Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ a2fc0f6a

History | View | Annotate | Download (31.7 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 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

    
64
void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
65
                              const float *src2, int src3, int blocksize, int step);
66
void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
67
                             const float *win, float add_bias, int len);
68
void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
69
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
70

    
71
/* encoding scans */
72
extern const uint8_t ff_alternate_horizontal_scan[64];
73
extern const uint8_t ff_alternate_vertical_scan[64];
74
extern const uint8_t ff_zigzag_direct[64];
75
extern const uint8_t ff_zigzag248_direct[64];
76

    
77
/* pixel operations */
78
#define MAX_NEG_CROP 1024
79

    
80
/* temporary */
81
extern uint32_t ff_squareTbl[512];
82
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
83

    
84
/* VP3 DSP functions */
85
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
86
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
87
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
88

    
89
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
90
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
91

    
92
/* 1/2^n downscaling functions from imgconvert.c */
93
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
94
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
95
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
96
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
97

    
98
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
99
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
100

    
101
/* minimum alignment rules ;)
102
If you notice errors in the align stuff, need more alignment for some ASM code
103
for some CPU or need to use a function with less aligned data then send a mail
104
to the ffmpeg-devel mailing list, ...
105

106
!warning These alignments might not match reality, (missing attribute((align))
107
stuff somewhere possible).
108
I (Michael) did not check them, these are just the alignments which I think
109
could be reached easily ...
110

111
!future video codecs might need functions with less strict alignment
112
*/
113

    
114
/*
115
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
116
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
117
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
118
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
119
void clear_blocks_c(DCTELEM *blocks);
120
*/
121

    
122
/* add and put pixel (decoding) */
123
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
124
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
125
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
126
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);
127
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
128
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
129
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
130
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
131

    
132
#define DEF_OLD_QPEL(name)\
133
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
134
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
135
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
136

    
137
DEF_OLD_QPEL(qpel16_mc11_old_c)
138
DEF_OLD_QPEL(qpel16_mc31_old_c)
139
DEF_OLD_QPEL(qpel16_mc12_old_c)
140
DEF_OLD_QPEL(qpel16_mc32_old_c)
141
DEF_OLD_QPEL(qpel16_mc13_old_c)
142
DEF_OLD_QPEL(qpel16_mc33_old_c)
143
DEF_OLD_QPEL(qpel8_mc11_old_c)
144
DEF_OLD_QPEL(qpel8_mc31_old_c)
145
DEF_OLD_QPEL(qpel8_mc12_old_c)
146
DEF_OLD_QPEL(qpel8_mc32_old_c)
147
DEF_OLD_QPEL(qpel8_mc13_old_c)
148
DEF_OLD_QPEL(qpel8_mc33_old_c)
149

    
150
#define CALL_2X_PIXELS(a, b, n)\
151
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
152
    b(block  , pixels  , line_size, h);\
153
    b(block+n, pixels+n, line_size, h);\
154
}
155

    
156
/* motion estimation */
157
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
158
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
159
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))*/;
160

    
161

    
162
// for snow slices
163
typedef struct slice_buffer_s slice_buffer;
164

    
165
/**
166
 * Scantable.
167
 */
168
typedef struct ScanTable{
169
    const uint8_t *scantable;
170
    uint8_t permutated[64];
171
    uint8_t raster_end[64];
172
#ifdef ARCH_POWERPC
173
                /** Used by dct_quantize_altivec to find last-non-zero */
174
    DECLARE_ALIGNED(16, uint8_t, inverse[64]);
175
#endif
176
} ScanTable;
177

    
178
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
179

    
180
void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
181
                         int block_w, int block_h,
182
                         int src_x, int src_y, int w, int h);
183

    
184
/**
185
 * DSPContext.
186
 */
187
typedef struct DSPContext {
188
    /* pixel ops : interface with DCT */
189
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
190
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
191
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
192
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
193
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
194
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
195
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
196
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
197
    /**
198
     * translational global motion compensation.
199
     */
200
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
201
    /**
202
     * global motion compensation.
203
     */
204
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
205
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
206
    void (*clear_block)(DCTELEM *block/*align 16*/);
207
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
208
    int (*pix_sum)(uint8_t * pix, int line_size);
209
    int (*pix_norm1)(uint8_t * pix, int line_size);
210
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
211

    
212
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
213
    me_cmp_func sse[5];
214
    me_cmp_func hadamard8_diff[5];
215
    me_cmp_func dct_sad[5];
216
    me_cmp_func quant_psnr[5];
217
    me_cmp_func bit[5];
218
    me_cmp_func rd[5];
219
    me_cmp_func vsad[5];
220
    me_cmp_func vsse[5];
221
    me_cmp_func nsse[5];
222
    me_cmp_func w53[5];
223
    me_cmp_func w97[5];
224
    me_cmp_func dct_max[5];
225
    me_cmp_func dct264_sad[5];
226

    
227
    me_cmp_func me_pre_cmp[5];
228
    me_cmp_func me_cmp[5];
229
    me_cmp_func me_sub_cmp[5];
230
    me_cmp_func mb_cmp[5];
231
    me_cmp_func ildct_cmp[5]; //only width 16 used
232
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
233

    
234
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
235
                             int size);
236

    
237
    /**
238
     * Halfpel motion compensation with rounding (a+b+1)>>1.
239
     * this is an array[4][4] of motion compensation functions for 4
240
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
241
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
242
     * @param block destination where the result is stored
243
     * @param pixels source
244
     * @param line_size number of bytes in a horizontal line of block
245
     * @param h height
246
     */
247
    op_pixels_func put_pixels_tab[4][4];
248

    
249
    /**
250
     * Halfpel motion compensation with rounding (a+b+1)>>1.
251
     * This is an array[4][4] of motion compensation functions for 4
252
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
253
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
254
     * @param block destination into which the result is averaged (a+b+1)>>1
255
     * @param pixels source
256
     * @param line_size number of bytes in a horizontal line of block
257
     * @param h height
258
     */
259
    op_pixels_func avg_pixels_tab[4][4];
260

    
261
    /**
262
     * Halfpel motion compensation with no rounding (a+b)>>1.
263
     * this is an array[2][4] of motion compensation functions for 2
264
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
265
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
266
     * @param block destination where the result is stored
267
     * @param pixels source
268
     * @param line_size number of bytes in a horizontal line of block
269
     * @param h height
270
     */
271
    op_pixels_func put_no_rnd_pixels_tab[4][4];
272

    
273
    /**
274
     * Halfpel motion compensation with no rounding (a+b)>>1.
275
     * this is an array[2][4] of motion compensation functions for 2
276
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
277
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
278
     * @param block destination into which the result is averaged (a+b)>>1
279
     * @param pixels source
280
     * @param line_size number of bytes in a horizontal line of block
281
     * @param h height
282
     */
283
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
284

    
285
    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);
286

    
287
    /**
288
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
289
     * this is an array[12] of motion compensation functions for the 9 thirdpe
290
     * positions<br>
291
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
292
     * @param block destination where the result is stored
293
     * @param pixels source
294
     * @param line_size number of bytes in a horizontal line of block
295
     * @param h height
296
     */
297
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
298
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
299

    
300
    qpel_mc_func put_qpel_pixels_tab[2][16];
301
    qpel_mc_func avg_qpel_pixels_tab[2][16];
302
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
303
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
304
    qpel_mc_func put_mspel_pixels_tab[8];
305

    
306
    /**
307
     * h264 Chroma MC
308
     */
309
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
310
    /* This is really one func used in VC-1 decoding */
311
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
312
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
313

    
314
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
315
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
316

    
317
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
318
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
319

    
320
    h264_weight_func weight_h264_pixels_tab[10];
321
    h264_biweight_func biweight_h264_pixels_tab[10];
322

    
323
    /* AVS specific */
324
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
325
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
326
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
327
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
328
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
329
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
330
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
331

    
332
    me_cmp_func pix_abs[2][4];
333

    
334
    /* huffyuv specific */
335
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
336
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
337
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
338
    /**
339
     * subtract huffyuv's variant of median prediction
340
     * note, this might read from src1[-1], src2[-1]
341
     */
342
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
343
    /* this might write to dst[w] */
344
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
345
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
346

    
347
    void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
348
    void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
349
    /* v/h_loop_filter_luma_intra: align 16 */
350
    void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
351
    void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
352
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
353
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
354
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
355
    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],
356
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
357

    
358
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
359
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
360

    
361
    void (*h261_loop_filter)(uint8_t *src, int stride);
362

    
363
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
364
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
365

    
366
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
367
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
368

    
369
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
370
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
371
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
372
    /* no alignment needed */
373
    void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
374
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
375
    void (*vector_fmul)(float *dst, const float *src, int len);
376
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
377
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
378
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
379
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
380
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
381
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
382
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
383

    
384
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
385
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
386
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
387
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
388

    
389
    /* (I)DCT */
390
    void (*fdct)(DCTELEM *block/* align 16*/);
391
    void (*fdct248)(DCTELEM *block/* align 16*/);
392

    
393
    /* IDCT really*/
394
    void (*idct)(DCTELEM *block/* align 16*/);
395

    
396
    /**
397
     * block -> idct -> clip to unsigned 8 bit -> dest.
398
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
399
     * @param line_size size in bytes of a horizontal line of dest
400
     */
401
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
402

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

    
409
    /**
410
     * idct input permutation.
411
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
412
     * IDCT)
413
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
414
     * with the zigzag/alternate scan<br>
415
     * an example to avoid confusion:
416
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
417
     * - (x -> referece dct -> reference idct -> x)
418
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
419
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
420
     */
421
    uint8_t idct_permutation[64];
422
    int idct_permutation_type;
423
#define FF_NO_IDCT_PERM 1
424
#define FF_LIBMPEG2_IDCT_PERM 2
425
#define FF_SIMPLE_IDCT_PERM 3
426
#define FF_TRANSPOSE_IDCT_PERM 4
427
#define FF_PARTTRANS_IDCT_PERM 5
428
#define FF_SSE2_IDCT_PERM 6
429

    
430
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
431
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
432
#define BASIS_SHIFT 16
433
#define RECON_SHIFT 6
434

    
435
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
436
#define EDGE_WIDTH 16
437

    
438
    /* h264 functions */
439
    void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
440
    void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
441
    void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
442
    void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
443
    void (*h264_dct)(DCTELEM block[4][4]);
444

    
445
    /* snow wavelet */
446
    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
447
    void (*horizontal_compose97i)(IDWTELEM *b, int width);
448
    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);
449

    
450
    void (*prefetch)(void *mem, int stride, int h);
451

    
452
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
453

    
454
    /* vc1 functions */
455
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
456
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
457
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
458
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
459
    void (*vc1_v_overlap)(uint8_t* src, int stride);
460
    void (*vc1_h_overlap)(uint8_t* src, int stride);
461
    /* put 8x8 block with bicubic interpolation and quarterpel precision
462
     * last argument is actually round value instead of height
463
     */
464
    op_pixels_func put_vc1_mspel_pixels_tab[16];
465

    
466
    /* intrax8 functions */
467
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
468
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
469
           int * range, int * sum,  int edges);
470

    
471
    /* ape functions */
472
    /**
473
     * Add contents of the second vector to the first one.
474
     * @param len length of vectors, should be multiple of 16
475
     */
476
    void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
477
    /**
478
     * Add contents of the second vector to the first one.
479
     * @param len length of vectors, should be multiple of 16
480
     */
481
    void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
482
    /**
483
     * Calculate scalar product of two vectors.
484
     * @param len length of vectors, should be multiple of 16
485
     * @param shift number of bits to discard from product
486
     */
487
    int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
488

    
489
    /* rv30 functions */
490
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
491
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
492

    
493
    /* rv40 functions */
494
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
495
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
496
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
497
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
498
} DSPContext;
499

    
500
void dsputil_static_init(void);
501
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
502

    
503
int ff_check_alignment(void);
504

    
505
/**
506
 * permute block according to permuatation.
507
 * @param last last non zero element in scantable order
508
 */
509
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
510

    
511
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
512

    
513
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
514

    
515
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
516
{
517
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
518
}
519

    
520
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
521
{
522
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
523
}
524

    
525
static inline int get_penalty_factor(int lambda, int lambda2, int type){
526
    switch(type&0xFF){
527
    default:
528
    case FF_CMP_SAD:
529
        return lambda>>FF_LAMBDA_SHIFT;
530
    case FF_CMP_DCT:
531
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
532
    case FF_CMP_W53:
533
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
534
    case FF_CMP_W97:
535
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
536
    case FF_CMP_SATD:
537
    case FF_CMP_DCT264:
538
        return (2*lambda)>>FF_LAMBDA_SHIFT;
539
    case FF_CMP_RD:
540
    case FF_CMP_PSNR:
541
    case FF_CMP_SSE:
542
    case FF_CMP_NSSE:
543
        return lambda2>>FF_LAMBDA_SHIFT;
544
    case FF_CMP_BIT:
545
        return 1;
546
    }
547
}
548

    
549
/**
550
 * Empty mmx state.
551
 * this must be called between any dsp function and float/double code.
552
 * for example sin(); dsp->idct_put(); emms_c(); cos()
553
 */
554
#define emms_c()
555

    
556
/* should be defined by architectures supporting
557
   one or more MultiMedia extension */
558
int mm_support(void);
559

    
560
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
561
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
562
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
563
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
564
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
565
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
566
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
567
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
568
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
569

    
570
#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
571

    
572
#if defined(HAVE_MMX)
573

    
574
#undef emms_c
575

    
576
extern int mm_flags;
577

    
578
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
579
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
580
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
581

    
582
static inline void emms(void)
583
{
584
    __asm__ volatile ("emms;":::"memory");
585
}
586

    
587

    
588
#define emms_c() \
589
{\
590
    if (mm_flags & FF_MM_MMX)\
591
        emms();\
592
}
593

    
594
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
595

    
596
#elif defined(ARCH_ARM)
597

    
598
extern int mm_flags;
599

    
600
#ifdef HAVE_NEON
601
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
602
#   define STRIDE_ALIGN 16
603
#endif
604

    
605
#elif defined(ARCH_POWERPC)
606

    
607
extern int mm_flags;
608

    
609
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
610
#define STRIDE_ALIGN 16
611

    
612
#elif defined(HAVE_MMI)
613

    
614
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
615
#define STRIDE_ALIGN 16
616

    
617
#else
618

    
619
#define mm_flags 0
620
#define mm_support() 0
621

    
622
#endif
623

    
624
#ifndef DECLARE_ALIGNED_8
625
#   define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
626
#endif
627

    
628
#ifndef STRIDE_ALIGN
629
#   define STRIDE_ALIGN 8
630
#endif
631

    
632
/* PSNR */
633
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
634
              int orig_linesize[3], int coded_linesize,
635
              AVCodecContext *avctx);
636

    
637
/* FFT computation */
638

    
639
/* NOTE: soon integer code will be added, so you must use the
640
   FFTSample type */
641
typedef float FFTSample;
642

    
643
struct MDCTContext;
644

    
645
typedef struct FFTComplex {
646
    FFTSample re, im;
647
} FFTComplex;
648

    
649
typedef struct FFTContext {
650
    int nbits;
651
    int inverse;
652
    uint16_t *revtab;
653
    FFTComplex *exptab;
654
    FFTComplex *exptab1; /* only used by SSE code */
655
    FFTComplex *tmp_buf;
656
    void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
657
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
658
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
659
    void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
660
} FFTContext;
661

    
662
int ff_fft_init(FFTContext *s, int nbits, int inverse);
663
void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
664
void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
665
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
666
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
667
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
668
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
669
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
670

    
671
static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
672
{
673
    s->fft_permute(s, z);
674
}
675
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
676
{
677
    s->fft_calc(s, z);
678
}
679
void ff_fft_end(FFTContext *s);
680

    
681
/* MDCT computation */
682

    
683
typedef struct MDCTContext {
684
    int n;  /* size of MDCT (i.e. number of input data * 2) */
685
    int nbits; /* n = 2^nbits */
686
    /* pre/post rotation tables */
687
    FFTSample *tcos;
688
    FFTSample *tsin;
689
    FFTContext fft;
690
} MDCTContext;
691

    
692
static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
693
{
694
    s->fft.imdct_calc(s, output, input);
695
}
696
static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
697
{
698
    s->fft.imdct_half(s, output, input);
699
}
700

    
701
/**
702
 * Generate a Kaiser-Bessel Derived Window.
703
 * @param   window  pointer to half window
704
 * @param   alpha   determines window shape
705
 * @param   n       size of half window
706
 */
707
void ff_kbd_window_init(float *window, float alpha, int n);
708

    
709
/**
710
 * Generate a sine window.
711
 * @param   window  pointer to half window
712
 * @param   n       size of half window
713
 */
714
void ff_sine_window_init(float *window, int n);
715
extern float ff_sine_128 [ 128];
716
extern float ff_sine_256 [ 256];
717
extern float ff_sine_512 [ 512];
718
extern float ff_sine_1024[1024];
719
extern float ff_sine_2048[2048];
720
extern float *ff_sine_windows[5];
721

    
722
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
723
void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
724
void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
725
void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
726
void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
727
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
728
void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
729
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
730
void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
731
void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
732
void ff_mdct_end(MDCTContext *s);
733

    
734
#define WRAPPER8_16(name8, name16)\
735
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
736
    return name8(s, dst           , src           , stride, h)\
737
          +name8(s, dst+8         , src+8         , stride, h);\
738
}
739

    
740
#define WRAPPER8_16_SQ(name8, name16)\
741
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
742
    int score=0;\
743
    score +=name8(s, dst           , src           , stride, 8);\
744
    score +=name8(s, dst+8         , src+8         , stride, 8);\
745
    if(h==16){\
746
        dst += 8*stride;\
747
        src += 8*stride;\
748
        score +=name8(s, dst           , src           , stride, 8);\
749
        score +=name8(s, dst+8         , src+8         , stride, 8);\
750
    }\
751
    return score;\
752
}
753

    
754

    
755
static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
756
{
757
    int i;
758
    for(i=0; i<h; i++)
759
    {
760
        AV_WN16(dst   , AV_RN16(src   ));
761
        dst+=dstStride;
762
        src+=srcStride;
763
    }
764
}
765

    
766
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
767
{
768
    int i;
769
    for(i=0; i<h; i++)
770
    {
771
        AV_WN32(dst   , AV_RN32(src   ));
772
        dst+=dstStride;
773
        src+=srcStride;
774
    }
775
}
776

    
777
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
778
{
779
    int i;
780
    for(i=0; i<h; i++)
781
    {
782
        AV_WN32(dst   , AV_RN32(src   ));
783
        AV_WN32(dst+4 , AV_RN32(src+4 ));
784
        dst+=dstStride;
785
        src+=srcStride;
786
    }
787
}
788

    
789
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
790
{
791
    int i;
792
    for(i=0; i<h; i++)
793
    {
794
        AV_WN32(dst   , AV_RN32(src   ));
795
        AV_WN32(dst+4 , AV_RN32(src+4 ));
796
        dst[8]= src[8];
797
        dst+=dstStride;
798
        src+=srcStride;
799
    }
800
}
801

    
802
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
803
{
804
    int i;
805
    for(i=0; i<h; i++)
806
    {
807
        AV_WN32(dst   , AV_RN32(src   ));
808
        AV_WN32(dst+4 , AV_RN32(src+4 ));
809
        AV_WN32(dst+8 , AV_RN32(src+8 ));
810
        AV_WN32(dst+12, AV_RN32(src+12));
811
        dst+=dstStride;
812
        src+=srcStride;
813
    }
814
}
815

    
816
static inline void copy_block17(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[16]= src[16];
826
        dst+=dstStride;
827
        src+=srcStride;
828
    }
829
}
830

    
831
#endif /* AVCODEC_DSPUTIL_H */