42 |
* * |
* * |
43 |
* Revision history: * |
* Revision history: * |
44 |
* * |
* * |
45 |
* 29.03.2002 interlacing speedup - used transfer strides instead of |
* 29.03.2002 interlacing speedup - used transfer strides instead of * |
46 |
* manual field-to-frame conversion |
* manual field-to-frame conversion * |
47 |
* 26.03.2002 interlacing support - moved transfers outside loops |
* 26.03.2002 interlacing support - moved transfers outside loops * |
48 |
* 22.12.2001 get_dc_scaler() moved to common.h |
* 22.12.2001 get_dc_scaler() moved to common.h * |
49 |
* 19.11.2001 introduced coefficient thresholding (Isibaar) * |
* 19.11.2001 introduced coefficient thresholding (Isibaar) * |
50 |
* 17.11.2001 initial version * |
* 17.11.2001 initial version * |
51 |
* * |
* * |
68 |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
69 |
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
70 |
|
|
71 |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
#define TOOSMALL_LIMIT 3 /* skip blocks having a coefficient sum below this value */ |
72 |
|
|
73 |
/* this isnt pretty, but its better than 20 ifdefs */ |
/* this isnt pretty, but its better than 20 ifdefs */ |
74 |
|
|
207 |
stop_quant_timer(); |
stop_quant_timer(); |
208 |
} |
} |
209 |
|
|
210 |
if (sum >= TOOSMALL_LIMIT) { // skip block ? |
if ((sum >= TOOSMALL_LIMIT) || (qcoeff[i*64] != 0) || |
211 |
|
(qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
212 |
|
|
213 |
if (pParam->m_quant_type == H263_QUANT) { |
if (pParam->m_quant_type == H263_QUANT) { |
214 |
start_timer(); |
start_timer(); |
252 |
|
|
253 |
} |
} |
254 |
|
|
255 |
|
void |
256 |
|
MBTransQuantIntra2(const MBParam * pParam, |
257 |
|
FRAMEINFO * frame, |
258 |
|
MACROBLOCK * pMB, |
259 |
|
const uint32_t x_pos, |
260 |
|
const uint32_t y_pos, |
261 |
|
int16_t data[6 * 64], |
262 |
|
int16_t qcoeff[6 * 64]) |
263 |
|
{ |
264 |
|
MBTrans(pParam,frame,pMB,x_pos,y_pos,data); |
265 |
|
MBfDCT(pParam,frame,pMB,data); |
266 |
|
MBQuantIntra(pParam,frame,pMB,data,qcoeff); |
267 |
|
MBDeQuantIntra(pParam,frame->quant,data,qcoeff); |
268 |
|
MBiDCT(data,0x3F); |
269 |
|
MBTransAdd(pParam,frame,pMB,x_pos,y_pos,data,0x3F); |
270 |
|
} |
271 |
|
|
272 |
|
|
273 |
|
uint8_t |
274 |
|
MBTransQuantInter2(const MBParam * pParam, |
275 |
|
FRAMEINFO * frame, |
276 |
|
MACROBLOCK * pMB, |
277 |
|
const uint32_t x_pos, |
278 |
|
const uint32_t y_pos, |
279 |
|
int16_t data[6 * 64], |
280 |
|
int16_t qcoeff[6 * 64]) |
281 |
|
{ |
282 |
|
uint8_t cbp; |
283 |
|
|
284 |
|
/* there is no MBTrans for Inter block, that's done in motion compensation already */ |
285 |
|
|
286 |
|
MBfDCT(pParam,frame,pMB,data); |
287 |
|
cbp = MBQuantInter(pParam,frame->quant,data,qcoeff); |
288 |
|
MBDeQuantInter(pParam,frame->quant,data,qcoeff,cbp); |
289 |
|
MBiDCT(data,cbp); |
290 |
|
MBTransAdd(pParam,frame,pMB,x_pos,y_pos,data,cbp); |
291 |
|
|
292 |
|
return cbp; |
293 |
|
} |
294 |
|
|
295 |
|
uint8_t |
296 |
|
MBTransQuantInterBVOP(const MBParam * pParam, |
297 |
|
FRAMEINFO * frame, |
298 |
|
MACROBLOCK * pMB, |
299 |
|
const uint32_t x_pos, |
300 |
|
const uint32_t y_pos, |
301 |
|
int16_t data[6 * 64], |
302 |
|
int16_t qcoeff[6 * 64]) |
303 |
|
{ |
304 |
|
uint8_t cbp; |
305 |
|
|
306 |
|
/* there is no MBTrans for Inter block, that's done in motion compensation already */ |
307 |
|
|
308 |
|
MBfDCT(pParam,frame,pMB,data); |
309 |
|
cbp = MBQuantInter(pParam,frame->quant,data,qcoeff); |
310 |
|
|
311 |
|
/* we don't have to DeQuant, iDCT and Transfer back data for B-frames */ |
312 |
|
|
313 |
|
return cbp; |
314 |
|
} |
315 |
|
|
316 |
|
|
317 |
|
void |
318 |
|
MBfDCT(const MBParam * pParam, |
319 |
|
FRAMEINFO * frame, |
320 |
|
MACROBLOCK * pMB, |
321 |
|
int16_t data[6 * 64]) |
322 |
|
{ |
323 |
|
int i; |
324 |
|
|
325 |
|
start_timer(); |
326 |
|
pMB->field_dct = 0; |
327 |
|
if ((frame->global_flags & XVID_INTERLACING)) { |
328 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
329 |
|
} |
330 |
|
stop_interlacing_timer(); |
331 |
|
|
332 |
|
for (i = 0; i < 6; i++) { |
333 |
|
start_timer(); |
334 |
|
fdct(&data[i * 64]); |
335 |
|
stop_dct_timer(); |
336 |
|
} |
337 |
|
} |
338 |
|
|
339 |
|
void |
340 |
|
MBQuantDeQuantIntra(const MBParam * pParam, |
341 |
|
FRAMEINFO * frame, |
342 |
|
MACROBLOCK * pMB, |
343 |
|
int16_t qcoeff[6 * 64], |
344 |
|
int16_t data[6*64]) |
345 |
|
{ |
346 |
|
int i; |
347 |
|
int iQuant = frame->quant; |
348 |
|
|
349 |
|
start_timer(); |
350 |
|
pMB->field_dct = 0; |
351 |
|
if ((frame->global_flags & XVID_INTERLACING)) { |
352 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
353 |
|
} |
354 |
|
stop_interlacing_timer(); |
355 |
|
|
356 |
|
for (i = 0; i < 6; i++) { |
357 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
358 |
|
|
359 |
|
if (pParam->m_quant_type == H263_QUANT) { |
360 |
|
start_timer(); |
361 |
|
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
362 |
|
stop_quant_timer(); |
363 |
|
|
364 |
|
start_timer(); |
365 |
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
366 |
|
stop_iquant_timer(); |
367 |
|
} else { |
368 |
|
start_timer(); |
369 |
|
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
370 |
|
stop_quant_timer(); |
371 |
|
|
372 |
|
start_timer(); |
373 |
|
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
374 |
|
stop_iquant_timer(); |
375 |
|
} |
376 |
|
} |
377 |
|
} |
378 |
|
|
379 |
|
void |
380 |
|
MBQuantIntra(const MBParam * pParam, |
381 |
|
FRAMEINFO * frame, |
382 |
|
MACROBLOCK *pMB, |
383 |
|
int16_t qcoeff[6 * 64], |
384 |
|
int16_t data[6*64]) |
385 |
|
{ |
386 |
|
int i; |
387 |
|
int iQuant = frame->quant; |
388 |
|
|
389 |
|
start_timer(); |
390 |
|
pMB->field_dct = 0; |
391 |
|
if ((frame->global_flags & XVID_INTERLACING)) { |
392 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
393 |
|
} |
394 |
|
stop_interlacing_timer(); |
395 |
|
|
396 |
|
for (i = 0; i < 6; i++) { |
397 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
398 |
|
|
399 |
|
if (pParam->m_quant_type == H263_QUANT) { |
400 |
|
start_timer(); |
401 |
|
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
402 |
|
stop_quant_timer(); |
403 |
|
} else { |
404 |
|
start_timer(); |
405 |
|
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
406 |
|
stop_quant_timer(); |
407 |
|
} |
408 |
|
} |
409 |
|
} |
410 |
|
|
411 |
|
void |
412 |
|
MBDeQuantIntra(const MBParam * pParam, |
413 |
|
const int iQuant, |
414 |
|
int16_t qcoeff[6 * 64], |
415 |
|
int16_t data[6*64]) |
416 |
|
{ |
417 |
|
int i; |
418 |
|
|
419 |
|
for (i = 0; i < 6; i++) { |
420 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
421 |
|
|
422 |
|
if (pParam->m_quant_type == H263_QUANT) { |
423 |
|
start_timer(); |
424 |
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
425 |
|
stop_iquant_timer(); |
426 |
|
} else { |
427 |
|
start_timer(); |
428 |
|
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
429 |
|
stop_iquant_timer(); |
430 |
|
} |
431 |
|
} |
432 |
|
} |
433 |
|
|
434 |
|
uint8_t |
435 |
|
MBQuantInter(const MBParam * pParam, |
436 |
|
const int iQuant, |
437 |
|
int16_t data[6 * 64], |
438 |
|
int16_t qcoeff[6 * 64]) |
439 |
|
{ |
440 |
|
|
441 |
|
int i; |
442 |
|
uint8_t cbp = 0; |
443 |
|
int sum; |
444 |
|
|
445 |
|
for (i = 0; i < 6; i++) { |
446 |
|
|
447 |
|
if (pParam->m_quant_type == 0) { |
448 |
|
start_timer(); |
449 |
|
sum = quant_inter(&qcoeff[i * 64], &data[i * 64], iQuant); |
450 |
|
stop_quant_timer(); |
451 |
|
} else { |
452 |
|
start_timer(); |
453 |
|
sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], iQuant); |
454 |
|
stop_quant_timer(); |
455 |
|
} |
456 |
|
|
457 |
|
if (sum >= TOOSMALL_LIMIT) { // skip block ? |
458 |
|
cbp |= 1 << (5 - i); |
459 |
|
} |
460 |
|
} |
461 |
|
return cbp; |
462 |
|
} |
463 |
|
|
464 |
|
void |
465 |
|
MBDeQuantInter( const MBParam * pParam, |
466 |
|
const int iQuant, |
467 |
|
int16_t data[6 * 64], |
468 |
|
int16_t qcoeff[6 * 64], |
469 |
|
const uint8_t cbp) |
470 |
|
{ |
471 |
|
int i; |
472 |
|
|
473 |
|
for (i = 0; i < 6; i++) { |
474 |
|
if (cbp & (1 << (5 - i))) |
475 |
|
{ |
476 |
|
if (pParam->m_quant_type == H263_QUANT) { |
477 |
|
start_timer(); |
478 |
|
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
479 |
|
stop_iquant_timer(); |
480 |
|
} else { |
481 |
|
start_timer(); |
482 |
|
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
483 |
|
stop_iquant_timer(); |
484 |
|
} |
485 |
|
} |
486 |
|
} |
487 |
|
} |
488 |
|
|
489 |
|
void |
490 |
|
MBiDCT( int16_t data[6 * 64], |
491 |
|
const uint8_t cbp) |
492 |
|
{ |
493 |
|
int i; |
494 |
|
|
495 |
|
for (i = 0; i < 6; i++) { |
496 |
|
if (cbp & (1 << (5 - i))) |
497 |
|
{ |
498 |
|
start_timer(); |
499 |
|
idct(&data[i * 64]); |
500 |
|
stop_idct_timer(); |
501 |
|
|
502 |
|
} |
503 |
|
} |
504 |
|
} |
505 |
|
|
506 |
|
|
507 |
|
void |
508 |
|
MBTrans(const MBParam * pParam, |
509 |
|
FRAMEINFO * frame, |
510 |
|
MACROBLOCK * pMB, |
511 |
|
const uint32_t x_pos, |
512 |
|
const uint32_t y_pos, |
513 |
|
int16_t data[6 * 64]) |
514 |
|
{ |
515 |
|
uint32_t stride = pParam->edged_width; |
516 |
|
uint32_t stride2 = stride / 2; |
517 |
|
uint32_t next_block = stride * 8; |
518 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
519 |
|
IMAGE *pCurrent = &frame->image; |
520 |
|
|
521 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
522 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
523 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
524 |
|
|
525 |
|
start_timer(); |
526 |
|
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
527 |
|
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
528 |
|
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
529 |
|
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
530 |
|
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
531 |
|
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
532 |
|
stop_transfer_timer(); |
533 |
|
} |
534 |
|
|
535 |
|
void |
536 |
|
MBTransAdd(const MBParam * pParam, |
537 |
|
FRAMEINFO * frame, |
538 |
|
MACROBLOCK * pMB, |
539 |
|
const uint32_t x_pos, |
540 |
|
const uint32_t y_pos, |
541 |
|
int16_t data[6 * 64], |
542 |
|
const uint8_t cbp) |
543 |
|
{ |
544 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
545 |
|
uint32_t stride = pParam->edged_width; |
546 |
|
uint32_t stride2 = stride / 2; |
547 |
|
uint32_t next_block = stride * 8; |
548 |
|
IMAGE *pCurrent = &frame->image; |
549 |
|
|
550 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
551 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
552 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
553 |
|
|
554 |
|
if (pMB->field_dct) { |
555 |
|
next_block = stride; |
556 |
|
stride *= 2; |
557 |
|
} |
558 |
|
|
559 |
|
start_timer(); |
560 |
|
if (cbp & 32) |
561 |
|
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
562 |
|
if (cbp & 16) |
563 |
|
transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride); |
564 |
|
if (cbp & 8) |
565 |
|
transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride); |
566 |
|
if (cbp & 4) |
567 |
|
transfer_16to8add(pY_Cur + next_block + 8, &data[3 * 64], stride); |
568 |
|
if (cbp & 2) |
569 |
|
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
570 |
|
if (cbp & 1) |
571 |
|
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
572 |
|
stop_transfer_timer(); |
573 |
|
} |
574 |
|
|
575 |
|
|
576 |
|
|
577 |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
578 |
|
|