Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ a41f4eac

History | View | Annotate | Download (27.6 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
typedef short IDWTELEM;
41

    
42
void fdct_ifast (DCTELEM *data);
43
void fdct_ifast248 (DCTELEM *data);
44
void ff_jpeg_fdct_islow (DCTELEM *data);
45
void ff_fdct248_islow (DCTELEM *data);
46

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

    
63
void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
64
                              const float *src2, int src3, int blocksize, int step);
65
void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
66

    
67
/* encoding scans */
68
extern const uint8_t ff_alternate_horizontal_scan[64];
69
extern const uint8_t ff_alternate_vertical_scan[64];
70
extern const uint8_t ff_zigzag_direct[64];
71
extern const uint8_t ff_zigzag248_direct[64];
72

    
73
/* pixel operations */
74
#define MAX_NEG_CROP 1024
75

    
76
/* temporary */
77
extern uint32_t ff_squareTbl[512];
78
extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
79

    
80
/* VP3 DSP functions */
81
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
82
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
83
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
84

    
85
/* 1/2^n downscaling functions from imgconvert.c */
86
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
87
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
88
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
89
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
90

    
91
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
92
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
93

    
94
/* minimum alignment rules ;)
95
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
96
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
97

98
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
99
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
100

101
!future video codecs might need functions with less strict alignment
102
*/
103

    
104
/*
105
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
106
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
107
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
108
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
109
void clear_blocks_c(DCTELEM *blocks);
110
*/
111

    
112
/* add and put pixel (decoding) */
113
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
114
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
115
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
116
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);
117
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
118
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
119
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
120
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
121

    
122
#define DEF_OLD_QPEL(name)\
123
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
124
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
125
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
126

    
127
DEF_OLD_QPEL(qpel16_mc11_old_c)
128
DEF_OLD_QPEL(qpel16_mc31_old_c)
129
DEF_OLD_QPEL(qpel16_mc12_old_c)
130
DEF_OLD_QPEL(qpel16_mc32_old_c)
131
DEF_OLD_QPEL(qpel16_mc13_old_c)
132
DEF_OLD_QPEL(qpel16_mc33_old_c)
133
DEF_OLD_QPEL(qpel8_mc11_old_c)
134
DEF_OLD_QPEL(qpel8_mc31_old_c)
135
DEF_OLD_QPEL(qpel8_mc12_old_c)
136
DEF_OLD_QPEL(qpel8_mc32_old_c)
137
DEF_OLD_QPEL(qpel8_mc13_old_c)
138
DEF_OLD_QPEL(qpel8_mc33_old_c)
139

    
140
#define CALL_2X_PIXELS(a, b, n)\
141
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
142
    b(block  , pixels  , line_size, h);\
143
    b(block+n, pixels+n, line_size, h);\
144
}
145

    
146
/* motion estimation */
147
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
148
// although currently h<4 is not used as functions with width <8 are neither used nor implemented
149
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))*/;
150

    
151

    
152
// for snow slices
153
typedef struct slice_buffer_s slice_buffer;
154

    
155
/**
156
 * DSPContext.
157
 */
158
typedef struct DSPContext {
159
    /* pixel ops : interface with DCT */
160
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
161
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
162
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
163
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
164
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
165
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
166
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
167
    int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
168
    /**
169
     * translational global motion compensation.
170
     */
171
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
172
    /**
173
     * global motion compensation.
174
     */
175
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
176
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
177
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
178
    int (*pix_sum)(uint8_t * pix, int line_size);
179
    int (*pix_norm1)(uint8_t * pix, int line_size);
180
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
181

    
182
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
183
    me_cmp_func sse[5];
184
    me_cmp_func hadamard8_diff[5];
185
    me_cmp_func dct_sad[5];
186
    me_cmp_func quant_psnr[5];
187
    me_cmp_func bit[5];
188
    me_cmp_func rd[5];
189
    me_cmp_func vsad[5];
190
    me_cmp_func vsse[5];
191
    me_cmp_func nsse[5];
192
    me_cmp_func w53[5];
193
    me_cmp_func w97[5];
194
    me_cmp_func dct_max[5];
195
    me_cmp_func dct264_sad[5];
196

    
197
    me_cmp_func me_pre_cmp[5];
198
    me_cmp_func me_cmp[5];
199
    me_cmp_func me_sub_cmp[5];
200
    me_cmp_func mb_cmp[5];
201
    me_cmp_func ildct_cmp[5]; //only width 16 used
202
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
203

    
204
    int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
205
                             int size);
206

    
207
    /**
208
     * Halfpel motion compensation with rounding (a+b+1)>>1.
209
     * this is an array[4][4] of motion compensation functions for 4
210
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
211
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
212
     * @param block destination where the result is stored
213
     * @param pixels source
214
     * @param line_size number of bytes in a horizontal line of block
215
     * @param h height
216
     */
217
    op_pixels_func put_pixels_tab[4][4];
218

    
219
    /**
220
     * Halfpel motion compensation with rounding (a+b+1)>>1.
221
     * This is an array[4][4] of motion compensation functions for 4
222
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
223
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
224
     * @param block destination into which the result is averaged (a+b+1)>>1
225
     * @param pixels source
226
     * @param line_size number of bytes in a horizontal line of block
227
     * @param h height
228
     */
229
    op_pixels_func avg_pixels_tab[4][4];
230

    
231
    /**
232
     * Halfpel motion compensation with no rounding (a+b)>>1.
233
     * this is an array[2][4] of motion compensation functions for 2
234
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
235
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
236
     * @param block destination where the result is stored
237
     * @param pixels source
238
     * @param line_size number of bytes in a horizontal line of block
239
     * @param h height
240
     */
241
    op_pixels_func put_no_rnd_pixels_tab[4][4];
242

    
243
    /**
244
     * Halfpel motion compensation with no rounding (a+b)>>1.
245
     * this is an array[2][4] of motion compensation functions for 2
246
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
247
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
248
     * @param block destination into which the result is averaged (a+b)>>1
249
     * @param pixels source
250
     * @param line_size number of bytes in a horizontal line of block
251
     * @param h height
252
     */
253
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
254

    
255
    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);
256

    
257
    /**
258
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
259
     * this is an array[12] of motion compensation functions for the 9 thirdpe
260
     * positions<br>
261
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
262
     * @param block destination where the result is stored
263
     * @param pixels source
264
     * @param line_size number of bytes in a horizontal line of block
265
     * @param h height
266
     */
267
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
268
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
269

    
270
    qpel_mc_func put_qpel_pixels_tab[2][16];
271
    qpel_mc_func avg_qpel_pixels_tab[2][16];
272
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
273
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
274
    qpel_mc_func put_mspel_pixels_tab[8];
275

    
276
    /**
277
     * h264 Chroma MC
278
     */
279
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
280
    /* This is really one func used in VC-1 decoding */
281
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
282
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
283

    
284
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
285
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
286

    
287
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
288
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
289

    
290
    h264_weight_func weight_h264_pixels_tab[10];
291
    h264_biweight_func biweight_h264_pixels_tab[10];
292

    
293
    /* AVS specific */
294
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
295
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
296
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
297
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
298
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
299
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
300
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
301

    
302
    me_cmp_func pix_abs[2][4];
303

    
304
    /* huffyuv specific */
305
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
306
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
307
    /**
308
     * subtract huffyuv's variant of median prediction
309
     * note, this might read from src1[-1], src2[-1]
310
     */
311
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
312
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
313

    
314
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
315
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
316
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
317
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
318
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
319
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
320
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
321
    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],
322
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);
323

    
324
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
325
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
326

    
327
    void (*h261_loop_filter)(uint8_t *src, int stride);
328

    
329
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
330
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
331
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
332
    void (*vector_fmul)(float *dst, const float *src, int len);
333
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
334
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
335
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
336

    
337
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
338
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
339
    void (*float_to_int16)(int16_t *dst, const float *src, int len);
340

    
341
    /* (I)DCT */
342
    void (*fdct)(DCTELEM *block/* align 16*/);
343
    void (*fdct248)(DCTELEM *block/* align 16*/);
344

    
345
    /* IDCT really*/
346
    void (*idct)(DCTELEM *block/* align 16*/);
347

    
348
    /**
349
     * block -> idct -> clip to unsigned 8 bit -> dest.
350
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
351
     * @param line_size size in bytes of a horizotal line of dest
352
     */
353
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
354

    
355
    /**
356
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
357
     * @param line_size size in bytes of a horizotal line of dest
358
     */
359
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
360

    
361
    /**
362
     * idct input permutation.
363
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
364
     * IDCT)
365
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
366
     * with the zigzag/alternate scan<br>
367
     * an example to avoid confusion:
368
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
369
     * - (x -> referece dct -> reference idct -> x)
370
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
371
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
372
     */
373
    uint8_t idct_permutation[64];
374
    int idct_permutation_type;
375
#define FF_NO_IDCT_PERM 1
376
#define FF_LIBMPEG2_IDCT_PERM 2
377
#define FF_SIMPLE_IDCT_PERM 3
378
#define FF_TRANSPOSE_IDCT_PERM 4
379
#define FF_PARTTRANS_IDCT_PERM 5
380

    
381
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
382
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
383
#define BASIS_SHIFT 16
384
#define RECON_SHIFT 6
385

    
386
    /* h264 functions */
387
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
388
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
389
    void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
390
    void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
391
    void (*h264_dct)(DCTELEM block[4][4]);
392

    
393
    /* snow wavelet */
394
    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
395
    void (*horizontal_compose97i)(IDWTELEM *b, int width);
396
    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);
397

    
398
    void (*prefetch)(void *mem, int stride, int h);
399

    
400
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
401

    
402
    /* vc1 functions */
403
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
404
    void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
405
    void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
406
    void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
407
    void (*vc1_v_overlap)(uint8_t* src, int stride);
408
    void (*vc1_h_overlap)(uint8_t* src, int stride);
409
    /* put 8x8 block with bicubic interpolation and quarterpel precision
410
     * last argument is actually round value instead of height
411
     */
412
    op_pixels_func put_vc1_mspel_pixels_tab[16];
413
} DSPContext;
414

    
415
void dsputil_static_init(void);
416
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
417

    
418
int ff_check_alignment(void);
419

    
420
/**
421
 * permute block according to permuatation.
422
 * @param last last non zero element in scantable order
423
 */
424
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
425

    
426
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
427

    
428
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
429

    
430
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
431
{
432
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
433
}
434

    
435
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
436
{
437
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
438
}
439

    
440
static inline int get_penalty_factor(int lambda, int lambda2, int type){
441
    switch(type&0xFF){
442
    default:
443
    case FF_CMP_SAD:
444
        return lambda>>FF_LAMBDA_SHIFT;
445
    case FF_CMP_DCT:
446
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
447
    case FF_CMP_W53:
448
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
449
    case FF_CMP_W97:
450
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
451
    case FF_CMP_SATD:
452
    case FF_CMP_DCT264:
453
        return (2*lambda)>>FF_LAMBDA_SHIFT;
454
    case FF_CMP_RD:
455
    case FF_CMP_PSNR:
456
    case FF_CMP_SSE:
457
    case FF_CMP_NSSE:
458
        return lambda2>>FF_LAMBDA_SHIFT;
459
    case FF_CMP_BIT:
460
        return 1;
461
    }
462
}
463

    
464
/**
465
 * Empty mmx state.
466
 * this must be called between any dsp function and float/double code.
467
 * for example sin(); dsp->idct_put(); emms_c(); cos()
468
 */
469
#define emms_c()
470

    
471
/* should be defined by architectures supporting
472
   one or more MultiMedia extension */
473
int mm_support(void);
474

    
475
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
476
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
477
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
478
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
479
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
480
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
481
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
482
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
483
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
484

    
485
#define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
486

    
487
#if defined(HAVE_MMX)
488

    
489
#undef emms_c
490

    
491
#define MM_MMX    0x0001 /* standard MMX */
492
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
493
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
494
#define MM_SSE    0x0008 /* SSE functions */
495
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
496
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
497
#define MM_SSE3   0x0040 /* Prescott SSE3 functions */
498
#define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
499

    
500
extern int mm_flags;
501

    
502
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
503
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
504
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
505

    
506
static inline void emms(void)
507
{
508
    __asm __volatile ("emms;":::"memory");
509
}
510

    
511

    
512
#define emms_c() \
513
{\
514
    if (mm_flags & MM_MMX)\
515
        emms();\
516
}
517

    
518
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
519

    
520
#define STRIDE_ALIGN 8
521

    
522
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
523

    
524
#elif defined(ARCH_ARMV4L)
525

    
526
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
527
   line optimizations */
528
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(4, t, v)
529
#define STRIDE_ALIGN 4
530

    
531
#define MM_IWMMXT    0x0100 /* XScale IWMMXT */
532

    
533
extern int mm_flags;
534

    
535
#elif defined(HAVE_MLIB)
536

    
537
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
538
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
539
#define STRIDE_ALIGN 8
540

    
541
#elif defined(ARCH_SPARC)
542

    
543
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
544
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
545
#define STRIDE_ALIGN 8
546

    
547
#elif defined(ARCH_ALPHA)
548

    
549
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
550
#define STRIDE_ALIGN 8
551

    
552
#elif defined(ARCH_POWERPC)
553

    
554
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
555

    
556
extern int mm_flags;
557

    
558
#if defined(HAVE_ALTIVEC) && !defined(__APPLE_CC__)
559
#define pixel altivec_pixel
560
#include <altivec.h>
561
#undef pixel
562
#endif
563

    
564
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
565
#define STRIDE_ALIGN 16
566

    
567
#elif defined(HAVE_MMI)
568

    
569
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
570
#define STRIDE_ALIGN 16
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
#elif defined(ARCH_BFIN)
578

    
579
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
580
#define STRIDE_ALIGN 8
581

    
582
#else
583

    
584
#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
585
#define STRIDE_ALIGN 8
586

    
587
#endif
588

    
589
/* PSNR */
590
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
591
              int orig_linesize[3], int coded_linesize,
592
              AVCodecContext *avctx);
593

    
594
/* FFT computation */
595

    
596
/* NOTE: soon integer code will be added, so you must use the
597
   FFTSample type */
598
typedef float FFTSample;
599

    
600
struct MDCTContext;
601

    
602
typedef struct FFTComplex {
603
    FFTSample re, im;
604
} FFTComplex;
605

    
606
typedef struct FFTContext {
607
    int nbits;
608
    int inverse;
609
    uint16_t *revtab;
610
    FFTComplex *exptab;
611
    FFTComplex *exptab1; /* only used by SSE code */
612
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
613
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
614
                       const FFTSample *input, FFTSample *tmp);
615
} FFTContext;
616

    
617
int ff_fft_init(FFTContext *s, int nbits, int inverse);
618
void ff_fft_permute(FFTContext *s, FFTComplex *z);
619
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
620
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
621
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
622
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
623
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
624

    
625
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
626
{
627
    s->fft_calc(s, z);
628
}
629
void ff_fft_end(FFTContext *s);
630

    
631
/* MDCT computation */
632

    
633
typedef struct MDCTContext {
634
    int n;  /* size of MDCT (i.e. number of input data * 2) */
635
    int nbits; /* n = 2^nbits */
636
    /* pre/post rotation tables */
637
    FFTSample *tcos;
638
    FFTSample *tsin;
639
    FFTContext fft;
640
} MDCTContext;
641

    
642
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
643
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
644
                const FFTSample *input, FFTSample *tmp);
645
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
646
                        const FFTSample *input, FFTSample *tmp);
647
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
648
                       const FFTSample *input, FFTSample *tmp);
649
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
650
               const FFTSample *input, FFTSample *tmp);
651
void ff_mdct_end(MDCTContext *s);
652

    
653
#define WARPER8_16(name8, name16)\
654
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
655
    return name8(s, dst           , src           , stride, h)\
656
          +name8(s, dst+8         , src+8         , stride, h);\
657
}
658

    
659
#define WARPER8_16_SQ(name8, name16)\
660
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
661
    int score=0;\
662
    score +=name8(s, dst           , src           , stride, 8);\
663
    score +=name8(s, dst+8         , src+8         , stride, 8);\
664
    if(h==16){\
665
        dst += 8*stride;\
666
        src += 8*stride;\
667
        score +=name8(s, dst           , src           , stride, 8);\
668
        score +=name8(s, dst+8         , src+8         , stride, 8);\
669
    }\
670
    return score;\
671
}
672

    
673

    
674
static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
675
{
676
    int i;
677
    for(i=0; i<h; i++)
678
    {
679
        AV_WN16(dst   , AV_RN16(src   ));
680
        dst+=dstStride;
681
        src+=srcStride;
682
    }
683
}
684

    
685
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
686
{
687
    int i;
688
    for(i=0; i<h; i++)
689
    {
690
        AV_WN32(dst   , AV_RN32(src   ));
691
        dst+=dstStride;
692
        src+=srcStride;
693
    }
694
}
695

    
696
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
697
{
698
    int i;
699
    for(i=0; i<h; i++)
700
    {
701
        AV_WN32(dst   , AV_RN32(src   ));
702
        AV_WN32(dst+4 , AV_RN32(src+4 ));
703
        dst+=dstStride;
704
        src+=srcStride;
705
    }
706
}
707

    
708
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
709
{
710
    int i;
711
    for(i=0; i<h; i++)
712
    {
713
        AV_WN32(dst   , AV_RN32(src   ));
714
        AV_WN32(dst+4 , AV_RN32(src+4 ));
715
        dst[8]= src[8];
716
        dst+=dstStride;
717
        src+=srcStride;
718
    }
719
}
720

    
721
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
722
{
723
    int i;
724
    for(i=0; i<h; i++)
725
    {
726
        AV_WN32(dst   , AV_RN32(src   ));
727
        AV_WN32(dst+4 , AV_RN32(src+4 ));
728
        AV_WN32(dst+8 , AV_RN32(src+8 ));
729
        AV_WN32(dst+12, AV_RN32(src+12));
730
        dst+=dstStride;
731
        src+=srcStride;
732
    }
733
}
734

    
735
static inline void copy_block17(uint8_t *dst, 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
        AV_WN32(dst+8 , AV_RN32(src+8 ));
743
        AV_WN32(dst+12, AV_RN32(src+12));
744
        dst[16]= src[16];
745
        dst+=dstStride;
746
        src+=srcStride;
747
    }
748
}
749

    
750
#endif