Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 80ba1ddb

History | View | Annotate | Download (31.5 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_h264_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul);
68
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
69
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
70

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

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

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

    
84
void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
85
void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
86
void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
87
void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
88

    
89
/* VP3 DSP functions */
90
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
91
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
92
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
93
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
94

    
95
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
96
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
97

    
98
/* Bink functions */
99
void ff_bink_idct_c    (DCTELEM *block);
100
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
101
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
102

    
103
/* EA functions */
104
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
105

    
106
/* 1/2^n downscaling functions from imgconvert.c */
107
#if LIBAVCODEC_VERSION_MAJOR < 53
108
/**
109
 * @deprecated Use av_image_copy_plane() instead.
110
 */
111
attribute_deprecated
112
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
113
#endif
114

    
115
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
116
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
117
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
118

    
119
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
120
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
121

    
122
/* minimum alignment rules ;)
123
If you notice errors in the align stuff, need more alignment for some ASM code
124
for some CPU or need to use a function with less aligned data then send a mail
125
to the ffmpeg-devel mailing list, ...
126

127
!warning These alignments might not match reality, (missing attribute((align))
128
stuff somewhere possible).
129
I (Michael) did not check them, these are just the alignments which I think
130
could be reached easily ...
131

132
!future video codecs might need functions with less strict alignment
133
*/
134

    
135
/*
136
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
137
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
138
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
139
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
140
void clear_blocks_c(DCTELEM *blocks);
141
*/
142

    
143
/* add and put pixel (decoding) */
144
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
145
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
146
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
147
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);
148
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
149
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
150

    
151
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
152

    
153
#define DEF_OLD_QPEL(name)\
154
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
155
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
156
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
157

    
158
DEF_OLD_QPEL(qpel16_mc11_old_c)
159
DEF_OLD_QPEL(qpel16_mc31_old_c)
160
DEF_OLD_QPEL(qpel16_mc12_old_c)
161
DEF_OLD_QPEL(qpel16_mc32_old_c)
162
DEF_OLD_QPEL(qpel16_mc13_old_c)
163
DEF_OLD_QPEL(qpel16_mc33_old_c)
164
DEF_OLD_QPEL(qpel8_mc11_old_c)
165
DEF_OLD_QPEL(qpel8_mc31_old_c)
166
DEF_OLD_QPEL(qpel8_mc12_old_c)
167
DEF_OLD_QPEL(qpel8_mc32_old_c)
168
DEF_OLD_QPEL(qpel8_mc13_old_c)
169
DEF_OLD_QPEL(qpel8_mc33_old_c)
170

    
171
#define CALL_2X_PIXELS(a, b, n)\
172
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
173
    b(block  , pixels  , line_size, h);\
174
    b(block+n, pixels+n, line_size, h);\
175
}
176

    
177
/* motion estimation */
178
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
179
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
180
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))*/;
181

    
182
/**
183
 * Scantable.
184
 */
185
typedef struct ScanTable{
186
    const uint8_t *scantable;
187
    uint8_t permutated[64];
188
    uint8_t raster_end[64];
189
#if ARCH_PPC
190
                /** Used by dct_quantize_altivec to find last-non-zero */
191
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
192
#endif
193
} ScanTable;
194

    
195
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
196

    
197
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
198
                         int block_w, int block_h,
199
                         int src_x, int src_y, int w, int h);
200

    
201
/**
202
 * DSPContext.
203
 */
204
typedef struct DSPContext {
205
    /* pixel ops : interface with DCT */
206
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
207
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
208
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
209
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
210
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
211
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
212
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
213
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
214
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
215
    /**
216
     * Motion estimation with emulated edge values.
217
     * @param buf pointer to destination buffer (unaligned)
218
     * @param src pointer to pixel source (unaligned)
219
     * @param linesize width (in pixels) for src/buf
220
     * @param block_w number of pixels (per row) to copy to buf
221
     * @param block_h nummber of pixel rows to copy to buf
222
     * @param src_x offset of src to start of row - this may be negative
223
     * @param src_y offset of src to top of image - this may be negative
224
     * @param w width of src in pixels
225
     * @param h height of src in pixels
226
     */
227
    void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
228
                             int block_w, int block_h,
229
                             int src_x, int src_y, int w, int h);
230
    /**
231
     * translational global motion compensation.
232
     */
233
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
234
    /**
235
     * global motion compensation.
236
     */
237
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
238
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
239
    void (*clear_block)(DCTELEM *block/*align 16*/);
240
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
241
    int (*pix_sum)(uint8_t * pix, int line_size);
242
    int (*pix_norm1)(uint8_t * pix, int line_size);
243
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
244

    
245
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
246
    me_cmp_func sse[6];
247
    me_cmp_func hadamard8_diff[6];
248
    me_cmp_func dct_sad[6];
249
    me_cmp_func quant_psnr[6];
250
    me_cmp_func bit[6];
251
    me_cmp_func rd[6];
252
    me_cmp_func vsad[6];
253
    me_cmp_func vsse[6];
254
    me_cmp_func nsse[6];
255
    me_cmp_func w53[6];
256
    me_cmp_func w97[6];
257
    me_cmp_func dct_max[6];
258
    me_cmp_func dct264_sad[6];
259

    
260
    me_cmp_func me_pre_cmp[6];
261
    me_cmp_func me_cmp[6];
262
    me_cmp_func me_sub_cmp[6];
263
    me_cmp_func mb_cmp[6];
264
    me_cmp_func ildct_cmp[6]; //only width 16 used
265
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
266

    
267
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
268
                             int size);
269

    
270
    /**
271
     * Halfpel motion compensation with rounding (a+b+1)>>1.
272
     * this is an array[4][4] of motion compensation functions for 4
273
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
274
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
275
     * @param block destination where the result is stored
276
     * @param pixels source
277
     * @param line_size number of bytes in a horizontal line of block
278
     * @param h height
279
     */
280
    op_pixels_func put_pixels_tab[4][4];
281

    
282
    /**
283
     * Halfpel motion compensation with rounding (a+b+1)>>1.
284
     * This is an array[4][4] of motion compensation functions for 4
285
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
286
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
287
     * @param block destination into which the result is averaged (a+b+1)>>1
288
     * @param pixels source
289
     * @param line_size number of bytes in a horizontal line of block
290
     * @param h height
291
     */
292
    op_pixels_func avg_pixels_tab[4][4];
293

    
294
    /**
295
     * Halfpel motion compensation with no rounding (a+b)>>1.
296
     * this is an array[2][4] of motion compensation functions for 2
297
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
298
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
299
     * @param block destination where the result is stored
300
     * @param pixels source
301
     * @param line_size number of bytes in a horizontal line of block
302
     * @param h height
303
     */
304
    op_pixels_func put_no_rnd_pixels_tab[4][4];
305

    
306
    /**
307
     * Halfpel motion compensation with no rounding (a+b)>>1.
308
     * this is an array[2][4] of motion compensation functions for 2
309
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
310
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
311
     * @param block destination into which the result is averaged (a+b)>>1
312
     * @param pixels source
313
     * @param line_size number of bytes in a horizontal line of block
314
     * @param h height
315
     */
316
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
317

    
318
    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);
319

    
320
    /**
321
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
322
     * this is an array[12] of motion compensation functions for the 9 thirdpe
323
     * positions<br>
324
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
325
     * @param block destination where the result is stored
326
     * @param pixels source
327
     * @param line_size number of bytes in a horizontal line of block
328
     * @param h height
329
     */
330
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
331
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
332

    
333
    qpel_mc_func put_qpel_pixels_tab[2][16];
334
    qpel_mc_func avg_qpel_pixels_tab[2][16];
335
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
336
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
337
    qpel_mc_func put_mspel_pixels_tab[8];
338

    
339
    /**
340
     * h264 Chroma MC
341
     */
342
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
343
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
344
    /* This is really one func used in VC-1 decoding */
345
    h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
346
    h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
347

    
348
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
349
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
350

    
351
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
352
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
353

    
354
    me_cmp_func pix_abs[2][4];
355

    
356
    /* huffyuv specific */
357
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
358
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
359
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
360
    /**
361
     * subtract huffyuv's variant of median prediction
362
     * note, this might read from src1[-1], src2[-1]
363
     */
364
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
365
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
366
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
367
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
368
    /* this might write to dst[w] */
369
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
370
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
371

    
372
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
373
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
374

    
375
    void (*h261_loop_filter)(uint8_t *src, int stride);
376

    
377
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
378
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
379

    
380
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
381
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
382
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
383

    
384
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
385
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
386
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
387
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
388
    void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
389
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
390
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
391
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
392
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
393
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
394
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
395
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
396
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
397
    /**
398
     * Multiply a vector of floats by a scalar float.  Source and
399
     * destination vectors must overlap exactly or not at all.
400
     * @param dst result vector, 16-byte aligned
401
     * @param src input vector, 16-byte aligned
402
     * @param mul scalar value
403
     * @param len length of vector, multiple of 4
404
     */
405
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
406
                               int len);
407
    /**
408
     * Multiply a vector of floats by concatenated short vectors of
409
     * floats and by a scalar float.  Source and destination vectors
410
     * must overlap exactly or not at all.
411
     * [0]: short vectors of length 2, 8-byte aligned
412
     * [1]: short vectors of length 4, 16-byte aligned
413
     * @param dst output vector, 16-byte aligned
414
     * @param src input vector, 16-byte aligned
415
     * @param sv  array of pointers to short vectors
416
     * @param mul scalar value
417
     * @param len number of elements in src and dst, multiple of 4
418
     */
419
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
420
                                     const float **sv, float mul, int len);
421
    /**
422
     * Multiply short vectors of floats by a scalar float, store
423
     * concatenated result.
424
     * [0]: short vectors of length 2, 8-byte aligned
425
     * [1]: short vectors of length 4, 16-byte aligned
426
     * @param dst output vector, 16-byte aligned
427
     * @param sv  array of pointers to short vectors
428
     * @param mul scalar value
429
     * @param len number of output elements, multiple of 4
430
     */
431
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
432
                              float mul, int len);
433
    /**
434
     * Calculate the scalar product of two vectors of floats.
435
     * @param v1  first vector, 16-byte aligned
436
     * @param v2  second vector, 16-byte aligned
437
     * @param len length of vectors, multiple of 4
438
     */
439
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
440
    /**
441
     * Calculate the sum and difference of two vectors of floats.
442
     * @param v1  first input vector, sum output, 16-byte aligned
443
     * @param v2  second input vector, difference output, 16-byte aligned
444
     * @param len length of vectors, multiple of 4
445
     */
446
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
447

    
448
    /* convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
449
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
450
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
451

    
452
    /* (I)DCT */
453
    void (*fdct)(DCTELEM *block/* align 16*/);
454
    void (*fdct248)(DCTELEM *block/* align 16*/);
455

    
456
    /* IDCT really*/
457
    void (*idct)(DCTELEM *block/* align 16*/);
458

    
459
    /**
460
     * block -> idct -> clip to unsigned 8 bit -> dest.
461
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
462
     * @param line_size size in bytes of a horizontal line of dest
463
     */
464
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
465

    
466
    /**
467
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
468
     * @param line_size size in bytes of a horizontal line of dest
469
     */
470
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
471

    
472
    /**
473
     * idct input permutation.
474
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
475
     * IDCT)
476
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
477
     * with the zigzag/alternate scan<br>
478
     * an example to avoid confusion:
479
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
480
     * - (x -> referece dct -> reference idct -> x)
481
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
482
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
483
     */
484
    uint8_t idct_permutation[64];
485
    int idct_permutation_type;
486
#define FF_NO_IDCT_PERM 1
487
#define FF_LIBMPEG2_IDCT_PERM 2
488
#define FF_SIMPLE_IDCT_PERM 3
489
#define FF_TRANSPOSE_IDCT_PERM 4
490
#define FF_PARTTRANS_IDCT_PERM 5
491
#define FF_SSE2_IDCT_PERM 6
492

    
493
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
494
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
495
#define BASIS_SHIFT 16
496
#define RECON_SHIFT 6
497

    
498
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
499
#define EDGE_WIDTH 16
500

    
501
    void (*prefetch)(void *mem, int stride, int h);
502

    
503
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
504

    
505
    /* mlp/truehd functions */
506
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
507
                               int firorder, int iirorder,
508
                               unsigned int filter_shift, int32_t mask, int blocksize,
509
                               int32_t *sample_buffer);
510

    
511
    /* vc1 functions */
512
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
513
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
514
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
515
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
516
    void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
517
    void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
518
    void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
519
    void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
520
    void (*vc1_v_overlap)(uint8_t* src, int stride);
521
    void (*vc1_h_overlap)(uint8_t* src, int stride);
522
    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
523
    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
524
    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
525
    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
526
    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
527
    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
528
    /* put 8x8 block with bicubic interpolation and quarterpel precision
529
     * last argument is actually round value instead of height
530
     */
531
    op_pixels_func put_vc1_mspel_pixels_tab[16];
532
    op_pixels_func avg_vc1_mspel_pixels_tab[16];
533

    
534
    /* intrax8 functions */
535
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
536
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
537
           int * range, int * sum,  int edges);
538

    
539
    /**
540
     * Calculate scalar product of two vectors.
541
     * @param len length of vectors, should be multiple of 16
542
     * @param shift number of bits to discard from product
543
     */
544
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
545
    /* ape functions */
546
    /**
547
     * Calculate scalar product of v1 and v2,
548
     * and v1[i] += v3[i] * mul
549
     * @param len length of vectors, should be multiple of 16
550
     */
551
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
552

    
553
    /* rv30 functions */
554
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
555
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
556

    
557
    /* rv40 functions */
558
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
559
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
560
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
561
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
562

    
563
    /* bink functions */
564
    op_fill_func fill_block_tab[2];
565
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
566
} DSPContext;
567

    
568
void dsputil_static_init(void);
569
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
570

    
571
int ff_check_alignment(void);
572

    
573
/**
574
 * permute block according to permuatation.
575
 * @param last last non zero element in scantable order
576
 */
577
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
578

    
579
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
580

    
581
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
582

    
583
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
584
{
585
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
586
}
587

    
588
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
589
{
590
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
591
}
592

    
593
static inline int get_penalty_factor(int lambda, int lambda2, int type){
594
    switch(type&0xFF){
595
    default:
596
    case FF_CMP_SAD:
597
        return lambda>>FF_LAMBDA_SHIFT;
598
    case FF_CMP_DCT:
599
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
600
    case FF_CMP_W53:
601
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
602
    case FF_CMP_W97:
603
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
604
    case FF_CMP_SATD:
605
    case FF_CMP_DCT264:
606
        return (2*lambda)>>FF_LAMBDA_SHIFT;
607
    case FF_CMP_RD:
608
    case FF_CMP_PSNR:
609
    case FF_CMP_SSE:
610
    case FF_CMP_NSSE:
611
        return lambda2>>FF_LAMBDA_SHIFT;
612
    case FF_CMP_BIT:
613
        return 1;
614
    }
615
}
616

    
617
/**
618
 * Empty mmx state.
619
 * this must be called between any dsp function and float/double code.
620
 * for example sin(); dsp->idct_put(); emms_c(); cos()
621
 */
622
#define emms_c()
623

    
624
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
625
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
626
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
627
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
628
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
629
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
630
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
631
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
632
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
633

    
634
void ff_dsputil_init_dwt(DSPContext *c);
635
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
636
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
637
void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
638
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
639
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
640
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
641

    
642
#if HAVE_MMX
643

    
644
#undef emms_c
645

    
646
static inline void emms(void)
647
{
648
    __asm__ volatile ("emms;":::"memory");
649
}
650

    
651
#define emms_c() emms()
652

    
653
#elif ARCH_ARM
654

    
655
#if HAVE_NEON
656
#   define STRIDE_ALIGN 16
657
#endif
658

    
659
#elif ARCH_PPC
660

    
661
#define STRIDE_ALIGN 16
662

    
663
#elif HAVE_MMI
664

    
665
#define STRIDE_ALIGN 16
666

    
667
#endif
668

    
669
#ifndef STRIDE_ALIGN
670
#   define STRIDE_ALIGN 8
671
#endif
672

    
673
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...)             \
674
    uint8_t la_##v[sizeof(t s o) + (a)];                \
675
    t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
676

    
677
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
678

    
679
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
680

    
681
#if HAVE_LOCAL_ALIGNED_8
682
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
683
#else
684
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
685
#endif
686

    
687
#if HAVE_LOCAL_ALIGNED_16
688
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
689
#else
690
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
691
#endif
692

    
693
/* PSNR */
694
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
695
              int orig_linesize[3], int coded_linesize,
696
              AVCodecContext *avctx);
697

    
698
#define WRAPPER8_16(name8, name16)\
699
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
700
    return name8(s, dst           , src           , stride, h)\
701
          +name8(s, dst+8         , src+8         , stride, h);\
702
}
703

    
704
#define WRAPPER8_16_SQ(name8, name16)\
705
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
706
    int score=0;\
707
    score +=name8(s, dst           , src           , stride, 8);\
708
    score +=name8(s, dst+8         , src+8         , stride, 8);\
709
    if(h==16){\
710
        dst += 8*stride;\
711
        src += 8*stride;\
712
        score +=name8(s, dst           , src           , stride, 8);\
713
        score +=name8(s, dst+8         , src+8         , stride, 8);\
714
    }\
715
    return score;\
716
}
717

    
718

    
719
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
720
{
721
    int i;
722
    for(i=0; i<h; i++)
723
    {
724
        AV_WN16(dst   , AV_RN16(src   ));
725
        dst+=dstStride;
726
        src+=srcStride;
727
    }
728
}
729

    
730
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
731
{
732
    int i;
733
    for(i=0; i<h; i++)
734
    {
735
        AV_WN32(dst   , AV_RN32(src   ));
736
        dst+=dstStride;
737
        src+=srcStride;
738
    }
739
}
740

    
741
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
742
{
743
    int i;
744
    for(i=0; i<h; i++)
745
    {
746
        AV_WN32(dst   , AV_RN32(src   ));
747
        AV_WN32(dst+4 , AV_RN32(src+4 ));
748
        dst+=dstStride;
749
        src+=srcStride;
750
    }
751
}
752

    
753
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
754
{
755
    int i;
756
    for(i=0; i<h; i++)
757
    {
758
        AV_WN32(dst   , AV_RN32(src   ));
759
        AV_WN32(dst+4 , AV_RN32(src+4 ));
760
        dst[8]= src[8];
761
        dst+=dstStride;
762
        src+=srcStride;
763
    }
764
}
765

    
766
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
767
{
768
    int i;
769
    for(i=0; i<h; i++)
770
    {
771
        AV_WN32(dst   , AV_RN32(src   ));
772
        AV_WN32(dst+4 , AV_RN32(src+4 ));
773
        AV_WN32(dst+8 , AV_RN32(src+8 ));
774
        AV_WN32(dst+12, AV_RN32(src+12));
775
        dst+=dstStride;
776
        src+=srcStride;
777
    }
778
}
779

    
780
static inline void copy_block17(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
        AV_WN32(dst+8 , AV_RN32(src+8 ));
788
        AV_WN32(dst+12, AV_RN32(src+12));
789
        dst[16]= src[16];
790
        dst+=dstStride;
791
        src+=srcStride;
792
    }
793
}
794

    
795
#endif /* AVCODEC_DSPUTIL_H */