Parent Directory | Revision Log
Revision 1.22 - (view) (download)
1 : | edgomez | 1.20 | /***************************************************************************** |
2 : | * | ||
3 : | * XVID MPEG-4 VIDEO CODEC | ||
4 : | * - Motion Compensation related code - | ||
5 : | * | ||
6 : | * Copyright(C) 2002 Peter Ross <pross@xvid.org> | ||
7 : | * 2003 Christoph Lampert <gruel@web.de> | ||
8 : | * | ||
9 : | * This program is free software ; you can redistribute it and/or modify | ||
10 : | * it under the terms of the GNU General Public License as published by | ||
11 : | * the Free Software Foundation ; either version 2 of the License, or | ||
12 : | * (at your option) any later version. | ||
13 : | * | ||
14 : | * This program is distributed in the hope that it will be useful, | ||
15 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
16 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
17 : | * GNU General Public License for more details. | ||
18 : | * | ||
19 : | * You should have received a copy of the GNU General Public License | ||
20 : | * along with this program ; if not, write to the Free Software | ||
21 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
22 : | * | ||
23 : | * $Id$ | ||
24 : | * | ||
25 : | ****************************************************************************/ | ||
26 : | edgomez | 1.17 | |
27 : | #include <stdio.h> | ||
28 : | suxen_drol | 1.3 | |
29 : | Isibaar | 1.1 | #include "../encoder.h" |
30 : | #include "../utils/mbfunctions.h" | ||
31 : | #include "../image/interpolate8x8.h" | ||
32 : | edgomez | 1.20 | #include "../image/qpel.h" |
33 : | edgomez | 1.17 | #include "../image/reduced.h" |
34 : | Isibaar | 1.1 | #include "../utils/timer.h" |
35 : | suxen_drol | 1.3 | #include "motion.h" |
36 : | Isibaar | 1.1 | |
37 : | edgomez | 1.17 | #ifndef RSHIFT |
38 : | #define RSHIFT(a,b) ((a) > 0 ? ((a) + (1<<((b)-1)))>>(b) : ((a) + (1<<((b)-1))-1)>>(b)) | ||
39 : | #endif | ||
40 : | |||
41 : | /* assume b>0 */ | ||
42 : | #ifndef RDIV | ||
43 : | #define RDIV(a,b) (((a)>0 ? (a) + ((b)>>1) : (a) - ((b)>>1))/(b)) | ||
44 : | #endif | ||
45 : | |||
46 : | |||
47 : | edgomez | 1.20 | /* This is borrowed from bitstream.c until we find a common solution */ |
48 : | edgomez | 1.17 | |
49 : | static uint32_t __inline | ||
50 : | log2bin(uint32_t value) | ||
51 : | { | ||
52 : | /* Changed by Chenm001 */ | ||
53 : | #if !defined(_MSC_VER) | ||
54 : | int n = 0; | ||
55 : | |||
56 : | while (value) { | ||
57 : | value >>= 1; | ||
58 : | n++; | ||
59 : | } | ||
60 : | return n; | ||
61 : | #else | ||
62 : | __asm { | ||
63 : | syskin | 1.19 | bsr eax, value |
64 : | edgomez | 1.17 | inc eax |
65 : | } | ||
66 : | #endif | ||
67 : | } | ||
68 : | |||
69 : | edgomez | 1.20 | /* |
70 : | * getref: calculate reference image pointer | ||
71 : | * the decision to use interpolation h/v/hv or the normal image is | ||
72 : | * based on dx & dy. | ||
73 : | */ | ||
74 : | |||
75 : | static __inline const uint8_t * | ||
76 : | get_ref(const uint8_t * const refn, | ||
77 : | const uint8_t * const refh, | ||
78 : | const uint8_t * const refv, | ||
79 : | const uint8_t * const refhv, | ||
80 : | const uint32_t x, | ||
81 : | const uint32_t y, | ||
82 : | const uint32_t block, | ||
83 : | const int32_t dx, | ||
84 : | const int32_t dy, | ||
85 : | const int32_t stride) | ||
86 : | { | ||
87 : | switch (((dx & 1) << 1) + (dy & 1)) { | ||
88 : | case 0: | ||
89 : | edgomez | 1.22 | return refn + (int) (((int)x * (int)block + dx / 2) + ((int)y * (int)block + dy / 2) * (int)stride); |
90 : | edgomez | 1.20 | case 1: |
91 : | edgomez | 1.22 | return refv + (int) (((int)x * (int)block + dx / 2) + ((int)y * (int)block + (dy - 1) / 2) * (int)stride); |
92 : | edgomez | 1.20 | case 2: |
93 : | edgomez | 1.22 | return refh + (int) (((int)x * (int)block + (dx - 1) / 2) + ((int)y * (int)block + dy / 2) * (int)stride); |
94 : | edgomez | 1.20 | default: |
95 : | edgomez | 1.22 | return refhv + (int) (((int)x * (int)block + (dx - 1) / 2) + ((int)y * (int)block + (dy - 1) / 2) * (int)stride); |
96 : | edgomez | 1.20 | } |
97 : | } | ||
98 : | Isibaar | 1.1 | |
99 : | edgomez | 1.5 | static __inline void |
100 : | edgomez | 1.17 | compensate16x16_interpolate(int16_t * const dct_codes, |
101 : | uint8_t * const cur, | ||
102 : | const uint8_t * const ref, | ||
103 : | const uint8_t * const refh, | ||
104 : | const uint8_t * const refv, | ||
105 : | const uint8_t * const refhv, | ||
106 : | uint8_t * const tmp, | ||
107 : | uint32_t x, | ||
108 : | uint32_t y, | ||
109 : | const int32_t dx, | ||
110 : | const int32_t dy, | ||
111 : | const int32_t stride, | ||
112 : | const int quarterpel, | ||
113 : | const int reduced_resolution, | ||
114 : | const int32_t rounding) | ||
115 : | Isibaar | 1.1 | { |
116 : | edgomez | 1.17 | const uint8_t * ptr; |
117 : | |||
118 : | if (!reduced_resolution) { | ||
119 : | |||
120 : | if(quarterpel) { | ||
121 : | if ((dx&3) | (dy&3)) { | ||
122 : | interpolate16x16_quarterpel(tmp - y * stride - x, | ||
123 : | (uint8_t *) ref, tmp + 32, | ||
124 : | tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); | ||
125 : | ptr = tmp; | ||
126 : | edgomez | 1.22 | } else ptr = ref + ((int)y + dy/4)*(int)stride + (int)x + dx/4; /* fullpixel position */ |
127 : | edgomez | 1.17 | |
128 : | } else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); | ||
129 : | Isibaar | 1.1 | |
130 : | edgomez | 1.5 | transfer_8to16sub(dct_codes, cur + y * stride + x, |
131 : | syskin | 1.19 | ptr, stride); |
132 : | edgomez | 1.17 | transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
133 : | syskin | 1.19 | ptr + 8, stride); |
134 : | edgomez | 1.17 | transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
135 : | syskin | 1.19 | ptr + 8*stride, stride); |
136 : | edgomez | 1.17 | transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
137 : | syskin | 1.19 | ptr + 8*stride + 8, stride); |
138 : | edgomez | 1.17 | |
139 : | edgomez | 1.20 | } else { /* reduced_resolution */ |
140 : | syskin | 1.19 | |
141 : | edgomez | 1.17 | x *= 2; y *= 2; |
142 : | |||
143 : | ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); | ||
144 : | syskin | 1.19 | |
145 : | edgomez | 1.17 | filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
146 : | filter_diff_18x18_to_8x8(dct_codes, ptr, stride); | ||
147 : | |||
148 : | filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); | ||
149 : | filter_diff_18x18_to_8x8(dct_codes+64, ptr + 16, stride); | ||
150 : | |||
151 : | filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); | ||
152 : | filter_diff_18x18_to_8x8(dct_codes+128, ptr + 16*stride, stride); | ||
153 : | |||
154 : | filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); | ||
155 : | filter_diff_18x18_to_8x8(dct_codes+192, ptr + 16*stride + 16, stride); | ||
156 : | |||
157 : | transfer32x32_copy(cur + y*stride + x, ptr, stride); | ||
158 : | } | ||
159 : | } | ||
160 : | |||
161 : | static __inline void | ||
162 : | compensate8x8_interpolate( int16_t * const dct_codes, | ||
163 : | uint8_t * const cur, | ||
164 : | const uint8_t * const ref, | ||
165 : | const uint8_t * const refh, | ||
166 : | const uint8_t * const refv, | ||
167 : | const uint8_t * const refhv, | ||
168 : | uint8_t * const tmp, | ||
169 : | uint32_t x, | ||
170 : | uint32_t y, | ||
171 : | const int32_t dx, | ||
172 : | const int32_t dy, | ||
173 : | const int32_t stride, | ||
174 : | const int32_t quarterpel, | ||
175 : | const int reduced_resolution, | ||
176 : | const int32_t rounding) | ||
177 : | { | ||
178 : | const uint8_t * ptr; | ||
179 : | |||
180 : | if (!reduced_resolution) { | ||
181 : | |||
182 : | if(quarterpel) { | ||
183 : | if ((dx&3) | (dy&3)) { | ||
184 : | interpolate8x8_quarterpel(tmp - y*stride - x, | ||
185 : | (uint8_t *) ref, tmp + 32, | ||
186 : | tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); | ||
187 : | ptr = tmp; | ||
188 : | edgomez | 1.22 | } else ptr = ref + ((int)y + dy/4)*(int)stride + (int)x + dx/4; /* fullpixel position */ |
189 : | edgomez | 1.17 | } else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
190 : | |||
191 : | transfer_8to16sub(dct_codes, cur + y * stride + x, ptr, stride); | ||
192 : | |||
193 : | edgomez | 1.20 | } else { /* reduced_resolution */ |
194 : | Isibaar | 1.1 | |
195 : | edgomez | 1.17 | x *= 2; y *= 2; |
196 : | Isibaar | 1.1 | |
197 : | edgomez | 1.17 | ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
198 : | Isibaar | 1.1 | |
199 : | edgomez | 1.17 | filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
200 : | filter_diff_18x18_to_8x8(dct_codes, ptr, stride); | ||
201 : | syskin | 1.19 | |
202 : | edgomez | 1.17 | transfer16x16_copy(cur + y*stride + x, ptr, stride); |
203 : | edgomez | 1.5 | } |
204 : | Isibaar | 1.1 | } |
205 : | |||
206 : | edgomez | 1.17 | /* XXX: slow, inelegant... */ |
207 : | static void | ||
208 : | interpolate18x18_switch(uint8_t * const cur, | ||
209 : | const uint8_t * const refn, | ||
210 : | const uint32_t x, | ||
211 : | const uint32_t y, | ||
212 : | const int32_t dx, | ||
213 : | const int dy, | ||
214 : | const int32_t stride, | ||
215 : | const int32_t rounding) | ||
216 : | { | ||
217 : | interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); | ||
218 : | interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); | ||
219 : | interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); | ||
220 : | |||
221 : | interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); | ||
222 : | interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); | ||
223 : | interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); | ||
224 : | |||
225 : | interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); | ||
226 : | interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); | ||
227 : | interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); | ||
228 : | } | ||
229 : | |||
230 : | static void | ||
231 : | CompensateChroma( int dx, int dy, | ||
232 : | const int i, const int j, | ||
233 : | IMAGE * const Cur, | ||
234 : | const IMAGE * const Ref, | ||
235 : | uint8_t * const temp, | ||
236 : | int16_t * const coeff, | ||
237 : | const int32_t stride, | ||
238 : | const int rounding, | ||
239 : | const int rrv) | ||
240 : | { /* uv-block-based compensation */ | ||
241 : | |||
242 : | if (!rrv) { | ||
243 : | transfer_8to16sub(coeff, Cur->u + 8 * j * stride + 8 * i, | ||
244 : | syskin | 1.19 | interpolate8x8_switch2(temp, Ref->u, 8 * i, 8 * j, |
245 : | edgomez | 1.17 | dx, dy, stride, rounding), |
246 : | stride); | ||
247 : | transfer_8to16sub(coeff + 64, Cur->v + 8 * j * stride + 8 * i, | ||
248 : | syskin | 1.19 | interpolate8x8_switch2(temp, Ref->v, 8 * i, 8 * j, |
249 : | edgomez | 1.17 | dx, dy, stride, rounding), |
250 : | stride); | ||
251 : | } else { | ||
252 : | uint8_t * current, * reference; | ||
253 : | |||
254 : | current = Cur->u + 16*j*stride + 16*i; | ||
255 : | reference = temp - 16*j*stride - 16*i; | ||
256 : | interpolate18x18_switch(reference, Ref->u, 16*i, 16*j, dx, dy, stride, rounding); | ||
257 : | filter_18x18_to_8x8(coeff, current, stride); | ||
258 : | filter_diff_18x18_to_8x8(coeff, temp, stride); | ||
259 : | transfer16x16_copy(current, temp, stride); | ||
260 : | |||
261 : | current = Cur->v + 16*j*stride + 16*i; | ||
262 : | interpolate18x18_switch(reference, Ref->v, 16*i, 16*j, dx, dy, stride, rounding); | ||
263 : | filter_18x18_to_8x8(coeff + 64, current, stride); | ||
264 : | filter_diff_18x18_to_8x8(coeff + 64, temp, stride); | ||
265 : | transfer16x16_copy(current, temp, stride); | ||
266 : | } | ||
267 : | } | ||
268 : | Isibaar | 1.1 | |
269 : | edgomez | 1.5 | void |
270 : | MBMotionCompensation(MACROBLOCK * const mb, | ||
271 : | syskin | 1.19 | const uint32_t i, |
272 : | const uint32_t j, | ||
273 : | const IMAGE * const ref, | ||
274 : | const IMAGE * const refh, | ||
275 : | const IMAGE * const refv, | ||
276 : | const IMAGE * const refhv, | ||
277 : | const IMAGE * const refGMC, | ||
278 : | IMAGE * const cur, | ||
279 : | int16_t * dct_codes, | ||
280 : | const uint32_t width, | ||
281 : | const uint32_t height, | ||
282 : | const uint32_t edged_width, | ||
283 : | const int32_t quarterpel, | ||
284 : | const int reduced_resolution, | ||
285 : | const int32_t rounding) | ||
286 : | edgomez | 1.17 | { |
287 : | int32_t dx; | ||
288 : | int32_t dy; | ||
289 : | |||
290 : | uint8_t * const tmp = refv->u; | ||
291 : | |||
292 : | if ( (!reduced_resolution) && (mb->mode == MODE_NOT_CODED) ) { /* quick copy for early SKIP */ | ||
293 : | /* early SKIP is only activated in P-VOPs, not in S-VOPs, so mcsel can never be 1 */ | ||
294 : | |||
295 : | transfer16x16_copy(cur->y + 16 * (i + j * edged_width), | ||
296 : | syskin | 1.19 | ref->y + 16 * (i + j * edged_width), |
297 : | edged_width); | ||
298 : | |||
299 : | edgomez | 1.17 | transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
300 : | ref->u + 8 * (i + j * edged_width/2), | ||
301 : | edged_width / 2); | ||
302 : | transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), | ||
303 : | ref->v + 8 * (i + j * edged_width/2), | ||
304 : | edged_width / 2); | ||
305 : | return; | ||
306 : | } | ||
307 : | |||
308 : | syskin | 1.19 | if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER |
309 : | edgomez | 1.17 | || mb->mode == MODE_INTER_Q)) { |
310 : | |||
311 : | /* reduced resolution + GMC: not possible */ | ||
312 : | |||
313 : | if (mb->mcsel) { | ||
314 : | syskin | 1.19 | |
315 : | edgomez | 1.17 | /* call normal routine once, easier than "if (mcsel)"ing all the time */ |
316 : | syskin | 1.19 | |
317 : | edgomez | 1.17 | transfer_8to16sub(&dct_codes[0*64], cur->y + 16*j*edged_width + 16*i, |
318 : | syskin | 1.19 | refGMC->y + 16*j*edged_width + 16*i, edged_width); |
319 : | edgomez | 1.17 | transfer_8to16sub(&dct_codes[1*64], cur->y + 16*j*edged_width + 16*i+8, |
320 : | syskin | 1.19 | refGMC->y + 16*j*edged_width + 16*i+8, edged_width); |
321 : | edgomez | 1.17 | transfer_8to16sub(&dct_codes[2*64], cur->y + (16*j+8)*edged_width + 16*i, |
322 : | syskin | 1.19 | refGMC->y + (16*j+8)*edged_width + 16*i, edged_width); |
323 : | edgomez | 1.17 | transfer_8to16sub(&dct_codes[3*64], cur->y + (16*j+8)*edged_width + 16*i+8, |
324 : | syskin | 1.19 | refGMC->y + (16*j+8)*edged_width + 16*i+8, edged_width); |
325 : | edgomez | 1.17 | |
326 : | /* lumi is needed earlier for mode decision, but chroma should be done block-based, but it isn't, yet. */ | ||
327 : | |||
328 : | transfer_8to16sub(&dct_codes[4 * 64], cur->u + 8 *j*edged_width/2 + 8*i, | ||
329 : | refGMC->u + 8 *j*edged_width/2 + 8*i, edged_width/2); | ||
330 : | |||
331 : | transfer_8to16sub(&dct_codes[5 * 64], cur->v + 8*j* edged_width/2 + 8*i, | ||
332 : | refGMC->v + 8*j* edged_width/2 + 8*i, edged_width/2); | ||
333 : | |||
334 : | return; | ||
335 : | } | ||
336 : | |||
337 : | /* ordinary compensation */ | ||
338 : | syskin | 1.19 | |
339 : | edgomez | 1.17 | dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
340 : | dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); | ||
341 : | |||
342 : | if (reduced_resolution) { | ||
343 : | dx = RRV_MV_SCALEUP(dx); | ||
344 : | dy = RRV_MV_SCALEUP(dy); | ||
345 : | } | ||
346 : | |||
347 : | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, | ||
348 : | refv->y, refhv->y, tmp, 16 * i, 16 * j, dx, dy, | ||
349 : | edged_width, quarterpel, reduced_resolution, rounding); | ||
350 : | |||
351 : | syskin | 1.19 | if (quarterpel) { dx /= 2; dy /= 2; } |
352 : | |||
353 : | edgomez | 1.17 | dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
354 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
355 : | |||
356 : | edgomez | 1.20 | } else { /* mode == MODE_INTER4V */ |
357 : | edgomez | 1.17 | int k, sumx = 0, sumy = 0; |
358 : | const VECTOR * const mvs = (quarterpel ? mb->qmvs : mb->mvs); | ||
359 : | |||
360 : | for (k = 0; k < 4; k++) { | ||
361 : | dx = mvs[k].x; | ||
362 : | dy = mvs[k].y; | ||
363 : | syskin | 1.19 | sumx += quarterpel ? dx/2 : dx; |
364 : | sumy += quarterpel ? dy/2 : dy; | ||
365 : | edgomez | 1.17 | |
366 : | if (reduced_resolution){ | ||
367 : | dx = RRV_MV_SCALEUP(dx); | ||
368 : | dy = RRV_MV_SCALEUP(dy); | ||
369 : | } | ||
370 : | |||
371 : | compensate8x8_interpolate(&dct_codes[k * 64], cur->y, ref->y, refh->y, | ||
372 : | refv->y, refhv->y, tmp, 16 * i + 8*(k&1), 16 * j + 8*(k>>1), dx, | ||
373 : | dy, edged_width, quarterpel, reduced_resolution, rounding); | ||
374 : | } | ||
375 : | dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; | ||
376 : | dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; | ||
377 : | } | ||
378 : | |||
379 : | CompensateChroma(dx, dy, i, j, cur, ref, tmp, | ||
380 : | &dct_codes[4 * 64], edged_width / 2, rounding, reduced_resolution); | ||
381 : | } | ||
382 : | |||
383 : | |||
384 : | void | ||
385 : | MBMotionCompensationBVOP(MBParam * pParam, | ||
386 : | MACROBLOCK * const mb, | ||
387 : | const uint32_t i, | ||
388 : | const uint32_t j, | ||
389 : | IMAGE * const cur, | ||
390 : | const IMAGE * const f_ref, | ||
391 : | const IMAGE * const f_refh, | ||
392 : | const IMAGE * const f_refv, | ||
393 : | const IMAGE * const f_refhv, | ||
394 : | const IMAGE * const b_ref, | ||
395 : | const IMAGE * const b_refh, | ||
396 : | const IMAGE * const b_refv, | ||
397 : | const IMAGE * const b_refhv, | ||
398 : | int16_t * dct_codes) | ||
399 : | { | ||
400 : | const uint32_t edged_width = pParam->edged_width; | ||
401 : | int32_t dx, dy, b_dx, b_dy, sumx, sumy, b_sumx, b_sumy; | ||
402 : | int k; | ||
403 : | edgomez | 1.20 | const int quarterpel = pParam->vol_flags & XVID_VOL_QUARTERPEL; |
404 : | edgomez | 1.17 | const uint8_t * ptr1, * ptr2; |
405 : | uint8_t * const tmp = f_refv->u; | ||
406 : | const VECTOR * const fmvs = (quarterpel ? mb->qmvs : mb->mvs); | ||
407 : | const VECTOR * const bmvs = (quarterpel ? mb->b_qmvs : mb->b_mvs); | ||
408 : | |||
409 : | switch (mb->mode) { | ||
410 : | case MODE_FORWARD: | ||
411 : | dx = fmvs->x; dy = fmvs->y; | ||
412 : | |||
413 : | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, | ||
414 : | f_refv->y, f_refhv->y, tmp, 16 * i, 16 * j, dx, | ||
415 : | dy, edged_width, quarterpel, 0, 0); | ||
416 : | |||
417 : | if (quarterpel) { dx /= 2; dy /= 2; } | ||
418 : | |||
419 : | CompensateChroma( (dx >> 1) + roundtab_79[dx & 0x3], | ||
420 : | (dy >> 1) + roundtab_79[dy & 0x3], | ||
421 : | i, j, cur, f_ref, tmp, | ||
422 : | &dct_codes[4 * 64], edged_width / 2, 0, 0); | ||
423 : | |||
424 : | return; | ||
425 : | |||
426 : | case MODE_BACKWARD: | ||
427 : | b_dx = bmvs->x; b_dy = bmvs->y; | ||
428 : | |||
429 : | syskin | 1.19 | compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
430 : | edgomez | 1.17 | b_refv->y, b_refhv->y, tmp, 16 * i, 16 * j, b_dx, |
431 : | syskin | 1.19 | b_dy, edged_width, quarterpel, 0, 0); |
432 : | edgomez | 1.17 | |
433 : | if (quarterpel) { b_dx /= 2; b_dy /= 2; } | ||
434 : | |||
435 : | CompensateChroma( (b_dx >> 1) + roundtab_79[b_dx & 0x3], | ||
436 : | (b_dy >> 1) + roundtab_79[b_dy & 0x3], | ||
437 : | i, j, cur, b_ref, tmp, | ||
438 : | &dct_codes[4 * 64], edged_width / 2, 0, 0); | ||
439 : | |||
440 : | return; | ||
441 : | |||
442 : | case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ | ||
443 : | case MODE_DIRECT_NO4V: | ||
444 : | dx = fmvs->x; dy = fmvs->y; | ||
445 : | b_dx = bmvs->x; b_dy = bmvs->y; | ||
446 : | |||
447 : | if (quarterpel) { | ||
448 : | syskin | 1.19 | |
449 : | edgomez | 1.17 | if ((dx&3) | (dy&3)) { |
450 : | interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width, | ||
451 : | (uint8_t *) f_ref->y, tmp + 32, | ||
452 : | tmp + 64, tmp + 96, 16*i, 16*j, dx, dy, edged_width, 0); | ||
453 : | ptr1 = tmp; | ||
454 : | edgomez | 1.22 | } else ptr1 = f_ref->y + (16*(int)j + dy/4)*(int)edged_width + 16*(int)i + dx/4; /* fullpixel position */ |
455 : | edgomez | 1.17 | |
456 : | if ((b_dx&3) | (b_dy&3)) { | ||
457 : | interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width + 16, | ||
458 : | (uint8_t *) b_ref->y, tmp + 32, | ||
459 : | tmp + 64, tmp + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0); | ||
460 : | ptr2 = tmp + 16; | ||
461 : | edgomez | 1.22 | } else ptr2 = b_ref->y + (16*(int)j + b_dy/4)*(int)edged_width + 16*(int)i + b_dx/4; /* fullpixel position */ |
462 : | edgomez | 1.17 | |
463 : | b_dx /= 2; | ||
464 : | b_dy /= 2; | ||
465 : | dx /= 2; | ||
466 : | dy /= 2; | ||
467 : | |||
468 : | } else { | ||
469 : | ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, | ||
470 : | i, j, 16, dx, dy, edged_width); | ||
471 : | |||
472 : | ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, | ||
473 : | i, j, 16, b_dx, b_dy, edged_width); | ||
474 : | } | ||
475 : | for (k = 0; k < 4; k++) | ||
476 : | transfer_8to16sub2(&dct_codes[k * 64], | ||
477 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
478 : | ptr1 + (k&1)*8 + (k>>1)*8*edged_width, | ||
479 : | ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width); | ||
480 : | |||
481 : | |||
482 : | dx = (dx >> 1) + roundtab_79[dx & 0x3]; | ||
483 : | dy = (dy >> 1) + roundtab_79[dy & 0x3]; | ||
484 : | |||
485 : | b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; | ||
486 : | b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; | ||
487 : | |||
488 : | break; | ||
489 : | syskin | 1.19 | |
490 : | edgomez | 1.20 | default: /* MODE_DIRECT (or MODE_DIRECT_NONE_MV in case of bframes decoding) */ |
491 : | edgomez | 1.17 | sumx = sumy = b_sumx = b_sumy = 0; |
492 : | |||
493 : | for (k = 0; k < 4; k++) { | ||
494 : | syskin | 1.19 | |
495 : | edgomez | 1.17 | dx = fmvs[k].x; dy = fmvs[k].y; |
496 : | b_dx = bmvs[k].x; b_dy = bmvs[k].y; | ||
497 : | |||
498 : | if (quarterpel) { | ||
499 : | sumx += dx/2; sumy += dy/2; | ||
500 : | b_sumx += b_dx/2; b_sumy += b_dy/2; | ||
501 : | |||
502 : | if ((dx&3) | (dy&3)) { | ||
503 : | interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width, | ||
504 : | syskin | 1.19 | (uint8_t *) f_ref->y, |
505 : | tmp + 32, tmp + 64, tmp + 96, | ||
506 : | edgomez | 1.17 | 16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
507 : | ptr1 = tmp; | ||
508 : | edgomez | 1.22 | } else ptr1 = f_ref->y + (16*(int)j + (k>>1)*8 + dy/4)*(int)edged_width + 16*(int)i + (k&1)*8 + dx/4; |
509 : | edgomez | 1.17 | |
510 : | if ((b_dx&3) | (b_dy&3)) { | ||
511 : | interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width + 16, | ||
512 : | (uint8_t *) b_ref->y, | ||
513 : | syskin | 1.19 | tmp + 16, tmp + 32, tmp + 48, |
514 : | edgomez | 1.17 | 16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
515 : | ptr2 = tmp + 16; | ||
516 : | edgomez | 1.22 | } else ptr2 = b_ref->y + (16*(int)j + (k>>1)*8 + b_dy/4)*(int)edged_width + 16*(int)i + (k&1)*8 + b_dx/4; |
517 : | edgomez | 1.17 | } else { |
518 : | sumx += dx; sumy += dy; | ||
519 : | b_sumx += b_dx; b_sumy += b_dy; | ||
520 : | |||
521 : | syskin | 1.19 | ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
522 : | edgomez | 1.17 | 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, edged_width); |
523 : | syskin | 1.19 | ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
524 : | edgomez | 1.17 | 2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy, edged_width); |
525 : | } | ||
526 : | transfer_8to16sub2(&dct_codes[k * 64], | ||
527 : | cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, | ||
528 : | ptr1, ptr2, edged_width); | ||
529 : | syskin | 1.19 | |
530 : | edgomez | 1.17 | } |
531 : | |||
532 : | dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; | ||
533 : | dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; | ||
534 : | b_dx = (b_sumx >> 3) + roundtab_76[b_sumx & 0xf]; | ||
535 : | b_dy = (b_sumy >> 3) + roundtab_76[b_sumy & 0xf]; | ||
536 : | |||
537 : | break; | ||
538 : | } | ||
539 : | |||
540 : | edgomez | 1.20 | /* v block-based chroma interpolation for direct and interpolate modes */ |
541 : | edgomez | 1.17 | transfer_8to16sub2(&dct_codes[4 * 64], |
542 : | cur->u + (j * 8) * edged_width / 2 + (i * 8), | ||
543 : | interpolate8x8_switch2(tmp, b_ref->u, 8 * i, 8 * j, | ||
544 : | b_dx, b_dy, edged_width / 2, 0), | ||
545 : | interpolate8x8_switch2(tmp + 8, f_ref->u, 8 * i, 8 * j, | ||
546 : | dx, dy, edged_width / 2, 0), | ||
547 : | edged_width / 2); | ||
548 : | |||
549 : | transfer_8to16sub2(&dct_codes[5 * 64], | ||
550 : | cur->v + (j * 8) * edged_width / 2 + (i * 8), | ||
551 : | interpolate8x8_switch2(tmp, b_ref->v, 8 * i, 8 * j, | ||
552 : | b_dx, b_dy, edged_width / 2, 0), | ||
553 : | interpolate8x8_switch2(tmp + 8, f_ref->v, 8 * i, 8 * j, | ||
554 : | dx, dy, edged_width / 2, 0), | ||
555 : | edged_width / 2); | ||
556 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |