1 |
|
// 01.05.2002 updated MBMotionCompensationBVOP |
2 |
|
// 14.04.2002 bframe compensation |
3 |
|
|
4 |
#include "../encoder.h" |
#include "../encoder.h" |
5 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
6 |
#include "../image/interpolate8x8.h" |
#include "../image/interpolate8x8.h" |
7 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
8 |
|
#include "motion.h" |
9 |
|
|
10 |
#define ABS(X) (((X)>0)?(X):-(X)) |
#define ABS(X) (((X)>0)?(X):-(X)) |
11 |
#define SIGN(X) (((X)>0)?1:-1) |
#define SIGN(X) (((X)>0)?1:-1) |
66 |
const IMAGE * const refv, |
const IMAGE * const refv, |
67 |
const IMAGE * const refhv, |
const IMAGE * const refhv, |
68 |
IMAGE * const cur, |
IMAGE * const cur, |
69 |
int16_t dct_codes[][64], |
int16_t *dct_codes, |
70 |
const uint32_t width, |
const uint32_t width, |
71 |
const uint32_t height, |
const uint32_t height, |
72 |
const uint32_t edged_width, |
const uint32_t edged_width, |
81 |
int32_t dx = mb->mvs[0].x; |
int32_t dx = mb->mvs[0].x; |
82 |
int32_t dy = mb->mvs[0].y; |
int32_t dy = mb->mvs[0].y; |
83 |
|
|
84 |
compensate8x8_halfpel(dct_codes[0], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[0*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
85 |
16*i, 16*j, dx, dy, edged_width); |
16*i, 16*j, dx, dy, edged_width); |
86 |
compensate8x8_halfpel(dct_codes[1], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[1*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
87 |
16*i + 8, 16*j, dx, dy, edged_width); |
16*i + 8, 16*j, dx, dy, edged_width); |
88 |
compensate8x8_halfpel(dct_codes[2], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[2*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
89 |
16*i, 16*j + 8, dx, dy, edged_width); |
16*i, 16*j + 8, dx, dy, edged_width); |
90 |
compensate8x8_halfpel(dct_codes[3], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[3*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
91 |
16*i + 8, 16*j + 8, dx, dy, edged_width); |
16*i + 8, 16*j + 8, dx, dy, edged_width); |
92 |
|
|
93 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
94 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
95 |
|
|
96 |
/* uv-image-based compensation |
/* uv-image-based compensation */ |
97 |
compensate8x8_halfpel(dct_codes[4], cur->u, ref->u, refh->u, refv->u, refhv->u, |
#ifdef BFRAMES |
98 |
|
compensate8x8_halfpel(&dct_codes[4*64], cur->u, ref->u, refh->u, refv->u, refhv->u, |
99 |
8*i, 8*j, dx, dy, edged_width/2); |
8*i, 8*j, dx, dy, edged_width/2); |
100 |
compensate8x8_halfpel(dct_codes[5], cur->v, ref->v, refh->v, refv->v, refhv->v, |
compensate8x8_halfpel(&dct_codes[5*64], cur->v, ref->v, refh->v, refv->v, refhv->v, |
101 |
8*i, 8*j, dx, dy, edged_width/2); */ |
8*i, 8*j, dx, dy, edged_width/2); |
102 |
|
#else |
103 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
104 |
interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
105 |
transfer_8to16sub(dct_codes[4], |
transfer_8to16sub(&dct_codes[4*64], |
106 |
cur->u + 8*j*edged_width/2 + 8*i, |
cur->u + 8*j*edged_width/2 + 8*i, |
107 |
refv->u + 8*j*edged_width/2 + 8*i, edged_width/2); |
refv->u + 8*j*edged_width/2 + 8*i, edged_width/2); |
108 |
|
|
109 |
interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
110 |
transfer_8to16sub(dct_codes[5], |
transfer_8to16sub(&dct_codes[5*64], |
111 |
cur->v + 8*j*edged_width/2 + 8*i, |
cur->v + 8*j*edged_width/2 + 8*i, |
112 |
refv->v + 8*j*edged_width/2 + 8*i, edged_width/2); |
refv->v + 8*j*edged_width/2 + 8*i, edged_width/2); |
113 |
|
#endif |
114 |
} |
} |
115 |
else // mode == MODE_INTER4V |
else // mode == MODE_INTER4V |
116 |
{ |
{ |
117 |
int32_t sum, dx, dy; |
int32_t sum, dx, dy; |
118 |
|
|
119 |
compensate8x8_halfpel(dct_codes[0], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[0*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
120 |
16*i, 16*j, mb->mvs[0].x, mb->mvs[0].y, edged_width); |
16*i, 16*j, mb->mvs[0].x, mb->mvs[0].y, edged_width); |
121 |
compensate8x8_halfpel(dct_codes[1], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[1*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
122 |
16*i + 8, 16*j, mb->mvs[1].x, mb->mvs[1].y, edged_width); |
16*i + 8, 16*j, mb->mvs[1].x, mb->mvs[1].y, edged_width); |
123 |
compensate8x8_halfpel(dct_codes[2], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[2*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
124 |
16*i, 16*j + 8, mb->mvs[2].x, mb->mvs[2].y, edged_width); |
16*i, 16*j + 8, mb->mvs[2].x, mb->mvs[2].y, edged_width); |
125 |
compensate8x8_halfpel(dct_codes[3], cur->y, ref->y, refh->y, refv->y, refhv->y, |
compensate8x8_halfpel(&dct_codes[3*64], cur->y, ref->y, refh->y, refv->y, refhv->y, |
126 |
16*i + 8, 16*j + 8, mb->mvs[3].x, mb->mvs[3].y, edged_width); |
16*i + 8, 16*j + 8, mb->mvs[3].x, mb->mvs[3].y, edged_width); |
127 |
|
|
128 |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
131 |
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; |
132 |
dy = (sum ? SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
dy = (sum ? SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
133 |
|
|
134 |
/* uv-image-based compensation |
/* uv-image-based compensation */ |
135 |
compensate8x8_halfpel(dct_codes[4], cur->u, ref->u, refh->u, refv->u, refhv->u, |
#ifdef BFRAMES |
136 |
|
compensate8x8_halfpel(&dct_codes[4*64], cur->u, ref->u, refh->u, refv->u, refhv->u, |
137 |
8*i, 8*j, dx, dy, edged_width/2); |
8*i, 8*j, dx, dy, edged_width/2); |
138 |
compensate8x8_halfpel(dct_codes[5], cur->v, ref->v, refh->v, refv->v, refhv->v, |
compensate8x8_halfpel(&dct_codes[5*64], cur->v, ref->v, refh->v, refv->v, refhv->v, |
139 |
8*i, 8*j, dx, dy, edged_width/2); */ |
8*i, 8*j, dx, dy, edged_width/2); |
140 |
|
#else |
141 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
142 |
interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
143 |
transfer_8to16sub(dct_codes[4], |
transfer_8to16sub(&dct_codes[4*64], |
144 |
cur->u + 8*j*edged_width/2 + 8*i, |
cur->u + 8*j*edged_width/2 + 8*i, |
145 |
refv->u + 8*j*edged_width/2 + 8*i, edged_width/2); |
refv->u + 8*j*edged_width/2 + 8*i, edged_width/2); |
146 |
|
|
147 |
interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
148 |
transfer_8to16sub(dct_codes[5], |
transfer_8to16sub(&dct_codes[5*64], |
149 |
cur->v + 8*j*edged_width/2 + 8*i, |
cur->v + 8*j*edged_width/2 + 8*i, |
150 |
refv->v + 8*j*edged_width/2 + 8*i, edged_width/2); |
refv->v + 8*j*edged_width/2 + 8*i, edged_width/2); |
151 |
|
#endif |
152 |
|
} |
153 |
|
} |
154 |
|
|
155 |
|
|
156 |
|
void MBMotionCompensationBVOP( |
157 |
|
MBParam * pParam, |
158 |
|
MACROBLOCK * const mb, |
159 |
|
const uint32_t i, |
160 |
|
const uint32_t j, |
161 |
|
IMAGE * const cur, |
162 |
|
