Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 19fb234e

History | View | Annotate | Download (31.2 KB)

1
/*
2
 * DSP utils
3
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22

    
23
/**
24
 * @file
25
 * DSP utils.
26
 * note, many functions in here may use MMX which trashes the FPU state, it is
27
 * absolutely necessary to call emms_c() between dsp & float/double code
28
 */
29

    
30
#ifndef AVCODEC_DSPUTIL_H
31
#define AVCODEC_DSPUTIL_H
32

    
33
#include "libavutil/intreadwrite.h"
34
#include "avcodec.h"
35

    
36

    
37
//#define DEBUG
38
/* dct code */
39
typedef short DCTELEM;
40

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

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

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

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

    
67
void ff_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
void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
72
                             const float *win, float add_bias, int len);
73
void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
74
void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
75

    
76
/* encoding scans */
77
extern const uint8_t ff_alternate_horizontal_scan[64];
78
extern const uint8_t ff_alternate_vertical_scan[64];
79
extern const uint8_t ff_zigzag_direct[64];
80
extern const uint8_t ff_zigzag248_direct[64];
81

    
82
/* pixel operations */
83
#define MAX_NEG_CROP 1024
84

    
85
/* temporary */
86
extern uint32_t ff_squareTbl[512];
87
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
88

    
89
void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
90
void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
91
void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
92
void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
93

    
94
/* VP3 DSP functions */
95
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
96
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
97
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
98
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
99

    
100
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
101
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
102

    
103
/* Bink functions */
104
void ff_bink_idct_c    (DCTELEM *block);
105
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
106
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
107

    
108
/* EA functions */
109
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
110

    
111
/* 1/2^n downscaling functions from imgconvert.c */
112
#if LIBAVCODEC_VERSION_MAJOR < 53
113
/**
114
 * @deprecated Use av_image_copy_plane() instead.
115
 */
116
attribute_deprecated
117
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
118
#endif
119

    
120
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
121
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
122
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
123

    
124
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
125
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
126

    
127
/* minimum alignment rules ;)
128
If you notice errors in the align stuff, need more alignment for some ASM code
129
for some CPU or need to use a function with less aligned data then send a mail
130
to the ffmpeg-devel mailing list, ...
131

132
!warning These alignments might not match reality, (missing attribute((align))
133
stuff somewhere possible).
134
I (Michael) did not check them, these are just the alignments which I think
135
could be reached easily ...
136

137
!future video codecs might need functions with less strict alignment
138
*/
139

    
140
/*
141
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
142
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
143
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
144
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
145
void clear_blocks_c(DCTELEM *blocks);
146
*/
147

    
148
/* add and put pixel (decoding) */
149
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
150
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
151
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
152
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);
153
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
154
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
155

    
156
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
157

    
158
#define DEF_OLD_QPEL(name)\
159
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
160
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
161
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
162

    
163
DEF_OLD_QPEL(qpel16_mc11_old_c)
164
DEF_OLD_QPEL(qpel16_mc31_old_c)
165
DEF_OLD_QPEL(qpel16_mc12_old_c)
166
DEF_OLD_QPEL(qpel16_mc32_old_c)
167
DEF_OLD_QPEL(qpel16_mc13_old_c)
168
DEF_OLD_QPEL(qpel16_mc33_old_c)
169
DEF_OLD_QPEL(qpel8_mc11_old_c)
170
DEF_OLD_QPEL(qpel8_mc31_old_c)
171
DEF_OLD_QPEL(qpel8_mc12_old_c)
172
DEF_OLD_QPEL(qpel8_mc32_old_c)
173
DEF_OLD_QPEL(qpel8_mc13_old_c)
174
DEF_OLD_QPEL(qpel8_mc33_old_c)
175

    
176
#define CALL_2X_PIXELS(a, b, n)\
177
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
178
    b(block  , pixels  , line_size, h);\
179
    b(block+n, pixels+n, line_size, h);\
180
}
181

    
182
/* motion estimation */
183
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
184
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
185
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))*/;
186

    
187
/**
188
 * Scantable.
189
 */
190
typedef struct ScanTable{
191
    const uint8_t *scantable;
192
    uint8_t permutated[64];
193
    uint8_t raster_end[64];
194
#if ARCH_PPC
195
                /** Used by dct_quantize_altivec to find last-non-zero */
196
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
197
#endif
198
} ScanTable;
199

    
200
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
201

    
202
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
203
                         int block_w, int block_h,
204
                         int src_x, int src_y, int w, int h);
205

    
206
/**
207
 * DSPContext.
208
 */
209
typedef struct DSPContext {
210
    /* pixel ops : interface with DCT */
211
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
212
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
213
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
214
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
215
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
216
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
217
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
218
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
219
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
220
    /**
221
     * translational global motion compensation.
222
     */
223
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
224
    /**
225
     * global motion compensation.
226
     */
227
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
228
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
229
    void (*clear_block)(DCTELEM *block/*align 16*/);
230
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
231
    int (*pix_sum)(uint8_t * pix, int line_size);
232
    int (*pix_norm1)(uint8_t * pix, int line_size);
233
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
234

    
235
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
236
    me_cmp_func sse[6];
237
    me_cmp_func hadamard8_diff[6];
238
    me_cmp_func dct_sad[6];
239
    me_cmp_func quant_psnr[6];
240
    me_cmp_func bit[6];
241
    me_cmp_func rd[6];
242
    me_cmp_func vsad[6];
243
    me_cmp_func vsse[6];
244
    me_cmp_func nsse[6];
245
    me_cmp_func w53[6];
246
    me_cmp_func w97[6];
247
    me_cmp_func dct_max[6];
248
    me_cmp_func dct264_sad[6];
249

    
250
    me_cmp_func me_pre_cmp[6];
251
    me_cmp_func me_cmp[6];
252
    me_cmp_func me_sub_cmp[6];
253
    me_cmp_func mb_cmp[6];
254
    me_cmp_func ildct_cmp[6]; //only width 16 used
255
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
256

    
257
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
258
                             int size);
259

    
260
    /**
261
     * Halfpel motion compensation with rounding (a+b+1)>>1.
262
     * this is an array[4][4] of motion compensation functions for 4
263
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
264
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
265
     * @param block destination where the result is stored
266
     * @param pixels source
267
     * @param line_size number of bytes in a horizontal line of block
268
     * @param h height
269
     */
270
    op_pixels_func put_pixels_tab[4][4];
271

    
272
    /**
273
     * Halfpel motion compensation with rounding (a+b+1)>>1.
274
     * This is an array[4][4] of motion compensation functions for 4
275
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
276
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
277
     * @param block destination into which the result is averaged (a+b+1)>>1
278
     * @param pixels source
279
     * @param line_size number of bytes in a horizontal line of block
280
     * @param h height
281
     */
282
    op_pixels_func avg_pixels_tab[4][4];
283

    
284
    /**
285
     * Halfpel motion compensation with no rounding (a+b)>>1.
286
     * this is an array[2][4] of motion compensation functions for 2
287
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
288
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
289
     * @param block destination where the result is stored
290
     * @param pixels source
291
     * @param line_size number of bytes in a horizontal line of block
292
     * @param h height
293
     */
294
    op_pixels_func put_no_rnd_pixels_tab[4][4];
295

    
296
    /**
297
     * Halfpel motion compensation with no rounding (a+b)>>1.
298
     * this is an array[2][4] of motion compensation functions for 2
299
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
300
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
301
     * @param block destination into which the result is averaged (a+b)>>1
302
     * @param pixels source
303
     * @param line_size number of bytes in a horizontal line of block
304
     * @param h height
305
     */
306
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
307

    
308
    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);
309

    
310
    /**
311
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
312
     * this is an array[12] of motion compensation functions for the 9 thirdpe
313
     * positions<br>
314
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
315
     * @param block destination where the result is stored
316
     * @param pixels source
317
     * @param line_size number of bytes in a horizontal line of block
318
     * @param h height
319
     */
320
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
321
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
322

    
323
    qpel_mc_func put_qpel_pixels_tab[2][16];
324
    qpel_mc_func avg_qpel_pixels_tab[2][16];
325
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
326
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
327
    qpel_mc_func put_mspel_pixels_tab[8];
328

    
329
    /**
330
     * h264 Chroma MC
331
     */
332
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
333
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
334
    /* This is really one func used in VC-1 decoding */
335
    h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
336
    h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
337

    
338
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
339
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
340

    
341
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
342
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
343

    
344
    me_cmp_func pix_abs[2][4];
345

    
346
    /* huffyuv specific */
347
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
348
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
349
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
350
    /**
351
     * subtract huffyuv's variant of median prediction
352
     * note, this might read from src1[-1], src2[-1]
353
     */
354
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
355
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
356
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
357
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
358
    /* this might write to dst[w] */
359
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
360
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
361

    
362
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
363
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
364

    
365
    void (*h261_loop_filter)(uint8_t *src, int stride);
366

    
367
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
368
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
369

    
370
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
371
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
372
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
373

    
374
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
375
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
376
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
377
    /* no alignment needed */
378
    void (*lpc_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
379
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
380
    void (*vector_fmul)(float *dst, const float *src, int len);
381
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
382
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
383
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
384
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
385
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
386
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
387
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
388
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
389
    /**
390
     * Multiply a vector of floats by a scalar float.  Source and
391
     * destination vectors must overlap exactly or not at all.
392
     * @param dst result vector, 16-byte aligned
393
     * @param src input vector, 16-byte aligned
394
     * @param mul scalar value
395
     * @param len length of vector, multiple of 4
396
     */
397
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
398
                               int len);
399
    /**
400
     * Multiply a vector of floats by concatenated short vectors of
401
     * floats and by a scalar float.  Source and destination vectors
402
     * must overlap exactly or not at all.
403
     * [0]: short vectors of length 2, 8-byte aligned
404
     * [1]: short vectors of length 4, 16-byte aligned
405
     * @param dst output vector, 16-byte aligned
406
     * @param src input vector, 16-byte aligned
407
     * @param sv  array of pointers to short vectors
408
     * @param mul scalar value
409
     * @param len number of elements in src and dst, multiple of 4
410
     */
411
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
412
                                     const float **sv, float mul, int len);
413
    /**
414
     * Multiply short vectors of floats by a scalar float, store
415
     * concatenated result.
416
     * [0]: short vectors of length 2, 8-byte aligned
417
     * [1]: short vectors of length 4, 16-byte aligned
418
     * @param dst output vector, 16-byte aligned
419
     * @param sv  array of pointers to short vectors
420
     * @param mul scalar value
421
     * @param len number of output elements, multiple of 4
422
     */
423
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
424
                              float mul, int len);
425
    /**
426
     * Calculate the scalar product of two vectors of floats.
427
     * @param v1  first vector, 16-byte aligned
428
     * @param v2  second vector, 16-byte aligned
429
     * @param len length of vectors, multiple of 4
430
     */
431
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
432
    /**
433
     * Calculate the sum and difference of two vectors of floats.
434
     * @param v1  first input vector, sum output, 16-byte aligned
435
     * @param v2  second input vector, difference output, 16-byte aligned
436
     * @param len length of vectors, multiple of 4
437
     */
438
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
439

    
440
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
441
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
442
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
443
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
444

    
445
    /* (I)DCT */
446
    void (*fdct)(DCTELEM *block/* align 16*/);
447
    void (*fdct248)(DCTELEM *block/* align 16*/);
448

    
449
    /* IDCT really*/
450
    void (*idct)(DCTELEM *block/* align 16*/);
451

    
452
    /**
453
     * block -> idct -> clip to unsigned 8 bit -> dest.
454
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
455
     * @param line_size size in bytes of a horizontal line of dest
456
     */
457
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
458

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

    
465
    /**
466
     * idct input permutation.
467
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
468
     * IDCT)
469
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
470
     * with the zigzag/alternate scan<br>
471
     * an example to avoid confusion:
472
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
473
     * - (x -> referece dct -> reference idct -> x)
474
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
475
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
476
     */
477
    uint8_t idct_permutation[64];
478
    int idct_permutation_type;
479
#define FF_NO_IDCT_PERM 1
480
#define FF_LIBMPEG2_IDCT_PERM 2
481
#define FF_SIMPLE_IDCT_PERM 3
482
#define FF_TRANSPOSE_IDCT_PERM 4
483
#define FF_PARTTRANS_IDCT_PERM 5
484
#define FF_SSE2_IDCT_PERM 6
485

    
486
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
487
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
488
#define BASIS_SHIFT 16
489
#define RECON_SHIFT 6
490

    
491
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
492
#define EDGE_WIDTH 16
493

    
494
    void (*prefetch)(void *mem, int stride, int h);
495

    
496
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
497

    
498
    /* mlp/truehd functions */
499
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
500
                               int firorder, int iirorder,
501
                               unsigned int filter_shift, int32_t mask, int blocksize,
502
                               int32_t *sample_buffer);
503

    
504
    /* vc1 functions */
505
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
506
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
507
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
508
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
509
    void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
510
    void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
511
    void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
512
    void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
513
    void (*vc1_v_overlap)(uint8_t* src, int stride);
514
    void (*vc1_h_overlap)(uint8_t* src, int stride);
515
    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
516
    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
517
    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
518
    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
519
    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
520
    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
521
    /* put 8x8 block with bicubic interpolation and quarterpel precision
522
     * last argument is actually round value instead of height
523
     */
524
    op_pixels_func put_vc1_mspel_pixels_tab[16];
525
    op_pixels_func avg_vc1_mspel_pixels_tab[16];
526

    
527
    /* intrax8 functions */
528
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
529
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
530
           int * range, int * sum,  int edges);
531

    
532
    /**
533
     * Calculate scalar product of two vectors.
534
     * @param len length of vectors, should be multiple of 16
535
     * @param shift number of bits to discard from product
536
     */
537
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
538
    /* ape functions */
539
    /**
540
     * Calculate scalar product of v1 and v2,
541
     * and v1[i] += v3[i] * mul
542
     * @param len length of vectors, should be multiple of 16
543
     */
544
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
545

    
546
    /* rv30 functions */
547
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
548
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
549

    
550
    /* rv40 functions */
551
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
552
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
553
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
554
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
555

    
556
    /* bink functions */
557
    op_fill_func fill_block_tab[2];
558
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
559
} DSPContext;
560

    
561
void dsputil_static_init(void);
562
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
563

    
564
int ff_check_alignment(void);
565

    
566
/**
567
 * permute block according to permuatation.
568
 * @param last last non zero element in scantable order
569
 */
570
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
571

    
572
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
573

    
574
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
575

    
576
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
577
{
578
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
579
}
580

    
581
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
582
{
583
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
584
}
585

    
586
static inline int get_penalty_factor(int lambda, int lambda2, int type){
587
    switch(type&0xFF){
588
    default:
589
    case FF_CMP_SAD:
590
        return lambda>>FF_LAMBDA_SHIFT;
591
    case FF_CMP_DCT:
592
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
593
    case FF_CMP_W53:
594
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
595
    case FF_CMP_W97:
596
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
597
    case FF_CMP_SATD:
598
    case FF_CMP_DCT264:
599
        return (2*lambda)>>FF_LAMBDA_SHIFT;
600
    case FF_CMP_RD:
601
    case FF_CMP_PSNR:
602
    case FF_CMP_SSE:
603
    case FF_CMP_NSSE:
604
        return lambda2>>FF_LAMBDA_SHIFT;
605
    case FF_CMP_BIT:
606
        return 1;
607
    }
608
}
609

    
610
/**
611
 * Empty mmx state.
612
 * this must be called between any dsp function and float/double code.
613
 * for example sin(); dsp->idct_put(); emms_c(); cos()
614
 */
615
#define emms_c()
616

    
617
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
618
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
619
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
620
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
621
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
622
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
623
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
624
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
625
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
626

    
627
void ff_dsputil_init_dwt(DSPContext *c);
628
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
629
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
630
void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
631
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
632
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
633
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
634

    
635
#if HAVE_MMX
636

    
637
#undef emms_c
638

    
639
static inline void emms(void)
640
{
641
    __asm__ volatile ("emms;":::"memory");
642
}
643

    
644
#define emms_c() emms()
645

    
646
#elif ARCH_ARM
647

    
648
#if HAVE_NEON
649
#   define STRIDE_ALIGN 16
650
#endif
651

    
652
#elif ARCH_PPC
653

    
654
#define STRIDE_ALIGN 16
655

    
656
#elif HAVE_MMI
657

    
658
#define STRIDE_ALIGN 16
659

    
660
#endif
661

    
662
#ifndef STRIDE_ALIGN
663
#   define STRIDE_ALIGN 8
664
#endif
665

    
666
#define LOCAL_ALIGNED(a, t, v, s, ...)                          \
667
    uint8_t la_##v[sizeof(t s __VA_ARGS__) + (a)];              \
668
    t (*v) __VA_ARGS__ = (void *)FFALIGN((uintptr_t)la_##v, a)
669

    
670
#if HAVE_LOCAL_ALIGNED_8
671
#   define LOCAL_ALIGNED_8(t, v, s, ...) DECLARE_ALIGNED(8, t, v) s __VA_ARGS__
672
#else
673
#   define LOCAL_ALIGNED_8(t, v, s, ...) LOCAL_ALIGNED(8, t, v, s, __VA_ARGS__)
674
#endif
675

    
676
#if HAVE_LOCAL_ALIGNED_16
677
#   define LOCAL_ALIGNED_16(t, v, s, ...) DECLARE_ALIGNED(16, t, v) s __VA_ARGS__
678
#else
679
#   define LOCAL_ALIGNED_16(t, v, s, ...) LOCAL_ALIGNED(16, t, v, s, __VA_ARGS__)
680
#endif
681

    
682
/* PSNR */
683
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
684
              int orig_linesize[3], int coded_linesize,
685
              AVCodecContext *avctx);
686

    
687
#define WRAPPER8_16(name8, name16)\
688
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
689
    return name8(s, dst           , src           , stride, h)\
690
          +name8(s, dst+8         , src+8         , stride, h);\
691
}
692

    
693
#define WRAPPER8_16_SQ(name8, name16)\
694
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
695
    int score=0;\
696
    score +=name8(s, dst           , src           , stride, 8);\
697
    score +=name8(s, dst+8         , src+8         , stride, 8);\
698
    if(h==16){\
699
        dst += 8*stride;\
700
        src += 8*stride;\
701
        score +=name8(s, dst           , src           , stride, 8);\
702
        score +=name8(s, dst+8         , src+8         , stride, 8);\
703
    }\
704
    return score;\
705
}
706

    
707

    
708
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
709
{
710
    int i;
711
    for(i=0; i<h; i++)
712
    {
713
        AV_WN16(dst   , AV_RN16(src   ));
714
        dst+=dstStride;
715
        src+=srcStride;
716
    }
717
}
718

    
719
static inline void copy_block4(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_WN32(dst   , AV_RN32(src   ));
725
        dst+=dstStride;
726
        src+=srcStride;
727
    }
728
}
729

    
730
static inline void copy_block8(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
        AV_WN32(dst+4 , AV_RN32(src+4 ));
737
        dst+=dstStride;
738
        src+=srcStride;
739
    }
740
}
741

    
742
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
743
{
744
    int i;
745
    for(i=0; i<h; i++)
746
    {
747
        AV_WN32(dst   , AV_RN32(src   ));
748
        AV_WN32(dst+4 , AV_RN32(src+4 ));
749
        dst[8]= src[8];
750
        dst+=dstStride;
751
        src+=srcStride;
752
    }
753
}
754

    
755
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
756
{
757
    int i;
758
    for(i=0; i<h; i++)
759
    {
760
        AV_WN32(dst   , AV_RN32(src   ));
761
        AV_WN32(dst+4 , AV_RN32(src+4 ));
762
        AV_WN32(dst+8 , AV_RN32(src+8 ));
763
        AV_WN32(dst+12, AV_RN32(src+12));
764
        dst+=dstStride;
765
        src+=srcStride;
766
    }
767
}
768

    
769
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
770
{
771
    int i;
772
    for(i=0; i<h; i++)
773
    {
774
        AV_WN32(dst   , AV_RN32(src   ));
775
        AV_WN32(dst+4 , AV_RN32(src+4 ));
776
        AV_WN32(dst+8 , AV_RN32(src+8 ));
777
        AV_WN32(dst+12, AV_RN32(src+12));
778
        dst[16]= src[16];
779
        dst+=dstStride;
780
        src+=srcStride;
781
    }
782
}
783

    
784
#endif /* AVCODEC_DSPUTIL_H */