1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* GMC interpolation module |
* - GMC interpolation module - |
5 |
|
* |
6 |
|
* Copyright(C) 2002-2003 Pascal Massimino <skal@planet-d.net> |
7 |
* |
* |
8 |
* This program is free software; you can redistribute it and/or modify |
* This program is free software; you can redistribute it and/or modify |
9 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
17 |
* |
* |
18 |
* You should have received a copy of the GNU General Public License |
* You should have received a copy of the GNU General Public License |
19 |
* along with this program; if not, write to the Free Software |
* along with this program; if not, write to the Free Software |
20 |
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
21 |
|
* |
22 |
|
* $Id$ |
23 |
* |
* |
24 |
*************************************************************************/ |
****************************************************************************/ |
25 |
|
|
26 |
#include "../portab.h" |
#include "../portab.h" |
27 |
#include "../global.h" |
#include "../global.h" |
30 |
|
|
31 |
#include <stdio.h> |
#include <stdio.h> |
32 |
|
|
33 |
/* These are mainly the new GMC routines by -Skal- (C) 2003 */ |
/* ************************************************************ |
34 |
|
* Pts = 2 or 3 |
35 |
////////////////////////////////////////////////////////// |
* |
36 |
// Pts = 2 or 3 |
* Warning! *src is the global frame pointer (that is: adress |
37 |
|
* of pixel 0,0), not the macroblock one. |
38 |
// Warning! *src is the global frame pointer (that is: adress |
* Conversely, *dst is the macroblock top-left adress. |
39 |
// of pixel 0,0), not the macroblock one. |
*/ |
|
// Conversely, *dst is the macroblock top-left adress. |
|
40 |
|
|
41 |
void Predict_16x16_C(const NEW_GMC_DATA * const This, |
void Predict_16x16_C(const NEW_GMC_DATA * const This, |
42 |
uint8_t *dst, const uint8_t *src, |
uint8_t *dst, const uint8_t *src, |
70 |
U += dUx; V += dVx; |
U += dUx; V += dVx; |
71 |
|
|
72 |
if (u > 0 && u <= W) { ri = MTab[u&15]; Offset = u>>4; } |
if (u > 0 && u <= W) { ri = MTab[u&15]; Offset = u>>4; } |
73 |
else if (u > W) Offset = W>>4; |
else { |
74 |
else Offset = -1; |
if (u > W) Offset = W>>4; |
75 |
|
else Offset = 0; |
76 |
|
ri = MTab[0]; |
77 |
|
} |
78 |
|
|
79 |
if (v > 0 && v <= H) { rj = MTab[v&15]; Offset += (v>>4)*srcstride; } |
if (v > 0 && v <= H) { rj = MTab[v&15]; Offset += (v>>4)*srcstride; } |
80 |
else if (v > H) Offset += (H>>4)*srcstride; |
else { |
81 |
else Offset -= srcstride; |
if (v > H) Offset += (H>>4)*srcstride; |
82 |
|
rj = MTab[0]; |
83 |
|
} |
84 |
|
|
85 |
f0 = src[Offset + 0]; |
f0 = src[Offset + 0]; |
86 |
f0 |= src[Offset + 1] << 16; |
f0 |= src[Offset + 1] << 16; |
136 |
ri = MTab[u&15]; |
ri = MTab[u&15]; |
137 |
Offset = u>>4; |
Offset = u>>4; |
138 |
} else { |
} else { |
|
ri = 16; |
|
139 |
if (u>W) Offset = W>>4; |
if (u>W) Offset = W>>4; |
140 |
else Offset = -1; |
else Offset = 0; |
141 |
|
ri = MTab[0]; |
142 |
} |
} |
143 |
|
|
144 |
if (v > 0 && v <= H) { |
if (v > 0 && v <= H) { |
145 |
rj = MTab[v&15]; |
rj = MTab[v&15]; |
146 |
Offset += (v>>4)*srcstride; |
Offset += (v>>4)*srcstride; |
147 |
} else { |
} else { |
|
rj = 16; |
|
148 |
if (v>H) Offset += (H>>4)*srcstride; |
if (v>H) Offset += (H>>4)*srcstride; |
149 |
else Offset -= srcstride; |
rj = MTab[0]; |
150 |
} |
} |
151 |
|
|
152 |
f0 = uSrc[Offset + 0]; |
f0 = uSrc[Offset + 0]; |
195 |
v = V >> 16; V += Dsp->dV[0]; vy += v; |
v = V >> 16; V += Dsp->dV[0]; vy += v; |
196 |
} |
} |
197 |
} |
} |
198 |
vx -= (256*x+120) << (5+Dsp->accuracy); // 120 = 15*16/2 |
vx -= (256*x+120) << (5+Dsp->accuracy); /* 120 = 15*16/2 */ |
199 |
vy -= (256*y+120) << (5+Dsp->accuracy); |
vy -= (256*y+120) << (5+Dsp->accuracy); |
200 |
|
|
201 |
mv->x = RSHIFT( vx, 8+Dsp->accuracy - qpel ); |
mv->x = RSHIFT( vx, 8+Dsp->accuracy - qpel ); |
202 |
mv->y = RSHIFT( vy, 8+Dsp->accuracy - qpel ); |
mv->y = RSHIFT( vy, 8+Dsp->accuracy - qpel ); |
203 |
} |
} |
204 |
|
|
205 |
////////////////////////////////////////////////////////// |
/* ************************************************************ |
206 |
// simplified version for 1 warp point |
* simplified version for 1 warp point |
207 |
|
*/ |
208 |
|
|
209 |
void Predict_1pt_16x16_C(const NEW_GMC_DATA * const This, |
void Predict_1pt_16x16_C(const NEW_GMC_DATA * const This, |
210 |
uint8_t *Dst, const uint8_t *Src, |
uint8_t *Dst, const uint8_t *Src, |
216 |
const int32_t Rounder = ( 128 - (rounding<<(2*rho)) ) << 16; |
const int32_t Rounder = ( 128 - (rounding<<(2*rho)) ) << 16; |
217 |
|
|
218 |
|
|
219 |
int32_t uo = This->Uo + (x<<8); // ((16*x)<<4) |
int32_t uo = This->Uo + (x<<8); /* ((16*x)<<4) */ |
220 |
int32_t vo = This->Vo + (y<<8); |
int32_t vo = This->Vo + (y<<8); |
221 |
const uint32_t ri = MTab[uo & 15]; |
uint32_t ri = MTab[uo & 15]; |
222 |
const uint32_t rj = MTab[vo & 15]; |
uint32_t rj = MTab[vo & 15]; |
223 |
int i, j; |
int i, j; |
224 |
|
|
225 |
int32_t Offset; |
int32_t Offset; |
226 |
if ((uint32_t)vo<=(uint32_t)H) Offset = (vo>>4)*srcstride; |
if (vo>=(-16*4) && vo<=H) Offset = (vo>>4)*srcstride; |
227 |
else if (vo>H) Offset = ( H>>4)*srcstride; |
else { |
228 |
|
if (vo>H) Offset = ( H>>4)*srcstride; |
229 |
else Offset =-16*srcstride; |
else Offset =-16*srcstride; |
230 |
if ((uint32_t)uo<=(uint32_t)W) Offset += (uo>>4); |
rj = MTab[0]; |
231 |
else if (uo>W) Offset += ( W>>4); |
} |
232 |
|
if (uo>=(-16*4) && uo<=W) Offset += (uo>>4); |
233 |
|
else { |
234 |
|
if (uo>W) Offset += (W>>4); |
235 |
else Offset -= 16; |
else Offset -= 16; |
236 |
|
ri = MTab[0]; |
237 |
|
} |
238 |
|
|
239 |
Dst += 16; |
Dst += 16; |
240 |
|
|
269 |
|
|
270 |
int32_t uo = This->Uco + (x<<7); |
int32_t uo = This->Uco + (x<<7); |
271 |
int32_t vo = This->Vco + (y<<7); |
int32_t vo = This->Vco + (y<<7); |
272 |
const uint32_t rri = MTab[uo & 15]; |
uint32_t rri = MTab[uo & 15]; |
273 |
const uint32_t rrj = MTab[vo & 15]; |
uint32_t rrj = MTab[vo & 15]; |
274 |
int i, j; |
int i, j; |
275 |
|
|
276 |
int32_t Offset; |
int32_t Offset; |
277 |
if ((uint32_t)vo<=(uint32_t)H) Offset = (vo>>4)*srcstride; |
if (vo>=(-8*4) && vo<=H) Offset = (vo>>4)*srcstride; |
278 |
else if (vo>H) Offset = ( H>>4)*srcstride; |
else { |
279 |
|
if (vo>H) Offset = ( H>>4)*srcstride; |
280 |
else Offset =-8*srcstride; |
else Offset =-8*srcstride; |
281 |
if ((uint32_t)uo<=(uint32_t)W) Offset += (uo>>4); |
rrj = MTab[0]; |
282 |
else if (uo>W) Offset += (W>>4); |
} |
283 |
|
if (uo>=(-8*4) && uo<=W) Offset += (uo>>4); |
284 |
|
else { |
285 |
|
if (uo>W) Offset += ( W>>4); |
286 |
else Offset -= 8; |
else Offset -= 8; |
287 |
|
rri = MTab[0]; |
288 |
|
} |
289 |
|
|
290 |
uDst += 8; |
uDst += 8; |
291 |
vDst += 8; |
vDst += 8; |
326 |
mv->y = RSHIFT(Dsp->Vo<<qpel, 3); |
mv->y = RSHIFT(Dsp->Vo<<qpel, 3); |
327 |
} |
} |
328 |
|
|
329 |
////////////////////////////////////////////////////////// |
/* ************************************************************* |
330 |
|
* Warning! It's Accuracy being passed, not 'resolution'! |
331 |
|
*/ |
|
// Warning! It's Accuracy being passed, not 'resolution'! |
|
332 |
|
|
333 |
void generate_GMCparameters( int nb_pts, const int accuracy, |
void generate_GMCparameters( int nb_pts, const int accuracy, |
334 |
const WARPPOINTS *const pts, |
const WARPPOINTS *const pts, |
340 |
gmc->accuracy = accuracy; |
gmc->accuracy = accuracy; |
341 |
gmc->num_wp = nb_pts; |
gmc->num_wp = nb_pts; |
342 |
|
|
343 |
// reduce the number of points, if possible |
/* reduce the number of points, if possible */ |
344 |
if (nb_pts<3 || (pts->duv[2].x==-pts->duv[1].y && pts->duv[2].y==pts->duv[1].x)) { |
if (nb_pts<2 || (pts->duv[2].x==0 && pts->duv[2].y==0 && pts->duv[1].x==0 && pts->duv[1].y==0 )) { |
345 |
if (nb_pts<2 || (pts->duv[1].x==0 && pts->duv[1].y==0)) { |
if (nb_pts<2 || (pts->duv[1].x==0 && pts->duv[1].y==0)) { |
346 |
if (nb_pts<1 || (pts->duv[0].x==0 && pts->duv[0].y==0)) { |
if (nb_pts<1 || (pts->duv[0].x==0 && pts->duv[0].y==0)) { |
347 |
nb_pts = 0; |
nb_pts = 0; |
350 |
} |
} |
351 |
else nb_pts = 2; |
else nb_pts = 2; |
352 |
} |
} |
|
else nb_pts = 3; |
|
353 |
|
|
354 |
// now, nb_pts stores the actual number of points required for interpolation |
/* now, nb_pts stores the actual number of points required for interpolation */ |
355 |
|
|
356 |
if (nb_pts<=1) |
if (nb_pts<=1) |
357 |
{ |
{ |
358 |
if (nb_pts==1) { |
if (nb_pts==1) { |
359 |
// store as 4b fixed point |
/* store as 4b fixed point */ |
360 |
gmc->Uo = pts->duv[0].x << accuracy; |
gmc->Uo = pts->duv[0].x << accuracy; |
361 |
gmc->Vo = pts->duv[0].y << accuracy; |
gmc->Vo = pts->duv[0].y << accuracy; |
362 |
gmc->Uco = ((pts->duv[0].x>>1) | (pts->duv[0].x&1)) << accuracy; // DIV2RND() |
gmc->Uco = ((pts->duv[0].x>>1) | (pts->duv[0].x&1)) << accuracy; /* DIV2RND() */ |
363 |
gmc->Vco = ((pts->duv[0].y>>1) | (pts->duv[0].y&1)) << accuracy; // DIV2RND() |
gmc->Vco = ((pts->duv[0].y>>1) | (pts->duv[0].y&1)) << accuracy; /* DIV2RND() */ |
364 |
} |
} |
365 |
else { // zero points?! |
else { /* zero points?! */ |
366 |
gmc->Uo = gmc->Vo = 0; |
gmc->Uo = gmc->Vo = 0; |
367 |
gmc->Uco = gmc->Vco = 0; |
gmc->Uco = gmc->Vco = 0; |
368 |
} |
} |
371 |
gmc->predict_8x8 = Predict_1pt_8x8_C; |
gmc->predict_8x8 = Predict_1pt_8x8_C; |
372 |
gmc->get_average_mv = get_average_mv_1pt_C; |
gmc->get_average_mv = get_average_mv_1pt_C; |
373 |
} |
} |
374 |
else { // 2 or 3 points |
else { /* 2 or 3 points */ |
375 |
const int rho = 3 - accuracy; // = {3,2,1,0} for Acc={0,1,2,3} |
const int rho = 3 - accuracy; /* = {3,2,1,0} for Acc={0,1,2,3} */ |
376 |
int Alpha = log2bin(width-1); |
int Alpha = log2bin(width-1); |
377 |
int Ws = 1 << Alpha; |
int Ws = 1 << Alpha; |
378 |
|
|
379 |
gmc->dU[0] = 16*Ws + RDIV( 8*Ws*pts->duv[1].x, width ); // dU/dx |
gmc->dU[0] = 16*Ws + RDIV( 8*Ws*pts->duv[1].x, width ); /* dU/dx */ |
380 |
gmc->dV[0] = RDIV( 8*Ws*pts->duv[1].y, width ); // dV/dx |
gmc->dV[0] = RDIV( 8*Ws*pts->duv[1].y, width ); /* dV/dx */ |
381 |
|
|
382 |
/* disabled, because possibly buggy? */ |
if (nb_pts==2) { |
383 |
|
gmc->dU[1] = -gmc->dV[0]; /* -Sin */ |
384 |
/* if (nb_pts==2) { |
gmc->dV[1] = gmc->dU[0] ; /* Cos */ |
|
gmc->dU[1] = -gmc->dV[0]; // -Sin |
|
|
gmc->dV[1] = gmc->dU[0] ; // Cos |
|
385 |
} |
} |
386 |
else */ |
else |
387 |
{ |
{ |
388 |
const int Beta = log2bin(height-1); |
const int Beta = log2bin(height-1); |
389 |
const int Hs = 1<<Beta; |
const int Hs = 1<<Beta; |
390 |
gmc->dU[1] = RDIV( 8*Hs*pts->duv[2].x, height ); // dU/dy |
gmc->dU[1] = RDIV( 8*Hs*pts->duv[2].x, height ); /* dU/dy */ |
391 |
gmc->dV[1] = 16*Hs + RDIV( 8*Hs*pts->duv[2].y, height ); // dV/dy |
gmc->dV[1] = 16*Hs + RDIV( 8*Hs*pts->duv[2].y, height ); /* dV/dy */ |
392 |
if (Beta>Alpha) { |
if (Beta>Alpha) { |
393 |
gmc->dU[0] <<= (Beta-Alpha); |
gmc->dU[0] <<= (Beta-Alpha); |
394 |
gmc->dV[0] <<= (Beta-Alpha); |
gmc->dV[0] <<= (Beta-Alpha); |
400 |
gmc->dV[1] <<= Alpha - Beta; |
gmc->dV[1] <<= Alpha - Beta; |
401 |
} |
} |
402 |
} |
} |
403 |
// upscale to 16b fixed-point |
/* upscale to 16b fixed-point */ |
404 |
gmc->dU[0] <<= (16-Alpha - rho); |
gmc->dU[0] <<= (16-Alpha - rho); |
405 |
gmc->dU[1] <<= (16-Alpha - rho); |
gmc->dU[1] <<= (16-Alpha - rho); |
406 |
gmc->dV[0] <<= (16-Alpha - rho); |
gmc->dV[0] <<= (16-Alpha - rho); |
419 |
} |
} |
420 |
} |
} |
421 |
|
|
422 |
////////////////////////////////////////////////////////// |
/* ******************************************************************* |
423 |
|
* quick and dirty routine to generate the full warped image |
424 |
/* quick and dirty routine to generate the full warped image (pGMC != NULL) |
* (pGMC != NULL) or just all average Motion Vectors (pGMC == NULL) */ |
|
or just all average Motion Vectors (pGMC == NULL) */ |
|
425 |
|
|
426 |
void |
void |
427 |
generate_GMCimage( const NEW_GMC_DATA *const gmc_data, // [input] precalculated data |
generate_GMCimage( const NEW_GMC_DATA *const gmc_data, /* [input] precalculated data */ |
428 |
const IMAGE *const pRef, // [input] |
const IMAGE *const pRef, /* [input] */ |
429 |
const int mb_width, |
const int mb_width, |
430 |
const int mb_height, |
const int mb_height, |
431 |
const int stride, |
const int stride, |
432 |
const int stride2, |
const int stride2, |
433 |
const int fcode, // [input] some parameters... |
const int fcode, /* [input] some parameters... */ |
434 |
const int32_t quarterpel, // [input] for rounding avgMV |
const int32_t quarterpel, /* [input] for rounding avgMV */ |
435 |
const int reduced_resolution, // [input] ignored |
const int reduced_resolution, /* [input] ignored */ |
436 |
const int32_t rounding, // [input] for rounding image data |
const int32_t rounding, /* [input] for rounding image data */ |
437 |
MACROBLOCK *const pMBs, // [output] average motion vectors |
MACROBLOCK *const pMBs, /* [output] average motion vectors */ |
438 |
IMAGE *const pGMC) // [output] full warped image |
IMAGE *const pGMC) /* [output] full warped image */ |
439 |
{ |
{ |
440 |
|
|
441 |
unsigned int mj,mi; |
unsigned int mj,mi; |