Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ eba9ae3c

History | View | Annotate | Download (21.7 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 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., 59 Temple Place, Suite 330, Boston, MA  02111-1307  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

    
39
void fdct_ifast (DCTELEM *data);
40
void fdct_ifast248 (DCTELEM *data);
41
void ff_jpeg_fdct_islow (DCTELEM *data);
42
void ff_fdct248_islow (DCTELEM *data);
43

    
44
void j_rev_dct (DCTELEM *data);
45
void j_rev_dct4 (DCTELEM *data);
46
void j_rev_dct2 (DCTELEM *data);
47
void j_rev_dct1 (DCTELEM *data);
48

    
49
void ff_fdct_mmx(DCTELEM *block);
50
void ff_fdct_mmx2(DCTELEM *block);
51
void ff_fdct_sse2(DCTELEM *block);
52

    
53
void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
54
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
55
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
56
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
57

    
58
/* encoding scans */
59
extern const uint8_t ff_alternate_horizontal_scan[64];
60
extern const uint8_t ff_alternate_vertical_scan[64];
61
extern const uint8_t ff_zigzag_direct[64];
62
extern const uint8_t ff_zigzag248_direct[64];
63

    
64
/* pixel operations */
65
#define MAX_NEG_CROP 1024
66

    
67
/* temporary */
68
extern uint32_t squareTbl[512];
69
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
70

    
71
/* VP3 DSP functions */
72
void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
73
void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
74
void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
75

    
76
/* minimum alignment rules ;)
77
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
78
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
79

80
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
81
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
82

83
!future video codecs might need functions with less strict alignment
84
*/
85

    
86
/*
87
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
88
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
89
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
90
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
91
void clear_blocks_c(DCTELEM *blocks);
92
*/
93

    
94
/* add and put pixel (decoding) */
95
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
96
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
97
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
98
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);
99
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
100
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
101
typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
102
typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offsetd, int offsets);
103

    
104
#define DEF_OLD_QPEL(name)\
105
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
106
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
107
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
108

    
109
DEF_OLD_QPEL(qpel16_mc11_old_c)
110
DEF_OLD_QPEL(qpel16_mc31_old_c)
111
DEF_OLD_QPEL(qpel16_mc12_old_c)
112
DEF_OLD_QPEL(qpel16_mc32_old_c)
113
DEF_OLD_QPEL(qpel16_mc13_old_c)
114
DEF_OLD_QPEL(qpel16_mc33_old_c)
115
DEF_OLD_QPEL(qpel8_mc11_old_c)
116
DEF_OLD_QPEL(qpel8_mc31_old_c)
117
DEF_OLD_QPEL(qpel8_mc12_old_c)
118
DEF_OLD_QPEL(qpel8_mc32_old_c)
119
DEF_OLD_QPEL(qpel8_mc13_old_c)
120
DEF_OLD_QPEL(qpel8_mc33_old_c)
121

    
122
#define CALL_2X_PIXELS(a, b, n)\
123
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
124
    b(block  , pixels  , line_size, h);\
125
    b(block+n, pixels+n, line_size, h);\
126
}
127

    
128
/* motion estimation */
129
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
130
// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
131
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))*/;
132

    
133

    
134
/**
135
 * DSPContext.
136
 */
137
typedef struct DSPContext {
138
    /* pixel ops : interface with DCT */
139
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
140
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
141
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
142
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
143
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
144
    void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
145
    void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
146
    /**
147
     * translational global motion compensation.
148
     */
149
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
150
    /**
151
     * global motion compensation.
152
     */
153
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
154
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
155
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
156
    int (*pix_sum)(uint8_t * pix, int line_size);
157
    int (*pix_norm1)(uint8_t * pix, int line_size);
158
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
159
    
160
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
161
    me_cmp_func sse[5];
162
    me_cmp_func hadamard8_diff[5];
163
    me_cmp_func dct_sad[5];
164
    me_cmp_func quant_psnr[5];
165
    me_cmp_func bit[5];
166
    me_cmp_func rd[5];
167
    me_cmp_func vsad[5];
168
    me_cmp_func vsse[5];
169
    me_cmp_func nsse[5];
170
    me_cmp_func w53[5];
171
    me_cmp_func w97[5];
172
    me_cmp_func dct_max[5];
173

    
174
    me_cmp_func me_pre_cmp[5];
175
    me_cmp_func me_cmp[5];
176
    me_cmp_func me_sub_cmp[5];
177
    me_cmp_func mb_cmp[5];
178
    me_cmp_func ildct_cmp[5]; //only width 16 used
179
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
180

    
181
    /**
182
     * Halfpel motion compensation with rounding (a+b+1)>>1.
183
     * this is an array[4][4] of motion compensation funcions for 4 
184
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
185
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
186
     * @param block destination where the result is stored
187
     * @param pixels source
188
     * @param line_size number of bytes in a horizontal line of block
189
     * @param h height
190
     */
191
    op_pixels_func put_pixels_tab[4][4];
192

    
193
    /**
194
     * Halfpel motion compensation with rounding (a+b+1)>>1.
195
     * This is an array[4][4] of motion compensation functions for 4 
196
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
197
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
198
     * @param block destination into which the result is averaged (a+b+1)>>1
199
     * @param pixels source
200
     * @param line_size number of bytes in a horizontal line of block
201
     * @param h height
202
     */
203
    op_pixels_func avg_pixels_tab[4][4];
204

    
205
    /**
206
     * Halfpel motion compensation with no rounding (a+b)>>1.
207
     * this is an array[2][4] of motion compensation funcions for 2 
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_no_rnd_pixels_tab[4][4];
216

    
217
    /**
218
     * Halfpel motion compensation with no rounding (a+b)>>1.
219
     * this is an array[2][4] of motion compensation funcions for 2 
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
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_no_rnd_pixels_tab[4][4];
228
    
229
    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);
230
    
231
    /**
232
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
233
     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
234
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
235
     * @param block destination where the result is stored
236
     * @param pixels source
237
     * @param line_size number of bytes in a horizontal line of block
238
     * @param h height
239
     */
240
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
241
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
242

    
243
    qpel_mc_func put_qpel_pixels_tab[2][16];
244
    qpel_mc_func avg_qpel_pixels_tab[2][16];
245
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
246
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
247
    qpel_mc_func put_mspel_pixels_tab[8];
248
    
249
    /**
250
     * h264 Chram MC
251
     */
252
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
253
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
254

    
255
    qpel_mc_func put_h264_qpel_pixels_tab[3][16];
256
    qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
257
    
258
    h264_weight_func weight_h264_pixels_tab[10];
259
    h264_biweight_func biweight_h264_pixels_tab[10];
260
    
261
    me_cmp_func pix_abs[2][4];
262
    
263
    /* huffyuv specific */
264
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
265
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
266
    /**
267
     * subtract huffyuv's variant of median prediction
268
     * note, this might read from src1[-1], src2[-1]
269
     */
270
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
271
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
272

    
273
    void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
274
    void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
275
    void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
276
    void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
277
    void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
278
    void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
279
    
280
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
281
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
282

    
283
    void (*h261_loop_filter)(uint8_t *src, int stride);
284

    
285
    /* (I)DCT */
286
    void (*fdct)(DCTELEM *block/* align 16*/);
287
    void (*fdct248)(DCTELEM *block/* align 16*/);
288
    
289
    /* IDCT really*/
290
    void (*idct)(DCTELEM *block/* align 16*/);
291
    
292
    /**
293
     * block -> idct -> clip to unsigned 8 bit -> dest.
294
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
295
     * @param line_size size in bytes of a horizotal line of dest
296
     */
297
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
298
    
299
    /**
300
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
301
     * @param line_size size in bytes of a horizotal line of dest
302
     */
303
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
304
    
305
    /**
306
     * idct input permutation.
307
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
308
     * IDCT)
309
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
310
     * with the zigzag/alternate scan<br>
311
     * an example to avoid confusion:
312
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
313
     * - (x -> referece dct -> reference idct -> x)
314
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
315
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
316
     */
317
    uint8_t idct_permutation[64];
318
    int idct_permutation_type;
319
#define FF_NO_IDCT_PERM 1
320
#define FF_LIBMPEG2_IDCT_PERM 2
321
#define FF_SIMPLE_IDCT_PERM 3
322
#define FF_TRANSPOSE_IDCT_PERM 4
323
#define FF_PARTTRANS_IDCT_PERM 5
324

    
325
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
326
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
327
#define BASIS_SHIFT 16
328
#define RECON_SHIFT 6
329
 
330
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
331
    void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
332
} DSPContext;
333

    
334
void dsputil_static_init(void);
335
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
336

    
337
/**
338
 * permute block according to permuatation.
339
 * @param last last non zero element in scantable order
340
 */
341
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
342

    
343
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
344

    
345
#define        BYTE_VEC32(c)        ((c)*0x01010101UL)
346

    
347
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
348
{
349
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
350
}
351

    
352
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
353
{
354
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
355
}
356

    
357
static inline int get_penalty_factor(int lambda, int lambda2, int type){
358
    switch(type&0xFF){
359
    default:
360
    case FF_CMP_SAD:
361
        return lambda>>FF_LAMBDA_SHIFT;
362
    case FF_CMP_DCT:
363
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
364
    case FF_CMP_W53:
365
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
366
    case FF_CMP_W97:
367
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
368
    case FF_CMP_SATD:
369
        return (2*lambda)>>FF_LAMBDA_SHIFT;
370
    case FF_CMP_RD:
371
    case FF_CMP_PSNR:
372
    case FF_CMP_SSE:
373
    case FF_CMP_NSSE:
374
        return lambda2>>FF_LAMBDA_SHIFT;
375
    case FF_CMP_BIT:
376
        return 1;
377
    }
378
}
379

    
380
/**
381
 * Empty mmx state.
382
 * this must be called between any dsp function and float/double code.
383
 * for example sin(); dsp->idct_put(); emms_c(); cos()
384
 */
385
#define emms_c()
386

    
387
/* should be defined by architectures supporting
388
   one or more MultiMedia extension */
389
int mm_support(void);
390

    
391
#define __align16 __attribute__ ((aligned (16)))
392

    
393
#if defined(HAVE_MMX)
394

    
395
#undef emms_c
396

    
397
#define MM_MMX    0x0001 /* standard MMX */
398
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
399
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
400
#define MM_SSE    0x0008 /* SSE functions */
401
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
402
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
403

    
404
extern int mm_flags;
405

    
406
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
407
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
408
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
409

    
410
static inline void emms(void)
411
{
412
    __asm __volatile ("emms;":::"memory");
413
}
414

    
415

    
416
#define emms_c() \
417
{\
418
    if (mm_flags & MM_MMX)\
419
        emms();\
420
}
421

    
422
#define __align8 __attribute__ ((aligned (8)))
423
#define STRIDE_ALIGN 8
424

    
425
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
426
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
427

    
428
#elif defined(ARCH_ARMV4L)
429

    
430
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
431
   line optimizations */
432
#define __align8 __attribute__ ((aligned (4)))
433
#define STRIDE_ALIGN 4
434

    
435
#define MM_IWMMXT    0x0100 /* XScale IWMMXT */
436

    
437
extern int mm_flags;
438

    
439
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
440

    
441
#elif defined(HAVE_MLIB)
442

    
443
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
444
#define __align8 __attribute__ ((aligned (8)))
445
#define STRIDE_ALIGN 8
446

    
447
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
448

    
449
#elif defined(ARCH_SPARC)
450

    
451
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
452
#define __align8 __attribute__ ((aligned (8)))
453
#define STRIDE_ALIGN 8
454
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
455

    
456
#elif defined(ARCH_ALPHA)
457

    
458
#define __align8 __attribute__ ((aligned (8)))
459
#define STRIDE_ALIGN 8
460

    
461
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
462

    
463
#elif defined(ARCH_POWERPC)
464

    
465
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
466

    
467
extern int mm_flags;
468

    
469
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
470
#define pixel altivec_pixel
471
#include <altivec.h>
472
#undef pixel
473
#endif
474

    
475
#define __align8 __attribute__ ((aligned (16)))
476
#define STRIDE_ALIGN 16
477

    
478
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
479

    
480
#elif defined(HAVE_MMI)
481

    
482
#define __align8 __attribute__ ((aligned (16)))
483
#define STRIDE_ALIGN 16
484

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

    
487
#elif defined(ARCH_SH4)
488

    
489
#define __align8 __attribute__ ((aligned (8)))
490
#define STRIDE_ALIGN 8
491

    
492
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
493

    
494
#else
495

    
496
#define __align8 __attribute__ ((aligned (8)))
497
#define STRIDE_ALIGN 8
498

    
499
#endif
500

    
501
#ifdef __GNUC__
502

    
503
struct unaligned_64 { uint64_t l; } __attribute__((packed));
504
struct unaligned_32 { uint32_t l; } __attribute__((packed));
505
struct unaligned_16 { uint16_t l; } __attribute__((packed));
506

    
507
#define LD16(a) (((const struct unaligned_16 *) (a))->l)
508
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
509
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
510

    
511
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
512

    
513
#else /* __GNUC__ */
514

    
515
#define LD16(a) (*((uint16_t*)(a)))
516
#define LD32(a) (*((uint32_t*)(a)))
517
#define LD64(a) (*((uint64_t*)(a)))
518

    
519
#define ST32(a, b) *((uint32_t*)(a)) = (b)
520

    
521
#endif /* !__GNUC__ */
522

    
523
/* PSNR */
524
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
525
              int orig_linesize[3], int coded_linesize,
526
              AVCodecContext *avctx);
527

    
528
/* FFT computation */
529

    
530
/* NOTE: soon integer code will be added, so you must use the
531
   FFTSample type */
532
typedef float FFTSample;
533

    
534
typedef struct FFTComplex {
535
    FFTSample re, im;
536
} FFTComplex;
537

    
538
typedef struct FFTContext {
539
    int nbits;
540
    int inverse;
541
    uint16_t *revtab;
542
    FFTComplex *exptab;
543
    FFTComplex *exptab1; /* only used by SSE code */
544
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
545
} FFTContext;
546

    
547
int ff_fft_init(FFTContext *s, int nbits, int inverse);
548
void ff_fft_permute(FFTContext *s, FFTComplex *z);
549
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
550
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
551
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
552

    
553
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
554
{
555
    s->fft_calc(s, z);
556
}
557
void ff_fft_end(FFTContext *s);
558

    
559
/* MDCT computation */
560

    
561
typedef struct MDCTContext {
562
    int n;  /* size of MDCT (i.e. number of input data * 2) */
563
    int nbits; /* n = 2^nbits */
564
    /* pre/post rotation tables */
565
    FFTSample *tcos;
566
    FFTSample *tsin;
567
    FFTContext fft;
568
} MDCTContext;
569

    
570
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
571
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
572
                const FFTSample *input, FFTSample *tmp);
573
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
574
               const FFTSample *input, FFTSample *tmp);
575
void ff_mdct_end(MDCTContext *s);
576

    
577
#define WARPER8_16(name8, name16)\
578
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
579
    return name8(s, dst           , src           , stride, h)\
580
          +name8(s, dst+8         , src+8         , stride, h);\
581
}
582

    
583
#define WARPER8_16_SQ(name8, name16)\
584
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
585
    int score=0;\
586
    score +=name8(s, dst           , src           , stride, 8);\
587
    score +=name8(s, dst+8         , src+8         , stride, 8);\
588
    if(h==16){\
589
        dst += 8*stride;\
590
        src += 8*stride;\
591
        score +=name8(s, dst           , src           , stride, 8);\
592
        score +=name8(s, dst+8         , src+8         , stride, 8);\
593
    }\
594
    return score;\
595
}
596

    
597
#ifndef HAVE_LRINTF
598
/* XXX: add ISOC specific test to avoid specific BSD testing. */
599
/* better than nothing implementation. */
600
/* btw, rintf() is existing on fbsd too -- alex */
601
static always_inline long int lrintf(float x)
602
{
603
#ifdef CONFIG_WIN32
604
#  ifdef ARCH_X86
605
    int32_t i;
606
    asm volatile(
607
        "fistpl %0\n\t"
608
        : "=m" (i) : "t" (x) : "st"
609
    );
610
    return i;
611
#  else
612
    /* XXX: incorrect, but make it compile */
613
    return (int)(x + (x < 0 ? -0.5 : 0.5));
614
#  endif
615
#else
616
    return (int)(rint(x));
617
#endif
618
}
619
#else
620
#ifndef _ISOC9X_SOURCE
621
#define _ISOC9X_SOURCE
622
#endif
623
#include <math.h>
624
#endif
625

    
626
#endif