Parent Directory | Revision Log
Revision 344 - (view) (download)
1 : | Isibaar | 3 | /************************************************************************** |
2 : | * | ||
3 : | chl | 259 | * 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 : | Isibaar | 3 | * Modifications: |
34 : | * | ||
35 : | suxen_drol | 152 | * 01.05.2002 updated MotionEstimationBVOP |
36 : | suxen_drol | 136 | * 25.04.2002 partial prevMB conversion |
37 : | chenm001 | 132 | * 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
38 : | * 14.04.2002 added MotionEstimationBVOP() | ||
39 : | chl | 96 | * 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
40 : | * EPZS and EPZS^2 | ||
41 : | Isibaar | 3 | * 08.02.2002 split up PMVfast into three routines: PMVFast, PMVFast_MainLoop |
42 : | * PMVFast_Refine to support multiple searches with different start points | ||
43 : | edgomez | 78 | * 07.01.2002 uv-block-based interpolation |
44 : | Isibaar | 3 | * 06.01.2002 INTER/INTRA-decision is now done before any SEARCH8 (speedup) |
45 : | edgomez | 78 | * changed INTER_BIAS to 150 (as suggested by suxen_drol) |
46 : | * removed halfpel refinement step in PMVfastSearch8 + quality=5 | ||
47 : | * added new quality mode = 6 which performs halfpel refinement | ||
48 : | * filesize difference between quality 5 and 6 is smaller than 1% | ||
49 : | Isibaar | 3 | * (Isibaar) |
50 : | * 31.12.2001 PMVfastSearch16 and PMVfastSearch8 (gruel) | ||
51 : | edgomez | 78 | * 30.12.2001 get_range/MotionSearchX simplified; blue/green bug fix |
52 : | * 22.12.2001 commented best_point==99 check | ||
53 : | * 19.12.2001 modified get_range (purple bug fix) | ||
54 : | Isibaar | 3 | * 15.12.2001 moved pmv displacement from mbprediction |
55 : | * 02.12.2001 motion estimation/compensation split (Isibaar) | ||
56 : | edgomez | 78 | * 16.11.2001 rewrote/tweaked search algorithms; pross@cs.rmit.edu.au |
57 : | Isibaar | 3 | * 10.11.2001 support for sad16/sad8 functions |
58 : | * 28.08.2001 reactivated MODE_INTER4V for EXT_MODE | ||
59 : | * 24.08.2001 removed MODE_INTER4V_Q, disabled MODE_INTER4V for EXT_MODE | ||
60 : | edgomez | 78 | * 22.08.2001 added MODE_INTER4V_Q |
61 : | Isibaar | 3 | * 20.08.2001 added pragma to get rid of internal compiler error with VC6 |
62 : | * idea by Cyril. Thanks. | ||
63 : | * | ||
64 : | * Michael Militzer <isibaar@videocoding.de> | ||
65 : | * | ||
66 : | **************************************************************************/ | ||
67 : | |||
68 : | #include <assert.h> | ||
69 : | #include <stdio.h> | ||
70 : | chl | 96 | #include <stdlib.h> |
71 : | Isibaar | 3 | |
72 : | #include "../encoder.h" | ||
73 : | #include "../utils/mbfunctions.h" | ||
74 : | #include "../prediction/mbprediction.h" | ||
75 : | #include "../global.h" | ||
76 : | #include "../utils/timer.h" | ||
77 : | suxen_drol | 118 | #include "motion.h" |
78 : | Isibaar | 3 | #include "sad.h" |
79 : | |||
80 : | |||
81 : | |||
82 : | edgomez | 195 | 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), | ||
84 : | (int) (1.49591 + 0.5), (int) (1.68601 + 0.5), | ||
85 : | (int) (1.89187 + 0.5), (int) (2.11542 + 0.5), (int) (2.35878 + 0.5), | ||
86 : | (int) (2.62429 + 0.5), (int) (2.91455 + 0.5), | ||
87 : | (int) (3.23253 + 0.5), (int) (3.58158 + 0.5), (int) (3.96555 + 0.5), | ||
88 : | (int) (4.38887 + 0.5), (int) (4.85673 + 0.5), | ||
89 : | (int) (5.37519 + 0.5), (int) (5.95144 + 0.5), (int) (6.59408 + 0.5), | ||
90 : | (int) (7.31349 + 0.5), (int) (8.12242 + 0.5), | ||
91 : | (int) (9.03669 + 0.5), (int) (10.0763 + 0.5), (int) (11.2669 + 0.5), | ||
92 : | (int) (12.6426 + 0.5), (int) (14.2493 + 0.5), | ||
93 : | (int) (16.1512 + 0.5), (int) (18.442 + 0.5), (int) (21.2656 + 0.5), | ||
94 : | (int) (24.8580 + 0.5), (int) (29.6436 + 0.5), | ||
95 : | (int) (36.4949 + 0.5) | ||
96 : | }; | ||
97 : | chl | 141 | |
98 : | edgomez | 195 | static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now */ |
99 : | chl | 141 | |
100 : | |||
101 : | |||
102 : | Isibaar | 3 | // mv.length table |
103 : | static const uint32_t mvtab[33] = { | ||
104 : | edgomez | 195 | 1, 2, 3, 4, 6, 7, 7, 7, |
105 : | 9, 9, 9, 10, 10, 10, 10, 10, | ||
106 : | 10, 10, 10, 10, 10, 10, 10, 10, | ||
107 : | 10, 11, 11, 11, 11, 11, 11, 12, 12 | ||
108 : | Isibaar | 3 | }; |
109 : | |||
110 : | |||
111 : | edgomez | 195 | static __inline uint32_t |
112 : | mv_bits(int32_t component, | ||
113 : | const uint32_t iFcode) | ||
114 : | Isibaar | 3 | { |
115 : | edgomez | 195 | if (component == 0) |
116 : | Isibaar | 3 | return 1; |
117 : | |||
118 : | edgomez | 195 | if (component < 0) |
119 : | Isibaar | 3 | component = -component; |
120 : | |||
121 : | edgomez | 195 | if (iFcode == 1) { |
122 : | Isibaar | 3 | if (component > 32) |
123 : | edgomez | 195 | component = 32; |
124 : | Isibaar | 3 | |
125 : | return mvtab[component] + 1; | ||
126 : | edgomez | 195 | } |
127 : | Isibaar | 3 | |
128 : | edgomez | 195 | component += (1 << (iFcode - 1)) - 1; |
129 : | component >>= (iFcode - 1); | ||
130 : | Isibaar | 3 | |
131 : | edgomez | 195 | if (component > 32) |
132 : | Isibaar | 3 | component = 32; |
133 : | |||
134 : | edgomez | 195 | return mvtab[component] + 1 + iFcode - 1; |
135 : | Isibaar | 3 | } |
136 : | |||
137 : | |||
138 : | edgomez | 195 | static __inline uint32_t |
139 : | calc_delta_16(const int32_t dx, | ||
140 : | const int32_t dy, | ||
141 : | const uint32_t iFcode, | ||
142 : | const uint32_t iQuant) | ||
143 : | Isibaar | 3 | { |
144 : | edgomez | 195 | return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + |
145 : | mv_bits(dy, iFcode)); | ||
146 : | Isibaar | 3 | } |
147 : | |||
148 : | edgomez | 195 | static __inline uint32_t |
149 : | calc_delta_8(const int32_t dx, | ||
150 : | const int32_t dy, | ||
151 : | const uint32_t iFcode, | ||
152 : | const uint32_t iQuant) | ||
153 : | Isibaar | 3 | { |
154 : | edgomez | 195 | return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + |
155 : | mv_bits(dy, iFcode)); | ||
156 : | Isibaar | 3 | } |
157 : | chl | 326 | |
158 : | edgomez | 195 | bool |
159 : | MotionEstimation(MBParam * const pParam, | ||
160 : | FRAMEINFO * const current, | ||
161 : | FRAMEINFO * const reference, | ||
162 : | const IMAGE * const pRefH, | ||
163 : | const IMAGE * const pRefV, | ||
164 : | const IMAGE * const pRefHV, | ||
165 : | const uint32_t iLimit) | ||
166 : | Isibaar | 3 | { |
167 : | edgomez | 78 | const uint32_t iWcount = pParam->mb_width; |
168 : | const uint32_t iHcount = pParam->mb_height; | ||
169 : | edgomez | 195 | MACROBLOCK *const pMBs = current->mbs; |
170 : | MACROBLOCK *const prevMBs = reference->mbs; | ||
171 : | const IMAGE *const pCurrent = ¤t->image; | ||
172 : | const IMAGE *const pRef = &reference->image; | ||
173 : | suxen_drol | 136 | |
174 : | chl | 317 | static const VECTOR zeroMV = { 0, 0 }; |
175 : | chl | 326 | VECTOR predMV; |
176 : | chl | 184 | |
177 : | chl | 172 | int32_t x, y; |
178 : | int32_t iIntra = 0; | ||
179 : | VECTOR pmv; | ||
180 : | suxen_drol | 136 | |
181 : | chl | 184 | if (sadInit) |
182 : | edgomez | 195 | (*sadInit) (); |
183 : | |||
184 : | chl | 259 | for (y = 0; y < iHcount; y++) { |
185 : | for (x = 0; x < iWcount; x ++) { | ||
186 : | |||
187 : | edgomez | 195 | MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
188 : | Isibaar | 3 | |
189 : | chl | 326 | predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
190 : | |||
191 : | edgomez | 195 | pMB->sad16 = |
192 : | chl | 326 | SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
193 : | x, y, predMV.x, predMV.y, predMV.x, predMV.y, | ||
194 : | current->motion_flags, current->quant, | ||
195 : | edgomez | 195 | current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
196 : | &pMB->pmvs[0]); | ||
197 : | chl | 184 | |
198 : | edgomez | 195 | if (0 < (pMB->sad16 - MV16_INTER_BIAS)) { |
199 : | chl | 172 | int32_t deviation; |
200 : | chl | 184 | |
201 : | edgomez | 195 | deviation = |
202 : | dev16(pCurrent->y + x * 16 + y * 16 * pParam->edged_width, | ||
203 : | pParam->edged_width); | ||
204 : | |||
205 : | if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) { | ||
206 : | chl | 172 | pMB->mode = MODE_INTRA; |
207 : | edgomez | 195 | pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
208 : | pMB->mvs[3] = zeroMV; | ||
209 : | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = | ||
210 : | pMB->sad8[3] = 0; | ||
211 : | |||
212 : | chl | 172 | iIntra++; |
213 : | if (iIntra >= iLimit) | ||
214 : | edgomez | 195 | return 1; |
215 : | |||
216 : | chl | 172 | continue; |
217 : | } | ||
218 : | edgomez | 195 | } |
219 : | |||
220 : | chl | 184 | pmv = pMB->pmvs[0]; |
221 : | if (current->global_flags & XVID_INTER4V) | ||
222 : | edgomez | 195 | if ((!(current->global_flags & XVID_LUMIMASKING) || |
223 : | pMB->dquant == NO_CHANGE)) { | ||
224 : | int32_t sad8 = IMV16X16 * current->quant; | ||
225 : | |||
226 : | chl | 326 | if (sad8 < pMB->sad16) { |
227 : | edgomez | 195 | sad8 += pMB->sad8[0] = |
228 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
229 : | chl | 326 | pCurrent, 2 * x, 2 * y, |
230 : | pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, | ||
231 : | current->motion_flags, | ||
232 : | edgomez | 195 | current->quant, current->fcode, pParam, |
233 : | pMBs, prevMBs, &pMB->mvs[0], | ||
234 : | &pMB->pmvs[0]); | ||
235 : | chl | 326 | } |
236 : | if (sad8 < pMB->sad16) { | ||
237 : | edgomez | 195 | |
238 : | chl | 326 | predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 1); |
239 : | edgomez | 195 | sad8 += pMB->sad8[1] = |
240 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
241 : | chl | 326 | pCurrent, 2 * x + 1, 2 * y, |
242 : | pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, | ||
243 : | current->motion_flags, | ||
244 : | edgomez | 195 | current->quant, current->fcode, pParam, |
245 : | pMBs, prevMBs, &pMB->mvs[1], | ||
246 : | &pMB->pmvs[1]); | ||
247 : | chl | 326 | } |
248 : | if (sad8 < pMB->sad16) { | ||
249 : | predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 2); | ||
250 : | edgomez | 195 | sad8 += pMB->sad8[2] = |
251 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
252 : | chl | 326 | pCurrent, 2 * x, 2 * y + 1, |
253 : | pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, | ||
254 : | current->motion_flags, | ||
255 : | edgomez | 195 | current->quant, current->fcode, pParam, |
256 : | pMBs, prevMBs, &pMB->mvs[2], | ||
257 : | &pMB->pmvs[2]); | ||
258 : | chl | 326 | } |
259 : | if (sad8 < pMB->sad16) { | ||
260 : | predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 3); | ||
261 : | edgomez | 195 | sad8 += pMB->sad8[3] = |
262 : | SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, | ||
263 : | pCurrent, 2 * x + 1, 2 * y + 1, | ||
264 : | chl | 326 | pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
265 : | current->motion_flags, | ||
266 : | current->quant, current->fcode, pParam, | ||
267 : | pMBs, prevMBs, | ||
268 : | &pMB->mvs[3], | ||
269 : | &pMB->pmvs[3]); | ||
270 : | } | ||
271 : | |||
272 : | chl | 184 | /* decide: MODE_INTER or MODE_INTER4V |
273 : | edgomez | 195 | mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
274 : | */ | ||
275 : | |||
276 : | if (sad8 < pMB->sad16) { | ||
277 : | chl | 175 | pMB->mode = MODE_INTER4V; |
278 : | edgomez | 195 | pMB->sad8[0] *= 4; |
279 : | chl | 175 | pMB->sad8[1] *= 4; |
280 : | pMB->sad8[2] *= 4; | ||
281 : | pMB->sad8[3] *= 4; | ||
282 : | continue; | ||
283 : | } | ||
284 : | |||
285 : | edgomez | 195 | } |
286 : | |||
287 : | pMB->mode = MODE_INTER; | ||
288 : | pMB->pmvs[0] = pmv; /* pMB->pmvs[1] = pMB->pmvs[2] = pMB->pmvs[3] are not needed for INTER */ | ||
289 : | 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] = | ||
291 : | pMB->sad16; | ||
292 : | chl | 259 | } |
293 : | } | ||
294 : | edgomez | 195 | |
295 : | return 0; | ||
296 : | Isibaar | 3 | } |
297 : | |||
298 : | chl | 326 | |
299 : | Isibaar | 3 | #define CHECK_MV16_ZERO {\ |
300 : | if ( (0 <= max_dx) && (0 >= min_dx) \ | ||
301 : | && (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); \ | ||
304 : | chl | 326 | iSAD += calc_delta_16(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
305 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
306 : | { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ | ||
307 : | } | ||
308 : | |||
309 : | chl | 96 | #define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
310 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
311 : | chl | 326 | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
312 : | chl | 96 | if (iSAD < iMinSAD) \ |
313 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ | ||
314 : | } | ||
315 : | Isibaar | 3 | |
316 : | #define CHECK_MV16_CANDIDATE(X,Y) { \ | ||
317 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
318 : | && ((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); \ | ||
321 : | chl | 326 | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
322 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
323 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
324 : | } | ||
325 : | |||
326 : | #define CHECK_MV16_CANDIDATE_DIR(X,Y,D) { \ | ||
327 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
328 : | && ((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); \ | ||
331 : | chl | 326 | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
332 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
333 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
334 : | } | ||
335 : | |||
336 : | #define CHECK_MV16_CANDIDATE_FOUND(X,Y,D) { \ | ||
337 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
338 : | && ((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); \ | ||
341 : | chl | 326 | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
342 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
343 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
344 : | } | ||
345 : | |||
346 : | |||
347 : | #define CHECK_MV8_ZERO {\ | ||
348 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ | ||
349 : | chl | 326 | iSAD += calc_delta_8(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
350 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
351 : | { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ | ||
352 : | } | ||
353 : | edgomez | 195 | |
354 : | chl | 96 | #define NOCHECK_MV8_CANDIDATE(X,Y) \ |
355 : | { \ | ||
356 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
357 : | chl | 326 | iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
358 : | chl | 96 | if (iSAD < iMinSAD) \ |
359 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ | ||
360 : | } | ||
361 : | Isibaar | 3 | |
362 : | #define CHECK_MV8_CANDIDATE(X,Y) { \ | ||
363 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
364 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
365 : | { \ | ||
366 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
367 : | chl | 326 | iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
368 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
369 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
370 : | } | ||
371 : | |||
372 : | #define CHECK_MV8_CANDIDATE_DIR(X,Y,D) { \ | ||
373 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
374 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
375 : | { \ | ||
376 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
377 : | chl | 326 | iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
378 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
379 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
380 : | } | ||
381 : | |||
382 : | #define CHECK_MV8_CANDIDATE_FOUND(X,Y,D) { \ | ||
383 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
384 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
385 : | { \ | ||
386 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
387 : | chl | 326 | iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
388 : | Isibaar | 3 | if (iSAD < iMinSAD) \ |
389 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
390 : | } | ||
391 : | |||
