Parent Directory | Revision Log
Revision 1077 - (view) (download)
1 : | chl | 1077 | /************************************************************************** |
2 : | * | ||
3 : | * XVID MPEG-4 VIDEO CODEC | ||
4 : | * GMC interpolation module | ||
5 : | * | ||
6 : | * This program is free software; you can redistribute it and/or modify | ||
7 : | * it under the terms of the GNU General Public License as published by | ||
8 : | * the Free Software Foundation; either version 2 of the License, or | ||
9 : | * (at your option) any later version. | ||
10 : | * | ||
11 : | * This program is distributed in the hope that it will be useful, | ||
12 : | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
13 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
14 : | * GNU General Public License for more details. | ||
15 : | * | ||
16 : | * You should have received a copy of the GNU General Public License | ||
17 : | * along with this program; if not, write to the Free Software | ||
18 : | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
19 : | * | ||
20 : | *************************************************************************/ | ||
21 : | |||
22 : | #include "../portab.h" | ||
23 : | #include "../global.h" | ||
24 : | #include "../encoder.h" | ||
25 : | #include "gmc.h" | ||
26 : | #include "motion_est.h" | ||
27 : | |||
28 : | #include <stdio.h> | ||
29 : | |||
30 : | /* These are mainly the new GMC routines by -Skal- (C) 2003 */ | ||
31 : | |||
32 : | ////////////////////////////////////////////////////////// | ||
33 : | // Pts = 2 or 3 | ||
34 : | |||
35 : | // Warning! *src is the global frame pointer (that is: adress | ||
36 : | // of pixel 0,0), not the macroblock one. | ||
37 : | // Conversely, *dst is the macroblock top-left adress. | ||
38 : | |||
39 : | |||
40 : | void Predict_16x16_C(const NEW_GMC_DATA * const This, | ||
41 : | uint8_t *dst, const uint8_t *src, | ||
42 : | int dststride, int srcstride, int x, int y, int rounding) | ||
43 : | { | ||
44 : | const int W = This->sW; | ||
45 : | const int H = This->sH; | ||
46 : | const int rho = 3 - This->accuracy; | ||
47 : | const int Rounder = ( (1<<7) - (rounding<<(2*rho)) ) << 16; | ||
48 : | |||
49 : | const int dUx = This->dU[0]; | ||
50 : | const int dVx = This->dV[0]; | ||
51 : | const int dUy = This->dU[1]; | ||
52 : | const int dVy = This->dV[1]; | ||
53 : | |||
54 : | int Uo = This->Uo + 16*(dUy*y + dUx*x); | ||
55 : | int Vo = This->Vo + 16*(dVy*y + dVx*x); | ||
56 : | |||
57 : | int i, j; | ||
58 : | |||
59 : | dst += 16; | ||
60 : | for (j=16; j>0; --j) | ||
61 : | { | ||
62 : | int U = Uo, V = Vo; | ||
63 : | Uo += dUy; Vo += dVy; | ||
64 : | for (i=-16; i<0; ++i) | ||
65 : | { | ||
66 : | unsigned int f0, f1, ri, rj; | ||
67 : | int Offset; | ||
68 : | |||
69 : | int u = ( U >> 16 ) << rho; | ||
70 : | int v = ( V >> 16 ) << rho; | ||
71 : | U += dUx; V += dVx; | ||
72 : | |||
