1 |
/************************************************************************** |
/************************************************************************** |
2 |
* |
* |
3 |
|
* XVID MPEG-4 VIDEO CODEC |
4 |
|
* motion estimation |
5 |
|
* |
6 |
|
* This program is an implementation of a part of one or more MPEG-4 |
7 |
|
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
8 |
|
* to use this software module in hardware or software products are |
9 |
|
* advised that its use may infringe existing patents or copyrights, and |
10 |
|
* any such use would be at such party's own risk. The original |
11 |
|
* developer of this software module and his/her company, and subsequent |
12 |
|
* editors and their companies, will have no liability for use of this |
13 |
|
* software or modifications or derivatives thereof. |
14 |
|
* |
15 |
|
* This program is free software; you can redistribute it and/or modify |
16 |
|
* it under the terms of the GNU General Public License as published by |
17 |
|
* the Free Software Foundation; either version 2 of the License, or |
18 |
|
* (at your option) any later version. |
19 |
|
* |
20 |
|
* This program is distributed in the hope that it will be useful, |
21 |
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
22 |
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
23 |
|
* GNU General Public License for more details. |
24 |
|
* |
25 |
|
* You should have received a copy of the GNU General Public License |
26 |
|
* along with this program; if not, write to the Free Software |
27 |
|
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
28 |
|
* |
29 |
|
*************************************************************************/ |
30 |
|
|
31 |
|
/************************************************************************** |
32 |
|
* |
33 |
* Modifications: |
* Modifications: |
34 |
* |
* |
35 |
* 01.05.2002 updated MotionEstimationBVOP |
* 01.05.2002 updated MotionEstimationBVOP |
77 |
#include "motion.h" |
#include "motion.h" |
78 |
#include "sad.h" |
#include "sad.h" |
79 |
|
|
|
// very large value |
|
|
#define MV_MAX_ERROR (4096 * 256) |
|
|
|
|
|
// stop search if sdelta < THRESHOLD |
|
|
#define MV16_THRESHOLD 192 |
|
|
#define MV8_THRESHOLD 56 |
|
|
|
|
|
#define NEIGH_MOVE_THRESH 0 |
|
|
// how much a block's MV must differ from his neighbour |
|
|
// to be search for INTER4V. The more, the faster... |
|
|
|
|
|
/* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ |
|
|
/* nb = vop pixels * 2^(bpp-8) */ |
|
|
#define MV16_00_BIAS (128+1) |
|
|
#define MV8_00_BIAS (0) |
|
|
|
|
|
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
|
|
#define MV16_INTER_BIAS 512 |
|
|
|
|
|
/* Parameters which control inter/inter4v decision */ |
|
|
#define IMV16X16 5 |
|
|
|
|
|
/* vector map (vlc delta size) smoother parameters */ |
|
|
#define NEIGH_TEND_16X16 2 |
|
|
#define NEIGH_TEND_8X8 2 |
|
|
|
|
|
// fast ((A)/2)*2 |
|
|
#define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) |
|
|
|
|
|
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
|
|
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
|
|
|
|
|
int32_t PMVfastSearch16(const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, |
|
|
const int y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
int32_t EPZSSearch16(const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, |
|
|
const int y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
|
|
|
int32_t PMVfastSearch8(const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, |
|
|
const int y, |
|
|
const int start_x, |
|
|
const int start_y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
int32_t EPZSSearch8(const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, |
|
|
const int y, |
|
|
const int start_x, |
|
|
const int start_y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
|
|
|
typedef int32_t(MainSearch16Func) (const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const uint8_t * const cur, |
|
|
const int x, |
|
|
const int y, |
|
|
int32_t startx, |
|
|
int32_t starty, |
|
|
int32_t iMinSAD, |
|
|
VECTOR * const currMV, |
|
|
const VECTOR * const pmv, |
|
|
const int32_t min_dx, |
|
|
const int32_t max_dx, |
|
|
const int32_t min_dy, |
|
|
const int32_t max_dy, |
|
|
const int32_t iEdgedWidth, |
|
|
const int32_t iDiamondSize, |
|
|
const int32_t iFcode, |
|
|
const int32_t iQuant, |
|
|
int iFound); |
|
|
|
|
|
typedef MainSearch16Func *MainSearch16FuncPtr; |
|
80 |
|
|
81 |
|
|
|
typedef int32_t(MainSearch8Func) (const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const uint8_t * const cur, |
|
|
const int x, |
|
|
const int y, |
|
|
int32_t startx, |
|
|
int32_t starty, |
|
|
int32_t iMinSAD, |
|
|
VECTOR * const currMV, |
|
|
const VECTOR * const pmv, |
|
|
const int32_t min_dx, |
|
|
const int32_t max_dx, |
|
|
const int32_t min_dy, |
|
|
const int32_t max_dy, |
|
|
const int32_t iEdgedWidth, |
|
|
const int32_t iDiamondSize, |
|
|
const int32_t iFcode, |
|
|
const int32_t iQuant, |
|
|
int iFound); |
|
|
|
|
|
typedef MainSearch8Func *MainSearch8FuncPtr; |
|
|
|
|
82 |
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
83 |
{ 0, (int) (1.00235 + 0.5), (int) (1.15582 + 0.5), (int) (1.31976 + 0.5), |
{ 0, (int) (1.00235 + 0.5), (int) (1.15582 + 0.5), (int) (1.31976 + 0.5), |
84 |
(int) (1.49591 + 0.5), (int) (1.68601 + 0.5), |
(int) (1.49591 + 0.5), (int) (1.68601 + 0.5), |
155 |
mv_bits(dy, iFcode)); |
mv_bits(dy, iFcode)); |
156 |
} |
} |
157 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
#ifndef SEARCH16 |
|
|
#define SEARCH16 PMVfastSearch16 |
|
|
//#define SEARCH16 FullSearch16 |
|
|
//#define SEARCH16 EPZSSearch16 |
|
|
#endif |
|
|
|
|
|
#ifndef SEARCH8 |
|
|
#define SEARCH8 PMVfastSearch8 |
|
|
//#define SEARCH8 EPZSSearch8 |
|
|
#endif |
|
|
|
|
158 |
bool |
bool |
159 |
MotionEstimation(MBParam * const pParam, |
MotionEstimation(MBParam * const pParam, |
160 |
FRAMEINFO * const current, |
FRAMEINFO * const current, |
171 |
const IMAGE *const pCurrent = ¤t->image; |
const IMAGE *const pCurrent = ¤t->image; |
172 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
173 |
|
|
174 |
const VECTOR zeroMV = { 0, 0 }; |
static const VECTOR zeroMV = { 0, 0 }; |
175 |
|
VECTOR predMV; |
176 |
|
|
177 |
int32_t x, y; |
int32_t x, y; |
178 |
int32_t iIntra = 0; |
int32_t iIntra = 0; |
181 |
if (sadInit) |
if (sadInit) |
182 |
(*sadInit) (); |
(*sadInit) (); |
183 |
|
|
184 |
for (y = 0; y < iHcount; y++) |
for (y = 0; y < iHcount; y++) { |
185 |
for (x = 0; x < iWcount; x++) { |
for (x = 0; x < iWcount; x++) { |
186 |
|
|
187 |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
188 |
|
|
189 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
190 |
|
|
191 |
pMB->sad16 = |
pMB->sad16 = |
192 |
SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
193 |
y, current->motion_flags, current->quant, |
x, y, predMV.x, predMV.y, predMV.x, predMV.y, |
194 |
|
current->motion_flags, current->quant, |
195 |
current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
196 |
&pMB->pmvs[0]); |
&pMB->pmvs[0]); |
197 |
|
|
223 |
pMB->dquant == NO_CHANGE)) { |
pMB->dquant == NO_CHANGE)) { |
224 |
int32_t sad8 = IMV16X16 * current->quant; |
int32_t sad8 = IMV16X16 * current->quant; |
225 |
|
|
226 |
if (sad8 < pMB->sad16) |
if (sad8 < pMB->sad16) { |
|
|
|
227 |
sad8 += pMB->sad8[0] = |
sad8 += pMB->sad8[0] = |
228 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
229 |
pCurrent, 2 * x, 2 * y, pMB->mv16.x, |
pCurrent, 2 * x, 2 * y, |
230 |
pMB->mv16.y, current->motion_flags, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
231 |
|
current->motion_flags, |
232 |
current->quant, current->fcode, pParam, |
current->quant, current->fcode, pParam, |
233 |
pMBs, prevMBs, &pMB->mvs[0], |
pMBs, prevMBs, &pMB->mvs[0], |
234 |
&pMB->pmvs[0]); |
&pMB->pmvs[0]); |
235 |
|
} |
236 |
|
if (sad8 < pMB->sad16) { |
237 |
|
|
238 |
if (sad8 < pMB->sad16) |
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 1); |
239 |
sad8 += pMB->sad8[1] = |
sad8 += pMB->sad8[1] = |
240 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
241 |
pCurrent, 2 * x + 1, 2 * y, pMB->mv16.x, |
pCurrent, 2 * x + 1, 2 * y, |
242 |
pMB->mv16.y, current->motion_flags, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
243 |
|
current->motion_flags, |
244 |
current->quant, current->fcode, pParam, |
current->quant, current->fcode, pParam, |
245 |
pMBs, prevMBs, &pMB->mvs[1], |
pMBs, prevMBs, &pMB->mvs[1], |
246 |
&pMB->pmvs[1]); |
&pMB->pmvs[1]); |
247 |
|
} |
248 |
if (sad8 < pMB->sad16) |
if (sad8 < pMB->sad16) { |
249 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 2); |
250 |
sad8 += pMB->sad8[2] = |
sad8 += pMB->sad8[2] = |
251 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
252 |
pCurrent, 2 * x, 2 * y + 1, pMB->mv16.x, |
pCurrent, 2 * x, 2 * y + 1, |
253 |
pMB->mv16.y, current->motion_flags, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
254 |
|
current->motion_flags, |
255 |
current->quant, current->fcode, pParam, |
current->quant, current->fcode, pParam, |
256 |
pMBs, prevMBs, &pMB->mvs[2], |
pMBs, prevMBs, &pMB->mvs[2], |
257 |
&pMB->pmvs[2]); |
&pMB->pmvs[2]); |
258 |
|
} |
259 |
if (sad8 < pMB->sad16) |
if (sad8 < pMB->sad16) { |
260 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 3); |
261 |
sad8 += pMB->sad8[3] = |
sad8 += pMB->sad8[3] = |
262 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
263 |
pCurrent, 2 * x + 1, 2 * y + 1, |
pCurrent, 2 * x + 1, 2 * y + 1, |
264 |
pMB->mv16.x, pMB->mv16.y, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
265 |
current->motion_flags, current->quant, |
current->motion_flags, |
266 |
current->fcode, pParam, pMBs, prevMBs, |
current->quant, current->fcode, pParam, |
267 |
&pMB->mvs[3], &pMB->pmvs[3]); |
pMBs, prevMBs, |
268 |
|
&pMB->mvs[3], |
269 |
|
&pMB->pmvs[3]); |
270 |
|
} |
271 |
|
|
272 |
/* decide: MODE_INTER or MODE_INTER4V |
/* decide: MODE_INTER or MODE_INTER4V |
273 |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
289 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
290 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = |
291 |
pMB->sad16; |
pMB->sad16; |
|
|
|
292 |
} |
} |
293 |
|
} |
294 |
|
|
295 |
return 0; |
return 0; |
296 |
} |
} |
297 |
|
|
298 |
|
|
299 |
#define CHECK_MV16_ZERO {\ |
#define CHECK_MV16_ZERO {\ |
300 |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
301 |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
302 |
{ \ |
{ \ |
303 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ |
304 |
iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
305 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
306 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
307 |
} |
} |
308 |
|
|
309 |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
310 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
311 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
312 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
313 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
314 |
} |
} |
318 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
319 |
{ \ |
{ \ |
320 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
321 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
322 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
323 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
324 |
} |
} |
328 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
329 |
{ \ |
{ \ |
330 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
331 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
332 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
333 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
334 |
} |
} |
338 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
339 |
{ \ |
{ \ |
340 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
341 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
342 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
343 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
344 |
} |
} |
346 |
|
|
347 |
#define CHECK_MV8_ZERO {\ |
#define CHECK_MV8_ZERO {\ |
348 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ |
349 |
iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
350 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
351 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
352 |
} |
} |
354 |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
355 |
{ \ |
{ \ |
356 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
357 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
358 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
359 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
360 |
} |
} |
364 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
365 |
{ \ |
{ \ |
366 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
367 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
368 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
369 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
370 |
} |
} |
374 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
375 |
{ \ |
{ \ |
376 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
377 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
378 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
379 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
380 |
} |
} |
384 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
385 |
{ \ |
{ \ |
386 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
387 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
388 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
389 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
390 |
} |
} |
410 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
411 |
const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; |
const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; |
412 |
int32_t iSAD; |
int32_t iSAD; |
413 |
int32_t pred_x,pred_y; |
VECTOR pred; |
414 |
|
|
415 |
|
|
416 |
get_pmv(pMBs, x, y, pParam->mb_width, 0, &pred_x, &pred_y); |
pred = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
417 |
|
|
418 |
iSAD = sad16( cur, |
iSAD = sad16( cur, |
419 |
get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), |
get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), |
423 |
|
|
424 |
currMV->x = 0; |
currMV->x = 0; |
425 |
currMV->y = 0; |
currMV->y = 0; |
426 |
currPMV->x = -pred_x; |
currPMV->x = -pred.x; |
427 |
currPMV->y = -pred_y; |
currPMV->y = -pred.y; |
428 |
|
|
429 |
return iSAD; |
return iSAD; |
430 |
|
|
439 |
const uint8_t * const cur, |
const uint8_t * const cur, |
440 |
const int x, |
const int x, |
441 |
const int y, |
const int y, |
442 |
int32_t startx, |
const int start_x, |
443 |
int32_t starty, |
const int start_y, |
444 |
int32_t iMinSAD, |
int iMinSAD, |
445 |
VECTOR * const currMV, |
VECTOR * const currMV, |
446 |
const VECTOR * const pmv, |
const int center_x, |
447 |
|
const int center_y, |
448 |
const int32_t min_dx, |
const int32_t min_dx, |
449 |
const int32_t max_dx, |
const int32_t max_dx, |
450 |
const int32_t min_dy, |
const int32_t min_dy, |
461 |
int32_t iSAD; |
int32_t iSAD; |
462 |
VECTOR backupMV; |
VECTOR backupMV; |
463 |
|
|
464 |
backupMV.x = startx; |
backupMV.x = start_x; |
465 |
backupMV.y = starty; |
backupMV.y = start_y; |
466 |
|
|
467 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
468 |
|
|
489 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
490 |
backupMV.y + iDiamondSize, 4); |
backupMV.y + iDiamondSize, 4); |
491 |
} else { |
} else { |
492 |
currMV->x = startx; |
currMV->x = start_x; |
493 |
currMV->y = starty; |
currMV->y = start_y; |
494 |
} |
} |
495 |
return iMinSAD; |
return iMinSAD; |
496 |
} |
} |
503 |
const uint8_t * const cur, |
const uint8_t * const cur, |
504 |
const int x, |
const int x, |
505 |
const int y, |
const int y, |
506 |
int32_t startx, |
const int start_x, |
507 |
int32_t starty, |
const int start_y, |
508 |
int32_t iMinSAD, |
int iMinSAD, |
509 |
VECTOR * const currMV, |
VECTOR * const currMV, |
510 |
const VECTOR * const pmv, |
const int center_x, |
511 |
|
const int center_y, |
512 |
const int32_t min_dx, |
const int32_t min_dx, |
513 |
const int32_t max_dx, |
const int32_t max_dx, |
514 |
const int32_t min_dy, |
const int32_t min_dy, |
525 |
int32_t iSAD; |
int32_t iSAD; |
526 |
VECTOR backupMV; |
VECTOR backupMV; |
527 |
|
|
528 |
backupMV.x = startx; |
backupMV.x = start_x; |
529 |
backupMV.y = starty; |
backupMV.y = start_y; |
530 |
|
|
531 |
/* It's one search with full square pattern, and new parts for all following diamonds */ |
/* It's one search with full square pattern, and new parts for all following diamonds */ |
532 |
|
|
666 |
break; |
break; |
667 |
} |
} |
668 |
} else { |
} else { |
669 |
currMV->x = startx; |
currMV->x = start_x; |
670 |
currMV->y = starty; |
currMV->y = start_y; |
671 |
} |
} |
672 |
return iMinSAD; |
return iMinSAD; |
673 |
} |
} |
681 |
const uint8_t * const cur, |
const uint8_t * const cur, |
682 |
const int x, |
const int x, |
683 |
const int y, |
const int y, |
684 |
int32_t startx, |
const int start_x, |
685 |
int32_t starty, |
const int start_y, |
686 |
int32_t iMinSAD, |
int iMinSAD, |
687 |
VECTOR * const currMV, |
VECTOR * const currMV, |
688 |
const VECTOR * const pmv, |
const int center_x, |
689 |
|
const int center_y, |
690 |
const int32_t min_dx, |
const int32_t min_dx, |
691 |
const int32_t max_dx, |
const int32_t max_dx, |
692 |
const int32_t min_dy, |
const int32_t min_dy, |
701 |
int32_t dx, dy; |
int32_t dx, dy; |
702 |
VECTOR backupMV; |
VECTOR backupMV; |
703 |
|
|
704 |
backupMV.x = startx; |
backupMV.x = start_x; |
705 |
backupMV.y = starty; |
backupMV.y = start_y; |
706 |
|
|
707 |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
708 |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
719 |
const uint8_t * const cur, |
const uint8_t * const cur, |
720 |
const int x, |
const int x, |
721 |
const int y, |
const int y, |
722 |
int32_t startx, |
int start_x, |
723 |
int32_t starty, |
int start_y, |
724 |
int32_t iMinSAD, |
int iMinSAD, |
725 |
VECTOR * const currMV, |
VECTOR * const currMV, |
726 |
const VECTOR * const pmv, |
const int center_x, |
727 |
|
const int center_y, |
728 |
const int32_t min_dx, |
const int32_t min_dx, |
729 |
const int32_t max_dx, |
const int32_t max_dx, |
730 |
const int32_t min_dy, |
const int32_t min_dy, |
741 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
742 |
|
|
743 |
if (iDirection) { |
if (iDirection) { |
744 |
CHECK_MV16_CANDIDATE(startx - iDiamondSize, starty); |
CHECK_MV16_CANDIDATE(start_x - iDiamondSize, start_y); |
745 |
CHECK_MV16_CANDIDATE(startx + iDiamondSize, starty); |
CHECK_MV16_CANDIDATE(start_x + iDiamondSize, start_y); |
746 |
CHECK_MV16_CANDIDATE(startx, starty - iDiamondSize); |
CHECK_MV16_CANDIDATE(start_x, start_y - iDiamondSize); |
747 |
CHECK_MV16_CANDIDATE(startx, starty + iDiamondSize); |
CHECK_MV16_CANDIDATE(start_x, start_y + iDiamondSize); |
748 |
} else { |
} else { |
749 |
int bDirection = 1 + 2 + 4 + 8; |
int bDirection = 1 + 2 + 4 + 8; |
750 |
|
|
751 |
do { |
do { |
752 |
iDirection = 0; |
iDirection = 0; |
753 |
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
754 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
755 |
|
|
756 |
if (bDirection & 2) |
if (bDirection & 2) |
757 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
758 |
|
|
759 |
if (bDirection & 4) |
if (bDirection & 4) |
760 |
CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
761 |
|
|
762 |
if (bDirection & 8) |
if (bDirection & 8) |
763 |
CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
764 |
|
|
765 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
766 |
|
|
768 |
{ |
{ |
769 |
bDirection = iDirection; |
bDirection = iDirection; |
770 |
iDirection = 0; |
iDirection = 0; |
771 |
startx = currMV->x; |
start_x = currMV->x; |
772 |
starty = currMV->y; |
start_y = currMV->y; |
773 |
if (bDirection & 3) //our candidate is left or right |
if (bDirection & 3) //our candidate is left or right |
774 |
{ |
{ |
775 |
CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
776 |
CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
777 |
} else // what remains here is up or down |
} else // what remains here is up or down |
778 |
{ |
{ |
779 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
780 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
781 |
} |
} |
782 |
|
|
783 |
if (iDirection) { |
if (iDirection) { |
784 |
bDirection += iDirection; |
bDirection += iDirection; |
785 |
startx = currMV->x; |
start_x = currMV->x; |
786 |
starty = currMV->y; |
start_y = currMV->y; |
787 |
} |
} |
788 |
} else //about to quit, eh? not so fast.... |
} else //about to quit, eh? not so fast.... |
789 |
{ |
{ |
790 |
switch (bDirection) { |
switch (bDirection) { |
791 |
case 2: |
case 2: |
792 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
793 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
794 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
795 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
796 |
break; |
break; |
797 |
case 1: |
case 1: |
798 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
|
799 |
starty - iDiamondSize, 1 + 4); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
800 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
start_y - iDiamondSize, 1 + 4); |
801 |
starty + iDiamondSize, 1 + 8); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
802 |
|
start_y + iDiamondSize, 1 + 8); |
803 |
break; |
break; |
804 |
case 2 + 4: |
case 2 + 4: |
805 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
806 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
807 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
808 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
809 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
810 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
811 |
break; |
break; |
812 |
case 4: |
case 4: |
813 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
814 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
815 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
816 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
817 |
break; |
break; |
818 |
case 8: |
case 8: |
819 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
820 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
821 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
822 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
823 |
break; |
break; |
824 |
case 1 + 4: |
case 1 + 4: |
825 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
826 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
827 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
828 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
829 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
830 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
831 |
break; |
break; |
832 |
case 2 + 8: |
case 2 + 8: |
833 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
834 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
835 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
836 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
837 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
838 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
839 |
break; |
break; |
840 |
case 1 + 8: |
case 1 + 8: |
841 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
842 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
843 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
844 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
845 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
846 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
847 |
break; |
break; |
848 |
default: //1+2+4+8 == we didn't find anything at all |
default: //1+2+4+8 == we didn't find anything at all |
849 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
850 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
851 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
852 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
853 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
854 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
855 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
856 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
857 |
break; |
break; |
858 |
} |
} |
859 |
if (!iDirection) |
if (!iDirection) |
860 |
break; //ok, the end. really |
break; //ok, the end. really |
861 |
else { |
else { |
862 |
bDirection = iDirection; |
bDirection = iDirection; |
863 |
startx = currMV->x; |
start_x = currMV->x; |
864 |
starty = currMV->y; |
start_y = currMV->y; |
865 |
} |
} |
866 |
} |
} |
867 |
} |
} |
870 |
return iMinSAD; |
return iMinSAD; |
871 |
} |
} |
872 |
|
|
873 |
|
|
874 |
|
#define CHECK_MV16_F_INTERPOL(X,Y,BX,BY) { \ |
875 |
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
876 |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
877 |
|
{ \ |
878 |
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
879 |
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
880 |
|
if (iSAD < iMinSAD) \ |
881 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
882 |
|
} |
883 |
|
|
884 |
|
#define CHECK_MV16_F_INTERPOL_DIR(X,Y,BX,BY,D) { \ |
885 |
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
886 |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
887 |
|
{ \ |
888 |
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
889 |
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
890 |
|
if (iSAD < iMinSAD) \ |
891 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
892 |
|
} |
893 |
|
|
894 |
|
#define CHECK_MV16_F_INTERPOL_FOUND(X,Y,BX,BY,D) { \ |
895 |
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
896 |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
897 |
|
{ \ |
898 |
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
899 |
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
900 |
|
if (iSAD < iMinSAD) \ |
901 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
902 |
|
} |
903 |
|
|
904 |
|
|
905 |
|
#define CHECK_MV16_B_INTERPOL(FX,FY,X,Y) { \ |
906 |
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
907 |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
908 |
|
{ \ |
909 |
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
910 |
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
911 |
|
if (iSAD < iMinSAD) \ |
912 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
913 |
|
} |
914 |
|
|
915 |
|
|
916 |
|
#define CHECK_MV16_B_INTERPOL_DIR(FX,FY,X,Y,D) { \ |
917 |
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
918 |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
919 |
|
{ \ |
920 |
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
921 |
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
922 |
|
if (iSAD < iMinSAD) \ |
923 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
924 |
|
} |
925 |
|
|
926 |
|
|
927 |
|
#define CHECK_MV16_B_INTERPOL_FOUND(FX,FY,X,Y,D) { \ |
928 |
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
929 |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
930 |
|
{ \ |
931 |
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
932 |
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
933 |
|
if (iSAD < iMinSAD) \ |
934 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
935 |
|
} |
936 |
|
|
937 |
|
|
938 |
|
#if (0==1) |
939 |
int32_t |
int32_t |
940 |
AdvDiamond8_MainSearch(const uint8_t * const pRef, |
Diamond16_InterpolMainSearch( |
941 |
const uint8_t * const pRefH, |
const uint8_t * const f_pRef, |
942 |
const uint8_t * const pRefV, |
const uint8_t * const f_pRefH, |
943 |
const uint8_t * const pRefHV, |
const uint8_t * const f_pRefV, |
944 |
|
const uint8_t * const f_pRefHV, |
945 |
const uint8_t * const cur, |
const uint8_t * const cur, |
946 |
|
|
947 |
|
const uint8_t * const b_pRef, |
948 |
|
const uint8_t * const b_pRefH, |
949 |
|
const uint8_t * const b_pRefV, |
950 |
|
const uint8_t * const b_pRefHV, |
951 |
|
|
952 |
const int x, |
const int x, |
953 |
const int y, |
const int y, |
954 |
int32_t startx, |
|
955 |
int32_t starty, |
const int f_start_x, |
956 |
int32_t iMinSAD, |
const int f_start_y, |
957 |
VECTOR * const currMV, |
const int b_start_x, |
958 |
const VECTOR * const pmv, |
const int b_start_y, |
959 |
|
|
960 |
|
int iMinSAD, |
961 |
|
VECTOR * const f_currMV, |
962 |
|
VECTOR * const b_currMV, |
963 |
|
|
964 |
|
const int f_center_x, |
965 |
|
const int f_center_y, |
966 |
|
const int b_center_x, |
967 |
|
const int b_center_y, |
968 |
|
|
969 |
const int32_t min_dx, |
const int32_t min_dx, |
970 |
const int32_t max_dx, |
const int32_t max_dx, |
971 |
const int32_t min_dy, |
const int32_t min_dy, |
972 |
const int32_t max_dy, |
const int32_t max_dy, |
973 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
974 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
975 |
const int32_t iFcode, |
|
976 |
|
const int32_t f_iFcode, |
977 |
|
const int32_t b_iFcode, |
978 |
|
|
979 |
const int32_t iQuant, |
const int32_t iQuant, |
980 |
int iDirection) |
int iFound) |
981 |
{ |
{ |
982 |
|
/* Do a diamond search around given starting point, return SAD of best */ |
983 |
|
|
984 |
|
int32_t f_iDirection = 0; |
985 |
|
int32_t b_iDirection = 0; |
986 |
int32_t iSAD; |
int32_t iSAD; |
987 |
|
|
988 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
VECTOR f_backupMV; |
989 |
|
VECTOR b_backupMV; |
|
if (iDirection) { |
|
|
CHECK_MV8_CANDIDATE(startx - iDiamondSize, starty); |
|
|
CHECK_MV8_CANDIDATE(startx + iDiamondSize, starty); |
|
|
CHECK_MV8_CANDIDATE(startx, starty - iDiamondSize); |
|
|
CHECK_MV8_CANDIDATE(startx, starty + iDiamondSize); |
|
|
} else { |
|
|
int bDirection = 1 + 2 + 4 + 8; |
|
|
|
|
|
do { |
|
|
iDirection = 0; |
|
|
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
|
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
|
|
|
|
|
if (bDirection & 2) |
|
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
|
990 |
|
|
991 |
if (bDirection & 4) |
f_backupMV.x = start_x; |
992 |
CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
f_backupMV.y = start_y; |
993 |
|
b_backupMV.x = start_x; |
994 |
|
b_backupMV.y = start_y; |
995 |
|
|
996 |
if (bDirection & 8) |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
|
CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
|
997 |
|
|
998 |
/* now we're doing diagonal checks near our candidate */ |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
999 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); |
1000 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
1001 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
1002 |
|
|
1003 |
if (iDirection) //checking if anything found |
if (iDirection) |
1004 |
{ |
while (!iFound) { |
1005 |
bDirection = iDirection; |
iFound = 1; |
1006 |
iDirection = 0; |
backupMV = *currMV; |
1007 |
startx = currMV->x; |
|
1008 |
starty = currMV->y; |
if (iDirection != 2) |
1009 |
if (bDirection & 3) //our candidate is left or right |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1010 |
{ |
backupMV.y, 1); |
1011 |
CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
if (iDirection != 1) |
1012 |
CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1013 |
} else // what remains here is up or down |
backupMV.y, 2); |
1014 |
{ |
if (iDirection != 4) |
1015 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
1016 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
backupMV.y - iDiamondSize, 3); |
1017 |
} |
if (iDirection != 3) |
1018 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
1019 |
|
backupMV.y + iDiamondSize, 4); |
1020 |
|
} else { |
1021 |
|
currMV->x = start_x; |
1022 |
|
currMV->y = start_y; |
1023 |
|
} |
1024 |
|
return iMinSAD; |
1025 |
|
} |
1026 |
|
#endif |
1027 |
|
|
1028 |
|
|
1029 |
|
int32_t |
1030 |
|
AdvDiamond8_MainSearch(const uint8_t * const pRef, |
1031 |
|
const uint8_t * const pRefH, |
1032 |
|
const uint8_t * const pRefV, |
1033 |
|
const uint8_t * const pRefHV, |
1034 |
|
const uint8_t * const cur, |
1035 |
|
const int x, |
1036 |
|
const int y, |
1037 |
|
int start_x, |
1038 |
|
int start_y, |
1039 |
|
int iMinSAD, |
1040 |
|
VECTOR * const currMV, |
1041 |
|
const int center_x, |
1042 |
|
const int center_y, |
1043 |
|
const int32_t min_dx, |
1044 |
|
const int32_t max_dx, |
1045 |
|
const int32_t min_dy, |
1046 |
|
const int32_t max_dy, |
1047 |
|
const int32_t iEdgedWidth, |
1048 |
|
const int32_t iDiamondSize, |
1049 |
|
const int32_t iFcode, |
1050 |
|
const int32_t iQuant, |
1051 |
|
int iDirection) |
1052 |
|
{ |
1053 |
|
|
1054 |
|
int32_t iSAD; |
1055 |
|
|
1056 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
1057 |
|
|
1058 |
|
if (iDirection) { |
1059 |
|
CHECK_MV8_CANDIDATE(start_x - iDiamondSize, start_y); |
1060 |
|
CHECK_MV8_CANDIDATE(start_x + iDiamondSize, start_y); |
1061 |
|
CHECK_MV8_CANDIDATE(start_x, start_y - iDiamondSize); |
1062 |
|
CHECK_MV8_CANDIDATE(start_x, start_y + iDiamondSize); |
1063 |
|
} else { |
1064 |
|
int bDirection = 1 + 2 + 4 + 8; |
1065 |
|
|
1066 |
|
do { |
1067 |
|
iDirection = 0; |
1068 |
|
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
1069 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
1070 |
|
|
1071 |
|
if (bDirection & 2) |
1072 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
1073 |
|
|
1074 |
|
if (bDirection & 4) |
1075 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
1076 |
|
|
1077 |
|
if (bDirection & 8) |
1078 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
1079 |
|
|
1080 |
|
/* now we're doing diagonal checks near our candidate */ |
1081 |
|
|
1082 |
|
if (iDirection) //checking if anything found |
1083 |
|
{ |
1084 |
|
bDirection = iDirection; |
1085 |
|
iDirection = 0; |
1086 |
|
start_x = currMV->x; |
1087 |
|
start_y = currMV->y; |
1088 |
|
if (bDirection & 3) //our candidate is left or right |
1089 |
|
{ |
1090 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
1091 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
1092 |
|
} else // what remains here is up or down |
1093 |
|
{ |
1094 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
1095 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
1096 |
|
} |
1097 |
|
|
1098 |
if (iDirection) { |
if (iDirection) { |
1099 |
bDirection += iDirection; |
bDirection += iDirection; |
1100 |
startx = currMV->x; |
start_x = currMV->x; |
1101 |
starty = currMV->y; |
start_y = currMV->y; |
1102 |
} |
} |
1103 |
} else //about to quit, eh? not so fast.... |
} else //about to quit, eh? not so fast.... |
1104 |
{ |
{ |
1105 |
switch (bDirection) { |
switch (bDirection) { |
1106 |
case 2: |
case 2: |
1107 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1108 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1109 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1110 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1111 |
break; |
break; |
1112 |
case 1: |
case 1: |
1113 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1114 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1115 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1116 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1117 |
break; |
break; |
1118 |
case 2 + 4: |
case 2 + 4: |
1119 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1120 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1121 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1122 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1123 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1124 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1125 |
break; |
break; |
1126 |
case 4: |
case 4: |
1127 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1128 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1129 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1130 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1131 |
break; |
break; |
1132 |
case 8: |
case 8: |
1133 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1134 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1135 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1136 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1137 |
break; |
break; |
1138 |
case 1 + 4: |
case 1 + 4: |
1139 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1140 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1141 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1142 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1143 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1144 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1145 |
break; |
break; |
1146 |
case 2 + 8: |
case 2 + 8: |
1147 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1148 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1149 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1150 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1151 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1152 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1153 |
break; |
break; |
1154 |
case 1 + 8: |
case 1 + 8: |
1155 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1156 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1157 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1158 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1159 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1160 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1161 |
break; |
break; |
1162 |
default: //1+2+4+8 == we didn't find anything at all |
default: //1+2+4+8 == we didn't find anything at all |
1163 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1164 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1165 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1166 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1167 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1168 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1169 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1170 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1171 |
break; |
break; |
1172 |
} |
} |
1173 |
if (!(iDirection)) |
if (!(iDirection)) |
1174 |
break; //ok, the end. really |
break; //ok, the end. really |
1175 |
else { |
else { |
1176 |
bDirection = iDirection; |
bDirection = iDirection; |
1177 |
startx = currMV->x; |
start_x = currMV->x; |
1178 |
starty = currMV->y; |
start_y = currMV->y; |
1179 |
} |
} |
1180 |
} |
} |
1181 |
} |
} |
1193 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1194 |
const int x, |
const int x, |
1195 |
const int y, |
const int y, |
1196 |
int32_t startx, |
const int start_x, |
1197 |
int32_t starty, |
const int start_y, |
1198 |
int32_t iMinSAD, |
int iMinSAD, |
1199 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1200 |
const VECTOR * const pmv, |
const int center_x, |
1201 |
|
const int center_y, |
1202 |
const int32_t min_dx, |
const int32_t min_dx, |
1203 |
const int32_t max_dx, |
const int32_t max_dx, |
1204 |
const int32_t min_dy, |
const int32_t min_dy, |
1213 |
int32_t dx, dy; |
int32_t dx, dy; |
1214 |
VECTOR backupMV; |
VECTOR backupMV; |
1215 |
|
|
1216 |
backupMV.x = startx; |
backupMV.x = start_x; |
1217 |
backupMV.y = starty; |
backupMV.y = start_y; |
1218 |
|
|
1219 |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
1220 |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
1223 |
return iMinSAD; |
return iMinSAD; |
1224 |
} |
} |
1225 |
|
|
1226 |
|
Halfpel8_RefineFuncPtr Halfpel8_Refine; |
1227 |
|
|
1228 |
int32_t |
int32_t |
1229 |
Halfpel16_Refine(const uint8_t * const pRef, |
Halfpel16_Refine(const uint8_t * const pRef, |
1235 |
const int y, |
const int y, |
1236 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1237 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1238 |
const VECTOR * const pmv, |
const int center_x, |
1239 |
|
const int center_y, |
1240 |
const int32_t min_dx, |
const int32_t min_dx, |
1241 |
const int32_t max_dx, |
const int32_t max_dx, |
1242 |
const int32_t min_dy, |
const int32_t min_dy, |
1265 |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
1266 |
|
|
1267 |
|
|
1268 |
|
|
1269 |
int32_t |
int32_t |
1270 |
PMVfastSearch16(const uint8_t * const pRef, |
PMVfastSearch16(const uint8_t * const pRef, |
1271 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1274 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1275 |
const int x, |
const int x, |
1276 |
const int y, |
const int y, |
1277 |
|
const int start_x, |
1278 |
|
const int start_y, |
1279 |
|
const int center_x, |
1280 |
|
const int center_y, |
1281 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1282 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1283 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1311 |
|
|
1312 |
MainSearch16FuncPtr MainSearchPtr; |
MainSearch16FuncPtr MainSearchPtr; |
1313 |
|
|
|
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
|
1314 |
const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
1315 |
|
|
1316 |
static int32_t threshA, threshB; |
int32_t threshA, threshB; |
1317 |
int32_t bPredEq; |
int32_t bPredEq; |
1318 |
int32_t iMinSAD, iSAD; |
int32_t iMinSAD, iSAD; |
1319 |
|
|
1331 |
} |
} |
1332 |
|
|
1333 |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
1334 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
1335 |
|
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
1336 |
|
|
1337 |
if ((x == 0) && (y == 0)) { |
if ((x == 0) && (y == 0)) { |
1338 |
threshA = 512; |
threshA = 512; |
1339 |
threshB = 1024; |
threshB = 1024; |
|
|
|
1340 |
} else { |
} else { |
1341 |
threshA = psad[0]; |
threshA = psad[0]; |
1342 |
threshB = threshA + 256; |
threshB = threshA + 256; |
1357 |
If SAD<=256 goto Step 10. |
If SAD<=256 goto Step 10. |
1358 |
*/ |
*/ |
1359 |
|
|
1360 |
*currMV = pmv[0]; /* current best := prediction */ |
currMV->x = start_x; |
1361 |
|
currMV->y = start_y; |
1362 |
|
|
1363 |
if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
1364 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
1365 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
1383 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
1384 |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
1385 |
iMinSAD += |
iMinSAD += |
1386 |
calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
1387 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
1388 |
|
|
1389 |
if ((iMinSAD < 256) || |
if ((iMinSAD < 256) || |
1390 |
((MVequal(*currMV, prevMB->mvs[0])) && |
((MVequal(*currMV, prevMB->mvs[0])) && |
1391 |
((uint32_t) iMinSAD < prevMB->sad16))) { |
((int32_t) iMinSAD < prevMB->sad16))) { |
1392 |
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
1393 |
{ |
{ |
1394 |
if (!MVzero(*currMV)) { |
if (!MVzero(*currMV)) { |
1492 |
|
|
1493 |
if ((iMinSAD <= threshA) || |
if ((iMinSAD <= threshA) || |
1494 |
(MVequal(*currMV, prevMB->mvs[0]) && |
(MVequal(*currMV, prevMB->mvs[0]) && |
1495 |
((uint32_t) iMinSAD < prevMB->sad16))) { |
((int32_t) iMinSAD < prevMB->sad16))) { |
1496 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1497 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1498 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1519 |
|
|
1520 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1521 |
|
|
1522 |
|
|
1523 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
1524 |
iSAD = |
iSAD = |
1525 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1526 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->x, currMV->y, iMinSAD, &newMV, center_x, center_y, |
1527 |
|
min_dx, max_dx, |
1528 |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1529 |
iQuant, iFound); |
iQuant, iFound); |
1530 |
|
|
1539 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
1540 |
iSAD = |
iSAD = |
1541 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1542 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
center_x, center_y, iMinSAD, &newMV, center_x, center_y, |
1543 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1544 |
iDiamondSize, iFcode, iQuant, iFound); |
iDiamondSize, iFcode, iQuant, iFound); |
1545 |
|
|
1552 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1553 |
iSAD = |
iSAD = |
1554 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1555 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, |
1556 |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dx, max_dx, min_dy, max_dy, |
1557 |
|
iEdgedWidth, iDiamondSize, iFcode, |
1558 |
iQuant, iFound); |
iQuant, iFound); |
1559 |
|
|
1560 |
if (iSAD < iMinSAD) { |
if (iSAD < iMinSAD) { |
1572 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
1573 |
iMinSAD = |
iMinSAD = |
1574 |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1575 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
1576 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
1577 |
|
|
1578 |
PMVfast16_Terminate_without_Refine: |
PMVfast16_Terminate_without_Refine: |
1579 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
1580 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
1581 |
return iMinSAD; |
return iMinSAD; |
1582 |
} |
} |
1583 |
|
|
1594 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1595 |
const int x, |
const int x, |
1596 |
const int y, |
const int y, |
1597 |
int32_t startx, |
int32_t start_x, |
1598 |
int32_t starty, |
int32_t start_y, |
1599 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1600 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1601 |
const VECTOR * const pmv, |
const int center_x, |
1602 |
|
const int center_y, |
1603 |
const int32_t min_dx, |
const int32_t min_dx, |
1604 |
const int32_t max_dx, |
const int32_t max_dx, |
1605 |
const int32_t min_dy, |
const int32_t min_dy, |
1616 |
int32_t iSAD; |
int32_t iSAD; |
1617 |
VECTOR backupMV; |
VECTOR backupMV; |
1618 |
|
|
1619 |
backupMV.x = startx; |
backupMV.x = start_x; |
1620 |
backupMV.y = starty; |
backupMV.y = start_y; |
1621 |
|
|
1622 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
1623 |
|
|
1644 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1645 |
backupMV.y + iDiamondSize, 4); |
backupMV.y + iDiamondSize, 4); |
1646 |
} else { |
} else { |
1647 |
currMV->x = startx; |
currMV->x = start_x; |
1648 |
currMV->y = starty; |
currMV->y = start_y; |
1649 |
} |
} |
1650 |
return iMinSAD; |
return iMinSAD; |
1651 |
} |
} |
1652 |
|
|
1653 |
int32_t |
int32_t |
1654 |
Halfpel8_Refine(const uint8_t * const pRef, |
Halfpel8_Refine_c(const uint8_t * const pRef, |
1655 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1656 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1657 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1660 |
const int y, |
const int y, |
1661 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1662 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1663 |
const VECTOR * const pmv, |
const int center_x, |
1664 |
|
const int center_y, |
1665 |
const int32_t min_dx, |
const int32_t min_dx, |
1666 |
const int32_t max_dx, |
const int32_t max_dx, |
1667 |
const int32_t min_dy, |
const int32_t min_dy, |
1700 |
const int y, |
const int y, |
1701 |
const int start_x, |
const int start_x, |
1702 |
const int start_y, |
const int start_y, |
1703 |
|
const int center_x, |
1704 |
|
const int center_y, |
1705 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1706 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1707 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1734 |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1735 |
const MACROBLOCK *const prevMB = prevMBs + (x >> 1) + (y >> 1) * iWcount; |
const MACROBLOCK *const prevMB = prevMBs + (x >> 1) + (y >> 1) * iWcount; |
1736 |
|
|
1737 |
static int32_t threshA, threshB; |
int32_t threshA, threshB; |
1738 |
int32_t iFound, bPredEq; |
int32_t iFound, bPredEq; |
1739 |
int32_t iMinSAD, iSAD; |
int32_t iMinSAD, iSAD; |
1740 |
|
|
1758 |
} |
} |
1759 |
|
|
1760 |
/* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
1761 |
bPredEq = |
//bPredEq = get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); |
1762 |
get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, (x >> 1), (y >> 1), iSubBlock, pmv, psad); |
1763 |
|
|
1764 |
if ((x == 0) && (y == 0)) { |
if ((x == 0) && (y == 0)) { |
1765 |
threshA = 512 / 4; |
threshA = 512 / 4; |
1766 |
threshB = 1024 / 4; |
threshB = 1024 / 4; |
1767 |
|
|
1768 |
} else { |
} else { |
1769 |
threshA = psad[0] / 4; /* good estimate */ |
threshA = psad[0] / 4; /* good estimate? */ |
1770 |
threshB = threshA + 256 / 4; |
threshB = threshA + 256 / 4; |
1771 |
if (threshA < 512 / 4) |
if (threshA < 512 / 4) |
1772 |
threshA = 512 / 4; |
threshA = 512 / 4; |
1805 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
1806 |
iEdgedWidth), iEdgedWidth); |
iEdgedWidth), iEdgedWidth); |
1807 |
iMinSAD += |
iMinSAD += |
1808 |
calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
1809 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
1810 |
|
|
1811 |
if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) |
if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) |
1812 |
&& ((uint32_t) iMinSAD < |
&& ((int32_t) iMinSAD < |
1813 |
prevMB->sad8[iSubBlock]))) { |
prevMB->sad8[iSubBlock]))) { |
1814 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1815 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1848 |
// the median prediction might be even better than mv16 |
// the median prediction might be even better than mv16 |
1849 |
|
|
1850 |
if (!MVequal(pmv[0], startMV)) |
if (!MVequal(pmv[0], startMV)) |
1851 |
CHECK_MV8_CANDIDATE(pmv[0].x, pmv[0].y); |
CHECK_MV8_CANDIDATE(center_x, center_y); |
1852 |
|
|
1853 |
// (0,0) if needed |
// (0,0) if needed |
1854 |
if (!MVzero(pmv[0])) |
if (!MVzero(pmv[0])) |
1864 |
|
|
1865 |
if ((iMinSAD <= threshA) || |
if ((iMinSAD <= threshA) || |
1866 |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
1867 |
((uint32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1868 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1869 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1870 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1923 |
|
|
1924 |
if ((iMinSAD <= threshA) || |
if ((iMinSAD <= threshA) || |
1925 |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
1926 |
((uint32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1927 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1928 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1929 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1945 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
1946 |
iSAD = |
iSAD = |
1947 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
1948 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
1949 |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1950 |
iQuant, iFound); |
iQuant, iFound); |
1951 |
|
|
1960 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
1961 |
iSAD = |
iSAD = |
1962 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1963 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
1964 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1965 |
iDiamondSize, iFcode, iQuant, iFound); |
iDiamondSize, iFcode, iQuant, iFound); |
1966 |
|
|
1973 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1974 |
iSAD = |
iSAD = |
1975 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1976 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
1977 |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1978 |
iQuant, iFound); |
iQuant, iFound); |
1979 |
|
|
1992 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
1993 |
iMinSAD = |
iMinSAD = |
1994 |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1995 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
1996 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
1997 |
|
|
1998 |
|
|
1999 |
PMVfast8_Terminate_without_Refine: |
PMVfast8_Terminate_without_Refine: |
2000 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2001 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2002 |
|
|
2003 |
return iMinSAD; |
return iMinSAD; |
2004 |
} |
} |
2011 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
2012 |
const int x, |
const int x, |
2013 |
const int y, |
const int y, |
2014 |
|
const int start_x, |
2015 |
|
const int start_y, |
2016 |
|
const int center_x, |
2017 |
|
const int center_y, |
2018 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2019 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2020 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2050 |
const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
2051 |
MACROBLOCK *oldMB = NULL; |
MACROBLOCK *oldMB = NULL; |
2052 |
|
|
2053 |
static int32_t thresh2; |
int32_t thresh2; |
2054 |
int32_t bPredEq; |
int32_t bPredEq; |
2055 |
int32_t iMinSAD, iSAD = 9999; |
int32_t iMinSAD, iSAD = 9999; |
2056 |
|
|
2073 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2074 |
} |
} |
2075 |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2076 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
2077 |
|
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
2078 |
|
|
2079 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2080 |
MinSAD=SAD |
MinSAD=SAD |
2085 |
|
|
2086 |
// Prepare for main loop |
// Prepare for main loop |
2087 |
|
|
2088 |
*currMV = pmv[0]; /* current best := median prediction */ |
currMV->x = start_x; |
2089 |
|
currMV->y = start_y; |
2090 |
|
|
2091 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2092 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
2093 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
2109 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
2110 |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2111 |
iMinSAD += |
iMinSAD += |
2112 |
calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2113 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
2114 |
|
|
2115 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2116 |
if ((iMinSAD < 256) || |
if ((iMinSAD < 256) || |
2117 |
((MVequal(*currMV, prevMB->mvs[0])) && |
((MVequal(*currMV, prevMB->mvs[0])) && |
2118 |
((uint32_t) iMinSAD < prevMB->sad16))) { |
((int32_t) iMinSAD < prevMB->sad16))) { |
2119 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2120 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2121 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2175 |
|
|
2176 |
if ((iMinSAD <= thresh2) |
if ((iMinSAD <= thresh2) |
2177 |
|| (MVequal(*currMV, prevMB->mvs[0]) && |
|| (MVequal(*currMV, prevMB->mvs[0]) && |
2178 |
((uint32_t) iMinSAD <= prevMB->sad16))) { |
((int32_t) iMinSAD <= prevMB->sad16))) { |
2179 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2180 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2181 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2221 |
|
|
2222 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2223 |
|
|
2224 |
if (MotionFlags & PMV_USESQUARES8) |
if (MotionFlags & PMV_USESQUARES16) |
2225 |
MainSearchPtr = Square16_MainSearch; |
MainSearchPtr = Square16_MainSearch; |
2226 |
else |
else |
2227 |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
2228 |
MainSearchPtr = AdvDiamond16_MainSearch; |
MainSearchPtr = AdvDiamond16_MainSearch; |
2229 |
else |
else |
2230 |
MainSearchPtr = Diamond16_MainSearch; |
MainSearchPtr = Diamond16_MainSearch; |
2233 |
|
|
2234 |
iSAD = |
iSAD = |
2235 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2236 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2237 |
min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2238 |
|
|
2239 |
if (iSAD < iMinSAD) { |
if (iSAD < iMinSAD) { |
2248 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
2249 |
iSAD = |
iSAD = |
2250 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2251 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2252 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2253 |
2, iFcode, iQuant, 0); |
2, iFcode, iQuant, 0); |
2254 |
} |
} |
2261 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2262 |
iSAD = |
iSAD = |
2263 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2264 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2265 |
max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2266 |
|
|
2267 |
if (iSAD < iMinSAD) { |
if (iSAD < iMinSAD) { |
2277 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
2278 |
iMinSAD = |
iMinSAD = |
2279 |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2280 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2281 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2282 |
|
|
2283 |
EPZS16_Terminate_without_Refine: |
EPZS16_Terminate_without_Refine: |
2284 |
|
|
2285 |
*oldMB = *prevMB; |
*oldMB = *prevMB; |
2286 |
|
|
2287 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2288 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2289 |
return iMinSAD; |
return iMinSAD; |
2290 |
} |
} |
2291 |
|
|
2300 |
const int y, |
const int y, |
2301 |
const int start_x, |
const int start_x, |
2302 |
const int start_y, |
const int start_y, |
2303 |
|
const int center_x, |
2304 |
|
const int center_y, |
2305 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2306 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2307 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2356 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2357 |
} |
} |
2358 |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2359 |
bPredEq = get_pmvdata(pMBs, x >> 1, y >> 1, iWcount, iSubBlock, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, x >> 1, y >> 1, iWcount, iSubBlock, pmv[0].x, pmv[0].y, psad); |
2360 |
|
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x >> 1, y >> 1, iSubBlock, pmv, psad); |
2361 |
|
|
2362 |
|
|
2363 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2392 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
2393 |
iEdgedWidth), iEdgedWidth); |
iEdgedWidth), iEdgedWidth); |
2394 |
iMinSAD += |
iMinSAD += |
2395 |
calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
2396 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
2397 |
|
|
2398 |
|
|
2479 |
|
|
2480 |
iSAD = |
iSAD = |
2481 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2482 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2483 |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2484 |
iQuant, 0); |
iQuant, 0); |
2485 |
|
|
2495 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
2496 |
iSAD = |
iSAD = |
2497 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2498 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2499 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2500 |
iDiamondSize, iFcode, iQuant, 0); |
iDiamondSize, iFcode, iQuant, 0); |
2501 |
|
|
2508 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2509 |
iSAD = |
iSAD = |
2510 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2511 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2512 |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2513 |
iQuant, 0); |
iQuant, 0); |
2514 |
|
|
2525 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
2526 |
iMinSAD = |
iMinSAD = |
2527 |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2528 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2529 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2530 |
|
|
2531 |
EPZS8_Terminate_without_Refine: |
EPZS8_Terminate_without_Refine: |
2532 |
|
|
2533 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2534 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2535 |
return iMinSAD; |
return iMinSAD; |
2536 |
} |
} |
2537 |
|
|
2538 |
|
|
2539 |
|
|
2540 |
|
int32_t |
2541 |
|
PMVfastIntSearch16(const uint8_t * const pRef, |
2542 |
|
const uint8_t * const pRefH, |
2543 |
|
const uint8_t * const pRefV, |
2544 |
|
const uint8_t * const pRefHV, |
2545 |
|
const IMAGE * const pCur, |
2546 |
|
const int x, |
2547 |
|
const int y, |
2548 |
|
const int start_x, |
2549 |
|
const int start_y, |
2550 |
|
const int center_x, |
2551 |
|
const int center_y, |
2552 |
|
const uint32_t MotionFlags, |
2553 |
|
const uint32_t iQuant, |
2554 |
|
const uint32_t iFcode, |
2555 |
|
const MBParam * const pParam, |
2556 |
|
const MACROBLOCK * const pMBs, |
2557 |
|
const MACROBLOCK * const prevMBs, |
2558 |
|
VECTOR * const currMV, |
2559 |
|
VECTOR * const currPMV) |
2560 |
|
{ |
2561 |
|
const uint32_t iWcount = pParam->mb_width; |
2562 |
|
const int32_t iWidth = pParam->width; |
2563 |
|
const int32_t iHeight = pParam->height; |
2564 |
|
const int32_t iEdgedWidth = pParam->edged_width; |
2565 |
|
|
2566 |
|
const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; |
2567 |
|
const VECTOR zeroMV = { 0, 0 }; |
2568 |
|
|
2569 |
|
int32_t iDiamondSize; |
2570 |
|
|
2571 |
|
int32_t min_dx; |
2572 |
|
int32_t max_dx; |
2573 |
|
int32_t min_dy; |
2574 |
|
int32_t max_dy; |
2575 |
|
|
2576 |
|
int32_t iFound; |
2577 |
|
|
2578 |
|
VECTOR newMV; |
2579 |
|
VECTOR backupMV; /* just for PMVFAST */ |
2580 |
|
|
2581 |
|
VECTOR pmv[4]; |
2582 |
|
int32_t psad[4]; |
2583 |
|
|
2584 |
|
MainSearch16FuncPtr MainSearchPtr; |
2585 |
|
|
2586 |
|
const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
2587 |
|
MACROBLOCK *const pMB = pMBs + x + y * iWcount; |
2588 |
|
|
2589 |
|
int32_t threshA, threshB; |
2590 |
|
int32_t bPredEq; |
2591 |
|
int32_t iMinSAD, iSAD; |
2592 |
|
|
2593 |
|
|
2594 |
|
/* Get maximum range */ |
2595 |
|
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
2596 |
|
iFcode); |
2597 |
|
|
2598 |
|
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
2599 |
|
|
2600 |
|
if ((x == 0) && (y == 0)) { |
2601 |
|
threshA = 512; |
2602 |
|
threshB = 1024; |
2603 |
|
|
2604 |
|
bPredEq = 0; |
2605 |
|
psad[0] = psad[1] = psad[2] = psad[3] = 0; |
2606 |
|
*currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; |
2607 |
|
|
2608 |
|
} else { |
2609 |
|
threshA = psad[0]; |
2610 |
|
threshB = threshA + 256; |
2611 |
|
if (threshA < 512) |
2612 |
|
threshA = 512; |
2613 |
|
if (threshA > 1024) |
2614 |
|
threshA = 1024; |
2615 |
|
if (threshB > 1792) |
2616 |
|
threshB = 1792; |
2617 |
|
|
2618 |
|
bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
2619 |
|
*currMV = pmv[0]; /* current best := prediction */ |
2620 |
|
} |
2621 |
|
|
2622 |
|
iFound = 0; |
2623 |
|
|
2624 |
|
/* Step 4: Calculate SAD around the Median prediction. |
2625 |
|
MinSAD=SAD |
2626 |
|
If Motion Vector equal to Previous frame motion vector |
2627 |
|
and MinSAD<PrevFrmSAD goto Step 10. |
2628 |
|
If SAD<=256 goto Step 10. |
2629 |
|
*/ |
2630 |
|
|
2631 |
|
if (currMV->x > max_dx) { |
2632 |
|
currMV->x = EVEN(max_dx); |
2633 |
|
} |
2634 |
|
if (currMV->x < min_dx) { |
2635 |
|
currMV->x = EVEN(min_dx); |
2636 |
|
} |
2637 |
|
if (currMV->y > max_dy) { |
2638 |
|
currMV->y = EVEN(max_dy); |
2639 |
|
} |
2640 |
|
if (currMV->y < min_dy) { |
2641 |
|
currMV->y = EVEN(min_dy); |
2642 |
|
} |
2643 |
|
|
2644 |
|
iMinSAD = |
2645 |
|
sad16(cur, |
2646 |
|
get_iref_mv(pRef, x, y, 16, currMV, |
2647 |
|
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2648 |
|
iMinSAD += |
2649 |
|
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2650 |
|
(uint8_t) iFcode, iQuant); |
2651 |
|
|
2652 |
|
if ((iMinSAD < 256) || |
2653 |
|
((MVequal(*currMV, prevMB->i_mvs[0])) && |
2654 |
|
((int32_t) iMinSAD < prevMB->i_sad16))) { |
2655 |
|
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
2656 |
|
{ |
2657 |
|
if (!MVzero(*currMV)) { |
2658 |
|
iMinSAD += MV16_00_BIAS; |
2659 |
|
CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures |
2660 |
|
iMinSAD -= MV16_00_BIAS; |
2661 |
|
} |
2662 |
|
} |
2663 |
|
|
2664 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
2665 |
|
goto PMVfastInt16_Terminate_with_Refine; |
2666 |
|
} |
2667 |
|
|
2668 |
|
|
2669 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
2670 |
|
vector of the median. |
2671 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
2672 |
|
*/ |
2673 |
|
|
2674 |
|
if ((bPredEq) && (MVequal(pmv[0], prevMB->i_mvs[0]))) |
2675 |
|
iFound = 2; |
2676 |
|
|
2677 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
2678 |
|
Otherwise select large Diamond Search. |
2679 |
|
*/ |
2680 |
|
|
2681 |
|
if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) |
2682 |
|
iDiamondSize = 2; // halfpel units! |
2683 |
|
else |
2684 |
|
iDiamondSize = 4; // halfpel units! |
2685 |
|
|
2686 |
|
/* |
2687 |
|
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
2688 |
|
Also calculate (0,0) but do not subtract offset. |
2689 |
|
Let MinSAD be the smallest SAD up to this point. |
2690 |
|
If MV is (0,0) subtract offset. |
2691 |
|
*/ |
2692 |
|
|
2693 |
|
// (0,0) is often a good choice |
2694 |
|
|
2695 |
|
if (!MVzero(pmv[0])) |
2696 |
|
CHECK_MV16_ZERO; |
2697 |
|
|
2698 |
|
// previous frame MV is always possible |
2699 |
|
|
2700 |
|
if (!MVzero(prevMB->i_mvs[0])) |
2701 |
|
if (!MVequal(prevMB->i_mvs[0], pmv[0])) |
2702 |
|
CHECK_MV16_CANDIDATE(prevMB->i_mvs[0].x, prevMB->i_mvs[0].y); |
2703 |
|
|
2704 |
|
// left neighbour, if allowed |
2705 |
|
|
2706 |
|
if (!MVzero(pmv[1])) |
2707 |
|
if (!MVequal(pmv[1], prevMB->i_mvs[0])) |
2708 |
|
if (!MVequal(pmv[1], pmv[0])) |
2709 |
|
CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); |
2710 |
|
|
2711 |
|
// top neighbour, if allowed |
2712 |
|
if (!MVzero(pmv[2])) |
2713 |
|
if (!MVequal(pmv[2], prevMB->i_mvs[0])) |
2714 |
|
if (!MVequal(pmv[2], pmv[0])) |
2715 |
|
if (!MVequal(pmv[2], pmv[1])) |
2716 |
|
CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); |
2717 |
|
|
2718 |
|
// top right neighbour, if allowed |
2719 |
|
if (!MVzero(pmv[3])) |
2720 |
|
if (!MVequal(pmv[3], prevMB->i_mvs[0])) |
2721 |
|
if (!MVequal(pmv[3], pmv[0])) |
2722 |
|
if (!MVequal(pmv[3], pmv[1])) |
2723 |
|
if (!MVequal(pmv[3], pmv[2])) |
2724 |
|
CHECK_MV16_CANDIDATE(pmv[3].x, |
2725 |
|
pmv[3].y); |
2726 |
|
|
2727 |
|
if ((MVzero(*currMV)) && |
2728 |
|
(!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
2729 |
|
iMinSAD -= MV16_00_BIAS; |
2730 |
|
|
2731 |
|
|
2732 |
|
/* Step 6: If MinSAD <= thresa goto Step 10. |
2733 |
|
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
2734 |
|
*/ |
2735 |
|
|
2736 |
|
if ((iMinSAD <= threshA) || |
2737 |
|
(MVequal(*currMV, prevMB->i_mvs[0]) && |
2738 |
|
((int32_t) iMinSAD < prevMB->i_sad16))) { |
2739 |
|
|
2740 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
2741 |
|
goto PMVfastInt16_Terminate_with_Refine; |
2742 |
|
} |
2743 |
|
|
2744 |
|
|
2745 |
|
/************ (Diamond Search) **************/ |
2746 |
|
/* |
2747 |
|
Step 7: Perform Diamond search, with either the small or large diamond. |
2748 |
|
If Found=2 only examine one Diamond pattern, and afterwards goto step 10 |
2749 |
|
Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. |
2750 |
|
If center then goto step 10. |
2751 |
|
Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. |
2752 |
|
Refine by using small diamond and goto step 10. |
2753 |
|
*/ |
2754 |
|
|
2755 |
|
if (MotionFlags & PMV_USESQUARES16) |
2756 |
|
MainSearchPtr = Square16_MainSearch; |
2757 |
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
2758 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
2759 |
|
else |
2760 |
|
MainSearchPtr = Diamond16_MainSearch; |
2761 |
|
|
2762 |
|
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2763 |
|
|
2764 |
|
|
2765 |
|
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
2766 |
|
iSAD = |
2767 |
|
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2768 |
|
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2769 |
|
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2770 |
|
iQuant, iFound); |
2771 |
|
|
2772 |
|
if (iSAD < iMinSAD) { |
2773 |
|
*currMV = newMV; |
2774 |
|
iMinSAD = iSAD; |
2775 |
|
} |
2776 |
|
|
2777 |
|
if (MotionFlags & PMV_EXTSEARCH16) { |
2778 |
|
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
2779 |
|
|
2780 |
|
if (!(MVequal(pmv[0], backupMV))) { |
2781 |
|
iSAD = |
2782 |
|
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2783 |
|
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2784 |
|
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2785 |
|
iDiamondSize, iFcode, iQuant, iFound); |
2786 |
|
|
2787 |
|
if (iSAD < iMinSAD) { |
2788 |
|
*currMV = newMV; |
2789 |
|
iMinSAD = iSAD; |
2790 |
|
} |
2791 |
|
} |
2792 |
|
|
2793 |
|
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2794 |
|
iSAD = |
2795 |
|
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2796 |
|
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2797 |
|
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2798 |
|
iQuant, iFound); |
2799 |
|
|
2800 |
|
if (iSAD < iMinSAD) { |
2801 |
|
*currMV = newMV; |
2802 |
|
iMinSAD = iSAD; |
2803 |
|
} |
2804 |
|
} |
2805 |
|
} |
2806 |
|
|
2807 |
|
/* |
2808 |
|
Step 10: The motion vector is chosen according to the block corresponding to MinSAD. |
2809 |
|
*/ |
2810 |
|
|
2811 |
|
PMVfastInt16_Terminate_with_Refine: |
2812 |
|
|
2813 |
|
pMB->i_mvs[0] = pMB->i_mvs[1] = pMB->i_mvs[2] = pMB->i_mvs[3] = pMB->i_mv16 = *currMV; |
2814 |
|
pMB->i_sad8[0] = pMB->i_sad8[1] = pMB->i_sad8[2] = pMB->i_sad8[3] = pMB->i_sad16 = iMinSAD; |
2815 |
|
|
2816 |
|
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
2817 |
|
iMinSAD = |
2818 |
|
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2819 |
|
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2820 |
|
iFcode, iQuant, iEdgedWidth); |
2821 |
|
|
2822 |
|
pmv[0] = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); // get _REAL_ prediction (halfpel possible) |
2823 |
|
|
2824 |
|
PMVfastInt16_Terminate_without_Refine: |
2825 |
|
currPMV->x = currMV->x - center_x; |
2826 |
|
currPMV->y = currMV->y - center_y; |
2827 |
|
return iMinSAD; |
2828 |
|
} |
2829 |
|
|
2830 |
|
|
2831 |
|
|
2832 |
/* *********************************************************** |
/* *********************************************************** |
2835 |
***************************************************************/ |
***************************************************************/ |
2836 |
|
|
2837 |
|
|
2838 |
|
#define DIRECT_PENALTY 0 |
2839 |
|
#define DIRECT_UPPERLIMIT 256 // never use direct mode if SAD is larger than this |
2840 |
|
|
2841 |
void |
void |
2842 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
2843 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
2844 |
|
const int32_t time_bp, |
2845 |
|
const int32_t time_pp, |
2846 |
// forward (past) reference |
// forward (past) reference |
2847 |
const MACROBLOCK * const f_mbs, |
const MACROBLOCK * const f_mbs, |
2848 |
const IMAGE * const f_ref, |
const IMAGE * const f_ref, |
2856 |
const IMAGE * const b_refV, |
const IMAGE * const b_refV, |
2857 |
const IMAGE * const b_refHV) |
const IMAGE * const b_refHV) |
2858 |
{ |
{ |
2859 |
const uint32_t mb_width = pParam->mb_width; |
const int mb_width = pParam->mb_width; |
2860 |
const uint32_t mb_height = pParam->mb_height; |
const int mb_height = pParam->mb_height; |
2861 |
const int32_t edged_width = pParam->edged_width; |
const int edged_width = pParam->edged_width; |
2862 |
|
|
2863 |
uint32_t i, j; |
int i, j, k; |
|
|
|
|
int32_t f_sad16; |
|
|
int32_t b_sad16; |
|
|
int32_t i_sad16; |
|
|
int32_t d_sad16; |
|
|
int32_t best_sad; |
|
2864 |
|
|
2865 |
|
static const VECTOR zeroMV={0,0}; |
2866 |
|
|
2867 |
|
int f_sad16; /* forward (as usual) search */ |
2868 |
|
int b_sad16; /* backward (only in b-frames) search */ |
2869 |
|
int i_sad16; /* interpolated (both direction, b-frames only) */ |
2870 |
|
int d_sad16; /* direct mode (assume linear motion) */ |
2871 |
|
|
2872 |
|
int best_sad; |
2873 |
|
|
2874 |
|
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
2875 |
VECTOR pmv_dontcare; |
VECTOR pmv_dontcare; |
2876 |
|
|
2877 |
|
int f_count=0; |
2878 |
|
int b_count=0; |
2879 |
|
int i_count=0; |
2880 |
|
int d_count=0; |
2881 |
|
int s_count=0; |
2882 |
|
|
2883 |
|
const int64_t TRB = (int32_t)time_pp - (int32_t)time_bp; |
2884 |
|
const int64_t TRD = (int32_t)time_pp; |
2885 |
|
|
2886 |
|
// fprintf(stderr,"TRB = %lld TRD = %lld time_bp =%d time_pp =%d\n\n",TRB,TRD,time_bp,time_pp); |
2887 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
2888 |
for (j = 0; j < mb_height; j++) { |
for (j = 0; j < mb_height; j++) { |
2889 |
|
|
2890 |
|
f_predMV = zeroMV; /* prediction is reset at left boundary */ |
2891 |
|
b_predMV = zeroMV; |
2892 |
|
|
2893 |
for (i = 0; i < mb_width; i++) { |
for (i = 0; i < mb_width; i++) { |
2894 |
MACROBLOCK *mb = &frame->mbs[i + j * mb_width]; |
MACROBLOCK *mb = &frame->mbs[i + j * mb_width]; |
2895 |
const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; |
const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; |
2896 |
const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; |
const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; |
2897 |
|
|
2898 |
|
mb->deltamv=zeroMV; |
2899 |
|
|
2900 |
|
/* special case, if collocated block is SKIPed: encoding is forward(0,0) */ |
2901 |
|
|
2902 |
|
#ifndef _DISABLE_SKIP |
2903 |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
2904 |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
2905 |
mb->mode = MODE_NOT_CODED; |
mb->mode = MODE_NOT_CODED; |
2909 |
mb->b_mvs[0].y = 0; |
mb->b_mvs[0].y = 0; |
2910 |
continue; |
continue; |
2911 |
} |
} |
2912 |
|
#endif |
2913 |
|
|
2914 |
|
d_sad16 = DIRECT_PENALTY; |
2915 |
|
|
2916 |
|
if (b_mb->mode == MODE_INTER4V) |
2917 |
|
{ |
2918 |
|
|
2919 |
|
/* same method of scaling as in decoder.c, so we copy from there */ |
2920 |
|
for (k = 0; k < 4; k++) { |
2921 |
|
|
2922 |
|
mb->directmv[k] = b_mb->mvs[k]; |
2923 |
|
|
2924 |
|
mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
2925 |
|
mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) |
2926 |
|
? ((TRB - TRD) * mb->directmv[k].x) / TRD |
2927 |
|
: mb->mvs[k].x - mb->directmv[k].x); |
2928 |
|
|
2929 |
|
mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
2930 |
|
mb->b_mvs[k].y = (int32_t) ((mb->directmv[k].y == 0) |
2931 |
|
? ((TRB - TRD) * mb->directmv[k].y) / TRD |
2932 |
|
: mb->mvs[k].y - mb->directmv[k].y); |
2933 |
|
|
2934 |
|
d_sad16 += |
2935 |
|
sad8bi(frame->image.y + 2*(i+(k&1))*8 + 2*(j+(k>>1))*8*edged_width, |
2936 |
|
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2937 |
|
2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->mvs[k], edged_width), |
2938 |
|
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
2939 |
|
2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->b_mvs[k], edged_width), |
2940 |
|
edged_width); |
2941 |
|
} |
2942 |
|
} |
2943 |
|
else |
2944 |
|
{ |
2945 |
|
mb->directmv[3] = mb->directmv[2] = mb->directmv[1] = |
2946 |
|
mb->directmv[0] = b_mb->mvs[0]; |
2947 |
|
|
2948 |
|
mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
2949 |
|
mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) |
2950 |
|
? ((TRB - TRD) * mb->directmv[0].x) / TRD |
2951 |
|
: mb->mvs[0].x - mb->directmv[0].x); |
2952 |
|
|
2953 |
|
mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
2954 |
|
mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) |
2955 |
|
? ((TRB - TRD) * mb->directmv[0].y) / TRD |
2956 |
|
: mb->mvs[0].y - mb->directmv[0].y); |
2957 |
|
|
2958 |
|
d_sad16 += sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
2959 |
|
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2960 |
|
i, j, 16, &mb->mvs[0], edged_width), |
2961 |
|
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
2962 |
|
i, j, 16, &mb->b_mvs[0], edged_width), |
2963 |
|
edged_width); |
2964 |
|
|
2965 |
|
} |
2966 |
|
d_sad16 += calc_delta_16(mb->deltamv.x, mb->deltamv.y, 1, frame->quant); |
2967 |
|
|
2968 |
// forward search |
// forward search |
2969 |
f_sad16 = |
f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2970 |
SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
&frame->image, i, j, |
2971 |
&frame->image, i, j, frame->motion_flags, |
mb->mvs[0].x, mb->mvs[0].y, /* start point f_directMV */ |
2972 |
frame->quant, frame->fcode, pParam, f_mbs, |
f_predMV.x, f_predMV.y, /* center is f-prediction */ |
2973 |
f_mbs /* todo */ , |
frame->motion_flags, |
2974 |
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
frame->quant, frame->fcode, pParam, |
2975 |
|
f_mbs, f_mbs, |
2976 |
|
&mb->mvs[0], &pmv_dontcare); |
2977 |
|
|
|
// backward search |
|
|
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, &frame->image, i, j, frame->motion_flags, frame->quant, frame->bcode, pParam, b_mbs, b_mbs, /* todo */ |
|
|
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
|
2978 |
|
|
2979 |
// interpolate search (simple, but effective) |
// backward search |
2980 |
i_sad16 = 65535; |
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
2981 |
|
&frame->image, i, j, |
2982 |
|
mb->b_mvs[0].x, mb->b_mvs[0].y, /* start point b_directMV */ |
2983 |
|
b_predMV.x, b_predMV.y, /* center is b-prediction */ |
2984 |
|
frame->motion_flags, |
2985 |
|
frame->quant, frame->bcode, pParam, |
2986 |
|
b_mbs, b_mbs, |
2987 |
|
&mb->b_mvs[0], &pmv_dontcare); |
2988 |
|
|
|
/* |
|
|
x/y range somewhat buggy |
|
2989 |
i_sad16 = |
i_sad16 = |
2990 |
sad16bi_c(frame->image.y + i * 16 + j * 16 * edged_width, |
sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
2991 |
get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2992 |
i, j, 16, mb->mvs[0].x, mb->mvs[0].y, |
i, j, 16, &mb->mvs[0], edged_width), |
2993 |
edged_width), get_ref(b_ref->y, b_refH->y, |
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
2994 |
b_refV->y, b_refHV->y, |
i, j, 16, &mb->b_mvs[0], edged_width), |
|
i, j, 16, |
|
|
mb->b_mvs[0].x, |
|
|
mb->b_mvs[0].x, |
|
|
edged_width), |
|
2995 |
edged_width); |
edged_width); |
2996 |
*/ |
i_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, |
2997 |
|
frame->fcode, frame->quant); |
2998 |
|
i_sad16 += calc_delta_16(mb->b_mvs[0].x-b_predMV.x, mb->b_mvs[0].y-b_predMV.y, |
2999 |
|
frame->bcode, frame->quant); |
3000 |
|
|
3001 |
// TODO: direct search |
// TODO: direct search |
3002 |
// predictor + range of [-32,32] |
// predictor + delta vector in range [-32,32] (fcode=1) |
|
d_sad16 = 65535; |
|
3003 |
|
|
3004 |
|
i_sad16 = 65535; |
3005 |
|
f_sad16 = 65535; |
3006 |
|
b_sad16 = 65535; |
3007 |
|
// d_sad16 = 65535; |
3008 |
|
|
3009 |
if (f_sad16 < b_sad16) { |
if (f_sad16 < b_sad16) { |
3010 |
best_sad = f_sad16; |
best_sad = f_sad16; |
3020 |
} |
} |
3021 |
|
|
3022 |
if (d_sad16 < best_sad) { |
if (d_sad16 < best_sad) { |
3023 |
|
|
3024 |
|
if (b_mb->mode == MODE_INTER4V) |
3025 |
|
{ |
3026 |
|
|
3027 |
|
/* same method of scaling as in decoder.c, so we copy from there */ |
3028 |
|
for (k = 0; k < 4; k++) { |
3029 |
|
|
3030 |
|
mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
3031 |
|
mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) |
3032 |
|
? ((TRB - TRD) * mb->directmv[k].x) / TRD |
3033 |
|
: mb->mvs[k].x - mb->directmv[k].x); |
3034 |
|
|
3035 |
|
mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
3036 |
|
mb->b_mvs[k].y = (int32_t) ((mb->directmv[k].y == 0) |
3037 |
|
? ((TRB - TRD) * mb->directmv[k].y) / TRD |
3038 |
|
: mb->mvs[k].y - mb->directmv[k].y); |
3039 |
|
} |
3040 |
|
} |
3041 |
|
else |
3042 |
|
{ |
3043 |
|
mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
3044 |
|
|
3045 |
|
mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) |
3046 |
|
? ((TRB - TRD) * mb->directmv[0].x) / TRD |
3047 |
|
: mb->mvs[0].x - mb->directmv[0].x); |
3048 |
|
|
3049 |
|
mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
3050 |
|
|
3051 |
|
mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) |
3052 |
|
? ((TRB - TRD) * mb->directmv[0].y) / TRD |
3053 |
|
: mb->mvs[0].y - mb->directmv[0].y); |
3054 |
|
|
3055 |
|
mb->mvs[3] = mb->mvs[2] = mb->mvs[1] = mb->mvs[0]; |
3056 |
|
mb->b_mvs[3] = mb->b_mvs[2] = mb->b_mvs[1] = mb->b_mvs[0]; |
3057 |
|
} |
3058 |
|
|
3059 |
best_sad = d_sad16; |
best_sad = d_sad16; |
3060 |
mb->mode = MODE_DIRECT; |
mb->mode = MODE_DIRECT; |
3061 |
|
mb->mode = MODE_INTERPOLATE; // direct mode still broken :-( |
3062 |
|
} |
3063 |
|
|
3064 |
|
switch (mb->mode) |
3065 |
|
{ |
3066 |
|
case MODE_FORWARD: |
3067 |
|
f_count++; |
3068 |
|
f_predMV = mb->mvs[0]; |
3069 |
|
break; |
3070 |
|
case MODE_BACKWARD: |
3071 |
|
b_count++; |
3072 |
|
b_predMV = mb->b_mvs[0]; |
3073 |
|
|
3074 |
|
break; |
3075 |
|
case MODE_INTERPOLATE: |
3076 |
|
i_count++; |
3077 |
|
f_predMV = mb->mvs[0]; |
3078 |
|
b_predMV = mb->b_mvs[0]; |
3079 |
|
break; |
3080 |
|
case MODE_DIRECT: |
3081 |
|
d_count++; |
3082 |
|
break; |
3083 |
|
default: |
3084 |
|
s_count++; // ??? |
3085 |
|
break; |
3086 |
} |
} |
3087 |
|
|
3088 |
} |
} |
3089 |
} |
} |
3090 |
|
|
3091 |
|
#ifdef _DEBUG_BFRAME_STAT |
3092 |
|
fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D: %04d S: %04d\n", |
3093 |
|
f_count,b_count,i_count,d_count,s_count); |
3094 |
|
#endif |
3095 |
|
|
3096 |
} |
} |