Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 94d85eaf

History | View | Annotate | Download (23.3 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 library is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2 of the License, or (at your option) any later version.
10
 *
11
 * This library is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with this library; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

    
21
/**
22
 * @file dsputil.h
23
 * DSP utils.
24
 * note, many functions in here may use MMX which trashes the FPU state, it is
25
 * absolutely necessary to call emms_c() between dsp & float/double code
26
 */
27

    
28
#ifndef DSPUTIL_H
29
#define DSPUTIL_H
30

    
31
#include "common.h"
32
#include "avcodec.h"
33

    
34

    
35
//#define DEBUG
36
/* dct code */
37
typedef short DCTELEM;
38
typedef int DWTELEM;
39

    
40
void fdct_ifast (DCTELEM *data);
41
void fdct_ifast248 (DCTELEM *data);
42
void ff_jpeg_fdct_islow (DCTELEM *data);
43
void ff_fdct248_islow (DCTELEM *data);
44

    
45
void j_rev_dct (DCTELEM *data);
46
void j_rev_dct4 (DCTELEM *data);
47
void j_rev_dct2 (DCTELEM *data);
48
void j_rev_dct1 (DCTELEM *data);
49

    
50
void ff_fdct_mmx(DCTELEM *block);
51
void ff_fdct_mmx2(DCTELEM *block);
52
void ff_fdct_sse2(DCTELEM *block);
53

    
54
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
55
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
56
void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
57
void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
58
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
59
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
60

    
61
/* encoding scans */
62
extern const uint8_t ff_alternate_horizontal_scan[64];
63
extern const uint8_t ff_alternate_vertical_scan[64];
64
extern const uint8_t ff_zigzag_direct[64];
65
extern const uint8_t ff_zigzag248_direct[64];
66

    
67
/* pixel operations */
68
#define MAX_NEG_CROP 1024
69

    
70
/* temporary */
71
extern uint32_t squareTbl[512];
72
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
73

    
74
/* VP3 DSP functions */
75
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
76
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
77
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
78

    
79
/* 1/2^n downscaling functions from imgconvert.c */
80
void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
81
void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
82
void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
83
void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
84

    
85
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
86
              int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
87

    
88
/* minimum alignment rules ;)
89
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
90
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
91

92
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
93
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
94

95
!future video codecs might need functions with less strict alignment
96
*/
97

    
98
/*
99
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
100
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
101
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
102
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
103
void clear_blocks_c(DCTELEM *blocks);
104
*/
105

    
106
/* add and put pixel (decoding) */
107
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
108
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
109
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
110
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);
111
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
112
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
113
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
114
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
115

    
116
#define DEF_OLD_QPEL(name)\
117
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
118
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
119
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
120

    
121
DEF_OLD_QPEL(qpel16_mc11_old_c)
122
DEF_OLD_QPEL(qpel16_mc31_old_c)
123
DEF_OLD_QPEL(qpel16_mc12_old_c)
124
DEF_OLD_QPEL(qpel16_mc32_old_c)
125
DEF_OLD_QPEL(qpel16_mc13_old_c)
126
DEF_OLD_QPEL(qpel16_mc33_old_c)
127
DEF_OLD_QPEL(qpel8_mc11_old_c)
128
DEF_OLD_QPEL(qpel8_mc31_old_c)
129
DEF_OLD_QPEL(qpel8_mc12_old_c)
130
DEF_OLD_QPEL(qpel8_mc32_old_c)
131
DEF_OLD_QPEL(qpel8_mc13_old_c)
132
DEF_OLD_QPEL(qpel8_mc33_old_c)
133

    
134
#define CALL_2X_PIXELS(a, b, n)\
135
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
136
    b(block  , pixels  , line_size, h);\
137
    b(block+n, pixels+n, line_size, h);\
138
}
139

    
140
/* motion estimation */
141
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
142
// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
143
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))*/;
144

    
145

    
146
// for snow slices
147
typedef struct slice_buffer_s slice_buffer;
148

    
149
/**
150
 * DSPContext.
151
 */
152
typedef struct DSPContext {
153
    /* pixel ops : interface with DCT */
154
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
155
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
156
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
157
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
158
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
159
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
160
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
161
    /**
162
     * translational global motion compensation.
163
     */
164
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
165
    /**
166
     * global motion compensation.
167
     */
168
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
169
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
170
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
171
    int (*pix_sum)(uint8_t * pix, int line_size);
172
    int (*pix_norm1)(uint8_t * pix, int line_size);
173
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
174

    
175
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
176
    me_cmp_func sse[5];
177
    me_cmp_func hadamard8_diff[5];
178
    me_cmp_func dct_sad[5];
179
    me_cmp_func quant_psnr[5];
180
    me_cmp_func bit[5];
181
    me_cmp_func rd[5];
182
    me_cmp_func vsad[5];
183
    me_cmp_func vsse[5];
184
    me_cmp_func nsse[5];
185
    me_cmp_func w53[5];
186
    me_cmp_func w97[5];
187
    me_cmp_func dct_max[5];
188
    me_cmp_func dct264_sad[5];
189

    
190
    me_cmp_func me_pre_cmp[5];
191
    me_cmp_func me_cmp[5];
192
    me_cmp_func me_sub_cmp[5];
193
    me_cmp_func mb_cmp[5];
194
    me_cmp_func ildct_cmp[5]; //only width 16 used
195
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
196

    
197
    /**
198
     * Halfpel motion compensation with rounding (a+b+1)>>1.
199
     * this is an array[4][4] of motion compensation funcions for 4
200
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
201
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
202
     * @param block destination where the result is stored
203
     * @param pixels source
204
     * @param line_size number of bytes in a horizontal line of block
205
     * @param h height
206
     */
207
    op_pixels_func put_pixels_tab[4][4];
208

    
209
    /**
210
     * Halfpel motion compensation with rounding (a+b+1)>>1.
211
     * This is an array[4][4] of motion compensation functions for 4
212
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
213
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
214
     * @param block destination into which the result is averaged (a+b+1)>>1
215
     * @param pixels source
216
     * @param line_size number of bytes in a horizontal line of block
217
     * @param h height
218
     */
219
    op_pixels_func avg_pixels_tab[4][4];
220

    
221
    /**
222
     * Halfpel motion compensation with no rounding (a+b)>>1.
223
     * this is an array[2][4] of motion compensation funcions for 2
224
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
225
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
226
     * @param block destination where the result is stored
227
     * @param pixels source
228
     * @param line_size number of bytes in a horizontal line of block
229
     * @param h height
230
     */
231
    op_pixels_func put_no_rnd_pixels_tab[4][4];
232

    
233
    /**
234
     * Halfpel motion compensation with no rounding (a+b)>>1.
235
     * this is an array[2][4] of motion compensation funcions for 2
236
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
237
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
238
     * @param block destination into which the result is averaged (a+b)>>1
239
     * @param pixels source
240
     * @param line_size number of bytes in a horizontal line of block
241
     * @param h height
242
     */
243
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
244

    
245
    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);
246

    
247
    /**
248
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
249
     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
250
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
251
     * @param block destination where the result is stored
252
     * @param pixels source
253
     * @param line_size number of bytes in a horizontal line of block
254
     * @param h height
255
     */
256
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
257
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
258

    
259
    qpel_mc_func put_qpel_pixels_tab[2][16];
260
    qpel_mc_func avg_qpel_pixels_tab[2][16];
261
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
262
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
263
    qpel_mc_func put_mspel_pixels_tab[8];
264

    
265
    /**
266
     * h264 Chram MC
267
     */
268
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
269
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
270

    
271
    qpel_mc_func put_h264_qpel_pixels_tab[4][16];
272
    qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
273

    
274
    h264_weight_func weight_h264_pixels_tab[10];
275
    h264_biweight_func biweight_h264_pixels_tab[10];
276

    
277
    me_cmp_func pix_abs[2][4];
278

    
279
    /* huffyuv specific */
280
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
281
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
282
    /**
283
     * subtract huffyuv's variant of median prediction
284
     * note, this might read from src1[-1], src2[-1]
285
     */
286
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
287
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
288

    
289
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
290
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
291
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
292
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
293
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
294
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
295

    
296
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
297
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
298

    
299
    void (*h261_loop_filter)(uint8_t *src, int stride);
300

    
301
    /* (I)DCT */
302
    void (*fdct)(DCTELEM *block/* align 16*/);
303
    void (*fdct248)(DCTELEM *block/* align 16*/);
304

    
305
    /* IDCT really*/
306
    void (*idct)(DCTELEM *block/* align 16*/);
307

    
308
    /**
309
     * block -> idct -> clip to unsigned 8 bit -> dest.
310
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
311
     * @param line_size size in bytes of a horizotal line of dest
312
     */
313
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
314

    
315
    /**
316
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
317
     * @param line_size size in bytes of a horizotal line of dest
318
     */
319
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
320

    
321
    /**
322
     * idct input permutation.
323
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
324
     * IDCT)
325
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
326
     * with the zigzag/alternate scan<br>
327
     * an example to avoid confusion:
328
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
329
     * - (x -> referece dct -> reference idct -> x)
330
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
331
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
332
     */
333
    uint8_t idct_permutation[64];
334
    int idct_permutation_type;
335
#define FF_NO_IDCT_PERM 1
336
#define FF_LIBMPEG2_IDCT_PERM 2
337
#define FF_SIMPLE_IDCT_PERM 3
338
#define FF_TRANSPOSE_IDCT_PERM 4
339
#define FF_PARTTRANS_IDCT_PERM 5
340

    
341
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
342
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
343
#define BASIS_SHIFT 16
344
#define RECON_SHIFT 6
345

    
346
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
347
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
348
    void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
349
    void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
350

    
351
    /* snow wavelet */
352
    void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
353
    void (*horizontal_compose97i)(DWTELEM *b, int width);
354
    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);
355

    
356
    void (*prefetch)(void *mem, int stride, int h);
357

    
358
    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
359
} DSPContext;
360

    
361
void dsputil_static_init(void);
362
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
363

    
364
/**
365
 * permute block according to permuatation.
366
 * @param last last non zero element in scantable order
367
 */
368
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
369

    
370
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
371

    
372
#define         BYTE_VEC32(c)   ((c)*0x01010101UL)
373

    
374
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
375
{
376
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
377
}
378

    
379
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
380
{
381
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
382
}
383

    
384
static inline int get_penalty_factor(int lambda, int lambda2, int type){
385
    switch(type&0xFF){
386
    default:
387
    case FF_CMP_SAD:
388
        return lambda>>FF_LAMBDA_SHIFT;
389
    case FF_CMP_DCT:
390
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
391
    case FF_CMP_W53:
392
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
393
    case FF_CMP_W97:
394
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
395
    case FF_CMP_SATD:
396
    case FF_CMP_DCT264:
397
        return (2*lambda)>>FF_LAMBDA_SHIFT;
398
    case FF_CMP_RD:
399
    case FF_CMP_PSNR:
400
    case FF_CMP_SSE:
401
    case FF_CMP_NSSE:
402
        return lambda2>>FF_LAMBDA_SHIFT;
403
    case FF_CMP_BIT:
404
        return 1;
405
    }
406
}
407

    
408
/**
409
 * Empty mmx state.
410
 * this must be called between any dsp function and float/double code.
411
 * for example sin(); dsp->idct_put(); emms_c(); cos()
412
 */
413
#define emms_c()
414

    
415
/* should be defined by architectures supporting
416
   one or more MultiMedia extension */
417
int mm_support(void);
418

    
419
#ifdef __GNUC__
420
  #define DECLARE_ALIGNED_16(t,v)       t v __attribute__ ((aligned (16)))
421
#else
422
  #define DECLARE_ALIGNED_16(t,v)      __declspec(align(16)) t v
423
#endif
424

    
425
#if defined(HAVE_MMX)
426

    
427
#undef emms_c
428

    
429
#define MM_MMX    0x0001 /* standard MMX */
430
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
431
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
432
#define MM_SSE    0x0008 /* SSE functions */
433
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
434
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
435
#define MM_SSE3   0x0040 /* Prescott SSE3 functions */
436

    
437
extern int mm_flags;
438

    
439
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
440
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
441
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
442

    
443
static inline void emms(void)
444
{
445
    __asm __volatile ("emms;":::"memory");
446
}
447

    
448

    
449
#define emms_c() \
450
{\
451
    if (mm_flags & MM_MMX)\
452
        emms();\
453
}
454

    
455
#ifdef __GNUC__
456
  #define DECLARE_ALIGNED_8(t,v)       t v __attribute__ ((aligned (8)))
457
#else
458
  #define DECLARE_ALIGNED_8(t,v)      __declspec(align(8)) t v
459
#endif
460

    
461
#define STRIDE_ALIGN 8
462

    
463
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
464
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
465

    
466
#elif defined(ARCH_ARMV4L)
467

    
468
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
469
   line optimizations */
470
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (4)))
471
#define STRIDE_ALIGN 4
472

    
473
#define MM_IWMMXT    0x0100 /* XScale IWMMXT */
474

    
475
extern int mm_flags;
476

    
477
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
478

    
479
#elif defined(HAVE_MLIB)
480

    
481
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
482
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
483
#define STRIDE_ALIGN 8
484

    
485
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
486

    
487
#elif defined(ARCH_SPARC)
488

    
489
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
490
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
491
#define STRIDE_ALIGN 8
492
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
493

    
494
#elif defined(ARCH_ALPHA)
495

    
496
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
497
#define STRIDE_ALIGN 8
498

    
499
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
500

    
501
#elif defined(ARCH_POWERPC)
502

    
503
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
504

    
505
extern int mm_flags;
506

    
507
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
508
#define pixel altivec_pixel
509
#include <altivec.h>
510
#undef pixel
511
#endif
512

    
513
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (16)))
514
#define STRIDE_ALIGN 16
515

    
516
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
517

    
518
#elif defined(HAVE_MMI)
519

    
520
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (16)))
521
#define STRIDE_ALIGN 16
522

    
523
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
524

    
525
#elif defined(ARCH_SH4)
526

    
527
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
528
#define STRIDE_ALIGN 8
529

    
530
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
531

    
532
#else
533

    
534
#define DECLARE_ALIGNED_8(t,v)    t v __attribute__ ((aligned (8)))
535
#define STRIDE_ALIGN 8
536

    
537
#endif
538

    
539
#ifdef __GNUC__
540

    
541
struct unaligned_64 { uint64_t l; } __attribute__((packed));
542
struct unaligned_32 { uint32_t l; } __attribute__((packed));
543
struct unaligned_16 { uint16_t l; } __attribute__((packed));
544

    
545
#define LD16(a) (((const struct unaligned_16 *) (a))->l)
546
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
547
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
548

    
549
#define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
550
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
551

    
552
#else /* __GNUC__ */
553

    
554
#define LD16(a) (*((uint16_t*)(a)))
555
#define LD32(a) (*((uint32_t*)(a)))
556
#define LD64(a) (*((uint64_t*)(a)))
557

    
558
#define ST16(a, b) *((uint16_t*)(a)) = (b)
559
#define ST32(a, b) *((uint32_t*)(a)) = (b)
560

    
561
#endif /* !__GNUC__ */
562

    
563
/* PSNR */
564
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
565
              int orig_linesize[3], int coded_linesize,
566
              AVCodecContext *avctx);
567

    
568
/* FFT computation */
569

    
570
/* NOTE: soon integer code will be added, so you must use the
571
   FFTSample type */
572
typedef float FFTSample;
573

    
574
typedef struct FFTComplex {
575
    FFTSample re, im;
576
} FFTComplex;
577

    
578
typedef struct FFTContext {
579
    int nbits;
580
    int inverse;
581
    uint16_t *revtab;
582
    FFTComplex *exptab;
583
    FFTComplex *exptab1; /* only used by SSE code */
584
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
585
} FFTContext;
586

    
587
int ff_fft_init(FFTContext *s, int nbits, int inverse);
588
void ff_fft_permute(FFTContext *s, FFTComplex *z);
589
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
590
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
591
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
592
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
593
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
594

    
595
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
596
{
597
    s->fft_calc(s, z);
598
}
599
void ff_fft_end(FFTContext *s);
600

    
601
/* MDCT computation */
602

    
603
typedef struct MDCTContext {
604
    int n;  /* size of MDCT (i.e. number of input data * 2) */
605
    int nbits; /* n = 2^nbits */
606
    /* pre/post rotation tables */
607
    FFTSample *tcos;
608
    FFTSample *tsin;
609
    FFTContext fft;
610
} MDCTContext;
611

    
612
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
613
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
614
                const FFTSample *input, FFTSample *tmp);
615
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
616
               const FFTSample *input, FFTSample *tmp);
617
void ff_mdct_end(MDCTContext *s);
618

    
619
#define WARPER8_16(name8, name16)\
620
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
621
    return name8(s, dst           , src           , stride, h)\
622
          +name8(s, dst+8         , src+8         , stride, h);\
623
}
624

    
625
#define WARPER8_16_SQ(name8, name16)\
626
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
627
    int score=0;\
628
    score +=name8(s, dst           , src           , stride, 8);\
629
    score +=name8(s, dst+8         , src+8         , stride, 8);\
630
    if(h==16){\
631
        dst += 8*stride;\
632
        src += 8*stride;\
633
        score +=name8(s, dst           , src           , stride, 8);\
634
        score +=name8(s, dst+8         , src+8         , stride, 8);\
635
    }\
636
    return score;\
637
}
638

    
639
#endif