Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 59006372

History | View | Annotate | Download (27.7 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 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 "common.h"
34
#include "avcodec.h"
35

    
36

    
37
//#define DEBUG
38
/* dct code */
39
typedef short DCTELEM;
40
typedef int DWTELEM;
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
// allthough currently h<4 is not used as functions with width <8 are not used and neither 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
    /**
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 funcions 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 funcions 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 funcions 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 funcions for the 9 thirdpel positions<br>
258
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
259
     * @param block destination where the result is stored
260
     * @param pixels source
261
     * @param line_size number of bytes in a horizontal line of block
262
     * @param h height
263
     */
264
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
265
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
266

    
267
    qpel_mc_func put_qpel_pixels_tab[2][16];
268
    qpel_mc_func avg_qpel_pixels_tab[2][16];
269
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
270
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
271
    qpel_mc_func put_mspel_pixels_tab[8];
272

    
273
    /**
274
     * h264 Chram MC
275
     */
276
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
277
    /* This is really one func used in VC-1 decoding */
278
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
279
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
280

    
281
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
282
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
283

    
284
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
285
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
286

    
287
    h264_weight_func weight_h264_pixels_tab[10];
288
    h264_biweight_func biweight_h264_pixels_tab[10];
289

    
290
    /* AVS specific */
291
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
292
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
293
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
294
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
295
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
296
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
297
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
298

    
299
    me_cmp_func pix_abs[2][4];
300

    
301
    /* huffyuv specific */
302
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
303
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
304
    /**
305
     * subtract huffyuv's variant of median prediction
306
     * note, this might read from src1[-1], src2[-1]
307
     */
308
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
309
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
310

    
311
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
312
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
313
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
314
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
315
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
316
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
317
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
318
    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],
319
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);
320

    
321
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
322
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
323

    
324
    void (*h261_loop_filter)(uint8_t *src, int stride);
325

    
326
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
327
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
328
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
329
    void (*vector_fmul)(float *dst, const float *src, int len);
330
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
331
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
332
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
333

    
334
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
335
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
336
    void (*float_to_int16)(int16_t *dst, const float *src, int len);
337

    
338
    /* (I)DCT */
339
    void (*fdct)(DCTELEM *block/* align 16*/);
340
    void (*fdct248)(DCTELEM *block/* align 16*/);
341

    
342
    /* IDCT really*/
343
    void (*idct)(DCTELEM *block/* align 16*/);
344

    
345
    /**
346
     * block -> idct -> clip to unsigned 8 bit -> dest.
347
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
348
     * @param line_size size in bytes of a horizotal line of dest
349
     */
350
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
351

    
352
    /**
353
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
354
     * @param line_size size in bytes of a horizotal line of dest
355
     */
356
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
357

    
358
    /**
359
     * idct input permutation.
360
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
361
     * IDCT)
362
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
363
     * with the zigzag/alternate scan<br>
364
     * an example to avoid confusion:
365
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
366
     * - (x -> referece dct -> reference idct -> x)
367
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
368
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
369
     */
370
    uint8_t idct_permutation[64];
371
    int idct_permutation_type;
372
#define FF_NO_IDCT_PERM 1
373
#define FF_LIBMPEG2_IDCT_PERM 2
374
#define FF_SIMPLE_IDCT_PERM 3
375
#define FF_TRANSPOSE_IDCT_PERM 4
376
#define FF_PARTTRANS_IDCT_PERM 5
377

    
378
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
379
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
380
#define BASIS_SHIFT 16
381
#define RECON_SHIFT 6
382

    
383
    /* h264 functions */
384
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
385
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
386
    void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
387
    void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
388
    void (*h264_dct)(DCTELEM block[4][4]);
389

    
390
    /* snow wavelet */
391
    void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
392
    void (*horizontal_compose97i)(DWTELEM *b, int width);
393
    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);
394

    
395
    void (*prefetch)(void *mem, int stride, int h);
396

    
397
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
398

    
399
    /* vc1 functions */
400
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
401
    void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
402
    void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
403
    void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
404
    void (*vc1_v_overlap)(uint8_t* src, int stride);
405
    void (*vc1_h_overlap)(uint8_t* src, int stride);
406
    /* put 8x8 block with bicubic interpolation and quarterpel precision
407
     * last argument is actually round value instead of height
408
     */
409
    op_pixels_func put_vc1_mspel_pixels_tab[16];
410
} DSPContext;
411

    
412
void dsputil_static_init(void);
413
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
414

    
415
int ff_check_alignment(void);
416

    
417
/**
418
 * permute block according to permuatation.
419
 * @param last last non zero element in scantable order
420
 */
421
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
422

    
423
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
424

    
425
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
426

    
427
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
428
{
429
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
430
}
431

    
432
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
433
{
434
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
435
}
436

    
437
static inline int get_penalty_factor(int lambda, int lambda2, int type){
438
    switch(type&0xFF){
439
    default:
440
    case FF_CMP_SAD:
441
        return lambda>>FF_LAMBDA_SHIFT;
442
    case FF_CMP_DCT:
443
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
444
    case FF_CMP_W53:
445
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
446
    case FF_CMP_W97:
447
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
448
    case FF_CMP_SATD:
449
    case FF_CMP_DCT264:
450
        return (2*lambda)>>FF_LAMBDA_SHIFT;
451
    case FF_CMP_RD:
452
    case FF_CMP_PSNR:
453
    case FF_CMP_SSE:
454
    case FF_CMP_NSSE:
455
        return lambda2>>FF_LAMBDA_SHIFT;
456
    case FF_CMP_BIT:
457
        return 1;
458
    }
459
}
460

    
461
/**
462
 * Empty mmx state.
463
 * this must be called between any dsp function and float/double code.
464
 * for example sin(); dsp->idct_put(); emms_c(); cos()
465
 */
466
#define emms_c()
467

    
468
/* should be defined by architectures supporting
469
   one or more MultiMedia extension */
470
int mm_support(void);
471

    
472
#ifdef __GNUC__
473
  #define DECLARE_ALIGNED_16(t,v)       t v __attribute__ ((aligned (16)))
474
#else
475
  #define DECLARE_ALIGNED_16(t,v)      __declspec(align(16)) t v
476
#endif
477

    
478
#if defined(HAVE_MMX)
479

    
480
#undef emms_c
481

    
482
#define MM_MMX    0x0001 /* standard MMX */
483
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
484
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
485
#define MM_SSE    0x0008 /* SSE functions */
486
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
487
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
488
#define MM_SSE3   0x0040 /* Prescott SSE3 functions */
489
#define MM_SSSE3  0x0080 /* Conroe SSSE3 functions */
490

    
491
extern int mm_flags;
492

    
493
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
494
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
495
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
496

    
497
static inline void emms(void)
498
{
499
    __asm __volatile ("emms;":::"memory");
500
}
501

    
502

    
503
#define emms_c() \
504
{\
505
    if (mm_flags & MM_MMX)\
506
        emms();\
507
}
508

    
509
#ifdef __GNUC__
510
  #define DECLARE_ALIGNED_8(t,v)       t v __attribute__ ((aligned (8)))
511
#else
512
  #define DECLARE_ALIGNED_8(t,v)      __declspec(align(8)) t v
513
#endif
514

    
515
#define STRIDE_ALIGN 8
516

    
517
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
518
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
519

    
520
#elif defined(ARCH_ARMV4L)
521

    
522
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
523
   line optimizations */
524
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (4)))
525
#define STRIDE_ALIGN 4
526

    
527
#define MM_IWMMXT    0x0100 /* XScale IWMMXT */
528

    
529
extern int mm_flags;
530

    
531
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
532

    
533
#elif defined(HAVE_MLIB)
534

    
535
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
536
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
537
#define STRIDE_ALIGN 8
538

    
539
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
540

    
541
#elif defined(ARCH_SPARC)
542

    
543
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
544
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
545
#define STRIDE_ALIGN 8
546
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
547

    
548
#elif defined(ARCH_ALPHA)
549

    
550
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
551
#define STRIDE_ALIGN 8
552

    
553
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
554

    
555
#elif defined(ARCH_POWERPC)
556

    
557
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
558

    
559
extern int mm_flags;
560

    
561
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
562
#define pixel altivec_pixel
563
#include <altivec.h>
564
#undef pixel
565
#endif
566

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

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

    
572
#elif defined(HAVE_MMI)
573

    
574
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (16)))
575
#define STRIDE_ALIGN 16
576

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

    
579
#elif defined(ARCH_SH4)
580

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

    
584
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
585

    
586
#elif defined(ARCH_BFIN)
587

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

    
591
void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
592

    
593
#else
594

    
595
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
596
#define STRIDE_ALIGN 8
597

    
598
#endif
599

    
600
/* PSNR */
601
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
602
              int orig_linesize[3], int coded_linesize,
603
              AVCodecContext *avctx);
604

    
605
/* FFT computation */
606

    
607
/* NOTE: soon integer code will be added, so you must use the
608
   FFTSample type */
609
typedef float FFTSample;
610

    
611
struct MDCTContext;
612

    
613
typedef struct FFTComplex {
614
    FFTSample re, im;
615
} FFTComplex;
616

    
617
typedef struct FFTContext {
618
    int nbits;
619
    int inverse;
620
    uint16_t *revtab;
621
    FFTComplex *exptab;
622
    FFTComplex *exptab1; /* only used by SSE code */
623
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
624
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
625
                       const FFTSample *input, FFTSample *tmp);
626
} FFTContext;
627

    
628
int ff_fft_init(FFTContext *s, int nbits, int inverse);
629
void ff_fft_permute(FFTContext *s, FFTComplex *z);
630
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
631
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
632
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
633
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
634
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
635

    
636
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
637
{
638
    s->fft_calc(s, z);
639
}
640
void ff_fft_end(FFTContext *s);
641

    
642
/* MDCT computation */
643

    
644
typedef struct MDCTContext {
645
    int n;  /* size of MDCT (i.e. number of input data * 2) */
646
    int nbits; /* n = 2^nbits */
647
    /* pre/post rotation tables */
648
    FFTSample *tcos;
649
    FFTSample *tsin;
650
    FFTContext fft;
651
} MDCTContext;
652

    
653
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
654
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
655
                const FFTSample *input, FFTSample *tmp);
656
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
657
                        const FFTSample *input, FFTSample *tmp);
658
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
659
                       const FFTSample *input, FFTSample *tmp);
660
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
661
               const FFTSample *input, FFTSample *tmp);
662
void ff_mdct_end(MDCTContext *s);
663

    
664
#define WARPER8_16(name8, name16)\
665
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
666
    return name8(s, dst           , src           , stride, h)\
667
          +name8(s, dst+8         , src+8         , stride, h);\
668
}
669

    
670
#define WARPER8_16_SQ(name8, name16)\
671
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
672
    int score=0;\
673
    score +=name8(s, dst           , src           , stride, 8);\
674
    score +=name8(s, dst+8         , src+8         , stride, 8);\
675
    if(h==16){\
676
        dst += 8*stride;\
677
        src += 8*stride;\
678
        score +=name8(s, dst           , src           , stride, 8);\
679
        score +=name8(s, dst+8         , src+8         , stride, 8);\
680
    }\
681
    return score;\
682
}
683

    
684

    
685
static inline void copy_block2(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
        ST16(dst   , LD16(src   ));
691
        dst+=dstStride;
692
        src+=srcStride;
693
    }
694
}
695

    
696
static inline void copy_block4(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
        ST32(dst   , LD32(src   ));
702
        dst+=dstStride;
703
        src+=srcStride;
704
    }
705
}
706

    
707
static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
708
{
709
    int i;
710
    for(i=0; i<h; i++)
711
    {
712
        ST32(dst   , LD32(src   ));
713
        ST32(dst+4 , LD32(src+4 ));
714
        dst+=dstStride;
715
        src+=srcStride;
716
    }
717
}
718

    
719
static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
720
{
721
    int i;
722
    for(i=0; i<h; i++)
723
    {
724
        ST32(dst   , LD32(src   ));
725
        ST32(dst+4 , LD32(src+4 ));
726
        dst[8]= src[8];
727
        dst+=dstStride;
728
        src+=srcStride;
729
    }
730
}
731

    
732
static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
733
{
734
    int i;
735
    for(i=0; i<h; i++)
736
    {
737
        ST32(dst   , LD32(src   ));
738
        ST32(dst+4 , LD32(src+4 ));
739
        ST32(dst+8 , LD32(src+8 ));
740
        ST32(dst+12, LD32(src+12));
741
        dst+=dstStride;
742
        src+=srcStride;
743
    }
744
}
745

    
746
static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
747
{
748
    int i;
749
    for(i=0; i<h; i++)
750
    {
751
        ST32(dst   , LD32(src   ));
752
        ST32(dst+4 , LD32(src+4 ));
753
        ST32(dst+8 , LD32(src+8 ));
754
        ST32(dst+12, LD32(src+12));
755
        dst[16]= src[16];
756
        dst+=dstStride;
757
        src+=srcStride;
758
    }
759
}
760

    
761
#endif