1 |
|
// 30.10.2002 corrected qpel chroma rounding |
2 |
|
// 04.10.2002 added qpel support to MBMotionCompensation |
3 |
// 01.05.2002 updated MBMotionCompensationBVOP |
// 01.05.2002 updated MBMotionCompensationBVOP |
4 |
// 14.04.2002 bframe compensation |
// 14.04.2002 bframe compensation |
5 |
|
|
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 |
compensate8x8_halfpel(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 int dy, |
const int32_t dy, |
28 |
const uint32_t stride) |
const uint32_t stride, |
29 |
|
const uint32_t quarterpel, |
30 |
|
const int reduced_resolution, |
31 |
|
const uint32_t rounding) |
32 |
{ |
{ |
|
int32_t ddx, ddy; |
|
33 |
|
|
34 |
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
if (reduced_resolution) |
35 |
{ |
{ |
36 |
case 0: |
const uint8_t * reference; |
37 |
ddx = dx / 2; |
x*=2; y*=2; |
|
ddy = dy / 2; |
|
|
transfer_8to16sub(dct_codes, cur + y * stride + x, |
|
|
ref + (int) ((y + ddy) * stride + x + ddx), stride); |
|
|
break; |
|
38 |
|
|
39 |
case 1: |
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
40 |
ddx = dx / 2; |
|
41 |
ddy = (dy - 1) / 2; |
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
42 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
43 |
refv + (int) ((y + ddy) * stride + x + ddx), stride); |
|
44 |
break; |
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) { |
57 |
|
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); |
59 |
|
|
|
case 2: |
|
|
ddx = (dx - 1) / 2; |
|
|
ddy = dy / 2; |
|
60 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
61 |
refh + (int) ((y + ddy) * stride + x + ddx), stride); |
refv + y*stride + x, stride); |
62 |
break; |
transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
63 |
|
refv + y*stride + x + 8, stride); |
64 |
|
transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
65 |
|
refv + y*stride + x + 8*stride, stride); |
66 |
|
transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
67 |
|
refv + y*stride + x + 8*stride+8, stride); |
68 |
|
|
69 |
|
} |
70 |
|
else |
71 |
|
{ |
72 |
|
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
73 |
|
|
|
default: // case 3: |
|
|
ddx = (dx - 1) / 2; |
|
|
ddy = (dy - 1) / 2; |
|
74 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
75 |
refhv + (int) ((y + ddy) * stride + x + ddx), stride); |
reference, stride); |
76 |
break; |
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
77 |
|
reference + 8, stride); |
78 |
|
transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
79 |
|
reference + 8*stride, stride); |
80 |
|
transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
81 |
|
reference + 8*stride + 8, stride); |
82 |
} |
} |
83 |
} |
} |
84 |
|
} |
85 |
|
|
86 |
|
static __inline void |
87 |
|
compensate8x8_interpolate(int16_t * const dct_codes, |
88 |
|
uint8_t * const cur, |
89 |
|
const uint8_t * const ref, |
90 |
|
const uint8_t * const refh, |
91 |
|
const uint8_t * const refv, |
92 |
|
const uint8_t * const refhv, |
93 |
|
uint32_t x, |
94 |
|
uint32_t y, |
95 |
|
const int32_t dx, |
96 |
|
const int32_t dy, |
97 |
|
const uint32_t stride, |
98 |
|
const uint32_t quarterpel, |
99 |
|
const int reduced_resolution, |
100 |
|
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) { |
117 |
|
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); |
119 |
|
|
120 |
|
transfer_8to16sub(dct_codes, cur + y*stride + x, |
121 |
|
refv + y*stride + x, stride); |
122 |
|
} |
123 |
|
else |
124 |
|
{ |
125 |
|
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
126 |
|
|
127 |
|
transfer_8to16sub(dct_codes, cur + y * stride + x, |
128 |
|
reference, stride); |
129 |
|
} |
130 |
|
} |
131 |
|
} |
132 |
|
|
133 |
void |
void |
134 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
143 |
const uint32_t width, |
const uint32_t width, |
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, |
147 |
|
const int reduced_resolution, |
148 |
const uint32_t rounding) |
const uint32_t rounding) |
149 |
{ |
{ |
|
static const uint32_t roundtab[16] = |
|
|
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
|
150 |
|
|
151 |
|
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
152 |
|
|
153 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
154 |
int32_t dx = mb->mvs[0].x; |
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
|
int32_t dy = mb->mvs[0].y; |
|
155 |
|
|
156 |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
157 |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
158 |
|
ref->y + 16 * (i + j * edged_width), |
159 |
edged_width); |
edged_width); |
160 |
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
|
161 |
refv->y, refhv->y, 16 * i + 8, 16 * j, dx, dy, |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
162 |
edged_width); |
ref->u + 8 * (i + j * edged_width/2), |
|
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i, 16 * j + 8, dx, dy, |
|
|
edged_width); |
|
|
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, dx, |
|
|
dy, edged_width); |
|
|
|
|
|
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
|
|
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
|
|
|
|
/* uv-image-based compensation */ |
|
|
/* Always do block-based compensation, until check for HALFPEL is possible |
|
|
#ifdef BFRAMES |
|
|
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, ref->u, refh->u, |
|
|
refv->u, refhv->u, 8 * i, 8 * j, dx, dy, |
|
163 |
edged_width / 2); |
edged_width / 2); |
164 |
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, ref->v, refh->v, |
transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), |
165 |
refv->v, refhv->v, 8 * i, 8 * j, dx, dy, |
ref->v + 8 * (i + j * edged_width/2), |
166 |
edged_width / 2); |
edged_width / 2); |
167 |
#else |
return; |
168 |
*/ |
} |
169 |
/* uv-block-based compensation */ |
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
170 |
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
|
171 |
edged_width / 2, rounding); |
if (reduced_resolution) |
172 |
|
{ |
173 |
|
dx = RRV_MV_SCALEUP(dx); |
174 |
|
dy = RRV_MV_SCALEUP(dy); |
175 |
|
} |
176 |
|
|
177 |
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
178 |
|
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
179 |
|
edged_width, quarterpel, reduced_resolution, rounding); |
180 |
|
|
181 |
|
if (quarterpel) |
182 |
|
{ |
183 |
|
dx /= 2; |
184 |
|
dy /= 2; |
185 |
|
} |
186 |
|
|
187 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
188 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
189 |
|
|
190 |
|
/* 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 |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
215 |
|
dx, dy, edged_width / 2, rounding), |
216 |
edged_width / 2); |
edged_width / 2); |
217 |
|
|
|
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, rounding); |
|
218 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
219 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
220 |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
221 |
|
dx, dy, edged_width / 2, rounding), |
222 |
edged_width / 2); |
edged_width / 2); |
223 |
/* |
} |
224 |
#endif |
|
225 |
*/ |
} else { // mode == MODE_INTER4V |
226 |
} else // mode == MODE_INTER4V |
int i; |
|
{ |
|
227 |
int32_t sum, dx, dy; |
int32_t sum, dx, dy; |
228 |
|
VECTOR mvs[4]; |
229 |
|
|
230 |
|
if(quarterpel) |
231 |
|
for (i = 0; i < 4; i++) mvs[i] = mb->qmvs[i]; |
232 |
|
else |
233 |
|
for (i = 0; i < 4; i++) mvs[i] = mb->mvs[i]; |
234 |
|
|
235 |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
if (reduced_resolution) |
236 |
refv->y, refhv->y, 16 * i, 16 * j, mb->mvs[0].x, |
{ |
237 |
mb->mvs[0].y, edged_width); |
for (i = 0; i < 4; i++) |
238 |
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
{ |
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, |
245 |
|
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
246 |
|
mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); |
247 |
|
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 |
mb->mvs[1].x, mb->mvs[1].y, edged_width); |
mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); |
250 |
compensate8x8_halfpel(&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 |
mb->mvs[2].x, mb->mvs[2].y, edged_width); |
mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); |
253 |
compensate8x8_halfpel(&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 |
mb->mvs[3].x, mb->mvs[3].y, edged_width); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
256 |
|
|
257 |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
if(quarterpel) |
258 |
dx = (sum ? SIGN(sum) * |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
259 |
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
else |
260 |
|
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
261 |
|
|
262 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
dy = (sum ? SIGN(sum) * |
|
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
263 |
|
|
264 |
/* uv-image-based compensation */ |
if(quarterpel) |
265 |
#ifdef BFRAMES |
sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
266 |
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, ref->u, refh->u, |
else |
267 |
refv->u, refhv->u, 8 * i, 8 * j, dx, dy, |
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
268 |
edged_width / 2); |
|
269 |
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, ref->v, refh->v, |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
270 |
refv->v, refhv->v, 8 * i, 8 * j, dx, dy, |
|
271 |
edged_width / 2); |
|
272 |
#else |
/* uv-block-based compensation |
273 |
/* uv-block-based compensation */ |
XXX: rrv doesnt work with 16x16 block-based intepolation. |
274 |
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
we need to use 18x18-block interpolation */ |
275 |
edged_width / 2, rounding); |
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 |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
298 |
|
dx, dy, edged_width / 2, rounding), |
299 |
edged_width / 2); |
edged_width / 2); |
300 |
|
|
|
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, rounding); |
|
301 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
302 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
303 |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
304 |
|
dx, dy, edged_width / 2, rounding), |
305 |
edged_width / 2); |
edged_width / 2); |
306 |
#endif |
} |
307 |
} |
} |
308 |
} |
} |
309 |
|
|
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_c(&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); |
352 |
i * 16, j * 16, 1, dx, dy, edged_width), |
|
353 |
edged_width); |
if (quarterpel) { |
354 |
|
dx /= 2; |
355 |
transfer_8to16sub(&dct_codes[1 * 64], |
dy /= 2; |
356 |
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); |
|
357 |
|
|
358 |
transfer_8to16sub_c(&dct_codes[2 * 64], |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
359 |
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
|
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); |
|
360 |
|
|
361 |
transfer_8to16sub(&dct_codes[3 * 64], |
/* uv-block-based compensation */ |
362 |
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
transfer_8to16sub(&dct_codes[4 * 64], |
363 |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
364 |
i * 16 + 8, j * 16 + 8, 1, dx, dy, |
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
365 |
edged_width), edged_width); |
dx, dy, edged_width / 2, 0), |
366 |
|
|
367 |
|
edged_width / 2); |
368 |
|
|
369 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
transfer_8to16sub(&dct_codes[5 * 64], |
370 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
371 |
|
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
372 |
|
dx, dy, edged_width / 2, 0), |
373 |
|
|
|
/* uv-image-based compensation */ |
|
|
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, f_ref->u, f_refh->u, |
|
|
f_refv->u, f_refhv->u, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2); |
|
|
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, f_ref->v, f_refh->v, |
|
|
f_refv->v, f_refhv->v, 8 * i, 8 * j, dx, dy, |
|
374 |
edged_width / 2); |
edged_width / 2); |
375 |
|
|
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_c(&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); |
390 |
i * 16, j * 16, 1, b_dx, b_dy, |
|
391 |
edged_width), edged_width); |
if (quarterpel) { |
392 |
|
b_dx /= 2; |
393 |
transfer_8to16sub(&dct_codes[1 * 64], |
b_dy /= 2; |
394 |
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); |
|
395 |
|
|
396 |
transfer_8to16sub_c(&dct_codes[2 * 64], |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
397 |
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
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, j * 16 + 8, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
398 |
|
|
|
transfer_8to16sub(&dct_codes[3 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
|
|
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); |
|
399 |
|
|
400 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
/* uv-block-based compensation */ |
401 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
transfer_8to16sub(&dct_codes[4 * 64], |
402 |
|
cur->u + 8 * j * edged_width / 2 + 8 * i, |
403 |
|
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
404 |
|
b_dx, b_dy, edged_width / 2, 0), |
405 |
|
|
|
/* uv-image-based compensation */ |
|
|
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, b_ref->u, b_refh->u, |
|
|
b_refv->u, b_refhv->u, 8 * i, 8 * j, b_dx, b_dy, |
|
406 |
edged_width / 2); |
edged_width / 2); |
407 |
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, b_ref->v, b_refh->v, |
|
408 |
b_refv->v, b_refhv->v, 8 * i, 8 * j, b_dx, b_dy, |
transfer_8to16sub(&dct_codes[5 * 64], |
409 |
|
cur->v + 8 * j * edged_width / 2 + 8 * i, |
410 |
|
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
411 |
|
b_dx, b_dy, edged_width / 2, 0), |
412 |
|
|
413 |
edged_width / 2); |
edged_width / 2); |
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: |
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 |
|
|
449 |
for (k=0;k<4;k++) |
for (k = 0; k < 4; k++) { |
450 |
{ |
transfer_8to16sub2(&dct_codes[k * 64], |
|
transfer_8to16sub2_c(&dct_codes[k * 64], |
|
451 |
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, |
452 |
get_ref(f_ref->y, f_refh->y, f_refv->y, |
get_ref(f_ref->y, f_refh->y, f_refv->y, |
453 |
f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
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 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
465 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
466 |
|
|
467 |
transfer_8to16sub2_c(&dct_codes[4 * 64], |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
468 |
|
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
469 |
|
|
470 |
|
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), |
472 |
get_ref(f_ref->u, f_refh->u, f_refv->u, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
473 |
f_refhv->u, i, j, 8, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
474 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, |
475 |
get_ref(b_ref->u, b_refh->u, b_refv->u, |
dx, dy, edged_width / 2, 0), |
|
b_refhv->u, i, j, 8, b_dx, b_dy, |
|
|
edged_width / 2), |
|
476 |
edged_width / 2); |
edged_width / 2); |
477 |
|
|
478 |
transfer_8to16sub2_c(&dct_codes[5 * 64], |
transfer_8to16sub2(&dct_codes[5 * 64], |
479 |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
480 |
get_ref(f_ref->v, f_refh->v, f_refv->v, |
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
481 |
f_refhv->v, 8 * i, 8 * j, 1, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
482 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, |
483 |
get_ref(b_ref->v, b_refh->v, b_refv->v, |
dx, dy, edged_width / 2, 0), |
|
b_refhv->v, 8 * i, 8 * j, 1, b_dx, b_dy, |
|
|
edged_width / 2), |
|
484 |
edged_width / 2); |
edged_width / 2); |
485 |
|
|
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 |
|
|
535 |
// fprintf(stderr,"Direct Vector %d -- %d:%d %d:%d\n",k,dx,dy,b_dx,b_dy); |
transfer_8to16sub2(&dct_codes[k * 64], |
|
|
|
|
transfer_8to16sub2_c(&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, |
538 |
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
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 |
} |
559 |
|
transfer_8to16sub2(&dct_codes[4 * 64], |
|
if (quarterpel) |
|
|
sum /= 2; |
|
|
*/ |
|
|
|
|
|
transfer_8to16sub2_c(&dct_codes[4 * 64], |
|
560 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
561 |
get_ref(f_ref->u, f_refh->u, f_refv->u, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
562 |
f_refhv->u, i, j, 8, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
563 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, dx, dy, |
564 |
get_ref(b_ref->u, b_refh->u, b_refv->u, |
edged_width / 2, 0), |
|
b_refhv->u, i, j, 8, b_dx, b_dy, |
|
|
edged_width / 2), |
|
565 |
edged_width / 2); |
edged_width / 2); |
566 |
|
|
567 |
transfer_8to16sub2_c(&dct_codes[5 * 64], |
transfer_8to16sub2(&dct_codes[5 * 64], |
568 |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
569 |
get_ref(f_ref->v, f_refh->v, f_refv->v, |
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
570 |
f_refhv->v, i, j, 8, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
571 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, |
572 |
get_ref(b_ref->v, b_refh->v, b_refv->v, |
dx, dy, edged_width / 2, 0), |
|
b_refhv->v, i, j, 8, b_dx, b_dy, |
|
|
edged_width / 2), |
|
573 |
edged_width / 2); |
edged_width / 2); |
574 |
|
|
575 |
|
|