392 : | /* too slow and not fully functional at the moment */ | ||
393 : | /* | ||
394 : | int32_t ZeroSearch16( | ||
395 : | const uint8_t * const pRef, | ||
396 : | const uint8_t * const pRefH, | ||
397 : | const uint8_t * const pRefV, | ||
398 : | const uint8_t * const pRefHV, | ||
399 : | const IMAGE * const pCur, | ||
400 : | const int x, const int y, | ||
401 : | const uint32_t MotionFlags, | ||
402 : | suxen_drol | 136 | const uint32_t iQuant, |
403 : | const uint32_t iFcode, | ||
404 : | Isibaar | 3 | MBParam * const pParam, |
405 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
406 : | const MACROBLOCK * const prevMBs, | ||
407 : | Isibaar | 3 | VECTOR * const currMV, |
408 : | VECTOR * const currPMV) | ||
409 : | { | ||
410 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
411 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
412 : | int32_t iSAD; | ||
413 : | suxen_drol | 254 | VECTOR pred; |
414 : | |||
415 : | Isibaar | 3 | |
416 : | suxen_drol | 254 | pred = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
417 : | Isibaar | 3 | |
418 : | iSAD = sad16( cur, | ||
419 : | get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), | ||
420 : | iEdgedWidth, MV_MAX_ERROR); | ||
421 : | if (iSAD <= iQuant * 96) | ||
422 : | iSAD -= MV16_00_BIAS; | ||
423 : | |||
424 : | currMV->x = 0; | ||
425 : | currMV->y = 0; | ||
426 : | suxen_drol | 254 | currPMV->x = -pred.x; |
427 : | currPMV->y = -pred.y; | ||
428 : | Isibaar | 3 | |
429 : | return iSAD; | ||
430 : | |||
431 : | } | ||
432 : | */ | ||
433 : | |||
434 : | edgomez | 195 | int32_t |
435 : | Diamond16_MainSearch(const uint8_t * const pRef, | ||
436 : | const uint8_t * const pRefH, | ||
437 : | const uint8_t * const pRefV, | ||
438 : | const uint8_t * const pRefHV, | ||
439 : | const uint8_t * const cur, | ||
440 : | const int x, | ||
441 : | const int y, | ||
442 : | chl | 326 | const int start_x, |
443 : | const int start_y, | ||
444 : | int iMinSAD, | ||
445 : | VECTOR * const currMV, | ||
446 : | const int center_x, | ||
447 : | const int center_y, | ||
448 : | edgomez | 195 | const int32_t min_dx, |
449 : | const int32_t max_dx, | ||
450 : | const int32_t min_dy, | ||
451 : | const int32_t max_dy, | ||
452 : | const int32_t iEdgedWidth, | ||
453 : | const int32_t iDiamondSize, | ||
454 : | const int32_t iFcode, | ||
455 : | const int32_t iQuant, | ||
456 : | int iFound) | ||
457 : | Isibaar | 3 | { |
458 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
459 : | |||
460 : | edgomez | 195 | int32_t iDirection = 0; |
461 : | chl | 344 | int32_t iDirectionBackup; |
462 : | Isibaar | 3 | int32_t iSAD; |
463 : | VECTOR backupMV; | ||
464 : | edgomez | 195 | |
465 : | chl | 326 | backupMV.x = start_x; |
466 : | backupMV.y = start_y; | ||
467 : | edgomez | 195 | |
468 : | Isibaar | 3 | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
469 : | |||
470 : | edgomez | 195 | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
471 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); | ||
472 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); | ||
473 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); | ||
474 : | Isibaar | 3 | |
475 : | chl | 344 | if (iDirection) { |
476 : | edgomez | 195 | while (!iFound) { |
477 : | iFound = 1; | ||
478 : | backupMV = *currMV; | ||
479 : | chl | 344 | iDirectionBackup = iDirection; |
480 : | edgomez | 195 | |
481 : | chl | 344 | if (iDirectionBackup != 2) |
482 : | edgomez | 195 | CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
483 : | backupMV.y, 1); | ||
484 : | chl | 344 | if (iDirectionBackup != 1) |
485 : | edgomez | 195 | CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
486 : | backupMV.y, 2); | ||
487 : | chl | 344 | if (iDirectionBackup != 4) |
488 : | edgomez | 195 | CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
489 : | backupMV.y - iDiamondSize, 3); | ||
490 : | chl | 344 | if (iDirectionBackup != 3) |
491 : | edgomez | 195 | CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
492 : | backupMV.y + iDiamondSize, 4); | ||
493 : | chl | 344 | } |
494 : | edgomez | 195 | } else { |
495 : | chl | 326 | currMV->x = start_x; |
496 : | currMV->y = start_y; | ||
497 : | edgomez | 78 | } |
498 : | Isibaar | 3 | return iMinSAD; |
499 : | } | ||
500 : | |||
501 : | edgomez | 195 | int32_t |
502 : | Square16_MainSearch(const uint8_t * const pRef, | ||
503 : | chl | 96 | const uint8_t * const pRefH, |
504 : | const uint8_t * const pRefV, | ||
505 : | const uint8_t * const pRefHV, | ||
506 : | const uint8_t * const cur, | ||
507 : | edgomez | 195 | const int x, |
508 : | const int y, | ||
509 : | chl | 326 | const int start_x, |
510 : | const int start_y, | ||
511 : | int iMinSAD, | ||
512 : | VECTOR * const currMV, | ||
513 : | const int center_x, | ||
514 : | const int center_y, | ||
515 : | edgomez | 195 | const int32_t min_dx, |
516 : | const int32_t max_dx, | ||
517 : | const int32_t min_dy, | ||
518 : | const int32_t max_dy, | ||
519 : | const int32_t iEdgedWidth, | ||
520 : | const int32_t iDiamondSize, | ||
521 : | chl | 96 | const int32_t iFcode, |
522 : | const int32_t iQuant, | ||
523 : | int iFound) | ||
524 : | { | ||
525 : | /* Do a square search around given starting point, return SAD of best */ | ||
526 : | |||
527 : | edgomez | 195 | int32_t iDirection = 0; |
528 : | chl | 96 | int32_t iSAD; |
529 : | VECTOR backupMV; | ||
530 : | edgomez | 195 | |
531 : | chl | 326 | backupMV.x = start_x; |
532 : | backupMV.y = start_y; | ||
533 : | edgomez | 195 | |
534 : | chl | 96 | /* It's one search with full square pattern, and new parts for all following diamonds */ |
535 : | |||
536 : | /* new direction are extra, so 1-4 is normal diamond | ||
537 : | 537 | ||
538 : | 1*2 | ||
539 : | 648 | ||
540 : | */ | ||
541 : | |||
542 : | edgomez | 195 | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
543 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); | ||
544 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); | ||
545 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); | ||
546 : | chl | 96 | |
547 : | edgomez | 195 | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
548 : | backupMV.y - iDiamondSize, 5); | ||
549 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
550 : | backupMV.y + iDiamondSize, 6); | ||
551 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
552 : | backupMV.y - iDiamondSize, 7); | ||
553 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
554 : | backupMV.y + iDiamondSize, 8); | ||
555 : | chl | 96 | |
556 : | edgomez | 195 | |
557 : | chl | 96 | if (iDirection) |
558 : | edgomez | 195 | while (!iFound) { |
559 : | iFound = 1; | ||
560 : | backupMV = *currMV; | ||
561 : | chl | 96 | |
562 : | edgomez | 195 | switch (iDirection) { |
563 : | case 1: | ||
564 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, | ||
565 : | backupMV.y, 1); | ||
566 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
567 : | backupMV.y - iDiamondSize, 5); | ||
568 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
569 : | backupMV.y - iDiamondSize, 7); | ||
570 : | break; | ||
571 : | case 2: | ||
572 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, | ||
573 : | 2); | ||
574 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
575 : | backupMV.y + iDiamondSize, 6); | ||
576 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
577 : | backupMV.y + iDiamondSize, 8); | ||
578 : | break; | ||
579 : | |||
580 : | case 3: | ||
581 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, | ||
582 : | 4); | ||
583 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
584 : | backupMV.y - iDiamondSize, 7); | ||
585 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
586 : | backupMV.y + iDiamondSize, 8); | ||
587 : | break; | ||
588 : | |||
589 : | case 4: | ||
590 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, | ||
591 : | 3); | ||
592 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
593 : | backupMV.y - iDiamondSize, 5); | ||
594 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
595 : | backupMV.y + iDiamondSize, 6); | ||
596 : | break; | ||
597 : | |||
598 : | case 5: | ||
599 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, | ||
600 : | 1); | ||
601 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, | ||
602 : | 3); | ||
603 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
604 : | backupMV.y - iDiamondSize, 5); | ||
605 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
606 : | backupMV.y + iDiamondSize, 6); | ||
607 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
608 : | backupMV.y - iDiamondSize, 7); | ||
609 : | break; | ||
610 : | |||
611 : | case 6: | ||
612 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, | ||
613 : | 2); | ||
614 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, | ||
615 : | 3); | ||
616 : | |||
617 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
618 : | backupMV.y - iDiamondSize, 5); | ||
619 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
620 : | backupMV.y + iDiamondSize, 6); | ||
621 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
622 : | backupMV.y + iDiamondSize, 8); | ||
623 : | |||
624 : | break; | ||
625 : | |||
626 : | case 7: | ||
627 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, | ||
628 : | backupMV.y, 1); | ||
629 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, | ||
630 : | 4); | ||
631 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
632 : | backupMV.y - iDiamondSize, 5); | ||
633 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
634 : | backupMV.y - iDiamondSize, 7); | ||
635 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
636 : | backupMV.y + iDiamondSize, 8); | ||
637 : | break; | ||
638 : | |||
639 : | case 8: | ||
640 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, | ||
641 : | 2); | ||
642 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, | ||
643 : | 4); | ||
644 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
645 : | backupMV.y + iDiamondSize, 6); | ||
646 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
647 : | backupMV.y - iDiamondSize, 7); | ||
648 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
649 : | backupMV.y + iDiamondSize, 8); | ||
650 : | break; | ||
651 : | chl | 96 | default: |
652 : | edgomez | 195 | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
653 : | 1); | ||
654 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, | ||
655 : | 2); | ||
656 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, | ||
657 : | 3); | ||
658 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, | ||
659 : | 4); | ||
660 : | |||
661 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
662 : | backupMV.y - iDiamondSize, 5); | ||
663 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, | ||
664 : | backupMV.y + iDiamondSize, 6); | ||
665 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
666 : | backupMV.y - iDiamondSize, 7); | ||
667 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, | ||
668 : | backupMV.y + iDiamondSize, 8); | ||
669 : | break; | ||
670 : | chl | 96 | } |
671 : | edgomez | 195 | } else { |
672 : | chl | 326 | currMV->x = start_x; |
673 : | currMV->y = start_y; | ||
674 : | edgomez | 195 | } |
675 : | chl | 96 | return iMinSAD; |
676 : | } | ||
677 : | |||
678 : | |||
679 : | edgomez | 195 | int32_t |
680 : | Full16_MainSearch(const uint8_t * const pRef, | ||
681 : | const uint8_t * const pRefH, | ||
682 : | const uint8_t * const pRefV, | ||
683 : | const uint8_t * const pRefHV, | ||
684 : | const uint8_t * const cur, | ||
685 : | const int x, | ||
686 : | const int y, | ||
687 : | chl | 326 | const int start_x, |
688 : | const int start_y, | ||
689 : | int iMinSAD, | ||
690 : | VECTOR * const currMV, | ||
691 : | const int center_x, | ||
692 : | const int center_y, | ||
693 : | edgomez | 195 | const int32_t min_dx, |
694 : | const int32_t max_dx, | ||
695 : | const int32_t min_dy, | ||
696 : | const int32_t max_dy, | ||
697 : | const int32_t iEdgedWidth, | ||
698 : | const int32_t iDiamondSize, | ||
699 : | const int32_t iFcode, | ||
700 : | const int32_t iQuant, | ||
701 : | int iFound) | ||
702 : | chl | 96 | { |
703 : | int32_t iSAD; | ||
704 : | edgomez | 195 | int32_t dx, dy; |
705 : | chl | 96 | VECTOR backupMV; |
706 : | edgomez | 195 | |
707 : | chl | 326 | backupMV.x = start_x; |
708 : | backupMV.y = start_y; | ||
709 : | chl | 96 | |
710 : | edgomez | 195 | for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
711 : | for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) | ||
712 : | NOCHECK_MV16_CANDIDATE(dx, dy); | ||
713 : | |||
714 : | chl | 96 | return iMinSAD; |
715 : | } | ||
716 : | |||
717 : | edgomez | 195 | int32_t |
718 : | AdvDiamond16_MainSearch(const uint8_t * const pRef, | ||
719 : | const uint8_t * const pRefH, | ||
720 : | const uint8_t * const pRefV, | ||
721 : | const uint8_t * const pRefHV, | ||
722 : | const uint8_t * const cur, | ||
723 : | const int x, | ||
724 : | const int y, | ||
725 : | chl | 326 | int start_x, |
726 : | int start_y, | ||
727 : | int iMinSAD, | ||
728 : | VECTOR * const currMV, | ||
729 : | const int center_x, | ||
730 : | const int center_y, | ||
731 : | edgomez | 195 | const int32_t min_dx, |
732 : | const int32_t max_dx, | ||
733 : | const int32_t min_dy, | ||
734 : | const int32_t max_dy, | ||
735 : | const int32_t iEdgedWidth, | ||
736 : | const int32_t iDiamondSize, | ||
737 : | const int32_t iFcode, | ||
738 : | const int32_t iQuant, | ||
739 : | int iDirection) | ||
740 : | chl | 181 | { |
741 : | |||
742 : | int32_t iSAD; | ||
743 : | |||
744 : | /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ | ||
745 : | |||
746 : | edgomez | 195 | if (iDirection) { |
747 : | chl | 326 | CHECK_MV16_CANDIDATE(start_x - iDiamondSize, start_y); |
748 : | CHECK_MV16_CANDIDATE(start_x + iDiamondSize, start_y); | ||
749 : | CHECK_MV16_CANDIDATE(start_x, start_y - iDiamondSize); | ||
750 : | CHECK_MV16_CANDIDATE(start_x, start_y + iDiamondSize); | ||
751 : | edgomez | 195 | } else { |
752 : | int bDirection = 1 + 2 + 4 + 8; | ||
753 : | |||
754 : | do { | ||
755 : | chl | 181 | iDirection = 0; |
756 : | edgomez | 195 | 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) |
757 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
758 : | chl | 181 | |
759 : | edgomez | 195 | if (bDirection & 2) |
760 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
761 : | chl | 181 | |
762 : | edgomez | 195 | if (bDirection & 4) |
763 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
764 : | chl | 181 | |
765 : | edgomez | 195 | if (bDirection & 8) |
766 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
767 : | chl | 181 | |
768 : | /* now we're doing diagonal checks near our candidate */ | ||
769 : | |||
770 : | edgomez | 195 | if (iDirection) //checking if anything found |
771 : | chl | 181 | { |
772 : | bDirection = iDirection; | ||
773 : | iDirection = 0; | ||
774 : | chl | 326 | start_x = currMV->x; |
775 : | start_y = currMV->y; | ||
776 : | edgomez | 195 | if (bDirection & 3) //our candidate is left or right |
777 : | chl | 181 | { |
778 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
779 : | CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); | ||
780 : | edgomez | 195 | } else // what remains here is up or down |
781 : | chl | 181 | { |
782 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
783 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); | ||
784 : | chl | 181 | } |
785 : | |||
786 : | edgomez | 195 | if (iDirection) { |
787 : | bDirection += iDirection; | ||
788 : | chl | 326 | start_x = currMV->x; |
789 : | start_y = currMV->y; | ||
790 : | chl | 181 | } |
791 : | edgomez | 195 | } else //about to quit, eh? not so fast.... |
792 : | chl | 181 | { |
793 : | edgomez | 195 | switch (bDirection) { |
794 : | chl | 181 | case 2: |
795 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
796 : | start_y - iDiamondSize, 2 + 4); | ||
797 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
798 : | start_y + iDiamondSize, 2 + 8); | ||
799 : | chl | 181 | break; |
800 : | case 1: | ||
801 : | chl | 326 | |
802 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
803 : | start_y - iDiamondSize, 1 + 4); | ||
804 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
805 : | start_y + iDiamondSize, 1 + 8); | ||
806 : | chl | 181 | break; |
807 : | edgomez | 195 | case 2 + 4: |
808 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
809 : | start_y - iDiamondSize, 1 + 4); | ||
810 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
811 : | start_y - iDiamondSize, 2 + 4); | ||
812 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
813 : | start_y + iDiamondSize, 2 + 8); | ||
814 : | chl | 181 | break; |
815 : | case 4: | ||
816 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
817 : | start_y - iDiamondSize, 2 + 4); | ||
818 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
819 : | start_y - iDiamondSize, 1 + 4); | ||
820 : | chl | 181 | break; |
821 : | case 8: | ||
822 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
823 : | start_y + iDiamondSize, 2 + 8); | ||
824 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
825 : | start_y + iDiamondSize, 1 + 8); | ||
826 : | chl | 181 | break; |
827 : | edgomez | 195 | case 1 + 4: |
828 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
829 : | start_y + iDiamondSize, 1 + 8); | ||
830 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
831 : | start_y - iDiamondSize, 1 + 4); | ||
832 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
833 : | start_y - iDiamondSize, 2 + 4); | ||
834 : | chl | 181 | break; |
835 : | edgomez | 195 | case 2 + 8: |
836 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
837 : | start_y - iDiamondSize, 1 + 4); | ||
838 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
839 : | start_y + iDiamondSize, 1 + 8); | ||
840 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
841 : | start_y + iDiamondSize, 2 + 8); | ||
842 : | chl | 181 | break; |
843 : | edgomez | 195 | case 1 + 8: |
844 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
845 : | start_y - iDiamondSize, 2 + 4); | ||
846 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
847 : | start_y + iDiamondSize, 2 + 8); | ||
848 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
849 : | start_y + iDiamondSize, 1 + 8); | ||
850 : | chl | 181 | break; |
851 : | edgomez | 195 | default: //1+2+4+8 == we didn't find anything at all |
852 : | chl | 326 | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
853 : | start_y - iDiamondSize, 1 + 4); | ||
854 : | CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, | ||
855 : | start_y + iDiamondSize, 1 + 8); | ||
856 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
857 : | start_y - iDiamondSize, 2 + 4); | ||
858 : | CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, | ||
859 : | start_y + iDiamondSize, 2 + 8); | ||
860 : | chl | 181 | break; |
861 : | } | ||
862 : | edgomez | 195 | if (!iDirection) |
863 : | break; //ok, the end. really | ||
864 : | else { | ||
865 : | bDirection = iDirection; | ||
866 : | chl | 326 | start_x = currMV->x; |
867 : | start_y = currMV->y; | ||
868 : | chl | 181 | } |
869 : | } | ||
870 : | } | ||
871 : | edgomez | 195 | while (1); //forever |
872 : | chl | 181 | } |
873 : | return iMinSAD; | ||
874 : | } | ||
875 : | |||
876 : | chl | 326 | |
877 : | #define CHECK_MV16_F_INTERPOL(X,Y,BX,BY) { \ | ||
878 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
879 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
880 : | { \ | ||
881 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
882 : | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ | ||
883 : | if (iSAD < iMinSAD) \ | ||
884 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
885 : | } | ||
886 : | |||
887 : | #define CHECK_MV16_F_INTERPOL_DIR(X,Y,BX,BY,D) { \ | ||
888 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
889 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
890 : | { \ | ||
891 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
892 : | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ | ||
893 : | if (iSAD < iMinSAD) \ | ||
894 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
895 : | } | ||
896 : | |||
897 : | #define CHECK_MV16_F_INTERPOL_FOUND(X,Y,BX,BY,D) { \ | ||
898 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
899 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
900 : | { \ | ||
901 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
902 : | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ | ||
903 : | if (iSAD < iMinSAD) \ | ||
904 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
905 : | } | ||
906 : | |||
907 : | |||
908 : | #define CHECK_MV16_B_INTERPOL(FX,FY,X,Y) { \ | ||
909 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
910 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
911 : | { \ | ||
912 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
913 : | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ | ||
914 : | if (iSAD < iMinSAD) \ | ||
915 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
916 : | } | ||
917 : | |||
918 : | |||
919 : | #define CHECK_MV16_B_INTERPOL_DIR(FX,FY,X,Y,D) { \ | ||
920 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
921 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
922 : | { \ | ||
923 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
924 : | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ | ||
925 : | if (iSAD < iMinSAD) \ | ||
926 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
927 : | } | ||
928 : | |||
929 : | |||
930 : | #define CHECK_MV16_B_INTERPOL_FOUND(FX,FY,X,Y,D) { \ | ||
931 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
932 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
933 : | { \ | ||
934 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
935 : | iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ | ||
936 : | if (iSAD < iMinSAD) \ | ||
937 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
938 : | } | ||
939 : | |||
940 : | |||
941 : | #if (0==1) | ||
942 : | edgomez | 195 | int32_t |
943 : | chl | 326 | Diamond16_InterpolMainSearch( |
944 : | const uint8_t * const f_pRef, | ||
945 : | const uint8_t * const f_pRefH, | ||
946 : | const uint8_t * const f_pRefV, | ||
947 : | const uint8_t * const f_pRefHV, | ||
948 : | const uint8_t * const cur, | ||
949 : | |||
950 : | const uint8_t * const b_pRef, | ||
951 : | const uint8_t * const b_pRefH, | ||
952 : | const uint8_t * const b_pRefV, | ||
953 : | const uint8_t * const b_pRefHV, | ||
954 : | |||
955 : | const int x, | ||
956 : | const int y, | ||
957 : | |||
958 : | const int f_start_x, | ||
959 : | const int f_start_y, | ||
960 : | const int b_start_x, | ||
961 : | const int b_start_y, | ||
962 : | |||
963 : | int iMinSAD, | ||
964 : | VECTOR * const f_currMV, | ||
965 : | VECTOR * const b_currMV, | ||
966 : | |||
967 : | const int f_center_x, | ||
968 : | const int f_center_y, | ||
969 : | const int b_center_x, | ||
970 : | const int b_center_y, | ||
971 : | |||
972 : | const int32_t min_dx, | ||
973 : | const int32_t max_dx, | ||
974 : | const int32_t min_dy, | ||
975 : | const int32_t max_dy, | ||
976 : | const int32_t iEdgedWidth, | ||
977 : | const int32_t iDiamondSize, | ||
978 : | |||
979 : | const int32_t f_iFcode, | ||
980 : | const int32_t b_iFcode, | ||
981 : | |||
982 : | const int32_t iQuant, | ||
983 : | int iFound) | ||
984 : | { | ||
985 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
986 : | |||
987 : | int32_t f_iDirection = 0; | ||
988 : | int32_t b_iDirection = 0; | ||
989 : | int32_t iSAD; | ||
990 : | |||
991 : | VECTOR f_backupMV; | ||
992 : | VECTOR b_backupMV; | ||
993 : | |||
994 : | f_backupMV.x = start_x; | ||
995 : | f_backupMV.y = start_y; | ||
996 : | b_backupMV.x = start_x; | ||
997 : | b_backupMV.y = start_y; | ||
998 : | |||
999 : | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ | ||
1000 : | |||
1001 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); | ||
1002 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); | ||
1003 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); | ||
1004 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); | ||
1005 : | |||
1006 : | if (iDirection) | ||
1007 : | while (!iFound) { | ||
1008 : | iFound = 1; | ||
1009 : | backupMV = *currMV; | ||
1010 : | |||
1011 : | if (iDirection != 2) | ||
1012 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, | ||
1013 : | backupMV.y, 1); | ||
1014 : | if (iDirection != 1) | ||
1015 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, | ||
1016 : | backupMV.y, 2); | ||
1017 : | if (iDirection != 4) | ||
1018 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x, | ||
1019 : | backupMV.y - iDiamondSize, 3); | ||
1020 : | if (iDirection != 3) | ||
1021 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x, | ||
1022 : | backupMV.y + iDiamondSize, 4); | ||
1023 : | } else { | ||
1024 : | currMV->x = start_x; | ||
1025 : | currMV->y = start_y; | ||
1026 : | } | ||
1027 : | return iMinSAD; | ||
1028 : | } | ||
1029 : | #endif | ||
1030 : | |||
1031 : | |||
1032 : | int32_t | ||
1033 : | edgomez | 195 | AdvDiamond8_MainSearch(const uint8_t * const pRef, |
1034 : | const uint8_t * const pRefH, | ||
1035 : | const uint8_t * const pRefV, | ||
1036 : | const uint8_t * const pRefHV, | ||
1037 : | const uint8_t * const cur, | ||
1038 : | const int x, | ||
1039 : | const int y, | ||
1040 : | chl | 326 | int start_x, |
1041 : | int start_y, | ||
1042 : | int iMinSAD, | ||
1043 : | edgomez | 195 | VECTOR * const currMV, |
1044 : | chl | 326 | const int center_x, |
1045 : | const int center_y, | ||
1046 : | edgomez | 195 | const int32_t min_dx, |
1047 : | const int32_t max_dx, | ||
1048 : | const int32_t min_dy, | ||
1049 : | const int32_t max_dy, | ||
1050 : | const int32_t iEdgedWidth, | ||
1051 : | const int32_t iDiamondSize, | ||
1052 : | const int32_t iFcode, | ||
1053 : | const int32_t iQuant, | ||
1054 : | int iDirection) | ||
1055 : | chl | 181 | { |
1056 : | |||
1057 : | int32_t iSAD; | ||
1058 : | |||
1059 : | /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ | ||
1060 : | |||
1061 : | edgomez | 195 | if (iDirection) { |
1062 : | chl | 326 | CHECK_MV8_CANDIDATE(start_x - iDiamondSize, start_y); |
1063 : | CHECK_MV8_CANDIDATE(start_x + iDiamondSize, start_y); | ||
1064 : | CHECK_MV8_CANDIDATE(start_x, start_y - iDiamondSize); | ||
1065 : | CHECK_MV8_CANDIDATE(start_x, start_y + iDiamondSize); | ||
1066 : | edgomez | 195 | } else { |
1067 : | int bDirection = 1 + 2 + 4 + 8; | ||
1068 : | |||
1069 : | do { | ||
1070 : | chl | 181 | iDirection = 0; |
1071 : | edgomez | 195 | 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) |
1072 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
1073 : | chl | 181 | |
1074 : | edgomez | 195 | if (bDirection & 2) |
1075 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
1076 : | chl | 181 | |
1077 : | edgomez | 195 | if (bDirection & 4) |
1078 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
1079 : | chl | 181 | |
1080 : | edgomez | 195 | if (bDirection & 8) |
1081 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
1082 : | chl | 181 | |
1083 : | /* now we're doing diagonal checks near our candidate */ | ||
1084 : | |||
1085 : | edgomez | 195 | if (iDirection) //checking if anything found |
1086 : | chl | 181 | { |
1087 : | bDirection = iDirection; | ||
1088 : | iDirection = 0; | ||
1089 : | chl | 326 | start_x = currMV->x; |
1090 : | start_y = currMV->y; | ||
1091 : | edgomez | 195 | if (bDirection & 3) //our candidate is left or right |
1092 : | chl | 181 | { |
1093 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
1094 : | CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); | ||
1095 : | edgomez | 195 | } else // what remains here is up or down |
1096 : | chl | 181 | { |
1097 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
1098 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); | ||
1099 : | chl | 181 | } |
1100 : | |||
1101 : | edgomez | 195 | if (iDirection) { |
1102 : | bDirection += iDirection; | ||
1103 : | chl | 326 | start_x = currMV->x; |
1104 : | start_y = currMV->y; | ||
1105 : | chl | 181 | } |
1106 : | edgomez | 195 | } else //about to quit, eh? not so fast.... |
1107 : | chl | 181 | { |
1108 : | edgomez | 195 | switch (bDirection) { |
1109 : | chl | 181 | case 2: |
1110 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1111 : | start_y - iDiamondSize, 2 + 4); | ||
1112 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1113 : | start_y + iDiamondSize, 2 + 8); | ||
1114 : | chl | 181 | break; |
1115 : | case 1: | ||
1116 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1117 : | start_y - iDiamondSize, 1 + 4); | ||
1118 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1119 : | start_y + iDiamondSize, 1 + 8); | ||
1120 : | chl | 181 | break; |
1121 : | edgomez | 195 | case 2 + 4: |
1122 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1123 : | start_y - iDiamondSize, 1 + 4); | ||
1124 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1125 : | start_y - iDiamondSize, 2 + 4); | ||
1126 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1127 : | start_y + iDiamondSize, 2 + 8); | ||
1128 : | chl | 181 | break; |
1129 : | case 4: | ||
1130 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1131 : | start_y - iDiamondSize, 2 + 4); | ||
1132 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1133 : | start_y - iDiamondSize, 1 + 4); | ||
1134 : | chl | 181 | break; |
1135 : | case 8: | ||
1136 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1137 : | start_y + iDiamondSize, 2 + 8); | ||
1138 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1139 : | start_y + iDiamondSize, 1 + 8); | ||
1140 : | chl | 181 | break; |
1141 : | edgomez | 195 | case 1 + 4: |
1142 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1143 : | start_y + iDiamondSize, 1 + 8); | ||
1144 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1145 : | start_y - iDiamondSize, 1 + 4); | ||
1146 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1147 : | start_y - iDiamondSize, 2 + 4); | ||
1148 : | chl | 181 | break; |
1149 : | edgomez | 195 | case 2 + 8: |
1150 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1151 : | start_y - iDiamondSize, 1 + 4); | ||
1152 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1153 : | start_y + iDiamondSize, 1 + 8); | ||
1154 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1155 : | start_y + iDiamondSize, 2 + 8); | ||
1156 : | chl | 181 | break; |
1157 : | edgomez | 195 | case 1 + 8: |
1158 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1159 : | start_y - iDiamondSize, 2 + 4); | ||
1160 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1161 : | start_y + iDiamondSize, 2 + 8); | ||
1162 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1163 : | start_y + iDiamondSize, 1 + 8); | ||
1164 : | chl | 181 | break; |
1165 : | edgomez | 195 | default: //1+2+4+8 == we didn't find anything at all |
1166 : | chl | 326 | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1167 : | start_y - iDiamondSize, 1 + 4); | ||
1168 : | CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, | ||
1169 : | start_y + iDiamondSize, 1 + 8); | ||
1170 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1171 : | start_y - iDiamondSize, 2 + 4); | ||
1172 : | CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, | ||
1173 : | start_y + iDiamondSize, 2 + 8); | ||
1174 : | chl | 181 | break; |
1175 : | } | ||
1176 : | edgomez | 195 | if (!(iDirection)) |
1177 : | break; //ok, the end. really | ||
1178 : | else { | ||
1179 : | bDirection = iDirection; | ||
1180 : | chl | 326 | start_x = currMV->x; |
1181 : | start_y = currMV->y; | ||
1182 : | chl | 181 | } |
1183 : | } | ||
1184 : | } | ||
1185 : | edgomez | 195 | while (1); //forever |
1186 : | chl | 181 | } |
1187 : | return iMinSAD; | ||
1188 : | } | ||
1189 : | |||
1190 : | |||
1191 : | edgomez | 195 | int32_t |
1192 : | Full8_MainSearch(const uint8_t * const pRef, | ||
1193 : | const uint8_t * const pRefH, | ||
1194 : | const uint8_t * const pRefV, | ||
1195 : | const uint8_t * const pRefHV, | ||
1196 : | const uint8_t * const cur, | ||
1197 : | const int x, | ||
1198 : | const int y, | ||
1199 : | chl | 326 | const int start_x, |
1200 : | const int start_y, | ||
1201 : | int iMinSAD, | ||
1202 : | VECTOR * const currMV, | ||
1203 : | const int center_x, | ||
1204 : | const int center_y, | ||
1205 : | edgomez | 195 | const int32_t min_dx, |
1206 : | const int32_t max_dx, | ||
1207 : | const int32_t min_dy, | ||
1208 : | const int32_t max_dy, | ||
1209 : | const int32_t iEdgedWidth, | ||
1210 : | const int32_t iDiamondSize, | ||
1211 : | const int32_t iFcode, | ||
1212 : | const int32_t iQuant, | ||
1213 : | int iFound) | ||
1214 : | chl | 96 | { |
1215 : | int32_t iSAD; | ||
1216 : | edgomez | 195 | int32_t dx, dy; |
1217 : | chl | 96 | VECTOR backupMV; |
1218 : | edgomez | 195 | |
1219 : | chl | 326 | backupMV.x = start_x; |
1220 : | backupMV.y = start_y; | ||
1221 : | chl | 96 | |
1222 : | edgomez | 195 | for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
1223 : | for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) | ||
1224 : | NOCHECK_MV8_CANDIDATE(dx, dy); | ||
1225 : | |||
1226 : | chl | 96 | return iMinSAD; |
1227 : | } | ||
1228 : | |||
1229 : | ia64p | 300 | Halfpel8_RefineFuncPtr Halfpel8_Refine; |
1230 : | chl | 96 | |
1231 : | edgomez | 195 | int32_t |
1232 : | Halfpel16_Refine(const uint8_t * const pRef, | ||
1233 : | const uint8_t * const pRefH, | ||
1234 : | const uint8_t * const pRefV, | ||
1235 : | const uint8_t * const pRefHV, | ||
1236 : | const uint8_t * const cur, | ||
1237 : | const int x, | ||
1238 : | const int y, | ||
1239 : | VECTOR * const currMV, | ||
1240 : | int32_t iMinSAD, | ||
1241 : | chl | 326 | const int center_x, |
1242 : | const int center_y, | ||
1243 : | edgomez | 195 | const int32_t min_dx, |
1244 : | const int32_t max_dx, | ||
1245 : | const int32_t min_dy, | ||
1246 : | const int32_t max_dy, | ||
1247 : | const int32_t iFcode, | ||
1248 : | const int32_t iQuant, | ||
1249 : | const int32_t iEdgedWidth) | ||
1250 : | Isibaar | 3 | { |
1251 : | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ | ||
1252 : | |||
1253 : | int32_t iSAD; | ||
1254 : | VECTOR backupMV = *currMV; | ||
1255 : | edgomez | 195 | |
1256 : | CHECK_MV16_CANDIDATE(backupMV.x - 1, backupMV.y - 1); | ||
1257 : | CHECK_MV16_CANDIDATE(backupMV.x, backupMV.y - 1); | ||
1258 : | CHECK_MV16_CANDIDATE(backupMV.x + 1, backupMV.y - 1); | ||
1259 : | CHECK_MV16_CANDIDATE(backupMV.x - 1, backupMV.y); | ||
1260 : | CHECK_MV16_CANDIDATE(backupMV.x + 1, backupMV.y); | ||
1261 : | CHECK_MV16_CANDIDATE(backupMV.x - 1, backupMV.y + 1); | ||
1262 : | CHECK_MV16_CANDIDATE(backupMV.x, backupMV.y + 1); | ||
1263 : | CHECK_MV16_CANDIDATE(backupMV.x + 1, backupMV.y + 1); | ||
1264 : | |||
1265 : | Isibaar | 3 | return iMinSAD; |
1266 : | } | ||
1267 : | |||
1268 : | #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) | ||
1269 : | |||
1270 : | chl | 96 | |
1271 : | chl | 326 | |
1272 : | edgomez | 195 | int32_t |
1273 : | PMVfastSearch16(const uint8_t * const pRef, | ||
1274 : | const uint8_t * const pRefH, | ||
1275 : | const uint8_t * const pRefV, | ||
1276 : | const uint8_t * const pRefHV, | ||
1277 : | const IMAGE * const pCur, | ||
1278 : | const int x, | ||
1279 : | const int y, | ||
1280 : | chl | 326 | const int start_x, |
1281 : | const int start_y, | ||
1282 : | const int center_x, | ||
1283 : | const int center_y, | ||
1284 : | edgomez | 195 | const uint32_t MotionFlags, |
1285 : | const uint32_t iQuant, | ||
1286 : | const uint32_t iFcode, | ||
1287 : | const MBParam * const pParam, | ||
1288 : | const MACROBLOCK * const pMBs, | ||
1289 : | const MACROBLOCK * const prevMBs, | ||
1290 : | VECTOR * const currMV, | ||
1291 : | VECTOR * const currPMV) | ||
1292 : | Isibaar | 3 | { |
1293 : | edgomez | 195 | const uint32_t iWcount = pParam->mb_width; |
1294 : | Isibaar | 3 | const int32_t iWidth = pParam->width; |
1295 : | const int32_t iHeight = pParam->height; | ||
1296 : | edgomez | 195 | const int32_t iEdgedWidth = pParam->edged_width; |
1297 : | Isibaar | 3 | |
1298 : | edgomez | 195 | const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; |
1299 : | Isibaar | 3 | |
1300 : | int32_t iDiamondSize; | ||
1301 : | edgomez | 195 | |
1302 : | Isibaar | 3 | int32_t min_dx; |
1303 : | int32_t max_dx; | ||
1304 : | int32_t min_dy; | ||
1305 : | int32_t max_dy; | ||
1306 : | edgomez | 195 | |
1307 : | Isibaar | 3 | int32_t iFound; |
1308 : | |||
1309 : | VECTOR newMV; | ||
1310 : | edgomez | 195 | VECTOR backupMV; /* just for PMVFAST */ |
1311 : | |||
1312 : | Isibaar | 3 | VECTOR pmv[4]; |
1313 : | int32_t psad[4]; | ||
1314 : | chl | 181 | |
1315 : | MainSearch16FuncPtr MainSearchPtr; | ||
1316 : | Isibaar | 3 | |
1317 : | edgomez | 195 | const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
1318 : | Isibaar | 3 | |
1319 : | chl | 259 | int32_t threshA, threshB; |
1320 : | edgomez | 195 | int32_t bPredEq; |
1321 : | int32_t iMinSAD, iSAD; | ||
1322 : | |||
1323 : | Isibaar | 3 | /* Get maximum range */ |
1324 : | edgomez | 195 | get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
1325 : | iFcode); | ||
1326 : | Isibaar | 3 | |
1327 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
1328 : | |||
1329 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPEL16)) { |
1330 : | min_dx = EVEN(min_dx); | ||
1331 : | max_dx = EVEN(max_dx); | ||
1332 : | min_dy = EVEN(min_dy); | ||
1333 : | max_dy = EVEN(max_dy); | ||
1334 : | } | ||
1335 : | Isibaar | 3 | |
1336 : | edgomez | 195 | /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
1337 : | chl | 285 | //bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
1338 : | bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); | ||
1339 : | Isibaar | 3 | |
1340 : | edgomez | 195 | if ((x == 0) && (y == 0)) { |
1341 : | threshA = 512; | ||
1342 : | Isibaar | 3 | threshB = 1024; |
1343 : | edgomez | 195 | } else { |
1344 : | Isibaar | 3 | threshA = psad[0]; |
1345 : | edgomez | 195 | threshB = threshA + 256; |
1346 : | if (threshA < 512) | ||
1347 : | threshA = 512; | ||
1348 : | if (threshA > 1024) | ||
1349 : | threshA = 1024; | ||
1350 : | if (threshB > 1792) | ||
1351 : | threshB = 1792; | ||
1352 : | Isibaar | 3 | } |
1353 : | |||
1354 : | edgomez | 195 | iFound = 0; |
1355 : | |||
1356 : | Isibaar | 3 | /* Step 4: Calculate SAD around the Median prediction. |
1357 : | edgomez | 78 | MinSAD=SAD |
1358 : | If Motion Vector equal to Previous frame motion vector | ||
1359 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1360 : | If SAD<=256 goto Step 10. | ||
1361 : | edgomez | 195 | */ |
1362 : | Isibaar | 3 | |
1363 : | chl | 326 | currMV->x = start_x; |
1364 : | currMV->y = start_y; | ||
1365 : | |||
1366 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
1367 : | Isibaar | 3 | currMV->x = EVEN(currMV->x); |
1368 : | currMV->y = EVEN(currMV->y); | ||
1369 : | } | ||
1370 : | edgomez | 195 | |
1371 : | if (currMV->x > max_dx) { | ||
1372 : | currMV->x = max_dx; | ||
1373 : | edgomez | 78 | } |
1374 : | edgomez | 195 | if (currMV->x < min_dx) { |
1375 : | currMV->x = min_dx; | ||
1376 : | edgomez | 78 | } |
1377 : | edgomez | 195 | if (currMV->y > max_dy) { |
1378 : | currMV->y = max_dy; | ||
1379 : | edgomez | 78 | } |
1380 : | edgomez | 195 | if (currMV->y < min_dy) { |
1381 : | currMV->y = min_dy; | ||
1382 : | edgomez | 78 | } |
1383 : | edgomez | 195 | |
1384 : | iMinSAD = | ||
1385 : | sad16(cur, | ||
1386 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, | ||
1387 : | iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); | ||
1388 : | iMinSAD += | ||
1389 : | chl | 326 | calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
1390 : | edgomez | 195 | (uint8_t) iFcode, iQuant); |
1391 : | |||
1392 : | if ((iMinSAD < 256) || | ||
1393 : | ((MVequal(*currMV, prevMB->mvs[0])) && | ||
1394 : | chl | 289 | ((int32_t) iMinSAD < prevMB->sad16))) { |
1395 : | edgomez | 195 | if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
1396 : | { | ||
1397 : | if (!MVzero(*currMV)) { | ||
1398 : | chl | 169 | iMinSAD += MV16_00_BIAS; |
1399 : | edgomez | 195 | CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures |
1400 : | chl | 169 | iMinSAD -= MV16_00_BIAS; |
1401 : | } | ||
1402 : | } | ||
1403 : | |||
1404 : | edgomez | 195 | if (MotionFlags & PMV_QUICKSTOP16) |
1405 : | chl | 96 | goto PMVfast16_Terminate_without_Refine; |
1406 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1407 : | goto PMVfast16_Terminate_with_Refine; | ||
1408 : | edgomez | 78 | } |
1409 : | Isibaar | 3 | |
1410 : | chl | 169 | |
1411 : | /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion | ||
1412 : | vector of the median. | ||
1413 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
1414 : | */ | ||
1415 : | |||
1416 : | edgomez | 195 | if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[0]))) |
1417 : | iFound = 2; | ||
1418 : | chl | 169 | |
1419 : | /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
1420 : | Otherwise select large Diamond Search. | ||
1421 : | */ | ||
1422 : | |||
1423 : | edgomez | 195 | if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) |
1424 : | iDiamondSize = 1; // halfpel! | ||
1425 : | chl | 169 | else |
1426 : | edgomez | 195 | iDiamondSize = 2; // halfpel! |
1427 : | chl | 169 | |
1428 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPELDIAMOND16)) |
1429 : | iDiamondSize *= 2; | ||
1430 : | chl | 169 | |
1431 : | Isibaar | 3 | /* |
1432 : | edgomez | 78 | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1433 : | Also calculate (0,0) but do not subtract offset. | ||
1434 : | Let MinSAD be the smallest SAD up to this point. | ||
1435 : | chl | 140 | If MV is (0,0) subtract offset. |
1436 : | Isibaar | 3 | */ |
1437 : | |||
1438 : | // (0,0) is always possible | ||
1439 : | |||
1440 : | chl | 169 | if (!MVzero(pmv[0])) |
1441 : | CHECK_MV16_ZERO; | ||
1442 : | Isibaar | 3 | |
1443 : | // previous frame MV is always possible | ||
1444 : | chl | 169 | |
1445 : | if (!MVzero(prevMB->mvs[0])) | ||
1446 : | edgomez | 195 | if (!MVequal(prevMB->mvs[0], pmv[0])) |
1447 : | CHECK_MV16_CANDIDATE(prevMB->mvs[0].x, prevMB->mvs[0].y); | ||
1448 : | |||
1449 : | Isibaar | 3 | // left neighbour, if allowed |
1450 : | chl | 169 | |
1451 : | if (!MVzero(pmv[1])) | ||
1452 : | edgomez | 195 | if (!MVequal(pmv[1], prevMB->mvs[0])) |
1453 : | if (!MVequal(pmv[1], pmv[0])) { | ||
1454 : | if (!(MotionFlags & PMV_HALFPEL16)) { | ||
1455 : | pmv[1].x = EVEN(pmv[1].x); | ||
1456 : | pmv[1].y = EVEN(pmv[1].y); | ||
1457 : | } | ||
1458 : | chl | 169 | |
1459 : | edgomez | 195 | CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); |
1460 : | } | ||
1461 : | Isibaar | 3 | // top neighbour, if allowed |
1462 : | chl | 169 | if (!MVzero(pmv[2])) |
1463 : | edgomez | 195 | if (!MVequal(pmv[2], prevMB->mvs[0])) |
1464 : | if (!MVequal(pmv[2], pmv[0])) | ||
1465 : | if (!MVequal(pmv[2], pmv[1])) { | ||
1466 : | if (!(MotionFlags & PMV_HALFPEL16)) { | ||
1467 : | pmv[2].x = EVEN(pmv[2].x); | ||
1468 : | pmv[2].y = EVEN(pmv[2].y); | ||
1469 : | } | ||
1470 : | CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); | ||
1471 : | |||
1472 : | Isibaar | 3 | // top right neighbour, if allowed |
1473 : | edgomez | 195 | if (!MVzero(pmv[3])) |
1474 : | if (!MVequal(pmv[3], prevMB->mvs[0])) | ||
1475 : | if (!MVequal(pmv[3], pmv[0])) | ||
1476 : | if (!MVequal(pmv[3], pmv[1])) | ||
1477 : | if (!MVequal(pmv[3], pmv[2])) { | ||
1478 : | if (!(MotionFlags & PMV_HALFPEL16)) { | ||
1479 : | pmv[3].x = EVEN(pmv[3].x); | ||
1480 : | pmv[3].y = EVEN(pmv[3].y); | ||
1481 : | } | ||
1482 : | CHECK_MV16_CANDIDATE(pmv[3].x, | ||
1483 : | pmv[3].y); | ||
1484 : | } | ||
1485 : | } | ||
1486 : | |||
1487 : | if ((MVzero(*currMV)) && | ||
1488 : | (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) | ||
1489 : | chl | 140 | iMinSAD -= MV16_00_BIAS; |
1490 : | Isibaar | 3 | |
1491 : | edgomez | 195 | |
1492 : | Isibaar | 3 | /* Step 6: If MinSAD <= thresa goto Step 10. |
1493 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
1494 : | */ | ||
1495 : | |||
1496 : | edgomez | 195 | if ((iMinSAD <= threshA) || |
1497 : | (MVequal(*currMV, prevMB->mvs[0]) && | ||
1498 : | chl | 289 | ((int32_t) iMinSAD < prevMB->sad16))) { |
1499 : | edgomez | 195 | if (MotionFlags & PMV_QUICKSTOP16) |
1500 : | chl | 96 | goto PMVfast16_Terminate_without_Refine; |
1501 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1502 : | goto PMVfast16_Terminate_with_Refine; | ||
1503 : | edgomez | 78 | } |
1504 : | Isibaar | 3 | |
1505 : | |||
1506 : | /************ (Diamond Search) **************/ | ||
1507 : | /* | ||
1508 : | edgomez | 78 | Step 7: Perform Diamond search, with either the small or large diamond. |
1509 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
1510 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
1511 : | If center then goto step 10. | ||
1512 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
1513 : | Refine by using small diamond and goto step 10. | ||
1514 : | Isibaar | 3 | */ |
1515 : | |||
1516 : | chl | 181 | if (MotionFlags & PMV_USESQUARES16) |
1517 : | MainSearchPtr = Square16_MainSearch; | ||
1518 : | edgomez | 195 | else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
1519 : | MainSearchPtr = AdvDiamond16_MainSearch; | ||
1520 : | chl | 181 | else |
1521 : | edgomez | 195 | MainSearchPtr = Diamond16_MainSearch; |
1522 : | chl | 181 | |
1523 : | edgomez | 195 | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1524 : | Isibaar | 3 | |
1525 : | chl | 259 | |
1526 : | Isibaar | 3 | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
1527 : | edgomez | 195 | iSAD = |
1528 : | chl | 326 | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1529 : | currMV->x, currMV->y, iMinSAD, &newMV, center_x, center_y, | ||
1530 : | min_dx, max_dx, | ||
1531 : | edgomez | 195 | min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1532 : | iQuant, iFound); | ||
1533 : | |||
1534 : | if (iSAD < iMinSAD) { | ||
1535 : | Isibaar | 3 | *currMV = newMV; |
1536 : | iMinSAD = iSAD; | ||
1537 : | } | ||
1538 : | |||
1539 : | edgomez | 195 | if (MotionFlags & PMV_EXTSEARCH16) { |
1540 : | Isibaar | 3 | /* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1541 : | |||
1542 : | edgomez | 195 | if (!(MVequal(pmv[0], backupMV))) { |
1543 : | iSAD = | ||
1544 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, | ||
1545 : | chl | 326 | center_x, center_y, iMinSAD, &newMV, center_x, center_y, |
1546 : | edgomez | 195 | min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1547 : | iDiamondSize, iFcode, iQuant, iFound); | ||
1548 : | |||
1549 : | if (iSAD < iMinSAD) { | ||
1550 : | *currMV = newMV; | ||
1551 : | iMinSAD = iSAD; | ||
1552 : | } | ||
1553 : | Isibaar | 3 | } |
1554 : | |||
1555 : | edgomez | 195 | if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1556 : | iSAD = | ||
1557 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, | ||
1558 : | chl | 326 | iMinSAD, &newMV, center_x, center_y, |
1559 : | min_dx, max_dx, min_dy, max_dy, | ||
1560 : | iEdgedWidth, iDiamondSize, iFcode, | ||
1561 : | edgomez | 195 | iQuant, iFound); |
1562 : | |||
1563 : | if (iSAD < iMinSAD) { | ||
1564 : | *currMV = newMV; | ||
1565 : | iMinSAD = iSAD; | ||
1566 : | } | ||
1567 : | Isibaar | 3 | } |
1568 : | } | ||
1569 : | |||
1570 : | /* | ||
1571 : | edgomez | 78 | Step 10: The motion vector is chosen according to the block corresponding to MinSAD. |
1572 : | Isibaar | 3 | */ |
1573 : | |||
1574 : | edgomez | 195 | PMVfast16_Terminate_with_Refine: |
1575 : | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step | ||
1576 : | iMinSAD = | ||
1577 : | Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, | ||
1578 : | chl | 326 | iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
1579 : | edgomez | 195 | iFcode, iQuant, iEdgedWidth); |
1580 : | Isibaar | 3 | |
1581 : | edgomez | 195 | PMVfast16_Terminate_without_Refine: |
1582 : | chl | 326 | currPMV->x = currMV->x - center_x; |
1583 : | currPMV->y = currMV->y - center_y; | ||
1584 : | Isibaar | 3 | return iMinSAD; |
1585 : | } | ||
1586 : | |||
1587 : | |||
1588 : | |||
1589 : | |||
1590 : | |||
1591 : | |||
1592 : | edgomez | 195 | int32_t |
1593 : | Diamond8_MainSearch(const uint8_t * const pRef, | ||
1594 : | const uint8_t * const pRefH, | ||
1595 : | const uint8_t * const pRefV, | ||
1596 : | const uint8_t * const pRefHV, | ||
1597 : | const uint8_t * const cur, | ||
1598 : | const int x, | ||
1599 : | const int y, | ||
1600 : | chl | 326 | int32_t start_x, |
1601 : | int32_t start_y, | ||
1602 : | edgomez | 195 | int32_t iMinSAD, |
1603 : | VECTOR * const currMV, | ||
1604 : | chl | 326 | const int center_x, |
1605 : | const int center_y, | ||
1606 : | edgomez | 195 | const int32_t min_dx, |
1607 : | const int32_t max_dx, | ||
1608 : | const int32_t min_dy, | ||
1609 : | const int32_t max_dy, | ||
1610 : | const int32_t iEdgedWidth, | ||
1611 : | const int32_t iDiamondSize, | ||
1612 : | const int32_t iFcode, | ||
1613 : | const int32_t iQuant, | ||
1614 : | int iFound) | ||
1615 : | Isibaar | 3 | { |
1616 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
1617 : | |||
1618 : | edgomez | 195 | int32_t iDirection = 0; |
1619 : | chl | 344 | int32_t iDirectionBackup; |
1620 : | Isibaar | 3 | int32_t iSAD; |
1621 : | VECTOR backupMV; | ||
1622 : | edgomez | 195 | |
1623 : | chl | 326 | backupMV.x = start_x; |
1624 : | backupMV.y = start_y; | ||
1625 : | edgomez | 195 | |
1626 : | Isibaar | 3 | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
1627 : | |||
1628 : | edgomez | 195 | CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
1629 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); | ||
1630 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); | ||
1631 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); | ||
1632 : | Isibaar | 3 | |
1633 : | chl | 344 | if (iDirection) { |
1634 : | edgomez | 195 | while (!iFound) { |
1635 : | iFound = 1; | ||
1636 : | backupMV = *currMV; // since iDirection!=0, this is well defined! | ||
1637 : | chl | 344 | iDirectionBackup = iDirection; |
1638 : | edgomez | 195 | |
1639 : | chl | 344 | if (iDirectionBackup != 2) |
1640 : | edgomez | 195 | CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1641 : | backupMV.y, 1); | ||
1642 : | chl | 344 | if (iDirectionBackup != 1) |
1643 : | edgomez | 195 | CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1644 : | backupMV.y, 2); | ||
1645 : | chl | 344 | if (iDirectionBackup != 4) |
1646 : | edgomez | 195 | CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1647 : | backupMV.y - iDiamondSize, 3); | ||
1648 : | chl | 344 | if (iDirectionBackup != 3) |
1649 : | edgomez | 195 | CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1650 : | backupMV.y + iDiamondSize, 4); | ||
1651 : | chl | 344 | } |
1652 : | edgomez | 195 | } else { |
1653 : | chl | 326 | currMV->x = start_x; |
1654 : | currMV->y = start_y; | ||
1655 : | edgomez | 78 | } |
1656 : | Isibaar | 3 | return iMinSAD; |
1657 : | } | ||
1658 : | |||
1659 : | edgomez | 195 | int32_t |
1660 : | ia64p | 300 | Halfpel8_Refine_c(const uint8_t * const pRef, |
1661 : | edgomez | 195 | const uint8_t * const pRefH, |
1662 : | const uint8_t * const pRefV, | ||
1663 : | const uint8_t * const pRefHV, | ||
1664 : | const uint8_t * const cur, | ||
1665 : | const int x, | ||
1666 : | const int y, | ||
1667 : | VECTOR * const currMV, | ||
1668 : | int32_t iMinSAD, | ||
1669 : | chl | 326 | const int center_x, |
1670 : | const int center_y, | ||
1671 : | edgomez | 195 | const int32_t min_dx, |
1672 : | const int32_t max_dx, | ||
1673 : | const int32_t min_dy, | ||
1674 : | const int32_t max_dy, | ||
1675 : | const int32_t iFcode, | ||
1676 : | const int32_t iQuant, | ||
1677 : | const int32_t iEdgedWidth) | ||
1678 : | Isibaar | 3 | { |
1679 : | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ | ||
1680 : | |||
1681 : | int32_t iSAD; | ||
1682 : | VECTOR backupMV = *currMV; | ||
1683 : | edgomez | 195 | |
1684 : | CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y - 1); | ||
1685 : | CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y - 1); | ||
1686 : | CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y - 1); | ||
1687 : | CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y); | ||
1688 : | CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y); | ||
1689 : | CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y + 1); | ||
1690 : | CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y + 1); | ||
1691 : | CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y + 1); | ||
1692 : | |||
1693 : | Isibaar | 3 | return iMinSAD; |
1694 : | } | ||
1695 : | |||
1696 : | |||
1697 : | #define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) | ||
1698 : | |||
1699 : | edgomez | 195 | int32_t |
1700 : | PMVfastSearch8(const uint8_t * const pRef, | ||
1701 : | const uint8_t * const pRefH, | ||
1702 : | const uint8_t * const pRefV, | ||
1703 : | const uint8_t * const pRefHV, | ||
1704 : | const IMAGE * const pCur, | ||
1705 : | const int x, | ||
1706 : | const int y, | ||
1707 : | const int start_x, | ||
1708 : | const int start_y, | ||
1709 : | chl | 326 | const int center_x, |
1710 : | const int center_y, | ||
1711 : | edgomez | 195 | const uint32_t MotionFlags, |
1712 : | const uint32_t iQuant, | ||
1713 : | const uint32_t iFcode, | ||
1714 : | const MBParam * const pParam, | ||
1715 : | const MACROBLOCK * const pMBs, | ||
1716 : | const MACROBLOCK * const prevMBs, | ||
1717 : | VECTOR * const currMV, | ||
1718 : | VECTOR * const currPMV) | ||
1719 : | Isibaar | 3 | { |
1720 : | edgomez | 195 | const uint32_t iWcount = pParam->mb_width; |
1721 : | Isibaar | 3 | const int32_t iWidth = pParam->width; |
1722 : | const int32_t iHeight = pParam->height; | ||
1723 : | edgomez | 195 | const int32_t iEdgedWidth = pParam->edged_width; |
1724 : | Isibaar | 3 | |
1725 : | edgomez | 195 | const uint8_t *cur = pCur->y + x * 8 + y * 8 * iEdgedWidth; |
1726 : | Isibaar | 3 | |
1727 : | int32_t iDiamondSize; | ||
1728 : | |||
1729 : | int32_t min_dx; | ||
1730 : | int32_t max_dx; | ||
1731 : | int32_t min_dy; | ||
1732 : | int32_t max_dy; | ||
1733 : | edgomez | 195 | |
1734 : | Isibaar | 3 | VECTOR pmv[4]; |
1735 : | int32_t psad[4]; | ||
1736 : | VECTOR newMV; | ||
1737 : | VECTOR backupMV; | ||
1738 : | edgomez | 170 | VECTOR startMV; |
1739 : | Isibaar | 3 | |
1740 : | edgomez | 195 | // const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1741 : | const MACROBLOCK *const prevMB = prevMBs + (x >> 1) + (y >> 1) * iWcount; | ||
1742 : | Isibaar | 3 | |
1743 : | chl | 259 | int32_t threshA, threshB; |
1744 : | edgomez | 195 | int32_t iFound, bPredEq; |
1745 : | int32_t iMinSAD, iSAD; | ||
1746 : | Isibaar | 3 | |
1747 : | edgomez | 195 | int32_t iSubBlock = (y & 1) + (y & 1) + (x & 1); |
1748 : | |||
1749 : | chl | 181 | MainSearch8FuncPtr MainSearchPtr; |
1750 : | |||
1751 : | edgomez | 170 | /* Init variables */ |
1752 : | startMV.x = start_x; | ||
1753 : | startMV.y = start_y; | ||
1754 : | |||
1755 : | /* Get maximum range */ | ||
1756 : | edgomez | 195 | get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, |
1757 : | iFcode); | ||
1758 : | Isibaar | 3 | |
1759 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPELDIAMOND8)) { |
1760 : | min_dx = EVEN(min_dx); | ||
1761 : | max_dx = EVEN(max_dx); | ||
1762 : | min_dy = EVEN(min_dy); | ||
1763 : | max_dy = EVEN(max_dy); | ||
1764 : | } | ||
1765 : | Isibaar | 3 | |
1766 : | edgomez | 195 | /* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
1767 : | chl | 285 | //bPredEq = get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); |
1768 : | bPredEq = get_pmvdata2(pMBs, iWcount, 0, (x >> 1), (y >> 1), iSubBlock, pmv, psad); | ||
1769 : | Isibaar | 3 | |
1770 : | edgomez | 195 | if ((x == 0) && (y == 0)) { |
1771 : | threshA = 512 / 4; | ||
1772 : | threshB = 1024 / 4; | ||
1773 : | |||
1774 : | } else { | ||
1775 : | chl | 326 | threshA = psad[0] / 4; /* good estimate? */ |
1776 : | edgomez | 195 | threshB = threshA + 256 / 4; |
1777 : | if (threshA < 512 / 4) | ||
1778 : | threshA = 512 / 4; | ||
1779 : | if (threshA > 1024 / 4) | ||
1780 : | threshA = 1024 / 4; | ||
1781 : | if (threshB > 1792 / 4) | ||
1782 : | threshB = 1792 / 4; | ||
1783 : | Isibaar | 3 | } |
1784 : | |||
1785 : | edgomez | 195 | iFound = 0; |
1786 : | |||
1787 : | Isibaar | 3 | /* Step 4: Calculate SAD around the Median prediction. |
1788 : | edgomez | 78 | MinSAD=SAD |
1789 : | If Motion Vector equal to Previous frame motion vector | ||
1790 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1791 : | If SAD<=256 goto Step 10. | ||
1792 : | edgomez | 195 | */ |
1793 : | Isibaar | 3 | |
1794 : | |||
1795 : | // Prepare for main loop | ||
1796 : | |||
1797 : | edgomez | 195 | // if (MotionFlags & PMV_USESQUARES8) |
1798 : | // MainSearchPtr = Square8_MainSearch; | ||
1799 : | // else | ||
1800 : | chl | 181 | |
1801 : | if (MotionFlags & PMV_ADVANCEDDIAMOND8) | ||
1802 : | MainSearchPtr = AdvDiamond8_MainSearch; | ||
1803 : | else | ||
1804 : | MainSearchPtr = Diamond8_MainSearch; | ||
1805 : | |||
1806 : | |||
1807 : | chl | 169 | *currMV = startMV; |
1808 : | edgomez | 195 | |
1809 : | iMinSAD = | ||
1810 : | sad8(cur, | ||
1811 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, | ||
1812 : | iEdgedWidth), iEdgedWidth); | ||
1813 : | iMinSAD += | ||
1814 : | chl | 326 | calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
1815 : | edgomez | 195 | (uint8_t) iFcode, iQuant); |
1816 : | |||
1817 : | if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) | ||
1818 : | chl | 289 | && ((int32_t) iMinSAD < |
1819 : | edgomez | 195 | prevMB->sad8[iSubBlock]))) { |
1820 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1821 : | chl | 96 | goto PMVfast8_Terminate_without_Refine; |
1822 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1823 : | goto PMVfast8_Terminate_with_Refine; | ||
1824 : | edgomez | 78 | } |
1825 : | Isibaar | 3 | |
1826 : | chl | 169 | /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1827 : | vector of the median. | ||
1828 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
1829 : | */ | ||
1830 : | chl | 96 | |
1831 : | edgomez | 195 | if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[iSubBlock]))) |
1832 : | iFound = 2; | ||
1833 : | chl | 169 | |
1834 : | /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
1835 : | Otherwise select large Diamond Search. | ||
1836 : | */ | ||
1837 : | |||
1838 : | edgomez | 195 | if ((!MVzero(pmv[0])) || (threshB < 1536 / 4) || (bPredEq)) |
1839 : | iDiamondSize = 1; // 1 halfpel! | ||
1840 : | chl | 169 | else |
1841 : | edgomez | 195 | iDiamondSize = 2; // 2 halfpel = 1 full pixel! |
1842 : | chl | 169 | |
1843 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPELDIAMOND8)) |
1844 : | iDiamondSize *= 2; | ||
1845 : | chl | 169 | |
1846 : | |||
1847 : | Isibaar | 3 | /* |
1848 : | edgomez | 78 | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1849 : | Also calculate (0,0) but do not subtract offset. | ||
1850 : | Let MinSAD be the smallest SAD up to this point. | ||
1851 : | chl | 140 | If MV is (0,0) subtract offset. |
1852 : | Isibaar | 3 | */ |
1853 : | |||
1854 : | chl | 169 | // the median prediction might be even better than mv16 |
1855 : | Isibaar | 3 | |
1856 : | edgomez | 195 | if (!MVequal(pmv[0], startMV)) |
1857 : | chl | 326 | CHECK_MV8_CANDIDATE(center_x, center_y); |
1858 : | chl | 169 | |
1859 : | // (0,0) if needed | ||
1860 : | if (!MVzero(pmv[0])) | ||
1861 : | edgomez | 195 | if (!MVzero(startMV)) |
1862 : | CHECK_MV8_ZERO; | ||
1863 : | Isibaar | 3 | |
1864 : | chl | 169 | // previous frame MV if needed |
1865 : | if (!MVzero(prevMB->mvs[iSubBlock])) | ||
1866 : | edgomez | 195 | if (!MVequal(prevMB->mvs[iSubBlock], startMV)) |
1867 : | if (!MVequal(prevMB->mvs[iSubBlock], pmv[0])) | ||
1868 : | CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x, | ||
1869 : | prevMB->mvs[iSubBlock].y); | ||
1870 : | chl | 169 | |
1871 : | edgomez | 195 | if ((iMinSAD <= threshA) || |
1872 : | (MVequal(*currMV, prevMB->mvs[iSubBlock]) && | ||
1873 : | chl | 289 | ((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1874 : | edgomez | 195 | if (MotionFlags & PMV_QUICKSTOP16) |
1875 : | chl | 169 | goto PMVfast8_Terminate_without_Refine; |
1876 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1877 : | goto PMVfast8_Terminate_with_Refine; | ||
1878 : | } | ||
1879 : | |||
1880 : | // left neighbour, if allowed and needed | ||
1881 : | if (!MVzero(pmv[1])) | ||
1882 : | edgomez | 195 | if (!MVequal(pmv[1], startMV)) |
1883 : | if (!MVequal(pmv[1], prevMB->mvs[iSubBlock])) | ||
1884 : | if (!MVequal(pmv[1], pmv[0])) { | ||
1885 : | if (!(MotionFlags & PMV_HALFPEL8)) { | ||
1886 : | pmv[1].x = EVEN(pmv[1].x); | ||
1887 : | pmv[1].y = EVEN(pmv[1].y); | ||
1888 : | } | ||
1889 : | CHECK_MV8_CANDIDATE(pmv[1].x, pmv[1].y); | ||
1890 : | } | ||
1891 : | chl | 169 | // top neighbour, if allowed and needed |
1892 : | if (!MVzero(pmv[2])) | ||
1893 : | edgomez | 195 | if (!MVequal(pmv[2], startMV)) |
1894 : | if (!MVequal(pmv[2], prevMB->mvs[iSubBlock])) | ||
1895 : | if (!MVequal(pmv[2], pmv[0])) | ||
1896 : | if (!MVequal(pmv[2], pmv[1])) { | ||
1897 : | if (!(MotionFlags & PMV_HALFPEL8)) { | ||
1898 : | pmv[2].x = EVEN(pmv[2].x); | ||
1899 : | pmv[2].y = EVEN(pmv[2].y); | ||
1900 : | } | ||
1901 : | CHECK_MV8_CANDIDATE(pmv[2].x, pmv[2].y); | ||
1902 : | |||
1903 : | chl | 169 | // top right neighbour, if allowed and needed |
1904 : | edgomez | 195 | if (!MVzero(pmv[3])) |
1905 : | if (!MVequal(pmv[3], startMV)) | ||
1906 : | if (!MVequal(pmv[3], prevMB->mvs[iSubBlock])) | ||
1907 : | if (!MVequal(pmv[3], pmv[0])) | ||
1908 : | if (!MVequal(pmv[3], pmv[1])) | ||
1909 : | if (!MVequal(pmv[3], pmv[2])) { | ||
1910 : | if (! | ||
1911 : | (MotionFlags & | ||
1912 : | PMV_HALFPEL8)) { | ||
1913 : | pmv[3].x = EVEN(pmv[3].x); | ||
1914 : | pmv[3].y = EVEN(pmv[3].y); | ||
1915 : | } | ||
1916 : | CHECK_MV8_CANDIDATE(pmv[3].x, | ||
1917 : | pmv[3].y); | ||
1918 : | } | ||
1919 : | } | ||
1920 : | Isibaar | 3 | |
1921 : | edgomez | 195 | if ((MVzero(*currMV)) && |
1922 : | (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) | ||
1923 : | chl | 140 | iMinSAD -= MV8_00_BIAS; |
1924 : | |||
1925 : | |||
1926 : | Isibaar | 3 | /* Step 6: If MinSAD <= thresa goto Step 10. |
1927 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
1928 : | */ | ||
1929 : | |||
1930 : | edgomez | 195 | if ((iMinSAD <= threshA) || |
1931 : | (MVequal(*currMV, prevMB->mvs[iSubBlock]) && | ||
1932 : | chl | 289 | ((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1933 : | edgomez | 195 | if (MotionFlags & PMV_QUICKSTOP16) |
1934 : | chl | 96 | goto PMVfast8_Terminate_without_Refine; |
1935 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1936 : | goto PMVfast8_Terminate_with_Refine; | ||
1937 : | edgomez | 78 | } |
1938 : | Isibaar | 3 | |
1939 : | /************ (Diamond Search) **************/ | ||
1940 : | /* | ||
1941 : | edgomez | 78 | Step 7: Perform Diamond search, with either the small or large diamond. |
1942 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
1943 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
1944 : | If center then goto step 10. | ||
1945 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
1946 : | Refine by using small diamond and goto step 10. | ||
1947 : | Isibaar | 3 | */ |
1948 : | |||
1949 : | edgomez | 195 | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1950 : | Isibaar | 3 | |
1951 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1952 : | edgomez | 195 | iSAD = |
1953 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, | ||
1954 : | chl | 326 | currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
1955 : | edgomez | 195 | min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1956 : | iQuant, iFound); | ||
1957 : | |||
1958 : | if (iSAD < iMinSAD) { | ||
1959 : | Isibaar | 3 | *currMV = newMV; |
1960 : | iMinSAD = iSAD; | ||
1961 : | } | ||
1962 : | |||
1963 : | edgomez | 195 | if (MotionFlags & PMV_EXTSEARCH8) { |
1964 : | Isibaar | 3 | /* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1965 : | |||
1966 : | edgomez | 195 | if (!(MVequal(pmv[0], backupMV))) { |
1967 : | iSAD = | ||
1968 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, | ||
1969 : | chl | 326 | pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
1970 : | edgomez | 195 | min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1971 : | iDiamondSize, iFcode, iQuant, iFound); | ||
1972 : | |||
1973 : | if (iSAD < iMinSAD) { | ||
1974 : | *currMV = newMV; | ||
1975 : | iMinSAD = iSAD; | ||
1976 : | } | ||
1977 : | Isibaar | 3 | } |
1978 : | |||
1979 : | edgomez | 195 | if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1980 : | iSAD = | ||
1981 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, | ||
1982 : | chl | 326 | iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
1983 : | edgomez | 195 | max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1984 : | iQuant, iFound); | ||
1985 : | |||
1986 : | if (iSAD < iMinSAD) { | ||
1987 : | *currMV = newMV; | ||
1988 : | iMinSAD = iSAD; | ||
1989 : | } | ||
1990 : | Isibaar | 3 | } |
1991 : | } | ||
1992 : | |||
1993 : | /* Step 10: The motion vector is chosen according to the block corresponding to MinSAD. | ||
1994 : | edgomez | 78 | By performing an optional local half-pixel search, we can refine this result even further. |
1995 : | Isibaar | 3 | */ |
1996 : | |||
1997 : | edgomez | 195 | PMVfast8_Terminate_with_Refine: |
1998 : | if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step | ||
1999 : | iMinSAD = | ||
2000 : | Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, | ||
2001 : | chl | 326 | iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2002 : | edgomez | 195 | iFcode, iQuant, iEdgedWidth); |
2003 : | Isibaar | 3 | |
2004 : | edgomez | 195 | |
2005 : | PMVfast8_Terminate_without_Refine: | ||
2006 : | chl | 326 | currPMV->x = currMV->x - center_x; |
2007 : | currPMV->y = currMV->y - center_y; | ||
2008 : | edgomez | 195 | |
2009 : | Isibaar | 3 | return iMinSAD; |
2010 : | } | ||
2011 : | chl | 96 | |
2012 : | edgomez | 195 | int32_t |
2013 : | EPZSSearch16(const uint8_t * const pRef, | ||
2014 : | const uint8_t * const pRefH, | ||
2015 : | const uint8_t * const pRefV, | ||
2016 : | const uint8_t * const pRefHV, | ||
2017 : | const IMAGE * const pCur, | ||
2018 : | const int x, | ||
2019 : | const int y, | ||
2020 : | chl | 326 | const int start_x, |
2021 : | const int start_y, | ||
2022 : | const int center_x, | ||
2023 : | const int center_y, | ||
2024 : | edgomez | 195 | const uint32_t MotionFlags, |
2025 : | const uint32_t iQuant, | ||
2026 : | const uint32_t iFcode, | ||
2027 : | const MBParam * const pParam, | ||
2028 : | const MACROBLOCK * const pMBs, | ||
2029 : | const MACROBLOCK * const prevMBs, | ||
2030 : | VECTOR * const currMV, | ||
2031 : | VECTOR * const currPMV) | ||
2032 : | chl | 96 | { |
2033 : | edgomez | 195 | const uint32_t iWcount = pParam->mb_width; |
2034 : | const uint32_t iHcount = pParam->mb_height; | ||
2035 : | chl | 96 | |
2036 : | const int32_t iWidth = pParam->width; | ||
2037 : | const int32_t iHeight = pParam->height; | ||
2038 : | edgomez | 195 | const int32_t iEdgedWidth = pParam->edged_width; |
2039 : | chl | 96 | |
2040 : | edgomez | 195 | const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; |
2041 : | chl | 96 | |
2042 : | int32_t min_dx; | ||
2043 : | int32_t max_dx; | ||
2044 : | int32_t min_dy; | ||
2045 : | int32_t max_dy; | ||
2046 : | edgomez | 195 | |
2047 : | chl | 96 | VECTOR newMV; |
2048 : | VECTOR backupMV; | ||
2049 : | edgomez | 195 | |
2050 : | chl | 96 | VECTOR pmv[4]; |
2051 : | int32_t psad[8]; | ||
2052 : | |||
2053 : | edgomez | 195 | static MACROBLOCK *oldMBs = NULL; |
2054 : | |||
2055 : | // const MACROBLOCK * const pMB = pMBs + x + y * iWcount; | ||
2056 : | const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; | ||
2057 : | MACROBLOCK *oldMB = NULL; | ||
2058 : | |||
2059 : | chl | 259 | int32_t thresh2; |
2060 : | edgomez | 195 | int32_t bPredEq; |
2061 : | int32_t iMinSAD, iSAD = 9999; | ||
2062 : | chl | 96 | |
2063 : | chl | 181 | MainSearch16FuncPtr MainSearchPtr; |
2064 : | chl | 96 | |
2065 : | edgomez | 195 | if (oldMBs == NULL) { |
2066 : | oldMBs = (MACROBLOCK *) calloc(iWcount * iHcount, sizeof(MACROBLOCK)); | ||
2067 : | // fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); | ||
2068 : | chl | 96 | } |
2069 : | oldMB = oldMBs + x + y * iWcount; | ||
2070 : | |||
2071 : | /* Get maximum range */ | ||
2072 : | edgomez | 195 | get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
2073 : | iFcode); | ||
2074 : | chl | 96 | |
2075 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPEL16)) { |
2076 : | min_dx = EVEN(min_dx); | ||
2077 : | max_dx = EVEN(max_dx); | ||
2078 : | min_dy = EVEN(min_dy); | ||
2079 : | max_dy = EVEN(max_dy); | ||
2080 : | } | ||
2081 : | /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ | ||
2082 : | chl | 285 | //bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
2083 : | bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); | ||
2084 : | chl | 96 | |
2085 : | /* Step 4: Calculate SAD around the Median prediction. | ||
2086 : | MinSAD=SAD | ||
2087 : | If Motion Vector equal to Previous frame motion vector | ||
2088 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
2089 : | If SAD<=256 goto Step 10. | ||
2090 : | edgomez | 195 | */ |
2091 : | chl | 96 | |
2092 : | // Prepare for main loop | ||
2093 : | |||
2094 : | chl | 326 | currMV->x = start_x; |
2095 : | currMV->y = start_y; | ||
2096 : | |||
2097 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPEL16)) { |
2098 : | chl | 96 | currMV->x = EVEN(currMV->x); |
2099 : | currMV->y = EVEN(currMV->y); | ||
2100 : | } | ||
2101 : | |||
2102 : | edgomez | 195 | if (currMV->x > max_dx) |
2103 : | currMV->x = max_dx; | ||
2104 : | if (currMV->x < min_dx) | ||
2105 : | currMV->x = min_dx; | ||
2106 : | if (currMV->y > max_dy) | ||
2107 : | currMV->y = max_dy; | ||
2108 : | if (currMV->y < min_dy) | ||
2109 : | currMV->y = min_dy; | ||
2110 : | |||
2111 : | /***************** This is predictor SET A: only median prediction ******************/ | ||
2112 : | |||
2113 : | iMinSAD = | ||
2114 : | sad16(cur, | ||
2115 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, | ||
2116 : | iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); | ||
2117 : | iMinSAD += | ||
2118 : | chl | 326 | calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2119 : | edgomez | 195 | (uint8_t) iFcode, iQuant); |
2120 : | |||
2121 : | chl | 96 | // thresh1 is fixed to 256 |
2122 : | edgomez | 195 | if ((iMinSAD < 256) || |
2123 : | ((MVequal(*currMV, prevMB->mvs[0])) && | ||
2124 : | chl | 289 | ((int32_t) iMinSAD < prevMB->sad16))) { |
2125 : | edgomez | 195 | if (MotionFlags & PMV_QUICKSTOP16) |
2126 : | goto EPZS16_Terminate_without_Refine; | ||
2127 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
2128 : | goto EPZS16_Terminate_with_Refine; | ||
2129 : | } | ||
2130 : | chl | 96 | |
2131 : | edgomez | 195 | /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ |
2132 : | chl | 96 | |
2133 : | // previous frame MV | ||
2134 : | edgomez | 195 | CHECK_MV16_CANDIDATE(prevMB->mvs[0].x, prevMB->mvs[0].y); |
2135 : | chl | 96 | |
2136 : | // set threshhold based on Min of Prediction and SAD of collocated block | ||
2137 : | // CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want | ||
2138 : | |||
2139 : | edgomez | 195 | if ((x == 0) && (y == 0)) { |
2140 : | thresh2 = 512; | ||
2141 : | } else { | ||
2142 : | chl | 96 | /* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ |
2143 : | |||
2144 : | edgomez | 195 | thresh2 = MIN(psad[0], iSAD) * 6 / 5 + 128; |
2145 : | chl | 96 | } |
2146 : | |||
2147 : | // MV=(0,0) is often a good choice | ||
2148 : | |||
2149 : | CHECK_MV16_ZERO; | ||
2150 : | |||
2151 : | edgomez | 195 | |
2152 : | chl | 96 | // left neighbour, if allowed |
2153 : | edgomez | 195 | if (x != 0) { |
2154 : | if (!(MotionFlags & PMV_HALFPEL16)) { | ||
2155 : | pmv[1].x = EVEN(pmv[1].x); | ||
2156 : | chl | 96 | pmv[1].y = EVEN(pmv[1].y); |
2157 : | } | ||
2158 : | edgomez | 195 | CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); |
2159 : | chl | 96 | } |
2160 : | // top neighbour, if allowed | ||
2161 : | edgomez | 195 | if (y != 0) { |
2162 : | if (!(MotionFlags & PMV_HALFPEL16)) { | ||
2163 : | pmv[2].x = EVEN(pmv[2].x); | ||
2164 : | chl | 96 | pmv[2].y = EVEN(pmv[2].y); |
2165 : | } | ||
2166 : | edgomez | 195 | CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); |
2167 : | |||
2168 : | chl | 96 | // top right neighbour, if allowed |
2169 : | edgomez | 195 | if ((uint32_t) x != (iWcount - 1)) { |
2170 : | if (!(MotionFlags & PMV_HALFPEL16)) { | ||
2171 : | pmv[3].x = EVEN(pmv[3].x); | ||
2172 : | chl | 96 | pmv[3].y = EVEN(pmv[3].y); |
2173 : | } | ||
2174 : | edgomez | 195 | CHECK_MV16_CANDIDATE(pmv[3].x, pmv[3].y); |
2175 : | chl | 96 | } |
2176 : | } | ||
2177 : | |||
2178 : | /* Terminate if MinSAD <= T_2 | ||
2179 : | Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] | ||
2180 : | */ | ||
2181 : | |||
2182 : | edgomez | 195 | if ((iMinSAD <= thresh2) |
2183 : | || (MVequal(*currMV, prevMB->mvs[0]) && | ||
2184 : | chl | 289 | ((int32_t) iMinSAD <= prevMB->sad16))) { |
2185 : | edgomez | 195 | if (MotionFlags & PMV_QUICKSTOP16) |
2186 : | goto EPZS16_Terminate_without_Refine; | ||
2187 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
2188 : | goto EPZS16_Terminate_with_Refine; | ||
2189 : | } | ||
2190 : | chl | 96 | |
2191 : | /***** predictor SET C: acceleration MV (new!), neighbours in prev. frame(new!) ****/ | ||
2192 : | |||
2193 : | edgomez | 195 | backupMV = prevMB->mvs[0]; // collocated MV |
2194 : | backupMV.x += (prevMB->mvs[0].x - oldMB->mvs[0].x); // acceleration X | ||
2195 : | backupMV.y += (prevMB->mvs[0].y - oldMB->mvs[0].y); // acceleration Y | ||
2196 : | chl | 96 | |
2197 : | edgomez | 195 | CHECK_MV16_CANDIDATE(backupMV.x, backupMV.y); |
2198 : | chl | 96 | |
2199 : | // left neighbour | ||
2200 : | edgomez | 195 | if (x != 0) |
2201 : | CHECK_MV16_CANDIDATE((prevMB - 1)->mvs[0].x, (prevMB - 1)->mvs[0].y); | ||
2202 : | chl | 96 | |
2203 : | // top neighbour | ||
2204 : | if (y != 0) | ||
2205 : | edgomez | 195 | CHECK_MV16_CANDIDATE((prevMB - iWcount)->mvs[0].x, |
2206 : | (prevMB - iWcount)->mvs[0].y); | ||
2207 : | chl | 96 | |
2208 : | // right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs | ||
2209 : | |||
2210 : | edgomez | 195 | if ((uint32_t) x != iWcount - 1) |
2211 : | CHECK_MV16_CANDIDATE((prevMB + 1)->mvs[0].x, (prevMB + 1)->mvs[0].y); | ||
2212 : | chl | 96 | |
2213 : | // bottom neighbour, dito | ||
2214 : | edgomez | 195 | if ((uint32_t) y != iHcount - 1) |
2215 : | CHECK_MV16_CANDIDATE((prevMB + iWcount)->mvs[0].x, | ||
2216 : | (prevMB + iWcount)->mvs[0].y); | ||
2217 : | chl | 96 | |
2218 : | /* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ | ||
2219 : | edgomez | 195 | if (iMinSAD <= thresh2) { |
2220 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
2221 : | goto EPZS16_Terminate_without_Refine; | ||
2222 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
2223 : | goto EPZS16_Terminate_with_Refine; | ||
2224 : | } | ||
2225 : | chl | 96 | |
2226 : | /************ (if Diamond Search) **************/ | ||
2227 : | |||
2228 : | edgomez | 195 | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2229 : | chl | 96 | |
2230 : | chl | 261 | if (MotionFlags & PMV_USESQUARES16) |
2231 : | chl | 184 | MainSearchPtr = Square16_MainSearch; |
2232 : | else | ||
2233 : | chl | 261 | if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
2234 : | chl | 184 | MainSearchPtr = AdvDiamond16_MainSearch; |
2235 : | else | ||
2236 : | MainSearchPtr = Diamond16_MainSearch; | ||
2237 : | |||
2238 : | chl | 96 | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
2239 : | |||
2240 : | edgomez | 195 | iSAD = |
2241 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, | ||
2242 : | chl | 326 | currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2243 : | edgomez | 195 | min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2244 : | |||
2245 : | if (iSAD < iMinSAD) { | ||
2246 : | chl | 96 | *currMV = newMV; |
2247 : | iMinSAD = iSAD; | ||
2248 : | } | ||
2249 : | |||
2250 : | |||
2251 : | edgomez | 195 | if (MotionFlags & PMV_EXTSEARCH16) { |
2252 : | chl | 96 | /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
2253 : | |||
2254 : | edgomez | 195 | if (!(MVequal(pmv[0], backupMV))) { |
2255 : | iSAD = | ||
2256 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, | ||
2257 : | chl | 326 | pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2258 : | edgomez | 195 | min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2259 : | 2, iFcode, iQuant, 0); | ||
2260 : | chl | 96 | } |
2261 : | edgomez | 195 | |
2262 : | if (iSAD < iMinSAD) { | ||
2263 : | chl | 96 | *currMV = newMV; |
2264 : | iMinSAD = iSAD; | ||
2265 : | } | ||
2266 : | edgomez | 195 | |
2267 : | if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { | ||
2268 : | iSAD = | ||
2269 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, | ||
2270 : | chl | 326 | iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2271 : | edgomez | 195 | max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2272 : | |||
2273 : | if (iSAD < iMinSAD) { | ||
2274 : | chl | 96 | *currMV = newMV; |
2275 : | iMinSAD = iSAD; | ||
2276 : | } | ||
2277 : | } | ||
2278 : | } | ||
2279 : | |||
2280 : | /*************** Choose best MV found **************/ | ||
2281 : | |||
2282 : | edgomez | 195 | EPZS16_Terminate_with_Refine: |
2283 : | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step | ||
2284 : | iMinSAD = | ||
2285 : | Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, | ||
2286 : | chl | 326 | iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2287 : | edgomez | 195 | iFcode, iQuant, iEdgedWidth); |
2288 : | chl | 96 | |
2289 : | edgomez | 195 | EPZS16_Terminate_without_Refine: |
2290 : | chl | 96 | |
2291 : | chl | 141 | *oldMB = *prevMB; |
2292 : | edgomez | 195 | |
2293 : | chl | 326 | currPMV->x = currMV->x - center_x; |
2294 : | currPMV->y = currMV->y - center_y; | ||
2295 : | chl | 96 | return iMinSAD; |
2296 : | } | ||
2297 : | |||
2298 : | |||
2299 : | edgomez | 195 | int32_t |
2300 : | EPZSSearch8(const uint8_t * const pRef, | ||
2301 : | const uint8_t * const pRefH, | ||
2302 : | const uint8_t * const pRefV, | ||
2303 : | const uint8_t * const pRefHV, | ||
2304 : | const IMAGE * const pCur, | ||
2305 : | const int x, | ||
2306 : | const int y, | ||
2307 : | const int start_x, | ||
2308 : | const int start_y, | ||
2309 : | chl | 326 | const int center_x, |
2310 : | const int center_y, | ||
2311 : | edgomez | 195 | const uint32_t MotionFlags, |
2312 : | const uint32_t iQuant, | ||
2313 : | const uint32_t iFcode, | ||
2314 : | const MBParam * const pParam, | ||
2315 : | const MACROBLOCK * const pMBs, | ||
2316 : | const MACROBLOCK * const prevMBs, | ||
2317 : | VECTOR * const currMV, | ||
2318 : | VECTOR * const currPMV) | ||
2319 : | chl | 96 | { |
2320 : | chl | 141 | /* Please not that EPZS might not be a good choice for 8x8-block motion search ! */ |
2321 : | |||
2322 : | const uint32_t iWcount = pParam->mb_width; | ||
2323 : | chl | 96 | const int32_t iWidth = pParam->width; |
2324 : | const int32_t iHeight = pParam->height; | ||
2325 : | edgomez | 195 | const int32_t iEdgedWidth = pParam->edged_width; |
2326 : | chl | 96 | |
2327 : | edgomez | 195 | const uint8_t *cur = pCur->y + x * 8 + y * 8 * iEdgedWidth; |
2328 : | chl | 96 | |
2329 : | edgomez | 195 | int32_t iDiamondSize = 1; |
2330 : | |||
2331 : | chl | 96 | int32_t min_dx; |
2332 : | int32_t max_dx; | ||
2333 : | int32_t min_dy; | ||
2334 : | int32_t max_dy; | ||
2335 : | edgomez | 195 | |
2336 : | chl | 96 | VECTOR newMV; |
2337 : | VECTOR backupMV; | ||
2338 : | edgomez | 195 | |
2339 : | chl | 96 | VECTOR pmv[4]; |
2340 : | int32_t psad[8]; | ||
2341 : | |||
2342 : | edgomez | 195 | const int32_t iSubBlock = ((y & 1) << 1) + (x & 1); |
2343 : | chl | 96 | |
2344 : | edgomez | 195 | // const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
2345 : | const MACROBLOCK *const prevMB = prevMBs + (x >> 1) + (y >> 1) * iWcount; | ||
2346 : | chl | 96 | |
2347 : | edgomez | 195 | int32_t bPredEq; |
2348 : | int32_t iMinSAD, iSAD = 9999; | ||
2349 : | |||
2350 : | chl | 181 | MainSearch8FuncPtr MainSearchPtr; |
2351 : | chl | 96 | |
2352 : | /* Get maximum range */ | ||
2353 : | edgomez | 195 | get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, |
2354 : | iFcode); | ||
2355 : | chl | 96 | |
2356 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
2357 : | |||
2358 : | edgomez | 195 | if (!(MotionFlags & PMV_HALFPEL8)) { |
2359 : | min_dx = EVEN(min_dx); | ||
2360 : | max_dx = EVEN(max_dx); | ||
2361 : | min_dy = EVEN(min_dy); | ||
2362 : | max_dy = EVEN(max_dy); | ||
2363 : | } | ||
2364 : | /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ | ||
2365 : | chl | 326 | //bPredEq = get_pmvdata(pMBs, x >> 1, y >> 1, iWcount, iSubBlock, pmv[0].x, pmv[0].y, psad); |
2366 : | chl | 285 | bPredEq = get_pmvdata2(pMBs, iWcount, 0, x >> 1, y >> 1, iSubBlock, pmv, psad); |
2367 : | chl | 96 | |
2368 : | |||
2369 : | /* Step 4: Calculate SAD around the Median prediction. | ||
2370 : | MinSAD=SAD | ||
2371 : | If Motion Vector equal to Previous frame motion vector | ||
2372 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
2373 : | If SAD<=256 goto Step 10. | ||
2374 : | */ | ||
2375 : | |||
2376 : | // Prepare for main loop | ||
2377 : | |||
2378 : | edgomez | 195 | |
2379 : | if (!(MotionFlags & PMV_HALFPEL8)) { | ||
2380 : | chl | 96 | currMV->x = EVEN(currMV->x); |
2381 : | currMV->y = EVEN(currMV->y); | ||
2382 : | } | ||
2383 : | |||
2384 : | edgomez | 195 | if (currMV->x > max_dx) |
2385 : | currMV->x = max_dx; | ||
2386 : | if (currMV->x < min_dx) | ||
2387 : | currMV->x = min_dx; | ||
2388 : | if (currMV->y > max_dy) | ||
2389 : | currMV->y = max_dy; | ||
2390 : | if (currMV->y < min_dy) | ||
2391 : | currMV->y = min_dy; | ||
2392 : | chl | 96 | |
2393 : | edgomez | 195 | /***************** This is predictor SET A: only median prediction ******************/ |
2394 : | chl | 96 | |
2395 : | edgomez | 195 | |
2396 : | iMinSAD = | ||
2397 : | sad8(cur, | ||
2398 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, | ||
2399 : | iEdgedWidth), iEdgedWidth); | ||
2400 : | iMinSAD += | ||
2401 : | chl | 326 | calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
2402 : | edgomez | 195 | (uint8_t) iFcode, iQuant); |
2403 : | |||
2404 : | |||
2405 : | chl | 96 | // thresh1 is fixed to 256 |
2406 : | edgomez | 195 | if (iMinSAD < 256 / 4) { |
2407 : | if (MotionFlags & PMV_QUICKSTOP8) | ||
2408 : | goto EPZS8_Terminate_without_Refine; | ||
2409 : | if (MotionFlags & PMV_EARLYSTOP8) | ||
2410 : | goto EPZS8_Terminate_with_Refine; | ||
2411 : | } | ||
2412 : | chl | 96 | |
2413 : | edgomez | 195 | /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ |
2414 : | chl | 96 | |
2415 : | |||
2416 : | // MV=(0,0) is often a good choice | ||
2417 : | CHECK_MV8_ZERO; | ||
2418 : | |||
2419 : | chl | 141 | // previous frame MV |
2420 : | edgomez | 195 | CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x, prevMB->mvs[iSubBlock].y); |
2421 : | |||
2422 : | chl | 141 | // left neighbour, if allowed |
2423 : | edgomez | 195 | if (psad[1] != MV_MAX_ERROR) { |
2424 : | if (!(MotionFlags & PMV_HALFPEL8)) { | ||
2425 : | pmv[1].x = EVEN(pmv[1].x); | ||
2426 : | chl | 141 | pmv[1].y = EVEN(pmv[1].y); |
2427 : | } | ||
2428 : | edgomez | 195 | CHECK_MV8_CANDIDATE(pmv[1].x, pmv[1].y); |
2429 : | chl | 141 | } |
2430 : | // top neighbour, if allowed | ||
2431 : | edgomez | 195 | if (psad[2] != MV_MAX_ERROR) { |
2432 : | if (!(MotionFlags & PMV_HALFPEL8)) { | ||
2433 : | pmv[2].x = EVEN(pmv[2].x); | ||
2434 : | chl | 141 | pmv[2].y = EVEN(pmv[2].y); |
2435 : | } | ||
2436 : | edgomez | 195 | CHECK_MV8_CANDIDATE(pmv[2].x, pmv[2].y); |
2437 : | |||
2438 : | chl | 141 | // top right neighbour, if allowed |
2439 : | edgomez | 195 | if (psad[3] != MV_MAX_ERROR) { |
2440 : | if (!(MotionFlags & PMV_HALFPEL8)) { | ||
2441 : | pmv[3].x = EVEN(pmv[3].x); | ||
2442 : | chl | 141 | pmv[3].y = EVEN(pmv[3].y); |
2443 : | } | ||
2444 : | edgomez | 195 | CHECK_MV8_CANDIDATE(pmv[3].x, pmv[3].y); |
2445 : | chl | 141 | } |
2446 : | } | ||
2447 : | |||
2448 : | /* // this bias is zero anyway, at the moment! | ||
2449 : | |||
2450 : | if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) ) // && (iMinSAD <= iQuant * 96) | ||
2451 : | iMinSAD -= MV8_00_BIAS; | ||
2452 : | |||
2453 : | edgomez | 195 | */ |
2454 : | chl | 141 | |
2455 : | chl | 96 | /* Terminate if MinSAD <= T_2 |
2456 : | Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] | ||
2457 : | */ | ||
2458 : | |||
2459 : | edgomez | 195 | if (iMinSAD < 512 / 4) { /* T_2 == 512/4 hardcoded */ |
2460 : | if (MotionFlags & PMV_QUICKSTOP8) | ||
2461 : | goto EPZS8_Terminate_without_Refine; | ||
2462 : | if (MotionFlags & PMV_EARLYSTOP8) | ||
2463 : | goto EPZS8_Terminate_with_Refine; | ||
2464 : | } | ||
2465 : | chl | 96 | |
2466 : | chl | 141 | /************ (Diamond Search) **************/ |
2467 : | chl | 96 | |
2468 : | edgomez | 195 | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2469 : | chl | 96 | |
2470 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8)) | ||
2471 : | iDiamondSize *= 2; | ||
2472 : | edgomez | 195 | |
2473 : | chl | 141 | /* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
2474 : | chl | 96 | |
2475 : | chl | 184 | // there is no EPZS^2 for inter4v at the moment |
2476 : | chl | 141 | |
2477 : | edgomez | 195 | // if (MotionFlags & PMV_USESQUARES8) |
2478 : | // MainSearchPtr = Square8_MainSearch; | ||
2479 : | // else | ||
2480 : | chl | 181 | |
2481 : | if (MotionFlags & PMV_ADVANCEDDIAMOND8) | ||
2482 : | MainSearchPtr = AdvDiamond8_MainSearch; | ||
2483 : | else | ||
2484 : | MainSearchPtr = Diamond8_MainSearch; | ||
2485 : | |||
2486 : | edgomez | 195 | iSAD = |
2487 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, | ||
2488 : | chl | 326 | currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2489 : | edgomez | 195 | min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2490 : | iQuant, 0); | ||
2491 : | chl | 96 | |
2492 : | chl | 141 | |
2493 : | edgomez | 195 | if (iSAD < iMinSAD) { |
2494 : | chl | 96 | *currMV = newMV; |
2495 : | iMinSAD = iSAD; | ||
2496 : | } | ||
2497 : | |||
2498 : | edgomez | 195 | if (MotionFlags & PMV_EXTSEARCH8) { |
2499 : | chl | 96 | /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
2500 : | |||
2501 : | edgomez | 195 | if (!(MVequal(pmv[0], backupMV))) { |
2502 : | iSAD = | ||
2503 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, | ||
2504 : | chl | 326 | pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2505 : | edgomez | 195 | min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2506 : | iDiamondSize, iFcode, iQuant, 0); | ||
2507 : | |||
2508 : | if (iSAD < iMinSAD) { | ||
2509 : | chl | 96 | *currMV = newMV; |
2510 : | iMinSAD = iSAD; | ||
2511 : | } | ||
2512 : | } | ||
2513 : | |||
2514 : | edgomez | 195 | if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2515 : | iSAD = | ||
2516 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, | ||
2517 : | chl | 326 | iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2518 : | edgomez | 195 | max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2519 : | iQuant, 0); | ||
2520 : | |||
2521 : | if (iSAD < iMinSAD) { | ||
2522 : | chl | 96 | *currMV = newMV; |
2523 : | iMinSAD = iSAD; | ||
2524 : | } | ||
2525 : | } | ||
2526 : | } | ||
2527 : | |||
2528 : | /*************** Choose best MV found **************/ | ||
2529 : | |||
2530 : | edgomez | 195 | EPZS8_Terminate_with_Refine: |
2531 : | if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step | ||
2532 : | iMinSAD = | ||
2533 : | Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, | ||
2534 : | chl | 326 | iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2535 : | edgomez | 195 | iFcode, iQuant, iEdgedWidth); |
2536 : | chl | 96 | |
2537 : | edgomez | 195 | EPZS8_Terminate_without_Refine: |
2538 : | chl | 96 | |
2539 : | chl | 326 | currPMV->x = currMV->x - center_x; |
2540 : | currPMV->y = currMV->y - center_y; | ||
2541 : | chl | 96 | return iMinSAD; |
2542 : | } | ||
2543 : | |||
2544 : | suxen_drol | 118 | |
2545 : | |||
2546 : | chl | 289 | int32_t |
2547 : | PMVfastIntSearch16(const uint8_t * const pRef, | ||
2548 : | const uint8_t * const pRefH, | ||
2549 : | const uint8_t * const pRefV, | ||
2550 : | const uint8_t * const pRefHV, | ||
2551 : | const IMAGE * const pCur, | ||
2552 : | const int x, | ||
2553 : | const int y, | ||
2554 : | chl | 326 | const int start_x, |
2555 : | const int start_y, | ||
2556 : | const int center_x, | ||
2557 : | const int center_y, | ||
2558 : | chl | 289 | const uint32_t MotionFlags, |
2559 : | const uint32_t iQuant, | ||
2560 : | const uint32_t iFcode, | ||
2561 : | const MBParam * const pParam, | ||
2562 : | const MACROBLOCK * const pMBs, | ||
2563 : | const MACROBLOCK * const prevMBs, | ||
2564 : | VECTOR * const currMV, | ||
2565 : | VECTOR * const currPMV) | ||
2566 : | { | ||
2567 : | const uint32_t iWcount = pParam->mb_width; | ||
2568 : | const int32_t iWidth = pParam->width; | ||
2569 : | const int32_t iHeight = pParam->height; | ||
2570 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
2571 : | suxen_drol | 118 | |
2572 : | chl | 289 | const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; |
2573 : | const VECTOR zeroMV = { 0, 0 }; | ||
2574 : | suxen_drol | 118 | |
2575 : | chl | 289 | int32_t iDiamondSize; |
2576 : | |||
2577 : | int32_t min_dx; | ||
2578 : | int32_t max_dx; | ||
2579 : | int32_t min_dy; | ||
2580 : | int32_t max_dy; | ||
2581 : | |||
2582 : | int32_t iFound; | ||
2583 : | |||
2584 : | VECTOR newMV; | ||
2585 : | VECTOR backupMV; /* just for PMVFAST */ | ||
2586 : | |||
2587 : | VECTOR pmv[4]; | ||
2588 : | int32_t psad[4]; | ||
2589 : | |||
2590 : | MainSearch16FuncPtr MainSearchPtr; | ||
2591 : | |||
2592 : | const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; | ||
2593 : | MACROBLOCK *const pMB = pMBs + x + y * iWcount; | ||
2594 : | |||
2595 : | int32_t threshA, threshB; | ||
2596 : | int32_t bPredEq; | ||
2597 : | int32_t iMinSAD, iSAD; | ||
2598 : | chl | 326 | |
2599 : | chl | 289 | |
2600 : | /* Get maximum range */ | ||
2601 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, | ||
2602 : | iFcode); | ||
2603 : | |||
2604 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
2605 : | |||
2606 : | if ((x == 0) && (y == 0)) { | ||
2607 : | threshA = 512; | ||
2608 : | threshB = 1024; | ||
2609 : | |||
2610 : | bPredEq = 0; | ||
2611 : | psad[0] = psad[1] = psad[2] = psad[3] = 0; | ||
2612 : | *currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; | ||
2613 : | |||
2614 : | } else { | ||
2615 : | threshA = psad[0]; | ||
2616 : | threshB = threshA + 256; | ||
2617 : | if (threshA < 512) | ||
2618 : | threshA = 512; | ||
2619 : | if (threshA > 1024) | ||
2620 : | threshA = 1024; | ||
2621 : | if (threshB > 1792) | ||
2622 : | threshB = 1792; | ||
2623 : | |||
2624 : | bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); | ||
2625 : | *currMV = pmv[0]; /* current best := prediction */ | ||
2626 : | } | ||
2627 : | |||
2628 : | iFound = 0; | ||
2629 : | |||
2630 : | /* Step 4: Calculate SAD around the Median prediction. | ||
2631 : | MinSAD=SAD | ||
2632 : | If Motion Vector equal to Previous frame motion vector | ||
2633 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
2634 : | If SAD<=256 goto Step 10. | ||
2635 : | */ | ||
2636 : | |||
2637 : | if (currMV->x > max_dx) { | ||
2638 : | currMV->x = EVEN(max_dx); | ||
2639 : | } | ||
2640 : | if (currMV->x < min_dx) { | ||
2641 : | currMV->x = EVEN(min_dx); | ||
2642 : | } | ||
2643 : | if (currMV->y > max_dy) { | ||
2644 : | currMV->y = EVEN(max_dy); | ||
2645 : | } | ||
2646 : | if (currMV->y < min_dy) { | ||
2647 : | currMV->y = EVEN(min_dy); | ||
2648 : | } | ||
2649 : | |||
2650 : | iMinSAD = | ||
2651 : | sad16(cur, | ||
2652 : | chl | 312 | get_iref_mv(pRef, x, y, 16, currMV, |
2653 : | chl | 289 | iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2654 : | iMinSAD += | ||
2655 : | chl | 326 | calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2656 : | chl | 289 | (uint8_t) iFcode, iQuant); |
2657 : | |||
2658 : | if ((iMinSAD < 256) || | ||
2659 : | ((MVequal(*currMV, prevMB->i_mvs[0])) && | ||
2660 : | ((int32_t) iMinSAD < prevMB->i_sad16))) { | ||
2661 : | if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode | ||
2662 : | { | ||
2663 : | if (!MVzero(*currMV)) { | ||
2664 : | iMinSAD += MV16_00_BIAS; | ||
2665 : | CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures | ||
2666 : | iMinSAD -= MV16_00_BIAS; | ||
2667 : | } | ||
2668 : | } | ||
2669 : | |||
2670 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
2671 : | goto PMVfastInt16_Terminate_with_Refine; | ||
2672 : | } | ||
2673 : | |||
2674 : | |||
2675 : | /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion | ||
2676 : | vector of the median. | ||
2677 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
2678 : | */ | ||
2679 : | |||
2680 : | if ((bPredEq) && (MVequal(pmv[0], prevMB->i_mvs[0]))) | ||
2681 : | iFound = 2; | ||
2682 : | |||
2683 : | /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
2684 : | Otherwise select large Diamond Search. | ||
2685 : | */ | ||
2686 : | |||
2687 : | if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) | ||
2688 : | iDiamondSize = 2; // halfpel units! | ||
2689 : | else | ||
2690 : | iDiamondSize = 4; // halfpel units! | ||
2691 : | |||
2692 : | /* | ||
2693 : | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. | ||
2694 : | Also calculate (0,0) but do not subtract offset. | ||
2695 : | Let MinSAD be the smallest SAD up to this point. | ||
2696 : | If MV is (0,0) subtract offset. | ||
2697 : | */ | ||
2698 : | |||
2699 : | // (0,0) is often a good choice | ||
2700 : | |||
2701 : | if (!MVzero(pmv[0])) | ||
2702 : | CHECK_MV16_ZERO; | ||
2703 : | |||
2704 : | // previous frame MV is always possible | ||
2705 : | |||
2706 : | if (!MVzero(prevMB->i_mvs[0])) | ||
2707 : | if (!MVequal(prevMB->i_mvs[0], pmv[0])) | ||
2708 : | CHECK_MV16_CANDIDATE(prevMB->i_mvs[0].x, prevMB->i_mvs[0].y); | ||
2709 : | |||
2710 : | // left neighbour, if allowed | ||
2711 : | |||
2712 : | if (!MVzero(pmv[1])) | ||
2713 : | if (!MVequal(pmv[1], prevMB->i_mvs[0])) | ||
2714 : | if (!MVequal(pmv[1], pmv[0])) | ||
2715 : | CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); | ||
2716 : | |||
2717 : | // top neighbour, if allowed | ||
2718 : | if (!MVzero(pmv[2])) | ||
2719 : | if (!MVequal(pmv[2], prevMB->i_mvs[0])) | ||
2720 : | if (!MVequal(pmv[2], pmv[0])) | ||
2721 : | if (!MVequal(pmv[2], pmv[1])) | ||
2722 : | CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); | ||
2723 : | |||
2724 : | // top right neighbour, if allowed | ||
2725 : | if (!MVzero(pmv[3])) | ||
2726 : | if (!MVequal(pmv[3], prevMB->i_mvs[0])) | ||
2727 : | if (!MVequal(pmv[3], pmv[0])) | ||
2728 : | if (!MVequal(pmv[3], pmv[1])) | ||
2729 : | if (!MVequal(pmv[3], pmv[2])) | ||
2730 : | CHECK_MV16_CANDIDATE(pmv[3].x, | ||
2731 : | pmv[3].y); | ||
2732 : | |||
2733 : | if ((MVzero(*currMV)) && | ||
2734 : | (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) | ||
2735 : | iMinSAD -= MV16_00_BIAS; | ||
2736 : | |||
2737 : | |||
2738 : | /* Step 6: If MinSAD <= thresa goto Step 10. | ||
2739 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
2740 : | */ | ||
2741 : | |||
2742 : | if ((iMinSAD <= threshA) || | ||
2743 : | (MVequal(*currMV, prevMB->i_mvs[0]) && | ||
2744 : | ((int32_t) iMinSAD < prevMB->i_sad16))) { | ||
2745 : | |||
2746 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
2747 : | goto PMVfastInt16_Terminate_with_Refine; | ||
2748 : | } | ||
2749 : | |||
2750 : | |||
2751 : | /************ (Diamond Search) **************/ | ||
2752 : | /* | ||
2753 : | Step 7: Perform Diamond search, with either the small or large diamond. | ||
2754 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
2755 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
2756 : | If center then goto step 10. | ||
2757 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
2758 : | Refine by using small diamond and goto step 10. | ||
2759 : | */ | ||
2760 : | |||
2761 : | if (MotionFlags & PMV_USESQUARES16) | ||
2762 : | MainSearchPtr = Square16_MainSearch; | ||
2763 : | else if (MotionFlags & PMV_ADVANCEDDIAMOND16) | ||
2764 : | MainSearchPtr = AdvDiamond16_MainSearch; | ||
2765 : | else | ||
2766 : | MainSearchPtr = Diamond16_MainSearch; | ||
2767 : | |||
2768 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
2769 : | |||
2770 : | |||
2771 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
2772 : | iSAD = | ||
2773 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, | ||
2774 : | chl | 326 | currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2775 : | chl | 289 | min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2776 : | iQuant, iFound); | ||
2777 : | |||
2778 : | if (iSAD < iMinSAD) { | ||
2779 : | *currMV = newMV; | ||
2780 : | iMinSAD = iSAD; | ||
2781 : | } | ||
2782 : | |||
2783 : | if (MotionFlags & PMV_EXTSEARCH16) { | ||
2784 : | /* extended: search (up to) two more times: orignal prediction and (0,0) */ | ||
2785 : | |||
2786 : | if (!(MVequal(pmv[0], backupMV))) { | ||
2787 : | iSAD = | ||
2788 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, | ||
2789 : | chl | 326 | pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2790 : | chl | 289 | min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2791 : | iDiamondSize, iFcode, iQuant, iFound); | ||
2792 : | |||
2793 : | if (iSAD < iMinSAD) { | ||
2794 : | *currMV = newMV; | ||
2795 : | iMinSAD = iSAD; | ||
2796 : | } | ||
2797 : | } | ||
2798 : | |||
2799 : | if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { | ||
2800 : | iSAD = | ||
2801 : | (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, | ||
2802 : | chl | 326 | iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2803 : | chl | 289 | max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2804 : | iQuant, iFound); | ||
2805 : | |||
2806 : | if (iSAD < iMinSAD) { | ||
2807 : | *currMV = newMV; | ||
2808 : | iMinSAD = iSAD; | ||
2809 : | } | ||
2810 : | } | ||
2811 : | } | ||
2812 : | |||
2813 : | /* | ||
2814 : | Step 10: The motion vector is chosen according to the block corresponding to MinSAD. | ||
2815 : | */ | ||
2816 : | |||
2817 : | PMVfastInt16_Terminate_with_Refine: | ||
2818 : | |||
2819 : | pMB->i_mvs[0] = pMB->i_mvs[1] = pMB->i_mvs[2] = pMB->i_mvs[3] = pMB->i_mv16 = *currMV; | ||
2820 : | pMB->i_sad8[0] = pMB->i_sad8[1] = pMB->i_sad8[2] = pMB->i_sad8[3] = pMB->i_sad16 = iMinSAD; | ||
2821 : | |||
2822 : | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step | ||
2823 : | iMinSAD = | ||
2824 : | Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, | ||
2825 : | chl | 326 | iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2826 : | chl | 289 | iFcode, iQuant, iEdgedWidth); |
2827 : | |||
2828 : | pmv[0] = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); // get _REAL_ prediction (halfpel possible) | ||
2829 : | |||
2830 : | PMVfastInt16_Terminate_without_Refine: | ||
2831 : | chl | 326 | currPMV->x = currMV->x - center_x; |
2832 : | currPMV->y = currMV->y - center_y; | ||
2833 : | chl | 289 | return iMinSAD; |
2834 : | } | ||
2835 : | |||
2836 : | |||
2837 : | |||
2838 : | suxen_drol | 118 | /* *********************************************************** |
2839 : | bvop motion estimation | ||
2840 : | // TODO: need to incorporate prediction here (eg. sad += calc_delta_16) | ||
2841 : | ***************************************************************/ | ||
2842 : | |||
2843 : | suxen_drol | 152 | |
2844 : | chl | 326 | #define DIRECT_PENALTY 0 |
2845 : | #define DIRECT_UPPERLIMIT 256 // never use direct mode if SAD is larger than this | ||
2846 : | |||
2847 : | edgomez | 195 | void |
2848 : | MotionEstimationBVOP(MBParam * const pParam, | ||
2849 : | FRAMEINFO * const frame, | ||
2850 : | chl | 317 | const int32_t time_bp, |
2851 : | const int32_t time_pp, | ||
2852 : | edgomez | 195 | // forward (past) reference |
2853 : | const MACROBLOCK * const f_mbs, | ||
2854 : | const IMAGE * const f_ref, | ||
2855 : | const IMAGE * const f_refH, | ||
2856 : | const IMAGE * const f_refV, | ||
2857 : | const IMAGE * const f_refHV, | ||
2858 : | // backward (future) reference | ||
2859 : | const MACROBLOCK * const b_mbs, | ||
2860 : | const IMAGE * const b_ref, | ||
2861 : | const IMAGE * const b_refH, | ||
2862 : | const IMAGE * const b_refV, | ||
2863 : | const IMAGE * const b_refHV) | ||
2864 : | suxen_drol | 118 | { |
2865 : | chl | 312 | const int mb_width = pParam->mb_width; |
2866 : | const int mb_height = pParam->mb_height; | ||
2867 : | const int edged_width = pParam->edged_width; | ||
2868 : | suxen_drol | 118 | |
2869 : | chl | 317 | int i, j, k; |
2870 : | edgomez | 195 | |
2871 : | chl | 317 | static const VECTOR zeroMV={0,0}; |
2872 : | chl | 312 | |
2873 : | int f_sad16; /* forward (as usual) search */ | ||
2874 : | int b_sad16; /* backward (only in b-frames) search */ | ||
2875 : | int i_sad16; /* interpolated (both direction, b-frames only) */ | ||
2876 : | int d_sad16; /* direct mode (assume linear motion) */ | ||
2877 : | suxen_drol | 118 | |
2878 : | chl | 312 | int best_sad; |
2879 : | |||
2880 : | chl | 341 | VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
2881 : | suxen_drol | 118 | VECTOR pmv_dontcare; |
2882 : | |||
2883 : | chl | 312 | int f_count=0; |
2884 : | int b_count=0; | ||
2885 : | int i_count=0; | ||
2886 : | int d_count=0; | ||
2887 : | int s_count=0; | ||
2888 : | chl | 341 | |
2889 : | chl | 317 | const int64_t TRB = (int32_t)time_pp - (int32_t)time_bp; |
2890 : | chl | 326 | const int64_t TRD = (int32_t)time_pp; |
2891 : | chl | 317 | |
2892 : | chl | 318 | // fprintf(stderr,"TRB = %lld TRD = %lld time_bp =%d time_pp =%d\n\n",TRB,TRD,time_bp,time_pp); |
2893 : | suxen_drol | 118 | // note: i==horizontal, j==vertical |
2894 : | edgomez | 195 | for (j = 0; j < mb_height; j++) { |
2895 : | chl | 326 | |
2896 : | f_predMV = zeroMV; /* prediction is reset at left boundary */ | ||
2897 : | b_predMV = zeroMV; | ||
2898 : | |||
2899 : | edgomez | 195 | for (i = 0; i < mb_width; i++) { |
2900 : | MACROBLOCK *mb = &frame->mbs[i + j * mb_width]; | ||
2901 : | const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; | ||
2902 : | const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; | ||
2903 : | suxen_drol | 118 | |
2904 : | chl | 341 | mb->deltamv=zeroMV; |
2905 : | chl | 317 | |
2906 : | chl | 312 | /* special case, if collocated block is SKIPed: encoding is forward(0,0) */ |
2907 : | |||
2908 : | chl | 337 | #ifndef _DISABLE_SKIP |
2909 : | edgomez | 195 | if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
2910 : | chl | 312 | b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
2911 : | suxen_drol | 152 | mb->mode = MODE_NOT_CODED; |
2912 : | suxen_drol | 118 | mb->mvs[0].x = 0; |
2913 : | mb->mvs[0].y = 0; | ||
2914 : | mb->b_mvs[0].x = 0; | ||
2915 : | mb->b_mvs[0].y = 0; | ||
2916 : | continue; | ||
2917 : | } | ||
2918 : | chl | 337 | #endif |
2919 : | suxen_drol | 118 | |
2920 : | chl | 341 | d_sad16 = DIRECT_PENALTY; |
2921 : | suxen_drol | 234 | |
2922 : | chl | 326 | if (b_mb->mode == MODE_INTER4V) |
2923 : | { | ||
2924 : | chl | 317 | |
2925 : | chl | 326 | /* same method of scaling as in decoder.c, so we copy from there */ |
2926 : | for (k = 0; k < 4; k++) { | ||
2927 : | |||
2928 : | chl | 341 | mb->directmv[k] = b_mb->mvs[k]; |
2929 : | chl | 326 | |
2930 : | chl | 341 | mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
2931 : | mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) | ||
2932 : | ? ((TRB - TRD) * mb->directmv[k].x) / TRD | ||
2933 : | : mb->mvs[k].x - mb->directmv[k].x); | ||
2934 : | chl | 326 | |
2935 : | chl | 341 | mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
2936 : | mb->b_mvs[k].y = (int32_t) ((mb->directmv[k].y == 0) | ||
2937 : | ? ((TRB - TRD) * mb->directmv[k].y) / TRD | ||
2938 : | : mb->mvs[k].y - mb->directmv[k].y); | ||
2939 : | |||
2940 : | d_sad16 += | ||
2941 : | chl | 326 | sad8bi(frame->image.y + 2*(i+(k&1))*8 + 2*(j+(k>>1))*8*edged_width, |
2942 : | chl | 312 | get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2943 : | chl | 326 | 2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->mvs[k], edged_width), |
2944 : | get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
2945 : | 2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->b_mvs[k], edged_width), | ||
2946 : | edged_width); | ||
2947 : | } | ||
2948 : | } | ||
2949 : | else | ||
2950 : | { | ||
2951 : | chl | 341 | mb->directmv[3] = mb->directmv[2] = mb->directmv[1] = |
2952 : | mb->directmv[0] = b_mb->mvs[0]; | ||
2953 : | chl | 326 | |
2954 : | chl | 341 | mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
2955 : | mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) | ||
2956 : | ? ((TRB - TRD) * mb->directmv[0].x) / TRD | ||
2957 : | : mb->mvs[0].x - mb->directmv[0].x); | ||
2958 : | chl | 326 | |
2959 : | chl | 341 | mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
2960 : | mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) | ||
2961 : | ? ((TRB - TRD) * mb->directmv[0].y) / TRD | ||
2962 : | : mb->mvs[0].y - mb->directmv[0].y); | ||
2963 : | |||
2964 : | d_sad16 += sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, | ||
2965 : | chl | 326 | get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2966 : | chl | 312 | i, j, 16, &mb->mvs[0], edged_width), |
2967 : | get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
2968 : | i, j, 16, &mb->b_mvs[0], edged_width), | ||
2969 : | edged_width); | ||
2970 : | chl | 326 | |
2971 : | } | ||
2972 : | chl | 341 | d_sad16 += calc_delta_16(mb->deltamv.x, mb->deltamv.y, 1, frame->quant); |
2973 : | chl | 326 | |
2974 : | suxen_drol | 118 | // forward search |
2975 : | chl | 312 | f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2976 : | chl | 326 | &frame->image, i, j, |
2977 : | mb->mvs[0].x, mb->mvs[0].y, /* start point f_directMV */ | ||
2978 : | f_predMV.x, f_predMV.y, /* center is f-prediction */ | ||
2979 : | chl | 341 | frame->motion_flags, |
2980 : | chl | 312 | frame->quant, frame->fcode, pParam, |
2981 : | chl | 326 | f_mbs, f_mbs, |
2982 : | &mb->mvs[0], &pmv_dontcare); | ||
2983 : | suxen_drol | 118 | |
2984 : | chl | 312 | |
2985 : | suxen_drol | 118 | // backward search |
2986 : | suxen_drol | 234 | b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
2987 : | chl | 326 | &frame->image, i, j, |
2988 : | mb->b_mvs[0].x, mb->b_mvs[0].y, /* start point b_directMV */ | ||
2989 : | b_predMV.x, b_predMV.y, /* center is b-prediction */ | ||
2990 : | chl | 341 | frame->motion_flags, |
2991 : | suxen_drol | 234 | frame->quant, frame->bcode, pParam, |
2992 : | chl | 326 | b_mbs, b_mbs, |
2993 : | &mb->b_mvs[0], &pmv_dontcare); | ||
2994 : | |||
2995 : | edgomez | 195 | i_sad16 = |
2996 : | chl | 312 | sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
2997 : | get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, | ||
2998 : | i, j, 16, &mb->mvs[0], edged_width), | ||
2999 : | chl | 326 | get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3000 : | i, j, 16, &mb->b_mvs[0], edged_width), | ||
3001 : | edgomez | 195 | edged_width); |
3002 : | chl | 326 | i_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, |
3003 : | frame->fcode, frame->quant); | ||
3004 : | chl | 341 | i_sad16 += calc_delta_16(mb->b_mvs[0].x-b_predMV.x, mb->b_mvs[0].y-b_predMV.y, |
3005 : | chl | 326 | frame->bcode, frame->quant); |
3006 : | suxen_drol | 118 | |
3007 : | // TODO: direct search | ||
3008 : | chl | 326 | // predictor + delta vector in range [-32,32] (fcode=1) |
3009 : | |||
3010 : | chl | 341 | i_sad16 = 65535; |
3011 : | f_sad16 = 65535; | ||
3012 : | b_sad16 = 65535; | ||
3013 : | // d_sad16 = 65535; | ||
3014 : | chl | 312 | |
3015 : | edgomez | 195 | if (f_sad16 < b_sad16) { |
3016 : | suxen_drol | 118 | best_sad = f_sad16; |
3017 : | suxen_drol | 152 | mb->mode = MODE_FORWARD; |
3018 : | edgomez | 195 | } else { |
3019 : | suxen_drol | 118 | best_sad = b_sad16; |
3020 : | suxen_drol | 152 | mb->mode = MODE_BACKWARD; |
3021 : | suxen_drol | 118 | } |
3022 : | edgomez | 195 | |
3023 : | if (i_sad16 < best_sad) { | ||
3024 : | suxen_drol | 118 | best_sad = i_sad16; |
3025 : | suxen_drol | 152 | mb->mode = MODE_INTERPOLATE; |
3026 : | suxen_drol | 118 | } |
3027 : | |||
3028 : | chl | 341 | if (d_sad16 < best_sad) { |
3029 : | chl | 326 | |
3030 : | chl | 341 | if (b_mb->mode == MODE_INTER4V) |
3031 : | { | ||
3032 : | chl | 326 | |
3033 : | chl | 341 | /* same method of scaling as in decoder.c, so we copy from there */ |
3034 : | for (k = 0; k < 4; k++) { | ||
3035 : | chl | 326 | |
3036 : | chl | 341 | mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
3037 : | mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) | ||
3038 : | ? ((TRB - TRD) * mb->directmv[k].x) / TRD | ||
3039 : | : mb->mvs[k].x - mb->directmv[k].x); | ||
3040 : | chl | 326 | |
3041 : | chl | 341 | mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
3042 : | mb->b_mvs[k].y = (int32_t) ((mb->directmv[k].y == 0) | ||
3043 : | ? ((TRB - TRD) * mb->directmv[k].y) / TRD | ||
3044 : | : mb->mvs[k].y - mb->directmv[k].y); | ||
3045 : | } | ||
3046 : | chl | 326 | } |
3047 : | else | ||
3048 : | { | ||
3049 : | chl | 341 | mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
3050 : | |||
3051 : | mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) | ||
3052 : | ? ((TRB - TRD) * mb->directmv[0].x) / TRD | ||
3053 : | : mb->mvs[0].x - mb->directmv[0].x); | ||
3054 : | |||
3055 : | mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); | ||
3056 : | |||
3057 : | mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) | ||
3058 : | ? ((TRB - TRD) * mb->directmv[0].y) / TRD | ||
3059 : | : mb->mvs[0].y - mb->directmv[0].y); | ||
3060 : | |||
3061 : | mb->mvs[3] = mb->mvs[2] = mb->mvs[1] = mb->mvs[0]; | ||
3062 : | mb->b_mvs[3] = mb->b_mvs[2] = mb->b_mvs[1] = mb->b_mvs[0]; | ||
3063 : | } | ||
3064 : | |||
3065 : | best_sad = d_sad16; | ||
3066 : | mb->mode = MODE_DIRECT; | ||
3067 : | mb->mode = MODE_INTERPOLATE; // direct mode still broken :-( | ||
3068 : | suxen_drol | 118 | } |
3069 : | |||
3070 : | chl | 312 | switch (mb->mode) |
3071 : | { | ||
3072 : | case MODE_FORWARD: | ||
3073 : | chl | 326 | f_count++; |
3074 : | f_predMV = mb->mvs[0]; | ||
3075 : | break; | ||
3076 : | chl | 312 | case MODE_BACKWARD: |
3077 : | chl | 326 | b_count++; |
3078 : | b_predMV = mb->b_mvs[0]; | ||
3079 : | |||
3080 : | break; | ||
3081 : | chl | 312 | case MODE_INTERPOLATE: |
3082 : | chl | 326 | i_count++; |
3083 : | f_predMV = mb->mvs[0]; | ||
3084 : | b_predMV = mb->b_mvs[0]; | ||
3085 : | break; | ||
3086 : | chl | 312 | case MODE_DIRECT: |
3087 : | chl | 341 | d_count++; |
3088 : | break; | ||
3089 : | chl | 312 | default: |
3090 : | chl | 341 | s_count++; // ??? |
3091 : | break; | ||
3092 : | chl | 312 | } |
3093 : | |||
3094 : | suxen_drol | 118 | } |
3095 : | } | ||
3096 : | chl | 312 | |
3097 : | #ifdef _DEBUG_BFRAME_STAT | ||
3098 : | chl | 341 | fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D: %04d S: %04d\n", |
3099 : | f_count,b_count,i_count,d_count,s_count); | ||
3100 : | chl | 312 | #endif |
3101 : | |||
3102 : | suxen_drol | 118 | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |