Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 3a088514

History | View | Annotate | Download (31.2 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
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

    
41
void fdct_ifast (DCTELEM *data);
42
void fdct_ifast248 (DCTELEM *data);
43
void ff_jpeg_fdct_islow (DCTELEM *data);
44
void ff_fdct248_islow (DCTELEM *data);
45

    
46
void j_rev_dct (DCTELEM *data);
47
void j_rev_dct4 (DCTELEM *data);
48
void j_rev_dct2 (DCTELEM *data);
49
void j_rev_dct1 (DCTELEM *data);
50
void ff_wmv2_idct_c(DCTELEM *data);
51

    
52
void ff_fdct_mmx(DCTELEM *block);
53
void ff_fdct_mmx2(DCTELEM *block);
54
void ff_fdct_sse2(DCTELEM *block);
55

    
56
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
57
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
58
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
59
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
60
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
61
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
62
void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
63
void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
64
void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
65
void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
66

    
67
void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
68
                             const float *win, float add_bias, int len);
69
void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
70
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
71

    
72
/* encoding scans */
73
extern const uint8_t ff_alternate_horizontal_scan[64];
74
extern const uint8_t ff_alternate_vertical_scan[64];
75
extern const uint8_t ff_zigzag_direct[64];
76
extern const uint8_t ff_zigzag248_direct[64];
77

    
78
/* pixel operations */
79
#define MAX_NEG_CROP 1024
80

    
81
/* temporary */
82
extern uint32_t ff_squareTbl[512];
83
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
84

    
85
/* VP3 DSP functions */
86
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
87
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
88
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
89
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
90

    
91
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
92
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
93

    
94
/* Bink functions */
95
void ff_bink_idct_c    (DCTELEM *block);
96
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
97
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
98

    
99
/* CAVS functions */
100
void ff_put_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride);
101
void ff_avg_cavs_qpel8_mc00_c(uint8_t *dst, uint8_t *src, int stride);
102
void ff_put_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride);
103
void ff_avg_cavs_qpel16_mc00_c(uint8_t *dst, uint8_t *src, int stride);
104

    
105
/* VC1 functions */
106
void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
107
void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
108

    
109
/* EA functions */
110
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
111

    
112
/* 1/2^n downscaling functions from imgconvert.c */
113
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
114
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
115
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
116
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
117

    
118
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
119
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
120

    
121
/* minimum alignment rules ;)
122
If you notice errors in the align stuff, need more alignment for some ASM code
123
for some CPU or need to use a function with less aligned data then send a mail
124
to the ffmpeg-devel mailing list, ...
125

126
!warning These alignments might not match reality, (missing attribute((align))
127
stuff somewhere possible).
128
I (Michael) did not check them, these are just the alignments which I think
129
could be reached easily ...
130

131
!future video codecs might need functions with less strict alignment
132
*/
133

    
134
/*
135
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
136
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
137
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
138
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
139
void clear_blocks_c(DCTELEM *blocks);
140
*/
141

    
142
/* add and put pixel (decoding) */
143
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
144
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
145
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
146
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);
147
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
148
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
149

    
150
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
151

    
152
#define DEF_OLD_QPEL(name)\
153
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
154
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
155
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
156

    
157
DEF_OLD_QPEL(qpel16_mc11_old_c)
158
DEF_OLD_QPEL(qpel16_mc31_old_c)
159
DEF_OLD_QPEL(qpel16_mc12_old_c)
160
DEF_OLD_QPEL(qpel16_mc32_old_c)
161
DEF_OLD_QPEL(qpel16_mc13_old_c)
162
DEF_OLD_QPEL(qpel16_mc33_old_c)
163
DEF_OLD_QPEL(qpel8_mc11_old_c)
164
DEF_OLD_QPEL(qpel8_mc31_old_c)
165
DEF_OLD_QPEL(qpel8_mc12_old_c)
166
DEF_OLD_QPEL(qpel8_mc32_old_c)
167
DEF_OLD_QPEL(qpel8_mc13_old_c)
168
DEF_OLD_QPEL(qpel8_mc33_old_c)
169

    
170
#define CALL_2X_PIXELS(a, b, n)\
171
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
172
    b(block  , pixels  , line_size, h);\
173
    b(block+n, pixels+n, line_size, h);\
174
}
175

    
176
/* motion estimation */
177
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
178
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
179
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))*/;
180

    
181
/**
182
 * Scantable.
183
 */
184
typedef struct ScanTable{
185
    const uint8_t *scantable;
186
    uint8_t permutated[64];
187
    uint8_t raster_end[64];
188
#if ARCH_PPC
189
                /** Used by dct_quantize_altivec to find last-non-zero */
190
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
191
#endif
192
} ScanTable;
193

    
194
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
195

    
196
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
197
                         int block_w, int block_h,
198
                         int src_x, int src_y, int w, int h);
199

    
200
/**
201
 * DSPContext.
202
 */
203
typedef struct DSPContext {
204
    /* pixel ops : interface with DCT */
205
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
206
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
207
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
208
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
209
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
210
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
211
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
212
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
213
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
214
    /**
215
     * translational global motion compensation.
216
     */
217
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
218
    /**
219
     * global motion compensation.
220
     */
221
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
222
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
223
    void (*clear_block)(DCTELEM *block/*align 16*/);
224
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
225
    int (*pix_sum)(uint8_t * pix, int line_size);
226
    int (*pix_norm1)(uint8_t * pix, int line_size);
227
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
228

    
229
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
230
    me_cmp_func sse[6];
231
    me_cmp_func hadamard8_diff[6];
232
    me_cmp_func dct_sad[6];
233
    me_cmp_func quant_psnr[6];
234
    me_cmp_func bit[6];
235
    me_cmp_func rd[6];
236
    me_cmp_func vsad[6];
237
    me_cmp_func vsse[6];
238
    me_cmp_func nsse[6];
239
    me_cmp_func w53[6];
240
    me_cmp_func w97[6];
241
    me_cmp_func dct_max[6];
242
    me_cmp_func dct264_sad[6];
243

    
244
    me_cmp_func me_pre_cmp[6];
245
    me_cmp_func me_cmp[6];
246
    me_cmp_func me_sub_cmp[6];
247
    me_cmp_func mb_cmp[6];
248
    me_cmp_func ildct_cmp[6]; //only width 16 used
249
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
250

    
251
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
252
                             int size);
253

    
254
    /**
255
     * Halfpel motion compensation with rounding (a+b+1)>>1.
256
     * this is an array[4][4] of motion compensation functions for 4
257
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
258
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
259
     * @param block destination where the result is stored
260
     * @param pixels source
261
     * @param line_size number of bytes in a horizontal line of block
262
     * @param h height
263
     */
264
    op_pixels_func put_pixels_tab[4][4];
265

    
266
    /**
267
     * Halfpel motion compensation with rounding (a+b+1)>>1.
268
     * This is an array[4][4] of motion compensation functions for 4
269
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
270
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
271
     * @param block destination into which the result is averaged (a+b+1)>>1
272
     * @param pixels source
273
     * @param line_size number of bytes in a horizontal line of block
274
     * @param h height
275
     */
276
    op_pixels_func avg_pixels_tab[4][4];
277

    
278
    /**
279
     * Halfpel motion compensation with no rounding (a+b)>>1.
280
     * this is an array[2][4] of motion compensation functions for 2
281
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
282
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
283
     * @param block destination where the result is stored
284
     * @param pixels source
285
     * @param line_size number of bytes in a horizontal line of block
286
     * @param h height
287
     */
288
    op_pixels_func put_no_rnd_pixels_tab[4][4];
289

    
290
    /**
291
     * Halfpel motion compensation with no rounding (a+b)>>1.
292
     * this is an array[2][4] of motion compensation functions for 2
293
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
294
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
295
     * @param block destination into which the result is averaged (a+b)>>1
296
     * @param pixels source
297
     * @param line_size number of bytes in a horizontal line of block
298
     * @param h height
299
     */
300
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
301

    
302
    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);
303

    
304
    /**
305
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
306
     * this is an array[12] of motion compensation functions for the 9 thirdpe
307
     * positions<br>
308
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
309
     * @param block destination where the result is stored
310
     * @param pixels source
311
     * @param line_size number of bytes in a horizontal line of block
312
     * @param h height
313
     */
314
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
315
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
316

    
317
    qpel_mc_func put_qpel_pixels_tab[2][16];
318
    qpel_mc_func avg_qpel_pixels_tab[2][16];
319
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
320
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
321
    qpel_mc_func put_mspel_pixels_tab[8];
322

    
323
    /**
324
     * h264 Chroma MC
325
     */
326
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
327
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
328
    /* This is really one func used in VC-1 decoding */
329
    h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
330
    h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
331

    
332
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
333
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
334

    
335
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
336
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
337

    
338
    me_cmp_func pix_abs[2][4];
339

    
340
    /* huffyuv specific */
341
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
342
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
343
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
344
    /**
345
     * subtract huffyuv's variant of median prediction
346
     * note, this might read from src1[-1], src2[-1]
347
     */
348
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
349
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
350
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
351
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
352
    /* this might write to dst[w] */
353
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
354
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
355

    
356
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
357
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
358

    
359
    void (*h261_loop_filter)(uint8_t *src, int stride);
360

    
361
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
362
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
363

    
364
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
365
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
366
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
367

    
368
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
369
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
370
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
371
    /* no alignment needed */
372
    void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
373
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
374
    void (*vector_fmul)(float *dst, const float *src, int len);
375
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
376
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
377
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
378
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
379
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
380
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
381
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
382
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
383
    /**
384
     * Multiply a vector of floats by a scalar float.  Source and
385
     * destination vectors must overlap exactly or not at all.
386
     * @param dst result vector, 16-byte aligned
387
     * @param src input vector, 16-byte aligned
388
     * @param mul scalar value
389
     * @param len length of vector, multiple of 4
390
     */
391
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
392
                               int len);
393
    /**
394
     * Multiply a vector of floats by concatenated short vectors of
395
     * floats and by a scalar float.  Source and destination vectors
396
     * must overlap exactly or not at all.
397
     * [0]: short vectors of length 2, 8-byte aligned
398
     * [1]: short vectors of length 4, 16-byte aligned
399
     * @param dst output vector, 16-byte aligned
400
     * @param src input vector, 16-byte aligned
401
     * @param sv  array of pointers to short vectors
402
     * @param mul scalar value
403
     * @param len number of elements in src and dst, multiple of 4
404
     */
405
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
406
                                     const float **sv, float mul, int len);
407
    /**
408
     * Multiply short vectors of floats by a scalar float, store
409
     * concatenated result.
410
     * [0]: short vectors of length 2, 8-byte aligned
411
     * [1]: short vectors of length 4, 16-byte aligned
412
     * @param dst output vector, 16-byte aligned
413
     * @param sv  array of pointers to short vectors
414
     * @param mul scalar value
415
     * @param len number of output elements, multiple of 4
416
     */
417
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
418
                              float mul, int len);
419
    /**
420
     * Calculate the scalar product of two vectors of floats.
421
     * @param v1  first vector, 16-byte aligned
422
     * @param v2  second vector, 16-byte aligned
423
     * @param len length of vectors, multiple of 4
424
     */
425
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
426
    /**
427
     * Calculate the sum and difference of two vectors of floats.
428
     * @param v1  first input vector, sum output, 16-byte aligned
429
     * @param v2  second input vector, difference output, 16-byte aligned
430
     * @param len length of vectors, multiple of 4
431
     */
432
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
433

    
434
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
435
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
436
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
437
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
438

    
439
    /* (I)DCT */
440
    void (*fdct)(DCTELEM *block/* align 16*/);
441
    void (*fdct248)(DCTELEM *block/* align 16*/);
442

    
443
    /* IDCT really*/
444
    void (*idct)(DCTELEM *block/* align 16*/);
445

    
446
    /**
447
     * block -> idct -> clip to unsigned 8 bit -> dest.
448
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
449
     * @param line_size size in bytes of a horizontal line of dest
450
     */
451
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
452

    
453
    /**
454
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
455
     * @param line_size size in bytes of a horizontal line of dest
456
     */
457
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
458

    
459
    /**
460
     * idct input permutation.
461
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
462
     * IDCT)
463
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
464
     * with the zigzag/alternate scan<br>
465
     * an example to avoid confusion:
466
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
467
     * - (x -> referece dct -> reference idct -> x)
468
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
469
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
470
     */
471
    uint8_t idct_permutation[64];
472
    int idct_permutation_type;
473
#define FF_NO_IDCT_PERM 1
474
#define FF_LIBMPEG2_IDCT_PERM 2
475
#define FF_SIMPLE_IDCT_PERM 3
476
#define FF_TRANSPOSE_IDCT_PERM 4
477
#define FF_PARTTRANS_IDCT_PERM 5
478
#define FF_SSE2_IDCT_PERM 6
479

    
480
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
481
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
482
#define BASIS_SHIFT 16
483
#define RECON_SHIFT 6
484

    
485
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
486
#define EDGE_WIDTH 16
487

    
488
    void (*prefetch)(void *mem, int stride, int h);
489

    
490
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
491

    
492
    /* mlp/truehd functions */
493
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
494
                               int firorder, int iirorder,
495
                               unsigned int filter_shift, int32_t mask, int blocksize,
496
                               int32_t *sample_buffer);
497

    
498
    /* vc1 functions */
499
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
500
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
501
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
502
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
503
    void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
504
    void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
505
    void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
506
    void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
507
    void (*vc1_v_overlap)(uint8_t* src, int stride);
508
    void (*vc1_h_overlap)(uint8_t* src, int stride);
509
    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
510
    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
511
    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
512
    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
513
    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
514
    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
515
    /* put 8x8 block with bicubic interpolation and quarterpel precision
516
     * last argument is actually round value instead of height
517
     */
518
    op_pixels_func put_vc1_mspel_pixels_tab[16];
519
    op_pixels_func avg_vc1_mspel_pixels_tab[16];
520

    
521
    /* intrax8 functions */
522
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
523
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
524
           int * range, int * sum,  int edges);
525

    
526
    /**
527
     * Calculate scalar product of two vectors.
528
     * @param len length of vectors, should be multiple of 16
529
     * @param shift number of bits to discard from product
530
     */
531
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
532
    /* ape functions */
533
    /**
534
     * Calculate scalar product of v1 and v2,
535
     * and v1[i] += v3[i] * mul
536
     * @param len length of vectors, should be multiple of 16
537
     */
538
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
539

    
540
    /* rv30 functions */
541
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
542
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
543

    
544
    /* rv40 functions */
545
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
546
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
547
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
548
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
549

    
550
    /* bink functions */
551
    op_fill_func fill_block_tab[2];
552
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
553
} DSPContext;
554

    
555
void dsputil_static_init(void);
556
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
557

    
558
int ff_check_alignment(void);
559

    
560
/**
561
 * permute block according to permuatation.
562
 * @param last last non zero element in scantable order
563
 */
564
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
565

    
566
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
567

    
568
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
569

    
570
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
571
{
572
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
573
}
574

    
575
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
576
{
577
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
578
}
579

    
580
static inline int get_penalty_factor(int lambda, int lambda2, int type){
581
    switch(type&0xFF){
582
    default:
583
    case FF_CMP_SAD:
584
        return lambda>>FF_LAMBDA_SHIFT;
585
    case FF_CMP_DCT:
586
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
587
    case FF_CMP_W53:
588
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
589
    case FF_CMP_W97:
590
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
591
    case FF_CMP_SATD:
592
    case FF_CMP_DCT264:
593
        return (2*lambda)>>FF_LAMBDA_SHIFT;
594
    case FF_CMP_RD:
595
    case FF_CMP_PSNR:
596
    case FF_CMP_SSE:
597
    case FF_CMP_NSSE:
598
        return lambda2>>FF_LAMBDA_SHIFT;
599
    case FF_CMP_BIT:
600
        return 1;
601
    }
602
}
603

    
604
/**
605
 * Empty mmx state.
606
 * this must be called between any dsp function and float/double code.
607
 * for example sin(); dsp->idct_put(); emms_c(); cos()
608
 */
609
#define emms_c()
610

    
611
/* should be defined by architectures supporting
612
   one or more MultiMedia extension */
613
int mm_support(void);
614

    
615
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
616
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
617
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
618
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
619
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
620
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
621
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
622
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
623
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
624

    
625
void ff_dsputil_init_dwt(DSPContext *c);
626
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
627
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
628
void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
629
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
630
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
631
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
632

    
633
#if HAVE_MMX
634

    
635
#undef emms_c
636

    
637
static inline void emms(void)
638
{
639
    __asm__ volatile ("emms;":::"memory");
640
}
641

    
642
#define emms_c() emms()
643

    
644
#elif ARCH_ARM
645

    
646
#if HAVE_NEON
647
#   define STRIDE_ALIGN 16
648
#endif
649

    
650
#elif ARCH_PPC
651

    
652
#define STRIDE_ALIGN 16
653

    
654
#elif HAVE_MMI
655

    
656
#define STRIDE_ALIGN 16
657

    
658
#else
659

    
660
#define mm_support() 0
661

    
662
#endif
663

    
664
#ifndef STRIDE_ALIGN
665
#   define STRIDE_ALIGN 8
666
#endif
667

    
668
#define LOCAL_ALIGNED(a, t, v, s, ...)                          \
669
    uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)];              \
670
    t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
671

    
672
#if HAVE_LOCAL_ALIGNED_8
673
#   define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
674
#else
675
#   define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
676
#endif
677

    
678
#if HAVE_LOCAL_ALIGNED_16
679
#   define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
680
#else
681
#   define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
682
#endif
683

    
684
/* PSNR */
685
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
686
              int orig_linesize[3], int coded_linesize,
687
              AVCodecContext *avctx);
688

    
689
#define WRAPPER8_16(name8, name16)\
690
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
691
    return name8(s, dst           , src           , stride, h)\
692
          +name8(s, dst+8         , src+8         , stride, h);\
693
}
694

    
695
#define WRAPPER8_16_SQ(name8, name16)\
696
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
697
    int score=0;\
698
    score +=name8(s, dst           , src           , stride, 8);\
699
    score +=name8(s, dst+8         , src+8         , stride, 8);\
700
    if(h==16){\
701
        dst += 8*stride;\
702
        src += 8*stride;\
703
        score +=name8(s, dst           , src           , stride, 8);\
704
        score +=name8(s, dst+8         , src+8         , stride, 8);\
705
    }\
706
    return score;\
707
}
708

    
709

    
710
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
711
{
712
    int i;
713
    for(i=0; i<h; i++)
714
    {
715
        AV_WN16(dst   , AV_RN16(src   ));
716
        dst+=dstStride;
717
        src+=srcStride;
718
    }
719
}
720

    
721
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
722
{
723
    int i;
724
    for(i=0; i<h; i++)
725
    {
726
        AV_WN32(dst   , AV_RN32(src   ));
727
        dst+=dstStride;
728
        src+=srcStride;
729
    }
730
}
731

    
732
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
733
{
734
    int i;
735
    for(i=0; i<h; i++)
736
    {
737
        AV_WN32(dst   , AV_RN32(src   ));
738
        AV_WN32(dst+4 , AV_RN32(src+4 ));
739
        dst+=dstStride;
740
        src+=srcStride;
741
    }
742
}
743

    
744
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
745
{
746
    int i;
747
    for(i=0; i<h; i++)
748
    {
749
        AV_WN32(dst   , AV_RN32(src   ));
750
        AV_WN32(dst+4 , AV_RN32(src+4 ));
751
        dst[8]= src[8];
752
        dst+=dstStride;
753
        src+=srcStride;
754
    }
755
}
756

    
757
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
758
{
759
    int i;
760
    for(i=0; i<h; i++)
761
    {
762
        AV_WN32(dst   , AV_RN32(src   ));
763
        AV_WN32(dst+4 , AV_RN32(src+4 ));
764
        AV_WN32(dst+8 , AV_RN32(src+8 ));
765
        AV_WN32(dst+12, AV_RN32(src+12));
766
        dst+=dstStride;
767
        src+=srcStride;
768
    }
769
}
770

    
771
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
772
{
773
    int i;
774
    for(i=0; i<h; i++)
775
    {
776
        AV_WN32(dst   , AV_RN32(src   ));
777
        AV_WN32(dst+4 , AV_RN32(src+4 ));
778
        AV_WN32(dst+8 , AV_RN32(src+8 ));
779
        AV_WN32(dst+12, AV_RN32(src+12));
780
        dst[16]= src[16];
781
        dst+=dstStride;
782
        src+=srcStride;
783
    }
784
}
785

    
786
#endif /* AVCODEC_DSPUTIL_H */