Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ af0b2d67

History | View | Annotate | Download (30.7 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_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
#if LIBAVCODEC_VERSION_MAJOR < 53
109
/**
110
 * @deprecated Use av_image_copy_plane() instead.
111
 */
112
attribute_deprecated
113
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
114
#endif
115

    
116
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
117
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
118
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
119

    
120
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
121
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
122

    
123
/* minimum alignment rules ;)
124
If you notice errors in the align stuff, need more alignment for some ASM code
125
for some CPU or need to use a function with less aligned data then send a mail
126
to the ffmpeg-devel mailing list, ...
127

128
!warning These alignments might not match reality, (missing attribute((align))
129
stuff somewhere possible).
130
I (Michael) did not check them, these are just the alignments which I think
131
could be reached easily ...
132

133
!future video codecs might need functions with less strict alignment
134
*/
135

    
136
/*
137
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
138
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
139
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
140
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
141
void clear_blocks_c(DCTELEM *blocks);
142
*/
143

    
144
/* add and put pixel (decoding) */
145
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
146
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
147
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
148
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);
149
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
150
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
151

    
152
typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
153

    
154
#define DEF_OLD_QPEL(name)\
155
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
156
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
157
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
158

    
159
DEF_OLD_QPEL(qpel16_mc11_old_c)
160
DEF_OLD_QPEL(qpel16_mc31_old_c)
161
DEF_OLD_QPEL(qpel16_mc12_old_c)
162
DEF_OLD_QPEL(qpel16_mc32_old_c)
163
DEF_OLD_QPEL(qpel16_mc13_old_c)
164
DEF_OLD_QPEL(qpel16_mc33_old_c)
165
DEF_OLD_QPEL(qpel8_mc11_old_c)
166
DEF_OLD_QPEL(qpel8_mc31_old_c)
167
DEF_OLD_QPEL(qpel8_mc12_old_c)
168
DEF_OLD_QPEL(qpel8_mc32_old_c)
169
DEF_OLD_QPEL(qpel8_mc13_old_c)
170
DEF_OLD_QPEL(qpel8_mc33_old_c)
171

    
172
#define CALL_2X_PIXELS(a, b, n)\
173
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
174
    b(block  , pixels  , line_size, h);\
175
    b(block+n, pixels+n, line_size, h);\
176
}
177

    
178
/* motion estimation */
179
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
180
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
181
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))*/;
182

    
183
/**
184
 * Scantable.
185
 */
186
typedef struct ScanTable{
187
    const uint8_t *scantable;
188
    uint8_t permutated[64];
189
    uint8_t raster_end[64];
190
#if ARCH_PPC
191
                /** Used by dct_quantize_altivec to find last-non-zero */
192
    DECLARE_ALIGNED(16, uint8_t, inverse)[64];
193
#endif
194
} ScanTable;
195

    
196
void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
197

    
198
void ff_emulated_edge_mc(uint8_t *buf, const uint8_t *src, int linesize,
199
                         int block_w, int block_h,
200
                         int src_x, int src_y, int w, int h);
201

    
202
void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
203
void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
204
void ff_put_signed_pixels_clamped_c(const DCTELEM *block, uint8_t *dest, int linesize);
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
     * Motion estimation with emulated edge values.
222
     * @param buf pointer to destination buffer (unaligned)
223
     * @param src pointer to pixel source (unaligned)
224
     * @param linesize width (in pixels) for src/buf
225
     * @param block_w number of pixels (per row) to copy to buf
226
     * @param block_h nummber of pixel rows to copy to buf
227
     * @param src_x offset of src to start of row - this may be negative
228
     * @param src_y offset of src to top of image - this may be negative
229
     * @param w width of src in pixels
230
     * @param h height of src in pixels
231
     */
232
    void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, int linesize,
233
                             int block_w, int block_h,
234
                             int src_x, int src_y, int w, int h);
235
    /**
236
     * translational global motion compensation.
237
     */
238
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
239
    /**
240
     * global motion compensation.
241
     */
242
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
243
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
244
    void (*clear_block)(DCTELEM *block/*align 16*/);
245
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
246
    int (*pix_sum)(uint8_t * pix, int line_size);
247
    int (*pix_norm1)(uint8_t * pix, int line_size);
248
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
249

    
250
    me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
251
    me_cmp_func sse[6];
252
    me_cmp_func hadamard8_diff[6];
253
    me_cmp_func dct_sad[6];
254
    me_cmp_func quant_psnr[6];
255
    me_cmp_func bit[6];
256
    me_cmp_func rd[6];
257
    me_cmp_func vsad[6];
258
    me_cmp_func vsse[6];
259
    me_cmp_func nsse[6];
260
    me_cmp_func w53[6];
261
    me_cmp_func w97[6];
262
    me_cmp_func dct_max[6];
263
    me_cmp_func dct264_sad[6];
264

    
265
    me_cmp_func me_pre_cmp[6];
266
    me_cmp_func me_cmp[6];
267
    me_cmp_func me_sub_cmp[6];
268
    me_cmp_func mb_cmp[6];
269
    me_cmp_func ildct_cmp[6]; //only width 16 used
270
    me_cmp_func frame_skip_cmp[6]; //only width 8 used
271

    
272
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
273
                             int size);
274

    
275
    /**
276
     * Halfpel motion compensation with rounding (a+b+1)>>1.
277
     * this is an array[4][4] of motion compensation functions for 4
278
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
279
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
280
     * @param block destination where the result is stored
281
     * @param pixels source
282
     * @param line_size number of bytes in a horizontal line of block
283
     * @param h height
284
     */
285
    op_pixels_func put_pixels_tab[4][4];
286

    
287
    /**
288
     * Halfpel motion compensation with rounding (a+b+1)>>1.
289
     * This is an array[4][4] of motion compensation functions for 4
290
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
291
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
292
     * @param block destination into which the result is averaged (a+b+1)>>1
293
     * @param pixels source
294
     * @param line_size number of bytes in a horizontal line of block
295
     * @param h height
296
     */
297
    op_pixels_func avg_pixels_tab[4][4];
298

    
299
    /**
300
     * Halfpel motion compensation with no rounding (a+b)>>1.
301
     * this is an array[2][4] of motion compensation functions for 2
302
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
303
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
304
     * @param block destination where the result is stored
305
     * @param pixels source
306
     * @param line_size number of bytes in a horizontal line of block
307
     * @param h height
308
     */
309
    op_pixels_func put_no_rnd_pixels_tab[4][4];
310

    
311
    /**
312
     * Halfpel motion compensation with no rounding (a+b)>>1.
313
     * this is an array[2][4] of motion compensation functions for 2
314
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
315
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
316
     * @param block destination into which the result is averaged (a+b)>>1
317
     * @param pixels source
318
     * @param line_size number of bytes in a horizontal line of block
319
     * @param h height
320
     */
321
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
322

    
323
    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);
324

    
325
    /**
326
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
327
     * this is an array[12] of motion compensation functions for the 9 thirdpe
328
     * positions<br>
329
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
330
     * @param block destination where the result is stored
331
     * @param pixels source
332
     * @param line_size number of bytes in a horizontal line of block
333
     * @param h height
334
     */
335
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
336
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
337

    
338
    qpel_mc_func put_qpel_pixels_tab[2][16];
339
    qpel_mc_func avg_qpel_pixels_tab[2][16];
340
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
341
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
342
    qpel_mc_func put_mspel_pixels_tab[8];
343

    
344
    /**
345
     * h264 Chroma MC
346
     */
347
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
348
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
349

    
350
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
351
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
352

    
353
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
354
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
355

    
356
    me_cmp_func pix_abs[2][4];
357

    
358
    /* huffyuv specific */
359
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
360
    void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
361
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
362
    /**
363
     * subtract huffyuv's variant of median prediction
364
     * note, this might read from src1[-1], src2[-1]
365
     */
366
    void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
367
    void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
368
    int  (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
369
    void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
370
    /* this might write to dst[w] */
371
    void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
372
    void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
373
    void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
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, int len);
397
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
398
    void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
399
    /**
400
     * Multiply a vector of floats by a scalar float.  Source and
401
     * destination vectors must overlap exactly or not at all.
402
     * @param dst result vector, 16-byte aligned
403
     * @param src input vector, 16-byte aligned
404
     * @param mul scalar value
405
     * @param len length of vector, multiple of 4
406
     */
407
    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
408
                               int len);
409
    /**
410
     * Multiply a vector of floats by concatenated short vectors of
411
     * floats and by a scalar float.  Source and destination vectors
412
     * must overlap exactly or not at all.
413
     * [0]: short vectors of length 2, 8-byte aligned
414
     * [1]: short vectors of length 4, 16-byte aligned
415
     * @param dst output vector, 16-byte aligned
416
     * @param src input vector, 16-byte aligned
417
     * @param sv  array of pointers to short vectors
418
     * @param mul scalar value
419
     * @param len number of elements in src and dst, multiple of 4
420
     */
421
    void (*vector_fmul_sv_scalar[2])(float *dst, const float *src,
422
                                     const float **sv, float mul, int len);
423
    /**
424
     * Multiply short vectors of floats by a scalar float, store
425
     * concatenated result.
426
     * [0]: short vectors of length 2, 8-byte aligned
427
     * [1]: short vectors of length 4, 16-byte aligned
428
     * @param dst output vector, 16-byte aligned
429
     * @param sv  array of pointers to short vectors
430
     * @param mul scalar value
431
     * @param len number of output elements, multiple of 4
432
     */
433
    void (*sv_fmul_scalar[2])(float *dst, const float **sv,
434
                              float mul, int len);
435
    /**
436
     * Calculate the scalar product of two vectors of floats.
437
     * @param v1  first vector, 16-byte aligned
438
     * @param v2  second vector, 16-byte aligned
439
     * @param len length of vectors, multiple of 4
440
     */
441
    float (*scalarproduct_float)(const float *v1, const float *v2, int len);
442
    /**
443
     * Calculate the sum and difference of two vectors of floats.
444
     * @param v1  first input vector, sum output, 16-byte aligned
445
     * @param v2  second input vector, difference output, 16-byte aligned
446
     * @param len length of vectors, multiple of 4
447
     */
448
    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
449

    
450
    /* (I)DCT */
451
    void (*fdct)(DCTELEM *block/* align 16*/);
452
    void (*fdct248)(DCTELEM *block/* align 16*/);
453

    
454
    /* IDCT really*/
455
    void (*idct)(DCTELEM *block/* align 16*/);
456

    
457
    /**
458
     * block -> idct -> clip to unsigned 8 bit -> dest.
459
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
460
     * @param line_size size in bytes of a horizontal line of dest
461
     */
462
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
463

    
464
    /**
465
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
466
     * @param line_size size in bytes of a horizontal line of dest
467
     */
468
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
469

    
470
    /**
471
     * idct input permutation.
472
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
473
     * IDCT)
474
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
475
     * with the zigzag/alternate scan<br>
476
     * an example to avoid confusion:
477
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
478
     * - (x -> referece dct -> reference idct -> x)
479
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
480
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
481
     */
482
    uint8_t idct_permutation[64];
483
    int idct_permutation_type;
484
#define FF_NO_IDCT_PERM 1
485
#define FF_LIBMPEG2_IDCT_PERM 2
486
#define FF_SIMPLE_IDCT_PERM 3
487
#define FF_TRANSPOSE_IDCT_PERM 4
488
#define FF_PARTTRANS_IDCT_PERM 5
489
#define FF_SSE2_IDCT_PERM 6
490

    
491
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
492
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
493
#define BASIS_SHIFT 16
494
#define RECON_SHIFT 6
495

    
496
    void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int sides);
497
#define EDGE_WIDTH 16
498
#define EDGE_TOP    1
499
#define EDGE_BOTTOM 2
500

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

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

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

    
511
    /* intrax8 functions */
512
    void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
513
    void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
514
           int * range, int * sum,  int edges);
515

    
516
    /**
517
     * Calculate scalar product of two vectors.
518
     * @param len length of vectors, should be multiple of 16
519
     * @param shift number of bits to discard from product
520
     */
521
    int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len, int shift);
522
    /* ape functions */
523
    /**
524
     * Calculate scalar product of v1 and v2,
525
     * and v1[i] += v3[i] * mul
526
     * @param len length of vectors, should be multiple of 16
527
     */
528
    int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
529

    
530
    /**
531
     * Apply symmetric window in 16-bit fixed-point.
532
     * @param output destination array
533
     *               constraints: 16-byte aligned
534
     * @param input  source array
535
     *               constraints: 16-byte aligned
536
     * @param window window array
537
     *               constraints: 16-byte aligned, at least len/2 elements
538
     * @param len    full window length
539
     *               constraints: multiple of ? greater than zero
540
     */
541
    void (*apply_window_int16)(int16_t *output, const int16_t *input,
542
                               const int16_t *window, unsigned int len);
543

    
544
    /* rv30 functions */
545
    qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
546
    qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
547

    
548
    /* rv40 functions */
549
    qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
550
    qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
551
    h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
552
    h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
553

    
554
    /* bink functions */
555
    op_fill_func fill_block_tab[2];
556
    void (*scale_block)(const uint8_t src[64]/*align 8*/, uint8_t *dst/*align 8*/, int linesize);
557
} DSPContext;
558

    
559
void dsputil_static_init(void);
560
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
561

    
562
int ff_check_alignment(void);
563

    
564
/**
565
 * permute block according to permuatation.
566
 * @param last last non zero element in scantable order
567
 */
568
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
569

    
570
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
571

    
572
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
573

    
574
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
575
{
576
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
577
}
578

    
579
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
580
{
581
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
582
}
583

    
584
static inline int get_penalty_factor(int lambda, int lambda2, int type){
585
    switch(type&0xFF){
586
    default:
587
    case FF_CMP_SAD:
588
        return lambda>>FF_LAMBDA_SHIFT;
589
    case FF_CMP_DCT:
590
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
591
    case FF_CMP_W53:
592
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
593
    case FF_CMP_W97:
594
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
595
    case FF_CMP_SATD:
596
    case FF_CMP_DCT264:
597
        return (2*lambda)>>FF_LAMBDA_SHIFT;
598
    case FF_CMP_RD:
599
    case FF_CMP_PSNR:
600
    case FF_CMP_SSE:
601
    case FF_CMP_NSSE:
602
        return lambda2>>FF_LAMBDA_SHIFT;
603
    case FF_CMP_BIT:
604
        return 1;
605
    }
606
}
607

    
608
/**
609
 * Empty mmx state.
610
 * this must be called between any dsp function and float/double code.
611
 * for example sin(); dsp->idct_put(); emms_c(); cos()
612
 */
613
#define emms_c()
614

    
615
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
616
void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
617
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
618
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
619
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
620
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
621
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
622
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
623
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
624

    
625
void ff_dsputil_init_dwt(DSPContext *c);
626
void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
627
void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
628
void ff_intrax8dsp_init(DSPContext* c, AVCodecContext *avctx);
629
void ff_mlp_init(DSPContext* c, AVCodecContext *avctx);
630
void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
631

    
632
#if HAVE_MMX
633

    
634
#undef emms_c
635

    
636
static inline void emms(void)
637
{
638
    __asm__ volatile ("emms;":::"memory");
639
}
640

    
641
#define emms_c() \
642
{\
643
    if(av_get_cpu_flags() & AV_CPU_FLAG_MMX)\
644
        emms();\
645
}
646

    
647
#elif ARCH_ARM
648

    
649
#if HAVE_NEON
650
#   define STRIDE_ALIGN 16
651
#endif
652

    
653
#elif ARCH_PPC
654

    
655
#define STRIDE_ALIGN 16
656

    
657
#elif HAVE_MMI
658

    
659
#define STRIDE_ALIGN 16
660

    
661
#endif
662

    
663
#ifndef STRIDE_ALIGN
664
#   define STRIDE_ALIGN 8
665
#endif
666

    
667
#define LOCAL_ALIGNED_A(a, t, v, s, o, ...)             \
668
    uint8_t la_##v[sizeof(t s o) + (a)];                \
669
    t (*v) o = (void *)FFALIGN((uintptr_t)la_##v, a)
670

    
671
#define LOCAL_ALIGNED_D(a, t, v, s, o, ...) DECLARE_ALIGNED(a, t, v) s o
672

    
673
#define LOCAL_ALIGNED(a, t, v, ...) LOCAL_ALIGNED_A(a, t, v, __VA_ARGS__,,)
674

    
675
#if HAVE_LOCAL_ALIGNED_8
676
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED_D(8, t, v, __VA_ARGS__,,)
677
#else
678
#   define LOCAL_ALIGNED_8(t, v, ...) LOCAL_ALIGNED(8, t, v, __VA_ARGS__)
679
#endif
680

    
681
#if HAVE_LOCAL_ALIGNED_16
682
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED_D(16, t, v, __VA_ARGS__,,)
683
#else
684
#   define LOCAL_ALIGNED_16(t, v, ...) LOCAL_ALIGNED(16, t, v, __VA_ARGS__)
685
#endif
686

    
687
/* PSNR */
688
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
689
              int orig_linesize[3], int coded_linesize,
690
              AVCodecContext *avctx);
691

    
692
#define WRAPPER8_16(name8, name16)\
693
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
694
    return name8(s, dst           , src           , stride, h)\
695
          +name8(s, dst+8         , src+8         , stride, h);\
696
}
697

    
698
#define WRAPPER8_16_SQ(name8, name16)\
699
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
700
    int score=0;\
701
    score +=name8(s, dst           , src           , stride, 8);\
702
    score +=name8(s, dst+8         , src+8         , stride, 8);\
703
    if(h==16){\
704
        dst += 8*stride;\
705
        src += 8*stride;\
706
        score +=name8(s, dst           , src           , stride, 8);\
707
        score +=name8(s, dst+8         , src+8         , stride, 8);\
708
    }\
709
    return score;\
710
}
711

    
712

    
713
static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
714
{
715
    int i;
716
    for(i=0; i<h; i++)
717
    {
718
        AV_WN16(dst   , AV_RN16(src   ));
719
        dst+=dstStride;
720
        src+=srcStride;
721
    }
722
}
723

    
724
static inline void copy_block4(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
725
{
726
    int i;
727
    for(i=0; i<h; i++)
728
    {
729
        AV_WN32(dst   , AV_RN32(src   ));
730
        dst+=dstStride;
731
        src+=srcStride;
732
    }
733
}
734

    
735
static inline void copy_block8(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+=dstStride;
743
        src+=srcStride;
744
    }
745
}
746

    
747
static inline void copy_block9(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
748
{
749
    int i;
750
    for(i=0; i<h; i++)
751
    {
752
        AV_WN32(dst   , AV_RN32(src   ));
753
        AV_WN32(dst+4 , AV_RN32(src+4 ));
754
        dst[8]= src[8];
755
        dst+=dstStride;
756
        src+=srcStride;
757
    }
758
}
759

    
760
static inline void copy_block16(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
761
{
762
    int i;
763
    for(i=0; i<h; i++)
764
    {
765
        AV_WN32(dst   , AV_RN32(src   ));
766
        AV_WN32(dst+4 , AV_RN32(src+4 ));
767
        AV_WN32(dst+8 , AV_RN32(src+8 ));
768
        AV_WN32(dst+12, AV_RN32(src+12));
769
        dst+=dstStride;
770
        src+=srcStride;
771
    }
772
}
773

    
774
static inline void copy_block17(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
775
{
776
    int i;
777
    for(i=0; i<h; i++)
778
    {
779
        AV_WN32(dst   , AV_RN32(src   ));
780
        AV_WN32(dst+4 , AV_RN32(src+4 ));
781
        AV_WN32(dst+8 , AV_RN32(src+8 ));
782
        AV_WN32(dst+12, AV_RN32(src+12));
783
        dst[16]= src[16];
784
        dst+=dstStride;
785
        src+=srcStride;
786
    }
787
}
788

    
789
#endif /* AVCODEC_DSPUTIL_H */