Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ e39e3aba

History | View | Annotate | Download (30.4 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 Libav.
7
 *
8
 * Libav 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
 * Libav 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 Libav; 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_chroma_dc_dequant_idct_c(DCTELEM *block, int qmul);
68
void ff_h264_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul);
69
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
70
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
71

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

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

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

    
85
void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
86
void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
87
void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
88
void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
89

    
90
/* VP3 DSP functions */
91
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
92
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
93
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
94
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
95

    
96
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
97
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
98

    
99
/* Bink functions */
100
void ff_bink_idct_c    (DCTELEM *block);
101
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
102
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
103

    
104
/* EA functions */
105
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
106

    
107
/* 1/2^n downscaling functions from imgconvert.c */
108
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
109
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
110
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
111

    
112
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
113
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
114

    
115
/* minimum alignment rules ;)
116
If you notice errors in the align stuff, need more alignment for some ASM code
117
for some CPU or need to use a function with less aligned data then send a mail
118
to the libav-devel mailing list, ...
119

120
!warning These alignments might not match reality, (missing attribute((align))
121
stuff somewhere possible).
122
I (Michael) did not check them, these are just the alignments which I think
123
could be reached easily ...
124

125
!future video codecs might need functions with less strict alignment
126
*/
127

    
128
/*
129
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
130
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
131
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
132
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
133
void clear_blocks_c(DCTELEM *blocks);
134
*/
135

    
136
/* add and put pixel (decoding) */
137
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
138
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
139
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
140
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);
141
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
142
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
143

    
144
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
145

    
146
#define DEF_OLD_QPEL(name)\
147
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
148
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
149
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
150

    
151
DEF_OLD_QPEL(qpel16_mc11_old_c)
152
DEF_OLD_QPEL(qpel16_mc31_old_c)
153
DEF_OLD_QPEL(qpel16_mc12_old_c)
154
DEF_OLD_QPEL(qpel16_mc32_old_c)
155
DEF_OLD_QPEL(qpel16_mc13_old_c)
156
DEF_OLD_QPEL(qpel16_mc33_old_c)
157
DEF_OLD_QPEL(qpel8_mc11_old_c)
158
DEF_OLD_QPEL(qpel8_mc31_old_c)
159
DEF_OLD_QPEL(qpel8_mc12_old_c)
160
DEF_OLD_QPEL(qpel8_mc32_old_c)
161
DEF_OLD_QPEL(qpel8_mc13_old_c)
162
DEF_OLD_QPEL(qpel8_mc33_old_c)
163

    
164
#define CALL_2X_PIXELS(a, b, n)\
165
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
166
    b(block  , pixels  , line_size, h);\
167
    b(block+n, pixels+n, line_size, h);\
168
}
169

    
170
/* motion estimation */
171
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
172
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
173
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))*/;
174

    
175
/**
176
 * Scantable.
177
 */
178
typedef struct ScanTable{
179
    const uint8_t *scantable;
180
    uint8_t permutated[64];
181
    uint8_t raster_end[64];
182
#if ARCH_PPC
183
                /** Used by dct_quantize_altivec to find last-non-zero */
184
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
185
#endif
186
} ScanTable;
187

    
188
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
189

    
190
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
191
                         int block_w, int block_h,
192
                         int src_x, int src_y, int w, int h);
193

    
194
void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
195
void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
196
void ff_put_signed_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
197

    
198
/**
199
 * DSPContext.
200
 */
201
typedef struct DSPContext {
202
    /* pixel ops : interface with DCT */
203
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
204
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
205
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
206
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
207
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
208
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
209
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
210
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
211
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
212
    /**
213
     * Motion estimation with emulated edge values.
214
     * @param buf pointer to destination buffer (unaligned)
215
     * @param src pointer to pixel source (unaligned)
216
     * @param linesize width (in pixels) for src/buf
217
     * @param block_w number of pixels (per row) to copy to buf
218
     * @param block_h nummber of pixel rows to copy to buf
219
     * @param src_x offset of src to start of row - this may be negative
220
     * @param src_y offset of src to top of image - this may be negative
221
     * @param w width of src in pixels
222
     * @param h height of src in pixels
223
     */
224
    void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
225
                             int block_w, int block_h,
226
                             int src_x, int src_y, int w, int h);
227
    /**
228
     * translational global motion compensation.
229
     */
230
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
231
    /**
232
     * global motion compensation.
233
     */
234
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
235
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
236
    void (*clear_block)(DCTELEM *block/*align 16*/);
237
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
238
    int (*pix_sum)(uint8_t * pix, int line_size);
239
    int (*pix_norm1)(uint8_t * pix, int line_size);
240
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
241

    
242
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
243
    me_cmp_func sse[6];
244
    me_cmp_func hadamard8_diff[6];
245
    me_cmp_func dct_sad[6];
246
    me_cmp_func quant_psnr[6];
247
    me_cmp_func bit[6];
248
    me_cmp_func rd[6];
249
    me_cmp_func vsad[6];
250
    me_cmp_func vsse[6];
251
    me_cmp_func nsse[6];
252
    me_cmp_func w53[6];
253
    me_cmp_func w97[6];
254
    me_cmp_func dct_max[6];
255
    me_cmp_func dct264_sad[6];
256

    
257
    me_cmp_func me_pre_cmp[6];
258
    me_cmp_func me_cmp[6];
259
    me_cmp_func me_sub_cmp[6];
260
    me_cmp_func mb_cmp[6];
261
    me_cmp_func ildct_cmp[6]; //only width 16 used
262
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
263

    
264
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
265
                             int size);
266

    
267
    /**
268
     * Halfpel motion compensation with rounding (a+b+1)>>1.
269
     * this is an array[4][4] of motion compensation functions for 4
270
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
271
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
272
     * @param block destination where the result is stored
273
     * @param pixels source
274
     * @param line_size number of bytes in a horizontal line of block
275
     * @param h height
276
     */
277
    op_pixels_func put_pixels_tab[4][4];
278

    
279
    /**
280
     * Halfpel motion compensation with rounding (a+b+1)>>1.
281
     * This is an array[4][4] of motion compensation functions for 4
282
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
283
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
284
     * @param block destination into which the result is averaged (a+b+1)>>1
285
     * @param pixels source
286
     * @param line_size number of bytes in a horizontal line of block
287
     * @param h height
288
     */
289
    op_pixels_func avg_pixels_tab[4][4];
290

    
291
    /**
292
     * Halfpel motion compensation with no rounding (a+b)>>1.
293
     * this is an array[2][4] of motion compensation functions for 2
294
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
295
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
296
     * @param block destination where the result is stored
297
     * @param pixels source
298
     * @param line_size number of bytes in a horizontal line of block
299
     * @param h height
300
     */
301
    op_pixels_func put_no_rnd_pixels_tab[4][4];
302

    
303
    /**
304
     * Halfpel motion compensation with no rounding (a+b)>>1.
305
     * this is an array[2][4] of motion compensation functions for 2
306
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
307
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
308
     * @param block destination into which the result is averaged (a+b)>>1
309
     * @param pixels source
310
     * @param line_size number of bytes in a horizontal line of block
311
     * @param h height
312
     */
313
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
314

    
315
    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);
316

    
317
    /**
318
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
319
     * this is an array[12] of motion compensation functions for the 9 thirdpe
320
     * positions<br>
321
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
322
     * @param block destination where the result is stored
323
     * @param pixels source
324
     * @param line_size number of bytes in a horizontal line of block
325
     * @param h height
326
     */
327
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
328
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
329

    
330
    qpel_mc_func put_qpel_pixels_tab[2][16];
331
    qpel_mc_func avg_qpel_pixels_tab[2][16];
332
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
333
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
334
    qpel_mc_func put_mspel_pixels_tab[8];
335

    
336
    /**
337
     * h264 Chroma MC
338
     */
339
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
340
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
341

    
342
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
343
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
344

    
345
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
346
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
347

    
348
    me_cmp_func pix_abs[2][4];
349

    
350
    /* huffyuv specific */
351
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
352
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
353
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
354
    /**
355
     * subtract huffyuv's variant of median prediction
356
     * note, this might read from src1[-1], src2[-1]
357
     */
358
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
359
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
360
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
361
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
362
    /* this might write to dst[w] */
363
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
364
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
365
    void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
366

    
367
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
368
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
369

    
370
    void (*h261_loop_filter)(uint8_t *src, int stride);
371

    
372
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
373
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
374

    
375
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
376
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
377
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
378

    
379
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
380
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
381
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
382
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
383
    void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
384
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
385
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
386
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
387
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
388
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
389
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
390
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
391
    /**
392
     * Multiply a vector of floats by a scalar float.  Source and
393
     * destination vectors must overlap exactly or not at all.
394
     * @param dst result vector, 16-byte aligned
395
     * @param src input vector, 16-byte aligned
396
     * @param mul scalar value
397
     * @param len length of vector, multiple of 4
398
     */
399
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
400
                               int len);
401
    /**
402
     * Multiply a vector of floats by concatenated short vectors of
403
     * floats and by a scalar float.  Source and destination vectors
404
     * must overlap exactly or not at all.
405
     * [0]: short vectors of length 2, 8-byte aligned
406
     * [1]: short vectors of length 4, 16-byte aligned
407
     * @param dst output vector, 16-byte aligned
408
     * @param src input vector, 16-byte aligned
409
     * @param sv  array of pointers to short vectors
410
     * @param mul scalar value
411
     * @param len number of elements in src and dst, multiple of 4
412
     */
413
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
414
                                     const float **sv, float mul, int len);
415
    /**
416
     * Multiply short vectors of floats by a scalar float, store
417
     * concatenated result.
418
     * [0]: short vectors of length 2, 8-byte aligned
419
     * [1]: short vectors of length 4, 16-byte aligned
420
     * @param dst output vector, 16-byte aligned
421
     * @param sv  array of pointers to short vectors
422
     * @param mul scalar value
423
     * @param len number of output elements, multiple of 4
424
     */
425
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
426
                              float mul, int len);
427
    /**
428
     * Calculate the scalar product of two vectors of floats.
429
     * @param v1  first vector, 16-byte aligned
430
     * @param v2  second vector, 16-byte aligned
431
     * @param len length of vectors, multiple of 4
432
     */
433
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
434
    /**
435
     * Calculate the sum and difference of two vectors of floats.
436
     * @param v1  first input vector, sum output, 16-byte aligned
437
     * @param v2  second input vector, difference output, 16-byte aligned
438
     * @param len length of vectors, multiple of 4
439
     */
440
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
441

    
442
    /* (I)DCT */
443
    void (*fdct)(DCTELEM *block/* align 16*/);
444
    void (*fdct248)(DCTELEM *block/* align 16*/);
445

    
446
    /* IDCT really*/
447
    void (*idct)(DCTELEM *block/* align 16*/);
448

    
449
    /**
450
     * block -> idct -> clip to unsigned 8 bit -> dest.
451
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
452
     * @param line_size size in bytes of a horizontal line of dest
453
     */
454
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
455

    
456
    /**
457
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
458
     * @param line_size size in bytes of a horizontal line of dest
459
     */
460
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
461

    
462
    /**
463
     * idct input permutation.
464
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
465
     * IDCT)
466
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
467
     * with the zigzag/alternate scan<br>
468
     * an example to avoid confusion:
469
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
470
     * - (x -> referece dct -> reference idct -> x)
471
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
472
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
473
     */
474
    uint8_t idct_permutation[64];
475
    int idct_permutation_type;
476
#define FF_NO_IDCT_PERM 1
477
#define FF_LIBMPEG2_IDCT_PERM 2
478
#define FF_SIMPLE_IDCT_PERM 3
479
#define FF_TRANSPOSE_IDCT_PERM 4
480
#define FF_PARTTRANS_IDCT_PERM 5
481
#define FF_SSE2_IDCT_PERM 6
482

    
483
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
484
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
485
#define BASIS_SHIFT 16
486
#define RECON_SHIFT 6
487

    
488
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int sides);
489
#define EDGE_WIDTH 16
490
#define EDGE_TOP    1
491
#define EDGE_BOTTOM 2
492

    
493
    void (*prefetch)(void *mem, int stride, int h);
494

    
495
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
496

    
497
    /* mlp/truehd functions */
498
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
499
                               int firorder, int iirorder,
500
                               unsigned int filter_shift, int32_t mask, int blocksize,
501
                               int32_t *sample_buffer);
502

    
503
    /* intrax8 functions */
504
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
505
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
506
           int * range, int * sum,  int edges);
507

    
508
    /**
509
     * Calculate scalar product of two vectors.
510
     * @param len length of vectors, should be multiple of 16
511
     * @param shift number of bits to discard from product
512
     */
513
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
514
    /* ape functions */
515
    /**
516
     * Calculate scalar product of v1 and v2,
517
     * and v1[i] += v3[i] * mul
518
     * @param len length of vectors, should be multiple of 16
519
     */
520
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
521

    
522
    /**
523
     * Apply symmetric window in 16-bit fixed-point.
524
     * @param output destination array
525
     *               constraints: 16-byte aligned
526
     * @param input  source array
527
     *               constraints: 16-byte aligned
528
     * @param window window array
529
     *               constraints: 16-byte aligned, at least len/2 elements
530
     * @param len    full window length
531
     *               constraints: multiple of ? greater than zero
532
     */
533
    void (*apply_window_int16)(int16_t *output, const int16_t *input,
534
                               const int16_t *window, unsigned int len);
535

    
536
    /* rv30 functions */
537
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
538
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
539

    
540
    /* rv40 functions */
541
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
542
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
543
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
544
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
545

    
546
    /* bink functions */
547
    op_fill_func fill_block_tab[2];
548
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
549
} DSPContext;
550

    
551
void dsputil_static_init(void);
552
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
553

    
554
int ff_check_alignment(void);
555

    
556
/**
557
 * permute block according to permuatation.
558
 * @param last last non zero element in scantable order
559
 */
560
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
561

    
562
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
563

    
564
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
565

    
566
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
567
{
568
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
569
}
570

    
571
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
572
{
573
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
574
}
575

    
576
static inline int get_penalty_factor(int lambda, int lambda2, int type){
577
    switch(type&0xFF){
578
    default:
579
    case FF_CMP_SAD:
580
        return lambda>>FF_LAMBDA_SHIFT;
581
    case FF_CMP_DCT:
582
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
583
    case FF_CMP_W53:
584
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
585
    case FF_CMP_W97:
586
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
587
    case FF_CMP_SATD:
588
    case FF_CMP_DCT264:
589
        return (2*lambda)>>FF_LAMBDA_SHIFT;
590
    case FF_CMP_RD:
591
    case FF_CMP_PSNR:
592
    case FF_CMP_SSE:
593
    case FF_CMP_NSSE:
594
        return lambda2>>FF_LAMBDA_SHIFT;
595
    case FF_CMP_BIT:
596
        return 1;
597
    }
598
}
599

    
600
/**
601
 * Empty mmx state.
602
 * this must be called between any dsp function and float/double code.
603
 * for example sin(); dsp->idct_put(); emms_c(); cos()
604
 */
605
#define emms_c()
606

    
607
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
608
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
609
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
610
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
611
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
612
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
613
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
614
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
615
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
616

    
617
void ff_dsputil_init_dwt(DSPContext *c);
618
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
619
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
620
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
621
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
622
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
623

    
624
#if HAVE_MMX
625

    
626
#undef emms_c
627

    
628
static inline void emms(void)
629
{
630
    __asm__ volatile ("emms;":::"memory");
631
}
632

    
633
#define emms_c() emms()
634

    
635
#elif ARCH_ARM
636

    
637
#if HAVE_NEON
638
#   define STRIDE_ALIGN 16
639
#endif
640

    
641
#elif ARCH_PPC
642

    
643
#define STRIDE_ALIGN 16
644

    
645
#elif HAVE_MMI
646

    
647
#define STRIDE_ALIGN 16
648

    
649
#endif
650

    
651
#ifndef STRIDE_ALIGN
652
#   define STRIDE_ALIGN 8
653
#endif
654

    
655
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...)             \
656
    uint8_t la_##v[sizeof(t s o) + (a)];                \
657
    t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
658

    
659
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
660

    
661
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
662

    
663
#if HAVE_LOCAL_ALIGNED_8
664
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
665
#else
666
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
667
#endif
668

    
669
#if HAVE_LOCAL_ALIGNED_16
670
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
671
#else
672
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
673
#endif
674

    
675
/* PSNR */
676
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
677
              int orig_linesize[3], int coded_linesize,
678
              AVCodecContext *avctx);
679

    
680
#define WRAPPER8_16(name8, name16)\
681
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
682
    return name8(s, dst           , src           , stride, h)\
683
          +name8(s, dst+8         , src+8         , stride, h);\
684
}
685

    
686
#define WRAPPER8_16_SQ(name8, name16)\
687
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
688
    int score=0;\
689
    score +=name8(s, dst           , src           , stride, 8);\
690
    score +=name8(s, dst+8         , src+8         , stride, 8);\
691
    if(h==16){\
692
        dst += 8*stride;\
693
        src += 8*stride;\
694
        score +=name8(s, dst           , src           , stride, 8);\
695
        score +=name8(s, dst+8         , src+8         , stride, 8);\
696
    }\
697
    return score;\
698
}
699

    
700

    
701
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
702
{
703
    int i;
704
    for(i=0; i<h; i++)
705
    {
706
        AV_WN16(dst   , AV_RN16(src   ));
707
        dst+=dstStride;
708
        src+=srcStride;
709
    }
710
}
711

    
712
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
713
{
714
    int i;
715
    for(i=0; i<h; i++)
716
    {
717
        AV_WN32(dst   , AV_RN32(src   ));
718
        dst+=dstStride;
719
        src+=srcStride;
720
    }
721
}
722

    
723
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
724
{
725
    int i;
726
    for(i=0; i<h; i++)
727
    {
728
        AV_WN32(dst   , AV_RN32(src   ));
729
        AV_WN32(dst+4 , AV_RN32(src+4 ));
730
        dst+=dstStride;
731
        src+=srcStride;
732
    }
733
}
734

    
735
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
736
{
737
    int i;
738
    for(i=0; i<h; i++)
739
    {
740
        AV_WN32(dst   , AV_RN32(src   ));
741
        AV_WN32(dst+4 , AV_RN32(src+4 ));
742
        dst[8]= src[8];
743
        dst+=dstStride;
744
        src+=srcStride;
745
    }
746
}
747

    
748
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
749
{
750
    int i;
751
    for(i=0; i<h; i++)
752
    {
753
        AV_WN32(dst   , AV_RN32(src   ));
754
        AV_WN32(dst+4 , AV_RN32(src+4 ));
755
        AV_WN32(dst+8 , AV_RN32(src+8 ));
756
        AV_WN32(dst+12, AV_RN32(src+12));
757
        dst+=dstStride;
758
        src+=srcStride;
759
    }
760
}
761

    
762
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
763
{
764
    int i;
765
    for(i=0; i<h; i++)
766
    {
767
        AV_WN32(dst   , AV_RN32(src   ));
768
        AV_WN32(dst+4 , AV_RN32(src+4 ));
769
        AV_WN32(dst+8 , AV_RN32(src+8 ));
770
        AV_WN32(dst+12, AV_RN32(src+12));
771
        dst[16]= src[16];
772
        dst+=dstStride;
773
        src+=srcStride;
774
    }
775
}
776

    
777
#endif /* AVCODEC_DSPUTIL_H */