Annotation of /xvidcore/src/motion/motion_est.c

Revision 1.44.2.14 - (view) (download)

1 :	Isibaar	1.1	/**************************************************************************
2 :			*
3 :	chl	1.26	* XVID MPEG-4 VIDEO CODEC
4 :			* motion estimation
5 :			*
6 :			* This program is an implementation of a part of one or more MPEG-4
7 :			* Video tools as specified in ISO/IEC 14496-2 standard. Those intending
8 :			* to use this software module in hardware or software products are
9 :			* advised that its use may infringe existing patents or copyrights, and
10 :			* any such use would be at such party's own risk. The original
11 :			* developer of this software module and his/her company, and subsequent
12 :			* editors and their companies, will have no liability for use of this
13 :			* software or modifications or derivatives thereof.
14 :			*
15 :			* This program is free software; you can redistribute it and/or modify
16 :			* it under the terms of the GNU General Public License as published by
17 :			* the Free Software Foundation; either version 2 of the License, or
18 :			* (at your option) any later version.
19 :			*
20 :			* This program is distributed in the hope that it will be useful,
21 :			* but WITHOUT ANY WARRANTY; without even the implied warranty of
22 :			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 :			* GNU General Public License for more details.
24 :			*
25 :			* You should have received a copy of the GNU General Public License
26 :			* along with this program; if not, write to the Free Software
27 :			* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 :			*
29 :			*************************************************************************/
30 :
31 :	Isibaar	1.1	#include <assert.h>
32 :			#include <stdio.h>
33 :	chl	1.3	#include <stdlib.h>
34 :	Isibaar	1.1
35 :			#include "../encoder.h"
36 :			#include "../utils/mbfunctions.h"
37 :			#include "../prediction/mbprediction.h"
38 :			#include "../global.h"
39 :			#include "../utils/timer.h"
40 :	Isibaar	1.44.2.7	#include "../image/interpolate8x8.h"
41 :	chl	1.44.2.1	#include "motion_est.h"
42 :	suxen_drol	1.6	#include "motion.h"
43 :	Isibaar	1.1	#include "sad.h"
44 :	Isibaar	1.44.2.2	#include "../utils/emms.h"
45 :	Isibaar	1.1
46 :	chl	1.44.2.1	#define INITIAL_SKIP_THRESH (10)
47 :			#define FINAL_SKIP_THRESH (50)
48 :			#define MAX_SAD00_FOR_SKIP (20)
49 :	Isibaar	1.44.2.2	#define MAX_CHROMA_SAD_FOR_SKIP (22)
50 :			#define SKIP_THRESH_B (25)
51 :	chl	1.44.2.1
52 :			#define CHECK_CANDIDATE(X,Y,D) { \
53 :			(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
54 :	edgomez	1.20
55 :	Isibaar	1.44.2.7	#define GET_REFERENCE(X, Y, REF) { \
56 :	chl	1.44.2.9	switch ( (((X)&1)<<1) + ((Y)&1) ) \
57 :	Isibaar	1.44.2.7	{ \
58 :	syskin	1.44.2.11	case 0 : REF = (uint8_t )data->Ref + (X)/2 + ((Y)/2)(data->iEdgedWidth); break; \
59 :			case 1 : REF = (uint8_t )data->RefV + (X)/2 + (((Y)-1)/2)(data->iEdgedWidth); break; \
60 :			case 2 : REF = (uint8_t )data->RefH + ((X)-1)/2 + ((Y)/2)(data->iEdgedWidth); break; \
61 :			default : REF = (uint8_t )data->RefHV + ((X)-1)/2 + (((Y)-1)/2)(data->iEdgedWidth); break; \
62 :	Isibaar	1.44.2.7	} \
63 :			}
64 :
65 :	chl	1.44.2.1	#define iDiamondSize 2
66 :	edgomez	1.20
67 :	chl	1.44.2.1	static __inline int
68 :			d_mv_bits(int x, int y, const uint32_t iFcode)
69 :			{
70 :			int xb, yb;
71 :
72 :			if (x == 0) xb = 1;
73 :			else {
74 :			if (x < 0) x = -x;
75 :			x += (1 << (iFcode - 1)) - 1;
76 :			x >>= (iFcode - 1);
77 :			if (x > 32) x = 32;
78 :			xb = mvtab[x] + iFcode;
79 :			}
80 :
81 :			if (y == 0) yb = 1;
82 :			else {
83 :			if (y < 0) y = -y;
84 :			y += (1 << (iFcode - 1)) - 1;
85 :			y >>= (iFcode - 1);
86 :			if (y > 32) y = 32;
87 :			yb = mvtab[y] + iFcode;
88 :			}
89 :			return xb + yb;
90 :			}
91 :
92 :	syskin	1.44.2.14	static int32_t
93 :			ChromaSAD(int dx, int dy, const SearchData * const data)
94 :	chl	1.44.2.1	{
95 :	syskin	1.44.2.14	int sad;
96 :			dx = (dx >> 1) + roundtab_79[dx & 0x3];
97 :			dy = (dy >> 1) + roundtab_79[dy & 0x3];
98 :
99 :			switch (((dx & 1) << 1) + (dy & 1)) { // ((dx%2)?2:0)+((dy%2)?1:0)
100 :			case 0:
101 :			sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
102 :			sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
103 :			break;
104 :			case 1:
105 :			dx = dx / 2; dy = (dy - 1) / 2;
106 :			sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
107 :			sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
108 :			break;
109 :			case 2:
110 :			dx = (dx - 1) / 2; dy = dy / 2;
111 :			sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
112 :			sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
113 :			break;
114 :			default:
115 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
116 :			interpolate8x8_halfpel_hv(data->RefQ,
117 :			data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
118 :			data->rounding);
119 :			sad = sad8(data->CurU, data->RefQ, data->iEdgedWidth/2);
120 :			interpolate8x8_halfpel_hv(data->RefQ,
121 :			data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
122 :			data->rounding);
123 :			sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2);
124 :			break;
125 :	chl	1.44.2.1	}
126 :	syskin	1.44.2.14	return sad;
127 :			}
128 :	chl	1.10
129 :
130 :	syskin	1.44.2.14	/* CHECK_CANDIATE FUNCTIONS START */
131 :	Isibaar	1.1
132 :
133 :	chl	1.44.2.1	static void
134 :	syskin	1.44.2.14	CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
135 :	Isibaar	1.44.2.12	{
136 :			int t;
137 :			const uint8_t * Reference;
138 :
139 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
140 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
141 :
142 :			switch ( ((x&1)<<1) + (y&1) ) {
143 :			case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
144 :			case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
145 :			case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
146 :			default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
147 :			}
148 :
149 :			data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
150 :
151 :	syskin	1.44.2.14	if (data->qpel) t = d_mv_bits(2x - data->predQMV.x, 2y - data->predQMV.y, data->iFcode);
152 :			else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
153 :
154 :			data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
155 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
156 :
157 :			if (data->chroma) data->temp[0] += ChromaSAD(x, y, data);
158 :	Isibaar	1.44.2.12
159 :			if (data->temp[0] < data->iMinSAD[0]) {
160 :			data->iMinSAD[0] = data->temp[0];
161 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
162 :			*dir = Direction; }
163 :
164 :			if (data->temp[1] < data->iMinSAD[1]) {
165 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
166 :			if (data->temp[2] < data->iMinSAD[2]) {
167 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
168 :			if (data->temp[3] < data->iMinSAD[3]) {
169 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
170 :			if (data->temp[4] < data->iMinSAD[4]) {
171 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
172 :
173 :			}
174 :
175 :			static void
176 :	chl	1.44.2.1	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
177 :	Isibaar	1.1	{
178 :	chl	1.44.2.1	int32_t sad;
179 :			const uint8_t * Reference;
180 :	Isibaar	1.1
181 :	chl	1.44.2.1	if (( x > data->max_dx) \|\| ( x < data->min_dx)
182 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
183 :	Isibaar	1.1
184 :	chl	1.44.2.1	switch ( ((x&1)<<1) + (y&1) )
185 :			{
186 :			case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
187 :			case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
188 :			case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
189 :			default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
190 :	edgomez	1.20	}
191 :	Isibaar	1.1
192 :	syskin	1.44.2.14	sad = sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR);
193 :			sad += (data->lambda16 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode) * sad)/1000;
194 :	Isibaar	1.1
195 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
196 :			*(data->iMinSAD) = sad;
197 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
198 :			*dir = Direction; }
199 :	Isibaar	1.1	}
200 :
201 :	chl	1.44.2.1	static void
202 :	Isibaar	1.44.2.7	CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
203 :
204 :			// CheckCandidate16 variant which expects x and y in quarter pixel resolution
205 :			// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
206 :			// around currentMV!
207 :			{
208 :			int t;
209 :	syskin	1.44.2.11	uint8_t * Reference = (uint8_t *)data->RefQ;
210 :	Isibaar	1.44.2.7	const uint8_t ref1, ref2, ref3, ref4;
211 :			VECTOR halfpelMV = *(data->currentMV);
212 :
213 :			int32_t iEdgedWidth = data->iEdgedWidth;
214 :			uint32_t rounding = data->rounding;
215 :
216 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
217 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
218 :
219 :	Isibaar	1.44.2.12	GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); // this refenrence is used in all cases
220 :	Isibaar	1.44.2.7	switch( ((x&1)<<1) + (y&1) )
221 :			{
222 :			case 0: // pure halfpel position - shouldn't happen during a refinement step
223 :	syskin	1.44.2.11	GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference);
224 :	Isibaar	1.44.2.7	break;
225 :
226 :			case 1: // x halfpel, y qpel - top or bottom during qpel refinement
227 :			GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
228 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
229 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
230 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
231 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
232 :			break;
233 :
234 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
235 :			GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
236 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
237 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
238 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
239 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
240 :			break;
241 :
242 :			default: // x and y in qpel resolution - the "corners" (top left/right and
243 :			// bottom left/right) during qpel refinement
244 :			GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
245 :			GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
246 :			GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
247 :
248 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
249 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
250 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
251 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
252 :			break;
253 :			}
254 :
255 :	syskin	1.44.2.8	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1);
256 :	Isibaar	1.44.2.7
257 :			t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
258 :	syskin	1.44.2.14	data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
259 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
260 :
261 :			if (data->chroma)
262 :			data->temp[0] += ChromaSAD(x/2, y/2, data);
263 :	Isibaar	1.44.2.7
264 :			if (data->temp[0] < data->iMinSAD[0]) {
265 :			data->iMinSAD[0] = data->temp[0];
266 :			data->currentQMV[0].x = x; data->currentQMV[0].y = y;
267 :	syskin	1.44.2.8	/* dir = Direction;/ }
268 :	Isibaar	1.44.2.7
269 :			if (data->temp[1] < data->iMinSAD[1]) {
270 :			data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; }
271 :			if (data->temp[2] < data->iMinSAD[2]) {
272 :			data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; }
273 :			if (data->temp[3] < data->iMinSAD[3]) {
274 :			data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; }
275 :			if (data->temp[4] < data->iMinSAD[4]) {
276 :			data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; }
277 :			}
278 :
279 :			static void
280 :	Isibaar	1.44.2.2	CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
281 :			{
282 :			int32_t sad;
283 :
284 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
285 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
286 :
287 :	syskin	1.44.2.11	sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
288 :	Isibaar	1.44.2.2	data->iEdgedWidth, 256*4096);
289 :
290 :			if (sad < *(data->iMinSAD)) {
291 :			*(data->iMinSAD) = sad;
292 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
293 :			*dir = Direction; }
294 :			}
295 :
296 :
297 :			static void
298 :	chl	1.44.2.1	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
299 :	Isibaar	1.1	{
300 :	chl	1.44.2.1	int32_t sad;
301 :			const int xb = data->currentMV[1].x;
302 :			const int yb = data->currentMV[1].y;
303 :			const uint8_t ReferenceF, ReferenceB;
304 :	suxen_drol	1.8
305 :	chl	1.44.2.1	if (( xf > data->max_dx) \|\| ( xf < data->min_dx)
306 :			\|\| ( yf > data->max_dy) \|\| (yf < data->min_dy)) return;
307 :	suxen_drol	1.8
308 :	chl	1.44.2.1	switch ( ((xf&1)<<1) + (yf&1) ) {
309 :			case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break;
310 :			case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
311 :			case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break;
312 :			default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
313 :			}
314 :	chl	1.19
315 :	chl	1.44.2.1	switch ( ((xb&1)<<1) + (yb&1) ) {
316 :			case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break;
317 :			case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
318 :			case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break;
319 :			default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
320 :			}
321 :	Isibaar	1.1
322 :	syskin	1.44.2.14	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
323 :	chl	1.19
324 :	syskin	1.44.2.14	sad += (data->lambda16 *
325 :			( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) +
326 :			d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode)) * sad)/1000;
327 :	edgomez	1.20
328 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
329 :			*(data->iMinSAD) = sad;
330 :			data->currentMV->x = xf; data->currentMV->y = yf;
331 :			*dir = Direction; }
332 :			}
333 :	edgomez	1.20
334 :	chl	1.44.2.1	static void
335 :			CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
336 :			{
337 :	syskin	1.44.2.14	int32_t sad = 0;
338 :	chl	1.44.2.1	int k;
339 :			const uint8_t *ReferenceF;
340 :			const uint8_t *ReferenceB;
341 :			VECTOR mvs, b_mvs;
342 :	edgomez	1.20
343 :	chl	1.44.2.1	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
344 :	edgomez	1.20
345 :	chl	1.44.2.1	for (k = 0; k < 4; k++) {
346 :			mvs.x = data->directmvF[k].x + x;
347 :			b_mvs.x = ((x == 0) ?
348 :			data->directmvB[k].x
349 :			: mvs.x - data->referencemv[k].x);
350 :	edgomez	1.20
351 :	chl	1.44.2.1	mvs.y = data->directmvF[k].y + y;
352 :			b_mvs.y = ((y == 0) ?
353 :			data->directmvB[k].y
354 :			: mvs.y - data->referencemv[k].y);
355 :
356 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
357 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
358 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
359 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
360 :
361 :			switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
362 :			case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
363 :			case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
364 :			case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
365 :			default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
366 :			}
367 :
368 :			switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
369 :			case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
370 :			case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
371 :			case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
372 :			default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
373 :			}
374 :
375 :			sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
376 :			ReferenceF + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
377 :			ReferenceB + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
378 :			data->iEdgedWidth);
379 :			if (sad > *(data->iMinSAD)) return;
380 :			}
381 :	chl	1.17
382 :	syskin	1.44.2.14	sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000;
383 :
384 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
385 :			*(data->iMinSAD) = sad;
386 :			data->currentMV->x = x; data->currentMV->y = y;
387 :			*dir = Direction; }
388 :	Isibaar	1.1	}
389 :
390 :	chl	1.44.2.1	static void
391 :			CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
392 :			{
393 :			int32_t sad;
394 :			const uint8_t *ReferenceF;
395 :			const uint8_t *ReferenceB;
396 :			VECTOR mvs, b_mvs;
397 :	chl	1.35
398 :	syskin	1.44.2.6	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
399 :	Isibaar	1.1
400 :	chl	1.44.2.1	mvs.x = data->directmvF[0].x + x;
401 :			b_mvs.x = ((x == 0) ?
402 :			data->directmvB[0].x
403 :			: mvs.x - data->referencemv[0].x);
404 :
405 :			mvs.y = data->directmvF[0].y + y;
406 :			b_mvs.y = ((y == 0) ?
407 :			data->directmvB[0].y
408 :			: mvs.y - data->referencemv[0].y);
409 :
410 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
411 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
412 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
413 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
414 :
415 :			switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
416 :			case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
417 :			case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
418 :			case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
419 :			default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
420 :			}
421 :
422 :			switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
423 :			case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
424 :			case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
425 :			case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
426 :			default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
427 :			}
428 :
429 :	syskin	1.44.2.14	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
430 :			sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000;
431 :	Isibaar	1.1
432 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
433 :			*(data->iMinSAD) = sad;
434 :			data->currentMV->x = x; data->currentMV->y = y;
435 :			*dir = Direction; }
436 :			}
437 :	Isibaar	1.1
438 :	chl	1.44.2.1	static void
439 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
440 :			{
441 :	syskin	1.44.2.14	int32_t sad; int t;
442 :	chl	1.44.2.1	const uint8_t * Reference;
443 :	Isibaar	1.1
444 :	chl	1.44.2.1	if (( x > data->max_dx) \|\| ( x < data->min_dx)
445 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
446 :	Isibaar	1.1
447 :	chl	1.44.2.1	switch ( ((x&1)<<1) + (y&1) )
448 :			{
449 :			case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
450 :			case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
451 :			case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
452 :			default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
453 :	edgomez	1.2	}
454 :	Isibaar	1.1
455 :	chl	1.44.2.1	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
456 :	syskin	1.44.2.14	if (data->qpel) t = d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode);
457 :			else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
458 :	Isibaar	1.44.2.12
459 :	syskin	1.44.2.14	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100;
460 :	Isibaar	1.44.2.12
461 :			if (sad < *(data->iMinSAD)) {
462 :			*(data->iMinSAD) = sad;
463 :			data->currentMV->x = x; data->currentMV->y = y;
464 :			*dir = Direction; }
465 :			}
466 :
467 :			static void
468 :	Isibaar	1.44.2.7	CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
469 :			// CheckCandidate16no4v variant which expects x and y in quarter pixel resolution
470 :			// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
471 :			// around currentMV!
472 :
473 :			{
474 :			int32_t sad;
475 :			uint8_t Reference = (uint8_t ) data->RefQ;
476 :			const uint8_t ref1, ref2, ref3, ref4;
477 :			VECTOR halfpelMV = *(data->currentMV);
478 :
479 :			int32_t iEdgedWidth = data->iEdgedWidth;
480 :			uint32_t rounding = data->rounding;
481 :
482 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
483 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
484 :
485 :	Isibaar	1.44.2.12	GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
486 :	Isibaar	1.44.2.7	switch( ((x&1)<<1) + (y&1) )
487 :			{
488 :			case 0: // pure halfpel position - shouldn't happen during a refinement step
489 :	syskin	1.44.2.11	GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference);
490 :	Isibaar	1.44.2.7	break;
491 :
492 :			case 1: // x halfpel, y qpel - top or bottom during qpel refinement
493 :			GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
494 :
495 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
496 :			break;
497 :
498 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
499 :			GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
500 :
501 :			interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
502 :			break;
503 :
504 :			default: // x and y in qpel resolution - the "corners" (top left/right and
505 :			// bottom left/right) during qpel refinement
506 :			GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
507 :			GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
508 :			GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
509 :
510 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
511 :			break;
512 :			}
513 :
514 :			sad = sad8(data->Cur, Reference, data->iEdgedWidth);
515 :	syskin	1.44.2.14	sad += (data->lambda8 * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode) * (sad+NEIGH_8X8_BIAS))/100;
516 :	Isibaar	1.44.2.7
517 :			if (sad < *(data->iMinSAD)) {
518 :			*(data->iMinSAD) = sad;
519 :			data->currentQMV->x = x; data->currentQMV->y = y;
520 :			*dir = Direction; }
521 :			}
522 :
523 :	Isibaar	1.44.2.2	/* CHECK_CANDIATE FUNCTIONS END */
524 :	chl	1.3
525 :	chl	1.44.2.1	/* MAINSEARCH FUNCTIONS START */
526 :	chl	1.18
527 :	chl	1.44.2.1	static void
528 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
529 :			{
530 :	chl	1.18
531 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
532 :
533 :	chl	1.44.2.1	int iDirection;
534 :	edgomez	1.20
535 :			do {
536 :	chl	1.18	iDirection = 0;
537 :	chl	1.44.2.1	if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
538 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
539 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
540 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
541 :	chl	1.18
542 :			/* now we're doing diagonal checks near our candidate */
543 :
544 :	chl	1.44.2.1	if (iDirection) { //checking if anything found
545 :	chl	1.18	bDirection = iDirection;
546 :			iDirection = 0;
547 :	chl	1.44.2.1	x = data->currentMV->x; y = data->currentMV->y;
548 :			if (bDirection & 3) { //our candidate is left or right
549 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
550 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
551 :			} else { // what remains here is up or down
552 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
553 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
554 :	chl	1.18
555 :	edgomez	1.20	if (iDirection) {
556 :			bDirection += iDirection;
557 :	chl	1.44.2.1	x = data->currentMV->x; y = data->currentMV->y; }
558 :			} else { //about to quit, eh? not so fast....
559 :	edgomez	1.20	switch (bDirection) {
560 :	chl	1.18	case 2:
561 :	chl	1.44.2.1	CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
562 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
563 :	chl	1.18	break;
564 :			case 1:
565 :	chl	1.44.2.1	CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
566 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
567 :	edgomez	1.20	break;
568 :			case 2 + 4:
569 :	chl	1.44.2.1	CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
570 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
571 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
572 :	chl	1.18	break;
573 :			case 4:
574 :	chl	1.44.2.1	CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
575 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
576 :	chl	1.18	break;
577 :			case 8:
578 :	chl	1.44.2.1	CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
579 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
580 :	edgomez	1.20	break;
581 :			case 1 + 4:
582 :	chl	1.44.2.1	CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
583 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
584 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
585 :	edgomez	1.20	break;
586 :			case 2 + 8:
587 :	chl	1.44.2.1	CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
588 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
589 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
590 :	edgomez	1.20	break;
591 :			case 1 + 8:
592 :	chl	1.44.2.1	CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
593 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
594 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
595 :	edgomez	1.20	break;
596 :			default: //1+2+4+8 == we didn't find anything at all
597 :	chl	1.44.2.1	CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
598 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
599 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
600 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
601 :	chl	1.18	break;
602 :			}
603 :	chl	1.44.2.1	if (!iDirection) break; //ok, the end. really
604 :			bDirection = iDirection;
605 :			x = data->currentMV->x; y = data->currentMV->y;
606 :	chl	1.18	}
607 :			}
608 :	edgomez	1.20	while (1); //forever
609 :	chl	1.18	}
610 :
611 :	chl	1.44.2.1	static void
612 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
613 :			{
614 :			int iDirection;
615 :	chl	1.35
616 :	chl	1.44.2.1	do {
617 :			iDirection = 0;
618 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
619 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
620 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
621 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
622 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
623 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
624 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
625 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
626 :
627 :			bDirection = iDirection;
628 :			x = data->currentMV->x; y = data->currentMV->y;
629 :			} while (iDirection);
630 :	chl	1.35	}
631 :
632 :	chl	1.44.2.1	static void
633 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
634 :	chl	1.18	{
635 :
636 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
637 :
638 :	chl	1.44.2.1	int iDirection;
639 :	edgomez	1.20
640 :			do {
641 :	chl	1.18	iDirection = 0;
642 :	chl	1.44.2.1	if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
643 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
644 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
645 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
646 :	chl	1.18
647 :			/* now we're doing diagonal checks near our candidate */
648 :
649 :	chl	1.44.2.1	if (iDirection) { //checking if anything found
650 :	chl	1.18	bDirection = iDirection;
651 :			iDirection = 0;
652 :	chl	1.44.2.1	x = data->currentMV->x; y = data->currentMV->y;
653 :			if (bDirection & 3) { //our candidate is left or right
654 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
655 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
656 :			} else { // what remains here is up or down
657 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
658 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
659 :	chl	1.18
660 :	chl	1.44.2.1	bDirection += iDirection;
661 :			x = data->currentMV->x; y = data->currentMV->y;
662 :	chl	1.18	}
663 :			}
664 :	chl	1.44.2.1	while (iDirection);
665 :	chl	1.18	}
666 :
667 :	chl	1.44.2.1	/* MAINSEARCH FUNCTIONS END */
668 :
669 :			/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
670 :	chl	1.18
671 :	chl	1.44.2.1	static void
672 :			HalfpelRefine(const SearchData * const data)
673 :	Isibaar	1.1	{
674 :			/* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */
675 :
676 :	chl	1.44.2.1	VECTOR backupMV = *(data->currentMV);
677 :			int iDirection; //not needed
678 :	edgomez	1.20
679 :	chl	1.44.2.1	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
680 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
681 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
682 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
683 :	edgomez	1.20
684 :	chl	1.44.2.1	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
685 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
686 :
687 :			CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
688 :			CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
689 :	Isibaar	1.1	}
690 :
691 :	Isibaar	1.44.2.7
692 :			static void
693 :			QuarterpelRefine(const SearchData * const data)
694 :			{
695 :			/* Perform quarter pixel refinement*/
696 :
697 :			VECTOR backupMV = *(data->currentQMV);
698 :			int iDirection; //not needed
699 :
700 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
701 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
702 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
703 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
704 :
705 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
706 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
707 :
708 :			CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
709 :			CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
710 :
711 :			}
712 :
713 :	chl	1.44.2.1	static __inline int
714 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
715 :			const int x, const int y,
716 :			const uint32_t iEdgedWidth, const uint32_t iQuant)
717 :	Isibaar	1.1
718 :	chl	1.44.2.1	{
719 :			/* keep repeating checks for all b-frames before this P frame,
720 :			to make sure that SKIP is possible (todo)
721 :			how: if skip is not possible set sad00 to a very high value */
722 :	chl	1.3
723 :	chl	1.44.2.1	uint32_t sadC = sad8(current->u + x8 + y(iEdgedWidth/2)*8,
724 :			reference->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2);
725 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
726 :	h	1.44.2.3	sadC += sad8(current->v + (x + y(iEdgedWidth/2))8,
727 :			reference->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
728 :	chl	1.44.2.1	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
729 :	chl	1.35
730 :	chl	1.44.2.1	return 1;
731 :			}
732 :	Isibaar	1.1
733 :	chl	1.44.2.1	static __inline void
734 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
735 :			{
736 :			pMB->mode = MODE_NOT_CODED;
737 :	Isibaar	1.44.2.2	pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
738 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
739 :	syskin	1.44.2.14
740 :	Isibaar	1.44.2.7	pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
741 :			pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
742 :
743 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
744 :			}
745 :
746 :	syskin	1.44.2.14	bool
747 :	chl	1.44.2.1	MotionEstimation(MBParam * const pParam,
748 :			FRAMEINFO * const current,
749 :			FRAMEINFO * const reference,
750 :			const IMAGE * const pRefH,
751 :			const IMAGE * const pRefV,
752 :			const IMAGE * const pRefHV,
753 :			const uint32_t iLimit)
754 :			{
755 :			MACROBLOCK *const pMBs = current->mbs;
756 :			const IMAGE *const pCurrent = &current->image;
757 :			const IMAGE *const pRef = &reference->image;
758 :	Isibaar	1.1
759 :	syskin	1.44.2.14	FILE * debug;
760 :	chl	1.44.2.1	const VECTOR zeroMV = { 0, 0 };
761 :	edgomez	1.20
762 :	chl	1.44.2.1	uint32_t x, y;
763 :			uint32_t iIntra = 0;
764 :	syskin	1.44.2.14	int32_t InterBias, quant = current->quant, sad00;
765 :	Isibaar	1.44.2.7	uint8_t *qimage;
766 :	chl	1.44.2.1
767 :	Isibaar	1.44.2.2	// some pre-initialized thingies for SearchP
768 :			int32_t temp[5];
769 :			VECTOR currentMV[5];
770 :	Isibaar	1.44.2.7	VECTOR currentQMV[5];
771 :	Isibaar	1.44.2.2	int32_t iMinSAD[5];
772 :			SearchData Data;
773 :			Data.iEdgedWidth = pParam->edged_width;
774 :			Data.currentMV = currentMV;
775 :	Isibaar	1.44.2.7	Data.currentQMV = currentQMV;
776 :	Isibaar	1.44.2.2	Data.iMinSAD = iMinSAD;
777 :			Data.temp = temp;
778 :			Data.iFcode = current->fcode;
779 :	Isibaar	1.44.2.7	Data.rounding = pParam->m_rounding_type;
780 :	syskin	1.44.2.14	Data.qpel = pParam->m_quarterpel;
781 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
782 :	Isibaar	1.44.2.7
783 :			if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
784 :	syskin	1.44.2.8	return 1; // allocate some mem for qpel interpolated blocks
785 :	Isibaar	1.44.2.7	// somehow this is dirty since I think we shouldn't use malloc outside
786 :			// encoder_create() - so please fix me!
787 :	syskin	1.44.2.14	Data.RefQ = qimage;
788 :	chl	1.44.2.1	if (sadInit) (*sadInit) ();
789 :
790 :			for (y = 0; y < pParam->mb_height; y++) {
791 :			for (x = 0; x < pParam->mb_width; x++) {
792 :			MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
793 :	syskin	1.44.2.14
794 :			pMB->sad16
795 :	chl	1.44.2.1	= sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
796 :			pRef->y + (x + y * pParam->edged_width) * 16,
797 :			pParam->edged_width, pMB->sad8 );
798 :
799 :	syskin	1.44.2.14	if (Data.chroma) {
800 :			pMB->sad16 += sad8(pCurrent->u + x8 + y(pParam->edged_width/2)*8,
801 :			pRef->u + x8 + y(pParam->edged_width/2)*8, pParam->edged_width/2);
802 :
803 :			pMB->sad16 += sad8(pCurrent->v + (x + y(pParam->edged_width/2))8,
804 :			pRef->v + (x + y(pParam->edged_width/2))8, pParam->edged_width/2);
805 :			}
806 :
807 :			sad00 = pMB->sad16; //if no gmc; else sad00 = (..)
808 :
809 :	chl	1.44.2.1	if (!(current->global_flags & XVID_LUMIMASKING)) {
810 :			pMB->dquant = NO_CHANGE;
811 :	syskin	1.44.2.11	pMB->quant = current->quant;
812 :			} else {
813 :	syskin	1.44.2.5	if (pMB->dquant != NO_CHANGE) {
814 :			quant += DQtab[pMB->dquant];
815 :			if (quant > 31) quant = 31;
816 :			else if (quant < 1) quant = 1;
817 :			}
818 :	syskin	1.44.2.11	pMB->quant = quant;
819 :			}
820 :	chl	1.44.2.1
821 :	syskin	1.44.2.14	//initial skip decision
822 :	chl	1.44.2.13	/* no early skip for GMC (global vector = skip vector is unknown!) */
823 :			if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
824 :			if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH)
825 :	syskin	1.44.2.14	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
826 :	chl	1.44.2.13	SkipMacroblockP(pMB, sad00);
827 :			continue;
828 :			}
829 :			}
830 :	syskin	1.44.2.14
831 :			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
832 :	chl	1.44.2.1	y, current->motion_flags, pMB->quant,
833 :	Isibaar	1.44.2.2	&Data, pParam, pMBs, reference->mbs,
834 :	chl	1.44.2.1	current->global_flags & XVID_INTER4V, pMB);
835 :
836 :			/* final skip decision, a.k.a. "the vector you found, really that good?" */
837 :	chl	1.44.2.13	if (current->coding_type == P_VOP) {
838 :			if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
839 :	syskin	1.44.2.11	&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) )
840 :	syskin	1.44.2.14	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
841 :	chl	1.44.2.13	SkipMacroblockP(pMB, sad00);
842 :			continue;
843 :			}
844 :			}
845 :	syskin	1.44.2.14
846 :	chl	1.44.2.1	/* finally, intra decision */
847 :	Isibaar	1.1
848 :	chl	1.44.2.1	InterBias = MV16_INTER_BIAS;
849 :	syskin	1.44.2.14	if (pMB->quant > 8) InterBias += 80 * (pMB->quant - 8); // to make high quants work
850 :	chl	1.44.2.1	if (y != 0)
851 :	syskin	1.44.2.14	if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 80;
852 :	chl	1.44.2.1	if (x != 0)
853 :	syskin	1.44.2.14	if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 80;
854 :
855 :			if (Data.chroma) InterBias += 50; // to compensate bigger SAD
856 :	chl	1.44.2.1
857 :			if (InterBias < pMB->sad16) {
858 :			const int32_t deviation =
859 :			dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
860 :			pParam->edged_width);
861 :	edgomez	1.20
862 :	chl	1.44.2.1	if (deviation < (pMB->sad16 - InterBias)) {
863 :	Isibaar	1.44.2.7	if (++iIntra >= iLimit) { free(qimage); return 1; }
864 :	chl	1.44.2.1	pMB->mode = MODE_INTRA;
865 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
866 :			pMB->mvs[3] = zeroMV;
867 :	Isibaar	1.44.2.7	pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
868 :			pMB->qmvs[3] = zeroMV;
869 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
870 :			pMB->sad8[3] = 0;
871 :			}
872 :			}
873 :			}
874 :			}
875 :	Isibaar	1.44.2.7	free(qimage);
876 :	chl	1.44.2.13
877 :			if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
878 :	syskin	1.44.2.14	current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
879 :	chl	1.44.2.13	else
880 :			current->GMC_MV = zeroMV;
881 :
882 :	syskin	1.44.2.14	return 0;
883 :	chl	1.44.2.1	}
884 :	chl	1.18
885 :	edgomez	1.20
886 :	chl	1.44.2.1	#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16\|PMV_HALFPELREFINE16)
887 :	edgomez	1.20
888 :	chl	1.44.2.1	static __inline int
889 :			make_mask(const VECTOR * const pmv, const int i)
890 :			{
891 :	Isibaar	1.44.2.2	int mask = 255, j;
892 :	chl	1.44.2.1	for (j = 0; j < i; j++) {
893 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
894 :			if (pmv[i].x == pmv[j].x) {
895 :			if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
896 :			if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
897 :			} else
898 :			if (pmv[i].y == pmv[j].y) {
899 :			if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
900 :			if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
901 :			}
902 :			}
903 :			return mask;
904 :			}
905 :
906 :			static __inline void
907 :			PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount,
908 :			const int iHcount, const MACROBLOCK * const prevMB)
909 :			{
910 :
911 :			//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
912 :
913 :			if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour
914 :			pmv[5].x = EVEN(pmv[3].x);
915 :	Isibaar	1.44.2.2	pmv[5].y = EVEN(pmv[3].y);
916 :			} else pmv[5].x = pmv[5].y = 0;
917 :	chl	1.44.2.1
918 :			if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
919 :			else pmv[3].x = pmv[3].y = 0;
920 :
921 :			if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
922 :			else pmv[4].x = pmv[4].y = 0;
923 :
924 :			// [1] median prediction
925 :			pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
926 :
927 :			pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
928 :
929 :			pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
930 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
931 :
932 :			if ((x != iWcount-1) && (y != iHcount-1)) {
933 :			pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
934 :	Isibaar	1.44.2.2	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
935 :			} else pmv[6].x = pmv[6].y = 0;
936 :	chl	1.44.2.1	}
937 :
938 :			static void
939 :	syskin	1.44.2.14	SearchP(const IMAGE * const pRef,
940 :	chl	1.44.2.1	const uint8_t * const pRefH,
941 :			const uint8_t * const pRefV,
942 :			const uint8_t * const pRefHV,
943 :			const IMAGE * const pCur,
944 :			const int x,
945 :			const int y,
946 :			const uint32_t MotionFlags,
947 :			const uint32_t iQuant,
948 :	Isibaar	1.44.2.2	SearchData * const Data,
949 :	chl	1.44.2.1	const MBParam * const pParam,
950 :			const MACROBLOCK * const pMBs,
951 :			const MACROBLOCK * const prevMBs,
952 :			int inter4v,
953 :			MACROBLOCK * const pMB)
954 :			{
955 :	Isibaar	1.1
956 :	chl	1.44.2.1	int i, iDirection = 255, mask, threshA;
957 :	Isibaar	1.44.2.2	VECTOR pmv[7];
958 :	chl	1.44.2.1
959 :	Isibaar	1.44.2.2	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
960 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
961 :	Isibaar	1.44.2.7	pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
962 :	Isibaar	1.44.2.2
963 :			Data->predMV = pmv[0];
964 :	Isibaar	1.44.2.7
965 :	Isibaar	1.44.2.2	Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
966 :	syskin	1.44.2.14	Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8;
967 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8;
968 :
969 :			Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16;
970 :	Isibaar	1.44.2.2	Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16;
971 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16;
972 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16;
973 :	syskin	1.44.2.14	Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
974 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
975 :	Isibaar	1.1
976 :	syskin	1.44.2.11	Data->lambda16 = lambda_vec16[iQuant];
977 :	Isibaar	1.44.2.12	Data->lambda8 = lambda_vec8[iQuant];
978 :	Isibaar	1.1
979 :	edgomez	1.20	if (!(MotionFlags & PMV_HALFPEL16)) {
980 :	Isibaar	1.44.2.2	Data->min_dx = EVEN(Data->min_dx);
981 :			Data->max_dx = EVEN(Data->max_dx);
982 :			Data->min_dy = EVEN(Data->min_dy);
983 :			Data->max_dy = EVEN(Data->max_dy); }
984 :	edgomez	1.44.2.4
985 :			if (pMB->dquant != NO_CHANGE) inter4v = 0;
986 :	Isibaar	1.44.2.2
987 :	Isibaar	1.44.2.7	for(i = 0; i < 5; i++)
988 :			Data->currentMV[i].x = Data->currentMV[i].y = 0;
989 :	chl	1.44.2.1
990 :	Isibaar	1.44.2.12	if (pParam->m_quarterpel) {
991 :			Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
992 :			i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode);
993 :			} else i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
994 :
995 :	syskin	1.44.2.14	Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000;
996 :			Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100;
997 :	Isibaar	1.44.2.2	Data->iMinSAD[2] = pMB->sad8[1];
998 :			Data->iMinSAD[3] = pMB->sad8[2];
999 :			Data->iMinSAD[4] = pMB->sad8[3];
1000 :	chl	1.44.2.1
1001 :			if ((x == 0) && (y == 0)) threshA = 512;
1002 :			else {
1003 :	edgomez	1.44.2.4	threshA = Data->temp[0]; // that's when we keep this SAD atm
1004 :	chl	1.44.2.1	if (threshA < 512) threshA = 512;
1005 :			if (threshA > 1024) threshA = 1024; }
1006 :
1007 :			PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1008 :			prevMBs + x + y * pParam->mb_width);
1009 :
1010 :	syskin	1.44.2.14	if (inter4v \|\| pParam->m_quarterpel \|\| Data->chroma) CheckCandidate = CheckCandidate16;
1011 :			else CheckCandidate = CheckCandidate16no4v;
1012 :	edgomez	1.44.2.4
1013 :	chl	1.44.2.1	/* main loop. checking all predictions */
1014 :
1015 :			for (i = 1; i < 7; i++) {
1016 :			if (!(mask = make_mask(pmv, i)) ) continue;
1017 :	edgomez	1.44.2.4	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1018 :	Isibaar	1.44.2.2	if (Data->iMinSAD[0] <= threshA) break;
1019 :	chl	1.44.2.1	}
1020 :
1021 :	Isibaar	1.44.2.2	if ((Data->iMinSAD[0] <= threshA) \|\|
1022 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1023 :			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1024 :	chl	1.44.2.1	inter4v = 0;
1025 :	Isibaar	1.44.2.2	} else {
1026 :	edgomez	1.20
1027 :	Isibaar	1.44.2.2	MainSearchFunc * MainSearchPtr;
1028 :			if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1029 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1030 :			else MainSearchPtr = DiamondSearch;
1031 :	chl	1.35
1032 :	Isibaar	1.44.2.2	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1033 :	chl	1.13
1034 :	chl	1.44.2.1	/* extended search, diamond starting in 0,0 and in prediction.
1035 :			note that this search is/might be done in halfpel positions,
1036 :			which makes it more different than the diamond above */
1037 :	Isibaar	1.1
1038 :	Isibaar	1.44.2.2	if (MotionFlags & PMV_EXTSEARCH16) {
1039 :			int32_t bSAD;
1040 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1041 :			if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1042 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1043 :			if (!(MVequal(startMV, backupMV))) {
1044 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1045 :
1046 :	Isibaar	1.44.2.12	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1047 :	Isibaar	1.44.2.2	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1048 :			if (bSAD < Data->iMinSAD[0]) {
1049 :			Data->currentMV[0] = backupMV;
1050 :			Data->iMinSAD[0] = bSAD; }
1051 :			}
1052 :	chl	1.13
1053 :	Isibaar	1.44.2.2	backupMV = Data->currentMV[0];
1054 :			if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1;
1055 :			else startMV.x = startMV.y = 0;
1056 :			if (!(MVequal(startMV, backupMV))) {
1057 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1058 :
1059 :	Isibaar	1.44.2.12	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1060 :	Isibaar	1.44.2.2	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1061 :			if (bSAD < Data->iMinSAD[0]) {
1062 :			Data->currentMV[0] = backupMV;
1063 :			Data->iMinSAD[0] = bSAD; }
1064 :			}
1065 :	chl	1.44.2.1	}
1066 :			}
1067 :	chl	1.13
1068 :	Isibaar	1.44.2.2	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1069 :	Isibaar	1.1
1070 :	Isibaar	1.44.2.7	for(i = 0; i < 5; i++) {
1071 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1072 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1073 :			}
1074 :
1075 :			if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1076 :
1077 :	Isibaar	1.44.2.12	CheckCandidate = CheckCandidate16_qpel;
1078 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1079 :	syskin	1.44.2.11	pParam->width, pParam->height, Data->iFcode, 0);
1080 :	syskin	1.44.2.8
1081 :	Isibaar	1.44.2.7	QuarterpelRefine(Data);
1082 :			}
1083 :
1084 :	syskin	1.44.2.14	if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0;
1085 :	Isibaar	1.44.2.2	if (inter4v) {
1086 :			SearchData Data8;
1087 :			Data8.iFcode = Data->iFcode;
1088 :	Isibaar	1.44.2.12	Data8.lambda8 = Data->lambda8;
1089 :	Isibaar	1.44.2.2	Data8.iEdgedWidth = Data->iEdgedWidth;
1090 :	Isibaar	1.44.2.12	Data8.RefQ = Data->RefQ;
1091 :	syskin	1.44.2.14	Data8.qpel = Data->qpel;
1092 :	Isibaar	1.44.2.2	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1093 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1094 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1095 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1096 :	syskin	1.44.2.14
1097 :			if (Data->chroma) {
1098 :			int sum, dx, dy;
1099 :
1100 :			if(pParam->m_quarterpel) {
1101 :			sum = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2;
1102 :			} else sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1103 :			dy = (sum >> 3) + roundtab_76[sum & 0xf];
1104 :
1105 :			if(pParam->m_quarterpel) {
1106 :			sum = pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2;
1107 :			} else sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1108 :			dx = (sum >> 3) + roundtab_76[sum & 0xf];
1109 :
1110 :			Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1111 :			}
1112 :	Isibaar	1.44.2.2	}
1113 :	chl	1.13
1114 :	chl	1.44.2.1	if (!(inter4v) \|\|
1115 :	Isibaar	1.44.2.2	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1116 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1117 :	chl	1.44.2.1	// INTER MODE
1118 :			pMB->mode = MODE_INTER;
1119 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1]
1120 :			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1121 :	edgomez	1.20
1122 :	Isibaar	1.44.2.7	pMB->qmvs[0] = pMB->qmvs[1]
1123 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1124 :
1125 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1126 :	Isibaar	1.44.2.2	pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1127 :	chl	1.13
1128 :	Isibaar	1.44.2.7	if(pParam->m_quarterpel) {
1129 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1130 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1131 :	syskin	1.44.2.14	} else {
1132 :	Isibaar	1.44.2.7	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1133 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1134 :			}
1135 :	chl	1.44.2.1	} else {
1136 :			// INTER4V MODE; all other things are already set in Search8
1137 :			pMB->mode = MODE_INTER4V;
1138 :	Isibaar	1.44.2.2	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1139 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1140 :	chl	1.44.2.1	}
1141 :			}
1142 :	edgomez	1.20
1143 :	chl	1.44.2.1	static void
1144 :			Search8(const SearchData * const OldData,
1145 :			const int x, const int y,
1146 :			const uint32_t MotionFlags,
1147 :			const MBParam * const pParam,
1148 :			MACROBLOCK * const pMB,
1149 :			const MACROBLOCK * const pMBs,
1150 :	Isibaar	1.44.2.2	const int block,
1151 :			SearchData * const Data)
1152 :	chl	1.44.2.1	{
1153 :	Isibaar	1.44.2.2	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1154 :			Data->currentMV = OldData->currentMV + 1 + block;
1155 :	Isibaar	1.44.2.7	Data->currentQMV = OldData->currentQMV + 1 + block;
1156 :
1157 :	syskin	1.44.2.11	if(pParam->m_quarterpel) {
1158 :	Isibaar	1.44.2.12	Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1159 :	syskin	1.44.2.14	if (block != 0) (Data->iMinSAD) += (Data->lambda8
1160 :	syskin	1.44.2.11	d_mv_bits( Data->currentQMV->x - Data->predQMV.x,
1161 :			Data->currentQMV->y - Data->predQMV.y,
1162 :	syskin	1.44.2.14	Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1163 :	Isibaar	1.44.2.12	} else {
1164 :			Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1165 :	syskin	1.44.2.14	if (block != 0) (Data->iMinSAD) += (Data->lambda8
1166 :	syskin	1.44.2.11	d_mv_bits( Data->currentMV->x - Data->predMV.x,
1167 :			Data->currentMV->y - Data->predMV.y,
1168 :	syskin	1.44.2.14	Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1169 :	Isibaar	1.44.2.12	}
1170 :	Isibaar	1.1
1171 :	chl	1.44.2.1	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8)) {
1172 :	edgomez	1.20
1173 :	Isibaar	1.44.2.2	Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1174 :			Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1175 :			Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1176 :			Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1177 :	Isibaar	1.1
1178 :	Isibaar	1.44.2.2	Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1179 :	chl	1.44.2.1
1180 :	Isibaar	1.44.2.2	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1181 :	Isibaar	1.44.2.7	pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1182 :	syskin	1.44.2.14	CheckCandidate = CheckCandidate8;
1183 :	Isibaar	1.1
1184 :	chl	1.44.2.1	if (MotionFlags & PMV_EXTSEARCH8) {
1185 :	Isibaar	1.44.2.7	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1186 :
1187 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1188 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1189 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1190 :			else MainSearchPtr = DiamondSearch;
1191 :	Isibaar	1.1
1192 :	Isibaar	1.44.2.7	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1193 :	Isibaar	1.1
1194 :	syskin	1.44.2.11	if(*(Data->iMinSAD) < temp_sad) {
1195 :	Isibaar	1.44.2.7	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1196 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1197 :			}
1198 :			}
1199 :
1200 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1201 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1202 :
1203 :			HalfpelRefine(Data); // perform halfpel refine of current best vector
1204 :
1205 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1206 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1207 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1208 :			}
1209 :			}
1210 :
1211 :	syskin	1.44.2.11	if(pParam->m_quarterpel) {
1212 :			if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1213 :			(MotionFlags & PMV_QUARTERPELREFINE8)) {
1214 :	Isibaar	1.44.2.10	CheckCandidate = CheckCandidate8_qpel;
1215 :	Isibaar	1.44.2.12	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1216 :			pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1217 :			QuarterpelRefine(Data);
1218 :	syskin	1.44.2.11	}
1219 :	Isibaar	1.44.2.7	}
1220 :			}
1221 :
1222 :			if(pParam->m_quarterpel) {
1223 :			pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
1224 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
1225 :			}
1226 :			else {
1227 :			pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1228 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1229 :	Isibaar	1.1	}
1230 :
1231 :	Isibaar	1.44.2.2	pMB->mvs[block] = *(Data->currentMV);
1232 :	Isibaar	1.44.2.7	pMB->qmvs[block] = *(Data->currentQMV);
1233 :	Isibaar	1.44.2.2
1234 :	syskin	1.44.2.11	pMB->sad8[block] = 4 * (*Data->iMinSAD);
1235 :	Isibaar	1.1	}
1236 :
1237 :	chl	1.44.2.1	/* B-frames code starts here */
1238 :	Isibaar	1.1
1239 :	chl	1.44.2.1	static __inline VECTOR
1240 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1241 :			{
1242 :			/* the stupidiest function ever */
1243 :			if (mode == MODE_FORWARD) return pMB->mvs[0];
1244 :			else return pMB->b_mvs[0];
1245 :			}
1246 :	Isibaar	1.1
1247 :	chl	1.44.2.1	static void __inline
1248 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1249 :			const uint32_t iWcount,
1250 :			const MACROBLOCK * const pMB,
1251 :			const uint32_t mode_curr)
1252 :			{
1253 :	Isibaar	1.1
1254 :	chl	1.44.2.1	// [0] is prediction
1255 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1256 :	Isibaar	1.1
1257 :	chl	1.44.2.1	pmv[1].x = pmv[1].y = 0; // [1] is zero
1258 :	Isibaar	1.1
1259 :	chl	1.44.2.1	pmv[2] = ChoosePred(pMB, mode_curr);
1260 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1261 :	Isibaar	1.1
1262 :	chl	1.44.2.1	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1263 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1264 :	Isibaar	1.44.2.2	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1265 :			} else pmv[3].x = pmv[3].y = 0;
1266 :	chl	1.39
1267 :	chl	1.44.2.1	if (y != 0) {
1268 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1269 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1270 :			} else pmv[4].x = pmv[4].y = 0;
1271 :	chl	1.39
1272 :	chl	1.44.2.1	if (x != 0) {
1273 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1274 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1275 :			} else pmv[5].x = pmv[5].y = 0;
1276 :	chl	1.39
1277 :	chl	1.44.2.1	if ((x != 0)&&(y != 0)) {
1278 :			pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1279 :			pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1280 :			} else pmv[6].x = pmv[6].y = 0;
1281 :	chl	1.39
1282 :	chl	1.44.2.1	// more?
1283 :	chl	1.39	}
1284 :
1285 :
1286 :	chl	1.44.2.1	/* search backward or forward, for b-frames */
1287 :			static void
1288 :			SearchBF( const uint8_t * const pRef,
1289 :			const uint8_t * const pRefH,
1290 :			const uint8_t * const pRefV,
1291 :			const uint8_t * const pRefHV,
1292 :			const IMAGE * const pCur,
1293 :			const int x, const int y,
1294 :			const uint32_t MotionFlags,
1295 :			const uint32_t iFcode,
1296 :			const MBParam * const pParam,
1297 :			MACROBLOCK * const pMB,
1298 :			const VECTOR * const predMV,
1299 :			int32_t * const best_sad,
1300 :	h	1.44.2.3	const int32_t mode_current,
1301 :			SearchData * const Data)
1302 :	Isibaar	1.1	{
1303 :
1304 :	chl	1.44.2.1	const int32_t iEdgedWidth = pParam->edged_width;
1305 :
1306 :			int i, iDirection, mask;
1307 :	h	1.44.2.3	VECTOR pmv[7];
1308 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1309 :	h	1.44.2.3	*Data->iMinSAD = MV_MAX_ERROR;
1310 :			Data->iFcode = iFcode;
1311 :	chl	1.44.2.1
1312 :	h	1.44.2.3	Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1313 :			Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1314 :			Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1315 :			Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1316 :	chl	1.44.2.1
1317 :	h	1.44.2.3	Data->predMV = *predMV;
1318 :	edgomez	1.20
1319 :	h	1.44.2.3	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1320 :	Isibaar	1.44.2.7	pParam->width, pParam->height, iFcode, pParam->m_quarterpel);
1321 :	Isibaar	1.1
1322 :	h	1.44.2.3	pmv[0] = Data->predMV;
1323 :			PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1324 :	Isibaar	1.1
1325 :	h	1.44.2.3	Data->currentMV->x = Data->currentMV->y = 0;
1326 :	Isibaar	1.1
1327 :	chl	1.44.2.1	CheckCandidate = CheckCandidate16no4v;
1328 :	edgomez	1.20
1329 :	chl	1.44.2.1	// main loop. checking all predictions
1330 :			for (i = 0; i < 8; i++) {
1331 :			if (!(mask = make_mask(pmv, i)) ) continue;
1332 :	h	1.44.2.3	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1333 :	edgomez	1.20	}
1334 :
1335 :	chl	1.44.2.1	if (MotionFlags & PMV_USESQUARES16)
1336 :			MainSearchPtr = SquareSearch;
1337 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1338 :			MainSearchPtr = AdvDiamondSearch;
1339 :			else MainSearchPtr = DiamondSearch;
1340 :	Isibaar	1.1
1341 :	h	1.44.2.3	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1342 :	Isibaar	1.1
1343 :	h	1.44.2.3	HalfpelRefine(Data);
1344 :	Isibaar	1.1
1345 :	chl	1.44.2.1	// three bits are needed to code backward mode. four for forward
1346 :			// we treat the bits just like they were vector's
1347 :	syskin	1.44.2.14	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1348 :			else Data->iMinSAD += 3 Data->lambda16;
1349 :	chl	1.44.2.1
1350 :	h	1.44.2.3	if (Data->iMinSAD < best_sad) {
1351 :			best_sad = Data->iMinSAD;
1352 :	chl	1.44.2.1	pMB->mode = mode_current;
1353 :	h	1.44.2.3	pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1354 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1355 :	syskin	1.44.2.11	if (mode_current == MODE_FORWARD) pMB->mvs[0] = (Data->currentMV+2) = Data->currentMV;
1356 :			else pMB->b_mvs[0] = (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1357 :	Isibaar	1.1	}
1358 :	chl	1.44.2.1
1359 :			}
1360 :
1361 :			static int32_t
1362 :	Isibaar	1.44.2.2	SearchDirect(const IMAGE * const f_Ref,
1363 :	chl	1.44.2.1	const uint8_t * const f_RefH,
1364 :			const uint8_t * const f_RefV,
1365 :			const uint8_t * const f_RefHV,
1366 :	Isibaar	1.44.2.2	const IMAGE * const b_Ref,
1367 :	chl	1.44.2.1	const uint8_t * const b_RefH,
1368 :			const uint8_t * const b_RefV,
1369 :			const uint8_t * const b_RefHV,
1370 :			const IMAGE * const pCur,
1371 :			const int x, const int y,
1372 :			const uint32_t MotionFlags,
1373 :			const int32_t TRB, const int32_t TRD,
1374 :			const MBParam * const pParam,
1375 :			MACROBLOCK * const pMB,
1376 :			const MACROBLOCK * const b_mb,
1377 :	h	1.44.2.3	int32_t * const best_sad,
1378 :			SearchData * const Data)
1379 :	chl	1.44.2.1
1380 :			{
1381 :	h	1.44.2.3	int32_t skip_sad;
1382 :	chl	1.44.2.1	int k;
1383 :	h	1.44.2.3
1384 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1385 :
1386 :	h	1.44.2.3	Data->iMinSAD = 2564096;
1387 :			Data->referencemv = b_mb->mvs;
1388 :	chl	1.44.2.1
1389 :	h	1.44.2.3	Data->Ref = f_Ref->y + (x + Data->iEdgedWidthy) 16;
1390 :			Data->RefH = f_RefH + (x + Data->iEdgedWidthy) 16;
1391 :			Data->RefV = f_RefV + (x + Data->iEdgedWidthy) 16;
1392 :			Data->RefHV = f_RefHV + (x + Data->iEdgedWidthy) 16;
1393 :			Data->bRef = b_Ref->y + (x + Data->iEdgedWidthy) 16;
1394 :			Data->bRefH = b_RefH + (x + Data->iEdgedWidthy) 16;
1395 :			Data->bRefV = b_RefV + (x + Data->iEdgedWidthy) 16;
1396 :			Data->bRefHV = b_RefHV + (x + Data->iEdgedWidthy) 16;
1397 :
1398 :			Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1399 :			Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1400 :			Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1401 :			Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1402 :	chl	1.44.2.1
1403 :			for (k = 0; k < 4; k++) {
1404 :	h	1.44.2.3	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1405 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1406 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1407 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1408 :
1409 :			if ( ( pMB->b_mvs[k].x > Data->max_dx ) \|\| ( pMB->b_mvs[k].x < Data->min_dx )
1410 :			\|\| ( pMB->b_mvs[k].y > Data->max_dy ) \|\| ( pMB->b_mvs[k].y < Data->min_dy )) {
1411 :	chl	1.44.2.1
1412 :	Isibaar	1.44.2.2	best_sad = 2564096; // in that case, we won't use direct mode
1413 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1414 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1415 :			return 0;
1416 :			}
1417 :			if (b_mb->mode != MODE_INTER4V) {
1418 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1419 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1420 :	h	1.44.2.3	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1421 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1422 :	Isibaar	1.44.2.2	break;
1423 :			}
1424 :			}
1425 :	chl	1.44.2.1
1426 :			if (b_mb->mode == MODE_INTER4V)
1427 :			CheckCandidate = CheckCandidateDirect;
1428 :			else CheckCandidate = CheckCandidateDirectno4v;
1429 :
1430 :	h	1.44.2.3	(*CheckCandidate)(0, 0, 255, &k, Data);
1431 :	Isibaar	1.44.2.2
1432 :			// skip decision
1433 :	syskin	1.44.2.14	if (Data->iMinSAD < pMB->quant SKIP_THRESH_B) {
1434 :	syskin	1.44.2.8	//possible skip - checking chroma. everything copied from MC
1435 :	Isibaar	1.44.2.2	//this is not full chroma compensation, only it's fullpel approximation. should work though
1436 :			int sum, dx, dy, b_dx, b_dy;
1437 :
1438 :			sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1439 :			dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1440 :
1441 :			sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1442 :			dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1443 :
1444 :			sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
1445 :			b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1446 :
1447 :			sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
1448 :			b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1449 :
1450 :	h	1.44.2.3	sum = sad8bi(pCur->u + 8x + 8y*(Data->iEdgedWidth/2),
1451 :			f_Ref->u + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1452 :			b_Ref->u + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1453 :			Data->iEdgedWidth/2);
1454 :			sum += sad8bi(pCur->v + 8x + 8y*(Data->iEdgedWidth/2),
1455 :			f_Ref->v + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1456 :			b_Ref->v + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1457 :			Data->iEdgedWidth/2);
1458 :	Isibaar	1.44.2.2
1459 :	syskin	1.44.2.11	if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) {
1460 :	Isibaar	1.44.2.2	pMB->mode = MODE_DIRECT_NONE_MV;
1461 :	h	1.44.2.3	return *Data->iMinSAD;
1462 :	Isibaar	1.44.2.2	}
1463 :			}
1464 :
1465 :	h	1.44.2.3	skip_sad = *Data->iMinSAD;
1466 :	Isibaar	1.44.2.2
1467 :	chl	1.44.2.1	// DIRECT MODE DELTA VECTOR SEARCH.
1468 :			// This has to be made more effective, but at the moment I'm happy it's running at all
1469 :
1470 :			if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1471 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1472 :			else MainSearchPtr = DiamondSearch;
1473 :
1474 :	h	1.44.2.3	(*MainSearchPtr)(0, 0, Data, 255);
1475 :	chl	1.44.2.1
1476 :	h	1.44.2.3	HalfpelRefine(Data);
1477 :	chl	1.44.2.1
1478 :	syskin	1.44.2.14	Data->iMinSAD += 1 Data->lambda16; // one bit is needed to code direct mode
1479 :	h	1.44.2.3	best_sad = Data->iMinSAD;
1480 :	chl	1.44.2.1
1481 :			if (b_mb->mode == MODE_INTER4V)
1482 :			pMB->mode = MODE_DIRECT;
1483 :			else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1484 :
1485 :	h	1.44.2.3	pMB->pmvs[3] = *Data->currentMV;
1486 :	chl	1.44.2.1
1487 :			for (k = 0; k < 4; k++) {
1488 :	h	1.44.2.3	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1489 :			pMB->b_mvs[k].x = ((Data->currentMV->x == 0)
1490 :			? Data->directmvB[k].x
1491 :			: pMB->mvs[k].x - Data->referencemv[k].x);
1492 :			pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1493 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1494 :			? Data->directmvB[k].y
1495 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1496 :	chl	1.44.2.1	if (b_mb->mode != MODE_INTER4V) {
1497 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1498 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1499 :			break;
1500 :			}
1501 :			}
1502 :	h	1.44.2.3	return skip_sad;
1503 :	chl	1.44.2.1	}
1504 :
1505 :	h	1.44.2.3
1506 :	chl	1.44.2.1	static __inline void
1507 :			SearchInterpolate(const uint8_t * const f_Ref,
1508 :			const uint8_t * const f_RefH,
1509 :			const uint8_t * const f_RefV,
1510 :			const uint8_t * const f_RefHV,
1511 :			const uint8_t * const b_Ref,
1512 :			const uint8_t * const b_RefH,
1513 :			const uint8_t * const b_RefV,
1514 :			const uint8_t * const b_RefHV,
1515 :			const IMAGE * const pCur,
1516 :			const int x, const int y,
1517 :			const uint32_t fcode,
1518 :			const uint32_t bcode,
1519 :			const uint32_t MotionFlags,
1520 :			const MBParam * const pParam,
1521 :			const VECTOR * const f_predMV,
1522 :			const VECTOR * const b_predMV,
1523 :			MACROBLOCK * const pMB,
1524 :	h	1.44.2.3	int32_t * const best_sad,
1525 :			SearchData * const fData)
1526 :	edgomez	1.20
1527 :	chl	1.3	{
1528 :
1529 :	edgomez	1.20	const int32_t iEdgedWidth = pParam->edged_width;
1530 :	chl	1.3
1531 :	chl	1.44.2.1	int iDirection, i, j;
1532 :	h	1.44.2.3	SearchData bData;
1533 :	edgomez	1.20
1534 :	syskin	1.44.2.11	(bData.iMinSAD = fData->iMinSAD) = 4096256;
1535 :	h	1.44.2.3	bData.Cur = fData->Cur;
1536 :			fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
1537 :			bData.currentMV = fData->currentMV + 1;
1538 :	syskin	1.44.2.11	bData.lambda16 = fData->lambda16;
1539 :	h	1.44.2.3	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1540 :
1541 :			bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16;
1542 :			bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16;
1543 :			bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16;
1544 :			bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
1545 :			bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16;
1546 :			bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
1547 :			bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
1548 :			bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
1549 :
1550 :			bData.bpredMV = fData->predMV = *f_predMV;
1551 :			fData->bpredMV = bData.predMV = *b_predMV;
1552 :
1553 :	syskin	1.44.2.11	fData->currentMV[0] = fData->currentMV[3]; //forward search stored it's vector here. backward stored it in the place it's needed
1554 :	Isibaar	1.44.2.7	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel);
1555 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel);
1556 :	chl	1.3
1557 :	h	1.44.2.3	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1558 :			if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy;
1559 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx;
1560 :			if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1561 :
1562 :			if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1563 :			if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy;
1564 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx;
1565 :			if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1566 :	Isibaar	1.44.2.2
1567 :	h	1.44.2.3	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1568 :	chl	1.3
1569 :	chl	1.44.2.1	//diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1570 :	chl	1.3
1571 :	chl	1.44.2.1	do {
1572 :			iDirection = 255;
1573 :			// forward MV moves
1574 :	h	1.44.2.3	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1575 :	chl	1.35
1576 :	h	1.44.2.3	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1577 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1578 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1579 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1580 :	edgomez	1.20
1581 :	chl	1.44.2.1	// backward MV moves
1582 :	h	1.44.2.3	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1583 :			fData->currentMV[2] = fData->currentMV[0];
1584 :	chl	1.3
1585 :	Isibaar	1.44.2.2	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1586 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1587 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1588 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1589 :	chl	1.3
1590 :	chl	1.44.2.1	} while (!(iDirection));
1591 :	chl	1.3
1592 :	syskin	1.44.2.14	fData->iMinSAD += 2 fData->lambda16; // two bits are needed to code interpolate mode.
1593 :	Isibaar	1.44.2.12
1594 :	h	1.44.2.3	if (fData->iMinSAD < best_sad) {
1595 :			best_sad = fData->iMinSAD;
1596 :			pMB->mvs[0] = fData->currentMV[0];
1597 :			pMB->b_mvs[0] = fData->currentMV[1];
1598 :	chl	1.44.2.1	pMB->mode = MODE_INTERPOLATE;
1599 :	edgomez	1.20
1600 :	chl	1.44.2.1	pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1601 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1602 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1603 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1604 :	edgomez	1.20	}
1605 :	chl	1.3	}
1606 :
1607 :	chl	1.44.2.1	void
1608 :			MotionEstimationBVOP(MBParam * const pParam,
1609 :			FRAMEINFO * const frame,
1610 :			const int32_t time_bp,
1611 :			const int32_t time_pp,
1612 :			// forward (past) reference
1613 :			const MACROBLOCK * const f_mbs,
1614 :			const IMAGE * const f_ref,
1615 :			const IMAGE * const f_refH,
1616 :			const IMAGE * const f_refV,
1617 :			const IMAGE * const f_refHV,
1618 :			// backward (future) reference
1619 :			const MACROBLOCK * const b_mbs,
1620 :			const IMAGE * const b_ref,
1621 :			const IMAGE * const b_refH,
1622 :			const IMAGE * const b_refV,
1623 :			const IMAGE * const b_refHV)
1624 :			{
1625 :			uint32_t i, j;
1626 :			int32_t best_sad, skip_sad;
1627 :			int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1628 :			static const VECTOR zeroMV={0,0};
1629 :	chl	1.3
1630 :	chl	1.44.2.1	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1631 :	chl	1.3
1632 :	chl	1.44.2.1	const int32_t TRB = time_pp - time_bp;
1633 :			const int32_t TRD = time_pp;
1634 :	chl	1.3
1635 :	h	1.44.2.3	// some pre-inintialized data for the rest of the search
1636 :
1637 :			SearchData Data;
1638 :			int32_t iMinSAD;
1639 :			VECTOR currentMV[3];
1640 :			Data.iEdgedWidth = pParam->edged_width;
1641 :			Data.currentMV = currentMV;
1642 :			Data.iMinSAD = &iMinSAD;
1643 :	syskin	1.44.2.11	Data.lambda16 = lambda_vec16[frame->quant];
1644 :	h	1.44.2.3
1645 :	chl	1.44.2.1	// note: i==horizontal, j==vertical
1646 :	edgomez	1.20
1647 :	chl	1.44.2.1	for (j = 0; j < pParam->mb_height; j++) {
1648 :	chl	1.3
1649 :	chl	1.44.2.1	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1650 :	edgomez	1.20
1651 :	chl	1.44.2.1	for (i = 0; i < pParam->mb_width; i++) {
1652 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1653 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1654 :	edgomez	1.20
1655 :	chl	1.44.2.13	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1656 :	syskin	1.44.2.14	if (b_mb->mode == MODE_NOT_CODED) {
1657 :	chl	1.44.2.1	pMB->mode = MODE_NOT_CODED;
1658 :			continue;
1659 :			}
1660 :	edgomez	1.20
1661 :	h	1.44.2.3	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1662 :	syskin	1.44.2.11	pMB->quant = frame->quant;
1663 :	syskin	1.44.2.14
1664 :	chl	1.44.2.1	/* direct search comes first, because it (1) checks for SKIP-mode
1665 :			and (2) sets very good predictions for forward and backward search */
1666 :	Isibaar	1.44.2.2	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1667 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1668 :	chl	1.44.2.1	&frame->image,
1669 :			i, j,
1670 :			frame->motion_flags,
1671 :			TRB, TRD,
1672 :			pParam,
1673 :			pMB, b_mb,
1674 :	h	1.44.2.3	&best_sad,
1675 :			&Data);
1676 :	edgomez	1.20
1677 :	Isibaar	1.44.2.2	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1678 :	chl	1.3
1679 :	chl	1.44.2.1	// forward search
1680 :			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1681 :			&frame->image, i, j,
1682 :			frame->motion_flags,
1683 :	h	1.44.2.3	frame->fcode, pParam,
1684 :	chl	1.44.2.1	pMB, &f_predMV, &best_sad,
1685 :	h	1.44.2.3	MODE_FORWARD, &Data);
1686 :	chl	1.10
1687 :	chl	1.44.2.1	// backward search
1688 :			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1689 :			&frame->image, i, j,
1690 :			frame->motion_flags,
1691 :	h	1.44.2.3	frame->bcode, pParam,
1692 :	chl	1.44.2.1	pMB, &b_predMV, &best_sad,
1693 :	h	1.44.2.3	MODE_BACKWARD, &Data);
1694 :	chl	1.10
1695 :	chl	1.44.2.1	// interpolate search comes last, because it uses data from forward and backward as prediction
1696 :	edgomez	1.20
1697 :	chl	1.44.2.1	SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1698 :			b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1699 :			&frame->image,
1700 :			i, j,
1701 :			frame->fcode, frame->bcode,
1702 :			frame->motion_flags,
1703 :	h	1.44.2.3	pParam,
1704 :	chl	1.44.2.1	&f_predMV, &b_predMV,
1705 :	h	1.44.2.3	pMB, &best_sad,
1706 :			&Data);
1707 :	edgomez	1.20
1708 :	chl	1.44.2.1	switch (pMB->mode) {
1709 :			case MODE_FORWARD:
1710 :			f_count++;
1711 :			f_predMV = pMB->mvs[0];
1712 :			break;
1713 :			case MODE_BACKWARD:
1714 :			b_count++;
1715 :			b_predMV = pMB->b_mvs[0];
1716 :			break;
1717 :			case MODE_INTERPOLATE:
1718 :			i_count++;
1719 :			f_predMV = pMB->mvs[0];
1720 :			b_predMV = pMB->b_mvs[0];
1721 :			break;
1722 :			case MODE_DIRECT:
1723 :			case MODE_DIRECT_NO4V:
1724 :			d_count++;
1725 :			break;
1726 :			default:
1727 :			break;
1728 :	chl	1.10	}
1729 :			}
1730 :			}
1731 :	chl	1.44.2.1	}
1732 :	chl	1.3
1733 :	chl	1.44.2.1	/* Hinted ME starts here */
1734 :	chl	1.18
1735 :	syskin	1.44.2.8	static void
1736 :	syskin	1.44.2.14	SearchPhinted ( const IMAGE * const pRef,
1737 :	chl	1.30	const uint8_t * const pRefH,
1738 :			const uint8_t * const pRefV,
1739 :			const uint8_t * const pRefHV,
1740 :			const IMAGE * const pCur,
1741 :			const int x,
1742 :			const int y,
1743 :			const uint32_t MotionFlags,
1744 :			const uint32_t iQuant,
1745 :			const MBParam * const pParam,
1746 :			const MACROBLOCK * const pMBs,
1747 :	chl	1.44.2.1	int inter4v,
1748 :	h	1.44.2.3	MACROBLOCK * const pMB,
1749 :	edgomez	1.44.2.4	SearchData * const Data)
1750 :	chl	1.44.2.1	{
1751 :	chl	1.30
1752 :	edgomez	1.44.2.4	int i, t;
1753 :	chl	1.44.2.1	MainSearchFunc * MainSearchPtr;
1754 :
1755 :	h	1.44.2.3	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1756 :	syskin	1.44.2.14	Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1757 :	h	1.44.2.3	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1758 :	Isibaar	1.44.2.7	pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
1759 :	chl	1.44.2.1
1760 :	syskin	1.44.2.14	Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
1761 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8;
1762 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8;
1763 :
1764 :			Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16;
1765 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16;
1766 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16;
1767 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16;
1768 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
1769 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
1770 :	edgomez	1.44.2.4
1771 :	chl	1.44.2.1	if (!(MotionFlags & PMV_HALFPEL16)) {
1772 :	h	1.44.2.3	Data->min_dx = EVEN(Data->min_dx);
1773 :			Data->max_dx = EVEN(Data->max_dx);
1774 :			Data->min_dy = EVEN(Data->min_dy);
1775 :			Data->max_dy = EVEN(Data->max_dy);
1776 :	chl	1.30	}
1777 :
1778 :	h	1.44.2.3	for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1779 :	chl	1.30
1780 :	chl	1.44.2.1	if (pMB->dquant != NO_CHANGE) inter4v = 0;
1781 :	chl	1.30
1782 :	syskin	1.44.2.14	if (inter4v \|\| pParam->m_quarterpel \|\| Data->chroma) CheckCandidate = CheckCandidate16;
1783 :			else CheckCandidate = CheckCandidate16no4v;
1784 :	chl	1.30
1785 :	chl	1.44.2.1	pMB->mvs[0].x = EVEN(pMB->mvs[0].x);
1786 :			pMB->mvs[0].y = EVEN(pMB->mvs[0].y);
1787 :	h	1.44.2.3	if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed
1788 :			if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx;
1789 :			if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy;
1790 :			if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy;
1791 :	chl	1.30
1792 :	edgomez	1.44.2.4	(*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data);
1793 :	chl	1.30
1794 :	chl	1.44.2.1	if (pMB->mode == MODE_INTER4V)
1795 :			for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search
1796 :			pMB->mvs[i].x = EVEN(pMB->mvs[i].x);
1797 :			pMB->mvs[i].y = EVEN(pMB->mvs[i].y);
1798 :			if (!(make_mask(pMB->mvs, i)))
1799 :	edgomez	1.44.2.4	(*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data);
1800 :	chl	1.44.2.1	}
1801 :	chl	1.30
1802 :			if (MotionFlags & PMV_USESQUARES16)
1803 :	chl	1.44.2.1	MainSearchPtr = SquareSearch;
1804 :	chl	1.30	else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1805 :	chl	1.44.2.1	MainSearchPtr = AdvDiamondSearch;
1806 :			else MainSearchPtr = DiamondSearch;
1807 :	chl	1.30
1808 :	h	1.44.2.3	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1809 :	chl	1.30
1810 :	h	1.44.2.3	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1811 :	chl	1.30
1812 :	syskin	1.44.2.8	for(i = 0; i < 5; i++) {
1813 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1814 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1815 :			}
1816 :
1817 :			if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1818 :	Isibaar	1.44.2.12	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1819 :			pParam->width, pParam->height, Data->iFcode, 0);
1820 :			CheckCandidate = CheckCandidate16_qpel;
1821 :	syskin	1.44.2.8	QuarterpelRefine(Data);
1822 :			}
1823 :
1824 :			if (inter4v) {
1825 :			SearchData Data8;
1826 :			Data8.iFcode = Data->iFcode;
1827 :	Isibaar	1.44.2.12	Data8.lambda8 = Data->lambda8;
1828 :	syskin	1.44.2.8	Data8.iEdgedWidth = Data->iEdgedWidth;
1829 :	Isibaar	1.44.2.12	Data8.RefQ = Data->RefQ;
1830 :	syskin	1.44.2.14	Data8.qpel = Data->qpel;
1831 :			Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1832 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1833 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1834 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1835 :
1836 :			if (Data->chroma) {
1837 :			int sum, dx, dy;
1838 :
1839 :			if(pParam->m_quarterpel)
1840 :			sum = (pMB->qmvs[0].y + pMB->qmvs[1].y + pMB->qmvs[2].y + pMB->qmvs[3].y)/2;
1841 :			else sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1842 :			dy = (sum ? SIGN(sum) *
1843 :			(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0);
1844 :
1845 :			if(pParam->m_quarterpel)
1846 :			sum = (pMB->qmvs[0].x + pMB->qmvs[1].x + pMB->qmvs[2].x + pMB->qmvs[3].x)/2;
1847 :			else sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1848 :			dx = (sum ? SIGN(sum) *
1849 :			(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0);
1850 :			Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1851 :			}
1852 :	syskin	1.44.2.8	}
1853 :	chl	1.30
1854 :	chl	1.44.2.1	if (!(inter4v) \|\|
1855 :	h	1.44.2.3	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] +
1856 :			Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1857 :	chl	1.44.2.1	// INTER MODE
1858 :			pMB->mode = MODE_INTER;
1859 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1]
1860 :	h	1.44.2.3	= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1861 :	chl	1.30
1862 :	Isibaar	1.44.2.12	pMB->qmvs[0] = pMB->qmvs[1]
1863 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1864 :
1865 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1866 :	h	1.44.2.3	pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1867 :	chl	1.30
1868 :	Isibaar	1.44.2.12	if(pParam->m_quarterpel) {
1869 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1870 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1871 :	syskin	1.44.2.14	} else {
1872 :	Isibaar	1.44.2.12	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1873 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1874 :			}
1875 :	chl	1.44.2.1	} else {
1876 :	syskin	1.44.2.14	// INTER4V MODE; all other things are already set in Search8
1877 :	chl	1.44.2.1	pMB->mode = MODE_INTER4V;
1878 :	h	1.44.2.3	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3]
1879 :			+ Data->iMinSAD[4] + IMV16X16 * iQuant;
1880 :	chl	1.44.2.1	}
1881 :	chl	1.30
1882 :			}
1883 :	suxen_drol	1.6
1884 :	edgomez	1.20	void
1885 :	chl	1.44.2.1	MotionEstimationHinted( MBParam * const pParam,
1886 :			FRAMEINFO * const current,
1887 :			FRAMEINFO * const reference,
1888 :			const IMAGE * const pRefH,
1889 :			const IMAGE * const pRefV,
1890 :			const IMAGE * const pRefHV)
1891 :	suxen_drol	1.6	{
1892 :	chl	1.44.2.1	MACROBLOCK *const pMBs = current->mbs;
1893 :			const IMAGE *const pCurrent = &current->image;
1894 :			const IMAGE *const pRef = &reference->image;
1895 :	chl	1.32
1896 :	chl	1.44.2.1	uint32_t x, y;
1897 :	syskin	1.44.2.11	uint8_t * qimage;
1898 :	syskin	1.44.2.5	int32_t temp[5], quant = current->quant;
1899 :	h	1.44.2.3	int32_t iMinSAD[5];
1900 :	Isibaar	1.44.2.12	VECTOR currentMV[5], currentQMV[5];
1901 :	h	1.44.2.3	SearchData Data;
1902 :			Data.iEdgedWidth = pParam->edged_width;
1903 :			Data.currentMV = currentMV;
1904 :	Isibaar	1.44.2.12	Data.currentQMV = currentQMV;
1905 :	h	1.44.2.3	Data.iMinSAD = iMinSAD;
1906 :			Data.temp = temp;
1907 :			Data.iFcode = current->fcode;
1908 :	syskin	1.44.2.8	Data.rounding = pParam->m_rounding_type;
1909 :	syskin	1.44.2.14	Data.qpel = pParam->m_quarterpel;
1910 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
1911 :	syskin	1.44.2.8
1912 :			if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
1913 :			return; // allocate some mem for qpel interpolated blocks
1914 :			// somehow this is dirty since I think we shouldn't use malloc outside
1915 :			// encoder_create() - so please fix me!
1916 :
1917 :	syskin	1.44.2.11	Data.RefQ = qimage;
1918 :
1919 :	chl	1.44.2.1	if (sadInit) (*sadInit) ();
1920 :	chl	1.32
1921 :	chl	1.44.2.1	for (y = 0; y < pParam->mb_height; y++) {
1922 :			for (x = 0; x < pParam->mb_width; x++) {
1923 :	suxen_drol	1.6
1924 :	chl	1.44.2.1	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
1925 :	suxen_drol	1.6
1926 :	chl	1.44.2.1	//intra mode is copied from the first pass. At least for the time being
1927 :			if ((pMB->mode == MODE_INTRA) \|\| (pMB->mode == MODE_NOT_CODED) ) continue;
1928 :	chl	1.37
1929 :	chl	1.44.2.1	if (!(current->global_flags & XVID_LUMIMASKING)) {
1930 :			pMB->dquant = NO_CHANGE;
1931 :			pMB->quant = current->quant; }
1932 :	Isibaar	1.44.2.12	else {
1933 :	syskin	1.44.2.5	if (pMB->dquant != NO_CHANGE) {
1934 :			quant += DQtab[pMB->dquant];
1935 :			if (quant > 31) quant = 31;
1936 :			else if (quant < 1) quant = 1;
1937 :			}
1938 :	Isibaar	1.44.2.12	pMB->quant = quant;
1939 :			}
1940 :	chl	1.33
1941 :	syskin	1.44.2.14	SearchPhinted(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
1942 :	Isibaar	1.44.2.2	y, current->motion_flags, pMB->quant,
1943 :	h	1.44.2.3	pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
1944 :			&Data);
1945 :	edgomez	1.20
1946 :	Isibaar	1.44.2.2	}
1947 :			}
1948 :	syskin	1.44.2.8	free(qimage);
1949 :	Isibaar	1.44.2.2	}
1950 :	chl	1.33
1951 :	Isibaar	1.44.2.2	static __inline int
1952 :			MEanalyzeMB ( const uint8_t * const pRef,
1953 :			const uint8_t * const pCur,
1954 :			const int x,
1955 :			const int y,
1956 :			const MBParam * const pParam,
1957 :			const MACROBLOCK * const pMBs,
1958 :	h	1.44.2.3	MACROBLOCK * const pMB,
1959 :			SearchData * const Data)
1960 :	Isibaar	1.44.2.2	{
1961 :	chl	1.32
1962 :	Isibaar	1.44.2.12	int i = 255, mask;
1963 :	h	1.44.2.3	VECTOR pmv[3];
1964 :	suxen_drol	1.22
1965 :	h	1.44.2.3	*(Data->iMinSAD) = MV_MAX_ERROR;
1966 :			Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1967 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1968 :	Isibaar	1.44.2.7	pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
1969 :	chl	1.35
1970 :	h	1.44.2.3	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1971 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1972 :	Isibaar	1.44.2.2
1973 :			CheckCandidate = CheckCandidate16no4vI;
1974 :	chl	1.44.2.1
1975 :	Isibaar	1.44.2.2	pmv[1].x = EVEN(pMB->mvs[0].x);
1976 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1977 :	h	1.44.2.3	pmv[0].x = EVEN(Data->predMV.x);
1978 :			pmv[0].y = EVEN(Data->predMV.y);
1979 :	Isibaar	1.44.2.2	pmv[2].x = pmv[2].y = 0;
1980 :
1981 :	h	1.44.2.3	CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data);
1982 :	Isibaar	1.44.2.2	if (!(mask = make_mask(pmv, 1)))
1983 :	h	1.44.2.3	CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
1984 :	Isibaar	1.44.2.2	if (!(mask = make_mask(pmv, 2)))
1985 :	h	1.44.2.3	CheckCandidate16no4vI(0, 0, mask, &i, Data);
1986 :	Isibaar	1.44.2.2
1987 :	Isibaar	1.44.2.12	DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1988 :	Isibaar	1.44.2.2
1989 :	Isibaar	1.44.2.12	pMB->mvs[0] = *Data->currentMV;
1990 :	syskin	1.44.2.14	pMB->mode = MODE_INTER;
1991 :	Isibaar	1.44.2.2
1992 :	h	1.44.2.3	return *(Data->iMinSAD);
1993 :	Isibaar	1.44.2.2	}
1994 :
1995 :			#define INTRA_THRESH 1350
1996 :			#define INTER_THRESH 900
1997 :
1998 :			int
1999 :			MEanalysis( const IMAGE * const pRef,
2000 :			const IMAGE * const pCurrent,
2001 :			MBParam * const pParam,
2002 :			MACROBLOCK * const pMBs,
2003 :			const uint32_t iFcode)
2004 :			{
2005 :			uint32_t x, y, intra = 0;
2006 :			int sSAD = 0;
2007 :	h	1.44.2.3
2008 :			VECTOR currentMV;
2009 :			int32_t iMinSAD;
2010 :			SearchData Data;
2011 :			Data.iEdgedWidth = pParam->edged_width;
2012 :			Data.currentMV = &currentMV;
2013 :			Data.iMinSAD = &iMinSAD;
2014 :			Data.iFcode = iFcode;
2015 :
2016 :	Isibaar	1.44.2.2	if (sadInit) (*sadInit) ();
2017 :
2018 :	syskin	1.44.2.11	for (y = 1; y < pParam->mb_height-1; y++) {
2019 :			for (x = 1; x < pParam->mb_width-1; x++) {
2020 :	Isibaar	1.44.2.2	int sad, dev;
2021 :			MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
2022 :
2023 :			sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
2024 :	h	1.44.2.3	pParam, pMBs, pMB, &Data);
2025 :	Isibaar	1.44.2.2
2026 :	syskin	1.44.2.11	if (sad > INTRA_THRESH) {
2027 :			dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
2028 :			pParam->edged_width);
2029 :	syskin	1.44.2.14	if (dev + INTRA_THRESH < sad) { intra++; pMB->mode = MODE_INTRA; }
2030 :	syskin	1.44.2.11	if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame
2031 :	Isibaar	1.44.2.2	}
2032 :	syskin	1.44.2.11	sSAD += sad;
2033 :	suxen_drol	1.6	}
2034 :			}
2035 :	Isibaar	1.44.2.2	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2036 :			if (sSAD > INTER_THRESH ) return 1; //P frame
2037 :			emms();
2038 :			return 0; // B frame
2039 :	chl	1.44.2.1
2040 :	syskin	1.44.2.6	}
2041 :
2042 :			int
2043 :			FindFcode( const MBParam * const pParam,
2044 :			const FRAMEINFO * const current)
2045 :			{
2046 :			uint32_t x, y;
2047 :			int max = 0, min = 0, i;
2048 :
2049 :			for (y = 0; y < pParam->mb_height; y++) {
2050 :			for (x = 0; x < pParam->mb_width; x++) {
2051 :
2052 :			MACROBLOCK pMB = &current->mbs[x + y pParam->mb_width];
2053 :			for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) {
2054 :			if (pMB->mvs[i].x > max) max = pMB->mvs[i].x;
2055 :			if (pMB->mvs[i].y > max) max = pMB->mvs[i].y;
2056 :
2057 :			if (pMB->mvs[i].x < min) min = pMB->mvs[i].x;
2058 :			if (pMB->mvs[i].y < min) min = pMB->mvs[i].y;
2059 :			}
2060 :			}
2061 :			}
2062 :
2063 :			min = -min;
2064 :			max += 1;
2065 :			if (min > max) max = min;
2066 :	Isibaar	1.44.2.12	if (pParam->m_quarterpel) max *= 2;
2067 :	syskin	1.44.2.6
2068 :			for (i = 1; (max > 32 << (i - 1)); i++);
2069 :			return i;
2070 :	chl	1.44.2.13	}
2071 :
2072 :	syskin	1.44.2.14	static void
2073 :			CheckGMC(int x, int y, const int dir, int * iDirection,
2074 :			const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
2075 :			const MBParam * const pParam)
2076 :			{
2077 :			uint32_t mx, my, a, count = 0;
2078 :
2079 :			for (my = 1; my < pParam->mb_height-1; my++)
2080 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2081 :			VECTOR mv;
2082 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2083 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED) continue;
2084 :			mv = pMB->mvs[0];
2085 :			a = ABS(mv.x - x) + ABS(mv.y - y);
2086 :			if (a < 6) count += 6 - a;
2087 :			}
2088 :
2089 :			if (count > *bestcount) {
2090 :			*bestcount = count;
2091 :			*iDirection = dir;
2092 :			GMC->x = x; GMC->y = y;
2093 :			}
2094 :			}
2095 :	chl	1.44.2.13
2096 :
2097 :			static VECTOR
2098 :			GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2099 :			{
2100 :
2101 :	syskin	1.44.2.14	uint32_t count, bestcount = 0;
2102 :	chl	1.44.2.13	int x, y;
2103 :	syskin	1.44.2.14	VECTOR gmc = {0,0};
2104 :	chl	1.44.2.13	int step, min_x, max_x, min_y, max_y;
2105 :			uint32_t mx, my;
2106 :	syskin	1.44.2.14	int iDirection, bDirection;
2107 :	chl	1.44.2.13
2108 :			min_x = min_y = -32<<iFcode;
2109 :			max_x = max_y = 32<<iFcode;
2110 :
2111 :	syskin	1.44.2.14	//step1: let's find a rough camera panning
2112 :	chl	1.44.2.13	for (step = 32; step >= 2; step /= 2) {
2113 :			bestcount = 0;
2114 :			for (y = min_y; y <= max_y; y += step)
2115 :			for (x = min_x ; x <= max_x; x += step) {
2116 :			count = 0;
2117 :			//for all macroblocks
2118 :			for (my = 1; my < pParam->mb_height-1; my++)
2119 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2120 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2121 :			VECTOR mv;
2122 :
2123 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2124 :			continue;
2125 :
2126 :			mv = pMB->mvs[0];
2127 :			if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2128 :			count++;
2129 :			}
2130 :			if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2131 :			}
2132 :			min_x = gmc.x - step;
2133 :			max_x = gmc.x + step;
2134 :			min_y = gmc.y - step;
2135 :			max_y = gmc.y + step;
2136 :
2137 :			}
2138 :	syskin	1.44.2.14
2139 :			if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2140 :			gmc.x = gmc.y = 0; //no camara pan, no GMC
2141 :
2142 :			// step2: let's refine camera panning using gradiend-descent approach.
2143 :			// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
2144 :			bestcount = 0;
2145 :			CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
2146 :			do {
2147 :			x = gmc.x; y = gmc.y;
2148 :			bDirection = iDirection; iDirection = 0;
2149 :			if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2150 :			if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2151 :			if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
2152 :			if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2153 :
2154 :			} while (iDirection);
2155 :
2156 :	chl	1.44.2.13	if (pParam->m_quarterpel) {
2157 :			gmc.x *= 2;
2158 :			gmc.y = 2; / we store the halfpel value as pseudo-qpel to make comparison easier */
2159 :			}
2160 :
2161 :			return gmc;
2162 :	Isibaar	1.44.2.2	}

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