Annotation of /branches/dev-api-3/xvidcore/src/motion/motion_est.c

Revision 721 - (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 :	suxen_drol	721	#include <string.h> // memcpy
35 :	Isibaar	3
36 :			#include "../encoder.h"
37 :			#include "../utils/mbfunctions.h"
38 :			#include "../prediction/mbprediction.h"
39 :			#include "../global.h"
40 :			#include "../utils/timer.h"
41 :	Isibaar	580	#include "../image/interpolate8x8.h"
42 :	chl	530	#include "motion_est.h"
43 :	suxen_drol	118	#include "motion.h"
44 :	Isibaar	3	#include "sad.h"
45 :	Isibaar	539	#include "../utils/emms.h"
46 :	Isibaar	3
47 :	chl	530	#define INITIAL_SKIP_THRESH (10)
48 :			#define FINAL_SKIP_THRESH (50)
49 :			#define MAX_SAD00_FOR_SKIP (20)
50 :	Isibaar	539	#define MAX_CHROMA_SAD_FOR_SKIP (22)
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 :	chl	530	static __inline int
56 :	syskin	712	d_mv_bits(int x, int y, const uint32_t iFcode, const int qpel, const int rrv)
57 :	Isibaar	3	{
58 :	chl	530	int xb, yb;
59 :	syskin	712	if (qpel) { x = 2; y = 2;}
60 :			else if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
61 :	chl	530
62 :			if (x == 0) xb = 1;
63 :			else {
64 :			if (x < 0) x = -x;
65 :			x += (1 << (iFcode - 1)) - 1;
66 :			x >>= (iFcode - 1);
67 :			if (x > 32) x = 32;
68 :			xb = mvtab[x] + iFcode;
69 :	edgomez	195	}
70 :	Isibaar	3
71 :	chl	530	if (y == 0) yb = 1;
72 :			else {
73 :			if (y < 0) y = -y;
74 :			y += (1 << (iFcode - 1)) - 1;
75 :			y >>= (iFcode - 1);
76 :			if (y > 32) y = 32;
77 :			yb = mvtab[y] + iFcode;
78 :			}
79 :			return xb + yb;
80 :	Isibaar	3	}
81 :
82 :	syskin	628	static int32_t
83 :			ChromaSAD(int dx, int dy, const SearchData * const data)
84 :	Isibaar	3	{
85 :	syskin	628	int sad;
86 :			dx = (dx >> 1) + roundtab_79[dx & 0x3];
87 :			dy = (dy >> 1) + roundtab_79[dy & 0x3];
88 :	Isibaar	3
89 :	syskin	712	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
90 :
91 :			switch (((dx & 1) << 1) \| (dy & 1)) {
92 :	syskin	628	case 0:
93 :			sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
94 :			sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
95 :			break;
96 :			case 1:
97 :			dx = dx / 2; dy = (dy - 1) / 2;
98 :			sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
99 :			sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
100 :			break;
101 :			case 2:
102 :			dx = (dx - 1) / 2; dy = dy / 2;
103 :			sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
104 :			sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
105 :			break;
106 :			default:
107 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
108 :			interpolate8x8_halfpel_hv(data->RefQ,
109 :			data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
110 :			data->rounding);
111 :			sad = sad8(data->CurU, data->RefQ, data->iEdgedWidth/2);
112 :			interpolate8x8_halfpel_hv(data->RefQ,
113 :			data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
114 :			data->rounding);
115 :			sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2);
116 :			break;
117 :	chl	530	}
118 :	syskin	712	data->temp[5] = dx; data->temp[6] = dy; data->temp[7] = sad; //backup
119 :	syskin	628	return sad;
120 :			}
121 :	Isibaar	3
122 :	syskin	663	static __inline const uint8_t *
123 :			GetReference(const int x, const int y, const int dir, const SearchData * const data)
124 :	Isibaar	606	{
125 :	syskin	663	// dir : 0 = forward, 1 = backward
126 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
127 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
128 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
129 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
130 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
131 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
132 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
133 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
134 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
135 :	Isibaar	606	}
136 :			}
137 :
138 :	syskin	663	static uint8_t *
139 :			Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data)
140 :	chl	530	{
141 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
142 :	syskin	668	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
143 :	syskin	663	const int32_t iEdgedWidth = data->iEdgedWidth;
144 :			const uint32_t rounding = data->rounding;
145 :			const int halfpel_x = x/2;
146 :			const int halfpel_y = y/2;
147 :			const uint8_t ref1, ref2, ref3, ref4;
148 :	chl	326
149 :	syskin	663	ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases
150 :			ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
151 :			switch( ((x&1)<<1) + (y&1) ) {
152 :			case 0: // pure halfpel position
153 :			Reference = (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data);
154 :			Reference += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
155 :			break;
156 :	Isibaar	3
157 :	syskin	663	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
158 :			ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
159 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
160 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
161 :	syskin	663	break;
162 :	Isibaar	3
163 :	syskin	663	case 2: // x qpel, y halfpel - left or right during qpel refinement
164 :			ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data);
165 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
166 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
167 :	syskin	663	break;
168 :	Isibaar	3
169 :	syskin	663	default: // x and y in qpel resolution - the "corners" (top left/right and
170 :			// bottom left/right) during qpel refinement
171 :			ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
172 :			ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data);
173 :			ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data);
174 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
175 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
176 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
177 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
178 :			break;
179 :			}
180 :			return Reference;
181 :	Isibaar	3	}
182 :
183 :	syskin	663	static uint8_t *
184 :			Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data)
185 :	Isibaar	580	{
186 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
187 :	syskin	668	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
188 :	syskin	663	const int32_t iEdgedWidth = data->iEdgedWidth;
189 :			const uint32_t rounding = data->rounding;
190 :			const int halfpel_x = x/2;
191 :			const int halfpel_y = y/2;
192 :	Isibaar	580	const uint8_t ref1, ref2, ref3, ref4;
193 :
194 :	syskin	663	ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases
195 :			switch( ((x&1)<<1) + (y&1) ) {
196 :			case 0: // pure halfpel position
197 :			return (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data);
198 :	Isibaar	580	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
199 :	syskin	663	ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
200 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
201 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
202 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
203 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
204 :	Isibaar	580	break;
205 :
206 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
207 :	syskin	663	ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data);
208 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
209 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
210 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
211 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
212 :	Isibaar	580	break;
213 :
214 :			default: // x and y in qpel resolution - the "corners" (top left/right and
215 :			// bottom left/right) during qpel refinement
216 :	syskin	663	ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
217 :			ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data);
218 :			ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data);
219 :	Isibaar	580	interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
220 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
221 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
222 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
223 :			break;
224 :			}
225 :	syskin	663	return Reference;
226 :			}
227 :
228 :			/* CHECK_CANDIATE FUNCTIONS START */
229 :
230 :			static void
231 :			CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
232 :			{
233 :			int t, xc, yc;
234 :			const uint8_t * Reference;
235 :			VECTOR * current;
236 :
237 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
238 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
239 :
240 :			if (data->qpel_precision) { // x and y are in 1/4 precision
241 :			Reference = Interpolate16x16qpel(x, y, 0, data);
242 :			xc = x/2; yc = y/2; //for chroma sad
243 :			current = data->currentQMV;
244 :			} else {
245 :	syskin	712	Reference = GetReference(x, y, 0, data);
246 :	syskin	663	current = data->currentMV;
247 :			xc = x; yc = y;
248 :			}
249 :	syskin	712	t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0);
250 :	Isibaar	580
251 :	syskin	663	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
252 :	Isibaar	580
253 :	syskin	628	data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
254 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
255 :	Isibaar	580
256 :	syskin	663	if (data->chroma) data->temp[0] += ChromaSAD(xc, yc, data);
257 :	syskin	628
258 :	Isibaar	580	if (data->temp[0] < data->iMinSAD[0]) {
259 :			data->iMinSAD[0] = data->temp[0];
260 :	syskin	663	current[0].x = x; current[0].y = y;
261 :			*dir = Direction; }
262 :	Isibaar	580
263 :			if (data->temp[1] < data->iMinSAD[1]) {
264 :	syskin	663	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; }
265 :	Isibaar	580	if (data->temp[2] < data->iMinSAD[2]) {
266 :	syskin	663	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
267 :	Isibaar	580	if (data->temp[3] < data->iMinSAD[3]) {
268 :	syskin	663	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
269 :	Isibaar	580	if (data->temp[4] < data->iMinSAD[4]) {
270 :	syskin	663	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
271 :
272 :	Isibaar	580	}
273 :
274 :			static void
275 :	syskin	704	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
276 :			{
277 :			int t;
278 :			const uint8_t * Reference;
279 :
280 :			if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) \|\| //non-zero integer value
281 :			( x > data->max_dx) \|\| ( x < data->min_dx)
282 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
283 :
284 :	syskin	712	Reference = GetReference(x, y, 0, data);
285 :			t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, 0, 1);
286 :	syskin	704
287 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
288 :
289 :			data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
290 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
291 :
292 :			if (data->temp[0] < data->iMinSAD[0]) {
293 :			data->iMinSAD[0] = data->temp[0];
294 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
295 :			*dir = Direction; }
296 :
297 :			if (data->temp[1] < data->iMinSAD[1]) {
298 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
299 :			if (data->temp[2] < data->iMinSAD[2]) {
300 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
301 :			if (data->temp[3] < data->iMinSAD[3]) {
302 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
303 :			if (data->temp[4] < data->iMinSAD[4]) {
304 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
305 :			}
306 :
307 :			static void
308 :	syskin	663	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
309 :	syskin	658	{
310 :			int32_t sad;
311 :	syskin	663	const uint8_t * Reference;
312 :			int t;
313 :			VECTOR * current;
314 :	syskin	658
315 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
316 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
317 :
318 :	syskin	712	if (data->rrv)
319 :	syskin	704	if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) ) return; //non-zero integer value
320 :
321 :	syskin	663	if (data->qpel_precision) { // x and y are in 1/4 precision
322 :			Reference = Interpolate16x16qpel(x, y, 0, data);
323 :			current = data->currentQMV;
324 :			} else {
325 :	syskin	712	Reference = GetReference(x, y, 0, data);
326 :	syskin	663	current = data->currentMV;
327 :	syskin	658	}
328 :	syskin	712	t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode,
329 :			data->qpel && !data->qpel_precision && !data->rrv, data->rrv);
330 :	syskin	658
331 :			sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
332 :	syskin	663	sad += (data->lambda16 * t * sad)/1000;
333 :	syskin	658
334 :	syskin	663	if (sad < *(data->iMinSAD)) {
335 :			*(data->iMinSAD) = sad;
336 :			current->x = x; current->y = y;
337 :			*dir = Direction; }
338 :	syskin	658	}
339 :
340 :			static void
341 :	Isibaar	539	CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
342 :			{
343 :	syskin	658	// maximum speed - for P/B/I decision
344 :	Isibaar	539	int32_t sad;
345 :
346 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
347 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
348 :
349 :	syskin	601	sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
350 :	Isibaar	539	data->iEdgedWidth, 256*4096);
351 :
352 :			if (sad < *(data->iMinSAD)) {
353 :			*(data->iMinSAD) = sad;
354 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
355 :			*dir = Direction; }
356 :			}
357 :
358 :
359 :			static void
360 :	chl	530	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
361 :			{
362 :			int32_t sad;
363 :	syskin	663	int xb, yb, t;
364 :	chl	530	const uint8_t ReferenceF, ReferenceB;
365 :	syskin	663	VECTOR *current;
366 :	Isibaar	3
367 :	chl	530	if (( xf > data->max_dx) \|\| ( xf < data->min_dx)
368 :			\|\| ( yf > data->max_dy) \|\| (yf < data->min_dy)) return;
369 :	Isibaar	3
370 :	syskin	663	if (data->qpel_precision) {
371 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
372 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
373 :			current = data->currentQMV;
374 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
375 :			} else {
376 :	syskin	712	ReferenceF = GetReference(xf, yf, 0, data);
377 :	syskin	663	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
378 :	syskin	712	ReferenceB = GetReference(xb, yb, 1, data);
379 :	syskin	663	current = data->currentMV;
380 :	chl	530	}
381 :	edgomez	195
382 :	syskin	712	t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0)
383 :			+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0);
384 :
385 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
386 :	syskin	663	sad += (data->lambda16 * t * sad)/1000;
387 :	syskin	628
388 :	chl	530	if (sad < *(data->iMinSAD)) {
389 :			*(data->iMinSAD) = sad;
390 :	syskin	663	current->x = xf; current->y = yf;
391 :	chl	530	*dir = Direction; }
392 :	Isibaar	3	}
393 :
394 :	chl	530	static void
395 :			CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
396 :	Isibaar	3	{
397 :	syskin	628	int32_t sad = 0;
398 :	chl	530	int k;
399 :			const uint8_t *ReferenceF;
400 :			const uint8_t *ReferenceB;
401 :			VECTOR mvs, b_mvs;
402 :	Isibaar	3
403 :	chl	530	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
404 :	Isibaar	3
405 :	chl	530	for (k = 0; k < 4; k++) {
406 :			mvs.x = data->directmvF[k].x + x;
407 :			b_mvs.x = ((x == 0) ?
408 :			data->directmvB[k].x
409 :			: mvs.x - data->referencemv[k].x);
410 :	Isibaar	3
411 :	chl	530	mvs.y = data->directmvF[k].y + y;
412 :			b_mvs.y = ((y == 0) ?
413 :			data->directmvB[k].y
414 :			: mvs.y - data->referencemv[k].y);
415 :
416 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
417 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
418 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
419 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
420 :	Isibaar	3
421 :	syskin	663	if (!data->qpel) {
422 :			mvs.x = 2; mvs.y = 2;
423 :			b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
424 :	chl	530	}
425 :	syskin	663	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
426 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
427 :	chl	530
428 :			sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
429 :	syskin	663	ReferenceF, ReferenceB,
430 :	chl	530	data->iEdgedWidth);
431 :			if (sad > *(data->iMinSAD)) return;
432 :	edgomez	78	}
433 :	chl	530
434 :	syskin	712	sad += (data->lambda16 * d_mv_bits(x, y, 1, 0, 0) * sad)/1000;
435 :	syskin	628
436 :	chl	530	if (sad < *(data->iMinSAD)) {
437 :			*(data->iMinSAD) = sad;
438 :			data->currentMV->x = x; data->currentMV->y = y;
439 :			*dir = Direction; }
440 :	Isibaar	3	}
441 :
442 :	chl	530	static void
443 :			CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
444 :	chl	96	{
445 :	chl	530	int32_t sad;
446 :			const uint8_t *ReferenceF;
447 :			const uint8_t *ReferenceB;
448 :			VECTOR mvs, b_mvs;
449 :	chl	96
450 :	syskin	576	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
451 :	edgomez	195
452 :	chl	530	mvs.x = data->directmvF[0].x + x;
453 :			b_mvs.x = ((x == 0) ?
454 :			data->directmvB[0].x
455 :			: mvs.x - data->referencemv[0].x);
456 :	edgomez	195
457 :	chl	530	mvs.y = data->directmvF[0].y + y;
458 :			b_mvs.y = ((y == 0) ?
459 :			data->directmvB[0].y
460 :			: mvs.y - data->referencemv[0].y);
461 :
462 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
463 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
464 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
465 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
466 :	chl	96
467 :	syskin	663	if (!data->qpel) {
468 :			mvs.x = 2; mvs.y = 2;
469 :			b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
470 :	syskin	712	}
471 :	syskin	663	ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
472 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
473 :	chl	96
474 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
475 :	syskin	712	sad += (data->lambda16 * d_mv_bits(x, y, 1, 0, 0) * sad)/1000;
476 :	chl	96
477 :	chl	530	if (sad < *(data->iMinSAD)) {
478 :			*(data->iMinSAD) = sad;
479 :			data->currentMV->x = x; data->currentMV->y = y;
480 :			*dir = Direction; }
481 :			}
482 :	chl	96
483 :	chl	530	static void
484 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
485 :			{
486 :	syskin	628	int32_t sad; int t;
487 :	chl	530	const uint8_t * Reference;
488 :	edgomez	195
489 :	chl	530	if (( x > data->max_dx) \|\| ( x < data->min_dx)
490 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
491 :	chl	96
492 :	syskin	663	if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data);
493 :	syskin	712	else Reference = GetReference(x, y, 0, data);
494 :	edgomez	195
495 :	chl	530	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
496 :	syskin	712	t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0);
497 :	edgomez	195
498 :	syskin	628	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100;
499 :	chl	96
500 :	Isibaar	606	if (sad < *(data->iMinSAD)) {
501 :			*(data->iMinSAD) = sad;
502 :			data->currentMV->x = x; data->currentMV->y = y;
503 :			*dir = Direction; }
504 :			}
505 :
506 :	Isibaar	539	/* CHECK_CANDIATE FUNCTIONS END */
507 :	chl	96
508 :	chl	530	/* MAINSEARCH FUNCTIONS START */
509 :	edgomez	195
510 :	chl	530	static void
511 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
512 :	chl	181	{
513 :
514 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
515 :
516 :	syskin	712	int iDirection;
517 :	edgomez	195
518 :	syskin	712	for(;;) { //forever
519 :			iDirection = 0;
520 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
521 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
522 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
523 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
524 :	chl	181
525 :	syskin	712	/* now we're doing diagonal checks near our candidate */
526 :	chl	181
527 :	syskin	712	if (iDirection) { //checking if anything found
528 :			bDirection = iDirection;
529 :			iDirection = 0;
530 :			x = data->currentMV->x; y = data->currentMV->y;
531 :			if (bDirection & 3) { //our candidate is left or right
532 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
533 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
534 :			} else { // what remains here is up or down
535 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
536 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1);
537 :			}
538 :	chl	181
539 :	syskin	712	if (iDirection) {
540 :			bDirection += iDirection;
541 :			x = data->currentMV->x; y = data->currentMV->y;
542 :	chl	181	}
543 :	syskin	712	} else { //about to quit, eh? not so fast....
544 :			switch (bDirection) {
545 :			case 2:
546 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
547 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
548 :			break;
549 :			case 1:
550 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
551 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
552 :			break;
553 :			case 2 + 4:
554 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
555 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
556 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
557 :			break;
558 :			case 4:
559 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
560 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
561 :			break;
562 :			case 8:
563 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
564 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
565 :			break;
566 :			case 1 + 4:
567 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
568 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
569 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
570 :			break;
571 :			case 2 + 8:
572 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
573 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
574 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
575 :			break;
576 :			case 1 + 8:
577 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
578 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
579 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
580 :			break;
581 :			default: //1+2+4+8 == we didn't find anything at all
582 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
583 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
584 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
585 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
586 :			break;
587 :			}
588 :			if (!iDirection) break; //ok, the end. really
589 :			bDirection = iDirection;
590 :			x = data->currentMV->x; y = data->currentMV->y;
591 :	chl	181	}
592 :	syskin	712	}
593 :	chl	181	}
594 :
595 :	chl	530	static void
596 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
597 :	chl	326	{
598 :	chl	530	int iDirection;
599 :	chl	326
600 :	chl	530	do {
601 :			iDirection = 0;
602 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
603 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
604 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
605 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
606 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
607 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
608 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
609 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
610 :	chl	326
611 :	chl	530	bDirection = iDirection;
612 :			x = data->currentMV->x; y = data->currentMV->y;
613 :			} while (iDirection);
614 :	chl	346	}
615 :
616 :	chl	530	static void
617 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
618 :	chl	346	{
619 :
620 :	chl	181	/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
621 :
622 :	syskin	712	int iDirection;
623 :	edgomez	195
624 :	syskin	712	do {
625 :			iDirection = 0;
626 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
627 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
628 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
629 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
630 :	chl	181
631 :	syskin	712	/* now we're doing diagonal checks near our candidate */
632 :	chl	181
633 :	syskin	712	if (iDirection) { //checking if anything found
634 :			bDirection = iDirection;
635 :			iDirection = 0;
636 :			x = data->currentMV->x; y = data->currentMV->y;
637 :			if (bDirection & 3) { //our candidate is left or right
638 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
639 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
640 :			} else { // what remains here is up or down
641 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
642 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1);
643 :	chl	181	}
644 :	syskin	712	bDirection += iDirection;
645 :			x = data->currentMV->x; y = data->currentMV->y;
646 :	chl	181	}
647 :	syskin	712	}
648 :			while (iDirection);
649 :	chl	181	}
650 :
651 :	chl	530	/* MAINSEARCH FUNCTIONS END */
652 :	chl	181
653 :	chl	530	/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
654 :
655 :			static void
656 :	syskin	663	SubpelRefine(const SearchData * const data)
657 :	chl	96	{
658 :	syskin	663	/* Do a half-pel or q-pel refinement */
659 :			VECTOR backupMV;
660 :	chl	530	int iDirection; //not needed
661 :	chl	96
662 :	syskin	663	if (data->qpel_precision)
663 :			backupMV = *(data->currentQMV);
664 :			else backupMV = *(data->currentMV);
665 :	edgomez	195
666 :	Isibaar	580	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
667 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
668 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
669 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
670 :
671 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
672 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
673 :
674 :			CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
675 :			CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
676 :			}
677 :
678 :	chl	530	static __inline int
679 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
680 :			const int x, const int y,
681 :	syskin	704	const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv)
682 :	chl	96
683 :	Isibaar	3	{
684 :	chl	530	/* keep repeating checks for all b-frames before this P frame,
685 :			to make sure that SKIP is possible (todo)
686 :			how: if skip is not possible set sad00 to a very high value */
687 :	syskin	704	if(rrv) {
688 :			uint32_t sadC = sad16(current->u + x16 + y(iEdgedWidth/2)*16,
689 :			reference->u + x16 + y(iEdgedWidth/2)16, iEdgedWidth/2, 2564096);
690 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
691 :			sadC += sad16(current->v + (x + y(iEdgedWidth/2))16,
692 :			reference->v + (x + y(iEdgedWidth/2))16, iEdgedWidth/2, 256*4096);
693 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
694 :			return 1;
695 :			} else {
696 :			uint32_t sadC = sad8(current->u + x8 + y(iEdgedWidth/2)*8,
697 :			reference->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2);
698 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
699 :			sadC += sad8(current->v + (x + y(iEdgedWidth/2))8,
700 :			reference->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
701 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
702 :			return 1;
703 :			}
704 :	chl	530	}
705 :	edgomez	195
706 :	chl	530	static __inline void
707 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
708 :			{
709 :			pMB->mode = MODE_NOT_CODED;
710 :	Isibaar	539	pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
711 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
712 :	syskin	628
713 :	Isibaar	580	pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
714 :			pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
715 :
716 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
717 :	Isibaar	3	}
718 :
719 :	syskin	628	bool
720 :	chl	530	MotionEstimation(MBParam * const pParam,
721 :			FRAMEINFO * const current,
722 :			FRAMEINFO * const reference,
723 :			const IMAGE * const pRefH,
724 :			const IMAGE * const pRefV,
725 :			const IMAGE * const pRefHV,
726 :			const uint32_t iLimit)
727 :	Isibaar	3	{
728 :	chl	530	MACROBLOCK *const pMBs = current->mbs;
729 :			const IMAGE *const pCurrent = &current->image;
730 :			const IMAGE *const pRef = &reference->image;
731 :	Isibaar	3
732 :	chl	530	const VECTOR zeroMV = { 0, 0 };
733 :	Isibaar	3
734 :	syskin	704	uint32_t mb_width = pParam->mb_width;
735 :			uint32_t mb_height = pParam->mb_height;
736 :	suxen_drol	702
737 :	chl	530	uint32_t x, y;
738 :			uint32_t iIntra = 0;
739 :	syskin	628	int32_t InterBias, quant = current->quant, sad00;
740 :	Isibaar	580	uint8_t *qimage;
741 :	edgomez	195
742 :	Isibaar	539	// some pre-initialized thingies for SearchP
743 :	syskin	712	int32_t temp[8];
744 :	Isibaar	539	VECTOR currentMV[5];
745 :	Isibaar	580	VECTOR currentQMV[5];
746 :	Isibaar	539	int32_t iMinSAD[5];
747 :			SearchData Data;
748 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
749 :	Isibaar	539	Data.iEdgedWidth = pParam->edged_width;
750 :			Data.currentMV = currentMV;
751 :	Isibaar	580	Data.currentQMV = currentQMV;
752 :	Isibaar	539	Data.iMinSAD = iMinSAD;
753 :			Data.temp = temp;
754 :			Data.iFcode = current->fcode;
755 :	Isibaar	580	Data.rounding = pParam->m_rounding_type;
756 :	syskin	628	Data.qpel = pParam->m_quarterpel;
757 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
758 :	syskin	704	Data.rrv = current->global_flags & XVID_REDUCED;
759 :	Isibaar	539
760 :	syskin	704	if ((current->global_flags & XVID_REDUCED)) {
761 :	suxen_drol	702	mb_width = (pParam->width + 31) / 32;
762 :			mb_height = (pParam->height + 31) / 32;
763 :	syskin	704	Data.qpel = Data.chroma = 0;
764 :	suxen_drol	702	}
765 :
766 :	Isibaar	580	if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
767 :	syskin	594	return 1; // allocate some mem for qpel interpolated blocks
768 :	Isibaar	580	// somehow this is dirty since I think we shouldn't use malloc outside
769 :			// encoder_create() - so please fix me!
770 :	syskin	628	Data.RefQ = qimage;
771 :	chl	530	if (sadInit) (*sadInit) ();
772 :	edgomez	195
773 :	suxen_drol	702	for (y = 0; y < mb_height; y++) {
774 :			for (x = 0; x < mb_width; x++) {
775 :	chl	530	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
776 :	syskin	628
777 :	syskin	704	if (Data.rrv) pMB->sad16 =
778 :			sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32,
779 :			pRef->y + (x + y * pParam->edged_width) * 32,
780 :			pParam->edged_width, pMB->sad8 );
781 :
782 :			else pMB->sad16 =
783 :			sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
784 :	chl	530	pRef->y + (x + y * pParam->edged_width) * 16,
785 :			pParam->edged_width, pMB->sad8 );
786 :	edgomez	195
787 :	syskin	628	if (Data.chroma) {
788 :			pMB->sad16 += sad8(pCurrent->u + x8 + y(pParam->edged_width/2)*8,
789 :			pRef->u + x8 + y(pParam->edged_width/2)*8, pParam->edged_width/2);
790 :
791 :			pMB->sad16 += sad8(pCurrent->v + (x + y(pParam->edged_width/2))8,
792 :			pRef->v + (x + y(pParam->edged_width/2))8, pParam->edged_width/2);
793 :			}
794 :
795 :			sad00 = pMB->sad16; //if no gmc; else sad00 = (..)
796 :
797 :	chl	530	if (!(current->global_flags & XVID_LUMIMASKING)) {
798 :			pMB->dquant = NO_CHANGE;
799 :	syskin	601	pMB->quant = current->quant;
800 :			} else {
801 :	syskin	574	if (pMB->dquant != NO_CHANGE) {
802 :			quant += DQtab[pMB->dquant];
803 :			if (quant > 31) quant = 31;
804 :			else if (quant < 1) quant = 1;
805 :			}
806 :	syskin	601	pMB->quant = quant;
807 :			}
808 :	chl	181
809 :	syskin	628	//initial skip decision
810 :	chl	619	/* no early skip for GMC (global vector = skip vector is unknown!) */
811 :			if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
812 :	syskin	704	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
813 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
814 :	chl	619	SkipMacroblockP(pMB, sad00);
815 :			continue;
816 :			}
817 :			}
818 :	syskin	628
819 :			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
820 :	chl	530	y, current->motion_flags, pMB->quant,
821 :	Isibaar	539	&Data, pParam, pMBs, reference->mbs,
822 :	chl	530	current->global_flags & XVID_INTER4V, pMB);
823 :	edgomez	195
824 :	chl	530	/* final skip decision, a.k.a. "the vector you found, really that good?" */
825 :	chl	619	if (current->coding_type == P_VOP) {
826 :			if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
827 :	syskin	704	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
828 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
829 :	chl	619	SkipMacroblockP(pMB, sad00);
830 :			continue;
831 :			}
832 :			}
833 :	syskin	628
834 :	chl	530	/* finally, intra decision */
835 :	Isibaar	3
836 :	chl	530	InterBias = MV16_INTER_BIAS;
837 :	syskin	639	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
838 :	chl	530	if (y != 0)
839 :	syskin	639	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
840 :	chl	530	if (x != 0)
841 :	syskin	639	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
842 :	syskin	628
843 :			if (Data.chroma) InterBias += 50; // to compensate bigger SAD
844 :	syskin	704	if (Data.rrv) InterBias *= 4; //??
845 :	Isibaar	3
846 :	chl	530	if (InterBias < pMB->sad16) {
847 :	syskin	704	int32_t deviation;
848 :			if (Data.rrv) {
849 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32,
850 :			pParam->edged_width)
851 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16,
852 :			pParam->edged_width)
853 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width,
854 :			pParam->edged_width)
855 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1),
856 :			pParam->edged_width);
857 :			} else
858 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
859 :	chl	530	pParam->edged_width);
860 :	Isibaar	3
861 :	syskin	704	if (deviation < (pMB->sad16 - InterBias)) {
862 :	Isibaar	580	if (++iIntra >= iLimit) { free(qimage); return 1; }
863 :	chl	530	pMB->mode = MODE_INTRA;
864 :	Isibaar	539	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
865 :			pMB->mvs[3] = zeroMV;
866 :	Isibaar	580	pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
867 :			pMB->qmvs[3] = zeroMV;
868 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
869 :			pMB->sad8[3] = 0;
870 :			}
871 :			}
872 :			}
873 :	Isibaar	3	}
874 :	Isibaar	580	free(qimage);
875 :	chl	619
876 :			if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
877 :	syskin	628	current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
878 :	chl	619	else
879 :			current->GMC_MV = zeroMV;
880 :
881 :	syskin	628	return 0;
882 :	chl	530	}
883 :	Isibaar	3
884 :	edgomez	195
885 :	chl	530	#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16\|PMV_HALFPELREFINE16)
886 :	Isibaar	3
887 :	chl	530	static __inline int
888 :			make_mask(const VECTOR * const pmv, const int i)
889 :			{
890 :	Isibaar	539	int mask = 255, j;
891 :	chl	530	for (j = 0; j < i; j++) {
892 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
893 :			if (pmv[i].x == pmv[j].x) {
894 :			if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
895 :			if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
896 :			} else
897 :			if (pmv[i].y == pmv[j].y) {
898 :			if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
899 :			if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
900 :			}
901 :	Isibaar	3	}
902 :	chl	530	return mask;
903 :			}
904 :	edgomez	195
905 :	chl	530	static __inline void
906 :	syskin	704	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
907 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
908 :	chl	530	{
909 :	edgomez	195
910 :	chl	530	//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
911 :	syskin	704	if (rrv) { iWcount /= 2; iHcount /= 2; }
912 :	edgomez	195
913 :	syskin	704	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
914 :	chl	530	pmv[5].x = EVEN(pmv[3].x);
915 :	Isibaar	539	pmv[5].y = EVEN(pmv[3].y);
916 :			} else pmv[5].x = pmv[5].y = 0;
917 :	chl	169
918 :	chl	530	if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
919 :			else pmv[3].x = pmv[3].y = 0;
920 :	Isibaar	3
921 :	chl	530	if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
922 :			else pmv[4].x = pmv[4].y = 0;
923 :	chl	169
924 :	chl	530	// [1] median prediction
925 :	syskin	704	if (rrv) { //median is in halfzero-precision
926 :			pmv[1].x = RRV_MV_SCALEUP(pmv[0].x);
927 :			pmv[1].y = RRV_MV_SCALEUP(pmv[0].y);
928 :			} else { pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); }
929 :	chl	169
930 :	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
931 :	chl	169
932 :	chl	530	pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
933 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
934 :	chl	169
935 :	syskin	704	if ((x < iWcount-1) && (y < iHcount-1)) {
936 :	chl	530	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
937 :	Isibaar	539	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
938 :			} else pmv[6].x = pmv[6].y = 0;
939 :	syskin	704
940 :			if (rrv) {
941 :			int i;
942 :			for (i = 0; i < 7; i++) {
943 :			pmv[i].x = RRV_MV_SCALEDOWN(pmv[i].x);
944 :			pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // a trick
945 :			}
946 :			}
947 :	chl	530	}
948 :	chl	169
949 :	chl	530	static void
950 :	syskin	628	SearchP(const IMAGE * const pRef,
951 :	chl	530	const uint8_t * const pRefH,
952 :			const uint8_t * const pRefV,
953 :			const uint8_t * const pRefHV,
954 :			const IMAGE * const pCur,
955 :			const int x,
956 :			const int y,
957 :			const uint32_t MotionFlags,
958 :			const uint32_t iQuant,
959 :	Isibaar	539	SearchData * const Data,
960 :	chl	530	const MBParam * const pParam,
961 :			const MACROBLOCK * const pMBs,
962 :			const MACROBLOCK * const prevMBs,
963 :			int inter4v,
964 :			MACROBLOCK * const pMB)
965 :			{
966 :	chl	169
967 :	chl	530	int i, iDirection = 255, mask, threshA;
968 :	Isibaar	539	VECTOR pmv[7];
969 :	Isibaar	3
970 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
971 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
972 :	Isibaar	3
973 :	syskin	712	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
974 :	syskin	628
975 :	syskin	712	Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache
976 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
977 :			Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
978 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
979 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
980 :	Isibaar	3
981 :	syskin	704	Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
982 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
983 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
984 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
985 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
986 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
987 :
988 :	syskin	601	Data->lambda16 = lambda_vec16[iQuant];
989 :	Isibaar	606	Data->lambda8 = lambda_vec8[iQuant];
990 :	syskin	663	Data->qpel_precision = 0;
991 :	chl	169
992 :	edgomez	568	if (pMB->dquant != NO_CHANGE) inter4v = 0;
993 :	edgomez	195
994 :	Isibaar	580	for(i = 0; i < 5; i++)
995 :			Data->currentMV[i].x = Data->currentMV[i].y = 0;
996 :	Isibaar	539
997 :	syskin	663	if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
998 :			else Data->predMV = pmv[0];
999 :	Isibaar	606
1000 :	syskin	712	i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode, 0, 0);
1001 :	syskin	628	Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000;
1002 :			Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100;
1003 :	Isibaar	539	Data->iMinSAD[2] = pMB->sad8[1];
1004 :			Data->iMinSAD[3] = pMB->sad8[2];
1005 :			Data->iMinSAD[4] = pMB->sad8[3];
1006 :
1007 :	chl	530	if ((x == 0) && (y == 0)) threshA = 512;
1008 :			else {
1009 :	edgomez	568	threshA = Data->temp[0]; // that's when we keep this SAD atm
1010 :	chl	530	if (threshA < 512) threshA = 512;
1011 :			if (threshA > 1024) threshA = 1024; }
1012 :	Isibaar	3
1013 :	chl	530	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1014 :	syskin	704	prevMBs + x + y * pParam->mb_width, Data->rrv);
1015 :	edgomez	195
1016 :	syskin	704	if (Data->rrv) CheckCandidate = CheckCandidate32;
1017 :			else if (inter4v \|\| Data->chroma) CheckCandidate = CheckCandidate16;
1018 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
1019 :	edgomez	568
1020 :	chl	530	/* main loop. checking all predictions */
1021 :
1022 :			for (i = 1; i < 7; i++) {
1023 :			if (!(mask = make_mask(pmv, i)) ) continue;
1024 :	edgomez	568	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1025 :	Isibaar	539	if (Data->iMinSAD[0] <= threshA) break;
1026 :	chl	530	}
1027 :
1028 :	Isibaar	539	if ((Data->iMinSAD[0] <= threshA) \|\|
1029 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1030 :			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1031 :	chl	530	inter4v = 0;
1032 :	Isibaar	539	} else {
1033 :	Isibaar	3
1034 :	syskin	712	MainSearchFunc * MainSearchPtr;
1035 :	Isibaar	539	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1036 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1037 :			else MainSearchPtr = DiamondSearch;
1038 :	chl	181
1039 :	Isibaar	539	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1040 :	Isibaar	3
1041 :	chl	530	/* extended search, diamond starting in 0,0 and in prediction.
1042 :			note that this search is/might be done in halfpel positions,
1043 :			which makes it more different than the diamond above */
1044 :	chl	259
1045 :	Isibaar	539	if (MotionFlags & PMV_EXTSEARCH16) {
1046 :			int32_t bSAD;
1047 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1048 :	syskin	704	if (Data->rrv) {
1049 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1050 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1051 :			} else
1052 :			if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1053 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1054 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1055 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1056 :	Isibaar	3
1057 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1058 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1059 :			if (bSAD < Data->iMinSAD[0]) {
1060 :			Data->currentMV[0] = backupMV;
1061 :			Data->iMinSAD[0] = bSAD; }
1062 :			}
1063 :	Isibaar	3
1064 :	Isibaar	539	backupMV = Data->currentMV[0];
1065 :	syskin	704	if (!MotionFlags & PMV_HALFPELREFINE16 \|\| Data->rrv) startMV.x = startMV.y = 0;
1066 :			else startMV.x = startMV.y = 1;
1067 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1068 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1069 :	edgomez	195
1070 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1071 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1072 :			if (bSAD < Data->iMinSAD[0]) {
1073 :			Data->currentMV[0] = backupMV;
1074 :			Data->iMinSAD[0] = bSAD; }
1075 :			}
1076 :	Isibaar	3	}
1077 :			}
1078 :
1079 :	syskin	663	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1080 :	Isibaar	3
1081 :	Isibaar	580	for(i = 0; i < 5; i++) {
1082 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1083 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1084 :			}
1085 :
1086 :	syskin	704	if((!Data->rrv) && (pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1087 :	Isibaar	580
1088 :	syskin	663	Data->qpel_precision = 1;
1089 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1090 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1091 :	syskin	601
1092 :	syskin	663	SubpelRefine(Data);
1093 :	Isibaar	580	}
1094 :
1095 :	syskin	628	if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0;
1096 :	Isibaar	539	if (inter4v) {
1097 :			SearchData Data8;
1098 :			Data8.iFcode = Data->iFcode;
1099 :	Isibaar	606	Data8.lambda8 = Data->lambda8;
1100 :	Isibaar	539	Data8.iEdgedWidth = Data->iEdgedWidth;
1101 :	Isibaar	606	Data8.RefQ = Data->RefQ;
1102 :	syskin	628	Data8.qpel = Data->qpel;
1103 :	syskin	704	Data8.rrv = Data->rrv;
1104 :	Isibaar	539	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1105 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1106 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1107 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1108 :	syskin	628
1109 :			if (Data->chroma) {
1110 :	syskin	663	int sumx, sumy, dx, dy;
1111 :	syskin	628
1112 :			if(pParam->m_quarterpel) {
1113 :	syskin	663	sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2;
1114 :			sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2;
1115 :			} else {
1116 :			sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1117 :			sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1118 :			}
1119 :			dx = (sumx >> 3) + roundtab_76[sumx & 0xf];
1120 :			dy = (sumy >> 3) + roundtab_76[sumy & 0xf];
1121 :	syskin	628
1122 :			Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1123 :			}
1124 :	Isibaar	539	}
1125 :	Isibaar	3
1126 :	syskin	704	if (Data->rrv) {
1127 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1128 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1129 :			}
1130 :	chl	530	if (!(inter4v) \|\|
1131 :	Isibaar	539	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1132 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1133 :	chl	530	// INTER MODE
1134 :			pMB->mode = MODE_INTER;
1135 :	Isibaar	539	pMB->mvs[0] = pMB->mvs[1]
1136 :			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1137 :	Isibaar	3
1138 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1139 :	Isibaar	539	pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1140 :	Isibaar	3
1141 :	Isibaar	580	if(pParam->m_quarterpel) {
1142 :	syskin	663	pMB->qmvs[0] = pMB->qmvs[1]
1143 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1144 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1145 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1146 :	syskin	628	} else {
1147 :	Isibaar	580	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1148 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1149 :			}
1150 :	edgomez	195	} else {
1151 :	chl	530	// INTER4V MODE; all other things are already set in Search8
1152 :			pMB->mode = MODE_INTER4V;
1153 :	Isibaar	539	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1154 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1155 :	edgomez	78	}
1156 :	Isibaar	3	}
1157 :
1158 :	chl	530	static void
1159 :			Search8(const SearchData * const OldData,
1160 :			const int x, const int y,
1161 :			const uint32_t MotionFlags,
1162 :			const MBParam * const pParam,
1163 :			MACROBLOCK * const pMB,
1164 :			const MACROBLOCK * const pMBs,
1165 :	Isibaar	539	const int block,
1166 :			SearchData * const Data)
1167 :	chl	345	{
1168 :	syskin	704	int i = 0;
1169 :	Isibaar	539	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1170 :			Data->currentMV = OldData->currentMV + 1 + block;
1171 :	Isibaar	580	Data->currentQMV = OldData->currentQMV + 1 + block;
1172 :	chl	345
1173 :	syskin	601	if(pParam->m_quarterpel) {
1174 :	syskin	704	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1175 :			if (block != 0) i = d_mv_bits( Data->currentQMV->x - Data->predMV.x,
1176 :	syskin	712	Data->currentQMV->y - Data->predMV.y, Data->iFcode, 0, 0);
1177 :
1178 :	Isibaar	606	} else {
1179 :	syskin	704	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1180 :			if (block != 0) {
1181 :	syskin	712	if (block != 0) i = d_mv_bits( Data->currentMV->x - Data->predMV.x,
1182 :			Data->currentMV->y - Data->predMV.y, Data->iFcode, 0, Data->rrv);
1183 :	syskin	704	}
1184 :	Isibaar	606	}
1185 :	chl	345
1186 :	syskin	704	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1187 :
1188 :	chl	530	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8)) {
1189 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
1190 :	chl	345
1191 :	syskin	704	Data->Ref = OldData->Ref + i8 ((block&1) + pParam->edged_width*(block>>1));
1192 :			Data->RefH = OldData->RefH + i8 ((block&1) + pParam->edged_width*(block>>1));
1193 :			Data->RefV = OldData->RefV + i8 ((block&1) + pParam->edged_width*(block>>1));
1194 :			Data->RefHV = OldData->RefHV + i8 ((block&1) + pParam->edged_width*(block>>1));
1195 :	chl	345
1196 :	syskin	704	Data->Cur = OldData->Cur + i8 ((block&1) + pParam->edged_width*(block>>1));
1197 :	syskin	663	Data->qpel_precision = 0;
1198 :	chl	345
1199 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1200 :			pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv);
1201 :
1202 :			if (Data->rrv) CheckCandidate = CheckCandidate16no4v;
1203 :			else CheckCandidate = CheckCandidate8;
1204 :
1205 :	chl	530	if (MotionFlags & PMV_EXTSEARCH8) {
1206 :	Isibaar	580	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1207 :
1208 :	chl	530	MainSearchFunc *MainSearchPtr;
1209 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1210 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1211 :			else MainSearchPtr = DiamondSearch;
1212 :	chl	345
1213 :	Isibaar	580	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1214 :	chl	345
1215 :	syskin	601	if(*(Data->iMinSAD) < temp_sad) {
1216 :	Isibaar	580	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1217 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1218 :			}
1219 :			}
1220 :
1221 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1222 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1223 :
1224 :	syskin	663	SubpelRefine(Data); // perform halfpel refine of current best vector
1225 :	Isibaar	580
1226 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1227 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1228 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1229 :			}
1230 :			}
1231 :
1232 :	syskin	712	if(!Data->rrv && Data->qpel) {
1233 :	syskin	601	if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1234 :			(MotionFlags & PMV_QUARTERPELREFINE8)) {
1235 :	syskin	663	Data->qpel_precision = 1;
1236 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1237 :			pParam->width, pParam->height, OldData->iFcode, 1, 0);
1238 :	syskin	663	SubpelRefine(Data);
1239 :	syskin	601	}
1240 :	Isibaar	580	}
1241 :	chl	345	}
1242 :	chl	530
1243 :	syskin	704	if (Data->rrv) {
1244 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1245 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1246 :			}
1247 :
1248 :	syskin	712	if(Data->qpel) {
1249 :	syskin	663	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1250 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1251 :			pMB->qmvs[block] = *(Data->currentQMV);
1252 :	syskin	704	} else {
1253 :	Isibaar	580	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1254 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1255 :			}
1256 :
1257 :	Isibaar	539	pMB->mvs[block] = *(Data->currentMV);
1258 :	syskin	601	pMB->sad8[block] = 4 * (*Data->iMinSAD);
1259 :	chl	345	}
1260 :
1261 :	chl	530	/* B-frames code starts here */
1262 :	chl	345
1263 :	chl	530	static __inline VECTOR
1264 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1265 :	Isibaar	3	{
1266 :	chl	530	/* the stupidiest function ever */
1267 :			if (mode == MODE_FORWARD) return pMB->mvs[0];
1268 :			else return pMB->b_mvs[0];
1269 :	Isibaar	3	}
1270 :
1271 :	chl	530	static void __inline
1272 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1273 :			const uint32_t iWcount,
1274 :			const MACROBLOCK * const pMB,
1275 :			const uint32_t mode_curr)
1276 :	Isibaar	3	{
1277 :
1278 :	chl	530	// [0] is prediction
1279 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1280 :	Isibaar	3
1281 :	chl	530	pmv[1].x = pmv[1].y = 0; // [1] is zero
1282 :	Isibaar	3
1283 :	chl	530	pmv[2] = ChoosePred(pMB, mode_curr);
1284 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1285 :	edgomez	195
1286 :	chl	530	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1287 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1288 :	Isibaar	539	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1289 :			} else pmv[3].x = pmv[3].y = 0;
1290 :	Isibaar	3
1291 :	chl	530	if (y != 0) {
1292 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1293 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1294 :			} else pmv[4].x = pmv[4].y = 0;
1295 :	Isibaar	3
1296 :	chl	530	if (x != 0) {
1297 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1298 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1299 :			} else pmv[5].x = pmv[5].y = 0;
1300 :	Isibaar	3
1301 :	chl	530	if ((x != 0)&&(y != 0)) {
1302 :			pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1303 :			pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1304 :			} else pmv[6].x = pmv[6].y = 0;
1305 :	edgomez	195
1306 :	chl	530	// more?
1307 :			}
1308 :	chl	181
1309 :	edgomez	170
1310 :	chl	530	/* search backward or forward, for b-frames */
1311 :			static void
1312 :			SearchBF( const uint8_t * const pRef,
1313 :			const uint8_t * const pRefH,
1314 :			const uint8_t * const pRefV,
1315 :			const uint8_t * const pRefHV,
1316 :			const IMAGE * const pCur,
1317 :			const int x, const int y,
1318 :			const uint32_t MotionFlags,
1319 :			const uint32_t iFcode,
1320 :			const MBParam * const pParam,
1321 :			MACROBLOCK * const pMB,
1322 :			const VECTOR * const predMV,
1323 :			int32_t * const best_sad,
1324 :	h	545	const int32_t mode_current,
1325 :			SearchData * const Data)
1326 :	chl	530	{
1327 :	Isibaar	3
1328 :	chl	530	const int32_t iEdgedWidth = pParam->edged_width;
1329 :
1330 :			int i, iDirection, mask;
1331 :	h	545	VECTOR pmv[7];
1332 :	chl	530	MainSearchFunc *MainSearchPtr;
1333 :	h	545	*Data->iMinSAD = MV_MAX_ERROR;
1334 :			Data->iFcode = iFcode;
1335 :	syskin	663	Data->qpel_precision = 0;
1336 :	Isibaar	3
1337 :	h	545	Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1338 :			Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1339 :			Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1340 :			Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1341 :	Isibaar	3
1342 :	h	545	Data->predMV = *predMV;
1343 :	edgomez	195
1344 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1345 :	syskin	712	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1346 :	Isibaar	3
1347 :	h	545	pmv[0] = Data->predMV;
1348 :	syskin	658	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1349 :	h	545	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1350 :	edgomez	195
1351 :	h	545	Data->currentMV->x = Data->currentMV->y = 0;
1352 :	chl	530	CheckCandidate = CheckCandidate16no4v;
1353 :	chl	181
1354 :	chl	530	// main loop. checking all predictions
1355 :			for (i = 0; i < 8; i++) {
1356 :			if (!(mask = make_mask(pmv, i)) ) continue;
1357 :	h	545	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1358 :	edgomez	78	}
1359 :	Isibaar	3
1360 :	chl	530	if (MotionFlags & PMV_USESQUARES16)
1361 :			MainSearchPtr = SquareSearch;
1362 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1363 :			MainSearchPtr = AdvDiamondSearch;
1364 :			else MainSearchPtr = DiamondSearch;
1365 :	chl	96
1366 :	h	545	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1367 :	chl	169
1368 :	syskin	663	SubpelRefine(Data);
1369 :	syskin	658
1370 :			if (Data->qpel) {
1371 :			Data->currentQMV->x = 2*Data->currentMV->x;
1372 :			Data->currentQMV->y = 2*Data->currentMV->y;
1373 :	syskin	663	Data->qpel_precision = 1;
1374 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1375 :			pParam->width, pParam->height, iFcode, 1, 0);
1376 :	syskin	663	SubpelRefine(Data);
1377 :	syskin	658	}
1378 :	chl	169
1379 :	chl	530	// three bits are needed to code backward mode. four for forward
1380 :			// we treat the bits just like they were vector's
1381 :	syskin	628	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1382 :			else Data->iMinSAD += 3 Data->lambda16;
1383 :	chl	169
1384 :	h	545	if (Data->iMinSAD < best_sad) {
1385 :			best_sad = Data->iMinSAD;
1386 :	chl	530	pMB->mode = mode_current;
1387 :	syskin	658	if (Data->qpel) {
1388 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1389 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1390 :			if (mode_current == MODE_FORWARD)
1391 :			pMB->qmvs[0] = *Data->currentQMV;
1392 :			else
1393 :			pMB->b_qmvs[0] = *Data->currentQMV;
1394 :			} else {
1395 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1396 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1397 :			}
1398 :			if (mode_current == MODE_FORWARD)
1399 :			pMB->mvs[0] = (Data->currentMV+2) = Data->currentMV;
1400 :			else
1401 :			pMB->b_mvs[0] = (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1402 :
1403 :	chl	169	}
1404 :	chl	530
1405 :			}
1406 :	chl	169
1407 :	syskin	712	static void
1408 :			SkipDecisionB(const IMAGE * const pCur,
1409 :			const IMAGE * const f_Ref,
1410 :			const IMAGE * const b_Ref,
1411 :			MACROBLOCK * const pMB,
1412 :			const uint32_t quant,
1413 :			const uint32_t x, const uint32_t y,
1414 :			const SearchData * const Data)
1415 :			{
1416 :			int dx, dy, b_dx, b_dy;
1417 :			uint32_t sum;
1418 :			//this is not full chroma compensation, only it's fullpel approximation. should work though
1419 :			if (Data->qpel) {
1420 :			dy = Data->directmvF[0].y/2 + Data->directmvF[1].y/2 +
1421 :			Data->directmvF[2].y/2 + Data->directmvF[3].y/2;
1422 :
1423 :			dx = Data->directmvF[0].x/2 + Data->directmvF[1].x/2 +
1424 :			Data->directmvF[2].x/2 + Data->directmvF[3].x/2;
1425 :
1426 :			b_dy = Data->directmvB[0].y/2 + Data->directmvB[1].y/2 +
1427 :			Data->directmvB[2].y/2 + Data->directmvB[3].y/2;
1428 :
1429 :			b_dx = Data->directmvB[0].x/2 + Data->directmvB[1].x/2 +
1430 :			Data->directmvB[2].x/2 + Data->directmvB[3].x/2;
1431 :
1432 :			} else {
1433 :			dy = Data->directmvF[0].y + Data->directmvF[1].y +
1434 :			Data->directmvF[2].y + Data->directmvF[3].y;
1435 :
1436 :			dx = Data->directmvF[0].x + Data->directmvF[1].x +
1437 :			Data->directmvF[2].x + Data->directmvF[3].x;
1438 :
1439 :			b_dy = Data->directmvB[0].y + Data->directmvB[1].y +
1440 :			Data->directmvB[2].y + Data->directmvB[3].y;
1441 :
1442 :			b_dx = Data->directmvB[0].x + Data->directmvB[1].x +
1443 :			Data->directmvB[2].x + Data->directmvB[3].x;
1444 :			}
1445 :
1446 :
1447 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1448 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1449 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1450 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1451 :
1452 :			sum = sad8bi(pCur->u + 8x + 8y*(Data->iEdgedWidth/2),
1453 :			f_Ref->u + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1454 :			b_Ref->u + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1455 :			Data->iEdgedWidth/2);
1456 :			sum += sad8bi(pCur->v + 8x + 8y*(Data->iEdgedWidth/2),
1457 :			f_Ref->v + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1458 :			b_Ref->v + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1459 :			Data->iEdgedWidth/2);
1460 :
1461 :			if (sum < 2MAX_CHROMA_SAD_FOR_SKIP quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1462 :			}
1463 :
1464 :
1465 :
1466 :			static __inline uint32_t
1467 :	Isibaar	539	SearchDirect(const IMAGE * const f_Ref,
1468 :	chl	530	const uint8_t * const f_RefH,
1469 :			const uint8_t * const f_RefV,
1470 :			const uint8_t * const f_RefHV,
1471 :	Isibaar	539	const IMAGE * const b_Ref,
1472 :	chl	530	const uint8_t * const b_RefH,
1473 :			const uint8_t * const b_RefV,
1474 :			const uint8_t * const b_RefHV,
1475 :			const IMAGE * const pCur,
1476 :			const int x, const int y,
1477 :			const uint32_t MotionFlags,
1478 :			const int32_t TRB, const int32_t TRD,
1479 :			const MBParam * const pParam,
1480 :			MACROBLOCK * const pMB,
1481 :			const MACROBLOCK * const b_mb,
1482 :	h	545	int32_t * const best_sad,
1483 :			SearchData * const Data)
1484 :	edgomez	195
1485 :	chl	530	{
1486 :	h	545	int32_t skip_sad;
1487 :	chl	530	int k;
1488 :	h	545
1489 :	chl	530	MainSearchFunc *MainSearchPtr;
1490 :	Isibaar	3
1491 :	h	545	Data->iMinSAD = 2564096;
1492 :	chl	140
1493 :	h	545	Data->Ref = f_Ref->y + (x + Data->iEdgedWidthy) 16;
1494 :			Data->RefH = f_RefH + (x + Data->iEdgedWidthy) 16;
1495 :			Data->RefV = f_RefV + (x + Data->iEdgedWidthy) 16;
1496 :			Data->RefHV = f_RefHV + (x + Data->iEdgedWidthy) 16;
1497 :			Data->bRef = b_Ref->y + (x + Data->iEdgedWidthy) 16;
1498 :			Data->bRefH = b_RefH + (x + Data->iEdgedWidthy) 16;
1499 :			Data->bRefV = b_RefV + (x + Data->iEdgedWidthy) 16;
1500 :			Data->bRefHV = b_RefHV + (x + Data->iEdgedWidthy) 16;
1501 :	chl	140
1502 :	h	545	Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1503 :			Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1504 :			Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1505 :			Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1506 :	syskin	658	if (Data->qpel) { //we measure in qpixels
1507 :			Data->max_dx *= 2;
1508 :			Data->max_dy *= 2;
1509 :			Data->min_dx *= 2;
1510 :			Data->min_dy *= 2;
1511 :			Data->referencemv = b_mb->qmvs;
1512 :			} else Data->referencemv = b_mb->mvs;
1513 :	syskin	672	Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here
1514 :	Isibaar	3
1515 :	chl	530	for (k = 0; k < 4; k++) {
1516 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1517 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1518 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1519 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1520 :	Isibaar	3
1521 :	h	545	if ( ( pMB->b_mvs[k].x > Data->max_dx ) \|\| ( pMB->b_mvs[k].x < Data->min_dx )
1522 :			\|\| ( pMB->b_mvs[k].y > Data->max_dy ) \|\| ( pMB->b_mvs[k].y < Data->min_dy )) {
1523 :	Isibaar	3
1524 :	Isibaar	539	best_sad = 2564096; // in that case, we won't use direct mode
1525 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1526 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1527 :	syskin	712	return 256*4096;
1528 :	Isibaar	539	}
1529 :			if (b_mb->mode != MODE_INTER4V) {
1530 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1531 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1532 :	h	545	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1533 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1534 :	Isibaar	539	break;
1535 :			}
1536 :			}
1537 :	Isibaar	3
1538 :	syskin	663
1539 :			if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect;
1540 :			else CheckCandidate = CheckCandidateDirectno4v;
1541 :	syskin	658
1542 :	h	545	(*CheckCandidate)(0, 0, 255, &k, Data);
1543 :	Isibaar	3
1544 :	syskin	712	// initial (fast) skip decision
1545 :			if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH*2) {
1546 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data->chroma, Data); //possible skip - checking chroma
1547 :			if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1548 :	Isibaar	539	}
1549 :
1550 :	h	545	skip_sad = *Data->iMinSAD;
1551 :	edgomez	195
1552 :	chl	530	// DIRECT MODE DELTA VECTOR SEARCH.
1553 :			// This has to be made more effective, but at the moment I'm happy it's running at all
1554 :	Isibaar	3
1555 :	chl	530	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1556 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1557 :			else MainSearchPtr = DiamondSearch;
1558 :	edgomez	195
1559 :	h	545	(*MainSearchPtr)(0, 0, Data, 255);
1560 :	Isibaar	3
1561 :	syskin	663	SubpelRefine(Data);
1562 :	Isibaar	3
1563 :	h	545	best_sad = Data->iMinSAD;
1564 :	Isibaar	3
1565 :	syskin	672	if (b_mb->mode == MODE_INTER4V)
1566 :	chl	530	pMB->mode = MODE_DIRECT;
1567 :	syskin	672	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1568 :	edgomez	195
1569 :	h	545	pMB->pmvs[3] = *Data->currentMV;
1570 :	edgomez	195
1571 :	chl	530	for (k = 0; k < 4; k++) {
1572 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1573 :	syskin	658	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1574 :	h	545	? Data->directmvB[k].x
1575 :	syskin	658	:pMB->mvs[k].x - Data->referencemv[k].x);
1576 :	h	545	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1577 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1578 :			? Data->directmvB[k].y
1579 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1580 :	syskin	658	if (Data->qpel) {
1581 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1582 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1583 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1584 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1585 :			}
1586 :
1587 :	chl	530	if (b_mb->mode != MODE_INTER4V) {
1588 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1589 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1590 :	syskin	658	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1591 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1592 :	chl	530	break;
1593 :			}
1594 :			}
1595 :	h	545	return skip_sad;
1596 :	Isibaar	3	}
1597 :	chl	96
1598 :	h	545
1599 :	syskin	712	static void
1600 :	chl	530	SearchInterpolate(const uint8_t * const f_Ref,
1601 :			const uint8_t * const f_RefH,
1602 :			const uint8_t * const f_RefV,
1603 :			const uint8_t * const f_RefHV,
1604 :			const uint8_t * const b_Ref,
1605 :			const uint8_t * const b_RefH,
1606 :			const uint8_t * const b_RefV,
1607 :			const uint8_t * const b_RefHV,
1608 :			const IMAGE * const pCur,
1609 :			const int x, const int y,
1610 :			const uint32_t fcode,
1611 :			const uint32_t bcode,
1612 :			const uint32_t MotionFlags,
1613 :			const MBParam * const pParam,
1614 :			const VECTOR * const f_predMV,
1615 :			const VECTOR * const b_predMV,
1616 :			MACROBLOCK * const pMB,
1617 :	h	545	int32_t * const best_sad,
1618 :			SearchData * const fData)
1619 :	chl	530
1620 :	chl	96	{
1621 :
1622 :	chl	530	int iDirection, i, j;
1623 :	h	545	SearchData bData;
1624 :	chl	96
1625 :	syskin	712	fData->qpel_precision = 0;
1626 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1627 :			fData->iMinSAD = 4096256;
1628 :			bData.currentMV ++; bData.currentQMV ++;
1629 :	h	545	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1630 :	edgomez	195
1631 :	syskin	712	i = (x + y * fData->iEdgedWidth) * 16;
1632 :			bData.bRef = fData->Ref = f_Ref + i;
1633 :			bData.bRefH = fData->RefH = f_RefH + i;
1634 :			bData.bRefV = fData->RefV = f_RefV + i;
1635 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1636 :			bData.Ref = fData->bRef = b_Ref + i;
1637 :			bData.RefH = fData->bRefH = b_RefH + i;
1638 :			bData.RefV = fData->bRefV = b_RefV + i;
1639 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1640 :	chl	96
1641 :	h	545	bData.bpredMV = fData->predMV = *f_predMV;
1642 :			fData->bpredMV = bData.predMV = *b_predMV;
1643 :	syskin	661	fData->currentMV[0] = fData->currentMV[2];
1644 :	chl	96
1645 :	syskin	712	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0);
1646 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0);
1647 :
1648 :	h	545	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1649 :	syskin	675	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1650 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1651 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1652 :	Isibaar	539
1653 :	h	545	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1654 :	syskin	675	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1655 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1656 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1657 :	Isibaar	539
1658 :	h	545	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1659 :	chl	96
1660 :	chl	530	//diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1661 :	chl	96
1662 :	chl	530	do {
1663 :			iDirection = 255;
1664 :			// forward MV moves
1665 :	h	545	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1666 :	chl	96
1667 :	h	545	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1668 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1669 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1670 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1671 :	chl	96
1672 :	chl	530	// backward MV moves
1673 :	h	545	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1674 :			fData->currentMV[2] = fData->currentMV[0];
1675 :	Isibaar	539	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1676 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1677 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1678 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1679 :	chl	96
1680 :	chl	530	} while (!(iDirection));
1681 :	edgomez	195
1682 :	syskin	658	if (fData->qpel) {
1683 :	syskin	675	CheckCandidate = CheckCandidateInt;
1684 :	syskin	663	fData->qpel_precision = bData.qpel_precision = 1;
1685 :	syskin	712	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0);
1686 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1687 :	syskin	658	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1688 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1689 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1690 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1691 :	syskin	663	SubpelRefine(fData);
1692 :	syskin	658	fData->currentQMV[2] = fData->currentQMV[0];
1693 :	syskin	663	SubpelRefine(&bData);
1694 :	syskin	658	}
1695 :	syskin	676
1696 :	syskin	712	fData->iMinSAD += (2+2) fData->lambda16; // two bits are needed to code interpolate mode.
1697 :	syskin	658
1698 :	h	545	if (fData->iMinSAD < best_sad) {
1699 :			best_sad = fData->iMinSAD;
1700 :			pMB->mvs[0] = fData->currentMV[0];
1701 :			pMB->b_mvs[0] = fData->currentMV[1];
1702 :	chl	530	pMB->mode = MODE_INTERPOLATE;
1703 :	syskin	658	if (fData->qpel) {
1704 :			pMB->qmvs[0] = fData->currentQMV[0];
1705 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1706 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1707 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1708 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1709 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1710 :			} else {
1711 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1712 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1713 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1714 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1715 :			}
1716 :	chl	530	}
1717 :			}
1718 :	chl	96
1719 :	chl	530	void
1720 :			MotionEstimationBVOP(MBParam * const pParam,
1721 :			FRAMEINFO * const frame,
1722 :			const int32_t time_bp,
1723 :			const int32_t time_pp,
1724 :			// forward (past) reference
1725 :			const MACROBLOCK * const f_mbs,
1726 :			const IMAGE * const f_ref,
1727 :			const IMAGE * const f_refH,
1728 :			const IMAGE * const f_refV,
1729 :			const IMAGE * const f_refHV,
1730 :			// backward (future) reference
1731 :	syskin	644	const FRAMEINFO * const b_reference,
1732 :	chl	530	const IMAGE * const b_ref,
1733 :			const IMAGE * const b_refH,
1734 :			const IMAGE * const b_refV,
1735 :			const IMAGE * const b_refHV)
1736 :			{
1737 :			uint32_t i, j;
1738 :	syskin	712	int32_t best_sad;
1739 :			uint32_t skip_sad;
1740 :	chl	530	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1741 :			static const VECTOR zeroMV={0,0};
1742 :	syskin	644	const MACROBLOCK * const b_mbs = b_reference->mbs;
1743 :	chl	96
1744 :	chl	530	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1745 :	chl	96
1746 :	chl	530	const int32_t TRB = time_pp - time_bp;
1747 :			const int32_t TRD = time_pp;
1748 :	syskin	658	uint8_t * qimage;
1749 :	chl	96
1750 :	h	545	// some pre-inintialized data for the rest of the search
1751 :
1752 :			SearchData Data;
1753 :			int32_t iMinSAD;
1754 :			VECTOR currentMV[3];
1755 :	syskin	658	VECTOR currentQMV[3];
1756 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
1757 :	h	545	Data.iEdgedWidth = pParam->edged_width;
1758 :	syskin	658	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1759 :	h	545	Data.iMinSAD = &iMinSAD;
1760 :	syskin	712	Data.lambda16 = lambda_vec16[frame->quant];
1761 :			Data.chroma = frame->quant;
1762 :	syskin	658	Data.qpel = pParam->m_quarterpel;
1763 :	syskin	663	Data.rounding = 0;
1764 :	h	545
1765 :	syskin	658	if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
1766 :			return; // allocate some mem for qpel interpolated blocks
1767 :			// somehow this is dirty since I think we shouldn't use malloc outside
1768 :			// encoder_create() - so please fix me!
1769 :			Data.RefQ = qimage;
1770 :
1771 :	chl	530	// note: i==horizontal, j==vertical
1772 :			for (j = 0; j < pParam->mb_height; j++) {
1773 :	chl	96
1774 :	chl	530	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1775 :	edgomez	195
1776 :	chl	530	for (i = 0; i < pParam->mb_width; i++) {
1777 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1778 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1779 :	edgomez	195
1780 :	chl	619	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1781 :	syskin	644	if (b_reference->coding_type != S_VOP)
1782 :			if (b_mb->mode == MODE_NOT_CODED) {
1783 :			pMB->mode = MODE_NOT_CODED;
1784 :			continue;
1785 :			}
1786 :	chl	96
1787 :	h	545	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1788 :	syskin	601	pMB->quant = frame->quant;
1789 :	syskin	628
1790 :	chl	530	/* direct search comes first, because it (1) checks for SKIP-mode
1791 :			and (2) sets very good predictions for forward and backward search */
1792 :	Isibaar	539	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1793 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1794 :	chl	530	&frame->image,
1795 :			i, j,
1796 :			frame->motion_flags,
1797 :			TRB, TRD,
1798 :			pParam,
1799 :			pMB, b_mb,
1800 :	h	545	&best_sad,
1801 :			&Data);
1802 :	chl	96
1803 :	Isibaar	539	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1804 :	syskin	658
1805 :	chl	530	// forward search
1806 :			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1807 :			&frame->image, i, j,
1808 :			frame->motion_flags,
1809 :	h	545	frame->fcode, pParam,
1810 :	chl	530	pMB, &f_predMV, &best_sad,
1811 :	h	545	MODE_FORWARD, &Data);
1812 :	chl	96
1813 :	chl	530	// backward search
1814 :			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1815 :			&frame->image, i, j,
1816 :			frame->motion_flags,
1817 :	h	545	frame->bcode, pParam,
1818 :	chl	530	pMB, &b_predMV, &best_sad,
1819 :	h	545	MODE_BACKWARD, &Data);
1820 :	chl	96
1821 :	chl	530	// interpolate search comes last, because it uses data from forward and backward as prediction
1822 :	syskin	662
1823 :	chl	530	SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1824 :			b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1825 :			&frame->image,
1826 :			i, j,
1827 :			frame->fcode, frame->bcode,
1828 :			frame->motion_flags,
1829 :	h	545	pParam,
1830 :	chl	530	&f_predMV, &b_predMV,
1831 :	h	545	pMB, &best_sad,
1832 :			&Data);
1833 :	syskin	662
1834 :	syskin	712	// final skip decision
1835 :			if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2)
1836 :			&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1837 :			SkipDecisionB(&frame->image, f_ref, b_ref, pMB,frame->quant, i, j, &Data);
1838 :
1839 :	chl	530	switch (pMB->mode) {
1840 :			case MODE_FORWARD:
1841 :			f_count++;
1842 :	syskin	712	if (Data.qpel) f_predMV = pMB->qmvs[0];
1843 :	syskin	658	else f_predMV = pMB->mvs[0];
1844 :	chl	530	break;
1845 :			case MODE_BACKWARD:
1846 :			b_count++;
1847 :	syskin	712	if (Data.qpel) b_predMV = pMB->b_qmvs[0];
1848 :	syskin	658	else b_predMV = pMB->b_mvs[0];
1849 :	chl	530	break;
1850 :			case MODE_INTERPOLATE:
1851 :			i_count++;
1852 :	syskin	712	if (Data.qpel) {
1853 :	syskin	658	f_predMV = pMB->qmvs[0];
1854 :			b_predMV = pMB->b_qmvs[0];
1855 :			} else {
1856 :			f_predMV = pMB->mvs[0];
1857 :			b_predMV = pMB->b_mvs[0];
1858 :			}
1859 :	chl	530	break;
1860 :			case MODE_DIRECT:
1861 :			case MODE_DIRECT_NO4V:
1862 :			d_count++;
1863 :			default:
1864 :			break;
1865 :	chl	96	}
1866 :			}
1867 :			}
1868 :	syskin	658	free(qimage);
1869 :	chl	96	}
1870 :
1871 :	Isibaar	539	static __inline int
1872 :			MEanalyzeMB ( const uint8_t * const pRef,
1873 :			const uint8_t * const pCur,
1874 :			const int x,
1875 :			const int y,
1876 :			const MBParam * const pParam,
1877 :			const MACROBLOCK * const pMBs,
1878 :	h	545	MACROBLOCK * const pMB,
1879 :			SearchData * const Data)
1880 :	Isibaar	539	{
1881 :	chl	317
1882 :	Isibaar	606	int i = 255, mask;
1883 :	h	545	VECTOR pmv[3];
1884 :	syskin	640	*(Data->iMinSAD) = MV_MAX_ERROR;
1885 :	suxen_drol	118
1886 :	syskin	640	//median is only used as prediction. it doesn't have to be real
1887 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1888 :			else
1889 :			if (x == 1) //left macroblock does not have any vector now
1890 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1891 :			else if (y == 1) // top macroblock don't have it's vector
1892 :			Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1893 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1894 :
1895 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1896 :	syskin	712	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, 0);
1897 :	chl	326
1898 :	h	545	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1899 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1900 :	Isibaar	539
1901 :			pmv[1].x = EVEN(pMB->mvs[0].x);
1902 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1903 :	syskin	640	pmv[2].x = EVEN(Data->predMV.x);
1904 :			pmv[2].y = EVEN(Data->predMV.y);
1905 :			pmv[0].x = pmv[0].y = 0;
1906 :	chl	326
1907 :	syskin	712	CheckCandidate16no4vI(0, 0, 255, &i, Data);
1908 :	syskin	640
1909 :			//early skip for 0,0
1910 :			if (Data->iMinSAD < MAX_SAD00_FOR_SKIP 4) {
1911 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1912 :			pMB->mode = MODE_NOT_CODED;
1913 :			return 0;
1914 :			}
1915 :
1916 :	Isibaar	539	if (!(mask = make_mask(pmv, 1)))
1917 :	syskin	712	CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
1918 :	Isibaar	539	if (!(mask = make_mask(pmv, 2)))
1919 :	syskin	712	CheckCandidate16no4vI(pmv[2].x, pmv[2].y, mask, &i, Data);
1920 :	Isibaar	539
1921 :	syskin	712	if (Data->iMinSAD > MAX_SAD00_FOR_SKIP 6) // diamond only if needed
1922 :	syskin	640	DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1923 :	Isibaar	539
1924 :	syskin	640	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1925 :	syskin	628	pMB->mode = MODE_INTER;
1926 :	h	545	return *(Data->iMinSAD);
1927 :	Isibaar	539	}
1928 :
1929 :			#define INTRA_THRESH 1350
1930 :	syskin	672	#define INTER_THRESH 1200
1931 :	Isibaar	539
1932 :	syskin	644
1933 :	Isibaar	539	int
1934 :			MEanalysis( const IMAGE * const pRef,
1935 :	syskin	644	FRAMEINFO * const Current,
1936 :	Isibaar	539	MBParam * const pParam,
1937 :	syskin	644	int maxIntra, //maximum number if non-I frames
1938 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1939 :			int bCount) // number if B frames in a row
1940 :	Isibaar	539	{
1941 :			uint32_t x, y, intra = 0;
1942 :			int sSAD = 0;
1943 :	syskin	644	MACROBLOCK * const pMBs = Current->mbs;
1944 :			const IMAGE * const pCurrent = &Current->image;
1945 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1946 :	h	545
1947 :			VECTOR currentMV;
1948 :			int32_t iMinSAD;
1949 :			SearchData Data;
1950 :			Data.iEdgedWidth = pParam->edged_width;
1951 :			Data.currentMV = &currentMV;
1952 :			Data.iMinSAD = &iMinSAD;
1953 :	syskin	644	Data.iFcode = Current->fcode;
1954 :	syskin	640	CheckCandidate = CheckCandidate16no4vI;
1955 :	h	545
1956 :	syskin	658	if (intraCount < 10) // we're right after an I frame
1957 :			IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
1958 :	syskin	644	else
1959 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
1960 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
1961 :
1962 :
1963 :	syskin	672	InterThresh += 400 * (1 - bCount);
1964 :	syskin	644	if (InterThresh < 200) InterThresh = 200;
1965 :
1966 :	Isibaar	539	if (sadInit) (*sadInit) ();
1967 :
1968 :	syskin	704	for (y = 1; y < pParam->mb_height-1; y++) {
1969 :			for (x = 1; x < pParam->mb_width-1; x++) {
1970 :	Isibaar	539	int sad, dev;
1971 :			MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
1972 :
1973 :			sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
1974 :	h	545	pParam, pMBs, pMB, &Data);
1975 :	Isibaar	539
1976 :	syskin	644	if (sad > IntraThresh) {
1977 :	syskin	601	dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
1978 :			pParam->edged_width);
1979 :	syskin	644	if (dev + IntraThresh < sad) {
1980 :			pMB->mode = MODE_INTRA;
1981 :	suxen_drol	709	if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
1982 :	Isibaar	539	}
1983 :	syskin	644	}
1984 :	syskin	601	sSAD += sad;
1985 :	suxen_drol	118	}
1986 :			}
1987 :	syskin	704	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
1988 :	suxen_drol	709	if (sSAD > InterThresh ) return P_VOP;
1989 :	Isibaar	539	emms();
1990 :	suxen_drol	709	return B_VOP;
1991 :	syskin	712
1992 :	suxen_drol	118	}
1993 :	syskin	576
1994 :	syskin	628	static void
1995 :			CheckGMC(int x, int y, const int dir, int * iDirection,
1996 :			const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
1997 :			const MBParam * const pParam)
1998 :			{
1999 :			uint32_t mx, my, a, count = 0;
2000 :	chl	619
2001 :	syskin	628	for (my = 1; my < pParam->mb_height-1; my++)
2002 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2003 :			VECTOR mv;
2004 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2005 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED) continue;
2006 :			mv = pMB->mvs[0];
2007 :			a = ABS(mv.x - x) + ABS(mv.y - y);
2008 :			if (a < 6) count += 6 - a;
2009 :			}
2010 :	chl	619
2011 :	syskin	628	if (count > *bestcount) {
2012 :			*bestcount = count;
2013 :			*iDirection = dir;
2014 :			GMC->x = x; GMC->y = y;
2015 :			}
2016 :			}
2017 :
2018 :
2019 :	suxen_drol	721	static VECTOR
2020 :	chl	619	GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2021 :			{
2022 :
2023 :	syskin	628	uint32_t count, bestcount = 0;
2024 :	chl	619	int x, y;
2025 :	syskin	628	VECTOR gmc = {0,0};
2026 :	chl	619	int step, min_x, max_x, min_y, max_y;
2027 :			uint32_t mx, my;
2028 :	syskin	628	int iDirection, bDirection;
2029 :	chl	619
2030 :			min_x = min_y = -32<<iFcode;
2031 :			max_x = max_y = 32<<iFcode;
2032 :
2033 :	syskin	628	//step1: let's find a rough camera panning
2034 :	chl	619	for (step = 32; step >= 2; step /= 2) {
2035 :			bestcount = 0;
2036 :			for (y = min_y; y <= max_y; y += step)
2037 :			for (x = min_x ; x <= max_x; x += step) {
2038 :			count = 0;
2039 :			//for all macroblocks
2040 :			for (my = 1; my < pParam->mb_height-1; my++)
2041 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2042 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2043 :			VECTOR mv;
2044 :
2045 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2046 :			continue;
2047 :
2048 :			mv = pMB->mvs[0];
2049 :			if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2050 :			count++;
2051 :			}
2052 :			if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2053 :			}
2054 :			min_x = gmc.x - step;
2055 :			max_x = gmc.x + step;
2056 :			min_y = gmc.y - step;
2057 :			max_y = gmc.y + step;
2058 :
2059 :			}
2060 :	syskin	628
2061 :			if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2062 :			gmc.x = gmc.y = 0; //no camara pan, no GMC
2063 :
2064 :			// step2: let's refine camera panning using gradiend-descent approach.
2065 :			// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
2066 :			bestcount = 0;
2067 :			CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
2068 :			do {
2069 :			x = gmc.x; y = gmc.y;
2070 :			bDirection = iDirection; iDirection = 0;
2071 :			if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2072 :			if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2073 :			if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
2074 :			if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2075 :
2076 :			} while (iDirection);
2077 :
2078 :	chl	619	if (pParam->m_quarterpel) {
2079 :			gmc.x *= 2;
2080 :			gmc.y = 2; / we store the halfpel value as pseudo-qpel to make comparison easier */
2081 :			}
2082 :
2083 :			return gmc;
2084 :			}

No admin address has been configured	ViewVC Help
Powered by ViewVC 1.0.4