31 |
const uint32_t rounding) |
const uint32_t rounding) |
32 |
{ |
{ |
33 |
|
|
34 |
if (reduced_resolution) |
if (reduced_resolution) { |
|
{ |
|
35 |
const uint8_t * reference; |
const uint8_t * reference; |
36 |
x*=2; y*=2; |
x*=2; y*=2; |
37 |
|
|
53 |
|
|
54 |
}else{ |
}else{ |
55 |
if(quarterpel) { |
if(quarterpel) { |
56 |
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
const uint8_t * ptr; |
57 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
if (dx&3 | dy&3) { |
58 |
|
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refv + 32, |
59 |
|
(uint8_t *) refv + 64, (uint8_t *) refv + 96, x, y, dx, dy, stride, rounding); |
60 |
|
ptr = refv + y*stride + x; |
61 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position |
62 |
|
|
63 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
64 |
refv + y*stride + x, stride); |
ptr, stride); |
65 |
transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
66 |
refv + y*stride + x + 8, stride); |
ptr + 8, stride); |
67 |
transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
68 |
refv + y*stride + x + 8*stride, stride); |
ptr + 8*stride, stride); |
69 |
transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
70 |
refv + y*stride + x + 8*stride+8, stride); |
ptr + 8*stride+8, stride); |
71 |
|
|
72 |
} |
} else { |
|
else |
|
|
{ |
|
73 |
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
74 |
|
|
75 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
100 |
const int reduced_resolution, |
const int reduced_resolution, |
101 |
const uint32_t rounding) |
const uint32_t rounding) |
102 |
{ |
{ |
103 |
if (reduced_resolution) |
if (reduced_resolution) { |
|
{ |
|
104 |
const uint8_t * reference; |
const uint8_t * reference; |
105 |
x*=2; y*=2; |
x*=2; y*=2; |
106 |
|
|
114 |
} else { |
} else { |
115 |
|
|
116 |
if(quarterpel) { |
if(quarterpel) { |
117 |
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
const uint8_t * ptr; |
118 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
if (dx&3 | dy&3) { |
119 |
|
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refv+32, |
120 |
|
(uint8_t *) refv + 64, (uint8_t *) refv+96, x, y, dx, dy, stride, rounding); |
121 |
|
ptr = refv + y*stride + x; |
122 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position |
123 |
|
|
124 |
|
|
125 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
126 |
refv + y*stride + x, stride); |
ptr, stride); |
127 |
} |
} else { |
|
else |
|
|
{ |
|
128 |
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
129 |
|
|
130 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
177 |
const int reduced_resolution, |
const int reduced_resolution, |
178 |
const uint32_t rounding) |
const uint32_t rounding) |
179 |
{ |
{ |
|
|
|
|
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
|
|
|
|
180 |
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
181 |
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
182 |
|
|
193 |
edged_width / 2); |
edged_width / 2); |
194 |
return; |
return; |
195 |
} |
} |
196 |
|
|
197 |
|
if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) /*&& !quarterpel*/) { |
198 |
|
|
199 |
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
200 |
|
|
201 |
if (reduced_resolution) |
if (reduced_resolution) { |
|
{ |
|
202 |
dx = RRV_MV_SCALEUP(dx); |
dx = RRV_MV_SCALEUP(dx); |
203 |
dy = RRV_MV_SCALEUP(dy); |
dy = RRV_MV_SCALEUP(dy); |
204 |
} |
} |
207 |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
208 |
edged_width, quarterpel, reduced_resolution, rounding); |
edged_width, quarterpel, reduced_resolution, rounding); |
209 |
|
|
210 |
if (quarterpel) |
if (quarterpel) { |
|
{ |
|
211 |
dx /= 2; |
dx /= 2; |
212 |
dy /= 2; |
dy /= 2; |
213 |
} |
} |
216 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
217 |
|
|
218 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
219 |
if (reduced_resolution) |
if (reduced_resolution) { |
|
{ |
|
220 |
const int stride = edged_width/2; |
const int stride = edged_width/2; |
221 |
uint8_t * current, * reference; |
uint8_t * current, * reference; |
222 |
|
|
249 |
|
|
250 |
} else { // mode == MODE_INTER4V |
} else { // mode == MODE_INTER4V |
251 |
int k; |
int k; |
|
int32_t sum, dx, dy; |
|
252 |
VECTOR mvs[4]; |
VECTOR mvs[4]; |
253 |
|
|
254 |
if(quarterpel) |
if(quarterpel) |
257 |
for (k = 0; k < 4; k++) mvs[k] = mb->mvs[k]; |
for (k = 0; k < 4; k++) mvs[k] = mb->mvs[k]; |
258 |
|
|
259 |
if (reduced_resolution) |
if (reduced_resolution) |
260 |
{ |
for (k = 0; k < 4; k++) { |
|
for (k = 0; k < 4; k++) |
|
|
{ |
|
261 |
mvs[k].x = RRV_MV_SCALEUP(mvs[k].x); |
mvs[k].x = RRV_MV_SCALEUP(mvs[k].x); |
262 |
mvs[k].y = RRV_MV_SCALEUP(mvs[k].y); |
mvs[k].y = RRV_MV_SCALEUP(mvs[k].y); |
263 |
} |
} |
|
} |
|
264 |
|
|
265 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
266 |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
275 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
276 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
277 |
|
|
278 |
if(quarterpel) |
if (quarterpel) { |
279 |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
dx = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
280 |
else |
dy = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
281 |
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
} else { |
282 |
|
dx = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
283 |
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
dy = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
284 |
|
} |
|
if(quarterpel) |
|
|
sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
|
|
else |
|
|
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
|
285 |
|
|
286 |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
dx = (dx >> 3) + roundtab_76[dx & 0xf]; |
287 |
|
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
288 |
|
|
289 |
|
|
290 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
291 |
if (reduced_resolution) |
if (reduced_resolution) { |
|
{ |
|
292 |
const int stride = edged_width/2; |
const int stride = edged_width/2; |
293 |
uint8_t * current, * reference; |
uint8_t * current, * reference; |
294 |
|
|
348 |
int k,sum; |
int k,sum; |
349 |
int x = i; |
int x = i; |
350 |
int y = j; |
int y = j; |
351 |
uint32_t quarterpel = pParam->m_quarterpel; |
const uint32_t quarterpel = pParam->m_quarterpel; |
352 |
|
const uint8_t * ptr1, * ptr2; |
353 |
|
|
354 |
switch (mb->mode) { |
switch (mb->mode) { |
355 |
case MODE_FORWARD: |
case MODE_FORWARD: |
364 |
|
|
365 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
366 |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
367 |
dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
dy, edged_width, quarterpel, 0, 0); |
368 |
|
|
369 |
if (quarterpel) { |
if (quarterpel) { |
370 |
dx /= 2; |
dx /= 2; |
379 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
380 |
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
381 |
dx, dy, edged_width / 2, 0), |
dx, dy, edged_width / 2, 0), |
|
|
|
382 |
edged_width / 2); |
edged_width / 2); |
383 |
|
|
384 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
385 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
386 |
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
387 |
dx, dy, edged_width / 2, 0), |
dx, dy, edged_width / 2, 0), |
|
|
|
388 |
edged_width / 2); |
edged_width / 2); |
389 |
|
|
390 |
break; |
break; |
400 |
|
|
401 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
402 |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
403 |
b_dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
b_dy, edged_width, quarterpel, 0, 0); |
404 |
|
|
405 |
if (quarterpel) { |
if (quarterpel) { |
406 |
b_dx /= 2; |
b_dx /= 2; |
437 |
b_dx = mb->b_qmvs[0].x; |
b_dx = mb->b_qmvs[0].x; |
438 |
b_dy = mb->b_qmvs[0].y; |
b_dy = mb->b_qmvs[0].y; |
439 |
|
|
440 |
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refh->y, |
if (dx&3 | dy&3) { |
441 |
(uint8_t *) f_refh->y + 64, (uint8_t *) f_refhv->y, 16*i, 16*j, dx, dy, edged_width, 0); |
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refv->y + 32, |
442 |
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) b_refh->y, |
(uint8_t *) f_refv->y + 64, (uint8_t *) f_refv->y + 96, 16*i, 16*j, dx, dy, edged_width, 0); |
443 |
(uint8_t *) b_refh->y + 64, (uint8_t *) b_refhv->y, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
ptr1 = f_refv->y + i * 16 + j * 16 * edged_width; |
444 |
|
} else ptr1 = f_ref->y + (16*j + dy/4)*edged_width + 16*i + dx/4; // fullpixel position |
445 |
|
|
446 |
|
if (b_dx&3 | b_dy&3) { |
447 |
|
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) f_refv->y + 32, |
448 |
|
(uint8_t *) f_refv->y + 64, (uint8_t *) f_refv->y + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
449 |
|
ptr2 = b_refv->y + i * 16 + j * 16 * edged_width; |
450 |
|
} else ptr2 = b_ref->y + (16*j + b_dy/4)*edged_width + 16*i + b_dx/4; // fullpixel position |
451 |
|
|
452 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
453 |
transfer_8to16sub2(&dct_codes[k * 64], |
transfer_8to16sub2(&dct_codes[k * 64], |
454 |
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
455 |
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
ptr1 + (k&1)*8 + (k>>1)*8*edged_width, |
456 |
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width); |
|
edged_width); |
|
457 |
} |
} |
458 |
b_dx /= 2; |
b_dx /= 2; |
459 |
b_dy /= 2; |
b_dy /= 2; |
506 |
break; |
break; |
507 |
|
|
508 |
case MODE_DIRECT: |
case MODE_DIRECT: |
509 |
|
|
510 |
if (quarterpel) { |
if (quarterpel) { |
511 |
for (k=0;k<4;k++) { |
for (k=0;k<4;k++) { |
|
|
|
512 |
dx = mb->qmvs[k].x; |
dx = mb->qmvs[k].x; |
513 |
dy = mb->qmvs[k].y; |
dy = mb->qmvs[k].y; |
514 |
b_dx = mb->b_qmvs[k].x; |
b_dx = mb->b_qmvs[k].x; |
515 |
b_dy = mb->b_qmvs[k].y; |
b_dy = mb->b_qmvs[k].y; |
516 |
|
|
517 |
|
if (dx&3 | dy&3) { |
518 |
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
519 |
(uint8_t *) f_ref->y, |
(uint8_t *) f_ref->y, |
520 |
(uint8_t *) f_refh->y, |
(uint8_t *) f_refv->y + 32, |
521 |
(uint8_t *) f_refh->y + 64, |
(uint8_t *) f_refv->y + 64, |
522 |
(uint8_t *) f_refhv->y, |
(uint8_t *) f_refv->y + 96, |
523 |
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
524 |
|
ptr1 = f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width; |
525 |
|
} else ptr1 = f_ref->y + (16*j + (k>>1)*8 + dy/4)*edged_width + 16*i + (k&1)*8 + dx/4; |
526 |
|
|
527 |
|
if (b_dx&3 | b_dy&3) { |
528 |
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
529 |
(uint8_t *) b_ref->y, |
(uint8_t *) b_ref->y, |
530 |
(uint8_t *) b_refh->y, |
(uint8_t *) f_refv->y + 32, |
531 |
(uint8_t *) b_refh->y + 64, |
(uint8_t *) f_refv->y + 64, |
532 |
(uint8_t *) b_refhv->y, |
(uint8_t *) f_refv->y + 96, |
533 |
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
534 |
|
ptr2 = b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width; |
535 |
|
} else ptr2 = b_ref->y + (16*j + (k>>1)*8 + b_dy/4)*edged_width + 16*i + (k&1)*8 + b_dx/4; |
536 |
|
|
537 |
|
|
538 |
transfer_8to16sub2(&dct_codes[k * 64], |
transfer_8to16sub2(&dct_codes[k * 64], |
539 |
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
540 |
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
ptr1, ptr2, edged_width); |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
edged_width); |
|
541 |
} |
} |
542 |
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
543 |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |