Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 8dbe5856

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
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
#if LIBAVCODEC_VERSION_MAJOR < 53
124
/**
125
 * @deprecated Use av_image_copy_plane() instead.
126
 */
127
attribute_deprecated
128
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
129
#endif
130

    
131
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
132
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
133
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
134

    
135
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
136
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
137

    
138
/* minimum alignment rules ;)
139
If you notice errors in the align stuff, need more alignment for some ASM code
140
for some CPU or need to use a function with less aligned data then send a mail
141
to the ffmpeg-devel mailing list, ...
142

143
!warning These alignments might not match reality, (missing attribute((align))
144
stuff somewhere possible).
145
I (Michael) did not check them, these are just the alignments which I think
146
could be reached easily ...
147

148
!future video codecs might need functions with less strict alignment
149
*/
150

    
151
/*
152
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
153
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
154
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
155
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
156
void clear_blocks_c(DCTELEM *blocks);
157
*/
158

    
159
/* add and put pixel (decoding) */
160
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
161
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
162
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
163
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);
164
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
165
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
166

    
167
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
168

    
169
#define DEF_OLD_QPEL(name)\
170
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
171
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
172
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
173

    
174
DEF_OLD_QPEL(qpel16_mc11_old_c)
175
DEF_OLD_QPEL(qpel16_mc31_old_c)
176
DEF_OLD_QPEL(qpel16_mc12_old_c)
177
DEF_OLD_QPEL(qpel16_mc32_old_c)
178
DEF_OLD_QPEL(qpel16_mc13_old_c)
179
DEF_OLD_QPEL(qpel16_mc33_old_c)
180
DEF_OLD_QPEL(qpel8_mc11_old_c)
181
DEF_OLD_QPEL(qpel8_mc31_old_c)
182
DEF_OLD_QPEL(qpel8_mc12_old_c)
183
DEF_OLD_QPEL(qpel8_mc32_old_c)
184
DEF_OLD_QPEL(qpel8_mc13_old_c)
185
DEF_OLD_QPEL(qpel8_mc33_old_c)
186

    
187
#define CALL_2X_PIXELS(a, b, n)\
188
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
189
    b(block  , pixels  , line_size, h);\
190
    b(block+n, pixels+n, line_size, h);\
191
}
192

    
193
/* motion estimation */
194
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
195
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
196
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))*/;
197

    
198
/**
199
 * Scantable.
200
 */
201
typedef struct ScanTable{
202
    const uint8_t *scantable;
203
    uint8_t permutated[64];
204
    uint8_t raster_end[64];
205
#if ARCH_PPC
206
                /** Used by dct_quantize_altivec to find last-non-zero */
207
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
208
#endif
209
} ScanTable;
210

    
211
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
212

    
213
#define EMULATED_EDGE(depth) \
214
void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, int linesize,\
215
                         int block_w, int block_h,\
216
                         int src_x, int src_y, int w, int h);
217

    
218
EMULATED_EDGE(8)
219
EMULATED_EDGE(9)
220
EMULATED_EDGE(10)
221

    
222
#define ff_emulated_edge_mc ff_emulated_edge_mc_8
223

    
224
void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
225
void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
226
void ff_put_signed_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
227

    
228
/**
229
 * DSPContext.
230
 */
231
typedef struct DSPContext {
232
    /* pixel ops : interface with DCT */
233
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
234
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
235
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
236
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
237
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
238
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
239
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
240
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
241
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
242
    /**
243
     * Motion estimation with emulated edge values.
244
     * @param buf pointer to destination buffer (unaligned)
245
     * @param src pointer to pixel source (unaligned)
246
     * @param linesize width (in pixels) for src/buf
247
     * @param block_w number of pixels (per row) to copy to buf
248
     * @param block_h nummber of pixel rows to copy to buf
249
     * @param src_x offset of src to start of row - this may be negative
250
     * @param src_y offset of src to top of image - this may be negative
251
     * @param w width of src in pixels
252
     * @param h height of src in pixels
253
     */
254
    void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
255
                             int block_w, int block_h,
256
                             int src_x, int src_y, int w, int h);
257
    /**
258
     * translational global motion compensation.
259
     */
260
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
261
    /**
262
     * global motion compensation.
263
     */
264
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
265
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
266
    void (*clear_block)(DCTELEM *block/*align 16*/);
267
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
268
    int (*pix_sum)(uint8_t * pix, int line_size);
269
    int (*pix_norm1)(uint8_t * pix, int line_size);
270
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
271

    
272
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
273
    me_cmp_func sse[6];
274
    me_cmp_func hadamard8_diff[6];
275
    me_cmp_func dct_sad[6];
276
    me_cmp_func quant_psnr[6];
277
    me_cmp_func bit[6];
278
    me_cmp_func rd[6];
279
    me_cmp_func vsad[6];
280
    me_cmp_func vsse[6];
281
    me_cmp_func nsse[6];
282
    me_cmp_func w53[6];
283
    me_cmp_func w97[6];
284
    me_cmp_func dct_max[6];
285
    me_cmp_func dct264_sad[6];
286

    
287
    me_cmp_func me_pre_cmp[6];
288
    me_cmp_func me_cmp[6];
289
    me_cmp_func me_sub_cmp[6];
290
    me_cmp_func mb_cmp[6];
291
    me_cmp_func ildct_cmp[6]; //only width 16 used
292
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
293

    
294
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
295
                             int size);
296

    
297
    /**
298
     * Halfpel motion compensation with rounding (a+b+1)>>1.
299
     * this is an array[4][4] of motion compensation functions for 4
300
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
301
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
302
     * @param block destination where the result is stored
303
     * @param pixels source
304
     * @param line_size number of bytes in a horizontal line of block
305
     * @param h height
306
     */
307
    op_pixels_func put_pixels_tab[4][4];
308

    
309
    /**
310
     * Halfpel motion compensation with rounding (a+b+1)>>1.
311
     * This is an array[4][4] of motion compensation functions for 4
312
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
313
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
314
     * @param block destination into which the result is averaged (a+b+1)>>1
315
     * @param pixels source
316
     * @param line_size number of bytes in a horizontal line of block
317
     * @param h height
318
     */
319
    op_pixels_func avg_pixels_tab[4][4];
320

    
321
    /**
322
     * Halfpel motion compensation with no rounding (a+b)>>1.
323
     * this is an array[2][4] of motion compensation functions for 2
324
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
325
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
326
     * @param block destination where the result is stored
327
     * @param pixels source
328
     * @param line_size number of bytes in a horizontal line of block
329
     * @param h height
330
     */
331
    op_pixels_func put_no_rnd_pixels_tab[4][4];
332

    
333
    /**
334
     * Halfpel motion compensation with no rounding (a+b)>>1.
335
     * this is an array[2][4] of motion compensation functions for 2
336
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
337
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
338
     * @param block destination into which the result is averaged (a+b)>>1
339
     * @param pixels source
340
     * @param line_size number of bytes in a horizontal line of block
341
     * @param h height
342
     */
343
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
344

    
345
    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);
346

    
347
    /**
348
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
349
     * this is an array[12] of motion compensation functions for the 9 thirdpe
350
     * positions<br>
351
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
352
     * @param block destination where the result is stored
353
     * @param pixels source
354
     * @param line_size number of bytes in a horizontal line of block
355
     * @param h height
356
     */
357
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
358
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
359

    
360
    qpel_mc_func put_qpel_pixels_tab[2][16];
361
    qpel_mc_func avg_qpel_pixels_tab[2][16];
362
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
363
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
364
    qpel_mc_func put_mspel_pixels_tab[8];
365

    
366
    /**
367
     * h264 Chroma MC
368
     */
369
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
370
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
371

    
372
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
373
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
374

    
375
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
376
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
377

    
378
    me_cmp_func pix_abs[2][4];
379

    
380
    /* huffyuv specific */
381
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
382
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
383
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
384
    /**
385
     * subtract huffyuv's variant of median prediction
386
     * note, this might read from src1[-1], src2[-1]
387
     */
388
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
389
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
390
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
391
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
392
    /* this might write to dst[w] */
393
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
394
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
395
    void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
396

    
397
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
398
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
399

    
400
    void (*h261_loop_filter)(uint8_t *src, int stride);
401

    
402
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
403
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
404

    
405
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
406
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
407
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
408

    
409
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
410
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
411
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
412
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
413
    void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
414
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
415
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
416
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
417
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
418
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
419
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
420
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
421
    /**
422
     * Multiply a vector of floats by a scalar float.  Source and
423
     * destination vectors must overlap exactly or not at all.
424
     * @param dst result vector, 16-byte aligned
425
     * @param src input vector, 16-byte aligned
426
     * @param mul scalar value
427
     * @param len length of vector, multiple of 4
428
     */
429
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
430
                               int len);
431
    /**
432
     * Multiply a vector of floats by concatenated short vectors of
433
     * floats and by a scalar float.  Source and destination vectors
434
     * must overlap exactly or not at all.
435
     * [0]: short vectors of length 2, 8-byte aligned
436
     * [1]: short vectors of length 4, 16-byte aligned
437
     * @param dst output vector, 16-byte aligned
438
     * @param src input vector, 16-byte aligned
439
     * @param sv  array of pointers to short vectors
440
     * @param mul scalar value
441
     * @param len number of elements in src and dst, multiple of 4
442
     */
443
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
444
                                     const float **sv, float mul, int len);
445
    /**
446
     * Multiply short vectors of floats by a scalar float, store
447
     * concatenated result.
448
     * [0]: short vectors of length 2, 8-byte aligned
449
     * [1]: short vectors of length 4, 16-byte aligned
450
     * @param dst output vector, 16-byte aligned
451
     * @param sv  array of pointers to short vectors
452
     * @param mul scalar value
453
     * @param len number of output elements, multiple of 4
454
     */
455
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
456
                              float mul, int len);
457
    /**
458
     * Calculate the scalar product of two vectors of floats.
459
     * @param v1  first vector, 16-byte aligned
460
     * @param v2  second vector, 16-byte aligned
461
     * @param len length of vectors, multiple of 4
462
     */
463
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
464
    /**
465
     * Calculate the sum and difference of two vectors of floats.
466
     * @param v1  first input vector, sum output, 16-byte aligned
467
     * @param v2  second input vector, difference output, 16-byte aligned
468
     * @param len length of vectors, multiple of 4
469
     */
470
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
471

    
472
    /* (I)DCT */
473
    void (*fdct)(DCTELEM *block/* align 16*/);
474
    void (*fdct248)(DCTELEM *block/* align 16*/);
475

    
476
    /* IDCT really*/
477
    void (*idct)(DCTELEM *block/* align 16*/);
478

    
479
    /**
480
     * block -> idct -> clip to unsigned 8 bit -> dest.
481
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
482
     * @param line_size size in bytes of a horizontal line of dest
483
     */
484
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
485

    
486
    /**
487
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
488
     * @param line_size size in bytes of a horizontal line of dest
489
     */
490
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
491

    
492
    /**
493
     * idct input permutation.
494
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
495
     * IDCT)
496
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
497
     * with the zigzag/alternate scan<br>
498
     * an example to avoid confusion:
499
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
500
     * - (x -> referece dct -> reference idct -> x)
501
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
502
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
503
     */
504
    uint8_t idct_permutation[64];
505
    int idct_permutation_type;
506
#define FF_NO_IDCT_PERM 1
507
#define FF_LIBMPEG2_IDCT_PERM 2
508
#define FF_SIMPLE_IDCT_PERM 3
509
#define FF_TRANSPOSE_IDCT_PERM 4
510
#define FF_PARTTRANS_IDCT_PERM 5
511
#define FF_SSE2_IDCT_PERM 6
512

    
513
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
514
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
515
#define BASIS_SHIFT 16
516
#define RECON_SHIFT 6
517

    
518
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int sides);
519
#define EDGE_WIDTH 16
520
#define EDGE_TOP    1
521
#define EDGE_BOTTOM 2
522

    
523
    void (*prefetch)(void *mem, int stride, int h);
524

    
525
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
526

    
527
    /* mlp/truehd functions */
528
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
529
                               int firorder, int iirorder,
530
                               unsigned int filter_shift, int32_t mask, int blocksize,
531
                               int32_t *sample_buffer);
532

    
533
    /* intrax8 functions */
534
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
535
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
536
           int * range, int * sum,  int edges);
537

    
538
    /**
539
     * Calculate scalar product of two vectors.
540
     * @param len length of vectors, should be multiple of 16
541
     * @param shift number of bits to discard from product
542
     */
543
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
544
    /* ape functions */
545
    /**
546
     * Calculate scalar product of v1 and v2,
547
     * and v1[i] += v3[i] * mul
548
     * @param len length of vectors, should be multiple of 16
549
     */
550
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
551

    
552
    /**
553
     * Apply symmetric window in 16-bit fixed-point.
554
     * @param output destination array
555
     *               constraints: 16-byte aligned
556
     * @param input  source array
557
     *               constraints: 16-byte aligned
558
     * @param window window array
559
     *               constraints: 16-byte aligned, at least len/2 elements
560
     * @param len    full window length
561
     *               constraints: multiple of ? greater than zero
562
     */
563
    void (*apply_window_int16)(int16_t *output, const int16_t *input,
564
                               const int16_t *window, unsigned int len);
565

    
566
    /* rv30 functions */
567
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
568
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
569

    
570
    /* rv40 functions */
571
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
572
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
573
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
574
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
575

    
576
    /* bink functions */
577
    op_fill_func fill_block_tab[2];
578
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
579
} DSPContext;
580

    
581
void dsputil_static_init(void);
582
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
583

    
584
int ff_check_alignment(void);
585

    
586
/**
587
 * permute block according to permuatation.
588
 * @param last last non zero element in scantable order
589
 */
590
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
591

    
592
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
593

    
594
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
595
#define         BYTE_VEC64(c)   ((c)*0x0001000100010001UL)
596

    
597
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
598
{
599
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
600
}
601

    
602
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
603
{
604
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
605
}
606

    
607
static inline uint64_t rnd_avg64(uint64_t a, uint64_t b)
608
{
609
    return (a | b) - (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
610
}
611

    
612
static inline uint64_t no_rnd_avg64(uint64_t a, uint64_t b)
613
{
614
    return (a & b) + (((a ^ b) & ~BYTE_VEC64(0x01)) >> 1);
615
}
616

    
617
static inline int get_penalty_factor(int lambda, int lambda2, int type){
618
    switch(type&0xFF){
619
    default:
620
    case FF_CMP_SAD:
621
        return lambda>>FF_LAMBDA_SHIFT;
622
    case FF_CMP_DCT:
623
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
624
    case FF_CMP_W53:
625
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
626
    case FF_CMP_W97:
627
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
628
    case FF_CMP_SATD:
629
    case FF_CMP_DCT264:
630
        return (2*lambda)>>FF_LAMBDA_SHIFT;
631
    case FF_CMP_RD:
632
    case FF_CMP_PSNR:
633
    case FF_CMP_SSE:
634
    case FF_CMP_NSSE:
635
        return lambda2>>FF_LAMBDA_SHIFT;
636
    case FF_CMP_BIT:
637
        return 1;
638
    }
639
}
640

    
641
/**
642
 * Empty mmx state.
643
 * this must be called between any dsp function and float/double code.
644
 * for example sin(); dsp->idct_put(); emms_c(); cos()
645
 */
646
#define emms_c()
647

    
648
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
649
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
650
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
651
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
652
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
653
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
654
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
655
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
656
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
657

    
658
void ff_dsputil_init_dwt(DSPContext *c);
659
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
660
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
661
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
662
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
663
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
664

    
665
#if HAVE_MMX
666

    
667
#undef emms_c
668

    
669
static inline void emms(void)
670
{
671
    __asm__ volatile ("emms;":::"memory");
672
}
673

    
674
#define emms_c() \
675
{\
676
    if(av_get_cpu_flags() & AV_CPU_FLAG_MMX)\
677
        emms();\
678
}
679

    
680
#elif ARCH_ARM
681

    
682
#if HAVE_NEON
683
#   define STRIDE_ALIGN 16
684
#endif
685

    
686
#elif ARCH_PPC
687

    
688
#define STRIDE_ALIGN 16
689

    
690
#elif HAVE_MMI
691

    
692
#define STRIDE_ALIGN 16
693

    
694
#endif
695

    
696
#ifndef STRIDE_ALIGN
697
#   define STRIDE_ALIGN 8
698
#endif
699

    
700
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...)             \
701
    uint8_t la_##v[sizeof(t s o) + (a)];                \
702
    t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
703

    
704
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
705

    
706
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
707

    
708
#if HAVE_LOCAL_ALIGNED_8
709
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
710
#else
711
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
712
#endif
713

    
714
#if HAVE_LOCAL_ALIGNED_16
715
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
716
#else
717
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
718
#endif
719

    
720
/* PSNR */
721
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
722
              int orig_linesize[3], int coded_linesize,
723
              AVCodecContext *avctx);
724

    
725
#define WRAPPER8_16(name8, name16)\
726
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
727
    return name8(s, dst           , src           , stride, h)\
728
          +name8(s, dst+8         , src+8         , stride, h);\
729
}
730

    
731
#define WRAPPER8_16_SQ(name8, name16)\
732
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
733
    int score=0;\
734
    score +=name8(s, dst           , src           , stride, 8);\
735
    score +=name8(s, dst+8         , src+8         , stride, 8);\
736
    if(h==16){\
737
        dst += 8*stride;\
738
        src += 8*stride;\
739
        score +=name8(s, dst           , src           , stride, 8);\
740
        score +=name8(s, dst+8         , src+8         , stride, 8);\
741
    }\
742
    return score;\
743
}
744

    
745

    
746
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
747
{
748
    int i;
749
    for(i=0; i<h; i++)
750
    {
751
        AV_WN16(dst   , AV_RN16(src   ));
752
        dst+=dstStride;
753
        src+=srcStride;
754
    }
755
}
756

    
757
static inline void copy_block4(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
        dst+=dstStride;
764
        src+=srcStride;
765
    }
766
}
767

    
768
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
769
{
770
    int i;
771
    for(i=0; i<h; i++)
772
    {
773
        AV_WN32(dst   , AV_RN32(src   ));
774
        AV_WN32(dst+4 , AV_RN32(src+4 ));
775
        dst+=dstStride;
776
        src+=srcStride;
777
    }
778
}
779

    
780
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
781
{
782
    int i;
783
    for(i=0; i<h; i++)
784
    {
785
        AV_WN32(dst   , AV_RN32(src   ));
786
        AV_WN32(dst+4 , AV_RN32(src+4 ));
787
        dst[8]= src[8];
788
        dst+=dstStride;
789
        src+=srcStride;
790
    }
791
}
792

    
793
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
794
{
795
    int i;
796
    for(i=0; i<h; i++)
797
    {
798
        AV_WN32(dst   , AV_RN32(src   ));
799
        AV_WN32(dst+4 , AV_RN32(src+4 ));
800
        AV_WN32(dst+8 , AV_RN32(src+8 ));
801
        AV_WN32(dst+12, AV_RN32(src+12));
802
        dst+=dstStride;
803
        src+=srcStride;
804
    }
805
}
806

    
807
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
808
{
809
    int i;
810
    for(i=0; i<h; i++)
811
    {
812
        AV_WN32(dst   , AV_RN32(src   ));
813
        AV_WN32(dst+4 , AV_RN32(src+4 ));
814
        AV_WN32(dst+8 , AV_RN32(src+8 ));
815
        AV_WN32(dst+12, AV_RN32(src+12));
816
        dst[16]= src[16];
817
        dst+=dstStride;
818
        src+=srcStride;
819
    }
820
}
821

    
822
#endif /* AVCODEC_DSPUTIL_H */