42 |
* * |
* * |
43 |
* Revision history: * |
* Revision history: * |
44 |
* * |
* * |
45 |
* 22.12.2001 get_dc_scaler() moved to common.h |
* 29.03.2002 interlacing speedup - used transfer strides instead of * |
46 |
|
* manual field-to-frame conversion * |
47 |
|
* 26.03.2002 interlacing support - moved transfers outside loops * |
48 |
|
* 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 |
* * |
* * |
52 |
******************************************************************************/ |
******************************************************************************/ |
53 |
|
|
54 |
|
#include <string.h> |
55 |
|
|
56 |
#include "../portab.h" |
#include "../portab.h" |
57 |
#include "mbfunctions.h" |
#include "mbfunctions.h" |
58 |
|
|
65 |
#include "../quant/quant_h263.h" |
#include "../quant/quant_h263.h" |
66 |
#include "../encoder.h" |
#include "../encoder.h" |
67 |
|
|
68 |
|
MBFIELDTEST_PTR MBFieldTest; |
69 |
|
|
70 |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
71 |
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
72 |
|
|
73 |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
#define TOOSMALL_LIMIT 2 /* skip blocks having a coefficient sum below this value */ |
74 |
|
|
75 |
/* this isnt pretty, but its better than 20 ifdefs */ |
/* this isnt pretty, but its better than 20 ifdefs */ |
76 |
|
|
77 |
void MBTransQuantIntra(const MBParam *pParam, |
void |
78 |
|
MBTransQuantIntra(const MBParam * pParam, |
79 |
|
FRAMEINFO * frame, |
80 |
|
MACROBLOCK * pMB, |
81 |
const uint32_t x_pos, |
const uint32_t x_pos, |
82 |
const uint32_t y_pos, |
const uint32_t y_pos, |
83 |
int16_t data[][64], |
int16_t data[6 * 64], |
84 |
int16_t qcoeff[][64], |
int16_t qcoeff[6 * 64]) |
|
IMAGE * const pCurrent) |
|
|
|
|
85 |
{ |
{ |
86 |
const uint32_t stride = pParam->edged_width; |
|
87 |
|
uint32_t stride = pParam->edged_width; |
88 |
|
uint32_t stride2 = stride / 2; |
89 |
|
uint32_t next_block = stride * 8; |
90 |
uint32_t i; |
uint32_t i; |
91 |
uint32_t iQuant = pParam->quant; |
uint32_t iQuant = frame->quant; |
92 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
93 |
|
IMAGE *pCurrent = &frame->image; |
94 |
|
|
95 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
96 |
pU_Cur = pCurrent->u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
97 |
pV_Cur = pCurrent->v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
|
|
for(i = 0; i < 6; i++) { |
|
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
|
98 |
|
|
99 |
start_timer(); |
start_timer(); |
100 |
|
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
101 |
|
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
102 |
|
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
103 |
|
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
104 |
|
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
105 |
|
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
106 |
|
stop_transfer_timer(); |
107 |
|
|
108 |
switch(i) { |
start_timer(); |
109 |
case 0 : |
pMB->field_dct = 0; |
110 |
transfer_8to16copy(data[0], pY_Cur, stride); |
if ((frame->global_flags & XVID_INTERLACING) && |
111 |
break; |
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
112 |
case 1 : |
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
113 |
transfer_8to16copy(data[1], pY_Cur + 8, stride); |
pMB->field_dct = MBDecideFieldDCT(data); |
|
break; |
|
|
case 2 : |
|
|
transfer_8to16copy(data[2], pY_Cur + 8 * stride, stride); |
|
|
break; |
|
|
case 3 : |
|
|
transfer_8to16copy(data[3], pY_Cur + 8 * stride + 8, stride); |
|
|
break; |
|
|
case 4 : |
|
|
transfer_8to16copy(data[4], pU_Cur, stride / 2); |
|
|
break; |
|
|
case 5 : |
|
|
transfer_8to16copy(data[5], pV_Cur, stride / 2); |
|
|
break; |
|
114 |
} |
} |
115 |
stop_transfer_timer(); |
stop_interlacing_timer(); |
116 |
|
|
117 |
|
for (i = 0; i < 6; i++) { |
118 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
119 |
|
|
120 |
start_timer(); |
start_timer(); |
121 |
fdct(data[i]); |
fdct(&data[i * 64]); |
122 |
stop_dct_timer(); |
stop_dct_timer(); |
123 |
|
|
124 |
if (pParam->quant_type == H263_QUANT) |
if (pParam->m_quant_type == H263_QUANT) { |
|
{ |
|
125 |
start_timer(); |
start_timer(); |
126 |
quant_intra(qcoeff[i], data[i], iQuant, iDcScaler); |
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
127 |
stop_quant_timer(); |
stop_quant_timer(); |
128 |
|
|
129 |
start_timer(); |
start_timer(); |
130 |
dequant_intra(data[i], qcoeff[i], iQuant, iDcScaler); |
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
131 |
stop_iquant_timer(); |
stop_iquant_timer(); |
132 |
} |
} else { |
|
else |
|
|
{ |
|
133 |
start_timer(); |
start_timer(); |
134 |
quant4_intra(qcoeff[i], data[i], iQuant, iDcScaler); |
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
135 |
stop_quant_timer(); |
stop_quant_timer(); |
136 |
|
|
137 |
start_timer(); |
start_timer(); |
138 |
dequant4_intra(data[i], qcoeff[i], iQuant, iDcScaler); |
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
139 |
stop_iquant_timer(); |
stop_iquant_timer(); |
140 |
} |
} |
141 |
|
|
142 |
start_timer(); |
start_timer(); |
143 |
idct(data[i]); |
idct(&data[i * 64]); |
144 |
stop_idct_timer(); |
stop_idct_timer(); |
145 |
|
} |
146 |
|
|
147 |
start_timer(); |
if (pMB->field_dct) { |
148 |
|
next_block = stride; |
149 |
switch(i) { |
stride *= 2; |
|
case 0: |
|
|
transfer_16to8copy(pY_Cur, data[0], stride); |
|
|
break; |
|
|
case 1: |
|
|
transfer_16to8copy(pY_Cur + 8, data[1], stride); |
|
|
break; |
|
|
case 2: |
|
|
transfer_16to8copy(pY_Cur + 8 * stride, data[2], stride); |
|
|
break; |
|
|
case 3: |
|
|
transfer_16to8copy(pY_Cur + 8 + 8 * stride, data[3], stride); |
|
|
break; |
|
|
case 4: |
|
|
transfer_16to8copy(pU_Cur, data[4], stride / 2); |
|
|
break; |
|
|
case 5: |
|
|
transfer_16to8copy(pV_Cur, data[5], stride / 2); |
|
|
break; |
|
150 |
} |
} |
151 |
|
|
152 |
|
start_timer(); |
153 |
|
transfer_16to8copy(pY_Cur, &data[0 * 64], stride); |
154 |
|
transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); |
155 |
|
transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); |
156 |
|
transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); |
157 |
|
transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); |
158 |
|
transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); |
159 |
stop_transfer_timer(); |
stop_transfer_timer(); |
|
} |
|
|
} |
|
160 |
|
|
161 |
|
} |
162 |
|
|
|
uint8_t MBTransQuantInter(const MBParam *pParam, |
|
|
const uint32_t x_pos, const uint32_t y_pos, |
|
|
int16_t data[][64], |
|
|
int16_t qcoeff[][64], |
|
|
IMAGE * const pCurrent) |
|
163 |
|
|
164 |
|
uint8_t |
165 |
|
MBTransQuantInter(const MBParam * pParam, |
166 |
|
FRAMEINFO * frame, |
167 |
|
MACROBLOCK * pMB, |
168 |
|
const uint32_t x_pos, |
169 |
|
const uint32_t y_pos, |
170 |
|
int16_t data[6 * 64], |
171 |
|
int16_t qcoeff[6 * 64]) |
172 |
{ |
{ |
173 |
const uint32_t stride = pParam->edged_width; |
|
174 |
uint8_t i; |
uint32_t stride = pParam->edged_width; |
175 |
uint8_t iQuant = pParam->quant; |
uint32_t stride2 = stride / 2; |
176 |
|
uint32_t next_block = stride * 8; |
177 |
|
uint32_t i; |
178 |
|
uint32_t iQuant = frame->quant; |
179 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
180 |
uint8_t cbp = 0; |
uint8_t cbp = 0; |
181 |
uint32_t sum; |
uint32_t sum; |
182 |
|
IMAGE *pCurrent = &frame->image; |
183 |
|
|
184 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
185 |
pU_Cur = pCurrent->u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
186 |
pV_Cur = pCurrent->v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
187 |
|
|
188 |
|
start_timer(); |
189 |
|
pMB->field_dct = 0; |
190 |
|
if ((frame->global_flags & XVID_INTERLACING) && |
191 |
|
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
192 |
|
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
193 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
194 |
|
} |
195 |
|
stop_interlacing_timer(); |
196 |
|
|
197 |
for(i = 0; i < 6; i++) { |
for(i = 0; i < 6; i++) { |
198 |
/* |
/* |
199 |
no need to transfer 8->16-bit |
* no need to transfer 8->16-bit |
200 |
(this is performed already in motion compensation) |
* (this is performed already in motion compensation) |
201 |
*/ |
*/ |
202 |
start_timer(); |
start_timer(); |
203 |
fdct(data[i]); |
fdct(&data[i * 64]); |
204 |
stop_dct_timer(); |
stop_dct_timer(); |
205 |
|
|
206 |
if (pParam->quant_type == 0) |
if (pParam->m_quant_type == 0) { |
207 |
{ |
start_timer(); |
208 |
|
sum = quant_inter(&qcoeff[i * 64], &data[i * 64], iQuant); |
209 |
|
stop_quant_timer(); |
210 |
|
} else { |
211 |
start_timer(); |
start_timer(); |
212 |
sum = quant_inter(qcoeff[i], data[i], iQuant); |
sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], iQuant); |
213 |
stop_quant_timer(); |
stop_quant_timer(); |
214 |
} |
} |
215 |
else |
|
216 |
|
if ((sum >= TOOSMALL_LIMIT) || (qcoeff[i*64] != 0) || |
217 |
|
(qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
218 |
|
|
219 |
|
if (pParam->m_quant_type == H263_QUANT) { |
220 |
|
start_timer(); |
221 |
|
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
222 |
|
stop_iquant_timer(); |
223 |
|
} else { |
224 |
|
start_timer(); |
225 |
|
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
226 |
|
stop_iquant_timer(); |
227 |
|
} |
228 |
|
|
229 |
|
cbp |= 1 << (5 - i); |
230 |
|
|
231 |
|
start_timer(); |
232 |
|
idct(&data[i * 64]); |
233 |
|
stop_idct_timer(); |
234 |
|
} |
235 |
|
} |
236 |
|
|
237 |
|
if (pMB->field_dct) { |
238 |
|
next_block = stride; |
239 |
|
stride *= 2; |
240 |
|
} |
241 |
|
|
242 |
|
start_timer(); |
243 |
|
if (cbp & 32) |
244 |
|
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
245 |
|
if (cbp & 16) |
246 |
|
transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride); |
247 |
|
if (cbp & 8) |
248 |
|
transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride); |
249 |
|
if (cbp & 4) |
250 |
|
transfer_16to8add(pY_Cur + next_block + 8, &data[3 * 64], stride); |
251 |
|
if (cbp & 2) |
252 |
|
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
253 |
|
if (cbp & 1) |
254 |
|
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
255 |
|
stop_transfer_timer(); |
256 |
|
|
257 |
|
return cbp; |
258 |
|
|
259 |
|
} |
260 |
|
|
261 |
|
void |
262 |
|
MBTransQuantIntra2(const MBParam * pParam, |
263 |
|
FRAMEINFO * frame, |
264 |
|
MACROBLOCK * pMB, |
265 |
|
const uint32_t x_pos, |
266 |
|
const uint32_t y_pos, |
267 |
|
int16_t data[6 * 64], |
268 |
|
int16_t qcoeff[6 * 64]) |
269 |
|
{ |
270 |
|
MBTrans(pParam,frame,pMB,x_pos,y_pos,data); |
271 |
|
MBfDCT(pParam,frame,pMB,data); |
272 |
|
MBQuantIntra(pParam,frame,pMB,data,qcoeff); |
273 |
|
MBDeQuantIntra(pParam,frame->quant,data,qcoeff); |
274 |
|
MBiDCT(data,0x3F); |
275 |
|
MBTransAdd(pParam,frame,pMB,x_pos,y_pos,data,0x3F); |
276 |
|
} |
277 |
|
|
278 |
|
|
279 |
|
uint8_t |
280 |
|
MBTransQuantInter2(const MBParam * pParam, |
281 |
|
FRAMEINFO * frame, |
282 |
|
MACROBLOCK * pMB, |
283 |
|
const uint32_t x_pos, |
284 |
|
const uint32_t y_pos, |
285 |
|
int16_t data[6 * 64], |
286 |
|
int16_t qcoeff[6 * 64]) |
287 |
{ |
{ |
288 |
|
uint8_t cbp; |
289 |
|
|
290 |
|
/* there is no MBTrans for Inter block, that's done in motion compensation already */ |
291 |
|
|
292 |
|
MBfDCT(pParam,frame,pMB,data); |
293 |
|
cbp = MBQuantInter(pParam,frame->quant,data,qcoeff); |
294 |
|
MBDeQuantInter(pParam,frame->quant,data,qcoeff,cbp); |
295 |
|
MBiDCT(data,cbp); |
296 |
|
MBTransAdd(pParam,frame,pMB,x_pos,y_pos,data,cbp); |
297 |
|
|
298 |
|
return cbp; |
299 |
|
} |
300 |
|
|
301 |
|
uint8_t |
302 |
|
MBTransQuantInterBVOP(const MBParam * pParam, |
303 |
|
FRAMEINFO * frame, |
304 |
|
MACROBLOCK * pMB, |
305 |
|
int16_t data[6 * 64], |
306 |
|
int16_t qcoeff[6 * 64]) |
307 |
|
{ |
308 |
|
uint8_t cbp; |
309 |
|
|
310 |
|
/* there is no MBTrans for Inter block, that's done in motion compensation already */ |
311 |
|
|
312 |
|
MBfDCT(pParam,frame,pMB,data); |
313 |
|
cbp = MBQuantInter(pParam,frame->quant,data,qcoeff); |
314 |
|
|
315 |
|
/* we don't have to DeQuant, iDCT and Transfer back data for B-frames */ |
316 |
|
|
317 |
|
return cbp; |
318 |
|
} |
319 |
|
|
320 |
|
|
321 |
|
void |
322 |
|
MBfDCT(const MBParam * pParam, |
323 |
|
FRAMEINFO * frame, |
324 |
|
MACROBLOCK * pMB, |
325 |
|
int16_t data[6 * 64]) |
326 |
|
{ |
327 |
|
int i; |
328 |
|
|
329 |
start_timer(); |
start_timer(); |
330 |
sum = quant4_inter(qcoeff[i], data[i], iQuant); |
pMB->field_dct = 0; |
331 |
stop_quant_timer(); |
if ((frame->global_flags & XVID_INTERLACING)) { |
332 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
333 |
} |
} |
334 |
|
stop_interlacing_timer(); |
335 |
|
|
336 |
if(sum >= TOOSMALL_LIMIT) { // skip block ? |
for (i = 0; i < 6; i++) { |
337 |
|
start_timer(); |
338 |
|
fdct(&data[i * 64]); |
339 |
|
stop_dct_timer(); |
340 |
|
} |
341 |
|
} |
342 |
|
|
343 |
if (pParam->quant_type == H263_QUANT) |
void |
344 |
|
MBQuantDeQuantIntra(const MBParam * pParam, |
345 |
|
FRAMEINFO * frame, |
346 |
|
MACROBLOCK * pMB, |
347 |
|
int16_t qcoeff[6 * 64], |
348 |
|
int16_t data[6*64]) |
349 |
{ |
{ |
350 |
|
int i; |
351 |
|
int iQuant = frame->quant; |
352 |
|
|
353 |
|
start_timer(); |
354 |
|
pMB->field_dct = 0; |
355 |
|
if ((frame->global_flags & XVID_INTERLACING)) { |
356 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
357 |
|
} |
358 |
|
stop_interlacing_timer(); |
359 |
|
|
360 |
|
for (i = 0; i < 6; i++) { |
361 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
362 |
|
|
363 |
|
if (pParam->m_quant_type == H263_QUANT) { |
364 |
start_timer(); |
start_timer(); |
365 |
dequant_inter(data[i], qcoeff[i], iQuant); |
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
366 |
|
stop_quant_timer(); |
367 |
|
|
368 |
|
start_timer(); |
369 |
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
370 |
stop_iquant_timer(); |
stop_iquant_timer(); |
371 |
|
} else { |
372 |
|
start_timer(); |
373 |
|
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
374 |
|
stop_quant_timer(); |
375 |
|
|
376 |
|
start_timer(); |
377 |
|
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
378 |
|
stop_iquant_timer(); |
379 |
|
} |
380 |
|
} |
381 |
} |
} |
382 |
else |
|
383 |
|
void |
384 |
|
MBQuantIntra(const MBParam * pParam, |
385 |
|
FRAMEINFO * frame, |
386 |
|
MACROBLOCK *pMB, |
387 |
|
int16_t qcoeff[6 * 64], |
388 |
|
int16_t data[6*64]) |
389 |
{ |
{ |
390 |
|
int i; |
391 |
|
int iQuant = frame->quant; |
392 |
|
|
393 |
|
start_timer(); |
394 |
|
pMB->field_dct = 0; |
395 |
|
if ((frame->global_flags & XVID_INTERLACING)) { |
396 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
397 |
|
} |
398 |
|
stop_interlacing_timer(); |
399 |
|
|
400 |
|
for (i = 0; i < 6; i++) { |
401 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
402 |
|
|
403 |
|
if (pParam->m_quant_type == H263_QUANT) { |
404 |
|
start_timer(); |
405 |
|
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
406 |
|
stop_quant_timer(); |
407 |
|
} else { |
408 |
start_timer(); |
start_timer(); |
409 |
dequant4_inter(data[i], qcoeff[i], iQuant); |
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
410 |
|
stop_quant_timer(); |
411 |
|
} |
412 |
|
} |
413 |
|
} |
414 |
|
|
415 |
|
void |
416 |
|
MBDeQuantIntra(const MBParam * pParam, |
417 |
|
const int iQuant, |
418 |
|
int16_t qcoeff[6 * 64], |
419 |
|
int16_t data[6*64]) |
420 |
|
{ |
421 |
|
int i; |
422 |
|
|
423 |
|
for (i = 0; i < 6; i++) { |
424 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
425 |
|
|
426 |
|
if (pParam->m_quant_type == H263_QUANT) { |
427 |
|
start_timer(); |
428 |
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
429 |
stop_iquant_timer(); |
stop_iquant_timer(); |
430 |
|
} else { |
431 |
|
start_timer(); |
432 |
|
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
433 |
|
stop_iquant_timer(); |
434 |
|
} |
435 |
|
} |
436 |
|
} |
437 |
|
|
438 |
|
uint8_t |
439 |
|
MBQuantInter(const MBParam * pParam, |
440 |
|
const int iQuant, |
441 |
|
int16_t data[6 * 64], |
442 |
|
int16_t qcoeff[6 * 64]) |
443 |
|
{ |
444 |
|
|
445 |
|
int i; |
446 |
|
uint8_t cbp = 0; |
447 |
|
int sum; |
448 |
|
|
449 |
|
for (i = 0; i < 6; i++) { |
450 |
|
|
451 |
|
if (pParam->m_quant_type == 0) { |
452 |
|
start_timer(); |
453 |
|
sum = quant_inter(&qcoeff[i * 64], &data[i * 64], iQuant); |
454 |
|
stop_quant_timer(); |
455 |
|
} else { |
456 |
|
start_timer(); |
457 |
|
sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], iQuant); |
458 |
|
stop_quant_timer(); |
459 |
} |
} |
460 |
|
|
461 |
|
if (sum >= TOOSMALL_LIMIT) { // skip block ? |
462 |
cbp |= 1 << (5 - i); |
cbp |= 1 << (5 - i); |
463 |
|
} |
464 |
|
} |
465 |
|
return cbp; |
466 |
|
} |
467 |
|
|
468 |
|
void |
469 |
|
MBDeQuantInter( const MBParam * pParam, |
470 |
|
const int iQuant, |
471 |
|
int16_t data[6 * 64], |
472 |
|
int16_t qcoeff[6 * 64], |
473 |
|
const uint8_t cbp) |
474 |
|
{ |
475 |
|
int i; |
476 |
|
|
477 |
|
for (i = 0; i < 6; i++) { |
478 |
|
if (cbp & (1 << (5 - i))) |
479 |
|
{ |
480 |
|
if (pParam->m_quant_type == H263_QUANT) { |
481 |
|
start_timer(); |
482 |
|
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
483 |
|
stop_iquant_timer(); |
484 |
|
} else { |
485 |
start_timer(); |
start_timer(); |
486 |
idct(data[i]); |
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
487 |
|
stop_iquant_timer(); |
488 |
|
} |
489 |
|
} |
490 |
|
} |
491 |
|
} |
492 |
|
|
493 |
|
void |
494 |
|
MBiDCT( int16_t data[6 * 64], |
495 |
|
const uint8_t cbp) |
496 |
|
{ |
497 |
|
int i; |
498 |
|
|
499 |
|
for (i = 0; i < 6; i++) { |
500 |
|
if (cbp & (1 << (5 - i))) |
501 |
|
{ |
502 |
|
start_timer(); |
503 |
|
idct(&data[i * 64]); |
504 |
stop_idct_timer(); |
stop_idct_timer(); |
505 |
|
|
506 |
|
} |
507 |
|
} |
508 |
|
} |
509 |
|
|
510 |
|
|
511 |
|
void |
512 |
|
MBTrans(const MBParam * pParam, |
513 |
|
FRAMEINFO * frame, |
514 |
|
MACROBLOCK * pMB, |
515 |
|
const uint32_t x_pos, |
516 |
|
const uint32_t y_pos, |
517 |
|
int16_t data[6 * 64]) |
518 |
|
{ |
519 |
|
uint32_t stride = pParam->edged_width; |
520 |
|
uint32_t stride2 = stride / 2; |
521 |
|
uint32_t next_block = stride * 8; |
522 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
523 |
|
IMAGE *pCurrent = &frame->image; |
524 |
|
|
525 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
526 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
527 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
528 |
|
|
529 |
start_timer(); |
start_timer(); |
530 |
|
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
531 |
|
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
532 |
|
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
533 |
|
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
534 |
|
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
535 |
|
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
536 |
|
stop_transfer_timer(); |
537 |
|
} |
538 |
|
|
539 |
switch(i) { |
void |
540 |
case 0: |
MBTransAdd(const MBParam * pParam, |
541 |
transfer_16to8add(pY_Cur, data[0], stride); |
FRAMEINFO * frame, |
542 |
break; |
MACROBLOCK * pMB, |
543 |
case 1: |
const uint32_t x_pos, |
544 |
transfer_16to8add(pY_Cur + 8, data[1], stride); |
const uint32_t y_pos, |
545 |
break; |
int16_t data[6 * 64], |
546 |
case 2: |
const uint8_t cbp) |
547 |
transfer_16to8add(pY_Cur + 8 * stride, data[2], stride); |
{ |
548 |
break; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
549 |
case 3: |
uint32_t stride = pParam->edged_width; |
550 |
transfer_16to8add(pY_Cur + 8 + 8 * stride, data[3], stride); |
uint32_t stride2 = stride / 2; |
551 |
break; |
uint32_t next_block = stride * 8; |
552 |
case 4: |
IMAGE *pCurrent = &frame->image; |
553 |
transfer_16to8add(pU_Cur, data[4], stride / 2); |
|
554 |
break; |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
555 |
case 5: |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
556 |
transfer_16to8add(pV_Cur, data[5], stride / 2); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
557 |
break; |
|
558 |
|
if (pMB->field_dct) { |
559 |
|
next_block = stride; |
560 |
|
stride *= 2; |
561 |
} |
} |
562 |
|
|
563 |
|
start_timer(); |
564 |
|
if (cbp & 32) |
565 |
|
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
566 |
|
if (cbp & 16) |
567 |
|
transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride); |
568 |
|
if (cbp & 8) |
569 |
|
transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride); |
570 |
|
if (cbp & 4) |
571 |
|
transfer_16to8add(pY_Cur + next_block + 8, &data[3 * 64], stride); |
572 |
|
if (cbp & 2) |
573 |
|
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
574 |
|
if (cbp & 1) |
575 |
|
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
576 |
stop_transfer_timer(); |
stop_transfer_timer(); |
577 |
} |
} |
578 |
|
|
579 |
|
|
580 |
|
|
581 |
|
/* permute block and return field dct choice */ |
582 |
|
|
583 |
|
|
584 |
|
uint32_t |
585 |
|
MBDecideFieldDCT(int16_t data[6 * 64]) |
586 |
|
{ |
587 |
|
uint32_t field = MBFieldTest(data); |
588 |
|
|
589 |
|
if (field) { |
590 |
|
MBFrameToField(data); |
591 |
} |
} |
592 |
return cbp; |
|
593 |
|
return field; |
594 |
|
} |
595 |
|
|
596 |
|
|
597 |
|
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
598 |
|
|
599 |
|
uint32_t |
600 |
|
MBFieldTest_c(int16_t data[6 * 64]) |
601 |
|
{ |
602 |
|
const uint8_t blocks[] = |
603 |
|
{ 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 }; |
604 |
|
const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 }; |
605 |
|
|
606 |
|
int frame = 0, field = 0; |
607 |
|
int i, j; |
608 |
|
|
609 |
|
for (i = 0; i < 7; ++i) { |
610 |
|
for (j = 0; j < 8; ++j) { |
611 |
|
frame += |
612 |
|
ABS(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]); |
613 |
|
frame += |
614 |
|
ABS(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]); |
615 |
|
frame += |
616 |
|
ABS(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]); |
617 |
|
frame += |
618 |
|
ABS(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]); |
619 |
|
|
620 |
|
field += |
621 |
|
ABS(data[blocks[i + 1] + lines[i + 1] + j] - |
622 |
|
data[blocks[i] + lines[i] + j]); |
623 |
|
field += |
624 |
|
ABS(data[blocks[i + 1] + lines[i + 1] + 8 + j] - |
625 |
|
data[blocks[i] + lines[i] + 8 + j]); |
626 |
|
field += |
627 |
|
ABS(data[blocks[i + 1] + 64 + lines[i + 1] + j] - |
628 |
|
data[blocks[i] + 64 + lines[i] + j]); |
629 |
|
field += |
630 |
|
ABS(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] - |
631 |
|
data[blocks[i] + 64 + lines[i] + 8 + j]); |
632 |
|
} |
633 |
|
} |
634 |
|
|
635 |
|
return (frame >= (field + 350)); |
636 |
|
} |
637 |
|
|
638 |
|
|
639 |
|
/* deinterlace Y blocks vertically */ |
640 |
|
|
641 |
|
#define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp)) |
642 |
|
#define LINE(X,Y) &data[X*64 + Y*8] |
643 |
|
|
644 |
|
void |
645 |
|
MBFrameToField(int16_t data[6 * 64]) |
646 |
|
{ |
647 |
|
int16_t tmp[8]; |
648 |
|
|
649 |
|
/* left blocks */ |
650 |
|
|
651 |
|
// 1=2, 2=4, 4=8, 8=1 |
652 |
|
MOVLINE(tmp, LINE(0, 1)); |
653 |
|
MOVLINE(LINE(0, 1), LINE(0, 2)); |
654 |
|
MOVLINE(LINE(0, 2), LINE(0, 4)); |
655 |
|
MOVLINE(LINE(0, 4), LINE(2, 0)); |
656 |
|
MOVLINE(LINE(2, 0), tmp); |
657 |
|
|
658 |
|
// 3=6, 6=12, 12=9, 9=3 |
659 |
|
MOVLINE(tmp, LINE(0, 3)); |
660 |
|
MOVLINE(LINE(0, 3), LINE(0, 6)); |
661 |
|
MOVLINE(LINE(0, 6), LINE(2, 4)); |
662 |
|
MOVLINE(LINE(2, 4), LINE(2, 1)); |
663 |
|
MOVLINE(LINE(2, 1), tmp); |
664 |
|
|
665 |
|
// 5=10, 10=5 |
666 |
|
MOVLINE(tmp, LINE(0, 5)); |
667 |
|
MOVLINE(LINE(0, 5), LINE(2, 2)); |
668 |
|
MOVLINE(LINE(2, 2), tmp); |
669 |
|
|
670 |
|
// 7=14, 14=13, 13=11, 11=7 |
671 |
|
MOVLINE(tmp, LINE(0, 7)); |
672 |
|
MOVLINE(LINE(0, 7), LINE(2, 6)); |
673 |
|
MOVLINE(LINE(2, 6), LINE(2, 5)); |
674 |
|
MOVLINE(LINE(2, 5), LINE(2, 3)); |
675 |
|
MOVLINE(LINE(2, 3), tmp); |
676 |
|
|
677 |
|
/* right blocks */ |
678 |
|
|
679 |
|
// 1=2, 2=4, 4=8, 8=1 |
680 |
|
MOVLINE(tmp, LINE(1, 1)); |
681 |
|
MOVLINE(LINE(1, 1), LINE(1, 2)); |
682 |
|
MOVLINE(LINE(1, 2), LINE(1, 4)); |
683 |
|
MOVLINE(LINE(1, 4), LINE(3, 0)); |
684 |
|
MOVLINE(LINE(3, 0), tmp); |
685 |
|
|
686 |
|
// 3=6, 6=12, 12=9, 9=3 |
687 |
|
MOVLINE(tmp, LINE(1, 3)); |
688 |
|
MOVLINE(LINE(1, 3), LINE(1, 6)); |
689 |
|
MOVLINE(LINE(1, 6), LINE(3, 4)); |
690 |
|
MOVLINE(LINE(3, 4), LINE(3, 1)); |
691 |
|
MOVLINE(LINE(3, 1), tmp); |
692 |
|
|
693 |
|
// 5=10, 10=5 |
694 |
|
MOVLINE(tmp, LINE(1, 5)); |
695 |
|
MOVLINE(LINE(1, 5), LINE(3, 2)); |
696 |
|
MOVLINE(LINE(3, 2), tmp); |
697 |
|
|
698 |
|
// 7=14, 14=13, 13=11, 11=7 |
699 |
|
MOVLINE(tmp, LINE(1, 7)); |
700 |
|
MOVLINE(LINE(1, 7), LINE(3, 6)); |
701 |
|
MOVLINE(LINE(3, 6), LINE(3, 5)); |
702 |
|
MOVLINE(LINE(3, 5), LINE(3, 3)); |
703 |
|
MOVLINE(LINE(3, 3), tmp); |
704 |
} |
} |