73 : | ri = 16; | ||
74 : | if ((uint32_t)u<=(uint32_t)W) { ri = MTab[u&15]; Offset = u>>4; } | ||
75 : | else if (u>W) Offset = W>>4; | ||
76 : | else Offset = -1; | ||
77 : | |||
78 : | rj = 16; | ||
79 : | if ((uint32_t)v<=(uint32_t)H) { rj = MTab[v&15]; Offset += (v>>4)*srcstride; } | ||
80 : | else if (v>H) Offset += (H>>4)*srcstride; | ||
81 : | else Offset -= srcstride; | ||
82 : | |||
83 : | f0 = src[ Offset +0 ]; | ||
84 : | f0 |= src[ Offset +1 ] << 16; | ||
85 : | f1 = src[ Offset+srcstride +0 ]; | ||
86 : | f1 |= src[ Offset+srcstride +1 ] << 16; | ||
87 : | f0 = (ri*f0)>>16; | ||
88 : | f1 = (ri*f1) & 0x0fff0000; | ||
89 : | f0 |= f1; | ||
90 : | f0 = ( rj*f0 + Rounder ) >> 24; | ||
91 : | |||
92 : | dst[i] = (uint8_t)f0; | ||
93 : | } | ||
94 : | dst += dststride; | ||
95 : | } | ||
96 : | } | ||
97 : | |||
98 : | |||
99 : | void Predict_8x8_C(const NEW_GMC_DATA * const This, | ||
100 : | uint8_t *uDst, const uint8_t *uSrc, | ||
101 : | uint8_t *vDst, const uint8_t *vSrc, | ||
102 : | int dststride, int srcstride, int x, int y, int rounding) | ||
103 : | { | ||
104 : | const int W = This->sW >> 1; | ||
105 : | const int H = This->sH >> 1; | ||
106 : | const int rho = 3-This->accuracy; | ||
107 : | const int32_t Rounder = ( 128 - (rounding<<(2*rho)) ) << 16; | ||
108 : | |||
109 : | const int32_t dUx = This->dU[0]; | ||
110 : | const int32_t dVx = This->dV[0]; | ||
111 : | const int32_t dUy = This->dU[1]; | ||
112 : | const int32_t dVy = This->dV[1]; | ||
113 : | |||
114 : | int32_t Uo = This->Uco + 8*(dUy*y + dUx*x); | ||
115 : | int32_t Vo = This->Vco + 8*(dVy*y + dVx*x); | ||
116 : | |||
117 : | int i, j; | ||
118 : | |||
119 : | uDst += 8; | ||
120 : | vDst += 8; | ||
121 : | for (j=8; j>0; --j) | ||
122 : | { | ||
123 : | int32_t U = Uo, V = Vo; | ||
124 : | Uo += dUy; Vo += dVy; | ||
125 : | |||
126 : | for (i=-8; i<0; ++i) | ||
127 : | { | ||
128 : | int Offset; | ||
129 : | uint32_t f0, f1, ri, rj; | ||
130 : | int32_t u, v; | ||
131 : | |||
132 : | u = ( U >> 16 ) << rho; | ||
133 : | v = ( V >> 16 ) << rho; | ||
134 : | U += dUx; V += dVx; | ||
135 : | |||
136 : | if ((uint32_t)u<=(uint32_t)W) { | ||
137 : | ri = MTab[u&15]; | ||
138 : | Offset = u>>4; | ||
139 : | } | ||
140 : | else { | ||
141 : | ri = 16; | ||
142 : | if (u>W) Offset = W>>4; | ||
143 : | else Offset = -1; | ||
144 : | } | ||
145 : | if ((uint32_t)v<=(uint32_t)H) { | ||
146 : | rj = MTab[v&15]; | ||
147 : | Offset += (v>>4)*srcstride; | ||
148 : | } | ||
149 : | else { | ||
150 : | rj = 16; | ||
151 : | if (v>H) Offset += (H>>4)*srcstride; | ||
152 : | else Offset -= srcstride; | ||
153 : | } | ||
154 : | |||
155 : | f0 = uSrc[ Offset +0 ]; | ||
156 : | f0 |= uSrc[ Offset +1 ] << 16; | ||
157 : | f1 = uSrc[ Offset+srcstride +0 ]; | ||
158 : | f1 |= uSrc[ Offset+srcstride +1 ] << 16; | ||
159 : | f0 = (ri*f0)>>16; | ||
160 : | f1 = (ri*f1) & 0x0fff0000; | ||
161 : | f0 |= f1; | ||
162 : | f0 = ( rj*f0 + Rounder ) >> 24; | ||
163 : | |||
164 : | uDst[i] = (uint8_t)f0; | ||
165 : | |||
166 : | f0 = vSrc[ Offset +0 ]; | ||
167 : | f0 |= vSrc[ Offset +1 ] << 16; | ||
168 : | f1 = vSrc[ Offset+srcstride +0 ]; | ||
169 : | f1 |= vSrc[ Offset+srcstride +1 ] << 16; | ||
170 : | f0 = (ri*f0)>>16; | ||
171 : | f1 = (ri*f1) & 0x0fff0000; | ||
172 : | f0 |= f1; | ||
173 : | f0 = ( rj*f0 + Rounder ) >> 24; | ||
174 : | |||
175 : | vDst[i] = (uint8_t)f0; | ||
176 : | } | ||
177 : | uDst += dststride; | ||
178 : | vDst += dststride; | ||
179 : | } | ||
180 : | } | ||
181 : | |||
182 : | |||
183 : | void get_average_mv_C(NEW_GMC_DATA *Dsp, VECTOR * const mv, | ||
184 : | int x, int y, int qpel) | ||
185 : | { | ||
186 : | int i, j; | ||
187 : | int vx = 0, vy = 0; | ||
188 : | int32_t uo = Dsp->Uo + 16*(Dsp->dU[1]*y + Dsp->dU[0]*x); | ||
189 : | int32_t vo = Dsp->Vo + 16*(Dsp->dV[1]*y + Dsp->dV[0]*x); | ||
190 : | for (j=16; j>0; --j) | ||
191 : | { | ||
192 : | int32_t U, V; | ||
193 : | U = uo; uo += Dsp->dU[1]; | ||
194 : | V = vo; vo += Dsp->dV[1]; | ||
195 : | for (i=16; i>0; --i) | ||
196 : | { | ||
197 : | int32_t u,v; | ||
198 : | u = U >> 16; U += Dsp->dU[0]; vx += u; | ||
199 : | v = V >> 16; V += Dsp->dV[0]; vy += v; | ||
200 : | } | ||
201 : | } | ||
202 : | vx -= (256*x+120) << (5+Dsp->accuracy); // 120 = 15*16/2 | ||
203 : | vy -= (256*y+120) << (5+Dsp->accuracy); | ||
204 : | |||
205 : | mv->x = RSHIFT( vx, 8+Dsp->accuracy - qpel ); | ||
206 : | mv->y = RSHIFT( vy, 8+Dsp->accuracy - qpel ); | ||
207 : | } | ||
208 : | |||
209 : | ////////////////////////////////////////////////////////// | ||
210 : | // simplified version for 1 warp point | ||
211 : | |||
212 : | |||
213 : | void Predict_1pt_16x16_C(const NEW_GMC_DATA * const This, | ||
214 : | uint8_t *Dst, const uint8_t *Src, | ||
215 : | int dststride, int srcstride, int x, int y, int rounding) | ||
216 : | { | ||
217 : | const int W = This->sW; | ||
218 : | const int H = This->sH; | ||
219 : | const int rho = 3-This->accuracy; | ||
220 : | const int32_t Rounder = ( 128 - (rounding<<(2*rho)) ) << 16; | ||
221 : | |||
222 : | |||
223 : | int32_t uo = This->Uo + (x<<8); // ((16*x)<<4) | ||
224 : | int32_t vo = This->Vo + (y<<8); | ||
225 : | const uint32_t ri = MTab[uo & 15]; | ||
226 : | const uint32_t rj = MTab[vo & 15]; | ||
227 : | int i, j; | ||
228 : | |||
229 : | int32_t Offset; | ||
230 : | if ((uint32_t)vo<=(uint32_t)H) Offset = (vo>>4)*srcstride; | ||
231 : | else if (vo>H) Offset = ( H>>4)*srcstride; | ||
232 : | else Offset =-16*srcstride; | ||
233 : | if ((uint32_t)uo<=(uint32_t)W) Offset += (uo>>4); | ||
234 : | else if (uo>W) Offset += ( W>>4); | ||
235 : | else Offset -= 16; | ||
236 : | |||
237 : | Dst += 16; | ||
238 : | |||
239 : | for(j=16; j>0; --j, Offset+=srcstride-16) | ||
240 : | { | ||
241 : | for(i=-16; i<0; ++i, ++Offset) | ||
242 : | { | ||
243 : | uint32_t f0, f1; | ||
244 : | f0 = Src[ Offset +0 ]; | ||
245 : | f0 |= Src[ Offset +1 ] << 16; | ||
246 : | f1 = Src[ Offset+srcstride +0 ]; | ||
247 : | f1 |= Src[ Offset+srcstride +1 ] << 16; | ||
248 : | f0 = (ri*f0)>>16; | ||
249 : | f1 = (ri*f1) & 0x0fff0000; | ||
250 : | f0 |= f1; | ||
251 : | f0 = ( rj*f0 + Rounder ) >> 24; | ||
252 : | Dst[i] = (uint8_t)f0; | ||
253 : | } | ||
254 : | Dst += dststride; | ||
255 : | } | ||
256 : | } | ||
257 : | |||
258 : | |||
259 : | void Predict_1pt_8x8_C(const NEW_GMC_DATA * const This, | ||
260 : | uint8_t *uDst, const uint8_t *uSrc, | ||
261 : | uint8_t *vDst, const uint8_t *vSrc, | ||
262 : | int dststride, int srcstride, int x, int y, int rounding) | ||
263 : | { | ||
264 : | const int W = This->sW >> 1; | ||
265 : | const int H = This->sH >> 1; | ||
266 : | const int rho = 3-This->accuracy; | ||
267 : | const int32_t Rounder = ( 128 - (rounding<<(2*rho)) ) << 16; | ||
268 : | |||
269 : | int32_t uo = This->Uco + (x<<7); | ||
270 : | int32_t vo = This->Vco + (y<<7); | ||
271 : | const uint32_t rri = MTab[uo & 15]; | ||
272 : | const uint32_t rrj = MTab[vo & 15]; | ||
273 : | int i, j; | ||
274 : | |||
275 : | int32_t Offset; | ||
276 : | if ((uint32_t)vo<=(uint32_t)H) Offset = (vo>>4)*srcstride; | ||
277 : | else if (vo>H) Offset = ( H>>4)*srcstride; | ||
278 : | else Offset =-8*srcstride; | ||
279 : | if ((uint32_t)uo<=(uint32_t)W) Offset += (uo>>4); | ||
280 : | else if (uo>W) Offset += ( W>>4); | ||
281 : | else Offset -= 8; | ||
282 : | |||
283 : | uDst += 8; | ||
284 : | vDst += 8; | ||
285 : | for(j=8; j>0; --j, Offset+=srcstride-8) | ||
286 : | { | ||
287 : | for(i=-8; i<0; ++i, Offset++) | ||
288 : | { | ||
289 : | uint32_t f0, f1; | ||
290 : | f0 = uSrc[ Offset +0 ]; | ||
291 : | f0 |= uSrc[ Offset +1 ] << 16; | ||
292 : | f1 = uSrc[ Offset+srcstride +0 ]; | ||
293 : | f1 |= uSrc[ Offset+srcstride +1 ] << 16; | ||
294 : | f0 = (rri*f0)>>16; | ||
295 : | f1 = (rri*f1) & 0x0fff0000; | ||
296 : | f0 |= f1; | ||
297 : | f0 = ( rrj*f0 + Rounder ) >> 24; | ||
298 : | uDst[i] = (uint8_t)f0; | ||
299 : | |||
300 : | f0 = vSrc[ Offset +0 ]; | ||
301 : | f0 |= vSrc[ Offset +1 ] << 16; | ||
302 : | f1 = vSrc[ Offset+srcstride +0 ]; | ||
303 : | f1 |= vSrc[ Offset+srcstride +1 ] << 16; | ||
304 : | f0 = (rri*f0)>>16; | ||
305 : | f1 = (rri*f1) & 0x0fff0000; | ||
306 : | f0 |= f1; | ||
307 : | f0 = ( rrj*f0 + Rounder ) >> 24; | ||
308 : | vDst[i] = (uint8_t)f0; | ||
309 : | } | ||
310 : | uDst += dststride; | ||
311 : | vDst += dststride; | ||
312 : | } | ||
313 : | } | ||
314 : | |||
315 : | |||
316 : | void get_average_mv_1pt_C(NEW_GMC_DATA *Dsp, VECTOR * const mv, | ||
317 : | int x, int y, int qpel) | ||
318 : | { | ||
319 : | mv->x = RSHIFT(Dsp->Uo<<qpel, 3); | ||
320 : | mv->y = RSHIFT(Dsp->Vo<<qpel, 3); | ||
321 : | } | ||
322 : | |||
323 : | ////////////////////////////////////////////////////////// | ||
324 : | |||
325 : | |||
326 : | // Warning! It's Accuracy being passed, not 'resolution'! | ||
327 : | |||
328 : | void generate_GMCparameters( int nb_pts, const int accuracy, | ||
329 : | const WARPPOINTS *const pts, | ||
330 : | const int width, const int height, | ||
331 : | NEW_GMC_DATA *const gmc) | ||
332 : | { | ||
333 : | gmc->sW = width << 4; | ||
334 : | gmc->sH = height << 4; | ||
335 : | gmc->accuracy = accuracy; | ||
336 : | gmc->num_wp = nb_pts; | ||
337 : | |||
338 : | // reduce the number of points, if possible | ||
339 : | if (nb_pts<3 || (pts->duv[2].x==-pts->duv[1].y && pts->duv[2].y==pts->duv[1].x)) { | ||
340 : | if (nb_pts<2 || (pts->duv[1].x==0 && pts->duv[1].y==0)) { | ||
341 : | if (nb_pts<1 || (pts->duv[0].x==0 && pts->duv[0].y==0)) { | ||
342 : | nb_pts = 0; | ||
343 : | } | ||
344 : | else nb_pts = 1; | ||
345 : | } | ||
346 : | else nb_pts = 2; | ||
347 : | } | ||
348 : | else nb_pts = 3; | ||
349 : | |||
350 : | // now, nb_pts stores the actual number of points required for interpolation | ||
351 : | |||
352 : | if (nb_pts<=1) | ||
353 : | { | ||
354 : | if (nb_pts==1) { | ||
355 : | // store as 4b fixed point | ||
356 : | gmc->Uo = pts->duv[0].x << accuracy; | ||
357 : | gmc->Vo = pts->duv[0].y << accuracy; | ||
358 : | gmc->Uco = ((pts->duv[0].x>>1) | (pts->duv[0].x&1)) << accuracy; // DIV2RND() | ||
359 : | gmc->Vco = ((pts->duv[0].y>>1) | (pts->duv[0].y&1)) << accuracy; // DIV2RND() | ||
360 : | } | ||
361 : | else { // zero points?! | ||
362 : | gmc->Uo = gmc->Vo = 0; | ||
363 : | gmc->Uco = gmc->Vco = 0; | ||
364 : | } | ||
365 : | |||
366 : | gmc->predict_16x16 = Predict_1pt_16x16_C; | ||
367 : | gmc->predict_8x8 = Predict_1pt_8x8_C; | ||
368 : | gmc->get_average_mv = get_average_mv_1pt_C; | ||
369 : | } | ||
370 : | else { // 2 or 3 points | ||
371 : | const int rho = 3 - accuracy; // = {3,2,1,0} for Acc={0,1,2,3} | ||
372 : | int Alpha = log2bin(width-1); | ||
373 : | int Ws = 1 << Alpha; | ||
374 : | |||
375 : | gmc->dU[0] = 16*Ws + RDIV( 8*Ws*pts->duv[1].x, width ); // dU/dx | ||
376 : | gmc->dV[0] = RDIV( 8*Ws*pts->duv[1].y, width ); // dV/dx | ||
377 : | |||
378 : | /* disabled, because possibly buggy? */ | ||
379 : | |||
380 : | /* if (nb_pts==2) { | ||
381 : | gmc->dU[1] = -gmc->dV[0]; // -Sin | ||
382 : | gmc->dV[1] = gmc->dU[0] ; // Cos | ||
383 : | } | ||
384 : | else */ | ||
385 : | { | ||
386 : | const int Beta = log2bin(height-1); | ||
387 : | const int Hs = 1<<Beta; | ||
388 : | gmc->dU[1] = RDIV( 8*Hs*pts->duv[2].x, height ); // dU/dy | ||
389 : | gmc->dV[1] = 16*Hs + RDIV( 8*Hs*pts->duv[2].y, height ); // dV/dy | ||
390 : | if (Beta>Alpha) { | ||
391 : | gmc->dU[0] <<= (Beta-Alpha); | ||
392 : | gmc->dV[0] <<= (Beta-Alpha); | ||
393 : | Alpha = Beta; | ||
394 : | Ws = Hs; | ||
395 : | } | ||
396 : | else { | ||
397 : | gmc->dU[1] <<= Alpha - Beta; | ||
398 : | gmc->dV[1] <<= Alpha - Beta; | ||
399 : | } | ||
400 : | } | ||
401 : | // upscale to 16b fixed-point | ||
402 : | gmc->dU[0] <<= (16-Alpha - rho); | ||
403 : | gmc->dU[1] <<= (16-Alpha - rho); | ||
404 : | gmc->dV[0] <<= (16-Alpha - rho); | ||
405 : | gmc->dV[1] <<= (16-Alpha - rho); | ||
406 : | |||
407 : | gmc->Uo = ( pts->duv[0].x <<(16+ accuracy)) + (1<<15); | ||
408 : | gmc->Vo = ( pts->duv[0].y <<(16+ accuracy)) + (1<<15); | ||
409 : | gmc->Uco = ((pts->duv[0].x-1)<<(17+ accuracy)) + (1<<17); | ||
410 : | gmc->Vco = ((pts->duv[0].y-1)<<(17+ accuracy)) + (1<<17); | ||
411 : | gmc->Uco = (gmc->Uco + gmc->dU[0] + gmc->dU[1])>>2; | ||
412 : | gmc->Vco = (gmc->Vco + gmc->dV[0] + gmc->dV[1])>>2; | ||
413 : | |||
414 : | gmc->predict_16x16 = Predict_16x16_C; | ||
415 : | gmc->predict_8x8 = Predict_8x8_C; | ||
416 : | gmc->get_average_mv = get_average_mv_C; | ||
417 : | } | ||
418 : | } | ||
419 : | |||
420 : | ////////////////////////////////////////////////////////// | ||
421 : | |||
422 : | |||
423 : | /* quick and dirty routine to generate the full warped image (pGMC != NULL) | ||
424 : | or just all average Motion Vectors (pGMC == NULL) */ | ||
425 : | |||
426 : | void | ||
427 : | generate_GMCimage( const NEW_GMC_DATA *const gmc_data, // [input] precalculated data | ||
428 : | const IMAGE *const pRef, // [input] | ||
429 : | const int mb_width, | ||
430 : | const int mb_height, | ||
431 : | const int stride, | ||
432 : | const int stride2, | ||
433 : | const int fcode, // [input] some parameters... | ||
434 : | const int32_t quarterpel, // [input] for rounding avgMV | ||
435 : | const int reduced_resolution, // [input] ignored | ||
436 : | const int32_t rounding, // [input] for rounding image data | ||
437 : | MACROBLOCK *const pMBs, // [output] average motion vectors | ||
438 : | IMAGE *const pGMC) // [output] full warped image | ||
439 : | { | ||
440 : | |||
441 : | unsigned int mj,mi; | ||
442 : | VECTOR avgMV; | ||
443 : | |||
444 : | for (mj = 0; mj < (unsigned int)mb_height; mj++) | ||
445 : | for (mi = 0; mi < (unsigned int)mb_width; mi++) { | ||
446 : | const int mbnum = mj*mb_width+mi; | ||
447 : | if (pGMC) | ||
448 : | { | ||
449 : | gmc_data->predict_16x16(gmc_data, | ||
450 : | pGMC->y + mj*16*stride + mi*16, pRef->y, | ||
451 : | stride, stride, mi, mj, rounding); | ||
452 : | |||
453 : | gmc_data->predict_8x8(gmc_data, | ||
454 : | pGMC->u + mj*8*stride2 + mi*8, pRef->u, | ||
455 : | pGMC->v + mj*8*stride2 + mi*8, pRef->v, | ||
456 : | stride2, stride2, mi, mj, rounding); | ||
457 : | } | ||
458 : | gmc_data->get_average_mv(gmc_data, &avgMV, mi, mj, quarterpel); | ||
459 : | |||
460 : | pMBs[mbnum].amv.x = gmc_sanitize(avgMV.x, quarterpel, fcode); | ||
461 : | pMBs[mbnum].amv.y = gmc_sanitize(avgMV.y, quarterpel, fcode); | ||
462 : | |||
463 : | pMBs[mbnum].mcsel = 0; /* until mode decision */ | ||
464 : | } | ||
465 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |