Parent Directory | Revision Log
Revision 1.15 - (view) (download)
1 : | Isibaar | 1.1 | /************************************************************************** |
2 : | * | ||
3 : | * Modifications: | ||
4 : | * | ||
5 : | suxen_drol | 1.11 | * 01.05.2002 updated MotionEstimationBVOP |
6 : | suxen_drol | 1.8 | * 25.04.2002 partial prevMB conversion |
7 : | chenm001 | 1.7 | * 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
8 : | * 14.04.2002 added MotionEstimationBVOP() | ||
9 : | chl | 1.3 | * 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
10 : | * EPZS and EPZS^2 | ||
11 : | Isibaar | 1.1 | * 08.02.2002 split up PMVfast into three routines: PMVFast, PMVFast_MainLoop |
12 : | * PMVFast_Refine to support multiple searches with different start points | ||
13 : | edgomez | 1.2 | * 07.01.2002 uv-block-based interpolation |
14 : | Isibaar | 1.1 | * 06.01.2002 INTER/INTRA-decision is now done before any SEARCH8 (speedup) |
15 : | edgomez | 1.2 | * changed INTER_BIAS to 150 (as suggested by suxen_drol) |
16 : | * removed halfpel refinement step in PMVfastSearch8 + quality=5 | ||
17 : | * added new quality mode = 6 which performs halfpel refinement | ||
18 : | * filesize difference between quality 5 and 6 is smaller than 1% | ||
19 : | Isibaar | 1.1 | * (Isibaar) |
20 : | * 31.12.2001 PMVfastSearch16 and PMVfastSearch8 (gruel) | ||
21 : | edgomez | 1.2 | * 30.12.2001 get_range/MotionSearchX simplified; blue/green bug fix |
22 : | * 22.12.2001 commented best_point==99 check | ||
23 : | * 19.12.2001 modified get_range (purple bug fix) | ||
24 : | Isibaar | 1.1 | * 15.12.2001 moved pmv displacement from mbprediction |
25 : | * 02.12.2001 motion estimation/compensation split (Isibaar) | ||
26 : | edgomez | 1.2 | * 16.11.2001 rewrote/tweaked search algorithms; pross@cs.rmit.edu.au |
27 : | Isibaar | 1.1 | * 10.11.2001 support for sad16/sad8 functions |
28 : | * 28.08.2001 reactivated MODE_INTER4V for EXT_MODE | ||
29 : | * 24.08.2001 removed MODE_INTER4V_Q, disabled MODE_INTER4V for EXT_MODE | ||
30 : | edgomez | 1.2 | * 22.08.2001 added MODE_INTER4V_Q |
31 : | Isibaar | 1.1 | * 20.08.2001 added pragma to get rid of internal compiler error with VC6 |
32 : | * idea by Cyril. Thanks. | ||
33 : | * | ||
34 : | * Michael Militzer <isibaar@videocoding.de> | ||
35 : | * | ||
36 : | **************************************************************************/ | ||
37 : | |||
38 : | #include <assert.h> | ||
39 : | #include <stdio.h> | ||
40 : | chl | 1.3 | #include <stdlib.h> |
41 : | Isibaar | 1.1 | |
42 : | #include "../encoder.h" | ||
43 : | #include "../utils/mbfunctions.h" | ||
44 : | #include "../prediction/mbprediction.h" | ||
45 : | #include "../global.h" | ||
46 : | #include "../utils/timer.h" | ||
47 : | suxen_drol | 1.6 | #include "motion.h" |
48 : | Isibaar | 1.1 | #include "sad.h" |
49 : | |||
50 : | // very large value | ||
51 : | #define MV_MAX_ERROR (4096 * 256) | ||
52 : | |||
53 : | // stop search if sdelta < THRESHOLD | ||
54 : | #define MV16_THRESHOLD 192 | ||
55 : | #define MV8_THRESHOLD 56 | ||
56 : | |||
57 : | /* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ | ||
58 : | /* nb = vop pixels * 2^(bpp-8) */ | ||
59 : | #define MV16_00_BIAS (128+1) | ||
60 : | chl | 1.9 | #define MV8_00_BIAS (0) |
61 : | Isibaar | 1.1 | |
62 : | /* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ | ||
63 : | chl | 1.15 | #define MV16_INTER_BIAS 512 |
64 : | Isibaar | 1.1 | |
65 : | /* Parameters which control inter/inter4v decision */ | ||
66 : | #define IMV16X16 5 | ||
67 : | |||
68 : | /* vector map (vlc delta size) smoother parameters */ | ||
69 : | #define NEIGH_TEND_16X16 2 | ||
70 : | #define NEIGH_TEND_8X8 2 | ||
71 : | |||
72 : | // fast ((A)/2)*2 | ||
73 : | #define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) | ||
74 : | |||
75 : | chl | 1.15 | #define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
76 : | #define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) | ||
77 : | Isibaar | 1.1 | |
78 : | chl | 1.3 | int32_t PMVfastSearch16( |
79 : | const uint8_t * const pRef, | ||
80 : | const uint8_t * const pRefH, | ||
81 : | const uint8_t * const pRefV, | ||
82 : | const uint8_t * const pRefHV, | ||
83 : | const IMAGE * const pCur, | ||
84 : | const int x, const int y, | ||
85 : | const uint32_t MotionFlags, | ||
86 : | suxen_drol | 1.8 | const uint32_t iQuant, |
87 : | const uint32_t iFcode, | ||
88 : | chl | 1.3 | const MBParam * const pParam, |
89 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
90 : | const MACROBLOCK * const prevMBs, | ||
91 : | chl | 1.3 | VECTOR * const currMV, |
92 : | VECTOR * const currPMV); | ||
93 : | |||
94 : | int32_t EPZSSearch16( | ||
95 : | const uint8_t * const pRef, | ||
96 : | const uint8_t * const pRefH, | ||
97 : | const uint8_t * const pRefV, | ||
98 : | const uint8_t * const pRefHV, | ||
99 : | const IMAGE * const pCur, | ||
100 : | const int x, const int y, | ||
101 : | const uint32_t MotionFlags, | ||
102 : | suxen_drol | 1.8 | const uint32_t iQuant, |
103 : | const uint32_t iFcode, | ||
104 : | chl | 1.3 | const MBParam * const pParam, |
105 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
106 : | const MACROBLOCK * const prevMBs, | ||
107 : | chl | 1.3 | VECTOR * const currMV, |
108 : | VECTOR * const currPMV); | ||
109 : | |||
110 : | Isibaar | 1.1 | |
111 : | int32_t PMVfastSearch8( | ||
112 : | chl | 1.3 | const uint8_t * const pRef, |
113 : | const uint8_t * const pRefH, | ||
114 : | const uint8_t * const pRefV, | ||
115 : | const uint8_t * const pRefHV, | ||
116 : | const IMAGE * const pCur, | ||
117 : | const int x, const int y, | ||
118 : | suxen_drol | 1.8 | const int start_x, const int start_y, |
119 : | chl | 1.3 | const uint32_t MotionFlags, |
120 : | suxen_drol | 1.8 | const uint32_t iQuant, |
121 : | const uint32_t iFcode, | ||
122 : | chl | 1.3 | const MBParam * const pParam, |
123 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
124 : | const MACROBLOCK * const prevMBs, | ||
125 : | chl | 1.3 | VECTOR * const currMV, |
126 : | VECTOR * const currPMV); | ||
127 : | |||
128 : | int32_t EPZSSearch8( | ||
129 : | const uint8_t * const pRef, | ||
130 : | const uint8_t * const pRefH, | ||
131 : | const uint8_t * const pRefV, | ||
132 : | const uint8_t * const pRefHV, | ||
133 : | const IMAGE * const pCur, | ||
134 : | const int x, const int y, | ||
135 : | suxen_drol | 1.8 | const int start_x, const int start_y, |
136 : | chl | 1.3 | const uint32_t MotionFlags, |
137 : | suxen_drol | 1.8 | const uint32_t iQuant, |
138 : | const uint32_t iFcode, | ||
139 : | chl | 1.3 | const MBParam * const pParam, |
140 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
141 : | const MACROBLOCK * const prevMBs, | ||
142 : | chl | 1.3 | VECTOR * const currMV, |
143 : | VECTOR * const currPMV); | ||
144 : | |||
145 : | |||
146 : | typedef int32_t (MainSearch16Func)( | ||
147 : | edgomez | 1.2 | const uint8_t * const pRef, |
148 : | const uint8_t * const pRefH, | ||
149 : | const uint8_t * const pRefV, | ||
150 : | const uint8_t * const pRefHV, | ||
151 : | chl | 1.3 | const uint8_t * const cur, |
152 : | edgomez | 1.2 | const int x, const int y, |
153 : | chl | 1.3 | int32_t startx, int32_t starty, |
154 : | int32_t iMinSAD, | ||
155 : | edgomez | 1.2 | VECTOR * const currMV, |
156 : | chl | 1.3 | const VECTOR * const pmv, |
157 : | const int32_t min_dx, const int32_t max_dx, | ||
158 : | const int32_t min_dy, const int32_t max_dy, | ||
159 : | const int32_t iEdgedWidth, | ||
160 : | const int32_t iDiamondSize, | ||
161 : | const int32_t iFcode, | ||
162 : | const int32_t iQuant, | ||
163 : | int iFound); | ||
164 : | |||
165 : | typedef MainSearch16Func* MainSearch16FuncPtr; | ||
166 : | Isibaar | 1.1 | |
167 : | chl | 1.3 | |
168 : | typedef int32_t (MainSearch8Func)( | ||
169 : | edgomez | 1.2 | const uint8_t * const pRef, |
170 : | const uint8_t * const pRefH, | ||
171 : | const uint8_t * const pRefV, | ||
172 : | const uint8_t * const pRefHV, | ||
173 : | chl | 1.3 | const uint8_t * const cur, |
174 : | edgomez | 1.2 | const int x, const int y, |
175 : | chl | 1.3 | int32_t startx, int32_t starty, |
176 : | int32_t iMinSAD, | ||
177 : | edgomez | 1.2 | VECTOR * const currMV, |
178 : | chl | 1.3 | const VECTOR * const pmv, |
179 : | const int32_t min_dx, const int32_t max_dx, | ||
180 : | const int32_t min_dy, const int32_t max_dy, | ||
181 : | const int32_t iEdgedWidth, | ||
182 : | const int32_t iDiamondSize, | ||
183 : | const int32_t iFcode, | ||
184 : | const int32_t iQuant, | ||
185 : | int iFound); | ||
186 : | Isibaar | 1.1 | |
187 : | chl | 1.3 | typedef MainSearch8Func* MainSearch8FuncPtr; |
188 : | Isibaar | 1.1 | |
189 : | chl | 1.10 | static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
190 : | { 0 ,(int)(1.00235+0.5), (int)(1.15582+0.5), (int)(1.31976+0.5), (int)(1.49591+0.5), (int)(1.68601+0.5), | ||
191 : | (int)(1.89187+0.5), (int)(2.11542+0.5), (int)(2.35878+0.5), (int)(2.62429+0.5), (int)(2.91455+0.5), | ||
192 : | (int)(3.23253+0.5), (int)(3.58158+0.5), (int)(3.96555+0.5), (int)(4.38887+0.5), (int)(4.85673+0.5), | ||
193 : | (int)(5.37519+0.5), (int)(5.95144+0.5), (int)(6.59408+0.5), (int)(7.31349+0.5), (int)(8.12242+0.5), | ||
194 : | (int)(9.03669+0.5), (int)(10.0763+0.5), (int)(11.2669+0.5), (int)(12.6426+0.5), (int)(14.2493+0.5), | ||
195 : | (int)(16.1512+0.5), (int)(18.442+0.5), (int)(21.2656+0.5), (int)(24.8580+0.5), (int)(29.6436+0.5), | ||
196 : | (int)(36.4949+0.5) }; | ||
197 : | |||
198 : | static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now*/ | ||
199 : | |||
200 : | |||
201 : | |||
202 : | Isibaar | 1.1 | // mv.length table |
203 : | static const uint32_t mvtab[33] = { | ||
204 : | 1, 2, 3, 4, 6, 7, 7, 7, | ||
205 : | 9, 9, 9, 10, 10, 10, 10, 10, | ||
206 : | 10, 10, 10, 10, 10, 10, 10, 10, | ||
207 : | 10, 11, 11, 11, 11, 11, 11, 12, 12 | ||
208 : | }; | ||
209 : | |||
210 : | |||
211 : | static __inline uint32_t mv_bits(int32_t component, const uint32_t iFcode) | ||
212 : | { | ||
213 : | if (component == 0) | ||
214 : | return 1; | ||
215 : | |||
216 : | if (component < 0) | ||
217 : | component = -component; | ||
218 : | |||
219 : | if (iFcode == 1) | ||
220 : | { | ||
221 : | if (component > 32) | ||
222 : | component = 32; | ||
223 : | |||
224 : | return mvtab[component] + 1; | ||
225 : | } | ||
226 : | |||
227 : | component += (1 << (iFcode - 1)) - 1; | ||
228 : | component >>= (iFcode - 1); | ||
229 : | |||
230 : | if (component > 32) | ||
231 : | component = 32; | ||
232 : | |||
233 : | return mvtab[component] + 1 + iFcode - 1; | ||
234 : | } | ||
235 : | |||
236 : | |||
237 : | chl | 1.10 | static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
238 : | Isibaar | 1.1 | { |
