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/qpel.h" |
#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 |
|
|
134 |
|
|
135 |
|
/* XXX: slow, inelegant... */ |
136 |
|
static void |
137 |
|
interpolate18x18_switch(uint8_t * const cur, |
138 |
|
const uint8_t * const refn, |
139 |
|
const uint32_t x, |
140 |
|
const uint32_t y, |
141 |
|
const int32_t dx, |
142 |
|
const int dy, |
143 |
|
const uint32_t stride, |
144 |
|
const uint32_t rounding) |
145 |
|
{ |
146 |
|
interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); |
147 |
|
interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); |
148 |
|
interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); |
149 |
|
|
150 |
|
interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); |
151 |
|
interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); |
152 |
|
interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); |
153 |
|
|
154 |
|
interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); |
155 |
|
interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); |
156 |
|
interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); |
157 |
|
} |
158 |
|
|
159 |
|
|
160 |
void |
void |
161 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
171 |
const uint32_t height, |
const uint32_t height, |
172 |
const uint32_t edged_width, |
const uint32_t edged_width, |
173 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
174 |
|
const int reduced_resolution, |
175 |
const uint32_t rounding) |
const uint32_t rounding) |
176 |
{ |
{ |
177 |
if (mb->mode == MODE_NOT_CODED) { |
|
178 |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
179 |
|
|
180 |
|
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); |
182 |
|
|
183 |
|
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
184 |
|
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
185 |
ref->y + 16 * (i + j * edged_width), |
ref->y + 16 * (i + j * edged_width), |
186 |
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); |
|
187 |
|
|
188 |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
189 |
ref->u + 8 * (i + j * edged_width/2), |
ref->u + 8 * (i + j * edged_width/2), |
193 |
edged_width / 2); |
edged_width / 2); |
194 |
return; |
return; |
195 |
} |
} |
196 |
|
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
197 |
|
|
198 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
if (reduced_resolution) |
199 |
int32_t dx = mb->mvs[0].x; |
{ |
200 |
int32_t dy = mb->mvs[0].y; |
dx = RRV_MV_SCALEUP(dx); |
201 |
|
dy = RRV_MV_SCALEUP(dy); |
|
if(quarterpel) { |
|
|
dx = mb->qmvs[0].x; |
|
|
dy = mb->qmvs[0].y; |
|
202 |
} |
} |
203 |
|
|
204 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
205 |
refv->y, refhv->y, 16 * i, 16 * j, dx, |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
206 |
dy, edged_width, quarterpel, rounding); |
edged_width, quarterpel, reduced_resolution, rounding); |
207 |
|
|
208 |
if (quarterpel) |
if (quarterpel) |
209 |
{ |
{ |
210 |
dx = (dx >> 2) + roundtab_78[dx & 0x7]; |
dx /= 2; |
211 |
dy = (dy >> 2) + roundtab_78[dy & 0x7]; |
dy /= 2; |
212 |
} |
} |
213 |
else { |
|
214 |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
215 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
|
} |
|
216 |
|
|
217 |
/* uv-block-based compensation */ |
/* uv-block-based compensation |
218 |
|
XXX: rrv doesnt work with 16x16 block-based intepolation. |
219 |
|
we need to use 18x18-block interpolation */ |
220 |
|
if (reduced_resolution) |
221 |
|
{ |
222 |
|
const stride = edged_width/2; |
223 |
|
uint8_t * current, * reference; |
224 |
|
|
225 |
|
current = cur->u + 16*j*stride + 16*i; |
226 |
|
reference = refv->u + 16*j*stride + 16*i; |
227 |
|
//interpolate16x16_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
228 |
|
interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
229 |
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
230 |
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
231 |
|
transfer16x16_copy(current, reference, stride); |
232 |
|
|
233 |
|
current = cur->v + 16*j*stride + 16*i; |
234 |
|
reference = refv->v + 16*j*stride + 16*i; |
235 |
|
//interpolate16x16_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
236 |
|
interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
237 |
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
238 |
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
239 |
|
transfer16x16_copy(current, reference, stride); |
240 |
|
}else{ |
241 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
242 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
243 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
249 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
250 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
251 |
edged_width / 2); |
edged_width / 2); |
252 |
|
} |
253 |
|
|
254 |
} else { // mode == MODE_INTER4V |
} else { // mode == MODE_INTER4V |
255 |
|
int i; |
256 |
int32_t sum, dx, dy; |
int32_t sum, dx, dy; |
257 |
VECTOR *mvs; |
VECTOR mvs[4]; |
258 |
|
|
259 |
if(quarterpel) |
if(quarterpel) |
260 |
mvs = mb->qmvs; |
for (i = 0; i < 4; i++) mvs[i] = mb->qmvs[i]; |
261 |
else |
else |
262 |
mvs = mb->mvs; |
for (i = 0; i < 4; i++) mvs[i] = mb->mvs[i]; |
263 |
|
|
264 |
|
if (reduced_resolution) |
265 |
|
{ |
266 |
|
for (i = 0; i < 4; i++) |
267 |
|
{ |
268 |
|
mvs[i].x = RRV_MV_SCALEUP(mvs[i].x); |
269 |
|
mvs[i].y = RRV_MV_SCALEUP(mvs[i].y); |
270 |
|
} |
271 |
|
} |
272 |
|
|
273 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
274 |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
275 |
mvs[0].y, edged_width, quarterpel, rounding); |
mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); |
276 |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
277 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
278 |
mvs[1].x, mvs[1].y, edged_width, quarterpel, rounding); |
mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); |
279 |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
280 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
281 |
mvs[2].x, mvs[2].y, edged_width, quarterpel, rounding); |
mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); |
282 |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
283 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
284 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
285 |
|
|
286 |
if(quarterpel) |
if(quarterpel) |
287 |
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); |
297 |
|
|
298 |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
299 |
|
|
300 |
/* uv-block-based compensation */ |
|
301 |
|
/* uv-block-based compensation |
302 |
|
XXX: rrv doesnt work with 16x16 block-based intepolation. |
303 |
|
we need to use 18x18-block interpolation */ |
304 |
|
if (reduced_resolution) |
305 |
|
{ |
306 |
|
const stride = edged_width/2; |
307 |
|
uint8_t * current, * reference; |
308 |
|
|
309 |
|
current = cur->u + 16*j*stride + 16*i; |
310 |
|
reference = refv->u + 16*j*stride + 16*i; |
311 |
|
interpolate16x16_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
312 |
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
313 |
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
314 |
|
transfer16x16_copy(current, reference, stride); |
315 |
|
|
316 |
|
current = cur->v + 16*j*stride + 16*i; |
317 |
|
reference = refv->v + 16*j*stride + 16*i; |
318 |
|
interpolate16x16_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
319 |
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
320 |
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
321 |
|
transfer16x16_copy(current, reference, stride); |
322 |
|
|
323 |
|
}else{ |
324 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
325 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
326 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
331 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
332 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
333 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
|
|
|
334 |
edged_width / 2); |
edged_width / 2); |
335 |
} |
} |
336 |
} |
} |
337 |
|
} |
338 |
|
|
339 |
|
|
340 |
void |
void |
362 |
int k,sum; |
int k,sum; |
363 |
int x = i; |
int x = i; |
364 |
int y = j; |
int y = j; |
365 |
|
uint32_t quarterpel = pParam->m_quarterpel; |
366 |
|
|
367 |
switch (mb->mode) { |
switch (mb->mode) { |
368 |
case MODE_FORWARD: |
case MODE_FORWARD: |
369 |
|
|
370 |
|
if (quarterpel) { |
371 |
|
dx = mb->qmvs[0].x; |
372 |
|
dy = mb->qmvs[0].y; |
373 |
|
} else { |
374 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
375 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
376 |
|
} |
377 |
|
|
378 |
transfer_8to16sub(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
379 |
cur->y + (j * 16) * edged_width + (i * 16), |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
380 |
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); |
|
381 |
|
|
382 |
|
if (quarterpel) { |
383 |
|
dx /= 2; |
384 |
|
dy /= 2; |
385 |
|
} |
386 |
|
|
387 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
388 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
389 |
|
|
390 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
391 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
405 |
break; |
break; |
406 |
|
|
407 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
408 |
|
if (quarterpel) { |
409 |
|
b_dx = mb->b_qmvs[0].x; |
410 |
|
b_dy = mb->b_qmvs[0].y; |
411 |
|
} else { |
412 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
413 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
414 |
|
} |
415 |
|
|
416 |
transfer_8to16sub(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
417 |
cur->y + (j * 16) * edged_width + (i * 16), |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
418 |
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); |
|
419 |
|
|
420 |
transfer_8to16sub(&dct_codes[1 * 64], |
if (quarterpel) { |
421 |
cur->y + (j * 16) * edged_width + (i * 16 + 8), |
b_dx /= 2; |
422 |
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
b_dy /= 2; |
423 |
i * 16 + 8, j * 16, 1, b_dx, b_dy, |
} |
|
edged_width), edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[2 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
|
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
i * 16, j * 16 + 8, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
424 |
|
|
425 |
transfer_8to16sub(&dct_codes[3 * 64], |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
426 |
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); |
|
427 |
|
|
|
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
|
|
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
|
428 |
|
|
429 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
430 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
443 |
|
|
444 |
break; |
break; |
445 |
|
|
|
|
|
446 |
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) */ |
447 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
448 |
|
|
449 |
|
if (quarterpel) { |
450 |
|
dx = mb->qmvs[0].x; |
451 |
|
dy = mb->qmvs[0].y; |
452 |
|
b_dx = mb->b_qmvs[0].x; |
453 |
|
b_dy = mb->b_qmvs[0].y; |
454 |
|
|
455 |
|
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refh->y, |
456 |
|
(uint8_t *) f_refh->y + 64, (uint8_t *) f_refhv->y, 16*i, 16*j, dx, dy, edged_width, 0); |
457 |
|
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) b_refh->y, |
458 |
|
(uint8_t *) b_refh->y + 64, (uint8_t *) b_refhv->y, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
459 |
|
|
460 |
|
for (k = 0; k < 4; k++) { |
461 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
462 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
463 |
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
464 |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
465 |
|
edged_width); |
466 |
|
} |
467 |
|
b_dx /= 2; |
468 |
|
b_dy /= 2; |
469 |
|
dx /= 2; |
470 |
|
dy /= 2; |
471 |
|
|
472 |
|
} else { |
473 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
474 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
|
|
|
475 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
476 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
477 |
|
|
487 |
edged_width); |
edged_width); |
488 |
} |
} |
489 |
|
|
490 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
} |
491 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
492 |
|
|
493 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
494 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
495 |
|
|
496 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
497 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
498 |
|
|
499 |
transfer_8to16sub2(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
500 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
515 |
break; |
break; |
516 |
|
|
517 |
case MODE_DIRECT: |
case MODE_DIRECT: |
518 |
|
if (quarterpel) { |
519 |
|
for (k=0;k<4;k++) { |
520 |
|
|
521 |
for (k=0;k<4;k++) |
dx = mb->qmvs[k].x; |
522 |
{ |
dy = mb->qmvs[k].y; |
523 |
|
b_dx = mb->b_qmvs[k].x; |
524 |
|
b_dy = mb->b_qmvs[k].y; |
525 |
|
|
526 |
|
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
527 |
|
(uint8_t *) f_ref->y, |
528 |
|
(uint8_t *) f_refh->y, |
529 |
|
(uint8_t *) f_refh->y + 64, |
530 |
|
(uint8_t *) f_refhv->y, |
531 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
532 |
|
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
533 |
|
(uint8_t *) b_ref->y, |
534 |
|
(uint8_t *) b_refh->y, |
535 |
|
(uint8_t *) b_refh->y + 64, |
536 |
|
(uint8_t *) b_refhv->y, |
537 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
538 |
|
|
539 |
|
|
540 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
541 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
542 |
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
543 |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
544 |
|
edged_width); |
545 |
|
} |
546 |
|
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
547 |
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
548 |
|
sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; |
549 |
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
550 |
|
|
551 |
|
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; |
552 |
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
553 |
|
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; |
554 |
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
555 |
|
|
556 |
|
} else { |
557 |
|
for (k=0;k<4;k++) { |
558 |
dx = mb->mvs[k].x; |
dx = mb->mvs[k].x; |
559 |
dy = mb->mvs[k].y; |
dy = mb->mvs[k].y; |
560 |
|
|
561 |
b_dx = mb->b_mvs[k].x; |
b_dx = mb->b_mvs[k].x; |
562 |
b_dy = mb->b_mvs[k].y; |
b_dy = mb->b_mvs[k].y; |
563 |
|
|
|
// fprintf(stderr,"Direct Vector %d -- %d:%d %d:%d\n",k,dx,dy,b_dx,b_dy); |
|
|
|
|
564 |
transfer_8to16sub2(&dct_codes[k * 64], |
transfer_8to16sub2(&dct_codes[k * 64], |
565 |
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, |
566 |
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, |
578 |
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; |
579 |
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)); |
580 |
|
|
|
|
|
581 |
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; |
582 |
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)); |
583 |
|
|
584 |
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; |
585 |
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)); |
586 |
|
|
587 |
/* // for QPel don't forget to always do |
} |
|
|
|
|
if (quarterpel) |
|
|
sum /= 2; |
|
|
*/ |
|
588 |
transfer_8to16sub2(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
589 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
590 |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |