Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 6d4c49a2

History | View | Annotate | Download (31.3 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
#define H264_IDCT(depth) \
57
void ff_h264_idct8_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
58
void ff_h264_idct_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
59
void ff_h264_idct8_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
60
void ff_h264_idct_dc_add_ ## depth ## _c(uint8_t *dst, DCTELEM *block, int stride);\
61
void ff_h264_lowres_idct_add_ ## depth ## _c(uint8_t *dst, int stride, DCTELEM *block);\
62
void ff_h264_lowres_idct_put_ ## depth ## _c(uint8_t *dst, int stride, DCTELEM *block);\
63
void ff_h264_idct_add16_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
64
void ff_h264_idct_add16intra_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
65
void ff_h264_idct8_add4_ ## depth ## _c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
66
void ff_h264_idct_add8_ ## depth ## _c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);\
67
void ff_h264_luma_dc_dequant_idct_ ## depth ## _c(DCTELEM *output, DCTELEM *input, int qmul);\
68
void ff_h264_chroma_dc_dequant_idct_ ## depth ## _c(DCTELEM *block, int qmul);
69

    
70
H264_IDCT( 8)
71
H264_IDCT( 9)
72
H264_IDCT(10)
73

    
74
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
75
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
76

    
77
/* encoding scans */
78
extern const uint8_t ff_alternate_horizontal_scan[64];
79
extern const uint8_t ff_alternate_vertical_scan[64];
80
extern const uint8_t ff_zigzag_direct[64];
81
extern const uint8_t ff_zigzag248_direct[64];
82

    
83
/* pixel operations */
84
#define MAX_NEG_CROP 1024
85

    
86
/* temporary */
87
extern uint32_t ff_squareTbl[512];
88
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
89

    
90
#define PUTAVG_PIXELS(depth)\
91
void ff_put_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
92
void ff_avg_pixels8x8_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
93
void ff_put_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);\
94
void ff_avg_pixels16x16_ ## depth ## _c(uint8_t *dst, uint8_t *src, int stride);
95

    
96
PUTAVG_PIXELS( 8)
97
PUTAVG_PIXELS( 9)
98
PUTAVG_PIXELS(10)
99

    
100
#define ff_put_pixels8x8_c ff_put_pixels8x8_8_c
101
#define ff_avg_pixels8x8_c ff_avg_pixels8x8_8_c
102
#define ff_put_pixels16x16_c ff_put_pixels16x16_8_c
103
#define ff_avg_pixels16x16_c ff_avg_pixels16x16_8_c
104

    
105
/* VP3 DSP functions */
106
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
107
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
108
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
109
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
110

    
111
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
112
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
113

    
114
/* Bink functions */
115
void ff_bink_idct_c    (DCTELEM *block);
116
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
117
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
118

    
119
/* EA functions */
120
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
121

    
122
/* 1/2^n downscaling functions from imgconvert.c */
123
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
124
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
125
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
126

    
127
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
128
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
129

    
130
/* minimum alignment rules ;)
131
If you notice errors in the align stuff, need more alignment for some ASM code
132
for some CPU or need to use a function with less aligned data then send a mail
133
to the ffmpeg-devel mailing list, ...
134

135
!warning These alignments might not match reality, (missing attribute((align))
136
stuff somewhere possible).
137
I (Michael) did not check them, these are just the alignments which I think
138
could be reached easily ...
139

140
!future video codecs might need functions with less strict alignment
141
*/
142

    
143
/*
144
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
145
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
146
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
147
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
148
void clear_blocks_c(DCTELEM *blocks);
149
*/
150

    
151
/* add and put pixel (decoding) */
152
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
153
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
154
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
155
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);
156
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
157
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
158

    
159
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
160

    
161
#define DEF_OLD_QPEL(name)\
162
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
163
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
164
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
165

    
166
DEF_OLD_QPEL(qpel16_mc11_old_c)
167
DEF_OLD_QPEL(qpel16_mc31_old_c)
168
DEF_OLD_QPEL(qpel16_mc12_old_c)
169
DEF_OLD_QPEL(qpel16_mc32_old_c)
170
DEF_OLD_QPEL(qpel16_mc13_old_c)
171
DEF_OLD_QPEL(qpel16_mc33_old_c)
172
DEF_OLD_QPEL(qpel8_mc11_old_c)
173
DEF_OLD_QPEL(qpel8_mc31_old_c)
174
DEF_OLD_QPEL(qpel8_mc12_old_c)
175
DEF_OLD_QPEL(qpel8_mc32_old_c)
176
DEF_OLD_QPEL(qpel8_mc13_old_c)
177
DEF_OLD_QPEL(qpel8_mc33_old_c)
178

    
179
#define CALL_2X_PIXELS(a, b, n)\
180
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
181
    b(block  , pixels  , line_size, h);\
182
    b(block+n, pixels+n, line_size, h);\
183
}
184

    
185
/* motion estimation */
186
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
187
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
188
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))*/;
189

    
190
/**
191
 * Scantable.
192
 */
193
typedef struct ScanTable{
194
    const uint8_t *scantable;
195
    uint8_t permutated[64];
196
    uint8_t raster_end[64];
197
#if ARCH_PPC
198
                /** Used by dct_quantize_altivec to find last-non-zero */
199
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
200
#endif
201
} ScanTable;
202

    
203
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
204

    
205
#define EMULATED_EDGE(depth) \
206
void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
207
                         int block_w, int block_h,\
208
                         int src_x, int src_y, int w, int h);
209

    
210
EMULATED_EDGE(8)
211
EMULATED_EDGE(9)
212
EMULATED_EDGE(10)
213

    
214
#define ff_emulated_edge_mc ff_emulated_edge_mc_8
215

    
216
void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
217
void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
218
void ff_put_signed_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
219

    
220
/**
221
 * DSPContext.
222
 */
223
typedef struct DSPContext {
224
    /* pixel ops : interface with DCT */
225
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
226
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
227
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
228
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
229
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
230
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
231
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
232
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
233
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
234
    /**
235
     * Motion estimation with emulated edge values.
236
     * @param buf pointer to destination buffer (unaligned)
237
     * @param src pointer to pixel source (unaligned)
238
     * @param linesize width (in pixels) for src/buf
239
     * @param block_w number of pixels (per row) to copy to buf
240
     * @param block_h nummber of pixel rows to copy to buf
241
     * @param src_x offset of src to start of row - this may be negative
242
     * @param src_y offset of src to top of image - this may be negative
243
     * @param w width of src in pixels
244
     * @param h height of src in pixels
245
     */
246
    void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
247
                             int block_w, int block_h,
248
                             int src_x, int src_y, int w, int h);
249
    /**
250
     * translational global motion compensation.
251
     */
252
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
253
    /**
254
     * global motion compensation.
255
     */
256
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
257
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
258
    void (*clear_block)(DCTELEM *block/*align 16*/);
259
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
260
    int (*pix_sum)(uint8_t * pix, int line_size);
261
    int (*pix_norm1)(uint8_t * pix, int line_size);
262
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
263

    
264
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
265
    me_cmp_func sse[6];
266
    me_cmp_func hadamard8_diff[6];
267
    me_cmp_func dct_sad[6];
268
    me_cmp_func quant_psnr[6];
269
    me_cmp_func bit[6];
270
    me_cmp_func rd[6];
271
    me_cmp_func vsad[6];
272
    me_cmp_func vsse[6];
273
    me_cmp_func nsse[6];
274
    me_cmp_func w53[6];
275
    me_cmp_func w97[6];
276
    me_cmp_func dct_max[6];
277
    me_cmp_func dct264_sad[6];
278

    
279
    me_cmp_func me_pre_cmp[6];
280
    me_cmp_func me_cmp[6];
281
    me_cmp_func me_sub_cmp[6];
282
    me_cmp_func mb_cmp[6];
283
    me_cmp_func ildct_cmp[6]; //only width 16 used
284
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
285

    
286
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
287
                             int size);
288

    
289
    /**
290
     * Halfpel motion compensation with rounding (a+b+1)>>1.
291
     * this is an array[4][4] of motion compensation functions for 4
292
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
293
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
294
     * @param block destination where the result is stored
295
     * @param pixels source
296
     * @param line_size number of bytes in a horizontal line of block
297
     * @param h height
298
     */
299
    op_pixels_func put_pixels_tab[4][4];
300

    
301
    /**
302
     * Halfpel motion compensation with rounding (a+b+1)>>1.
303
     * This is an array[4][4] of motion compensation functions for 4
304
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
305
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
306
     * @param block destination into which the result is averaged (a+b+1)>>1
307
     * @param pixels source
308
     * @param line_size number of bytes in a horizontal line of block
309
     * @param h height
310
     */
311
    op_pixels_func avg_pixels_tab[4][4];
312

    
313
    /**
314
     * Halfpel motion compensation with no rounding (a+b)>>1.
315
     * this is an array[2][4] of motion compensation functions for 2
316
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
317
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
318
     * @param block destination where the result is stored
319
     * @param pixels source
320
     * @param line_size number of bytes in a horizontal line of block
321
     * @param h height
322
     */
323
    op_pixels_func put_no_rnd_pixels_tab[4][4];
324

    
325
    /**
326
     * Halfpel motion compensation with no rounding (a+b)>>1.
327
     * this is an array[2][4] of motion compensation functions for 2
328
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
329
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
330
     * @param block destination into which the result is averaged (a+b)>>1
331
     * @param pixels source
332
     * @param line_size number of bytes in a horizontal line of block
333
     * @param h height
334
     */
335
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
336

    
337
    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);
338

    
339
    /**
340
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
341
     * this is an array[12] of motion compensation functions for the 9 thirdpe
342
     * positions<br>
343
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
344
     * @param block destination where the result is stored
345
     * @param pixels source
346
     * @param line_size number of bytes in a horizontal line of block
347
     * @param h height
348
     */
349
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
350
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
351

    
352
    qpel_mc_func put_qpel_pixels_tab[2][16];
353
    qpel_mc_func avg_qpel_pixels_tab[2][16];
354
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
355
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
356
    qpel_mc_func put_mspel_pixels_tab[8];
357

    
358
    /**
359
     * h264 Chroma MC
360
     */
361
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
362
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
363

    
364
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
365
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
366

    
367
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
368
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
369

    
370
    me_cmp_func pix_abs[2][4];
371

    
372
    /* huffyuv specific */
373
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
374
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
375
    /**
376
     * subtract huffyuv's variant of median prediction
377
     * note, this might read from src1[-1], src2[-1]
378
     */
379
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
380
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
381
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
382
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
383
    /* this might write to dst[w] */
384
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
385
    void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
386

    
387
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
388
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
389

    
390
    void (*h261_loop_filter)(uint8_t *src, int stride);
391

    
392
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
393
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
394

    
395
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
396
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
397
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
398

    
399
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
400
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
401
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
402
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
403
    void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
404
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
405
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
406
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
407
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
408
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
409
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
410
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
411
    /**
412
     * Multiply a vector of floats by a scalar float.  Source and
413
     * destination vectors must overlap exactly or not at all.
414
     * @param dst result vector, 16-byte aligned
415
     * @param src input vector, 16-byte aligned
416
     * @param mul scalar value
417
     * @param len length of vector, multiple of 4
418
     */
419
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
420
                               int len);
421
    /**
422
     * Multiply a vector of floats by concatenated short vectors of
423
     * floats and by a scalar float.  Source and destination vectors
424
     * must overlap exactly or not at all.
425
     * [0]: short vectors of length 2, 8-byte aligned
426
     * [1]: short vectors of length 4, 16-byte aligned
427
     * @param dst output vector, 16-byte aligned
428
     * @param src input vector, 16-byte aligned
429
     * @param sv  array of pointers to short vectors
430
     * @param mul scalar value
431
     * @param len number of elements in src and dst, multiple of 4
432
     */
433
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
434
                                     const float **sv, float mul, int len);
435
    /**
436
     * Multiply short vectors of floats by a scalar float, store
437
     * concatenated result.
438
     * [0]: short vectors of length 2, 8-byte aligned
439
     * [1]: short vectors of length 4, 16-byte aligned
440
     * @param dst output vector, 16-byte aligned
441
     * @param sv  array of pointers to short vectors
442
     * @param mul scalar value
443
     * @param len number of output elements, multiple of 4
444
     */
445
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
446
                              float mul, int len);
447
    /**
448
     * Calculate the scalar product of two vectors of floats.
449
     * @param v1  first vector, 16-byte aligned
450
     * @param v2  second vector, 16-byte aligned
451
     * @param len length of vectors, multiple of 4
452
     */
453
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
454
    /**
455
     * Calculate the sum and difference of two vectors of floats.
456
     * @param v1  first input vector, sum output, 16-byte aligned
457
     * @param v2  second input vector, difference output, 16-byte aligned
458
     * @param len length of vectors, multiple of 4
459
     */
460
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
461

    
462
    /* (I)DCT */
463
    void (*fdct)(DCTELEM *block/* align 16*/);
464
    void (*fdct248)(DCTELEM *block/* align 16*/);
465

    
466
    /* IDCT really*/
467
    void (*idct)(DCTELEM *block/* align 16*/);
468

    
469
    /**
470
     * block -> idct -> clip to unsigned 8 bit -> dest.
471
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
472
     * @param line_size size in bytes of a horizontal line of dest
473
     */
474
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
475

    
476
    /**
477
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
478
     * @param line_size size in bytes of a horizontal line of dest
479
     */
480
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
481

    
482
    /**
483
     * idct input permutation.
484
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
485
     * IDCT)
486
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
487
     * with the zigzag/alternate scan<br>
488
     * an example to avoid confusion:
489
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
490
     * - (x -> referece dct -> reference idct -> x)
491
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
492
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
493
     */
494
    uint8_t idct_permutation[64];
495
    int idct_permutation_type;
496
#define FF_NO_IDCT_PERM 1
497
#define FF_LIBMPEG2_IDCT_PERM 2
498
#define FF_SIMPLE_IDCT_PERM 3
499
#define FF_TRANSPOSE_IDCT_PERM 4
500
#define FF_PARTTRANS_IDCT_PERM 5
501
#define FF_SSE2_IDCT_PERM 6
502

    
503
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
504
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
505
#define BASIS_SHIFT 16
506
#define RECON_SHIFT 6
507

    
508
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int sides);
509
#define EDGE_WIDTH 16
510
#define EDGE_TOP    1
511
#define EDGE_BOTTOM 2
512

    
513
    void (*prefetch)(void *mem, int stride, int h);
514

    
515
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
516

    
517
    /* mlp/truehd functions */
518
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
519
                               int firorder, int iirorder,
520
                               unsigned int filter_shift, int32_t mask, int blocksize,
521
                               int32_t *sample_buffer);
522

    
523
    /* intrax8 functions */
524
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
525
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
526
           int * range, int * sum,  int edges);
527

    
528
    /**
529
     * Calculate scalar product of two vectors.
530
     * @param len length of vectors, should be multiple of 16
531
     * @param shift number of bits to discard from product
532
     */
533
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
534
    /* ape functions */
535
    /**
536
     * Calculate scalar product of v1 and v2,
537
     * and v1[i] += v3[i] * mul
538
     * @param len length of vectors, should be multiple of 16
539
     */
540
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
541

    
542
    /**
543
     * Apply symmetric window in 16-bit fixed-point.
544
     * @param output destination array
545
     *               constraints: 16-byte aligned
546
     * @param input  source array
547
     *               constraints: 16-byte aligned
548
     * @param window window array
549
     *               constraints: 16-byte aligned, at least len/2 elements
550
     * @param len    full window length
551
     *               constraints: multiple of ? greater than zero
552
     */
553
    void (*apply_window_int16)(int16_t *output, const int16_t *input,
554
                               const int16_t *window, unsigned int len);
555

    
556
    /* rv30 functions */
557
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
558
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
559

    
560
    /* rv40 functions */
561
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
562
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
563
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
564
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
565

    
566
    /* bink functions */
567
    op_fill_func fill_block_tab[2];
568
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
569
} DSPContext;
570

    
571
void dsputil_static_init(void);
572
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
573

    
574
int ff_check_alignment(void);
575

    
576
/**
577
 * permute block according to permuatation.
578
 * @param last last non zero element in scantable order
579
 */
580
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
581

    
582
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
583

    
584
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
585
#define         BYTE_VEC64(c)   ((c)*0x0001000100010001UL)
586

    
587
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
588
{
589
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
590
}
591

    
592
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
593
{
594
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
595
}
596

    
597
static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
598
{
599
    return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
600
}
601

    
602
static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
603
{
604
    return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
605
}
606

    
607
static inline int get_penalty_factor(int lambda, int lambda2, int type){
608
    switch(type&0xFF){
609
    default:
610
    case FF_CMP_SAD:
611
        return lambda>>FF_LAMBDA_SHIFT;
612
    case FF_CMP_DCT:
613
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
614
    case FF_CMP_W53:
615
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
616
    case FF_CMP_W97:
617
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
618
    case FF_CMP_SATD:
619
    case FF_CMP_DCT264:
620
        return (2*lambda)>>FF_LAMBDA_SHIFT;
621
    case FF_CMP_RD:
622
    case FF_CMP_PSNR:
623
    case FF_CMP_SSE:
624
    case FF_CMP_NSSE:
625
        return lambda2>>FF_LAMBDA_SHIFT;
626
    case FF_CMP_BIT:
627
        return 1;
628
    }
629
}
630

    
631
/**
632
 * Empty mmx state.
633
 * this must be called between any dsp function and float/double code.
634
 * for example sin(); dsp->idct_put(); emms_c(); cos()
635
 */
636
#define emms_c()
637

    
638
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
639
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
640
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
641
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
642
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
643
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
644
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
645
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
646
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
647

    
648
void ff_dsputil_init_dwt(DSPContext *c);
649
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
650
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
651
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
652
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
653
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
654

    
655
#if HAVE_MMX
656

    
657
#undef emms_c
658

    
659
static inline void emms(void)
660
{
661
    __asm__ volatile ("emms;":::"memory");
662
}
663

    
664
#define emms_c() \
665
{\
666
    if(av_get_cpu_flags() & AV_CPU_FLAG_MMX)\
667
        emms();\
668
}
669

    
670
#elif ARCH_ARM
671

    
672
#if HAVE_NEON
673
#   define STRIDE_ALIGN 16
674
#endif
675

    
676
#elif ARCH_PPC
677

    
678
#define STRIDE_ALIGN 16
679

    
680
#elif HAVE_MMI
681

    
682
#define STRIDE_ALIGN 16
683

    
684
#endif
685

    
686
#ifndef STRIDE_ALIGN
687
#   define STRIDE_ALIGN 8
688
#endif
689

    
690
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...)             \
691
    uint8_t la_##v[sizeof(t s o) + (a)];                \
692
    t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
693

    
694
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
695

    
696
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
697

    
698
#if HAVE_LOCAL_ALIGNED_8
699
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
700
#else
701
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
702
#endif
703

    
704
#if HAVE_LOCAL_ALIGNED_16
705
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
706
#else
707
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
708
#endif
709

    
710
/* PSNR */
711
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
712
              int orig_linesize[3], int coded_linesize,
713
              AVCodecContext *avctx);
714

    
715
#define WRAPPER8_16(name8, name16)\
716
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
717
    return name8(s, dst           , src           , stride, h)\
718
          +name8(s, dst+8         , src+8         , stride, h);\
719
}
720

    
721
#define WRAPPER8_16_SQ(name8, name16)\
722
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
723
    int score=0;\
724
    score +=name8(s, dst           , src           , stride, 8);\
725
    score +=name8(s, dst+8         , src+8         , stride, 8);\
726
    if(h==16){\
727
        dst += 8*stride;\
728
        src += 8*stride;\
729
        score +=name8(s, dst           , src           , stride, 8);\
730
        score +=name8(s, dst+8         , src+8         , stride, 8);\
731
    }\
732
    return score;\
733
}
734

    
735

    
736
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
737
{
738
    int i;
739
    for(i=0; i<h; i++)
740
    {
741
        AV_WN16(dst   , AV_RN16(src   ));
742
        dst+=dstStride;
743
        src+=srcStride;
744
    }
745
}
746

    
747
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
748
{
749
    int i;
750
    for(i=0; i<h; i++)
751
    {
752
        AV_WN32(dst   , AV_RN32(src   ));
753
        dst+=dstStride;
754
        src+=srcStride;
755
    }
756
}
757

    
758
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
759
{
760
    int i;
761
    for(i=0; i<h; i++)
762
    {
763
        AV_WN32(dst   , AV_RN32(src   ));
764
        AV_WN32(dst+4 , AV_RN32(src+4 ));
765
        dst+=dstStride;
766
        src+=srcStride;
767
    }
768
}
769

    
770
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
771
{
772
    int i;
773
    for(i=0; i<h; i++)
774
    {
775
        AV_WN32(dst   , AV_RN32(src   ));
776
        AV_WN32(dst+4 , AV_RN32(src+4 ));
777
        dst[8]= src[8];
778
        dst+=dstStride;
779
        src+=srcStride;
780
    }
781
}
782

    
783
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
784
{
785
    int i;
786
    for(i=0; i<h; i++)
787
    {
788
        AV_WN32(dst   , AV_RN32(src   ));
789
        AV_WN32(dst+4 , AV_RN32(src+4 ));
790
        AV_WN32(dst+8 , AV_RN32(src+8 ));
791
        AV_WN32(dst+12, AV_RN32(src+12));
792
        dst+=dstStride;
793
        src+=srcStride;
794
    }
795
}
796

    
797
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
798
{
799
    int i;
800
    for(i=0; i<h; i++)
801
    {
802
        AV_WN32(dst   , AV_RN32(src   ));
803
        AV_WN32(dst+4 , AV_RN32(src+4 ));
804
        AV_WN32(dst+8 , AV_RN32(src+8 ));
805
        AV_WN32(dst+12, AV_RN32(src+12));
806
        dst[16]= src[16];
807
        dst+=dstStride;
808
        src+=srcStride;
809
    }
810
}
811

    
812
#endif /* AVCODEC_DSPUTIL_H */