42 |
* * |
* * |
43 |
* Revision history: * |
* Revision history: * |
44 |
* * |
* * |
45 |
* 29.03.2002 interlacing speedup - used transfer strides instead of |
* 29.03.2002 interlacing speedup - used transfer strides instead of * |
46 |
* manual field-to-frame conversion |
* manual field-to-frame conversion * |
47 |
* 26.03.2002 interlacing support - moved transfers outside loops |
* 26.03.2002 interlacing support - moved transfers outside loops * |
48 |
* 22.12.2001 get_dc_scaler() moved to common.h |
* 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 |
* * |
* * |
65 |
#include "../quant/quant_h263.h" |
#include "../quant/quant_h263.h" |
66 |
#include "../encoder.h" |
#include "../encoder.h" |
67 |
|
|
68 |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
#include "../image/reduced.h" |
|
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
|
69 |
|
|
70 |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
MBFIELDTEST_PTR MBFieldTest; |
71 |
|
|
72 |
/* this isnt pretty, but its better than 20 ifdefs */ |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
73 |
|
|
74 |
void |
void |
75 |
MBTransQuantIntra(const MBParam * pParam, |
MBTransQuantIntra(const MBParam * pParam, |
83 |
|
|
84 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
85 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
86 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); |
87 |
uint32_t i; |
uint32_t i; |
88 |
uint32_t iQuant = frame->quant; |
uint32_t iQuant = frame->quant; |
89 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
90 |
IMAGE *pCurrent = &frame->image; |
IMAGE *pCurrent = &frame->image; |
91 |
|
|
92 |
|
start_timer(); |
93 |
|
if ((frame->global_flags & XVID_REDUCED)) |
94 |
|
{ |
95 |
|
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
96 |
|
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
97 |
|
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
98 |
|
|
99 |
|
filter_18x18_to_8x8(&data[0 * 64], pY_Cur, stride); |
100 |
|
filter_18x18_to_8x8(&data[1 * 64], pY_Cur + 16, stride); |
101 |
|
filter_18x18_to_8x8(&data[2 * 64], pY_Cur + next_block, stride); |
102 |
|
filter_18x18_to_8x8(&data[3 * 64], pY_Cur + next_block + 16, stride); |
103 |
|
filter_18x18_to_8x8(&data[4 * 64], pU_Cur, stride2); |
104 |
|
filter_18x18_to_8x8(&data[5 * 64], pV_Cur, stride2); |
105 |
|
}else{ |
106 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
107 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
108 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
109 |
|
|
|
start_timer(); |
|
110 |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
111 |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
112 |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
113 |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
114 |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
115 |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
116 |
|
} |
117 |
stop_transfer_timer(); |
stop_transfer_timer(); |
118 |
|
|
119 |
|
/* XXX: rrv+interlacing is buggy */ |
120 |
start_timer(); |
start_timer(); |
121 |
pMB->field_dct = 0; |
pMB->field_dct = 0; |
122 |
if ((frame->global_flags & XVID_INTERLACING)) { |
if ((frame->global_flags & XVID_INTERLACING) && |
123 |
|
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
124 |
|
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
125 |
pMB->field_dct = MBDecideFieldDCT(data); |
pMB->field_dct = MBDecideFieldDCT(data); |
126 |
} |
} |
127 |
stop_interlacing_timer(); |
stop_interlacing_timer(); |
137 |
start_timer(); |
start_timer(); |
138 |
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
139 |
stop_quant_timer(); |
stop_quant_timer(); |
|
|
|
|
start_timer(); |
|
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
|
|
stop_iquant_timer(); |
|
140 |
} else { |
} else { |
141 |
start_timer(); |
start_timer(); |
142 |
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
143 |
stop_quant_timer(); |
stop_quant_timer(); |
144 |
|
} |
145 |
|
|
146 |
|
/* speedup: dont decode when encoding only ivops */ |
147 |
|
if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) |
148 |
|
{ |
149 |
|
if (pParam->m_quant_type == H263_QUANT) { |
150 |
|
start_timer(); |
151 |
|
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
152 |
|
stop_iquant_timer(); |
153 |
|
} else { |
154 |
start_timer(); |
start_timer(); |
155 |
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
156 |
stop_iquant_timer(); |
stop_iquant_timer(); |
160 |
idct(&data[i * 64]); |
idct(&data[i * 64]); |
161 |
stop_idct_timer(); |
stop_idct_timer(); |
162 |
} |
} |
163 |
|
} |
164 |
|
|
165 |
|
/* speedup: dont decode when encoding only ivops */ |
166 |
|
if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) |
167 |
|
{ |
168 |
|
|
169 |
if (pMB->field_dct) { |
if (pMB->field_dct) { |
170 |
next_block = stride; |
next_block = stride; |
172 |
} |
} |
173 |
|
|
174 |
start_timer(); |
start_timer(); |
175 |
|
if ((frame->global_flags & XVID_REDUCED)) |
176 |
|
{ |
177 |
|
copy_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); |
178 |
|
copy_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); |
179 |
|
copy_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); |
180 |
|
copy_upsampled_8x8_16to8(pY_Cur + next_block + 16, &data[3 * 64], stride); |
181 |
|
copy_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); |
182 |
|
copy_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); |
183 |
|
|
184 |
|
}else{ |
185 |
transfer_16to8copy(pY_Cur, &data[0 * 64], stride); |
transfer_16to8copy(pY_Cur, &data[0 * 64], stride); |
186 |
transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); |
transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); |
187 |
transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); |
transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); |
188 |
transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); |
transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); |
189 |
transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); |
transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); |
190 |
transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); |
transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); |
191 |
|
} |
192 |
stop_transfer_timer(); |
stop_transfer_timer(); |
193 |
|
} |
194 |
|
|
195 |
} |
} |
196 |
|
|
207 |
|
|
208 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
209 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
210 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); |
211 |
uint32_t i; |
uint32_t i; |
212 |
uint32_t iQuant = frame->quant; |
uint32_t iQuant = frame->quant; |
213 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
215 |
uint32_t sum; |
uint32_t sum; |
216 |
IMAGE *pCurrent = &frame->image; |
IMAGE *pCurrent = &frame->image; |
217 |
|
|
218 |
|
if ((frame->global_flags & XVID_REDUCED)) |
219 |
|
{ |
220 |
|
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
221 |
|
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
222 |
|
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
223 |
|
}else{ |
224 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
225 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
226 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
227 |
|
} |
228 |
|
|
229 |
start_timer(); |
start_timer(); |
230 |
pMB->field_dct = 0; |
pMB->field_dct = 0; |
231 |
if ((frame->global_flags & XVID_INTERLACING)) { |
if ((frame->global_flags & XVID_INTERLACING) && |
232 |
|
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
233 |
|
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
234 |
pMB->field_dct = MBDecideFieldDCT(data); |
pMB->field_dct = MBDecideFieldDCT(data); |
235 |
} |
} |
236 |
stop_interlacing_timer(); |
stop_interlacing_timer(); |
237 |
|
|
238 |
for (i = 0; i < 6; i++) { |
for (i = 0; i < 6; i++) { |
239 |
|
uint32_t increase_limit = (iQuant == 1) ? 1 : 0; |
240 |
|
|
241 |
/* |
/* |
242 |
* no need to transfer 8->16-bit |
* no need to transfer 8->16-bit |
243 |
* (this is performed already in motion compensation) |
* (this is performed already in motion compensation) |
256 |
stop_quant_timer(); |
stop_quant_timer(); |
257 |
} |
} |
258 |
|
|
259 |
if (sum >= TOOSMALL_LIMIT) { // skip block ? |
if ((sum >= TOOSMALL_LIMIT + increase_limit) || (qcoeff[i*64] != 0) || |
260 |
|
(qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
261 |
|
|
262 |
if (pParam->m_quant_type == H263_QUANT) { |
if (pParam->m_quant_type == H263_QUANT) { |
263 |
start_timer(); |
start_timer(); |
283 |
} |
} |
284 |
|
|
285 |
start_timer(); |
start_timer(); |
286 |
|
if ((frame->global_flags & XVID_REDUCED)) |
287 |
|
{ |
288 |
|
if (cbp & 32) |
289 |
|
add_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); |
290 |
|
if (cbp & 16) |
291 |
|
add_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); |
292 |
|
if (cbp & 8) |
293 |
|
add_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); |
294 |
|
if (cbp & 4) |
295 |
|
add_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride); |
296 |
|
if (cbp & 2) |
297 |
|
add_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); |
298 |
|
if (cbp & 1) |
299 |
|
add_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); |
300 |
|
}else{ |
301 |
if (cbp & 32) |
if (cbp & 32) |
302 |
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
303 |
if (cbp & 16) |
if (cbp & 16) |
310 |
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
311 |
if (cbp & 1) |
if (cbp & 1) |
312 |
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
313 |
|
} |
314 |
stop_transfer_timer(); |
stop_transfer_timer(); |
315 |
|
|
316 |
return cbp; |
return cbp; |
361 |
MBTransQuantInterBVOP(const MBParam * pParam, |
MBTransQuantInterBVOP(const MBParam * pParam, |
362 |
FRAMEINFO * frame, |
FRAMEINFO * frame, |
363 |
MACROBLOCK * pMB, |
MACROBLOCK * pMB, |
|
const uint32_t x_pos, |
|
|
const uint32_t y_pos, |
|
364 |
int16_t data[6 * 64], |
int16_t data[6 * 64], |
365 |
int16_t qcoeff[6 * 64]) |
int16_t qcoeff[6 * 64]) |
366 |
{ |
{ |
637 |
|
|
638 |
|
|
639 |
|
|
640 |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
/* permute block and return field dct choice */ |
641 |
|
|
642 |
|
|
643 |
uint32_t |
uint32_t |
644 |
MBDecideFieldDCT(int16_t data[6 * 64]) |
MBDecideFieldDCT(int16_t data[6 * 64]) |
645 |
{ |
{ |
646 |
|
uint32_t field = MBFieldTest(data); |
647 |
|
|
648 |
|
if (field) { |
649 |
|
MBFrameToField(data); |
650 |
|
} |
651 |
|
|
652 |
|
return field; |
653 |
|
} |
654 |
|
|
655 |
|
|
656 |
|
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
657 |
|
|
658 |
|
uint32_t |
659 |
|
MBFieldTest_c(int16_t data[6 * 64]) |
660 |
|
{ |
661 |
const uint8_t blocks[] = |
const uint8_t blocks[] = |
662 |
{ 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 }; |
663 |
const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 }; |
const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 }; |
691 |
} |
} |
692 |
} |
} |
693 |
|
|
694 |
if (frame > field) { |
return (frame >= (field + 350)); |
|
MBFrameToField(data); |
|
|
} |
|
|
|
|
|
return (frame > field); |
|
695 |
} |
} |
696 |
|
|
697 |
|
|