Parent Directory
|
Revision Log
Revision 580 - (view) (download)
1 : | Isibaar | 3 | /************************************************************************** |
2 : | * | ||
3 : | chl | 259 | * XVID MPEG-4 VIDEO CODEC |
4 : | * motion estimation | ||
5 : | * | ||
6 : | * This program is an implementation of a part of one or more MPEG-4 | ||
7 : | * Video tools as specified in ISO/IEC 14496-2 standard. Those intending | ||
8 : | * to use this software module in hardware or software products are | ||
9 : | * advised that its use may infringe existing patents or copyrights, and | ||
10 : | * any such use would be at such party's own risk. The original | ||
11 : | * developer of this software module and his/her company, and subsequent | ||
12 : | * editors and their companies, will have no liability for use of this | ||
13 : | * software or modifications or derivatives thereof. | ||
14 : | * | ||
15 : | * This program is free software; you can redistribute it and/or modify | ||
16 : | * it under the terms of the GNU General Public License as published by | ||
17 : | * the Free Software Foundation; either version 2 of the License, or | ||
18 : | * (at your option) any later version. | ||
19 : | * | ||
20 : | * This program is distributed in the hope that it will be useful, | ||
21 : | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
22 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
23 : | * GNU General Public License for more details. | ||
24 : | * | ||
25 : | * You should have received a copy of the GNU General Public License | ||
26 : | * along with this program; if not, write to the Free Software | ||
27 : | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
28 : | * | ||
29 : | *************************************************************************/ | ||
30 : | |||
31 : | Isibaar | 3 | #include <assert.h> |
32 : | #include <stdio.h> | ||
33 : | chl | 96 | #include <stdlib.h> |
34 : | Isibaar | 3 | |
35 : | #include "../encoder.h" | ||
36 : | #include "../utils/mbfunctions.h" | ||
37 : | #include "../prediction/mbprediction.h" | ||
38 : | #include "../global.h" | ||
39 : | #include "../utils/timer.h" | ||
40 : | Isibaar | 580 | #include "../image/interpolate8x8.h" |
41 : | chl | 530 | #include "motion_est.h" |
42 : | suxen_drol | 118 | #include "motion.h" |
43 : | Isibaar | 3 | #include "sad.h" |
44 : | Isibaar | 539 | #include "../utils/emms.h" |
45 : | Isibaar | 3 | |
46 : | chl | 530 | #define INITIAL_SKIP_THRESH (10) |
47 : | #define FINAL_SKIP_THRESH (50) | ||
48 : | #define MAX_SAD00_FOR_SKIP (20) | ||
49 : | Isibaar | 539 | #define MAX_CHROMA_SAD_FOR_SKIP (22) |
50 : | #define SKIP_THRESH_B (25) | ||
51 : | Isibaar | 3 | |
52 : | chl | 530 | #define CHECK_CANDIDATE(X,Y,D) { \ |
53 : | (*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } | ||
54 : | Isibaar | 3 | |
55 : | Isibaar | 580 | #define GET_REFERENCE(X, Y, REF) { \ |
56 : | switch ( ((X&1)<<1) + (Y&1) ) \ | ||
57 : | { \ | ||
58 : | case 0 : REF = data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \ | ||
59 : | case 1 : REF = data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \ | ||
60 : | case 2 : REF = data->RefH + ((X)-1)/2 + ((Y)/2)*(data->iEdgedWidth); break; \ | ||
61 : | default : REF = data->RefHV + ((X)-1)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \ | ||
62 : | } \ | ||
63 : | } | ||
64 : | |||
65 : | chl | 530 | #define iDiamondSize 2 |
66 : | chl | 141 | |
67 : | chl | 530 | static __inline int |
68 : | d_mv_bits(int x, int y, const uint32_t iFcode) | ||
69 : | Isibaar | 3 | { |
70 : | chl | 530 | int xb, yb; |
71 : | |||
72 : | if (x == 0) xb = 1; | ||
73 : | else { | ||
74 : | if (x < 0) x = -x; | ||
75 : | x += (1 << (iFcode - 1)) - 1; | ||
76 : | x >>= (iFcode - 1); | ||
77 : | if (x > 32) x = 32; | ||
78 : | xb = mvtab[x] + iFcode; | ||
79 : | edgomez | 195 | } |
80 : | Isibaar | 3 | |
81 : | chl | 530 | if (y == 0) yb = 1; |
82 : | else { | ||
83 : | if (y < 0) y = -y; | ||
84 : | y += (1 << (iFcode - 1)) - 1; | ||
85 : | y >>= (iFcode - 1); | ||
86 : | if (y > 32) y = 32; | ||
87 : | yb = mvtab[y] + iFcode; | ||
88 : | } | ||
89 : | return xb + yb; | ||
90 : | Isibaar | 3 | } |
91 : | |||
92 : | |||
93 : | Isibaar | 539 | /* CHECK_CANDIATE FUNCTIONS START */ |
94 : | |||
95 : | chl | 530 | static void |
96 : | CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
97 : | Isibaar | 3 | { |
98 : | chl | 530 | int32_t * const sad = data->temp; |
99 : | int t; | ||
100 : | const uint8_t * Reference; | ||
101 : | Isibaar | 3 | |
102 : | chl | 530 | if (( x > data->max_dx) || ( x < data->min_dx) |
103 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
104 : | suxen_drol | 136 | |
105 : | chl | 530 | switch ( ((x&1)<<1) + (y&1) ) { |
106 : | case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; | ||
107 : | case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; | ||
108 : | case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; | ||
109 : | default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; | ||
110 : | } | ||
111 : | chl | 259 | |
112 : | chl | 530 | data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, sad+1); |
113 : | Isibaar | 3 | |
114 : | chl | 530 | t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
115 : | data->temp[0] += lambda_vec16[data->iQuant] * t; | ||
116 : | data->temp[1] += lambda_vec8[data->iQuant] * t; | ||
117 : | chl | 326 | |
118 : | chl | 530 | if (data->temp[0] < data->iMinSAD[0]) { |
119 : | data->iMinSAD[0] = data->temp[0]; | ||
120 : | data->currentMV[0].x = x; data->currentMV[0].y = y; | ||
121 : | *dir = Direction; } | ||
122 : | chl | 184 | |
123 : | chl | 530 | if (data->temp[1] < data->iMinSAD[1]) { |
124 : | data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } | ||
125 : | if (data->temp[2] < data->iMinSAD[2]) { | ||
126 : | data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } | ||
127 : | if (data->temp[3] < data->iMinSAD[3]) { | ||
128 : | data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } | ||
129 : | if (data->temp[4] < data->iMinSAD[4]) { | ||
130 : | data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } | ||
131 : | chl | 184 | |
132 : | Isibaar | 3 | } |
133 : | |||
134 : | chl | 530 | static void |
135 : | CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
136 : | { | ||
137 : | int32_t sad; | ||
138 : | const uint8_t * Reference; | ||
139 : | chl | 326 | |
140 : | chl | 530 | if (( x > data->max_dx) || ( x < data->min_dx) |
141 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
142 : | Isibaar | 3 | |
143 : | chl | 530 | switch ( ((x&1)<<1) + (y&1) ) |
144 : | { | ||
145 : | case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; | ||
146 : | case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; | ||
147 : | case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; | ||
148 : | default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; | ||
149 : | } | ||
150 : | Isibaar | 3 | |
151 : | chl | 530 | sad = lambda_vec16[data->iQuant] * |
152 : | d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); | ||
153 : | Isibaar | 580 | sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR); |
154 : | Isibaar | 3 | |
155 : | chl | 530 | if (sad < *(data->iMinSAD)) { |
156 : | *(data->iMinSAD) = sad; | ||
157 : | data->currentMV[0].x = x; data->currentMV[0].y = y; | ||
158 : | *dir = Direction; } | ||
159 : | Isibaar | 3 | } |
160 : | |||
161 : | chl | 530 | static void |
162 : | Isibaar | 580 | CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
163 : | |||
164 : | // CheckCandidate16 variant which expects x and y in quarter pixel resolution | ||
165 : | // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) | ||
166 : | // around currentMV! | ||
167 : | { | ||
168 : | int32_t * const sad = data->temp; | ||
169 : | int t; | ||
170 : | uint8_t * Reference = (uint8_t *) data->RefQ; | ||
171 : | const uint8_t *ref1, *ref2, *ref3, *ref4; | ||
172 : | VECTOR halfpelMV = *(data->currentMV); | ||
173 : | |||
174 : | int32_t iEdgedWidth = data->iEdgedWidth; | ||
175 : | uint32_t rounding = data->rounding; | ||
176 : | |||
177 : | if (( x > data->max_dx) || ( x < data->min_dx) | ||
178 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
179 : | |||
180 : | switch( ((x&1)<<1) + (y&1) ) | ||
181 : | { | ||
182 : | case 0: // pure halfpel position - shouldn't happen during a refinement step | ||
183 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, (const uint8_t *) Reference); | ||
184 : | break; | ||
185 : | |||
186 : | case 1: // x halfpel, y qpel - top or bottom during qpel refinement | ||
187 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
188 : | GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); | ||
189 : | |||
190 : | interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); | ||
191 : | interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); | ||
192 : | interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); | ||
193 : | interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); | ||
194 : | break; | ||
195 : | |||
196 : | case 2: // x qpel, y halfpel - left or right during qpel refinement | ||
197 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
198 : | GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); | ||
199 : | |||
200 : | interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); | ||
201 : | interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); | ||
202 : | interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); | ||
203 : | interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); | ||
204 : | break; | ||
205 : | |||
206 : | default: // x and y in qpel resolution - the "corners" (top left/right and | ||
207 : | // bottom left/right) during qpel refinement | ||
208 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
209 : | GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); | ||
210 : | GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); | ||
211 : | GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); | ||
212 : | |||
213 : | interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); | ||
214 : | interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); | ||
215 : | interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); | ||
216 : | interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); | ||
217 : | break; | ||
218 : | } | ||
219 : | |||
220 : | data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, sad+1); | ||
221 : | |||
222 : | t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode); | ||
223 : | data->temp[0] += lambda_vec16[data->iQuant] * t; | ||
224 : | data->temp[1] += lambda_vec8[data->iQuant] * t; | ||
225 : | |||
226 : | if (data->temp[0] < data->iMinSAD[0]) { | ||
227 : | data->iMinSAD[0] = data->temp[0]; | ||
228 : | data->currentQMV[0].x = x; data->currentQMV[0].y = y; | ||
229 : | *dir = Direction; } | ||
230 : | |||
231 : | if (data->temp[1] < data->iMinSAD[1]) { | ||
232 : | data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; } | ||
233 : | if (data->temp[2] < data->iMinSAD[2]) { | ||
234 : | data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; } | ||
235 : | if (data->temp[3] < data->iMinSAD[3]) { | ||
236 : | data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; } | ||
237 : | if (data->temp[4] < data->iMinSAD[4]) { | ||
238 : | data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; } | ||
239 : | } | ||
240 : | |||
241 : | static void | ||
242 : | CheckCandidate16no4v_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
243 : | |||
244 : | // CheckCandidate16no4v variant which expects x and y in quarter pixel resolution | ||
245 : | // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) | ||
246 : | // around currentMV! | ||
247 : | { | ||
248 : | int32_t sad; | ||
249 : | uint8_t * Reference = (uint8_t *) data->RefQ; | ||
250 : | const uint8_t *ref1, *ref2, *ref3, *ref4; | ||
251 : | VECTOR halfpelMV = *(data->currentMV); | ||
252 : | |||
253 : | int32_t iEdgedWidth = data->iEdgedWidth; | ||
254 : | uint32_t rounding = data->rounding; | ||
255 : | |||
256 : | if (( x > data->max_dx) || ( x < data->min_dx) | ||
257 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
258 : | |||
259 : | switch( ((x&1)<<1) + (y&1) ) | ||
260 : | { | ||
261 : | case 0: // pure halfpel position - shouldn't happen during a refinement step | ||
262 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, (const uint8_t *) Reference); | ||
263 : | break; | ||
264 : | |||
265 : | case 1: // x halfpel, y qpel - top or bottom during qpel refinement | ||
266 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
267 : | GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); | ||
268 : | |||
269 : | interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); | ||
270 : | interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); | ||
271 : | interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); | ||
272 : | interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); | ||
273 : | break; | ||
274 : | |||
275 : | case 2: // x qpel, y halfpel - left or right during qpel refinement | ||
276 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
277 : | GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); | ||
278 : | |||
279 : | interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); | ||
280 : | interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); | ||
281 : | interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); | ||
282 : | interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); | ||
283 : | break; | ||
284 : | |||
285 : | default: // x and y in qpel resolution - the "corners" (top left/right and | ||
286 : | // bottom left/right) during qpel refinement | ||
287 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
288 : | GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); | ||
289 : | GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); | ||
290 : | GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); | ||
291 : | |||
292 : | interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); | ||
293 : | interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); | ||
294 : | interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); | ||
295 : | interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); | ||
296 : | break; | ||
297 : | } | ||
298 : | |||
299 : | sad = lambda_vec16[data->iQuant] * | ||
300 : | d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode); | ||
301 : | sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR); | ||
302 : | |||
303 : | if (sad < *(data->iMinSAD)) { | ||
304 : | *(data->iMinSAD) = sad; | ||
305 : | data->currentQMV[0].x = x; data->currentQMV[0].y = y; | ||
306 : | *dir = Direction; | ||
307 : | } | ||
308 : | } | ||
309 : | |||
310 : | static void | ||
311 : | Isibaar | 539 | CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
312 : | { | ||
313 : | int32_t sad; | ||
314 : | |||
315 : | if (( x > data->max_dx) || ( x < data->min_dx) | ||
316 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
317 : | |||
318 : | sad = lambda_vec16[data->iQuant] * | ||
319 : | d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); | ||
320 : | |||
321 : | sad += sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), | ||
322 : | data->iEdgedWidth, 256*4096); | ||
323 : | |||
324 : | if (sad < *(data->iMinSAD)) { | ||
325 : | *(data->iMinSAD) = sad; | ||
326 : | data->currentMV[0].x = x; data->currentMV[0].y = y; | ||
327 : | *dir = Direction; } | ||
328 : | } | ||
329 : | |||
330 : | |||
331 : | static void | ||
332 : | chl | 530 | CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data) |
333 : | { | ||
334 : | int32_t sad; | ||
335 : | const int xb = data->currentMV[1].x; | ||
336 : | const int yb = data->currentMV[1].y; | ||
337 : | const uint8_t *ReferenceF, *ReferenceB; | ||
338 : | Isibaar | 3 | |
339 : | chl | 530 | if (( xf > data->max_dx) || ( xf < data->min_dx) |
340 : | || ( yf > data->max_dy) || (yf < data->min_dy)) return; | ||
341 : | Isibaar | 3 | |
342 : | chl | 530 | switch ( ((xf&1)<<1) + (yf&1) ) { |
343 : | case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break; | ||
344 : | case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break; | ||
345 : | case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break; | ||
346 : | default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break; | ||
347 : | } | ||
348 : | edgomez | 195 | |
349 : | chl | 530 | switch ( ((xb&1)<<1) + (yb&1) ) { |
350 : | case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break; | ||
351 : | case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break; | ||
352 : | case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break; | ||
353 : | default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break; | ||
354 : | } | ||
355 : | Isibaar | 3 | |
356 : | chl | 530 | sad = lambda_vec16[data->iQuant] * |
357 : | ( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) + | ||
358 : | d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) ); | ||
359 : | Isibaar | 3 | |
360 : | chl | 530 | sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
361 : | Isibaar | 3 | |
362 : | chl | 530 | if (sad < *(data->iMinSAD)) { |
363 : | *(data->iMinSAD) = sad; | ||
364 : | data->currentMV->x = xf; data->currentMV->y = yf; | ||
365 : | *dir = Direction; } | ||
366 : | Isibaar | 3 | } |
367 : | |||
368 : | chl | 530 | static void |
369 : | CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
370 : | Isibaar | 3 | { |
371 : | chl | 530 | int32_t sad; |
372 : | int k; | ||
373 : | const uint8_t *ReferenceF; | ||
374 : | const uint8_t *ReferenceB; | ||
375 : | VECTOR mvs, b_mvs; | ||
376 : | Isibaar | 3 | |
377 : | chl | 530 | if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
378 : | Isibaar | 3 | |
379 : | chl | 530 | sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
380 : | Isibaar | 3 | |
381 : | chl | 530 | for (k = 0; k < 4; k++) { |
382 : | mvs.x = data->directmvF[k].x + x; | ||
383 : | b_mvs.x = ((x == 0) ? | ||
384 : | data->directmvB[k].x | ||
385 : | : mvs.x - data->referencemv[k].x); | ||
386 : | Isibaar | 3 | |
387 : | chl | 530 | mvs.y = data->directmvF[k].y + y; |
388 : | b_mvs.y = ((y == 0) ? | ||
389 : | data->directmvB[k].y | ||
390 : | : mvs.y - data->referencemv[k].y); | ||
391 : | |||
392 : | if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx ) | ||
393 : | || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy ) | ||
394 : | || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) | ||
395 : | || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return; | ||
396 : | Isibaar | 3 | |
397 : | chl | 530 | switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) { |
398 : | case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break; | ||
399 : | case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
400 : | case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break; | ||
401 : | default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
402 : | } | ||
403 : | Isibaar | 3 | |
404 : | chl | 530 | switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) { |
405 : | case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; | ||
406 : | case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
407 : | case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; | ||
408 : | default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
409 : | chl | 344 | } |
410 : | chl | 530 | |
411 : | sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), | ||
412 : | ReferenceF + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), | ||
413 : | ReferenceB + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), | ||
414 : | data->iEdgedWidth); | ||
415 : | if (sad > *(data->iMinSAD)) return; | ||
416 : | edgomez | 78 | } |
417 : | chl | 530 | |
418 : | if (sad < *(data->iMinSAD)) { | ||
419 : | *(data->iMinSAD) = sad; | ||
420 : | data->currentMV->x = x; data->currentMV->y = y; | ||
421 : | *dir = Direction; } | ||
422 : | Isibaar | 3 | } |
423 : | |||
424 : | chl | 530 | static void |
425 : | CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
426 : | chl | 96 | { |
427 : | chl | 530 | int32_t sad; |
428 : | const uint8_t *ReferenceF; | ||
429 : | const uint8_t *ReferenceB; | ||
430 : | VECTOR mvs, b_mvs; | ||
431 : | chl | 96 | |
432 : | syskin | 576 | if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
433 : | chl | 530 | |
434 : | sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); | ||
435 : | edgomez | 195 | |
436 : | chl | 530 | mvs.x = data->directmvF[0].x + x; |
437 : | b_mvs.x = ((x == 0) ? | ||
438 : | data->directmvB[0].x | ||
439 : | : mvs.x - data->referencemv[0].x); | ||
440 : | edgomez | 195 | |
441 : | chl | 530 | mvs.y = data->directmvF[0].y + y; |
442 : | b_mvs.y = ((y == 0) ? | ||
443 : | data->directmvB[0].y | ||
444 : | : mvs.y - data->referencemv[0].y); | ||
445 : | |||
446 : | if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx ) | ||
447 : | || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy ) | ||
448 : | || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) | ||
449 : | || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return; | ||
450 : | chl | 96 | |
451 : | chl | 530 | switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) { |
452 : | case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break; | ||
453 : | case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
454 : | case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break; | ||
455 : | default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
456 : | } | ||
457 : | chl | 96 | |
458 : | chl | 530 | switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) { |
459 : | case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; | ||
460 : | case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
461 : | case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; | ||
462 : | default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; | ||
463 : | } | ||
464 : | |||
465 : | sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); | ||
466 : | chl | 96 | |
467 : | chl | 530 | if (sad < *(data->iMinSAD)) { |
468 : | *(data->iMinSAD) = sad; | ||
469 : | data->currentMV->x = x; data->currentMV->y = y; | ||
470 : | *dir = Direction; } | ||
471 : | } | ||
472 : | chl | 96 | |
473 : | chl | 530 | static void |
474 : | CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
475 : | { | ||
476 : | int32_t sad; | ||
477 : | const uint8_t * Reference; | ||
478 : | edgomez | 195 | |
479 : | chl | 530 | if (( x > data->max_dx) || ( x < data->min_dx) |
480 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
481 : | chl | 96 | |
482 : | chl | 530 | switch ( ((x&1)<<1) + (y&1) ) |
483 : | { | ||
484 : | case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; | ||
485 : | case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; | ||
486 : | case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; | ||
487 : | default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; | ||
488 : | } | ||
489 : | edgomez | 195 | |
490 : | chl | 530 | sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
491 : | sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); | ||
492 : | edgomez | 195 | |
493 : | chl | 530 | if (sad < *(data->iMinSAD)) { |
494 : | *(data->iMinSAD) = sad; | ||
495 : | data->currentMV->x = x; data->currentMV->y = y; | ||
496 : | *dir = Direction; } | ||
497 : | chl | 96 | } |
498 : | |||
499 : | Isibaar | 580 | static void |
500 : | CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) | ||
501 : | // CheckCandidate16no4v variant which expects x and y in quarter pixel resolution | ||
502 : | // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) | ||
503 : | // around currentMV! | ||
504 : | |||
505 : | { | ||
506 : | int32_t sad; | ||
507 : | uint8_t *Reference = (uint8_t *) data->RefQ; | ||
508 : | const uint8_t *ref1, *ref2, *ref3, *ref4; | ||
509 : | VECTOR halfpelMV = *(data->currentMV); | ||
510 : | |||
511 : | int32_t iEdgedWidth = data->iEdgedWidth; | ||
512 : | uint32_t rounding = data->rounding; | ||
513 : | |||
514 : | if (( x > data->max_dx) || ( x < data->min_dx) | ||
515 : | || ( y > data->max_dy) || (y < data->min_dy)) return; | ||
516 : | |||
517 : | switch( ((x&1)<<1) + (y&1) ) | ||
518 : | { | ||
519 : | case 0: // pure halfpel position - shouldn't happen during a refinement step | ||
520 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, (const uint8_t *) Reference); | ||
521 : | break; | ||
522 : | |||
523 : | case 1: // x halfpel, y qpel - top or bottom during qpel refinement | ||
524 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
525 : | GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); | ||
526 : | |||
527 : | interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); | ||
528 : | break; | ||
529 : | |||
530 : | case 2: // x qpel, y halfpel - left or right during qpel refinement | ||
531 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
532 : | GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); | ||
533 : | |||
534 : | interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); | ||
535 : | break; | ||
536 : | |||
537 : | default: // x and y in qpel resolution - the "corners" (top left/right and | ||
538 : | // bottom left/right) during qpel refinement | ||
539 : | GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); | ||
540 : | GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); | ||
541 : | GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); | ||
542 : | GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); | ||
543 : | |||
544 : | interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); | ||
545 : | break; | ||
546 : | } | ||
547 : | |||
548 : | sad = sad8(data->Cur, Reference, data->iEdgedWidth); | ||
549 : | sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode); | ||
550 : | |||
551 : | if (sad < *(data->iMinSAD)) { | ||
552 : | *(data->iMinSAD) = sad; | ||
553 : | data->currentQMV->x = x; data->currentQMV->y = y; | ||
554 : | *dir = Direction; } | ||
555 : | } | ||
556 : | |||
557 : | Isibaar | 539 | /* CHECK_CANDIATE FUNCTIONS END */ |
558 : | chl | 96 | |
559 : | chl | 530 | /* MAINSEARCH FUNCTIONS START */ |
560 : | edgomez | 195 | |
561 : | chl | 530 | static void |
562 : | AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection) | ||
563 : | chl | 181 | { |
564 : | |||
565 : | /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ | ||
566 : | |||
567 : | chl | 530 | int iDirection; |
568 : | edgomez | 195 | |
569 : | do { | ||
570 : | chl | 181 | iDirection = 0; |
571 : | chl | 530 | if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
572 : | if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); | ||
573 : | if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); | ||
574 : | if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8); | ||
575 : | chl | 181 | |
576 : | /* now we're doing diagonal checks near our candidate */ | ||
577 : | |||
578 : | chl | 530 | if (iDirection) { //checking if anything found |
579 : | chl | 181 | bDirection = iDirection; |
580 : | iDirection = 0; | ||
581 : | chl | 530 | x = data->currentMV->x; y = data->currentMV->y; |
582 : | if (bDirection & 3) { //our candidate is left or right | ||
583 : | CHECK_CANDIDATE(x, y + iDiamondSize, 8); | ||
584 : | CHECK_CANDIDATE(x, y - iDiamondSize, 4); | ||
585 : | } else { // what remains here is up or down | ||
586 : | CHECK_CANDIDATE(x + iDiamondSize, y, 2); | ||
587 : | CHECK_CANDIDATE(x - iDiamondSize, y, 1); } | ||
588 : | chl | 181 | |
589 : | edgomez | 195 | if (iDirection) { |
590 : | bDirection += iDirection; | ||
591 : | chl | 530 | x = data->currentMV->x; y = data->currentMV->y; } |
592 : | } else { //about to quit, eh? not so fast.... | ||
593 : | edgomez | 195 | switch (bDirection) { |
594 : | chl | 181 | case 2: |
595 : | chl | 530 | CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
596 : | CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); | ||
597 : | chl | 181 | break; |
598 : | case 1: | ||
599 : | chl | 530 | CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); |
600 : | CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); | ||
601 : | chl | 181 | break; |
602 : | edgomez | 195 | case 2 + 4: |
603 : | chl | 530 | CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); |
604 : | CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); | ||
605 : | CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); | ||
606 : | chl | 181 | break; |
607 : | case 4: | ||
608 : | chl | 530 | CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
609 : | CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); | ||
610 : | chl | 181 | break; |
611 : | case 8: | ||
612 : | chl | 530 | CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
613 : | CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); | ||
614 : | chl | 181 | break; |
615 : | edgomez | 195 | case 1 + 4: |
616 : | chl | 530 | CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); |
617 : | CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); | ||
618 : | CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); | ||
619 : | chl | 181 | break; |
620 : | edgomez | 195 | case 2 + 8: |
621 : | chl | 530 | CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); |
622 : | CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); | ||
623 : | CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); | ||
624 : | chl | 181 | break; |
625 : | edgomez | 195 | case 1 + 8: |
626 : | chl | 530 | CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
627 : | CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); | ||
628 : | CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); | ||
629 : | chl | 181 | break; |
630 : | edgomez | 195 | default: //1+2+4+8 == we didn't find anything at all |
631 : | chl | 530 | CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); |
632 : | CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); | ||
633 : | CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); | ||
634 : | CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); | ||
635 : | chl | 181 | break; |
636 : | } | ||
637 : | chl | 530 | if (!iDirection) break; //ok, the end. really |
638 : | bDirection = iDirection; | ||
639 : | x = data->currentMV->x; y = data->currentMV->y; | ||
640 : | chl | 181 | } |
641 : | } | ||
642 : | edgomez | 195 | while (1); //forever |
643 : | chl | 181 | } |
644 : | |||
645 : | chl | 530 | static void |
646 : | SquareSearch(int x, int y, const SearchData * const data, int bDirection) | ||
647 : | chl | 326 | { |
648 : | chl | 530 | int iDirection; |
649 : | chl | 326 | |
650 : | chl | 530 | do { |
651 : | iDirection = 0; | ||
652 : | if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64); | ||
653 : | if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128); | ||
654 : | if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32); | ||
655 : | if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128); | ||
656 : | if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64); | ||
657 : | if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128); | ||
658 : | if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128); | ||
659 : | if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128); | ||
660 : | chl | 326 | |
661 : | chl | 530 | bDirection = iDirection; |
662 : | x = data->currentMV->x; y = data->currentMV->y; | ||
663 : | } while (iDirection); | ||
664 : | chl | 346 | } |
665 : | |||
666 : | chl | 530 | static void |
667 : | DiamondSearch(int x, int y, const SearchData * const data, int bDirection) | ||
668 : | chl | 346 | { |
669 : | |||
670 : | chl | 181 | /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
671 : | |||
672 : | chl | 530 | int iDirection; |
673 : | edgomez | 195 | |
674 : | do { | ||
675 : | chl | 181 | iDirection = 0; |
676 : | chl | 530 | if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
677 : | if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); | ||
678 : | if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); | ||
679 : | if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8); | ||
680 : | chl | 181 | |
681 : | /* now we're doing diagonal checks near our candidate */ | ||
682 : | |||
683 : | chl | 530 | if (iDirection) { //checking if anything found |
684 : | chl | 181 | bDirection = iDirection; |
685 : | iDirection = 0; | ||
686 : | chl | 530 | x = data->currentMV->x; y = data->currentMV->y; |
687 : | if (bDirection & 3) { //our candidate is left or right | ||
688 : | CHECK_CANDIDATE(x, y + iDiamondSize, 8); | ||
689 : | CHECK_CANDIDATE(x, y - iDiamondSize, 4); | ||
690 : | } else { // what remains here is up or down | ||
691 : | CHECK_CANDIDATE(x + iDiamondSize, y, 2); | ||
692 : | CHECK_CANDIDATE(x - iDiamondSize, y, 1); } | ||
693 : | chl | 181 | |
694 : | chl | 530 | bDirection += iDirection; |
695 : | x = data->currentMV->x; y = data->currentMV->y; | ||
696 : | chl | 181 | } |
697 : | } | ||
698 : | chl | 530 | while (iDirection); |
699 : | chl | 181 | } |
700 : | |||
701 : | chl | 530 | /* MAINSEARCH FUNCTIONS END */ |
702 : | chl | 181 | |
703 : | chl | 530 | /* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */ |
704 : | |||
705 : | static void | ||
706 : | HalfpelRefine(const SearchData * const data) | ||
707 : | chl | 96 | { |
708 : | chl | 530 | /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ |
709 : | edgomez | 195 | |
710 : | chl | 530 | VECTOR backupMV = *(data->currentMV); |
711 : | int iDirection; //not needed | ||
712 : | chl | 96 | |
713 : | chl | 530 | CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
714 : | CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); | ||
715 : | CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); | ||
716 : | CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); | ||
717 : | edgomez | 195 | |
718 : | chl | 530 | CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); |
719 : | CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); | ||
720 : | |||
721 : | CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); | ||
722 : | CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); | ||
723 : | chl | 96 | } |
724 : | |||
725 : | Isibaar | 580 | |
726 : | static void | ||
727 : | QuarterpelRefine(const SearchData * const data) | ||
728 : | { | ||
729 : | /* Perform quarter pixel refinement*/ | ||
730 : | |||
731 : | VECTOR backupMV = *(data->currentQMV); | ||
732 : | int iDirection; //not needed | ||
733 : | |||
734 : | CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); | ||
735 : | CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); | ||
736 : | CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); | ||
737 : | CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); | ||
738 : | |||
739 : | CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); | ||
740 : | CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); | ||
741 : | |||
742 : | CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); | ||
743 : | CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); | ||
744 : | |||
745 : | } | ||
746 : | |||
747 : | chl | 530 | static __inline int |
748 : | SkipDecisionP(const IMAGE * current, const IMAGE * reference, | ||
749 : | const int x, const int y, | ||
750 : | const uint32_t iEdgedWidth, const uint32_t iQuant) | ||
751 : | chl | 96 | |
752 : | Isibaar | 3 | { |
753 : | chl | 530 | /* keep repeating checks for all b-frames before this P frame, |
754 : | to make sure that SKIP is possible (todo) | ||
755 : | how: if skip is not possible set sad00 to a very high value */ | ||
756 : | Isibaar | 3 | |
757 : | chl | 530 | uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
758 : | reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); | ||
759 : | if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; | ||
760 : | h | 545 | sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
761 : | reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); | ||
762 : | chl | 530 | if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
763 : | edgomez | 195 | |
764 : | chl | 530 | return 1; |
765 : | } | ||
766 : | edgomez | 195 | |
767 : | chl | 530 | static __inline void |
768 : | SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) | ||
769 : | { | ||
770 : | pMB->mode = MODE_NOT_CODED; | ||
771 : | Isibaar | 539 | pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
772 : | pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; | ||
773 : | Isibaar | 580 | |
774 : | pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0; | ||
775 : | pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0; | ||
776 : | |||
777 : | chl | 530 | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
778 : | Isibaar | 3 | } |
779 : | |||
780 : | chl | 530 | bool |
781 : | MotionEstimation(MBParam * const pParam, | ||
782 : | FRAMEINFO * const current, | ||
783 : | FRAMEINFO * const reference, | ||
784 : | const IMAGE * const pRefH, | ||
785 : | const IMAGE * const pRefV, | ||
786 : | const IMAGE * const pRefHV, | ||
787 : | const uint32_t iLimit) | ||
788 : | Isibaar | 3 | { |
789 : | chl | 530 | MACROBLOCK *const pMBs = current->mbs; |
790 : | const IMAGE *const pCurrent = ¤t->image; | ||
791 : | const IMAGE *const pRef = &reference->image; | ||
792 : | Isibaar | 3 | |
793 : | chl | 530 | const VECTOR zeroMV = { 0, 0 }; |
794 : | Isibaar | 3 | |
795 : | chl | 530 | uint32_t x, y; |
796 : | uint32_t iIntra = 0; | ||
797 : | syskin | 574 | int32_t InterBias, quant = current->quant; |
798 : | Isibaar | 580 | uint8_t *qimage; |
799 : | edgomez | 195 | |
800 : | Isibaar | 539 | // some pre-initialized thingies for SearchP |
801 : | int32_t temp[5]; | ||
802 : | VECTOR currentMV[5]; | ||
803 : | Isibaar | 580 | VECTOR currentQMV[5]; |
804 : | Isibaar | 539 | int32_t iMinSAD[5]; |
805 : | SearchData Data; | ||
806 : | Data.iEdgedWidth = pParam->edged_width; | ||
807 : | Data.currentMV = currentMV; | ||
808 : | Isibaar | 580 | Data.currentQMV = currentQMV; |
809 : | Isibaar | 539 | Data.iMinSAD = iMinSAD; |
810 : | Data.temp = temp; | ||
811 : | Data.iFcode = current->fcode; | ||
812 : | Isibaar | 580 | Data.rounding = pParam->m_rounding_type; |
813 : | Isibaar | 539 | |
814 : | Isibaar | 580 | if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
815 : | return 0; // allocate some mem for qpel interpolated blocks | ||
816 : | // somehow this is dirty since I think we shouldn't use malloc outside | ||
817 : | // encoder_create() - so please fix me! | ||
818 : | |||
819 : | chl | 530 | if (sadInit) (*sadInit) (); |
820 : | edgomez | 195 | |
821 : | chl | 530 | for (y = 0; y < pParam->mb_height; y++) { |
822 : | for (x = 0; x < pParam->mb_width; x++) { | ||
823 : | Isibaar | 3 | |
824 : | chl | 530 | MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
825 : | int32_t sad00 = pMB->sad16 | ||
826 : | = sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16, | ||
827 : | pRef->y + (x + y * pParam->edged_width) * 16, | ||
828 : | pParam->edged_width, pMB->sad8 ); | ||
829 : | edgomez | 195 | |
830 : | chl | 530 | if (!(current->global_flags & XVID_LUMIMASKING)) { |
831 : | pMB->dquant = NO_CHANGE; | ||
832 : | pMB->quant = current->quant; } | ||
833 : | syskin | 574 | else |
834 : | if (pMB->dquant != NO_CHANGE) { | ||
835 : | quant += DQtab[pMB->dquant]; | ||
836 : | if (quant > 31) quant = 31; | ||
837 : | else if (quant < 1) quant = 1; | ||
838 : | pMB->quant = quant; | ||
839 : | } | ||
840 : | chl | 181 | |
841 : | chl | 530 | //initial skip decision |
842 : | Isibaar | 3 | |
843 : | chl | 530 | if ((pMB->dquant == NO_CHANGE) && (sad00 <= MAX_SAD00_FOR_SKIP * pMB->quant) |
844 : | && (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) ) { | ||
845 : | if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) { | ||
846 : | SkipMacroblockP(pMB, sad00); | ||
847 : | continue; | ||
848 : | Isibaar | 580 | sad00 = 256 * 4096; |
849 : | chl | 530 | } |
850 : | } else sad00 = 256*4096; // skip not allowed - for final skip decision | ||
851 : | Isibaar | 3 | |
852 : | Isibaar | 580 | SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x, |
853 : | chl | 530 | y, current->motion_flags, pMB->quant, |
854 : | Isibaar | 539 | &Data, pParam, pMBs, reference->mbs, |
855 : | chl | 530 | current->global_flags & XVID_INTER4V, pMB); |
856 : | edgomez | 195 | |
857 : | chl | 530 | /* final skip decision, a.k.a. "the vector you found, really that good?" */ |
858 : | if (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) | ||
859 : | if ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) | ||
860 : | { SkipMacroblockP(pMB, sad00); continue; } | ||
861 : | |||
862 : | /* finally, intra decision */ | ||
863 : | Isibaar | 3 | |
864 : | chl | 530 | InterBias = MV16_INTER_BIAS; |
865 : | if (pMB->quant > 8) InterBias += 50 * (pMB->quant - 8); // to make high quants work | ||
866 : | if (y != 0) | ||
867 : | if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 50; | ||
868 : | if (x != 0) | ||
869 : | if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 50; | ||
870 : | Isibaar | 3 | |
871 : | chl | 530 | if (InterBias < pMB->sad16) { |
872 : | const int32_t deviation = | ||
873 : | dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, | ||
874 : | pParam->edged_width); | ||
875 : | Isibaar | 3 | |
876 : | chl | 530 | if (deviation < (pMB->sad16 - InterBias)) { |
877 : | Isibaar | 580 | if (++iIntra >= iLimit) { free(qimage); return 1; } |
878 : | chl | 530 | pMB->mode = MODE_INTRA; |
879 : | Isibaar | 539 | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
880 : | pMB->mvs[3] = zeroMV; | ||
881 : | Isibaar | 580 | pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = |
882 : | pMB->qmvs[3] = zeroMV; | ||
883 : | chl | 530 | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
884 : | pMB->sad8[3] = 0; | ||
885 : | } | ||
886 : | } | ||
887 : | } | ||
888 : | Isibaar | 3 | } |
889 : | Isibaar | 580 | free(qimage); |
890 : | chl | 530 | return 0; |
891 : | } | ||
892 : | Isibaar | 3 | |
893 : | edgomez | 195 | |
894 : | chl | 530 | #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
895 : | Isibaar | 3 | |
896 : | chl | 530 | static __inline int |
897 : | make_mask(const VECTOR * const pmv, const int i) | ||
898 : | { | ||
899 : | Isibaar | 539 | int mask = 255, j; |
900 : | chl | 530 | for (j = 0; j < i; j++) { |
901 : | if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already | ||
902 : | if (pmv[i].x == pmv[j].x) { | ||
903 : | if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; } | ||
904 : | if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; } | ||
905 : | } else | ||
906 : | if (pmv[i].y == pmv[j].y) { | ||
907 : | if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; } | ||
908 : | if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; } | ||
909 : | } | ||
910 : | Isibaar | 3 | } |
911 : | chl | 530 | return mask; |
912 : | } | ||
913 : | edgomez | 195 | |
914 : | chl | 530 | static __inline void |
915 : | PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount, | ||
916 : | const int iHcount, const MACROBLOCK * const prevMB) | ||
917 : | { | ||
918 : | edgomez | 195 | |
919 : | chl | 530 | //this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself |
920 : | edgomez | 195 | |
921 : | chl | 530 | if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour |
922 : | pmv[5].x = EVEN(pmv[3].x); | ||
923 : | Isibaar | 539 | pmv[5].y = EVEN(pmv[3].y); |
924 : | } else pmv[5].x = pmv[5].y = 0; | ||
925 : | chl | 169 | |
926 : | chl | 530 | if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour |
927 : | else pmv[3].x = pmv[3].y = 0; | ||
928 : | Isibaar | 3 | |
929 : | chl | 530 | if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour |
930 : | else pmv[4].x = pmv[4].y = 0; | ||
931 : | chl | 169 | |
932 : | chl | 530 | // [1] median prediction |
933 : | pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); | ||
934 : | chl | 169 | |
935 : | chl | 530 | pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask |
936 : | chl | 169 | |
937 : | chl | 530 | pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame |
938 : | pmv[2].y = EVEN(prevMB->mvs[0].y); | ||
939 : | chl | 169 | |
940 : | chl | 530 | if ((x != iWcount-1) && (y != iHcount-1)) { |
941 : | pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame | ||
942 : | Isibaar | 539 | pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
943 : | } else pmv[6].x = pmv[6].y = 0; | ||
944 : | chl | 530 | } |
945 : | chl | 169 | |
946 : | chl | 530 | static void |
947 : | SearchP(const uint8_t * const pRef, | ||
948 : | const uint8_t * const pRefH, | ||
949 : | const uint8_t * const pRefV, | ||
950 : | const uint8_t * const pRefHV, | ||
951 : | Isibaar | 580 | const uint8_t * const pRefQ, |
952 : | chl | 530 | const IMAGE * const pCur, |
953 : | const int x, | ||
954 : | const int y, | ||
955 : | const uint32_t MotionFlags, | ||
956 : | const uint32_t iQuant, | ||
957 : | Isibaar | 539 | SearchData * const Data, |
958 : | chl | 530 | const MBParam * const pParam, |
959 : | const MACROBLOCK * const pMBs, | ||
960 : | const MACROBLOCK * const prevMBs, | ||
961 : | int inter4v, | ||
962 : | MACROBLOCK * const pMB) | ||
963 : | { | ||
964 : | chl | 169 | |
965 : | chl | 530 | int i, iDirection = 255, mask, threshA; |
966 : | Isibaar | 539 | VECTOR pmv[7]; |
967 : | Isibaar | 3 | |
968 : | Isibaar | 580 | Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
969 : | |||
970 : | Isibaar | 539 | get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
971 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, | ||
972 : | Isibaar | 580 | pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
973 : | Isibaar | 3 | |
974 : | Isibaar | 539 | Data->predMV = pmv[0]; |
975 : | Isibaar | 580 | |
976 : | Isibaar | 539 | Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
977 : | Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16; | ||
978 : | Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; | ||
979 : | Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; | ||
980 : | Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; | ||
981 : | Isibaar | 580 | Data->RefQ = pRefQ; |
982 : | Isibaar | 3 | |
983 : | Isibaar | 539 | Data->iQuant = iQuant; |
984 : | chl | 169 | |
985 : | chl | 530 | if (!(MotionFlags & PMV_HALFPEL16)) { |
986 : | Isibaar | 539 | Data->min_dx = EVEN(Data->min_dx); |
987 : | Data->max_dx = EVEN(Data->max_dx); | ||
988 : | Data->min_dy = EVEN(Data->min_dy); | ||
989 : | Data->max_dy = EVEN(Data->max_dy); } | ||
990 : | edgomez | 568 | |
991 : | if (pMB->dquant != NO_CHANGE) inter4v = 0; | ||
992 : | edgomez | 195 | |
993 : | Isibaar | 539 | if (inter4v) CheckCandidate = CheckCandidate16; |
994 : | else CheckCandidate = CheckCandidate16no4v; | ||
995 : | chl | 169 | |
996 : | Isibaar | 580 | for(i = 0; i < 5; i++) |
997 : | Data->currentMV[i].x = Data->currentMV[i].y = 0; | ||
998 : | Isibaar | 539 | |
999 : | i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); | ||
1000 : | Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; | ||
1001 : | Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; | ||
1002 : | Data->iMinSAD[2] = pMB->sad8[1]; | ||
1003 : | Data->iMinSAD[3] = pMB->sad8[2]; | ||
1004 : | Data->iMinSAD[4] = pMB->sad8[3]; | ||
1005 : | |||
1006 : | chl | 530 | if ((x == 0) && (y == 0)) threshA = 512; |
1007 : | else { | ||
1008 : | edgomez | 568 | threshA = Data->temp[0]; // that's when we keep this SAD atm |
1009 : | chl | 530 | if (threshA < 512) threshA = 512; |
1010 : | if (threshA > 1024) threshA = 1024; } | ||
1011 : | Isibaar | 3 | |
1012 : | chl | 530 | PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
1013 : | prevMBs + x + y * pParam->mb_width); | ||
1014 : | edgomez | 195 | |
1015 : | edgomez | 568 | if (inter4v) CheckCandidate = CheckCandidate16; |
1016 : | else CheckCandidate = CheckCandidate16no4v; | ||
1017 : | |||
1018 : | |||
1019 : | chl | 530 | /* main loop. checking all predictions */ |
1020 : | |||
1021 : | for (i = 1; i < 7; i++) { | ||
1022 : | if (!(mask = make_mask(pmv, i)) ) continue; | ||
1023 : | edgomez | 568 | (*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1024 : | Isibaar | 539 | if (Data->iMinSAD[0] <= threshA) break; |
1025 : | chl | 530 | } |
1026 : | |||
1027 : | Isibaar | 539 | if ((Data->iMinSAD[0] <= threshA) || |
1028 : | (MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && | ||
1029 : | (Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) { | ||
1030 : | chl | 530 | inter4v = 0; |
1031 : | Isibaar | 539 | } else { |
1032 : | Isibaar | 3 | |
1033 : | Isibaar | 539 | MainSearchFunc * MainSearchPtr; |
1034 : | if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch; | ||
1035 : | else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; | ||
1036 : | else MainSearchPtr = DiamondSearch; | ||
1037 : | chl | 181 | |
1038 : | Isibaar | 539 | (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
1039 : | Isibaar | 3 | |
1040 : | chl | 530 | /* extended search, diamond starting in 0,0 and in prediction. |
1041 : | note that this search is/might be done in halfpel positions, | ||
1042 : | which makes it more different than the diamond above */ | ||
1043 : | chl | 259 | |
1044 : | Isibaar | 539 | if (MotionFlags & PMV_EXTSEARCH16) { |
1045 : | int32_t bSAD; | ||
1046 : | VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0]; | ||
1047 : | if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? | ||
1048 : | startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); | ||
1049 : | if (!(MVequal(startMV, backupMV))) { | ||
1050 : | bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; | ||
1051 : | Isibaar | 3 | |
1052 : | Isibaar | 539 | CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data); |
1053 : | (*MainSearchPtr)(startMV.x, startMV.y, Data, 255); | ||
1054 : | if (bSAD < Data->iMinSAD[0]) { | ||
1055 : | Data->currentMV[0] = backupMV; | ||
1056 : | Data->iMinSAD[0] = bSAD; } | ||
1057 : | } | ||
1058 : | Isibaar | 3 | |
1059 : | Isibaar | 539 | backupMV = Data->currentMV[0]; |
1060 : | if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1; | ||
1061 : | else startMV.x = startMV.y = 0; | ||
1062 : | if (!(MVequal(startMV, backupMV))) { | ||
1063 : | bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; | ||
1064 : | edgomez | 195 | |
1065 : | Isibaar | 539 | CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data); |
1066 : | (*MainSearchPtr)(startMV.x, startMV.y, Data, 255); | ||
1067 : | if (bSAD < Data->iMinSAD[0]) { | ||
1068 : | Data->currentMV[0] = backupMV; | ||
1069 : | Data->iMinSAD[0] = bSAD; } | ||
1070 : | } | ||
1071 : | Isibaar | 3 | } |
1072 : | } | ||
1073 : | |||
1074 : | Isibaar | 539 | if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1075 : | Isibaar | 3 | |
1076 : | Isibaar | 580 | for(i = 0; i < 5; i++) { |
1077 : | Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors | ||
1078 : | Data->currentQMV[i].y = 2 * Data->currentMV[i].y; | ||
1079 : | } | ||
1080 : | |||
1081 : | if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { | ||
1082 : | |||
1083 : | if(inter4v) | ||
1084 : | CheckCandidate = CheckCandidate16_qpel; | ||
1085 : | else | ||
1086 : | CheckCandidate = CheckCandidate16no4v_qpel; | ||
1087 : | |||
1088 : | QuarterpelRefine(Data); | ||
1089 : | } | ||
1090 : | |||
1091 : | Isibaar | 539 | if (inter4v) { |
1092 : | SearchData Data8; | ||
1093 : | Data8.iFcode = Data->iFcode; | ||
1094 : | Data8.iQuant = Data->iQuant; | ||
1095 : | Data8.iEdgedWidth = Data->iEdgedWidth; | ||
1096 : | Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); | ||
1097 : | Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); | ||
1098 : | Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); | ||
1099 : | Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); | ||
1100 : | } | ||
1101 : | Isibaar | 3 | |
1102 : | chl | 530 | if (!(inter4v) || |
1103 : | Isibaar | 539 | (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1104 : | Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { | ||
1105 : | chl | 530 | // INTER MODE |
1106 : | pMB->mode = MODE_INTER; | ||
1107 : | Isibaar | 539 | pMB->mvs[0] = pMB->mvs[1] |
1108 : | = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; | ||
1109 : | Isibaar | 3 | |
1110 : | Isibaar | 580 | pMB->qmvs[0] = pMB->qmvs[1] |
1111 : | = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; | ||
1112 : | |||
1113 : | chl | 530 | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1114 : | Isibaar | 539 | pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1115 : | Isibaar | 3 | |
1116 : | Isibaar | 580 | if(pParam->m_quarterpel) { |
1117 : | pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x; | ||
1118 : | pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y; | ||
1119 : | } | ||
1120 : | else { | ||
1121 : | pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; | ||
1122 : | pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; | ||
1123 : | } | ||
1124 : | edgomez | 195 | } else { |
1125 : | chl | 530 | // INTER4V MODE; all other things are already set in Search8 |
1126 : | pMB->mode = MODE_INTER4V; | ||
1127 : | Isibaar | 539 | pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
1128 : | Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; | ||
1129 : | edgomez | 78 | } |
1130 : | Isibaar | 3 | } |
1131 : | |||
1132 : | chl | 530 | static void |
1133 : | Search8(const SearchData * const OldData, | ||
1134 : | const int x, const int y, | ||
1135 : | const uint32_t MotionFlags, | ||
1136 : | const MBParam * const pParam, | ||
1137 : | MACROBLOCK * const pMB, | ||
1138 : | const MACROBLOCK * const pMBs, | ||
1139 : | Isibaar | 539 | const int block, |
1140 : | SearchData * const Data) | ||
1141 : | chl | 345 | { |
1142 : | Isibaar | 539 | Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1143 : | Isibaar | 580 | Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1144 : | Isibaar | 539 | Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1145 : | Data->currentMV = OldData->currentMV + 1 + block; | ||
1146 : | Isibaar | 580 | Data->currentQMV = OldData->currentQMV + 1 + block; |
1147 : | chl | 345 | |
1148 : | Isibaar | 580 | if (block != 0) { |
1149 : | if(pParam->m_quarterpel) { | ||
1150 : | *(Data->iMinSAD) += lambda_vec8[Data->iQuant] * | ||
1151 : | d_mv_bits( Data->currentQMV->x - Data->predQMV.x, | ||
1152 : | Data->currentQMV->y - Data->predQMV.y, | ||
1153 : | Data->iFcode); | ||
1154 : | } | ||
1155 : | else { | ||
1156 : | *(Data->iMinSAD) += lambda_vec8[Data->iQuant] * | ||
1157 : | d_mv_bits( Data->currentMV->x - Data->predMV.x, | ||
1158 : | Data->currentMV->y - Data->predMV.y, | ||
1159 : | Data->iFcode); | ||
1160 : | } | ||
1161 : | } | ||
1162 : | chl | 345 | |
1163 : | Isibaar | 580 | |
1164 : | chl | 530 | if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
1165 : | chl | 345 | |
1166 : | Isibaar | 539 | Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1167 : | Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1168 : | Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1169 : | Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1170 : | Isibaar | 580 | Data->RefQ = OldData->RefQ; |
1171 : | chl | 345 | |
1172 : | Isibaar | 539 | Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1173 : | chl | 530 | |
1174 : | Isibaar | 539 | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1175 : | Isibaar | 580 | pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
1176 : | chl | 345 | |
1177 : | chl | 530 | CheckCandidate = CheckCandidate8; |
1178 : | chl | 345 | |
1179 : | chl | 530 | if (MotionFlags & PMV_EXTSEARCH8) { |
1180 : | Isibaar | 580 | int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1181 : | |||
1182 : | chl | 530 | MainSearchFunc *MainSearchPtr; |
1183 : | if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; | ||
1184 : | else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; | ||
1185 : | else MainSearchPtr = DiamondSearch; | ||
1186 : | chl | 345 | |
1187 : | Isibaar | 580 | (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1188 : | chl | 345 | |
1189 : | Isibaar | 580 | if(*(Data->iMinSAD) < temp_sad) { |
1190 : | Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector | ||
1191 : | Data->currentQMV->y = 2 * Data->currentMV->y; | ||
1192 : | } | ||
1193 : | } | ||
1194 : | |||
1195 : | if (MotionFlags & PMV_HALFPELREFINE8) { | ||
1196 : | int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD | ||
1197 : | |||
1198 : | HalfpelRefine(Data); // perform halfpel refine of current best vector | ||
1199 : | |||
1200 : | if(*(Data->iMinSAD) < temp_sad) { // we have found a better match | ||
1201 : | Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector | ||
1202 : | Data->currentQMV->y = 2 * Data->currentMV->y; | ||
1203 : | } | ||
1204 : | } | ||
1205 : | |||
1206 : | if(pParam->m_quarterpel) { | ||
1207 : | if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && | ||
1208 : | (MotionFlags & PMV_QUARTERPELREFINE8)) { | ||
1209 : | |||
1210 : | CheckCandidate = CheckCandidate8_qpel; | ||
1211 : | QuarterpelRefine(Data); | ||
1212 : | } | ||
1213 : | } | ||
1214 : | chl | 345 | } |
1215 : | chl | 530 | |
1216 : | Isibaar | 580 | if(pParam->m_quarterpel) { |
1217 : | pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x; | ||
1218 : | pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y; | ||
1219 : | } | ||
1220 : | else { | ||
1221 : | pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; | ||
1222 : | pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; | ||
1223 : | } | ||
1224 : | |||
1225 : | Isibaar | 539 | pMB->mvs[block] = *(Data->currentMV); |
1226 : | Isibaar | 580 | pMB->qmvs[block] = *(Data->currentQMV); |
1227 : | |||
1228 : | Isibaar | 539 | pMB->sad8[block] = 4 * (*Data->iMinSAD); |
1229 : | chl | 345 | } |
1230 : | |||
1231 : | chl | 530 | /* B-frames code starts here */ |
1232 : | chl | 345 | |
1233 : | chl | 530 | static __inline VECTOR |
1234 : | ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode) | ||
1235 : | Isibaar | 3 | { |
1236 : | chl | 530 | /* the stupidiest function ever */ |
1237 : | if (mode == MODE_FORWARD) return pMB->mvs[0]; | ||
1238 : | else return pMB->b_mvs[0]; | ||
1239 : | Isibaar | 3 | } |
1240 : | |||
1241 : | chl | 530 | static void __inline |
1242 : | PreparePredictionsBF(VECTOR * const pmv, const int x, const int y, | ||
1243 : | const uint32_t iWcount, | ||
1244 : | const MACROBLOCK * const pMB, | ||
1245 : | const uint32_t mode_curr) | ||
1246 : | Isibaar | 3 | { |
1247 : | |||
1248 : | chl | 530 | // [0] is prediction |
1249 : | pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y); | ||
1250 : | Isibaar | 3 | |
1251 : | chl | 530 | pmv[1].x = pmv[1].y = 0; // [1] is zero |
1252 : | Isibaar | 3 | |
1253 : | chl | 530 | pmv[2] = ChoosePred(pMB, mode_curr); |
1254 : | pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); | ||
1255 : | edgomez | 195 | |
1256 : | chl | 530 | if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour |
1257 : | pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr); | ||
1258 : | Isibaar | 539 | pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); |
1259 : | } else pmv[3].x = pmv[3].y = 0; | ||
1260 : | Isibaar | 3 | |
1261 : | chl | 530 | if (y != 0) { |
1262 : | pmv[4] = ChoosePred(pMB-iWcount, mode_curr); | ||
1263 : | pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y); | ||
1264 : | } else pmv[4].x = pmv[4].y = 0; | ||
1265 : | Isibaar | 3 | |
1266 : | chl | 530 | if (x != 0) { |
1267 : | pmv[5] = ChoosePred(pMB-1, mode_curr); | ||
1268 : | pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y); | ||
1269 : | } else pmv[5].x = pmv[5].y = 0; | ||
1270 : | Isibaar | 3 | |
1271 : | chl | 530 | if ((x != 0)&&(y != 0)) { |
1272 : | pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr); | ||
1273 : | pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y); | ||
1274 : | } else pmv[6].x = pmv[6].y = 0; | ||
1275 : | edgomez | 195 | |
1276 : | chl | 530 | // more? |
1277 : | } | ||
1278 : | chl | 181 | |
1279 : | edgomez | 170 | |
1280 : | chl | 530 | /* search backward or forward, for b-frames */ |
1281 : | static void | ||
1282 : | SearchBF( const uint8_t * const pRef, | ||
1283 : | const uint8_t * const pRefH, | ||
1284 : | const uint8_t * const pRefV, | ||
1285 : | const uint8_t * const pRefHV, | ||
1286 : | const IMAGE * const pCur, | ||
1287 : | const int x, const int y, | ||
1288 : | const uint32_t MotionFlags, | ||
1289 : | const uint32_t iFcode, | ||
1290 : | const MBParam * const pParam, | ||
1291 : | MACROBLOCK * const pMB, | ||
1292 : | const VECTOR * const predMV, | ||
1293 : | int32_t * const best_sad, | ||
1294 : | h | 545 | const int32_t mode_current, |
1295 : | SearchData * const Data) | ||
1296 : | chl | 530 | { |
1297 : | Isibaar | 3 | |
1298 : | chl | 530 | const int32_t iEdgedWidth = pParam->edged_width; |
1299 : | |||
1300 : | int i, iDirection, mask; | ||
1301 : | h | 545 | VECTOR pmv[7]; |
1302 : | chl | 530 | MainSearchFunc *MainSearchPtr; |
1303 : | h | 545 | *Data->iMinSAD = MV_MAX_ERROR; |
1304 : | Data->iFcode = iFcode; | ||
1305 : | Isibaar | 3 | |
1306 : | h | 545 | Data->Ref = pRef + (x + y * iEdgedWidth) * 16; |
1307 : | Data->RefH = pRefH + (x + y * iEdgedWidth) * 16; | ||
1308 : | Data->RefV = pRefV + (x + y * iEdgedWidth) * 16; | ||
1309 : | Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16; | ||
1310 : | Isibaar | 3 | |
1311 : | h | 545 | Data->predMV = *predMV; |
1312 : | edgomez | 195 | |
1313 : | h | 545 | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1314 : | Isibaar | 580 | pParam->width, pParam->height, iFcode, pParam->m_quarterpel); |
1315 : | Isibaar | 3 | |
1316 : | h | 545 | pmv[0] = Data->predMV; |
1317 : | PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); | ||
1318 : | edgomez | 195 | |
1319 : | h | 545 | Data->currentMV->x = Data->currentMV->y = 0; |
1320 : | Isibaar | 3 | |
1321 : | chl | 530 | CheckCandidate = CheckCandidate16no4v; |
1322 : | chl | 181 | |
1323 : | chl | 530 | // main loop. checking all predictions |
1324 : | for (i = 0; i < 8; i++) { | ||
1325 : | if (!(mask = make_mask(pmv, i)) ) continue; | ||
1326 : | h | 545 | CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1327 : | edgomez | 78 | } |
1328 : | Isibaar | 3 | |
1329 : | chl | 530 | if (MotionFlags & PMV_USESQUARES16) |
1330 : | MainSearchPtr = SquareSearch; | ||
1331 : | else if (MotionFlags & PMV_ADVANCEDDIAMOND16) | ||
1332 : | MainSearchPtr = AdvDiamondSearch; | ||
1333 : | else MainSearchPtr = DiamondSearch; | ||
1334 : | chl | 96 | |
1335 : | h | 545 | (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1336 : | chl | 169 | |
1337 : | h | 545 | HalfpelRefine(Data); |
1338 : | chl | 169 | |
1339 : | chl | 530 | // three bits are needed to code backward mode. four for forward |
1340 : | // we treat the bits just like they were vector's | ||
1341 : | h | 545 | if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * lambda_vec16[Data->iQuant]; |
1342 : | else *Data->iMinSAD += 3 * lambda_vec16[Data->iQuant]; | ||
1343 : | chl | 169 | |
1344 : | |||
1345 : | h | 545 | if (*Data->iMinSAD < *best_sad) { |
1346 : | *best_sad = *Data->iMinSAD; | ||
1347 : | chl | 530 | pMB->mode = mode_current; |
1348 : | h | 545 | pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
1349 : | pMB->pmvs[0].y = Data->currentMV->y - predMV->y; | ||
1350 : | if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; | ||
1351 : | else pMB->b_mvs[0] = *Data->currentMV; | ||
1352 : | chl | 169 | } |
1353 : | chl | 530 | |
1354 : | } | ||
1355 : | chl | 169 | |
1356 : | chl | 530 | static int32_t |
1357 : | Isibaar | 539 | SearchDirect(const IMAGE * const f_Ref, |
1358 : | chl | 530 | const uint8_t * const f_RefH, |
1359 : | const uint8_t * const f_RefV, | ||
1360 : | const uint8_t * const f_RefHV, | ||
1361 : | Isibaar | 539 | const IMAGE * const b_Ref, |
1362 : | chl | 530 | const uint8_t * const b_RefH, |
1363 : | const uint8_t * const b_RefV, | ||
1364 : | const uint8_t * const b_RefHV, | ||
1365 : | const IMAGE * const pCur, | ||
1366 : | const int x, const int y, | ||
1367 : | const uint32_t MotionFlags, | ||
1368 : | const int32_t TRB, const int32_t TRD, | ||
1369 : | const MBParam * const pParam, | ||
1370 : | MACROBLOCK * const pMB, | ||
1371 : | const MACROBLOCK * const b_mb, | ||
1372 : | h | 545 | int32_t * const best_sad, |
1373 : | SearchData * const Data) | ||
1374 : | edgomez | 195 | |
1375 : | chl | 530 | { |
1376 : | h | 545 | int32_t skip_sad; |
1377 : | chl | 530 | int k; |
1378 : | h | 545 | |
1379 : | chl | 530 | MainSearchFunc *MainSearchPtr; |
1380 : | Isibaar | 3 | |
1381 : | h | 545 | *Data->iMinSAD = 256*4096; |
1382 : | Data->referencemv = b_mb->mvs; | ||
1383 : | chl | 140 | |
1384 : | h | 545 | Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
1385 : | Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16; | ||
1386 : | Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16; | ||
1387 : | Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16; | ||
1388 : | Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16; | ||
1389 : | Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16; | ||
1390 : | Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16; | ||
1391 : | Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16; | ||
1392 : | chl | 140 | |
1393 : | h | 545 | Data->max_dx = 2 * pParam->width - 2 * (x) * 16; |
1394 : | Data->max_dy = 2 * pParam->height - 2 * (y) * 16; | ||
1395 : | Data->min_dx = -(2 * 16 + 2 * (x) * 16); | ||
1396 : | Data->min_dy = -(2 * 16 + 2 * (y) * 16); | ||
1397 : | Isibaar | 3 | |
1398 : | chl | 530 | for (k = 0; k < 4; k++) { |
1399 : | h | 545 | pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); |
1400 : | pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD; | ||
1401 : | pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD); | ||
1402 : | pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD; | ||
1403 : | Isibaar | 3 | |
1404 : | h | 545 | if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx ) |
1405 : | || ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) { | ||
1406 : | Isibaar | 3 | |
1407 : | Isibaar | 539 | *best_sad = 256*4096; // in that case, we won't use direct mode |
1408 : | pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" | ||
1409 : | pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; | ||
1410 : | return 0; | ||
1411 : | } | ||
1412 : | if (b_mb->mode != MODE_INTER4V) { | ||
1413 : | pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; | ||
1414 : | pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0]; | ||
1415 : | h | 545 | Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0]; |
1416 : | Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0]; | ||
1417 : | Isibaar | 539 | break; |
1418 : | } | ||
1419 : | } | ||
1420 : | Isibaar | 3 | |
1421 : | Isibaar | 539 | if (b_mb->mode == MODE_INTER4V) |
1422 : | CheckCandidate = CheckCandidateDirect; | ||
1423 : | else CheckCandidate = CheckCandidateDirectno4v; | ||
1424 : | edgomez | 195 | |
1425 : | h | 545 | (*CheckCandidate)(0, 0, 255, &k, Data); |
1426 : | Isibaar | 3 | |
1427 : | chl | 530 | // skip decision |
1428 : | h | 545 | if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) { |
1429 : | Isibaar | 539 | //checking chroma. everything copied from MC |
1430 : | //this is not full chroma compensation, only it's fullpel approximation. should work though | ||
1431 : | int sum, dx, dy, b_dx, b_dy; | ||
1432 : | Isibaar | 3 | |
1433 : | Isibaar | 539 | sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1434 : | dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
1435 : | |||
1436 : | sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; | ||
1437 : | dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
1438 : | |||
1439 : | sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; | ||
1440 : | b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
1441 : | |||
1442 : | sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; | ||
1443 : | b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); | ||
1444 : | |||
1445 : | h | 545 | sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
1446 : | f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, | ||
1447 : | b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, | ||
1448 : | Data->iEdgedWidth/2); | ||
1449 : | sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), | ||
1450 : | f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, | ||
1451 : | b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, | ||
1452 : | Data->iEdgedWidth/2); | ||
1453 : | Isibaar | 539 | |
1454 : | h | 545 | if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) { |
1455 : | Isibaar | 539 | pMB->mode = MODE_DIRECT_NONE_MV; |
1456 : | h | 545 | return *Data->iMinSAD; |
1457 : | Isibaar | 539 | } |
1458 : | } | ||
1459 : | |||
1460 : | h | 545 | skip_sad = *Data->iMinSAD; |
1461 : | edgomez | 195 | |
1462 : | chl | 530 | // DIRECT MODE DELTA VECTOR SEARCH. |
1463 : | // This has to be made more effective, but at the moment I'm happy it's running at all | ||
1464 : | Isibaar | 3 | |
1465 : | chl | 530 | if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch; |
1466 : | else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; | ||
1467 : | else MainSearchPtr = DiamondSearch; | ||
1468 : | edgomez | 195 | |
1469 : | h | 545 | (*MainSearchPtr)(0, 0, Data, 255); |
1470 : | Isibaar | 3 | |
1471 : | h | 545 | HalfpelRefine(Data); |
1472 : | Isibaar | 3 | |
1473 : | h | 545 | *Data->iMinSAD += 1 * lambda_vec16[Data->iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's |
1474 : | *best_sad = *Data->iMinSAD; | ||
1475 : | Isibaar | 3 | |
1476 : | chl | 530 | if (b_mb->mode == MODE_INTER4V) |
1477 : | pMB->mode = MODE_DIRECT; | ||
1478 : | else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation | ||
1479 : | edgomez | 195 | |
1480 : | h | 545 | pMB->pmvs[3] = *Data->currentMV; |
1481 : | edgomez | 195 | |
1482 : | chl | 530 | for (k = 0; k < 4; k++) { |
1483 : | h | 545 | pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x; |
1484 : | pMB->b_mvs[k].x = ((Data->currentMV->x == 0) | ||
1485 : | ? Data->directmvB[k].x | ||
1486 : | : pMB->mvs[k].x - Data->referencemv[k].x); | ||
1487 : | pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y); | ||
1488 : | pMB->b_mvs[k].y = ((Data->currentMV->y == 0) | ||
1489 : | ? Data->directmvB[k].y | ||
1490 : | : pMB->mvs[k].y - Data->referencemv[k].y); | ||
1491 : | chl | 530 | if (b_mb->mode != MODE_INTER4V) { |
1492 : | pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; | ||
1493 : | pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0]; | ||
1494 : | break; | ||
1495 : | } | ||
1496 : | } | ||
1497 : | h | 545 | return skip_sad; |
1498 : | Isibaar | 3 | } |
1499 : | chl | 96 | |
1500 : | h | 545 | |
1501 : | chl | 530 | static __inline void |
1502 : | SearchInterpolate(const uint8_t * const f_Ref, | ||
1503 : | const uint8_t * const f_RefH, | ||
1504 : | const uint8_t * const f_RefV, | ||
1505 : | const uint8_t * const f_RefHV, | ||
1506 : | const uint8_t * const b_Ref, | ||
1507 : | const uint8_t * const b_RefH, | ||
1508 : | const uint8_t * const b_RefV, | ||
1509 : | const uint8_t * const b_RefHV, | ||
1510 : | const IMAGE * const pCur, | ||
1511 : | const int x, const int y, | ||
1512 : | const uint32_t fcode, | ||
1513 : | const uint32_t bcode, | ||
1514 : | const uint32_t MotionFlags, | ||
1515 : | const MBParam * const pParam, | ||
1516 : | const VECTOR * const f_predMV, | ||
1517 : | const VECTOR * const b_predMV, | ||
1518 : | MACROBLOCK * const pMB, | ||
1519 : | h | 545 | int32_t * const best_sad, |
1520 : | SearchData * const fData) | ||
1521 : | chl | 530 | |
1522 : | chl | 96 | { |
1523 : | |||
1524 : | edgomez | 195 | const int32_t iEdgedWidth = pParam->edged_width; |
1525 : | chl | 96 | |
1526 : | chl | 530 | int iDirection, i, j; |
1527 : | h | 545 | SearchData bData; |
1528 : | chl | 96 | |
1529 : | h | 545 | bData.iMinSAD = fData->iMinSAD; |
1530 : | *bData.iMinSAD = 4096*256; | ||
1531 : | bData.Cur = fData->Cur; | ||
1532 : | fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; | ||
1533 : | bData.currentMV = fData->currentMV + 1; | ||
1534 : | bData.iQuant = fData->iQuant; | ||
1535 : | fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; | ||
1536 : | edgomez | 195 | |
1537 : | h | 545 | bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
1538 : | bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16; | ||
1539 : | bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16; | ||
1540 : | bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; | ||
1541 : | bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16; | ||
1542 : | bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; | ||
1543 : | bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; | ||
1544 : | bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; | ||
1545 : | chl | 96 | |
1546 : | h | 545 | bData.bpredMV = fData->predMV = *f_predMV; |
1547 : | fData->bpredMV = bData.predMV = *b_predMV; | ||
1548 : | edgomez | 195 | |
1549 : | h | 545 | fData->currentMV[0] = pMB->mvs[0]; |
1550 : | fData->currentMV[1] = pMB->b_mvs[0]; | ||
1551 : | Isibaar | 580 | get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel); |
1552 : | get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel); | ||
1553 : | chl | 96 | |
1554 : | h | 545 | if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1555 : | if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; | ||
1556 : | if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx; | ||
1557 : | if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy; | ||
1558 : | Isibaar | 539 | |
1559 : | h | 545 | if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
1560 : | if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy; | ||
1561 : | if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx; | ||
1562 : | if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy; | ||
1563 : | Isibaar | 539 | |
1564 : | h | 545 | CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1565 : | chl | 96 | |
1566 : | chl | 530 | //diamond. I wish we could use normal mainsearch functions (square, advdiamond) |
1567 : | chl | 96 | |
1568 : | chl | 530 | do { |
1569 : | iDirection = 255; | ||
1570 : | // forward MV moves | ||
1571 : | h | 545 | i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
1572 : | chl | 96 | |
1573 : | h | 545 | CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
1574 : | CheckCandidateInt(i, j + 1, 0, &iDirection, fData); | ||
1575 : | CheckCandidateInt(i - 1, j, 0, &iDirection, fData); | ||
1576 : | CheckCandidateInt(i, j - 1, 0, &iDirection, fData); | ||
1577 : | chl | 96 | |
1578 : | chl | 530 | // backward MV moves |
1579 : | h | 545 | i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
1580 : | fData->currentMV[2] = fData->currentMV[0]; | ||
1581 : | chl | 326 | |
1582 : | Isibaar | 539 | CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
1583 : | CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); | ||
1584 : | CheckCandidateInt(i - 1, j, 0, &iDirection, &bData); | ||
1585 : | CheckCandidateInt(i, j - 1, 0, &iDirection, &bData); | ||
1586 : | chl | 96 | |
1587 : | chl | 530 | } while (!(iDirection)); |
1588 : | edgomez | 195 | |
1589 : | chl | 530 | // two bits are needed to code interpolate mode. we treat the bits just like they were vector's |
1590 : | h | 545 | *fData->iMinSAD += 2 * lambda_vec16[fData->iQuant]; |
1591 : | if (*fData->iMinSAD < *best_sad) { | ||
1592 : | *best_sad = *fData->iMinSAD; | ||
1593 : | pMB->mvs[0] = fData->currentMV[0]; | ||
1594 : | pMB->b_mvs[0] = fData->currentMV[1]; | ||
1595 : | chl | 530 | pMB->mode = MODE_INTERPOLATE; |
1596 : | chl | 96 | |
1597 : | chl | 530 | pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
1598 : | pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y; | ||
1599 : | pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x; | ||
1600 : | pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y; | ||
1601 : | } | ||
1602 : | } | ||
1603 : | chl | 96 | |
1604 : | h | 545 | |
1605 : | chl | 530 | void |
1606 : | MotionEstimationBVOP(MBParam * const pParam, | ||
1607 : | FRAMEINFO * const frame, | ||
1608 : | const int32_t time_bp, | ||
1609 : | const int32_t time_pp, | ||
1610 : | // forward (past) reference | ||
1611 : | const MACROBLOCK * const f_mbs, | ||
1612 : | const IMAGE * const f_ref, | ||
1613 : | const IMAGE * const f_refH, | ||
1614 : | const IMAGE * const f_refV, | ||
1615 : | const IMAGE * const f_refHV, | ||
1616 : | // backward (future) reference | ||
1617 : | const MACROBLOCK * const b_mbs, | ||
1618 : | const IMAGE * const b_ref, | ||
1619 : | const IMAGE * const b_refH, | ||
1620 : | const IMAGE * const b_refV, | ||
1621 : | const IMAGE * const b_refHV) | ||
1622 : | { | ||
1623 : | uint32_t i, j; | ||
1624 : | int32_t best_sad, skip_sad; | ||
1625 : | int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; | ||
1626 : | static const VECTOR zeroMV={0,0}; | ||
1627 : | chl | 96 | |
1628 : | chl | 530 | VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
1629 : | chl | 96 | |
1630 : | chl | 530 | const int32_t TRB = time_pp - time_bp; |
1631 : | const int32_t TRD = time_pp; | ||
1632 : | chl | 96 | |
1633 : | h | 545 | // some pre-inintialized data for the rest of the search |
1634 : | |||
1635 : | SearchData Data; | ||
1636 : | int32_t iMinSAD; | ||
1637 : | VECTOR currentMV[3]; | ||
1638 : | Data.iEdgedWidth = pParam->edged_width; | ||
1639 : | Data.currentMV = currentMV; | ||
1640 : | Data.iMinSAD = &iMinSAD; | ||
1641 : | Data.iQuant = frame->quant; | ||
1642 : | |||
1643 : | chl | 530 | // note: i==horizontal, j==vertical |
1644 : | chl | 96 | |
1645 : | chl | 530 | for (j = 0; j < pParam->mb_height; j++) { |
1646 : | chl | 96 | |
1647 : | chl | 530 | f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */ |
1648 : | edgomez | 195 | |
1649 : | chl | 530 | for (i = 0; i < pParam->mb_width; i++) { |
1650 : | MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; | ||
1651 : | const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; | ||
1652 : | edgomez | 195 | |
1653 : | chl | 530 | /* special case, if collocated block is SKIPed: encoding is forward (0,0), cpb=0 without further ado */ |
1654 : | if (b_mb->mode == MODE_NOT_CODED) { | ||
1655 : | pMB->mode = MODE_NOT_CODED; | ||
1656 : | continue; | ||
1657 : | chl | 96 | } |
1658 : | |||
1659 : | h | 545 | Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
1660 : | chl | 530 | /* direct search comes first, because it (1) checks for SKIP-mode |
1661 : | and (2) sets very good predictions for forward and backward search */ | ||
1662 : | chl | 96 | |
1663 : | Isibaar | 539 | skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
1664 : | b_ref, b_refH->y, b_refV->y, b_refHV->y, | ||
1665 : | chl | 530 | &frame->image, |
1666 : | i, j, | ||
1667 : | frame->motion_flags, | ||
1668 : | TRB, TRD, | ||
1669 : | pParam, | ||
1670 : | pMB, b_mb, | ||
1671 : | h | 545 | &best_sad, |
1672 : | &Data); | ||
1673 : | chl | 96 | |
1674 : | Isibaar | 539 | if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
1675 : | chl | 96 | |
1676 : | chl | 530 | // best_sad = 256*4096; //uncomment to disable Directsearch. |
1677 : | // To disable any other mode, just comment the function call | ||
1678 : | chl | 96 | |
1679 : | chl | 530 | // forward search |
1680 : | SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, | ||
1681 : | &frame->image, i, j, | ||
1682 : | frame->motion_flags, | ||
1683 : | h | 545 | frame->fcode, pParam, |
1684 : | chl | 530 | pMB, &f_predMV, &best_sad, |
1685 : | h | 545 | MODE_FORWARD, &Data); |
1686 : | chl | 96 | |
1687 : | chl | 530 | // backward search |
1688 : | SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
1689 : | &frame->image, i, j, | ||
1690 : | frame->motion_flags, | ||
1691 : | h | 545 | frame->bcode, pParam, |
1692 : | chl | 530 | pMB, &b_predMV, &best_sad, |
1693 : | h | 545 | MODE_BACKWARD, &Data); |
1694 : | chl | 96 | |
1695 : | chl | 530 | // interpolate search comes last, because it uses data from forward and backward as prediction |
1696 : | chl | 96 | |
1697 : | chl | 530 | SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
1698 : | b_ref->y, b_refH->y, b_refV->y, b_refHV->y, | ||
1699 : | &frame->image, | ||
1700 : | i, j, | ||
1701 : | frame->fcode, frame->bcode, | ||
1702 : | frame->motion_flags, | ||
1703 : | h | 545 | pParam, |
1704 : | chl | 530 | &f_predMV, &b_predMV, |
1705 : | h | 545 | pMB, &best_sad, |
1706 : | &Data); | ||
1707 : | chl | 96 | |
1708 : | chl | 530 | switch (pMB->mode) { |
1709 : | case MODE_FORWARD: | ||
1710 : | f_count++; | ||
1711 : | f_predMV = pMB->mvs[0]; | ||
1712 : | break; | ||
1713 : | case MODE_BACKWARD: | ||
1714 : | b_count++; | ||
1715 : | b_predMV = pMB->b_mvs[0]; | ||
1716 : | break; | ||
1717 : | case MODE_INTERPOLATE: | ||
1718 : | i_count++; | ||
1719 : | f_predMV = pMB->mvs[0]; | ||
1720 : | b_predMV = pMB->b_mvs[0]; | ||
1721 : | break; | ||
1722 : | case MODE_DIRECT: | ||
1723 : | case MODE_DIRECT_NO4V: | ||
1724 : | d_count++; | ||
1725 : | break; | ||
1726 : | default: | ||
1727 : | break; | ||
1728 : | chl | 96 | } |
1729 : | } | ||
1730 : | } | ||
1731 : | |||
1732 : | chl | 530 | // fprintf(debug,"B-Stat: F: %04d B: %04d I: %04d D: %04d, N: %04d\n", |
1733 : | // f_count,b_count,i_count,d_count,n_count); | ||
1734 : | chl | 96 | |
1735 : | } | ||
1736 : | |||
1737 : | chl | 530 | /* Hinted ME starts here */ |
1738 : | chl | 96 | |
1739 : | chl | 530 | static __inline void |
1740 : | Search8hinted( const SearchData * const OldData, | ||
1741 : | const int x, const int y, | ||
1742 : | const uint32_t MotionFlags, | ||
1743 : | const MBParam * const pParam, | ||
1744 : | MACROBLOCK * const pMB, | ||
1745 : | const MACROBLOCK * const pMBs, | ||
1746 : | const int block) | ||
1747 : | chl | 96 | { |
1748 : | chl | 530 | SearchData Data; |
1749 : | MainSearchFunc *MainSearchPtr; | ||
1750 : | chl | 141 | |
1751 : | chl | 530 | Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1752 : | Data.iMinSAD = OldData->iMinSAD + 1 + block; | ||
1753 : | Data.currentMV = OldData->currentMV+1+block; | ||
1754 : | Data.iFcode = OldData->iFcode; | ||
1755 : | Data.iQuant = OldData->iQuant; | ||
1756 : | chl | 96 | |
1757 : | chl | 530 | Data.Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1758 : | Data.RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1759 : | Data.RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1760 : | Data.RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1761 : | Data.iEdgedWidth = pParam->edged_width; | ||
1762 : | Data.Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); | ||
1763 : | chl | 96 | |
1764 : | chl | 530 | CheckCandidate = CheckCandidate8; |
1765 : | edgomez | 195 | |
1766 : | chl | 530 | if (block != 0) |
1767 : | *(Data.iMinSAD) += lambda_vec8[Data.iQuant] * | ||
1768 : | d_mv_bits( Data.currentMV->x - Data.predMV.x, | ||
1769 : | Data.currentMV->y - Data.predMV.y, | ||
1770 : | Data.iFcode); | ||
1771 : | edgomez | 195 | |
1772 : | |||
1773 : | chl | 530 | get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8, |
1774 : | Isibaar | 580 | pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
1775 : | chl | 96 | |
1776 : | chl | 530 | if (pMB->mode == MODE_INTER4V) { |
1777 : | int dummy; | ||
1778 : | CheckCandidate8(pMB->mvs[block].x, pMB->mvs[block].y, 0, &dummy, &Data); } | ||
1779 : | chl | 96 | |
1780 : | chl | 530 | if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
1781 : | else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; | ||
1782 : | else MainSearchPtr = DiamondSearch; | ||
1783 : | chl | 96 | |
1784 : | chl | 530 | (*MainSearchPtr)(Data.currentMV->x, Data.currentMV->y, &Data, 255); |
1785 : | edgomez | 195 | |
1786 : | chl | 530 | if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data); |
1787 : | chl | 96 | |
1788 : | chl | 530 | pMB->pmvs[block].x = Data.currentMV->x - Data.predMV.x; |
1789 : | pMB->pmvs[block].y = Data.currentMV->y - Data.predMV.y; | ||
1790 : | pMB->mvs[block] = *(Data.currentMV); | ||
1791 : | pMB->sad8[block] = 4 * (*(Data.iMinSAD)); | ||
1792 : | chl | 96 | } |
1793 : | |||
1794 : | suxen_drol | 118 | |
1795 : | chl | 530 | static void |
1796 : | SearchPhinted ( const uint8_t * const pRef, | ||
1797 : | chl | 289 | const uint8_t * const pRefH, |
1798 : | const uint8_t * const pRefV, | ||
1799 : | const uint8_t * const pRefHV, | ||
1800 : | const IMAGE * const pCur, | ||
1801 : | const int x, | ||
1802 : | const int y, | ||
1803 : | const uint32_t MotionFlags, | ||
1804 : | const uint32_t iQuant, | ||
1805 : | const MBParam * const pParam, | ||
1806 : | const MACROBLOCK * const pMBs, | ||
1807 : | chl | 530 | int inter4v, |
1808 : | h | 545 | MACROBLOCK * const pMB, |
1809 : | edgomez | 568 | SearchData * const Data) |
1810 : | chl | 289 | { |
1811 : | chl | 530 | |
1812 : | chl | 289 | const int32_t iEdgedWidth = pParam->edged_width; |
1813 : | chl | 530 | |
1814 : | edgomez | 568 | int i, t; |
1815 : | chl | 530 | MainSearchFunc * MainSearchPtr; |
1816 : | suxen_drol | 118 | |
1817 : | h | 545 | Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1818 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, | ||
1819 : | Isibaar | 580 | pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
1820 : | suxen_drol | 118 | |
1821 : | h | 545 | Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
1822 : | Data->Ref = pRef + (x + iEdgedWidth*y)*16; | ||
1823 : | Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; | ||
1824 : | Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; | ||
1825 : | Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; | ||
1826 : | Data->iQuant = iQuant; | ||
1827 : | edgomez | 568 | |
1828 : | chl | 530 | if (!(MotionFlags & PMV_HALFPEL16)) { |
1829 : | h | 545 | Data->min_dx = EVEN(Data->min_dx); |
1830 : | Data->max_dx = EVEN(Data->max_dx); | ||
1831 : | Data->min_dy = EVEN(Data->min_dy); | ||
1832 : | Data->max_dy = EVEN(Data->max_dy); | ||
1833 : | chl | 530 | } |
1834 : | chl | 289 | |
1835 : | h | 545 | for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
1836 : | chl | 289 | |
1837 : | chl | 530 | if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1838 : | chl | 289 | |
1839 : | chl | 530 | if (inter4v) |
1840 : | CheckCandidate = CheckCandidate16; | ||
1841 : | else CheckCandidate = CheckCandidate16no4v; | ||
1842 : | chl | 289 | |
1843 : | |||
1844 : | chl | 530 | pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
1845 : | pMB->mvs[0].y = EVEN(pMB->mvs[0].y); | ||
1846 : | h | 545 | if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed |
1847 : | if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx; | ||
1848 : | if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy; | ||
1849 : | if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy; | ||
1850 : | chl | 370 | |
1851 : | edgomez | 568 | (*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data); |
1852 : | chl | 289 | |
1853 : | chl | 530 | if (pMB->mode == MODE_INTER4V) |
1854 : | for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search | ||
1855 : | pMB->mvs[i].x = EVEN(pMB->mvs[i].x); | ||
1856 : | pMB->mvs[i].y = EVEN(pMB->mvs[i].y); | ||
1857 : | if (!(make_mask(pMB->mvs, i))) | ||
1858 : | edgomez | 568 | (*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data); |
1859 : | chl | 289 | } |
1860 : | |||
1861 : | if (MotionFlags & PMV_USESQUARES16) | ||
1862 : | chl | 530 | MainSearchPtr = SquareSearch; |
1863 : | chl | 289 | else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
1864 : | chl | 530 | MainSearchPtr = AdvDiamondSearch; |
1865 : | else MainSearchPtr = DiamondSearch; | ||
1866 : | chl | 289 | |
1867 : | h | 545 | (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1868 : | chl | 289 | |
1869 : | h | 545 | if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1870 : | chl | 289 | |
1871 : | chl | 530 | if (inter4v) |
1872 : | for(i = 0; i < 4; i++) | ||
1873 : | h | 545 | Search8hinted(Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i); |
1874 : | chl | 289 | |
1875 : | chl | 530 | if (!(inter4v) || |
1876 : | h | 545 | (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
1877 : | Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { | ||
1878 : | chl | 530 | // INTER MODE |
1879 : | chl | 289 | |
1880 : | chl | 530 | pMB->mode = MODE_INTER; |
1881 : | Isibaar | 539 | pMB->mvs[0] = pMB->mvs[1] |
1882 : | h | 545 | = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1883 : | chl | 289 | |
1884 : | chl | 530 | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1885 : | h | 545 | pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1886 : | chl | 289 | |
1887 : | h | 545 | pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1888 : | pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; | ||
1889 : | chl | 530 | } else { |
1890 : | // INTER4V MODE; all other things are already set in Search8hinted | ||
1891 : | pMB->mode = MODE_INTER4V; | ||
1892 : | h | 545 | pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] |
1893 : | + Data->iMinSAD[4] + IMV16X16 * iQuant; | ||
1894 : | chl | 289 | } |
1895 : | |||
1896 : | } | ||
1897 : | |||
1898 : | edgomez | 195 | void |
1899 : | chl | 530 | MotionEstimationHinted( MBParam * const pParam, |
1900 : | FRAMEINFO * const current, | ||
1901 : | FRAMEINFO * const reference, | ||
1902 : | const IMAGE * const pRefH, | ||
1903 : | const IMAGE * const pRefV, | ||
1904 : | const IMAGE * const pRefHV) | ||
1905 : | suxen_drol | 118 | { |
1906 : | chl | 530 | MACROBLOCK *const pMBs = current->mbs; |
1907 : | const IMAGE *const pCurrent = ¤t->image; | ||
1908 : | const IMAGE *const pRef = &reference->image; | ||
1909 : | suxen_drol | 118 | |
1910 : | chl | 530 | uint32_t x, y; |
1911 : | syskin | 574 | int32_t temp[5], quant = current->quant; |
1912 : | h | 545 | int32_t iMinSAD[5]; |
1913 : | VECTOR currentMV[5]; | ||
1914 : | SearchData Data; | ||
1915 : | Data.iEdgedWidth = pParam->edged_width; | ||
1916 : | Data.currentMV = currentMV; | ||
1917 : | Data.iMinSAD = iMinSAD; | ||
1918 : | Data.temp = temp; | ||
1919 : | Data.iFcode = current->fcode; | ||
1920 : | |||
1921 : | chl | 530 | if (sadInit) (*sadInit) (); |
1922 : | suxen_drol | 118 | |
1923 : | chl | 530 | for (y = 0; y < pParam->mb_height; y++) { |
1924 : | for (x = 0; x < pParam->mb_width; x++) { | ||
1925 : | chl | 312 | |
1926 : | chl | 530 | MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
1927 : | suxen_drol | 118 | |
1928 : | chl | 530 | //intra mode is copied from the first pass. At least for the time being |
1929 : | if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue; | ||
1930 : | chl | 346 | |
1931 : | syskin | 574 | |
1932 : | chl | 530 | if (!(current->global_flags & XVID_LUMIMASKING)) { |
1933 : | pMB->dquant = NO_CHANGE; | ||
1934 : | pMB->quant = current->quant; } | ||
1935 : | syskin | 574 | else |
1936 : | if (pMB->dquant != NO_CHANGE) { | ||
1937 : | quant += DQtab[pMB->dquant]; | ||
1938 : | if (quant > 31) quant = 31; | ||
1939 : | else if (quant < 1) quant = 1; | ||
1940 : | pMB->quant = quant; | ||
1941 : | } | ||
1942 : | chl | 341 | |
1943 : | Isibaar | 539 | SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1944 : | y, current->motion_flags, pMB->quant, | ||
1945 : | h | 545 | pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
1946 : | &Data); | ||
1947 : | chl | 317 | |
1948 : | Isibaar | 539 | } |
1949 : | } | ||
1950 : | } | ||
1951 : | suxen_drol | 118 | |
1952 : | Isibaar | 539 | static __inline int |
1953 : | MEanalyzeMB ( const uint8_t * const pRef, | ||
1954 : | const uint8_t * const pCur, | ||
1955 : | const int x, | ||
1956 : | const int y, | ||
1957 : | const MBParam * const pParam, | ||
1958 : | const MACROBLOCK * const pMBs, | ||
1959 : | h | 545 | MACROBLOCK * const pMB, |
1960 : | SearchData * const Data) | ||
1961 : | Isibaar | 539 | { |
1962 : | chl | 317 | |
1963 : | Isibaar | 539 | int i, mask; |
1964 : | h | 545 | VECTOR pmv[3]; |
1965 : | suxen_drol | 118 | |
1966 : | h | 545 | *(Data->iMinSAD) = MV_MAX_ERROR; |
1967 : | Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); | ||
1968 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, | ||
1969 : | Isibaar | 580 | pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
1970 : | chl | 326 | |
1971 : | h | 545 | Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
1972 : | Data->Ref = pRef + (x + y * pParam->edged_width) * 16; | ||
1973 : | Isibaar | 539 | |
1974 : | CheckCandidate = CheckCandidate16no4vI; | ||
1975 : | chl | 326 | |
1976 : | Isibaar | 539 | pmv[1].x = EVEN(pMB->mvs[0].x); |
1977 : | pmv[1].y = EVEN(pMB->mvs[0].y); | ||
1978 : | h | 545 | pmv[0].x = EVEN(Data->predMV.x); |
1979 : | pmv[0].y = EVEN(Data->predMV.y); | ||
1980 : | Isibaar | 539 | pmv[2].x = pmv[2].y = 0; |
1981 : | chl | 326 | |
1982 : | h | 545 | CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data); |
1983 : | Isibaar | 539 | if (!(mask = make_mask(pmv, 1))) |
1984 : | h | 545 | CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data); |
1985 : | Isibaar | 539 | if (!(mask = make_mask(pmv, 2))) |
1986 : | h | 545 | CheckCandidate16no4vI(0, 0, mask, &i, Data); |
1987 : | Isibaar | 539 | |
1988 : | h | 545 | DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
1989 : | Isibaar | 539 | |
1990 : | pMB->mvs[0] = pMB->mvs[1] | ||
1991 : | h | 545 | = pMB->mvs[2] = pMB->mvs[3] = *Data->currentMV; // all, for future get_pmv() |
1992 : | Isibaar | 539 | |
1993 : | h | 545 | return *(Data->iMinSAD); |
1994 : | Isibaar | 539 | } |
1995 : | |||
1996 : | #define INTRA_THRESH 1350 | ||
1997 : | #define INTER_THRESH 900 | ||
1998 : | |||
1999 : | int | ||
2000 : | MEanalysis( const IMAGE * const pRef, | ||
2001 : | const IMAGE * const pCurrent, | ||
2002 : | MBParam * const pParam, | ||
2003 : | MACROBLOCK * const pMBs, | ||
2004 : | const uint32_t iFcode) | ||
2005 : | { | ||
2006 : | uint32_t x, y, intra = 0; | ||
2007 : | int sSAD = 0; | ||
2008 : | h | 545 | |
2009 : | VECTOR currentMV; | ||
2010 : | int32_t iMinSAD; | ||
2011 : | SearchData Data; | ||
2012 : | Data.iEdgedWidth = pParam->edged_width; | ||
2013 : | Data.currentMV = ¤tMV; | ||
2014 : | Data.iMinSAD = &iMinSAD; | ||
2015 : | Data.iFcode = iFcode; | ||
2016 : | Data.iQuant = 2; | ||
2017 : | |||
2018 : | Isibaar | 539 | if (sadInit) (*sadInit) (); |
2019 : | |||
2020 : | for (y = 0; y < pParam->mb_height-1; y++) { | ||
2021 : | for (x = 0; x < pParam->mb_width; x++) { | ||
2022 : | int sad, dev; | ||
2023 : | MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; | ||
2024 : | |||
2025 : | sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, | ||
2026 : | h | 545 | pParam, pMBs, pMB, &Data); |
2027 : | Isibaar | 539 | |
2028 : | if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work | ||
2029 : | if (sad > INTRA_THRESH) { | ||
2030 : | dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, | ||
2031 : | pParam->edged_width); | ||
2032 : | if (dev + INTRA_THRESH < sad) intra++; | ||
2033 : | if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame | ||
2034 : | } | ||
2035 : | sSAD += sad; | ||
2036 : | } | ||
2037 : | |||
2038 : | suxen_drol | 118 | } |
2039 : | } | ||
2040 : | Isibaar | 539 | sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
2041 : | if (sSAD > INTER_THRESH ) return 1; //P frame | ||
2042 : | emms(); | ||
2043 : | return 0; // B frame | ||
2044 : | |||
2045 : | suxen_drol | 118 | } |
2046 : | syskin | 576 | |
2047 : | int | ||
2048 : | FindFcode( const MBParam * const pParam, | ||
2049 : | const FRAMEINFO * const current) | ||
2050 : | { | ||
2051 : | uint32_t x, y; | ||
2052 : | int max = 0, min = 0, i; | ||
2053 : | |||
2054 : | for (y = 0; y < pParam->mb_height; y++) { | ||
2055 : | for (x = 0; x < pParam->mb_width; x++) { | ||
2056 : | |||
2057 : | MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; | ||
2058 : | for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) { | ||
2059 : | if (pMB->mvs[i].x > max) max = pMB->mvs[i].x; | ||
2060 : | if (pMB->mvs[i].y > max) max = pMB->mvs[i].y; | ||
2061 : | |||
2062 : | if (pMB->mvs[i].x < min) min = pMB->mvs[i].x; | ||
2063 : | if (pMB->mvs[i].y < min) min = pMB->mvs[i].y; | ||
2064 : | } | ||
2065 : | } | ||
2066 : | } | ||
2067 : | |||
2068 : | min = -min; | ||
2069 : | max += 1; | ||
2070 : | if (min > max) max = min; | ||
2071 : | |||
2072 : | for (i = 1; (max > 32 << (i - 1)); i++); | ||
2073 : | return i; | ||
2074 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |