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

Revision 1.44.2.40 - (view) (download)

1 :	Isibaar	1.1	/**************************************************************************
2 :			*
3 :	chl	1.26	* XVID MPEG-4 VIDEO CODEC
4 :	syskin	1.44.2.40	* motion estimation
5 :	chl	1.26	*
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 :	suxen_drol	1.44.2.35	#include <string.h> // memcpy
35 :	Isibaar	1.1
36 :			#include "../encoder.h"
37 :			#include "../utils/mbfunctions.h"
38 :			#include "../prediction/mbprediction.h"
39 :			#include "../global.h"
40 :			#include "../utils/timer.h"
41 :	Isibaar	1.44.2.7	#include "../image/interpolate8x8.h"
42 :	chl	1.44.2.1	#include "motion_est.h"
43 :	suxen_drol	1.6	#include "motion.h"
44 :	Isibaar	1.1	#include "sad.h"
45 :	Isibaar	1.44.2.2	#include "../utils/emms.h"
46 :	Isibaar	1.1
47 :	chl	1.44.2.1	#define INITIAL_SKIP_THRESH (10)
48 :			#define FINAL_SKIP_THRESH (50)
49 :			#define MAX_SAD00_FOR_SKIP (20)
50 :	Isibaar	1.44.2.2	#define MAX_CHROMA_SAD_FOR_SKIP (22)
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 :	chl	1.44.2.1	static __inline int
56 :	syskin	1.44.2.36	d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
57 :	chl	1.44.2.1	{
58 :			int xb, yb;
59 :	syskin	1.44.2.34	if (qpel) { x = 2; y = 2;}
60 :			else if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
61 :	syskin	1.44.2.40	x -= pred.x;
62 :			y -= pred.y;
63 :	chl	1.44.2.1
64 :	syskin	1.44.2.40	if (x) {
65 :	chl	1.44.2.1	if (x < 0) x = -x;
66 :			x += (1 << (iFcode - 1)) - 1;
67 :			x >>= (iFcode - 1);
68 :			if (x > 32) x = 32;
69 :			xb = mvtab[x] + iFcode;
70 :	syskin	1.44.2.40	} else xb = 1;
71 :	chl	1.44.2.1
72 :	syskin	1.44.2.40	if (y) {
73 :	chl	1.44.2.1	if (y < 0) y = -y;
74 :			y += (1 << (iFcode - 1)) - 1;
75 :			y >>= (iFcode - 1);
76 :			if (y > 32) y = 32;
77 :			yb = mvtab[y] + iFcode;
78 :	syskin	1.44.2.40	} else yb = 1;
79 :	chl	1.44.2.1	return xb + yb;
80 :			}
81 :
82 :	syskin	1.44.2.40	static int32_t
83 :	syskin	1.44.2.14	ChromaSAD(int dx, int dy, const SearchData * const data)
84 :	chl	1.44.2.1	{
85 :	syskin	1.44.2.14	int sad;
86 :	syskin	1.44.2.40	const uint32_t stride = data->iEdgedWidth/2;
87 :	syskin	1.44.2.14	dx = (dx >> 1) + roundtab_79[dx & 0x3];
88 :			dy = (dy >> 1) + roundtab_79[dy & 0x3];
89 :
90 :	syskin	1.44.2.34	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
91 :	syskin	1.44.2.40	data->temp[5] = dx; data->temp[6] = dy; // backup
92 :	syskin	1.44.2.34
93 :			switch (((dx & 1) << 1) \| (dy & 1)) {
94 :	syskin	1.44.2.14	case 0:
95 :	syskin	1.44.2.40	dx = dx / 2; dy = dy / 2;
96 :			sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride);
97 :			sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride);
98 :	syskin	1.44.2.14	break;
99 :			case 1:
100 :			dx = dx / 2; dy = (dy - 1) / 2;
101 :	syskin	1.44.2.40	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride);
102 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride);
103 :	syskin	1.44.2.14	break;
104 :			case 2:
105 :			dx = (dx - 1) / 2; dy = dy / 2;
106 :	syskin	1.44.2.40	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride);
107 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride);
108 :	syskin	1.44.2.14	break;
109 :			default:
110 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
111 :	syskin	1.44.2.40	interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding);
112 :			sad = sad8(data->CurU, data->RefQ, stride);
113 :
114 :			interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding);
115 :			sad += sad8(data->CurV, data->RefQ, stride);
116 :	syskin	1.44.2.14	break;
117 :	chl	1.44.2.1	}
118 :	syskin	1.44.2.40	data->temp[7] = sad; //backup, part 2
119 :	syskin	1.44.2.14	return sad;
120 :			}
121 :	chl	1.10
122 :	syskin	1.44.2.22	static __inline const uint8_t *
123 :	syskin	1.44.2.40	GetReferenceB(const int x, const int y, const int dir, const SearchData * const data)
124 :	Isibaar	1.44.2.12	{
125 :	syskin	1.44.2.22	// dir : 0 = forward, 1 = backward
126 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
127 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
128 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
129 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
130 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
131 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
132 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
133 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
134 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
135 :	Isibaar	1.44.2.12	}
136 :			}
137 :
138 :	syskin	1.44.2.40	// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate
139 :			static __inline const uint8_t *
140 :			GetReference(const int x, const int y, const SearchData * const data)
141 :			{
142 :			switch ( ((x&1)<<1) \| (y&1) ) {
143 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
144 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
145 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
146 :			default : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
147 :			}
148 :			}
149 :
150 :			static uint8_t *
151 :	syskin	1.44.2.22	Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data)
152 :	Isibaar	1.1	{
153 :	syskin	1.44.2.22	// create or find a qpel-precision reference picture; return pointer to it
154 :	syskin	1.44.2.24	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
155 :	syskin	1.44.2.22	const int32_t iEdgedWidth = data->iEdgedWidth;
156 :			const uint32_t rounding = data->rounding;
157 :			const int halfpel_x = x/2;
158 :			const int halfpel_y = y/2;
159 :			const uint8_t ref1, ref2, ref3, ref4;
160 :	Isibaar	1.1
161 :	syskin	1.44.2.40	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
162 :	syskin	1.44.2.22	ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
163 :			switch( ((x&1)<<1) + (y&1) ) {
164 :			case 0: // pure halfpel position
165 :	syskin	1.44.2.38	return (uint8_t *) ref1;
166 :	syskin	1.44.2.22	break;
167 :	Isibaar	1.1
168 :	syskin	1.44.2.22	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
169 :	syskin	1.44.2.40	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
170 :	syskin	1.44.2.22	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
171 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
172 :	syskin	1.44.2.22	break;
173 :	Isibaar	1.1
174 :	syskin	1.44.2.22	case 2: // x qpel, y halfpel - left or right during qpel refinement
175 :	syskin	1.44.2.40	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
176 :	syskin	1.44.2.22	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
177 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
178 :	syskin	1.44.2.22	break;
179 :	Isibaar	1.1
180 :	syskin	1.44.2.22	default: // x and y in qpel resolution - the "corners" (top left/right and
181 :			// bottom left/right) during qpel refinement
182 :	syskin	1.44.2.40	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
183 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
184 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
185 :	syskin	1.44.2.22	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
186 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
187 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
188 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
189 :			break;
190 :			}
191 :			return Reference;
192 :	Isibaar	1.1	}
193 :
194 :	syskin	1.44.2.40	static uint8_t *
195 :	syskin	1.44.2.22	Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data)
196 :	Isibaar	1.44.2.7	{
197 :	syskin	1.44.2.22	// create or find a qpel-precision reference picture; return pointer to it
198 :	syskin	1.44.2.24	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
199 :	syskin	1.44.2.22	const int32_t iEdgedWidth = data->iEdgedWidth;
200 :			const uint32_t rounding = data->rounding;
201 :			const int halfpel_x = x/2;
202 :			const int halfpel_y = y/2;
203 :	Isibaar	1.44.2.7	const uint8_t ref1, ref2, ref3, ref4;
204 :
205 :	syskin	1.44.2.40	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
206 :	syskin	1.44.2.22	switch( ((x&1)<<1) + (y&1) ) {
207 :			case 0: // pure halfpel position
208 :	syskin	1.44.2.38	return (uint8_t *) ref1;
209 :	Isibaar	1.44.2.7	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
210 :	syskin	1.44.2.40	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
211 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
212 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
213 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
214 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
215 :	Isibaar	1.44.2.7	break;
216 :
217 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
218 :	syskin	1.44.2.40	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
219 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
220 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
221 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
222 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
223 :	Isibaar	1.44.2.7	break;
224 :
225 :			default: // x and y in qpel resolution - the "corners" (top left/right and
226 :			// bottom left/right) during qpel refinement
227 :	syskin	1.44.2.40	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
228 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
229 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
230 :	Isibaar	1.44.2.7	interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
231 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
232 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
233 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
234 :			break;
235 :			}
236 :	syskin	1.44.2.22	return Reference;
237 :			}
238 :
239 :			/* CHECK_CANDIATE FUNCTIONS START */
240 :
241 :	syskin	1.44.2.40	static void
242 :	syskin	1.44.2.22	CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
243 :			{
244 :			int t, xc, yc;
245 :			const uint8_t * Reference;
246 :			VECTOR * current;
247 :
248 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
249 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
250 :
251 :			if (data->qpel_precision) { // x and y are in 1/4 precision
252 :			Reference = Interpolate16x16qpel(x, y, 0, data);
253 :			xc = x/2; yc = y/2; //for chroma sad
254 :			current = data->currentQMV;
255 :			} else {
256 :	syskin	1.44.2.40	Reference = GetReference(x, y, data);
257 :	syskin	1.44.2.22	current = data->currentMV;
258 :			xc = x; yc = y;
259 :			}
260 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
261 :	Isibaar	1.44.2.7
262 :	syskin	1.44.2.22	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
263 :	Isibaar	1.44.2.7
264 :	syskin	1.44.2.14	data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
265 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
266 :
267 :	syskin	1.44.2.22	if (data->chroma) data->temp[0] += ChromaSAD(xc, yc, data);
268 :	Isibaar	1.44.2.7
269 :			if (data->temp[0] < data->iMinSAD[0]) {
270 :			data->iMinSAD[0] = data->temp[0];
271 :	syskin	1.44.2.22	current[0].x = x; current[0].y = y;
272 :			*dir = Direction; }
273 :	Isibaar	1.44.2.7
274 :			if (data->temp[1] < data->iMinSAD[1]) {
275 :	syskin	1.44.2.22	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; }
276 :	Isibaar	1.44.2.7	if (data->temp[2] < data->iMinSAD[2]) {
277 :	syskin	1.44.2.22	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
278 :	Isibaar	1.44.2.7	if (data->temp[3] < data->iMinSAD[3]) {
279 :	syskin	1.44.2.22	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
280 :	Isibaar	1.44.2.7	if (data->temp[4] < data->iMinSAD[4]) {
281 :	syskin	1.44.2.22	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
282 :
283 :	Isibaar	1.44.2.7	}
284 :
285 :	syskin	1.44.2.40	static void
286 :	syskin	1.44.2.32	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
287 :			{
288 :			int t;
289 :			const uint8_t * Reference;
290 :
291 :			if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) \|\| //non-zero integer value
292 :			( x > data->max_dx) \|\| ( x < data->min_dx)
293 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
294 :
295 :	syskin	1.44.2.40	Reference = GetReference(x, y, data);
296 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
297 :	syskin	1.44.2.32
298 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
299 :
300 :			data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
301 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
302 :
303 :			if (data->temp[0] < data->iMinSAD[0]) {
304 :			data->iMinSAD[0] = data->temp[0];
305 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
306 :			*dir = Direction; }
307 :
308 :			if (data->temp[1] < data->iMinSAD[1]) {
309 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
310 :			if (data->temp[2] < data->iMinSAD[2]) {
311 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
312 :			if (data->temp[3] < data->iMinSAD[3]) {
313 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
314 :			if (data->temp[4] < data->iMinSAD[4]) {
315 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
316 :			}
317 :
318 :	syskin	1.44.2.40	static void
319 :	syskin	1.44.2.22	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
320 :	syskin	1.44.2.19	{
321 :			int32_t sad;
322 :	syskin	1.44.2.22	const uint8_t * Reference;
323 :			int t;
324 :			VECTOR * current;
325 :	syskin	1.44.2.19
326 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
327 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
328 :
329 :	syskin	1.44.2.34	if (data->rrv)
330 :	syskin	1.44.2.32	if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) ) return; //non-zero integer value
331 :
332 :	syskin	1.44.2.22	if (data->qpel_precision) { // x and y are in 1/4 precision
333 :			Reference = Interpolate16x16qpel(x, y, 0, data);
334 :			current = data->currentQMV;
335 :			} else {
336 :	syskin	1.44.2.40	Reference = GetReference(x, y, data);
337 :	syskin	1.44.2.22	current = data->currentMV;
338 :	syskin	1.44.2.19	}
339 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode,
340 :	syskin	1.44.2.40	data->qpel && !data->qpel_precision, data->rrv);
341 :	syskin	1.44.2.19
342 :			sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
343 :	syskin	1.44.2.22	sad += (data->lambda16 * t * sad)/1000;
344 :	syskin	1.44.2.19
345 :	syskin	1.44.2.22	if (sad < *(data->iMinSAD)) {
346 :			*(data->iMinSAD) = sad;
347 :			current->x = x; current->y = y;
348 :			*dir = Direction; }
349 :	syskin	1.44.2.19	}
350 :
351 :	syskin	1.44.2.40	static void
352 :	syskin	1.44.2.37	CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
353 :	Isibaar	1.44.2.2	{
354 :	syskin	1.44.2.19	// maximum speed - for P/B/I decision
355 :	Isibaar	1.44.2.2
356 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
357 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
358 :
359 :	syskin	1.44.2.37	data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
360 :			data->iEdgedWidth, data->temp+1);
361 :			if (data->temp[0] < *(data->iMinSAD)) {
362 :			*(data->iMinSAD) = data->temp[0];
363 :	Isibaar	1.44.2.2	data->currentMV[0].x = x; data->currentMV[0].y = y;
364 :			*dir = Direction; }
365 :	syskin	1.44.2.37	if (data->temp[1] < data->iMinSAD[1]) {
366 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
367 :			if (data->temp[2] < data->iMinSAD[2]) {
368 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
369 :			if (data->temp[3] < data->iMinSAD[3]) {
370 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
371 :			if (data->temp[4] < data->iMinSAD[4]) {
372 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
373 :
374 :	Isibaar	1.44.2.2	}
375 :
376 :	syskin	1.44.2.40	static void
377 :	chl	1.44.2.1	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
378 :	Isibaar	1.1	{
379 :	chl	1.44.2.1	int32_t sad;
380 :	syskin	1.44.2.22	int xb, yb, t;
381 :	chl	1.44.2.1	const uint8_t ReferenceF, ReferenceB;
382 :	syskin	1.44.2.22	VECTOR *current;
383 :	suxen_drol	1.8
384 :	chl	1.44.2.1	if (( xf > data->max_dx) \|\| ( xf < data->min_dx)
385 :			\|\| ( yf > data->max_dy) \|\| (yf < data->min_dy)) return;
386 :	suxen_drol	1.8
387 :	syskin	1.44.2.22	if (data->qpel_precision) {
388 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
389 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
390 :			current = data->currentQMV;
391 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
392 :			} else {
393 :	syskin	1.44.2.40	ReferenceF = GetReference(xf, yf, data);
394 :	syskin	1.44.2.22	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
395 :	syskin	1.44.2.40	ReferenceB = GetReferenceB(xb, yb, 1, data);
396 :	syskin	1.44.2.22	current = data->currentMV;
397 :	syskin	1.44.2.19	}
398 :
399 :	syskin	1.44.2.36	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0)
400 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
401 :	syskin	1.44.2.34
402 :	syskin	1.44.2.19	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
403 :	syskin	1.44.2.22	sad += (data->lambda16 * t * sad)/1000;
404 :	syskin	1.44.2.19
405 :			if (sad < *(data->iMinSAD)) {
406 :			*(data->iMinSAD) = sad;
407 :	syskin	1.44.2.22	current->x = xf; current->y = yf;
408 :	syskin	1.44.2.19	*dir = Direction; }
409 :			}
410 :
411 :	syskin	1.44.2.40	static void
412 :	chl	1.44.2.1	CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
413 :			{
414 :	syskin	1.44.2.14	int32_t sad = 0;
415 :	chl	1.44.2.1	int k;
416 :			const uint8_t *ReferenceF;
417 :			const uint8_t *ReferenceB;
418 :			VECTOR mvs, b_mvs;
419 :	syskin	1.44.2.36	const VECTOR zeroMV={0,0};
420 :	edgomez	1.20
421 :	chl	1.44.2.1	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
422 :	edgomez	1.20
423 :	chl	1.44.2.1	for (k = 0; k < 4; k++) {
424 :			mvs.x = data->directmvF[k].x + x;
425 :			b_mvs.x = ((x == 0) ?
426 :			data->directmvB[k].x
427 :			: mvs.x - data->referencemv[k].x);
428 :	edgomez	1.20
429 :	chl	1.44.2.1	mvs.y = data->directmvF[k].y + y;
430 :			b_mvs.y = ((y == 0) ?
431 :			data->directmvB[k].y
432 :			: mvs.y - data->referencemv[k].y);
433 :
434 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
435 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
436 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
437 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
438 :
439 :	syskin	1.44.2.40	if (!data->qpel) {
440 :			mvs.x = 2; mvs.y = 2;
441 :	syskin	1.44.2.22	b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
442 :	chl	1.44.2.1	}
443 :	syskin	1.44.2.22	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
444 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
445 :	chl	1.44.2.1
446 :			sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
447 :	syskin	1.44.2.22	ReferenceF, ReferenceB,
448 :	chl	1.44.2.1	data->iEdgedWidth);
449 :			if (sad > *(data->iMinSAD)) return;
450 :			}
451 :	chl	1.17
452 :	syskin	1.44.2.36	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000;
453 :	syskin	1.44.2.14
454 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
455 :			*(data->iMinSAD) = sad;
456 :			data->currentMV->x = x; data->currentMV->y = y;
457 :			*dir = Direction; }
458 :	Isibaar	1.1	}
459 :
460 :	syskin	1.44.2.40	static void
461 :	chl	1.44.2.1	CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
462 :			{
463 :			int32_t sad;
464 :			const uint8_t *ReferenceF;
465 :			const uint8_t *ReferenceB;
466 :			VECTOR mvs, b_mvs;
467 :	syskin	1.44.2.36	const VECTOR zeroMV = {0,0};
468 :	chl	1.35
469 :	syskin	1.44.2.6	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
470 :	Isibaar	1.1
471 :	chl	1.44.2.1	mvs.x = data->directmvF[0].x + x;
472 :			b_mvs.x = ((x == 0) ?
473 :			data->directmvB[0].x
474 :			: mvs.x - data->referencemv[0].x);
475 :
476 :			mvs.y = data->directmvF[0].y + y;
477 :			b_mvs.y = ((y == 0) ?
478 :			data->directmvB[0].y
479 :			: mvs.y - data->referencemv[0].y);
480 :
481 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
482 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
483 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
484 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
485 :
486 :	syskin	1.44.2.40	if (!data->qpel) {
487 :			mvs.x = 2; mvs.y = 2;
488 :	syskin	1.44.2.22	b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
489 :	syskin	1.44.2.34	}
490 :	syskin	1.44.2.22	ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
491 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
492 :	chl	1.44.2.1
493 :	syskin	1.44.2.14	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
494 :	syskin	1.44.2.36	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000;
495 :	Isibaar	1.1
496 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
497 :			*(data->iMinSAD) = sad;
498 :			data->currentMV->x = x; data->currentMV->y = y;
499 :			*dir = Direction; }
500 :			}
501 :	Isibaar	1.1
502 :	chl	1.44.2.1	static void
503 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
504 :			{
505 :	syskin	1.44.2.14	int32_t sad; int t;
506 :	chl	1.44.2.1	const uint8_t * Reference;
507 :	Isibaar	1.1
508 :	chl	1.44.2.1	if (( x > data->max_dx) \|\| ( x < data->min_dx)
509 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
510 :	Isibaar	1.1
511 :	syskin	1.44.2.22	if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data);
512 :	syskin	1.44.2.40	else Reference = GetReference(x, y, data);
513 :	Isibaar	1.1
514 :	chl	1.44.2.1	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
515 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
516 :	Isibaar	1.44.2.12
517 :	syskin	1.44.2.14	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100;
518 :	Isibaar	1.44.2.12
519 :			if (sad < *(data->iMinSAD)) {
520 :			*(data->iMinSAD) = sad;
521 :			data->currentMV->x = x; data->currentMV->y = y;
522 :			*dir = Direction; }
523 :			}
524 :
525 :	Isibaar	1.44.2.2	/* CHECK_CANDIATE FUNCTIONS END */
526 :	chl	1.3
527 :	chl	1.44.2.1	/* MAINSEARCH FUNCTIONS START */
528 :	chl	1.18
529 :	chl	1.44.2.1	static void
530 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
531 :			{
532 :	chl	1.18
533 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
534 :
535 :	syskin	1.44.2.34	int iDirection;
536 :
537 :			for(;;) { //forever
538 :			iDirection = 0;
539 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
540 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
541 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
542 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
543 :	edgomez	1.20
544 :	syskin	1.44.2.34	/* now we're doing diagonal checks near our candidate */
545 :
546 :			if (iDirection) { //checking if anything found
547 :			bDirection = iDirection;
548 :	chl	1.18	iDirection = 0;
549 :	syskin	1.44.2.34	x = data->currentMV->x; y = data->currentMV->y;
550 :			if (bDirection & 3) { //our candidate is left or right
551 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
552 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
553 :			} else { // what remains here is up or down
554 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
555 :	syskin	1.44.2.40	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
556 :	syskin	1.44.2.34	}
557 :
558 :			if (iDirection) {
559 :			bDirection += iDirection;
560 :	syskin	1.44.2.40	x = data->currentMV->x; y = data->currentMV->y;
561 :	syskin	1.44.2.34	}
562 :			} else { //about to quit, eh? not so fast....
563 :			switch (bDirection) {
564 :			case 2:
565 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
566 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
567 :			break;
568 :			case 1:
569 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
570 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
571 :			break;
572 :			case 2 + 4:
573 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
574 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
575 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
576 :			break;
577 :			case 4:
578 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
579 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
580 :			break;
581 :			case 8:
582 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
583 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
584 :			break;
585 :			case 1 + 4:
586 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
587 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
588 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
589 :			break;
590 :			case 2 + 8:
591 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
592 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
593 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
594 :			break;
595 :			case 1 + 8:
596 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
597 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
598 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
599 :			break;
600 :			default: //1+2+4+8 == we didn't find anything at all
601 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
602 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
603 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
604 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
605 :			break;
606 :	chl	1.18	}
607 :	syskin	1.44.2.34	if (!iDirection) break; //ok, the end. really
608 :			bDirection = iDirection;
609 :			x = data->currentMV->x; y = data->currentMV->y;
610 :	chl	1.18	}
611 :	syskin	1.44.2.34	}
612 :	chl	1.18	}
613 :
614 :	chl	1.44.2.1	static void
615 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
616 :			{
617 :			int iDirection;
618 :	chl	1.35
619 :	chl	1.44.2.1	do {
620 :			iDirection = 0;
621 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
622 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
623 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
624 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
625 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
626 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
627 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
628 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
629 :
630 :			bDirection = iDirection;
631 :			x = data->currentMV->x; y = data->currentMV->y;
632 :			} while (iDirection);
633 :	chl	1.35	}
634 :
635 :	chl	1.44.2.1	static void
636 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
637 :	chl	1.18	{
638 :
639 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
640 :
641 :	syskin	1.44.2.34	int iDirection;
642 :	edgomez	1.20
643 :	syskin	1.44.2.34	do {
644 :			iDirection = 0;
645 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
646 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
647 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
648 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
649 :	chl	1.18
650 :	syskin	1.44.2.34	/* now we're doing diagonal checks near our candidate */
651 :
652 :			if (iDirection) { //checking if anything found
653 :			bDirection = iDirection;
654 :			iDirection = 0;
655 :			x = data->currentMV->x; y = data->currentMV->y;
656 :			if (bDirection & 3) { //our candidate is left or right
657 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
658 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
659 :			} else { // what remains here is up or down
660 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
661 :	syskin	1.44.2.40	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
662 :	chl	1.18	}
663 :	syskin	1.44.2.34	bDirection += iDirection;
664 :			x = data->currentMV->x; y = data->currentMV->y;
665 :	chl	1.18	}
666 :	syskin	1.44.2.34	}
667 :			while (iDirection);
668 :	chl	1.18	}
669 :
670 :	chl	1.44.2.1	/* MAINSEARCH FUNCTIONS END */
671 :
672 :			/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
673 :	chl	1.18
674 :	chl	1.44.2.1	static void
675 :	syskin	1.44.2.22	SubpelRefine(const SearchData * const data)
676 :	Isibaar	1.1	{
677 :	syskin	1.44.2.22	/* Do a half-pel or q-pel refinement */
678 :			VECTOR backupMV;
679 :	chl	1.44.2.1	int iDirection; //not needed
680 :	edgomez	1.20
681 :	syskin	1.44.2.22	if (data->qpel_precision)
682 :			backupMV = *(data->currentQMV);
683 :			else backupMV = *(data->currentMV);
684 :	Isibaar	1.44.2.7
685 :	syskin	1.44.2.40	CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
686 :	Isibaar	1.44.2.7	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
687 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
688 :	syskin	1.44.2.40	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
689 :	Isibaar	1.44.2.7	CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
690 :	syskin	1.44.2.40	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
691 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
692 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
693 :	Isibaar	1.44.2.7	}
694 :
695 :	chl	1.44.2.1	static __inline int
696 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
697 :			const int x, const int y,
698 :	syskin	1.44.2.32	const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv)
699 :	Isibaar	1.1
700 :	chl	1.44.2.1	{
701 :			/* keep repeating checks for all b-frames before this P frame,
702 :			to make sure that SKIP is possible (todo)
703 :			how: if skip is not possible set sad00 to a very high value */
704 :	syskin	1.44.2.32	if(rrv) {
705 :			uint32_t sadC = sad16(current->u + x16 + y(iEdgedWidth/2)*16,
706 :			reference->u + x16 + y(iEdgedWidth/2)16, iEdgedWidth/2, 2564096);
707 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
708 :			sadC += sad16(current->v + (x + y(iEdgedWidth/2))16,
709 :			reference->v + (x + y(iEdgedWidth/2))16, iEdgedWidth/2, 256*4096);
710 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
711 :			return 1;
712 :			} else {
713 :			uint32_t sadC = sad8(current->u + x8 + y(iEdgedWidth/2)*8,
714 :			reference->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2);
715 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
716 :			sadC += sad8(current->v + (x + y(iEdgedWidth/2))8,
717 :			reference->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
718 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
719 :			return 1;
720 :			}
721 :	chl	1.44.2.1	}
722 :	Isibaar	1.1
723 :	chl	1.44.2.1	static __inline void
724 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
725 :			{
726 :			pMB->mode = MODE_NOT_CODED;
727 :	Isibaar	1.44.2.2	pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
728 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
729 :	syskin	1.44.2.14
730 :	Isibaar	1.44.2.7	pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
731 :			pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
732 :
733 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
734 :			}
735 :
736 :	syskin	1.44.2.14	bool
737 :	chl	1.44.2.1	MotionEstimation(MBParam * const pParam,
738 :			FRAMEINFO * const current,
739 :			FRAMEINFO * const reference,
740 :			const IMAGE * const pRefH,
741 :			const IMAGE * const pRefV,
742 :			const IMAGE * const pRefHV,
743 :			const uint32_t iLimit)
744 :			{
745 :			MACROBLOCK *const pMBs = current->mbs;
746 :			const IMAGE *const pCurrent = &current->image;
747 :			const IMAGE *const pRef = &reference->image;
748 :	Isibaar	1.1
749 :	chl	1.44.2.1	const VECTOR zeroMV = { 0, 0 };
750 :	edgomez	1.20
751 :	syskin	1.44.2.32	uint32_t mb_width = pParam->mb_width;
752 :			uint32_t mb_height = pParam->mb_height;
753 :	suxen_drol	1.44.2.31
754 :	chl	1.44.2.1	uint32_t x, y;
755 :			uint32_t iIntra = 0;
756 :	syskin	1.44.2.14	int32_t InterBias, quant = current->quant, sad00;
757 :	Isibaar	1.44.2.7	uint8_t *qimage;
758 :	chl	1.44.2.1
759 :	Isibaar	1.44.2.2	// some pre-initialized thingies for SearchP
760 :	syskin	1.44.2.34	int32_t temp[8];
761 :	Isibaar	1.44.2.2	VECTOR currentMV[5];
762 :	Isibaar	1.44.2.7	VECTOR currentQMV[5];
763 :	Isibaar	1.44.2.2	int32_t iMinSAD[5];
764 :			SearchData Data;
765 :	syskin	1.44.2.32	memset(&Data, 0, sizeof(SearchData));
766 :	Isibaar	1.44.2.2	Data.iEdgedWidth = pParam->edged_width;
767 :			Data.currentMV = currentMV;
768 :	Isibaar	1.44.2.7	Data.currentQMV = currentQMV;
769 :	Isibaar	1.44.2.2	Data.iMinSAD = iMinSAD;
770 :			Data.temp = temp;
771 :			Data.iFcode = current->fcode;
772 :	Isibaar	1.44.2.7	Data.rounding = pParam->m_rounding_type;
773 :	syskin	1.44.2.14	Data.qpel = pParam->m_quarterpel;
774 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
775 :	syskin	1.44.2.32	Data.rrv = current->global_flags & XVID_REDUCED;
776 :	Isibaar	1.44.2.7
777 :	syskin	1.44.2.32	if ((current->global_flags & XVID_REDUCED)) {
778 :	suxen_drol	1.44.2.31	mb_width = (pParam->width + 31) / 32;
779 :			mb_height = (pParam->height + 31) / 32;
780 :	syskin	1.44.2.32	Data.qpel = Data.chroma = 0;
781 :	suxen_drol	1.44.2.31	}
782 :
783 :	Isibaar	1.44.2.7	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 :	suxen_drol	1.44.2.31	for (y = 0; y < mb_height; y++) {
791 :			for (x = 0; x < mb_width; x++) {
792 :	chl	1.44.2.1	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
793 :	syskin	1.44.2.14
794 :	syskin	1.44.2.40	if (Data.rrv) pMB->sad16 =
795 :	syskin	1.44.2.32	sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32,
796 :			pRef->y + (x + y * pParam->edged_width) * 32,
797 :			pParam->edged_width, pMB->sad8 );
798 :
799 :	syskin	1.44.2.40	else pMB->sad16 =
800 :	syskin	1.44.2.32	sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
801 :	chl	1.44.2.1	pRef->y + (x + y * pParam->edged_width) * 16,
802 :			pParam->edged_width, pMB->sad8 );
803 :
804 :	syskin	1.44.2.14	if (Data.chroma) {
805 :			pMB->sad16 += sad8(pCurrent->u + x8 + y(pParam->edged_width/2)*8,
806 :			pRef->u + x8 + y(pParam->edged_width/2)*8, pParam->edged_width/2);
807 :
808 :			pMB->sad16 += sad8(pCurrent->v + (x + y(pParam->edged_width/2))8,
809 :			pRef->v + (x + y(pParam->edged_width/2))8, pParam->edged_width/2);
810 :			}
811 :
812 :			sad00 = pMB->sad16; //if no gmc; else sad00 = (..)
813 :
814 :	chl	1.44.2.1	if (!(current->global_flags & XVID_LUMIMASKING)) {
815 :			pMB->dquant = NO_CHANGE;
816 :	syskin	1.44.2.40	pMB->quant = current->quant;
817 :	syskin	1.44.2.11	} else {
818 :	syskin	1.44.2.5	if (pMB->dquant != NO_CHANGE) {
819 :			quant += DQtab[pMB->dquant];
820 :			if (quant > 31) quant = 31;
821 :			else if (quant < 1) quant = 1;
822 :			}
823 :	syskin	1.44.2.11	pMB->quant = quant;
824 :			}
825 :	chl	1.44.2.1
826 :	syskin	1.44.2.14	//initial skip decision
827 :	chl	1.44.2.13	/* no early skip for GMC (global vector = skip vector is unknown!) */
828 :			if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
829 :	syskin	1.44.2.32	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
830 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
831 :	chl	1.44.2.13	SkipMacroblockP(pMB, sad00);
832 :			continue;
833 :			}
834 :			}
835 :	syskin	1.44.2.14
836 :			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
837 :	chl	1.44.2.1	y, current->motion_flags, pMB->quant,
838 :	Isibaar	1.44.2.2	&Data, pParam, pMBs, reference->mbs,
839 :	chl	1.44.2.1	current->global_flags & XVID_INTER4V, pMB);
840 :
841 :			/* final skip decision, a.k.a. "the vector you found, really that good?" */
842 :	syskin	1.44.2.40	if (current->coding_type == P_VOP) {
843 :	chl	1.44.2.13	if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
844 :	syskin	1.44.2.32	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
845 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
846 :	chl	1.44.2.13	SkipMacroblockP(pMB, sad00);
847 :			continue;
848 :			}
849 :			}
850 :	syskin	1.44.2.14
851 :	chl	1.44.2.1	/* finally, intra decision */
852 :	Isibaar	1.1
853 :	chl	1.44.2.1	InterBias = MV16_INTER_BIAS;
854 :	syskin	1.44.2.16	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
855 :	chl	1.44.2.1	if (y != 0)
856 :	syskin	1.44.2.16	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
857 :	chl	1.44.2.1	if (x != 0)
858 :	syskin	1.44.2.16	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
859 :	syskin	1.44.2.14
860 :			if (Data.chroma) InterBias += 50; // to compensate bigger SAD
861 :	syskin	1.44.2.32	if (Data.rrv) InterBias *= 4; //??
862 :	chl	1.44.2.1
863 :			if (InterBias < pMB->sad16) {
864 :	syskin	1.44.2.32	int32_t deviation;
865 :			if (Data.rrv) {
866 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32,
867 :			pParam->edged_width)
868 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16,
869 :			pParam->edged_width)
870 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width,
871 :			pParam->edged_width)
872 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1),
873 :			pParam->edged_width);
874 :			} else
875 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
876 :	chl	1.44.2.1	pParam->edged_width);
877 :	edgomez	1.20
878 :	syskin	1.44.2.32	if (deviation < (pMB->sad16 - InterBias)) {
879 :	Isibaar	1.44.2.7	if (++iIntra >= iLimit) { free(qimage); return 1; }
880 :	chl	1.44.2.1	pMB->mode = MODE_INTRA;
881 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
882 :			pMB->mvs[3] = zeroMV;
883 :	Isibaar	1.44.2.7	pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
884 :			pMB->qmvs[3] = zeroMV;
885 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
886 :			pMB->sad8[3] = 0;
887 :			}
888 :			}
889 :			}
890 :			}
891 :	Isibaar	1.44.2.7	free(qimage);
892 :	chl	1.44.2.13
893 :			if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
894 :	syskin	1.44.2.14	current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
895 :	chl	1.44.2.13	else
896 :			current->GMC_MV = zeroMV;
897 :
898 :	syskin	1.44.2.14	return 0;
899 :	chl	1.44.2.1	}
900 :	chl	1.18
901 :	edgomez	1.20
902 :	chl	1.44.2.1	#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16\|PMV_HALFPELREFINE16)
903 :	edgomez	1.20
904 :	chl	1.44.2.1	static __inline int
905 :			make_mask(const VECTOR * const pmv, const int i)
906 :			{
907 :	Isibaar	1.44.2.2	int mask = 255, j;
908 :	chl	1.44.2.1	for (j = 0; j < i; j++) {
909 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
910 :			if (pmv[i].x == pmv[j].x) {
911 :			if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
912 :			if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
913 :	syskin	1.44.2.40	} else
914 :	chl	1.44.2.1	if (pmv[i].y == pmv[j].y) {
915 :			if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
916 :			if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
917 :			}
918 :			}
919 :			return mask;
920 :			}
921 :
922 :	syskin	1.44.2.40	static __inline void
923 :	syskin	1.44.2.32	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
924 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
925 :	chl	1.44.2.1	{
926 :
927 :			//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
928 :	syskin	1.44.2.32	if (rrv) { iWcount /= 2; iHcount /= 2; }
929 :	chl	1.44.2.1
930 :	syskin	1.44.2.32	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
931 :	chl	1.44.2.1	pmv[5].x = EVEN(pmv[3].x);
932 :	syskin	1.44.2.40	pmv[5].y = EVEN(pmv[3].y);
933 :	Isibaar	1.44.2.2	} else pmv[5].x = pmv[5].y = 0;
934 :	chl	1.44.2.1
935 :			if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
936 :			else pmv[3].x = pmv[3].y = 0;
937 :
938 :			if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
939 :			else pmv[4].x = pmv[4].y = 0;
940 :
941 :			// [1] median prediction
942 :	syskin	1.44.2.32	if (rrv) { //median is in halfzero-precision
943 :			pmv[1].x = RRV_MV_SCALEUP(pmv[0].x);
944 :			pmv[1].y = RRV_MV_SCALEUP(pmv[0].y);
945 :			} else { pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); }
946 :	chl	1.44.2.1
947 :			pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
948 :
949 :			pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
950 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
951 :
952 :	syskin	1.44.2.32	if ((x < iWcount-1) && (y < iHcount-1)) {
953 :	chl	1.44.2.1	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
954 :	syskin	1.44.2.40	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
955 :	Isibaar	1.44.2.2	} else pmv[6].x = pmv[6].y = 0;
956 :	syskin	1.44.2.32
957 :			if (rrv) {
958 :			int i;
959 :			for (i = 0; i < 7; i++) {
960 :			pmv[i].x = RRV_MV_SCALEDOWN(pmv[i].x);
961 :			pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // a trick
962 :			}
963 :			}
964 :	chl	1.44.2.1	}
965 :
966 :			static void
967 :	syskin	1.44.2.14	SearchP(const IMAGE * const pRef,
968 :	chl	1.44.2.1	const uint8_t * const pRefH,
969 :			const uint8_t * const pRefV,
970 :			const uint8_t * const pRefHV,
971 :			const IMAGE * const pCur,
972 :			const int x,
973 :			const int y,
974 :			const uint32_t MotionFlags,
975 :			const uint32_t iQuant,
976 :	Isibaar	1.44.2.2	SearchData * const Data,
977 :	chl	1.44.2.1	const MBParam * const pParam,
978 :			const MACROBLOCK * const pMBs,
979 :			const MACROBLOCK * const prevMBs,
980 :			int inter4v,
981 :			MACROBLOCK * const pMB)
982 :			{
983 :	Isibaar	1.1
984 :	chl	1.44.2.1	int i, iDirection = 255, mask, threshA;
985 :	Isibaar	1.44.2.2	VECTOR pmv[7];
986 :	chl	1.44.2.1
987 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
988 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
989 :	syskin	1.44.2.32
990 :	syskin	1.44.2.34	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
991 :	syskin	1.44.2.32
992 :	syskin	1.44.2.34	Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache
993 :	syskin	1.44.2.32	if (Data->rrv) i = 2; else i = 1;
994 :			Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
995 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
996 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
997 :
998 :			Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
999 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
1000 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
1001 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
1002 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1003 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1004 :	Isibaar	1.1
1005 :	syskin	1.44.2.11	Data->lambda16 = lambda_vec16[iQuant];
1006 :	Isibaar	1.44.2.12	Data->lambda8 = lambda_vec8[iQuant];
1007 :	syskin	1.44.2.22	Data->qpel_precision = 0;
1008 :	Isibaar	1.1
1009 :	edgomez	1.44.2.4	if (pMB->dquant != NO_CHANGE) inter4v = 0;
1010 :	Isibaar	1.44.2.2
1011 :	Isibaar	1.44.2.7	for(i = 0; i < 5; i++)
1012 :			Data->currentMV[i].x = Data->currentMV[i].y = 0;
1013 :	chl	1.44.2.1
1014 :	syskin	1.44.2.22	if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1015 :			else Data->predMV = pmv[0];
1016 :	Isibaar	1.44.2.12
1017 :	syskin	1.44.2.36	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
1018 :	syskin	1.44.2.14	Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000;
1019 :			Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100;
1020 :	Isibaar	1.44.2.2	Data->iMinSAD[2] = pMB->sad8[1];
1021 :			Data->iMinSAD[3] = pMB->sad8[2];
1022 :			Data->iMinSAD[4] = pMB->sad8[3];
1023 :	chl	1.44.2.1
1024 :			if ((x == 0) && (y == 0)) threshA = 512;
1025 :			else {
1026 :	edgomez	1.44.2.4	threshA = Data->temp[0]; // that's when we keep this SAD atm
1027 :	chl	1.44.2.1	if (threshA < 512) threshA = 512;
1028 :			if (threshA > 1024) threshA = 1024; }
1029 :
1030 :			PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1031 :	syskin	1.44.2.32	prevMBs + x + y * pParam->mb_width, Data->rrv);
1032 :	chl	1.44.2.1
1033 :	syskin	1.44.2.32	if (Data->rrv) CheckCandidate = CheckCandidate32;
1034 :			else if (inter4v \|\| Data->chroma) CheckCandidate = CheckCandidate16;
1035 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
1036 :	edgomez	1.44.2.4
1037 :	chl	1.44.2.1	/* main loop. checking all predictions */
1038 :
1039 :			for (i = 1; i < 7; i++) {
1040 :			if (!(mask = make_mask(pmv, i)) ) continue;
1041 :	edgomez	1.44.2.4	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1042 :	Isibaar	1.44.2.2	if (Data->iMinSAD[0] <= threshA) break;
1043 :	chl	1.44.2.1	}
1044 :
1045 :	Isibaar	1.44.2.2	if ((Data->iMinSAD[0] <= threshA) \|\|
1046 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1047 :			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1048 :	chl	1.44.2.1	inter4v = 0;
1049 :	Isibaar	1.44.2.2	} else {
1050 :	edgomez	1.20
1051 :	syskin	1.44.2.40	MainSearchFunc * MainSearchPtr;
1052 :	Isibaar	1.44.2.2	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1053 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1054 :			else MainSearchPtr = DiamondSearch;
1055 :	chl	1.35
1056 :	Isibaar	1.44.2.2	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1057 :	chl	1.13
1058 :	chl	1.44.2.1	/* extended search, diamond starting in 0,0 and in prediction.
1059 :			note that this search is/might be done in halfpel positions,
1060 :			which makes it more different than the diamond above */
1061 :	Isibaar	1.1
1062 :	Isibaar	1.44.2.2	if (MotionFlags & PMV_EXTSEARCH16) {
1063 :			int32_t bSAD;
1064 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1065 :	syskin	1.44.2.32	if (Data->rrv) {
1066 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1067 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1068 :	syskin	1.44.2.40	} else
1069 :	syskin	1.44.2.32	if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1070 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1071 :	Isibaar	1.44.2.2	if (!(MVequal(startMV, backupMV))) {
1072 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1073 :
1074 :	Isibaar	1.44.2.12	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1075 :	Isibaar	1.44.2.2	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1076 :			if (bSAD < Data->iMinSAD[0]) {
1077 :			Data->currentMV[0] = backupMV;
1078 :			Data->iMinSAD[0] = bSAD; }
1079 :			}
1080 :	chl	1.13
1081 :	Isibaar	1.44.2.2	backupMV = Data->currentMV[0];
1082 :	syskin	1.44.2.32	if (!MotionFlags & PMV_HALFPELREFINE16 \|\| Data->rrv) startMV.x = startMV.y = 0;
1083 :			else startMV.x = startMV.y = 1;
1084 :	Isibaar	1.44.2.2	if (!(MVequal(startMV, backupMV))) {
1085 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1086 :
1087 :	Isibaar	1.44.2.12	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1088 :	Isibaar	1.44.2.2	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1089 :			if (bSAD < Data->iMinSAD[0]) {
1090 :			Data->currentMV[0] = backupMV;
1091 :			Data->iMinSAD[0] = bSAD; }
1092 :			}
1093 :	chl	1.44.2.1	}
1094 :			}
1095 :	chl	1.13
1096 :	syskin	1.44.2.22	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1097 :	Isibaar	1.1
1098 :	Isibaar	1.44.2.7	for(i = 0; i < 5; i++) {
1099 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1100 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1101 :			}
1102 :
1103 :	syskin	1.44.2.40	if((Data->qpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1104 :	Isibaar	1.44.2.7
1105 :	syskin	1.44.2.22	Data->qpel_precision = 1;
1106 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1107 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1108 :	syskin	1.44.2.8
1109 :	syskin	1.44.2.22	SubpelRefine(Data);
1110 :	Isibaar	1.44.2.7	}
1111 :
1112 :	syskin	1.44.2.14	if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0;
1113 :	Isibaar	1.44.2.2	if (inter4v) {
1114 :			SearchData Data8;
1115 :			Data8.iFcode = Data->iFcode;
1116 :	Isibaar	1.44.2.12	Data8.lambda8 = Data->lambda8;
1117 :	Isibaar	1.44.2.2	Data8.iEdgedWidth = Data->iEdgedWidth;
1118 :	Isibaar	1.44.2.12	Data8.RefQ = Data->RefQ;
1119 :	syskin	1.44.2.14	Data8.qpel = Data->qpel;
1120 :	syskin	1.44.2.32	Data8.rrv = Data->rrv;
1121 :	Isibaar	1.44.2.2	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1122 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1123 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1124 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1125 :	syskin	1.44.2.14
1126 :			if (Data->chroma) {
1127 :	syskin	1.44.2.22	int sumx, sumy, dx, dy;
1128 :	syskin	1.44.2.14
1129 :			if(pParam->m_quarterpel) {
1130 :	syskin	1.44.2.22	sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2;
1131 :			sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2;
1132 :			} else {
1133 :			sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1134 :			sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1135 :			}
1136 :			dx = (sumx >> 3) + roundtab_76[sumx & 0xf];
1137 :			dy = (sumy >> 3) + roundtab_76[sumy & 0xf];
1138 :	syskin	1.44.2.14
1139 :			Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1140 :			}
1141 :	Isibaar	1.44.2.2	}
1142 :	chl	1.13
1143 :	syskin	1.44.2.32	if (Data->rrv) {
1144 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1145 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1146 :			}
1147 :	syskin	1.44.2.38
1148 :	chl	1.44.2.1	if (!(inter4v) \|\|
1149 :	syskin	1.44.2.40	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1150 :	Isibaar	1.44.2.2	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1151 :	chl	1.44.2.1	// INTER MODE
1152 :			pMB->mode = MODE_INTER;
1153 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1]
1154 :			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1155 :	edgomez	1.20
1156 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1157 :	Isibaar	1.44.2.2	pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1158 :	chl	1.13
1159 :	Isibaar	1.44.2.7	if(pParam->m_quarterpel) {
1160 :	syskin	1.44.2.22	pMB->qmvs[0] = pMB->qmvs[1]
1161 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1162 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1163 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1164 :	syskin	1.44.2.14	} else {
1165 :	Isibaar	1.44.2.7	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1166 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1167 :			}
1168 :	chl	1.44.2.1	} else {
1169 :	syskin	1.44.2.40	// INTER4V MODE; all other things are already set in Search8
1170 :	chl	1.44.2.1	pMB->mode = MODE_INTER4V;
1171 :	syskin	1.44.2.40	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1172 :	Isibaar	1.44.2.2	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1173 :	chl	1.44.2.1	}
1174 :			}
1175 :	edgomez	1.20
1176 :	chl	1.44.2.1	static void
1177 :			Search8(const SearchData * const OldData,
1178 :			const int x, const int y,
1179 :			const uint32_t MotionFlags,
1180 :			const MBParam * const pParam,
1181 :			MACROBLOCK * const pMB,
1182 :			const MACROBLOCK * const pMBs,
1183 :	Isibaar	1.44.2.2	const int block,
1184 :			SearchData * const Data)
1185 :	chl	1.44.2.1	{
1186 :	syskin	1.44.2.32	int i = 0;
1187 :	Isibaar	1.44.2.2	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1188 :			Data->currentMV = OldData->currentMV + 1 + block;
1189 :	Isibaar	1.44.2.7	Data->currentQMV = OldData->currentQMV + 1 + block;
1190 :
1191 :	syskin	1.44.2.11	if(pParam->m_quarterpel) {
1192 :	syskin	1.44.2.32	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1193 :	syskin	1.44.2.36	if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1194 :			Data->predMV, Data->iFcode, 0, 0);
1195 :	syskin	1.44.2.34
1196 :	Isibaar	1.44.2.12	} else {
1197 :	syskin	1.44.2.32	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1198 :			if (block != 0) {
1199 :	syskin	1.44.2.36	if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1200 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1201 :	syskin	1.44.2.32	}
1202 :	Isibaar	1.44.2.12	}
1203 :	Isibaar	1.1
1204 :	syskin	1.44.2.32	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1205 :
1206 :	chl	1.44.2.1	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8)) {
1207 :	syskin	1.44.2.32	if (Data->rrv) i = 2; else i = 1;
1208 :	edgomez	1.20
1209 :	syskin	1.44.2.32	Data->Ref = OldData->Ref + i8 ((block&1) + pParam->edged_width*(block>>1));
1210 :			Data->RefH = OldData->RefH + i8 ((block&1) + pParam->edged_width*(block>>1));
1211 :			Data->RefV = OldData->RefV + i8 ((block&1) + pParam->edged_width*(block>>1));
1212 :			Data->RefHV = OldData->RefHV + i8 ((block&1) + pParam->edged_width*(block>>1));
1213 :	Isibaar	1.1
1214 :	syskin	1.44.2.32	Data->Cur = OldData->Cur + i8 ((block&1) + pParam->edged_width*(block>>1));
1215 :	syskin	1.44.2.22	Data->qpel_precision = 0;
1216 :	syskin	1.44.2.34
1217 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1218 :			pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv);
1219 :
1220 :			if (Data->rrv) CheckCandidate = CheckCandidate16no4v;
1221 :			else CheckCandidate = CheckCandidate8;
1222 :	Isibaar	1.1
1223 :	chl	1.44.2.1	if (MotionFlags & PMV_EXTSEARCH8) {
1224 :	Isibaar	1.44.2.7	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1225 :
1226 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1227 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1228 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1229 :			else MainSearchPtr = DiamondSearch;
1230 :	Isibaar	1.1
1231 :	Isibaar	1.44.2.7	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1232 :	Isibaar	1.1
1233 :	syskin	1.44.2.11	if(*(Data->iMinSAD) < temp_sad) {
1234 :	Isibaar	1.44.2.7	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1235 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1236 :			}
1237 :			}
1238 :
1239 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1240 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1241 :
1242 :	syskin	1.44.2.22	SubpelRefine(Data); // perform halfpel refine of current best vector
1243 :	Isibaar	1.44.2.7
1244 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1245 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1246 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1247 :			}
1248 :			}
1249 :
1250 :	syskin	1.44.2.40	if(Data->qpel) {
1251 :	syskin	1.44.2.11	if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1252 :			(MotionFlags & PMV_QUARTERPELREFINE8)) {
1253 :	syskin	1.44.2.22	Data->qpel_precision = 1;
1254 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1255 :			pParam->width, pParam->height, OldData->iFcode, 1, 0);
1256 :	syskin	1.44.2.22	SubpelRefine(Data);
1257 :	syskin	1.44.2.11	}
1258 :	Isibaar	1.44.2.7	}
1259 :			}
1260 :
1261 :	syskin	1.44.2.32	if (Data->rrv) {
1262 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1263 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1264 :			}
1265 :
1266 :	syskin	1.44.2.34	if(Data->qpel) {
1267 :	syskin	1.44.2.22	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1268 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1269 :			pMB->qmvs[block] = *(Data->currentQMV);
1270 :	syskin	1.44.2.32	} else {
1271 :	Isibaar	1.44.2.7	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1272 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1273 :	Isibaar	1.1	}
1274 :
1275 :	Isibaar	1.44.2.2	pMB->mvs[block] = *(Data->currentMV);
1276 :	syskin	1.44.2.11	pMB->sad8[block] = 4 * (*Data->iMinSAD);
1277 :	Isibaar	1.1	}
1278 :
1279 :	chl	1.44.2.1	/* B-frames code starts here */
1280 :	Isibaar	1.1
1281 :	chl	1.44.2.1	static __inline VECTOR
1282 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1283 :			{
1284 :			/* the stupidiest function ever */
1285 :			if (mode == MODE_FORWARD) return pMB->mvs[0];
1286 :			else return pMB->b_mvs[0];
1287 :			}
1288 :	Isibaar	1.1
1289 :	chl	1.44.2.1	static void __inline
1290 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1291 :			const uint32_t iWcount,
1292 :			const MACROBLOCK * const pMB,
1293 :			const uint32_t mode_curr)
1294 :			{
1295 :	Isibaar	1.1
1296 :	chl	1.44.2.1	// [0] is prediction
1297 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1298 :	Isibaar	1.1
1299 :	chl	1.44.2.1	pmv[1].x = pmv[1].y = 0; // [1] is zero
1300 :	Isibaar	1.1
1301 :	chl	1.44.2.1	pmv[2] = ChoosePred(pMB, mode_curr);
1302 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1303 :	Isibaar	1.1
1304 :	chl	1.44.2.1	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1305 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1306 :	syskin	1.44.2.40	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1307 :	Isibaar	1.44.2.2	} else pmv[3].x = pmv[3].y = 0;
1308 :	chl	1.39
1309 :	chl	1.44.2.1	if (y != 0) {
1310 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1311 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1312 :			} else pmv[4].x = pmv[4].y = 0;
1313 :	chl	1.39
1314 :	chl	1.44.2.1	if (x != 0) {
1315 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1316 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1317 :			} else pmv[5].x = pmv[5].y = 0;
1318 :	chl	1.39
1319 :	chl	1.44.2.1	if ((x != 0)&&(y != 0)) {
1320 :			pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1321 :			pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1322 :			} else pmv[6].x = pmv[6].y = 0;
1323 :	chl	1.39
1324 :	chl	1.44.2.1	// more?
1325 :	chl	1.39	}
1326 :
1327 :
1328 :	chl	1.44.2.1	/* search backward or forward, for b-frames */
1329 :			static void
1330 :			SearchBF( const uint8_t * const pRef,
1331 :			const uint8_t * const pRefH,
1332 :			const uint8_t * const pRefV,
1333 :			const uint8_t * const pRefHV,
1334 :			const IMAGE * const pCur,
1335 :			const int x, const int y,
1336 :			const uint32_t MotionFlags,
1337 :			const uint32_t iFcode,
1338 :			const MBParam * const pParam,
1339 :			MACROBLOCK * const pMB,
1340 :			const VECTOR * const predMV,
1341 :			int32_t * const best_sad,
1342 :	h	1.44.2.3	const int32_t mode_current,
1343 :			SearchData * const Data)
1344 :	Isibaar	1.1	{
1345 :
1346 :	chl	1.44.2.1	const int32_t iEdgedWidth = pParam->edged_width;
1347 :
1348 :	syskin	1.44.2.38	int i, iDirection = 255, mask;
1349 :	h	1.44.2.3	VECTOR pmv[7];
1350 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1351 :	h	1.44.2.3	*Data->iMinSAD = MV_MAX_ERROR;
1352 :			Data->iFcode = iFcode;
1353 :	syskin	1.44.2.22	Data->qpel_precision = 0;
1354 :	chl	1.44.2.1
1355 :	h	1.44.2.3	Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1356 :			Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1357 :			Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1358 :			Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1359 :	chl	1.44.2.1
1360 :	h	1.44.2.3	Data->predMV = *predMV;
1361 :	edgomez	1.20
1362 :	h	1.44.2.3	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1363 :	syskin	1.44.2.34	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1364 :	Isibaar	1.1
1365 :	h	1.44.2.3	pmv[0] = Data->predMV;
1366 :	syskin	1.44.2.19	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1367 :	h	1.44.2.3	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1368 :	Isibaar	1.1
1369 :	h	1.44.2.3	Data->currentMV->x = Data->currentMV->y = 0;
1370 :	chl	1.44.2.1	CheckCandidate = CheckCandidate16no4v;
1371 :	edgomez	1.20
1372 :	chl	1.44.2.1	// main loop. checking all predictions
1373 :	syskin	1.44.2.38	for (i = 0; i < 7; i++) {
1374 :	chl	1.44.2.1	if (!(mask = make_mask(pmv, i)) ) continue;
1375 :	h	1.44.2.3	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1376 :	edgomez	1.20	}
1377 :
1378 :	chl	1.44.2.1	if (MotionFlags & PMV_USESQUARES16)
1379 :			MainSearchPtr = SquareSearch;
1380 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1381 :			MainSearchPtr = AdvDiamondSearch;
1382 :			else MainSearchPtr = DiamondSearch;
1383 :	Isibaar	1.1
1384 :	syskin	1.44.2.38	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1385 :	Isibaar	1.1
1386 :	syskin	1.44.2.22	SubpelRefine(Data);
1387 :	syskin	1.44.2.19
1388 :	syskin	1.44.2.38	if (Data->qpel && Data->iMinSAD < best_sad + 300) {
1389 :	syskin	1.44.2.19	Data->currentQMV->x = 2*Data->currentMV->x;
1390 :			Data->currentQMV->y = 2*Data->currentMV->y;
1391 :	syskin	1.44.2.22	Data->qpel_precision = 1;
1392 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1393 :			pParam->width, pParam->height, iFcode, 1, 0);
1394 :	syskin	1.44.2.22	SubpelRefine(Data);
1395 :	syskin	1.44.2.19	}
1396 :	Isibaar	1.1
1397 :	chl	1.44.2.1	// three bits are needed to code backward mode. four for forward
1398 :	syskin	1.44.2.38
1399 :	syskin	1.44.2.14	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1400 :			else Data->iMinSAD += 3 Data->lambda16;
1401 :	chl	1.44.2.1
1402 :	h	1.44.2.3	if (Data->iMinSAD < best_sad) {
1403 :			best_sad = Data->iMinSAD;
1404 :	chl	1.44.2.1	pMB->mode = mode_current;
1405 :	syskin	1.44.2.19	if (Data->qpel) {
1406 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1407 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1408 :	syskin	1.44.2.40	if (mode_current == MODE_FORWARD)
1409 :	syskin	1.44.2.19	pMB->qmvs[0] = *Data->currentQMV;
1410 :	syskin	1.44.2.40	else
1411 :	syskin	1.44.2.19	pMB->b_qmvs[0] = *Data->currentQMV;
1412 :			} else {
1413 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1414 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1415 :			}
1416 :	syskin	1.44.2.37	if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1417 :			else pMB->b_mvs[0] = *Data->currentMV;
1418 :	Isibaar	1.1	}
1419 :	syskin	1.44.2.37	if (mode_current == MODE_FORWARD) (Data->currentMV+2) = Data->currentMV;
1420 :			else (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1421 :
1422 :	chl	1.44.2.1	}
1423 :
1424 :	syskin	1.44.2.40	static void
1425 :	syskin	1.44.2.34	SkipDecisionB(const IMAGE * const pCur,
1426 :			const IMAGE * const f_Ref,
1427 :	syskin	1.44.2.40	const IMAGE * const b_Ref,
1428 :	syskin	1.44.2.34	MACROBLOCK * const pMB,
1429 :			const uint32_t quant,
1430 :			const uint32_t x, const uint32_t y,
1431 :			const SearchData * const Data)
1432 :			{
1433 :			int dx, dy, b_dx, b_dy;
1434 :			uint32_t sum;
1435 :			//this is not full chroma compensation, only it's fullpel approximation. should work though
1436 :			if (Data->qpel) {
1437 :			dy = Data->directmvF[0].y/2 + Data->directmvF[1].y/2 +
1438 :			Data->directmvF[2].y/2 + Data->directmvF[3].y/2;
1439 :
1440 :			dx = Data->directmvF[0].x/2 + Data->directmvF[1].x/2 +
1441 :			Data->directmvF[2].x/2 + Data->directmvF[3].x/2;
1442 :
1443 :			b_dy = Data->directmvB[0].y/2 + Data->directmvB[1].y/2 +
1444 :			Data->directmvB[2].y/2 + Data->directmvB[3].y/2;
1445 :
1446 :			b_dx = Data->directmvB[0].x/2 + Data->directmvB[1].x/2 +
1447 :			Data->directmvB[2].x/2 + Data->directmvB[3].x/2;
1448 :
1449 :			} else {
1450 :			dy = Data->directmvF[0].y + Data->directmvF[1].y +
1451 :			Data->directmvF[2].y + Data->directmvF[3].y;
1452 :
1453 :			dx = Data->directmvF[0].x + Data->directmvF[1].x +
1454 :			Data->directmvF[2].x + Data->directmvF[3].x;
1455 :
1456 :			b_dy = Data->directmvB[0].y + Data->directmvB[1].y +
1457 :			Data->directmvB[2].y + Data->directmvB[3].y;
1458 :
1459 :			b_dx = Data->directmvB[0].x + Data->directmvB[1].x +
1460 :			Data->directmvB[2].x + Data->directmvB[3].x;
1461 :			}
1462 :
1463 :
1464 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1465 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1466 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1467 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1468 :
1469 :			sum = sad8bi(pCur->u + 8x + 8y*(Data->iEdgedWidth/2),
1470 :			f_Ref->u + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1471 :			b_Ref->u + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1472 :			Data->iEdgedWidth/2);
1473 :			sum += sad8bi(pCur->v + 8x + 8y*(Data->iEdgedWidth/2),
1474 :			f_Ref->v + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1475 :			b_Ref->v + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1476 :			Data->iEdgedWidth/2);
1477 :
1478 :			if (sum < 2MAX_CHROMA_SAD_FOR_SKIP quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1479 :			}
1480 :
1481 :
1482 :
1483 :			static __inline uint32_t
1484 :	Isibaar	1.44.2.2	SearchDirect(const IMAGE * const f_Ref,
1485 :	chl	1.44.2.1	const uint8_t * const f_RefH,
1486 :			const uint8_t * const f_RefV,
1487 :			const uint8_t * const f_RefHV,
1488 :	Isibaar	1.44.2.2	const IMAGE * const b_Ref,
1489 :	chl	1.44.2.1	const uint8_t * const b_RefH,
1490 :			const uint8_t * const b_RefV,
1491 :			const uint8_t * const b_RefHV,
1492 :			const IMAGE * const pCur,
1493 :			const int x, const int y,
1494 :			const uint32_t MotionFlags,
1495 :			const int32_t TRB, const int32_t TRD,
1496 :			const MBParam * const pParam,
1497 :			MACROBLOCK * const pMB,
1498 :			const MACROBLOCK * const b_mb,
1499 :	h	1.44.2.3	int32_t * const best_sad,
1500 :			SearchData * const Data)
1501 :	chl	1.44.2.1
1502 :			{
1503 :	h	1.44.2.3	int32_t skip_sad;
1504 :	chl	1.44.2.1	int k;
1505 :	h	1.44.2.3
1506 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1507 :
1508 :	h	1.44.2.3	Data->iMinSAD = 2564096;
1509 :	chl	1.44.2.1
1510 :	h	1.44.2.3	Data->Ref = f_Ref->y + (x + Data->iEdgedWidthy) 16;
1511 :			Data->RefH = f_RefH + (x + Data->iEdgedWidthy) 16;
1512 :			Data->RefV = f_RefV + (x + Data->iEdgedWidthy) 16;
1513 :			Data->RefHV = f_RefHV + (x + Data->iEdgedWidthy) 16;
1514 :			Data->bRef = b_Ref->y + (x + Data->iEdgedWidthy) 16;
1515 :			Data->bRefH = b_RefH + (x + Data->iEdgedWidthy) 16;
1516 :			Data->bRefV = b_RefV + (x + Data->iEdgedWidthy) 16;
1517 :			Data->bRefHV = b_RefHV + (x + Data->iEdgedWidthy) 16;
1518 :
1519 :			Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1520 :			Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1521 :			Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1522 :			Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1523 :	syskin	1.44.2.19	if (Data->qpel) { //we measure in qpixels
1524 :			Data->max_dx *= 2;
1525 :			Data->max_dy *= 2;
1526 :			Data->min_dx *= 2;
1527 :			Data->min_dy *= 2;
1528 :			Data->referencemv = b_mb->qmvs;
1529 :			} else Data->referencemv = b_mb->mvs;
1530 :	syskin	1.44.2.26	Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here
1531 :	chl	1.44.2.1
1532 :			for (k = 0; k < 4; k++) {
1533 :	h	1.44.2.3	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1534 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1535 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1536 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1537 :
1538 :			if ( ( pMB->b_mvs[k].x > Data->max_dx ) \|\| ( pMB->b_mvs[k].x < Data->min_dx )
1539 :			\|\| ( pMB->b_mvs[k].y > Data->max_dy ) \|\| ( pMB->b_mvs[k].y < Data->min_dy )) {
1540 :	chl	1.44.2.1
1541 :	Isibaar	1.44.2.2	best_sad = 2564096; // in that case, we won't use direct mode
1542 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1543 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1544 :	syskin	1.44.2.34	return 256*4096;
1545 :	Isibaar	1.44.2.2	}
1546 :			if (b_mb->mode != MODE_INTER4V) {
1547 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1548 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1549 :	h	1.44.2.3	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1550 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1551 :	Isibaar	1.44.2.2	break;
1552 :			}
1553 :			}
1554 :	chl	1.44.2.1
1555 :	syskin	1.44.2.22	if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect;
1556 :			else CheckCandidate = CheckCandidateDirectno4v;
1557 :	syskin	1.44.2.19
1558 :	h	1.44.2.3	(*CheckCandidate)(0, 0, 255, &k, Data);
1559 :	Isibaar	1.44.2.2
1560 :	syskin	1.44.2.34	// initial (fast) skip decision
1561 :			if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH*2) {
1562 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data->chroma, Data); //possible skip - checking chroma
1563 :			if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1564 :	Isibaar	1.44.2.2	}
1565 :
1566 :	h	1.44.2.3	skip_sad = *Data->iMinSAD;
1567 :	Isibaar	1.44.2.2
1568 :	chl	1.44.2.1	// DIRECT MODE DELTA VECTOR SEARCH.
1569 :			// This has to be made more effective, but at the moment I'm happy it's running at all
1570 :
1571 :			if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1572 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1573 :			else MainSearchPtr = DiamondSearch;
1574 :
1575 :	h	1.44.2.3	(*MainSearchPtr)(0, 0, Data, 255);
1576 :	chl	1.44.2.1
1577 :	syskin	1.44.2.22	SubpelRefine(Data);
1578 :	chl	1.44.2.1
1579 :	h	1.44.2.3	best_sad = Data->iMinSAD;
1580 :	chl	1.44.2.1
1581 :	syskin	1.44.2.39	if (b_mb->mode == MODE_INTER4V \|\| Data->qpel) pMB->mode = MODE_DIRECT;
1582 :	syskin	1.44.2.26	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1583 :	chl	1.44.2.1
1584 :	h	1.44.2.3	pMB->pmvs[3] = *Data->currentMV;
1585 :	chl	1.44.2.1
1586 :			for (k = 0; k < 4; k++) {
1587 :	h	1.44.2.3	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1588 :	syskin	1.44.2.19	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1589 :	h	1.44.2.3	? Data->directmvB[k].x
1590 :	syskin	1.44.2.19	:pMB->mvs[k].x - Data->referencemv[k].x);
1591 :	h	1.44.2.3	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1592 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1593 :			? Data->directmvB[k].y
1594 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1595 :	syskin	1.44.2.19	if (Data->qpel) {
1596 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1597 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1598 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1599 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1600 :			}
1601 :
1602 :	chl	1.44.2.1	if (b_mb->mode != MODE_INTER4V) {
1603 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1604 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1605 :	syskin	1.44.2.19	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1606 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1607 :	chl	1.44.2.1	break;
1608 :			}
1609 :			}
1610 :	h	1.44.2.3	return skip_sad;
1611 :	chl	1.44.2.1	}
1612 :
1613 :	h	1.44.2.3
1614 :	syskin	1.44.2.34	static void
1615 :	chl	1.44.2.1	SearchInterpolate(const uint8_t * const f_Ref,
1616 :			const uint8_t * const f_RefH,
1617 :			const uint8_t * const f_RefV,
1618 :			const uint8_t * const f_RefHV,
1619 :			const uint8_t * const b_Ref,
1620 :			const uint8_t * const b_RefH,
1621 :			const uint8_t * const b_RefV,
1622 :			const uint8_t * const b_RefHV,
1623 :			const IMAGE * const pCur,
1624 :			const int x, const int y,
1625 :			const uint32_t fcode,
1626 :			const uint32_t bcode,
1627 :			const uint32_t MotionFlags,
1628 :			const MBParam * const pParam,
1629 :			const VECTOR * const f_predMV,
1630 :			const VECTOR * const b_predMV,
1631 :			MACROBLOCK * const pMB,
1632 :	h	1.44.2.3	int32_t * const best_sad,
1633 :			SearchData * const fData)
1634 :	edgomez	1.20
1635 :	chl	1.3	{
1636 :
1637 :	chl	1.44.2.1	int iDirection, i, j;
1638 :	h	1.44.2.3	SearchData bData;
1639 :	edgomez	1.20
1640 :	syskin	1.44.2.34	fData->qpel_precision = 0;
1641 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1642 :			fData->iMinSAD = 4096256;
1643 :	syskin	1.44.2.37	bData.currentMV++; bData.currentQMV++;
1644 :	h	1.44.2.3	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1645 :
1646 :	syskin	1.44.2.34	i = (x + y * fData->iEdgedWidth) * 16;
1647 :			bData.bRef = fData->Ref = f_Ref + i;
1648 :			bData.bRefH = fData->RefH = f_RefH + i;
1649 :			bData.bRefV = fData->RefV = f_RefV + i;
1650 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1651 :			bData.Ref = fData->bRef = b_Ref + i;
1652 :			bData.RefH = fData->bRefH = b_RefH + i;
1653 :			bData.RefV = fData->bRefV = b_RefV + i;
1654 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1655 :	h	1.44.2.3
1656 :			bData.bpredMV = fData->predMV = *f_predMV;
1657 :			fData->bpredMV = bData.predMV = *b_predMV;
1658 :	syskin	1.44.2.20	fData->currentMV[0] = fData->currentMV[2];
1659 :	syskin	1.44.2.34
1660 :			get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0);
1661 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0);
1662 :	chl	1.3
1663 :	h	1.44.2.3	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1664 :	syskin	1.44.2.27	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1665 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1666 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1667 :	h	1.44.2.3
1668 :			if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1669 :	syskin	1.44.2.27	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1670 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1671 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1672 :	Isibaar	1.44.2.2
1673 :	h	1.44.2.3	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1674 :	chl	1.3
1675 :	chl	1.44.2.1	//diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1676 :	chl	1.3
1677 :	chl	1.44.2.1	do {
1678 :			iDirection = 255;
1679 :			// forward MV moves
1680 :	h	1.44.2.3	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1681 :	chl	1.35
1682 :	h	1.44.2.3	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1683 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1684 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1685 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1686 :	edgomez	1.20
1687 :	chl	1.44.2.1	// backward MV moves
1688 :	h	1.44.2.3	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1689 :			fData->currentMV[2] = fData->currentMV[0];
1690 :	Isibaar	1.44.2.2	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1691 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1692 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1693 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1694 :	chl	1.3
1695 :	chl	1.44.2.1	} while (!(iDirection));
1696 :	chl	1.3
1697 :	syskin	1.44.2.19	if (fData->qpel) {
1698 :	syskin	1.44.2.38	if (fData->iMinSAD > best_sad + 500) return;
1699 :	syskin	1.44.2.27	CheckCandidate = CheckCandidateInt;
1700 :	syskin	1.44.2.22	fData->qpel_precision = bData.qpel_precision = 1;
1701 :	syskin	1.44.2.34	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0);
1702 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1703 :	syskin	1.44.2.19	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1704 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1705 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1706 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1707 :	syskin	1.44.2.22	SubpelRefine(fData);
1708 :	syskin	1.44.2.38	if (fData->iMinSAD > best_sad + 300) return;
1709 :	syskin	1.44.2.19	fData->currentQMV[2] = fData->currentQMV[0];
1710 :	syskin	1.44.2.22	SubpelRefine(&bData);
1711 :	syskin	1.44.2.19	}
1712 :	syskin	1.44.2.28
1713 :	syskin	1.44.2.38	fData->iMinSAD += (2+3) fData->lambda16; // two bits are needed to code interpolate mode.
1714 :	syskin	1.44.2.19
1715 :	h	1.44.2.3	if (fData->iMinSAD < best_sad) {
1716 :			best_sad = fData->iMinSAD;
1717 :			pMB->mvs[0] = fData->currentMV[0];
1718 :			pMB->b_mvs[0] = fData->currentMV[1];
1719 :	chl	1.44.2.1	pMB->mode = MODE_INTERPOLATE;
1720 :	syskin	1.44.2.19	if (fData->qpel) {
1721 :			pMB->qmvs[0] = fData->currentQMV[0];
1722 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1723 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1724 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1725 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1726 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1727 :			} else {
1728 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1729 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1730 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1731 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1732 :			}
1733 :	edgomez	1.20	}
1734 :	chl	1.3	}
1735 :
1736 :	chl	1.44.2.1	void
1737 :			MotionEstimationBVOP(MBParam * const pParam,
1738 :			FRAMEINFO * const frame,
1739 :			const int32_t time_bp,
1740 :			const int32_t time_pp,
1741 :			// forward (past) reference
1742 :			const MACROBLOCK * const f_mbs,
1743 :			const IMAGE * const f_ref,
1744 :			const IMAGE * const f_refH,
1745 :			const IMAGE * const f_refV,
1746 :			const IMAGE * const f_refHV,
1747 :			// backward (future) reference
1748 :	syskin	1.44.2.18	const FRAMEINFO * const b_reference,
1749 :	chl	1.44.2.1	const IMAGE * const b_ref,
1750 :			const IMAGE * const b_refH,
1751 :			const IMAGE * const b_refV,
1752 :			const IMAGE * const b_refHV)
1753 :			{
1754 :			uint32_t i, j;
1755 :	syskin	1.44.2.34	int32_t best_sad;
1756 :			uint32_t skip_sad;
1757 :	chl	1.44.2.1	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1758 :			static const VECTOR zeroMV={0,0};
1759 :	syskin	1.44.2.18	const MACROBLOCK * const b_mbs = b_reference->mbs;
1760 :	chl	1.3
1761 :	chl	1.44.2.1	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1762 :	chl	1.3
1763 :	chl	1.44.2.1	const int32_t TRB = time_pp - time_bp;
1764 :			const int32_t TRD = time_pp;
1765 :	syskin	1.44.2.19	uint8_t * qimage;
1766 :	chl	1.3
1767 :	h	1.44.2.3	// some pre-inintialized data for the rest of the search
1768 :
1769 :			SearchData Data;
1770 :			int32_t iMinSAD;
1771 :			VECTOR currentMV[3];
1772 :	syskin	1.44.2.19	VECTOR currentQMV[3];
1773 :	syskin	1.44.2.32	memset(&Data, 0, sizeof(SearchData));
1774 :	h	1.44.2.3	Data.iEdgedWidth = pParam->edged_width;
1775 :	syskin	1.44.2.19	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1776 :	h	1.44.2.3	Data.iMinSAD = &iMinSAD;
1777 :	syskin	1.44.2.34	Data.lambda16 = lambda_vec16[frame->quant];
1778 :			Data.chroma = frame->quant;
1779 :	syskin	1.44.2.19	Data.qpel = pParam->m_quarterpel;
1780 :	syskin	1.44.2.22	Data.rounding = 0;
1781 :	h	1.44.2.3
1782 :	syskin	1.44.2.19	if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
1783 :			return; // allocate some mem for qpel interpolated blocks
1784 :			// somehow this is dirty since I think we shouldn't use malloc outside
1785 :			// encoder_create() - so please fix me!
1786 :			Data.RefQ = qimage;
1787 :	edgomez	1.20
1788 :	syskin	1.44.2.19	// note: i==horizontal, j==vertical
1789 :	chl	1.44.2.1	for (j = 0; j < pParam->mb_height; j++) {
1790 :	chl	1.3
1791 :	chl	1.44.2.1	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1792 :	edgomez	1.20
1793 :	chl	1.44.2.1	for (i = 0; i < pParam->mb_width; i++) {
1794 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1795 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1796 :	edgomez	1.20
1797 :	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 */
1798 :	syskin	1.44.2.18	if (b_reference->coding_type != S_VOP)
1799 :			if (b_mb->mode == MODE_NOT_CODED) {
1800 :			pMB->mode = MODE_NOT_CODED;
1801 :			continue;
1802 :			}
1803 :	edgomez	1.20
1804 :	h	1.44.2.3	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1805 :	syskin	1.44.2.11	pMB->quant = frame->quant;
1806 :	syskin	1.44.2.14
1807 :	chl	1.44.2.1	/* direct search comes first, because it (1) checks for SKIP-mode
1808 :			and (2) sets very good predictions for forward and backward search */
1809 :	Isibaar	1.44.2.2	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1810 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1811 :	chl	1.44.2.1	&frame->image,
1812 :			i, j,
1813 :			frame->motion_flags,
1814 :			TRB, TRD,
1815 :			pParam,
1816 :			pMB, b_mb,
1817 :	h	1.44.2.3	&best_sad,
1818 :			&Data);
1819 :	edgomez	1.20
1820 :	Isibaar	1.44.2.2	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1821 :	syskin	1.44.2.37
1822 :	chl	1.44.2.1	// forward search
1823 :			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1824 :			&frame->image, i, j,
1825 :			frame->motion_flags,
1826 :	h	1.44.2.3	frame->fcode, pParam,
1827 :	chl	1.44.2.1	pMB, &f_predMV, &best_sad,
1828 :	h	1.44.2.3	MODE_FORWARD, &Data);
1829 :	chl	1.10
1830 :	chl	1.44.2.1	// backward search
1831 :			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1832 :			&frame->image, i, j,
1833 :			frame->motion_flags,
1834 :	h	1.44.2.3	frame->bcode, pParam,
1835 :	chl	1.44.2.1	pMB, &b_predMV, &best_sad,
1836 :	h	1.44.2.3	MODE_BACKWARD, &Data);
1837 :	chl	1.10
1838 :	chl	1.44.2.1	// interpolate search comes last, because it uses data from forward and backward as prediction
1839 :			SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1840 :			b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1841 :			&frame->image,
1842 :			i, j,
1843 :			frame->fcode, frame->bcode,
1844 :			frame->motion_flags,
1845 :	h	1.44.2.3	pParam,
1846 :	chl	1.44.2.1	&f_predMV, &b_predMV,
1847 :	h	1.44.2.3	pMB, &best_sad,
1848 :			&Data);
1849 :	syskin	1.44.2.21
1850 :	syskin	1.44.2.34	// final skip decision
1851 :			if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2)
1852 :			&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1853 :			SkipDecisionB(&frame->image, f_ref, b_ref, pMB,frame->quant, i, j, &Data);
1854 :
1855 :	chl	1.44.2.1	switch (pMB->mode) {
1856 :			case MODE_FORWARD:
1857 :			f_count++;
1858 :	syskin	1.44.2.34	if (Data.qpel) f_predMV = pMB->qmvs[0];
1859 :	syskin	1.44.2.19	else f_predMV = pMB->mvs[0];
1860 :	chl	1.44.2.1	break;
1861 :			case MODE_BACKWARD:
1862 :			b_count++;
1863 :	syskin	1.44.2.34	if (Data.qpel) b_predMV = pMB->b_qmvs[0];
1864 :	syskin	1.44.2.19	else b_predMV = pMB->b_mvs[0];
1865 :	chl	1.44.2.1	break;
1866 :			case MODE_INTERPOLATE:
1867 :			i_count++;
1868 :	syskin	1.44.2.34	if (Data.qpel) {
1869 :	syskin	1.44.2.19	f_predMV = pMB->qmvs[0];
1870 :			b_predMV = pMB->b_qmvs[0];
1871 :	syskin	1.44.2.40	} else {
1872 :	syskin	1.44.2.19	f_predMV = pMB->mvs[0];
1873 :			b_predMV = pMB->b_mvs[0];
1874 :			}
1875 :	chl	1.44.2.1	break;
1876 :			case MODE_DIRECT:
1877 :			case MODE_DIRECT_NO4V:
1878 :			d_count++;
1879 :			default:
1880 :			break;
1881 :	chl	1.10	}
1882 :			}
1883 :			}
1884 :	syskin	1.44.2.19	free(qimage);
1885 :	chl	1.44.2.1	}
1886 :	chl	1.3
1887 :	syskin	1.44.2.37	static __inline void
1888 :	Isibaar	1.44.2.2	MEanalyzeMB ( const uint8_t * const pRef,
1889 :			const uint8_t * const pCur,
1890 :			const int x,
1891 :			const int y,
1892 :			const MBParam * const pParam,
1893 :	syskin	1.44.2.37	MACROBLOCK * const pMBs,
1894 :	h	1.44.2.3	SearchData * const Data)
1895 :	Isibaar	1.44.2.2	{
1896 :	chl	1.32
1897 :	syskin	1.44.2.37	int i, mask;
1898 :	h	1.44.2.3	VECTOR pmv[3];
1899 :	syskin	1.44.2.37	MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width];
1900 :
1901 :			for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1902 :	syskin	1.44.2.17
1903 :			//median is only used as prediction. it doesn't have to be real
1904 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1905 :			else
1906 :			if (x == 1) //left macroblock does not have any vector now
1907 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1908 :	syskin	1.44.2.40	else if (y == 1) // top macroblock doesn't have it's vector
1909 :	syskin	1.44.2.17	Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1910 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1911 :
1912 :	h	1.44.2.3	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1913 :	syskin	1.44.2.37	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv);
1914 :	chl	1.35
1915 :	h	1.44.2.3	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1916 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1917 :	syskin	1.44.2.37
1918 :	Isibaar	1.44.2.2	pmv[1].x = EVEN(pMB->mvs[0].x);
1919 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1920 :	syskin	1.44.2.17	pmv[2].x = EVEN(Data->predMV.x);
1921 :			pmv[2].y = EVEN(Data->predMV.y);
1922 :			pmv[0].x = pmv[0].y = 0;
1923 :
1924 :	syskin	1.44.2.37	CheckCandidate32I(0, 0, 255, &i, Data);
1925 :	syskin	1.44.2.17
1926 :	syskin	1.44.2.37	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) {
1927 :	Isibaar	1.44.2.2
1928 :	syskin	1.44.2.37	if (!(mask = make_mask(pmv, 1)))
1929 :			CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data);
1930 :			if (!(mask = make_mask(pmv, 2)))
1931 :			CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data);
1932 :
1933 :			if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) // diamond only if needed
1934 :			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1935 :
1936 :			for (i = 0; i < 4; i++) {
1937 :			MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1) * pParam->mb_width)];
1938 :			MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
1939 :			MB->mode = MODE_INTER;
1940 :			MB->sad16 = Data->iMinSAD[i+1];
1941 :			}
1942 :			}
1943 :	Isibaar	1.44.2.2	}
1944 :
1945 :	syskin	1.44.2.37	#define INTRA_BIAS 2500
1946 :			#define INTRA_THRESH 1500
1947 :			#define INTER_THRESH 1400
1948 :	Isibaar	1.44.2.2
1949 :	syskin	1.44.2.18
1950 :	Isibaar	1.44.2.2	int
1951 :			MEanalysis( const IMAGE * const pRef,
1952 :	syskin	1.44.2.18	FRAMEINFO * const Current,
1953 :	Isibaar	1.44.2.2	MBParam * const pParam,
1954 :	syskin	1.44.2.18	int maxIntra, //maximum number if non-I frames
1955 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1956 :			int bCount) // number if B frames in a row
1957 :	Isibaar	1.44.2.2	{
1958 :			uint32_t x, y, intra = 0;
1959 :			int sSAD = 0;
1960 :	syskin	1.44.2.18	MACROBLOCK * const pMBs = Current->mbs;
1961 :			const IMAGE * const pCurrent = &Current->image;
1962 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1963 :	syskin	1.44.2.37	const VECTOR zeroMV = {0,0};
1964 :	h	1.44.2.3
1965 :	syskin	1.44.2.37	int32_t iMinSAD[5], temp[5];
1966 :			VECTOR currentMV[5];
1967 :	h	1.44.2.3	SearchData Data;
1968 :			Data.iEdgedWidth = pParam->edged_width;
1969 :	syskin	1.44.2.37	Data.currentMV = currentMV;
1970 :			Data.iMinSAD = iMinSAD;
1971 :	syskin	1.44.2.18	Data.iFcode = Current->fcode;
1972 :	syskin	1.44.2.37	Data.rrv = Current->global_flags & XVID_REDUCED;
1973 :			Data.temp = temp;
1974 :			CheckCandidate = CheckCandidate32I;
1975 :	h	1.44.2.3
1976 :	syskin	1.44.2.19	if (intraCount < 10) // we're right after an I frame
1977 :			IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
1978 :	syskin	1.44.2.18	else
1979 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
1980 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
1981 :
1982 :	syskin	1.44.2.26	InterThresh += 400 * (1 - bCount);
1983 :	syskin	1.44.2.37	if (InterThresh < 300) InterThresh = 300;
1984 :	syskin	1.44.2.18
1985 :	Isibaar	1.44.2.2	if (sadInit) (*sadInit) ();
1986 :
1987 :	syskin	1.44.2.37	for (y = 1; y < pParam->mb_height-1; y+=2) {
1988 :			for (x = 1; x < pParam->mb_width-1; x+=2) {
1989 :			int i;
1990 :
1991 :			if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
1992 :	Isibaar	1.44.2.2
1993 :	syskin	1.44.2.37	MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
1994 :	Isibaar	1.44.2.2
1995 :	syskin	1.44.2.37	for (i = 0; i < 4; i++) {
1996 :			int dev;
1997 :			MACROBLOCK pMB = &pMBs[x+(i&1) + y+(i>>1) pParam->mb_width];
1998 :			if (pMB->sad16 > IntraThresh) {
1999 :			dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1))* pParam->edged_width) * 16,
2000 :			pParam->edged_width);
2001 :			if (dev + IntraThresh < pMB->sad16) {
2002 :			pMB->mode = MODE_INTRA;
2003 :			if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
2004 :			}
2005 :	Isibaar	1.44.2.2	}
2006 :	syskin	1.44.2.37	sSAD += pMB->sad16;
2007 :	syskin	1.44.2.18	}
2008 :	suxen_drol	1.6	}
2009 :			}
2010 :	syskin	1.44.2.32	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2011 :	syskin	1.44.2.37	if (sSAD > IntraThresh + INTRA_BIAS ) return I_VOP;
2012 :	suxen_drol	1.44.2.33	if (sSAD > InterThresh ) return P_VOP;
2013 :	Isibaar	1.44.2.2	emms();
2014 :	suxen_drol	1.44.2.33	return B_VOP;
2015 :	syskin	1.44.2.34
2016 :	chl	1.44.2.13	}
2017 :
2018 :	syskin	1.44.2.14	static void
2019 :	syskin	1.44.2.40	CheckGMC(int x, int y, const int dir, int * iDirection,
2020 :	syskin	1.44.2.14	const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
2021 :			const MBParam * const pParam)
2022 :			{
2023 :			uint32_t mx, my, a, count = 0;
2024 :
2025 :			for (my = 1; my < pParam->mb_height-1; my++)
2026 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2027 :			VECTOR mv;
2028 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2029 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED) continue;
2030 :			mv = pMB->mvs[0];
2031 :			a = ABS(mv.x - x) + ABS(mv.y - y);
2032 :			if (a < 6) count += 6 - a;
2033 :			}
2034 :
2035 :			if (count > *bestcount) {
2036 :			*bestcount = count;
2037 :			*iDirection = dir;
2038 :	syskin	1.44.2.40	GMC->x = x; GMC->y = y;
2039 :	syskin	1.44.2.14	}
2040 :			}
2041 :	chl	1.44.2.13
2042 :
2043 :	suxen_drol	1.44.2.35	static VECTOR
2044 :	chl	1.44.2.13	GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2045 :			{
2046 :
2047 :	syskin	1.44.2.14	uint32_t count, bestcount = 0;
2048 :	chl	1.44.2.13	int x, y;
2049 :	syskin	1.44.2.14	VECTOR gmc = {0,0};
2050 :	chl	1.44.2.13	int step, min_x, max_x, min_y, max_y;
2051 :			uint32_t mx, my;
2052 :	syskin	1.44.2.14	int iDirection, bDirection;
2053 :	chl	1.44.2.13
2054 :			min_x = min_y = -32<<iFcode;
2055 :			max_x = max_y = 32<<iFcode;
2056 :
2057 :	syskin	1.44.2.14	//step1: let's find a rough camera panning
2058 :	chl	1.44.2.13	for (step = 32; step >= 2; step /= 2) {
2059 :			bestcount = 0;
2060 :			for (y = min_y; y <= max_y; y += step)
2061 :			for (x = min_x ; x <= max_x; x += step) {
2062 :			count = 0;
2063 :			//for all macroblocks
2064 :			for (my = 1; my < pParam->mb_height-1; my++)
2065 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2066 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2067 :			VECTOR mv;
2068 :
2069 :	syskin	1.44.2.40	if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2070 :	chl	1.44.2.13	continue;
2071 :
2072 :			mv = pMB->mvs[0];
2073 :			if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2074 :			count++;
2075 :			}
2076 :			if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2077 :			}
2078 :			min_x = gmc.x - step;
2079 :			max_x = gmc.x + step;
2080 :			min_y = gmc.y - step;
2081 :			max_y = gmc.y + step;
2082 :			}
2083 :	syskin	1.44.2.14
2084 :	syskin	1.44.2.40	if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2085 :	syskin	1.44.2.14	gmc.x = gmc.y = 0; //no camara pan, no GMC
2086 :
2087 :	syskin	1.44.2.40	// step2: let's refine camera panning using gradiend-descent approach
2088 :	syskin	1.44.2.14	// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
2089 :			bestcount = 0;
2090 :			CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
2091 :			do {
2092 :			x = gmc.x; y = gmc.y;
2093 :			bDirection = iDirection; iDirection = 0;
2094 :			if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2095 :			if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2096 :			if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
2097 :			if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2098 :
2099 :			} while (iDirection);
2100 :
2101 :	chl	1.44.2.13	if (pParam->m_quarterpel) {
2102 :			gmc.x *= 2;
2103 :			gmc.y = 2; / we store the halfpel value as pseudo-qpel to make comparison easier */
2104 :			}
2105 :
2106 :			return gmc;
2107 :	Isibaar	1.44.2.2	}

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