Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 755bfeab

History | View | Annotate | Download (27.4 KB)

1
/*
2
 * DSP utils
3
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6
 * This file is part of 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 dsputil.h
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 DSPUTIL_H
31
#define DSPUTIL_H
32

    
33
#include "avcodec.h"
34

    
35

    
36
//#define DEBUG
37
/* dct code */
38
typedef short DCTELEM;
39
typedef int DWTELEM;
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

    
51
void ff_fdct_mmx(DCTELEM *block);
52
void ff_fdct_mmx2(DCTELEM *block);
53
void ff_fdct_sse2(DCTELEM *block);
54

    
55
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
56
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
57
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
58
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
59
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
60
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
61

    
62
void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
63
                              const float *src2, int src3, int blocksize, int step);
64
void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
65

    
66
/* encoding scans */
67
extern const uint8_t ff_alternate_horizontal_scan[64];
68
extern const uint8_t ff_alternate_vertical_scan[64];
69
extern const uint8_t ff_zigzag_direct[64];
70
extern const uint8_t ff_zigzag248_direct[64];
71

    
72
/* pixel operations */
73
#define MAX_NEG_CROP 1024
74

    
75
/* temporary */
76
extern uint32_t ff_squareTbl[512];
77
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
78

    
79
/* VP3 DSP functions */
80
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
81
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
82
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
83

    
84
/* 1/2^n downscaling functions from imgconvert.c */
85
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
86
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
87
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
88
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
89

    
90
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
91
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
92

    
93
/* minimum alignment rules ;)
94
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
95
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
96

97
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
98
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
99

100
!future video codecs might need functions with less strict alignment
101
*/
102

    
103
/*
104
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
105
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
106
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
107
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
108
void clear_blocks_c(DCTELEM *blocks);
109
*/
110

    
111
/* add and put pixel (decoding) */
112
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
113
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
114
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
115
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);
116
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
117
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
118
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
119
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
120

    
121
#define DEF_OLD_QPEL(name)\
122
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
123
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
124
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
125

    
126
DEF_OLD_QPEL(qpel16_mc11_old_c)
127
DEF_OLD_QPEL(qpel16_mc31_old_c)
128
DEF_OLD_QPEL(qpel16_mc12_old_c)
129
DEF_OLD_QPEL(qpel16_mc32_old_c)
130
DEF_OLD_QPEL(qpel16_mc13_old_c)
131
DEF_OLD_QPEL(qpel16_mc33_old_c)
132
DEF_OLD_QPEL(qpel8_mc11_old_c)
133
DEF_OLD_QPEL(qpel8_mc31_old_c)
134
DEF_OLD_QPEL(qpel8_mc12_old_c)
135
DEF_OLD_QPEL(qpel8_mc32_old_c)
136
DEF_OLD_QPEL(qpel8_mc13_old_c)
137
DEF_OLD_QPEL(qpel8_mc33_old_c)
138

    
139
#define CALL_2X_PIXELS(a, b, n)\
140
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
141
    b(block  , pixels  , line_size, h);\
142
    b(block+n, pixels+n, line_size, h);\
143
}
144

    
145
/* motion estimation */
146
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
147
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
148
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))*/;
149

    
150

    
151
// for snow slices
152
typedef struct slice_buffer_s slice_buffer;
153

    
154
/**
155
 * DSPContext.
156
 */
157
typedef struct DSPContext {
158
    /* pixel ops : interface with DCT */
159
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
160
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
161
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
162
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
163
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
164
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
165
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
166
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
167
    /**
168
     * translational global motion compensation.
169
     */
170
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
171
    /**
172
     * global motion compensation.
173
     */
174
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
175
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
176
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
177
    int (*pix_sum)(uint8_t * pix, int line_size);
178
    int (*pix_norm1)(uint8_t * pix, int line_size);
179
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
180

    
181
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
182
    me_cmp_func sse[5];
183
    me_cmp_func hadamard8_diff[5];
184
    me_cmp_func dct_sad[5];
185
    me_cmp_func quant_psnr[5];
186
    me_cmp_func bit[5];
187
    me_cmp_func rd[5];
188
    me_cmp_func vsad[5];
189
    me_cmp_func vsse[5];
190
    me_cmp_func nsse[5];
191
    me_cmp_func w53[5];
192
    me_cmp_func w97[5];
193
    me_cmp_func dct_max[5];
194
    me_cmp_func dct264_sad[5];
195

    
196
    me_cmp_func me_pre_cmp[5];
197
    me_cmp_func me_cmp[5];
198
    me_cmp_func me_sub_cmp[5];
199
    me_cmp_func mb_cmp[5];
200
    me_cmp_func ildct_cmp[5]; //only width 16 used
201
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
202

    
203
    int (*ssd_int8_vs_int16)(int8_t *pix1, int16_t *pix2, int size);
204

    
205
    /**
206
     * Halfpel motion compensation with rounding (a+b+1)>>1.
207
     * this is an array[4][4] of motion compensation functions for 4
208
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
209
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
210
     * @param block destination where the result is stored
211
     * @param pixels source
212
     * @param line_size number of bytes in a horizontal line of block
213
     * @param h height
214
     */
215
    op_pixels_func put_pixels_tab[4][4];
216

    
217
    /**
218
     * Halfpel motion compensation with rounding (a+b+1)>>1.
219
     * This is an array[4][4] of motion compensation functions for 4
220
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
221
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
222
     * @param block destination into which the result is averaged (a+b+1)>>1
223
     * @param pixels source
224
     * @param line_size number of bytes in a horizontal line of block
225
     * @param h height
226
     */
227
    op_pixels_func avg_pixels_tab[4][4];
228

    
229
    /**
230
     * Halfpel motion compensation with no rounding (a+b)>>1.
231
     * this is an array[2][4] of motion compensation functions for 2
232
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
233
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
234
     * @param block destination where the result is stored
235
     * @param pixels source
236
     * @param line_size number of bytes in a horizontal line of block
237
     * @param h height
238
     */
239
    op_pixels_func put_no_rnd_pixels_tab[4][4];
240

    
241
    /**
242
     * Halfpel motion compensation with no rounding (a+b)>>1.
243
     * this is an array[2][4] of motion compensation functions for 2
244
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
245
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
246
     * @param block destination into which the result is averaged (a+b)>>1
247
     * @param pixels source
248
     * @param line_size number of bytes in a horizontal line of block
249
     * @param h height
250
     */
251
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
252

    
253
    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);
254

    
255
    /**
256
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
257
     * this is an array[12] of motion compensation functions for the 9 thirdpe
258
     * positions<br>
259
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
260
     * @param block destination where the result is stored
261
     * @param pixels source
262
     * @param line_size number of bytes in a horizontal line of block
263
     * @param h height
264
     */
265
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
266
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
267

    
268
    qpel_mc_func put_qpel_pixels_tab[2][16];
269
    qpel_mc_func avg_qpel_pixels_tab[2][16];
270
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
271
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
272
    qpel_mc_func put_mspel_pixels_tab[8];
273

    
274
    /**
275
     * h264 Chroma MC
276
     */
277
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
278
    /* This is really one func used in VC-1 decoding */
279
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
280
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
281

    
282
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
283
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
284

    
285
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
286
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
287

    
288
    h264_weight_func weight_h264_pixels_tab[10];
289
    h264_biweight_func biweight_h264_pixels_tab[10];
290

    
291
    /* AVS specific */
292
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
293
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
294
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
295
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
296
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
297
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
298
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
299

    
300
    me_cmp_func pix_abs[2][4];
301

    
302
    /* huffyuv specific */
303
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
304
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
305
    /**
306
     * subtract huffyuv's variant of median prediction
307
     * note, this might read from src1[-1], src2[-1]
308
     */
309
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
310
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
311

    
312
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
313
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
314
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
315
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
316
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
317
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
318
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
319
    void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
320
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);
321

    
322
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
323
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
324

    
325
    void (*h261_loop_filter)(uint8_t *src, int stride);
326

    
327
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
328
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
329
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
330
    void (*vector_fmul)(float *dst, const float *src, int len);
331
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
332
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
333
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
334

    
335
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
336
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
337
    void (*float_to_int16)(int16_t *dst, const float *src, int len);
338

    
339
    /* (I)DCT */
340
    void (*fdct)(DCTELEM *block/* align 16*/);
341
    void (*fdct248)(DCTELEM *block/* align 16*/);
342

    
343
    /* IDCT really*/
344
    void (*idct)(DCTELEM *block/* align 16*/);
345

    
346
    /**
347
     * block -> idct -> clip to unsigned 8 bit -> dest.
348
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
349
     * @param line_size size in bytes of a horizotal line of dest
350
     */
351
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
352

    
353
    /**
354
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
355
     * @param line_size size in bytes of a horizotal line of dest
356
     */
357
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
358

    
359
    /**
360
     * idct input permutation.
361
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
362
     * IDCT)
363
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
364
     * with the zigzag/alternate scan<br>
365
     * an example to avoid confusion:
366
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
367
     * - (x -> referece dct -> reference idct -> x)
368
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
369
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
370
     */
371
    uint8_t idct_permutation[64];
372
    int idct_permutation_type;
373
#define FF_NO_IDCT_PERM 1
374
#define FF_LIBMPEG2_IDCT_PERM 2
375
#define FF_SIMPLE_IDCT_PERM 3
376
#define FF_TRANSPOSE_IDCT_PERM 4
377
#define FF_PARTTRANS_IDCT_PERM 5
378

    
379
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
380
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
381
#define BASIS_SHIFT 16
382
#define RECON_SHIFT 6
383

    
384
    /* h264 functions */
385
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
386
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
387
    void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
388
    void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
389
    void (*h264_dct)(DCTELEM block[4][4]);
390

    
391
    /* snow wavelet */
392
    void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
393
    void (*horizontal_compose97i)(DWTELEM *b, int width);
394
    void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
395

    
396
    void (*prefetch)(void *mem, int stride, int h);
397

    
398
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
399

    
400
    /* vc1 functions */
401
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
402
    void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
403
    void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
404
    void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
405
    void (*vc1_v_overlap)(uint8_t* src, int stride);
406
    void (*vc1_h_overlap)(uint8_t* src, int stride);
407
    /* put 8x8 block with bicubic interpolation and quarterpel precision
408
     * last argument is actually round value instead of height
409
     */
410
    op_pixels_func put_vc1_mspel_pixels_tab[16];
411
} DSPContext;
412

    
413
void dsputil_static_init(void);
414
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
415

    
416
int ff_check_alignment(void);
417

    
418
/**
419
 * permute block according to permuatation.
420
 * @param last last non zero element in scantable order
421
 */
422
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
423

    
424
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
425

    
426
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
427

    
428
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
429
{
430
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
431
}
432

    
433
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
434
{
435
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
436
}
437

    
438
static inline int get_penalty_factor(int lambda, int lambda2, int type){
439
    switch(type&0xFF){
440
    default:
441
    case FF_CMP_SAD:
442
        return lambda>>FF_LAMBDA_SHIFT;
443
    case FF_CMP_DCT:
444
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
445
    case FF_CMP_W53:
446
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
447
    case FF_CMP_W97:
448
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
449
    case FF_CMP_SATD:
450
    case FF_CMP_DCT264:
451
        return (2*lambda)>>FF_LAMBDA_SHIFT;
452
    case FF_CMP_RD:
453
    case FF_CMP_PSNR:
454
    case FF_CMP_SSE:
455
    case FF_CMP_NSSE:
456
        return lambda2>>FF_LAMBDA_SHIFT;
457
    case FF_CMP_BIT:
458
        return 1;
459
    }
460
}
461

    
462
/**
463
 * Empty mmx state.
464
 * this must be called between any dsp function and float/double code.
465
 * for example sin(); dsp->idct_put(); emms_c(); cos()
466
 */
467
#define emms_c()
468

    
469
/* should be defined by architectures supporting
470
   one or more MultiMedia extension */
471
int mm_support(void);
472

    
473
#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
474

    
475
#if defined(HAVE_MMX)
476

    
477
#undef emms_c
478

    
479
#define MM_MMX    0x0001 /* standard MMX */
480
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
481
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
482
#define MM_SSE    0x0008 /* SSE functions */
483
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
484
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
485
#define MM_SSE3   0x0040 /* Prescott SSE3 functions */
486
#define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
487

    
488
extern int mm_flags;
489

    
490
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
491
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
492
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
493

    
494
static inline void emms(void)
495
{
496
    __asm __volatile ("emms;":::"memory");
497
}
498

    
499

    
500
#define emms_c() \
501
{\
502
    if (mm_flags & MM_MMX)\
503
        emms();\
504
}
505

    
506
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
507

    
508
#define STRIDE_ALIGN 8
509

    
510
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
511
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
512

    
513
#elif defined(ARCH_ARMV4L)
514

    
515
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
516
   line optimizations */
517
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(4, t, v)
518
#define STRIDE_ALIGN 4
519

    
520
#define MM_IWMMXT    0x0100 /* XScale IWMMXT */
521

    
522
extern int mm_flags;
523

    
524
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
525

    
526
#elif defined(HAVE_MLIB)
527

    
528
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
529
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
530
#define STRIDE_ALIGN 8
531

    
532
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
533

    
534
#elif defined(ARCH_SPARC)
535

    
536
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
537
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
538
#define STRIDE_ALIGN 8
539
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
540

    
541
#elif defined(ARCH_ALPHA)
542

    
543
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
544
#define STRIDE_ALIGN 8
545

    
546
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
547

    
548
#elif defined(ARCH_POWERPC)
549

    
550
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
551

    
552
extern int mm_flags;
553

    
554
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
555
#define pixel altivec_pixel
556
#include <altivec.h>
557
#undef pixel
558
#endif
559

    
560
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
561
#define STRIDE_ALIGN 16
562

    
563
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
564

    
565
#elif defined(HAVE_MMI)
566

    
567
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
568
#define STRIDE_ALIGN 16
569

    
570
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
571

    
572
#elif defined(ARCH_SH4)
573

    
574
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
575
#define STRIDE_ALIGN 8
576

    
577
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
578

    
579
#elif defined(ARCH_BFIN)
580

    
581
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
582
#define STRIDE_ALIGN 8
583

    
584
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
585

    
586
#else
587

    
588
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
589
#define STRIDE_ALIGN 8
590

    
591
#endif
592

    
593
/* PSNR */
594
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
595
              int orig_linesize[3], int coded_linesize,
596
              AVCodecContext *avctx);
597

    
598
/* FFT computation */
599

    
600
/* NOTE: soon integer code will be added, so you must use the
601
   FFTSample type */
602
typedef float FFTSample;
603

    
604
struct MDCTContext;
605

    
606
typedef struct FFTComplex {
607
    FFTSample re, im;
608
} FFTComplex;
609

    
610
typedef struct FFTContext {
611
    int nbits;
612
    int inverse;
613
    uint16_t *revtab;
614
    FFTComplex *exptab;
615
    FFTComplex *exptab1; /* only used by SSE code */
616
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
617
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
618
                       const FFTSample *input, FFTSample *tmp);
619
} FFTContext;
620

    
621
int ff_fft_init(FFTContext *s, int nbits, int inverse);
622
void ff_fft_permute(FFTContext *s, FFTComplex *z);
623
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
624
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
625
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
626
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
627
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
628

    
629
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
630
{
631
    s->fft_calc(s, z);
632
}
633
void ff_fft_end(FFTContext *s);
634

    
635
/* MDCT computation */
636

    
637
typedef struct MDCTContext {
638
    int n;  /* size of MDCT (i.e. number of input data * 2) */
639
    int nbits; /* n = 2^nbits */
640
    /* pre/post rotation tables */
641
    FFTSample *tcos;
642
    FFTSample *tsin;
643
    FFTContext fft;
644
} MDCTContext;
645

    
646
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
647
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
648
                const FFTSample *input, FFTSample *tmp);
649
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
650
                        const FFTSample *input, FFTSample *tmp);
651
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
652
                       const FFTSample *input, FFTSample *tmp);
653
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
654
               const FFTSample *input, FFTSample *tmp);
655
void ff_mdct_end(MDCTContext *s);
656

    
657
#define WARPER8_16(name8, name16)\
658
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
659
    return name8(s, dst           , src           , stride, h)\
660
          +name8(s, dst+8         , src+8         , stride, h);\
661
}
662

    
663
#define WARPER8_16_SQ(name8, name16)\
664
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
665
    int score=0;\
666
    score +=name8(s, dst           , src           , stride, 8);\
667
    score +=name8(s, dst+8         , src+8         , stride, 8);\
668
    if(h==16){\
669
        dst += 8*stride;\
670
        src += 8*stride;\
671
        score +=name8(s, dst           , src           , stride, 8);\
672
        score +=name8(s, dst+8         , src+8         , stride, 8);\
673
    }\
674
    return score;\
675
}
676

    
677

    
678
static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
679
{
680
    int i;
681
    for(i=0; i<h; i++)
682
    {
683
        ST16(dst   , LD16(src   ));
684
        dst+=dstStride;
685
        src+=srcStride;
686
    }
687
}
688

    
689
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
690
{
691
    int i;
692
    for(i=0; i<h; i++)
693
    {
694
        ST32(dst   , LD32(src   ));
695
        dst+=dstStride;
696
        src+=srcStride;
697
    }
698
}
699

    
700
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
701
{
702
    int i;
703
    for(i=0; i<h; i++)
704
    {
705
        ST32(dst   , LD32(src   ));
706
        ST32(dst+4 , LD32(src+4 ));
707
        dst+=dstStride;
708
        src+=srcStride;
709
    }
710
}
711

    
712
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
713
{
714
    int i;
715
    for(i=0; i<h; i++)
716
    {
717
        ST32(dst   , LD32(src   ));
718
        ST32(dst+4 , LD32(src+4 ));
719
        dst[8]= src[8];
720
        dst+=dstStride;
721
        src+=srcStride;
722
    }
723
}
724

    
725
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
726
{
727
    int i;
728
    for(i=0; i<h; i++)
729
    {
730
        ST32(dst   , LD32(src   ));
731
        ST32(dst+4 , LD32(src+4 ));
732
        ST32(dst+8 , LD32(src+8 ));
733
        ST32(dst+12, LD32(src+12));
734
        dst+=dstStride;
735
        src+=srcStride;
736
    }
737
}
738

    
739
static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
740
{
741
    int i;
742
    for(i=0; i<h; i++)
743
    {
744
        ST32(dst   , LD32(src   ));
745
        ST32(dst+4 , LD32(src+4 ));
746
        ST32(dst+8 , LD32(src+8 ));
747
        ST32(dst+12, LD32(src+12));
748
        dst[16]= src[16];
749
        dst+=dstStride;
750
        src+=srcStride;
751
    }
752
}
753

    
754
#endif