239 : | chl | 1.10 | return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
240 : | Isibaar | 1.1 | } |
241 : | |||
242 : | chl | 1.10 | static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
243 : | Isibaar | 1.1 | |
244 : | { | ||
245 : | chl | 1.10 | return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
246 : | Isibaar | 1.1 | } |
247 : | |||
248 : | |||
249 : | |||
250 : | |||
251 : | |||
252 : | #ifndef SEARCH16 | ||
253 : | #define SEARCH16 PMVfastSearch16 | ||
254 : | chl | 1.3 | //#define SEARCH16 FullSearch16 |
255 : | //#define SEARCH16 EPZSSearch16 | ||
256 : | Isibaar | 1.1 | #endif |
257 : | |||
258 : | #ifndef SEARCH8 | ||
259 : | #define SEARCH8 PMVfastSearch8 | ||
260 : | chl | 1.3 | //#define SEARCH8 EPZSSearch8 |
261 : | Isibaar | 1.1 | #endif |
262 : | |||
263 : | bool MotionEstimation( | ||
264 : | edgomez | 1.2 | MBParam * const pParam, |
265 : | suxen_drol | 1.8 | FRAMEINFO * const current, |
266 : | FRAMEINFO * const reference, | ||
267 : | edgomez | 1.2 | const IMAGE * const pRefH, |
268 : | const IMAGE * const pRefV, | ||
269 : | const IMAGE * const pRefHV, | ||
270 : | const uint32_t iLimit) | ||
271 : | Isibaar | 1.1 | |
272 : | { | ||
273 : | edgomez | 1.2 | const uint32_t iWcount = pParam->mb_width; |
274 : | const uint32_t iHcount = pParam->mb_height; | ||
275 : | chl | 1.15 | MACROBLOCK * const pMBs = current->mbs; |
276 : | MACROBLOCK * const prevMBs = reference->mbs; // previous frame | ||
277 : | suxen_drol | 1.8 | |
278 : | chl | 1.15 | const IMAGE * const pCurrent = ¤t->image; |
279 : | const IMAGE * const pRef = &reference->image; | ||
280 : | suxen_drol | 1.8 | |
281 : | chl | 1.15 | const VECTOR zeroMV = {0,0}; |
282 : | Isibaar | 1.1 | |
283 : | chl | 1.15 | int32_t x, y; |
284 : | int32_t iIntra = 0; | ||
285 : | VECTOR pmv; | ||
286 : | |||
287 : | canard | 1.5 | if (sadInit) |
288 : | canard | 1.4 | (*sadInit)(); |
289 : | suxen_drol | 1.8 | |
290 : | chl | 1.15 | for (y = 0; y < iHcount; y++) |
291 : | for (x = 0; x < iWcount; x++) | ||
292 : | Isibaar | 1.1 | { |
293 : | chl | 1.15 | MACROBLOCK* const pMB = &pMBs[x + y * iWcount]; |
294 : | Isibaar | 1.1 | |
295 : | chl | 1.15 | pMB->sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
296 : | x, y, current->motion_flags, current->quant, current->fcode, | ||
297 : | pParam, pMBs, prevMBs, &pMB->mv16, &pMB->pmvs[0]); | ||
298 : | } | ||
299 : | Isibaar | 1.1 | |
300 : | chl | 1.15 | for (y = 0; y < iHcount; y++) |
301 : | for (x = 0; x < iWcount; x++) | ||
302 : | { | ||
303 : | MACROBLOCK* const pMB = &pMBs[x + y * iWcount]; | ||
304 : | Isibaar | 1.1 | |
305 : | chl | 1.15 | if (0 < (pMB->sad16 - MV16_INTER_BIAS)) |
306 : | { | ||
307 : | int32_t deviation; | ||
308 : | deviation = dev16(pCurrent->y + x*16 + y*16*pParam->edged_width, | ||
309 : | pParam->edged_width); | ||
310 : | Isibaar | 1.1 | |
311 : | chl | 1.15 | if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) |
312 : | { | ||
313 : | pMB->mode = MODE_INTRA; | ||
314 : | pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] | ||
315 : | = pMB->mvs[2] = pMB->mvs[3] = zeroMV; | ||
316 : | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] | ||
317 : | = pMB->sad8[2] = pMB->sad8[3] = 0; | ||
318 : | |||
319 : | iIntra++; | ||
320 : | if (iIntra >= iLimit) | ||
321 : | return 1; | ||
322 : | edgomez | 1.2 | |
323 : | chl | 1.15 | continue; |
324 : | } | ||
325 : | edgomez | 1.2 | } |
326 : | Isibaar | 1.1 | |
327 : | chl | 1.15 | if ( (current->global_flags & XVID_INTER4V) |
328 : | && (!(current->global_flags & XVID_LUMIMASKING) | ||
329 : | || pMB->dquant == NO_CHANGE) ) | ||
330 : | edgomez | 1.2 | { |
331 : | chl | 1.15 | int32_t sad8 = 129; //IMV16X16 * current->quant; |
332 : | |||
333 : | if (sad8 < pMB->sad16) | ||
334 : | sad8 += pMB->sad8[0] | ||
335 : | = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
336 : | 2*x, 2*y, pMB->mv16.x, pMB->mv16.y, | ||
337 : | suxen_drol | 1.8 | current->motion_flags, current->quant, current->fcode, |
338 : | pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); | ||
339 : | edgomez | 1.2 | |
340 : | chl | 1.15 | if (sad8 < pMB->sad16) |
341 : | sad8 += pMB->sad8[1] | ||
342 : | = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
343 : | 2*x+1, 2*y, pMB->mv16.x, pMB->mv16.y, | ||
344 : | current->motion_flags, current->quant, current->fcode, | ||
345 : | pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); | ||
346 : | |||
347 : | if (sad8 < pMB->sad16) | ||
348 : | sad8 += pMB->sad8[2] | ||
349 : | = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
350 : | 2*x, 2*y+1, pMB->mv16.x, pMB->mv16.y, | ||
351 : | current->motion_flags, current->quant, current->fcode, | ||
352 : | pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); | ||
353 : | Isibaar | 1.1 | |
354 : | chl | 1.15 | if (sad8 < pMB->sad16) |
355 : | sad8 += pMB->sad8[3] | ||
356 : | = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
357 : | 2*x+1, 2*y+1, pMB->mv16.x, pMB->mv16.y, | ||
358 : | current->motion_flags, current->quant, current->fcode, | ||
359 : | pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); | ||
360 : | Isibaar | 1.1 | |
361 : | edgomez | 1.2 | /* decide: MODE_INTER or MODE_INTER4V |
362 : | chl | 1.15 | mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
363 : | edgomez | 1.2 | */ |
364 : | |||
365 : | chl | 1.15 | if (sad8 < pMB->sad16) |
366 : | chl | 1.12 | { |
367 : | edgomez | 1.2 | pMB->mode = MODE_INTER4V; |
368 : | chl | 1.15 | pMB->sad8[0] *= 4; |
369 : | chl | 1.12 | pMB->sad8[1] *= 4; |
370 : | pMB->sad8[2] *= 4; | ||
371 : | pMB->sad8[3] *= 4; | ||
372 : | chl | 1.15 | continue; |
373 : | chl | 1.12 | } |
374 : | edgomez | 1.2 | } |
375 : | chl | 1.12 | |
376 : | chl | 1.15 | pMB->mode = MODE_INTER; |
377 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; | ||
378 : | pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = pMB->sad16; | ||
379 : | |||
380 : | if (current->global_flags & XVID_INTER4V) | ||
381 : | { pmv = get_pmv(pMBs, x, y, pParam->mb_width, 0); | ||
382 : | // get_pmv has to be called again. inter4v changes predictors | ||
383 : | |||
384 : | pMB->pmvs[0].x = pMB->mv16.x - pmv.x; | ||
385 : | pMB->pmvs[0].y = pMB->mv16.y - pmv.y; | ||
386 : | Isibaar | 1.1 | } |
387 : | } | ||
388 : | return 0; | ||
389 : | } | ||
390 : | |||
391 : | #define CHECK_MV16_ZERO {\ | ||
392 : | if ( (0 <= max_dx) && (0 >= min_dx) \ | ||
393 : | && (0 <= max_dy) && (0 >= min_dy) ) \ | ||
394 : | { \ | ||
395 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ | ||
396 : | chl | 1.10 | iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
397 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
398 : | { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ | ||
399 : | } | ||
400 : | |||
401 : | chl | 1.3 | #define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
402 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
403 : | chl | 1.10 | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
404 : | chl | 1.3 | if (iSAD < iMinSAD) \ |
405 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ | ||
406 : | } | ||
407 : | Isibaar | 1.1 | |
408 : | #define CHECK_MV16_CANDIDATE(X,Y) { \ | ||
409 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
410 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
411 : | { \ | ||
412 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
413 : | chl | 1.10 | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
414 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
415 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
416 : | } | ||
417 : | |||
418 : | #define CHECK_MV16_CANDIDATE_DIR(X,Y,D) { \ | ||
419 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
420 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
421 : | { \ | ||
422 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
423 : | chl | 1.10 | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
424 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
425 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
426 : | } | ||
427 : | |||
428 : | #define CHECK_MV16_CANDIDATE_FOUND(X,Y,D) { \ | ||
429 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
430 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
431 : | { \ | ||
432 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
433 : | chl | 1.10 | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
434 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
435 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
436 : | } | ||
437 : | |||
438 : | |||
439 : | #define CHECK_MV8_ZERO {\ | ||
440 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ | ||
441 : | chl | 1.10 | iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
442 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
443 : | { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ | ||
444 : | } | ||
445 : | |||
446 : | chl | 1.3 | #define NOCHECK_MV8_CANDIDATE(X,Y) \ |
447 : | { \ | ||
448 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
449 : | chl | 1.10 | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
450 : | chl | 1.3 | if (iSAD < iMinSAD) \ |
451 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ | ||
452 : | } | ||
453 : | Isibaar | 1.1 | |
454 : | #define CHECK_MV8_CANDIDATE(X,Y) { \ | ||
455 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
456 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
457 : | { \ | ||
458 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
459 : | chl | 1.10 | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
460 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
461 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
462 : | } | ||
463 : | |||
464 : | #define CHECK_MV8_CANDIDATE_DIR(X,Y,D) { \ | ||
465 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
466 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
467 : | { \ | ||
468 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
469 : | chl | 1.10 | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
470 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
471 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
472 : | } | ||
473 : | |||
474 : | #define CHECK_MV8_CANDIDATE_FOUND(X,Y,D) { \ | ||
475 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
476 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
477 : | { \ | ||
478 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
479 : | chl | 1.10 | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
480 : | Isibaar | 1.1 | if (iSAD < iMinSAD) \ |
481 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
482 : | } | ||
483 : | |||
484 : | /* too slow and not fully functional at the moment */ | ||
485 : | /* | ||
486 : | int32_t ZeroSearch16( | ||
487 : | const uint8_t * const pRef, | ||
488 : | const uint8_t * const pRefH, | ||
489 : | const uint8_t * const pRefV, | ||
490 : | const uint8_t * const pRefHV, | ||
491 : | const IMAGE * const pCur, | ||
492 : | const int x, const int y, | ||
493 : | const uint32_t MotionFlags, | ||
494 : | suxen_drol | 1.8 | const uint32_t iQuant, |
495 : | const uint32_t iFcode, | ||
496 : | Isibaar | 1.1 | MBParam * const pParam, |
497 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
498 : | const MACROBLOCK * const prevMBs, | ||
499 : | Isibaar | 1.1 | VECTOR * const currMV, |
500 : | VECTOR * const currPMV) | ||
501 : | { | ||
502 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
503 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
504 : | int32_t iSAD; | ||
505 : | int32_t pred_x,pred_y; | ||
506 : | |||
507 : | get_pmv(pMBs, x, y, pParam->mb_width, 0, &pred_x, &pred_y); | ||
508 : | |||
509 : | iSAD = sad16( cur, | ||
510 : | get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), | ||
511 : | iEdgedWidth, MV_MAX_ERROR); | ||
512 : | if (iSAD <= iQuant * 96) | ||
513 : | iSAD -= MV16_00_BIAS; | ||
514 : | |||
515 : | currMV->x = 0; | ||
516 : | currMV->y = 0; | ||
517 : | currPMV->x = -pred_x; | ||
518 : | currPMV->y = -pred_y; | ||
519 : | |||
520 : | return iSAD; | ||
521 : | |||
522 : | } | ||
523 : | */ | ||
524 : | |||
525 : | chl | 1.3 | int32_t Diamond16_MainSearch( |
526 : | edgomez | 1.2 | const uint8_t * const pRef, |
527 : | const uint8_t * const pRefH, | ||
528 : | const uint8_t * const pRefV, | ||
529 : | const uint8_t * const pRefHV, | ||
530 : | const uint8_t * const cur, | ||
531 : | const int x, const int y, | ||
532 : | int32_t startx, int32_t starty, | ||
533 : | int32_t iMinSAD, | ||
534 : | VECTOR * const currMV, | ||
535 : | const VECTOR * const pmv, | ||
536 : | const int32_t min_dx, const int32_t max_dx, | ||
537 : | const int32_t min_dy, const int32_t max_dy, | ||
538 : | const int32_t iEdgedWidth, | ||
539 : | const int32_t iDiamondSize, | ||
540 : | const int32_t iFcode, | ||
541 : | const int32_t iQuant, | ||
542 : | int iFound) | ||
543 : | Isibaar | 1.1 | { |
544 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
545 : | |||
546 : | int32_t iDirection=0; | ||
547 : | int32_t iSAD; | ||
548 : | VECTOR backupMV; | ||
549 : | backupMV.x = startx; | ||
550 : | backupMV.y = starty; | ||
551 : | |||
552 : | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ | ||
553 : | |||
554 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
555 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
556 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
557 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
558 : | |||
559 : | if (iDirection) | ||
560 : | while (!iFound) | ||
561 : | { | ||
562 : | iFound = 1; | ||
563 : | backupMV=*currMV; | ||
564 : | |||
565 : | if ( iDirection != 2) | ||
566 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
567 : | if ( iDirection != 1) | ||
568 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); | ||
569 : | if ( iDirection != 4) | ||
570 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); | ||
571 : | if ( iDirection != 3) | ||
572 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); | ||
573 : | } | ||
574 : | else | ||
575 : | edgomez | 1.2 | { |
576 : | currMV->x = startx; | ||
577 : | currMV->y = starty; | ||
578 : | } | ||
579 : | Isibaar | 1.1 | return iMinSAD; |
580 : | } | ||
581 : | |||
582 : | chl | 1.3 | int32_t Square16_MainSearch( |
583 : | const uint8_t * const pRef, | ||
584 : | const uint8_t * const pRefH, | ||
585 : | const uint8_t * const pRefV, | ||
586 : | const uint8_t * const pRefHV, | ||
587 : | const uint8_t * const cur, | ||
588 : | const int x, const int y, | ||
589 : | int32_t startx, int32_t starty, | ||
590 : | int32_t iMinSAD, | ||
591 : | VECTOR * const currMV, | ||
592 : | const VECTOR * const pmv, | ||
593 : | const int32_t min_dx, const int32_t max_dx, | ||
594 : | const int32_t min_dy, const int32_t max_dy, | ||
595 : | const int32_t iEdgedWidth, | ||
596 : | const int32_t iDiamondSize, | ||
597 : | const int32_t iFcode, | ||
598 : | const int32_t iQuant, | ||
599 : | int iFound) | ||
600 : | { | ||
601 : | /* Do a square search around given starting point, return SAD of best */ | ||
602 : | |||
603 : | int32_t iDirection=0; | ||
604 : | int32_t iSAD; | ||
605 : | VECTOR backupMV; | ||
606 : | backupMV.x = startx; | ||
607 : | backupMV.y = starty; | ||
608 : | |||
609 : | /* It's one search with full square pattern, and new parts for all following diamonds */ | ||
610 : | |||
611 : | /* new direction are extra, so 1-4 is normal diamond | ||
612 : | 537 | ||
613 : | 1*2 | ||
614 : | 648 | ||
615 : | */ | ||
616 : | |||
617 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
618 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
619 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
620 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
621 : | |||
622 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
623 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
624 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
625 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
626 : | |||
627 : | |||
628 : | if (iDirection) | ||
629 : | while (!iFound) | ||
630 : | { | ||
631 : | iFound = 1; | ||
632 : | backupMV=*currMV; | ||
633 : | |||
634 : | switch (iDirection) | ||
635 : | { | ||
636 : | case 1: | ||
637 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
638 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
639 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
640 : | break; | ||
641 : | case 2: | ||
642 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
643 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
644 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
645 : | break; | ||
646 : | |||
647 : | case 3: | ||
648 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
649 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
650 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
651 : | break; | ||
652 : | |||
653 : | case 4: | ||
654 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
655 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
656 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
657 : | break; | ||
658 : | |||
659 : | case 5: | ||
660 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
661 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
662 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
663 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
664 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
665 : | break; | ||
666 : | |||
667 : | case 6: | ||
668 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
669 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
670 : | |||
671 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
672 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
673 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
674 : | |||
675 : | break; | ||
676 : | |||
677 : | case 7: | ||
678 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
679 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
680 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
681 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
682 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
683 : | break; | ||
684 : | |||
685 : | case 8: | ||
686 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
687 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
688 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
689 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
690 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
691 : | break; | ||
692 : | default: | ||
693 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
694 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
695 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
696 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
697 : | |||
698 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
699 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
700 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
701 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
702 : | break; | ||
703 : | } | ||
704 : | } | ||
705 : | else | ||
706 : | { | ||
707 : | currMV->x = startx; | ||
708 : | currMV->y = starty; | ||
709 : | } | ||
710 : | return iMinSAD; | ||
711 : | } | ||
712 : | |||
713 : | |||
714 : | int32_t Full16_MainSearch( | ||
715 : | const uint8_t * const pRef, | ||
716 : | const uint8_t * const pRefH, | ||
717 : | const uint8_t * const pRefV, | ||
718 : | const uint8_t * const pRefHV, | ||
719 : | const uint8_t * const cur, | ||
720 : | const int x, const int y, | ||
721 : | int32_t startx, int32_t starty, | ||
722 : | int32_t iMinSAD, | ||
723 : | VECTOR * const currMV, | ||
724 : | const VECTOR * const pmv, | ||
725 : | const int32_t min_dx, const int32_t max_dx, | ||
726 : | const int32_t min_dy, const int32_t max_dy, | ||
727 : | const int32_t iEdgedWidth, | ||
728 : | const int32_t iDiamondSize, | ||
729 : | const int32_t iFcode, | ||
730 : | const int32_t iQuant, | ||
731 : | int iFound) | ||
732 : | { | ||
733 : | int32_t iSAD; | ||
734 : | int32_t dx,dy; | ||
735 : | VECTOR backupMV; | ||
736 : | backupMV.x = startx; | ||
737 : | backupMV.y = starty; | ||
738 : | |||
739 : | for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) | ||
740 : | for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) | ||
741 : | NOCHECK_MV16_CANDIDATE(dx,dy); | ||
742 : | |||
743 : | return iMinSAD; | ||
744 : | } | ||
745 : | |||
746 : | int32_t Full8_MainSearch( | ||
747 : | const uint8_t * const pRef, | ||
748 : | const uint8_t * const pRefH, | ||
749 : | const uint8_t * const pRefV, | ||
750 : | const uint8_t * const pRefHV, | ||
751 : | const uint8_t * const cur, | ||
752 : | const int x, const int y, | ||
753 : | int32_t startx, int32_t starty, | ||
754 : | int32_t iMinSAD, | ||
755 : | VECTOR * const currMV, | ||
756 : | const VECTOR * const pmv, | ||
757 : | const int32_t min_dx, const int32_t max_dx, | ||
758 : | const int32_t min_dy, const int32_t max_dy, | ||
759 : | const int32_t iEdgedWidth, | ||
760 : | const int32_t iDiamondSize, | ||
761 : | const int32_t iFcode, | ||
762 : | const int32_t iQuant, | ||
763 : | int iFound) | ||
764 : | { | ||
765 : | int32_t iSAD; | ||
766 : | int32_t dx,dy; | ||
767 : | VECTOR backupMV; | ||
768 : | backupMV.x = startx; | ||
769 : | backupMV.y = starty; | ||
770 : | |||
771 : | for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) | ||
772 : | for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) | ||
773 : | NOCHECK_MV8_CANDIDATE(dx,dy); | ||
774 : | |||
775 : | return iMinSAD; | ||
776 : | } | ||
777 : | |||
778 : | |||
779 : | |||
780 : | int32_t Halfpel16_Refine( | ||
781 : | edgomez | 1.2 | const uint8_t * const pRef, |
782 : | const uint8_t * const pRefH, | ||
783 : | const uint8_t * const pRefV, | ||
784 : | const uint8_t * const pRefHV, | ||
785 : | const uint8_t * const cur, | ||
786 : | const int x, const int y, | ||
787 : | VECTOR * const currMV, | ||
788 : | int32_t iMinSAD, | ||
789 : | const VECTOR * const pmv, | ||
790 : | const int32_t min_dx, const int32_t max_dx, | ||
791 : | const int32_t min_dy, const int32_t max_dy, | ||
792 : | const int32_t iFcode, | ||
793 : | const int32_t iQuant, | ||
794 : | const int32_t iEdgedWidth) | ||
795 : | Isibaar | 1.1 | { |
796 : | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ | ||
797 : | |||
798 : | int32_t iSAD; | ||
799 : | VECTOR backupMV = *currMV; | ||
800 : | |||
801 : | CHECK_MV16_CANDIDATE(backupMV.x-1,backupMV.y-1); | ||
802 : | CHECK_MV16_CANDIDATE(backupMV.x ,backupMV.y-1); | ||
803 : | CHECK_MV16_CANDIDATE(backupMV.x+1,backupMV.y-1); | ||
804 : | CHECK_MV16_CANDIDATE(backupMV.x-1,backupMV.y); | ||
805 : | CHECK_MV16_CANDIDATE(backupMV.x+1,backupMV.y); | ||
806 : | CHECK_MV16_CANDIDATE(backupMV.x-1,backupMV.y+1); | ||
807 : | CHECK_MV16_CANDIDATE(backupMV.x ,backupMV.y+1); | ||
808 : | CHECK_MV16_CANDIDATE(backupMV.x+1,backupMV.y+1); | ||
809 : | |||
810 : | return iMinSAD; | ||
811 : | } | ||
812 : | |||
813 : | #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) | ||
814 : | |||
815 : | chl | 1.3 | |
816 : | Isibaar | 1.1 | int32_t PMVfastSearch16( |
817 : | chl | 1.3 | const uint8_t * const pRef, |
818 : | const uint8_t * const pRefH, | ||
819 : | const uint8_t * const pRefV, | ||
820 : | const uint8_t * const pRefHV, | ||
821 : | const IMAGE * const pCur, | ||
822 : | const int x, const int y, | ||
823 : | const uint32_t MotionFlags, | ||
824 : | suxen_drol | 1.8 | const uint32_t iQuant, |
825 : | const uint32_t iFcode, | ||
826 : | chl | 1.3 | const MBParam * const pParam, |
827 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
828 : | const MACROBLOCK * const prevMBs, | ||
829 : | chl | 1.3 | VECTOR * const currMV, |
830 : | VECTOR * const currPMV) | ||
831 : | Isibaar | 1.1 | { |
832 : | chenm001 | 1.7 | const uint32_t iWcount = pParam->mb_width; |
833 : | Isibaar | 1.1 | const int32_t iWidth = pParam->width; |
834 : | const int32_t iHeight = pParam->height; | ||
835 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
836 : | |||
837 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
838 : | |||
839 : | int32_t iDiamondSize; | ||
840 : | |||
841 : | int32_t min_dx; | ||
842 : | int32_t max_dx; | ||
843 : | int32_t min_dy; | ||
844 : | int32_t max_dy; | ||
845 : | |||
846 : | int32_t iFound; | ||
847 : | |||
848 : | VECTOR newMV; | ||
849 : | VECTOR backupMV; /* just for PMVFAST */ | ||
850 : | |||
851 : | VECTOR pmv[4]; | ||
852 : | int32_t psad[4]; | ||
853 : | |||
854 : | chl | 1.15 | // const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
855 : | suxen_drol | 1.8 | const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
856 : | Isibaar | 1.1 | |
857 : | static int32_t threshA,threshB; | ||
858 : | int32_t bPredEq; | ||
859 : | int32_t iMinSAD,iSAD; | ||
860 : | |||
861 : | /* Get maximum range */ | ||
862 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, | ||
863 : | edgomez | 1.2 | x, y, 16, iWidth, iHeight, iFcode); |
864 : | Isibaar | 1.1 | |
865 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
866 : | |||
867 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
868 : | { min_dx = EVEN(min_dx); | ||
869 : | edgomez | 1.2 | max_dx = EVEN(max_dx); |
870 : | min_dy = EVEN(min_dy); | ||
871 : | max_dy = EVEN(max_dy); | ||
872 : | Isibaar | 1.1 | } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
873 : | |||
874 : | |||
875 : | bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); | ||
876 : | |||
877 : | if ((x==0) && (y==0) ) | ||
878 : | { | ||
879 : | threshA = 512; | ||
880 : | threshB = 1024; | ||
881 : | |||
882 : | } | ||
883 : | else | ||
884 : | { | ||
885 : | threshA = psad[0]; | ||
886 : | threshB = threshA+256; | ||
887 : | if (threshA< 512) threshA = 512; | ||
888 : | if (threshA>1024) threshA = 1024; | ||
889 : | if (threshB>1792) threshB = 1792; | ||
890 : | } | ||
891 : | |||
892 : | iFound=0; | ||
893 : | |||
894 : | /* Step 4: Calculate SAD around the Median prediction. | ||
895 : | edgomez | 1.2 | MinSAD=SAD |
896 : | If Motion Vector equal to Previous frame motion vector | ||
897 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
898 : | If SAD<=256 goto Step 10. | ||
899 : | Isibaar | 1.1 | */ |
900 : | |||
901 : | *currMV=pmv[0]; /* current best := prediction */ | ||
902 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
903 : | { /* This should NOT be necessary! */ | ||
904 : | currMV->x = EVEN(currMV->x); | ||
905 : | currMV->y = EVEN(currMV->y); | ||
906 : | } | ||
907 : | |||
908 : | if (currMV->x > max_dx) | ||
909 : | edgomez | 1.2 | { |
910 : | currMV->x=max_dx; | ||
911 : | } | ||
912 : | Isibaar | 1.1 | if (currMV->x < min_dx) |
913 : | edgomez | 1.2 | { |
914 : | currMV->x=min_dx; | ||
915 : | } | ||
916 : | Isibaar | 1.1 | if (currMV->y > max_dy) |
917 : | edgomez | 1.2 | { |
918 : | currMV->y=max_dy; | ||
919 : | } | ||
920 : | Isibaar | 1.1 | if (currMV->y < min_dy) |
921 : | edgomez | 1.2 | { |
922 : | currMV->y=min_dy; | ||
923 : | } | ||
924 : | Isibaar | 1.1 | |
925 : | iMinSAD = sad16( cur, | ||
926 : | edgomez | 1.2 | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
927 : | iEdgedWidth, MV_MAX_ERROR); | ||
928 : | chl | 1.10 | iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
929 : | Isibaar | 1.1 | |
930 : | suxen_drol | 1.8 | if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
931 : | edgomez | 1.2 | { |
932 : | chl | 1.13 | if (iMinSAD < 2*iQuant) // high chances for SKIP-mode |
933 : | { | ||
934 : | if (!MVzero(*currMV)) | ||
935 : | { | ||
936 : | iMinSAD += MV16_00_BIAS; | ||
937 : | CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures | ||
938 : | iMinSAD -= MV16_00_BIAS; | ||
939 : | } | ||
940 : | } | ||
941 : | |||
942 : | edgomez | 1.2 | if (MotionFlags & PMV_QUICKSTOP16) |
943 : | chl | 1.3 | goto PMVfast16_Terminate_without_Refine; |
944 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
945 : | goto PMVfast16_Terminate_with_Refine; | ||
946 : | edgomez | 1.2 | } |
947 : | Isibaar | 1.1 | |
948 : | chl | 1.13 | |
949 : | /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion | ||
950 : | vector of the median. | ||
951 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
952 : | */ | ||
953 : | |||
954 : | if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[0]) ) ) | ||
955 : | iFound=2; | ||
956 : | |||
957 : | /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
958 : | Otherwise select large Diamond Search. | ||
959 : | */ | ||
960 : | |||
961 : | if ( (!MVzero(pmv[0])) || (threshB<1536) || (bPredEq) ) | ||
962 : | iDiamondSize=1; // halfpel! | ||
963 : | else | ||
964 : | iDiamondSize=2; // halfpel! | ||
965 : | |||
966 : | if (!(MotionFlags & PMV_HALFPELDIAMOND16) ) | ||
967 : | iDiamondSize*=2; | ||
968 : | |||
969 : | Isibaar | 1.1 | /* |
970 : | edgomez | 1.2 | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
971 : | Also calculate (0,0) but do not subtract offset. | ||
972 : | Let MinSAD be the smallest SAD up to this point. | ||
973 : | chl | 1.9 | If MV is (0,0) subtract offset. |
974 : | Isibaar | 1.1 | */ |
975 : | |||
976 : | // (0,0) is always possible | ||
977 : | |||
978 : | chl | 1.13 | if (!MVzero(pmv[0])) |
979 : | CHECK_MV16_ZERO; | ||
980 : | Isibaar | 1.1 | |
981 : | // previous frame MV is always possible | ||
982 : | chl | 1.13 | |
983 : | if (!MVzero(prevMB->mvs[0])) | ||
984 : | if (!MVequal(prevMB->mvs[0],pmv[0])) | ||
985 : | CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); | ||
986 : | Isibaar | 1.1 | |
987 : | // left neighbour, if allowed | ||
988 : | chl | 1.13 | |
989 : | if (!MVzero(pmv[1])) | ||
990 : | if (!MVequal(pmv[1],prevMB->mvs[0])) | ||
991 : | if (!MVequal(pmv[1],pmv[0])) | ||
992 : | Isibaar | 1.1 | { |
993 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
994 : | { pmv[1].x = EVEN(pmv[1].x); | ||
995 : | chl | 1.13 | pmv[1].y = EVEN(pmv[1].y); |
996 : | Isibaar | 1.1 | } |
997 : | chl | 1.13 | |
998 : | Isibaar | 1.1 | CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
999 : | } | ||
1000 : | |||
1001 : | // top neighbour, if allowed | ||
1002 : | chl | 1.13 | if (!MVzero(pmv[2])) |
1003 : | if (!MVequal(pmv[2],prevMB->mvs[0])) | ||
1004 : | if (!MVequal(pmv[2],pmv[0])) | ||
1005 : | if (!MVequal(pmv[2],pmv[1])) | ||
1006 : | Isibaar | 1.1 | { |
1007 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1008 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1009 : | chl | 1.13 | pmv[2].y = EVEN(pmv[2].y); |
1010 : | Isibaar | 1.1 | } |
1011 : | CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1012 : | |||
1013 : | // top right neighbour, if allowed | ||
1014 : | chl | 1.13 | if (!MVzero(pmv[3])) |
1015 : | if (!MVequal(pmv[3],prevMB->mvs[0])) | ||
1016 : | if (!MVequal(pmv[3],pmv[0])) | ||
1017 : | if (!MVequal(pmv[3],pmv[1])) | ||
1018 : | if (!MVequal(pmv[3],pmv[2])) | ||
1019 : | Isibaar | 1.1 | { |
1020 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1021 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1022 : | chl | 1.13 | pmv[3].y = EVEN(pmv[3].y); |
1023 : | Isibaar | 1.1 | } |
1024 : | CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); | ||
1025 : | } | ||
1026 : | } | ||
1027 : | chl | 1.9 | |
1028 : | chl | 1.10 | if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96)*/ ) |
1029 : | chl | 1.9 | iMinSAD -= MV16_00_BIAS; |
1030 : | Isibaar | 1.1 | |
1031 : | chl | 1.9 | |
1032 : | Isibaar | 1.1 | /* Step 6: If MinSAD <= thresa goto Step 10. |
1033 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
1034 : | */ | ||
1035 : | |||
1036 : | suxen_drol | 1.8 | if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
1037 : | edgomez | 1.2 | { |
1038 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1039 : | chl | 1.3 | goto PMVfast16_Terminate_without_Refine; |
1040 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1041 : | goto PMVfast16_Terminate_with_Refine; | ||
1042 : | edgomez | 1.2 | } |
1043 : | Isibaar | 1.1 | |
1044 : | |||
1045 : | /************ (Diamond Search) **************/ | ||
1046 : | /* | ||
1047 : | edgomez | 1.2 | Step 7: Perform Diamond search, with either the small or large diamond. |
1048 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
1049 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
1050 : | If center then goto step 10. | ||
1051 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
1052 : | Refine by using small diamond and goto step 10. | ||
1053 : | Isibaar | 1.1 | */ |
1054 : | |||
1055 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1056 : | |||
1057 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1058 : | chl | 1.3 | iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1059 : | edgomez | 1.2 | x, y, |
1060 : | currMV->x, currMV->y, iMinSAD, &newMV, | ||
1061 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1062 : | Isibaar | 1.1 | |
1063 : | if (iSAD < iMinSAD) | ||
1064 : | { | ||
1065 : | *currMV = newMV; | ||
1066 : | iMinSAD = iSAD; | ||
1067 : | } | ||
1068 : | |||
1069 : | if (MotionFlags & PMV_EXTSEARCH16) | ||
1070 : | { | ||
1071 : | /* extended: search (up to) two more times: orignal prediction and (0,0) */ | ||
1072 : | |||
1073 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1074 : | chl | 1.3 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1075 : | edgomez | 1.2 | x, y, |
1076 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1077 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1078 : | Isibaar | 1.1 | |
1079 : | edgomez | 1.2 | if (iSAD < iMinSAD) |
1080 : | { | ||
1081 : | *currMV = newMV; | ||
1082 : | iMinSAD = iSAD; | ||
1083 : | } | ||
1084 : | Isibaar | 1.1 | } |
1085 : | |||
1086 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1087 : | chl | 1.3 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1088 : | edgomez | 1.2 | x, y, |
1089 : | 0, 0, iMinSAD, &newMV, | ||
1090 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1091 : | Isibaar | 1.1 | |
1092 : | edgomez | 1.2 | if (iSAD < iMinSAD) |
1093 : | { | ||
1094 : | *currMV = newMV; | ||
1095 : | iMinSAD = iSAD; | ||
1096 : | } | ||
1097 : | Isibaar | 1.1 | } |
1098 : | } | ||
1099 : | |||
1100 : | /* | ||
1101 : | edgomez | 1.2 | Step 10: The motion vector is chosen according to the block corresponding to MinSAD. |
1102 : | Isibaar | 1.1 | */ |
1103 : | |||
1104 : | chl | 1.3 | PMVfast16_Terminate_with_Refine: |
1105 : | Isibaar | 1.1 | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
1106 : | chl | 1.3 | iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
1107 : | x, y, | ||
1108 : | currMV, iMinSAD, | ||
1109 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1110 : | Isibaar | 1.1 | |
1111 : | chl | 1.3 | PMVfast16_Terminate_without_Refine: |
1112 : | Isibaar | 1.1 | currPMV->x = currMV->x - pmv[0].x; |
1113 : | currPMV->y = currMV->y - pmv[0].y; | ||
1114 : | return iMinSAD; | ||
1115 : | } | ||
1116 : | |||
1117 : | |||
1118 : | |||
1119 : | |||
1120 : | |||
1121 : | |||
1122 : | chl | 1.3 | int32_t Diamond8_MainSearch( |
1123 : | edgomez | 1.2 | const uint8_t * const pRef, |
1124 : | const uint8_t * const pRefH, | ||
1125 : | const uint8_t * const pRefV, | ||
1126 : | const uint8_t * const pRefHV, | ||
1127 : | const uint8_t * const cur, | ||
1128 : | const int x, const int y, | ||
1129 : | int32_t startx, int32_t starty, | ||
1130 : | int32_t iMinSAD, | ||
1131 : | VECTOR * const currMV, | ||
1132 : | const VECTOR * const pmv, | ||
1133 : | const int32_t min_dx, const int32_t max_dx, | ||
1134 : | const int32_t min_dy, const int32_t max_dy, | ||
1135 : | const int32_t iEdgedWidth, | ||
1136 : | const int32_t iDiamondSize, | ||
1137 : | const int32_t iFcode, | ||
1138 : | const int32_t iQuant, | ||
1139 : | int iFound) | ||
1140 : | Isibaar | 1.1 | { |
1141 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
1142 : | |||
1143 : | int32_t iDirection=0; | ||
1144 : | int32_t iSAD; | ||
1145 : | VECTOR backupMV; | ||
1146 : | backupMV.x = startx; | ||
1147 : | backupMV.y = starty; | ||
1148 : | |||
1149 : | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ | ||
1150 : | |||
1151 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
1152 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
1153 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
1154 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
1155 : | |||
1156 : | if (iDirection) | ||
1157 : | while (!iFound) | ||
1158 : | { | ||
1159 : | iFound = 1; | ||
1160 : | backupMV=*currMV; // since iDirection!=0, this is well defined! | ||
1161 : | |||
1162 : | if ( iDirection != 2) | ||
1163 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
1164 : | if ( iDirection != 1) | ||
1165 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); | ||
1166 : | if ( iDirection != 4) | ||
1167 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); | ||
1168 : | if ( iDirection != 3) | ||
1169 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); | ||
1170 : | } | ||
1171 : | else | ||
1172 : | edgomez | 1.2 | { |
1173 : | currMV->x = startx; | ||
1174 : | currMV->y = starty; | ||
1175 : | } | ||
1176 : | Isibaar | 1.1 | return iMinSAD; |
1177 : | } | ||
1178 : | |||
1179 : | chl | 1.3 | int32_t Halfpel8_Refine( |
1180 : | edgomez | 1.2 | const uint8_t * const pRef, |
1181 : | const uint8_t * const pRefH, | ||
1182 : | const uint8_t * const pRefV, | ||
1183 : | const uint8_t * const pRefHV, | ||
1184 : | const uint8_t * const cur, | ||
1185 : | const int x, const int y, | ||
1186 : | VECTOR * const currMV, | ||
1187 : | int32_t iMinSAD, | ||
1188 : | const VECTOR * const pmv, | ||
1189 : | const int32_t min_dx, const int32_t max_dx, | ||
1190 : | const int32_t min_dy, const int32_t max_dy, | ||
1191 : | const int32_t iFcode, | ||
1192 : | const int32_t iQuant, | ||
1193 : | const int32_t iEdgedWidth) | ||
1194 : | Isibaar | 1.1 | { |
1195 : | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ | ||
1196 : | |||
1197 : | int32_t iSAD; | ||
1198 : | VECTOR backupMV = *currMV; | ||
1199 : | |||
1200 : | CHECK_MV8_CANDIDATE(backupMV.x-1,backupMV.y-1); | ||
1201 : | CHECK_MV8_CANDIDATE(backupMV.x ,backupMV.y-1); | ||
1202 : | CHECK_MV8_CANDIDATE(backupMV.x+1,backupMV.y-1); | ||
1203 : | CHECK_MV8_CANDIDATE(backupMV.x-1,backupMV.y); | ||
1204 : | CHECK_MV8_CANDIDATE(backupMV.x+1,backupMV.y); | ||
1205 : | CHECK_MV8_CANDIDATE(backupMV.x-1,backupMV.y+1); | ||
1206 : | CHECK_MV8_CANDIDATE(backupMV.x ,backupMV.y+1); | ||
1207 : | CHECK_MV8_CANDIDATE(backupMV.x+1,backupMV.y+1); | ||
1208 : | |||
1209 : | return iMinSAD; | ||
1210 : | } | ||
1211 : | |||
1212 : | |||
1213 : | #define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) | ||
1214 : | |||
1215 : | int32_t PMVfastSearch8( | ||
1216 : | chl | 1.3 | const uint8_t * const pRef, |
1217 : | const uint8_t * const pRefH, | ||
1218 : | const uint8_t * const pRefV, | ||
1219 : | const uint8_t * const pRefHV, | ||
1220 : | const IMAGE * const pCur, | ||
1221 : | const int x, const int y, | ||
1222 : | suxen_drol | 1.8 | const int start_x, const int start_y, |
1223 : | chl | 1.3 | const uint32_t MotionFlags, |
1224 : | suxen_drol | 1.8 | const uint32_t iQuant, |
1225 : | const uint32_t iFcode, | ||
1226 : | chl | 1.3 | const MBParam * const pParam, |
1227 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
1228 : | const MACROBLOCK * const prevMBs, | ||
1229 : | chl | 1.3 | VECTOR * const currMV, |
1230 : | VECTOR * const currPMV) | ||
1231 : | Isibaar | 1.1 | { |
1232 : | suxen_drol | 1.8 | const uint32_t iWcount = pParam->mb_width; |
1233 : | Isibaar | 1.1 | const int32_t iWidth = pParam->width; |
1234 : | const int32_t iHeight = pParam->height; | ||
1235 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
1236 : | |||
1237 : | const uint8_t * cur = pCur->y + x*8 + y*8*iEdgedWidth; | ||
1238 : | |||
1239 : | int32_t iDiamondSize; | ||
1240 : | |||
1241 : | int32_t min_dx; | ||
1242 : | int32_t max_dx; | ||
1243 : | int32_t min_dy; | ||
1244 : | int32_t max_dy; | ||
1245 : | |||
1246 : | VECTOR pmv[4]; | ||
1247 : | int32_t psad[4]; | ||
1248 : | VECTOR newMV; | ||
1249 : | VECTOR backupMV; | ||
1250 : | edgomez | 1.14 | VECTOR startMV; |
1251 : | Isibaar | 1.1 | |
1252 : | chl | 1.15 | // const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1253 : | suxen_drol | 1.8 | const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1254 : | Isibaar | 1.1 | |
1255 : | static int32_t threshA,threshB; | ||
1256 : | int32_t iFound,bPredEq; | ||
1257 : | int32_t iMinSAD,iSAD; | ||
1258 : | |||
1259 : | chl | 1.13 | int32_t iSubBlock = (y&1)+(y&1) + (x&1); |
1260 : | Isibaar | 1.1 | |
1261 : | edgomez | 1.14 | /* Init variables */ |
1262 : | startMV.x = start_x; | ||
1263 : | startMV.y = start_y; | ||
1264 : | |||
1265 : | /* Get maximum range */ | ||
1266 : | edgomez | 1.2 | get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1267 : | x, y, 8, iWidth, iHeight, iFcode); | ||
1268 : | Isibaar | 1.1 | |
1269 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) | ||
1270 : | { min_dx = EVEN(min_dx); | ||
1271 : | chl | 1.13 | max_dx = EVEN(max_dx); |
1272 : | min_dy = EVEN(min_dy); | ||
1273 : | max_dy = EVEN(max_dy); | ||
1274 : | Isibaar | 1.1 | } /* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
1275 : | |||
1276 : | |||
1277 : | bPredEq = get_pmvdata(pMBs, (x>>1), (y>>1), iWcount, iSubBlock, pmv, psad); | ||
1278 : | |||
1279 : | if ((x==0) && (y==0) ) | ||
1280 : | { | ||
1281 : | threshA = 512/4; | ||
1282 : | threshB = 1024/4; | ||
1283 : | |||
1284 : | } | ||
1285 : | else | ||
1286 : | { | ||
1287 : | threshA = psad[0]/4; /* good estimate */ | ||
1288 : | threshB = threshA+256/4; | ||
1289 : | if (threshA< 512/4) threshA = 512/4; | ||
1290 : | if (threshA>1024/4) threshA = 1024/4; | ||
1291 : | if (threshB>1792/4) threshB = 1792/4; | ||
1292 : | } | ||
1293 : | |||
1294 : | iFound=0; | ||
1295 : | |||
1296 : | /* Step 4: Calculate SAD around the Median prediction. | ||
1297 : | edgomez | 1.2 | MinSAD=SAD |
1298 : | If Motion Vector equal to Previous frame motion vector | ||
1299 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1300 : | If SAD<=256 goto Step 10. | ||
1301 : | Isibaar | 1.1 | */ |
1302 : | |||
1303 : | |||
1304 : | // Prepare for main loop | ||
1305 : | |||
1306 : | chl | 1.13 | *currMV = startMV; |
1307 : | Isibaar | 1.1 | |
1308 : | iMinSAD = sad8( cur, | ||
1309 : | edgomez | 1.2 | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
1310 : | iEdgedWidth); | ||
1311 : | chl | 1.10 | iMinSAD += calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, (uint8_t)iFcode, iQuant); |
1312 : | Isibaar | 1.1 | |
1313 : | chl | 1.13 | if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,prevMB->mvs[iSubBlock])) |
1314 : | && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) | ||
1315 : | edgomez | 1.2 | { |
1316 : | chl | 1.3 | if (MotionFlags & PMV_QUICKSTOP16) |
1317 : | goto PMVfast8_Terminate_without_Refine; | ||
1318 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1319 : | goto PMVfast8_Terminate_with_Refine; | ||
1320 : | edgomez | 1.2 | } |
1321 : | Isibaar | 1.1 | |
1322 : | chl | 1.13 | /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1323 : | vector of the median. | ||
1324 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
1325 : | */ | ||
1326 : | |||
1327 : | if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[iSubBlock]) ) ) | ||
1328 : | iFound=2; | ||
1329 : | |||
1330 : | /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
1331 : | Otherwise select large Diamond Search. | ||
1332 : | */ | ||
1333 : | |||
1334 : | if ( (!MVzero(pmv[0])) || (threshB<1536/4) || (bPredEq) ) | ||
1335 : | iDiamondSize=1; // 1 halfpel! | ||
1336 : | else | ||
1337 : | iDiamondSize=2; // 2 halfpel = 1 full pixel! | ||
1338 : | |||
1339 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8) ) | ||
1340 : | iDiamondSize*=2; | ||
1341 : | |||
1342 : | chl | 1.3 | |
1343 : | Isibaar | 1.1 | /* |
1344 : | edgomez | 1.2 | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1345 : | Also calculate (0,0) but do not subtract offset. | ||
1346 : | Let MinSAD be the smallest SAD up to this point. | ||
1347 : | chl | 1.9 | If MV is (0,0) subtract offset. |
1348 : | Isibaar | 1.1 | */ |
1349 : | |||
1350 : | chl | 1.13 | // the median prediction might be even better than mv16 |
1351 : | |||
1352 : | if (!MVequal(pmv[0],startMV)) | ||
1353 : | CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); | ||
1354 : | Isibaar | 1.1 | |
1355 : | chl | 1.13 | // (0,0) if needed |
1356 : | if (!MVzero(pmv[0])) | ||
1357 : | if (!MVzero(startMV)) | ||
1358 : | Isibaar | 1.1 | CHECK_MV8_ZERO; |
1359 : | |||
1360 : | chl | 1.13 | // previous frame MV if needed |
1361 : | if (!MVzero(prevMB->mvs[iSubBlock])) | ||
1362 : | if (!MVequal(prevMB->mvs[iSubBlock],startMV)) | ||
1363 : | if (!MVequal(prevMB->mvs[iSubBlock],pmv[0])) | ||
1364 : | chl | 1.9 | CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
1365 : | chl | 1.13 | |
1366 : | if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) | ||
1367 : | { | ||
1368 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1369 : | goto PMVfast8_Terminate_without_Refine; | ||
1370 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1371 : | goto PMVfast8_Terminate_with_Refine; | ||
1372 : | } | ||
1373 : | |||
1374 : | |||
1375 : | // left neighbour, if allowed and needed | ||
1376 : | if (!MVzero(pmv[1])) | ||
1377 : | if (!MVequal(pmv[1],startMV)) | ||
1378 : | if (!MVequal(pmv[1],prevMB->mvs[iSubBlock])) | ||
1379 : | if (!MVequal(pmv[1],pmv[0])) | ||
1380 : | Isibaar | 1.1 | { |
1381 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1382 : | { pmv[1].x = EVEN(pmv[1].x); | ||
1383 : | chl | 1.13 | pmv[1].y = EVEN(pmv[1].y); |
1384 : | Isibaar | 1.1 | } |
1385 : | CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); | ||
1386 : | } | ||
1387 : | |||
1388 : | chl | 1.13 | // top neighbour, if allowed and needed |
1389 : | if (!MVzero(pmv[2])) | ||
1390 : | if (!MVequal(pmv[2],startMV)) | ||
1391 : | if (!MVequal(pmv[2],prevMB->mvs[iSubBlock])) | ||
1392 : | if (!MVequal(pmv[2],pmv[0])) | ||
1393 : | if (!MVequal(pmv[2],pmv[1])) | ||
1394 : | Isibaar | 1.1 | { |
1395 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1396 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1397 : | chl | 1.13 | pmv[2].y = EVEN(pmv[2].y); |
1398 : | Isibaar | 1.1 | } |
1399 : | CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1400 : | |||
1401 : | chl | 1.13 | // top right neighbour, if allowed and needed |
1402 : | if (!MVzero(pmv[3])) | ||
1403 : | if (!MVequal(pmv[3],startMV)) | ||
1404 : | if (!MVequal(pmv[3],prevMB->mvs[iSubBlock])) | ||
1405 : | if (!MVequal(pmv[3],pmv[0])) | ||
1406 : | if (!MVequal(pmv[3],pmv[1])) | ||
1407 : | if (!MVequal(pmv[3],pmv[2])) | ||
1408 : | Isibaar | 1.1 | { |
1409 : | edgomez | 1.2 | if (!(MotionFlags & PMV_HALFPEL8 )) |
1410 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1411 : | chl | 1.13 | pmv[3].y = EVEN(pmv[3].y); |
1412 : | edgomez | 1.2 | } |
1413 : | Isibaar | 1.1 | CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
1414 : | } | ||
1415 : | } | ||
1416 : | |||
1417 : | chl | 1.10 | if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
1418 : | chl | 1.9 | iMinSAD -= MV8_00_BIAS; |
1419 : | |||
1420 : | |||
1421 : | Isibaar | 1.1 | /* Step 6: If MinSAD <= thresa goto Step 10. |
1422 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
1423 : | */ | ||
1424 : | |||
1425 : | chl | 1.9 | if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1426 : | edgomez | 1.2 | { |
1427 : | chl | 1.3 | if (MotionFlags & PMV_QUICKSTOP16) |
1428 : | goto PMVfast8_Terminate_without_Refine; | ||
1429 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1430 : | goto PMVfast8_Terminate_with_Refine; | ||
1431 : | edgomez | 1.2 | } |
1432 : | Isibaar | 1.1 | |
1433 : | /************ (Diamond Search) **************/ | ||
1434 : | /* | ||
1435 : | edgomez | 1.2 | Step 7: Perform Diamond search, with either the small or large diamond. |
1436 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
1437 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
1438 : | If center then goto step 10. | ||
1439 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
1440 : | Refine by using small diamond and goto step 10. | ||
1441 : | Isibaar | 1.1 | */ |
1442 : | |||
1443 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1444 : | |||
1445 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1446 : | chl | 1.3 | iSAD = Diamond8_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1447 : | edgomez | 1.2 | x, y, |
1448 : | currMV->x, currMV->y, iMinSAD, &newMV, | ||
1449 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1450 : | Isibaar | 1.1 | |
1451 : | if (iSAD < iMinSAD) | ||
1452 : | { | ||
1453 : | *currMV = newMV; | ||
1454 : | iMinSAD = iSAD; | ||
1455 : | } | ||
1456 : | |||
1457 : | if (MotionFlags & PMV_EXTSEARCH8) | ||
1458 : | { | ||
1459 : | /* extended: search (up to) two more times: orignal prediction and (0,0) */ | ||
1460 : | |||
1461 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1462 : | chl | 1.3 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1463 : | edgomez | 1.2 | x, y, |
1464 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1465 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1466 : | Isibaar | 1.1 | |
1467 : | edgomez | 1.2 | if (iSAD < iMinSAD) |
1468 : | { | ||
1469 : | *currMV = newMV; | ||
1470 : | iMinSAD = iSAD; | ||
1471 : | } | ||
1472 : | Isibaar | 1.1 | } |
1473 : | |||
1474 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1475 : | chl | 1.3 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1476 : | edgomez | 1.2 | x, y, |
1477 : | 0, 0, iMinSAD, &newMV, | ||
1478 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1479 : | Isibaar | 1.1 | |
1480 : | edgomez | 1.2 | if (iSAD < iMinSAD) |
1481 : | { | ||
1482 : | *currMV = newMV; | ||
1483 : | iMinSAD = iSAD; | ||
1484 : | } | ||
1485 : | Isibaar | 1.1 | } |
1486 : | } | ||
1487 : | |||
1488 : | /* Step 10: The motion vector is chosen according to the block corresponding to MinSAD. | ||
1489 : | edgomez | 1.2 | By performing an optional local half-pixel search, we can refine this result even further. |
1490 : | Isibaar | 1.1 | */ |
1491 : | |||
1492 : | chl | 1.3 | PMVfast8_Terminate_with_Refine: |
1493 : | Isibaar | 1.1 | if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
1494 : | chl | 1.3 | iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
1495 : | edgomez | 1.2 | x, y, |
1496 : | currMV, iMinSAD, | ||
1497 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1498 : | Isibaar | 1.1 | |
1499 : | |||
1500 : | chl | 1.3 | PMVfast8_Terminate_without_Refine: |
1501 : | Isibaar | 1.1 | currPMV->x = currMV->x - pmv[0].x; |
1502 : | currPMV->y = currMV->y - pmv[0].y; | ||
1503 : | |||
1504 : | return iMinSAD; | ||
1505 : | } | ||
1506 : | chl | 1.3 | |
1507 : | int32_t EPZSSearch16( | ||
1508 : | const uint8_t * const pRef, | ||
1509 : | const uint8_t * const pRefH, | ||
1510 : | const uint8_t * const pRefV, | ||
1511 : | const uint8_t * const pRefHV, | ||
1512 : | const IMAGE * const pCur, | ||
1513 : | const int x, const int y, | ||
1514 : | const uint32_t MotionFlags, | ||
1515 : | suxen_drol | 1.8 | const uint32_t iQuant, |
1516 : | const uint32_t iFcode, | ||
1517 : | chl | 1.3 | const MBParam * const pParam, |
1518 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
1519 : | const MACROBLOCK * const prevMBs, | ||
1520 : | chl | 1.3 | VECTOR * const currMV, |
1521 : | VECTOR * const currPMV) | ||
1522 : | { | ||
1523 : | chenm001 | 1.7 | const uint32_t iWcount = pParam->mb_width; |
1524 : | const uint32_t iHcount = pParam->mb_height; | ||
1525 : | chl | 1.3 | |
1526 : | const int32_t iWidth = pParam->width; | ||
1527 : | const int32_t iHeight = pParam->height; | ||
1528 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
1529 : | |||
1530 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
1531 : | |||
1532 : | int32_t min_dx; | ||
1533 : | int32_t max_dx; | ||
1534 : | int32_t min_dy; | ||
1535 : | int32_t max_dy; | ||
1536 : | |||
1537 : | VECTOR newMV; | ||
1538 : | VECTOR backupMV; | ||
1539 : | |||
1540 : | VECTOR pmv[4]; | ||
1541 : | int32_t psad[8]; | ||
1542 : | |||
1543 : | static MACROBLOCK * oldMBs = NULL; | ||
1544 : | chl | 1.15 | // const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1545 : | suxen_drol | 1.8 | const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1546 : | chl | 1.3 | MACROBLOCK * oldMB = NULL; |
1547 : | |||
1548 : | static int32_t thresh2; | ||
1549 : | int32_t bPredEq; | ||
1550 : | int32_t iMinSAD,iSAD=9999; | ||
1551 : | |||
1552 : | MainSearch16FuncPtr EPZSMainSearchPtr; | ||
1553 : | |||
1554 : | if (oldMBs == NULL) | ||
1555 : | chl | 1.10 | { oldMBs = (MACROBLOCK*) calloc(iWcount*iHcount,sizeof(MACROBLOCK)); |
1556 : | chl | 1.9 | // fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
1557 : | chl | 1.3 | } |
1558 : | oldMB = oldMBs + x + y * iWcount; | ||
1559 : | |||
1560 : | /* Get maximum range */ | ||
1561 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, | ||
1562 : | x, y, 16, iWidth, iHeight, iFcode); | ||
1563 : | |||
1564 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1565 : | { min_dx = EVEN(min_dx); | ||
1566 : | max_dx = EVEN(max_dx); | ||
1567 : | min_dy = EVEN(min_dy); | ||
1568 : | max_dy = EVEN(max_dy); | ||
1569 : | } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ | ||
1570 : | |||
1571 : | bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); | ||
1572 : | |||
1573 : | /* Step 4: Calculate SAD around the Median prediction. | ||
1574 : | MinSAD=SAD | ||
1575 : | If Motion Vector equal to Previous frame motion vector | ||
1576 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1577 : | If SAD<=256 goto Step 10. | ||
1578 : | */ | ||
1579 : | |||
1580 : | // Prepare for main loop | ||
1581 : | |||
1582 : | *currMV=pmv[0]; /* current best := median prediction */ | ||
1583 : | if (!(MotionFlags & PMV_HALFPEL16)) | ||
1584 : | { | ||
1585 : | currMV->x = EVEN(currMV->x); | ||
1586 : | currMV->y = EVEN(currMV->y); | ||
1587 : | } | ||
1588 : | |||
1589 : | if (currMV->x > max_dx) | ||
1590 : | currMV->x=max_dx; | ||
1591 : | if (currMV->x < min_dx) | ||
1592 : | currMV->x=min_dx; | ||
1593 : | if (currMV->y > max_dy) | ||
1594 : | currMV->y=max_dy; | ||
1595 : | if (currMV->y < min_dy) | ||
1596 : | currMV->y=min_dy; | ||
1597 : | |||
1598 : | /***************** This is predictor SET A: only median prediction ******************/ | ||
1599 : | |||
1600 : | iMinSAD = sad16( cur, | ||
1601 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), | ||
1602 : | iEdgedWidth, MV_MAX_ERROR); | ||
1603 : | chl | 1.10 | iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
1604 : | chl | 1.3 | |
1605 : | // thresh1 is fixed to 256 | ||
1606 : | chl | 1.10 | if ( (iMinSAD < 256 ) || ( (MVequal(*currMV, prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
1607 : | chl | 1.3 | { |
1608 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1609 : | goto EPZS16_Terminate_without_Refine; | ||
1610 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1611 : | goto EPZS16_Terminate_with_Refine; | ||
1612 : | } | ||
1613 : | |||
1614 : | /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ | ||
1615 : | |||
1616 : | // previous frame MV | ||
1617 : | chl | 1.10 | CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1618 : | chl | 1.3 | |
1619 : | // set threshhold based on Min of Prediction and SAD of collocated block | ||
1620 : | // CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want | ||
1621 : | |||
1622 : | if ((x==0) && (y==0) ) | ||
1623 : | { | ||
1624 : | thresh2 = 512; | ||
1625 : | } | ||
1626 : | else | ||
1627 : | { | ||
1628 : | /* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ | ||
1629 : | |||
1630 : | thresh2 = MIN(psad[0],iSAD)*6/5 + 128; | ||
1631 : | } | ||
1632 : | |||
1633 : | // MV=(0,0) is often a good choice | ||
1634 : | |||
1635 : | CHECK_MV16_ZERO; | ||
1636 : | |||
1637 : | |||
1638 : | // left neighbour, if allowed | ||
1639 : | if (x != 0) | ||
1640 : | { | ||
1641 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1642 : | { pmv[1].x = EVEN(pmv[1].x); | ||
1643 : | pmv[1].y = EVEN(pmv[1].y); | ||
1644 : | } | ||
1645 : | CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); | ||
1646 : | } | ||
1647 : | |||
1648 : | // top neighbour, if allowed | ||
1649 : | if (y != 0) | ||
1650 : | { | ||
1651 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1652 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1653 : | pmv[2].y = EVEN(pmv[2].y); | ||
1654 : | } | ||
1655 : | CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1656 : | |||
1657 : | // top right neighbour, if allowed | ||
1658 : | chenm001 | 1.7 | if ((uint32_t)x != (iWcount-1)) |
1659 : | chl | 1.3 | { |
1660 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1661 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1662 : | pmv[3].y = EVEN(pmv[3].y); | ||
1663 : | } | ||
1664 : | CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); | ||
1665 : | } | ||
1666 : | } | ||
1667 : | |||
1668 : | /* Terminate if MinSAD <= T_2 | ||
1669 : | Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] | ||
1670 : | */ | ||
1671 : | |||
1672 : | if ( (iMinSAD <= thresh2) | ||
1673 : | chl | 1.10 | || ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD <= prevMB->sad16) ) ) |
1674 : | chl | 1.3 | { |
1675 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1676 : | goto EPZS16_Terminate_without_Refine; | ||
1677 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1678 : | goto EPZS16_Terminate_with_Refine; | ||
1679 : | } | ||
1680 : | |||
1681 : | /***** predictor SET C: acceleration MV (new!), neighbours in prev. frame(new!) ****/ | ||
1682 : | |||
1683 : | chl | 1.10 | backupMV = prevMB->mvs[0]; // collocated MV |
1684 : | backupMV.x += (prevMB->mvs[0].x - oldMB->mvs[0].x ); // acceleration X | ||
1685 : | backupMV.y += (prevMB->mvs[0].y - oldMB->mvs[0].y ); // acceleration Y | ||
1686 : | chl | 1.3 | |
1687 : | chl | 1.10 | CHECK_MV16_CANDIDATE(backupMV.x,backupMV.y); |
1688 : | chl | 1.3 | |
1689 : | // left neighbour | ||
1690 : | if (x != 0) | ||
1691 : | chl | 1.10 | CHECK_MV16_CANDIDATE((prevMB-1)->mvs[0].x,(prevMB-1)->mvs[0].y); |
1692 : | chl | 1.3 | |
1693 : | // top neighbour | ||
1694 : | if (y != 0) | ||
1695 : | chl | 1.10 | CHECK_MV16_CANDIDATE((prevMB-iWcount)->mvs[0].x,(prevMB-iWcount)->mvs[0].y); |
1696 : | chl | 1.3 | |
1697 : | // right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs | ||
1698 : | |||
1699 : | chenm001 | 1.7 | if ((uint32_t)x != iWcount-1) |
1700 : | chl | 1.10 | CHECK_MV16_CANDIDATE((prevMB+1)->mvs[0].x,(prevMB+1)->mvs[0].y); |
1701 : | chl | 1.3 | |
1702 : | // bottom neighbour, dito | ||
1703 : | chenm001 | 1.7 | if ((uint32_t)y != iHcount-1) |
1704 : | chl | 1.10 | CHECK_MV16_CANDIDATE((prevMB+iWcount)->mvs[0].x,(prevMB+iWcount)->mvs[0].y); |
1705 : | chl | 1.3 | |
1706 : | /* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ | ||
1707 : | if (iMinSAD <= thresh2) | ||
1708 : | { | ||
1709 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1710 : | goto EPZS16_Terminate_without_Refine; | ||
1711 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1712 : | goto EPZS16_Terminate_with_Refine; | ||
1713 : | } | ||
1714 : | |||
1715 : | /************ (if Diamond Search) **************/ | ||
1716 : | |||
1717 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1718 : | |||
1719 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1720 : | |||
1721 : | if (MotionFlags & PMV_USESQUARES16) | ||
1722 : | EPZSMainSearchPtr = Square16_MainSearch; | ||
1723 : | else | ||
1724 : | EPZSMainSearchPtr = Diamond16_MainSearch; | ||
1725 : | |||
1726 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1727 : | x, y, | ||
1728 : | currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, | ||
1729 : | 2, iFcode, iQuant, 0); | ||
1730 : | |||
1731 : | if (iSAD < iMinSAD) | ||
1732 : | { | ||
1733 : | *currMV = newMV; | ||
1734 : | iMinSAD = iSAD; | ||
1735 : | } | ||
1736 : | |||
1737 : | |||
1738 : | if (MotionFlags & PMV_EXTSEARCH16) | ||
1739 : | { | ||
1740 : | /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ | ||
1741 : | |||
1742 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1743 : | { | ||
1744 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1745 : | x, y, | ||
1746 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1747 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); | ||
1748 : | } | ||
1749 : | |||
1750 : | if (iSAD < iMinSAD) | ||
1751 : | { | ||
1752 : | *currMV = newMV; | ||
1753 : | iMinSAD = iSAD; | ||
1754 : | } | ||
1755 : | |||
1756 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1757 : | { | ||
1758 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1759 : | x, y, | ||
1760 : | 0, 0, iMinSAD, &newMV, | ||
1761 : | chl | 1.10 | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
1762 : | chl | 1.3 | |
1763 : | if (iSAD < iMinSAD) | ||
1764 : | { | ||
1765 : | *currMV = newMV; | ||
1766 : | iMinSAD = iSAD; | ||
1767 : | } | ||
1768 : | } | ||
1769 : | } | ||
1770 : | |||
1771 : | /*************** Choose best MV found **************/ | ||
1772 : | |||
1773 : | EPZS16_Terminate_with_Refine: | ||
1774 : | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step | ||
1775 : | iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, | ||
1776 : | x, y, | ||
1777 : | currMV, iMinSAD, | ||
1778 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1779 : | |||
1780 : | EPZS16_Terminate_without_Refine: | ||
1781 : | |||
1782 : | chl | 1.10 | *oldMB = *prevMB; |
1783 : | chl | 1.3 | |
1784 : | currPMV->x = currMV->x - pmv[0].x; | ||
1785 : | currPMV->y = currMV->y - pmv[0].y; | ||
1786 : | return iMinSAD; | ||
1787 : | } | ||
1788 : | |||
1789 : | |||
1790 : | int32_t EPZSSearch8( | ||
1791 : | const uint8_t * const pRef, | ||
1792 : | const uint8_t * const pRefH, | ||
1793 : | const uint8_t * const pRefV, | ||
1794 : | const uint8_t * const pRefHV, | ||
1795 : | const IMAGE * const pCur, | ||
1796 : | const int x, const int y, | ||
1797 : | const int start_x, const int start_y, | ||
1798 : | const uint32_t MotionFlags, | ||
1799 : | suxen_drol | 1.8 | const uint32_t iQuant, |
1800 : | const uint32_t iFcode, | ||
1801 : | chl | 1.3 | const MBParam * const pParam, |
1802 : | suxen_drol | 1.8 | const MACROBLOCK * const pMBs, |
1803 : | const MACROBLOCK * const prevMBs, | ||
1804 : | chl | 1.3 | VECTOR * const currMV, |
1805 : | VECTOR * const currPMV) | ||
1806 : | { | ||
1807 : | chl | 1.10 | /* Please not that EPZS might not be a good choice for 8x8-block motion search ! */ |
1808 : | |||
1809 : | const uint32_t iWcount = pParam->mb_width; | ||
1810 : | chl | 1.3 | const int32_t iWidth = pParam->width; |
1811 : | const int32_t iHeight = pParam->height; | ||
1812 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
1813 : | |||
1814 : | const uint8_t * cur = pCur->y + x*8 + y*8*iEdgedWidth; | ||
1815 : | |||
1816 : | int32_t iDiamondSize=1; | ||
1817 : | |||
1818 : | int32_t min_dx; | ||
1819 : | int32_t max_dx; | ||
1820 : | int32_t min_dy; | ||
1821 : | int32_t max_dy; | ||
1822 : | |||
1823 : | VECTOR newMV; | ||
1824 : | VECTOR backupMV; | ||
1825 : | |||
1826 : | VECTOR pmv[4]; | ||
1827 : | int32_t psad[8]; | ||
1828 : | |||
1829 : | const int32_t iSubBlock = ((y&1)<<1) + (x&1); | ||
1830 : | |||
1831 : | chl | 1.15 | // const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1832 : | suxen_drol | 1.8 | const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1833 : | chl | 1.3 | |
1834 : | int32_t bPredEq; | ||
1835 : | int32_t iMinSAD,iSAD=9999; | ||
1836 : | |||
1837 : | MainSearch8FuncPtr EPZSMainSearchPtr; | ||
1838 : | |||
1839 : | /* Get maximum range */ | ||
1840 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, | ||
1841 : | x, y, 8, iWidth, iHeight, iFcode); | ||
1842 : | |||
1843 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
1844 : | |||
1845 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1846 : | { min_dx = EVEN(min_dx); | ||
1847 : | max_dx = EVEN(max_dx); | ||
1848 : | min_dy = EVEN(min_dy); | ||
1849 : | max_dy = EVEN(max_dy); | ||
1850 : | } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ | ||
1851 : | |||
1852 : | bPredEq = get_pmvdata(pMBs, x>>1, y>>1, iWcount, iSubBlock, pmv, psad); | ||
1853 : | |||
1854 : | |||
1855 : | /* Step 4: Calculate SAD around the Median prediction. | ||
1856 : | MinSAD=SAD | ||
1857 : | If Motion Vector equal to Previous frame motion vector | ||
1858 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1859 : | If SAD<=256 goto Step 10. | ||
1860 : | */ | ||
1861 : | |||
1862 : | // Prepare for main loop | ||
1863 : | |||
1864 : | |||
1865 : | if (!(MotionFlags & PMV_HALFPEL8)) | ||
1866 : | { | ||
1867 : | currMV->x = EVEN(currMV->x); | ||
1868 : | currMV->y = EVEN(currMV->y); | ||
1869 : | } | ||
1870 : | |||
1871 : | if (currMV->x > max_dx) | ||
1872 : | currMV->x=max_dx; | ||
1873 : | if (currMV->x < min_dx) | ||
1874 : | currMV->x=min_dx; | ||
1875 : | if (currMV->y > max_dy) | ||
1876 : | currMV->y=max_dy; | ||
1877 : | if (currMV->y < min_dy) | ||
1878 : | currMV->y=min_dy; | ||
1879 : | |||
1880 : | /***************** This is predictor SET A: only median prediction ******************/ | ||
1881 : | |||
1882 : | |||
1883 : | iMinSAD = sad8( cur, | ||
1884 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), | ||
1885 : | iEdgedWidth); | ||
1886 : | chl | 1.10 | iMinSAD += calc_delta_8(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
1887 : | chl | 1.3 | |
1888 : | |||
1889 : | // thresh1 is fixed to 256 | ||
1890 : | if (iMinSAD < 256/4 ) | ||
1891 : | { | ||
1892 : | if (MotionFlags & PMV_QUICKSTOP8) | ||
1893 : | goto EPZS8_Terminate_without_Refine; | ||
1894 : | if (MotionFlags & PMV_EARLYSTOP8) | ||
1895 : | goto EPZS8_Terminate_with_Refine; | ||
1896 : | } | ||
1897 : | |||
1898 : | /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ | ||
1899 : | |||
1900 : | chl | 1.10 | |
1901 : | // MV=(0,0) is often a good choice | ||
1902 : | CHECK_MV8_ZERO; | ||
1903 : | |||
1904 : | chl | 1.3 | // previous frame MV |
1905 : | chl | 1.10 | CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
1906 : | |||
1907 : | // left neighbour, if allowed | ||
1908 : | if (psad[1] != MV_MAX_ERROR) | ||
1909 : | { | ||
1910 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1911 : | { pmv[1].x = EVEN(pmv[1].x); | ||
1912 : | pmv[1].y = EVEN(pmv[1].y); | ||
1913 : | } | ||
1914 : | CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); | ||
1915 : | } | ||
1916 : | chl | 1.3 | |
1917 : | chl | 1.10 | // top neighbour, if allowed |
1918 : | if (psad[2] != MV_MAX_ERROR) | ||
1919 : | { | ||
1920 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1921 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1922 : | pmv[2].y = EVEN(pmv[2].y); | ||
1923 : | } | ||
1924 : | CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1925 : | |||
1926 : | // top right neighbour, if allowed | ||
1927 : | if (psad[3] != MV_MAX_ERROR) | ||
1928 : | { | ||
1929 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1930 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1931 : | pmv[3].y = EVEN(pmv[3].y); | ||
1932 : | } | ||
1933 : | CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); | ||
1934 : | } | ||
1935 : | } | ||
1936 : | |||
1937 : | /* // this bias is zero anyway, at the moment! | ||
1938 : | |||
1939 : | if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) ) // && (iMinSAD <= iQuant * 96) | ||
1940 : | iMinSAD -= MV8_00_BIAS; | ||
1941 : | chl | 1.3 | |
1942 : | chl | 1.10 | */ |
1943 : | chl | 1.3 | |
1944 : | /* Terminate if MinSAD <= T_2 | ||
1945 : | Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] | ||
1946 : | */ | ||
1947 : | |||
1948 : | if (iMinSAD < 512/4) /* T_2 == 512/4 hardcoded */ | ||
1949 : | { | ||
1950 : | if (MotionFlags & PMV_QUICKSTOP8) | ||
1951 : | goto EPZS8_Terminate_without_Refine; | ||
1952 : | if (MotionFlags & PMV_EARLYSTOP8) | ||
1953 : | goto EPZS8_Terminate_with_Refine; | ||
1954 : | } | ||
1955 : | |||
1956 : | chl | 1.10 | /************ (Diamond Search) **************/ |
1957 : | chl | 1.3 | |
1958 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1959 : | |||
1960 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8)) | ||
1961 : | iDiamondSize *= 2; | ||
1962 : | |||
1963 : | chl | 1.10 | /* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
1964 : | |||
1965 : | /* // there is no EPZS^2 for inter4v at the moment | ||
1966 : | |||
1967 : | if (MotionFlags & PMV_USESQUARES8) | ||
1968 : | EPZSMainSearchPtr = Square8_MainSearch; | ||
1969 : | else | ||
1970 : | */ | ||
1971 : | chl | 1.3 | |
1972 : | chl | 1.10 | EPZSMainSearchPtr = Diamond8_MainSearch; |
1973 : | chl | 1.3 | |
1974 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1975 : | x, y, | ||
1976 : | currMV->x, currMV->y, iMinSAD, &newMV, | ||
1977 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, | ||
1978 : | chl | 1.10 | iDiamondSize, iFcode, iQuant, 0); |
1979 : | |||
1980 : | chl | 1.3 | |
1981 : | if (iSAD < iMinSAD) | ||
1982 : | { | ||
1983 : | *currMV = newMV; | ||
1984 : | iMinSAD = iSAD; | ||
1985 : | } | ||
1986 : | |||
1987 : | if (MotionFlags & PMV_EXTSEARCH8) | ||
1988 : | { | ||
1989 : | /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ | ||
1990 : | |||
1991 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1992 : | { | ||
1993 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1994 : | x, y, | ||
1995 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1996 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); | ||
1997 : | |||
1998 : | if (iSAD < iMinSAD) | ||
1999 : | { | ||
2000 : | *currMV = newMV; | ||
2001 : | iMinSAD = iSAD; | ||
2002 : | } | ||
2003 : | } | ||
2004 : | |||
2005 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
2006 : | { | ||
2007 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
2008 : | x, y, | ||
2009 : | 0, 0, iMinSAD, &newMV, | ||
2010 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); | ||
2011 : | |||
2012 : | if (iSAD < iMinSAD) | ||
2013 : | { | ||
2014 : | *currMV = newMV; | ||
2015 : | iMinSAD = iSAD; | ||
2016 : | } | ||
2017 : | } | ||
2018 : | } | ||
2019 : | |||
2020 : | /*************** Choose best MV found **************/ | ||
2021 : | |||
2022 : | EPZS8_Terminate_with_Refine: | ||
2023 : | if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step | ||
2024 : | iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, | ||
2025 : | x, y, | ||
2026 : | currMV, iMinSAD, | ||
2027 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
2028 : | |||
2029 : | EPZS8_Terminate_without_Refine: | ||
2030 : | |||
2031 : | currPMV->x = currMV->x - pmv[0].x; | ||
2032 : | currPMV->y = currMV->y - pmv[0].y; | ||
2033 : | return iMinSAD; | ||
2034 : | } | ||
2035 : | |||
2036 : | suxen_drol | 1.6 | |
2037 : | |||
2038 : | |||
2039 : | |||
2040 : | /* *********************************************************** | ||
2041 : | bvop motion estimation | ||
2042 : | // TODO: need to incorporate prediction here (eg. sad += calc_delta_16) | ||
2043 : | ***************************************************************/ | ||
2044 : | |||
2045 : | suxen_drol | 1.11 | |
2046 : | suxen_drol | 1.6 | void MotionEstimationBVOP( |
2047 : | MBParam * const pParam, | ||
2048 : | FRAMEINFO * const frame, | ||
2049 : | |||
2050 : | // forward (past) reference | ||
2051 : | const MACROBLOCK * const f_mbs, | ||
2052 : | const IMAGE * const f_ref, | ||
2053 : | const IMAGE * const f_refH, | ||
2054 : | const IMAGE * const f_refV, | ||
2055 : | const IMAGE * const f_refHV, | ||
2056 : | // backward (future) reference | ||
2057 : | const MACROBLOCK * const b_mbs, | ||
2058 : | const IMAGE * const b_ref, | ||
2059 : | const IMAGE * const b_refH, | ||
2060 : | const IMAGE * const b_refV, | ||
2061 : | const IMAGE * const b_refHV) | ||
2062 : | { | ||
2063 : | const uint32_t mb_width = pParam->mb_width; | ||
2064 : | const uint32_t mb_height = pParam->mb_height; | ||
2065 : | const int32_t edged_width = pParam->edged_width; | ||
2066 : | |||
2067 : | suxen_drol | 1.11 | uint32_t i,j; |
2068 : | suxen_drol | 1.6 | |
2069 : | int32_t f_sad16; | ||
2070 : | int32_t b_sad16; | ||
2071 : | int32_t i_sad16; | ||
2072 : | int32_t d_sad16; | ||
2073 : | int32_t best_sad; | ||
2074 : | |||
2075 : | VECTOR pmv_dontcare; | ||
2076 : | |||
2077 : | // note: i==horizontal, j==vertical | ||
2078 : | for (j = 0; j < mb_height; j++) | ||
2079 : | { | ||
2080 : | for (i = 0; i < mb_width; i++) | ||
2081 : | { | ||
2082 : | MACROBLOCK *mb = &frame->mbs[i + j*mb_width]; | ||
2083 : | const MACROBLOCK *f_mb = &f_mbs[i + j*mb_width]; | ||
2084 : | const MACROBLOCK *b_mb = &b_mbs[i + j*mb_width]; | ||
2085 : | |||
2086 : | if (b_mb->mode == MODE_INTER | ||
2087 : | && b_mb->cbp == 0 | ||
2088 : | && b_mb->mvs[0].x == 0 | ||
2089 : | && b_mb->mvs[0].y == 0) | ||
2090 : | { | ||
2091 : | suxen_drol | 1.11 | mb->mode = MODE_NOT_CODED; |
2092 : | suxen_drol | 1.6 | mb->mvs[0].x = 0; |
2093 : | mb->mvs[0].y = 0; | ||
2094 : | mb->b_mvs[0].x = 0; | ||
2095 : | mb->b_mvs[0].y = 0; | ||
2096 : | continue; | ||
2097 : | } | ||
2098 : | |||
2099 : | |||
2100 : | // forward search | ||
2101 : | f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, | ||
2102 : | &frame->image, | ||
2103 : | i, j, | ||
2104 : | frame->motion_flags, frame->quant, frame->fcode, | ||
2105 : | pParam, | ||
2106 : | suxen_drol | 1.11 | f_mbs, f_mbs /* todo */, |
2107 : | suxen_drol | 1.6 | &mb->mvs[0], &pmv_dontcare); // ignore pmv |
2108 : | |||
2109 : | // backward search | ||
2110 : | b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
2111 : | &frame->image, | ||
2112 : | i, j, | ||
2113 : | frame->motion_flags, frame->quant, frame->bcode, | ||
2114 : | pParam, | ||
2115 : | suxen_drol | 1.11 | b_mbs, b_mbs, /* todo */ |
2116 : | suxen_drol | 1.6 | &mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
2117 : | |||
2118 : | // interpolate search (simple, but effective) | ||
2119 : | i_sad16 = sad16bi_c( | ||
2120 : | frame->image.y + i*16 + j*16*edged_width, | ||
2121 : | get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, | ||
2122 : | i, j, 16, mb->mvs[0].x, mb->mvs[0].y, edged_width), | ||
2123 : | get_ref(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
2124 : | i, j, 16, mb->b_mvs[0].x, mb->b_mvs[0].x, edged_width), | ||
2125 : | edged_width); | ||
2126 : | |||
2127 : | // TODO: direct search | ||
2128 : | // predictor + range of [-32,32] | ||
2129 : | d_sad16 = 65535; | ||
2130 : | |||
2131 : | |||
2132 : | if (f_sad16 < b_sad16) | ||
2133 : | { | ||
2134 : | best_sad = f_sad16; | ||
2135 : | suxen_drol | 1.11 | mb->mode = MODE_FORWARD; |
2136 : | suxen_drol | 1.6 | } |
2137 : | else | ||
2138 : | { | ||
2139 : | best_sad = b_sad16; | ||
2140 : | suxen_drol | 1.11 | mb->mode = MODE_BACKWARD; |
2141 : | suxen_drol | 1.6 | } |
2142 : | |||
2143 : | if (i_sad16 < best_sad) | ||
2144 : | { | ||
2145 : | best_sad = i_sad16; | ||
2146 : | suxen_drol | 1.11 | mb->mode = MODE_INTERPOLATE; |
2147 : | suxen_drol | 1.6 | } |
2148 : | |||
2149 : | if (d_sad16 < best_sad) | ||
2150 : | { | ||
2151 : | best_sad = d_sad16; | ||
2152 : | suxen_drol | 1.11 | mb->mode = MODE_DIRECT; |
2153 : | suxen_drol | 1.6 | } |
2154 : | |||
2155 : | } | ||
2156 : | } | ||
2157 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |