Annotation of /branches/dev-api-4/xvidcore/src/motion/motion_est.c

Revision 1136 - (view) (download)

1 :	edgomez	1054	/*****************************************************************************
2 :	Isibaar	3	*
3 :	edgomez	1054	* XVID MPEG-4 VIDEO CODEC
4 :			* - Motion Estimation related code -
5 :	chl	259	*
6 :	edgomez	1054	* Copyright(C) 2002 Christoph Lampert <gruel@web.de>
7 :			* 2002 Michael Militzer <michael@xvid.org>
8 :			* 2002-2003 Radoslaw Czyz <xvid@syskin.cjb.net>
9 :	chl	259	*
10 :	edgomez	1054	* This program is free software ; you can redistribute it and/or modify
11 :			* it under the terms of the GNU General Public License as published by
12 :			* the Free Software Foundation ; either version 2 of the License, or
13 :			* (at your option) any later version.
14 :	chl	259	*
15 :	edgomez	1054	* This program is distributed in the hope that it will be useful,
16 :			* but WITHOUT ANY WARRANTY ; without even the implied warranty of
17 :			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 :			* GNU General Public License for more details.
19 :	chl	259	*
20 :	edgomez	1054	* You should have received a copy of the GNU General Public License
21 :			* along with this program ; if not, write to the Free Software
22 :			* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 :	chl	259	*
24 :	Isibaar	1136	* $Id: motion_est.c,v 1.58.2.33 2003-09-04 18:40:02 Isibaar Exp $
25 :	edgomez	1054	*
26 :			****************************************************************************/
27 :	chl	259
28 :	Isibaar	3	#include <assert.h>
29 :			#include <stdio.h>
30 :	chl	96	#include <stdlib.h>
31 :	edgomez	1053	#include <string.h> /* memcpy */
32 :			#include <math.h> /* lrint */
33 :	Isibaar	3
34 :			#include "../encoder.h"
35 :			#include "../utils/mbfunctions.h"
36 :			#include "../prediction/mbprediction.h"
37 :	syskin	1084	#include "../global.h"
38 :	Isibaar	3	#include "../utils/timer.h"
39 :	edgomez	851	#include "../image/interpolate8x8.h"
40 :			#include "motion_est.h"
41 :	suxen_drol	118	#include "motion.h"
42 :	Isibaar	3	#include "sad.h"
43 :	chl	1077	#include "gmc.h"
44 :	edgomez	851	#include "../utils/emms.h"
45 :			#include "../dct/fdct.h"
46 :	Isibaar	3
47 :	edgomez	872	/*****************************************************************************
48 :			* Modified rounding tables -- declared in motion.h
49 :			* Original tables see ISO spec tables 7-6 -> 7-9
50 :			****************************************************************************/
51 :
52 :			const uint32_t roundtab[16] =
53 :			{0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 };
54 :
55 :			/* K = 4 */
56 :			const uint32_t roundtab_76[16] =
57 :			{ 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1 };
58 :
59 :			/* K = 2 */
60 :			const uint32_t roundtab_78[8] =
61 :			{ 0, 0, 1, 1, 0, 0, 0, 1 };
62 :
63 :			/* K = 1 */
64 :			const uint32_t roundtab_79[4] =
65 :			{ 0, 1, 0, 0 };
66 :
67 :	edgomez	851	#define INITIAL_SKIP_THRESH (10)
68 :			#define FINAL_SKIP_THRESH (50)
69 :			#define MAX_SAD00_FOR_SKIP (20)
70 :			#define MAX_CHROMA_SAD_FOR_SKIP (22)
71 :	Isibaar	3
72 :	edgomez	851	#define CHECK_CANDIDATE(X,Y,D) { \
73 :	syskin	1132	CheckCandidate((X),(Y), data, (D) ); }
74 :	Isibaar	3
75 :	syskin	1084
76 :	edgomez	872	/*****************************************************************************
77 :			* Code
78 :			****************************************************************************/
79 :
80 :	edgomez	851	static __inline uint32_t
81 :			d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
82 :			{
83 :	edgomez	974	int bits;
84 :			const int q = (1 << (iFcode - 1)) - 1;
85 :
86 :			x <<= qpel;
87 :			y <<= qpel;
88 :	edgomez	851	if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
89 :	chl	141
90 :	edgomez	851	x -= pred.x;
91 :	edgomez	974	bits = (x != 0 ? iFcode:0);
92 :			x = abs(x);
93 :			x += q;
94 :			x >>= (iFcode - 1);
95 :			bits += mvtab[x];
96 :
97 :	edgomez	851	y -= pred.y;
98 :	edgomez	974	bits += (y != 0 ? iFcode:0);
99 :			y = abs(y);
100 :			y += q;
101 :			y >>= (iFcode - 1);
102 :			bits += mvtab[y];
103 :	chl	141
104 :	edgomez	974	return bits;
105 :	edgomez	851	}
106 :	chl	141
107 :	edgomez	978	static int32_t ChromaSAD2(const int fx, const int fy, const int bx, const int by,
108 :			const SearchData * const data)
109 :	edgomez	851	{
110 :			int sad;
111 :			const uint32_t stride = data->iEdgedWidth/2;
112 :	syskin	1129	uint8_t f_refu, f_refv, b_refu, b_refv;
113 :
114 :			const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = {
115 :			NULL,
116 :			interpolate8x8_halfpel_v,
117 :			interpolate8x8_halfpel_h,
118 :			interpolate8x8_halfpel_hv
119 :			};
120 :
121 :	edgomez	978	int offset = (fx>>1) + (fy>>1)*stride;
122 :	syskin	1129	int filter = ((fx & 1) << 1) \| (fy & 1);
123 :	Isibaar	3
124 :	syskin	1129	if (filter != 0) {
125 :			f_refu = data->RefQ;
126 :			f_refv = data->RefQ + 8;
127 :			interpolate8x8_halfpel[filter](f_refu, data->RefP[4] + offset, stride, data->rounding);
128 :			interpolate8x8_halfpel[filter](f_refv, data->RefP[5] + offset, stride, data->rounding);
129 :			} else {
130 :			f_refu = (uint8_t*)data->RefP[4] + offset;
131 :			f_refv = (uint8_t*)data->RefP[5] + offset;
132 :	edgomez	851	}
133 :	Isibaar	3
134 :	edgomez	978	offset = (bx>>1) + (by>>1)*stride;
135 :	syskin	1129	filter = ((bx & 1) << 1) \| (by & 1);
136 :
137 :			if (filter != 0) {
138 :			b_refu = data->RefQ + 16;
139 :			b_refv = data->RefQ + 24;
140 :			interpolate8x8_halfpel[filter](b_refu, data->b_RefP[4] + offset, stride, data->rounding);
141 :			interpolate8x8_halfpel[filter](b_refv, data->b_RefP[5] + offset, stride, data->rounding);
142 :			} else {
143 :			b_refu = (uint8_t*)data->b_RefP[4] + offset;
144 :			b_refv = (uint8_t*)data->b_RefP[5] + offset;
145 :	edgomez	851	}
146 :	Isibaar	3
147 :	edgomez	851	sad = sad8bi(data->CurU, b_refu, f_refu, stride);
148 :			sad += sad8bi(data->CurV, b_refv, f_refv, stride);
149 :	Isibaar	3
150 :	edgomez	851	return sad;
151 :			}
152 :	Isibaar	3
153 :	edgomez	851	static int32_t
154 :	edgomez	978	ChromaSAD(const int dx, const int dy, const SearchData * const data)
155 :	edgomez	851	{
156 :			int sad;
157 :			const uint32_t stride = data->iEdgedWidth/2;
158 :	edgomez	978	int offset = (dx>>1) + (dy>>1)*stride;
159 :	syskin	1132	int next = 1;
160 :	Isibaar	3
161 :	edgomez	1053	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; /* it has been checked recently */
162 :			data->temp[5] = dx; data->temp[6] = dy; /* backup */
163 :	Isibaar	3
164 :	edgomez	851	switch (((dx & 1) << 1) \| (dy & 1)) {
165 :			case 0:
166 :	edgomez	978	sad = sad8(data->CurU, data->RefP[4] + offset, stride);
167 :			sad += sad8(data->CurV, data->RefP[5] + offset, stride);
168 :	edgomez	851	break;
169 :			case 1:
170 :	syskin	1132	next = stride;
171 :	edgomez	851	case 2:
172 :	syskin	1132	sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + next, stride);
173 :			sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + next, stride);
174 :	edgomez	851	break;
175 :			default:
176 :	edgomez	978	interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding);
177 :	edgomez	851	sad = sad8(data->CurU, data->RefQ, stride);
178 :
179 :	edgomez	978	interpolate8x8_halfpel_hv(data->RefQ, data->RefP[5] + offset, stride, data->rounding);
180 :	edgomez	851	sad += sad8(data->CurV, data->RefQ, stride);
181 :			break;
182 :			}
183 :	edgomez	1053	data->temp[7] = sad; /* backup, part 2 */
184 :	edgomez	851	return sad;
185 :	Isibaar	3	}
186 :
187 :	edgomez	851	static __inline const uint8_t *
188 :			GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data)
189 :	Isibaar	3	{
190 :	edgomez	1053	/* dir : 0 = forward, 1 = backward */
191 :	edgomez	987	const uint8_t const const direction = ( dir == 0 ? data->RefP : data->b_RefP );
192 :	edgomez	978	const int picture = ((x&1)<<1) \| (y&1);
193 :			const int offset = (x>>1) + (y>>1)*data->iEdgedWidth;
194 :			return direction[picture] + offset;
195 :	Isibaar	3	}
196 :
197 :	edgomez	1053	/* this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate */
198 :	edgomez	851	static __inline const uint8_t *
199 :			GetReference(const int x, const int y, const SearchData * const data)
200 :	Isibaar	3	{
201 :	edgomez	978	const int picture = ((x&1)<<1) \| (y&1);
202 :			const int offset = (x>>1) + (y>>1)*data->iEdgedWidth;
203 :			return data->RefP[picture] + offset;
204 :	Isibaar	3	}
205 :	edgomez	851
206 :			static uint8_t *
207 :			Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data)
208 :	Isibaar	3	{
209 :	edgomez	1053	/* create or find a qpel-precision reference picture; return pointer to it */
210 :	edgomez	851	uint8_t * Reference = data->RefQ + 16*dir;
211 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
212 :			const uint32_t rounding = data->rounding;
213 :			const int halfpel_x = x/2;
214 :			const int halfpel_y = y/2;
215 :			const uint8_t ref1, ref2, ref3, ref4;
216 :	suxen_drol	136
217 :	edgomez	851	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
218 :			ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
219 :			switch( ((x&1)<<1) + (y&1) ) {
220 :	edgomez	1053	case 3: /* x and y in qpel resolution - the "corners" (top left/right and */
221 :			/* bottom left/right) during qpel refinement */
222 :	edgomez	963	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
223 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
224 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
225 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
226 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
227 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
228 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
229 :	edgomez	851	break;
230 :	chl	184
231 :	edgomez	1053	case 1: /* x halfpel, y qpel - top or bottom during qpel refinement */
232 :	edgomez	851	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
233 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
234 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
235 :			break;
236 :	suxen_drol	136
237 :	edgomez	1053	case 2: /* x qpel, y halfpel - left or right during qpel refinement */
238 :	edgomez	851	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
239 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
240 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
241 :			break;
242 :	edgomez	195
243 :	edgomez	1053	default: /* pure halfpel position */
244 :	edgomez	963	return (uint8_t *) ref1;
245 :	edgomez	978
246 :	edgomez	851	}
247 :			return Reference;
248 :			}
249 :	Isibaar	3
250 :	edgomez	851	static uint8_t *
251 :			Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data)
252 :			{
253 :	edgomez	1053	/* create or find a qpel-precision reference picture; return pointer to it */
254 :	edgomez	851	uint8_t * Reference = data->RefQ + 16*dir;
255 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
256 :			const uint32_t rounding = data->rounding;
257 :			const int halfpel_x = x/2;
258 :			const int halfpel_y = y/2;
259 :			const uint8_t ref1, ref2, ref3, ref4;
260 :	chl	326
261 :	edgomez	851	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
262 :			switch( ((x&1)<<1) + (y&1) ) {
263 :	edgomez	1053	case 3:
264 :			/*
265 :			* x and y in qpel resolution - the "corners" (top left/right and
266 :			* bottom left/right) during qpel refinement
267 :			*/
268 :	edgomez	851	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
269 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
270 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
271 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
272 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
273 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
274 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
275 :			break;
276 :	chl	184
277 :	edgomez	1053	case 1: /* x halfpel, y qpel - top or bottom during qpel refinement */
278 :	edgomez	851	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
279 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
280 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
281 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
282 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
283 :			break;
284 :	chl	184
285 :	edgomez	1053	case 2: /* x qpel, y halfpel - left or right during qpel refinement */
286 :	edgomez	851	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
287 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
288 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
289 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
290 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
291 :			break;
292 :	edgomez	195
293 :	syskin	1129
294 :	edgomez	1053	default: /* pure halfpel position */
295 :	edgomez	851	return (uint8_t *) ref1;
296 :			}
297 :			return Reference;
298 :			}
299 :	edgomez	195
300 :	edgomez	851	/* CHECK_CANDIATE FUNCTIONS START */
301 :	edgomez	195
302 :	edgomez	851	static void
303 :	syskin	1132	CheckCandidate16(const int x, const int y, const SearchData * const data, const int Direction)
304 :	edgomez	851	{
305 :			int xc, yc;
306 :			const uint8_t * Reference;
307 :			VECTOR * current;
308 :			int32_t sad; uint32_t t;
309 :	edgomez	195
310 :	edgomez	851	if ( (x > data->max_dx) \|\| (x < data->min_dx)
311 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
312 :	edgomez	195
313 :	edgomez	851	if (!data->qpel_precision) {
314 :			Reference = GetReference(x, y, data);
315 :			current = data->currentMV;
316 :			xc = x; yc = y;
317 :	edgomez	1053	} else { /* x and y are in 1/4 precision */
318 :	edgomez	851	Reference = Interpolate16x16qpel(x, y, 0, data);
319 :	edgomez	1053	xc = x/2; yc = y/2; /* for chroma sad */
320 :	edgomez	851	current = data->currentQMV;
321 :			}
322 :	edgomez	195
323 :	syskin	1132	sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp);
324 :	edgomez	851	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
325 :	edgomez	195
326 :	edgomez	851	sad += (data->lambda16 * t * sad)>>10;
327 :	syskin	1132	data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10;
328 :	chl	175
329 :	syskin	1084	if (data->chroma && sad < data->iMinSAD[0])
330 :			sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
331 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
332 :	edgomez	195
333 :	edgomez	851	if (sad < data->iMinSAD[0]) {
334 :			data->iMinSAD[0] = sad;
335 :			current[0].x = x; current[0].y = y;
336 :	syskin	1132	*data->dir = Direction;
337 :	edgomez	851	}
338 :	edgomez	195
339 :	syskin	1132	if (data->temp[0] < data->iMinSAD[1]) {
340 :			data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; }
341 :			if (data->temp[1] < data->iMinSAD[2]) {
342 :			data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; }
343 :			if (data->temp[2] < data->iMinSAD[3]) {
344 :			data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; }
345 :			if (data->temp[3] < data->iMinSAD[4]) {
346 :			data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; }
347 :	Isibaar	3	}
348 :
349 :	edgomez	851	static void
350 :	Isibaar	1136	CheckCandidate16_subpel(const int x, const int y, const SearchData * const data, const int Direction)
351 :			{
352 :			int xc, yc;
353 :			const uint8_t *Reference;
354 :			VECTOR current, current2;
355 :			int32_t sad; uint32_t t;
356 :
357 :			if ( (x > data->max_dx) \|\| (x < data->min_dx)
358 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
359 :
360 :			if (!data->qpel_precision) {
361 :			Reference = GetReference(x, y, data);
362 :			current = data->currentMV;
363 :			current2 = data->currentMV2;
364 :			xc = x; yc = y;
365 :			} else { /* x and y are in 1/4 precision */
366 :			Reference = Interpolate16x16qpel(x, y, 0, data);
367 :			xc = x/2; yc = y/2; /* for chroma sad */
368 :			current = data->currentQMV;
369 :			current2 = data->currentQMV2;
370 :			}
371 :
372 :			sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp);
373 :			t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
374 :
375 :			sad += (data->lambda16 * t * sad)>>10;
376 :			data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10;
377 :
378 :			if (data->chroma && sad < data->iMinSAD[0])
379 :			sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
380 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
381 :
382 :			if (data->temp[0] < data->iMinSAD[1]) {
383 :			data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; }
384 :			if (data->temp[1] < data->iMinSAD[2]) {
385 :			data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; }
386 :			if (data->temp[2] < data->iMinSAD[3]) {
387 :			data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; }
388 :			if (data->temp[3] < data->iMinSAD[4]) {
389 :			data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; }
390 :
391 :			if (sad < data->iMinSAD[0]) {
392 :			(data->iMinSAD2) = (data->iMinSAD);
393 :			current2->x = current->x; current2->y = current->y;
394 :
395 :			data->iMinSAD[0] = sad;
396 :			current[0].x = x; current[0].y = y;
397 :			*data->dir = Direction;
398 :			return;
399 :			}
400 :
401 :			if (sad < *(data->iMinSAD2)) {
402 :			*(data->iMinSAD2) = sad;
403 :			current2->x = x; current2->y = y;
404 :			*data->dir = Direction;
405 :			}
406 :			}
407 :
408 :			static void
409 :	syskin	1132	CheckCandidate8(const int x, const int y, const SearchData * const data, const int Direction)
410 :	edgomez	851	{
411 :			int32_t sad; uint32_t t;
412 :			const uint8_t * Reference;
413 :	edgomez	974	VECTOR * current;
414 :	chl	326
415 :	edgomez	851	if ( (x > data->max_dx) \|\| (x < data->min_dx)
416 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
417 :	Isibaar	3
418 :	edgomez	974	if (!data->qpel_precision) {
419 :			Reference = GetReference(x, y, data);
420 :			current = data->currentMV;
421 :	edgomez	1053	} else { /* x and y are in 1/4 precision */
422 :	edgomez	974	Reference = Interpolate8x8qpel(x, y, 0, 0, data);
423 :			current = data->currentQMV;
424 :			}
425 :	Isibaar	3
426 :	edgomez	851	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
427 :			t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
428 :	Isibaar	3
429 :	edgomez	851	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10;
430 :	Isibaar	3
431 :	edgomez	851	if (sad < *(data->iMinSAD)) {
432 :			*(data->iMinSAD) = sad;
433 :	edgomez	974	current->x = x; current->y = y;
434 :	syskin	1132	*data->dir = Direction;
435 :	edgomez	851	}
436 :	Isibaar	3	}
437 :
438 :	edgomez	851	static void
439 :	syskin	1132	CheckCandidate32(const int x, const int y, const SearchData * const data, const int Direction)
440 :	edgomez	851	{
441 :			uint32_t t;
442 :			const uint8_t * Reference;
443 :	syskin	1132	int sad;
444 :	edgomez	851
445 :	edgomez	1053	if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) \|\| /* non-zero even value */
446 :	edgomez	851	(x > data->max_dx) \|\| (x < data->min_dx)
447 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
448 :
449 :			Reference = GetReference(x, y, data);
450 :			t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
451 :
452 :	syskin	1132	sad = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp);
453 :	edgomez	851
454 :	syskin	1132	sad += (data->lambda16 * t * sad) >> 10;
455 :			data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10;
456 :	edgomez	851
457 :	syskin	1132	if (sad < data->iMinSAD[0]) {
458 :			data->iMinSAD[0] = sad;
459 :	edgomez	851	data->currentMV[0].x = x; data->currentMV[0].y = y;
460 :	syskin	1132	*data->dir = Direction;
461 :			}
462 :	edgomez	851
463 :	syskin	1132	if (data->temp[0] < data->iMinSAD[1]) {
464 :			data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
465 :			if (data->temp[1] < data->iMinSAD[2]) {
466 :			data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
467 :			if (data->temp[2] < data->iMinSAD[3]) {
468 :			data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
469 :			if (data->temp[3] < data->iMinSAD[4]) {
470 :			data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
471 :	Isibaar	3	}
472 :	edgomez	195
473 :	edgomez	851	static void
474 :	syskin	1132	CheckCandidate16no4v(const int x, const int y, const SearchData * const data, const int Direction)
475 :	edgomez	851	{
476 :			int32_t sad, xc, yc;
477 :			const uint8_t * Reference;
478 :			uint32_t t;
479 :			VECTOR * current;
480 :
481 :	edgomez	963	if ( (x > data->max_dx) \|\| ( x < data->min_dx)
482 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
483 :	edgomez	851
484 :	edgomez	1053	if (data->rrv && (!(x&1) && x !=0) \| (!(y&1) && y !=0) ) return; /* non-zero even value */
485 :	edgomez	851
486 :	edgomez	1053	if (data->qpel_precision) { /* x and y are in 1/4 precision */
487 :	edgomez	851	Reference = Interpolate16x16qpel(x, y, 0, data);
488 :			current = data->currentQMV;
489 :			xc = x/2; yc = y/2;
490 :			} else {
491 :			Reference = GetReference(x, y, data);
492 :			current = data->currentMV;
493 :			xc = x; yc = y;
494 :			}
495 :			t = d_mv_bits(x, y, data->predMV, data->iFcode,
496 :			data->qpel^data->qpel_precision, data->rrv);
497 :
498 :			sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
499 :			sad += (data->lambda16 * t * sad)>>10;
500 :
501 :	syskin	1084	if (data->chroma && sad < *data->iMinSAD)
502 :			sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
503 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
504 :	edgomez	851
505 :			if (sad < *(data->iMinSAD)) {
506 :			*(data->iMinSAD) = sad;
507 :			current->x = x; current->y = y;
508 :	syskin	1132	*data->dir = Direction;
509 :	edgomez	851	}
510 :	chl	96	}
511 :	Isibaar	3
512 :	edgomez	851	static void
513 :	syskin	1132	CheckCandidate16I(const int x, const int y, const SearchData * const data, const int Direction)
514 :	chl	1077	{
515 :			int sad;
516 :			// int xc, yc;
517 :			const uint8_t * Reference;
518 :			// VECTOR * current;
519 :
520 :			if ( (x > data->max_dx) \|\| ( x < data->min_dx)
521 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
522 :
523 :			Reference = GetReference(x, y, data);
524 :			// xc = x; yc = y;
525 :
526 :			sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
527 :			// sad += d_mv_bits(x, y, data->predMV, data->iFcode, 0, 0);
528 :
529 :			/* if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
530 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
531 :			*/
532 :
533 :			if (sad < data->iMinSAD[0]) {
534 :			data->iMinSAD[0] = sad;
535 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
536 :	syskin	1132	*data->dir = Direction;
537 :	chl	1077	}
538 :			}
539 :
540 :			static void
541 :	syskin	1132	CheckCandidate32I(const int x, const int y, const SearchData * const data, const int Direction)
542 :	edgomez	851	{
543 :	edgomez	1053	/* maximum speed - for P/B/I decision */
544 :	edgomez	851	int32_t sad;
545 :
546 :			if ( (x > data->max_dx) \|\| (x < data->min_dx)
547 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
548 :
549 :	syskin	1135	sad = sad32v_c(data->Cur, data->RefP[0] + x + y*((int)data->iEdgedWidth),
550 :	syskin	1132	data->iEdgedWidth, data->temp);
551 :	edgomez	851
552 :			if (sad < *(data->iMinSAD)) {
553 :			*(data->iMinSAD) = sad;
554 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
555 :	syskin	1132	*data->dir = Direction;
556 :	edgomez	851	}
557 :	syskin	1132	if (data->temp[0] < data->iMinSAD[1]) {
558 :			data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
559 :			if (data->temp[1] < data->iMinSAD[2]) {
560 :			data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
561 :			if (data->temp[2] < data->iMinSAD[3]) {
562 :			data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
563 :			if (data->temp[3] < data->iMinSAD[4]) {
564 :			data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
565 :	edgomez	851
566 :	Isibaar	3	}
567 :
568 :	edgomez	851	static void
569 :	syskin	1132	CheckCandidateInt(const int xf, const int yf, const SearchData * const data, const int Direction)
570 :	edgomez	851	{
571 :			int32_t sad, xb, yb, xcf, ycf, xcb, ycb;
572 :			uint32_t t;
573 :			const uint8_t ReferenceF, ReferenceB;
574 :			VECTOR *current;
575 :
576 :	edgomez	953	if ((xf > data->max_dx) \|\| (xf < data->min_dx) \|\|
577 :			(yf > data->max_dy) \|\| (yf < data->min_dy))
578 :			return;
579 :	edgomez	851
580 :			if (!data->qpel_precision) {
581 :			ReferenceF = GetReference(xf, yf, data);
582 :			xb = data->currentMV[1].x; yb = data->currentMV[1].y;
583 :			ReferenceB = GetReferenceB(xb, yb, 1, data);
584 :			current = data->currentMV;
585 :			xcf = xf; ycf = yf;
586 :			xcb = xb; ycb = yb;
587 :			} else {
588 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
589 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
590 :			current = data->currentQMV;
591 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
592 :			xcf = xf/2; ycf = yf/2;
593 :			xcb = xb/2; ycb = yb/2;
594 :			}
595 :
596 :			t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0)
597 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0);
598 :
599 :			sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
600 :			sad += (data->lambda16 * t * sad)>>10;
601 :
602 :	syskin	1084	if (data->chroma && sad < *data->iMinSAD)
603 :			sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3],
604 :			(ycf >> 1) + roundtab_79[ycf & 0x3],
605 :			(xcb >> 1) + roundtab_79[xcb & 0x3],
606 :			(ycb >> 1) + roundtab_79[ycb & 0x3], data);
607 :	edgomez	851
608 :			if (sad < *(data->iMinSAD)) {
609 :			*(data->iMinSAD) = sad;
610 :			current->x = xf; current->y = yf;
611 :	syskin	1132	*data->dir = Direction;
612 :	edgomez	851	}
613 :	Isibaar	3	}
614 :
615 :	edgomez	851	static void
616 :	syskin	1132	CheckCandidateDirect(const int x, const int y, const SearchData * const data, const int Direction)
617 :	edgomez	851	{
618 :			int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0;
619 :			uint32_t k;
620 :			const uint8_t *ReferenceF;
621 :			const uint8_t *ReferenceB;
622 :			VECTOR mvs, b_mvs;
623 :
624 :	edgomez	953	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
625 :	edgomez	851
626 :			for (k = 0; k < 4; k++) {
627 :			mvs.x = data->directmvF[k].x + x;
628 :			b_mvs.x = ((x == 0) ?
629 :			data->directmvB[k].x
630 :			: mvs.x - data->referencemv[k].x);
631 :
632 :			mvs.y = data->directmvF[k].y + y;
633 :			b_mvs.y = ((y == 0) ?
634 :			data->directmvB[k].y
635 :			: mvs.y - data->referencemv[k].y);
636 :
637 :	edgomez	953	if ((mvs.x > data->max_dx) \|\| (mvs.x < data->min_dx) \|\|
638 :			(mvs.y > data->max_dy) \|\| (mvs.y < data->min_dy) \|\|
639 :			(b_mvs.x > data->max_dx) \|\| (b_mvs.x < data->min_dx) \|\|
640 :			(b_mvs.y > data->max_dy) \|\| (b_mvs.y < data->min_dy) )
641 :			return;
642 :	edgomez	851
643 :			if (data->qpel) {
644 :			xcf += mvs.x/2; ycf += mvs.y/2;
645 :			xcb += b_mvs.x/2; ycb += b_mvs.y/2;
646 :			} else {
647 :			xcf += mvs.x; ycf += mvs.y;
648 :			xcb += b_mvs.x; ycb += b_mvs.y;
649 :	edgomez	1053	mvs.x = 2; mvs.y = 2; /* we move to qpel precision anyway */
650 :	edgomez	851	b_mvs.x = 2; b_mvs.y = 2;
651 :			}
652 :
653 :			ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
654 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
655 :
656 :			sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
657 :			ReferenceF, ReferenceB, data->iEdgedWidth);
658 :			if (sad > *(data->iMinSAD)) return;
659 :			}
660 :
661 :			sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
662 :
663 :	syskin	1084	if (data->chroma && sad < *data->iMinSAD)
664 :			sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
665 :			(ycf >> 3) + roundtab_76[ycf & 0xf],
666 :			(xcb >> 3) + roundtab_76[xcb & 0xf],
667 :			(ycb >> 3) + roundtab_76[ycb & 0xf], data);
668 :	edgomez	851
669 :			if (sad < *(data->iMinSAD)) {
670 :			*(data->iMinSAD) = sad;
671 :			data->currentMV->x = x; data->currentMV->y = y;
672 :	syskin	1132	*data->dir = Direction;
673 :	edgomez	851	}
674 :	Isibaar	3	}
675 :
676 :	edgomez	851	static void
677 :	syskin	1132	CheckCandidateDirectno4v(const int x, const int y, const SearchData * const data, const int Direction)
678 :	Isibaar	3	{
679 :	edgomez	851	int32_t sad, xcf, ycf, xcb, ycb;
680 :			const uint8_t *ReferenceF;
681 :			const uint8_t *ReferenceB;
682 :			VECTOR mvs, b_mvs;
683 :	Isibaar	3
684 :	edgomez	963	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
685 :	Isibaar	3
686 :	edgomez	851	mvs.x = data->directmvF[0].x + x;
687 :			b_mvs.x = ((x == 0) ?
688 :			data->directmvB[0].x
689 :			: mvs.x - data->referencemv[0].x);
690 :	Isibaar	3
691 :	edgomez	851	mvs.y = data->directmvF[0].y + y;
692 :			b_mvs.y = ((y == 0) ?
693 :			data->directmvB[0].y
694 :			: mvs.y - data->referencemv[0].y);
695 :	Isibaar	3
696 :	edgomez	963	if ( (mvs.x > data->max_dx) \|\| (mvs.x < data->min_dx)
697 :			\|\| (mvs.y > data->max_dy) \|\| (mvs.y < data->min_dy)
698 :			\|\| (b_mvs.x > data->max_dx) \|\| (b_mvs.x < data->min_dx)
699 :			\|\| (b_mvs.y > data->max_dy) \|\| (b_mvs.y < data->min_dy) ) return;
700 :	Isibaar	3
701 :	edgomez	851	if (data->qpel) {
702 :			xcf = 4(mvs.x/2); ycf = 4(mvs.y/2);
703 :			xcb = 4(b_mvs.x/2); ycb = 4(b_mvs.y/2);
704 :			ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
705 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
706 :			} else {
707 :			xcf = 4mvs.x; ycf = 4mvs.y;
708 :			xcb = 4b_mvs.x; ycb = 4b_mvs.y;
709 :			ReferenceF = GetReference(mvs.x, mvs.y, data);
710 :			ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data);
711 :			}
712 :
713 :			sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
714 :			sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
715 :
716 :	syskin	1084	if (data->chroma && sad < *data->iMinSAD)
717 :			sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
718 :			(ycf >> 3) + roundtab_76[ycf & 0xf],
719 :			(xcb >> 3) + roundtab_76[xcb & 0xf],
720 :			(ycb >> 3) + roundtab_76[ycb & 0xf], data);
721 :	edgomez	851
722 :			if (sad < *(data->iMinSAD)) {
723 :			*(data->iMinSAD) = sad;
724 :			data->currentMV->x = x; data->currentMV->y = y;
725 :	syskin	1132	*data->dir = Direction;
726 :	edgomez	851	}
727 :	Isibaar	3	}
728 :
729 :	edgomez	851
730 :			static void
731 :	syskin	1132	CheckCandidateRD16(const int x, const int y, const SearchData * const data, const int Direction)
732 :	Isibaar	3	{
733 :
734 :	edgomez	962	int16_t in = data->dctSpace, coeff = data->dctSpace + 64;
735 :	syskin	1108	int32_t rd = 0;
736 :	edgomez	851	VECTOR * current;
737 :			const uint8_t * ptr;
738 :			int i, cbp = 0, t, xc, yc;
739 :	edgomez	195
740 :	edgomez	851	if ( (x > data->max_dx) \|\| (x < data->min_dx)
741 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
742 :	edgomez	195
743 :	edgomez	851	if (!data->qpel_precision) {
744 :			ptr = GetReference(x, y, data);
745 :			current = data->currentMV;
746 :			xc = x; yc = y;
747 :	edgomez	1053	} else { /* x and y are in 1/4 precision */
748 :	edgomez	851	ptr = Interpolate16x16qpel(x, y, 0, data);
749 :			current = data->currentQMV;
750 :			xc = x/2; yc = y/2;
751 :			}
752 :	Isibaar	3
753 :	edgomez	851	for(i = 0; i < 4; i++) {
754 :			int s = 8((i&1) + (i>>1)data->iEdgedWidth);
755 :			transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth);
756 :	syskin	1108	rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i);
757 :	edgomez	851	}
758 :	Isibaar	3
759 :	syskin	1108	rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
760 :	edgomez	195
761 :	syskin	1071	if (data->temp[0] + t < data->iMinSAD[1]) {
762 :	chl	1115	data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) \| (cbp&32); }
763 :	syskin	1071	if (data->temp[1] < data->iMinSAD[2]) {
764 :	chl	1115	data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) \| (cbp&16); }
765 :	syskin	1071	if (data->temp[2] < data->iMinSAD[3]) {
766 :	chl	1115	data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) \| (cbp&8); }
767 :	syskin	1071	if (data->temp[3] < data->iMinSAD[4]) {
768 :	chl	1115	data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) \| (cbp&4); }
769 :	syskin	1071
770 :	syskin	1108	rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
771 :	edgomez	851
772 :	syskin	1108	if (rd >= data->iMinSAD[0]) return;
773 :	edgomez	851
774 :	edgomez	1053	/* chroma */
775 :	edgomez	1022	xc = (xc >> 1) + roundtab_79[xc & 0x3];
776 :			yc = (yc >> 1) + roundtab_79[yc & 0x3];
777 :	edgomez	851
778 :	edgomez	1053	/* chroma U */
779 :	syskin	1084	ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
780 :			transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2);
781 :	syskin	1108	rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4);
782 :			if (rd >= data->iMinSAD[0]) return;
783 :	edgomez	1022
784 :	edgomez	1053	/* chroma V */
785 :	syskin	1084	ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
786 :			transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2);
787 :	syskin	1108	rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5);
788 :	edgomez	1022
789 :	syskin	1108	rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) \| ((cbp & 3) << 3)].len;
790 :	edgomez	851
791 :	syskin	1108	if (rd < data->iMinSAD[0]) {
792 :			data->iMinSAD[0] = rd;
793 :	edgomez	851	current[0].x = x; current[0].y = y;
794 :	syskin	1132	*data->dir = Direction;
795 :	syskin	1084	*data->cbp = cbp;
796 :	edgomez	851	}
797 :	syskin	1071	}
798 :	edgomez	851
799 :			static void
800 :	syskin	1132	CheckCandidateRD8(const int x, const int y, const SearchData * const data, const int Direction)
801 :	chl	96	{
802 :
803 :	edgomez	962	int16_t in = data->dctSpace, coeff = data->dctSpace + 64;
804 :	syskin	1108	int32_t rd;
805 :	edgomez	851	VECTOR * current;
806 :			const uint8_t * ptr;
807 :	edgomez	1022	int cbp = 0;
808 :	edgomez	195
809 :	edgomez	851	if ( (x > data->max_dx) \|\| (x < data->min_dx)
810 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
811 :	edgomez	195
812 :	edgomez	851	if (!data->qpel_precision) {
813 :			ptr = GetReference(x, y, data);
814 :			current = data->currentMV;
815 :	edgomez	1053	} else { /* x and y are in 1/4 precision */
816 :	edgomez	851	ptr = Interpolate8x8qpel(x, y, 0, 0, data);
817 :			current = data->currentQMV;
818 :			}
819 :	chl	96
820 :	edgomez	851	transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth);
821 :	syskin	1108	rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5);
822 :			rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
823 :	chl	96
824 :	syskin	1108	if (rd < data->iMinSAD[0]) {
825 :	syskin	1084	*data->cbp = cbp;
826 :	syskin	1108	data->iMinSAD[0] = rd;
827 :	edgomez	851	current[0].x = x; current[0].y = y;
828 :	syskin	1132	*data->dir = Direction;
829 :	edgomez	851	}
830 :			}
831 :	chl	96
832 :	edgomez	851	/* CHECK_CANDIATE FUNCTIONS END */
833 :	edgomez	195
834 :	edgomez	851	/* MAINSEARCH FUNCTIONS START */
835 :	chl	96
836 :	edgomez	851	static void
837 :	syskin	1132	AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate)
838 :	edgomez	851	{
839 :	edgomez	195
840 :	edgomez	851	/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
841 :	edgomez	195
842 :	syskin	1132	unsigned int * const iDirection = data->dir;
843 :	edgomez	195
844 :	edgomez	1053	for(;;) { /* forever */
845 :	syskin	1132	*iDirection = 0;
846 :	edgomez	851	if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
847 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
848 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
849 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
850 :	edgomez	195
851 :	edgomez	851	/* now we're doing diagonal checks near our candidate */
852 :	edgomez	195
853 :	syskin	1132	if (iDirection) { / if anything found */
854 :			bDirection = *iDirection;
855 :			*iDirection = 0;
856 :	edgomez	851	x = data->currentMV->x; y = data->currentMV->y;
857 :	edgomez	1053	if (bDirection & 3) { /* our candidate is left or right */
858 :	edgomez	851	CHECK_CANDIDATE(x, y + iDiamondSize, 8);
859 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
860 :	edgomez	1053	} else { /* what remains here is up or down */
861 :	edgomez	851	CHECK_CANDIDATE(x + iDiamondSize, y, 2);
862 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1);
863 :			}
864 :	edgomez	195
865 :	syskin	1132	if (*iDirection) {
866 :			bDirection += *iDirection;
867 :	edgomez	851	x = data->currentMV->x; y = data->currentMV->y;
868 :			}
869 :	edgomez	1053	} else { /* about to quit, eh? not so fast.... */
870 :	edgomez	851	switch (bDirection) {
871 :			case 2:
872 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
873 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
874 :	edgomez	195	break;
875 :	edgomez	851	case 1:
876 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
877 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
878 :	edgomez	195	break;
879 :	edgomez	851	case 2 + 4:
880 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
881 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
882 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
883 :			break;
884 :			case 4:
885 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
886 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
887 :			break;
888 :	edgomez	195	case 8:
889 :	edgomez	851	CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
890 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
891 :	edgomez	195	break;
892 :	edgomez	851	case 1 + 4:
893 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
894 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
895 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
896 :	edgomez	195	break;
897 :	edgomez	851	case 2 + 8:
898 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
899 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
900 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
901 :			break;
902 :			case 1 + 8:
903 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
904 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
905 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
906 :			break;
907 :	edgomez	1053	default: /* 1+2+4+8 == we didn't find anything at all */
908 :	edgomez	851	CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
909 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
910 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
911 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
912 :			break;
913 :	chl	96	}
914 :	syskin	1132	if (!iDirection) break; / ok, the end. really */
915 :			bDirection = *iDirection;
916 :	edgomez	851	x = data->currentMV->x; y = data->currentMV->y;
917 :	chl	345	}
918 :	edgomez	195	}
919 :	chl	96	}
920 :
921 :	edgomez	851	static void
922 :	syskin	1132	SquareSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate)
923 :	chl	96	{
924 :	syskin	1132	unsigned int * const iDirection = data->dir;
925 :	edgomez	195
926 :	edgomez	851	do {
927 :	syskin	1132	*iDirection = 0;
928 :	edgomez	851	if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
929 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
930 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
931 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
932 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
933 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
934 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
935 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
936 :	chl	96
937 :	syskin	1132	bDirection = *iDirection;
938 :	edgomez	851	x = data->currentMV->x; y = data->currentMV->y;
939 :	syskin	1132	} while (*iDirection);
940 :	chl	96	}
941 :
942 :	edgomez	851	static void
943 :	syskin	1132	DiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate)
944 :	chl	181	{
945 :
946 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
947 :
948 :	syskin	1132	unsigned int * const iDirection = data->dir;
949 :	edgomez	195
950 :	edgomez	851	do {
951 :	syskin	1132	*iDirection = 0;
952 :	edgomez	851	if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
953 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
954 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
955 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
956 :
957 :			/* now we're doing diagonal checks near our candidate */
958 :
959 :	syskin	1132	if (iDirection) { / checking if anything found */
960 :			bDirection = *iDirection;
961 :			*iDirection = 0;
962 :	edgomez	851	x = data->currentMV->x; y = data->currentMV->y;
963 :	edgomez	1053	if (bDirection & 3) { /* our candidate is left or right */
964 :	edgomez	851	CHECK_CANDIDATE(x, y + iDiamondSize, 8);
965 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
966 :	edgomez	1053	} else { /* what remains here is up or down */
967 :	edgomez	851	CHECK_CANDIDATE(x + iDiamondSize, y, 2);
968 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1);
969 :			}
970 :	syskin	1132	bDirection += *iDirection;
971 :	edgomez	851	x = data->currentMV->x; y = data->currentMV->y;
972 :			}
973 :			}
974 :	syskin	1132	while (*iDirection);
975 :	edgomez	851	}
976 :	chl	181
977 :	edgomez	851	/* MAINSEARCH FUNCTIONS END */
978 :	chl	181
979 :	edgomez	851	static void
980 :	Isibaar	1136	SubpelRefine_Fast(SearchData * data, CheckFunc * CheckCandidate)
981 :			{
982 :			/* Do a half-pel or q-pel refinement */
983 :			VECTOR centerMV;
984 :			VECTOR second_best;
985 :			int best_sad = *data->iMinSAD;
986 :			int xo, yo, xo2, yo2;
987 :			int size = 2;
988 :			CheckFunc *backupFunc = CheckCandidate;
989 :
990 :			if(data->qpel_precision)
991 :			size = 1;
992 :
993 :			centerMV = *data->currentMV;
994 :			data->iMinSAD = 256 4096;
995 :
996 :			CHECK_CANDIDATE(centerMV.x, centerMV.y - size, 0);
997 :			CHECK_CANDIDATE(centerMV.x + size, centerMV.y - size, 0);
998 :			CHECK_CANDIDATE(centerMV.x + size, centerMV.y, 0);
999 :			CHECK_CANDIDATE(centerMV.x + size, centerMV.y + size, 0);
1000 :
1001 :			CHECK_CANDIDATE(centerMV.x, centerMV.y + size, 0);
1002 :			CHECK_CANDIDATE(centerMV.x - size, centerMV.y + size, 0);
1003 :			CHECK_CANDIDATE(centerMV.x - size, centerMV.y, 0);
1004 :			CHECK_CANDIDATE(centerMV.x - size, centerMV.y - size, 0);
1005 :
1006 :			second_best = *data->currentMV;
1007 :
1008 :			if(data->qpel_precision)
1009 :			second_best.x = 2; second_best.y = 2;
1010 :
1011 :			data->currentMV[0] = centerMV;
1012 :			*data->iMinSAD = best_sad;
1013 :
1014 :			centerMV = data->qpel_precision ? data->currentQMV : data->currentMV;
1015 :
1016 :			xo = centerMV.x;
1017 :			yo = centerMV.y;
1018 :			xo2 = second_best.x;
1019 :			yo2 = second_best.y;
1020 :
1021 :			CheckCandidate = CheckCandidate16_subpel;
1022 :			data->iMinSAD2 = 256 4096;
1023 :
1024 :			if (yo == yo2)
1025 :			{
1026 :			CHECK_CANDIDATE((xo+xo2)>>1, yo, 0);
1027 :			CHECK_CANDIDATE(xo, yo-1, 0);
1028 :			CHECK_CANDIDATE(xo, yo+1, 0);
1029 :
1030 :			if(best_sad <= *data->iMinSAD2)
1031 :			goto ende;
1032 :
1033 :			if(data->currentQMV[0].x == data->currentQMV2[0].x) {
1034 :			CHECK_CANDIDATE((xo+xo2)>>1, yo-1, 0);
1035 :			CHECK_CANDIDATE((xo+xo2)>>1, yo+1, 0);
1036 :			goto ende;
1037 :			}
1038 :			else {
1039 :			CHECK_CANDIDATE((xo+xo2)>>1,
1040 :			(data->currentQMV[0].x == xo) ? data->currentQMV[0].y : data->currentQMV2[0].y,
1041 :			0);
1042 :			goto ende;
1043 :			}
1044 :			}
1045 :
1046 :			if (xo == xo2)
1047 :			{
1048 :			CHECK_CANDIDATE(xo, (yo+yo2)>>1, 0);
1049 :			CHECK_CANDIDATE(xo-1, yo, 0);
1050 :			CHECK_CANDIDATE(xo+1, yo, 0);
1051 :
1052 :			if(best_sad < *data->iMinSAD2)
1053 :			goto ende;
1054 :
1055 :			if(data->currentQMV[0].y == data->currentQMV2[0].y) {
1056 :			CHECK_CANDIDATE(xo-1, (yo+yo2)>>1, 0);
1057 :			CHECK_CANDIDATE(xo+1, (yo+yo2)>>1, 0);
1058 :			goto ende;
1059 :			}
1060 :			else {
1061 :			CHECK_CANDIDATE((data->currentQMV[0].y == yo) ? data->currentQMV[0].x : data->currentQMV2[0].x, (yo+yo2)>>1, 0);
1062 :			goto ende;
1063 :			}
1064 :			}
1065 :
1066 :			CHECK_CANDIDATE(xo, (yo+yo2)>>1, 0);
1067 :			CHECK_CANDIDATE((xo+xo2)>>1, yo, 0);
1068 :
1069 :			if(best_sad <= *data->iMinSAD2)
1070 :			goto ende;
1071 :
1072 :			CHECK_CANDIDATE((xo+xo2)>>1, (yo+yo2)>>1, 0);
1073 :
1074 :			ende:
1075 :			CheckCandidate = backupFunc;
1076 :			}
1077 :
1078 :			static void
1079 :	syskin	1132	SubpelRefine(const SearchData * const data, CheckFunc * const CheckCandidate)
1080 :	edgomez	851	{
1081 :			/* Do a half-pel or q-pel refinement */
1082 :			const VECTOR centerMV = data->qpel_precision ? data->currentQMV : data->currentMV;
1083 :	chl	181
1084 :	edgomez	851	CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0);
1085 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0);
1086 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, 0);
1087 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, 0);
1088 :			CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, 0);
1089 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, 0);
1090 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, 0);
1091 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, 0);
1092 :			}
1093 :	chl	181
1094 :	edgomez	851	static __inline int
1095 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
1096 :			const int x, const int y,
1097 :			const uint32_t stride, const uint32_t iQuant, int rrv)
1098 :	chl	181
1099 :	edgomez	851	{
1100 :	edgomez	963	int offset = (x + ystride)8;
1101 :	edgomez	851	if(!rrv) {
1102 :	edgomez	963	uint32_t sadC = sad8(current->u + offset,
1103 :			reference->u + offset, stride);
1104 :	edgomez	851	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
1105 :	edgomez	963	sadC += sad8(current->v + offset,
1106 :			reference->v + offset, stride);
1107 :	edgomez	851	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
1108 :			return 1;
1109 :	chl	181
1110 :	edgomez	851	} else {
1111 :	edgomez	963	uint32_t sadC = sad16(current->u + 2*offset,
1112 :			reference->u + 2offset, stride, 2564096);
1113 :	edgomez	851	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
1114 :	edgomez	963	sadC += sad16(current->v + 2*offset,
1115 :			reference->v + 2offset, stride, 2564096);
1116 :	edgomez	851	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
1117 :			return 1;
1118 :	chl	181	}
1119 :	edgomez	851	}
1120 :	edgomez	504
1121 :	edgomez	851	static __inline void
1122 :	Isibaar	1095	ZeroMacroblockP(MACROBLOCK *pMB, const int32_t sad)
1123 :			{
1124 :			pMB->mode = MODE_INTER;
1125 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV;
1126 :			pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV;
1127 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
1128 :			}
1129 :
1130 :			static __inline void
1131 :	Isibaar	1136	ModeDecision_Fast(SearchData * const Data,
1132 :			MACROBLOCK * const pMB,
1133 :			const MACROBLOCK * const pMBs,
1134 :			const int x, const int y,
1135 :			const MBParam * const pParam,
1136 :			const uint32_t MotionFlags,
1137 :			const uint32_t VopFlags,
1138 :			const uint32_t VolFlags,
1139 :			const IMAGE * const pCurrent,
1140 :			const IMAGE * const pRef,
1141 :			const IMAGE * const vGMC,
1142 :			const int coding_type)
1143 :			{
1144 :			int mode = MODE_INTER;
1145 :			int mcsel = 0;
1146 :			int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
1147 :			const uint32_t iQuant = pMB->quant;
1148 :			const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0);
1149 :			int sad;
1150 :			int min_rd = -1, intra_rd, i, cbp, c[2] = {0, 0};
1151 :			VECTOR backup[5], *v;
1152 :			int sad_backup[5];
1153 :			int InterBias = MV16_INTER_BIAS;
1154 :			int thresh = 0;
1155 :			int count = 0;
1156 :			int top = 0, top_right = 0, left = 0;
1157 :
1158 :			pMB->mcsel = 0;
1159 :
1160 :			/* INTER <-> INTER4V decision */
1161 :			if ((Data->iMinSAD[0] + 125 < Data->iMinSAD[1] +
1162 :			Data->iMinSAD[2] + Data->iMinSAD[3] + Data->iMinSAD[4])) { /* normal, fast, SAD-based mode decision */
1163 :			if (inter4v == 0 \|\| Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1164 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
1165 :			mode = MODE_INTER;
1166 :			sad = Data->iMinSAD[0];
1167 :			} else {
1168 :			mode = MODE_INTER4V;
1169 :			sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
1170 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
1171 :			Data->iMinSAD[0] = sad;
1172 :			}
1173 :
1174 :			/* final skip decision, a.k.a. "the vector you found, really that good?" */
1175 :			if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
1176 :			if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH)
1177 :			if (Data->chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) {
1178 :			mode = MODE_NOT_CODED;
1179 :			goto early_out;
1180 :			}
1181 :
1182 :			/* mcsel */
1183 :			if (coding_type == S_VOP) {
1184 :
1185 :			int32_t iSAD = sad16(Data->Cur,
1186 :			vGMC->y + 16yData->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
1187 :
1188 :			if (Data->chroma) {
1189 :			iSAD += sad8(Data->CurU, vGMC->u + 8y(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
1190 :			iSAD += sad8(Data->CurV, vGMC->v + 8y(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
1191 :			}
1192 :
1193 :			if (iSAD <= sad) { /* mode decision GMC */
1194 :			mode = MODE_INTER;
1195 :			mcsel = 1;
1196 :			sad = iSAD;
1197 :			}
1198 :
1199 :			}
1200 :			} else { /* Rate-Distortion INTER<->INTER4V */
1201 :			Data->iQuant = iQuant;
1202 :			Data->cbp = c;
1203 :			v = Data->qpel ? Data->currentQMV : Data->currentMV;
1204 :
1205 :			/* final skip decision, a.k.a. "the vector you found, really that good?" */
1206 :			if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
1207 :			if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH)
1208 :			if (Data->chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) {
1209 :			mode = MODE_NOT_CODED;
1210 :			goto early_out;
1211 :			}
1212 :
1213 :			for (i = 0; i < 5; i++) {
1214 :			sad_backup[i] = Data->iMinSAD[i];
1215 :			Data->iMinSAD[i] = 256*4096;
1216 :			backup[i] = v[i];
1217 :			}
1218 :
1219 :			min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags);
1220 :			cbp = *Data->cbp;
1221 :			sad = sad_backup[0];
1222 :
1223 :			if (coding_type == S_VOP) {
1224 :			int gmc_rd;
1225 :			int32_t iSAD = sad16(Data->Cur,
1226 :			vGMC->y + 16yData->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
1227 :
1228 :			if (Data->chroma) {
1229 :			iSAD += sad8(Data->CurU, vGMC->u + 8y(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
1230 :			iSAD += sad8(Data->CurV, vGMC->v + 8y(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
1231 :			}
1232 :
1233 :			Data->iMinSAD = min_rd += BITS_MULT1; /* mcsel */
1234 :			gmc_rd = findRDgmc(Data, vGMC, x, y);
1235 :			if (gmc_rd < min_rd) {
1236 :			mcsel = 1;
1237 :			*Data->iMinSAD = min_rd = gmc_rd;
1238 :			mode = MODE_INTER;
1239 :			cbp = *Data->cbp;
1240 :			sad = iSAD;
1241 :			}
1242 :			}
1243 :
1244 :			if (inter4v) {
1245 :			int v4_rd;
1246 :			v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
1247 :			if (v4_rd < min_rd) {
1248 :			Data->iMinSAD[0] = min_rd = v4_rd;
1249 :			mode = MODE_INTER4V;
1250 :			cbp = *Data->cbp;
1251 :			sad = sad_backup[1] + sad_backup[2] +
1252 :			sad_backup[3] + sad_backup[4] + IMV16X16 * (int32_t)iQuant;
1253 :			}
1254 :			}
1255 :			}
1256 :
1257 :			left = top = top_right = -1;
1258 :			thresh = 0;
1259 :
1260 :			if(x > 0 && y > 0 && x < pParam->mb_width) {
1261 :			left = (&pMBs[(x-1) + y * pParam->mb_width])->sad16; // left
1262 :			top = (&pMBs[x + (y-1) * pParam->mb_width])->sad16; // top
1263 :			top_right = (&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16; // top right
1264 :
1265 :			if(((&pMBs[(x-1) + y * pParam->mb_width])->mode != MODE_INTRA) &&
1266 :			((&pMBs[x + (y-1) * pParam->mb_width])->mode != MODE_INTRA) &&
1267 :			((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mode != MODE_INTRA))
1268 :			thresh = MAX(MAX(top, left), top_right);
1269 :			else
1270 :			thresh = MIN(MIN(top, left), top_right);
1271 :			}
1272 :
1273 :			/* INTRA <-> INTER decision */
1274 :			if (sad < thresh) { /* normal, fast, SAD-based mode decision */
1275 :			/* intra decision */
1276 :
1277 :			if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */
1278 :			if (y != 0)
1279 :			if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
1280 :			if (x != 0)
1281 :			if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
1282 :
1283 :			if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */
1284 :			if (Data->rrv) InterBias *= 4;
1285 :
1286 :			if (InterBias < sad) {
1287 :			int32_t deviation;
1288 :			if (!Data->rrv)
1289 :			deviation = dev16(Data->Cur, Data->iEdgedWidth);
1290 :			else
1291 :			deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */
1292 :			dev16(Data->Cur+16, Data->iEdgedWidth) +
1293 :			dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) +
1294 :			dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth);
1295 :
1296 :			if (deviation < (sad - InterBias)) mode = MODE_INTRA;
1297 :			}
1298 :
1299 :			pMB->cbp = 63;
1300 :			} else { /* Rate-Distortion INTRA<->INTER */
1301 :			if(min_rd < 0) {
1302 :			Data->iQuant = iQuant;
1303 :			Data->cbp = c;
1304 :			v = Data->qpel ? Data->currentQMV : Data->currentMV;
1305 :
1306 :			for (i = 0; i < 5; i++) {
1307 :			Data->iMinSAD[i] = 256*4096;
1308 :			backup[i] = v[i];
1309 :			}
1310 :
1311 :			if(mode == MODE_INTER) {
1312 :			min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags);
1313 :			cbp = *Data->cbp;
1314 :
1315 :			if (coding_type == S_VOP) {
1316 :			int gmc_rd;
1317 :
1318 :			Data->iMinSAD = min_rd += BITS_MULT1; /* mcsel */
1319 :			gmc_rd = findRDgmc(Data, vGMC, x, y);
1320 :			if (gmc_rd < min_rd) {
1321 :			mcsel = 1;
1322 :			*Data->iMinSAD = min_rd = gmc_rd;
1323 :			mode = MODE_INTER;
1324 :			cbp = *Data->cbp;
1325 :			}
1326 :			}
1327 :			}
1328 :
1329 :			if(mode == MODE_INTER4V) {
1330 :			int v4_rd;
1331 :			v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
1332 :			if (v4_rd < min_rd) {
1333 :			Data->iMinSAD[0] = min_rd = v4_rd;
1334 :			mode = MODE_INTER4V;
1335 :			cbp = *Data->cbp;
1336 :			}
1337 :			}
1338 :			}
1339 :
1340 :			intra_rd = findRDintra(Data);
1341 :			if (intra_rd < min_rd) {
1342 :			*Data->iMinSAD = min_rd = intra_rd;
1343 :			mode = MODE_INTRA;
1344 :			}
1345 :
1346 :			pMB->cbp = cbp;
1347 :			}
1348 :
1349 :			early_out:
1350 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
1351 :
1352 :			if (Data->rrv) {
1353 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1354 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1355 :			}
1356 :
1357 :			if (mode == MODE_INTER && mcsel == 0) {
1358 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1359 :
1360 :			if(Data->qpel) {
1361 :			pMB->qmvs[0] = pMB->qmvs[1]
1362 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1363 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1364 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1365 :			} else {
1366 :			pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1367 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1368 :			}
1369 :
1370 :			} else if (mode == MODE_INTER ) { // but mcsel == 1
1371 :
1372 :			pMB->mcsel = 1;
1373 :			if (Data->qpel) {
1374 :			pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
1375 :			pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
1376 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
1377 :			} else
1378 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
1379 :
1380 :			} else
1381 :			if (mode == MODE_INTER4V) ; /* anything here? */
1382 :			else /* INTRA, NOT_CODED */
1383 :			ZeroMacroblockP(pMB, 0);
1384 :
1385 :			pMB->mode = mode;
1386 :			}
1387 :
1388 :			static __inline void
1389 :	edgomez	1022	ModeDecision(SearchData * const Data,
1390 :			MACROBLOCK * const pMB,
1391 :			const MACROBLOCK * const pMBs,
1392 :			const int x, const int y,
1393 :			const MBParam * const pParam,
1394 :			const uint32_t MotionFlags,
1395 :			const uint32_t VopFlags,
1396 :			const uint32_t VolFlags,
1397 :			const IMAGE * const pCurrent,
1398 :	syskin	1084	const IMAGE * const pRef,
1399 :			const IMAGE * const vGMC,
1400 :			const int coding_type)
1401 :	edgomez	1022	{
1402 :			int mode = MODE_INTER;
1403 :	syskin	1084	int mcsel = 0;
1404 :	edgomez	1022	int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
1405 :			const uint32_t iQuant = pMB->quant;
1406 :
1407 :	syskin	1084	const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0);
1408 :	edgomez	1022
1409 :	syskin	1084	pMB->mcsel = 0;
1410 :
1411 :	edgomez	1107	if (!(VopFlags & XVID_VOP_MODEDECISION_RD)) { /* normal, fast, SAD-based mode decision */
1412 :	edgomez	1022	int sad;
1413 :			int InterBias = MV16_INTER_BIAS;
1414 :			if (inter4v == 0 \|\| Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1415 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
1416 :			mode = MODE_INTER;
1417 :			sad = Data->iMinSAD[0];
1418 :			} else {
1419 :			mode = MODE_INTER4V;
1420 :			sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
1421 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
1422 :			Data->iMinSAD[0] = sad;
1423 :			}
1424 :
1425 :			/* final skip decision, a.k.a. "the vector you found, really that good?" */
1426 :			if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
1427 :			if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH)
1428 :			if (Data->chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) {
1429 :			mode = MODE_NOT_CODED;
1430 :			sad = 0;
1431 :			}
1432 :
1433 :	syskin	1084	/* mcsel */
1434 :			if (coding_type == S_VOP) {
1435 :
1436 :			int32_t iSAD = sad16(Data->Cur,
1437 :			vGMC->y + 16yData->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
1438 :
1439 :			if (Data->chroma) {
1440 :			iSAD += sad8(Data->CurU, vGMC->u + 8y(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
1441 :			iSAD += sad8(Data->CurV, vGMC->v + 8y(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
1442 :			}
1443 :
1444 :			if (iSAD <= sad) { /* mode decision GMC */
1445 :			mode = MODE_INTER;
1446 :			mcsel = 1;
1447 :			sad = iSAD;
1448 :			}
1449 :
1450 :			}
1451 :
1452 :	edgomez	1022	/* intra decision */
1453 :
1454 :	edgomez	1053	if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */
1455 :	edgomez	1022	if (y != 0)
1456 :			if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
1457 :			if (x != 0)
1458 :			if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
1459 :
1460 :	syskin	1084	if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */
1461 :	edgomez	1022	if (Data->rrv) InterBias *= 4;
1462 :
1463 :	syskin	1084	if (InterBias < sad) {
1464 :	edgomez	1022	int32_t deviation;
1465 :	syskin	1084	if (!Data->rrv)
1466 :			deviation = dev16(Data->Cur, Data->iEdgedWidth);
1467 :			else
1468 :			deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */
1469 :			dev16(Data->Cur+16, Data->iEdgedWidth) +
1470 :			dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) +
1471 :			dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth);
1472 :	edgomez	1022
1473 :			if (deviation < (sad - InterBias)) mode = MODE_INTRA;
1474 :			}
1475 :
1476 :	syskin	1084	pMB->cbp = 63;
1477 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
1478 :
1479 :	syskin	1108	} else { /* Rate-Distortion */
1480 :	edgomez	1022
1481 :	syskin	1108	int min_rd, intra_rd, i, cbp, c[2] = {0, 0};
1482 :	edgomez	1022	VECTOR backup[5], *v;
1483 :			Data->iQuant = iQuant;
1484 :	syskin	1084	Data->cbp = c;
1485 :	edgomez	1022
1486 :			v = Data->qpel ? Data->currentQMV : Data->currentMV;
1487 :			for (i = 0; i < 5; i++) {
1488 :			Data->iMinSAD[i] = 256*4096;
1489 :			backup[i] = v[i];
1490 :			}
1491 :
1492 :	syskin	1108	min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags);
1493 :	syskin	1084	cbp = *Data->cbp;
1494 :
1495 :			if (coding_type == S_VOP) {
1496 :	syskin	1108	int gmc_rd;
1497 :			Data->iMinSAD = min_rd += BITS_MULT1; /* mcsel */
1498 :			gmc_rd = findRDgmc(Data, vGMC, x, y);
1499 :			if (gmc_rd < min_rd) {
1500 :	syskin	1084	mcsel = 1;
1501 :	syskin	1108	*Data->iMinSAD = min_rd = gmc_rd;
1502 :	syskin	1084	mode = MODE_INTER;
1503 :			cbp = *Data->cbp;
1504 :	edgomez	1022	}
1505 :	syskin	1084	}
1506 :	edgomez	1022
1507 :	syskin	1084	if (inter4v) {
1508 :	syskin	1108	int v4_rd;
1509 :			v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
1510 :			if (v4_rd < min_rd) {
1511 :			Data->iMinSAD[0] = min_rd = v4_rd;
1512 :	syskin	1084	mode = MODE_INTER4V;
1513 :			cbp = *Data->cbp;
1514 :			}
1515 :			}
1516 :	edgomez	1022
1517 :	syskin	1108	intra_rd = findRDintra(Data);
1518 :			if (intra_rd < min_rd) {
1519 :			*Data->iMinSAD = min_rd = intra_rd;
1520 :	syskin	1084	mode = MODE_INTRA;
1521 :	edgomez	1022	}
1522 :	syskin	1084
1523 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0;
1524 :			pMB->cbp = cbp;
1525 :	edgomez	1022	}
1526 :
1527 :			if (Data->rrv) {
1528 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1529 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1530 :			}
1531 :
1532 :	syskin	1084	if (mode == MODE_INTER && mcsel == 0) {
1533 :	edgomez	1022	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1534 :
1535 :			if(Data->qpel) {
1536 :			pMB->qmvs[0] = pMB->qmvs[1]
1537 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1538 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1539 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1540 :			} else {
1541 :			pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1542 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1543 :			}
1544 :
1545 :	syskin	1084	} else if (mode == MODE_INTER ) { // but mcsel == 1
1546 :
1547 :			pMB->mcsel = 1;
1548 :			if (Data->qpel) {
1549 :			pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
1550 :			pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
1551 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
1552 :			} else
1553 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
1554 :
1555 :			} else
1556 :			if (mode == MODE_INTER4V) ; /* anything here? */
1557 :			else /* INTRA, NOT_CODED */
1558 :	syskin	1108	ZeroMacroblockP(pMB, 0);
1559 :	edgomez	1022
1560 :			pMB->mode = mode;
1561 :			}
1562 :
1563 :	edgomez	851	bool
1564 :			MotionEstimation(MBParam * const pParam,
1565 :			FRAMEINFO * const current,
1566 :			FRAMEINFO * const reference,
1567 :			const IMAGE * const pRefH,
1568 :			const IMAGE * const pRefV,
1569 :			const IMAGE * const pRefHV,
1570 :	chl	1077	const IMAGE * const pGMC,
1571 :	edgomez	851	const uint32_t iLimit)
1572 :	chl	181	{
1573 :	edgomez	851	MACROBLOCK *const pMBs = current->mbs;
1574 :			const IMAGE *const pCurrent = &current->image;
1575 :			const IMAGE *const pRef = &reference->image;
1576 :	chl	181
1577 :	edgomez	851	uint32_t mb_width = pParam->mb_width;
1578 :			uint32_t mb_height = pParam->mb_height;
1579 :			const uint32_t iEdgedWidth = pParam->edged_width;
1580 :	suxen_drol	890	const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags);
1581 :	Isibaar	1136	int stat_thresh = 0;
1582 :	chl	181
1583 :	edgomez	851	uint32_t x, y;
1584 :			uint32_t iIntra = 0;
1585 :	syskin	1108	int32_t sad00;
1586 :	chl	1077	int skip_thresh = INITIAL_SKIP_THRESH * \
1587 :	edgomez	953	(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \
1588 :	edgomez	1107	(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1);
1589 :	chl	181
1590 :	edgomez	1053	/* some pre-initialized thingies for SearchP */
1591 :	syskin	1132	int32_t temp[8]; uint32_t dir;
1592 :	edgomez	851	VECTOR currentMV[5];
1593 :			VECTOR currentQMV[5];
1594 :	Isibaar	1136	VECTOR currentMV2[5];
1595 :			VECTOR currentQMV2[5];
1596 :	edgomez	851	int32_t iMinSAD[5];
1597 :	Isibaar	1136	int32_t iMinSAD2[5];
1598 :	syskin	1075	DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE);
1599 :	edgomez	851	SearchData Data;
1600 :			memset(&Data, 0, sizeof(SearchData));
1601 :			Data.iEdgedWidth = iEdgedWidth;
1602 :			Data.currentMV = currentMV;
1603 :			Data.currentQMV = currentQMV;
1604 :	Isibaar	1136	Data.currentMV2 = currentMV2;
1605 :			Data.currentQMV2 = currentQMV2;
1606 :	edgomez	851	Data.iMinSAD = iMinSAD;
1607 :	Isibaar	1136	Data.iMinSAD2 = iMinSAD2;
1608 :	edgomez	851	Data.temp = temp;
1609 :	syskin	1132	Data.dir = &dir;
1610 :	edgomez	851	Data.iFcode = current->fcode;
1611 :			Data.rounding = pParam->m_rounding_type;
1612 :	edgomez	963	Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0);
1613 :	edgomez	1107	Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP;
1614 :	chl	1077	Data.rrv = (current->vop_flags & XVID_VOP_REDUCED) ? 1:0;
1615 :	edgomez	962	Data.dctSpace = dct_space;
1616 :	edgomez	1022	Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT);
1617 :	edgomez	195
1618 :	edgomez	949	if ((current->vop_flags & XVID_VOP_REDUCED)) {
1619 :	edgomez	851	mb_width = (pParam->width + 31) / 32;
1620 :			mb_height = (pParam->height + 31) / 32;
1621 :			Data.qpel = 0;
1622 :			}
1623 :	chl	181
1624 :	edgomez	1053	Data.RefQ = pRefV->u; /* a good place, also used in MC (for similar purpose) */
1625 :	edgomez	851	if (sadInit) (*sadInit) ();
1626 :	chl	181
1627 :	edgomez	851	for (y = 0; y < mb_height; y++) {
1628 :			for (x = 0; x < mb_width; x++) {
1629 :			MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
1630 :	Isibaar	1136	MACROBLOCK prevMB = &reference->mbs[x + y pParam->mb_width];
1631 :	chl	181
1632 :	edgomez	851	if (!Data.rrv) pMB->sad16 =
1633 :			sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16,
1634 :			pRef->y + (x + y * iEdgedWidth) * 16,
1635 :			pParam->edged_width, pMB->sad8 );
1636 :	chl	181
1637 :	edgomez	851	else pMB->sad16 =
1638 :			sad32v_c(pCurrent->y + (x + y * iEdgedWidth) * 32,
1639 :			pRef->y + (x + y * iEdgedWidth) * 32,
1640 :			pParam->edged_width, pMB->sad8 );
1641 :	chl	181
1642 :	edgomez	851	if (Data.chroma) {
1643 :			Data.temp[7] = sad8(pCurrent->u + x8 + y(iEdgedWidth/2)*8,
1644 :			pRef->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2)
1645 :			+ sad8(pCurrent->v + (x + y(iEdgedWidth/2))8,
1646 :			pRef->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
1647 :			pMB->sad16 += Data.temp[7];
1648 :			}
1649 :
1650 :			sad00 = pMB->sad16;
1651 :
1652 :	edgomez	1053	/* initial skip decision */
1653 :			/* no early skip for GMC (global vector = skip vector is unknown!) */
1654 :	syskin	1084	if (current->coding_type != S_VOP) { /* no fast SKIP for S(GMC)-VOPs */
1655 :	edgomez	953	if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh)
1656 :	edgomez	851	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
1657 :	syskin	1108	ZeroMacroblockP(pMB, sad00);
1658 :			pMB->mode = MODE_NOT_CODED;
1659 :	edgomez	851	continue;
1660 :			}
1661 :			}
1662 :
1663 :	Isibaar	1136	if(MotionFlags & XVID_ME_DETECT_STATIC_MOTION) {
1664 :			if(x > 0 && y > 0 && x < pParam->mb_width)
1665 :			if(MVequal((&pMBs[(x-1) + y * pParam->mb_width])->mvs[0], zeroMV) &&
1666 :			MVequal((&pMBs[x + (y-1) * pParam->mb_width])->mvs[0], zeroMV) &&
1667 :			MVequal((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mvs[0], zeroMV) &&
1668 :			MVequal(prevMB->mvs[0], zeroMV)) {
1669 :			stat_thresh = MAX((&pMBs[(x-1) + y * pParam->mb_width])->sad16,
1670 :			MAX((&pMBs[x + (y-1) * pParam->mb_width])->sad16,
1671 :			MAX((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16,
1672 :			prevMB->sad16)));
1673 :			}
1674 :			else
1675 :			stat_thresh = MIN((&pMBs[(x-1) + y * pParam->mb_width])->sad16,
1676 :			MIN((&pMBs[x + (y-1) * pParam->mb_width])->sad16,
1677 :			MIN((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16,
1678 :			prevMB->sad16)));
1679 :			}
1680 :
1681 :	Isibaar	1095	if ((current->vop_flags & XVID_VOP_CARTOON) &&
1682 :	Isibaar	1136	(sad00 < pMB->quant * 4 * skip_thresh) \|\| (sad00 < stat_thresh)) { /* favorize (0,0) vector for cartoons */
1683 :	Isibaar	1095	ZeroMacroblockP(pMB, sad00);
1684 :			continue;
1685 :			}
1686 :
1687 :	edgomez	851	SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
1688 :	edgomez	1022	y, MotionFlags, current->vop_flags, current->vol_flags,
1689 :			&Data, pParam, pMBs, reference->mbs, pMB);
1690 :	edgomez	851
1691 :	Isibaar	1136	if(current->vop_flags & XVID_VOP_FAST_MODEDECISION_RD) {
1692 :			ModeDecision_Fast(&Data, pMB, pMBs, x, y, pParam,
1693 :			MotionFlags, current->vop_flags, current->vol_flags,
1694 :			pCurrent, pRef, pGMC, current->coding_type);
1695 :			}
1696 :			else {
1697 :			ModeDecision(&Data, pMB, pMBs, x, y, pParam,
1698 :			MotionFlags, current->vop_flags, current->vol_flags,
1699 :			pCurrent, pRef, pGMC, current->coding_type);
1700 :			}
1701 :	edgomez	1022
1702 :	edgomez	851	if (pMB->mode == MODE_INTRA)
1703 :			if (++iIntra > iLimit) return 1;
1704 :	chl	181	}
1705 :			}
1706 :	edgomez	851
1707 :			return 0;
1708 :	chl	181	}
1709 :
1710 :	syskin	1132	/* check if given vector is equal to any vector checked before */
1711 :			static __inline int
1712 :			vector_repeats(const VECTOR * const pmv, const int i)
1713 :			{
1714 :			unsigned int j;
1715 :			for (j = 0; j < i; j++)
1716 :			if (MVequal(pmv[i], pmv[j])) return 1; /* same vector has been checked already */
1717 :			return 0;
1718 :			}
1719 :	chl	181
1720 :	syskin	1132	/* make a binary mask that prevents diamonds/squares
1721 :			from checking a vector which has been checked as a prediction */
1722 :	edgomez	851	static __inline int
1723 :	syskin	1132	make_mask(const VECTOR * const pmv, const int i, const int current)
1724 :	chl	96	{
1725 :	syskin	1132	unsigned int mask = 255, j;
1726 :	edgomez	851	for (j = 0; j < i; j++) {
1727 :	syskin	1132	if (pmv[current].x == pmv[j].x) {
1728 :			if (pmv[current].y == pmv[j].y + iDiamondSize) mask &= ~4;
1729 :			else if (pmv[current].y == pmv[j].y - iDiamondSize) mask &= ~8;
1730 :	edgomez	851	} else
1731 :	syskin	1132	if (pmv[current].y == pmv[j].y) {
1732 :			if (pmv[current].x == pmv[j].x + iDiamondSize) mask &= ~1;
1733 :			else if (pmv[current].x == pmv[j].x - iDiamondSize) mask &= ~2;
1734 :	edgomez	851	}
1735 :			}
1736 :			return mask;
1737 :			}
1738 :	edgomez	195
1739 :	edgomez	851	static __inline void
1740 :			PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
1741 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
1742 :			{
1743 :	edgomez	1053	/* this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself */
1744 :	edgomez	851	if (rrv) { iWcount /= 2; iHcount /= 2; }
1745 :	edgomez	195
1746 :	edgomez	1053	if ( (y != 0) && (x < (iWcount-1)) ) { /* [5] top-right neighbour */
1747 :	edgomez	851	pmv[5].x = EVEN(pmv[3].x);
1748 :			pmv[5].y = EVEN(pmv[3].y);
1749 :			} else pmv[5].x = pmv[5].y = 0;
1750 :	chl	96
1751 :	edgomez	1053	if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }/* pmv[3] is left neighbour */
1752 :	edgomez	851	else pmv[3].x = pmv[3].y = 0;
1753 :	chl	96
1754 :	edgomez	1053	if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }/* [4] top neighbour */
1755 :	edgomez	851	else pmv[4].x = pmv[4].y = 0;
1756 :	Isibaar	3
1757 :	edgomez	1053	/* [1] median prediction */
1758 :	edgomez	851	pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
1759 :	edgomez	195
1760 :	edgomez	1053	pmv[0].x = pmv[0].y = 0; /* [0] is zero; not used in the loop (checked before) but needed here for make_mask */
1761 :	edgomez	195
1762 :	edgomez	1053	pmv[2].x = EVEN(prevMB->mvs[0].x); /* [2] is last frame */
1763 :	edgomez	851	pmv[2].y = EVEN(prevMB->mvs[0].y);
1764 :
1765 :			if ((x < iWcount-1) && (y < iHcount-1)) {
1766 :	edgomez	1053	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); /* [6] right-down neighbour in last frame */
1767 :	edgomez	851	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
1768 :			} else pmv[6].x = pmv[6].y = 0;
1769 :
1770 :			if (rrv) {
1771 :			int i;
1772 :			for (i = 0; i < 7; i++) {
1773 :			pmv[i].x = RRV_MV_SCALEUP(pmv[i].x);
1774 :			pmv[i].y = RRV_MV_SCALEUP(pmv[i].y);
1775 :			}
1776 :			}
1777 :	Isibaar	3	}
1778 :
1779 :	edgomez	851	static void
1780 :			SearchP(const IMAGE * const pRef,
1781 :			const uint8_t * const pRefH,
1782 :			const uint8_t * const pRefV,
1783 :			const uint8_t * const pRefHV,
1784 :			const IMAGE * const pCur,
1785 :			const int x,
1786 :			const int y,
1787 :			const uint32_t MotionFlags,
1788 :	suxen_drol	890	const uint32_t VopFlags,
1789 :	edgomez	1022	const uint32_t VolFlags,
1790 :	edgomez	851	SearchData * const Data,
1791 :			const MBParam * const pParam,
1792 :			const MACROBLOCK * const pMBs,
1793 :			const MACROBLOCK * const prevMBs,
1794 :			MACROBLOCK * const pMB)
1795 :			{
1796 :	Isibaar	3
1797 :	syskin	1132	int i, threshA;
1798 :	edgomez	851	VECTOR pmv[7];
1799 :	edgomez	1022	int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
1800 :	syskin	1132	CheckFunc * CheckCandidate;
1801 :	Isibaar	3
1802 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
1803 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
1804 :	edgomez	195
1805 :	edgomez	851	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp);
1806 :	Isibaar	3
1807 :	edgomez	1053	Data->temp[5] = Data->temp[6] = 0; /* chroma-sad cache */
1808 :	edgomez	851	i = Data->rrv ? 2 : 1;
1809 :			Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
1810 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1811 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1812 :	chl	326
1813 :	edgomez	978	Data->RefP[0] = pRef->y + (x + Data->iEdgedWidthy) 16*i;
1814 :			Data->RefP[2] = pRefH + (x + Data->iEdgedWidthy) 16*i;
1815 :			Data->RefP[1] = pRefV + (x + Data->iEdgedWidthy) 16*i;
1816 :			Data->RefP[3] = pRefHV + (x + Data->iEdgedWidthy) 16*i;
1817 :			Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1818 :			Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1819 :	edgomez	195
1820 :	edgomez	1022	Data->lambda16 = lambda_vec16[pMB->quant];
1821 :			Data->lambda8 = lambda_vec8[pMB->quant];
1822 :	edgomez	851	Data->qpel_precision = 0;
1823 :	syskin	1132	*Data->dir = 0;
1824 :	edgomez	195
1825 :	edgomez	963	memset(Data->currentMV, 0, 5*sizeof(VECTOR));
1826 :	chl	169
1827 :	edgomez	851	if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1828 :			else Data->predMV = pmv[0];
1829 :	Isibaar	3
1830 :	edgomez	851	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
1831 :			Data->iMinSAD[0] = pMB->sad16 + ((Data->lambda16 * i * pMB->sad16)>>10);
1832 :			Data->iMinSAD[1] = pMB->sad8[0] + ((Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS)) >> 10);
1833 :			Data->iMinSAD[2] = pMB->sad8[1];
1834 :			Data->iMinSAD[3] = pMB->sad8[2];
1835 :			Data->iMinSAD[4] = pMB->sad8[3];
1836 :	chl	169
1837 :	edgomez	1107	if ((!(VopFlags & XVID_VOP_MODEDECISION_RD)) && (x \| y)) {
1838 :	edgomez	1053	threshA = Data->temp[0]; /* that's where we keep this SAD atm */
1839 :	edgomez	851	if (threshA < 512) threshA = 512;
1840 :			else if (threshA > 1024) threshA = 1024;
1841 :			} else
1842 :			threshA = 512;
1843 :	chl	169
1844 :	edgomez	851	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1845 :			prevMBs + x + y * pParam->mb_width, Data->rrv);
1846 :	chl	169
1847 :	edgomez	851	if (!Data->rrv) {
1848 :	syskin	1132	if (inter4v) CheckCandidate = CheckCandidate16;
1849 :	edgomez	1053	else CheckCandidate = CheckCandidate16no4v; /* for extra speed */
1850 :	edgomez	851	} else CheckCandidate = CheckCandidate32;
1851 :	chl	169
1852 :	edgomez	851	/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/
1853 :	chl	169
1854 :	syskin	1132	for (i = 1; i < 7; i++)
1855 :			if (!vector_repeats(pmv, i)) {
1856 :			CheckCandidate(pmv[i].x, pmv[i].y, Data, i);
1857 :			if (Data->iMinSAD[0] <= threshA) { i++; break; }
1858 :			}
1859 :	chl	169
1860 :	edgomez	851	if ((Data->iMinSAD[0] <= threshA) \|\|
1861 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1862 :	edgomez	1022	(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16)))
1863 :			inter4v = 0;
1864 :	edgomez	851	else {
1865 :	Isibaar	3
1866 :	edgomez	851	MainSearchFunc * MainSearchPtr;
1867 :	syskin	1132	int mask = make_mask(pmv, i, *Data->dir); // all vectors pmv[0..i-1] have been checked
1868 :
1869 :	edgomez	949	if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch;
1870 :			else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1871 :	edgomez	851	else MainSearchPtr = DiamondSearch;
1872 :	Isibaar	3
1873 :	syskin	1132	MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate);
1874 :	Isibaar	3
1875 :	edgomez	851	/* extended search, diamond starting in 0,0 and in prediction.
1876 :			note that this search is/might be done in halfpel positions,
1877 :			which makes it more different than the diamond above */
1878 :	chl	169
1879 :	edgomez	949	if (MotionFlags & XVID_ME_EXTSEARCH16) {
1880 :	edgomez	851	int32_t bSAD;
1881 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1882 :			if (Data->rrv) {
1883 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1884 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1885 :			}
1886 :			if (!(MVequal(startMV, backupMV))) {
1887 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1888 :	edgomez	195
1889 :	syskin	1132	CheckCandidate(startMV.x, startMV.y, Data, 255);
1890 :			MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate);
1891 :	edgomez	851	if (bSAD < Data->iMinSAD[0]) {
1892 :			Data->currentMV[0] = backupMV;
1893 :			Data->iMinSAD[0] = bSAD; }
1894 :			}
1895 :	chl	169
1896 :	edgomez	851	backupMV = Data->currentMV[0];
1897 :			startMV.x = startMV.y = 1;
1898 :			if (!(MVequal(startMV, backupMV))) {
1899 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1900 :	chl	169
1901 :	syskin	1132	CheckCandidate(startMV.x, startMV.y, Data, 255);
1902 :			MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate);
1903 :	edgomez	851	if (bSAD < Data->iMinSAD[0]) {
1904 :			Data->currentMV[0] = backupMV;
1905 :	syskin	1132	Data->iMinSAD[0] = bSAD;
1906 :			}
1907 :	edgomez	195	}
1908 :	edgomez	851	}
1909 :			}
1910 :	edgomez	195
1911 :	edgomez	949	if (MotionFlags & XVID_ME_HALFPELREFINE16)
1912 :	syskin	1132	SubpelRefine(Data, CheckCandidate);
1913 :	edgomez	195
1914 :	edgomez	851	for(i = 0; i < 5; i++) {
1915 :	edgomez	1053	Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */
1916 :	edgomez	851	Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1917 :			}
1918 :	Isibaar	3
1919 :	edgomez	1023	if (Data->qpel) {
1920 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
1921 :			pParam->width, pParam->height, Data->iFcode, 2, 0);
1922 :	edgomez	987	Data->qpel_precision = 1;
1923 :	edgomez	1023	if (MotionFlags & XVID_ME_QUARTERPELREFINE16)
1924 :	Isibaar	1136	if(MotionFlags & XVID_ME_FASTREFINE16)
1925 :			SubpelRefine_Fast(Data, CheckCandidate);
1926 :			else
1927 :			SubpelRefine(Data, CheckCandidate);
1928 :	edgomez	953	}
1929 :	Isibaar	3
1930 :	edgomez	1023	if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30)
1931 :	edgomez	1022	inter4v = 0;
1932 :	Isibaar	3
1933 :	edgomez	987	if (inter4v) {
1934 :	edgomez	851	SearchData Data8;
1935 :	edgomez	1053	memcpy(&Data8, Data, sizeof(SearchData)); /* quick copy of common data */
1936 :	Isibaar	3
1937 :	edgomez	851	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1938 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1939 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1940 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1941 :	chl	181
1942 :	edgomez	1107	if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_RD))) {
1943 :	edgomez	1053	/* chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, it will not be used */
1944 :	edgomez	851	int sumx = 0, sumy = 0;
1945 :	Isibaar	3
1946 :	edgomez	1022	if (Data->qpel)
1947 :			for (i = 1; i < 5; i++) {
1948 :			sumx += Data->currentQMV[i].x/2;
1949 :			sumy += Data->currentQMV[i].y/2;
1950 :			}
1951 :			else
1952 :			for (i = 1; i < 5; i++) {
1953 :			sumx += Data->currentMV[i].x;
1954 :			sumy += Data->currentMV[i].y;
1955 :			}
1956 :	chl	259
1957 :	edgomez	851	Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf],
1958 :			(sumy >> 3) + roundtab_76[sumy & 0xf], Data);
1959 :			}
1960 :	edgomez	1022	} else Data->iMinSAD[1] = 4096*256;
1961 :	edgomez	851	}
1962 :
1963 :			static void
1964 :			Search8(const SearchData * const OldData,
1965 :			const int x, const int y,
1966 :			const uint32_t MotionFlags,
1967 :			const MBParam * const pParam,
1968 :			MACROBLOCK * const pMB,
1969 :			const MACROBLOCK * const pMBs,
1970 :			const int block,
1971 :			SearchData * const Data)
1972 :			{
1973 :			int i = 0;
1974 :	syskin	1132	CheckFunc * CheckCandidate;
1975 :	edgomez	851	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1976 :			Data->currentMV = OldData->currentMV + 1 + block;
1977 :			Data->currentQMV = OldData->currentQMV + 1 + block;
1978 :
1979 :			if(Data->qpel) {
1980 :			Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1981 :			if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1982 :			Data->predMV, Data->iFcode, 0, 0);
1983 :			} else {
1984 :			Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1985 :			if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1986 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1987 :	Isibaar	3	}
1988 :
1989 :	edgomez	851	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
1990 :	Isibaar	3
1991 :	edgomez	949	if (MotionFlags & (XVID_ME_EXTSEARCH8\|XVID_ME_HALFPELREFINE8\|XVID_ME_QUARTERPELREFINE8)) {
1992 :	Isibaar	3
1993 :	edgomez	978	if (Data->rrv) i = 16; else i = 8;
1994 :	Isibaar	3
1995 :	edgomez	978	Data->RefP[0] = OldData->RefP[0] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
1996 :			Data->RefP[1] = OldData->RefP[1] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
1997 :			Data->RefP[2] = OldData->RefP[2] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
1998 :			Data->RefP[3] = OldData->RefP[3] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
1999 :
2000 :			Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1));
2001 :	edgomez	851	Data->qpel_precision = 0;
2002 :	Isibaar	3
2003 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
2004 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
2005 :	Isibaar	3
2006 :	edgomez	851	if (!Data->rrv) CheckCandidate = CheckCandidate8;
2007 :			else CheckCandidate = CheckCandidate16no4v;
2008 :	Isibaar	3
2009 :	edgomez	1107	if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_RD))) {
2010 :	edgomez	1053	int32_t temp_sad = (Data->iMinSAD); / store current MinSAD */
2011 :	Isibaar	3
2012 :	edgomez	851	MainSearchFunc *MainSearchPtr;
2013 :	edgomez	949	if (MotionFlags & XVID_ME_USESQUARES8) MainSearchPtr = SquareSearch;
2014 :			else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
2015 :	edgomez	851	else MainSearchPtr = DiamondSearch;
2016 :	Isibaar	3
2017 :	syskin	1132	MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate);
2018 :	Isibaar	3
2019 :	edgomez	851	if(*(Data->iMinSAD) < temp_sad) {
2020 :	edgomez	1053	Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */
2021 :	edgomez	851	Data->currentQMV->y = 2 * Data->currentMV->y;
2022 :			}
2023 :			}
2024 :	edgomez	195
2025 :	edgomez	949	if (MotionFlags & XVID_ME_HALFPELREFINE8) {
2026 :	edgomez	1053	int32_t temp_sad = (Data->iMinSAD); / store current MinSAD */
2027 :	edgomez	195
2028 :	syskin	1132	SubpelRefine(Data, CheckCandidate); /* perform halfpel refine of current best vector */
2029 :	Isibaar	3
2030 :	edgomez	1053	if((Data->iMinSAD) < temp_sad) { / we have found a better match */
2031 :			Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */
2032 :	edgomez	851	Data->currentQMV->y = 2 * Data->currentMV->y;
2033 :			}
2034 :			}
2035 :	Isibaar	3
2036 :	edgomez	949	if (Data->qpel && MotionFlags & XVID_ME_QUARTERPELREFINE8) {
2037 :	edgomez	851	Data->qpel_precision = 1;
2038 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
2039 :			pParam->width, pParam->height, Data->iFcode, 2, 0);
2040 :	syskin	1132	SubpelRefine(Data, CheckCandidate);
2041 :	edgomez	851	}
2042 :			}
2043 :	edgomez	195
2044 :	edgomez	851	if (Data->rrv) {
2045 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
2046 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
2047 :			}
2048 :
2049 :			if(Data->qpel) {
2050 :			pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
2051 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
2052 :			pMB->qmvs[block] = *Data->currentQMV;
2053 :	edgomez	195	} else {
2054 :	edgomez	851	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
2055 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
2056 :	edgomez	78	}
2057 :	edgomez	851
2058 :			pMB->mvs[block] = *Data->currentMV;
2059 :			pMB->sad8[block] = 4 * *Data->iMinSAD;
2060 :	Isibaar	3	}
2061 :
2062 :	edgomez	851	/* motion estimation for B-frames */
2063 :	chl	345
2064 :	edgomez	851	static __inline VECTOR
2065 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
2066 :			{
2067 :			/* the stupidiest function ever */
2068 :			return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]);
2069 :			}
2070 :	chl	345
2071 :	edgomez	851	static void __inline
2072 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
2073 :			const uint32_t iWcount,
2074 :			const MACROBLOCK * const pMB,
2075 :			const uint32_t mode_curr)
2076 :	chl	345	{
2077 :
2078 :	edgomez	1053	/* [0] is prediction */
2079 :	edgomez	851	pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
2080 :	chl	345
2081 :	edgomez	1053	pmv[1].x = pmv[1].y = 0; /* [1] is zero */
2082 :	chl	345
2083 :	edgomez	851	pmv[2] = ChoosePred(pMB, mode_curr);
2084 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
2085 :	chl	345
2086 :	edgomez	1053	if ((y != 0)&&(x != (int)(iWcount+1))) { /* [3] top-right neighbour */
2087 :	edgomez	851	pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
2088 :			pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
2089 :			} else pmv[3].x = pmv[3].y = 0;
2090 :	chl	345
2091 :	edgomez	851	if (y != 0) {
2092 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
2093 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
2094 :			} else pmv[4].x = pmv[4].y = 0;
2095 :	chl	345
2096 :	edgomez	851	if (x != 0) {
2097 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
2098 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
2099 :			} else pmv[5].x = pmv[5].y = 0;
2100 :	chl	345
2101 :	edgomez	851	if (x != 0 && y != 0) {
2102 :			pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
2103 :			pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y);
2104 :			} else pmv[6].x = pmv[6].y = 0;
2105 :			}
2106 :	chl	345
2107 :
2108 :	edgomez	851	/* search backward or forward */
2109 :			static void
2110 :			SearchBF( const IMAGE * const pRef,
2111 :			const uint8_t * const pRefH,
2112 :			const uint8_t * const pRefV,
2113 :			const uint8_t * const pRefHV,
2114 :			const IMAGE * const pCur,
2115 :			const int x, const int y,
2116 :			const uint32_t MotionFlags,
2117 :			const uint32_t iFcode,
2118 :			const MBParam * const pParam,
2119 :			MACROBLOCK * const pMB,
2120 :			const VECTOR * const predMV,
2121 :			int32_t * const best_sad,
2122 :			const int32_t mode_current,
2123 :			SearchData * const Data)
2124 :			{
2125 :	chl	345
2126 :	syskin	1135	int i;
2127 :	edgomez	851	VECTOR pmv[7];
2128 :			MainSearchFunc *MainSearchPtr;
2129 :			*Data->iMinSAD = MV_MAX_ERROR;
2130 :			Data->iFcode = iFcode;
2131 :			Data->qpel_precision = 0;
2132 :	edgomez	1053	Data->temp[5] = Data->temp[6] = Data->temp[7] = 2564096; / reset chroma-sad cache */
2133 :	chl	345
2134 :	edgomez	978	Data->RefP[0] = pRef->y + (x + Data->iEdgedWidthy) 16;
2135 :			Data->RefP[2] = pRefH + (x + Data->iEdgedWidthy) 16;
2136 :			Data->RefP[1] = pRefV + (x + Data->iEdgedWidthy) 16;
2137 :			Data->RefP[3] = pRefHV + (x + Data->iEdgedWidthy) 16;
2138 :			Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
2139 :			Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
2140 :	chl	345
2141 :	edgomez	851	Data->predMV = *predMV;
2142 :	chl	345
2143 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
2144 :			pParam->width, pParam->height, iFcode - Data->qpel, 1, 0);
2145 :	chl	345
2146 :	edgomez	851	pmv[0] = Data->predMV;
2147 :			if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
2148 :	chl	345
2149 :	edgomez	851	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
2150 :	chl	345
2151 :	edgomez	851	Data->currentMV->x = Data->currentMV->y = 0;
2152 :	chl	345
2153 :	edgomez	1053	/* main loop. checking all predictions */
2154 :	syskin	1132	for (i = 0; i < 7; i++)
2155 :			if (!vector_repeats(pmv, i) )
2156 :			CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i);
2157 :	chl	345
2158 :	edgomez	949	if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch;
2159 :			else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
2160 :	edgomez	851	else MainSearchPtr = DiamondSearch;
2161 :	chl	345
2162 :	syskin	1135	if (*Data->iMinSAD > 512) {
2163 :			unsigned int mask = make_mask(pmv, 7, *Data->dir);
2164 :			MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v);
2165 :			}
2166 :	chl	345
2167 :	syskin	1132	SubpelRefine(Data, CheckCandidate16no4v);
2168 :
2169 :	edgomez	851	if (Data->qpel && Data->iMinSAD < best_sad + 300) {
2170 :			Data->currentQMV->x = 2*Data->currentMV->x;
2171 :			Data->currentQMV->y = 2*Data->currentMV->y;
2172 :			Data->qpel_precision = 1;
2173 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
2174 :			pParam->width, pParam->height, iFcode, 2, 0);
2175 :	syskin	1132	SubpelRefine(Data, CheckCandidate16no4v);
2176 :	edgomez	851	}
2177 :	chl	345
2178 :	edgomez	1053	/* three bits are needed to code backward mode. four for forward */
2179 :	Isibaar	3
2180 :	edgomez	851	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
2181 :			else Data->iMinSAD += 3 Data->lambda16;
2182 :	edgomez	195
2183 :	edgomez	851	if (Data->iMinSAD < best_sad) {
2184 :			best_sad = Data->iMinSAD;
2185 :			pMB->mode = mode_current;
2186 :			if (Data->qpel) {
2187 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
2188 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
2189 :			if (mode_current == MODE_FORWARD)
2190 :			pMB->qmvs[0] = *Data->currentQMV;
2191 :			else
2192 :			pMB->b_qmvs[0] = *Data->currentQMV;
2193 :			} else {
2194 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
2195 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
2196 :			}
2197 :			if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
2198 :			else pMB->b_mvs[0] = *Data->currentMV;
2199 :			}
2200 :	edgomez	195
2201 :	edgomez	851	if (mode_current == MODE_FORWARD) (Data->currentMV+2) = Data->currentMV;
2202 :	edgomez	1053	else (Data->currentMV+1) = Data->currentMV; /* we store currmv for interpolate search */
2203 :	Isibaar	3	}
2204 :
2205 :	edgomez	851	static void
2206 :			SkipDecisionB(const IMAGE * const pCur,
2207 :			const IMAGE * const f_Ref,
2208 :			const IMAGE * const b_Ref,
2209 :			MACROBLOCK * const pMB,
2210 :			const uint32_t x, const uint32_t y,
2211 :			const SearchData * const Data)
2212 :			{
2213 :			int dx = 0, dy = 0, b_dx = 0, b_dy = 0;
2214 :			int32_t sum;
2215 :			int k;
2216 :			const uint32_t stride = Data->iEdgedWidth/2;
2217 :	edgomez	1053	/* this is not full chroma compensation, only it's fullpel approximation. should work though */
2218 :	Isibaar	3
2219 :	edgomez	851	for (k = 0; k < 4; k++) {
2220 :	syskin	1135	dy += Data->directmvF[k].y >> Data->qpel;
2221 :			dx += Data->directmvF[k].x >> Data->qpel;
2222 :			b_dy += Data->directmvB[k].y >> Data->qpel;
2223 :			b_dx += Data->directmvB[k].x >> Data->qpel;
2224 :	edgomez	851	}
2225 :	Isibaar	3
2226 :	edgomez	851	dy = (dy >> 3) + roundtab_76[dy & 0xf];
2227 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
2228 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
2229 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
2230 :	Isibaar	3
2231 :	edgomez	851	sum = sad8bi(pCur->u + 8 * x + 8 * y * stride,
2232 :			f_Ref->u + (y8 + dy/2) stride + x*8 + dx/2,
2233 :			b_Ref->u + (y8 + b_dy/2) stride + x*8 + b_dx/2,
2234 :			stride);
2235 :	Isibaar	3
2236 :	syskin	1135	if (sum >= MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) return; /* no skip */
2237 :	Isibaar	3
2238 :	edgomez	851	sum += sad8bi(pCur->v + 8x + 8 y * stride,
2239 :			f_Ref->v + (y8 + dy/2) stride + x*8 + dx/2,
2240 :			b_Ref->v + (y8 + b_dy/2) stride + x*8 + b_dx/2,
2241 :			stride);
2242 :	edgomez	195
2243 :	syskin	1135	if (sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) {
2244 :	edgomez	1053	pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */
2245 :	edgomez	978	for (k = 0; k < 4; k++) {
2246 :	edgomez	963	pMB->qmvs[k] = pMB->mvs[k];
2247 :			pMB->b_qmvs[k] = pMB->b_mvs[k];
2248 :			}
2249 :			}
2250 :	edgomez	851	}
2251 :	Isibaar	3
2252 :	edgomez	851	static __inline uint32_t
2253 :			SearchDirect(const IMAGE * const f_Ref,
2254 :			const uint8_t * const f_RefH,
2255 :			const uint8_t * const f_RefV,
2256 :			const uint8_t * const f_RefHV,
2257 :			const IMAGE * const b_Ref,
2258 :			const uint8_t * const b_RefH,
2259 :			const uint8_t * const b_RefV,
2260 :			const uint8_t * const b_RefHV,
2261 :			const IMAGE * const pCur,
2262 :			const int x, const int y,
2263 :			const uint32_t MotionFlags,
2264 :			const int32_t TRB, const int32_t TRD,
2265 :			const MBParam * const pParam,
2266 :			MACROBLOCK * const pMB,
2267 :			const MACROBLOCK * const b_mb,
2268 :			int32_t * const best_sad,
2269 :			SearchData * const Data)
2270 :	Isibaar	3
2271 :	edgomez	851	{
2272 :			int32_t skip_sad;
2273 :			int k = (x + Data->iEdgedWidthy) 16;
2274 :			MainSearchFunc *MainSearchPtr;
2275 :	syskin	1132	CheckFunc * CheckCandidate;
2276 :	Isibaar	3
2277 :	edgomez	851	Data->iMinSAD = 2564096;
2278 :	edgomez	978	Data->RefP[0] = f_Ref->y + k;
2279 :			Data->RefP[2] = f_RefH + k;
2280 :			Data->RefP[1] = f_RefV + k;
2281 :			Data->RefP[3] = f_RefHV + k;
2282 :			Data->b_RefP[0] = b_Ref->y + k;
2283 :			Data->b_RefP[2] = b_RefH + k;
2284 :			Data->b_RefP[1] = b_RefV + k;
2285 :			Data->b_RefP[3] = b_RefHV + k;
2286 :			Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
2287 :			Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
2288 :			Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
2289 :			Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
2290 :	edgomez	195
2291 :	edgomez	851	k = Data->qpel ? 4 : 2;
2292 :			Data->max_dx = k * (pParam->width - x * 16);
2293 :			Data->max_dy = k * (pParam->height - y * 16);
2294 :			Data->min_dx = -k * (16 + x * 16);
2295 :			Data->min_dy = -k * (16 + y * 16);
2296 :	chl	181
2297 :	edgomez	851	Data->referencemv = Data->qpel ? b_mb->qmvs : b_mb->mvs;
2298 :			Data->qpel_precision = 0;
2299 :	edgomez	170
2300 :	edgomez	851	for (k = 0; k < 4; k++) {
2301 :			pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
2302 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
2303 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
2304 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
2305 :	Isibaar	3
2306 :	edgomez	851	if ( (pMB->b_mvs[k].x > Data->max_dx) \| (pMB->b_mvs[k].x < Data->min_dx)
2307 :			\| (pMB->b_mvs[k].y > Data->max_dy) \| (pMB->b_mvs[k].y < Data->min_dy) ) {
2308 :
2309 :	edgomez	1053	best_sad = 2564096; /* in that case, we won't use direct mode */
2310 :			pMB->mode = MODE_DIRECT; /* just to make sure it doesn't say "MODE_DIRECT_NONE_MV" */
2311 :	edgomez	851	pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
2312 :			return 256*4096;
2313 :			}
2314 :			if (b_mb->mode != MODE_INTER4V) {
2315 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
2316 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
2317 :			Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
2318 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
2319 :			break;
2320 :			}
2321 :	edgomez	195	}
2322 :	Isibaar	3
2323 :	edgomez	851	CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v;
2324 :	Isibaar	3
2325 :	syskin	1132	CheckCandidate(0, 0, Data, 255);
2326 :	edgomez	195
2327 :	edgomez	1053	/* initial (fast) skip decision */
2328 :	syskin	1135	if (Data->iMinSAD < Data->iQuant INITIAL_SKIP_THRESH * (Data->chroma?3:2)) {
2329 :	edgomez	1053	/* possible skip */
2330 :	edgomez	851	if (Data->chroma) {
2331 :			pMB->mode = MODE_DIRECT_NONE_MV;
2332 :	edgomez	1053	return Data->iMinSAD; / skip. */
2333 :	edgomez	851	} else {
2334 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data);
2335 :	edgomez	1053	if (pMB->mode == MODE_DIRECT_NONE_MV) return Data->iMinSAD; / skip. */
2336 :	edgomez	851	}
2337 :	Isibaar	3	}
2338 :
2339 :	edgomez	953	*Data->iMinSAD += Data->lambda16;
2340 :	edgomez	851	skip_sad = *Data->iMinSAD;
2341 :	syskin	1084
2342 :	edgomez	1053	/*
2343 :			* DIRECT MODE DELTA VECTOR SEARCH.
2344 :			* This has to be made more effective, but at the moment I'm happy it's running at all
2345 :			*/
2346 :	edgomez	195
2347 :	edgomez	949	if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch;
2348 :			else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
2349 :	edgomez	851	else MainSearchPtr = DiamondSearch;
2350 :	Isibaar	3
2351 :	syskin	1132	MainSearchPtr(0, 0, Data, 255, CheckCandidate);
2352 :	Isibaar	3
2353 :	syskin	1132	SubpelRefine(Data, CheckCandidate);
2354 :	chl	181
2355 :	edgomez	851	best_sad = Data->iMinSAD;
2356 :	chl	181
2357 :	edgomez	851	if (Data->qpel \|\| b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT;
2358 :	edgomez	1053	else pMB->mode = MODE_DIRECT_NO4V; /* for faster compensation */
2359 :	chl	181
2360 :	edgomez	851	pMB->pmvs[3] = *Data->currentMV;
2361 :	edgomez	195
2362 :	edgomez	851	for (k = 0; k < 4; k++) {
2363 :			pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
2364 :			pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
2365 :			? Data->directmvB[k].x
2366 :			:pMB->mvs[k].x - Data->referencemv[k].x);
2367 :			pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
2368 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
2369 :			? Data->directmvB[k].y
2370 :			: pMB->mvs[k].y - Data->referencemv[k].y);
2371 :			if (Data->qpel) {
2372 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
2373 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
2374 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
2375 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
2376 :			}
2377 :	edgomez	195
2378 :	edgomez	851	if (b_mb->mode != MODE_INTER4V) {
2379 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
2380 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
2381 :			pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
2382 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
2383 :			break;
2384 :			}
2385 :	edgomez	78	}
2386 :	edgomez	851	return skip_sad;
2387 :			}
2388 :	Isibaar	3
2389 :	edgomez	851	static void
2390 :			SearchInterpolate(const IMAGE * const f_Ref,
2391 :			const uint8_t * const f_RefH,
2392 :			const uint8_t * const f_RefV,
2393 :			const uint8_t * const f_RefHV,
2394 :			const IMAGE * const b_Ref,
2395 :			const uint8_t * const b_RefH,
2396 :			const uint8_t * const b_RefV,
2397 :			const uint8_t * const b_RefHV,
2398 :			const IMAGE * const pCur,
2399 :			const int x, const int y,
2400 :			const uint32_t fcode,
2401 :			const uint32_t bcode,
2402 :			const uint32_t MotionFlags,
2403 :			const MBParam * const pParam,
2404 :			const VECTOR * const f_predMV,
2405 :			const VECTOR * const b_predMV,
2406 :			MACROBLOCK * const pMB,
2407 :			int32_t * const best_sad,
2408 :			SearchData * const fData)
2409 :	chl	96
2410 :	edgomez	851	{
2411 :	chl	169
2412 :	syskin	1132	int i, j;
2413 :	edgomez	851	SearchData bData;
2414 :	chl	169
2415 :	edgomez	851	fData->qpel_precision = 0;
2416 :	edgomez	1053	memcpy(&bData, fData, sizeof(SearchData)); /* quick copy of common data */
2417 :	edgomez	851	fData->iMinSAD = 4096256;
2418 :			bData.currentMV++; bData.currentQMV++;
2419 :			fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
2420 :	chl	169
2421 :	edgomez	851	i = (x + y * fData->iEdgedWidth) * 16;
2422 :	chl	169
2423 :	edgomez	978	bData.b_RefP[0] = fData->RefP[0] = f_Ref->y + i;
2424 :			bData.b_RefP[2] = fData->RefP[2] = f_RefH + i;
2425 :			bData.b_RefP[1] = fData->RefP[1] = f_RefV + i;
2426 :			bData.b_RefP[3] = fData->RefP[3] = f_RefHV + i;
2427 :			bData.RefP[0] = fData->b_RefP[0] = b_Ref->y + i;
2428 :			bData.RefP[2] = fData->b_RefP[2] = b_RefH + i;
2429 :			bData.RefP[1] = fData->b_RefP[1] = b_RefV + i;
2430 :			bData.RefP[3] = fData->b_RefP[3] = b_RefHV + i;
2431 :			bData.b_RefP[4] = fData->RefP[4] = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
2432 :			bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
2433 :			bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
2434 :			bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
2435 :	syskin	1132	bData.dir = fData->dir;
2436 :	chl	169
2437 :	edgomez	851	bData.bpredMV = fData->predMV = *f_predMV;
2438 :			fData->bpredMV = bData.predMV = *b_predMV;
2439 :			fData->currentMV[0] = fData->currentMV[2];
2440 :	Isibaar	3
2441 :	syskin	1129	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode - fData->qpel, 1, 0);
2442 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode - fData->qpel, 1, 0);
2443 :	Isibaar	3
2444 :	edgomez	851	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
2445 :			if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
2446 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
2447 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
2448 :	chl	169
2449 :	edgomez	851	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
2450 :			if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
2451 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
2452 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
2453 :	Isibaar	3
2454 :	syskin	1132	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, fData, 255);
2455 :	chl	169
2456 :	edgomez	1053	/* diamond */
2457 :	edgomez	851	do {
2458 :	syskin	1132	*fData->dir = 255;
2459 :	edgomez	1053	/* forward MV moves */
2460 :	edgomez	851	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
2461 :	chl	169
2462 :	syskin	1132	CheckCandidateInt(i + 1, j, fData, 0);
2463 :			CheckCandidateInt(i, j + 1, fData, 0);
2464 :			CheckCandidateInt(i - 1, j, fData, 0);
2465 :			CheckCandidateInt(i, j - 1, fData, 0);
2466 :	edgomez	195
2467 :	edgomez	1053	/* backward MV moves */
2468 :	edgomez	851	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
2469 :			fData->currentMV[2] = fData->currentMV[0];
2470 :	syskin	1132	CheckCandidateInt(i + 1, j, &bData, 0);
2471 :			CheckCandidateInt(i, j + 1, &bData, 0);
2472 :			CheckCandidateInt(i - 1, j, &bData, 0);
2473 :			CheckCandidateInt(i, j - 1, &bData, 0);
2474 :	Isibaar	3
2475 :	syskin	1132	} while (!(*fData->dir));
2476 :	chl	140
2477 :	edgomez	1053	/* qpel refinement */
2478 :	edgomez	851	if (fData->qpel) {
2479 :			if (fData->iMinSAD > best_sad + 500) return;
2480 :			fData->qpel_precision = bData.qpel_precision = 1;
2481 :	syskin	1129	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0);
2482 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0);
2483 :	edgomez	851	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
2484 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
2485 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
2486 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
2487 :	syskin	1132	SubpelRefine(fData, CheckCandidateInt);
2488 :	edgomez	851	if (fData->iMinSAD > best_sad + 300) return;
2489 :			fData->currentQMV[2] = fData->currentQMV[0];
2490 :	syskin	1132	SubpelRefine(&bData, CheckCandidateInt);
2491 :	edgomez	851	}
2492 :	chl	140
2493 :	edgomez	1053	fData->iMinSAD += (2+3) fData->lambda16; /* two bits are needed to code interpolate mode. */
2494 :	Isibaar	3
2495 :	edgomez	851	if (fData->iMinSAD < best_sad) {
2496 :			best_sad = fData->iMinSAD;
2497 :			pMB->mvs[0] = fData->currentMV[0];
2498 :			pMB->b_mvs[0] = fData->currentMV[1];
2499 :			pMB->mode = MODE_INTERPOLATE;
2500 :			if (fData->qpel) {
2501 :			pMB->qmvs[0] = fData->currentQMV[0];
2502 :			pMB->b_qmvs[0] = fData->currentQMV[1];
2503 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
2504 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
2505 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
2506 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
2507 :			} else {
2508 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
2509 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
2510 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
2511 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
2512 :			}
2513 :	edgomez	78	}
2514 :	edgomez	851	}
2515 :	Isibaar	3
2516 :	edgomez	851	void
2517 :			MotionEstimationBVOP(MBParam * const pParam,
2518 :			FRAMEINFO * const frame,
2519 :			const int32_t time_bp,
2520 :			const int32_t time_pp,
2521 :	edgomez	1053	/* forward (past) reference */
2522 :	edgomez	851	const MACROBLOCK * const f_mbs,
2523 :			const IMAGE * const f_ref,
2524 :			const IMAGE * const f_refH,
2525 :			const IMAGE * const f_refV,
2526 :			const IMAGE * const f_refHV,
2527 :	edgomez	1053	/* backward (future) reference */
2528 :	edgomez	851	const FRAMEINFO * const b_reference,
2529 :			const IMAGE * const b_ref,
2530 :			const IMAGE * const b_refH,
2531 :			const IMAGE * const b_refV,
2532 :			const IMAGE * const b_refHV)
2533 :			{
2534 :			uint32_t i, j;
2535 :			int32_t best_sad;
2536 :			uint32_t skip_sad;
2537 :			int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
2538 :			const MACROBLOCK * const b_mbs = b_reference->mbs;
2539 :	Isibaar	3
2540 :	edgomez	851	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
2541 :	Isibaar	3
2542 :	edgomez	851	const int32_t TRB = time_pp - time_bp;
2543 :			const int32_t TRD = time_pp;
2544 :	edgomez	195
2545 :	edgomez	1053	/* some pre-inintialized data for the rest of the search */
2546 :	Isibaar	3
2547 :	edgomez	851	SearchData Data;
2548 :			int32_t iMinSAD;
2549 :	syskin	1132	uint32_t dir;
2550 :	edgomez	851	VECTOR currentMV[3];
2551 :			VECTOR currentQMV[3];
2552 :			int32_t temp[8];
2553 :			memset(&Data, 0, sizeof(SearchData));
2554 :			Data.iEdgedWidth = pParam->edged_width;
2555 :			Data.currentMV = currentMV; Data.currentQMV = currentQMV;
2556 :			Data.iMinSAD = &iMinSAD;
2557 :	syskin	1135	Data.lambda16 = lambda_vec16[MAX(frame->quant-2, 2)];
2558 :	syskin	1112	Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0;
2559 :	edgomez	851	Data.rounding = 0;
2560 :	edgomez	1107	Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP;
2561 :	edgomez	851	Data.temp = temp;
2562 :	syskin	1132	Data.dir = &dir;
2563 :	syskin	1135	Data.iQuant = frame->quant;
2564 :	Isibaar	3
2565 :	edgomez	1053	Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */
2566 :
2567 :			/* note: i==horizontal, j==vertical */
2568 :	edgomez	851	for (j = 0; j < pParam->mb_height; j++) {
2569 :	edgomez	195
2570 :	edgomez	851	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
2571 :	Isibaar	3
2572 :	edgomez	851	for (i = 0; i < pParam->mb_width; i++) {
2573 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
2574 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
2575 :	edgomez	195
2576 :	edgomez	851	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
2577 :			if (b_reference->coding_type != S_VOP)
2578 :			if (b_mb->mode == MODE_NOT_CODED) {
2579 :			pMB->mode = MODE_NOT_CODED;
2580 :			continue;
2581 :			}
2582 :	Isibaar	3
2583 :	edgomez	851	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
2584 :			Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8;
2585 :			Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8;
2586 :	Isibaar	3
2587 :	edgomez	851	/* direct search comes first, because it (1) checks for SKIP-mode
2588 :			and (2) sets very good predictions for forward and backward search */
2589 :			skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
2590 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
2591 :			&frame->image,
2592 :			i, j,
2593 :			frame->motion_flags,
2594 :			TRB, TRD,
2595 :			pParam,
2596 :			pMB, b_mb,
2597 :			&best_sad,
2598 :			&Data);
2599 :	Isibaar	3
2600 :	edgomez	851	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
2601 :	edgomez	195
2602 :	edgomez	1053	/* forward search */
2603 :	edgomez	851	SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y,
2604 :			&frame->image, i, j,
2605 :			frame->motion_flags,
2606 :			frame->fcode, pParam,
2607 :			pMB, &f_predMV, &best_sad,
2608 :			MODE_FORWARD, &Data);
2609 :	edgomez	195
2610 :	edgomez	1053	/* backward search */
2611 :	edgomez	851	SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y,
2612 :			&frame->image, i, j,
2613 :			frame->motion_flags,
2614 :			frame->bcode, pParam,
2615 :			pMB, &b_predMV, &best_sad,
2616 :			MODE_BACKWARD, &Data);
2617 :
2618 :	edgomez	1053	/* interpolate search comes last, because it uses data from forward and backward as prediction */
2619 :	edgomez	851	SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y,
2620 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
2621 :			&frame->image,
2622 :			i, j,
2623 :			frame->fcode, frame->bcode,
2624 :			frame->motion_flags,
2625 :			pParam,
2626 :			&f_predMV, &b_predMV,
2627 :			pMB, &best_sad,
2628 :			&Data);
2629 :
2630 :	edgomez	1053	/* final skip decision */
2631 :	syskin	1135	if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * 2)
2632 :	edgomez	851	&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
2633 :			SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data);
2634 :
2635 :			switch (pMB->mode) {
2636 :			case MODE_FORWARD:
2637 :			f_count++;
2638 :			f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
2639 :			break;
2640 :			case MODE_BACKWARD:
2641 :			b_count++;
2642 :			b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
2643 :			break;
2644 :			case MODE_INTERPOLATE:
2645 :			i_count++;
2646 :			f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
2647 :			b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
2648 :			break;
2649 :			case MODE_DIRECT:
2650 :			case MODE_DIRECT_NO4V:
2651 :			d_count++;
2652 :			default:
2653 :			break;
2654 :			}
2655 :			}
2656 :			}
2657 :	Isibaar	3	}
2658 :	chl	96
2659 :	edgomez	851	static __inline void
2660 :			MEanalyzeMB ( const uint8_t * const pRef,
2661 :			const uint8_t * const pCur,
2662 :			const int x,
2663 :			const int y,
2664 :			const MBParam * const pParam,
2665 :			MACROBLOCK * const pMBs,
2666 :			SearchData * const Data)
2667 :	chl	96	{
2668 :
2669 :	syskin	1132	int i;
2670 :	edgomez	851	VECTOR pmv[3];
2671 :	edgomez	953	MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width];
2672 :	chl	96
2673 :	syskin	1135	unsigned int simplicity = 0;
2674 :
2675 :	edgomez	851	for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
2676 :	chl	96
2677 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
2678 :	syskin	1135	pParam->width, pParam->height, Data->iFcode - Data->qpel - 1, 0, 0);
2679 :	edgomez	195
2680 :	edgomez	851	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
2681 :	edgomez	978	Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16;
2682 :	chl	96
2683 :	syskin	1135	pmv[0].x = pMB->mvs[0].x;
2684 :			pmv[0].y = pMB->mvs[0].y;
2685 :	edgomez	195
2686 :	syskin	1135	CheckCandidate32I(pmv[0].x, pmv[0].y, Data, 0);
2687 :	edgomez	195
2688 :	syskin	1135	if (*Data->iMinSAD > 200) {
2689 :	chl	96
2690 :	syskin	1135	pmv[1].x = pmv[1].y = 0;
2691 :
2692 :			/* median is only used as prediction. it doesn't have to be real */
2693 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
2694 :			else
2695 :			if (x == 1) /* left macroblock does not have any vector now */
2696 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; /* top instead of median */
2697 :			else if (y == 1) /* top macroblock doesn't have it's vector */
2698 :			Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */
2699 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); /* else median */
2700 :
2701 :			pmv[2].x = Data->predMV.x;
2702 :			pmv[2].y = Data->predMV.y;
2703 :
2704 :	syskin	1133	if (!vector_repeats(pmv, 1))
2705 :	syskin	1132	CheckCandidate32I(pmv[1].x, pmv[1].y, Data, 1);
2706 :	syskin	1133	if (!vector_repeats(pmv, 2))
2707 :	syskin	1132	CheckCandidate32I(pmv[2].x, pmv[2].y, Data, 2);
2708 :	chl	96
2709 :	syskin	1135	if (Data->iMinSAD > 500) { / diamond only if needed */
2710 :	syskin	1133	unsigned int mask = make_mask(pmv, 3, *Data->dir);
2711 :	syskin	1134	DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate32I);
2712 :	syskin	1135	} else simplicity++;
2713 :	edgomez	851
2714 :	syskin	1135	if (Data->iMinSAD > 500) / refinement from 2-pixel to 1-pixel */
2715 :			SubpelRefine(Data, CheckCandidate32I);
2716 :			else simplicity++;
2717 :			} else simplicity++;
2718 :
2719 :	edgomez	953	for (i = 0; i < 4; i++) {
2720 :			MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width];
2721 :			MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
2722 :			MB->mode = MODE_INTER;
2723 :	syskin	1135	/* if we skipped some search steps, we have to assume that SAD would be lower with them */
2724 :			MB->sad16 = Data->iMinSAD[i+1] - (simplicity<<7);
2725 :	chl	96	}
2726 :	edgomez	851	}
2727 :	chl	96
2728 :	syskin	1075	#define INTRA_THRESH 2200
2729 :	syskin	1135	#define INTER_THRESH 40
2730 :	syskin	1075	#define INTRA_THRESH2 95
2731 :	chl	96
2732 :	edgomez	851	int
2733 :			MEanalysis( const IMAGE * const pRef,
2734 :	edgomez	953	const FRAMEINFO * const Current,
2735 :			const MBParam * const pParam,
2736 :	syskin	1075	const int maxIntra, //maximum number if non-I frames
2737 :			const int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
2738 :	syskin	1084	const int bCount, // number of B frames in a row
2739 :	edgomez	953	const int b_thresh)
2740 :	edgomez	851	{
2741 :			uint32_t x, y, intra = 0;
2742 :			int sSAD = 0;
2743 :			MACROBLOCK * const pMBs = Current->mbs;
2744 :			const IMAGE * const pCurrent = &Current->image;
2745 :	syskin	1075	int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + b_thresh;
2746 :	chl	1077	int blocks = 0;
2747 :	syskin	1075	int complexity = 0;
2748 :	chl	96
2749 :	edgomez	851	int32_t iMinSAD[5], temp[5];
2750 :	syskin	1132	uint32_t dir;
2751 :	edgomez	851	VECTOR currentMV[5];
2752 :			SearchData Data;
2753 :			Data.iEdgedWidth = pParam->edged_width;
2754 :			Data.currentMV = currentMV;
2755 :			Data.iMinSAD = iMinSAD;
2756 :			Data.iFcode = Current->fcode;
2757 :			Data.temp = temp;
2758 :	syskin	1132	Data.dir = &dir;
2759 :			Data.qpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0;
2760 :	syskin	1135	Data.qpel_precision = 0;
2761 :	chl	96
2762 :	edgomez	1023	if (intraCount != 0) {
2763 :	syskin	1075	if (intraCount < 10) // we're right after an I frame
2764 :	edgomez	1023	IntraThresh += 15* (intraCount - 10) * (intraCount - 10);
2765 :			else
2766 :	syskin	1075	if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
2767 :	edgomez	1023	IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra;
2768 :			}
2769 :	chl	96
2770 :	syskin	1135	InterThresh -= 20 * bCount;
2771 :			if (InterThresh < 10 + b_thresh) InterThresh = 10 + b_thresh;
2772 :	chl	326
2773 :	edgomez	851	if (sadInit) (*sadInit) ();
2774 :	chl	96
2775 :	edgomez	851	for (y = 1; y < pParam->mb_height-1; y += 2) {
2776 :			for (x = 1; x < pParam->mb_width-1; x += 2) {
2777 :			int i;
2778 :	syskin	1075	blocks += 10;
2779 :	edgomez	195
2780 :	edgomez	851	if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
2781 :	syskin	1075	else { //extrapolation of the vector found for last frame
2782 :	edgomez	978	pMBs[x + y * pParam->mb_width].mvs[0].x =
2783 :	edgomez	953	(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount;
2784 :	edgomez	978	pMBs[x + y * pParam->mb_width].mvs[0].y =
2785 :	edgomez	953	(pMBs[x + y * pParam->mb_width].mvs[0].y * (bCount+1) ) / bCount;
2786 :			}
2787 :	edgomez	195
2788 :	edgomez	851	MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
2789 :	edgomez	195
2790 :	edgomez	851	for (i = 0; i < 4; i++) {
2791 :			int dev;
2792 :			MACROBLOCK pMB = &pMBs[x+(i&1) + (y+(i>>1)) pParam->mb_width];
2793 :	syskin	1075	dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16,
2794 :			pParam->edged_width);
2795 :	syskin	1084
2796 :	syskin	1108	complexity += MAX(dev, 300);
2797 :	syskin	1075	if (dev + IntraThresh < pMB->sad16) {
2798 :			pMB->mode = MODE_INTRA;
2799 :			if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP;
2800 :	edgomez	851	}
2801 :	edgomez	953
2802 :	syskin	1084	if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0)
2803 :	syskin	1135	if (dev > 1000 && pMB->sad16 < 1000)
2804 :	syskin	1075	sSAD += 1000;
2805 :
2806 :	syskin	1135	sSAD += (dev < 4000) ? pMB->sad16 : pMB->sad16/2; /* blocks with big contrast differences usually have large SAD - while they look very good in b-frames */
2807 :	edgomez	851	}
2808 :			}
2809 :	edgomez	195	}
2810 :	syskin	1075	complexity >>= 7;
2811 :	edgomez	953
2812 :	syskin	1075	sSAD /= complexity + 4*blocks;
2813 :	edgomez	953
2814 :	syskin	1108	if (intraCount > 80 && sSAD > INTRA_THRESH2 ) return I_VOP;
2815 :	edgomez	851	if (sSAD > InterThresh ) return P_VOP;
2816 :			emms();
2817 :			return B_VOP;
2818 :			}
2819 :	chl	96
2820 :
2821 :	edgomez	1053	/* functions which perform BITS-based search/bitcount */
2822 :	chl	96
2823 :	edgomez	851	static int
2824 :	syskin	1108	findRDinter(SearchData * const Data,
2825 :			const MACROBLOCK * const pMBs, const int x, const int y,
2826 :			const MBParam * const pParam,
2827 :			const uint32_t MotionFlags)
2828 :	chl	96	{
2829 :	syskin	1132	int i;
2830 :	edgomez	851	int32_t bsad[5];
2831 :	chl	141
2832 :	edgomez	851	if (Data->qpel) {
2833 :			for(i = 0; i < 5; i++) {
2834 :			Data->currentMV[i].x = Data->currentQMV[i].x/2;
2835 :			Data->currentMV[i].y = Data->currentQMV[i].y/2;
2836 :			}
2837 :			Data->qpel_precision = 1;
2838 :	syskin	1132	CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255);
2839 :	chl	96
2840 :	edgomez	1107	if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD \| XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */
2841 :	edgomez	851	for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i];
2842 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
2843 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
2844 :	edgomez	851	Data->qpel_precision = 0;
2845 :			if (Data->currentQMV->x & 1 \|\| Data->currentQMV->y & 1)
2846 :	syskin	1132	CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
2847 :	edgomez	851	}
2848 :	edgomez	195
2849 :	edgomez	1053	} else { /* not qpel */
2850 :	edgomez	195
2851 :	syskin	1132	CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
2852 :	edgomez	851	}
2853 :	chl	96
2854 :	syskin	1132	if (MotionFlags&XVID_ME_EXTSEARCH_RD)
2855 :			SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16);
2856 :	chl	96
2857 :	syskin	1132	if (MotionFlags&XVID_ME_HALFPELREFINE16_RD)
2858 :			SubpelRefine(Data, CheckCandidateRD16);
2859 :	chl	96
2860 :	edgomez	851	if (Data->qpel) {
2861 :	edgomez	1107	if (MotionFlags&(XVID_ME_EXTSEARCH_RD \| XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */
2862 :	edgomez	851	for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) {
2863 :	edgomez	1053	Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */
2864 :	edgomez	851	Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
2865 :			}
2866 :	edgomez	195
2867 :	edgomez	1053	/* preparing for qpel-precision search */
2868 :	edgomez	851	Data->qpel_precision = 1;
2869 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
2870 :			pParam->width, pParam->height, Data->iFcode, 2, 0);
2871 :	edgomez	851	}
2872 :	syskin	1132	if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD)
2873 :			SubpelRefine(Data, CheckCandidateRD16);
2874 :	edgomez	851	}
2875 :	chl	96
2876 :	edgomez	1107	if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */
2877 :	edgomez	851	VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV;
2878 :			if (!(Data->predMV.x == v->x && Data->predMV.y == v->y))
2879 :	syskin	1132	CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255);
2880 :	edgomez	851	}
2881 :			return Data->iMinSAD[0];
2882 :			}
2883 :	chl	96
2884 :	edgomez	851	static int
2885 :	syskin	1108	findRDinter4v(const SearchData * const Data,
2886 :			MACROBLOCK * const pMB, const MACROBLOCK * const pMBs,
2887 :			const int x, const int y,
2888 :			const MBParam * const pParam, const uint32_t MotionFlags,
2889 :			const VECTOR * const backup)
2890 :	edgomez	851	{
2891 :	chl	96
2892 :	syskin	1132	int cbp = 0, bits = 0, t = 0, i;
2893 :	edgomez	851	SearchData Data2, *Data8 = &Data2;
2894 :			int sumx = 0, sumy = 0;
2895 :	edgomez	962	int16_t in = Data->dctSpace, coeff = Data->dctSpace + 64;
2896 :	edgomez	1022	uint8_t * ptr;
2897 :	chl	96
2898 :	edgomez	851	memcpy(Data8, Data, sizeof(SearchData));
2899 :	syskin	1132
2900 :	edgomez	1053	for (i = 0; i < 4; i++) { /* for all luma blocks */
2901 :	edgomez	1022
2902 :	edgomez	851	Data8->iMinSAD = Data->iMinSAD + i + 1;
2903 :			Data8->currentMV = Data->currentMV + i + 1;
2904 :			Data8->currentQMV = Data->currentQMV + i + 1;
2905 :			Data8->Cur = Data->Cur + 8((i&1) + (i>>1)Data->iEdgedWidth);
2906 :	edgomez	978	Data8->RefP[0] = Data->RefP[0] + 8((i&1) + (i>>1)Data->iEdgedWidth);
2907 :			Data8->RefP[2] = Data->RefP[2] + 8((i&1) + (i>>1)Data->iEdgedWidth);
2908 :			Data8->RefP[1] = Data->RefP[1] + 8((i&1) + (i>>1)Data->iEdgedWidth);
2909 :			Data8->RefP[3] = Data->RefP[3] + 8((i&1) + (i>>1)Data->iEdgedWidth);
2910 :	syskin	1084	*Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; // copy corresponding cbp bit
2911 :	edgomez	195
2912 :	edgomez	851	if(Data->qpel) {
2913 :			Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i);
2914 :			if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y,
2915 :			Data8->predMV, Data8->iFcode, 0, 0);
2916 :			} else {
2917 :			Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i);
2918 :			if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y,
2919 :			Data8->predMV, Data8->iFcode, 0, 0);
2920 :			}
2921 :	chl	96
2922 :	syskin	1129	get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2x + (i&1), 2y + (i>>1), 3,
2923 :			pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1, 0);
2924 :	chl	96
2925 :	edgomez	1023	Data8->iMinSAD += BITS_MULTt;
2926 :	chl	96
2927 :	edgomez	851	Data8->qpel_precision = Data8->qpel;
2928 :	edgomez	1053	/* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */
2929 :	edgomez	987	{
2930 :			VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV;
2931 :	edgomez	1022	if (!MVequal (*v, backup[i+1]) )
2932 :	syskin	1132	CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255);
2933 :	edgomez	851	}
2934 :	edgomez	195
2935 :	edgomez	851	if (Data8->qpel) {
2936 :	edgomez	1107	if (MotionFlags&XVID_ME_HALFPELREFINE8_RD \|\| (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */
2937 :	edgomez	851	int32_t s = *Data8->iMinSAD;
2938 :			Data8->currentMV->x = Data8->currentQMV->x/2;
2939 :			Data8->currentMV->y = Data8->currentQMV->y/2;
2940 :			Data8->qpel_precision = 0;
2941 :	syskin	1129	get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2x + (i&1), 2y + (i>>1), 3,
2942 :			pParam->width, pParam->height, Data8->iFcode - 1, 1, 0);
2943 :	edgomez	195
2944 :	edgomez	851	if (Data8->currentQMV->x & 1 \|\| Data8->currentQMV->y & 1)
2945 :	syskin	1132	CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255);
2946 :	edgomez	195
2947 :	edgomez	1107	if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD)
2948 :	syskin	1132	SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
2949 :	chl	96
2950 :	edgomez	1107	if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
2951 :	syskin	1132	SubpelRefine(Data8, CheckCandidateRD8);
2952 :	chl	96
2953 :	edgomez	1053	if(s > Data8->iMinSAD) { / we have found a better match */
2954 :	edgomez	851	Data8->currentQMV->x = 2*Data8->currentMV->x;
2955 :			Data8->currentQMV->y = 2*Data8->currentMV->y;
2956 :			}
2957 :	chl	96
2958 :	edgomez	851	Data8->qpel_precision = 1;
2959 :	syskin	1129	get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2x + (i&1), 2y + (i>>1), 3,
2960 :			pParam->width, pParam->height, Data8->iFcode, 2, 0);
2961 :	chl	96
2962 :	edgomez	851	}
2963 :	syskin	1132	if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD)
2964 :			SubpelRefine(Data8, CheckCandidateRD8);
2965 :	edgomez	195
2966 :	edgomez	1053	} else { /* not qpel */
2967 :	edgomez	851
2968 :	edgomez	1107	if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */
2969 :	syskin	1132	SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
2970 :	edgomez	1022
2971 :	edgomez	1107	if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
2972 :	syskin	1132	SubpelRefine(Data8, CheckCandidateRD8); /* halfpel refinement */
2973 :	edgomez	1022	}
2974 :
2975 :	edgomez	1053	/* checking vector equal to predicion */
2976 :	edgomez	1107	if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) {
2977 :	edgomez	851	const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV;
2978 :	edgomez	1022	if (!MVequal(*v, Data8->predMV))
2979 :	syskin	1132	CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255);
2980 :	chl	141	}
2981 :	edgomez	195
2982 :	edgomez	851	bits += *Data8->iMinSAD;
2983 :	edgomez	1053	if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */
2984 :	edgomez	851
2985 :	edgomez	1053	/* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */
2986 :	edgomez	851	if(Data->qpel) {
2987 :			pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x;
2988 :			pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y;
2989 :			pMB->qmvs[i] = *Data8->currentQMV;
2990 :			sumx += Data8->currentQMV->x/2;
2991 :			sumy += Data8->currentQMV->y/2;
2992 :			} else {
2993 :			pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x;
2994 :			pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y;
2995 :			sumx += Data8->currentMV->x;
2996 :			sumy += Data8->currentMV->y;
2997 :	chl	141	}
2998 :	edgomez	851	pMB->mvs[i] = *Data8->currentMV;
2999 :			pMB->sad8[i] = 4 * *Data8->iMinSAD;
3000 :	syskin	1084	if (Data8->cbp[0]) cbp \|= 1 << (5 - i);
3001 :	chl	141
3002 :	syskin	1084	} /* end - for all luma blocks */
3003 :	chl	141
3004 :	edgomez	1023	bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
3005 :	chl	141
3006 :	edgomez	1053	/* let's check chroma */
3007 :	edgomez	1022	sumx = (sumx >> 3) + roundtab_76[sumx & 0xf];
3008 :			sumy = (sumy >> 3) + roundtab_76[sumy & 0xf];
3009 :	chl	96
3010 :	edgomez	1053	/* chroma U */
3011 :	edgomez	1022	ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
3012 :			transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2);
3013 :	syskin	1075	bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4);
3014 :	edgomez	1022
3015 :			if (bits >= *Data->iMinSAD) return bits;
3016 :
3017 :	edgomez	1053	/* chroma V */
3018 :	edgomez	1022	ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
3019 :			transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2);
3020 :	syskin	1075	bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5);
3021 :	edgomez	1022
3022 :	edgomez	1023	bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) \| ((cbp & 3) << 3)].len;
3023 :	edgomez	1022
3024 :	syskin	1084	*Data->cbp = cbp;
3025 :	edgomez	851	return bits;
3026 :			}
3027 :	chl	96
3028 :	edgomez	851	static int
3029 :	syskin	1108	findRDintra(const SearchData * const Data)
3030 :	edgomez	851	{
3031 :	edgomez	1053	int bits = BITS_MULT1; / this one is ac/dc prediction flag bit */
3032 :	edgomez	1023	int cbp = 0, i, dc = 0;
3033 :	edgomez	962	int16_t in = Data->dctSpace, coeff = Data->dctSpace + 64;
3034 :	edgomez	195
3035 :	edgomez	851	for(i = 0; i < 4; i++) {
3036 :			int s = 8((i&1) + (i>>1)Data->iEdgedWidth);
3037 :			transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth);
3038 :	syskin	1075	bits += Block_CalcBitsIntra(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, &dc);
3039 :	chl	141
3040 :	edgomez	1022	if (bits >= Data->iMinSAD[0]) return bits;
3041 :	edgomez	851	}
3042 :	chl	181
3043 :	edgomez	1023	bits += BITS_MULT*xvid_cbpy_tab[cbp>>2].len;
3044 :	chl	96
3045 :	edgomez	1053	/chroma U /
3046 :	edgomez	1022	transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2);
3047 :	syskin	1075	bits += Block_CalcBitsIntra(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, &dc);
3048 :	syskin	1084
3049 :	edgomez	1022	if (bits >= Data->iMinSAD[0]) return bits;
3050 :
3051 :	edgomez	1053	/* chroma V */
3052 :	edgomez	1022	transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2);
3053 :	syskin	1075	bits += Block_CalcBitsIntra(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, &dc);
3054 :	edgomez	1022
3055 :	edgomez	1023	bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) \| ((cbp & 3) << 3)].len;
3056 :	edgomez	1022
3057 :	edgomez	851	return bits;
3058 :	chl	96	}
3059 :	chl	1077
3060 :	syskin	1084	static int
3061 :	syskin	1108	findRDgmc(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y)
3062 :	syskin	1084	{
3063 :			int bits = BITS_MULT1; / this one is mcsel */
3064 :			int cbp = 0, i;
3065 :			int16_t in = Data->dctSpace, coeff = Data->dctSpace + 64;
3066 :	chl	1077
3067 :	syskin	1084	for(i = 0; i < 4; i++) {
3068 :			int s = 8((i&1) + (i>>1)Data->iEdgedWidth);
3069 :			transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16(x+yData->iEdgedWidth), Data->iEdgedWidth);
3070 :			bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i);
3071 :			if (bits >= Data->iMinSAD[0]) return bits;
3072 :			}
3073 :	chl	1077
3074 :	syskin	1084	bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
3075 :	chl	1077
3076 :	syskin	1084	/chroma U /
3077 :			transfer_8to16subro(in, Data->CurU, vGMC->u + 8(x+y(Data->iEdgedWidth/2)), Data->iEdgedWidth/2);
3078 :			bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4);
3079 :	chl	1077
3080 :	syskin	1084	if (bits >= Data->iMinSAD[0]) return bits;
3081 :
3082 :			/* chroma V */
3083 :			transfer_8to16subro(in, Data->CurV , vGMC->v + 8(x+y(Data->iEdgedWidth/2)), Data->iEdgedWidth/2);
3084 :			bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5);
3085 :
3086 :			bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) \| ((cbp & 3) << 3)].len;
3087 :
3088 :			*Data->cbp = cbp;
3089 :
3090 :			return bits;
3091 :			}
3092 :
3093 :
3094 :
3095 :
3096 :	chl	1077	static __inline void
3097 :			GMEanalyzeMB ( const uint8_t * const pCur,
3098 :			const uint8_t * const pRef,
3099 :			const uint8_t * const pRefH,
3100 :			const uint8_t * const pRefV,
3101 :			const uint8_t * const pRefHV,
3102 :			const int x,
3103 :			const int y,
3104 :			const MBParam * const pParam,
3105 :			MACROBLOCK * const pMBs,
3106 :			SearchData * const Data)
3107 :			{
3108 :
3109 :			int i=0;
3110 :			MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width];
3111 :
3112 :			Data->iMinSAD[0] = MV_MAX_ERROR;
3113 :
3114 :	syskin	1084	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
3115 :	chl	1077
3116 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
3117 :			pParam->width, pParam->height, 16, 1, 0);
3118 :	chl	1077
3119 :			Data->Cur = pCur + 16(x + y pParam->edged_width);
3120 :			Data->RefP[0] = pRef + 16(x + y pParam->edged_width);
3121 :			Data->RefP[1] = pRefV + 16(x + y pParam->edged_width);
3122 :			Data->RefP[2] = pRefH + 16(x + y pParam->edged_width);
3123 :			Data->RefP[3] = pRefHV + 16(x + y pParam->edged_width);
3124 :
3125 :			Data->currentMV[0].x = Data->currentMV[0].y = 0;
3126 :	syskin	1132	CheckCandidate16I(0, 0, Data, 255);
3127 :	chl	1077
3128 :			if ( (Data->predMV.x !=0) \|\| (Data->predMV.y != 0) )
3129 :	syskin	1132	CheckCandidate16I(Data->predMV.x, Data->predMV.y, Data, 255);
3130 :	chl	1077
3131 :	syskin	1132	DiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255, CheckCandidate16I);
3132 :	chl	1077
3133 :	syskin	1132	SubpelRefine(Data, CheckCandidate16I);
3134 :	syskin	1084
3135 :
3136 :	chl	1077	/* for QPel halfpel positions are worse than in halfpel mode :( */
3137 :			/* if (Data->qpel) {
3138 :			Data->currentQMV->x = 2*Data->currentMV->x;
3139 :			Data->currentQMV->y = 2*Data->currentMV->y;
3140 :			Data->qpel_precision = 1;
3141 :	syskin	1129	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
3142 :			pParam->width, pParam->height, iFcode, 2, 0);
3143 :	chl	1077	SubpelRefine(Data);
3144 :			}
3145 :			*/
3146 :
3147 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
3148 :			pMB->sad16 = Data->iMinSAD[0];
3149 :	syskin	1084	pMB->mode = MODE_INTER;
3150 :			pMB->sad16 += 10*d_mv_bits(pMB->mvs[0].x, pMB->mvs[0].y, Data->predMV, Data->iFcode, 0, 0);
3151 :	chl	1077	return;
3152 :			}
3153 :
3154 :			void
3155 :			GMEanalysis(const MBParam * const pParam,
3156 :			const FRAMEINFO * const current,
3157 :			const FRAMEINFO * const reference,
3158 :			const IMAGE * const pRefH,
3159 :			const IMAGE * const pRefV,
3160 :			const IMAGE * const pRefHV)
3161 :			{
3162 :			uint32_t x, y;
3163 :			MACROBLOCK * const pMBs = current->mbs;
3164 :			const IMAGE * const pCurrent = &current->image;
3165 :			const IMAGE * const pReference = &reference->image;
3166 :
3167 :			int32_t iMinSAD[5], temp[5];
3168 :			VECTOR currentMV[5];
3169 :	syskin	1132	uint32_t dir;
3170 :	chl	1077	SearchData Data;
3171 :			memset(&Data, 0, sizeof(SearchData));
3172 :	syskin	1084
3173 :	chl	1077	Data.iEdgedWidth = pParam->edged_width;
3174 :			Data.rounding = pParam->m_rounding_type;
3175 :	syskin	1084
3176 :	chl	1077	Data.currentMV = &currentMV[0];
3177 :			Data.iMinSAD = &iMinSAD[0];
3178 :			Data.iFcode = current->fcode;
3179 :			Data.temp = temp;
3180 :	syskin	1132	Data.dir = &dir;
3181 :	chl	1077
3182 :			if (sadInit) (*sadInit) ();
3183 :
3184 :			for (y = 0; y < pParam->mb_height; y ++) {
3185 :			for (x = 0; x < pParam->mb_width; x ++) {
3186 :			GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data);
3187 :			}
3188 :			}
3189 :			return;
3190 :			}
3191 :
3192 :
3193 :			WARPPOINTS
3194 :			GlobalMotionEst(MACROBLOCK * const pMBs,
3195 :			const MBParam * const pParam,
3196 :			const FRAMEINFO * const current,
3197 :			const FRAMEINFO * const reference,
3198 :			const IMAGE * const pRefH,
3199 :			const IMAGE * const pRefV,
3200 :			const IMAGE * const pRefHV)
3201 :			{
3202 :
3203 :	syskin	1084	const int deltax=8; // upper bound for difference between a MV and it's neighbour MVs
3204 :			const int deltay=8;
3205 :	chl	1077	const unsigned int gradx=512; // lower bound for gradient in MB (ignore "flat" blocks)
3206 :	syskin	1084	const unsigned int grady=512;
3207 :	chl	1077
3208 :			double sol[4] = { 0., 0., 0., 0. };
3209 :	syskin	1084
3210 :	chl	1077	WARPPOINTS gmc;
3211 :
3212 :			uint32_t mx, my;
3213 :
3214 :			int MBh = pParam->mb_height;
3215 :			int MBw = pParam->mb_width;
3216 :			const int minblocks = 9; //MBhMBw/32+3; / just some reasonable number 3% + 3 */
3217 :			const int maxblocks = MBhMBw/4; / just some reasonable number 3% + 3 */
3218 :
3219 :			int num=0;
3220 :			int oldnum;
3221 :
3222 :			gmc.duv[0].x = gmc.duv[0].y = gmc.duv[1].x = gmc.duv[1].y = gmc.duv[2].x = gmc.duv[2].y = 0;
3223 :
3224 :			GMEanalysis(pParam,current, reference, pRefH, pRefV, pRefHV);
3225 :
3226 :			/* block based ME isn't done, yet, so do a quick presearch */
3227 :
3228 :			// filter mask of all blocks
3229 :
3230 :			for (my = 0; my < (uint32_t)MBh; my++)
3231 :			for (mx = 0; mx < (uint32_t)MBw; mx++)
3232 :			{
3233 :			const int mbnum = mx + my * MBw;
3234 :			pMBs[mbnum].mcsel = 0;
3235 :			}
3236 :	syskin	1084
3237 :
3238 :			for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */
3239 :	chl	1077	for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */
3240 :			{
3241 :			const int mbnum = mx + my * MBw;
3242 :			MACROBLOCK *const pMB = &pMBs[mbnum];
3243 :			const VECTOR mv = pMB->mvs[0];
3244 :
3245 :			/* don't use object boundaries */
3246 :	syskin	1084	if ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax)
3247 :	chl	1077	&& (abs(mv.y - (pMB-1)->mvs[0].y) < deltay)
3248 :	syskin	1084	&& (abs(mv.x - (pMB+1)->mvs[0].x) < deltax)
3249 :	chl	1077	&& (abs(mv.y - (pMB+1)->mvs[0].y) < deltay)
3250 :	syskin	1084	&& (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax)
3251 :	chl	1077	&& (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay)
3252 :	syskin	1084	&& (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax)
3253 :	chl	1077	&& (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) )
3254 :			{ const int iEdgedWidth = pParam->edged_width;
3255 :			const uint8_t const pCur = current->image.y + 16(my*iEdgedWidth + mx);
3256 :			if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx )
3257 :			&& (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) )
3258 :	syskin	1084	{ pMB->mcsel = 1;
3259 :	chl	1077	num++;
3260 :			}
3261 :
3262 :			/* only use "structured" blocks */
3263 :			}
3264 :			}
3265 :			emms();
3266 :
3267 :	syskin	1084	/* further filtering would be possible, but during iteration, remaining
3268 :	chl	1077	outliers usually are removed, too */
3269 :
3270 :			if (num>= minblocks)
3271 :			do { /* until convergence */
3272 :			double DtimesF[4];
3273 :			double a,b,c,n,invdenom;
3274 :			double meanx,meany;
3275 :
3276 :			a = b = c = n = 0;
3277 :			DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.;
3278 :			for (my = 1; my < (uint32_t)MBh-1; my++)
3279 :			for (mx = 1; mx < (uint32_t)MBw-1; mx++)
3280 :			{
3281 :			const int mbnum = mx + my * MBw;
3282 :			const VECTOR mv = pMBs[mbnum].mvs[0];
3283 :
3284 :			if (!pMBs[mbnum].mcsel)
3285 :			continue;
3286 :
3287 :			n++;
3288 :			a += 16*mx+8;
3289 :			b += 16*my+8;
3290 :			c += (16mx+8)(16mx+8)+(16my+8)(16my+8);
3291 :
3292 :			DtimesF[0] += (double)mv.x;
3293 :			DtimesF[1] += (double)mv.x(16mx+8) + (double)mv.y(16my+8);
3294 :			DtimesF[2] += (double)mv.x(16my+8) - (double)mv.y(16mx+8);
3295 :			DtimesF[3] += (double)mv.y;
3296 :			}
3297 :	syskin	1084
3298 :	chl	1077	invdenom = aa+bb-c*n;
3299 :
3300 :			/* Solve the system: sol = (D'ED)^{-1} D'EF */
3301 :			/* D'EF has been calculated in the same loop as matrix */
3302 :
3303 :			sol[0] = -cDtimesF[0] + aDtimesF[1] + b*DtimesF[2];
3304 :			sol[1] = aDtimesF[0] - nDtimesF[1] + b*DtimesF[3];
3305 :			sol[2] = bDtimesF[0] - nDtimesF[2] - a*DtimesF[3];
3306 :			sol[3] = bDtimesF[1] - aDtimesF[2] - c*DtimesF[3];
3307 :
3308 :			sol[0] /= invdenom;
3309 :			sol[1] /= invdenom;
3310 :			sol[2] /= invdenom;
3311 :			sol[3] /= invdenom;
3312 :
3313 :			meanx = meany = 0.;
3314 :			oldnum = 0;
3315 :			for (my = 1; my < (uint32_t)MBh-1; my++)
3316 :			for (mx = 1; mx < (uint32_t)MBw-1; mx++)
3317 :			{
3318 :			const int mbnum = mx + my * MBw;
3319 :			const VECTOR mv = pMBs[mbnum].mvs[0];
3320 :
3321 :			if (!pMBs[mbnum].mcsel)
3322 :			continue;
3323 :
3324 :			oldnum++;
3325 :			meanx += fabs(( sol[0] + (16mx+8)sol[1] + (16my+8)sol[2] ) - (double)mv.x );
3326 :			meany += fabs(( sol[3] - (16mx+8)sol[2] + (16my+8)sol[1] ) - (double)mv.y );
3327 :			}
3328 :
3329 :			if (4meanx > oldnum) / better fit than 0.25 (=1/4pel) is useless */
3330 :			meanx /= oldnum;
3331 :			else
3332 :			meanx = 0.25;
3333 :
3334 :			if (4*meany > oldnum)
3335 :			meany /= oldnum;
3336 :			else
3337 :			meany = 0.25;
3338 :
3339 :			num = 0;
3340 :			for (my = 0; my < (uint32_t)MBh; my++)
3341 :			for (mx = 0; mx < (uint32_t)MBw; mx++)
3342 :			{
3343 :			const int mbnum = mx + my * MBw;
3344 :			const VECTOR mv = pMBs[mbnum].mvs[0];
3345 :
3346 :			if (!pMBs[mbnum].mcsel)
3347 :			continue;
3348 :
3349 :			if ( ( fabs(( sol[0] + (16mx+8)sol[1] + (16my+8)sol[2] ) - (double)mv.x ) > meanx )
3350 :			\|\| ( fabs(( sol[3] - (16mx+8)sol[2] + (16my+8)sol[1] ) - (double)mv.y ) > meany ) )
3351 :			pMBs[mbnum].mcsel=0;
3352 :			else
3353 :			num++;
3354 :			}
3355 :
3356 :			} while ( (oldnum != num) && (num>= minblocks) );
3357 :
3358 :			if (num < minblocks)
3359 :			{
3360 :			const int iEdgedWidth = pParam->edged_width;
3361 :			num = 0;
3362 :
3363 :			/* fprintf(stderr,"Warning! Unreliable GME (%d/%d blocks), falling back to translation.\n",num,MBh*MBw);
3364 :	syskin	1084	*/
3365 :	chl	1077	gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0;
3366 :
3367 :	edgomez	1107	if (!(current->motion_flags & XVID_ME_GME_REFINE))
3368 :	chl	1077	return gmc;
3369 :	syskin	1084
3370 :			for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */
3371 :	chl	1077	for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */
3372 :			{
3373 :			const int mbnum = mx + my * MBw;
3374 :			MACROBLOCK *const pMB = &pMBs[mbnum];
3375 :			const uint8_t const pCur = current->image.y + 16(my*iEdgedWidth + mx);
3376 :			if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx )
3377 :			&& (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) )
3378 :	syskin	1084	{ pMB->mcsel = 1;
3379 :	chl	1077	gmc.duv[0].x += pMB->mvs[0].x;
3380 :			gmc.duv[0].y += pMB->mvs[0].y;
3381 :			num++;
3382 :			}
3383 :			}
3384 :	syskin	1084
3385 :	chl	1077	if (gmc.duv[0].x)
3386 :			gmc.duv[0].x /= num;
3387 :	syskin	1084	if (gmc.duv[0].y)
3388 :	chl	1077	gmc.duv[0].y /= num;
3389 :			} else {
3390 :
3391 :			gmc.duv[0].x=(int)(sol[0]+0.5);
3392 :			gmc.duv[0].y=(int)(sol[3]+0.5);
3393 :
3394 :			gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5);
3395 :			gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5);
3396 :
3397 :			gmc.duv[2].x=-gmc.duv[1].y; /* two warp points only */
3398 :			gmc.duv[2].y=gmc.duv[1].x;
3399 :			}
3400 :			if (num>maxblocks)
3401 :			{ for (my = 1; my < (uint32_t)MBh-1; my++)
3402 :			for (mx = 1; mx < (uint32_t)MBw-1; mx++)
3403 :			{
3404 :			const int mbnum = mx + my * MBw;
3405 :			if (pMBs[mbnum-1].mcsel)
3406 :			pMBs[mbnum].mcsel=0;
3407 :	syskin	1084	else
3408 :	chl	1077	if (pMBs[mbnum-MBw].mcsel)
3409 :			pMBs[mbnum].mcsel=0;
3410 :			}
3411 :			}
3412 :			return gmc;
3413 :			}
3414 :
3415 :			int
3416 :			GlobalMotionEstRefine(
3417 :			WARPPOINTS *const startwp,
3418 :			MACROBLOCK * const pMBs,
3419 :			const MBParam * const pParam,
3420 :			const FRAMEINFO * const current,
3421 :			const FRAMEINFO * const reference,
3422 :			const IMAGE * const pCurr,
3423 :			const IMAGE * const pRef,
3424 :			const IMAGE * const pRefH,
3425 :			const IMAGE * const pRefV,
3426 :			const IMAGE * const pRefHV)
3427 :			{
3428 :			uint8_t* GMCblock = (uint8_t)malloc(16pParam->edged_width);
3429 :			WARPPOINTS bestwp=*startwp;
3430 :			WARPPOINTS centerwp,currwp;
3431 :			int gmcminSAD=0;
3432 :			int gmcSAD=0;
3433 :			int direction;
3434 :			// int mx,my;
3435 :
3436 :			/* use many blocks... */
3437 :			/* for (my = 0; my < (uint32_t)pParam->mb_height; my++)
3438 :			for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++)
3439 :			{
3440 :			const int mbnum = mx + my * pParam->mb_width;
3441 :	syskin	1084	pMBs[mbnum].mcsel=1;
3442 :	chl	1077	}
3443 :			*/
3444 :
3445 :			/* or rather don't use too many blocks... */
3446 :			/*
3447 :			for (my = 1; my < (uint32_t)MBh-1; my++)
3448 :			for (mx = 1; mx < (uint32_t)MBw-1; mx++)
3449 :			{
3450 :			const int mbnum = mx + my * MBw;
3451 :			if (MBmask[mbnum-1])
3452 :			MBmask[mbnum-1]=0;
3453 :	syskin	1084	else
3454 :	chl	1077	if (MBmask[mbnum-MBw])
3455 :			MBmask[mbnum-1]=0;
3456 :	syskin	1084
3457 :	chl	1077	}
3458 :			*/
3459 :			gmcminSAD = globalSAD(&bestwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3460 :
3461 :	syskin	1084	if ( (reference->coding_type == S_VOP)
3462 :			&& ( (reference->warp.duv[1].x != bestwp.duv[1].x)
3463 :			\|\| (reference->warp.duv[1].y != bestwp.duv[1].y)
3464 :			\|\| (reference->warp.duv[0].x != bestwp.duv[0].x)
3465 :			\|\| (reference->warp.duv[0].y != bestwp.duv[0].y)
3466 :			\|\| (reference->warp.duv[2].x != bestwp.duv[2].x)
3467 :			\|\| (reference->warp.duv[2].y != bestwp.duv[2].y) ) )
3468 :	chl	1077	{
3469 :	syskin	1084	gmcSAD = globalSAD(&reference->warp, pParam, pMBs,
3470 :	chl	1077	current, pRef, pCurr, GMCblock);
3471 :	syskin	1084
3472 :	chl	1077	if (gmcSAD < gmcminSAD)
3473 :			{ bestwp = reference->warp;
3474 :			gmcminSAD = gmcSAD;
3475 :			}
3476 :			}
3477 :
3478 :			do {
3479 :			direction = 0;
3480 :			centerwp = bestwp;
3481 :
3482 :	syskin	1084	currwp = centerwp;
3483 :
3484 :	chl	1077	currwp.duv[0].x--;
3485 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3486 :			if (gmcSAD < gmcminSAD)
3487 :			{ bestwp = currwp;
3488 :			gmcminSAD = gmcSAD;
3489 :	syskin	1084	direction = 1;
3490 :	chl	1077	}
3491 :	syskin	1084	else
3492 :			{
3493 :	chl	1077	currwp = centerwp; currwp.duv[0].x++;
3494 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3495 :			if (gmcSAD < gmcminSAD)
3496 :			{ bestwp = currwp;
3497 :			gmcminSAD = gmcSAD;
3498 :	syskin	1084	direction = 2;
3499 :	chl	1077	}
3500 :			}
3501 :			if (direction) continue;
3502 :
3503 :			currwp = centerwp; currwp.duv[0].y--;
3504 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3505 :			if (gmcSAD < gmcminSAD)
3506 :			{ bestwp = currwp;
3507 :			gmcminSAD = gmcSAD;
3508 :	syskin	1084	direction = 4;
3509 :			}
3510 :			else
3511 :	chl	1077	{
3512 :			currwp = centerwp; currwp.duv[0].y++;
3513 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3514 :			if (gmcSAD < gmcminSAD)
3515 :			{ bestwp = currwp;
3516 :			gmcminSAD = gmcSAD;
3517 :	syskin	1084	direction = 8;
3518 :	chl	1077	}
3519 :	syskin	1084	}
3520 :	chl	1077	if (direction) continue;
3521 :	syskin	1084
3522 :	chl	1077	currwp = centerwp; currwp.duv[1].x++;
3523 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3524 :			if (gmcSAD < gmcminSAD)
3525 :			{ bestwp = currwp;
3526 :			gmcminSAD = gmcSAD;
3527 :	syskin	1084	direction = 32;
3528 :			}
3529 :	chl	1077	currwp.duv[2].y++;
3530 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3531 :			if (gmcSAD < gmcminSAD)
3532 :			{ bestwp = currwp;
3533 :			gmcminSAD = gmcSAD;
3534 :	syskin	1084	direction = 1024;
3535 :			}
3536 :	chl	1077
3537 :			currwp = centerwp; currwp.duv[1].x--;
3538 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3539 :			if (gmcSAD < gmcminSAD)
3540 :			{ bestwp = currwp;
3541 :			gmcminSAD = gmcSAD;
3542 :	syskin	1084	direction = 16;
3543 :			}
3544 :			else
3545 :	chl	1077	{
3546 :			currwp = centerwp; currwp.duv[1].x++;
3547 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3548 :			if (gmcSAD < gmcminSAD)
3549 :			{ bestwp = currwp;
3550 :			gmcminSAD = gmcSAD;
3551 :	syskin	1084	direction = 32;
3552 :	chl	1077	}
3553 :	syskin	1084	}
3554 :	chl	1077	if (direction) continue;
3555 :
3556 :
3557 :			currwp = centerwp; currwp.duv[1].y--;
3558 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3559 :			if (gmcSAD < gmcminSAD)
3560 :			{ bestwp = currwp;
3561 :			gmcminSAD = gmcSAD;
3562 :	syskin	1084	direction = 64;
3563 :			}
3564 :			else
3565 :	chl	1077	{
3566 :			currwp = centerwp; currwp.duv[1].y++;
3567 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3568 :			if (gmcSAD < gmcminSAD)
3569 :			{ bestwp = currwp;
3570 :			gmcminSAD = gmcSAD;
3571 :	syskin	1084	direction = 128;
3572 :	chl	1077	}
3573 :			}
3574 :			if (direction) continue;
3575 :	syskin	1084
3576 :	chl	1077	currwp = centerwp; currwp.duv[2].x--;
3577 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3578 :			if (gmcSAD < gmcminSAD)
3579 :			{ bestwp = currwp;
3580 :			gmcminSAD = gmcSAD;
3581 :	syskin	1084	direction = 256;
3582 :			}
3583 :			else
3584 :	chl	1077	{
3585 :			currwp = centerwp; currwp.duv[2].x++;
3586 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3587 :			if (gmcSAD < gmcminSAD)
3588 :			{ bestwp = currwp;
3589 :			gmcminSAD = gmcSAD;
3590 :	syskin	1084	direction = 512;
3591 :	chl	1077	}
3592 :			}
3593 :			if (direction) continue;
3594 :	syskin	1084
3595 :	chl	1077	currwp = centerwp; currwp.duv[2].y--;
3596 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3597 :			if (gmcSAD < gmcminSAD)
3598 :			{ bestwp = currwp;
3599 :			gmcminSAD = gmcSAD;
3600 :	syskin	1084	direction = 1024;
3601 :			}
3602 :	chl	1077	else
3603 :	syskin	1084	{
3604 :	chl	1077	currwp = centerwp; currwp.duv[2].y++;
3605 :			gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock);
3606 :			if (gmcSAD < gmcminSAD)
3607 :			{ bestwp = currwp;
3608 :			gmcminSAD = gmcSAD;
3609 :	syskin	1084	direction = 2048;
3610 :	chl	1077	}
3611 :			}
3612 :			} while (direction);
3613 :			free(GMCblock);
3614 :	syskin	1084
3615 :	chl	1077	*startwp = bestwp;
3616 :
3617 :			return gmcminSAD;
3618 :			}
3619 :
3620 :			int
3621 :	syskin	1084	globalSAD(const WARPPOINTS *const wp,
3622 :			const MBParam * const pParam,
3623 :	chl	1077	const MACROBLOCK * const pMBs,
3624 :			const FRAMEINFO * const current,
3625 :			const IMAGE * const pRef,
3626 :			const IMAGE * const pCurr,
3627 :			uint8_t *const GMCblock)
3628 :			{
3629 :			NEW_GMC_DATA gmc_data;
3630 :			int iSAD, gmcSAD=0;
3631 :			int num=0;
3632 :			unsigned int mx, my;
3633 :	syskin	1084
3634 :	chl	1077	generate_GMCparameters( 3, 3, wp, pParam->width, pParam->height, &gmc_data);
3635 :	syskin	1084
3636 :	chl	1077	for (my = 0; my < (uint32_t)pParam->mb_height; my++)
3637 :			for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) {
3638 :
3639 :			const int mbnum = mx + my * pParam->mb_width;
3640 :			const int iEdgedWidth = pParam->edged_width;
3641 :	syskin	1084
3642 :	chl	1077	if (!pMBs[mbnum].mcsel)
3643 :			continue;
3644 :	syskin	1084
3645 :			gmc_data.predict_16x16(&gmc_data, GMCblock,
3646 :	chl	1077	pRef->y,
3647 :	syskin	1084	iEdgedWidth,
3648 :			iEdgedWidth,
3649 :			mx, my,
3650 :	chl	1077	pParam->m_rounding_type);
3651 :
3652 :			iSAD = sad16 ( pCurr->y + 16(myiEdgedWidth + mx),
3653 :	syskin	1084	GMCblock , iEdgedWidth, 65536);
3654 :	chl	1077	iSAD -= pMBs[mbnum].sad16;
3655 :	syskin	1084
3656 :			if (iSAD<0)
3657 :	chl	1077	gmcSAD += iSAD;
3658 :			num++;
3659 :			}
3660 :			return gmcSAD;
3661 :			}
3662 :

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