39 |
#include "../bitstream/zigzag.h" |
#include "../bitstream/zigzag.h" |
40 |
#include "../dct/fdct.h" |
#include "../dct/fdct.h" |
41 |
#include "../dct/idct.h" |
#include "../dct/idct.h" |
42 |
#include "../quant/quant_mpeg4.h" |
#include "../quant/quant.h" |
|
#include "../quant/quant_h263.h" |
|
43 |
#include "../encoder.h" |
#include "../encoder.h" |
44 |
|
|
45 |
#include "../image/reduced.h" |
#include "../image/reduced.h" |
122 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64], |
123 |
int16_t data[6*64]) |
int16_t data[6*64]) |
124 |
{ |
{ |
125 |
int i; |
int mpeg; |
126 |
|
int scaler_lum, scaler_chr; |
127 |
|
|
128 |
for (i = 0; i < 6; i++) { |
quant_intraFuncPtr const quant[2] = |
129 |
uint32_t iDcScaler = get_dc_scaler(pMB->quant, i < 4); |
{ |
130 |
|
quant_h263_intra, |
131 |
|
quant_mpeg_intra |
132 |
|
}; |
133 |
|
|
134 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
135 |
|
scaler_lum = get_dc_scaler(pMB->quant, 1); |
136 |
|
scaler_chr = get_dc_scaler(pMB->quant, 0); |
137 |
|
|
138 |
/* Quantize the block */ |
/* Quantize the block */ |
139 |
start_timer(); |
start_timer(); |
140 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) { |
quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum); |
141 |
quant_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); |
quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum); |
142 |
} else { |
quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum); |
143 |
quant4_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); |
quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum); |
144 |
} |
quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr); |
145 |
|
quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr); |
146 |
stop_quant_timer(); |
stop_quant_timer(); |
147 |
} |
} |
|
} |
|
148 |
|
|
149 |
/* DeQuantize all blocks -- Intra mode */ |
/* DeQuantize all blocks -- Intra mode */ |
150 |
static __inline void |
static __inline void |
153 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64], |
154 |
int16_t data[6*64]) |
int16_t data[6*64]) |
155 |
{ |
{ |
156 |
int i; |
int mpeg; |
157 |
|
int scaler_lum, scaler_chr; |
158 |
|
|
159 |
for (i = 0; i < 6; i++) { |
quant_intraFuncPtr const dequant[2] = |
160 |
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
{ |
161 |
|
dequant_h263_intra, |
162 |
|
dequant_mpeg_intra |
163 |
|
}; |
164 |
|
|
165 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
166 |
|
scaler_lum = get_dc_scaler(iQuant, 1); |
167 |
|
scaler_chr = get_dc_scaler(iQuant, 0); |
168 |
|
|
169 |
start_timer(); |
start_timer(); |
170 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
dequant[mpeg](&qcoeff[0 * 64], &data[0 * 64], iQuant, scaler_lum); |
171 |
dequant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum); |
172 |
else |
dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum); |
173 |
dequant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum); |
174 |
|
dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr); |
175 |
|
dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr); |
176 |
stop_iquant_timer(); |
stop_iquant_timer(); |
177 |
} |
} |
|
} |
|
178 |
|
|
179 |
|
|
180 |
|
typedef int (*trellis_func_ptr_t)(int16_t *const Out, |
181 |
|
const int16_t *const In, |
182 |
|
int Q, |
183 |
|
const uint16_t * const Zigzag, |
184 |
|
int Non_Zero); |
185 |
|
|
186 |
static int |
static int |
187 |
dct_quantize_trellis_h263_c(int16_t *const Out, |
dct_quantize_trellis_h263_c(int16_t *const Out, |
188 |
const int16_t *const In, |
const int16_t *const In, |
190 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
191 |
int Non_Zero); |
int Non_Zero); |
192 |
|
|
|
#if 0 |
|
193 |
static int |
static int |
194 |
dct_quantize_trellis_mpeg_c(int16_t *const Out, |
dct_quantize_trellis_mpeg_c(int16_t *const Out, |
195 |
const int16_t *const In, |
const int16_t *const In, |
196 |
int Q, |
int Q, |
197 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
198 |
int Non_Zero); |
int Non_Zero); |
|
#endif |
|
199 |
|
|
200 |
/* Quantize all blocks -- Inter mode */ |
/* Quantize all blocks -- Inter mode */ |
201 |
static __inline uint8_t |
static __inline uint8_t |
211 |
int i; |
int i; |
212 |
uint8_t cbp = 0; |
uint8_t cbp = 0; |
213 |
int sum; |
int sum; |
214 |
int code_block; |
int code_block, mpeg; |
215 |
|
|
216 |
|
quant_interFuncPtr const quant[2] = |
217 |
|
{ |
218 |
|
quant_h263_inter, |
219 |
|
quant_mpeg_inter |
220 |
|
}; |
221 |
|
|
222 |
|
trellis_func_ptr_t const trellis[2] = |
223 |
|
{ |
224 |
|
dct_quantize_trellis_h263_c, |
225 |
|
dct_quantize_trellis_mpeg_c |
226 |
|
}; |
227 |
|
|
228 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
229 |
|
|
230 |
for (i = 0; i < 6; i++) { |
for (i = 0; i < 6; i++) { |
231 |
|
|
232 |
/* Quantize the block */ |
/* Quantize the block */ |
233 |
start_timer(); |
start_timer(); |
234 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) { |
|
235 |
sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant); |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant); |
236 |
if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) { |
|
237 |
sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1; |
if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
238 |
/* limit = 1; // Isibaar: why? deactivated so far - so please complain! ;-) */ |
sum = trellis[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63); |
|
} |
|
|
} else { |
|
|
sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant); |
|
|
#if 0 |
|
|
if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) |
|
|
sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1; |
|
|
#endif |
|
239 |
} |
} |
240 |
stop_quant_timer(); |
stop_quant_timer(); |
241 |
|
|
274 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64], |
275 |
const uint8_t cbp) |
const uint8_t cbp) |
276 |
{ |
{ |
277 |
int i; |
int mpeg; |
278 |
|
|
279 |
|
quant_interFuncPtr const dequant[2] = |
280 |
|
{ |
281 |
|
dequant_h263_inter, |
282 |
|
dequant_mpeg_inter |
283 |
|
}; |
284 |
|
|
285 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
286 |
|
|
|
for (i = 0; i < 6; i++) { |
|
|
if (cbp & (1 << (5 - i))) { |
|
287 |
start_timer(); |
start_timer(); |
288 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant); |
289 |
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant); |
290 |
else |
if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant); |
291 |
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant); |
292 |
|
if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant); |
293 |
|
if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant); |
294 |
stop_iquant_timer(); |
stop_iquant_timer(); |
295 |
} |
} |
|
} |
|
|
} |
|
296 |
|
|
297 |
typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS); |
typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS); |
298 |
typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS); |
typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS); |
310 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
311 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
312 |
int32_t cst; |
int32_t cst; |
313 |
|
int vop_reduced; |
314 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
315 |
const IMAGE * const pCurrent = &frame->image; |
const IMAGE * const pCurrent = &frame->image; |
316 |
|
transfer_operation_8to16_t * const functions[2] = |
317 |
|
{ |
318 |
|
(transfer_operation_8to16_t *)transfer_8to16copy, |
319 |
|
(transfer_operation_8to16_t *)filter_18x18_to_8x8 |
320 |
|
}; |
321 |
transfer_operation_8to16_t *transfer_op = NULL; |
transfer_operation_8to16_t *transfer_op = NULL; |
322 |
|
|
323 |
if ((frame->vop_flags & XVID_VOP_REDUCED)) { |
vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED); |
324 |
|
|
325 |
/* Image pointers */ |
/* Image pointers */ |
326 |
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced)); |
327 |
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
328 |
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
329 |
|
|
330 |
/* Block size */ |
/* Block size */ |
331 |
cst = 16; |
cst = 8<<vop_reduced; |
332 |
|
|
333 |
/* Operation function */ |
/* Operation function */ |
334 |
transfer_op = (transfer_operation_8to16_t*)filter_18x18_to_8x8; |
transfer_op = functions[vop_reduced]; |
|
} else { |
|
|
|
|
|
/* Image pointers */ |
|
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
|
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
|
|
|
/* Block size */ |
|
|
cst = 8; |
|
|
|
|
|
/* Operation function */ |
|
|
transfer_op = (transfer_operation_8to16_t*)transfer_8to16copy; |
|
|
} |
|
335 |
|
|
336 |
/* Do the transfer */ |
/* Do the transfer */ |
337 |
start_timer(); |
start_timer(); |
351 |
const uint32_t x_pos, |
const uint32_t x_pos, |
352 |
const uint32_t y_pos, |
const uint32_t y_pos, |
353 |
int16_t data[6 * 64], |
int16_t data[6 * 64], |
354 |
const uint32_t add, |
const uint32_t add, /* Must be 1 or 0 */ |
355 |
const uint8_t cbp) |
const uint8_t cbp) |
356 |
{ |
{ |
357 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
359 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
360 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
361 |
uint32_t cst; |
uint32_t cst; |
362 |
|
int vop_reduced; |
363 |
const IMAGE * const pCurrent = &frame->image; |
const IMAGE * const pCurrent = &frame->image; |
364 |
|
|
365 |
|
/* Array of function pointers, indexed by [vop_reduced<<1+add] */ |
366 |
|
transfer_operation_16to8_t * const functions[4] = |
367 |
|
{ |
368 |
|
(transfer_operation_16to8_t*)transfer_16to8copy, |
369 |
|
(transfer_operation_16to8_t*)transfer_16to8add, |
370 |
|
(transfer_operation_16to8_t*)copy_upsampled_8x8_16to8, |
371 |
|
(transfer_operation_16to8_t*)add_upsampled_8x8_16to8 |
372 |
|
}; |
373 |
|
|
374 |
transfer_operation_16to8_t *transfer_op = NULL; |
transfer_operation_16to8_t *transfer_op = NULL; |
375 |
|
|
376 |
if (pMB->field_dct) { |
if (pMB->field_dct) { |
378 |
stride *= 2; |
stride *= 2; |
379 |
} |
} |
380 |
|
|
381 |
if ((frame->vop_flags & XVID_VOP_REDUCED)) { |
/* Makes this vars booleans */ |
382 |
|
vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED); |
383 |
|
|
384 |
/* Image pointers */ |
/* Image pointers */ |
385 |
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced)); |
386 |
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
387 |
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
388 |
|
|
389 |
/* Block size */ |
/* Block size */ |
390 |
cst = 16; |
cst = 8<<vop_reduced; |
391 |
|
|
392 |
/* Operation function */ |
/* Operation function */ |
393 |
if(add) |
transfer_op = functions[(vop_reduced<<1) + add]; |
|
transfer_op = (transfer_operation_16to8_t*)add_upsampled_8x8_16to8; |
|
|
else |
|
|
transfer_op = (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8; |
|
|
} else { |
|
|
|
|
|
/* Image pointers */ |
|
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
|
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
|
|
|
/* Block size */ |
|
|
cst = 8; |
|
|
|
|
|
/* Operation function */ |
|
|
if(add) |
|
|
transfer_op = (transfer_operation_16to8_t*)transfer_16to8add; |
|
|
else |
|
|
transfer_op = (transfer_operation_16to8_t*)transfer_16to8copy; |
|
|
} |
|
394 |
|
|
395 |
/* Do the operation */ |
/* Do the operation */ |
396 |
start_timer(); |
start_timer(); |
449 |
uint8_t cbp; |
uint8_t cbp; |
450 |
uint32_t limit; |
uint32_t limit; |
451 |
|
|
452 |
/* |
/* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
453 |
* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
* already */ |
|
* already |
|
|
*/ |
|
454 |
|
|
455 |
/* Perform DCT (and field decision) */ |
/* Perform DCT (and field decision) */ |
456 |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
488 |
uint8_t cbp; |
uint8_t cbp; |
489 |
uint32_t limit; |
uint32_t limit; |
490 |
|
|
491 |
/* |
/* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
492 |
* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
* already */ |
|
* already |
|
|
*/ |
|
493 |
|
|
494 |
/* Perform DCT (and field decision) */ |
/* Perform DCT (and field decision) */ |
495 |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
507 |
* History comment: |
* History comment: |
508 |
* We don't have to DeQuant, iDCT and Transfer back data for B-frames. |
* We don't have to DeQuant, iDCT and Transfer back data for B-frames. |
509 |
* |
* |
510 |
* BUT some plugins require the original frame to be passed so we have |
* BUT some plugins require the rebuilt original frame to be passed so we |
511 |
* to take care of that here |
* have to take care of that here |
512 |
*/ |
*/ |
513 |
if((pParam->plugin_flags & XVID_REQORIGINAL)) { |
if((pParam->plugin_flags & XVID_REQORIGINAL)) { |
514 |
|
|
633 |
MOVLINE(LINE(3, 3), tmp); |
MOVLINE(LINE(3, 3), tmp); |
634 |
} |
} |
635 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
636 |
/***************************************************************************** |
/***************************************************************************** |
637 |
* Trellis based R-D optimal quantization |
* Trellis based R-D optimal quantization |
638 |
* |
* |
640 |
* |
* |
641 |
****************************************************************************/ |
****************************************************************************/ |
642 |
|
|
|
|
|
|
#if 0 |
|
|
static int |
|
|
dct_quantize_trellis_mpeg_c(int16_t *const Out, |
|
|
const int16_t *const In, |
|
|
int Q, |
|
|
const uint16_t * const Zigzag, |
|
|
int Non_Zero) |
|
|
{ |
|
|
return 63; |
|
|
} |
|
|
#endif |
|
|
|
|
643 |
/*---------------------------------------------------------------------------- |
/*---------------------------------------------------------------------------- |
644 |
* |
* |
645 |
* Trellis-Based quantization |
* Trellis-Based quantization |
765 |
}; |
}; |
766 |
#undef TL |
#undef TL |
767 |
|
|
768 |
static __inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i) |
static int __inline |
769 |
|
Find_Last(const int16_t *C, const uint16_t *Zigzag, int i) |
770 |
{ |
{ |
771 |
while(i>=0) |
while(i>=0) |
772 |
if (C[Zigzag[i]]) |
if (C[Zigzag[i]]) |
775 |
return -1; |
return -1; |
776 |
} |
} |
777 |
|
|
778 |
|
static int __inline |
779 |
|
Compute_Sum(const int16_t *C, int last) |
780 |
|
{ |
781 |
|
int sum = 0; |
782 |
|
|
783 |
|
while(last--) |
784 |
|
sum += abs(C[last]); |
785 |
|
|
786 |
|
return(sum); |
787 |
|
} |
788 |
/* this routine has been strippen of all debug code */ |
/* this routine has been strippen of all debug code */ |
789 |
|
|
790 |
static int |
static int |
813 |
int Last_Node = -1; |
int Last_Node = -1; |
814 |
uint32_t Last_Cost = 0; |
uint32_t Last_Cost = 0; |
815 |
|
|
816 |
int i, j; |
int i, j, sum; |
817 |
Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */ |
Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */ |
818 |
|
|
819 |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
820 |
if (Non_Zero<0) |
if (Non_Zero<0) |
821 |
return -1; |
return 0; /* Sum is zero if there are only zero coeffs */ |
822 |
|
|
823 |
for(i=0; i<=Non_Zero; i++) |
for(i=0; i<=Non_Zero; i++) { |
|
{ |
|
824 |
const int AC = In[Zigzag[i]]; |
const int AC = In[Zigzag[i]]; |
825 |
const int Level1 = Out[Zigzag[i]]; |
const int Level1 = Out[Zigzag[i]]; |
826 |
const int Dist0 = Lambda* AC*AC; |
const int Dist0 = Lambda* AC*AC; |
827 |
uint32_t Best_Cost = 0xf0000000; |
uint32_t Best_Cost = 0xf0000000; |
828 |
Last_Cost += Dist0; |
Last_Cost += Dist0; |
829 |
|
|
830 |
if ((uint32_t)(Level1+1)<3) /* very specialized loop for -1,0,+1 */ |
/* very specialized loop for -1,0,+1 */ |
831 |
{ |
if ((uint32_t)(Level1+1)<3) { |
832 |
int dQ; |
int dQ; |
833 |
int Run; |
int Run; |
834 |
uint32_t Cost0; |
uint32_t Cost0; |
844 |
|
|
845 |
Nodes[i].Run = 1; |
Nodes[i].Run = 1; |
846 |
Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0; |
Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0; |
847 |
for(Run=i-Run_Start; Run>0; --Run) |
for(Run=i-Run_Start; Run>0; --Run) { |
|
{ |
|
848 |
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
849 |
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16); |
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16); |
850 |
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16); |
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16); |
870 |
} |
} |
871 |
if (Last_Node==i) |
if (Last_Node==i) |
872 |
Last.Level = Nodes[i].Level; |
Last.Level = Nodes[i].Level; |
873 |
} |
} else { /* "big" levels */ |
|
else /* "big" levels */ |
|
|
{ |
|
874 |
const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last; |
const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last; |
875 |
int Level2; |
int Level2; |
876 |
int dQ1, dQ2; |
int dQ1, dQ2; |
895 |
Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0; |
Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0; |
896 |
Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0; |
Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0; |
897 |
} |
} |
898 |
|
|
899 |
Dist1 = Lambda*dQ1*dQ1; |
Dist1 = Lambda*dQ1*dQ1; |
900 |
Dist2 = Lambda*dQ2*dQ2; |
Dist2 = Lambda*dQ2*dQ2; |
901 |
dDist21 = Dist2-Dist1; |
dDist21 = Dist2-Dist1; |
918 |
if (Cost2<Cost1) { |
if (Cost2<Cost1) { |
919 |
Cost1 = Cost2; |
Cost1 = Cost2; |
920 |
bLevel = Level2; |
bLevel = Level2; |
921 |
} else |
} else { |
922 |
bLevel = Level1; |
bLevel = Level1; |
923 |
|
} |
924 |
|
|
925 |
if (Cost1<Best_Cost) { |
if (Cost1<Best_Cost) { |
926 |
Best_Cost = Cost1; |
Best_Cost = Cost1; |
934 |
if (Cost2<Cost1) { |
if (Cost2<Cost1) { |
935 |
Cost1 = Cost2; |
Cost1 = Cost2; |
936 |
bLevel = Level2; |
bLevel = Level2; |
937 |
} else |
} else { |
938 |
bLevel = Level1; |
bLevel = Level1; |
939 |
|
} |
940 |
|
|
941 |
if (Cost1<Last_Cost) { |
if (Cost1<Last_Cost) { |
942 |
Last_Cost = Cost1; |
Last_Cost = Cost1; |
953 |
if (Best_Cost < Min_Cost + Dist0) { |
if (Best_Cost < Min_Cost + Dist0) { |
954 |
Min_Cost = Best_Cost; |
Min_Cost = Best_Cost; |
955 |
Run_Start = i; |
Run_Start = i; |
956 |
} |
} else { |
|
else |
|
|
{ |
|
957 |
/* |
/* |
958 |
* as noticed by Michael Niedermayer (michaelni at gmx.at), there's |
* as noticed by Michael Niedermayer (michaelni at gmx.at), there's |
959 |
* a code shorter by 1 bit for a larger run (!), same level. We give |
* a code shorter by 1 bit for a larger run (!), same level. We give |
969 |
} |
} |
970 |
} |
} |
971 |
|
|
972 |
|
/* It seems trellis doesn't give good results... just compute the Out sum and |
973 |
|
* quit (even if we did not modify it, upperlayer relies on this data) */ |
974 |
if (Last_Node<0) |
if (Last_Node<0) |
975 |
return -1; |
return Compute_Sum(Out, Non_Zero); |
976 |
|
|
977 |
/* reconstruct optimal sequence backward with surviving paths */ |
/* reconstruct optimal sequence backward with surviving paths */ |
978 |
memset(Out, 0x00, 64*sizeof(*Out)); |
memset(Out, 0x00, 64*sizeof(*Out)); |
979 |
Out[Zigzag[Last_Node]] = Last.Level; |
Out[Zigzag[Last_Node]] = Last.Level; |
980 |
i = Last_Node - Last.Run; |
i = Last_Node - Last.Run; |
981 |
|
sum = 0; |
982 |
while(i>=0) { |
while(i>=0) { |
983 |
Out[Zigzag[i]] = Nodes[i].Level; |
Out[Zigzag[i]] = Nodes[i].Level; |
984 |
|
sum += abs(Nodes[i].Level); |
985 |
i -= Nodes[i].Run; |
i -= Nodes[i].Run; |
986 |
} |
} |
|
return Last_Node; |
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
987 |
|
|
988 |
|
return sum; |
989 |
|
} |
990 |
|
|
991 |
|
static int |
992 |
|
dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero) |
993 |
|
{ |
994 |
|
/* ToDo: Ok ok it's just a place holder for Gruel -- damn write this one :-) */ |
995 |
|
return Compute_Sum(Out, 63); |
996 |
|
} |
997 |
|
|
998 |
/* original version including heavy debugging info */ |
/* original version including heavy debugging info */ |
999 |
|
|