Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 55fde95e

History | View | Annotate | Download (26.5 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 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
    /**
204
     * Halfpel motion compensation with rounding (a+b+1)>>1.
205
     * this is an array[4][4] of motion compensation funcions for 4
206
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
207
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
208
     * @param block destination where the result is stored
209
     * @param pixels source
210
     * @param line_size number of bytes in a horizontal line of block
211
     * @param h height
212
     */
213
    op_pixels_func put_pixels_tab[4][4];
214

    
215
    /**
216
     * Halfpel motion compensation with rounding (a+b+1)>>1.
217
     * This is an array[4][4] of motion compensation functions for 4
218
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
219
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
220
     * @param block destination into which the result is averaged (a+b+1)>>1
221
     * @param pixels source
222
     * @param line_size number of bytes in a horizontal line of block
223
     * @param h height
224
     */
225
    op_pixels_func avg_pixels_tab[4][4];
226

    
227
    /**
228
     * Halfpel motion compensation with no rounding (a+b)>>1.
229
     * this is an array[2][4] of motion compensation funcions for 2
230
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
231
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
232
     * @param block destination where the result is stored
233
     * @param pixels source
234
     * @param line_size number of bytes in a horizontal line of block
235
     * @param h height
236
     */
237
    op_pixels_func put_no_rnd_pixels_tab[4][4];
238

    
239
    /**
240
     * Halfpel motion compensation with no rounding (a+b)>>1.
241
     * this is an array[2][4] of motion compensation funcions for 2
242
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
243
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
244
     * @param block destination into which the result is averaged (a+b)>>1
245
     * @param pixels source
246
     * @param line_size number of bytes in a horizontal line of block
247
     * @param h height
248
     */
249
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
250

    
251
    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);
252

    
253
    /**
254
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
255
     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
256
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
257
     * @param block destination where the result is stored
258
     * @param pixels source
259
     * @param line_size number of bytes in a horizontal line of block
260
     * @param h height
261
     */
262
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
263
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
264

    
265
    qpel_mc_func put_qpel_pixels_tab[2][16];
266
    qpel_mc_func avg_qpel_pixels_tab[2][16];
267
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
268
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
269
    qpel_mc_func put_mspel_pixels_tab[8];
270

    
271
    /**
272
     * h264 Chram MC
273
     */
274
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
275
    /* This is really one func used in VC-1 decoding */
276
    h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
277
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
278

    
279
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
280
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
281

    
282
    qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
283
    qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
284

    
285
    h264_weight_func weight_h264_pixels_tab[10];
286
    h264_biweight_func biweight_h264_pixels_tab[10];
287

    
288
    /* AVS specific */
289
    qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
290
    qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
291
    void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
292
    void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
293
    void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
294
    void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
295
    void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
296

    
297
    me_cmp_func pix_abs[2][4];
298

    
299
    /* huffyuv specific */
300
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
301
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
302
    /**
303
     * subtract huffyuv's variant of median prediction
304
     * note, this might read from src1[-1], src2[-1]
305
     */
306
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
307
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
308

    
309
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
310
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
311
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
312
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
313
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
314
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
315
    // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
316
    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],
317
                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);
318

    
319
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
320
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
321

    
322
    void (*h261_loop_filter)(uint8_t *src, int stride);
323

    
324
    /* assume len is a multiple of 4, and arrays are 16-byte aligned */
325
    void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
326
    /* assume len is a multiple of 8, and arrays are 16-byte aligned */
327
    void (*vector_fmul)(float *dst, const float *src, int len);
328
    void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
329
    /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
330
    void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
331

    
332
    /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
333
     * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
334
    void (*float_to_int16)(int16_t *dst, const float *src, int len);
335

    
336
    /* (I)DCT */
337
    void (*fdct)(DCTELEM *block/* align 16*/);
338
    void (*fdct248)(DCTELEM *block/* align 16*/);
339

    
340
    /* IDCT really*/
341
    void (*idct)(DCTELEM *block/* align 16*/);
342

    
343
    /**
344
     * block -> idct -> clip to unsigned 8 bit -> dest.
345
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
346
     * @param line_size size in bytes of a horizotal line of dest
347
     */
348
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
349

    
350
    /**
351
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
352
     * @param line_size size in bytes of a horizotal line of dest
353
     */
354
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
355

    
356
    /**
357
     * idct input permutation.
358
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
359
     * IDCT)
360
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
361
     * with the zigzag/alternate scan<br>
362
     * an example to avoid confusion:
363
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
364
     * - (x -> referece dct -> reference idct -> x)
365
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
366
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
367
     */
368
    uint8_t idct_permutation[64];
369
    int idct_permutation_type;
370
#define FF_NO_IDCT_PERM 1
371
#define FF_LIBMPEG2_IDCT_PERM 2
372
#define FF_SIMPLE_IDCT_PERM 3
373
#define FF_TRANSPOSE_IDCT_PERM 4
374
#define FF_PARTTRANS_IDCT_PERM 5
375

    
376
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
377
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
378
#define BASIS_SHIFT 16
379
#define RECON_SHIFT 6
380

    
381
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
382
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
383
    void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
384
    void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
385

    
386
    /* snow wavelet */
387
    void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
388
    void (*horizontal_compose97i)(DWTELEM *b, int width);
389
    void (*inner_add_yblock)(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);
390

    
391
    void (*prefetch)(void *mem, int stride, int h);
392

    
393
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
394

    
395
    /* vc1 functions */
396
    void (*vc1_inv_trans_8x8)(DCTELEM *b);
397
    void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
398
    void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
399
    void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
400
    void (*vc1_v_overlap)(uint8_t* src, int stride, int rnd);
401
    void (*vc1_h_overlap)(uint8_t* src, int stride, int rnd);
402
    /* put 8x8 block with bicubic interpolation and quarterpel precision
403
     * last argument is actually round value instead of height
404
     */
405
    op_pixels_func put_vc1_mspel_pixels_tab[16];
406
} DSPContext;
407

    
408
void dsputil_static_init(void);
409
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
410

    
411
/**
412
 * permute block according to permuatation.
413
 * @param last last non zero element in scantable order
414
 */
415
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
416

    
417
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
418

    
419
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
420

    
421
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
422
{
423
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
424
}
425

    
426
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
427
{
428
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
429
}
430

    
431
static inline int get_penalty_factor(int lambda, int lambda2, int type){
432
    switch(type&0xFF){
433
    default:
434
    case FF_CMP_SAD:
435
        return lambda>>FF_LAMBDA_SHIFT;
436
    case FF_CMP_DCT:
437
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
438
    case FF_CMP_W53:
439
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
440
    case FF_CMP_W97:
441
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
442
    case FF_CMP_SATD:
443
    case FF_CMP_DCT264:
444
        return (2*lambda)>>FF_LAMBDA_SHIFT;
445
    case FF_CMP_RD:
446
    case FF_CMP_PSNR:
447
    case FF_CMP_SSE:
448
    case FF_CMP_NSSE:
449
        return lambda2>>FF_LAMBDA_SHIFT;
450
    case FF_CMP_BIT:
451
        return 1;
452
    }
453
}
454

    
455
/**
456
 * Empty mmx state.
457
 * this must be called between any dsp function and float/double code.
458
 * for example sin(); dsp->idct_put(); emms_c(); cos()
459
 */
460
#define emms_c()
461

    
462
/* should be defined by architectures supporting
463
   one or more MultiMedia extension */
464
int mm_support(void);
465

    
466
#ifdef __GNUC__
467
  #define DECLARE_ALIGNED_16(t,v)       t v __attribute__ ((aligned (16)))
468
#else
469
  #define DECLARE_ALIGNED_16(t,v)      __declspec(align(16)) t v
470
#endif
471

    
472
#if defined(HAVE_MMX)
473

    
474
#undef emms_c
475

    
476
#define MM_MMX    0x0001 /* standard MMX */
477
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
478
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
479
#define MM_SSE    0x0008 /* SSE functions */
480
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
481
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
482
#define MM_SSE3   0x0040 /* Prescott SSE3 functions */
483

    
484
extern int mm_flags;
485

    
486
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
487
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
488
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
489

    
490
static inline void emms(void)
491
{
492
    __asm __volatile ("emms;":::"memory");
493
}
494

    
495

    
496
#define emms_c() \
497
{\
498
    if (mm_flags & MM_MMX)\
499
        emms();\
500
}
501

    
502
#ifdef __GNUC__
503
  #define DECLARE_ALIGNED_8(t,v)       t v __attribute__ ((aligned (8)))
504
#else
505
  #define DECLARE_ALIGNED_8(t,v)      __declspec(align(8)) t v
506
#endif
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)    t v __attribute__ ((aligned (4)))
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)    t v __attribute__ ((aligned (8)))
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)    t v __attribute__ ((aligned (8)))
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)    t v __attribute__ ((aligned (8)))
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)    t v __attribute__ ((aligned (16)))
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)    t v __attribute__ ((aligned (16)))
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)    t v __attribute__ ((aligned (8)))
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)    t v __attribute__ ((aligned (8)))
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)    t v __attribute__ ((aligned (8)))
589
#define STRIDE_ALIGN 8
590

    
591
#endif
592

    
593
#ifdef __GNUC__
594

    
595
struct unaligned_64 { uint64_t l; } __attribute__((packed));
596
struct unaligned_32 { uint32_t l; } __attribute__((packed));
597
struct unaligned_16 { uint16_t l; } __attribute__((packed));
598

    
599
#define LD16(a) (((const struct unaligned_16 *) (a))->l)
600
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
601
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
602

    
603
#define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
604
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
605

    
606
#else /* __GNUC__ */
607

    
608
#define LD16(a) (*((uint16_t*)(a)))
609
#define LD32(a) (*((uint32_t*)(a)))
610
#define LD64(a) (*((uint64_t*)(a)))
611

    
612
#define ST16(a, b) *((uint16_t*)(a)) = (b)
613
#define ST32(a, b) *((uint32_t*)(a)) = (b)
614

    
615
#endif /* !__GNUC__ */
616

    
617
/* PSNR */
618
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
619
              int orig_linesize[3], int coded_linesize,
620
              AVCodecContext *avctx);
621

    
622
/* FFT computation */
623

    
624
/* NOTE: soon integer code will be added, so you must use the
625
   FFTSample type */
626
typedef float FFTSample;
627

    
628
struct MDCTContext;
629

    
630
typedef struct FFTComplex {
631
    FFTSample re, im;
632
} FFTComplex;
633

    
634
typedef struct FFTContext {
635
    int nbits;
636
    int inverse;
637
    uint16_t *revtab;
638
    FFTComplex *exptab;
639
    FFTComplex *exptab1; /* only used by SSE code */
640
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
641
    void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
642
                       const FFTSample *input, FFTSample *tmp);
643
} FFTContext;
644

    
645
int ff_fft_init(FFTContext *s, int nbits, int inverse);
646
void ff_fft_permute(FFTContext *s, FFTComplex *z);
647
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
648
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
649
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
650
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
651
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
652

    
653
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
654
{
655
    s->fft_calc(s, z);
656
}
657
void ff_fft_end(FFTContext *s);
658

    
659
/* MDCT computation */
660

    
661
typedef struct MDCTContext {
662
    int n;  /* size of MDCT (i.e. number of input data * 2) */
663
    int nbits; /* n = 2^nbits */
664
    /* pre/post rotation tables */
665
    FFTSample *tcos;
666
    FFTSample *tsin;
667
    FFTContext fft;
668
} MDCTContext;
669

    
670
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
671
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
672
                const FFTSample *input, FFTSample *tmp);
673
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
674
                        const FFTSample *input, FFTSample *tmp);
675
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
676
                       const FFTSample *input, FFTSample *tmp);
677
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
678
               const FFTSample *input, FFTSample *tmp);
679
void ff_mdct_end(MDCTContext *s);
680

    
681
#define WARPER8_16(name8, name16)\
682
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
683
    return name8(s, dst           , src           , stride, h)\
684
          +name8(s, dst+8         , src+8         , stride, h);\
685
}
686

    
687
#define WARPER8_16_SQ(name8, name16)\
688
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
689
    int score=0;\
690
    score +=name8(s, dst           , src           , stride, 8);\
691
    score +=name8(s, dst+8         , src+8         , stride, 8);\
692
    if(h==16){\
693
        dst += 8*stride;\
694
        src += 8*stride;\
695
        score +=name8(s, dst           , src           , stride, 8);\
696
        score +=name8(s, dst+8         , src+8         , stride, 8);\
697
    }\
698
    return score;\
699
}
700

    
701
#endif