6 |
#include "../encoder.h" |
#include "../encoder.h" |
7 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
8 |
#include "../image/interpolate8x8.h" |
#include "../image/interpolate8x8.h" |
9 |
|
#include "../image/reduced.h" |
10 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
11 |
#include "motion.h" |
#include "motion.h" |
12 |
|
|
13 |
#define ABS(X) (((X)>0)?(X):-(X)) |
#define ABS(X) (((X)>0)?(X):-(X)) |
14 |
#define SIGN(X) (((X)>0)?1:-1) |
#define SIGN(X) (((X)>0)?1:-1) |
15 |
|
|
16 |
|
|
17 |
static __inline void |
static __inline void |
18 |
compensate16x16_interpolate(int16_t * const dct_codes, |
compensate16x16_interpolate(int16_t * const dct_codes, |
19 |
uint8_t * const cur, |
uint8_t * const cur, |
20 |
const uint8_t * const ref, |
const uint8_t * const ref, |
21 |
const uint8_t * const refh, |
const uint8_t * const refh, |
22 |
const uint8_t * const refv, |
uint8_t * const refv, |
23 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
24 |
const uint32_t x, |
uint32_t x, |
25 |
const uint32_t y, |
uint32_t y, |
26 |
const int32_t dx, |
const int32_t dx, |
27 |
const int32_t dy, |
const int32_t dy, |
28 |
const uint32_t stride, |
const uint32_t stride, |
29 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
30 |
|
const int reduced_resolution, |
31 |
const uint32_t rounding) |
const uint32_t rounding) |
32 |
{ |
{ |
33 |
|
|
34 |
|
if (reduced_resolution) |
35 |
|
{ |
36 |
|
const uint8_t * reference; |
37 |
|
x*=2; y*=2; |
38 |
|
|
39 |
|
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
40 |
|
|
41 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
42 |
|
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
43 |
|
|
44 |
|
filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
45 |
|
filter_diff_18x18_to_8x8(dct_codes+64, reference + 16, stride); |
46 |
|
|
47 |
|
filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
48 |
|
filter_diff_18x18_to_8x8(dct_codes+128, reference + 16*stride, stride); |
49 |
|
|
50 |
|
filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
51 |
|
filter_diff_18x18_to_8x8(dct_codes+192, reference + 16*stride + 16, stride); |
52 |
|
|
53 |
|
transfer32x32_copy(cur + y*stride + x, reference, stride); |
54 |
|
|
55 |
|
}else{ |
56 |
if(quarterpel) { |
if(quarterpel) { |
57 |
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
58 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
69 |
} |
} |
70 |
else |
else |
71 |
{ |
{ |
72 |
const uint8_t * reference; |
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
73 |
|
|
|
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
|
|
{ |
|
|
case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; |
|
|
case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; |
|
|
case 2: reference = refh + ((y + dy / 2) * stride + x + (dx-1) / 2); break; |
|
|
default: // case 3: |
|
|
reference = refhv + ((y + (dy-1) / 2) * stride + x + (dx-1) / 2); break; |
|
|
} |
|
74 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
75 |
reference, stride); |
reference, stride); |
76 |
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
81 |
reference + 8*stride + 8, stride); |
reference + 8*stride + 8, stride); |
82 |
} |
} |
83 |
} |
} |
84 |
|
} |
85 |
|
|
86 |
static __inline void |
static __inline void |
87 |
compensate8x8_interpolate(int16_t * const dct_codes, |
compensate8x8_interpolate(int16_t * const dct_codes, |
90 |
const uint8_t * const refh, |
const uint8_t * const refh, |
91 |
const uint8_t * const refv, |
const uint8_t * const refv, |
92 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
93 |
const uint32_t x, |
uint32_t x, |
94 |
const uint32_t y, |
uint32_t y, |
95 |
const int32_t dx, |
const int32_t dx, |
96 |
const int32_t dy, |
const int32_t dy, |
97 |
const uint32_t stride, |
const uint32_t stride, |
98 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
99 |
|
const int reduced_resolution, |
100 |
const uint32_t rounding) |
const uint32_t rounding) |
101 |
{ |
{ |
102 |
|
if (reduced_resolution) |
103 |
|
{ |
104 |
|
const uint8_t * reference; |
105 |
|
x*=2; y*=2; |
106 |
|
|
107 |
|
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
108 |
|
|
109 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
110 |
|
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
111 |
|
|
112 |
|
transfer16x16_copy(cur + y*stride + x, reference, stride); |
113 |
|
|
114 |
|
} else { |
115 |
|
|
116 |
if(quarterpel) { |
if(quarterpel) { |
117 |
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
118 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
122 |
} |
} |
123 |
else |
else |
124 |
{ |
{ |
125 |
const uint8_t * reference; |
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
126 |
|
|
|
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
|
|
{ |
|
|
case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; |
|
|
case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; |
|
|
case 2: reference = refh + ((y + dy / 2) * stride + x + (dx-1) / 2); break; |
|
|
default: // case 3: |
|
|
reference = refhv + ((y + (dy-1) / 2) * stride + x + (dx-1) / 2); break; |
|
|
} |
|
127 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
128 |
reference, stride); |
reference, stride); |
129 |
} |
} |
130 |
} |
} |
131 |
|
} |
132 |
|
|
133 |
void |
void |
134 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
144 |
const uint32_t height, |
const uint32_t height, |
145 |
const uint32_t edged_width, |
const uint32_t edged_width, |
146 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
147 |
|
const int reduced_resolution, |
148 |
const uint32_t rounding) |
const uint32_t rounding) |
149 |
{ |
{ |
150 |
if (mb->mode == MODE_NOT_CODED) { |
|
151 |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
152 |
|
|
153 |
|
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
154 |
|
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
155 |
|
|
156 |
|
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
157 |
|
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
158 |
ref->y + 16 * (i + j * edged_width), |
ref->y + 16 * (i + j * edged_width), |
159 |
edged_width); |
edged_width); |
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8, |
|
|
ref->y + 16 * (i + j * edged_width) + 8, |
|
|
edged_width); |
|
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * edged_width, |
|
|
ref->y + 16 * (i + j * edged_width) + 8 * edged_width, |
|
|
edged_width); |
|
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), |
|
|
ref->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), |
|
|
edged_width); |
|
160 |
|
|
161 |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
162 |
ref->u + 8 * (i + j * edged_width/2), |
ref->u + 8 * (i + j * edged_width/2), |
166 |
edged_width / 2); |
edged_width / 2); |
167 |
return; |
return; |
168 |
} |
} |
169 |
|
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
170 |
|
|
171 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
if (reduced_resolution) |
172 |
int32_t dx = mb->mvs[0].x; |
{ |
173 |
int32_t dy = mb->mvs[0].y; |
dx = RRV_MV_SCALEUP(dx); |
174 |
|
dy = RRV_MV_SCALEUP(dy); |
|
if(quarterpel) { |
|
|
dx = mb->qmvs[0].x; |
|
|
dy = mb->qmvs[0].y; |
|
175 |
} |
} |
176 |
|
|
177 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
178 |
refv->y, refhv->y, 16 * i, 16 * j, dx, |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
179 |
dy, edged_width, quarterpel, rounding); |
edged_width, quarterpel, reduced_resolution, rounding); |
180 |
|
|
181 |
if (quarterpel) |
if (quarterpel) |
182 |
{ |
{ |
187 |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
188 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
189 |
|
|
190 |
/* uv-block-based compensation */ |
/* uv-block-based compensation |
191 |
|
XXX: rrv doesnt work with 16x16 block-based intepolation. |
192 |
|
we need to use 18x18-block interpolation */ |
193 |
|
if (reduced_resolution) |
194 |
|
{ |
195 |
|
const stride = edged_width/2; |
196 |
|
uint8_t * current, * reference; |
197 |
|
|
198 |
|
current = cur->u + 16*j*stride + 16*i; |
199 |
|
reference = refv->u + 16*j*stride + 16*i; |
200 |
|
interpolate16x16_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
201 |
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
202 |
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
203 |
|
transfer16x16_copy(current, reference, stride); |
204 |
|
|
205 |
|
current = cur->v + 16*j*stride + 16*i; |
206 |
|
reference = refv->v + 16*j*stride + 16*i; |
207 |
|
interpolate16x16_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
208 |
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
209 |
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
210 |
|
transfer16x16_copy(current, reference, stride); |
211 |
|
}else{ |
212 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
213 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
214 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
220 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
221 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
222 |
edged_width / 2); |
edged_width / 2); |
223 |
|
} |
224 |
|
|
225 |
} else { // mode == MODE_INTER4V |
} else { // mode == MODE_INTER4V |
226 |
|
int i; |
227 |
int32_t sum, dx, dy; |
int32_t sum, dx, dy; |
228 |
VECTOR *mvs; |
VECTOR mvs[4]; |
229 |
|
|
230 |
if(quarterpel) |
if(quarterpel) |
231 |
mvs = mb->qmvs; |
for (i = 0; i < 4; i++) mvs[i] = mb->qmvs[i]; |
232 |
else |
else |
233 |
mvs = mb->mvs; |
for (i = 0; i < 4; i++) mvs[i] = mb->mvs[i]; |
234 |
|
|
235 |
|
if (reduced_resolution) |
236 |
|
{ |
237 |
|
for (i = 0; i < 4; i++) |
238 |
|
{ |
239 |
|
mvs[i].x = RRV_MV_SCALEUP(mvs[i].x); |
240 |
|
mvs[i].y = RRV_MV_SCALEUP(mvs[i].y); |
241 |
|
} |
242 |
|
} |
243 |
|
|
244 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
245 |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
246 |
mvs[0].y, edged_width, quarterpel, rounding); |
mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); |
247 |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
248 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
249 |
mvs[1].x, mvs[1].y, edged_width, quarterpel, rounding); |
mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); |
250 |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
251 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
252 |
mvs[2].x, mvs[2].y, edged_width, quarterpel, rounding); |
mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); |
253 |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
254 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
255 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
256 |
|
|
257 |
if(quarterpel) |
if(quarterpel) |
258 |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
268 |
|
|
269 |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
270 |
|
|
271 |
/* uv-block-based compensation */ |
|
272 |
|
/* uv-block-based compensation |
273 |
|
XXX: rrv doesnt work with 16x16 block-based intepolation. |
274 |
|
we need to use 18x18-block interpolation */ |
275 |
|
if (reduced_resolution) |
276 |
|
{ |
277 |
|
const stride = edged_width/2; |
278 |
|
uint8_t * current, * reference; |
279 |
|
|
280 |
|
current = cur->u + 16*j*stride + 16*i; |
281 |
|
reference = refv->u + 16*j*stride + 16*i; |
282 |
|
interpolate16x16_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
283 |
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
284 |
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
285 |
|
transfer16x16_copy(current, reference, stride); |
286 |
|
|
287 |
|
current = cur->v + 16*j*stride + 16*i; |
288 |
|
reference = refv->v + 16*j*stride + 16*i; |
289 |
|
interpolate16x16_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
290 |
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
291 |
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
292 |
|
transfer16x16_copy(current, reference, stride); |
293 |
|
|
294 |
|
}else{ |
295 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
296 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
297 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
302 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
303 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
304 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
|
|
|
305 |
edged_width / 2); |
edged_width / 2); |
306 |
} |
} |
307 |
} |
} |
308 |
|
} |
309 |
|
|
310 |
|
|
311 |
void |
void |
333 |
int k,sum; |
int k,sum; |
334 |
int x = i; |
int x = i; |
335 |
int y = j; |
int y = j; |
336 |
|
uint32_t quarterpel = pParam->m_quarterpel; |
337 |
|
|
338 |
switch (mb->mode) { |
switch (mb->mode) { |
339 |
case MODE_FORWARD: |
case MODE_FORWARD: |
340 |
|
|
341 |
|
if (quarterpel) { |
342 |
|
dx = mb->qmvs[0].x; |
343 |
|
dy = mb->qmvs[0].y; |
344 |
|
} else { |
345 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
346 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
347 |
|
} |
348 |
|
|
349 |
transfer_8to16sub(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
350 |
cur->y + (j * 16) * edged_width + (i * 16), |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
351 |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
|
i * 16, j * 16, 1, dx, dy, edged_width), |
|
|
edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[1 * 64], |
|
|
cur->y + (j * 16) * edged_width + (i * 16 + 8), |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
i * 16 + 8, j * 16, 1, dx, dy, edged_width), |
|
|
edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[2 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
i * 16, j * 16 + 8, 1, dx, dy, edged_width), |
|
|
edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[3 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
i * 16 + 8, j * 16 + 8, 1, dx, dy, edged_width), |
|
|
edged_width); |
|
352 |
|
|
353 |
|
if (quarterpel) { |
354 |
|
dx /= 2; |
355 |
|
dy /= 2; |
356 |
|
} |
357 |
|
|
358 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
359 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
360 |
|
|
361 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
362 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
376 |
break; |
break; |
377 |
|
|
378 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
379 |
|
if (quarterpel) { |
380 |
|
b_dx = mb->b_qmvs[0].x; |
381 |
|
b_dy = mb->b_qmvs[0].y; |
382 |
|
} else { |
383 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
384 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
385 |
|
} |
386 |
|
|
387 |
transfer_8to16sub(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
388 |
cur->y + (j * 16) * edged_width + (i * 16), |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
389 |
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
b_dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
|
i * 16, j * 16, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[1 * 64], |
|
|
cur->y + (j * 16) * edged_width + (i * 16 + 8), |
|
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
i * 16 + 8, j * 16, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
390 |
|
|
391 |
transfer_8to16sub(&dct_codes[2 * 64], |
if (quarterpel) { |
392 |
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
b_dx /= 2; |
393 |
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
b_dy /= 2; |
394 |
i * 16, j * 16 + 8, 1, b_dx, b_dy, |
} |
|
edged_width), edged_width); |
|
395 |
|
|
396 |
transfer_8to16sub(&dct_codes[3 * 64], |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
397 |
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
i * 16 + 8, j * 16 + 8, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
398 |
|
|
|
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
|
|
