Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ d23e3e5f

History | View | Annotate | Download (31.7 KB)

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

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

    
30
#ifndef AVCODEC_DSPUTIL_H
31
#define AVCODEC_DSPUTIL_H
32

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

    
36

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

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

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

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

    
56
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

    
74
/* encoding scans */
75
extern const uint8_t ff_alternate_horizontal_scan[64];
76
extern const uint8_t ff_alternate_vertical_scan[64];
77
extern const uint8_t ff_zigzag_direct[64];
78
extern const uint8_t ff_zigzag248_direct[64];
79

    
80
/* pixel operations */
81
#define MAX_NEG_CROP 1024
82

    
83
/* temporary */
84
extern uint32_t ff_squareTbl[512];
85
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
86

    
87
void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
88
void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, int stride);
89
void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
90
void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, int stride);
91

    
92
/* VP3 DSP functions */
93
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
94
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
95
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
96
void ff_vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
97

    
98
void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
99
void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
100

    
101
/* Bink functions */
102
void ff_bink_idct_c    (DCTELEM *block);
103
void ff_bink_idct_add_c(uint8_t *dest, int linesize, DCTELEM *block);
104
void ff_bink_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
105

    
106
/* EA functions */
107
void ff_ea_idct_put_c(uint8_t *dest, int linesize, DCTELEM *block);
108

    
109
/* 1/2^n downscaling functions from imgconvert.c */
110
#if LIBAVCODEC_VERSION_MAJOR < 53
111
/**
112
 * @deprecated Use av_image_copy_plane() instead.
113
 */
114
attribute_deprecated
115
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
116
#endif
117

    
118
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
119
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
120
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
121

    
122
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
123
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
124

    
125
/* minimum alignment rules ;)
126
If you notice errors in the align stuff, need more alignment for some ASM code
127
for some CPU or need to use a function with less aligned data then send a mail
128
to the ffmpeg-devel mailing list, ...
129

130
!warning These alignments might not match reality, (missing attribute((align))
131
stuff somewhere possible).
132
I (Michael) did not check them, these are just the alignments which I think
133
could be reached easily ...
134

135
!future video codecs might need functions with less strict alignment
136
*/
137

    
138
/*
139
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
140
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
141
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
142
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
143
void clear_blocks_c(DCTELEM *blocks);
144
*/
145

    
146
/* add and put pixel (decoding) */
147
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
148
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
149
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
150
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);
151
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
152
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
153

    
154
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
155

    
156
#define DEF_OLD_QPEL(name)\
157
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
158
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
159
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
160

    
161
DEF_OLD_QPEL(qpel16_mc11_old_c)
162
DEF_OLD_QPEL(qpel16_mc31_old_c)
163
DEF_OLD_QPEL(qpel16_mc12_old_c)
164
DEF_OLD_QPEL(qpel16_mc32_old_c)
165
DEF_OLD_QPEL(qpel16_mc13_old_c)
166
DEF_OLD_QPEL(qpel16_mc33_old_c)
167
DEF_OLD_QPEL(qpel8_mc11_old_c)
168
DEF_OLD_QPEL(qpel8_mc31_old_c)
169
DEF_OLD_QPEL(qpel8_mc12_old_c)
170
DEF_OLD_QPEL(qpel8_mc32_old_c)
171
DEF_OLD_QPEL(qpel8_mc13_old_c)
172
DEF_OLD_QPEL(qpel8_mc33_old_c)
173

    
174
#define CALL_2X_PIXELS(a, b, n)\
175
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
176
    b(block  , pixels  , line_size, h);\
177
    b(block+n, pixels+n, line_size, h);\
178
}
179

    
180
/* motion estimation */
181
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
182
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
183
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))*/;
184

    
185
/**
186
 * Scantable.
187
 */
188
typedef struct ScanTable{
189
    const uint8_t *scantable;
190
    uint8_t permutated[64];
191
    uint8_t raster_end[64];
192
#if ARCH_PPC
193
                /** Used by dct_quantize_altivec to find last-non-zero */
194
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
195
#endif
196
} ScanTable;
197

    
198
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
199

    
200
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
201
                         int block_w, int block_h,
202
                         int src_x, int src_y, int w, int h);
203

    
204
/**
205
 * DSPContext.
206
 */
207
typedef struct DSPContext {
208
    /* pixel ops : interface with DCT */
209
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
210
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
211
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
212
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
213
    void (*put_pixels_nonclamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
214
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
215
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
216
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
217
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
218
    /**
219
     * Motion estimation with emulated edge values.
220
     * @param buf pointer to destination buffer (unaligned)
221
     * @param src pointer to pixel source (unaligned)
222
     * @param linesize width (in pixels) for src/buf
223
     * @param block_w number of pixels (per row) to copy to buf
224
     * @param block_h nummber of pixel rows to copy to buf
225
     * @param src_x offset of src to start of row - this may be negative
226
     * @param src_y offset of src to top of image - this may be negative
227
     * @param w width of src in pixels
228
     * @param h height of src in pixels
229
     */
230
    void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
231
                             int block_w, int block_h,
232
                             int src_x, int src_y, int w, int h);
233
    /**
234
     * translational global motion compensation.
235
     */
236
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
237
    /**
238
     * global motion compensation.
239
     */
240
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
241
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
242
    void (*clear_block)(DCTELEM *block/*align 16*/);
243
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
244
    int (*pix_sum)(uint8_t * pix, int line_size);
245
    int (*pix_norm1)(uint8_t * pix, int line_size);
246
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
247

    
248
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
249
    me_cmp_func sse[6];
250
    me_cmp_func hadamard8_diff[6];
251
    me_cmp_func dct_sad[6];
252
    me_cmp_func quant_psnr[6];
253
    me_cmp_func bit[6];
254
    me_cmp_func rd[6];
255
    me_cmp_func vsad[6];
256
    me_cmp_func vsse[6];
257
    me_cmp_func nsse[6];
258
    me_cmp_func w53[6];
259
    me_cmp_func w97[6];
260
    me_cmp_func dct_max[6];
261
    me_cmp_func dct264_sad[6];
262

    
263
    me_cmp_func me_pre_cmp[6];
264
    me_cmp_func me_cmp[6];
265
    me_cmp_func me_sub_cmp[6];
266
    me_cmp_func mb_cmp[6];
267
    me_cmp_func ildct_cmp[6]; //only width 16 used
268
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
269

    
270
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
271
                             int size);
272

    
273
    /**
274
     * Halfpel motion compensation with rounding (a+b+1)>>1.
275
     * this is an array[4][4] of motion compensation functions for 4
276
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
277
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
278
     * @param block destination where the result is stored
279
     * @param pixels source
280
     * @param line_size number of bytes in a horizontal line of block
281
     * @param h height
282
     */
283
    op_pixels_func put_pixels_tab[4][4];
284

    
285
    /**
286
     * Halfpel motion compensation with rounding (a+b+1)>>1.
287
     * This is an array[4][4] of motion compensation functions for 4
288
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
289
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
290
     * @param block destination into which the result is averaged (a+b+1)>>1
291
     * @param pixels source
292
     * @param line_size number of bytes in a horizontal line of block
293
     * @param h height
294
     */
295
    op_pixels_func avg_pixels_tab[4][4];
296

    
297
    /**
298
     * Halfpel motion compensation with no rounding (a+b)>>1.
299
     * this is an array[2][4] of motion compensation functions for 2
300
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
301
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
302
     * @param block destination where the result is stored
303
     * @param pixels source
304
     * @param line_size number of bytes in a horizontal line of block
305
     * @param h height
306
     */
307
    op_pixels_func put_no_rnd_pixels_tab[4][4];
308

    
309
    /**
310
     * Halfpel motion compensation with no rounding (a+b)>>1.
311
     * this is an array[2][4] of motion compensation functions for 2
312
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
313
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
314
     * @param block destination into which the result is averaged (a+b)>>1
315
     * @param pixels source
316
     * @param line_size number of bytes in a horizontal line of block
317
     * @param h height
318
     */
319
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
320

    
321
    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);
322

    
323
    /**
324
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
325
     * this is an array[12] of motion compensation functions for the 9 thirdpe
326
     * positions<br>
327
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
328
     * @param block destination where the result is stored
329
     * @param pixels source
330
     * @param line_size number of bytes in a horizontal line of block
331
     * @param h height
332
     */
333
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
334
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
335

    
336
    qpel_mc_func put_qpel_pixels_tab[2][16];
337
    qpel_mc_func avg_qpel_pixels_tab[2][16];
338
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
339
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
340
    qpel_mc_func put_mspel_pixels_tab[8];
341

    
342
    /**
343
     * h264 Chroma MC
344
     */
345
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
346
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
347
    /* This is really one func used in VC-1 decoding */
348
    h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3];
349
    h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3];
350

    
351
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
352
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
353

    
354
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
355
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
356

    
357
    me_cmp_func pix_abs[2][4];
358

    
359
    /* huffyuv specific */
360
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
361
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
362
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
363
    /**
364
     * subtract huffyuv's variant of median prediction
365
     * note, this might read from src1[-1], src2[-1]
366
     */
367
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
368
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
369
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
370
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
371
    /* this might write to dst[w] */
372
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
373
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
374

    
375
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
376
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
377

    
378
    void (*h261_loop_filter)(uint8_t *src, int stride);
379

    
380
    void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
381
    void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
382

    
383
    void (*vp3_idct_dc_add)(uint8_t *dest/*align 8*/, int line_size, const DCTELEM *block/*align 16*/);
384
    void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
385
    void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
386

    
387
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
388
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
389
    void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
390
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
391
    void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
392
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
393
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
394
    void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
395
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
396
    void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
397
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
398
    void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
399
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
400
    /**
401
     * Multiply a vector of floats by a scalar float.  Source and
402
     * destination vectors must overlap exactly or not at all.
403
     * @param dst result vector, 16-byte aligned
404
     * @param src input vector, 16-byte aligned
405
     * @param mul scalar value
406
     * @param len length of vector, multiple of 4
407
     */
408
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
409
                               int len);
410
    /**
411
     * Multiply a vector of floats by concatenated short vectors of
412
     * floats and by a scalar float.  Source and destination vectors
413
     * must overlap exactly or not at all.
414
     * [0]: short vectors of length 2, 8-byte aligned
415
     * [1]: short vectors of length 4, 16-byte aligned
416
     * @param dst output vector, 16-byte aligned
417
     * @param src input vector, 16-byte aligned
418
     * @param sv  array of pointers to short vectors
419
     * @param mul scalar value
420
     * @param len number of elements in src and dst, multiple of 4
421
     */
422
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
423
                                     const float **sv, float mul, int len);
424
    /**
425
     * Multiply short vectors of floats by a scalar float, store
426
     * concatenated result.
427
     * [0]: short vectors of length 2, 8-byte aligned
428
     * [1]: short vectors of length 4, 16-byte aligned
429
     * @param dst output vector, 16-byte aligned
430
     * @param sv  array of pointers to short vectors
431
     * @param mul scalar value
432
     * @param len number of output elements, multiple of 4
433
     */
434
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
435
                              float mul, int len);
436
    /**
437
     * Calculate the scalar product of two vectors of floats.
438
     * @param v1  first vector, 16-byte aligned
439
     * @param v2  second vector, 16-byte aligned
440
     * @param len length of vectors, multiple of 4
441
     */
442
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
443
    /**
444
     * Calculate the sum and difference of two vectors of floats.
445
     * @param v1  first input vector, sum output, 16-byte aligned
446
     * @param v2  second input vector, difference output, 16-byte aligned
447
     * @param len length of vectors, multiple of 4
448
     */
449
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
450

    
451
    /* convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
452
    void (*float_to_int16)(int16_t *dst, const float *src, long len);
453
    void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
454

    
455
    /* (I)DCT */
456
    void (*fdct)(DCTELEM *block/* align 16*/);
457
    void (*fdct248)(DCTELEM *block/* align 16*/);
458

    
459
    /* IDCT really*/
460
    void (*idct)(DCTELEM *block/* align 16*/);
461

    
462
    /**
463
     * block -> idct -> clip to unsigned 8 bit -> dest.
464
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
465
     * @param line_size size in bytes of a horizontal line of dest
466
     */
467
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
468

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

    
475
    /**
476
     * idct input permutation.
477
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
478
     * IDCT)
479
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
480
     * with the zigzag/alternate scan<br>
481
     * an example to avoid confusion:
482
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
483
     * - (x -> referece dct -> reference idct -> x)
484
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
485
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
486
     */
487
    uint8_t idct_permutation[64];
488
    int idct_permutation_type;
489
#define FF_NO_IDCT_PERM 1
490
#define FF_LIBMPEG2_IDCT_PERM 2
491
#define FF_SIMPLE_IDCT_PERM 3
492
#define FF_TRANSPOSE_IDCT_PERM 4
493
#define FF_PARTTRANS_IDCT_PERM 5
494
#define FF_SSE2_IDCT_PERM 6
495

    
496
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
497
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
498
#define BASIS_SHIFT 16
499
#define RECON_SHIFT 6
500

    
501
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
502
#define EDGE_WIDTH 16
503

    
504
    void (*prefetch)(void *mem, int stride, int h);
505

    
506
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
507

    
508
    /* mlp/truehd functions */
509
    void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff,
510
                               int firorder, int iirorder,
511
                               unsigned int filter_shift, int32_t mask, int blocksize,
512
                               int32_t *sample_buffer);
513

    
514
    /* vc1 functions */
515
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
516
    void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
517
    void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
518
    void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
519
    void (*vc1_inv_trans_8x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
520
    void (*vc1_inv_trans_8x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
521
    void (*vc1_inv_trans_4x8_dc)(uint8_t *dest, int line_size, DCTELEM *block);
522
    void (*vc1_inv_trans_4x4_dc)(uint8_t *dest, int line_size, DCTELEM *block);
523
    void (*vc1_v_overlap)(uint8_t* src, int stride);
524
    void (*vc1_h_overlap)(uint8_t* src, int stride);
525
    void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
526
    void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
527
    void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);
528
    void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq);
529
    void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq);
530
    void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq);
531
    /* put 8x8 block with bicubic interpolation and quarterpel precision
532
     * last argument is actually round value instead of height
533
     */
534
    op_pixels_func put_vc1_mspel_pixels_tab[16];
535
    op_pixels_func avg_vc1_mspel_pixels_tab[16];
536

    
537
    /* intrax8 functions */
538
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
539
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
540
           int * range, int * sum,  int edges);
541

    
542
    /**
543
     * Calculate scalar product of two vectors.
544
     * @param len length of vectors, should be multiple of 16
545
     * @param shift number of bits to discard from product
546
     */
547
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
548
    /* ape functions */
549
    /**
550
     * Calculate scalar product of v1 and v2,
551
     * and v1[i] += v3[i] * mul
552
     * @param len length of vectors, should be multiple of 16
553
     */
554
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
555

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

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

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

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

    
574
int ff_check_alignment(void);
575

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

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

    
584
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
585

    
586
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
587
{
588
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
589
}
590

    
591
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
592
{
593
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
594
}
595

    
596
static inline int get_penalty_factor(int lambda, int lambda2, int type){
597
    switch(type&0xFF){
598
    default:
599
    case FF_CMP_SAD:
600
        return lambda>>FF_LAMBDA_SHIFT;
601
    case FF_CMP_DCT:
602
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
603
    case FF_CMP_W53:
604
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
605
    case FF_CMP_W97:
606
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
607
    case FF_CMP_SATD:
608
    case FF_CMP_DCT264:
609
        return (2*lambda)>>FF_LAMBDA_SHIFT;
610
    case FF_CMP_RD:
611
    case FF_CMP_PSNR:
612
    case FF_CMP_SSE:
613
    case FF_CMP_NSSE:
614
        return lambda2>>FF_LAMBDA_SHIFT;
615
    case FF_CMP_BIT:
616
        return 1;
617
    }
618
}
619

    
620
/**
621
 * Empty mmx state.
622
 * this must be called between any dsp function and float/double code.
623
 * for example sin(); dsp->idct_put(); emms_c(); cos()
624
 */
625
#define emms_c()
626

    
627
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
628
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
629
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
630
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
631
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
632
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
633
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
634
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
635
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
636

    
637
void ff_dsputil_init_dwt(DSPContext *c);
638
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
639
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
640
void ff_vc1dsp_init(DSPContext* c, AVCodecContext *avctx);
641
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
642
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
643
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
644

    
645
#if HAVE_MMX
646

    
647
#undef emms_c
648

    
649
static inline void emms(void)
650
{
651
    __asm__ volatile ("emms;":::"memory");
652
}
653

    
654
#define emms_c() emms()
655

    
656
#elif ARCH_ARM
657

    
658
#if HAVE_NEON
659
#   define STRIDE_ALIGN 16
660
#endif
661

    
662
#elif ARCH_PPC
663

    
664
#define STRIDE_ALIGN 16
665

    
666
#elif HAVE_MMI
667

    
668
#define STRIDE_ALIGN 16
669

    
670
#endif
671

    
672
#ifndef STRIDE_ALIGN
673
#   define STRIDE_ALIGN 8
674
#endif
675

    
676
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...)             \
677
    uint8_t la_##v[sizeof(t s o) + (a)];                \
678
    t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
679

    
680
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
681

    
682
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
683

    
684
#if HAVE_LOCAL_ALIGNED_8
685
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
686
#else
687
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
688
#endif
689

    
690
#if HAVE_LOCAL_ALIGNED_16
691
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
692
#else
693
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
694
#endif
695

    
696
/* PSNR */
697
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
698
              int orig_linesize[3], int coded_linesize,
699
              AVCodecContext *avctx);
700

    
701
#define WRAPPER8_16(name8, name16)\
702
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
703
    return name8(s, dst           , src           , stride, h)\
704
          +name8(s, dst+8         , src+8         , stride, h);\
705
}
706

    
707
#define WRAPPER8_16_SQ(name8, name16)\
708
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
709
    int score=0;\
710
    score +=name8(s, dst           , src           , stride, 8);\
711
    score +=name8(s, dst+8         , src+8         , stride, 8);\
712
    if(h==16){\
713
        dst += 8*stride;\
714
        src += 8*stride;\
715
        score +=name8(s, dst           , src           , stride, 8);\
716
        score +=name8(s, dst+8         , src+8         , stride, 8);\
717
    }\
718
    return score;\
719
}
720

    
721

    
722
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
723
{
724
    int i;
725
    for(i=0; i<h; i++)
726
    {
727
        AV_WN16(dst   , AV_RN16(src   ));
728
        dst+=dstStride;
729
        src+=srcStride;
730
    }
731
}
732

    
733
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
734
{
735
    int i;
736
    for(i=0; i<h; i++)
737
    {
738
        AV_WN32(dst   , AV_RN32(src   ));
739
        dst+=dstStride;
740
        src+=srcStride;
741
    }
742
}
743

    
744
static inline void copy_block8(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
745
{
746
    int i;
747
    for(i=0; i<h; i++)
748
    {
749
        AV_WN32(dst   , AV_RN32(src   ));
750
        AV_WN32(dst+4 , AV_RN32(src+4 ));
751
        dst+=dstStride;
752
        src+=srcStride;
753
    }
754
}
755

    
756
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
757
{
758
    int i;
759
    for(i=0; i<h; i++)
760
    {
761
        AV_WN32(dst   , AV_RN32(src   ));
762
        AV_WN32(dst+4 , AV_RN32(src+4 ));
763
        dst[8]= src[8];
764
        dst+=dstStride;
765
        src+=srcStride;
766
    }
767
}
768

    
769
static inline void copy_block16(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+=dstStride;
779
        src+=srcStride;
780
    }
781
}
782

    
783
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
784
{
785
    int i;
786
    for(i=0; i<h; i++)
787
    {
788
        AV_WN32(dst   , AV_RN32(src   ));
789
        AV_WN32(dst+4 , AV_RN32(src+4 ));
790
        AV_WN32(dst+8 , AV_RN32(src+8 ));
791
        AV_WN32(dst+12, AV_RN32(src+12));
792
        dst[16]= src[16];
793
        dst+=dstStride;
794
        src+=srcStride;
795
    }
796
}
797

    
798
#endif /* AVCODEC_DSPUTIL_H */