32 |
* |
* |
33 |
* History: |
* History: |
34 |
* |
* |
35 |
|
* 26.03.2002 interlacing support - moved transfers outside decode loop |
36 |
* 26.12.2001 decoder_mbinter: dequant/idct moved within if(coded) block |
* 26.12.2001 decoder_mbinter: dequant/idct moved within if(coded) block |
37 |
* 22.12.2001 block based interpolation |
* 22.12.2001 block based interpolation |
38 |
* 01.12.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
* 01.12.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
55 |
#include "dct/fdct.h" |
#include "dct/fdct.h" |
56 |
#include "utils/mem_transfer.h" |
#include "utils/mem_transfer.h" |
57 |
#include "image/interpolate8x8.h" |
#include "image/interpolate8x8.h" |
58 |
|
#include "utils/mbfunctions.h" |
59 |
|
|
60 |
#include "bitstream/mbcoding.h" |
#include "bitstream/mbcoding.h" |
61 |
#include "prediction/mbprediction.h" |
#include "prediction/mbprediction.h" |
70 |
{ |
{ |
71 |
DECODER * dec; |
DECODER * dec; |
72 |
|
|
73 |
dec = xvid_malloc(sizeof(DECODER), 16); |
dec = xvid_malloc(sizeof(DECODER), CACHE_LINE); |
74 |
if (dec == NULL) |
if (dec == NULL) |
75 |
{ |
{ |
76 |
return XVID_ERR_MEMORY; |
return XVID_ERR_MEMORY; |
99 |
return XVID_ERR_MEMORY; |
return XVID_ERR_MEMORY; |
100 |
} |
} |
101 |
|
|
102 |
dec->mbs = xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height, 16); |
dec->mbs = xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height, CACHE_LINE); |
103 |
if (dec->mbs == NULL) |
if (dec->mbs == NULL) |
104 |
{ |
{ |
105 |
image_destroy(&dec->cur, dec->edged_width, dec->edged_height); |
image_destroy(&dec->cur, dec->edged_width, dec->edged_height); |
137 |
|
|
138 |
// decode an intra macroblock |
// decode an intra macroblock |
139 |
|
|
140 |
void decoder_mbintra(DECODER * dec, MACROBLOCK * mb, int x, int y, uint32_t acpred_flag, uint32_t cbp, Bitstream * bs, int quant, int intra_dc_threshold) |
void decoder_mbintra(DECODER * dec, |
141 |
|
MACROBLOCK * pMB, |
142 |
|
const uint32_t x_pos, |
143 |
|
const uint32_t y_pos, |
144 |
|
const uint32_t acpred_flag, |
145 |
|
const uint32_t cbp, |
146 |
|
Bitstream * bs, |
147 |
|
const uint32_t quant, |
148 |
|
const uint32_t intra_dc_threshold) |
149 |
{ |
{ |
150 |
uint32_t k; |
CACHE_ALIGN int16_t block[6][64]; |
151 |
|
CACHE_ALIGN int16_t data[6][64]; |
152 |
|
|
153 |
for (k = 0; k < 6; k++) |
const uint32_t stride = dec->edged_width; |
154 |
|
uint32_t i; |
155 |
|
uint32_t iQuant = pMB->quant; |
156 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
157 |
|
|
158 |
|
pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); |
159 |
|
pU_Cur = dec->cur.u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
160 |
|
pV_Cur = dec->cur.v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
161 |
|
|
162 |
|
memset(block, 0, sizeof(block)); // clear |
163 |
|
|
164 |
|
for (i = 0; i < 6; i++) |
165 |
{ |
{ |
166 |
uint32_t dcscalar; |
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
|
int16_t block[64]; |
|
|
int16_t data[64]; |
|
167 |
int16_t predictors[8]; |
int16_t predictors[8]; |
168 |
int start_coeff; |
int start_coeff; |
169 |
|
|
|
dcscalar = get_dc_scaler(mb->quant, k < 4); |
|
|
|
|
170 |
start_timer(); |
start_timer(); |
171 |
predict_acdc(dec->mbs, x, y, dec->mb_width, k, block, mb->quant, dcscalar, predictors); |
predict_acdc(dec->mbs, x_pos, y_pos, dec->mb_width, i, block[i], iQuant, iDcScaler, predictors); |
172 |
if (!acpred_flag) |
if (!acpred_flag) |
173 |
{ |
{ |
174 |
mb->acpred_directions[k] = 0; |
pMB->acpred_directions[i] = 0; |
175 |
} |
} |
176 |
stop_prediction_timer(); |
stop_prediction_timer(); |
177 |
|
|
|
memset(block, 0, 64*sizeof(int16_t)); // clear |
|
|
|
|
178 |
if (quant < intra_dc_threshold) |
if (quant < intra_dc_threshold) |
179 |
{ |
{ |
180 |
int dc_size; |
int dc_size; |
181 |
int dc_dif; |
int dc_dif; |
182 |
|
|
183 |
dc_size = k < 4 ? get_dc_size_lum(bs) : get_dc_size_chrom(bs); |
dc_size = i < 4 ? get_dc_size_lum(bs) : get_dc_size_chrom(bs); |
184 |
dc_dif = dc_size ? get_dc_dif(bs, dc_size) : 0 ; |
dc_dif = dc_size ? get_dc_dif(bs, dc_size) : 0 ; |
185 |
|
|
186 |
if (dc_size > 8) |
if (dc_size > 8) |
188 |
BitstreamSkip(bs, 1); // marker |
BitstreamSkip(bs, 1); // marker |
189 |
} |
} |
190 |
|
|
191 |
block[0] = dc_dif; |
block[i][0] = dc_dif; |
192 |
start_coeff = 1; |
start_coeff = 1; |
193 |
} |
} |
194 |
else |
else |
197 |
} |
} |
198 |
|
|
199 |
start_timer(); |
start_timer(); |
200 |
if (cbp & (1 << (5-k))) // coded |
if (cbp & (1 << (5-i))) // coded |
201 |
{ |
{ |
202 |
get_intra_block(bs, block, mb->acpred_directions[k], start_coeff); |
get_intra_block(bs, block[i], pMB->acpred_directions[i], start_coeff); |
203 |
} |
} |
204 |
stop_coding_timer(); |
stop_coding_timer(); |
205 |
|
|
206 |
start_timer(); |
start_timer(); |
207 |
add_acdc(mb, k, block, dcscalar, predictors); |
add_acdc(pMB, i, block[i], iDcScaler, predictors); |
208 |
stop_prediction_timer(); |
stop_prediction_timer(); |
209 |
|
|
210 |
start_timer(); |
start_timer(); |
211 |
if (dec->quant_type == 0) |
if (dec->quant_type == 0) |
212 |
{ |
{ |
213 |
dequant_intra(data, block, mb->quant, dcscalar); |
dequant_intra(data[i], block[i], iQuant, iDcScaler); |
214 |
} |
} |
215 |
else |
else |
216 |
{ |
{ |
217 |
dequant4_intra(data, block, mb->quant, dcscalar); |
dequant4_intra(data[i], block[i], iQuant, iDcScaler); |
218 |
} |
} |
219 |
stop_iquant_timer(); |
stop_iquant_timer(); |
220 |
|
|
221 |
start_timer(); |
start_timer(); |
222 |
idct(data); |
idct(data[i]); |
223 |
stop_idct_timer(); |
stop_idct_timer(); |
224 |
|
} |
225 |
|
|
226 |
start_timer(); |
start_timer(); |
227 |
if (k < 4) |
if (dec->interlacing && pMB->field_dct) |
|
{ |
|
|
transfer_16to8copy(dec->cur.y + (16*y*dec->edged_width) + 16*x + (4*(k&2)*dec->edged_width) + 8*(k&1), data, dec->edged_width); |
|
|
} |
|
|
else if (k == 4) |
|
228 |
{ |
{ |
229 |
transfer_16to8copy(dec->cur.u+ 8*y*(dec->edged_width/2) + 8*x, data, (dec->edged_width/2)); |
MBFieldToFrame(data); |
|
} |
|
|
else // if (k == 5) |
|
|
{ |
|
|
transfer_16to8copy(dec->cur.v + 8*y*(dec->edged_width/2) + 8*x, data, (dec->edged_width/2)); |
|
230 |
} |
} |
231 |
|
stop_interlacing_timer(); |
232 |
|
|
233 |
|
start_timer(); |
234 |
|
transfer_16to8copy(pY_Cur, data[0], stride); |
235 |
|
transfer_16to8copy(pY_Cur + 8, data[1], stride); |
236 |
|
transfer_16to8copy(pY_Cur + 8 * stride, data[2], stride); |
237 |
|
transfer_16to8copy(pY_Cur + 8 + 8 * stride, data[3], stride); |
238 |
|
transfer_16to8copy(pU_Cur, data[4], stride / 2); |
239 |
|
transfer_16to8copy(pV_Cur, data[5], stride / 2); |
240 |
stop_transfer_timer(); |
stop_transfer_timer(); |
241 |
} |
} |
|
} |
|
242 |
|
|
243 |
|
|
244 |
|
|
252 |
|
|
253 |
// decode an inter macroblock |
// decode an inter macroblock |
254 |
|
|
255 |
void decoder_mbinter(DECODER * dec, MACROBLOCK * mb, int x, int y, uint32_t acpred_flag, uint32_t cbp, Bitstream * bs, int quant, int rounding) |
void decoder_mbinter(DECODER * dec, |
256 |
|
const MACROBLOCK * pMB, |
257 |
|
const uint32_t x_pos, |
258 |
|
const uint32_t y_pos, |
259 |
|
const uint32_t acpred_flag, |
260 |
|
const uint32_t cbp, |
261 |
|
Bitstream * bs, |
262 |
|
const uint32_t quant, |
263 |
|
const uint32_t rounding) |
264 |
{ |
{ |
265 |
|
CACHE_ALIGN int16_t block[6][64]; |
266 |
|
CACHE_ALIGN int16_t data[6][64]; |
267 |
|
|
268 |
const uint32_t stride = dec->edged_width; |
const uint32_t stride = dec->edged_width; |
269 |
const uint32_t stride2 = dec->edged_width / 2; |
const uint32_t stride2 = dec->edged_width / 2; |
270 |
|
uint32_t i; |
271 |
|
uint32_t iQuant = pMB->quant; |
272 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
273 |
int uv_dx, uv_dy; |
int uv_dx, uv_dy; |
|
uint32_t k; |
|
274 |
|
|
275 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) |
pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); |
276 |
|
pU_Cur = dec->cur.u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
277 |
|
pV_Cur = dec->cur.v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
278 |
|
|
279 |
|
if (pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q) |
280 |
{ |
{ |
281 |
uv_dx = mb->mvs[0].x; |
uv_dx = pMB->mvs[0].x; |
282 |
uv_dy = mb->mvs[0].y; |
uv_dy = pMB->mvs[0].y; |
283 |
|
|
284 |
uv_dx = (uv_dx & 3) ? (uv_dx >> 1) | 1 : uv_dx / 2; |
uv_dx = (uv_dx & 3) ? (uv_dx >> 1) | 1 : uv_dx / 2; |
285 |
uv_dy = (uv_dy & 3) ? (uv_dy >> 1) | 1 : uv_dy / 2; |
uv_dy = (uv_dy & 3) ? (uv_dy >> 1) | 1 : uv_dy / 2; |
287 |
else |
else |
288 |
{ |
{ |
289 |
int sum; |
int sum; |
290 |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
291 |
uv_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) ); |
uv_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) ); |
292 |
|
|
293 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
294 |
uv_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) ); |
uv_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) ); |
295 |
} |
} |
296 |
|
|
297 |
start_timer(); |
start_timer(); |
298 |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x, 16*y , mb->mvs[0].x, mb->mvs[0].y, stride, rounding); |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos, 16*y_pos , pMB->mvs[0].x, pMB->mvs[0].y, stride, rounding); |
299 |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x + 8, 16*y , mb->mvs[1].x, mb->mvs[1].y, stride, rounding); |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos + 8, 16*y_pos , pMB->mvs[1].x, pMB->mvs[1].y, stride, rounding); |
300 |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x, 16*y + 8, mb->mvs[2].x, mb->mvs[2].y, stride, rounding); |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos, 16*y_pos + 8, pMB->mvs[2].x, pMB->mvs[2].y, stride, rounding); |
301 |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x + 8, 16*y + 8, mb->mvs[3].x, mb->mvs[3].y, stride, rounding); |
interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos + 8, 16*y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, rounding); |
302 |
interpolate8x8_switch(dec->cur.u, dec->refn.u, 8*x, 8*y, uv_dx, uv_dy, stride2, rounding); |
interpolate8x8_switch(dec->cur.u, dec->refn.u, 8*x_pos, 8*y_pos, uv_dx, uv_dy, stride2, rounding); |
303 |
interpolate8x8_switch(dec->cur.v, dec->refn.v, 8*x, 8*y, uv_dx, uv_dy, stride2, rounding); |
interpolate8x8_switch(dec->cur.v, dec->refn.v, 8*x_pos, 8*y_pos, uv_dx, uv_dy, stride2, rounding); |
304 |
stop_comp_timer(); |
stop_comp_timer(); |
305 |
|
|
306 |
|
for (i = 0; i < 6; i++) |
|
for (k = 0; k < 6; k++) |
|
307 |
{ |
{ |
308 |
int16_t block[64]; |
if (cbp & (1 << (5-i))) // coded |
|
int16_t data[64]; |
|
|
|
|
|
if (cbp & (1 << (5-k))) // coded |
|
309 |
{ |
{ |
310 |
memset(block, 0, 64 * sizeof(int16_t)); // clear |
memset(block[i], 0, 64 * sizeof(int16_t)); // clear |
311 |
|
|
312 |
start_timer(); |
start_timer(); |
313 |
get_inter_block(bs, block); |
get_inter_block(bs, block[i]); |
314 |
stop_coding_timer(); |
stop_coding_timer(); |
315 |
|
|
316 |
start_timer(); |
start_timer(); |
317 |
if (dec->quant_type == 0) |
if (dec->quant_type == 0) |
318 |
{ |
{ |
319 |
dequant_inter(data, block, mb->quant); |
dequant_inter(data[i], block[i], iQuant); |
320 |
} |
} |
321 |
else |
else |
322 |
{ |
{ |
323 |
dequant4_inter(data, block, mb->quant); |
dequant4_inter(data[i], block[i], iQuant); |
324 |
} |
} |
325 |
stop_iquant_timer(); |
stop_iquant_timer(); |
326 |
|
|
327 |
start_timer(); |
start_timer(); |
328 |
idct(data); |
idct(data[i]); |
329 |
stop_idct_timer(); |
stop_idct_timer(); |
|
|
|
|
start_timer(); |
|
|
if (k < 4) |
|
|
{ |
|
|
transfer_16to8add(dec->cur.y + (16*y + 4*(k&2))*stride + 16*x + 8*(k&1), data, stride); |
|
330 |
} |
} |
|
else if (k == 4) |
|
|
{ |
|
|
transfer_16to8add(dec->cur.u + 8*y*stride2 + 8*x, data, stride2); |
|
331 |
} |
} |
332 |
else // k == 5 |
|
333 |
|
start_timer(); |
334 |
|
if (pMB->field_dct) |
335 |
{ |
{ |
336 |
transfer_16to8add(dec->cur.v + 8*y*stride2 + 8*x, data, stride2); |
MBFieldToFrame(data); |
337 |
} |
} |
338 |
|
stop_interlacing_timer(); |
339 |
|
|
340 |
|
start_timer(); |
341 |
|
if (cbp & 32) |
342 |
|
transfer_16to8add(pY_Cur, data[0], stride); |
343 |
|
if (cbp & 16) |
344 |
|
transfer_16to8add(pY_Cur + 8, data[1], stride); |
345 |
|
if (cbp & 8) |
346 |
|
transfer_16to8add(pY_Cur + 8 * stride, data[2], stride); |
347 |
|
if (cbp & 4) |
348 |
|
transfer_16to8add(pY_Cur + 8 + 8 * stride, data[3], stride); |
349 |
|
if (cbp & 2) |
350 |
|
transfer_16to8add(pU_Cur, data[4], stride / 2); |
351 |
|
if (cbp & 1) |
352 |
|
transfer_16to8add(pV_Cur, data[5], stride / 2); |
353 |
stop_transfer_timer(); |
stop_transfer_timer(); |
354 |
} |
} |
|
} |
|
|
} |
|
|
|
|
355 |
|
|
356 |
|
|
357 |
void decoder_iframe(DECODER * dec, Bitstream * bs, int quant, int intra_dc_threshold) |
void decoder_iframe(DECODER * dec, Bitstream * bs, int quant, int intra_dc_threshold) |
399 |
} |
} |
400 |
mb->quant = quant; |
mb->quant = quant; |
401 |
|
|
402 |
|
if (dec->interlacing) |
403 |
|
{ |
404 |
|
mb->field_dct = BitstreamGetBit(bs); |
405 |
|
DEBUG1("deci: field_dct: ", mb->field_dct); |
406 |
|
} |
407 |
|
|
408 |
decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, intra_dc_threshold); |
decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, intra_dc_threshold); |
409 |
} |
} |
467 |
image_swap(&dec->cur, &dec->refn); |
image_swap(&dec->cur, &dec->refn); |
468 |
|
|
469 |
start_timer(); |
start_timer(); |
470 |
image_setedges(&dec->refn, dec->edged_width, dec->edged_height, dec->width, dec->height); |
image_setedges(&dec->refn, dec->edged_width, dec->edged_height, dec->width, dec->height, dec->interlacing); |
471 |
stop_edges_timer(); |
stop_edges_timer(); |
472 |
|
|
473 |
for (y = 0; y < dec->mb_height; y++) |
for (y = 0; y < dec->mb_height; y++) |
520 |
} |
} |
521 |
mb->quant = quant; |
mb->quant = quant; |
522 |
|
|
523 |
|
if (dec->interlacing) |
524 |
|
{ |
525 |
|
mb->field_dct = BitstreamGetBit(bs); |
526 |
|
DEBUG1("decp: field_dct: ", mb->field_dct); |
527 |
|
|
528 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) |
529 |
{ |
{ |
530 |
|
mb->field_pred = BitstreamGetBit(bs); |
531 |
|
DEBUG1("decp: field_pred: ", mb->field_pred); |
532 |
|
|
533 |
|
if (mb->field_pred) |
534 |
|
{ |
535 |
|
mb->field_for_top = BitstreamGetBit(bs); |
536 |
|
DEBUG1("decp: field_for_top: ", mb->field_for_top); |
537 |
|
mb->field_for_bot = BitstreamGetBit(bs); |
538 |
|
DEBUG1("decp: field_for_bot: ", mb->field_for_bot); |
539 |
|
} |
540 |
|
} |
541 |
|
} |
542 |
|
|
543 |
|
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) |
544 |
|
{ |
545 |
|
if (dec->interlacing && mb->field_pred) |
546 |
|
{ |
547 |
|
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); |
548 |
|
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[1], fcode); |
549 |
|
} |
550 |
|
else |
551 |
|
{ |
552 |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); |
553 |
mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = mb->mvs[0].x; |
mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = mb->mvs[0].x; |
554 |
mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = mb->mvs[0].y; |
mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = mb->mvs[0].y; |
555 |
} |
} |
556 |
|
} |
557 |
else if (mb->mode == MODE_INTER4V /* || mb->mode == MODE_INTER4V_Q */) |
else if (mb->mode == MODE_INTER4V /* || mb->mode == MODE_INTER4V_Q */) |
558 |
{ |
{ |
559 |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); |
get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); |