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
|
399 |
|
|
400 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
401 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
414 |
|
|
415 |
break; |
break; |
416 |
|
|
|
|
|
417 |
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
418 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
419 |
|
|
420 |
|
if (quarterpel) { |
421 |
|
dx = mb->qmvs[0].x; |
422 |
|
dy = mb->qmvs[0].y; |
423 |
|
b_dx = mb->b_qmvs[0].x; |
424 |
|
b_dy = mb->b_qmvs[0].y; |
425 |
|
|
426 |
|
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refh->y, |
427 |
|
(uint8_t *) f_refh->y + 64, (uint8_t *) f_refhv->y, 16*i, 16*j, dx, dy, edged_width, 0); |
428 |
|
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) b_refh->y, |
429 |
|
(uint8_t *) b_refh->y + 64, (uint8_t *) b_refhv->y, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
430 |
|
|
431 |
|
for (k = 0; k < 4; k++) { |
432 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
433 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
434 |
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
435 |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
436 |
|
edged_width); |
437 |
|
} |
438 |
|
b_dx /= 2; |
439 |
|
b_dy /= 2; |
440 |
|
dx /= 2; |
441 |
|
dy /= 2; |
442 |
|
|
443 |
|
} else { |
444 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
445 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
|
|
|
446 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
447 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
448 |
|
|
458 |
edged_width); |
edged_width); |
459 |
} |
} |
460 |
|
|
461 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
} |
462 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
463 |
|
|
464 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
465 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
466 |
|
|
467 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
468 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
469 |
|
|
470 |
transfer_8to16sub2(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
471 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
486 |
break; |
break; |
487 |
|
|
488 |
case MODE_DIRECT: |
case MODE_DIRECT: |
489 |
|
if (quarterpel) { |
490 |
|
for (k=0;k<4;k++) { |
491 |
|
|
492 |
for (k=0;k<4;k++) |
dx = mb->qmvs[k].x; |
493 |
{ |
dy = mb->qmvs[k].y; |
494 |
|
b_dx = mb->b_qmvs[k].x; |
495 |
|
b_dy = mb->b_qmvs[k].y; |
496 |
|
|
497 |
|
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
498 |
|
(uint8_t *) f_ref->y, |
499 |
|
(uint8_t *) f_refh->y, |
500 |
|
(uint8_t *) f_refh->y + 64, |
501 |
|
(uint8_t *) f_refhv->y, |
502 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
503 |
|
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
504 |
|
(uint8_t *) b_ref->y, |
505 |
|
(uint8_t *) b_refh->y, |
506 |
|
(uint8_t *) b_refh->y + 64, |
507 |
|
(uint8_t *) b_refhv->y, |
508 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
509 |
|
|
510 |
|
|
511 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
512 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
513 |
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
514 |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
515 |
|
edged_width); |
516 |
|
} |
517 |
|
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
518 |
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
519 |
|
sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; |
520 |
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
521 |
|
|
522 |
|
sum = mb->b_qmvs[0].y/2 + mb->b_qmvs[1].y/2 + mb->b_qmvs[2].y/2 + mb->b_qmvs[3].y/2; |
523 |
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
524 |
|
sum = mb->b_qmvs[0].x/2 + mb->b_qmvs[1].x/2 + mb->b_qmvs[2].x/2 + mb->b_qmvs[3].x/2; |
525 |
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
526 |
|
|
527 |
|
} else { |
528 |
|
for (k=0;k<4;k++) { |
529 |
dx = mb->mvs[k].x; |
dx = mb->mvs[k].x; |
530 |
dy = mb->mvs[k].y; |
dy = mb->mvs[k].y; |
531 |
|
|
532 |
b_dx = mb->b_mvs[k].x; |
b_dx = mb->b_mvs[k].x; |
533 |
b_dy = mb->b_mvs[k].y; |
b_dy = mb->b_mvs[k].y; |
534 |
|
|
|
// fprintf(stderr,"Direct Vector %d -- %d:%d %d:%d\n",k,dx,dy,b_dx,b_dy); |
|
|
|
|
535 |
transfer_8to16sub2(&dct_codes[k * 64], |
transfer_8to16sub2(&dct_codes[k * 64], |
536 |
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, |
537 |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
549 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
550 |
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
551 |
|
|
|
|
|
552 |
sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; |
sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; |
553 |
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
554 |
|
|
555 |
sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; |
sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; |
556 |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
557 |
|
|
558 |
/* // for QPel don't forget to always do |
} |
|
|
|
|
if (quarterpel) |
|
|
sum /= 2; |
|
|
*/ |
|
559 |
transfer_8to16sub2(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
560 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
561 |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |