Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 79cb09b2

History | View | Annotate | Download (33.2 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 AVCODEC_DSPUTIL_H
31
#define AVCODEC_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
void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
64
void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
65
void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
66
void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
67

    
68
void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
69
                              const float *src2, int src3, int blocksize, int step);
70
void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
71
                             const float *win, float add_bias, int len);
72
void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
73
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
74

    
75
/* encoding scans */
76
extern const uint8_t ff_alternate_horizontal_scan[64];
77
extern const uint8_t ff_alternate_vertical_scan[64];
78
extern const uint8_t ff_zigzag_direct[64];
79
extern const uint8_t ff_zigzag248_direct[64];
80

    
81
/* pixel operations */
82
#define MAX_NEG_CROP 1024
83

    
84
/* temporary */
85
extern uint32_t ff_squareTbl[512];
86
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
87

    
88
/* VP3 DSP functions */
89
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
90
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
91
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
92

    
93
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
94
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
95

    
96
/* 1/2^n downscaling functions from imgconvert.c */
97
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
98
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
99
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
100
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
101

    
102
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
103
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
104

    
105
/* minimum alignment rules ;)
106
If you notice errors in the align stuff, need more alignment for some ASM code
107
for some CPU or need to use a function with less aligned data then send a mail
108
to the ffmpeg-devel mailing list, ...
109

110
!warning These alignments might not match reality, (missing attribute((align))
111
stuff somewhere possible).
112
I (Michael) did not check them, these are just the alignments which I think
113
could be reached easily ...
114

115
!future video codecs might need functions with less strict alignment
116
*/
117

    
118
/*
119
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
120
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
121
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
122
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
123
void clear_blocks_c(DCTELEM *blocks);
124
*/
125

    
126
/* add and put pixel (decoding) */
127
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
128
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
129
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
130
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);
131
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
132
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
133
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
134
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
135

    
136
#define DEF_OLD_QPEL(name)\
137
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
138
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
139
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
140

    
141
DEF_OLD_QPEL(qpel16_mc11_old_c)
142
DEF_OLD_QPEL(qpel16_mc31_old_c)
143
DEF_OLD_QPEL(qpel16_mc12_old_c)
144
DEF_OLD_QPEL(qpel16_mc32_old_c)
145
DEF_OLD_QPEL(qpel16_mc13_old_c)
146
DEF_OLD_QPEL(qpel16_mc33_old_c)
147
DEF_OLD_QPEL(qpel8_mc11_old_c)
148
DEF_OLD_QPEL(qpel8_mc31_old_c)
149
DEF_OLD_QPEL(qpel8_mc12_old_c)
150
DEF_OLD_QPEL(qpel8_mc32_old_c)
151
DEF_OLD_QPEL(qpel8_mc13_old_c)
152
DEF_OLD_QPEL(qpel8_mc33_old_c)
153

    
154
#define CALL_2X_PIXELS(a, b, n)\
155
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
156
    b(block  , pixels  , line_size, h);\
157
    b(block+n, pixels+n, line_size, h);\
158
}
159

    
160
/* motion estimation */
161
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
162
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
163
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))*/;
164

    
165

    
166
// for snow slices
167
typedef struct slice_buffer_s slice_buffer;
168

    
169
/**
170
 * Scantable.
171
 */
172
typedef struct ScanTable{
173
    const uint8_t *scantable;
174
    uint8_t permutated[64];
175
    uint8_t raster_end[64];
176
#ifdef ARCH_PPC
177
                /** Used by dct_quantize_altivec to find last-non-zero */
178
    DECLARE_ALIGNED(16, uint8_t, inverse[64]);
179
#endif
180
} ScanTable;
181

    
182
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
183

    
184
void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
185
                         int block_w, int block_h,
186
                         int src_x, int src_y, int w, int h);
187

    
188
/**
189
 * DSPContext.
190
 */
191
typedef struct DSPContext {
192
    /* pixel ops : interface with DCT */
193
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
194
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
195
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
196
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
197
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
198
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
199
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
200
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
201
    /**
202
     * translational global motion compensation.
203
     */
204
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
205
    /**
206
     * global motion compensation.
207
     */
208
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
209
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
210
    void (*clear_block)(DCTELEM *block/*align 16*/);
211
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
212
    int (*pix_sum)(uint8_t * pix, int line_size);
213
    int (*pix_norm1)(uint8_t * pix, int line_size);
214
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
215

    
216
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
217
    me_cmp_func sse[5];
218
    me_cmp_func hadamard8_diff[5];
219
    me_cmp_func dct_sad[5];
220
    me_cmp_func quant_psnr[5];
221
    me_cmp_func bit[5];
222
    me_cmp_func rd[5];
223
    me_cmp_func vsad[5];
224
    me_cmp_func vsse[5];
225
    me_cmp_func nsse[5];
226
    me_cmp_func w53[5];
227
    me_cmp_func w97[5];
228
    me_cmp_func dct_max[5];
229
    me_cmp_func dct264_sad[5];
230

    
231
    me_cmp_func me_pre_cmp[5];
232
    me_cmp_func me_cmp[5];
233
    me_cmp_func me_sub_cmp[5];
234
    me_cmp_func mb_cmp[5];
235
    me_cmp_func ildct_cmp[5]; //only width 16 used
236
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
237

    
238
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
239
                             int size);
240

    
241
    /**
242
     * Halfpel motion compensation with rounding (a+b+1)>>1.
243
     * this is an array[4][4] of motion compensation functions for 4
244
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
245
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
246
     * @param block destination where the result is stored
247
     * @param pixels source
248
     * @param line_size number of bytes in a horizontal line of block
249
     * @param h height
250
     */
251
    op_pixels_func put_pixels_tab[4][4];
252

    
253
    /**
254
     * Halfpel motion compensation with rounding (a+b+1)>>1.
255
     * This is an array[4][4] of motion compensation functions for 4
256
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
257
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
258
     * @param block destination into which the result is averaged (a+b+1)>>1
259
     * @param pixels source
260
     * @param line_size number of bytes in a horizontal line of block
261
     * @param h height
262
     */
263
    op_pixels_func avg_pixels_tab[4][4];
264

    
265
    /**
266
     * Halfpel motion compensation with no rounding (a+b)>>1.
267
     * this is an array[2][4] of motion compensation functions for 2
268
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
269
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
270
     * @param block destination where the result is stored
271
     * @param pixels source
272
     * @param line_size number of bytes in a horizontal line of block
273
     * @param h height
274
     */
275
    op_pixels_func put_no_rnd_pixels_tab[4][4];
276

    
277
    /**
278
     * Halfpel motion compensation with no rounding (a+b)>>1.
279
     * this is an array[2][4] of motion compensation functions for 2
280
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
281
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
282
     * @param block destination into which the result is averaged (a+b)>>1
283
     * @param pixels source
284
     * @param line_size number of bytes in a horizontal line of block
285
     * @param h height
286
     */
287
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
288

    
289
    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);
290

    
291
    /**
292
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
293
     * this is an array[12] of motion compensation functions for the 9 thirdpe
294
     * positions<br>
295
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
296
     * @param block destination where the result is stored
297
     * @param pixels source
298
     * @param line_size number of bytes in a horizontal line of block
299
     * @param h height
300
     */
301
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
302
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
303

    
304
    qpel_mc_func put_qpel_pixels_tab[2][16];
305
    qpel_mc_func avg_qpel_pixels_tab[2][16];
306
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
307
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
308
    qpel_mc_func put_mspel_pixels_tab[8];
309

    
310
    /**
311
     * h264 Chroma MC
312
     */
313
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
314
    /* This is really one func used in VC-1 decoding */
315
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
316
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
317

    
318
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
319
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
320

    
321
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
322
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
323

    
324
    h264_weight_func weight_h264_pixels_tab[10];
325
    h264_biweight_func biweight_h264_pixels_tab[10];
326

    
327
    /* AVS specific */
328
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
329
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
330
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
331
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
332
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
333
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
334
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
335

    
336
    me_cmp_func pix_abs[2][4];
337

    
338
    /* huffyuv specific */
339
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
340
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
341
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
342
    /**
343
     * subtract huffyuv's variant of median prediction
344
     * note, this might read from src1[-1], src2[-1]
345
     */
346
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
347
    /* this might write to dst[w] */
348
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
349
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
350

    
351
    void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
352
    void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
353
    /* v/h_loop_filter_luma_intra: align 16 */
354
    void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
355
    void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
356
    void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
357
    void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
358
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
359
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
360
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
361
    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],
362
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
363

    
364
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
365
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
366

    
367
    void (*h261_loop_filter)(uint8_t *src, int stride);
368

    
369
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
370
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
371

    
372
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
373
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
374

    
375
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
376
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
377
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
378
    /* no alignment needed */
379
    void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
380
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
381
    void (*vector_fmul)(float *dst, const float *src, int len);
382
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
383
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
384
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
385
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
386
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
387
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
388
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
389

    
390
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
391
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
392
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
393
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
394

    
395
    /* (I)DCT */
396
    void (*fdct)(DCTELEM *block/* align 16*/);
397
    void (*fdct248)(DCTELEM *block/* align 16*/);
398

    
399
    /* IDCT really*/
400
    void (*idct)(DCTELEM *block/* align 16*/);
401

    
402
    /**
403
     * block -> idct -> clip to unsigned 8 bit -> dest.
404
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
405
     * @param line_size size in bytes of a horizontal line of dest
406
     */
407
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
408

    
409
    /**
410
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
411
     * @param line_size size in bytes of a horizontal line of dest
412
     */
413
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
414

    
415
    /**
416
     * idct input permutation.
417
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
418
     * IDCT)
419
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
420
     * with the zigzag/alternate scan<br>
421
     * an example to avoid confusion:
422
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
423
     * - (x -> referece dct -> reference idct -> x)
424
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
425
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
426
     */
427
    uint8_t idct_permutation[64];
428
    int idct_permutation_type;
429
#define FF_NO_IDCT_PERM 1
430
#define FF_LIBMPEG2_IDCT_PERM 2
431
#define FF_SIMPLE_IDCT_PERM 3
432
#define FF_TRANSPOSE_IDCT_PERM 4
433
#define FF_PARTTRANS_IDCT_PERM 5
434
#define FF_SSE2_IDCT_PERM 6
435

    
436
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
437
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
438
#define BASIS_SHIFT 16
439
#define RECON_SHIFT 6
440

    
441
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
442
#define EDGE_WIDTH 16
443

    
444
    /* h264 functions */
445
    /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
446
       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
447
        The reason for above, is that no 2 out of one list may use a different permutation.
448
    */
449
    void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
450
    void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
451
    void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
452
    void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
453
    void (*h264_dct)(DCTELEM block[4][4]);
454
    void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
455
    void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
456
    void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
457
    void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
458

    
459
    /* snow wavelet */
460
    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
461
    void (*horizontal_compose97i)(IDWTELEM *b, int width);
462
    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);
463

    
464
    void (*prefetch)(void *mem, int stride, int h);
465

    
466
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
467

    
468
    /* vc1 functions */
469
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
470
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
471
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
472
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
473
    void (*vc1_v_overlap)(uint8_t* src, int stride);
474
    void (*vc1_h_overlap)(uint8_t* src, int stride);
475
    /* put 8x8 block with bicubic interpolation and quarterpel precision
476
     * last argument is actually round value instead of height
477
     */
478
    op_pixels_func put_vc1_mspel_pixels_tab[16];
479

    
480
    /* intrax8 functions */
481
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
482
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
483
           int * range, int * sum,  int edges);
484

    
485
    /* ape functions */
486
    /**
487
     * Add contents of the second vector to the first one.
488
     * @param len length of vectors, should be multiple of 16
489
     */
490
    void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
491
    /**
492
     * Add contents of the second vector to the first one.
493
     * @param len length of vectors, should be multiple of 16
494
     */
495
    void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
496
    /**
497
     * Calculate scalar product of two vectors.
498
     * @param len length of vectors, should be multiple of 16
499
     * @param shift number of bits to discard from product
500
     */
501
    int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
502

    
503
    /* rv30 functions */
504
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
505
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
506

    
507
    /* rv40 functions */
508
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
509
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
510
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
511
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
512
} DSPContext;
513

    
514
void dsputil_static_init(void);
515
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
516

    
517
int ff_check_alignment(void);
518

    
519
/**
520
 * permute block according to permuatation.
521
 * @param last last non zero element in scantable order
522
 */
523
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
524

    
525
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
526

    
527
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
528

    
529
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
530
{
531
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
532
}
533

    
534
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
535
{
536
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
537
}
538

    
539
static inline int get_penalty_factor(int lambda, int lambda2, int type){
540
    switch(type&0xFF){
541
    default:
542
    case FF_CMP_SAD:
543
        return lambda>>FF_LAMBDA_SHIFT;
544
    case FF_CMP_DCT:
545
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
546
    case FF_CMP_W53:
547
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
548
    case FF_CMP_W97:
549
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
550
    case FF_CMP_SATD:
551
    case FF_CMP_DCT264:
552
        return (2*lambda)>>FF_LAMBDA_SHIFT;
553
    case FF_CMP_RD:
554
    case FF_CMP_PSNR:
555
    case FF_CMP_SSE:
556
    case FF_CMP_NSSE:
557
        return lambda2>>FF_LAMBDA_SHIFT;
558
    case FF_CMP_BIT:
559
        return 1;
560
    }
561
}
562

    
563
/**
564
 * Empty mmx state.
565
 * this must be called between any dsp function and float/double code.
566
 * for example sin(); dsp->idct_put(); emms_c(); cos()
567
 */
568
#define emms_c()
569

    
570
/* should be defined by architectures supporting
571
   one or more MultiMedia extension */
572
int mm_support(void);
573

    
574
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
575
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
576
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
577
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
578
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
579
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
580
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
581
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
582
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
583

    
584
#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
585

    
586
#if defined(HAVE_MMX)
587

    
588
#undef emms_c
589

    
590
extern int mm_flags;
591

    
592
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
593
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
594
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
595

    
596
static inline void emms(void)
597
{
598
    __asm__ volatile ("emms;":::"memory");
599
}
600

    
601

    
602
#define emms_c() \
603
{\
604
    if (mm_flags & FF_MM_MMX)\
605
        emms();\
606
}
607

    
608
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
609

    
610
#elif defined(ARCH_ARM)
611

    
612
extern int mm_flags;
613

    
614
#ifdef HAVE_NEON
615
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
616
#   define STRIDE_ALIGN 16
617
#endif
618

    
619
#elif defined(ARCH_PPC)
620

    
621
extern int mm_flags;
622

    
623
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
624
#define STRIDE_ALIGN 16
625

    
626
#elif defined(HAVE_MMI)
627

    
628
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
629
#define STRIDE_ALIGN 16
630

    
631
#else
632

    
633
#define mm_flags 0
634
#define mm_support() 0
635

    
636
#endif
637

    
638
#ifndef DECLARE_ALIGNED_8
639
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
640
#endif
641

    
642
#ifndef STRIDE_ALIGN
643
#   define STRIDE_ALIGN 8
644
#endif
645

    
646
/* PSNR */
647
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
648
              int orig_linesize[3], int coded_linesize,
649
              AVCodecContext *avctx);
650

    
651
/* FFT computation */
652

    
653
/* NOTE: soon integer code will be added, so you must use the
654
   FFTSample type */
655
typedef float FFTSample;
656

    
657
struct MDCTContext;
658

    
659
typedef struct FFTComplex {
660
    FFTSample re, im;
661
} FFTComplex;
662

    
663
typedef struct FFTContext {
664
    int nbits;
665
    int inverse;
666
    uint16_t *revtab;
667
    FFTComplex *exptab;
668
    FFTComplex *exptab1; /* only used by SSE code */
669
    FFTComplex *tmp_buf;
670
    void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
671
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
672
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
673
    void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
674
} FFTContext;
675

    
676
int ff_fft_init(FFTContext *s, int nbits, int inverse);
677
void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
678
void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
679
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
680
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
681
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
682
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
683
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
684

    
685
static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
686
{
687
    s->fft_permute(s, z);
688
}
689
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
690
{
691
    s->fft_calc(s, z);
692
}
693
void ff_fft_end(FFTContext *s);
694

    
695
/* MDCT computation */
696

    
697
typedef struct MDCTContext {
698
    int n;  /* size of MDCT (i.e. number of input data * 2) */
699
    int nbits; /* n = 2^nbits */
700
    /* pre/post rotation tables */
701
    FFTSample *tcos;
702
    FFTSample *tsin;
703
    FFTContext fft;
704
} MDCTContext;
705

    
706
static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
707
{
708
    s->fft.imdct_calc(s, output, input);
709
}
710
static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
711
{
712
    s->fft.imdct_half(s, output, input);
713
}
714

    
715
/**
716
 * Generate a Kaiser-Bessel Derived Window.
717
 * @param   window  pointer to half window
718
 * @param   alpha   determines window shape
719
 * @param   n       size of half window
720
 */
721
void ff_kbd_window_init(float *window, float alpha, int n);
722

    
723
/**
724
 * Generate a sine window.
725
 * @param   window  pointer to half window
726
 * @param   n       size of half window
727
 */
728
void ff_sine_window_init(float *window, int n);
729
extern float ff_sine_128 [ 128];
730
extern float ff_sine_256 [ 256];
731
extern float ff_sine_512 [ 512];
732
extern float ff_sine_1024[1024];
733
extern float ff_sine_2048[2048];
734
extern float *ff_sine_windows[5];
735

    
736
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
737
void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
738
void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
739
void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
740
void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
741
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
742
void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
743
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
744
void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
745
void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
746
void ff_mdct_end(MDCTContext *s);
747

    
748
#define WRAPPER8_16(name8, name16)\
749
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
750
    return name8(s, dst           , src           , stride, h)\
751
          +name8(s, dst+8         , src+8         , stride, h);\
752
}
753

    
754
#define WRAPPER8_16_SQ(name8, name16)\
755
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
756
    int score=0;\
757
    score +=name8(s, dst           , src           , stride, 8);\
758
    score +=name8(s, dst+8         , src+8         , stride, 8);\
759
    if(h==16){\
760
        dst += 8*stride;\
761
        src += 8*stride;\
762
        score +=name8(s, dst           , src           , stride, 8);\
763
        score +=name8(s, dst+8         , src+8         , stride, 8);\
764
    }\
765
    return score;\
766
}
767

    
768

    
769
static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
770
{
771
    int i;
772
    for(i=0; i<h; i++)
773
    {
774
        AV_WN16(dst   , AV_RN16(src   ));
775
        dst+=dstStride;
776
        src+=srcStride;
777
    }
778
}
779

    
780
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
781
{
782
    int i;
783
    for(i=0; i<h; i++)
784
    {
785
        AV_WN32(dst   , AV_RN32(src   ));
786
        dst+=dstStride;
787
        src+=srcStride;
788
    }
789
}
790

    
791
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
792
{
793
    int i;
794
    for(i=0; i<h; i++)
795
    {
796
        AV_WN32(dst   , AV_RN32(src   ));
797
        AV_WN32(dst+4 , AV_RN32(src+4 ));
798
        dst+=dstStride;
799
        src+=srcStride;
800
    }
801
}
802

    
803
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
804
{
805
    int i;
806
    for(i=0; i<h; i++)
807
    {
808
        AV_WN32(dst   , AV_RN32(src   ));
809
        AV_WN32(dst+4 , AV_RN32(src+4 ));
810
        dst[8]= src[8];
811
        dst+=dstStride;
812
        src+=srcStride;
813
    }
814
}
815

    
816
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
817
{
818
    int i;
819
    for(i=0; i<h; i++)
820
    {
821
        AV_WN32(dst   , AV_RN32(src   ));
822
        AV_WN32(dst+4 , AV_RN32(src+4 ));
823
        AV_WN32(dst+8 , AV_RN32(src+8 ));
824
        AV_WN32(dst+12, AV_RN32(src+12));
825
        dst+=dstStride;
826
        src+=srcStride;
827
    }
828
}
829

    
830
static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
831
{
832
    int i;
833
    for(i=0; i<h; i++)
834
    {
835
        AV_WN32(dst   , AV_RN32(src   ));
836
        AV_WN32(dst+4 , AV_RN32(src+4 ));
837
        AV_WN32(dst+8 , AV_RN32(src+8 ));
838
        AV_WN32(dst+12, AV_RN32(src+12));
839
        dst[16]= src[16];
840
        dst+=dstStride;
841
        src+=srcStride;
842
    }
843
}
844

    
845
#endif /* AVCODEC_DSPUTIL_H */