Parent Directory
|
Revision Log
Revision 140 - (view) (download)
1 : | Isibaar | 3 | /************************************************************************** |
2 : | * | ||
3 : | * Modifications: | ||
4 : | * | ||
5 : | suxen_drol | 136 | * 25.04.2002 partial prevMB conversion |
6 : | chenm001 | 132 | * 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
7 : | * 14.04.2002 added MotionEstimationBVOP() | ||
8 : | chl | 96 | * 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
9 : | * EPZS and EPZS^2 | ||
10 : | Isibaar | 3 | * 08.02.2002 split up PMVfast into three routines: PMVFast, PMVFast_MainLoop |
11 : | * PMVFast_Refine to support multiple searches with different start points | ||
12 : | edgomez | 78 | * 07.01.2002 uv-block-based interpolation |
13 : | Isibaar | 3 | * 06.01.2002 INTER/INTRA-decision is now done before any SEARCH8 (speedup) |
14 : | edgomez | 78 | * changed INTER_BIAS to 150 (as suggested by suxen_drol) |
15 : | * removed halfpel refinement step in PMVfastSearch8 + quality=5 | ||
16 : | * added new quality mode = 6 which performs halfpel refinement | ||
17 : | * filesize difference between quality 5 and 6 is smaller than 1% | ||
18 : | Isibaar | 3 | * (Isibaar) |
19 : | * 31.12.2001 PMVfastSearch16 and PMVfastSearch8 (gruel) | ||
20 : | edgomez | 78 | * 30.12.2001 get_range/MotionSearchX simplified; blue/green bug fix |
21 : | * 22.12.2001 commented best_point==99 check | ||
22 : | * 19.12.2001 modified get_range (purple bug fix) | ||
23 : | Isibaar | 3 | * 15.12.2001 moved pmv displacement from mbprediction |
24 : | * 02.12.2001 motion estimation/compensation split (Isibaar) | ||
25 : | edgomez | 78 | * 16.11.2001 rewrote/tweaked search algorithms; pross@cs.rmit.edu.au |
26 : | Isibaar | 3 | * 10.11.2001 support for sad16/sad8 functions |
27 : | * 28.08.2001 reactivated MODE_INTER4V for EXT_MODE | ||
28 : | * 24.08.2001 removed MODE_INTER4V_Q, disabled MODE_INTER4V for EXT_MODE | ||
29 : | edgomez | 78 | * 22.08.2001 added MODE_INTER4V_Q |
30 : | Isibaar | 3 | * 20.08.2001 added pragma to get rid of internal compiler error with VC6 |
31 : | * idea by Cyril. Thanks. | ||
32 : | * | ||
33 : | * Michael Militzer <isibaar@videocoding.de> | ||
34 : | * | ||
35 : | **************************************************************************/ | ||
36 : | |||
37 : | #include <assert.h> | ||
38 : | #include <stdio.h> | ||
39 : | chl | 96 | #include <stdlib.h> |
40 : | Isibaar | 3 | |
41 : | #include "../encoder.h" | ||
42 : | #include "../utils/mbfunctions.h" | ||
43 : | #include "../prediction/mbprediction.h" | ||
44 : | #include "../global.h" | ||
45 : | #include "../utils/timer.h" | ||
46 : | suxen_drol | 118 | #include "motion.h" |
47 : | Isibaar | 3 | #include "sad.h" |
48 : | |||
49 : | // very large value | ||
50 : | #define MV_MAX_ERROR (4096 * 256) | ||
51 : | |||
52 : | // stop search if sdelta < THRESHOLD | ||
53 : | #define MV16_THRESHOLD 192 | ||
54 : | #define MV8_THRESHOLD 56 | ||
55 : | |||
56 : | /* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ | ||
57 : | /* nb = vop pixels * 2^(bpp-8) */ | ||
58 : | #define MV16_00_BIAS (128+1) | ||
59 : | chl | 140 | #define MV8_00_BIAS (0) |
60 : | Isibaar | 3 | |
61 : | /* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ | ||
62 : | #define INTER_BIAS 512 | ||
63 : | |||
64 : | /* Parameters which control inter/inter4v decision */ | ||
65 : | #define IMV16X16 5 | ||
66 : | |||
67 : | /* vector map (vlc delta size) smoother parameters */ | ||
68 : | #define NEIGH_TEND_16X16 2 | ||
69 : | #define NEIGH_TEND_8X8 2 | ||
70 : | |||
71 : | |||
72 : | // fast ((A)/2)*2 | ||
73 : | #define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) | ||
74 : | |||
75 : | |||
76 : | chl | 96 | int32_t PMVfastSearch16( |
77 : | const uint8_t * const pRef, | ||
78 : | const uint8_t * const pRefH, | ||
79 : | const uint8_t * const pRefV, | ||
80 : | const uint8_t * const pRefHV, | ||
81 : | const IMAGE * const pCur, | ||
82 : | const int x, const int y, | ||
83 : | const uint32_t MotionFlags, | ||
84 : | suxen_drol | 136 | const uint32_t iQuant, |
85 : | const uint32_t iFcode, | ||
86 : | chl | 96 | const MBParam * const pParam, |
87 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
88 : | const MACROBLOCK * const prevMBs, | ||
89 : | chl | 96 | VECTOR * const currMV, |
90 : | VECTOR * const currPMV); | ||
91 : | Isibaar | 3 | |
92 : | chl | 96 | int32_t EPZSSearch16( |
93 : | const uint8_t * const pRef, | ||
94 : | const uint8_t * const pRefH, | ||
95 : | const uint8_t * const pRefV, | ||
96 : | const uint8_t * const pRefHV, | ||
97 : | const IMAGE * const pCur, | ||
98 : | const int x, const int y, | ||
99 : | const uint32_t MotionFlags, | ||
100 : | suxen_drol | 136 | const uint32_t iQuant, |
101 : | const uint32_t iFcode, | ||
102 : | chl | 96 | const MBParam * const pParam, |
103 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
104 : | const MACROBLOCK * const prevMBs, | ||
105 : | chl | 96 | VECTOR * const currMV, |
106 : | VECTOR * const currPMV); | ||
107 : | |||
108 : | |||
109 : | Isibaar | 3 | int32_t PMVfastSearch8( |
110 : | chl | 96 | const uint8_t * const pRef, |
111 : | const uint8_t * const pRefH, | ||
112 : | const uint8_t * const pRefV, | ||
113 : | const uint8_t * const pRefHV, | ||
114 : | const IMAGE * const pCur, | ||
115 : | const int x, const int y, | ||
116 : | suxen_drol | 136 | const int start_x, const int start_y, |
117 : | chl | 96 | const uint32_t MotionFlags, |
118 : | suxen_drol | 136 | const uint32_t iQuant, |
119 : | const uint32_t iFcode, | ||
120 : | chl | 96 | const MBParam * const pParam, |
121 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
122 : | const MACROBLOCK * const prevMBs, | ||
123 : | chl | 96 | VECTOR * const currMV, |
124 : | VECTOR * const currPMV); | ||
125 : | |||
126 : | int32_t EPZSSearch8( | ||
127 : | const uint8_t * const pRef, | ||
128 : | const uint8_t * const pRefH, | ||
129 : | const uint8_t * const pRefV, | ||
130 : | const uint8_t * const pRefHV, | ||
131 : | const IMAGE * const pCur, | ||
132 : | const int x, const int y, | ||
133 : | suxen_drol | 136 | const int start_x, const int start_y, |
134 : | chl | 96 | const uint32_t MotionFlags, |
135 : | suxen_drol | 136 | const uint32_t iQuant, |
136 : | const uint32_t iFcode, | ||
137 : | chl | 96 | const MBParam * const pParam, |
138 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
139 : | const MACROBLOCK * const prevMBs, | ||
140 : | chl | 96 | VECTOR * const currMV, |
141 : | VECTOR * const currPMV); | ||
142 : | |||
143 : | |||
144 : | typedef int32_t (MainSearch16Func)( | ||
145 : | edgomez | 78 | const uint8_t * const pRef, |
146 : | const uint8_t * const pRefH, | ||
147 : | const uint8_t * const pRefV, | ||
148 : | const uint8_t * const pRefHV, | ||
149 : | chl | 96 | const uint8_t * const cur, |
150 : | edgomez | 78 | const int x, const int y, |
151 : | chl | 96 | int32_t startx, int32_t starty, |
152 : | int32_t iMinSAD, | ||
153 : | edgomez | 78 | VECTOR * const currMV, |
154 : | chl | 96 | const VECTOR * const pmv, |
155 : | const int32_t min_dx, const int32_t max_dx, | ||
156 : | const int32_t min_dy, const int32_t max_dy, | ||
157 : | const int32_t iEdgedWidth, | ||
158 : | const int32_t iDiamondSize, | ||
159 : | const int32_t iFcode, | ||
160 : | const int32_t iQuant, | ||
161 : | int iFound); | ||
162 : | Isibaar | 3 | |
163 : | chl | 96 | typedef MainSearch16Func* MainSearch16FuncPtr; |
164 : | |||
165 : | |||
166 : | typedef int32_t (MainSearch8Func)( | ||
167 : | edgomez | 78 | const uint8_t * const pRef, |
168 : | const uint8_t * const pRefH, | ||
169 : | const uint8_t * const pRefV, | ||
170 : | const uint8_t * const pRefHV, | ||
171 : | chl | 96 | const uint8_t * const cur, |
172 : | edgomez | 78 | const int x, const int y, |
173 : | chl | 96 | int32_t startx, int32_t starty, |
174 : | int32_t iMinSAD, | ||
175 : | edgomez | 78 | VECTOR * const currMV, |
176 : | chl | 96 | const VECTOR * const pmv, |
177 : | const int32_t min_dx, const int32_t max_dx, | ||
178 : | const int32_t min_dy, const int32_t max_dy, | ||
179 : | const int32_t iEdgedWidth, | ||
180 : | const int32_t iDiamondSize, | ||
181 : | const int32_t iFcode, | ||
182 : | const int32_t iQuant, | ||
183 : | int iFound); | ||
184 : | Isibaar | 3 | |
185 : | chl | 96 | typedef MainSearch8Func* MainSearch8FuncPtr; |
186 : | Isibaar | 3 | |
187 : | // mv.length table | ||
188 : | static const uint32_t mvtab[33] = { | ||
189 : | 1, 2, 3, 4, 6, 7, 7, 7, | ||
190 : | 9, 9, 9, 10, 10, 10, 10, 10, | ||
191 : | 10, 10, 10, 10, 10, 10, 10, 10, | ||
192 : | 10, 11, 11, 11, 11, 11, 11, 12, 12 | ||
193 : | }; | ||
194 : | |||
195 : | |||
196 : | static __inline uint32_t mv_bits(int32_t component, const uint32_t iFcode) | ||
197 : | { | ||
198 : | if (component == 0) | ||
199 : | return 1; | ||
200 : | |||
201 : | if (component < 0) | ||
202 : | component = -component; | ||
203 : | |||
204 : | if (iFcode == 1) | ||
205 : | { | ||
206 : | if (component > 32) | ||
207 : | component = 32; | ||
208 : | |||
209 : | return mvtab[component] + 1; | ||
210 : | } | ||
211 : | |||
212 : | component += (1 << (iFcode - 1)) - 1; | ||
213 : | component >>= (iFcode - 1); | ||
214 : | |||
215 : | if (component > 32) | ||
216 : | component = 32; | ||
217 : | |||
218 : | return mvtab[component] + 1 + iFcode - 1; | ||
219 : | } | ||
220 : | |||
221 : | |||
222 : | static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode) | ||
223 : | { | ||
224 : | return NEIGH_TEND_16X16 * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); | ||
225 : | } | ||
226 : | |||
227 : | static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode) | ||
228 : | |||
229 : | { | ||
230 : | return NEIGH_TEND_8X8 * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); | ||
231 : | } | ||
232 : | |||
233 : | |||
234 : | |||
235 : | |||
236 : | |||
237 : | #ifndef SEARCH16 | ||
238 : | #define SEARCH16 PMVfastSearch16 | ||
239 : | chl | 96 | //#define SEARCH16 FullSearch16 |
240 : | //#define SEARCH16 EPZSSearch16 | ||
241 : | Isibaar | 3 | #endif |
242 : | |||
243 : | #ifndef SEARCH8 | ||
244 : | #define SEARCH8 PMVfastSearch8 | ||
245 : | chl | 96 | //#define SEARCH8 EPZSSearch8 |
246 : | Isibaar | 3 | #endif |
247 : | |||
248 : | bool MotionEstimation( | ||
249 : | edgomez | 78 | MBParam * const pParam, |
250 : | suxen_drol | 136 | FRAMEINFO * const current, |
251 : | FRAMEINFO * const reference, | ||
252 : | edgomez | 78 | const IMAGE * const pRefH, |
253 : | const IMAGE * const pRefV, | ||
254 : | const IMAGE * const pRefHV, | ||
255 : | const uint32_t iLimit) | ||
256 : | Isibaar | 3 | |
257 : | { | ||
258 : | edgomez | 78 | const uint32_t iWcount = pParam->mb_width; |
259 : | const uint32_t iHcount = pParam->mb_height; | ||
260 : | suxen_drol | 136 | MACROBLOCK * pMBs = current->mbs; |
261 : | IMAGE * pCurrent = ¤t->image; | ||
262 : | |||
263 : | MACROBLOCK * prevMBs = reference->mbs; // previous frame | ||
264 : | IMAGE * pRef = &reference->image; | ||
265 : | |||
266 : | Isibaar | 3 | |
267 : | uint32_t i, j, iIntra = 0; | ||
268 : | |||
269 : | edgomez | 78 | VECTOR mv16; |
270 : | VECTOR pmv16; | ||
271 : | Isibaar | 3 | |
272 : | edgomez | 78 | int32_t sad8 = 0; |
273 : | int32_t sad16; | ||
274 : | int32_t deviation; | ||
275 : | Isibaar | 3 | |
276 : | canard | 117 | if (sadInit) |
277 : | canard | 115 | (*sadInit)(); |
278 : | suxen_drol | 136 | |
279 : | |||
280 : | /* eventhough we have a seperate prevMBs, | ||
281 : | pmvfast/epsz does something "funny" with the previous frames data */ | ||
282 : | |||
283 : | chl | 140 | /* for (i = 0; i < iHcount; i++) |
284 : | suxen_drol | 136 | for (j = 0; j < iWcount; j++) |
285 : | { | ||
286 : | pMBs[j + i * iWcount].mvs[0] = prevMBs[j + i * iWcount].mvs[0]; | ||
287 : | pMBs[j + i * iWcount].mvs[1] = prevMBs[j + i * iWcount].mvs[1]; | ||
288 : | pMBs[j + i * iWcount].mvs[2] = prevMBs[j + i * iWcount].mvs[2]; | ||
289 : | pMBs[j + i * iWcount].mvs[3] = prevMBs[j + i * iWcount].mvs[3]; | ||
290 : | } | ||
291 : | chl | 140 | */ |
292 : | suxen_drol | 136 | /*dprintf("*** BEFORE ***"); |
293 : | for (i = 0; i < iHcount; i++) | ||
294 : | for (j = 0; j < iWcount; j++) | ||
295 : | { | ||
296 : | dprintf(" [%i,%i] mode=%i dquant=%i mvs=(%i %i %i %i) sad8=(%i %i %i %i) sad16=(%i)", j,i, | ||
297 : | pMBs[j + i * iWcount].mode, | ||
298 : | pMBs[j + i * iWcount].dquant, | ||
299 : | pMBs[j + i * iWcount].mvs[0], | ||
300 : | pMBs[j + i * iWcount].mvs[1], | ||
301 : | pMBs[j + i * iWcount].mvs[2], | ||
302 : | pMBs[j + i * iWcount].mvs[3], | ||
303 : | prevMBs[j + i * iWcount].sad8[0], | ||
304 : | prevMBs[j + i * iWcount].sad8[1], | ||
305 : | prevMBs[j + i * iWcount].sad8[2], | ||
306 : | prevMBs[j + i * iWcount].sad8[3], | ||
307 : | prevMBs[j + i * iWcount].sad16); | ||
308 : | } | ||
309 : | */ | ||
310 : | canard | 115 | |
311 : | Isibaar | 3 | // note: i==horizontal, j==vertical |
312 : | edgomez | 78 | for (i = 0; i < iHcount; i++) |
313 : | Isibaar | 3 | for (j = 0; j < iWcount; j++) |
314 : | { | ||
315 : | MACROBLOCK *pMB = &pMBs[j + i * iWcount]; | ||
316 : | suxen_drol | 136 | MACROBLOCK *prevMB = &prevMBs[j + i * iWcount]; |
317 : | Isibaar | 3 | |
318 : | sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
319 : | suxen_drol | 136 | j, i, current->motion_flags, current->quant, current->fcode, |
320 : | pParam, pMBs, prevMBs, &mv16, &pmv16); | ||
321 : | Isibaar | 3 | pMB->sad16=sad16; |
322 : | |||
323 : | |||
324 : | edgomez | 78 | /* decide: MODE_INTER or MODE_INTRA |
325 : | if (dev_intra < sad_inter - 2 * nb) use_intra | ||
326 : | */ | ||
327 : | Isibaar | 3 | |
328 : | edgomez | 78 | deviation = dev16(pCurrent->y + j*16 + i*16*pParam->edged_width, pParam->edged_width); |
329 : | Isibaar | 3 | |
330 : | edgomez | 78 | if (deviation < (sad16 - INTER_BIAS)) |
331 : | { | ||
332 : | pMB->mode = MODE_INTRA; | ||
333 : | pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; | ||
334 : | pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; | ||
335 : | Isibaar | 3 | |
336 : | suxen_drol | 136 | pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
337 : | |||
338 : | edgomez | 78 | iIntra++; |
339 : | if(iIntra >= iLimit) | ||
340 : | return 1; | ||
341 : | Isibaar | 3 | |
342 : | edgomez | 78 | continue; |
343 : | } | ||
344 : | Isibaar | 3 | |
345 : | suxen_drol | 136 | if (current->global_flags & XVID_INTER4V) |
346 : | edgomez | 78 | { |
347 : | pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, | ||
348 : | suxen_drol | 136 | 2 * j, 2 * i, mv16.x, mv16.y, |
349 : | current->motion_flags, current->quant, current->fcode, | ||
350 : | pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); | ||
351 : | Isibaar | 3 | |
352 : | edgomez | 78 | pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
353 : | suxen_drol | 136 | 2 * j + 1, 2 * i, mv16.x, mv16.y, |
354 : | current->motion_flags, current->quant, current->fcode, | ||
355 : | pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); | ||
356 : | Isibaar | 3 | |
357 : | edgomez | 78 | pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
358 : | suxen_drol | 136 | 2 * j, 2 * i + 1, mv16.x, mv16.y, |
359 : | current->motion_flags, current->quant, current->fcode, | ||
360 : | pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); | ||
361 : | Isibaar | 3 | |
362 : | edgomez | 78 | pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
363 : | suxen_drol | 136 | 2 * j + 1, 2 * i + 1, mv16.x, mv16.y, |
364 : | current->motion_flags, current->quant, current->fcode, | ||
365 : | pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); | ||
366 : | Isibaar | 3 | |
367 : | edgomez | 78 | sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
368 : | } | ||
369 : | Isibaar | 3 | |
370 : | |||
371 : | edgomez | 78 | /* decide: MODE_INTER or MODE_INTER4V |
372 : | mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v | ||
373 : | */ | ||
374 : | Isibaar | 3 | |
375 : | suxen_drol | 136 | if (!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE) |
376 : | { | ||
377 : | if (((current->global_flags & XVID_INTER4V)==0) || | ||
378 : | (sad16 < (sad8 + (int32_t)(IMV16X16 * current->quant)))) | ||
379 : | { | ||
380 : | Isibaar | 3 | |
381 : | edgomez | 78 | sad8 = sad16; |
382 : | pMB->mode = MODE_INTER; | ||
383 : | pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; | ||
384 : | pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; | ||
385 : | pMB->pmvs[0].x = pmv16.x; | ||
386 : | pMB->pmvs[0].y = pmv16.y; | ||
387 : | } | ||
388 : | else | ||
389 : | pMB->mode = MODE_INTER4V; | ||
390 : | } | ||
391 : | else | ||
392 : | { | ||
393 : | Isibaar | 3 | sad8 = sad16; |
394 : | pMB->mode = MODE_INTER; | ||
395 : | pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; | ||
396 : | pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; | ||
397 : | pMB->pmvs[0].x = pmv16.x; | ||
398 : | pMB->pmvs[0].y = pmv16.y; | ||
399 : | } | ||
400 : | } | ||
401 : | |||
402 : | suxen_drol | 136 | /* dprintf("*** AFTER ***", pMBs[0].b_mvs[0].x); |
403 : | for (i = 0; i < iHcount; i++) | ||
404 : | for (j = 0; j < iWcount; j++) | ||
405 : | { | ||
406 : | dprintf(" [%i,%i] mode=%i dquant=%i mvs=(%i %i %i %i) sad8=(%i %i %i %i) sad16=(%i)", j,i, | ||
407 : | pMBs[j + i * iWcount].mode, | ||
408 : | pMBs[j + i * iWcount].dquant, | ||
409 : | pMBs[j + i * iWcount].mvs[0], | ||
410 : | pMBs[j + i * iWcount].mvs[1], | ||
411 : | pMBs[j + i * iWcount].mvs[2], | ||
412 : | pMBs[j + i * iWcount].mvs[3], | ||
413 : | pMBs[j + i * iWcount].sad8[0], | ||
414 : | pMBs[j + i * iWcount].sad8[1], | ||
415 : | pMBs[j + i * iWcount].sad8[2], | ||
416 : | pMBs[j + i * iWcount].sad8[3], | ||
417 : | pMBs[j + i * iWcount].sad16); | ||
418 : | } | ||
419 : | */ | ||
420 : | |||
421 : | Isibaar | 3 | return 0; |
422 : | } | ||
423 : | |||
424 : | #define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) | ||
425 : | |||
426 : | #define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) | ||
427 : | |||
428 : | |||
429 : | #define CHECK_MV16_ZERO {\ | ||
430 : | if ( (0 <= max_dx) && (0 >= min_dx) \ | ||
431 : | && (0 <= max_dy) && (0 >= min_dy) ) \ | ||
432 : | { \ | ||
433 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ | ||
434 : | iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
435 : | if (iSAD < iMinSAD) \ | ||
436 : | { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ | ||
437 : | } | ||
438 : | |||
439 : | chl | 96 | #define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
440 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
441 : | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
442 : | if (iSAD < iMinSAD) \ | ||
443 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ | ||
444 : | } | ||
445 : | Isibaar | 3 | |
446 : | #define CHECK_MV16_CANDIDATE(X,Y) { \ | ||
447 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
448 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
449 : | { \ | ||
450 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
451 : | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
452 : | if (iSAD < iMinSAD) \ | ||
453 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
454 : | } | ||
455 : | |||
456 : | #define CHECK_MV16_CANDIDATE_DIR(X,Y,D) { \ | ||
457 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
458 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
459 : | { \ | ||
460 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
461 : | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
462 : | if (iSAD < iMinSAD) \ | ||
463 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
464 : | } | ||
465 : | |||
466 : | #define CHECK_MV16_CANDIDATE_FOUND(X,Y,D) { \ | ||
467 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
468 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
469 : | { \ | ||
470 : | iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ | ||
471 : | iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
472 : | if (iSAD < iMinSAD) \ | ||
473 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
474 : | } | ||
475 : | |||
476 : | |||
477 : | #define CHECK_MV8_ZERO {\ | ||
478 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ | ||
479 : | iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
480 : | if (iSAD < iMinSAD) \ | ||
481 : | { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ | ||
482 : | } | ||
483 : | |||
484 : | chl | 96 | #define NOCHECK_MV8_CANDIDATE(X,Y) \ |
485 : | { \ | ||
486 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
487 : | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
488 : | if (iSAD < iMinSAD) \ | ||
489 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ | ||
490 : | } | ||
491 : | Isibaar | 3 | |
492 : | #define CHECK_MV8_CANDIDATE(X,Y) { \ | ||
493 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
494 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
495 : | { \ | ||
496 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
497 : | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
498 : | if (iSAD < iMinSAD) \ | ||
499 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ | ||
500 : | } | ||
501 : | |||
502 : | #define CHECK_MV8_CANDIDATE_DIR(X,Y,D) { \ | ||
503 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
504 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
505 : | { \ | ||
506 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
507 : | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
508 : | if (iSAD < iMinSAD) \ | ||
509 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ | ||
510 : | } | ||
511 : | |||
512 : | #define CHECK_MV8_CANDIDATE_FOUND(X,Y,D) { \ | ||
513 : | if ( ((X) <= max_dx) && ((X) >= min_dx) \ | ||
514 : | && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ | ||
515 : | { \ | ||
516 : | iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ | ||
517 : | iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ | ||
518 : | if (iSAD < iMinSAD) \ | ||
519 : | { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ | ||
520 : | } | ||
521 : | |||
522 : | /* too slow and not fully functional at the moment */ | ||
523 : | /* | ||
524 : | int32_t ZeroSearch16( | ||
525 : | const uint8_t * const pRef, | ||
526 : | const uint8_t * const pRefH, | ||
527 : | const uint8_t * const pRefV, | ||
528 : | const uint8_t * const pRefHV, | ||
529 : | const IMAGE * const pCur, | ||
530 : | const int x, const int y, | ||
531 : | const uint32_t MotionFlags, | ||
532 : | suxen_drol | 136 | const uint32_t iQuant, |
533 : | const uint32_t iFcode, | ||
534 : | Isibaar | 3 | MBParam * const pParam, |
535 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
536 : | const MACROBLOCK * const prevMBs, | ||
537 : | Isibaar | 3 | VECTOR * const currMV, |
538 : | VECTOR * const currPMV) | ||
539 : | { | ||
540 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
541 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
542 : | int32_t iSAD; | ||
543 : | int32_t pred_x,pred_y; | ||
544 : | |||
545 : | get_pmv(pMBs, x, y, pParam->mb_width, 0, &pred_x, &pred_y); | ||
546 : | |||
547 : | iSAD = sad16( cur, | ||
548 : | get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), | ||
549 : | iEdgedWidth, MV_MAX_ERROR); | ||
550 : | if (iSAD <= iQuant * 96) | ||
551 : | iSAD -= MV16_00_BIAS; | ||
552 : | |||
553 : | currMV->x = 0; | ||
554 : | currMV->y = 0; | ||
555 : | currPMV->x = -pred_x; | ||
556 : | currPMV->y = -pred_y; | ||
557 : | |||
558 : | return iSAD; | ||
559 : | |||
560 : | } | ||
561 : | */ | ||
562 : | |||
563 : | chl | 96 | int32_t Diamond16_MainSearch( |
564 : | edgomez | 78 | const uint8_t * const pRef, |
565 : | const uint8_t * const pRefH, | ||
566 : | const uint8_t * const pRefV, | ||
567 : | const uint8_t * const pRefHV, | ||
568 : | const uint8_t * const cur, | ||
569 : | const int x, const int y, | ||
570 : | int32_t startx, int32_t starty, | ||
571 : | int32_t iMinSAD, | ||
572 : | VECTOR * const currMV, | ||
573 : | const VECTOR * const pmv, | ||
574 : | const int32_t min_dx, const int32_t max_dx, | ||
575 : | const int32_t min_dy, const int32_t max_dy, | ||
576 : | const int32_t iEdgedWidth, | ||
577 : | const int32_t iDiamondSize, | ||
578 : | const int32_t iFcode, | ||
579 : | const int32_t iQuant, | ||
580 : | int iFound) | ||
581 : | Isibaar | 3 | { |
582 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
583 : | |||
584 : | int32_t iDirection=0; | ||
585 : | int32_t iSAD; | ||
586 : | VECTOR backupMV; | ||
587 : | backupMV.x = startx; | ||
588 : | backupMV.y = starty; | ||
589 : | |||
590 : | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ | ||
591 : | |||
592 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
593 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
594 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
595 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
596 : | |||
597 : | if (iDirection) | ||
598 : | while (!iFound) | ||
599 : | { | ||
600 : | iFound = 1; | ||
601 : | backupMV=*currMV; | ||
602 : | |||
603 : | if ( iDirection != 2) | ||
604 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
605 : | if ( iDirection != 1) | ||
606 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); | ||
607 : | if ( iDirection != 4) | ||
608 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); | ||
609 : | if ( iDirection != 3) | ||
610 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); | ||
611 : | } | ||
612 : | else | ||
613 : | edgomez | 78 | { |
614 : | currMV->x = startx; | ||
615 : | currMV->y = starty; | ||
616 : | } | ||
617 : | Isibaar | 3 | return iMinSAD; |
618 : | } | ||
619 : | |||
620 : | chl | 96 | int32_t Square16_MainSearch( |
621 : | const uint8_t * const pRef, | ||
622 : | const uint8_t * const pRefH, | ||
623 : | const uint8_t * const pRefV, | ||
624 : | const uint8_t * const pRefHV, | ||
625 : | const uint8_t * const cur, | ||
626 : | const int x, const int y, | ||
627 : | int32_t startx, int32_t starty, | ||
628 : | int32_t iMinSAD, | ||
629 : | VECTOR * const currMV, | ||
630 : | const VECTOR * const pmv, | ||
631 : | const int32_t min_dx, const int32_t max_dx, | ||
632 : | const int32_t min_dy, const int32_t max_dy, | ||
633 : | const int32_t iEdgedWidth, | ||
634 : | const int32_t iDiamondSize, | ||
635 : | const int32_t iFcode, | ||
636 : | const int32_t iQuant, | ||
637 : | int iFound) | ||
638 : | { | ||
639 : | /* Do a square search around given starting point, return SAD of best */ | ||
640 : | |||
641 : | int32_t iDirection=0; | ||
642 : | int32_t iSAD; | ||
643 : | VECTOR backupMV; | ||
644 : | backupMV.x = startx; | ||
645 : | backupMV.y = starty; | ||
646 : | |||
647 : | /* It's one search with full square pattern, and new parts for all following diamonds */ | ||
648 : | |||
649 : | /* new direction are extra, so 1-4 is normal diamond | ||
650 : | 537 | ||
651 : | 1*2 | ||
652 : | 648 | ||
653 : | */ | ||
654 : | |||
655 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
656 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
657 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
658 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
659 : | |||
660 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
661 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
662 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
663 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
664 : | |||
665 : | |||
666 : | if (iDirection) | ||
667 : | while (!iFound) | ||
668 : | { | ||
669 : | iFound = 1; | ||
670 : | backupMV=*currMV; | ||
671 : | |||
672 : | switch (iDirection) | ||
673 : | { | ||
674 : | case 1: | ||
675 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
676 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
677 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
678 : | break; | ||
679 : | case 2: | ||
680 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
681 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
682 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
683 : | break; | ||
684 : | |||
685 : | case 3: | ||
686 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
687 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
688 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
689 : | break; | ||
690 : | |||
691 : | case 4: | ||
692 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
693 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
694 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
695 : | break; | ||
696 : | |||
697 : | case 5: | ||
698 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
699 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
700 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
701 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
702 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
703 : | break; | ||
704 : | |||
705 : | case 6: | ||
706 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
707 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
708 : | |||
709 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
710 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
711 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
712 : | |||
713 : | break; | ||
714 : | |||
715 : | case 7: | ||
716 : | CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
717 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
718 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
719 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
720 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
721 : | break; | ||
722 : | |||
723 : | case 8: | ||
724 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
725 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
726 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
727 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
728 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
729 : | break; | ||
730 : | default: | ||
731 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
732 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
733 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
734 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
735 : | |||
736 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); | ||
737 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); | ||
738 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); | ||
739 : | CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); | ||
740 : | break; | ||
741 : | } | ||
742 : | } | ||
743 : | else | ||
744 : | { | ||
745 : | currMV->x = startx; | ||
746 : | currMV->y = starty; | ||
747 : | } | ||
748 : | return iMinSAD; | ||
749 : | } | ||
750 : | |||
751 : | |||
752 : | int32_t Full16_MainSearch( | ||
753 : | const uint8_t * const pRef, | ||
754 : | const uint8_t * const pRefH, | ||
755 : | const uint8_t * const pRefV, | ||
756 : | const uint8_t * const pRefHV, | ||
757 : | const uint8_t * const cur, | ||
758 : | const int x, const int y, | ||
759 : | int32_t startx, int32_t starty, | ||
760 : | int32_t iMinSAD, | ||
761 : | VECTOR * const currMV, | ||
762 : | const VECTOR * const pmv, | ||
763 : | const int32_t min_dx, const int32_t max_dx, | ||
764 : | const int32_t min_dy, const int32_t max_dy, | ||
765 : | const int32_t iEdgedWidth, | ||
766 : | const int32_t iDiamondSize, | ||
767 : | const int32_t iFcode, | ||
768 : | const int32_t iQuant, | ||
769 : | int iFound) | ||
770 : | { | ||
771 : | int32_t iSAD; | ||
772 : | int32_t dx,dy; | ||
773 : | VECTOR backupMV; | ||
774 : | backupMV.x = startx; | ||
775 : | backupMV.y = starty; | ||
776 : | |||
777 : | for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) | ||
778 : | for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) | ||
779 : | NOCHECK_MV16_CANDIDATE(dx,dy); | ||
780 : | |||
781 : | return iMinSAD; | ||
782 : | } | ||
783 : | |||
784 : | int32_t Full8_MainSearch( | ||
785 : | const uint8_t * const pRef, | ||
786 : | const uint8_t * const pRefH, | ||
787 : | const uint8_t * const pRefV, | ||
788 : | const uint8_t * const pRefHV, | ||
789 : | const uint8_t * const cur, | ||
790 : | const int x, const int y, | ||
791 : | int32_t startx, int32_t starty, | ||
792 : | int32_t iMinSAD, | ||
793 : | VECTOR * const currMV, | ||
794 : | const VECTOR * const pmv, | ||
795 : | const int32_t min_dx, const int32_t max_dx, | ||
796 : | const int32_t min_dy, const int32_t max_dy, | ||
797 : | const int32_t iEdgedWidth, | ||
798 : | const int32_t iDiamondSize, | ||
799 : | const int32_t iFcode, | ||
800 : | const int32_t iQuant, | ||
801 : | int iFound) | ||
802 : | { | ||
803 : | int32_t iSAD; | ||
804 : | int32_t dx,dy; | ||
805 : | VECTOR backupMV; | ||
806 : | backupMV.x = startx; | ||
807 : | backupMV.y = starty; | ||
808 : | |||
809 : | for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) | ||
810 : | for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) | ||
811 : | NOCHECK_MV8_CANDIDATE(dx,dy); | ||
812 : | |||
813 : | return iMinSAD; | ||
814 : | } | ||
815 : | |||
816 : | |||
817 : | |||
818 : | int32_t Halfpel16_Refine( | ||
819 : | edgomez | 78 | const uint8_t * const pRef, |
820 : | const uint8_t * const pRefH, | ||
821 : | const uint8_t * const pRefV, | ||
822 : | const uint8_t * const pRefHV, | ||
823 : | const uint8_t * const cur, | ||
824 : | const int x, const int y, | ||
825 : | VECTOR * const currMV, | ||
826 : | int32_t iMinSAD, | ||
827 : | const VECTOR * const pmv, | ||
828 : | const int32_t min_dx, const int32_t max_dx, | ||
829 : | const int32_t min_dy, const int32_t max_dy, | ||
830 : | const int32_t iFcode, | ||
831 : | const int32_t iQuant, | ||
832 : | const int32_t iEdgedWidth) | ||
833 : | Isibaar | 3 | { |
834 : | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ | ||
835 : | |||
836 : | int32_t iSAD; | ||
837 : | VECTOR backupMV = *currMV; | ||
838 : | |||
839 : | CHECK_MV16_CANDIDATE(backupMV.x-1,backupMV.y-1); | ||
840 : | CHECK_MV16_CANDIDATE(backupMV.x ,backupMV.y-1); | ||
841 : | CHECK_MV16_CANDIDATE(backupMV.x+1,backupMV.y-1); | ||
842 : | CHECK_MV16_CANDIDATE(backupMV.x-1,backupMV.y); | ||
843 : | CHECK_MV16_CANDIDATE(backupMV.x+1,backupMV.y); | ||
844 : | CHECK_MV16_CANDIDATE(backupMV.x-1,backupMV.y+1); | ||
845 : | CHECK_MV16_CANDIDATE(backupMV.x ,backupMV.y+1); | ||
846 : | CHECK_MV16_CANDIDATE(backupMV.x+1,backupMV.y+1); | ||
847 : | |||
848 : | return iMinSAD; | ||
849 : | } | ||
850 : | |||
851 : | #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) | ||
852 : | |||
853 : | chl | 96 | |
854 : | Isibaar | 3 | int32_t PMVfastSearch16( |
855 : | chl | 96 | const uint8_t * const pRef, |
856 : | const uint8_t * const pRefH, | ||
857 : | const uint8_t * const pRefV, | ||
858 : | const uint8_t * const pRefHV, | ||
859 : | const IMAGE * const pCur, | ||
860 : | const int x, const int y, | ||
861 : | const uint32_t MotionFlags, | ||
862 : | suxen_drol | 136 | const uint32_t iQuant, |
863 : | const uint32_t iFcode, | ||
864 : | chl | 96 | const MBParam * const pParam, |
865 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
866 : | const MACROBLOCK * const prevMBs, | ||
867 : | chl | 96 | VECTOR * const currMV, |
868 : | VECTOR * const currPMV) | ||
869 : | Isibaar | 3 | { |
870 : | chenm001 | 132 | const uint32_t iWcount = pParam->mb_width; |
871 : | Isibaar | 3 | const int32_t iWidth = pParam->width; |
872 : | const int32_t iHeight = pParam->height; | ||
873 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
874 : | |||
875 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
876 : | |||
877 : | int32_t iDiamondSize; | ||
878 : | |||
879 : | int32_t min_dx; | ||
880 : | int32_t max_dx; | ||
881 : | int32_t min_dy; | ||
882 : | int32_t max_dy; | ||
883 : | |||
884 : | int32_t iFound; | ||
885 : | |||
886 : | VECTOR newMV; | ||
887 : | VECTOR backupMV; /* just for PMVFAST */ | ||
888 : | |||
889 : | VECTOR pmv[4]; | ||
890 : | int32_t psad[4]; | ||
891 : | |||
892 : | suxen_drol | 136 | const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
893 : | const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; | ||
894 : | Isibaar | 3 | |
895 : | static int32_t threshA,threshB; | ||
896 : | int32_t bPredEq; | ||
897 : | int32_t iMinSAD,iSAD; | ||
898 : | |||
899 : | /* Get maximum range */ | ||
900 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, | ||
901 : | edgomez | 78 | x, y, 16, iWidth, iHeight, iFcode); |
902 : | Isibaar | 3 | |
903 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
904 : | |||
905 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
906 : | { min_dx = EVEN(min_dx); | ||
907 : | edgomez | 78 | max_dx = EVEN(max_dx); |
908 : | min_dy = EVEN(min_dy); | ||
909 : | max_dy = EVEN(max_dy); | ||
910 : | Isibaar | 3 | } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
911 : | |||
912 : | |||
913 : | bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); | ||
914 : | |||
915 : | if ((x==0) && (y==0) ) | ||
916 : | { | ||
917 : | threshA = 512; | ||
918 : | threshB = 1024; | ||
919 : | |||
920 : | } | ||
921 : | else | ||
922 : | { | ||
923 : | threshA = psad[0]; | ||
924 : | threshB = threshA+256; | ||
925 : | if (threshA< 512) threshA = 512; | ||
926 : | if (threshA>1024) threshA = 1024; | ||
927 : | if (threshB>1792) threshB = 1792; | ||
928 : | } | ||
929 : | |||
930 : | iFound=0; | ||
931 : | |||
932 : | /* Step 2: Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion | ||
933 : | edgomez | 78 | vector of the median. |
934 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
935 : | Isibaar | 3 | */ |
936 : | |||
937 : | suxen_drol | 136 | if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[0]) ) ) |
938 : | Isibaar | 3 | iFound=2; |
939 : | |||
940 : | /* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
941 : | edgomez | 78 | Otherwise select large Diamond Search. |
942 : | Isibaar | 3 | */ |
943 : | |||
944 : | if ( (pmv[0].x != 0) || (pmv[0].y != 0) || (threshB<1536) || (bPredEq) ) | ||
945 : | iDiamondSize=1; // halfpel! | ||
946 : | else | ||
947 : | iDiamondSize=2; // halfpel! | ||
948 : | |||
949 : | if (!(MotionFlags & PMV_HALFPELDIAMOND16) ) | ||
950 : | iDiamondSize*=2; | ||
951 : | |||
952 : | /* Step 4: Calculate SAD around the Median prediction. | ||
953 : | edgomez | 78 | MinSAD=SAD |
954 : | If Motion Vector equal to Previous frame motion vector | ||
955 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
956 : | If SAD<=256 goto Step 10. | ||
957 : | Isibaar | 3 | */ |
958 : | |||
959 : | |||
960 : | // Prepare for main loop | ||
961 : | |||
962 : | *currMV=pmv[0]; /* current best := prediction */ | ||
963 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
964 : | { /* This should NOT be necessary! */ | ||
965 : | currMV->x = EVEN(currMV->x); | ||
966 : | currMV->y = EVEN(currMV->y); | ||
967 : | } | ||
968 : | |||
969 : | if (currMV->x > max_dx) | ||
970 : | edgomez | 78 | { |
971 : | currMV->x=max_dx; | ||
972 : | } | ||
973 : | Isibaar | 3 | if (currMV->x < min_dx) |
974 : | edgomez | 78 | { |
975 : | currMV->x=min_dx; | ||
976 : | } | ||
977 : | Isibaar | 3 | if (currMV->y > max_dy) |
978 : | edgomez | 78 | { |
979 : | currMV->y=max_dy; | ||
980 : | } | ||
981 : | Isibaar | 3 | if (currMV->y < min_dy) |
982 : | edgomez | 78 | { |
983 : | currMV->y=min_dy; | ||
984 : | } | ||
985 : | Isibaar | 3 | |
986 : | iMinSAD = sad16( cur, | ||
987 : | edgomez | 78 | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
988 : | iEdgedWidth, MV_MAX_ERROR); | ||
989 : | Isibaar | 3 | iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; |
990 : | |||
991 : | suxen_drol | 136 | if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
992 : | edgomez | 78 | { |
993 : | Isibaar | 3 | |
994 : | edgomez | 78 | if (MotionFlags & PMV_QUICKSTOP16) |
995 : | chl | 96 | goto PMVfast16_Terminate_without_Refine; |
996 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
997 : | goto PMVfast16_Terminate_with_Refine; | ||
998 : | edgomez | 78 | } |
999 : | Isibaar | 3 | |
1000 : | /* | ||
1001 : | edgomez | 78 | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1002 : | Also calculate (0,0) but do not subtract offset. | ||
1003 : | Let MinSAD be the smallest SAD up to this point. | ||
1004 : | chl | 140 | If MV is (0,0) subtract offset. |
1005 : | Isibaar | 3 | */ |
1006 : | |||
1007 : | // (0,0) is always possible | ||
1008 : | |||
1009 : | CHECK_MV16_ZERO; | ||
1010 : | |||
1011 : | // previous frame MV is always possible | ||
1012 : | suxen_drol | 136 | CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1013 : | Isibaar | 3 | |
1014 : | // left neighbour, if allowed | ||
1015 : | if (x != 0) | ||
1016 : | { | ||
1017 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1018 : | { pmv[1].x = EVEN(pmv[1].x); | ||
1019 : | edgomez | 78 | pmv[1].y = EVEN(pmv[1].y); |
1020 : | Isibaar | 3 | } |
1021 : | CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); | ||
1022 : | } | ||
1023 : | |||
1024 : | // top neighbour, if allowed | ||
1025 : | if (y != 0) | ||
1026 : | { | ||
1027 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1028 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1029 : | edgomez | 78 | pmv[2].y = EVEN(pmv[2].y); |
1030 : | Isibaar | 3 | } |
1031 : | CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1032 : | |||
1033 : | // top right neighbour, if allowed | ||
1034 : | chenm001 | 132 | if ((uint32_t)x != (iWcount-1)) |
1035 : | Isibaar | 3 | { |
1036 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1037 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1038 : | edgomez | 78 | pmv[3].y = EVEN(pmv[3].y); |
1039 : | Isibaar | 3 | } |
1040 : | CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); | ||
1041 : | } | ||
1042 : | } | ||
1043 : | chl | 140 | |
1044 : | if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) && (iSAD <= iQuant * 96) ) | ||
1045 : | iMinSAD -= MV16_00_BIAS; | ||
1046 : | Isibaar | 3 | |
1047 : | chl | 140 | |
1048 : | Isibaar | 3 | /* Step 6: If MinSAD <= thresa goto Step 10. |
1049 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
1050 : | */ | ||
1051 : | |||
1052 : | suxen_drol | 136 | if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
1053 : | edgomez | 78 | { |
1054 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1055 : | chl | 96 | goto PMVfast16_Terminate_without_Refine; |
1056 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1057 : | goto PMVfast16_Terminate_with_Refine; | ||
1058 : | edgomez | 78 | } |
1059 : | Isibaar | 3 | |
1060 : | |||
1061 : | /************ (Diamond Search) **************/ | ||
1062 : | /* | ||
1063 : | edgomez | 78 | Step 7: Perform Diamond search, with either the small or large diamond. |
1064 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
1065 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
1066 : | If center then goto step 10. | ||
1067 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
1068 : | Refine by using small diamond and goto step 10. | ||
1069 : | Isibaar | 3 | */ |
1070 : | |||
1071 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1072 : | |||
1073 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1074 : | chl | 96 | iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1075 : | edgomez | 78 | x, y, |
1076 : | currMV->x, currMV->y, iMinSAD, &newMV, | ||
1077 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1078 : | Isibaar | 3 | |
1079 : | if (iSAD < iMinSAD) | ||
1080 : | { | ||
1081 : | *currMV = newMV; | ||
1082 : | iMinSAD = iSAD; | ||
1083 : | } | ||
1084 : | |||
1085 : | if (MotionFlags & PMV_EXTSEARCH16) | ||
1086 : | { | ||
1087 : | /* extended: search (up to) two more times: orignal prediction and (0,0) */ | ||
1088 : | |||
1089 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1090 : | chl | 96 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1091 : | edgomez | 78 | x, y, |
1092 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1093 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1094 : | Isibaar | 3 | |
1095 : | edgomez | 78 | if (iSAD < iMinSAD) |
1096 : | { | ||
1097 : | *currMV = newMV; | ||
1098 : | iMinSAD = iSAD; | ||
1099 : | Isibaar | 3 | } |
1100 : | edgomez | 78 | } |
1101 : | Isibaar | 3 | |
1102 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1103 : | chl | 96 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1104 : | edgomez | 78 | x, y, |
1105 : | 0, 0, iMinSAD, &newMV, | ||
1106 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1107 : | Isibaar | 3 | |
1108 : | edgomez | 78 | if (iSAD < iMinSAD) |
1109 : | { | ||
1110 : | *currMV = newMV; | ||
1111 : | iMinSAD = iSAD; | ||
1112 : | Isibaar | 3 | } |
1113 : | edgomez | 78 | } |
1114 : | Isibaar | 3 | } |
1115 : | |||
1116 : | /* | ||
1117 : | edgomez | 78 | Step 10: The motion vector is chosen according to the block corresponding to MinSAD. |
1118 : | Isibaar | 3 | */ |
1119 : | |||
1120 : | chl | 96 | PMVfast16_Terminate_with_Refine: |
1121 : | Isibaar | 3 | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
1122 : | chl | 96 | iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
1123 : | x, y, | ||
1124 : | currMV, iMinSAD, | ||
1125 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1126 : | Isibaar | 3 | |
1127 : | chl | 96 | PMVfast16_Terminate_without_Refine: |
1128 : | Isibaar | 3 | currPMV->x = currMV->x - pmv[0].x; |
1129 : | currPMV->y = currMV->y - pmv[0].y; | ||
1130 : | return iMinSAD; | ||
1131 : | } | ||
1132 : | |||
1133 : | |||
1134 : | |||
1135 : | |||
1136 : | |||
1137 : | |||
1138 : | chl | 96 | int32_t Diamond8_MainSearch( |
1139 : | edgomez | 78 | const uint8_t * const pRef, |
1140 : | const uint8_t * const pRefH, | ||
1141 : | const uint8_t * const pRefV, | ||
1142 : | const uint8_t * const pRefHV, | ||
1143 : | const uint8_t * const cur, | ||
1144 : | const int x, const int y, | ||
1145 : | int32_t startx, int32_t starty, | ||
1146 : | int32_t iMinSAD, | ||
1147 : | VECTOR * const currMV, | ||
1148 : | const VECTOR * const pmv, | ||
1149 : | const int32_t min_dx, const int32_t max_dx, | ||
1150 : | const int32_t min_dy, const int32_t max_dy, | ||
1151 : | const int32_t iEdgedWidth, | ||
1152 : | const int32_t iDiamondSize, | ||
1153 : | const int32_t iFcode, | ||
1154 : | const int32_t iQuant, | ||
1155 : | int iFound) | ||
1156 : | Isibaar | 3 | { |
1157 : | /* Do a diamond search around given starting point, return SAD of best */ | ||
1158 : | |||
1159 : | int32_t iDirection=0; | ||
1160 : | int32_t iSAD; | ||
1161 : | VECTOR backupMV; | ||
1162 : | backupMV.x = startx; | ||
1163 : | backupMV.y = starty; | ||
1164 : | |||
1165 : | /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ | ||
1166 : | |||
1167 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); | ||
1168 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); | ||
1169 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); | ||
1170 : | CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); | ||
1171 : | |||
1172 : | if (iDirection) | ||
1173 : | while (!iFound) | ||
1174 : | { | ||
1175 : | iFound = 1; | ||
1176 : | backupMV=*currMV; // since iDirection!=0, this is well defined! | ||
1177 : | |||
1178 : | if ( iDirection != 2) | ||
1179 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); | ||
1180 : | if ( iDirection != 1) | ||
1181 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); | ||
1182 : | if ( iDirection != 4) | ||
1183 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); | ||
1184 : | if ( iDirection != 3) | ||
1185 : | CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); | ||
1186 : | } | ||
1187 : | else | ||
1188 : | edgomez | 78 | { |
1189 : | currMV->x = startx; | ||
1190 : | currMV->y = starty; | ||
1191 : | } | ||
1192 : | Isibaar | 3 | return iMinSAD; |
1193 : | } | ||
1194 : | |||
1195 : | chl | 96 | int32_t Halfpel8_Refine( |
1196 : | edgomez | 78 | const uint8_t * const pRef, |
1197 : | const uint8_t * const pRefH, | ||
1198 : | const uint8_t * const pRefV, | ||
1199 : | const uint8_t * const pRefHV, | ||
1200 : | const uint8_t * const cur, | ||
1201 : | const int x, const int y, | ||
1202 : | VECTOR * const currMV, | ||
1203 : | int32_t iMinSAD, | ||
1204 : | const VECTOR * const pmv, | ||
1205 : | const int32_t min_dx, const int32_t max_dx, | ||
1206 : | const int32_t min_dy, const int32_t max_dy, | ||
1207 : | const int32_t iFcode, | ||
1208 : | const int32_t iQuant, | ||
1209 : | const int32_t iEdgedWidth) | ||
1210 : | Isibaar | 3 | { |
1211 : | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ | ||
1212 : | |||
1213 : | int32_t iSAD; | ||
1214 : | VECTOR backupMV = *currMV; | ||
1215 : | |||
1216 : | CHECK_MV8_CANDIDATE(backupMV.x-1,backupMV.y-1); | ||
1217 : | CHECK_MV8_CANDIDATE(backupMV.x ,backupMV.y-1); | ||
1218 : | CHECK_MV8_CANDIDATE(backupMV.x+1,backupMV.y-1); | ||
1219 : | CHECK_MV8_CANDIDATE(backupMV.x-1,backupMV.y); | ||
1220 : | CHECK_MV8_CANDIDATE(backupMV.x+1,backupMV.y); | ||
1221 : | CHECK_MV8_CANDIDATE(backupMV.x-1,backupMV.y+1); | ||
1222 : | CHECK_MV8_CANDIDATE(backupMV.x ,backupMV.y+1); | ||
1223 : | CHECK_MV8_CANDIDATE(backupMV.x+1,backupMV.y+1); | ||
1224 : | |||
1225 : | return iMinSAD; | ||
1226 : | } | ||
1227 : | |||
1228 : | |||
1229 : | #define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) | ||
1230 : | |||
1231 : | int32_t PMVfastSearch8( | ||
1232 : | chl | 96 | 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 IMAGE * const pCur, | ||
1237 : | const int x, const int y, | ||
1238 : | suxen_drol | 136 | const int start_x, const int start_y, |
1239 : | chl | 96 | const uint32_t MotionFlags, |
1240 : | suxen_drol | 136 | const uint32_t iQuant, |
1241 : | const uint32_t iFcode, | ||
1242 : | chl | 96 | const MBParam * const pParam, |
1243 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
1244 : | const MACROBLOCK * const prevMBs, | ||
1245 : | chl | 96 | VECTOR * const currMV, |
1246 : | VECTOR * const currPMV) | ||
1247 : | Isibaar | 3 | { |
1248 : | suxen_drol | 136 | const uint32_t iWcount = pParam->mb_width; |
1249 : | Isibaar | 3 | const int32_t iWidth = pParam->width; |
1250 : | const int32_t iHeight = pParam->height; | ||
1251 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
1252 : | |||
1253 : | const uint8_t * cur = pCur->y + x*8 + y*8*iEdgedWidth; | ||
1254 : | |||
1255 : | int32_t iDiamondSize; | ||
1256 : | |||
1257 : | int32_t min_dx; | ||
1258 : | int32_t max_dx; | ||
1259 : | int32_t min_dy; | ||
1260 : | int32_t max_dy; | ||
1261 : | |||
1262 : | VECTOR pmv[4]; | ||
1263 : | int32_t psad[4]; | ||
1264 : | VECTOR newMV; | ||
1265 : | VECTOR backupMV; | ||
1266 : | |||
1267 : | suxen_drol | 136 | const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1268 : | const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; | ||
1269 : | Isibaar | 3 | |
1270 : | static int32_t threshA,threshB; | ||
1271 : | int32_t iFound,bPredEq; | ||
1272 : | int32_t iMinSAD,iSAD; | ||
1273 : | |||
1274 : | int32_t iSubBlock = ((y&1)<<1) + (x&1); | ||
1275 : | |||
1276 : | /* Get maximum range */ | ||
1277 : | edgomez | 78 | get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1278 : | x, y, 8, iWidth, iHeight, iFcode); | ||
1279 : | Isibaar | 3 | |
1280 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) | ||
1281 : | { min_dx = EVEN(min_dx); | ||
1282 : | edgomez | 78 | max_dx = EVEN(max_dx); |
1283 : | min_dy = EVEN(min_dy); | ||
1284 : | max_dy = EVEN(max_dy); | ||
1285 : | Isibaar | 3 | } /* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
1286 : | |||
1287 : | |||
1288 : | bPredEq = get_pmvdata(pMBs, (x>>1), (y>>1), iWcount, iSubBlock, pmv, psad); | ||
1289 : | |||
1290 : | if ((x==0) && (y==0) ) | ||
1291 : | { | ||
1292 : | threshA = 512/4; | ||
1293 : | threshB = 1024/4; | ||
1294 : | |||
1295 : | } | ||
1296 : | else | ||
1297 : | { | ||
1298 : | threshA = psad[0]/4; /* good estimate */ | ||
1299 : | threshB = threshA+256/4; | ||
1300 : | if (threshA< 512/4) threshA = 512/4; | ||
1301 : | if (threshA>1024/4) threshA = 1024/4; | ||
1302 : | if (threshB>1792/4) threshB = 1792/4; | ||
1303 : | } | ||
1304 : | |||
1305 : | iFound=0; | ||
1306 : | |||
1307 : | /* Step 2: Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion | ||
1308 : | edgomez | 78 | vector of the median. |
1309 : | If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 | ||
1310 : | Isibaar | 3 | */ |
1311 : | |||
1312 : | chl | 140 | if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[iSubBlock]) ) ) |
1313 : | Isibaar | 3 | iFound=2; |
1314 : | |||
1315 : | /* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. | ||
1316 : | edgomez | 78 | Otherwise select large Diamond Search. |
1317 : | Isibaar | 3 | */ |
1318 : | |||
1319 : | if ( (pmv[0].x != 0) || (pmv[0].y != 0) || (threshB<1536/4) || (bPredEq) ) | ||
1320 : | iDiamondSize=1; // 1 halfpel! | ||
1321 : | else | ||
1322 : | iDiamondSize=2; // 2 halfpel = 1 full pixel! | ||
1323 : | |||
1324 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8) ) | ||
1325 : | iDiamondSize*=2; | ||
1326 : | |||
1327 : | /* Step 4: Calculate SAD around the Median prediction. | ||
1328 : | edgomez | 78 | MinSAD=SAD |
1329 : | If Motion Vector equal to Previous frame motion vector | ||
1330 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1331 : | If SAD<=256 goto Step 10. | ||
1332 : | Isibaar | 3 | */ |
1333 : | |||
1334 : | |||
1335 : | // Prepare for main loop | ||
1336 : | |||
1337 : | currMV->x=start_x; /* start with mv16 */ | ||
1338 : | currMV->y=start_y; | ||
1339 : | |||
1340 : | iMinSAD = sad8( cur, | ||
1341 : | edgomez | 78 | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
1342 : | iEdgedWidth); | ||
1343 : | Isibaar | 3 | iMinSAD += calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, (uint8_t)iFcode) * iQuant; |
1344 : | |||
1345 : | chl | 140 | if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,prevMB->mvs[iSubBlock])) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1346 : | edgomez | 78 | { |
1347 : | chl | 96 | if (MotionFlags & PMV_QUICKSTOP16) |
1348 : | goto PMVfast8_Terminate_without_Refine; | ||
1349 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1350 : | goto PMVfast8_Terminate_with_Refine; | ||
1351 : | edgomez | 78 | } |
1352 : | Isibaar | 3 | |
1353 : | chl | 96 | |
1354 : | Isibaar | 3 | /* |
1355 : | edgomez | 78 | Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1356 : | Also calculate (0,0) but do not subtract offset. | ||
1357 : | Let MinSAD be the smallest SAD up to this point. | ||
1358 : | chl | 140 | If MV is (0,0) subtract offset. |
1359 : | Isibaar | 3 | */ |
1360 : | |||
1361 : | // the prediction might be even better than mv16 | ||
1362 : | CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); | ||
1363 : | |||
1364 : | // (0,0) is always possible | ||
1365 : | CHECK_MV8_ZERO; | ||
1366 : | |||
1367 : | // previous frame MV is always possible | ||
1368 : | chl | 140 | CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
1369 : | Isibaar | 3 | |
1370 : | // left neighbour, if allowed | ||
1371 : | if (psad[1] != MV_MAX_ERROR) | ||
1372 : | { | ||
1373 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1374 : | { pmv[1].x = EVEN(pmv[1].x); | ||
1375 : | edgomez | 78 | pmv[1].y = EVEN(pmv[1].y); |
1376 : | Isibaar | 3 | } |
1377 : | CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); | ||
1378 : | } | ||
1379 : | |||
1380 : | // top neighbour, if allowed | ||
1381 : | if (psad[2] != MV_MAX_ERROR) | ||
1382 : | { | ||
1383 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1384 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1385 : | edgomez | 78 | pmv[2].y = EVEN(pmv[2].y); |
1386 : | Isibaar | 3 | } |
1387 : | CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1388 : | |||
1389 : | // top right neighbour, if allowed | ||
1390 : | if (psad[3] != MV_MAX_ERROR) | ||
1391 : | { | ||
1392 : | edgomez | 78 | if (!(MotionFlags & PMV_HALFPEL8 )) |
1393 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1394 : | Isibaar | 3 | pmv[3].y = EVEN(pmv[3].y); |
1395 : | edgomez | 78 | } |
1396 : | Isibaar | 3 | CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
1397 : | } | ||
1398 : | } | ||
1399 : | |||
1400 : | chl | 140 | if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) && (iSAD <= iQuant * 96) ) |
1401 : | iMinSAD -= MV8_00_BIAS; | ||
1402 : | |||
1403 : | |||
1404 : | Isibaar | 3 | /* Step 6: If MinSAD <= thresa goto Step 10. |
1405 : | If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. | ||
1406 : | */ | ||
1407 : | |||
1408 : | chl | 140 | if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1409 : | edgomez | 78 | { |
1410 : | chl | 96 | if (MotionFlags & PMV_QUICKSTOP16) |
1411 : | goto PMVfast8_Terminate_without_Refine; | ||
1412 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1413 : | goto PMVfast8_Terminate_with_Refine; | ||
1414 : | edgomez | 78 | } |
1415 : | Isibaar | 3 | |
1416 : | /************ (Diamond Search) **************/ | ||
1417 : | /* | ||
1418 : | edgomez | 78 | Step 7: Perform Diamond search, with either the small or large diamond. |
1419 : | If Found=2 only examine one Diamond pattern, and afterwards goto step 10 | ||
1420 : | Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. | ||
1421 : | If center then goto step 10. | ||
1422 : | Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. | ||
1423 : | Refine by using small diamond and goto step 10. | ||
1424 : | Isibaar | 3 | */ |
1425 : | |||
1426 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1427 : | |||
1428 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1429 : | chl | 96 | iSAD = Diamond8_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1430 : | edgomez | 78 | x, y, |
1431 : | currMV->x, currMV->y, iMinSAD, &newMV, | ||
1432 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1433 : | Isibaar | 3 | |
1434 : | if (iSAD < iMinSAD) | ||
1435 : | { | ||
1436 : | *currMV = newMV; | ||
1437 : | iMinSAD = iSAD; | ||
1438 : | } | ||
1439 : | |||
1440 : | if (MotionFlags & PMV_EXTSEARCH8) | ||
1441 : | { | ||
1442 : | /* extended: search (up to) two more times: orignal prediction and (0,0) */ | ||
1443 : | |||
1444 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1445 : | chl | 96 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1446 : | edgomez | 78 | x, y, |
1447 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1448 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1449 : | Isibaar | 3 | |
1450 : | edgomez | 78 | if (iSAD < iMinSAD) |
1451 : | { | ||
1452 : | *currMV = newMV; | ||
1453 : | iMinSAD = iSAD; | ||
1454 : | Isibaar | 3 | } |
1455 : | edgomez | 78 | } |
1456 : | Isibaar | 3 | |
1457 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1458 : | chl | 96 | { iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
1459 : | edgomez | 78 | x, y, |
1460 : | 0, 0, iMinSAD, &newMV, | ||
1461 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); | ||
1462 : | Isibaar | 3 | |
1463 : | edgomez | 78 | if (iSAD < iMinSAD) |
1464 : | { | ||
1465 : | *currMV = newMV; | ||
1466 : | iMinSAD = iSAD; | ||
1467 : | Isibaar | 3 | } |
1468 : | edgomez | 78 | } |
1469 : | Isibaar | 3 | } |
1470 : | |||
1471 : | /* Step 10: The motion vector is chosen according to the block corresponding to MinSAD. | ||
1472 : | edgomez | 78 | By performing an optional local half-pixel search, we can refine this result even further. |
1473 : | Isibaar | 3 | */ |
1474 : | |||
1475 : | chl | 96 | PMVfast8_Terminate_with_Refine: |
1476 : | Isibaar | 3 | if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
1477 : | chl | 96 | iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
1478 : | edgomez | 78 | x, y, |
1479 : | currMV, iMinSAD, | ||
1480 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1481 : | Isibaar | 3 | |
1482 : | |||
1483 : | chl | 96 | PMVfast8_Terminate_without_Refine: |
1484 : | Isibaar | 3 | currPMV->x = currMV->x - pmv[0].x; |
1485 : | currPMV->y = currMV->y - pmv[0].y; | ||
1486 : | |||
1487 : | return iMinSAD; | ||
1488 : | } | ||
1489 : | chl | 96 | |
1490 : | int32_t EPZSSearch16( | ||
1491 : | const uint8_t * const pRef, | ||
1492 : | const uint8_t * const pRefH, | ||
1493 : | const uint8_t * const pRefV, | ||
1494 : | const uint8_t * const pRefHV, | ||
1495 : | const IMAGE * const pCur, | ||
1496 : | const int x, const int y, | ||
1497 : | const uint32_t MotionFlags, | ||
1498 : | suxen_drol | 136 | const uint32_t iQuant, |
1499 : | const uint32_t iFcode, | ||
1500 : | chl | 96 | const MBParam * const pParam, |
1501 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
1502 : | const MACROBLOCK * const prevMBs, | ||
1503 : | chl | 96 | VECTOR * const currMV, |
1504 : | VECTOR * const currPMV) | ||
1505 : | { | ||
1506 : | chenm001 | 132 | const uint32_t iWcount = pParam->mb_width; |
1507 : | const uint32_t iHcount = pParam->mb_height; | ||
1508 : | chl | 96 | |
1509 : | const int32_t iWidth = pParam->width; | ||
1510 : | const int32_t iHeight = pParam->height; | ||
1511 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
1512 : | |||
1513 : | const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; | ||
1514 : | |||
1515 : | int32_t min_dx; | ||
1516 : | int32_t max_dx; | ||
1517 : | int32_t min_dy; | ||
1518 : | int32_t max_dy; | ||
1519 : | |||
1520 : | VECTOR newMV; | ||
1521 : | VECTOR backupMV; | ||
1522 : | |||
1523 : | VECTOR pmv[4]; | ||
1524 : | int32_t psad[8]; | ||
1525 : | |||
1526 : | static MACROBLOCK * oldMBs = NULL; | ||
1527 : | suxen_drol | 136 | const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1528 : | const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; | ||
1529 : | chl | 96 | MACROBLOCK * oldMB = NULL; |
1530 : | |||
1531 : | static int32_t thresh2; | ||
1532 : | int32_t bPredEq; | ||
1533 : | int32_t iMinSAD,iSAD=9999; | ||
1534 : | |||
1535 : | MainSearch16FuncPtr EPZSMainSearchPtr; | ||
1536 : | |||
1537 : | if (oldMBs == NULL) | ||
1538 : | { oldMBs = (MACROBLOCK*) calloc(1,iWcount*iHcount*sizeof(MACROBLOCK)); | ||
1539 : | chl | 140 | // fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
1540 : | chl | 96 | } |
1541 : | oldMB = oldMBs + x + y * iWcount; | ||
1542 : | |||
1543 : | /* Get maximum range */ | ||
1544 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, | ||
1545 : | x, y, 16, iWidth, iHeight, iFcode); | ||
1546 : | |||
1547 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1548 : | { min_dx = EVEN(min_dx); | ||
1549 : | max_dx = EVEN(max_dx); | ||
1550 : | min_dy = EVEN(min_dy); | ||
1551 : | max_dy = EVEN(max_dy); | ||
1552 : | } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ | ||
1553 : | |||
1554 : | bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); | ||
1555 : | |||
1556 : | /* Step 4: Calculate SAD around the Median prediction. | ||
1557 : | MinSAD=SAD | ||
1558 : | If Motion Vector equal to Previous frame motion vector | ||
1559 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1560 : | If SAD<=256 goto Step 10. | ||
1561 : | */ | ||
1562 : | |||
1563 : | // Prepare for main loop | ||
1564 : | |||
1565 : | *currMV=pmv[0]; /* current best := median prediction */ | ||
1566 : | if (!(MotionFlags & PMV_HALFPEL16)) | ||
1567 : | { | ||
1568 : | currMV->x = EVEN(currMV->x); | ||
1569 : | currMV->y = EVEN(currMV->y); | ||
1570 : | } | ||
1571 : | |||
1572 : | if (currMV->x > max_dx) | ||
1573 : | currMV->x=max_dx; | ||
1574 : | if (currMV->x < min_dx) | ||
1575 : | currMV->x=min_dx; | ||
1576 : | if (currMV->y > max_dy) | ||
1577 : | currMV->y=max_dy; | ||
1578 : | if (currMV->y < min_dy) | ||
1579 : | currMV->y=min_dy; | ||
1580 : | |||
1581 : | /***************** This is predictor SET A: only median prediction ******************/ | ||
1582 : | |||
1583 : | iMinSAD = sad16( cur, | ||
1584 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), | ||
1585 : | iEdgedWidth, MV_MAX_ERROR); | ||
1586 : | iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; | ||
1587 : | |||
1588 : | // thresh1 is fixed to 256 | ||
1589 : | suxen_drol | 136 | if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,pMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
1590 : | chl | 96 | { |
1591 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1592 : | goto EPZS16_Terminate_without_Refine; | ||
1593 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1594 : | goto EPZS16_Terminate_with_Refine; | ||
1595 : | } | ||
1596 : | |||
1597 : | /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ | ||
1598 : | |||
1599 : | // previous frame MV | ||
1600 : | CHECK_MV16_CANDIDATE(pMB->mvs[0].x,pMB->mvs[0].y); | ||
1601 : | |||
1602 : | // set threshhold based on Min of Prediction and SAD of collocated block | ||
1603 : | // CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want | ||
1604 : | |||
1605 : | if ((x==0) && (y==0) ) | ||
1606 : | { | ||
1607 : | thresh2 = 512; | ||
1608 : | } | ||
1609 : | else | ||
1610 : | { | ||
1611 : | /* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ | ||
1612 : | |||
1613 : | thresh2 = MIN(psad[0],iSAD)*6/5 + 128; | ||
1614 : | } | ||
1615 : | |||
1616 : | // MV=(0,0) is often a good choice | ||
1617 : | |||
1618 : | CHECK_MV16_ZERO; | ||
1619 : | |||
1620 : | |||
1621 : | // left neighbour, if allowed | ||
1622 : | if (x != 0) | ||
1623 : | { | ||
1624 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1625 : | { pmv[1].x = EVEN(pmv[1].x); | ||
1626 : | pmv[1].y = EVEN(pmv[1].y); | ||
1627 : | } | ||
1628 : | CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); | ||
1629 : | } | ||
1630 : | |||
1631 : | // top neighbour, if allowed | ||
1632 : | if (y != 0) | ||
1633 : | { | ||
1634 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1635 : | { pmv[2].x = EVEN(pmv[2].x); | ||
1636 : | pmv[2].y = EVEN(pmv[2].y); | ||
1637 : | } | ||
1638 : | CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); | ||
1639 : | |||
1640 : | // top right neighbour, if allowed | ||
1641 : | chenm001 | 132 | if ((uint32_t)x != (iWcount-1)) |
1642 : | chl | 96 | { |
1643 : | if (!(MotionFlags & PMV_HALFPEL16 )) | ||
1644 : | { pmv[3].x = EVEN(pmv[3].x); | ||
1645 : | pmv[3].y = EVEN(pmv[3].y); | ||
1646 : | } | ||
1647 : | CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); | ||
1648 : | } | ||
1649 : | } | ||
1650 : | |||
1651 : | /* Terminate if MinSAD <= T_2 | ||
1652 : | Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] | ||
1653 : | */ | ||
1654 : | |||
1655 : | if ( (iMinSAD <= thresh2) | ||
1656 : | suxen_drol | 136 | || ( MVequal(*currMV,pMB->mvs[0]) && ((uint32_t)iMinSAD <= prevMB->sad16) ) ) |
1657 : | chl | 96 | { |
1658 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1659 : | goto EPZS16_Terminate_without_Refine; | ||
1660 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1661 : | goto EPZS16_Terminate_with_Refine; | ||
1662 : | } | ||
1663 : | |||
1664 : | /***** predictor SET C: acceleration MV (new!), neighbours in prev. frame(new!) ****/ | ||
1665 : | |||
1666 : | backupMV = pMB->mvs[0]; // last MV | ||
1667 : | backupMV.x += (pMB->mvs[0].x - oldMB->mvs[0].x ); // acceleration X | ||
1668 : | backupMV.y += (pMB->mvs[0].y - oldMB->mvs[0].y ); // acceleration Y | ||
1669 : | |||
1670 : | CHECK_MV16_CANDIDATE(backupMV.x,backupMV.y); | ||
1671 : | |||
1672 : | // left neighbour | ||
1673 : | if (x != 0) | ||
1674 : | CHECK_MV16_CANDIDATE((oldMB-1)->mvs[0].x,oldMB->mvs[0].y); | ||
1675 : | |||
1676 : | // top neighbour | ||
1677 : | if (y != 0) | ||
1678 : | CHECK_MV16_CANDIDATE((oldMB-iWcount)->mvs[0].x,oldMB->mvs[0].y); | ||
1679 : | |||
1680 : | // right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs | ||
1681 : | |||
1682 : | chenm001 | 132 | if ((uint32_t)x != iWcount-1) |
1683 : | chl | 96 | CHECK_MV16_CANDIDATE((pMB+1)->mvs[0].x,oldMB->mvs[0].y); |
1684 : | |||
1685 : | // bottom neighbour, dito | ||
1686 : | chenm001 | 132 | if ((uint32_t)y != iHcount-1) |
1687 : | chl | 96 | CHECK_MV16_CANDIDATE((pMB+iWcount)->mvs[0].x,oldMB->mvs[0].y); |
1688 : | |||
1689 : | /* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ | ||
1690 : | if (iMinSAD <= thresh2) | ||
1691 : | { | ||
1692 : | if (MotionFlags & PMV_QUICKSTOP16) | ||
1693 : | goto EPZS16_Terminate_without_Refine; | ||
1694 : | if (MotionFlags & PMV_EARLYSTOP16) | ||
1695 : | goto EPZS16_Terminate_with_Refine; | ||
1696 : | } | ||
1697 : | |||
1698 : | /************ (if Diamond Search) **************/ | ||
1699 : | |||
1700 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1701 : | |||
1702 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1703 : | |||
1704 : | if (MotionFlags & PMV_USESQUARES16) | ||
1705 : | EPZSMainSearchPtr = Square16_MainSearch; | ||
1706 : | else | ||
1707 : | EPZSMainSearchPtr = Diamond16_MainSearch; | ||
1708 : | |||
1709 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1710 : | x, y, | ||
1711 : | currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, | ||
1712 : | 2, iFcode, iQuant, 0); | ||
1713 : | |||
1714 : | if (iSAD < iMinSAD) | ||
1715 : | { | ||
1716 : | *currMV = newMV; | ||
1717 : | iMinSAD = iSAD; | ||
1718 : | } | ||
1719 : | |||
1720 : | |||
1721 : | if (MotionFlags & PMV_EXTSEARCH16) | ||
1722 : | { | ||
1723 : | /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ | ||
1724 : | |||
1725 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1726 : | { | ||
1727 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1728 : | x, y, | ||
1729 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1730 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); | ||
1731 : | } | ||
1732 : | |||
1733 : | if (iSAD < iMinSAD) | ||
1734 : | { | ||
1735 : | *currMV = newMV; | ||
1736 : | iMinSAD = iSAD; | ||
1737 : | } | ||
1738 : | |||
1739 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1740 : | { | ||
1741 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1742 : | x, y, | ||
1743 : | 0, 0, iMinSAD, &newMV, | ||
1744 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, /*iDiamondSize*/ 2, iFcode, iQuant, 0); | ||
1745 : | |||
1746 : | if (iSAD < iMinSAD) | ||
1747 : | { | ||
1748 : | *currMV = newMV; | ||
1749 : | iMinSAD = iSAD; | ||
1750 : | } | ||
1751 : | } | ||
1752 : | } | ||
1753 : | |||
1754 : | /*************** Choose best MV found **************/ | ||
1755 : | |||
1756 : | EPZS16_Terminate_with_Refine: | ||
1757 : | if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step | ||
1758 : | iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, | ||
1759 : | x, y, | ||
1760 : | currMV, iMinSAD, | ||
1761 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1762 : | |||
1763 : | EPZS16_Terminate_without_Refine: | ||
1764 : | |||
1765 : | *oldMB = *pMB; | ||
1766 : | |||
1767 : | currPMV->x = currMV->x - pmv[0].x; | ||
1768 : | currPMV->y = currMV->y - pmv[0].y; | ||
1769 : | return iMinSAD; | ||
1770 : | } | ||
1771 : | |||
1772 : | |||
1773 : | int32_t EPZSSearch8( | ||
1774 : | const uint8_t * const pRef, | ||
1775 : | const uint8_t * const pRefH, | ||
1776 : | const uint8_t * const pRefV, | ||
1777 : | const uint8_t * const pRefHV, | ||
1778 : | const IMAGE * const pCur, | ||
1779 : | const int x, const int y, | ||
1780 : | const int start_x, const int start_y, | ||
1781 : | const uint32_t MotionFlags, | ||
1782 : | suxen_drol | 136 | const uint32_t iQuant, |
1783 : | const uint32_t iFcode, | ||
1784 : | chl | 96 | const MBParam * const pParam, |
1785 : | suxen_drol | 136 | const MACROBLOCK * const pMBs, |
1786 : | const MACROBLOCK * const prevMBs, | ||
1787 : | chl | 96 | VECTOR * const currMV, |
1788 : | VECTOR * const currPMV) | ||
1789 : | { | ||
1790 : | chenm001 | 132 | const uint32_t iWcount = pParam->mb_width; |
1791 : | chl | 96 | const int32_t iWidth = pParam->width; |
1792 : | const int32_t iHeight = pParam->height; | ||
1793 : | const int32_t iEdgedWidth = pParam->edged_width; | ||
1794 : | |||
1795 : | const uint8_t * cur = pCur->y + x*8 + y*8*iEdgedWidth; | ||
1796 : | |||
1797 : | int32_t iDiamondSize=1; | ||
1798 : | |||
1799 : | int32_t min_dx; | ||
1800 : | int32_t max_dx; | ||
1801 : | int32_t min_dy; | ||
1802 : | int32_t max_dy; | ||
1803 : | |||
1804 : | VECTOR newMV; | ||
1805 : | VECTOR backupMV; | ||
1806 : | |||
1807 : | VECTOR pmv[4]; | ||
1808 : | int32_t psad[8]; | ||
1809 : | |||
1810 : | const int32_t iSubBlock = ((y&1)<<1) + (x&1); | ||
1811 : | |||
1812 : | suxen_drol | 136 | const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1813 : | const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; | ||
1814 : | chl | 96 | |
1815 : | int32_t bPredEq; | ||
1816 : | int32_t iMinSAD,iSAD=9999; | ||
1817 : | |||
1818 : | MainSearch8FuncPtr EPZSMainSearchPtr; | ||
1819 : | |||
1820 : | /* Get maximum range */ | ||
1821 : | get_range(&min_dx, &max_dx, &min_dy, &max_dy, | ||
1822 : | x, y, 8, iWidth, iHeight, iFcode); | ||
1823 : | |||
1824 : | /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ | ||
1825 : | |||
1826 : | if (!(MotionFlags & PMV_HALFPEL8 )) | ||
1827 : | { min_dx = EVEN(min_dx); | ||
1828 : | max_dx = EVEN(max_dx); | ||
1829 : | min_dy = EVEN(min_dy); | ||
1830 : | max_dy = EVEN(max_dy); | ||
1831 : | } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ | ||
1832 : | |||
1833 : | bPredEq = get_pmvdata(pMBs, x>>1, y>>1, iWcount, iSubBlock, pmv, psad); | ||
1834 : | |||
1835 : | |||
1836 : | /* Step 4: Calculate SAD around the Median prediction. | ||
1837 : | MinSAD=SAD | ||
1838 : | If Motion Vector equal to Previous frame motion vector | ||
1839 : | and MinSAD<PrevFrmSAD goto Step 10. | ||
1840 : | If SAD<=256 goto Step 10. | ||
1841 : | */ | ||
1842 : | |||
1843 : | // Prepare for main loop | ||
1844 : | |||
1845 : | |||
1846 : | if (!(MotionFlags & PMV_HALFPEL8)) | ||
1847 : | { | ||
1848 : | currMV->x = EVEN(currMV->x); | ||
1849 : | currMV->y = EVEN(currMV->y); | ||
1850 : | } | ||
1851 : | |||
1852 : | if (currMV->x > max_dx) | ||
1853 : | currMV->x=max_dx; | ||
1854 : | if (currMV->x < min_dx) | ||
1855 : | currMV->x=min_dx; | ||
1856 : | if (currMV->y > max_dy) | ||
1857 : | currMV->y=max_dy; | ||
1858 : | if (currMV->y < min_dy) | ||
1859 : | currMV->y=min_dy; | ||
1860 : | |||
1861 : | /***************** This is predictor SET A: only median prediction ******************/ | ||
1862 : | |||
1863 : | |||
1864 : | iMinSAD = sad8( cur, | ||
1865 : | get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), | ||
1866 : | iEdgedWidth); | ||
1867 : | iMinSAD += calc_delta_8(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; | ||
1868 : | |||
1869 : | |||
1870 : | // thresh1 is fixed to 256 | ||
1871 : | if (iMinSAD < 256/4 ) | ||
1872 : | { | ||
1873 : | if (MotionFlags & PMV_QUICKSTOP8) | ||
1874 : | goto EPZS8_Terminate_without_Refine; | ||
1875 : | if (MotionFlags & PMV_EARLYSTOP8) | ||
1876 : | goto EPZS8_Terminate_with_Refine; | ||
1877 : | } | ||
1878 : | |||
1879 : | /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ | ||
1880 : | |||
1881 : | // previous frame MV | ||
1882 : | CHECK_MV8_CANDIDATE(pMB->mvs[0].x,pMB->mvs[0].y); | ||
1883 : | |||
1884 : | // MV=(0,0) is often a good choice | ||
1885 : | |||
1886 : | CHECK_MV8_ZERO; | ||
1887 : | |||
1888 : | /* Terminate if MinSAD <= T_2 | ||
1889 : | Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] | ||
1890 : | */ | ||
1891 : | |||
1892 : | if (iMinSAD < 512/4) /* T_2 == 512/4 hardcoded */ | ||
1893 : | { | ||
1894 : | if (MotionFlags & PMV_QUICKSTOP8) | ||
1895 : | goto EPZS8_Terminate_without_Refine; | ||
1896 : | if (MotionFlags & PMV_EARLYSTOP8) | ||
1897 : | goto EPZS8_Terminate_with_Refine; | ||
1898 : | } | ||
1899 : | |||
1900 : | /************ (if Diamond Search) **************/ | ||
1901 : | |||
1902 : | backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ | ||
1903 : | |||
1904 : | if (!(MotionFlags & PMV_HALFPELDIAMOND8)) | ||
1905 : | iDiamondSize *= 2; | ||
1906 : | |||
1907 : | /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ | ||
1908 : | |||
1909 : | // if (MotionFlags & PMV_USESQUARES8) | ||
1910 : | // EPZSMainSearchPtr = Square8_MainSearch; | ||
1911 : | // else | ||
1912 : | EPZSMainSearchPtr = Diamond8_MainSearch; | ||
1913 : | |||
1914 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1915 : | x, y, | ||
1916 : | currMV->x, currMV->y, iMinSAD, &newMV, | ||
1917 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, | ||
1918 : | iDiamondSize, iFcode, iQuant, 00); | ||
1919 : | |||
1920 : | |||
1921 : | if (iSAD < iMinSAD) | ||
1922 : | { | ||
1923 : | *currMV = newMV; | ||
1924 : | iMinSAD = iSAD; | ||
1925 : | } | ||
1926 : | |||
1927 : | if (MotionFlags & PMV_EXTSEARCH8) | ||
1928 : | { | ||
1929 : | /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ | ||
1930 : | |||
1931 : | if (!(MVequal(pmv[0],backupMV)) ) | ||
1932 : | { | ||
1933 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1934 : | x, y, | ||
1935 : | pmv[0].x, pmv[0].y, iMinSAD, &newMV, | ||
1936 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); | ||
1937 : | |||
1938 : | if (iSAD < iMinSAD) | ||
1939 : | { | ||
1940 : | *currMV = newMV; | ||
1941 : | iMinSAD = iSAD; | ||
1942 : | } | ||
1943 : | } | ||
1944 : | |||
1945 : | if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) | ||
1946 : | { | ||
1947 : | iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, | ||
1948 : | x, y, | ||
1949 : | 0, 0, iMinSAD, &newMV, | ||
1950 : | pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); | ||
1951 : | |||
1952 : | if (iSAD < iMinSAD) | ||
1953 : | { | ||
1954 : | *currMV = newMV; | ||
1955 : | iMinSAD = iSAD; | ||
1956 : | } | ||
1957 : | } | ||
1958 : | } | ||
1959 : | |||
1960 : | /*************** Choose best MV found **************/ | ||
1961 : | |||
1962 : | EPZS8_Terminate_with_Refine: | ||
1963 : | if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step | ||
1964 : | iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, | ||
1965 : | x, y, | ||
1966 : | currMV, iMinSAD, | ||
1967 : | pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); | ||
1968 : | |||
1969 : | EPZS8_Terminate_without_Refine: | ||
1970 : | |||
1971 : | currPMV->x = currMV->x - pmv[0].x; | ||
1972 : | currPMV->y = currMV->y - pmv[0].y; | ||
1973 : | return iMinSAD; | ||
1974 : | } | ||
1975 : | |||
1976 : | suxen_drol | 118 | |
1977 : | |||
1978 : | |||
1979 : | |||
1980 : | /* *********************************************************** | ||
1981 : | bvop motion estimation | ||
1982 : | // TODO: need to incorporate prediction here (eg. sad += calc_delta_16) | ||
1983 : | ***************************************************************/ | ||
1984 : | |||
1985 : | /* | ||
1986 : | void MotionEstimationBVOP( | ||
1987 : | MBParam * const pParam, | ||
1988 : | FRAMEINFO * const frame, | ||
1989 : | |||
1990 : | // forward (past) reference | ||
1991 : | const MACROBLOCK * const f_mbs, | ||
1992 : | const IMAGE * const f_ref, | ||
1993 : | const IMAGE * const f_refH, | ||
1994 : | const IMAGE * const f_refV, | ||
1995 : | const IMAGE * const f_refHV, | ||
1996 : | // backward (future) reference | ||
1997 : | const MACROBLOCK * const b_mbs, | ||
1998 : | const IMAGE * const b_ref, | ||
1999 : | const IMAGE * const b_refH, | ||
2000 : | const IMAGE * const b_refV, | ||
2001 : | const IMAGE * const b_refHV) | ||
2002 : | { | ||
2003 : | const uint32_t mb_width = pParam->mb_width; | ||
2004 : | const uint32_t mb_height = pParam->mb_height; | ||
2005 : | const int32_t edged_width = pParam->edged_width; | ||
2006 : | |||
2007 : | int32_t i,j; | ||
2008 : | |||
2009 : | int32_t f_sad16; | ||
2010 : | int32_t b_sad16; | ||
2011 : | int32_t i_sad16; | ||
2012 : | int32_t d_sad16; | ||
2013 : | int32_t best_sad; | ||
2014 : | |||
2015 : | VECTOR pmv_dontcare; | ||
2016 : | |||
2017 : | // note: i==horizontal, j==vertical | ||
2018 : | for (j = 0; j < mb_height; j++) | ||
2019 : | { | ||
2020 : | for (i = 0; i < mb_width; i++) | ||
2021 : | { | ||
2022 : | MACROBLOCK *mb = &frame->mbs[i + j*mb_width]; | ||
2023 : | const MACROBLOCK *f_mb = &f_mbs[i + j*mb_width]; | ||
2024 : | const MACROBLOCK *b_mb = &b_mbs[i + j*mb_width]; | ||
2025 : | |||
2026 : | if (b_mb->mode == MODE_INTER | ||
2027 : | && b_mb->cbp == 0 | ||
2028 : | && b_mb->mvs[0].x == 0 | ||
2029 : | && b_mb->mvs[0].y == 0) | ||
2030 : | { | ||
2031 : | mb->mode = MB_IGNORE; | ||
2032 : | mb->mvs[0].x = 0; | ||
2033 : | mb->mvs[0].y = 0; | ||
2034 : | mb->b_mvs[0].x = 0; | ||
2035 : | mb->b_mvs[0].y = 0; | ||
2036 : | continue; | ||
2037 : | } | ||
2038 : | |||
2039 : | |||
2040 : | // forward search | ||
2041 : | f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, | ||
2042 : | &frame->image, | ||
2043 : | i, j, | ||
2044 : | frame->motion_flags, frame->quant, frame->fcode, | ||
2045 : | pParam, | ||
2046 : | f_mbs, | ||
2047 : | &mb->mvs[0], &pmv_dontcare); // ignore pmv | ||
2048 : | |||
2049 : | // backward search | ||
2050 : | b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
2051 : | &frame->image, | ||
2052 : | i, j, | ||
2053 : | frame->motion_flags, frame->quant, frame->bcode, | ||
2054 : | pParam, | ||
2055 : | b_mbs, | ||
2056 : | &mb->b_mvs[0], &pmv_dontcare); // ignore pmv | ||
2057 : | |||
2058 : | // interpolate search (simple, but effective) | ||
2059 : | i_sad16 = sad16bi_c( | ||
2060 : | frame->image.y + i*16 + j*16*edged_width, | ||
2061 : | get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, | ||
2062 : | i, j, 16, mb->mvs[0].x, mb->mvs[0].y, edged_width), | ||
2063 : | get_ref(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
2064 : | i, j, 16, mb->b_mvs[0].x, mb->b_mvs[0].x, edged_width), | ||
2065 : | edged_width); | ||
2066 : | |||
2067 : | // TODO: direct search | ||
2068 : | // predictor + range of [-32,32] | ||
2069 : | d_sad16 = 65535; | ||
2070 : | |||
2071 : | |||
2072 : | if (f_sad16 < b_sad16) | ||
2073 : | { | ||
2074 : | best_sad = f_sad16; | ||
2075 : | mb->mode = MB_FORWARD; | ||
2076 : | } | ||
2077 : | else | ||
2078 : | { | ||
2079 : | best_sad = b_sad16; | ||
2080 : | mb->mode = MB_BACKWARD; | ||
2081 : | } | ||
2082 : | |||
2083 : | if (i_sad16 < best_sad) | ||
2084 : | { | ||
2085 : | best_sad = i_sad16; | ||
2086 : | mb->mode = MB_INTERPOLATE; | ||
2087 : | } | ||
2088 : | |||
2089 : | if (d_sad16 < best_sad) | ||
2090 : | { | ||
2091 : | best_sad = d_sad16; | ||
2092 : | mb->mode = MB_DIRECT; | ||
2093 : | } | ||
2094 : | |||
2095 : | } | ||
2096 : | } | ||
2097 : | } | ||
2098 : | |||
2099 : | chenm001 | 132 | */ |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |