65 |
#include "../quant/quant_h263.h" |
#include "../quant/quant_h263.h" |
66 |
#include "../encoder.h" |
#include "../encoder.h" |
67 |
|
|
68 |
|
#include "../image/reduced.h" |
69 |
|
|
70 |
|
MBFIELDTEST_PTR MBFieldTest; |
71 |
|
|
72 |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
73 |
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
74 |
|
|
75 |
#define TOOSMALL_LIMIT 3 /* skip blocks having a coefficient sum below this value */ |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
|
|
|
|
/* this isnt pretty, but its better than 20 ifdefs */ |
|
76 |
|
|
77 |
void |
void |
78 |
MBTransQuantIntra(const MBParam * pParam, |
MBTransQuantIntra(const MBParam * pParam, |
86 |
|
|
87 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
88 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
89 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); |
90 |
uint32_t i; |
uint32_t i; |
91 |
uint32_t iQuant = frame->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; |
IMAGE *pCurrent = &frame->image; |
94 |
|
|
95 |
|
start_timer(); |
96 |
|
if ((frame->global_flags & XVID_REDUCED)) |
97 |
|
{ |
98 |
|
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
99 |
|
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
100 |
|
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
101 |
|
|
102 |
|
filter_18x18_to_8x8(&data[0 * 64], pY_Cur, stride); |
103 |
|
filter_18x18_to_8x8(&data[1 * 64], pY_Cur + 16, stride); |
104 |
|
filter_18x18_to_8x8(&data[2 * 64], pY_Cur + next_block, stride); |
105 |
|
filter_18x18_to_8x8(&data[3 * 64], pY_Cur + next_block + 16, stride); |
106 |
|
filter_18x18_to_8x8(&data[4 * 64], pU_Cur, stride2); |
107 |
|
filter_18x18_to_8x8(&data[5 * 64], pV_Cur, stride2); |
108 |
|
}else{ |
109 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
110 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
111 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
112 |
|
|
|
start_timer(); |
|
113 |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
114 |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
115 |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
116 |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
117 |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
118 |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
119 |
|
} |
120 |
stop_transfer_timer(); |
stop_transfer_timer(); |
121 |
|
|
122 |
|
/* XXX: rrv+interlacing is buggy */ |
123 |
start_timer(); |
start_timer(); |
124 |
pMB->field_dct = 0; |
pMB->field_dct = 0; |
125 |
if ((frame->global_flags & XVID_INTERLACING) && |
if ((frame->global_flags & XVID_INTERLACING) && |
140 |
start_timer(); |
start_timer(); |
141 |
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
142 |
stop_quant_timer(); |
stop_quant_timer(); |
|
|
|
|
start_timer(); |
|
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
|
|
stop_iquant_timer(); |
|
143 |
} else { |
} else { |
144 |
start_timer(); |
start_timer(); |
145 |
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
146 |
stop_quant_timer(); |
stop_quant_timer(); |
147 |
|
} |
148 |
|
|
149 |
|
/* speedup: dont decode when encoding only ivops */ |
150 |
|
if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) |
151 |
|
{ |
152 |
|
if (pParam->m_quant_type == H263_QUANT) { |
153 |
|
start_timer(); |
154 |
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
155 |
|
stop_iquant_timer(); |
156 |
|
} else { |
157 |
start_timer(); |
start_timer(); |
158 |
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
159 |
stop_iquant_timer(); |
stop_iquant_timer(); |
163 |
idct(&data[i * 64]); |
idct(&data[i * 64]); |
164 |
stop_idct_timer(); |
stop_idct_timer(); |
165 |
} |
} |
166 |
|
} |
167 |
|
|
168 |
|
/* speedup: dont decode when encoding only ivops */ |
169 |
|
if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) |
170 |
|
{ |
171 |
|
|
172 |
if (pMB->field_dct) { |
if (pMB->field_dct) { |
173 |
next_block = stride; |
next_block = stride; |
175 |
} |
} |
176 |
|
|
177 |
start_timer(); |
start_timer(); |
178 |
|
if ((frame->global_flags & XVID_REDUCED)) |
179 |
|
{ |
180 |
|
copy_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); |
181 |
|
copy_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); |
182 |
|
copy_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); |
183 |
|
copy_upsampled_8x8_16to8(pY_Cur + next_block + 16, &data[3 * 64], stride); |
184 |
|
copy_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); |
185 |
|
copy_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); |
186 |
|
|
187 |
|
}else{ |
188 |
transfer_16to8copy(pY_Cur, &data[0 * 64], stride); |
transfer_16to8copy(pY_Cur, &data[0 * 64], stride); |
189 |
transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); |
transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); |
190 |
transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); |
transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); |
191 |
transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); |
transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); |
192 |
transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); |
transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); |
193 |
transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); |
transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); |
194 |
|
} |
195 |
stop_transfer_timer(); |
stop_transfer_timer(); |
196 |
|
} |
197 |
|
|
198 |
} |
} |
199 |
|
|
210 |
|
|
211 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
212 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
213 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); |
214 |
uint32_t i; |
uint32_t i; |
215 |
uint32_t iQuant = frame->quant; |
uint32_t iQuant = frame->quant; |
216 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
218 |
uint32_t sum; |
uint32_t sum; |
219 |
IMAGE *pCurrent = &frame->image; |
IMAGE *pCurrent = &frame->image; |
220 |
|
|
221 |
|
if ((frame->global_flags & XVID_REDUCED)) |
222 |
|
{ |
223 |
|
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
224 |
|
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
225 |
|
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
226 |
|
}else{ |
227 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
228 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
229 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
230 |
|
} |
231 |
|
|
232 |
start_timer(); |
start_timer(); |
233 |
pMB->field_dct = 0; |
pMB->field_dct = 0; |
239 |
stop_interlacing_timer(); |
stop_interlacing_timer(); |
240 |
|
|
241 |
for (i = 0; i < 6; i++) { |
for (i = 0; i < 6; i++) { |
242 |
|
uint32_t increase_limit = (iQuant == 1) ? 1 : 0; |
243 |
|
|
244 |
/* |
/* |
245 |
* no need to transfer 8->16-bit |
* no need to transfer 8->16-bit |
246 |
* (this is performed already in motion compensation) |
* (this is performed already in motion compensation) |
259 |
stop_quant_timer(); |
stop_quant_timer(); |
260 |
} |
} |
261 |
|
|
262 |
if ((sum >= TOOSMALL_LIMIT) || (qcoeff[i*64] != 0) || |
if ((sum >= TOOSMALL_LIMIT + increase_limit) || (qcoeff[i*64] != 0) || |
263 |
(qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
(qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
264 |
|
|
265 |
if (pParam->m_quant_type == H263_QUANT) { |
if (pParam->m_quant_type == H263_QUANT) { |
286 |
} |
} |
287 |
|
|
288 |
start_timer(); |
start_timer(); |
289 |
|
if ((frame->global_flags & XVID_REDUCED)) |
290 |
|
{ |
291 |
|
if (cbp & 32) |
292 |
|
add_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); |
293 |
|
if (cbp & 16) |
294 |
|
add_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); |
295 |
|
if (cbp & 8) |
296 |
|
add_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); |
297 |
|
if (cbp & 4) |
298 |
|
add_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride); |
299 |
|
if (cbp & 2) |
300 |
|
add_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); |
301 |
|
if (cbp & 1) |
302 |
|
add_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); |
303 |
|
}else{ |
304 |
if (cbp & 32) |
if (cbp & 32) |
305 |
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
306 |
if (cbp & 16) |
if (cbp & 16) |
313 |
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
314 |
if (cbp & 1) |
if (cbp & 1) |
315 |
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
316 |
|
} |
317 |
stop_transfer_timer(); |
stop_transfer_timer(); |
318 |
|
|
319 |
return cbp; |
return cbp; |
640 |
|
|
641 |
|
|
642 |
|
|
643 |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
/* permute block and return field dct choice */ |
644 |
|
|
645 |
|
|
646 |
uint32_t |
uint32_t |
647 |
MBDecideFieldDCT(int16_t data[6 * 64]) |
MBDecideFieldDCT(int16_t data[6 * 64]) |
648 |
{ |
{ |
649 |
|
uint32_t field = MBFieldTest(data); |
650 |
|
|
651 |
|
if (field) { |
652 |
|
MBFrameToField(data); |
653 |
|
} |
654 |
|
|
655 |
|
return field; |
656 |
|
} |
657 |
|
|
658 |
|
|
659 |
|
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
660 |
|
|
661 |
|
uint32_t |
662 |
|
MBFieldTest_c(int16_t data[6 * 64]) |
663 |
|
{ |
664 |
const uint8_t blocks[] = |
const uint8_t blocks[] = |
665 |
{ 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 }; |
{ 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 }; |
666 |
const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 }; |
const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 }; |
694 |
} |
} |
695 |
} |
} |
696 |
|
|
697 |
if (frame > field) { |
return (frame >= (field + 350)); |
|
MBFrameToField(data); |
|
|
} |
|
|
|
|
|
return (frame > field); |
|
698 |
} |
} |
699 |
|
|
700 |
|
|