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

Revision 768 - (view) (download)

1 :	Isibaar	3	/**************************************************************************
2 :			*
3 :	chl	259	* XVID MPEG-4 VIDEO CODEC
4 :	syskin	748	* motion estimation
5 :	chl	259	*
6 :			* This program is an implementation of a part of one or more MPEG-4
7 :			* Video tools as specified in ISO/IEC 14496-2 standard. Those intending
8 :			* to use this software module in hardware or software products are
9 :			* advised that its use may infringe existing patents or copyrights, and
10 :			* any such use would be at such party's own risk. The original
11 :			* developer of this software module and his/her company, and subsequent
12 :			* editors and their companies, will have no liability for use of this
13 :			* software or modifications or derivatives thereof.
14 :			*
15 :			* This program is free software; you can redistribute it and/or modify
16 :			* it under the terms of the GNU General Public License as published by
17 :			* the Free Software Foundation; either version 2 of the License, or
18 :			* (at your option) any later version.
19 :			*
20 :			* This program is distributed in the hope that it will be useful,
21 :			* but WITHOUT ANY WARRANTY; without even the implied warranty of
22 :			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 :			* GNU General Public License for more details.
24 :			*
25 :			* You should have received a copy of the GNU General Public License
26 :			* along with this program; if not, write to the Free Software
27 :			* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 :			*
29 :			*************************************************************************/
30 :
31 :	Isibaar	3	#include <assert.h>
32 :			#include <stdio.h>
33 :	chl	96	#include <stdlib.h>
34 :	suxen_drol	721	#include <string.h> // memcpy
35 :	Isibaar	3
36 :			#include "../encoder.h"
37 :			#include "../utils/mbfunctions.h"
38 :			#include "../prediction/mbprediction.h"
39 :			#include "../global.h"
40 :			#include "../utils/timer.h"
41 :	Isibaar	580	#include "../image/interpolate8x8.h"
42 :	chl	530	#include "motion_est.h"
43 :	suxen_drol	118	#include "motion.h"
44 :	Isibaar	3	#include "sad.h"
45 :	Isibaar	539	#include "../utils/emms.h"
46 :	Isibaar	3
47 :	chl	530	#define INITIAL_SKIP_THRESH (10)
48 :			#define FINAL_SKIP_THRESH (50)
49 :			#define MAX_SAD00_FOR_SKIP (20)
50 :	Isibaar	539	#define MAX_CHROMA_SAD_FOR_SKIP (22)
51 :	Isibaar	3
52 :	chl	530	#define CHECK_CANDIDATE(X,Y,D) { \
53 :			(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
54 :	Isibaar	3
55 :	syskin	768	static __inline uint32_t
56 :	syskin	724	d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
57 :	Isibaar	3	{
58 :	chl	530	int xb, yb;
59 :	syskin	768	x += x * qpel; y += y * qpel;
60 :			if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
61 :	syskin	748	x -= pred.x;
62 :			y -= pred.y;
63 :	syskin	768
64 :	syskin	748	if (x) {
65 :	syskin	768	x = ABS(x);
66 :	chl	530	x += (1 << (iFcode - 1)) - 1;
67 :			x >>= (iFcode - 1);
68 :			if (x > 32) x = 32;
69 :			xb = mvtab[x] + iFcode;
70 :	syskin	748	} else xb = 1;
71 :	Isibaar	3
72 :	syskin	748	if (y) {
73 :	syskin	768	y = ABS(y);
74 :	chl	530	y += (1 << (iFcode - 1)) - 1;
75 :			y >>= (iFcode - 1);
76 :			if (y > 32) y = 32;
77 :			yb = mvtab[y] + iFcode;
78 :	syskin	748	} else yb = 1;
79 :	chl	530	return xb + yb;
80 :	Isibaar	3	}
81 :
82 :	syskin	748	static int32_t
83 :	syskin	628	ChromaSAD(int dx, int dy, const SearchData * const data)
84 :	Isibaar	3	{
85 :	syskin	628	int sad;
86 :	syskin	748	const uint32_t stride = data->iEdgedWidth/2;
87 :	Isibaar	3
88 :	syskin	712	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
89 :	syskin	768	data->temp[5] = dx; data->temp[6] = dy; // backup
90 :	syskin	712
91 :			switch (((dx & 1) << 1) \| (dy & 1)) {
92 :	syskin	628	case 0:
93 :	syskin	748	dx = dx / 2; dy = dy / 2;
94 :			sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride);
95 :			sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride);
96 :	syskin	628	break;
97 :			case 1:
98 :			dx = dx / 2; dy = (dy - 1) / 2;
99 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride);
100 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride);
101 :	syskin	628	break;
102 :			case 2:
103 :			dx = (dx - 1) / 2; dy = dy / 2;
104 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride);
105 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride);
106 :	syskin	628	break;
107 :			default:
108 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
109 :	syskin	748	interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding);
110 :			sad = sad8(data->CurU, data->RefQ, stride);
111 :
112 :			interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding);
113 :			sad += sad8(data->CurV, data->RefQ, stride);
114 :	syskin	628	break;
115 :	chl	530	}
116 :	syskin	748	data->temp[7] = sad; //backup, part 2
117 :	syskin	628	return sad;
118 :			}
119 :	Isibaar	3
120 :	syskin	663	static __inline const uint8_t *
121 :	syskin	768	GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data)
122 :	Isibaar	606	{
123 :	syskin	663	// dir : 0 = forward, 1 = backward
124 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
125 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
126 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
127 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
128 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
129 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
130 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
131 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
132 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
133 :	Isibaar	606	}
134 :			}
135 :
136 :	syskin	748	// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate
137 :			static __inline const uint8_t *
138 :			GetReference(const int x, const int y, const SearchData * const data)
139 :			{
140 :			switch ( ((x&1)<<1) \| (y&1) ) {
141 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
142 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
143 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
144 :			default : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
145 :			}
146 :			}
147 :
148 :			static uint8_t *
149 :	syskin	768	Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data)
150 :	chl	530	{
151 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
152 :	syskin	768	uint8_t * Reference = data->RefQ + 16*dir;
153 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
154 :	syskin	663	const uint32_t rounding = data->rounding;
155 :			const int halfpel_x = x/2;
156 :			const int halfpel_y = y/2;
157 :			const uint8_t ref1, ref2, ref3, ref4;
158 :	chl	326
159 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
160 :	syskin	663	ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
161 :			switch( ((x&1)<<1) + (y&1) ) {
162 :			case 0: // pure halfpel position
163 :	syskin	740	return (uint8_t *) ref1;
164 :	syskin	663	break;
165 :	Isibaar	3
166 :	syskin	663	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
167 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
168 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
169 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
170 :	syskin	663	break;
171 :	Isibaar	3
172 :	syskin	663	case 2: // x qpel, y halfpel - left or right during qpel refinement
173 :	syskin	748	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
174 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
175 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
176 :	syskin	663	break;
177 :	Isibaar	3
178 :	syskin	663	default: // x and y in qpel resolution - the "corners" (top left/right and
179 :			// bottom left/right) during qpel refinement
180 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
181 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
182 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
183 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
184 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
185 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
186 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
187 :			break;
188 :			}
189 :			return Reference;
190 :	Isibaar	3	}
191 :
192 :	syskin	748	static uint8_t *
193 :	syskin	768	Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data)
194 :	Isibaar	580	{
195 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
196 :	syskin	768	uint8_t * Reference = data->RefQ + 16*dir;
197 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
198 :	syskin	663	const uint32_t rounding = data->rounding;
199 :			const int halfpel_x = x/2;
200 :			const int halfpel_y = y/2;
201 :	Isibaar	580	const uint8_t ref1, ref2, ref3, ref4;
202 :
203 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
204 :	syskin	663	switch( ((x&1)<<1) + (y&1) ) {
205 :			case 0: // pure halfpel position
206 :	syskin	740	return (uint8_t *) ref1;
207 :	Isibaar	580	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
208 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
209 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
210 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
211 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
212 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
213 :	Isibaar	580	break;
214 :
215 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
216 :	syskin	768	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
217 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
218 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
219 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
220 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
221 :	Isibaar	580	break;
222 :
223 :			default: // x and y in qpel resolution - the "corners" (top left/right and
224 :			// bottom left/right) during qpel refinement
225 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
226 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
227 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
228 :	Isibaar	580	interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
229 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
230 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
231 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
232 :			break;
233 :			}
234 :	syskin	663	return Reference;
235 :			}
236 :
237 :			/* CHECK_CANDIATE FUNCTIONS START */
238 :
239 :	syskin	748	static void
240 :	syskin	663	CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
241 :			{
242 :			int t, xc, yc;
243 :			const uint8_t * Reference;
244 :			VECTOR * current;
245 :
246 :	syskin	768	if ( (x > data->max_dx) \| (x < data->min_dx)
247 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
248 :	syskin	663
249 :			if (data->qpel_precision) { // x and y are in 1/4 precision
250 :			Reference = Interpolate16x16qpel(x, y, 0, data);
251 :			xc = x/2; yc = y/2; //for chroma sad
252 :			current = data->currentQMV;
253 :			} else {
254 :	syskin	748	Reference = GetReference(x, y, data);
255 :	syskin	663	current = data->currentMV;
256 :			xc = x; yc = y;
257 :			}
258 :	syskin	768	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
259 :
260 :	syskin	663	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
261 :	Isibaar	580
262 :	syskin	768	data->temp[0] += (data->lambda16 * t * data->temp[0])>>10;
263 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
264 :	Isibaar	580
265 :	syskin	753	if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
266 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
267 :	syskin	628
268 :	Isibaar	580	if (data->temp[0] < data->iMinSAD[0]) {
269 :			data->iMinSAD[0] = data->temp[0];
270 :	syskin	663	current[0].x = x; current[0].y = y;
271 :			*dir = Direction; }
272 :	Isibaar	580
273 :			if (data->temp[1] < data->iMinSAD[1]) {
274 :	syskin	663	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; }
275 :	Isibaar	580	if (data->temp[2] < data->iMinSAD[2]) {
276 :	syskin	663	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
277 :	Isibaar	580	if (data->temp[3] < data->iMinSAD[3]) {
278 :	syskin	663	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
279 :	Isibaar	580	if (data->temp[4] < data->iMinSAD[4]) {
280 :	syskin	663	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
281 :
282 :	Isibaar	580	}
283 :
284 :	syskin	748	static void
285 :	syskin	704	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
286 :			{
287 :	syskin	768	uint32_t t;
288 :	syskin	704	const uint8_t * Reference;
289 :
290 :	syskin	768	if ( (!(x&1) && x !=0) \| (!(y&1) && y !=0) \|\| //non-zero integer value
291 :			(x > data->max_dx) \| (x < data->min_dx)
292 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
293 :	syskin	704
294 :	syskin	748	Reference = GetReference(x, y, data);
295 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
296 :	syskin	704
297 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
298 :
299 :	syskin	768	data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10;
300 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
301 :	syskin	704
302 :			if (data->temp[0] < data->iMinSAD[0]) {
303 :			data->iMinSAD[0] = data->temp[0];
304 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
305 :			*dir = Direction; }
306 :
307 :			if (data->temp[1] < data->iMinSAD[1]) {
308 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
309 :			if (data->temp[2] < data->iMinSAD[2]) {
310 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
311 :			if (data->temp[3] < data->iMinSAD[3]) {
312 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
313 :			if (data->temp[4] < data->iMinSAD[4]) {
314 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
315 :			}
316 :
317 :	syskin	748	static void
318 :	syskin	663	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
319 :	syskin	658	{
320 :			int32_t sad;
321 :	syskin	663	const uint8_t * Reference;
322 :	syskin	768	uint32_t t;
323 :	syskin	663	VECTOR * current;
324 :	syskin	658
325 :	syskin	768	if ( (x > data->max_dx) \| ( x < data->min_dx)
326 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
327 :	syskin	658
328 :	syskin	768	if (data->rrv && (!(x&1) && x !=0) \| (!(y&1) && y !=0) ) return; //non-zero even value
329 :	syskin	704
330 :	syskin	663	if (data->qpel_precision) { // x and y are in 1/4 precision
331 :			Reference = Interpolate16x16qpel(x, y, 0, data);
332 :			current = data->currentQMV;
333 :			} else {
334 :	syskin	748	Reference = GetReference(x, y, data);
335 :	syskin	663	current = data->currentMV;
336 :	syskin	658	}
337 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode,
338 :	syskin	768	data->qpel^data->qpel_precision, data->rrv);
339 :
340 :	syskin	658	sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
341 :	syskin	768	sad += (data->lambda16 * t * sad)>>10;
342 :	syskin	658
343 :	syskin	663	if (sad < *(data->iMinSAD)) {
344 :			*(data->iMinSAD) = sad;
345 :			current->x = x; current->y = y;
346 :			*dir = Direction; }
347 :	syskin	658	}
348 :
349 :	syskin	748	static void
350 :	syskin	739	CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
351 :	Isibaar	539	{
352 :	syskin	658	// maximum speed - for P/B/I decision
353 :	Isibaar	539
354 :	syskin	768	if ( (x > data->max_dx) \| (x < data->min_dx)
355 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
356 :	Isibaar	539
357 :	syskin	739	data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
358 :			data->iEdgedWidth, data->temp+1);
359 :	syskin	768
360 :	syskin	739	if (data->temp[0] < *(data->iMinSAD)) {
361 :			*(data->iMinSAD) = data->temp[0];
362 :	Isibaar	539	data->currentMV[0].x = x; data->currentMV[0].y = y;
363 :			*dir = Direction; }
364 :	syskin	739	if (data->temp[1] < data->iMinSAD[1]) {
365 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
366 :			if (data->temp[2] < data->iMinSAD[2]) {
367 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
368 :			if (data->temp[3] < data->iMinSAD[3]) {
369 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
370 :			if (data->temp[4] < data->iMinSAD[4]) {
371 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
372 :
373 :	Isibaar	539	}
374 :
375 :	syskin	748	static void
376 :	chl	530	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
377 :			{
378 :	syskin	768	int32_t sad, xb, yb;
379 :			uint32_t t;
380 :	chl	530	const uint8_t ReferenceF, ReferenceB;
381 :	syskin	663	VECTOR *current;
382 :	Isibaar	3
383 :	syskin	768	if ( (xf > data->max_dx) \| (xf < data->min_dx)
384 :			\| (yf > data->max_dy) \| (yf < data->min_dy) ) return;
385 :	Isibaar	3
386 :	syskin	768	if (!data->qpel_precision) {
387 :	syskin	748	ReferenceF = GetReference(xf, yf, data);
388 :	syskin	663	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
389 :	syskin	748	ReferenceB = GetReferenceB(xb, yb, 1, data);
390 :	syskin	663	current = data->currentMV;
391 :	syskin	768	} else {
392 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
393 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
394 :			current = data->currentQMV;
395 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
396 :	chl	530	}
397 :	edgomez	195
398 :	syskin	768	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0)
399 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0);
400 :
401 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
402 :	syskin	768	sad += (data->lambda16 * t * sad)>>10;
403 :	syskin	628
404 :	chl	530	if (sad < *(data->iMinSAD)) {
405 :			*(data->iMinSAD) = sad;
406 :	syskin	663	current->x = xf; current->y = yf;
407 :	chl	530	*dir = Direction; }
408 :	Isibaar	3	}
409 :
410 :	syskin	748	static void
411 :	chl	530	CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
412 :	Isibaar	3	{
413 :	syskin	628	int32_t sad = 0;
414 :	syskin	768	uint32_t k;
415 :	chl	530	const uint8_t *ReferenceF;
416 :			const uint8_t *ReferenceB;
417 :			VECTOR mvs, b_mvs;
418 :	Isibaar	3
419 :	syskin	768	if (( x > 31) \| ( x < -32) \| ( y > 31) \| (y < -32)) return;
420 :	Isibaar	3
421 :	chl	530	for (k = 0; k < 4; k++) {
422 :			mvs.x = data->directmvF[k].x + x;
423 :			b_mvs.x = ((x == 0) ?
424 :			data->directmvB[k].x
425 :			: mvs.x - data->referencemv[k].x);
426 :	Isibaar	3
427 :	chl	530	mvs.y = data->directmvF[k].y + y;
428 :			b_mvs.y = ((y == 0) ?
429 :			data->directmvB[k].y
430 :			: mvs.y - data->referencemv[k].y);
431 :	syskin	768
432 :	chl	530	if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
433 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
434 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
435 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
436 :	Isibaar	3
437 :	syskin	768
438 :			mvs.x = 2 - data->qpel; mvs.y = 2 - data->qpel;
439 :			b_mvs.x = 2 - data->qpel; b_mvs.y = 2 - data->qpel; //we move to qpel precision anyway
440 :
441 :	syskin	663	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
442 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
443 :	syskin	768
444 :	chl	530	sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
445 :	syskin	768	ReferenceF, ReferenceB, data->iEdgedWidth);
446 :	chl	530	if (sad > *(data->iMinSAD)) return;
447 :	edgomez	78	}
448 :	chl	530
449 :	syskin	768	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
450 :	syskin	628
451 :	chl	530	if (sad < *(data->iMinSAD)) {
452 :			*(data->iMinSAD) = sad;
453 :			data->currentMV->x = x; data->currentMV->y = y;
454 :			*dir = Direction; }
455 :	Isibaar	3	}
456 :
457 :	syskin	748	static void
458 :	chl	530	CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
459 :	chl	96	{
460 :	chl	530	int32_t sad;
461 :			const uint8_t *ReferenceF;
462 :			const uint8_t *ReferenceB;
463 :			VECTOR mvs, b_mvs;
464 :	chl	96
465 :	syskin	768	if (( x > 31) \| ( x < -32) \| ( y > 31) \| (y < -32)) return;
466 :	edgomez	195
467 :	chl	530	mvs.x = data->directmvF[0].x + x;
468 :			b_mvs.x = ((x == 0) ?
469 :			data->directmvB[0].x
470 :			: mvs.x - data->referencemv[0].x);
471 :	edgomez	195
472 :	chl	530	mvs.y = data->directmvF[0].y + y;
473 :			b_mvs.y = ((y == 0) ?
474 :			data->directmvB[0].y
475 :			: mvs.y - data->referencemv[0].y);
476 :	syskin	768
477 :	chl	530	if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
478 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
479 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
480 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
481 :	chl	96
482 :	syskin	768	mvs.x = 2 - data->qpel; mvs.y = 2 - data->qpel;
483 :			b_mvs.x = 2 - data->qpel; b_mvs.y = 2 - data->qpel; //we move to qpel precision anyway
484 :
485 :	syskin	663	ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
486 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
487 :	chl	96
488 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
489 :	syskin	768	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
490 :	chl	96
491 :	chl	530	if (sad < *(data->iMinSAD)) {
492 :			*(data->iMinSAD) = sad;
493 :			data->currentMV->x = x; data->currentMV->y = y;
494 :			*dir = Direction; }
495 :			}
496 :	chl	96
497 :	chl	530	static void
498 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
499 :			{
500 :	syskin	768	int32_t sad; uint32_t t;
501 :	chl	530	const uint8_t * Reference;
502 :	edgomez	195
503 :	syskin	768	if ( (x > data->max_dx) \| (x < data->min_dx)
504 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
505 :	chl	96
506 :	syskin	663	if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data);
507 :	syskin	768	else Reference = GetReference(x, y, data);
508 :	edgomez	195
509 :	chl	530	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
510 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
511 :	edgomez	195
512 :	syskin	768	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10;
513 :	chl	96
514 :	Isibaar	606	if (sad < *(data->iMinSAD)) {
515 :			*(data->iMinSAD) = sad;
516 :			data->currentMV->x = x; data->currentMV->y = y;
517 :			*dir = Direction; }
518 :			}
519 :
520 :	Isibaar	539	/* CHECK_CANDIATE FUNCTIONS END */
521 :	chl	96
522 :	chl	530	/* MAINSEARCH FUNCTIONS START */
523 :	edgomez	195
524 :	chl	530	static void
525 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
526 :	chl	181	{
527 :
528 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
529 :
530 :	syskin	712	int iDirection;
531 :	edgomez	195
532 :	syskin	712	for(;;) { //forever
533 :			iDirection = 0;
534 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
535 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
536 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
537 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
538 :	chl	181
539 :	syskin	712	/* now we're doing diagonal checks near our candidate */
540 :	chl	181
541 :	syskin	768	if (iDirection) { //if anything found
542 :	syskin	712	bDirection = iDirection;
543 :			iDirection = 0;
544 :			x = data->currentMV->x; y = data->currentMV->y;
545 :			if (bDirection & 3) { //our candidate is left or right
546 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
547 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
548 :			} else { // what remains here is up or down
549 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
550 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
551 :	syskin	712	}
552 :	chl	181
553 :	syskin	712	if (iDirection) {
554 :			bDirection += iDirection;
555 :	syskin	748	x = data->currentMV->x; y = data->currentMV->y;
556 :	chl	181	}
557 :	syskin	712	} else { //about to quit, eh? not so fast....
558 :			switch (bDirection) {
559 :			case 2:
560 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
561 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
562 :			break;
563 :			case 1:
564 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
565 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
566 :			break;
567 :			case 2 + 4:
568 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
569 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
570 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
571 :			break;
572 :			case 4:
573 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
574 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
575 :			break;
576 :			case 8:
577 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
578 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
579 :			break;
580 :			case 1 + 4:
581 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
582 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
583 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
584 :			break;
585 :			case 2 + 8:
586 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
587 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
588 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
589 :			break;
590 :			case 1 + 8:
591 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
592 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
593 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
594 :			break;
595 :			default: //1+2+4+8 == we didn't find anything at all
596 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
597 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
598 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
599 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
600 :			break;
601 :			}
602 :			if (!iDirection) break; //ok, the end. really
603 :			bDirection = iDirection;
604 :			x = data->currentMV->x; y = data->currentMV->y;
605 :	chl	181	}
606 :	syskin	712	}
607 :	chl	181	}
608 :
609 :	chl	530	static void
610 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
611 :	chl	326	{
612 :	chl	530	int iDirection;
613 :	chl	326
614 :	chl	530	do {
615 :			iDirection = 0;
616 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
617 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
618 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
619 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
620 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
621 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
622 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
623 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
624 :	chl	326
625 :	chl	530	bDirection = iDirection;
626 :			x = data->currentMV->x; y = data->currentMV->y;
627 :			} while (iDirection);
628 :	chl	346	}
629 :
630 :	chl	530	static void
631 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
632 :	chl	346	{
633 :
634 :	chl	181	/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
635 :
636 :	syskin	712	int iDirection;
637 :	edgomez	195
638 :	syskin	712	do {
639 :			iDirection = 0;
640 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
641 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
642 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
643 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
644 :	chl	181
645 :	syskin	712	/* now we're doing diagonal checks near our candidate */
646 :	chl	181
647 :	syskin	712	if (iDirection) { //checking if anything found
648 :			bDirection = iDirection;
649 :			iDirection = 0;
650 :			x = data->currentMV->x; y = data->currentMV->y;
651 :			if (bDirection & 3) { //our candidate is left or right
652 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
653 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
654 :			} else { // what remains here is up or down
655 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
656 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
657 :	chl	181	}
658 :	syskin	712	bDirection += iDirection;
659 :			x = data->currentMV->x; y = data->currentMV->y;
660 :	chl	181	}
661 :	syskin	712	}
662 :			while (iDirection);
663 :	chl	181	}
664 :
665 :	chl	530	/* MAINSEARCH FUNCTIONS END */
666 :	chl	181
667 :	chl	530	static void
668 :	syskin	663	SubpelRefine(const SearchData * const data)
669 :	chl	96	{
670 :	syskin	663	/* Do a half-pel or q-pel refinement */
671 :	syskin	768	const VECTOR centerMV = data->qpel_precision ? data->currentQMV : data->currentMV;
672 :			int iDirection; //only needed because macro expects it
673 :	chl	96
674 :	syskin	768	CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0);
675 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0);
676 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, 0);
677 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, 0);
678 :			CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, 0);
679 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, 0);
680 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, 0);
681 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, 0);
682 :	Isibaar	580	}
683 :
684 :	chl	530	static __inline int
685 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
686 :			const int x, const int y,
687 :	syskin	768	const uint32_t stride, const uint32_t iQuant, int rrv)
688 :	chl	96
689 :	Isibaar	3	{
690 :	syskin	768	if(!rrv) {
691 :			uint32_t sadC = sad8(current->u + x8 + ystride*8,
692 :			reference->u + x8 + ystride*8, stride);
693 :	syskin	704	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
694 :	syskin	768	sadC += sad8(current->v + (x + ystride)8,
695 :			reference->v + (x + ystride)8, stride);
696 :	syskin	704	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
697 :			return 1;
698 :	syskin	768
699 :			} else {
700 :			uint32_t sadC = sad16(current->u + x16 + ystride*16,
701 :			reference->u + x16 + ystride16, stride, 2564096);
702 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
703 :			sadC += sad16(current->v + (x + ystride)16,
704 :			reference->v + (x + ystride)16, stride, 256*4096);
705 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
706 :			return 1;
707 :	syskin	704	}
708 :	chl	530	}
709 :	edgomez	195
710 :	chl	530	static __inline void
711 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
712 :			{
713 :			pMB->mode = MODE_NOT_CODED;
714 :	syskin	768	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV;
715 :			pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV;
716 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
717 :	Isibaar	3	}
718 :
719 :	syskin	628	bool
720 :	chl	530	MotionEstimation(MBParam * const pParam,
721 :			FRAMEINFO * const current,
722 :			FRAMEINFO * const reference,
723 :			const IMAGE * const pRefH,
724 :			const IMAGE * const pRefV,
725 :			const IMAGE * const pRefHV,
726 :			const uint32_t iLimit)
727 :	Isibaar	3	{
728 :	chl	530	MACROBLOCK *const pMBs = current->mbs;
729 :			const IMAGE *const pCurrent = &current->image;
730 :			const IMAGE *const pRef = &reference->image;
731 :	Isibaar	3
732 :	syskin	704	uint32_t mb_width = pParam->mb_width;
733 :			uint32_t mb_height = pParam->mb_height;
734 :	syskin	768	const uint32_t iEdgedWidth = pParam->edged_width;
735 :	suxen_drol	702
736 :	chl	530	uint32_t x, y;
737 :			uint32_t iIntra = 0;
738 :	syskin	628	int32_t InterBias, quant = current->quant, sad00;
739 :	edgomez	195
740 :	Isibaar	539	// some pre-initialized thingies for SearchP
741 :	syskin	712	int32_t temp[8];
742 :	Isibaar	539	VECTOR currentMV[5];
743 :	Isibaar	580	VECTOR currentQMV[5];
744 :	Isibaar	539	int32_t iMinSAD[5];
745 :			SearchData Data;
746 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
747 :	syskin	768	Data.iEdgedWidth = iEdgedWidth;
748 :	Isibaar	539	Data.currentMV = currentMV;
749 :	Isibaar	580	Data.currentQMV = currentQMV;
750 :	Isibaar	539	Data.iMinSAD = iMinSAD;
751 :			Data.temp = temp;
752 :			Data.iFcode = current->fcode;
753 :	Isibaar	580	Data.rounding = pParam->m_rounding_type;
754 :	syskin	628	Data.qpel = pParam->m_quarterpel;
755 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
756 :	syskin	704	Data.rrv = current->global_flags & XVID_REDUCED;
757 :	Isibaar	539
758 :	syskin	704	if ((current->global_flags & XVID_REDUCED)) {
759 :	suxen_drol	702	mb_width = (pParam->width + 31) / 32;
760 :			mb_height = (pParam->height + 31) / 32;
761 :	syskin	704	Data.qpel = Data.chroma = 0;
762 :	suxen_drol	702	}
763 :
764 :	syskin	754	Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose)
765 :	chl	530	if (sadInit) (*sadInit) ();
766 :	edgomez	195
767 :	suxen_drol	702	for (y = 0; y < mb_height; y++) {
768 :			for (x = 0; x < mb_width; x++) {
769 :	chl	530	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
770 :	syskin	768
771 :			if (!Data.rrv) pMB->sad16 =
772 :			sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16,
773 :			pRef->y + (x + y * iEdgedWidth) * 16,
774 :	syskin	704	pParam->edged_width, pMB->sad8 );
775 :	syskin	768
776 :	syskin	748	else pMB->sad16 =
777 :	syskin	768	sad32v_c(pCurrent->y + (x + y * iEdgedWidth) * 32,
778 :			pRef->y + (x + y * iEdgedWidth) * 32,
779 :	chl	530	pParam->edged_width, pMB->sad8 );
780 :	edgomez	195
781 :	syskin	628	if (Data.chroma) {
782 :	syskin	768	Data.temp[7] = sad8(pCurrent->u + x8 + y(iEdgedWidth/2)*8,
783 :			pRef->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2)
784 :			+ sad8(pCurrent->v + (x + y(iEdgedWidth/2))8,
785 :			pRef->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
786 :			pMB->sad16 += Data.temp[7];
787 :	syskin	628	}
788 :
789 :	syskin	768	sad00 = pMB->sad16;
790 :	syskin	628
791 :	chl	530	if (!(current->global_flags & XVID_LUMIMASKING)) {
792 :			pMB->dquant = NO_CHANGE;
793 :	syskin	601	} else {
794 :	syskin	574	if (pMB->dquant != NO_CHANGE) {
795 :			quant += DQtab[pMB->dquant];
796 :			if (quant > 31) quant = 31;
797 :			else if (quant < 1) quant = 1;
798 :			}
799 :	syskin	601	}
800 :	syskin	768	pMB->quant = current->quant;
801 :	chl	181
802 :	syskin	628	//initial skip decision
803 :	syskin	768	/* no early skip for GMC (global vector = skip vector is unknown!) */
804 :	chl	619	if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
805 :	syskin	768	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
806 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
807 :	chl	619	SkipMacroblockP(pMB, sad00);
808 :			continue;
809 :			}
810 :			}
811 :	syskin	768
812 :	syskin	628	SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
813 :	chl	530	y, current->motion_flags, pMB->quant,
814 :	Isibaar	539	&Data, pParam, pMBs, reference->mbs,
815 :	chl	530	current->global_flags & XVID_INTER4V, pMB);
816 :	edgomez	195
817 :	chl	530	/* final skip decision, a.k.a. "the vector you found, really that good?" */
818 :	syskin	748	if (current->coding_type == P_VOP) {
819 :	chl	619	if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
820 :	syskin	704	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
821 :	syskin	768	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
822 :	chl	619	SkipMacroblockP(pMB, sad00);
823 :			continue;
824 :			}
825 :			}
826 :	syskin	768
827 :	chl	530	/* finally, intra decision */
828 :	Isibaar	3
829 :	chl	530	InterBias = MV16_INTER_BIAS;
830 :	syskin	768	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
831 :	chl	530	if (y != 0)
832 :	syskin	639	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
833 :	chl	530	if (x != 0)
834 :	syskin	639	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
835 :	syskin	768
836 :	syskin	628	if (Data.chroma) InterBias += 50; // to compensate bigger SAD
837 :	syskin	768	if (Data.rrv) InterBias *= 4;
838 :	Isibaar	3
839 :	syskin	768	if (InterBias < pMB->sad16) {
840 :	syskin	704	int32_t deviation;
841 :	syskin	768	if (!Data.rrv)
842 :			deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 16, iEdgedWidth);
843 :			else {
844 :			deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 32, iEdgedWidth)
845 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16, iEdgedWidth)
846 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * iEdgedWidth, iEdgedWidth)
847 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * (iEdgedWidth+1), iEdgedWidth);
848 :			}
849 :			if (deviation < (pMB->sad16 - InterBias)) {
850 :	syskin	754	if (++iIntra >= iLimit) return 1;
851 :	syskin	768	SkipMacroblockP(pMB, 0); //same thing
852 :	chl	530	pMB->mode = MODE_INTRA;
853 :			}
854 :			}
855 :			}
856 :	Isibaar	3	}
857 :	chl	619
858 :			if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
859 :	syskin	628	current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
860 :	chl	619	else
861 :			current->GMC_MV = zeroMV;
862 :
863 :	syskin	628	return 0;
864 :	chl	530	}
865 :	Isibaar	3
866 :	edgomez	195
867 :	chl	530	static __inline int
868 :			make_mask(const VECTOR * const pmv, const int i)
869 :			{
870 :	Isibaar	539	int mask = 255, j;
871 :	chl	530	for (j = 0; j < i; j++) {
872 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
873 :			if (pmv[i].x == pmv[j].x) {
874 :	syskin	768	if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4;
875 :			else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8;
876 :	syskin	748	} else
877 :	chl	530	if (pmv[i].y == pmv[j].y) {
878 :	syskin	768	if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1;
879 :			else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2;
880 :	chl	530	}
881 :	Isibaar	3	}
882 :	chl	530	return mask;
883 :			}
884 :	edgomez	195
885 :	syskin	748	static __inline void
886 :	syskin	704	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
887 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
888 :	chl	530	{
889 :	edgomez	195
890 :	chl	530	//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
891 :	syskin	704	if (rrv) { iWcount /= 2; iHcount /= 2; }
892 :	edgomez	195
893 :	syskin	704	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
894 :	chl	530	pmv[5].x = EVEN(pmv[3].x);
895 :	syskin	748	pmv[5].y = EVEN(pmv[3].y);
896 :	Isibaar	539	} else pmv[5].x = pmv[5].y = 0;
897 :	chl	169
898 :	chl	530	if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
899 :			else pmv[3].x = pmv[3].y = 0;
900 :	Isibaar	3
901 :	chl	530	if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
902 :	syskin	768	else pmv[4].x = pmv[4].y = 0;
903 :	chl	169
904 :	chl	530	// [1] median prediction
905 :	syskin	767	pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
906 :	chl	169
907 :	chl	530	pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
908 :	chl	169
909 :	chl	530	pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
910 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
911 :	chl	169
912 :	syskin	704	if ((x < iWcount-1) && (y < iHcount-1)) {
913 :	chl	530	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
914 :	syskin	748	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
915 :	Isibaar	539	} else pmv[6].x = pmv[6].y = 0;
916 :	syskin	704
917 :			if (rrv) {
918 :			int i;
919 :			for (i = 0; i < 7; i++) {
920 :	syskin	768	pmv[i].x = RRV_MV_SCALEUP(pmv[i].x);
921 :	syskin	767	pmv[i].y = RRV_MV_SCALEUP(pmv[i].y);
922 :	syskin	704	}
923 :			}
924 :	chl	530	}
925 :	chl	169
926 :	chl	530	static void
927 :	syskin	628	SearchP(const IMAGE * const pRef,
928 :	chl	530	const uint8_t * const pRefH,
929 :			const uint8_t * const pRefV,
930 :			const uint8_t * const pRefHV,
931 :			const IMAGE * const pCur,
932 :			const int x,
933 :			const int y,
934 :			const uint32_t MotionFlags,
935 :			const uint32_t iQuant,
936 :	Isibaar	539	SearchData * const Data,
937 :	chl	530	const MBParam * const pParam,
938 :			const MACROBLOCK * const pMBs,
939 :			const MACROBLOCK * const prevMBs,
940 :			int inter4v,
941 :			MACROBLOCK * const pMB)
942 :			{
943 :	chl	169
944 :	chl	530	int i, iDirection = 255, mask, threshA;
945 :	Isibaar	539	VECTOR pmv[7];
946 :	Isibaar	3
947 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
948 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
949 :	Isibaar	3
950 :	syskin	768	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp);
951 :	syskin	628
952 :	syskin	768	Data->temp[5] = Data->temp[6] = 0; // chroma-sad cache
953 :			i = Data->rrv ? 2 : 1;
954 :	syskin	704	Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
955 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
956 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
957 :	Isibaar	3
958 :	syskin	704	Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
959 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
960 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
961 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
962 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
963 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
964 :
965 :	syskin	601	Data->lambda16 = lambda_vec16[iQuant];
966 :	Isibaar	606	Data->lambda8 = lambda_vec8[iQuant];
967 :	syskin	663	Data->qpel_precision = 0;
968 :	chl	169
969 :	edgomez	568	if (pMB->dquant != NO_CHANGE) inter4v = 0;
970 :	edgomez	195
971 :	syskin	768	for(i = 0; i < 5; i++)
972 :	Isibaar	580	Data->currentMV[i].x = Data->currentMV[i].y = 0;
973 :	Isibaar	539
974 :	syskin	768	if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
975 :	syskin	663	else Data->predMV = pmv[0];
976 :	Isibaar	606
977 :	syskin	724	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
978 :	syskin	768	Data->iMinSAD[0] = pMB->sad16 + ((Data->lambda16 * i * pMB->sad16)>>10);
979 :			Data->iMinSAD[1] = pMB->sad8[0] + ((Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS)) >> 10);
980 :	Isibaar	539	Data->iMinSAD[2] = pMB->sad8[1];
981 :			Data->iMinSAD[3] = pMB->sad8[2];
982 :			Data->iMinSAD[4] = pMB->sad8[3];
983 :
984 :	syskin	768	if (x \| y) {
985 :	edgomez	568	threshA = Data->temp[0]; // that's when we keep this SAD atm
986 :	chl	530	if (threshA < 512) threshA = 512;
987 :	syskin	768	else if (threshA > 1024) threshA = 1024;
988 :			} else threshA = 512;
989 :	Isibaar	3
990 :	chl	530	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
991 :	syskin	704	prevMBs + x + y * pParam->mb_width, Data->rrv);
992 :	edgomez	195
993 :	syskin	768	if (!Data->rrv) {
994 :			if (inter4v \| Data->chroma) CheckCandidate = CheckCandidate16;
995 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
996 :			} else CheckCandidate = CheckCandidate32;
997 :	edgomez	568
998 :	syskin	768	/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/
999 :	chl	530
1000 :			for (i = 1; i < 7; i++) {
1001 :			if (!(mask = make_mask(pmv, i)) ) continue;
1002 :	edgomez	568	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1003 :	Isibaar	539	if (Data->iMinSAD[0] <= threshA) break;
1004 :	chl	530	}
1005 :
1006 :	Isibaar	539	if ((Data->iMinSAD[0] <= threshA) \|\|
1007 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1008 :	syskin	768	(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16)))
1009 :	chl	530	inter4v = 0;
1010 :	syskin	768	else {
1011 :	Isibaar	3
1012 :	syskin	748	MainSearchFunc * MainSearchPtr;
1013 :	Isibaar	539	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1014 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1015 :			else MainSearchPtr = DiamondSearch;
1016 :	chl	181
1017 :	Isibaar	539	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1018 :	Isibaar	3
1019 :	chl	530	/* extended search, diamond starting in 0,0 and in prediction.
1020 :			note that this search is/might be done in halfpel positions,
1021 :			which makes it more different than the diamond above */
1022 :	chl	259
1023 :	Isibaar	539	if (MotionFlags & PMV_EXTSEARCH16) {
1024 :			int32_t bSAD;
1025 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1026 :	syskin	704	if (Data->rrv) {
1027 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1028 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1029 :	syskin	748	} else
1030 :	syskin	704	if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1031 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1032 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1033 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1034 :	Isibaar	3
1035 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1036 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1037 :			if (bSAD < Data->iMinSAD[0]) {
1038 :			Data->currentMV[0] = backupMV;
1039 :			Data->iMinSAD[0] = bSAD; }
1040 :			}
1041 :	Isibaar	3
1042 :	Isibaar	539	backupMV = Data->currentMV[0];
1043 :	syskin	768	if (MotionFlags & PMV_HALFPELREFINE16 && !Data->rrv) startMV.x = startMV.y = 1;
1044 :			else startMV.x = startMV.y = 0;
1045 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1046 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1047 :	edgomez	195
1048 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1049 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1050 :			if (bSAD < Data->iMinSAD[0]) {
1051 :			Data->currentMV[0] = backupMV;
1052 :			Data->iMinSAD[0] = bSAD; }
1053 :			}
1054 :	Isibaar	3	}
1055 :			}
1056 :
1057 :	syskin	663	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1058 :	Isibaar	3
1059 :	Isibaar	580	for(i = 0; i < 5; i++) {
1060 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1061 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1062 :			}
1063 :
1064 :	syskin	768	if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE16) {
1065 :	syskin	663	Data->qpel_precision = 1;
1066 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1067 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1068 :	syskin	601
1069 :	syskin	663	SubpelRefine(Data);
1070 :	Isibaar	580	}
1071 :
1072 :	syskin	768	if (Data->iMinSAD[0] < (int32_t)iQuant * 30) inter4v = 0;
1073 :	Isibaar	539	if (inter4v) {
1074 :			SearchData Data8;
1075 :	syskin	768	memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data
1076 :
1077 :	Isibaar	539	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1078 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1079 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1080 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1081 :	syskin	768
1082 :	syskin	628	if (Data->chroma) {
1083 :	syskin	768	int sumx = 0, sumy = 0;
1084 :			const int div = 1 + Data->qpel;
1085 :			const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs;
1086 :	syskin	628
1087 :	syskin	768	for (i = 0; i < 4; i++) {
1088 :			sumx += mv[i].x / div;
1089 :			sumy += mv[i].y / div;
1090 :	syskin	663	}
1091 :	syskin	768
1092 :	syskin	753	Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf],
1093 :			(sumy >> 3) + roundtab_76[sumy & 0xf], Data);
1094 :	syskin	628	}
1095 :	Isibaar	539	}
1096 :	Isibaar	3
1097 :	syskin	704	if (Data->rrv) {
1098 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1099 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1100 :			}
1101 :	syskin	740
1102 :	chl	530	if (!(inter4v) \|\|
1103 :	syskin	748	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1104 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1105 :	chl	530	// INTER MODE
1106 :			pMB->mode = MODE_INTER;
1107 :	syskin	768	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1108 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1109 :	Isibaar	3
1110 :	syskin	768	if(Data->qpel) {
1111 :	syskin	663	pMB->qmvs[0] = pMB->qmvs[1]
1112 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1113 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1114 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1115 :	syskin	628	} else {
1116 :	Isibaar	580	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1117 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1118 :			}
1119 :	edgomez	195	} else {
1120 :	syskin	748	// INTER4V MODE; all other things are already set in Search8
1121 :	chl	530	pMB->mode = MODE_INTER4V;
1122 :	syskin	748	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1123 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1124 :	edgomez	78	}
1125 :	Isibaar	3	}
1126 :
1127 :	chl	530	static void
1128 :			Search8(const SearchData * const OldData,
1129 :			const int x, const int y,
1130 :			const uint32_t MotionFlags,
1131 :			const MBParam * const pParam,
1132 :			MACROBLOCK * const pMB,
1133 :			const MACROBLOCK * const pMBs,
1134 :	Isibaar	539	const int block,
1135 :			SearchData * const Data)
1136 :	chl	345	{
1137 :	syskin	704	int i = 0;
1138 :	Isibaar	539	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1139 :			Data->currentMV = OldData->currentMV + 1 + block;
1140 :	Isibaar	580	Data->currentQMV = OldData->currentQMV + 1 + block;
1141 :	chl	345
1142 :	syskin	768	if(Data->qpel) {
1143 :	syskin	704	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1144 :	syskin	724	if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1145 :			Data->predMV, Data->iFcode, 0, 0);
1146 :	Isibaar	606	} else {
1147 :	syskin	704	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1148 :	syskin	768	if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1149 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1150 :	Isibaar	606	}
1151 :	chl	345
1152 :	syskin	768	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
1153 :	syskin	704
1154 :	syskin	768	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8\|PMV_QUARTERPELREFINE8)) {
1155 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
1156 :	chl	345
1157 :	syskin	768	Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1158 :			Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1159 :			Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1160 :			Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1161 :	chl	345
1162 :	syskin	768	Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1163 :	syskin	663	Data->qpel_precision = 0;
1164 :	chl	345
1165 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1166 :	syskin	768	pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
1167 :	syskin	712
1168 :	syskin	768	if (!Data->rrv) CheckCandidate = CheckCandidate8;
1169 :			else CheckCandidate = CheckCandidate16no4v;
1170 :	syskin	712
1171 :	chl	530	if (MotionFlags & PMV_EXTSEARCH8) {
1172 :	Isibaar	580	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1173 :	syskin	768
1174 :	chl	530	MainSearchFunc *MainSearchPtr;
1175 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1176 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1177 :			else MainSearchPtr = DiamondSearch;
1178 :	chl	345
1179 :	Isibaar	580	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1180 :	chl	345
1181 :	syskin	601	if(*(Data->iMinSAD) < temp_sad) {
1182 :	Isibaar	580	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1183 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1184 :			}
1185 :			}
1186 :
1187 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1188 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1189 :
1190 :	syskin	663	SubpelRefine(Data); // perform halfpel refine of current best vector
1191 :	Isibaar	580
1192 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1193 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1194 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1195 :			}
1196 :			}
1197 :
1198 :	syskin	768	if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE8) {
1199 :			Data->qpel_precision = 1;
1200 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1201 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1202 :			SubpelRefine(Data);
1203 :	Isibaar	580	}
1204 :	chl	345	}
1205 :	chl	530
1206 :	syskin	704	if (Data->rrv) {
1207 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1208 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1209 :			}
1210 :	syskin	768
1211 :	syskin	712	if(Data->qpel) {
1212 :	syskin	663	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1213 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1214 :	syskin	768	pMB->qmvs[block] = *Data->currentQMV;
1215 :	syskin	704	} else {
1216 :	Isibaar	580	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1217 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1218 :			}
1219 :
1220 :	syskin	768	pMB->mvs[block] = *Data->currentMV;
1221 :			pMB->sad8[block] = 4 * *Data->iMinSAD;
1222 :	chl	345	}
1223 :
1224 :	syskin	768	/* motion estimation for B-frames */
1225 :	chl	345
1226 :	chl	530	static __inline VECTOR
1227 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1228 :	Isibaar	3	{
1229 :	chl	530	/* the stupidiest function ever */
1230 :	syskin	768	return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]);
1231 :	Isibaar	3	}
1232 :
1233 :	chl	530	static void __inline
1234 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1235 :			const uint32_t iWcount,
1236 :			const MACROBLOCK * const pMB,
1237 :			const uint32_t mode_curr)
1238 :	Isibaar	3	{
1239 :
1240 :	chl	530	// [0] is prediction
1241 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1242 :	Isibaar	3
1243 :	chl	530	pmv[1].x = pmv[1].y = 0; // [1] is zero
1244 :	Isibaar	3
1245 :	chl	530	pmv[2] = ChoosePred(pMB, mode_curr);
1246 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1247 :	edgomez	195
1248 :	chl	530	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1249 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1250 :	syskin	748	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1251 :	Isibaar	539	} else pmv[3].x = pmv[3].y = 0;
1252 :	Isibaar	3
1253 :	chl	530	if (y != 0) {
1254 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1255 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1256 :			} else pmv[4].x = pmv[4].y = 0;
1257 :	Isibaar	3
1258 :	chl	530	if (x != 0) {
1259 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1260 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1261 :			} else pmv[5].x = pmv[5].y = 0;
1262 :	Isibaar	3
1263 :	syskin	768	if (x != 0 && y != 0) {
1264 :	chl	530	pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1265 :	syskin	767	pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y);
1266 :	chl	530	} else pmv[6].x = pmv[6].y = 0;
1267 :			}
1268 :	chl	181
1269 :	edgomez	170
1270 :	syskin	768	/* search backward or forward */
1271 :	chl	530	static void
1272 :			SearchBF( const uint8_t * const pRef,
1273 :			const uint8_t * const pRefH,
1274 :			const uint8_t * const pRefV,
1275 :			const uint8_t * const pRefHV,
1276 :			const IMAGE * const pCur,
1277 :			const int x, const int y,
1278 :			const uint32_t MotionFlags,
1279 :			const uint32_t iFcode,
1280 :			const MBParam * const pParam,
1281 :			MACROBLOCK * const pMB,
1282 :			const VECTOR * const predMV,
1283 :			int32_t * const best_sad,
1284 :	h	545	const int32_t mode_current,
1285 :			SearchData * const Data)
1286 :	chl	530	{
1287 :	Isibaar	3
1288 :	syskin	740	int i, iDirection = 255, mask;
1289 :	h	545	VECTOR pmv[7];
1290 :	chl	530	MainSearchFunc *MainSearchPtr;
1291 :	h	545	*Data->iMinSAD = MV_MAX_ERROR;
1292 :			Data->iFcode = iFcode;
1293 :	syskin	663	Data->qpel_precision = 0;
1294 :	Isibaar	3
1295 :	syskin	768	Data->Ref = pRef + (x + y * Data->iEdgedWidth) * 16;
1296 :			Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16;
1297 :			Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16;
1298 :			Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16;
1299 :	Isibaar	3
1300 :	h	545	Data->predMV = *predMV;
1301 :	edgomez	195
1302 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1303 :	syskin	712	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1304 :	Isibaar	3
1305 :	h	545	pmv[0] = Data->predMV;
1306 :	syskin	658	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1307 :	syskin	768
1308 :	h	545	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1309 :	edgomez	195
1310 :	h	545	Data->currentMV->x = Data->currentMV->y = 0;
1311 :	chl	530	CheckCandidate = CheckCandidate16no4v;
1312 :	chl	181
1313 :	chl	530	// main loop. checking all predictions
1314 :	syskin	740	for (i = 0; i < 7; i++) {
1315 :	chl	530	if (!(mask = make_mask(pmv, i)) ) continue;
1316 :	h	545	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1317 :	edgomez	78	}
1318 :	Isibaar	3
1319 :	syskin	768	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1320 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1321 :	chl	530	else MainSearchPtr = DiamondSearch;
1322 :	chl	96
1323 :	syskin	740	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1324 :	chl	169
1325 :	syskin	663	SubpelRefine(Data);
1326 :	syskin	768
1327 :	syskin	740	if (Data->qpel && Data->iMinSAD < best_sad + 300) {
1328 :	syskin	658	Data->currentQMV->x = 2*Data->currentMV->x;
1329 :			Data->currentQMV->y = 2*Data->currentMV->y;
1330 :	syskin	663	Data->qpel_precision = 1;
1331 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1332 :			pParam->width, pParam->height, iFcode, 1, 0);
1333 :	syskin	663	SubpelRefine(Data);
1334 :	syskin	658	}
1335 :	chl	169
1336 :	chl	530	// three bits are needed to code backward mode. four for forward
1337 :	syskin	740
1338 :	syskin	768	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1339 :			else Data->iMinSAD += 3 Data->lambda16;
1340 :	chl	169
1341 :	h	545	if (Data->iMinSAD < best_sad) {
1342 :			best_sad = Data->iMinSAD;
1343 :	chl	530	pMB->mode = mode_current;
1344 :	syskin	658	if (Data->qpel) {
1345 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1346 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1347 :	syskin	748	if (mode_current == MODE_FORWARD)
1348 :	syskin	658	pMB->qmvs[0] = *Data->currentQMV;
1349 :	syskin	748	else
1350 :	syskin	658	pMB->b_qmvs[0] = *Data->currentQMV;
1351 :			} else {
1352 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1353 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1354 :			}
1355 :	syskin	739	if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1356 :			else pMB->b_mvs[0] = *Data->currentMV;
1357 :			}
1358 :	syskin	658
1359 :	syskin	768	if (mode_current == MODE_FORWARD) (Data->currentMV+2) = Data->currentMV;
1360 :			else (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1361 :	chl	530	}
1362 :	chl	169
1363 :	syskin	748	static void
1364 :	syskin	712	SkipDecisionB(const IMAGE * const pCur,
1365 :	syskin	767	const IMAGE * const f_Ref,
1366 :			const IMAGE * const b_Ref,
1367 :			MACROBLOCK * const pMB,
1368 :			const uint32_t x, const uint32_t y,
1369 :			const SearchData * const Data)
1370 :	syskin	712	{
1371 :	syskin	767	int dx = 0, dy = 0, b_dx = 0, b_dy = 0;
1372 :	syskin	768	int32_t sum;
1373 :	syskin	767	const int div = 1 + Data->qpel;
1374 :			int k;
1375 :	syskin	768	const uint32_t stride = Data->iEdgedWidth/2;
1376 :	syskin	712	//this is not full chroma compensation, only it's fullpel approximation. should work though
1377 :
1378 :	syskin	767	for (k = 0; k < 4; k++) {
1379 :			dy += Data->directmvF[k].y / div;
1380 :			dx += Data->directmvF[0].x / div;
1381 :			b_dy += Data->directmvB[0].y / div;
1382 :			b_dx += Data->directmvB[0].x / div;
1383 :	syskin	712	}
1384 :
1385 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1386 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1387 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1388 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1389 :
1390 :	syskin	768	sum = sad8bi(pCur->u + 8 * x + 8 * y * stride,
1391 :			f_Ref->u + (y8 + dy/2) stride + x*8 + dx/2,
1392 :			b_Ref->u + (y8 + b_dy/2) stride + x*8 + b_dx/2,
1393 :			stride);
1394 :	syskin	767
1395 :	syskin	768	if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; //no skip
1396 :	syskin	767
1397 :	syskin	768	sum += sad8bi(pCur->v + 8x + 8 y * stride,
1398 :			f_Ref->v + (y8 + dy/2) stride + x*8 + dx/2,
1399 :			b_Ref->v + (y8 + b_dy/2) stride + x*8 + b_dx/2,
1400 :			stride);
1401 :	syskin	712
1402 :	syskin	768	if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1403 :	syskin	712	}
1404 :
1405 :			static __inline uint32_t
1406 :	Isibaar	539	SearchDirect(const IMAGE * const f_Ref,
1407 :	chl	530	const uint8_t * const f_RefH,
1408 :			const uint8_t * const f_RefV,
1409 :			const uint8_t * const f_RefHV,
1410 :	Isibaar	539	const IMAGE * const b_Ref,
1411 :	chl	530	const uint8_t * const b_RefH,
1412 :			const uint8_t * const b_RefV,
1413 :			const uint8_t * const b_RefHV,
1414 :			const IMAGE * const pCur,
1415 :			const int x, const int y,
1416 :			const uint32_t MotionFlags,
1417 :			const int32_t TRB, const int32_t TRD,
1418 :			const MBParam * const pParam,
1419 :			MACROBLOCK * const pMB,
1420 :			const MACROBLOCK * const b_mb,
1421 :	h	545	int32_t * const best_sad,
1422 :			SearchData * const Data)
1423 :	edgomez	195
1424 :	chl	530	{
1425 :	h	545	int32_t skip_sad;
1426 :	syskin	768	int k = (x + Data->iEdgedWidthy) 16;
1427 :	chl	530	MainSearchFunc *MainSearchPtr;
1428 :	Isibaar	3
1429 :	h	545	Data->iMinSAD = 2564096;
1430 :	syskin	768	Data->Ref = f_Ref->y + k;
1431 :			Data->RefH = f_RefH + k;
1432 :			Data->RefV = f_RefV + k;
1433 :			Data->RefHV = f_RefHV + k;
1434 :			Data->bRef = b_Ref->y + k;
1435 :			Data->bRefH = b_RefH + k;
1436 :			Data->bRefV = b_RefV + k;
1437 :			Data->bRefHV = b_RefHV + k;
1438 :	chl	140
1439 :	syskin	768	k = Data->qpel ? 4 : 2;
1440 :			Data->max_dx = k * (pParam->width - x * 16);
1441 :			Data->max_dy = k * (pParam->height - y * 16);
1442 :			Data->min_dx = -k * (16 + x * 16);
1443 :			Data->min_dy = -k * (16 + y * 16);
1444 :	chl	140
1445 :	syskin	768	Data->referencemv = Data->qpel ? b_mb->qmvs : b_mb->mvs;
1446 :			Data->qpel_precision = 0;
1447 :	Isibaar	3
1448 :	chl	530	for (k = 0; k < 4; k++) {
1449 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1450 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1451 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1452 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1453 :	Isibaar	3
1454 :	syskin	768	if ( (pMB->b_mvs[k].x > Data->max_dx) \| (pMB->b_mvs[k].x < Data->min_dx)
1455 :			\| (pMB->b_mvs[k].y > Data->max_dy) \| (pMB->b_mvs[k].y < Data->min_dy) ) {
1456 :	Isibaar	3
1457 :	Isibaar	539	best_sad = 2564096; // in that case, we won't use direct mode
1458 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1459 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1460 :	syskin	712	return 256*4096;
1461 :	Isibaar	539	}
1462 :			if (b_mb->mode != MODE_INTER4V) {
1463 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1464 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1465 :	h	545	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1466 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1467 :	Isibaar	539	break;
1468 :			}
1469 :			}
1470 :	Isibaar	3
1471 :	syskin	768	CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v;
1472 :
1473 :	h	545	(*CheckCandidate)(0, 0, 255, &k, Data);
1474 :	Isibaar	3
1475 :	syskin	712	// initial (fast) skip decision
1476 :	syskin	768	if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH * 2) {
1477 :			//possible skip - checking chroma
1478 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data);
1479 :	syskin	712	if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1480 :	Isibaar	539	}
1481 :
1482 :	h	545	skip_sad = *Data->iMinSAD;
1483 :	edgomez	195
1484 :	syskin	768	// DIRECT MODE DELTA VECTOR SEARCH.
1485 :	chl	530	// This has to be made more effective, but at the moment I'm happy it's running at all
1486 :	Isibaar	3
1487 :	chl	530	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1488 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1489 :			else MainSearchPtr = DiamondSearch;
1490 :	edgomez	195
1491 :	h	545	(*MainSearchPtr)(0, 0, Data, 255);
1492 :	Isibaar	3
1493 :	syskin	663	SubpelRefine(Data);
1494 :	Isibaar	3
1495 :	h	545	best_sad = Data->iMinSAD;
1496 :	Isibaar	3
1497 :	syskin	768	if (Data->qpel \|\| b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT;
1498 :	syskin	672	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1499 :	edgomez	195
1500 :	h	545	pMB->pmvs[3] = *Data->currentMV;
1501 :	edgomez	195
1502 :	chl	530	for (k = 0; k < 4; k++) {
1503 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1504 :	syskin	658	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1505 :	h	545	? Data->directmvB[k].x
1506 :	syskin	658	:pMB->mvs[k].x - Data->referencemv[k].x);
1507 :	h	545	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1508 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1509 :			? Data->directmvB[k].y
1510 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1511 :	syskin	658	if (Data->qpel) {
1512 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1513 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1514 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1515 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1516 :			}
1517 :
1518 :	chl	530	if (b_mb->mode != MODE_INTER4V) {
1519 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1520 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1521 :	syskin	658	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1522 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1523 :	chl	530	break;
1524 :			}
1525 :			}
1526 :	h	545	return skip_sad;
1527 :	Isibaar	3	}
1528 :	chl	96
1529 :	syskin	712	static void
1530 :	chl	530	SearchInterpolate(const uint8_t * const f_Ref,
1531 :			const uint8_t * const f_RefH,
1532 :			const uint8_t * const f_RefV,
1533 :			const uint8_t * const f_RefHV,
1534 :			const uint8_t * const b_Ref,
1535 :			const uint8_t * const b_RefH,
1536 :			const uint8_t * const b_RefV,
1537 :			const uint8_t * const b_RefHV,
1538 :			const IMAGE * const pCur,
1539 :			const int x, const int y,
1540 :			const uint32_t fcode,
1541 :			const uint32_t bcode,
1542 :			const uint32_t MotionFlags,
1543 :			const MBParam * const pParam,
1544 :			const VECTOR * const f_predMV,
1545 :			const VECTOR * const b_predMV,
1546 :			MACROBLOCK * const pMB,
1547 :	h	545	int32_t * const best_sad,
1548 :			SearchData * const fData)
1549 :	chl	530
1550 :	chl	96	{
1551 :
1552 :	chl	530	int iDirection, i, j;
1553 :	h	545	SearchData bData;
1554 :	chl	96
1555 :	syskin	712	fData->qpel_precision = 0;
1556 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1557 :			fData->iMinSAD = 4096256;
1558 :	syskin	739	bData.currentMV++; bData.currentQMV++;
1559 :	h	545	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1560 :	edgomez	195
1561 :	syskin	712	i = (x + y * fData->iEdgedWidth) * 16;
1562 :			bData.bRef = fData->Ref = f_Ref + i;
1563 :			bData.bRefH = fData->RefH = f_RefH + i;
1564 :			bData.bRefV = fData->RefV = f_RefV + i;
1565 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1566 :			bData.Ref = fData->bRef = b_Ref + i;
1567 :			bData.RefH = fData->bRefH = b_RefH + i;
1568 :			bData.RefV = fData->bRefV = b_RefV + i;
1569 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1570 :	chl	96
1571 :	h	545	bData.bpredMV = fData->predMV = *f_predMV;
1572 :			fData->bpredMV = bData.predMV = *b_predMV;
1573 :	syskin	661	fData->currentMV[0] = fData->currentMV[2];
1574 :	chl	96
1575 :	syskin	712	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0);
1576 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0);
1577 :
1578 :	h	545	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1579 :	syskin	675	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1580 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1581 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1582 :	Isibaar	539
1583 :	h	545	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1584 :	syskin	675	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1585 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1586 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1587 :	Isibaar	539
1588 :	h	545	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1589 :	chl	96
1590 :	syskin	768	//diamond
1591 :	chl	530	do {
1592 :			iDirection = 255;
1593 :			// forward MV moves
1594 :	h	545	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1595 :	chl	96
1596 :	h	545	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1597 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1598 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1599 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1600 :	chl	96
1601 :	chl	530	// backward MV moves
1602 :	h	545	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1603 :			fData->currentMV[2] = fData->currentMV[0];
1604 :	Isibaar	539	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1605 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1606 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1607 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1608 :	chl	96
1609 :	chl	530	} while (!(iDirection));
1610 :	edgomez	195
1611 :	syskin	768	//qpel refinement
1612 :	syskin	658	if (fData->qpel) {
1613 :	syskin	740	if (fData->iMinSAD > best_sad + 500) return;
1614 :	syskin	675	CheckCandidate = CheckCandidateInt;
1615 :	syskin	663	fData->qpel_precision = bData.qpel_precision = 1;
1616 :	syskin	712	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0);
1617 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1618 :	syskin	658	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1619 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1620 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1621 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1622 :	syskin	663	SubpelRefine(fData);
1623 :	syskin	740	if (fData->iMinSAD > best_sad + 300) return;
1624 :	syskin	658	fData->currentQMV[2] = fData->currentQMV[0];
1625 :	syskin	663	SubpelRefine(&bData);
1626 :	syskin	658	}
1627 :
1628 :	syskin	768	fData->iMinSAD += (2+3) fData->lambda16; // two bits are needed to code interpolate mode.
1629 :
1630 :	h	545	if (fData->iMinSAD < best_sad) {
1631 :			best_sad = fData->iMinSAD;
1632 :			pMB->mvs[0] = fData->currentMV[0];
1633 :			pMB->b_mvs[0] = fData->currentMV[1];
1634 :	chl	530	pMB->mode = MODE_INTERPOLATE;
1635 :	syskin	658	if (fData->qpel) {
1636 :			pMB->qmvs[0] = fData->currentQMV[0];
1637 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1638 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1639 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1640 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1641 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1642 :			} else {
1643 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1644 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1645 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1646 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1647 :			}
1648 :	chl	530	}
1649 :			}
1650 :	chl	96
1651 :	chl	530	void
1652 :			MotionEstimationBVOP(MBParam * const pParam,
1653 :			FRAMEINFO * const frame,
1654 :			const int32_t time_bp,
1655 :			const int32_t time_pp,
1656 :			// forward (past) reference
1657 :			const MACROBLOCK * const f_mbs,
1658 :			const IMAGE * const f_ref,
1659 :			const IMAGE * const f_refH,
1660 :			const IMAGE * const f_refV,
1661 :			const IMAGE * const f_refHV,
1662 :			// backward (future) reference
1663 :	syskin	644	const FRAMEINFO * const b_reference,
1664 :	chl	530	const IMAGE * const b_ref,
1665 :			const IMAGE * const b_refH,
1666 :			const IMAGE * const b_refV,
1667 :			const IMAGE * const b_refHV)
1668 :			{
1669 :			uint32_t i, j;
1670 :	syskin	712	int32_t best_sad;
1671 :			uint32_t skip_sad;
1672 :	chl	530	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1673 :	syskin	644	const MACROBLOCK * const b_mbs = b_reference->mbs;
1674 :	chl	96
1675 :	chl	530	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1676 :	chl	96
1677 :	chl	530	const int32_t TRB = time_pp - time_bp;
1678 :			const int32_t TRD = time_pp;
1679 :	chl	96
1680 :	h	545	// some pre-inintialized data for the rest of the search
1681 :
1682 :			SearchData Data;
1683 :			int32_t iMinSAD;
1684 :			VECTOR currentMV[3];
1685 :	syskin	658	VECTOR currentQMV[3];
1686 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
1687 :	h	545	Data.iEdgedWidth = pParam->edged_width;
1688 :	syskin	658	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1689 :	h	545	Data.iMinSAD = &iMinSAD;
1690 :	syskin	712	Data.lambda16 = lambda_vec16[frame->quant];
1691 :			Data.chroma = frame->quant;
1692 :	syskin	658	Data.qpel = pParam->m_quarterpel;
1693 :	syskin	663	Data.rounding = 0;
1694 :	h	545
1695 :	syskin	754	Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose)
1696 :	chl	530	// note: i==horizontal, j==vertical
1697 :			for (j = 0; j < pParam->mb_height; j++) {
1698 :	chl	96
1699 :	chl	530	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1700 :	edgomez	195
1701 :	chl	530	for (i = 0; i < pParam->mb_width; i++) {
1702 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1703 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1704 :	edgomez	195
1705 :	chl	619	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1706 :	syskin	644	if (b_reference->coding_type != S_VOP)
1707 :			if (b_mb->mode == MODE_NOT_CODED) {
1708 :			pMB->mode = MODE_NOT_CODED;
1709 :			continue;
1710 :			}
1711 :	chl	96
1712 :	h	545	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1713 :	syskin	601	pMB->quant = frame->quant;
1714 :	syskin	628
1715 :	chl	530	/* direct search comes first, because it (1) checks for SKIP-mode
1716 :			and (2) sets very good predictions for forward and backward search */
1717 :	Isibaar	539	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1718 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1719 :	chl	530	&frame->image,
1720 :			i, j,
1721 :			frame->motion_flags,
1722 :			TRB, TRD,
1723 :			pParam,
1724 :			pMB, b_mb,
1725 :	h	545	&best_sad,
1726 :			&Data);
1727 :	chl	96
1728 :	Isibaar	539	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1729 :	syskin	739
1730 :	chl	530	// forward search
1731 :			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1732 :			&frame->image, i, j,
1733 :			frame->motion_flags,
1734 :	h	545	frame->fcode, pParam,
1735 :	chl	530	pMB, &f_predMV, &best_sad,
1736 :	h	545	MODE_FORWARD, &Data);
1737 :	chl	96
1738 :	chl	530	// backward search
1739 :			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1740 :			&frame->image, i, j,
1741 :			frame->motion_flags,
1742 :	h	545	frame->bcode, pParam,
1743 :	chl	530	pMB, &b_predMV, &best_sad,
1744 :	h	545	MODE_BACKWARD, &Data);
1745 :	chl	96
1746 :	chl	530	// interpolate search comes last, because it uses data from forward and backward as prediction
1747 :			SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1748 :			b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1749 :			&frame->image,
1750 :			i, j,
1751 :			frame->fcode, frame->bcode,
1752 :			frame->motion_flags,
1753 :	h	545	pParam,
1754 :	chl	530	&f_predMV, &b_predMV,
1755 :	h	545	pMB, &best_sad,
1756 :			&Data);
1757 :	syskin	662
1758 :	syskin	712	// final skip decision
1759 :			if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2)
1760 :			&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1761 :	syskin	767	SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data);
1762 :	syskin	712
1763 :	chl	530	switch (pMB->mode) {
1764 :			case MODE_FORWARD:
1765 :			f_count++;
1766 :	syskin	768	f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
1767 :	chl	530	break;
1768 :			case MODE_BACKWARD:
1769 :			b_count++;
1770 :	syskin	768	b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
1771 :	chl	530	break;
1772 :			case MODE_INTERPOLATE:
1773 :			i_count++;
1774 :	syskin	768	f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
1775 :			b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
1776 :	chl	530	break;
1777 :			case MODE_DIRECT:
1778 :			case MODE_DIRECT_NO4V:
1779 :			d_count++;
1780 :			default:
1781 :			break;
1782 :	chl	96	}
1783 :			}
1784 :			}
1785 :			}
1786 :
1787 :	syskin	739	static __inline void
1788 :	Isibaar	539	MEanalyzeMB ( const uint8_t * const pRef,
1789 :			const uint8_t * const pCur,
1790 :			const int x,
1791 :			const int y,
1792 :			const MBParam * const pParam,
1793 :	syskin	739	MACROBLOCK * const pMBs,
1794 :	h	545	SearchData * const Data)
1795 :	Isibaar	539	{
1796 :	chl	317
1797 :	syskin	739	int i, mask;
1798 :	h	545	VECTOR pmv[3];
1799 :	syskin	739	MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width];
1800 :	syskin	768
1801 :	syskin	739	for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1802 :	suxen_drol	118
1803 :	syskin	640	//median is only used as prediction. it doesn't have to be real
1804 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1805 :			else
1806 :			if (x == 1) //left macroblock does not have any vector now
1807 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1808 :	syskin	748	else if (y == 1) // top macroblock doesn't have it's vector
1809 :	syskin	640	Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1810 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1811 :
1812 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1813 :	syskin	739	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv);
1814 :	chl	326
1815 :	h	545	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1816 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1817 :	syskin	739
1818 :	Isibaar	539	pmv[1].x = EVEN(pMB->mvs[0].x);
1819 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1820 :	syskin	640	pmv[2].x = EVEN(Data->predMV.x);
1821 :			pmv[2].y = EVEN(Data->predMV.y);
1822 :			pmv[0].x = pmv[0].y = 0;
1823 :	chl	326
1824 :	syskin	739	CheckCandidate32I(0, 0, 255, &i, Data);
1825 :	syskin	640
1826 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) {
1827 :	syskin	640
1828 :	syskin	739	if (!(mask = make_mask(pmv, 1)))
1829 :			CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data);
1830 :			if (!(mask = make_mask(pmv, 2)))
1831 :			CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data);
1832 :	Isibaar	539
1833 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) // diamond only if needed
1834 :			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1835 :	Isibaar	539
1836 :	syskin	739	for (i = 0; i < 4; i++) {
1837 :	syskin	768	MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width];
1838 :	syskin	739	MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
1839 :			MB->mode = MODE_INTER;
1840 :			MB->sad16 = Data->iMinSAD[i+1];
1841 :			}
1842 :			}
1843 :	Isibaar	539	}
1844 :
1845 :	syskin	739	#define INTRA_BIAS 2500
1846 :			#define INTRA_THRESH 1500
1847 :			#define INTER_THRESH 1400
1848 :	Isibaar	539
1849 :			int
1850 :	syskin	768	MEanalysis( const IMAGE * const pRef,
1851 :	syskin	644	FRAMEINFO * const Current,
1852 :	Isibaar	539	MBParam * const pParam,
1853 :	syskin	644	int maxIntra, //maximum number if non-I frames
1854 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1855 :	syskin	768	int bCount) // number of B frames in a row
1856 :	Isibaar	539	{
1857 :			uint32_t x, y, intra = 0;
1858 :			int sSAD = 0;
1859 :	syskin	644	MACROBLOCK * const pMBs = Current->mbs;
1860 :			const IMAGE * const pCurrent = &Current->image;
1861 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1862 :	h	545
1863 :	syskin	739	int32_t iMinSAD[5], temp[5];
1864 :			VECTOR currentMV[5];
1865 :	h	545	SearchData Data;
1866 :			Data.iEdgedWidth = pParam->edged_width;
1867 :	syskin	739	Data.currentMV = currentMV;
1868 :			Data.iMinSAD = iMinSAD;
1869 :	syskin	644	Data.iFcode = Current->fcode;
1870 :	syskin	739	Data.rrv = Current->global_flags & XVID_REDUCED;
1871 :			Data.temp = temp;
1872 :			CheckCandidate = CheckCandidate32I;
1873 :	h	545
1874 :	syskin	768	if (intraCount != 0 && intraCount < 10) // we're right after an I frame
1875 :	syskin	658	IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
1876 :	syskin	644	else
1877 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
1878 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
1879 :
1880 :	syskin	672	InterThresh += 400 * (1 - bCount);
1881 :	syskin	739	if (InterThresh < 300) InterThresh = 300;
1882 :	syskin	644
1883 :	Isibaar	539	if (sadInit) (*sadInit) ();
1884 :
1885 :	syskin	768	for (y = 1; y < pParam->mb_height-1; y += 2) {
1886 :			for (x = 1; x < pParam->mb_width-1; x += 2) {
1887 :	syskin	739	int i;
1888 :	Isibaar	539
1889 :	syskin	739	if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
1890 :
1891 :			MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
1892 :	syskin	768
1893 :	syskin	739	for (i = 0; i < 4; i++) {
1894 :			int dev;
1895 :	syskin	768	MACROBLOCK pMB = &pMBs[x+(i&1) + (y+(i>>1)) pParam->mb_width];
1896 :	syskin	739	if (pMB->sad16 > IntraThresh) {
1897 :	syskin	768	dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16,
1898 :			pParam->edged_width);
1899 :	syskin	739	if (dev + IntraThresh < pMB->sad16) {
1900 :			pMB->mode = MODE_INTRA;
1901 :			if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
1902 :			}
1903 :	Isibaar	539	}
1904 :	syskin	739	sSAD += pMB->sad16;
1905 :	syskin	644	}
1906 :	suxen_drol	118	}
1907 :			}
1908 :	syskin	704	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
1909 :	syskin	768	if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP;
1910 :	suxen_drol	709	if (sSAD > InterThresh ) return P_VOP;
1911 :	Isibaar	539	emms();
1912 :	suxen_drol	709	return B_VOP;
1913 :	syskin	712
1914 :	suxen_drol	118	}
1915 :	syskin	576
1916 :	syskin	628	static void
1917 :	syskin	748	CheckGMC(int x, int y, const int dir, int * iDirection,
1918 :	syskin	628	const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
1919 :			const MBParam * const pParam)
1920 :			{
1921 :			uint32_t mx, my, a, count = 0;
1922 :	chl	619
1923 :	syskin	628	for (my = 1; my < pParam->mb_height-1; my++)
1924 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
1925 :			VECTOR mv;
1926 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
1927 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED) continue;
1928 :			mv = pMB->mvs[0];
1929 :			a = ABS(mv.x - x) + ABS(mv.y - y);
1930 :			if (a < 6) count += 6 - a;
1931 :			}
1932 :	chl	619
1933 :	syskin	628	if (count > *bestcount) {
1934 :			*bestcount = count;
1935 :			*iDirection = dir;
1936 :	syskin	748	GMC->x = x; GMC->y = y;
1937 :	syskin	628	}
1938 :			}
1939 :
1940 :
1941 :	suxen_drol	721	static VECTOR
1942 :	chl	619	GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
1943 :			{
1944 :
1945 :	syskin	628	uint32_t count, bestcount = 0;
1946 :	chl	619	int x, y;
1947 :	syskin	628	VECTOR gmc = {0,0};
1948 :	chl	619	int step, min_x, max_x, min_y, max_y;
1949 :			uint32_t mx, my;
1950 :	syskin	628	int iDirection, bDirection;
1951 :	chl	619
1952 :			min_x = min_y = -32<<iFcode;
1953 :			max_x = max_y = 32<<iFcode;
1954 :
1955 :	syskin	628	//step1: let's find a rough camera panning
1956 :	chl	619	for (step = 32; step >= 2; step /= 2) {
1957 :			bestcount = 0;
1958 :			for (y = min_y; y <= max_y; y += step)
1959 :			for (x = min_x ; x <= max_x; x += step) {
1960 :			count = 0;
1961 :			//for all macroblocks
1962 :			for (my = 1; my < pParam->mb_height-1; my++)
1963 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
1964 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
1965 :			VECTOR mv;
1966 :	syskin	768
1967 :	syskin	748	if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
1968 :	chl	619	continue;
1969 :	syskin	768
1970 :	chl	619	mv = pMB->mvs[0];
1971 :			if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
1972 :			count++;
1973 :			}
1974 :			if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
1975 :			}
1976 :			min_x = gmc.x - step;
1977 :			max_x = gmc.x + step;
1978 :			min_y = gmc.y - step;
1979 :			max_y = gmc.y + step;
1980 :			}
1981 :	syskin	768
1982 :	syskin	748	if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
1983 :	syskin	628	gmc.x = gmc.y = 0; //no camara pan, no GMC
1984 :
1985 :	syskin	748	// step2: let's refine camera panning using gradiend-descent approach
1986 :	syskin	628	// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
1987 :			bestcount = 0;
1988 :			CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
1989 :			do {
1990 :			x = gmc.x; y = gmc.y;
1991 :			bDirection = iDirection; iDirection = 0;
1992 :			if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
1993 :			if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
1994 :			if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
1995 :			if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
1996 :
1997 :			} while (iDirection);
1998 :
1999 :	chl	619	if (pParam->m_quarterpel) {
2000 :			gmc.x *= 2;
2001 :			gmc.y = 2; / we store the halfpel value as pseudo-qpel to make comparison easier */
2002 :			}
2003 :
2004 :			return gmc;
2005 :			}

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