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