const IMAGE * const f_ref, |
163 |
|
const IMAGE * const f_refh, |
164 |
|
const IMAGE * const f_refv, |
165 |
|
const IMAGE * const f_refhv, |
166 |
|
const IMAGE * const b_ref, |
167 |
|
const IMAGE * const b_refh, |
168 |
|
const IMAGE * const b_refv, |
169 |
|
const IMAGE * const b_refhv, |
170 |
|
int16_t * dct_codes) |
171 |
|
{ |
172 |
|
static const uint32_t roundtab[16] = |
173 |
|
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
174 |
|
|
175 |
|
const int32_t edged_width = pParam->edged_width; |
176 |
|
int32_t dx, dy; |
177 |
|
int32_t b_dx, b_dy; |
178 |
|
int x = i; |
179 |
|
int y = j; |
180 |
|
|
181 |
|
|
182 |
|
|
183 |
|
switch(mb->mode) |
184 |
|
{ |
185 |
|
case MODE_FORWARD : |
186 |
|
dx = mb->mvs[0].x; |
187 |
|
dy = mb->mvs[0].y; |
188 |
|
|
189 |
|
transfer_8to16sub_c( |
190 |
|
&dct_codes[0*64], |
191 |
|
cur->y + (j*16)*edged_width + (i*16), |
192 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
193 |
|
i*16, j*16, 1, dx, dy, edged_width), |
194 |
|
edged_width); |
195 |
|
|
196 |
|
transfer_8to16sub( |
197 |
|
&dct_codes[1*64], |
198 |
|
cur->y + (j*16)*edged_width + (i*16+8), |
199 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
200 |
|
i*16+8, j*16, 1, dx, dy, edged_width), |
201 |
|
edged_width); |
202 |
|
|
203 |
|
transfer_8to16sub_c( |
204 |
|
&dct_codes[2*64], |
205 |
|
cur->y + (j*16+8)*edged_width + (i*16), |
206 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
207 |
|
i*16, j*16+8, 1, dx, dy, edged_width), |
208 |
|
edged_width); |
209 |
|
|
210 |
|
transfer_8to16sub( |
211 |
|
&dct_codes[3*64], |
212 |
|
cur->y + (j*16+8)*edged_width + (i*16+8), |
213 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
214 |
|
i*16+8, j*16+8, 1, dx, dy, edged_width), |
215 |
|
edged_width); |
216 |
|
|
217 |
|
|
218 |
|
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
219 |
|
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
220 |
|
|
221 |
|
/* uv-image-based compensation */ |
222 |
|
compensate8x8_halfpel(&dct_codes[4*64], cur->u, f_ref->u, f_refh->u, f_refv->u, f_refhv->u, |
223 |
|
8*i, 8*j, dx, dy, edged_width/2); |
224 |
|
compensate8x8_halfpel(&dct_codes[5*64], cur->v, f_ref->v, f_refh->v, f_refv->v, f_refhv->v, |
225 |
|
8*i, 8*j, dx, dy, edged_width/2); |
226 |
|
|
227 |
|
break; |
228 |
|
|
229 |
|
case MODE_BACKWARD : |
230 |
|
b_dx = mb->b_mvs[0].x; |
231 |
|
b_dy = mb->b_mvs[0].y; |
232 |
|
|
233 |
|
transfer_8to16sub_c( |
234 |
|
&dct_codes[0*64], |
235 |
|
cur->y + (j*16)*edged_width + (i*16), |
236 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
237 |
|
i*16, j*16, 1, b_dx, b_dy, edged_width), |
238 |
|
edged_width); |
239 |
|
|
240 |
|
transfer_8to16sub( |
241 |
|
&dct_codes[1*64], |
242 |
|
cur->y + (j*16)*edged_width + (i*16+8), |
243 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
244 |
|
i*16+8, j*16, 1, b_dx, b_dy, edged_width), |
245 |
|
edged_width); |
246 |
|
|
247 |
|
transfer_8to16sub_c( |
248 |
|
&dct_codes[2*64], |
249 |
|
cur->y + (j*16+8)*edged_width + (i*16), |
250 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
251 |
|
i*16, j*16+8, 1, b_dx, b_dy, edged_width), |
252 |
|
edged_width); |
253 |
|
|
254 |
|
transfer_8to16sub( |
255 |
|
&dct_codes[3*64], |
256 |
|
cur->y + (j*16+8)*edged_width + (i*16+8), |
257 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
258 |
|
i*16+8, j*16+8, 1, b_dx, b_dy, edged_width), |
259 |
|
edged_width); |
260 |
|
|
261 |
|
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
262 |
|
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
263 |
|
|
264 |
|
/* uv-image-based compensation */ |
265 |
|
compensate8x8_halfpel(&dct_codes[4*64], cur->u, |
266 |
|
b_ref->u, b_refh->u, b_refv->u, b_refhv->u, |
267 |
|
8*i, 8*j, b_dx, b_dy, edged_width/2); |
268 |
|
compensate8x8_halfpel(&dct_codes[5*64], cur->v, |
269 |
|
b_ref->v, b_refh->v, b_refv->v, b_refhv->v, |
270 |
|
8*i, 8*j, b_dx, b_dy, edged_width/2); |
271 |
|
|
272 |
|
break; |
273 |
|
|
274 |
|
|
275 |
|
case MODE_INTERPOLATE : |
276 |
|
dx = mb->mvs[0].x; |
277 |
|
dy = mb->mvs[0].y; |
278 |
|
b_dx = mb->b_mvs[0].x; |
279 |
|
b_dy = mb->b_mvs[0].y; |
280 |
|
|
281 |
|
transfer_8to16sub2_c( |
282 |
|
&dct_codes[0*64], |
283 |
|
cur->y + (i*16) + (j*16)*edged_width, |
284 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
285 |
|
16*i, 16*j, 1, dx, dy, edged_width), |
286 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
287 |
|
16*i, 16*j, 1, b_dx, b_dy, edged_width), |
288 |
|
edged_width); |
289 |
|
|
290 |
|
transfer_8to16sub2_c( |
291 |
|
&dct_codes[1*64], |
292 |
|
cur->y + (i*16+8) + (j*16)*edged_width, |
293 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
294 |
|
16*i+8, 16*j, 1, dx, dy, edged_width), |
295 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
296 |
|
16*i+8, 16*j, 1, b_dx, b_dy, edged_width), |
297 |
|
edged_width); |
298 |
|
|
299 |
|
transfer_8to16sub2_c( |
300 |
|
&dct_codes[2*64], |
301 |
|
cur->y + (i*16) + (j*16+8)*edged_width, |
302 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
303 |
|
16*i, 16*j+8, 1, dx, dy, edged_width), |
304 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
305 |
|
16*i, 16*j+8, 1, b_dx, b_dy, edged_width), |
306 |
|
edged_width); |
307 |
|
|
308 |
|
transfer_8to16sub2_c( |
309 |
|
&dct_codes[3*64], |
310 |
|
cur->y + (i*16+8) + (j*16+8)*edged_width, |
311 |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
312 |
|
16*i + 8, 16*j + 8, 1, dx, dy, edged_width), |
313 |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
314 |
|
16*i + 8, 16*j + 8, 1, b_dx, b_dy, edged_width), |
315 |
|
edged_width); |
316 |
|
|
317 |
|
|
318 |
|
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
319 |
|
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
320 |
|
|
321 |
|
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
322 |
|
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
323 |
|
|
324 |
|
transfer_8to16sub2_c( |
325 |
|
&dct_codes[4*64], |
326 |
|
cur->u + (y*8)*edged_width/2 + (x*8), |
327 |
|
get_ref(f_ref->u, f_refh->u, f_refv->u, f_refhv->u, |
328 |
|
8*i, 8*j, 1, dx, dy, edged_width/2), |
329 |
|
get_ref(b_ref->u, b_refh->u, b_refv->u, b_refhv->u, |
330 |
|
8*i, 8*j, 1, b_dx, b_dy, edged_width/2), |
331 |
|
edged_width/2); |
332 |
|
|
333 |
|
transfer_8to16sub2_c( |
334 |
|
&dct_codes[5*64], |
335 |
|
cur->v + (y*8)*edged_width/2 + (x*8), |
336 |
|
get_ref(f_ref->v, f_refh->v, f_refv->v, f_refhv->v, |
337 |
|
8*i, 8*j, 1, dx, dy, edged_width/2), |
338 |
|
get_ref(b_ref->v, b_refh->v, b_refv->v, b_refhv->v, |
339 |
|
8*i, 8*j, 1, b_dx, b_dy, edged_width/2), |
340 |
|
edged_width/2); |
341 |
|
|
342 |
|
break; |
343 |
|
|
344 |
|
case MODE_DIRECT : |
345 |
|
// todo |
346 |
|
break; |
347 |
} |
} |
348 |
|
|
349 |
} |
} |