-sws 0 && -ssf chs fixed
[libav.git] / libavcodec / dsputil.h
1 /*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20 /**
21 * @file dsputil.h
22 * DSP utils.
23 * note, many functions in here may use MMX which trashes the FPU state, it is
24 * absolutely necessary to call emms_c() between dsp & float/double code
25 */
26
27 #ifndef DSPUTIL_H
28 #define DSPUTIL_H
29
30 #include "common.h"
31 #include "avcodec.h"
32
33
34 //#define DEBUG
35 /* dct code */
36 typedef short DCTELEM;
37
38 void fdct_ifast (DCTELEM *data);
39 void fdct_ifast248 (DCTELEM *data);
40 void ff_jpeg_fdct_islow (DCTELEM *data);
41 void ff_fdct248_islow (DCTELEM *data);
42
43 void j_rev_dct (DCTELEM *data);
44
45 void ff_fdct_mmx(DCTELEM *block);
46 void ff_fdct_mmx2(DCTELEM *block);
47
48 /* encoding scans */
49 extern const uint8_t ff_alternate_horizontal_scan[64];
50 extern const uint8_t ff_alternate_vertical_scan[64];
51 extern const uint8_t ff_zigzag_direct[64];
52 extern const uint8_t ff_zigzag248_direct[64];
53
54 /* pixel operations */
55 #define MAX_NEG_CROP 384
56
57 /* temporary */
58 extern uint32_t squareTbl[512];
59 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
60
61
62 /* minimum alignment rules ;)
63 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
64 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
65
66 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
67 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
68
69 !future video codecs might need functions with less strict alignment
70 */
71
72 /*
73 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
74 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
75 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
76 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
77 void clear_blocks_c(DCTELEM *blocks);
78 */
79
80 /* add and put pixel (decoding) */
81 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
82 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
83 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);
84 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
85 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
86
87 #define DEF_OLD_QPEL(name)\
88 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
89 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
90 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
91
92 DEF_OLD_QPEL(qpel16_mc11_old_c)
93 DEF_OLD_QPEL(qpel16_mc31_old_c)
94 DEF_OLD_QPEL(qpel16_mc12_old_c)
95 DEF_OLD_QPEL(qpel16_mc32_old_c)
96 DEF_OLD_QPEL(qpel16_mc13_old_c)
97 DEF_OLD_QPEL(qpel16_mc33_old_c)
98 DEF_OLD_QPEL(qpel8_mc11_old_c)
99 DEF_OLD_QPEL(qpel8_mc31_old_c)
100 DEF_OLD_QPEL(qpel8_mc12_old_c)
101 DEF_OLD_QPEL(qpel8_mc32_old_c)
102 DEF_OLD_QPEL(qpel8_mc13_old_c)
103 DEF_OLD_QPEL(qpel8_mc33_old_c)
104
105 #define CALL_2X_PIXELS(a, b, n)\
106 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
107 b(block , pixels , line_size, h);\
108 b(block+n, pixels+n, line_size, h);\
109 }
110
111 /* motion estimation */
112
113 typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
114
115 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
116
117
118 /**
119 * DSPContext.
120 */
121 typedef struct DSPContext {
122 /* pixel ops : interface with DCT */
123 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
124 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
125 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
126 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
127 /**
128 * translational global motion compensation.
129 */
130 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
131 /**
132 * global motion compensation.
133 */
134 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
135 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
136 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
137 int (*pix_sum)(uint8_t * pix, int line_size);
138 int (*pix_norm1)(uint8_t * pix, int line_size);
139 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
140 me_cmp_func sse[2];
141 me_cmp_func hadamard8_diff[2];
142 me_cmp_func dct_sad[2];
143 me_cmp_func quant_psnr[2];
144 me_cmp_func bit[2];
145 me_cmp_func rd[2];
146 int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
147
148 me_cmp_func me_pre_cmp[11];
149 me_cmp_func me_cmp[11];
150 me_cmp_func me_sub_cmp[11];
151 me_cmp_func mb_cmp[11];
152
153 /* maybe create an array for 16/8/4/2 functions */
154 /**
155 * Halfpel motion compensation with rounding (a+b+1)>>1.
156 * this is an array[4][4] of motion compensation funcions for 4
157 * horizontal blocksizes (2,4,8,16) and the 4 halfpel positions<br>
158 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
159 * @param block destination where the result is stored
160 * @param pixels source
161 * @param line_size number of bytes in a horizontal line of block
162 * @param h height
163 */
164 op_pixels_func put_pixels_tab[4][4];
165
166 /**
167 * Halfpel motion compensation with rounding (a+b+1)>>1.
168 * This is an array[4][4] of motion compensation functions for 4
169 * horizontal blocksizes (2,4,8,16) and the 4 halfpel positions<br>
170 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
171 * @param block destination into which the result is averaged (a+b+1)>>1
172 * @param pixels source
173 * @param line_size number of bytes in a horizontal line of block
174 * @param h height
175 */
176 op_pixels_func avg_pixels_tab[4][4];
177
178 /**
179 * Halfpel motion compensation with no rounding (a+b)>>1.
180 * this is an array[2][4] of motion compensation funcions for 2
181 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
182 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
183 * @param block destination where the result is stored
184 * @param pixels source
185 * @param line_size number of bytes in a horizontal line of block
186 * @param h height
187 */
188 op_pixels_func put_no_rnd_pixels_tab[2][4];
189
190 /**
191 * Halfpel motion compensation with no rounding (a+b)>>1.
192 * this is an array[2][4] of motion compensation funcions for 2
193 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
194 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
195 * @param block destination into which the result is averaged (a+b)>>1
196 * @param pixels source
197 * @param line_size number of bytes in a horizontal line of block
198 * @param h height
199 */
200 op_pixels_func avg_no_rnd_pixels_tab[2][4];
201
202 /**
203 * Thirdpel motion compensation with rounding (a+b+1)>>1.
204 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
205 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
206 * @param block destination where the result is stored
207 * @param pixels source
208 * @param line_size number of bytes in a horizontal line of block
209 * @param h height
210 */
211 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
212 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
213
214 qpel_mc_func put_qpel_pixels_tab[2][16];
215 qpel_mc_func avg_qpel_pixels_tab[2][16];
216 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
217 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
218 qpel_mc_func put_mspel_pixels_tab[8];
219
220 /**
221 * h264 Chram MC
222 */
223 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
224 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
225
226 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
227 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
228
229 op_pixels_abs_func pix_abs16x16;
230 op_pixels_abs_func pix_abs16x16_x2;
231 op_pixels_abs_func pix_abs16x16_y2;
232 op_pixels_abs_func pix_abs16x16_xy2;
233 op_pixels_abs_func pix_abs8x8;
234 op_pixels_abs_func pix_abs8x8_x2;
235 op_pixels_abs_func pix_abs8x8_y2;
236 op_pixels_abs_func pix_abs8x8_xy2;
237
238 /* huffyuv specific */
239 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
240 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
241 /**
242 * subtract huffyuv's variant of median prediction
243 * note, this might read from src1[-1], src2[-1]
244 */
245 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
246 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
247
248 /* (I)DCT */
249 void (*fdct)(DCTELEM *block/* align 16*/);
250 void (*fdct248)(DCTELEM *block/* align 16*/);
251
252 /* IDCT really*/
253 void (*idct)(DCTELEM *block/* align 16*/);
254
255 /**
256 * block -> idct -> clip to unsigned 8 bit -> dest.
257 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
258 * @param line_size size in bytes of a horizotal line of dest
259 */
260 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
261
262 /**
263 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
264 * @param line_size size in bytes of a horizotal line of dest
265 */
266 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
267
268 /**
269 * idct input permutation.
270 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
271 * IDCT)
272 * this permutation must be performed before the idct_put/add, note, normally this can be merged
273 * with the zigzag/alternate scan<br>
274 * an example to avoid confusion:
275 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
276 * - (x -> referece dct -> reference idct -> x)
277 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
278 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
279 */
280 uint8_t idct_permutation[64];
281 int idct_permutation_type;
282 #define FF_NO_IDCT_PERM 1
283 #define FF_LIBMPEG2_IDCT_PERM 2
284 #define FF_SIMPLE_IDCT_PERM 3
285 #define FF_TRANSPOSE_IDCT_PERM 4
286
287 } DSPContext;
288
289 void dsputil_static_init(void);
290 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
291
292 /**
293 * permute block according to permuatation.
294 * @param last last non zero element in scantable order
295 */
296 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
297
298 #define BYTE_VEC32(c) ((c)*0x01010101UL)
299
300 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
301 {
302 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
303 }
304
305 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
306 {
307 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
308 }
309
310 /**
311 * Empty mmx state.
312 * this must be called between any dsp function and float/double code.
313 * for example sin(); dsp->idct_put(); emms_c(); cos()
314 */
315 #define emms_c()
316
317 /* should be defined by architectures supporting
318 one or more MultiMedia extension */
319 int mm_support(void);
320
321 #if defined(HAVE_MMX)
322
323 #undef emms_c
324
325 #define MM_MMX 0x0001 /* standard MMX */
326 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
327 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
328 #define MM_SSE 0x0008 /* SSE functions */
329 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
330
331 extern int mm_flags;
332
333 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
334 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
335
336 static inline void emms(void)
337 {
338 __asm __volatile ("emms;":::"memory");
339 }
340
341
342 #define emms_c() \
343 {\
344 if (mm_flags & MM_MMX)\
345 emms();\
346 }
347
348 #define __align8 __attribute__ ((aligned (8)))
349
350 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
351 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
352
353 #elif defined(ARCH_ARMV4L)
354
355 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
356 line ptimizations */
357 #define __align8 __attribute__ ((aligned (4)))
358
359 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
360
361 #elif defined(HAVE_MLIB)
362
363 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
364 #define __align8 __attribute__ ((aligned (8)))
365
366 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
367
368 #elif defined(ARCH_ALPHA)
369
370 #define __align8 __attribute__ ((aligned (8)))
371
372 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
373
374 #elif defined(ARCH_POWERPC)
375
376 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
377
378 extern int mm_flags;
379
380 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
381 #include <altivec.h>
382 #endif
383
384 #define __align8 __attribute__ ((aligned (16)))
385
386 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
387
388 #elif defined(HAVE_MMI)
389
390 #define __align8 __attribute__ ((aligned (16)))
391
392 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
393
394 #elif defined(ARCH_SH4)
395
396 #define __align8 __attribute__ ((aligned (8)))
397
398 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
399
400 #else
401
402 #define __align8
403
404 #endif
405
406 #ifdef __GNUC__
407
408 struct unaligned_64 { uint64_t l; } __attribute__((packed));
409 struct unaligned_32 { uint32_t l; } __attribute__((packed));
410 struct unaligned_16 { uint16_t l; } __attribute__((packed));
411
412 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
413 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
414 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
415
416 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
417
418 #else /* __GNUC__ */
419
420 #define LD16(a) (*((uint16_t*)(a)))
421 #define LD32(a) (*((uint32_t*)(a)))
422 #define LD64(a) (*((uint64_t*)(a)))
423
424 #define ST32(a, b) *((uint32_t*)(a)) = (b)
425
426 #endif /* !__GNUC__ */
427
428 /* PSNR */
429 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
430 int orig_linesize[3], int coded_linesize,
431 AVCodecContext *avctx);
432
433 /* FFT computation */
434
435 /* NOTE: soon integer code will be added, so you must use the
436 FFTSample type */
437 typedef float FFTSample;
438
439 typedef struct FFTComplex {
440 FFTSample re, im;
441 } FFTComplex;
442
443 typedef struct FFTContext {
444 int nbits;
445 int inverse;
446 uint16_t *revtab;
447 FFTComplex *exptab;
448 FFTComplex *exptab1; /* only used by SSE code */
449 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
450 } FFTContext;
451
452 int fft_init(FFTContext *s, int nbits, int inverse);
453 void fft_permute(FFTContext *s, FFTComplex *z);
454 void fft_calc_c(FFTContext *s, FFTComplex *z);
455 void fft_calc_sse(FFTContext *s, FFTComplex *z);
456 void fft_calc_altivec(FFTContext *s, FFTComplex *z);
457
458 static inline void fft_calc(FFTContext *s, FFTComplex *z)
459 {
460 s->fft_calc(s, z);
461 }
462 void fft_end(FFTContext *s);
463
464 /* MDCT computation */
465
466 typedef struct MDCTContext {
467 int n; /* size of MDCT (i.e. number of input data * 2) */
468 int nbits; /* n = 2^nbits */
469 /* pre/post rotation tables */
470 FFTSample *tcos;
471 FFTSample *tsin;
472 FFTContext fft;
473 } MDCTContext;
474
475 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
476 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
477 const FFTSample *input, FFTSample *tmp);
478 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
479 const FFTSample *input, FFTSample *tmp);
480 void ff_mdct_end(MDCTContext *s);
481
482 #define WARPER88_1616(name8, name16)\
483 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
484 return name8(s, dst , src , stride)\
485 +name8(s, dst+8 , src+8 , stride)\
486 +name8(s, dst +8*stride, src +8*stride, stride)\
487 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
488 }
489
490 #ifndef HAVE_LRINTF
491 /* XXX: add ISOC specific test to avoid specific BSD testing. */
492 /* better than nothing implementation. */
493 /* btw, rintf() is existing on fbsd too -- alex */
494 static inline long int lrintf(float x)
495 {
496 #ifdef CONFIG_WIN32
497 /* XXX: incorrect, but make it compile */
498 return (int)(x);
499 #else
500 return (int)(rint(x));
501 #endif
502 }
503 #endif
504
505 #if defined(CONFIG_OS2) || defined(CONFIG_SUNOS)
506 static inline float floorf(float f) {
507 return floor(f);
508 }
509 #endif
510
511 #endif