[6];
228
    me_cmp_func rd[6];
229
    me_cmp_func vsad[6];
230
    me_cmp_func vsse[6];
231
    me_cmp_func nsse[6];
232
    me_cmp_func w53[6];
233
    me_cmp_func w97[6];
234
    me_cmp_func dct_max[6];
235
    me_cmp_func dct264_sad[6];
236

    
237
    me_cmp_func me_pre_cmp[6];
238
    me_cmp_func me_cmp[6];
239
    me_cmp_func me_sub_cmp[6];
240
    me_cmp_func mb_cmp[6];
241
    me_cmp_func ildct_cmp[6]; //only width 16 used
242
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
243

    
244
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
245
                             int size);
246

    
247
    /**
248
     * Halfpel motion compensation with rounding (a+b+1)>>1.
249
     * this is an array[4][4] of motion compensation functions for 4
250
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
251
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
252
     * @param block destination where the result is stored
253
     * @param pixels source
254
     * @param line_size number of bytes in a horizontal line of block
255
     * @param h height
256
     */
257
    op_pixels_func put_pixels_tab[4][4];
258

    
259
    /**
260
     * Halfpel motion compensation with rounding (a+b+1)>>1.
261
     * This is an array[4][4] of motion compensation functions for 4
262
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
263
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
264
     * @param block destination into which the result is averaged (a+b+1)>>1
265
     * @param pixels source
266
     * @param line_size number of bytes in a horizontal line of block
267
     * @param h height
268
     */
269
    op_pixels_func avg_pixels_tab[4][4];
270

    
271
    /**
272
     * Halfpel motion compensation with no rounding (a+b)>>1.
273
     * this is an array[2][4] of motion compensation functions for 2
274
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
275
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
276
     * @param block destination where the result is stored
277
     * @param pixels source
278
     * @param line_size number of bytes in a horizontal line of block
279
     * @param h height
280
     */
281
    op_pixels_func put_no_rnd_pixels_tab[4][4];
282

    
283
    /**
284
     * Halfpel motion compensation with no rounding (a+b)>>1.
285
     * this is an array[2][4] of motion compensation functions for 2
286
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
287
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
288
     * @param block destination into which the result is averaged (a+b)>>1
289
     * @param pixels source
290
     * @param line_size number of bytes in a horizontal line of block
291
     * @param h height
292
     */
293
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
294

    
295
    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);
296

    
297
    /**
298
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
299
     * this is an array[12] of motion compensation functions for the 9 thirdpe
300
     * positions<br>
301
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
302
     * @param block destination where the result is stored
303
     * @param pixels source
304
     * @param line_size number of bytes in a horizontal line of block
305
     * @param h height
306
     */
307
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
308
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
309

    
310
    qpel_mc_func put_qpel_pixels_tab[2][16];
311
    qpel_mc_func avg_qpel_pixels_tab[2][16];
312
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
313
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
314
    qpel_mc_func put_mspel_pixels_tab[8];
315

    
316
    /**
317
     * h264 Chroma MC
318
     */
319
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
320
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
321
    /* This is really one func used in VC-1 decoding */
322
    h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
323
    h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
324

    
325
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
326
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
327

    
328
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
329
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
330

    
331
    h264_weight_func weight_h264_pixels_tab[10];
332
    h264_biweight_func biweight_h264_pixels_tab[10];
333

    
334
    /* AVS specific */
335
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
336
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
337
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
338
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
339
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
340
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
341
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
342

    
343
    me_cmp_func pix_abs[2][4];
344

    
345
    /* huffyuv specific */
346
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
347
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
348
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
349
    /**
350
     * subtract huffyuv's variant of median prediction
351
     * note, this might read from src1[-1], src2[-1]
352
     */
353
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
354
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
355
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
356
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
357
    /* this might write to dst[w] */
358
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
359
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
360

    
361
    void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
362
    void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
363
    /* v/h_loop_filter_luma_intra: align 16 */
364
    void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
365
    void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
366
    void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
367
    void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
368
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
369
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
370
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
371
    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],
372
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
373

    
374
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
375
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
376

    
377
    void (*h261_loop_filter)(uint8_t *src, int stride);
378

    
379
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
380
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
381

    
382
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
383
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
384

    
385
    void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride,
386
                             const int16_t *h_weights,const int16_t *v_weights);
387

    
388
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
389
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
390
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
391
    /* no alignment needed */
392
    void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
393
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
394
    void (*vector_fmul)(float *dst, const float *src, int len);
395
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
396
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
397
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
398
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
399
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
400
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
401
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
402
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
403
    /**
404
     * Multiply a vector of floats by a scalar float.  Source and
405
     * destination vectors must overlap exactly or not at all.
406
     * @param dst result vector, 16-byte aligned
407
     * @param src input vector, 16-byte aligned
408
     * @param mul scalar value
409
     * @param len length of vector, multiple of 4
410
     */
411
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
412
                               int len);
413
    /**
414
     * Multiply a vector of floats by concatenated short vectors of
415
     * floats and by a scalar float.  Source and destination vectors
416
     * must overlap exactly or not at all.
417
     * [0]: short vectors of length 2, 8-byte aligned
418
     * [1]: short vectors of length 4, 16-byte aligned
419
     * @param dst output vector, 16-byte aligned
420
     * @param src input vector, 16-byte aligned
421
     * @param sv  array of pointers to short vectors
422
     * @param mul scalar value
423
     * @param len number of elements in src and dst, multiple of 4
424
     */
425
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
426
                                     const float **sv, float mul, int len);
427
    /**
428
     * Multiply short vectors of floats by a scalar float, store
429
     * concatenated result.
430
     * [0]: short vectors of length 2, 8-byte aligned
431
     * [1]: short vectors of length 4, 16-byte aligned
432
     * @param dst output vector, 16-byte aligned
433
     * @param sv  array of pointers to short vectors
434
     * @param mul scalar value
435
     * @param len number of output elements, multiple of 4
436
     */
437
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
438
                              float mul, int len);
439
    /**
440
     * Calculate the scalar product of two vectors of floats.
441
     * @param v1  first vector, 16-byte aligned
442
     * @param v2  second vector, 16-byte aligned
443
     * @param len length of vectors, multiple of 4
444
     */
445
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
446
    /**
447
     * Calculate the sum and difference of two vectors of floats.
448
     * @param v1  first input vector, sum output, 16-byte aligned
449
     * @param v2  second input vector, difference output, 16-byte aligned
450
     * @param len length of vectors, multiple of 4
451
     */
452
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
453

    
454
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
455
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
456
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
457
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
458

    
459
    /* (I)DCT */
460
    void (*fdct)(DCTELEM *block/* align 16*/);
461
    void (*fdct248)(DCTELEM *block/* align 16*/);
462

    
463
    /* IDCT really*/
464
    void (*idct)(DCTELEM *block/* align 16*/);
465

    
466
    /**
467
     * block -> idct -> clip to unsigned 8 bit -> dest.
468
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
469
     * @param line_size size in bytes of a horizontal line of dest
470
     */
471
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
472

    
473
    /**
474
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
475
     * @param line_size size in bytes of a horizontal line of dest
476
     */
477
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
478

    
479
    /**
480
     * idct input permutation.
481
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
482
     * IDCT)
483
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
484
     * with the zigzag/alternate scan<br>
485
     * an example to avoid confusion:
486
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
487
     * - (x -> referece dct -> reference idct -> x)
488
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
489
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
490
     */
491
    uint8_t idct_permutation[64];
492
    int idct_permutation_type;
493
#define FF_NO_IDCT_PERM 1
494
#define FF_LIBMPEG2_IDCT_PERM 2
495
#define FF_SIMPLE_IDCT_PERM 3
496
#define FF_TRANSPOSE_IDCT_PERM 4
497
#define FF_PARTTRANS_IDCT_PERM 5
498
#define FF_SSE2_IDCT_PERM 6
499

    
500
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
501
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
502
#define BASIS_SHIFT 16
503
#define RECON_SHIFT 6
504

    
505
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
506
#define EDGE_WIDTH 16
507

    
508
    /* h264 functions */
509
    /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
510
       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
511
        The reason for above, is that no 2 out of one list may use a different permutation.
512
    */
513
    void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
514
    void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
515
    void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
516
    void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
517
    void (*h264_dct)(DCTELEM block[4][4]);
518
    void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
519
    void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
520
    void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
521
    void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
522

    
523
    /* snow wavelet */
524
    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
525
    void (*horizontal_compose97i)(IDWTELEM *b, int width);
526
    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);
527

    
528
    void (*prefetch)(void *mem, int stride, int h);
529

    
530
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
531

    
532
    /* mlp/truehd functions */
533
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
534
                               int firorder, int iirorder,
535
                               unsigned int filter_shift, int32_t mask, int blocksize,
536
                               int32_t *sample_buffer);
537

    
538
    /* vc1 functions */
539
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
540
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
541
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
542
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
543
    void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
544
    void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
545
    void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
546
    void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
547
    void (*vc1_v_overlap)(uint8_t* src, int stride);
548
    void (*vc1_h_overlap)(uint8_t* src, int stride);
549
    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
550
    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
551
    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
552
    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
553
    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
554
    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
555
    /* put 8x8 block with bicubic interpolation and quarterpel precision
556
     * last argument is actually round value instead of height
557
     */
558
    op_pixels_func put_vc1_mspel_pixels_tab[16];
559
    op_pixels_func avg_vc1_mspel_pixels_tab[16];
560

    
561
    /* intrax8 functions */
562
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
563
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
564
           int * range, int * sum,  int edges);
565

    
566
    /**
567
     * Calculate scalar product of two vectors.
568
     * @param len length of vectors, should be multiple of 16
569
     * @param shift number of bits to discard from product
570
     */
571
    int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
572
    /* ape functions */
573
    /**
574
     * Calculate scalar product of v1 and v2,
575
     * and v1[i] += v3[i] * mul
576
     * @param len length of vectors, should be multiple of 16
577
     */
578
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, int16_t *v2, int16_t *v3, int len, int mul);
579

    
580
    /* rv30 functions */
581
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
582
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
583

    
584
    /* rv40 functions */
585
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
586
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
587
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
588
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
589

    
590
    /* bink functions */
591
    op_fill_func fill_block_tab[2];
592
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
593
} DSPContext;
594

    
595
void dsputil_static_init(void);
596
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
597

    
598
int ff_check_alignment(void);
599

    
600
/**
601
 * permute block according to permuatation.
602
 * @param last last non zero element in scantable order
603
 */
604
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
605

    
606
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
607

    
608
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
609

    
610
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
611
{
612
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
613
}
614

    
615
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
616
{
617
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
618
}
619

    
620
static inline int get_penalty_factor(int lambda, int lambda2, int type){
621
    switch(type&0xFF){
622
    default:
623
    case FF_CMP_SAD:
624
        return lambda>>FF_LAMBDA_SHIFT;
625
    case FF_CMP_DCT:
626
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
627
    case FF_CMP_W53:
628
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
629
    case FF_CMP_W97:
630
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
631
    case FF_CMP_SATD:
632
    case FF_CMP_DCT264:
633
        return (2*lambda)>>FF_LAMBDA_SHIFT;
634
    case FF_CMP_RD:
635
    case FF_CMP_PSNR:
636
    case FF_CMP_SSE:
637
    case FF_CMP_NSSE:
638
        return lambda2>>FF_LAMBDA_SHIFT;
639
    case FF_CMP_BIT:
640
        return 1;
641
    }
642
}
643

    
644
/**
645
 * Empty mmx state.
646
 * this must be called between any dsp function and float/double code.
647
 * for example sin(); dsp->idct_put(); emms_c(); cos()
648
 */
649
#define emms_c()
650

    
651
/* should be defined by architectures supporting
652
   one or more MultiMedia extension */
653
int mm_support(void);
654
extern int mm_flags;
655

    
656
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
657
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
658
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
659
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
660
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
661
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
662
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
663
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
664
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
665

    
666
#if HAVE_MMX
667

    
668
#undef emms_c
669

    
670
static inline void emms(void)
671
{
672
    __asm__ volatile ("emms;":::"memory");
673
}
674

    
675

    
676
#define emms_c() \
677
{\
678
    if (mm_flags & FF_MM_MMX)\
679
        emms();\
680
}
681

    
682
#elif ARCH_ARM
683

    
684
#if HAVE_NEON
685
#   define STRIDE_ALIGN 16
686
#endif
687

    
688
#elif ARCH_PPC
689

    
690
#define STRIDE_ALIGN 16
691

    
692
#elif HAVE_MMI
693

    
694
#define STRIDE_ALIGN 16
695

    
696
#else
697

    
698
#define mm_flags 0
699
#define mm_support() 0
700

    
701
#endif
702

    
703
#ifndef STRIDE_ALIGN
704
#   define STRIDE_ALIGN 8
705
#endif
706

    
707
#define LOCAL_ALIGNED(a, t, v, s, ...)                          \
708
    uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)];              \
709
    t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
710

    
711
#if HAVE_LOCAL_ALIGNED_8
712
#   define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
713
#else
714
#   define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
715
#endif
716

    
717
#if HAVE_LOCAL_ALIGNED_16
718
#   define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
719
#else
720
#   define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
721
#endif
722

    
723
/* PSNR */
724
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
725
              int orig_linesize[3], int coded_linesize,
726
              AVCodecContext *avctx);
727

    
728
/* FFT computation */
729

    
730
/* NOTE: soon integer code will be added, so you must use the
731
   FFTSample type */
732
typedef float FFTSample;
733

    
734
typedef struct FFTComplex {
735
    FFTSample re, im;
736
} FFTComplex;
737

    
738
typedef struct FFTContext {
739
    int nbits;
740
    int inverse;
741
    uint16_t *revtab;
742
    FFTComplex *exptab;
743
    FFTComplex *exptab1; /* only used by SSE code */
744
    FFTComplex *tmp_buf;
745
    int mdct_size; /* size of MDCT (i.e. number of input data * 2) */
746
    int mdct_bits; /* n = 2^nbits */
747
    /* pre/post rotation tables */
748
    FFTSample *tcos;
749
    FFTSample *tsin;
750
    void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
751
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
752
    void (*imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
753
    void (*imdct_half)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
754
    void (*mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
755
    int split_radix;
756
    int permutation;
757
#define FF_MDCT_PERM_NONE       0
758
#define FF_MDCT_PERM_INTERLEAVE 1
759
} FFTContext;
760

    
761
#if CONFIG_HARDCODED_TABLES
762
#define COSTABLE_CONST const
763
#define SINTABLE_CONST const
764
#define SINETABLE_CONST const
765
#else
766
#define COSTABLE_CONST
767
#define SINTABLE_CONST
768
#define SINETABLE_CONST
769
#endif
770

    
771
#define COSTABLE(size) \
772
    COSTABLE_CONST DECLARE_ALIGNED(16, FFTSample, ff_cos_##size)[size/2]
773
#define SINTABLE(size) \
774
    SINTABLE_CONST DECLARE_ALIGNED(16, FFTSample, ff_sin_##size)[size/2]
775
#define SINETABLE(size) \
776
    SINETABLE_CONST DECLARE_ALIGNED(16, float, ff_sine_##size)[size]
777
extern COSTABLE(16);
778
extern COSTABLE(32);
779
extern COSTABLE(64);
780
extern COSTABLE(128);
781
extern COSTABLE(256);
782
extern COSTABLE(512);
783
extern COSTABLE(1024);
784
extern COSTABLE(2048);
785
extern COSTABLE(4096);
786
extern COSTABLE(8192);
787
extern COSTABLE(16384);
788
extern COSTABLE(32768);
789
extern COSTABLE(65536);
790
extern COSTABLE_CONST FFTSample* const ff_cos_tabs[17];
791

    
792
/**
793
 * Initializes the cosine table in ff_cos_tabs[index]
794
 * \param index index in ff_cos_tabs array of the table to initialize
795
 */
796
void ff_init_ff_cos_tabs(int index);
797

    
798
extern SINTABLE(16);
799
extern SINTABLE(32);
800
extern SINTABLE(64);
801
extern SINTABLE(128);
802
extern SINTABLE(256);
803
extern SINTABLE(512);
804
extern SINTABLE(1024);
805
extern SINTABLE(2048);
806
extern SINTABLE(4096);
807
extern SINTABLE(8192);
808
extern SINTABLE(16384);
809
extern SINTABLE(32768);
810
extern SINTABLE(65536);
811

    
812
/**
813
 * Sets up a complex FFT.
814
 * @param nbits           log2 of the length of the input array
815
 * @param inverse         if 0 perform the forward transform, if 1 perform the inverse
816
 */
817
int ff_fft_init(FFTContext *s, int nbits, int inverse);
818
void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
819
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
820

    
821
void ff_fft_init_altivec(FFTContext *s);
822
void ff_fft_init_mmx(FFTContext *s);
823
void ff_fft_init_arm(FFTContext *s);
824

    
825
/**
826
 * Do the permutation needed BEFORE calling ff_fft_calc().
827
 */
828
static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
829
{
830
    s->fft_permute(s, z);
831
}
832
/**
833
 * Do a complex FFT with the parameters defined in ff_fft_init(). The
834
 * input data must be permuted before. No 1.0/sqrt(n) normalization is done.
835
 */
836
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
837
{
838
    s->fft_calc(s, z);
839
}
840
void ff_fft_end(FFTContext *s);
841

    
842
/* MDCT computation */
843

    
844
static inline void ff_imdct_calc(FFTContext *s, FFTSample *output, const FFTSample *input)
845
{
846
    s->imdct_calc(s, output, input);
847
}
848
static inline void ff_imdct_half(FFTContext *s, FFTSample *output, const FFTSample *input)
849
{
850
    s->imdct_half(s, output, input);
851
}
852

    
853
static inline void ff_mdct_calc(FFTContext *s, FFTSample *output,
854
                                const FFTSample *input)
855
{
856
    s->mdct_calc(s, output, input);
857
}
858

    
859
/**
860
 * Generate a Kaiser-Bessel Derived Window.
861
 * @param   window  pointer to half window
862
 * @param   alpha   determines window shape
863
 * @param   n       size of half window
864
 */
865
void ff_kbd_window_init(float *window, float alpha, int n);
866

    
867
/**
868
 * Generate a sine window.
869
 * @param   window  pointer to half window
870
 * @param   n       size of half window
871
 */
872
void ff_sine_window_init(float *window, int n);
873
/**
874
 * initialize the specified entry of ff_sine_windows
875
 */
876
void ff_init_ff_sine_windows(int index);
877
extern SINETABLE(  32);
878
extern SINETABLE(  64);
879
extern SINETABLE( 128);
880
extern SINETABLE( 256);
881
extern SINETABLE( 512);
882
extern SINETABLE(1024);
883
extern SINETABLE(2048);
884
extern SINETABLE(4096);
885
extern SINETABLE_CONST float * const ff_sine_windows[13];
886

    
887
int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale);
888
void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
889
void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input);
890
void ff_mdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
891
void ff_mdct_end(FFTContext *s);
892

    
893
/* Real Discrete Fourier Transform */
894

    
895
enum RDFTransformType {
896
    RDFT,
897
    IRDFT,
898
    RIDFT,
899
    IRIDFT,
900
};
901

    
902
typedef struct {
903
    int nbits;
904
    int inverse;
905
    int sign_convention;
906

    
907
    /* pre/post rotation tables */
908
    const FFTSample *tcos;
909
    SINTABLE_CONST FFTSample *tsin;
910
    FFTContext fft;
911
} RDFTContext;
912

    
913
/**
914
 * Sets up a real FFT.
915
 * @param nbits           log2 of the length of the input array
916
 * @param trans           the type of transform
917
 */
918
int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans);
919
void ff_rdft_calc(RDFTContext *s, FFTSample *data);
920
void ff_rdft_end(RDFTContext *s);
921

    
922
/* Discrete Cosine Transform */
923

    
924
typedef struct {
925
    int nbits;
926
    int inverse;
927
    FFTSample *data;
928
    RDFTContext rdft;
929
    const float *costab;
930
    FFTSample *csc2;
931
} DCTContext;
932

    
933
/**
934
 * Sets up (Inverse)DCT.
935
 * @param nbits           log2 of the length of the input array
936
 * @param inverse         >0 forward transform, <0 inverse transform
937
 */
938
int  ff_dct_init(DCTContext *s, int nbits, int inverse);
939
void ff_dct_calc(DCTContext *s, FFTSample *data);
940
void ff_dct_end (DCTContext *s);
941

    
942
#define WRAPPER8_16(name8, name16)\
943
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
944
    return name8(s, dst           , src           , stride, h)\
945
          +name8(s, dst+8         , src+8         , stride, h);\
946
}
947

    
948
#define WRAPPER8_16_SQ(name8, name16)\
949
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
950
    int score=0;\
951
    score +=name8(s, dst           , src           , stride, 8);\
952
    score +=name8(s, dst+8         , src+8         , stride, 8);\
953
    if(h==16){\
954
        dst += 8*stride;\
955
        src += 8*stride;\
956
        score +=name8(s, dst           , src           , stride, 8);\
957
        score +=name8(s, dst+8         , src+8         , stride, 8);\
958
    }\
959
    return score;\
960
}
961

    
962

    
963
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
964
{
965
    int i;
966
    for(i=0; i<h; i++)
967
    {
968
        AV_WN16(dst   , AV_RN16(src   ));
969
        dst+=dstStride;
970
        src+=srcStride;
971
    }
972
}
973

    
974
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
975
{
976
    int i;
977
    for(i=0; i<h; i++)
978
    {
979
        AV_WN32(dst   , AV_RN32(src   ));
980
        dst+=dstStride;
981
        src+=srcStride;
982
    }
983
}
984

    
985
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
986
{
987
    int i;
988
    for(i=0; i<h; i++)
989
    {
990
        AV_WN32(dst   , AV_RN32(src   ));
991
        AV_WN32(dst+4 , AV_RN32(src+4 ));
992
        dst+=dstStride;
993
        src+=srcStride;
994
    }
995
}
996

    
997
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
998
{
999
    int i;
1000
    for(i=0; i<h; i++)
1001
    {
1002
        AV_WN32(dst   , AV_RN32(src   ));
1003
        AV_WN32(dst+4 , AV_RN32(src+4 ));
1004
        dst[8]= src[8];
1005
        dst+=dstStride;
1006
        src+=srcStride;
1007
    }
1008
}
1009

    
1010
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
1011
{
1012
    int i;
1013
    for(i=0; i<h; i++)
1014
    {
1015
        AV_WN32(dst   , AV_RN32(src   ));
1016
        AV_WN32(dst+4 , AV_RN32(src+4 ));
1017
        AV_WN32(dst+8 , AV_RN32(src+8 ));
1018
        AV_WN32(dst+12, AV_RN32(src+12));
1019
        dst+=dstStride;
1020
        src+=srcStride;
1021
    }
1022
}
1023

    
1024
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
1025
{
1026
    int i;
1027
    for(i=0; i<h; i++)
1028
    {
1029
        AV_WN32(dst   , AV_RN32(src   ));
1030
        AV_WN32(dst+4 , AV_RN32(src+4 ));
1031
        AV_WN32(dst+8 , AV_RN32(src+8 ));
1032
        AV_WN32(dst+12, AV_RN32(src+12));
1033
        dst[16]= src[16];
1034
        dst+=dstStride;
1035
        src+=srcStride;
1036
    }
1037
}
1038

    
1039
#endif /* AVCODEC_DSPUTIL_H */