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 "../quant/quant_matrix.h" |
46 |
|
|
47 |
MBFIELDTEST_PTR MBFieldTest; |
MBFIELDTEST_PTR MBFieldTest; |
48 |
|
|
90 |
|
|
91 |
/* Perform DCT */ |
/* Perform DCT */ |
92 |
start_timer(); |
start_timer(); |
93 |
fdct(&data[0 * 64]); |
fdct((short * const)&data[0 * 64]); |
94 |
fdct(&data[1 * 64]); |
fdct((short * const)&data[1 * 64]); |
95 |
fdct(&data[2 * 64]); |
fdct((short * const)&data[2 * 64]); |
96 |
fdct(&data[3 * 64]); |
fdct((short * const)&data[3 * 64]); |
97 |
fdct(&data[4 * 64]); |
fdct((short * const)&data[4 * 64]); |
98 |
fdct(&data[5 * 64]); |
fdct((short * const)&data[5 * 64]); |
99 |
stop_dct_timer(); |
stop_dct_timer(); |
100 |
} |
} |
101 |
|
|
105 |
const uint8_t cbp) |
const uint8_t cbp) |
106 |
{ |
{ |
107 |
start_timer(); |
start_timer(); |
108 |
if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]); |
if(cbp & (1 << (5 - 0))) idct((short * const)&data[0 * 64]); |
109 |
if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]); |
if(cbp & (1 << (5 - 1))) idct((short * const)&data[1 * 64]); |
110 |
if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]); |
if(cbp & (1 << (5 - 2))) idct((short * const)&data[2 * 64]); |
111 |
if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]); |
if(cbp & (1 << (5 - 3))) idct((short * const)&data[3 * 64]); |
112 |
if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]); |
if(cbp & (1 << (5 - 4))) idct((short * const)&data[4 * 64]); |
113 |
if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]); |
if(cbp & (1 << (5 - 5))) idct((short * const)&data[5 * 64]); |
114 |
stop_idct_timer(); |
stop_idct_timer(); |
115 |
} |
} |
116 |
|
|
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, pParam->mpeg_quant_matrices); |
141 |
quant_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); |
quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
142 |
} else { |
quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
143 |
quant4_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); |
quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
144 |
} |
quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
145 |
|
quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
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, pParam->mpeg_quant_matrices); |
171 |
dequant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
172 |
else |
dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
173 |
dequant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
174 |
|
dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr, pParam->mpeg_quant_matrices); |
175 |
|
dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr, pParam->mpeg_quant_matrices); |
176 |
stop_iquant_timer(); |
stop_iquant_timer(); |
177 |
} |
} |
|
} |
|
|
|
|
178 |
|
|
179 |
static int |
static int |
180 |
dct_quantize_trellis_h263_c(int16_t *const Out, |
dct_quantize_trellis_c(int16_t *const Out, |
181 |
const int16_t *const In, |
const int16_t *const In, |
182 |
int Q, |
int Q, |
183 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
184 |
int Non_Zero); |
const uint16_t * const QuantMatrix, |
185 |
|
int Non_Zero, |
186 |
#if 0 |
int Sum, |
187 |
static int |
int Lambda_Mod); |
|
dct_quantize_trellis_mpeg_c(int16_t *const Out, |
|
|
const int16_t *const In, |
|
|
int Q, |
|
|
const uint16_t * const Zigzag, |
|
|
int Non_Zero); |
|
|
#endif |
|
188 |
|
|
189 |
/* Quantize all blocks -- Inter mode */ |
/* Quantize all blocks -- Inter mode */ |
190 |
static __inline uint8_t |
static __inline uint8_t |
200 |
int i; |
int i; |
201 |
uint8_t cbp = 0; |
uint8_t cbp = 0; |
202 |
int sum; |
int sum; |
203 |
int code_block; |
int code_block, mpeg; |
204 |
|
|
205 |
|
quant_interFuncPtr const quant[2] = |
206 |
|
{ |
207 |
|
quant_h263_inter, |
208 |
|
quant_mpeg_inter |
209 |
|
}; |
210 |
|
|
211 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
212 |
|
|
213 |
for (i = 0; i < 6; i++) { |
for (i = 0; i < 6; i++) { |
214 |
|
|
215 |
/* Quantize the block */ |
/* Quantize the block */ |
216 |
start_timer(); |
start_timer(); |
217 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) { |
|
218 |
sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant); |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices); |
219 |
if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) { |
|
220 |
sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1; |
if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
221 |
limit = 1; |
const uint16_t *matrix; |
222 |
} |
const static uint16_t h263matrix[] = |
223 |
} else { |
{ |
224 |
sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant); |
16, 16, 16, 16, 16, 16, 16, 16, |
225 |
#if 0 |
16, 16, 16, 16, 16, 16, 16, 16, |
226 |
if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) |
16, 16, 16, 16, 16, 16, 16, 16, |
227 |
sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1; |
16, 16, 16, 16, 16, 16, 16, 16, |
228 |
#endif |
16, 16, 16, 16, 16, 16, 16, 16, |
229 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
230 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
231 |
|
16, 16, 16, 16, 16, 16, 16, 16 |
232 |
|
}; |
233 |
|
|
234 |
|
matrix = (mpeg)?get_inter_matrix(pParam->mpeg_quant_matrices):h263matrix; |
235 |
|
sum = dct_quantize_trellis_c(&qcoeff[i*64], &data[i*64], |
236 |
|
pMB->quant, &scan_tables[0][0], |
237 |
|
matrix, |
238 |
|
63, |
239 |
|
sum, |
240 |
|
pMB->lambda[i]); |
241 |
} |
} |
242 |
stop_quant_timer(); |
stop_quant_timer(); |
243 |
|
|
276 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64], |
277 |
const uint8_t cbp) |
const uint8_t cbp) |
278 |
{ |
{ |
279 |
int i; |
int mpeg; |
280 |
|
|
281 |
|
quant_interFuncPtr const dequant[2] = |
282 |
|
{ |
283 |
|
dequant_h263_inter, |
284 |
|
dequant_mpeg_inter |
285 |
|
}; |
286 |
|
|
287 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
288 |
|
|
|
for (i = 0; i < 6; i++) { |
|
|
if (cbp & (1 << (5 - i))) { |
|
289 |
start_timer(); |
start_timer(); |
290 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant, pParam->mpeg_quant_matrices); |
291 |
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant, pParam->mpeg_quant_matrices); |
292 |
else |
if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant, pParam->mpeg_quant_matrices); |
293 |
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant, pParam->mpeg_quant_matrices); |
294 |
|
if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant, pParam->mpeg_quant_matrices); |
295 |
|
if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant, pParam->mpeg_quant_matrices); |
296 |
stop_iquant_timer(); |
stop_iquant_timer(); |
297 |
} |
} |
|
} |
|
|
} |
|
298 |
|
|
299 |
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); |
300 |
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); |
311 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
312 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
313 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
|
int32_t cst; |
|
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; |
|
transfer_operation_8to16_t *transfer_op = NULL; |
|
|
|
|
|
if ((frame->vop_flags & XVID_VOP_REDUCED)) { |
|
|
|
|
|
/* Image pointers */ |
|
|
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
|
|
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
|
|
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
|
|
|
|
|
/* Block size */ |
|
|
cst = 16; |
|
|
|
|
|
/* Operation function */ |
|
|
transfer_op = (transfer_operation_8to16_t*)filter_18x18_to_8x8; |
|
|
} else { |
|
316 |
|
|
317 |
/* Image pointers */ |
/* Image pointers */ |
318 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
319 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
320 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
321 |
|
|
|
/* Block size */ |
|
|
cst = 8; |
|
|
|
|
|
/* Operation function */ |
|
|
transfer_op = (transfer_operation_8to16_t*)transfer_8to16copy; |
|
|
} |
|
|
|
|
322 |
/* Do the transfer */ |
/* Do the transfer */ |
323 |
start_timer(); |
start_timer(); |
324 |
transfer_op(&data[0 * 64], pY_Cur, stride); |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
325 |
transfer_op(&data[1 * 64], pY_Cur + cst, stride); |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
326 |
transfer_op(&data[2 * 64], pY_Cur + next_block, stride); |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
327 |
transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride); |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
328 |
transfer_op(&data[4 * 64], pU_Cur, stride2); |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
329 |
transfer_op(&data[5 * 64], pV_Cur, stride2); |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
330 |
stop_transfer_timer(); |
stop_transfer_timer(); |
331 |
} |
} |
332 |
|
|
337 |
const uint32_t x_pos, |
const uint32_t x_pos, |
338 |
const uint32_t y_pos, |
const uint32_t y_pos, |
339 |
int16_t data[6 * 64], |
int16_t data[6 * 64], |
340 |
const uint32_t add, |
const uint32_t add, /* Must be 1 or 0 */ |
341 |
const uint8_t cbp) |
const uint8_t cbp) |
342 |
{ |
{ |
343 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
344 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
345 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
346 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
|
uint32_t cst; |
|
347 |
const IMAGE * const pCurrent = &frame->image; |
const IMAGE * const pCurrent = &frame->image; |
|
transfer_operation_16to8_t *transfer_op = NULL; |
|
|
|
|
|
if (pMB->field_dct) { |
|
|
next_block = stride; |
|
|
stride *= 2; |
|
|
} |
|
|
|
|
|
if ((frame->vop_flags & XVID_VOP_REDUCED)) { |
|
348 |
|
|
349 |
/* Image pointers */ |
/* Array of function pointers, indexed by [add] */ |
350 |
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
transfer_operation_16to8_t * const functions[2] = |
351 |
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
{ |
352 |
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
(transfer_operation_16to8_t*)transfer_16to8copy, |
353 |
|
(transfer_operation_16to8_t*)transfer_16to8add, |
354 |
/* Block size */ |
}; |
|
cst = 16; |
|
355 |
|
|
356 |
/* Operation function */ |
transfer_operation_16to8_t *transfer_op = NULL; |
|
if(add) |
|
|
transfer_op = (transfer_operation_16to8_t*)add_upsampled_8x8_16to8; |
|
|
else |
|
|
transfer_op = (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8; |
|
|
} else { |
|
357 |
|
|
358 |
/* Image pointers */ |
/* Image pointers */ |
359 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
360 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
361 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
362 |
|
|
363 |
/* Block size */ |
if (pMB->field_dct) { |
364 |
cst = 8; |
next_block = stride; |
365 |
|
stride *= 2; |
366 |
|
} |
367 |
|
|
368 |
/* Operation function */ |
/* Operation function */ |
369 |
if(add) |
transfer_op = functions[add]; |
|
transfer_op = (transfer_operation_16to8_t*)transfer_16to8add; |
|
|
else |
|
|
transfer_op = (transfer_operation_16to8_t*)transfer_16to8copy; |
|
|
} |
|
370 |
|
|
371 |
/* Do the operation */ |
/* Do the operation */ |
372 |
start_timer(); |
start_timer(); |
373 |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
374 |
if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride); |
if (cbp&16) transfer_op(pY_Cur + 8, &data[1 * 64], stride); |
375 |
if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); |
if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); |
376 |
if (cbp& 4) transfer_op(pY_Cur + next_block + cst, &data[3 * 64], stride); |
if (cbp& 4) transfer_op(pY_Cur + next_block + 8, &data[3 * 64], stride); |
377 |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
378 |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
379 |
stop_transfer_timer(); |
stop_transfer_timer(); |
425 |
uint8_t cbp; |
uint8_t cbp; |
426 |
uint32_t limit; |
uint32_t limit; |
427 |
|
|
428 |
/* |
/* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
429 |
* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
* already */ |
|
* already |
|
|
*/ |
|
430 |
|
|
431 |
/* Perform DCT (and field decision) */ |
/* Perform DCT (and field decision) */ |
432 |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
434 |
/* Set the limit threshold */ |
/* Set the limit threshold */ |
435 |
limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0); |
limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0); |
436 |
|
|
437 |
|
if (frame->vop_flags & XVID_VOP_CARTOON) |
438 |
|
limit *= 3; |
439 |
|
|
440 |
/* Quantize the block */ |
/* Quantize the block */ |
441 |
cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit); |
cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit); |
442 |
|
|
464 |
uint8_t cbp; |
uint8_t cbp; |
465 |
uint32_t limit; |
uint32_t limit; |
466 |
|
|
467 |
/* |
/* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
468 |
* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
* already */ |
|
* already |
|
|
*/ |
|
469 |
|
|
470 |
/* Perform DCT (and field decision) */ |
/* Perform DCT (and field decision) */ |
471 |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
473 |
/* Set the limit threshold */ |
/* Set the limit threshold */ |
474 |
limit = BVOP_TOOSMALL_LIMIT; |
limit = BVOP_TOOSMALL_LIMIT; |
475 |
|
|
476 |
|
if (frame->vop_flags & XVID_VOP_CARTOON) |
477 |
|
limit *= 2; |
478 |
|
|
479 |
/* Quantize the block */ |
/* Quantize the block */ |
480 |
cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit); |
cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit); |
481 |
|
|
483 |
* History comment: |
* History comment: |
484 |
* 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. |
485 |
* |
* |
486 |
* 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 |
487 |
* to take care of that here |
* have to take care of that here |
488 |
*/ |
*/ |
489 |
if((pParam->plugin_flags & XVID_REQORIGINAL)) { |
if((pParam->plugin_flags & XVID_REQORIGINAL)) { |
490 |
|
|
609 |
MOVLINE(LINE(3, 3), tmp); |
MOVLINE(LINE(3, 3), tmp); |
610 |
} |
} |
611 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
612 |
/***************************************************************************** |
/***************************************************************************** |
613 |
* Trellis based R-D optimal quantization |
* Trellis based R-D optimal quantization |
614 |
* |
* |
616 |
* |
* |
617 |
****************************************************************************/ |
****************************************************************************/ |
618 |
|
|
|
|
|
|
#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 |
|
|
|
|
619 |
/*---------------------------------------------------------------------------- |
/*---------------------------------------------------------------------------- |
620 |
* |
* |
621 |
* Trellis-Based quantization |
* Trellis-Based quantization |
627 |
* IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000. |
* IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000. |
628 |
* |
* |
629 |
* we are at stake with a simplified Bellmand-Ford / Dijkstra Single |
* we are at stake with a simplified Bellmand-Ford / Dijkstra Single |
630 |
* Source Shorted Path algo. But due to the underlying graph structure |
* Source Shortest Path algo. But due to the underlying graph structure |
631 |
* ("Trellis"), it can be turned into a dynamic programming algo, |
* ("Trellis"), it can be turned into a dynamic programming algo, |
632 |
* partially saving the explicit graph's nodes representation. And |
* partially saving the explicit graph's nodes representation. And |
633 |
* without using a heap, since the open frontier of the DAG is always |
* without using a heap, since the open frontier of the DAG is always |
634 |
* known, and of fixed sized. |
* known, and of fixed size. |
635 |
*--------------------------------------------------------------------------*/ |
*--------------------------------------------------------------------------*/ |
636 |
|
|
637 |
|
|
731 |
Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1, |
Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1, |
732 |
}; |
}; |
733 |
|
|
734 |
#define TL(q) 0xfe00/(q*q) |
/* TL_SHIFT controls the precision of the RD optimizations in trellis |
735 |
|
* valid range is [10..16]. The bigger, the more trellis is vulnerable |
736 |
|
* to overflows in cost formulas. |
737 |
|
* - 10 allows ac values up to 2^11 == 2048 |
738 |
|
* - 16 allows ac values up to 2^8 == 256 |
739 |
|
*/ |
740 |
|
#define TL_SHIFT 11 |
741 |
|
#define TL(q) ((0xfe00>>(16-TL_SHIFT))/(q*q)) |
742 |
|
|
743 |
static const int Trellis_Lambda_Tabs[31] = { |
static const int Trellis_Lambda_Tabs[31] = { |
744 |
TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7), |
TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7), |
748 |
}; |
}; |
749 |
#undef TL |
#undef TL |
750 |
|
|
751 |
static __inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i) |
static int __inline |
752 |
|
Find_Last(const int16_t *C, const uint16_t *Zigzag, int i) |
753 |
{ |
{ |
754 |
while(i>=0) |
while(i>=0) |
755 |
if (C[Zigzag[i]]) |
if (C[Zigzag[i]]) |
759 |
} |
} |
760 |
|
|
761 |
/* this routine has been strippen of all debug code */ |
/* this routine has been strippen of all debug code */ |
|
|
|
762 |
static int |
static int |
763 |
dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero) |
dct_quantize_trellis_c(int16_t *const Out, |
764 |
|
const int16_t *const In, |
765 |
|
int Q, |
766 |
|
const uint16_t * const Zigzag, |
767 |
|
const uint16_t * const QuantMatrix, |
768 |
|
int Non_Zero, |
769 |
|
int Sum, |
770 |
|
int Lambda_Mod) |
771 |
{ |
{ |
772 |
|
|
773 |
/* |
/* Note: We should search last non-zero coeffs on *real* DCT input coeffs |
774 |
* Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]), |
* (In[]), not quantized one (Out[]). However, it only improves the result |
775 |
* not quantized one (Out[]). However, it only improves the result *very* |
* *very* slightly (~0.01dB), whereas speed drops to crawling level :) |
776 |
* slightly (~0.01dB), whereas speed drops to crawling level :) |
* Well, actually, taking 1 more coeff past Non_Zero into account sometimes |
777 |
* Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps. |
* helps. */ |
|
*/ |
|
778 |
typedef struct { int16_t Run, Level; } NODE; |
typedef struct { int16_t Run, Level; } NODE; |
779 |
|
|
780 |
NODE Nodes[65], Last; |
NODE Nodes[65], Last = { 0, 0}; |
781 |
uint32_t Run_Costs0[64+1]; |
uint32_t Run_Costs0[64+1]; |
782 |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
783 |
const int Mult = 2*Q; |
|
784 |
const int Bias = (Q-1) | 1; |
/* it's 1/lambda, actually */ |
785 |
const int Lev0 = Mult + Bias; |
const int Lambda = (Lambda_Mod*Trellis_Lambda_Tabs[Q-1])>>LAMBDA_EXP; |
|
const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */ |
|
786 |
|
|
787 |
int Run_Start = -1; |
int Run_Start = -1; |
788 |
uint32_t Min_Cost = 2<<16; |
uint32_t Min_Cost = 2<<TL_SHIFT; |
789 |
|
|
790 |
int Last_Node = -1; |
int Last_Node = -1; |
791 |
uint32_t Last_Cost = 0; |
uint32_t Last_Cost = 0; |
792 |
|
|
793 |
int i, j; |
int i, j; |
794 |
Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */ |
|
795 |
|
/* source (w/ CBP penalty) */ |
796 |
|
Run_Costs[-1] = 2<<TL_SHIFT; |
797 |
|
|
798 |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
799 |
if (Non_Zero<0) |
if (Non_Zero<0) |
800 |
return -1; |
return 0; /* Sum is zero if there are only zero coeffs */ |
801 |
|
|
802 |
|
for(i=0; i<=Non_Zero; i++) { |
803 |
|
const int q = ((Q*QuantMatrix[Zigzag[i]])>>4); |
804 |
|
const int Mult = 2*q; |
805 |
|
const int Bias = (q-1) | 1; |
806 |
|
const int Lev0 = Mult + Bias; |
807 |
|
|
|
for(i=0; i<=Non_Zero; i++) |
|
|
{ |
|
808 |
const int AC = In[Zigzag[i]]; |
const int AC = In[Zigzag[i]]; |
809 |
const int Level1 = Out[Zigzag[i]]; |
const int Level1 = Out[Zigzag[i]]; |
810 |
const int Dist0 = Lambda* AC*AC; |
const unsigned int Dist0 = Lambda* AC*AC; |
811 |
uint32_t Best_Cost = 0xf0000000; |
uint32_t Best_Cost = 0xf0000000; |
812 |
Last_Cost += Dist0; |
Last_Cost += Dist0; |
813 |
|
|
814 |
if ((uint32_t)(Level1+1)<3) /* very specialized loop for -1,0,+1 */ |
/* very specialized loop for -1,0,+1 */ |
815 |
{ |
if ((uint32_t)(Level1+1)<3) { |
816 |
int dQ; |
int dQ; |
817 |
int Run; |
int Run; |
818 |
uint32_t Cost0; |
uint32_t Cost0; |
827 |
Cost0 = Lambda*dQ*dQ; |
Cost0 = Lambda*dQ*dQ; |
828 |
|
|
829 |
Nodes[i].Run = 1; |
Nodes[i].Run = 1; |
830 |
Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0; |
Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0; |
831 |
for(Run=i-Run_Start; Run>0; --Run) |
for(Run=i-Run_Start; Run>0; --Run) { |
|
{ |
|
832 |
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
833 |
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16); |
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT); |
834 |
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16); |
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT); |
835 |
|
|
836 |
/* |
/* TODO: what about tie-breaks? Should we favor short runs or |
|
* TODO: what about tie-breaks? Should we favor short runs or |
|
837 |
* long runs? Although the error is the same, it would not be |
* long runs? Although the error is the same, it would not be |
838 |
* spread the same way along high and low frequencies... |
* spread the same way along high and low frequencies... */ |
|
*/ |
|
839 |
|
|
840 |
/* (I'd say: favour short runs => hifreq errors (HVS) -- gruel ) */ |
/* Gruel: I'd say, favour short runs => hifreq errors (HVS) */ |
841 |
|
|
842 |
if (Cost<Best_Cost) { |
if (Cost<Best_Cost) { |
843 |
Best_Cost = Cost; |
Best_Cost = Cost; |
852 |
} |
} |
853 |
if (Last_Node==i) |
if (Last_Node==i) |
854 |
Last.Level = Nodes[i].Level; |
Last.Level = Nodes[i].Level; |
855 |
} |
} else if (51U>(uint32_t)(Level1+25)) { |
856 |
else /* "big" levels */ |
/* "big" levels (not less than ESC3, though) */ |
|
{ |
|
857 |
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; |
858 |
int Level2; |
int Level2; |
859 |
int dQ1, dQ2; |
int dQ1, dQ2; |
878 |
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; |
879 |
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; |
880 |
} |
} |
881 |
|
|
882 |
Dist1 = Lambda*dQ1*dQ1; |
Dist1 = Lambda*dQ1*dQ1; |
883 |
Dist2 = Lambda*dQ2*dQ2; |
Dist2 = Lambda*dQ2*dQ2; |
884 |
dDist21 = Dist2-Dist1; |
dDist21 = Dist2-Dist1; |
889 |
uint32_t Cost1, Cost2; |
uint32_t Cost1, Cost2; |
890 |
int bLevel; |
int bLevel; |
891 |
|
|
892 |
/* |
/* for sub-optimal (but slightly worth it, speed-wise) search, |
893 |
* for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following: |
* uncomment the following: |
894 |
* if (Cost_Base>=Best_Cost) continue; |
* if (Cost_Base>=Best_Cost) continue; |
895 |
* (? doesn't seem to have any effect -- gruel ) |
* (? doesn't seem to have any effect -- gruel ) */ |
|
*/ |
|
896 |
|
|
897 |
Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16); |
Cost1 = Cost_Base + (Tbl_L1[Run-1]<<TL_SHIFT); |
898 |
Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21; |
Cost2 = Cost_Base + (Tbl_L2[Run-1]<<TL_SHIFT) + dDist21; |
899 |
|
|
900 |
if (Cost2<Cost1) { |
if (Cost2<Cost1) { |
901 |
Cost1 = Cost2; |
Cost1 = Cost2; |
902 |
bLevel = Level2; |
bLevel = Level2; |
903 |
} else |
} else { |
904 |
bLevel = Level1; |
bLevel = Level1; |
905 |
|
} |
906 |
|
|
907 |
if (Cost1<Best_Cost) { |
if (Cost1<Best_Cost) { |
908 |
Best_Cost = Cost1; |
Best_Cost = Cost1; |
910 |
Nodes[i].Level = bLevel; |
Nodes[i].Level = bLevel; |
911 |
} |
} |
912 |
|
|
913 |
Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16); |
Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<TL_SHIFT); |
914 |
Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21; |
Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<TL_SHIFT) + dDist21; |
915 |
|
|
916 |
if (Cost2<Cost1) { |
if (Cost2<Cost1) { |
917 |
Cost1 = Cost2; |
Cost1 = Cost2; |
918 |
bLevel = Level2; |
bLevel = Level2; |
919 |
} else |
} else { |
920 |
bLevel = Level1; |
bLevel = Level1; |
921 |
|
} |
922 |
|
|
923 |
if (Cost1<Last_Cost) { |
if (Cost1<Last_Cost) { |
924 |
Last_Cost = Cost1; |
Last_Cost = Cost1; |
927 |
Last_Node = i; |
Last_Node = i; |
928 |
} |
} |
929 |
} /* end of "for Run" */ |
} /* end of "for Run" */ |
930 |
|
} else { |
931 |
|
/* Very very high levels, with no chance of being optimizable |
932 |
|
* => Simply pick best Run. */ |
933 |
|
int Run; |
934 |
|
for(Run=i-Run_Start; Run>0; --Run) { |
935 |
|
/* 30 bits + no distortion */ |
936 |
|
const uint32_t Cost = (30<<TL_SHIFT) + Run_Costs[i-Run]; |
937 |
|
if (Cost<Best_Cost) { |
938 |
|
Best_Cost = Cost; |
939 |
|
Nodes[i].Run = Run; |
940 |
|
Nodes[i].Level = Level1; |
941 |
|
} |
942 |
|
|
943 |
|
if (Cost<Last_Cost) { |
944 |
|
Last_Cost = Cost; |
945 |
|
Last.Run = Run; |
946 |
|
Last.Level = Level1; |
947 |
|
Last_Node = i; |
948 |
|
} |
949 |
} |
} |
950 |
|
} |
951 |
|
|
952 |
|
|
953 |
Run_Costs[i] = Best_Cost; |
Run_Costs[i] = Best_Cost; |
954 |
|
|
955 |
if (Best_Cost < Min_Cost + Dist0) { |
if (Best_Cost < Min_Cost + Dist0) { |
956 |
Min_Cost = Best_Cost; |
Min_Cost = Best_Cost; |
957 |
Run_Start = i; |
Run_Start = i; |
958 |
} |
} else { |
959 |
else |
/* as noticed by Michael Niedermayer (michaelni at gmx.at), |
960 |
{ |
* there's a code shorter by 1 bit for a larger run (!), same |
961 |
/* |
* level. We give it a chance by not moving the left barrier too |
962 |
* as noticed by Michael Niedermayer (michaelni at gmx.at), there's |
* much. */ |
963 |
* a code shorter by 1 bit for a larger run (!), same level. We give |
while( Run_Costs[Run_Start]>Min_Cost+(1<<TL_SHIFT) ) |
|
* it a chance by not moving the left barrier too much. |
|
|
*/ |
|
|
|
|
|
while( Run_Costs[Run_Start]>Min_Cost+(1<<16) ) |
|
964 |
Run_Start++; |
Run_Start++; |
965 |
|
|
966 |
/* spread on preceding coeffs the cost incurred by skipping this one */ |
/* spread on preceding coeffs the cost incurred by skipping this |
967 |
|
* one */ |
968 |
for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0; |
for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0; |
969 |
Min_Cost += Dist0; |
Min_Cost += Dist0; |
970 |
} |
} |
971 |
} |
} |
972 |
|
|
973 |
|
/* It seems trellis doesn't give good results... just leave the block untouched |
974 |
|
* and return the original sum value */ |
975 |
if (Last_Node<0) |
if (Last_Node<0) |
976 |
return -1; |
return Sum; |
977 |
|
|
978 |
/* reconstruct optimal sequence backward with surviving paths */ |
/* reconstruct optimal sequence backward with surviving paths */ |
979 |
memset(Out, 0x00, 64*sizeof(*Out)); |
memset(Out, 0x00, 64*sizeof(*Out)); |
980 |
Out[Zigzag[Last_Node]] = Last.Level; |
Out[Zigzag[Last_Node]] = Last.Level; |
981 |
i = Last_Node - Last.Run; |
i = Last_Node - Last.Run; |
982 |
|
Sum = abs(Last.Level); |
983 |
while(i>=0) { |
while(i>=0) { |
984 |
Out[Zigzag[i]] = Nodes[i].Level; |
Out[Zigzag[i]] = Nodes[i].Level; |
985 |
|
Sum += abs(Nodes[i].Level); |
986 |
i -= Nodes[i].Run; |
i -= Nodes[i].Run; |
987 |
} |
} |
|
return Last_Node; |
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
988 |
|
|
989 |
|
return Sum; |
990 |
|
} |
991 |
|
|
992 |
/* original version including heavy debugging info */ |
/* original version including heavy debugging info */ |
993 |
|
|
1045 |
V -= Ref[Zigzag[i]]; |
V -= Ref[Zigzag[i]]; |
1046 |
Dist += V*V; |
Dist += V*V; |
1047 |
} |
} |
1048 |
Cost = Lambda*Dist + (Bits<<16); |
Cost = Lambda*Dist + (Bits<<TL_SHIFT); |
1049 |
if (DBG==1) |
if (DBG==1) |
1050 |
printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 ); |
printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 ); |
1051 |
return Cost; |
return Cost; |
1077 |
const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */ |
const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */ |
1078 |
|
|
1079 |
int Run_Start = -1; |
int Run_Start = -1; |
1080 |
Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */ |
Run_Costs[-1] = 2<<TL_SHIFT; /* source (w/ CBP penalty) */ |
1081 |
uint32_t Min_Cost = 2<<16; |
uint32_t Min_Cost = 2<<TL_SHIFT; |
1082 |
|
|
1083 |
int Last_Node = -1; |
int Last_Node = -1; |
1084 |
uint32_t Last_Cost = 0; |
uint32_t Last_Cost = 0; |
1117 |
Cost0 = Lambda*dQ*dQ; |
Cost0 = Lambda*dQ*dQ; |
1118 |
|
|
1119 |
Nodes[i].Run = 1; |
Nodes[i].Run = 1; |
1120 |
Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0; |
Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0; |
1121 |
for(Run=i-Run_Start; Run>0; --Run) |
for(Run=i-Run_Start; Run>0; --Run) |
1122 |
{ |
{ |
1123 |
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
1124 |
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16); |
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT); |
1125 |
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16); |
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT); |
1126 |
|
|
1127 |
/* |
/* |
1128 |
* TODO: what about tie-breaks? Should we favor short runs or |
* TODO: what about tie-breaks? Should we favor short runs or |
1198 |
* for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following: |
* for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following: |
1199 |
* if (Cost_Base>=Best_Cost) continue; |
* if (Cost_Base>=Best_Cost) continue; |
1200 |
*/ |
*/ |
1201 |
Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16); |
Cost1 = Cost_Base + (Tbl_L1[Run-1]<<TL_SHIFT); |
1202 |
Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21; |
Cost2 = Cost_Base + (Tbl_L2[Run-1]<<TL_SHIFT) + dDist21; |
1203 |
|
|
1204 |
if (Cost2<Cost1) { |
if (Cost2<Cost1) { |
1205 |
Cost1 = Cost2; |
Cost1 = Cost2; |
1213 |
Nodes[i].Level = bLevel; |
Nodes[i].Level = bLevel; |
1214 |
} |
} |
1215 |
|
|
1216 |
Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16); |
Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<TL_SHIFT); |
1217 |
Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21; |
Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<TL_SHIFT) + dDist21; |
1218 |
|
|
1219 |
if (Cost2<Cost1) { |
if (Cost2<Cost1) { |
1220 |
Cost1 = Cost2; |
Cost1 = Cost2; |
1260 |
* it a chance by not moving the left barrier too much. |
* it a chance by not moving the left barrier too much. |
1261 |
*/ |
*/ |
1262 |
|
|
1263 |
while( Run_Costs[Run_Start]>Min_Cost+(1<<16) ) |
while( Run_Costs[Run_Start]>Min_Cost+(1<<TL_SHIFT) ) |
1264 |
Run_Start++; |
Run_Start++; |
1265 |
|
|
1266 |
/* spread on preceding coeffs the cost incurred by skipping this one */ |
/* spread on preceding coeffs the cost incurred by skipping this one */ |