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

Revision 1.44.2.36 - (view) (download)

1 :	Isibaar	1.1	/**************************************************************************
2 :			*
3 :	chl	1.26	* XVID MPEG-4 VIDEO CODEC
4 :			* motion estimation
5 :			*
6 :			* This program is an implementation of a part of one or more MPEG-4
7 :			* Video tools as specified in ISO/IEC 14496-2 standard. Those intending
8 :			* to use this software module in hardware or software products are
9 :			* advised that its use may infringe existing patents or copyrights, and
10 :			* any such use would be at such party's own risk. The original
11 :			* developer of this software module and his/her company, and subsequent
12 :			* editors and their companies, will have no liability for use of this
13 :			* software or modifications or derivatives thereof.
14 :			*
15 :			* This program is free software; you can redistribute it and/or modify
16 :			* it under the terms of the GNU General Public License as published by
17 :			* the Free Software Foundation; either version 2 of the License, or
18 :			* (at your option) any later version.
19 :			*
20 :			* This program is distributed in the hope that it will be useful,
21 :			* but WITHOUT ANY WARRANTY; without even the implied warranty of
22 :			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 :			* GNU General Public License for more details.
24 :			*
25 :			* You should have received a copy of the GNU General Public License
26 :			* along with this program; if not, write to the Free Software
27 :			* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 :			*
29 :			*************************************************************************/
30 :
31 :	Isibaar	1.1	#include <assert.h>
32 :			#include <stdio.h>
33 :	chl	1.3	#include <stdlib.h>
34 :	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	1.44.2.36	x = pred.x - x;
62 :			y = pred.y - y;
63 :	chl	1.44.2.1
64 :			if (x == 0) xb = 1;
65 :			else {
66 :			if (x < 0) x = -x;
67 :			x += (1 << (iFcode - 1)) - 1;
68 :			x >>= (iFcode - 1);
69 :			if (x > 32) x = 32;
70 :			xb = mvtab[x] + iFcode;
71 :			}
72 :
73 :			if (y == 0) yb = 1;
74 :			else {
75 :			if (y < 0) y = -y;
76 :			y += (1 << (iFcode - 1)) - 1;
77 :			y >>= (iFcode - 1);
78 :			if (y > 32) y = 32;
79 :			yb = mvtab[y] + iFcode;
80 :			}
81 :			return xb + yb;
82 :			}
83 :
84 :	syskin	1.44.2.14	static int32_t
85 :			ChromaSAD(int dx, int dy, const SearchData * const data)
86 :	chl	1.44.2.1	{
87 :	syskin	1.44.2.14	int sad;
88 :			dx = (dx >> 1) + roundtab_79[dx & 0x3];
89 :			dy = (dy >> 1) + roundtab_79[dy & 0x3];
90 :
91 :	syskin	1.44.2.34	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
92 :
93 :			switch (((dx & 1) << 1) \| (dy & 1)) {
94 :	syskin	1.44.2.14	case 0:
95 :			sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
96 :			sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
97 :			break;
98 :			case 1:
99 :			dx = dx / 2; dy = (dy - 1) / 2;
100 :			sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
101 :			sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
102 :			break;
103 :			case 2:
104 :			dx = (dx - 1) / 2; dy = dy / 2;
105 :			sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
106 :			sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
107 :			break;
108 :			default:
109 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
110 :			interpolate8x8_halfpel_hv(data->RefQ,
111 :			data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
112 :			data->rounding);
113 :			sad = sad8(data->CurU, data->RefQ, data->iEdgedWidth/2);
114 :			interpolate8x8_halfpel_hv(data->RefQ,
115 :			data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
116 :			data->rounding);
117 :			sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2);
118 :			break;
119 :	chl	1.44.2.1	}
120 :	syskin	1.44.2.34	data->temp[5] = dx; data->temp[6] = dy; data->temp[7] = sad; //backup
121 :	syskin	1.44.2.14	return sad;
122 :			}
123 :	chl	1.10
124 :	syskin	1.44.2.22	static __inline const uint8_t *
125 :			GetReference(const int x, const int y, const int dir, const SearchData * const data)
126 :	Isibaar	1.44.2.12	{
127 :	syskin	1.44.2.22	// dir : 0 = forward, 1 = backward
128 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
129 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
130 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
131 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
132 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
133 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
134 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
135 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
136 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
137 :	Isibaar	1.44.2.12	}
138 :			}
139 :
140 :	syskin	1.44.2.22	static uint8_t *
141 :			Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data)
142 :	Isibaar	1.1	{
143 :	syskin	1.44.2.22	// create or find a qpel-precision reference picture; return pointer to it
144 :	syskin	1.44.2.24	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
145 :	syskin	1.44.2.22	const int32_t iEdgedWidth = data->iEdgedWidth;
146 :			const uint32_t rounding = data->rounding;
147 :			const int halfpel_x = x/2;
148 :			const int halfpel_y = y/2;
149 :			const uint8_t ref1, ref2, ref3, ref4;
150 :	Isibaar	1.1
151 :	syskin	1.44.2.22	ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases
152 :			ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
153 :			switch( ((x&1)<<1) + (y&1) ) {
154 :			case 0: // pure halfpel position
155 :			Reference = (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data);
156 :			Reference += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
157 :			break;
158 :	Isibaar	1.1
159 :	syskin	1.44.2.22	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
160 :			ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
161 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
162 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
163 :	syskin	1.44.2.22	break;
164 :	Isibaar	1.1
165 :	syskin	1.44.2.22	case 2: // x qpel, y halfpel - left or right during qpel refinement
166 :			ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data);
167 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
168 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
169 :	syskin	1.44.2.22	break;
170 :	Isibaar	1.1
171 :	syskin	1.44.2.22	default: // x and y in qpel resolution - the "corners" (top left/right and
172 :			// bottom left/right) during qpel refinement
173 :			ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
174 :			ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data);
175 :			ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data);
176 :			ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
177 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
178 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
179 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
180 :			break;
181 :			}
182 :			return Reference;
183 :	Isibaar	1.1	}
184 :
185 :	syskin	1.44.2.22	static uint8_t *
186 :			Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data)
187 :	Isibaar	1.44.2.7	{
188 :	syskin	1.44.2.22	// create or find a qpel-precision reference picture; return pointer to it
189 :	syskin	1.44.2.24	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
190 :	syskin	1.44.2.22	const int32_t iEdgedWidth = data->iEdgedWidth;
191 :			const uint32_t rounding = data->rounding;
192 :			const int halfpel_x = x/2;
193 :			const int halfpel_y = y/2;
194 :	Isibaar	1.44.2.7	const uint8_t ref1, ref2, ref3, ref4;
195 :
196 :	syskin	1.44.2.22	ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases
197 :			switch( ((x&1)<<1) + (y&1) ) {
198 :			case 0: // pure halfpel position
199 :			return (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data);
200 :	Isibaar	1.44.2.7	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
201 :	syskin	1.44.2.22	ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
202 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
203 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
204 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
205 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
206 :	Isibaar	1.44.2.7	break;
207 :
208 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
209 :	syskin	1.44.2.22	ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data);
210 :	Isibaar	1.44.2.23	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
211 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
212 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
213 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
214 :	Isibaar	1.44.2.7	break;
215 :
216 :			default: // x and y in qpel resolution - the "corners" (top left/right and
217 :			// bottom left/right) during qpel refinement
218 :	syskin	1.44.2.22	ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data);
219 :			ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data);
220 :			ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data);
221 :	Isibaar	1.44.2.7	interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
222 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
223 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
224 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
225 :			break;
226 :			}
227 :	syskin	1.44.2.22	return Reference;
228 :			}
229 :
230 :			/* CHECK_CANDIATE FUNCTIONS START */
231 :
232 :			static void
233 :			CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
234 :			{
235 :			int t, xc, yc;
236 :			const uint8_t * Reference;
237 :			VECTOR * current;
238 :
239 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
240 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
241 :
242 :			if (data->qpel_precision) { // x and y are in 1/4 precision
243 :			Reference = Interpolate16x16qpel(x, y, 0, data);
244 :			xc = x/2; yc = y/2; //for chroma sad
245 :			current = data->currentQMV;
246 :			} else {
247 :	syskin	1.44.2.34	Reference = GetReference(x, y, 0, data);
248 :	syskin	1.44.2.22	current = data->currentMV;
249 :			xc = x; yc = y;
250 :			}
251 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
252 :	Isibaar	1.44.2.7
253 :	syskin	1.44.2.22	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
254 :	Isibaar	1.44.2.7
255 :	syskin	1.44.2.14	data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
256 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
257 :
258 :	syskin	1.44.2.22	if (data->chroma) data->temp[0] += ChromaSAD(xc, yc, data);
259 :	Isibaar	1.44.2.7
260 :			if (data->temp[0] < data->iMinSAD[0]) {
261 :			data->iMinSAD[0] = data->temp[0];
262 :	syskin	1.44.2.22	current[0].x = x; current[0].y = y;
263 :			*dir = Direction; }
264 :	Isibaar	1.44.2.7
265 :			if (data->temp[1] < data->iMinSAD[1]) {
266 :	syskin	1.44.2.22	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; }
267 :	Isibaar	1.44.2.7	if (data->temp[2] < data->iMinSAD[2]) {
268 :	syskin	1.44.2.22	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
269 :	Isibaar	1.44.2.7	if (data->temp[3] < data->iMinSAD[3]) {
270 :	syskin	1.44.2.22	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
271 :	Isibaar	1.44.2.7	if (data->temp[4] < data->iMinSAD[4]) {
272 :	syskin	1.44.2.22	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
273 :
274 :	Isibaar	1.44.2.7	}
275 :
276 :			static void
277 :	syskin	1.44.2.32	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
278 :			{
279 :			int t;
280 :			const uint8_t * Reference;
281 :
282 :			if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) \|\| //non-zero integer value
283 :			( x > data->max_dx) \|\| ( x < data->min_dx)
284 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
285 :
286 :	syskin	1.44.2.34	Reference = GetReference(x, y, 0, data);
287 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
288 :	syskin	1.44.2.32
289 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
290 :
291 :			data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
292 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
293 :
294 :			if (data->temp[0] < data->iMinSAD[0]) {
295 :			data->iMinSAD[0] = data->temp[0];
296 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
297 :			*dir = Direction; }
298 :
299 :			if (data->temp[1] < data->iMinSAD[1]) {
300 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
301 :			if (data->temp[2] < data->iMinSAD[2]) {
302 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
303 :			if (data->temp[3] < data->iMinSAD[3]) {
304 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
305 :			if (data->temp[4] < data->iMinSAD[4]) {
306 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
307 :			}
308 :
309 :			static void
310 :	syskin	1.44.2.22	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
311 :	syskin	1.44.2.19	{
312 :			int32_t sad;
313 :	syskin	1.44.2.22	const uint8_t * Reference;
314 :			int t;
315 :			VECTOR * current;
316 :	syskin	1.44.2.19
317 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
318 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
319 :
320 :	syskin	1.44.2.34	if (data->rrv)
321 :	syskin	1.44.2.32	if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) ) return; //non-zero integer value
322 :
323 :	syskin	1.44.2.22	if (data->qpel_precision) { // x and y are in 1/4 precision
324 :			Reference = Interpolate16x16qpel(x, y, 0, data);
325 :			current = data->currentQMV;
326 :			} else {
327 :	syskin	1.44.2.34	Reference = GetReference(x, y, 0, data);
328 :	syskin	1.44.2.22	current = data->currentMV;
329 :	syskin	1.44.2.19	}
330 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode,
331 :	syskin	1.44.2.34	data->qpel && !data->qpel_precision && !data->rrv, data->rrv);
332 :	syskin	1.44.2.19
333 :			sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
334 :	syskin	1.44.2.22	sad += (data->lambda16 * t * sad)/1000;
335 :	syskin	1.44.2.19
336 :	syskin	1.44.2.22	if (sad < *(data->iMinSAD)) {
337 :			*(data->iMinSAD) = sad;
338 :			current->x = x; current->y = y;
339 :			*dir = Direction; }
340 :	syskin	1.44.2.19	}
341 :
342 :			static void
343 :	Isibaar	1.44.2.2	CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
344 :			{
345 :	syskin	1.44.2.19	// maximum speed - for P/B/I decision
346 :	Isibaar	1.44.2.2	int32_t sad;
347 :
348 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
349 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
350 :
351 :	syskin	1.44.2.11	sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
352 :	Isibaar	1.44.2.2	data->iEdgedWidth, 256*4096);
353 :
354 :			if (sad < *(data->iMinSAD)) {
355 :			*(data->iMinSAD) = sad;
356 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
357 :			*dir = Direction; }
358 :			}
359 :
360 :
361 :			static void
362 :	chl	1.44.2.1	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
363 :	Isibaar	1.1	{
364 :	chl	1.44.2.1	int32_t sad;
365 :	syskin	1.44.2.22	int xb, yb, t;
366 :	chl	1.44.2.1	const uint8_t ReferenceF, ReferenceB;
367 :	syskin	1.44.2.22	VECTOR *current;
368 :	suxen_drol	1.8
369 :	chl	1.44.2.1	if (( xf > data->max_dx) \|\| ( xf < data->min_dx)
370 :			\|\| ( yf > data->max_dy) \|\| (yf < data->min_dy)) return;
371 :	suxen_drol	1.8
372 :	syskin	1.44.2.22	if (data->qpel_precision) {
373 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
374 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
375 :			current = data->currentQMV;
376 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
377 :			} else {
378 :	syskin	1.44.2.34	ReferenceF = GetReference(xf, yf, 0, data);
379 :	syskin	1.44.2.22	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
380 :	syskin	1.44.2.34	ReferenceB = GetReference(xb, yb, 1, data);
381 :	syskin	1.44.2.22	current = data->currentMV;
382 :	syskin	1.44.2.19	}
383 :
384 :	syskin	1.44.2.36	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0)
385 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
386 :	syskin	1.44.2.34
387 :	syskin	1.44.2.19	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
388 :	syskin	1.44.2.22	sad += (data->lambda16 * t * sad)/1000;
389 :	syskin	1.44.2.19
390 :			if (sad < *(data->iMinSAD)) {
391 :			*(data->iMinSAD) = sad;
392 :	syskin	1.44.2.22	current->x = xf; current->y = yf;
393 :	syskin	1.44.2.19	*dir = Direction; }
394 :			}
395 :
396 :	chl	1.44.2.1	static void
397 :			CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
398 :			{
399 :	syskin	1.44.2.14	int32_t sad = 0;
400 :	chl	1.44.2.1	int k;
401 :			const uint8_t *ReferenceF;
402 :			const uint8_t *ReferenceB;
403 :			VECTOR mvs, b_mvs;
404 :	syskin	1.44.2.36	const VECTOR zeroMV={0,0};
405 :	edgomez	1.20
406 :	chl	1.44.2.1	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
407 :	edgomez	1.20
408 :	chl	1.44.2.1	for (k = 0; k < 4; k++) {
409 :			mvs.x = data->directmvF[k].x + x;
410 :			b_mvs.x = ((x == 0) ?
411 :			data->directmvB[k].x
412 :			: mvs.x - data->referencemv[k].x);
413 :	edgomez	1.20
414 :	chl	1.44.2.1	mvs.y = data->directmvF[k].y + y;
415 :			b_mvs.y = ((y == 0) ?
416 :			data->directmvB[k].y
417 :			: mvs.y - data->referencemv[k].y);
418 :
419 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
420 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
421 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
422 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
423 :
424 :	syskin	1.44.2.22	if (!data->qpel) {
425 :			mvs.x = 2; mvs.y = 2;
426 :			b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
427 :	chl	1.44.2.1	}
428 :	syskin	1.44.2.22	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
429 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
430 :	chl	1.44.2.1
431 :			sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
432 :	syskin	1.44.2.22	ReferenceF, ReferenceB,
433 :	chl	1.44.2.1	data->iEdgedWidth);
434 :			if (sad > *(data->iMinSAD)) return;
435 :			}
436 :	chl	1.17
437 :	syskin	1.44.2.36	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000;
438 :	syskin	1.44.2.14
439 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
440 :			*(data->iMinSAD) = sad;
441 :			data->currentMV->x = x; data->currentMV->y = y;
442 :			*dir = Direction; }
443 :	Isibaar	1.1	}
444 :
445 :	chl	1.44.2.1	static void
446 :			CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
447 :			{
448 :			int32_t sad;
449 :			const uint8_t *ReferenceF;
450 :			const uint8_t *ReferenceB;
451 :			VECTOR mvs, b_mvs;
452 :	syskin	1.44.2.36	const VECTOR zeroMV = {0,0};
453 :	chl	1.35
454 :	syskin	1.44.2.6	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
455 :	Isibaar	1.1
456 :	chl	1.44.2.1	mvs.x = data->directmvF[0].x + x;
457 :			b_mvs.x = ((x == 0) ?
458 :			data->directmvB[0].x
459 :			: mvs.x - data->referencemv[0].x);
460 :
461 :			mvs.y = data->directmvF[0].y + y;
462 :			b_mvs.y = ((y == 0) ?
463 :			data->directmvB[0].y
464 :			: mvs.y - data->referencemv[0].y);
465 :
466 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
467 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
468 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
469 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
470 :
471 :	syskin	1.44.2.22	if (!data->qpel) {
472 :			mvs.x = 2; mvs.y = 2;
473 :			b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
474 :	syskin	1.44.2.34	}
475 :	syskin	1.44.2.22	ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
476 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
477 :	chl	1.44.2.1
478 :	syskin	1.44.2.14	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
479 :	syskin	1.44.2.36	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000;
480 :	Isibaar	1.1
481 :	chl	1.44.2.1	if (sad < *(data->iMinSAD)) {
482 :			*(data->iMinSAD) = sad;
483 :			data->currentMV->x = x; data->currentMV->y = y;
484 :			*dir = Direction; }
485 :			}
486 :	Isibaar	1.1
487 :	chl	1.44.2.1	static void
488 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
489 :			{
490 :	syskin	1.44.2.14	int32_t sad; int t;
491 :	chl	1.44.2.1	const uint8_t * Reference;
492 :	Isibaar	1.1
493 :	chl	1.44.2.1	if (( x > data->max_dx) \|\| ( x < data->min_dx)
494 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
495 :	Isibaar	1.1
496 :	syskin	1.44.2.22	if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data);
497 :	syskin	1.44.2.34	else Reference = GetReference(x, y, 0, data);
498 :	Isibaar	1.1
499 :	chl	1.44.2.1	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
500 :	syskin	1.44.2.36	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
501 :	Isibaar	1.44.2.12
502 :	syskin	1.44.2.14	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100;
503 :	Isibaar	1.44.2.12
504 :			if (sad < *(data->iMinSAD)) {
505 :			*(data->iMinSAD) = sad;
506 :			data->currentMV->x = x; data->currentMV->y = y;
507 :			*dir = Direction; }
508 :			}
509 :
510 :	Isibaar	1.44.2.2	/* CHECK_CANDIATE FUNCTIONS END */
511 :	chl	1.3
512 :	chl	1.44.2.1	/* MAINSEARCH FUNCTIONS START */
513 :	chl	1.18
514 :	chl	1.44.2.1	static void
515 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
516 :			{
517 :	chl	1.18
518 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
519 :
520 :	syskin	1.44.2.34	int iDirection;
521 :
522 :			for(;;) { //forever
523 :			iDirection = 0;
524 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
525 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
526 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
527 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
528 :	edgomez	1.20
529 :	syskin	1.44.2.34	/* now we're doing diagonal checks near our candidate */
530 :
531 :			if (iDirection) { //checking if anything found
532 :			bDirection = iDirection;
533 :	chl	1.18	iDirection = 0;
534 :	syskin	1.44.2.34	x = data->currentMV->x; y = data->currentMV->y;
535 :			if (bDirection & 3) { //our candidate is left or right
536 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
537 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
538 :			} else { // what remains here is up or down
539 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
540 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1);
541 :			}
542 :
543 :			if (iDirection) {
544 :			bDirection += iDirection;
545 :			x = data->currentMV->x; y = data->currentMV->y;
546 :			}
547 :			} else { //about to quit, eh? not so fast....
548 :			switch (bDirection) {
549 :			case 2:
550 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
551 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
552 :			break;
553 :			case 1:
554 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
555 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
556 :			break;
557 :			case 2 + 4:
558 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
559 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
560 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
561 :			break;
562 :			case 4:
563 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
564 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
565 :			break;
566 :			case 8:
567 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
568 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
569 :			break;
570 :			case 1 + 4:
571 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
572 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
573 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
574 :			break;
575 :			case 2 + 8:
576 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
577 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
578 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
579 :			break;
580 :			case 1 + 8:
581 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
582 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
583 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
584 :			break;
585 :			default: //1+2+4+8 == we didn't find anything at all
586 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
587 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
588 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
589 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
590 :			break;
591 :	chl	1.18	}
592 :	syskin	1.44.2.34	if (!iDirection) break; //ok, the end. really
593 :			bDirection = iDirection;
594 :			x = data->currentMV->x; y = data->currentMV->y;
595 :	chl	1.18	}
596 :	syskin	1.44.2.34	}
597 :	chl	1.18	}
598 :
599 :	chl	1.44.2.1	static void
600 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
601 :			{
602 :			int iDirection;
603 :	chl	1.35
604 :	chl	1.44.2.1	do {
605 :			iDirection = 0;
606 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
607 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
608 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
609 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
610 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
611 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
612 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
613 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
614 :
615 :			bDirection = iDirection;
616 :			x = data->currentMV->x; y = data->currentMV->y;
617 :			} while (iDirection);
618 :	chl	1.35	}
619 :
620 :	chl	1.44.2.1	static void
621 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
622 :	chl	1.18	{
623 :
624 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
625 :
626 :	syskin	1.44.2.34	int iDirection;
627 :	edgomez	1.20
628 :	syskin	1.44.2.34	do {
629 :			iDirection = 0;
630 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
631 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
632 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
633 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
634 :	chl	1.18
635 :	syskin	1.44.2.34	/* now we're doing diagonal checks near our candidate */
636 :
637 :			if (iDirection) { //checking if anything found
638 :			bDirection = iDirection;
639 :			iDirection = 0;
640 :			x = data->currentMV->x; y = data->currentMV->y;
641 :			if (bDirection & 3) { //our candidate is left or right
642 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
643 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
644 :			} else { // what remains here is up or down
645 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
646 :			CHECK_CANDIDATE(x - iDiamondSize, y, 1);
647 :	chl	1.18	}
648 :	syskin	1.44.2.34	bDirection += iDirection;
649 :			x = data->currentMV->x; y = data->currentMV->y;
650 :	chl	1.18	}
651 :	syskin	1.44.2.34	}
652 :			while (iDirection);
653 :	chl	1.18	}
654 :
655 :	chl	1.44.2.1	/* MAINSEARCH FUNCTIONS END */
656 :
657 :			/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
658 :	chl	1.18
659 :	chl	1.44.2.1	static void
660 :	syskin	1.44.2.22	SubpelRefine(const SearchData * const data)
661 :	Isibaar	1.1	{
662 :	syskin	1.44.2.22	/* Do a half-pel or q-pel refinement */
663 :			VECTOR backupMV;
664 :	chl	1.44.2.1	int iDirection; //not needed
665 :	edgomez	1.20
666 :	syskin	1.44.2.22	if (data->qpel_precision)
667 :			backupMV = *(data->currentQMV);
668 :			else backupMV = *(data->currentMV);
669 :	Isibaar	1.44.2.7
670 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
671 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
672 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
673 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
674 :
675 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
676 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
677 :
678 :			CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
679 :			CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
680 :			}
681 :
682 :	chl	1.44.2.1	static __inline int
683 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
684 :			const int x, const int y,
685 :	syskin	1.44.2.32	const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv)
686 :	Isibaar	1.1
687 :	chl	1.44.2.1	{
688 :			/* keep repeating checks for all b-frames before this P frame,
689 :			to make sure that SKIP is possible (todo)
690 :			how: if skip is not possible set sad00 to a very high value */
691 :	syskin	1.44.2.32	if(rrv) {
692 :			uint32_t sadC = sad16(current->u + x16 + y(iEdgedWidth/2)*16,
693 :			reference->u + x16 + y(iEdgedWidth/2)16, iEdgedWidth/2, 2564096);
694 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
695 :			sadC += sad16(current->v + (x + y(iEdgedWidth/2))16,
696 :			reference->v + (x + y(iEdgedWidth/2))16, iEdgedWidth/2, 256*4096);
697 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
698 :			return 1;
699 :			} else {
700 :			uint32_t sadC = sad8(current->u + x8 + y(iEdgedWidth/2)*8,
701 :			reference->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2);
702 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
703 :			sadC += sad8(current->v + (x + y(iEdgedWidth/2))8,
704 :			reference->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
705 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
706 :			return 1;
707 :			}
708 :	chl	1.44.2.1	}
709 :	Isibaar	1.1
710 :	chl	1.44.2.1	static __inline void
711 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
712 :			{
713 :			pMB->mode = MODE_NOT_CODED;
714 :	Isibaar	1.44.2.2	pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
715 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
716 :	syskin	1.44.2.14
717 :	Isibaar	1.44.2.7	pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
718 :			pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
719 :
720 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
721 :			}
722 :
723 :	syskin	1.44.2.14	bool
724 :	chl	1.44.2.1	MotionEstimation(MBParam * const pParam,
725 :			FRAMEINFO * const current,
726 :			FRAMEINFO * const reference,
727 :			const IMAGE * const pRefH,
728 :			const IMAGE * const pRefV,
729 :			const IMAGE * const pRefHV,
730 :			const uint32_t iLimit)
731 :			{
732 :			MACROBLOCK *const pMBs = current->mbs;
733 :			const IMAGE *const pCurrent = &current->image;
734 :			const IMAGE *const pRef = &reference->image;
735 :	Isibaar	1.1
736 :	chl	1.44.2.1	const VECTOR zeroMV = { 0, 0 };
737 :	edgomez	1.20
738 :	syskin	1.44.2.32	uint32_t mb_width = pParam->mb_width;
739 :			uint32_t mb_height = pParam->mb_height;
740 :	suxen_drol	1.44.2.31
741 :	chl	1.44.2.1	uint32_t x, y;
742 :			uint32_t iIntra = 0;
743 :	syskin	1.44.2.14	int32_t InterBias, quant = current->quant, sad00;
744 :	Isibaar	1.44.2.7	uint8_t *qimage;
745 :	chl	1.44.2.1
746 :	Isibaar	1.44.2.2	// some pre-initialized thingies for SearchP
747 :	syskin	1.44.2.34	int32_t temp[8];
748 :	Isibaar	1.44.2.2	VECTOR currentMV[5];
749 :	Isibaar	1.44.2.7	VECTOR currentQMV[5];
750 :	Isibaar	1.44.2.2	int32_t iMinSAD[5];
751 :			SearchData Data;
752 :	syskin	1.44.2.32	memset(&Data, 0, sizeof(SearchData));
753 :	Isibaar	1.44.2.2	Data.iEdgedWidth = pParam->edged_width;
754 :			Data.currentMV = currentMV;
755 :	Isibaar	1.44.2.7	Data.currentQMV = currentQMV;
756 :	Isibaar	1.44.2.2	Data.iMinSAD = iMinSAD;
757 :			Data.temp = temp;
758 :			Data.iFcode = current->fcode;
759 :	Isibaar	1.44.2.7	Data.rounding = pParam->m_rounding_type;
760 :	syskin	1.44.2.14	Data.qpel = pParam->m_quarterpel;
761 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
762 :	syskin	1.44.2.32	Data.rrv = current->global_flags & XVID_REDUCED;
763 :	Isibaar	1.44.2.7
764 :	syskin	1.44.2.32	if ((current->global_flags & XVID_REDUCED)) {
765 :	suxen_drol	1.44.2.31	mb_width = (pParam->width + 31) / 32;
766 :			mb_height = (pParam->height + 31) / 32;
767 :	syskin	1.44.2.32	Data.qpel = Data.chroma = 0;
768 :	suxen_drol	1.44.2.31	}
769 :
770 :	Isibaar	1.44.2.7	if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
771 :	syskin	1.44.2.8	return 1; // allocate some mem for qpel interpolated blocks
772 :	Isibaar	1.44.2.7	// somehow this is dirty since I think we shouldn't use malloc outside
773 :			// encoder_create() - so please fix me!
774 :	syskin	1.44.2.14	Data.RefQ = qimage;
775 :	chl	1.44.2.1	if (sadInit) (*sadInit) ();
776 :
777 :	suxen_drol	1.44.2.31	for (y = 0; y < mb_height; y++) {
778 :			for (x = 0; x < mb_width; x++) {
779 :	chl	1.44.2.1	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
780 :	syskin	1.44.2.14
781 :	syskin	1.44.2.32	if (Data.rrv) pMB->sad16 =
782 :			sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32,
783 :			pRef->y + (x + y * pParam->edged_width) * 32,
784 :			pParam->edged_width, pMB->sad8 );
785 :
786 :			else pMB->sad16 =
787 :			sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
788 :	chl	1.44.2.1	pRef->y + (x + y * pParam->edged_width) * 16,
789 :			pParam->edged_width, pMB->sad8 );
790 :
791 :	syskin	1.44.2.14	if (Data.chroma) {
792 :			pMB->sad16 += sad8(pCurrent->u + x8 + y(pParam->edged_width/2)*8,
793 :			pRef->u + x8 + y(pParam->edged_width/2)*8, pParam->edged_width/2);
794 :
795 :			pMB->sad16 += sad8(pCurrent->v + (x + y(pParam->edged_width/2))8,
796 :			pRef->v + (x + y(pParam->edged_width/2))8, pParam->edged_width/2);
797 :			}
798 :
799 :			sad00 = pMB->sad16; //if no gmc; else sad00 = (..)
800 :
801 :	chl	1.44.2.1	if (!(current->global_flags & XVID_LUMIMASKING)) {
802 :			pMB->dquant = NO_CHANGE;
803 :	syskin	1.44.2.11	pMB->quant = current->quant;
804 :			} else {
805 :	syskin	1.44.2.5	if (pMB->dquant != NO_CHANGE) {
806 :			quant += DQtab[pMB->dquant];
807 :			if (quant > 31) quant = 31;
808 :			else if (quant < 1) quant = 1;
809 :			}
810 :	syskin	1.44.2.11	pMB->quant = quant;
811 :			}
812 :	chl	1.44.2.1
813 :	syskin	1.44.2.14	//initial skip decision
814 :	chl	1.44.2.13	/* no early skip for GMC (global vector = skip vector is unknown!) */
815 :			if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
816 :	syskin	1.44.2.32	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
817 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
818 :	chl	1.44.2.13	SkipMacroblockP(pMB, sad00);
819 :			continue;
820 :			}
821 :			}
822 :	syskin	1.44.2.14
823 :			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
824 :	chl	1.44.2.1	y, current->motion_flags, pMB->quant,
825 :	Isibaar	1.44.2.2	&Data, pParam, pMBs, reference->mbs,
826 :	chl	1.44.2.1	current->global_flags & XVID_INTER4V, pMB);
827 :
828 :			/* final skip decision, a.k.a. "the vector you found, really that good?" */
829 :	chl	1.44.2.13	if (current->coding_type == P_VOP) {
830 :			if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
831 :	syskin	1.44.2.32	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
832 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
833 :	chl	1.44.2.13	SkipMacroblockP(pMB, sad00);
834 :			continue;
835 :			}
836 :			}
837 :	syskin	1.44.2.14
838 :	chl	1.44.2.1	/* finally, intra decision */
839 :	Isibaar	1.1
840 :	chl	1.44.2.1	InterBias = MV16_INTER_BIAS;
841 :	syskin	1.44.2.16	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
842 :	chl	1.44.2.1	if (y != 0)
843 :	syskin	1.44.2.16	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
844 :	chl	1.44.2.1	if (x != 0)
845 :	syskin	1.44.2.16	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
846 :	syskin	1.44.2.14
847 :			if (Data.chroma) InterBias += 50; // to compensate bigger SAD
848 :	syskin	1.44.2.32	if (Data.rrv) InterBias *= 4; //??
849 :	chl	1.44.2.1
850 :			if (InterBias < pMB->sad16) {
851 :	syskin	1.44.2.32	int32_t deviation;
852 :			if (Data.rrv) {
853 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32,
854 :			pParam->edged_width)
855 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16,
856 :			pParam->edged_width)
857 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width,
858 :			pParam->edged_width)
859 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1),
860 :			pParam->edged_width);
861 :			} else
862 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
863 :	chl	1.44.2.1	pParam->edged_width);
864 :	edgomez	1.20
865 :	syskin	1.44.2.32	if (deviation < (pMB->sad16 - InterBias)) {
866 :	Isibaar	1.44.2.7	if (++iIntra >= iLimit) { free(qimage); return 1; }
867 :	chl	1.44.2.1	pMB->mode = MODE_INTRA;
868 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
869 :			pMB->mvs[3] = zeroMV;
870 :	Isibaar	1.44.2.7	pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
871 :			pMB->qmvs[3] = zeroMV;
872 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
873 :			pMB->sad8[3] = 0;
874 :			}
875 :			}
876 :			}
877 :			}
878 :	Isibaar	1.44.2.7	free(qimage);
879 :	chl	1.44.2.13
880 :			if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
881 :	syskin	1.44.2.14	current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
882 :	chl	1.44.2.13	else
883 :			current->GMC_MV = zeroMV;
884 :
885 :	syskin	1.44.2.14	return 0;
886 :	chl	1.44.2.1	}
887 :	chl	1.18
888 :	edgomez	1.20
889 :	chl	1.44.2.1	#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16\|PMV_HALFPELREFINE16)
890 :	edgomez	1.20
891 :	chl	1.44.2.1	static __inline int
892 :			make_mask(const VECTOR * const pmv, const int i)
893 :			{
894 :	Isibaar	1.44.2.2	int mask = 255, j;
895 :	chl	1.44.2.1	for (j = 0; j < i; j++) {
896 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
897 :			if (pmv[i].x == pmv[j].x) {
898 :			if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
899 :			if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
900 :			} else
901 :			if (pmv[i].y == pmv[j].y) {
902 :			if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
903 :			if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
904 :			}
905 :			}
906 :			return mask;
907 :			}
908 :
909 :			static __inline void
910 :	syskin	1.44.2.32	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
911 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
912 :	chl	1.44.2.1	{
913 :
914 :			//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
915 :	syskin	1.44.2.32	if (rrv) { iWcount /= 2; iHcount /= 2; }
916 :	chl	1.44.2.1
917 :	syskin	1.44.2.32	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
918 :	chl	1.44.2.1	pmv[5].x = EVEN(pmv[3].x);
919 :	Isibaar	1.44.2.2	pmv[5].y = EVEN(pmv[3].y);
920 :			} else pmv[5].x = pmv[5].y = 0;
921 :	chl	1.44.2.1
922 :			if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
923 :			else pmv[3].x = pmv[3].y = 0;
924 :
925 :			if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
926 :			else pmv[4].x = pmv[4].y = 0;
927 :
928 :			// [1] median prediction
929 :	syskin	1.44.2.32	if (rrv) { //median is in halfzero-precision
930 :			pmv[1].x = RRV_MV_SCALEUP(pmv[0].x);
931 :			pmv[1].y = RRV_MV_SCALEUP(pmv[0].y);
932 :			} else { pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); }
933 :	chl	1.44.2.1
934 :			pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
935 :
936 :			pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
937 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
938 :
939 :	syskin	1.44.2.32	if ((x < iWcount-1) && (y < iHcount-1)) {
940 :	chl	1.44.2.1	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
941 :	Isibaar	1.44.2.2	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
942 :			} else pmv[6].x = pmv[6].y = 0;
943 :	syskin	1.44.2.32
944 :			if (rrv) {
945 :			int i;
946 :			for (i = 0; i < 7; i++) {
947 :			pmv[i].x = RRV_MV_SCALEDOWN(pmv[i].x);
948 :			pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // a trick
949 :			}
950 :			}
951 :	chl	1.44.2.1	}
952 :
953 :			static void
954 :	syskin	1.44.2.14	SearchP(const IMAGE * const pRef,
955 :	chl	1.44.2.1	const uint8_t * const pRefH,
956 :			const uint8_t * const pRefV,
957 :			const uint8_t * const pRefHV,
958 :			const IMAGE * const pCur,
959 :			const int x,
960 :			const int y,
961 :			const uint32_t MotionFlags,
962 :			const uint32_t iQuant,
963 :	Isibaar	1.44.2.2	SearchData * const Data,
964 :	chl	1.44.2.1	const MBParam * const pParam,
965 :			const MACROBLOCK * const pMBs,
966 :			const MACROBLOCK * const prevMBs,
967 :			int inter4v,
968 :			MACROBLOCK * const pMB)
969 :			{
970 :	Isibaar	1.1
971 :	chl	1.44.2.1	int i, iDirection = 255, mask, threshA;
972 :	Isibaar	1.44.2.2	VECTOR pmv[7];
973 :	chl	1.44.2.1
974 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
975 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
976 :	syskin	1.44.2.32
977 :	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)()
978 :	syskin	1.44.2.32
979 :	syskin	1.44.2.34	Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache
980 :	syskin	1.44.2.32	if (Data->rrv) i = 2; else i = 1;
981 :			Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
982 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
983 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
984 :
985 :			Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
986 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
987 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
988 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
989 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
990 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
991 :	Isibaar	1.1
992 :	syskin	1.44.2.11	Data->lambda16 = lambda_vec16[iQuant];
993 :	Isibaar	1.44.2.12	Data->lambda8 = lambda_vec8[iQuant];
994 :	syskin	1.44.2.22	Data->qpel_precision = 0;
995 :	Isibaar	1.1
996 :	edgomez	1.44.2.4	if (pMB->dquant != NO_CHANGE) inter4v = 0;
997 :	Isibaar	1.44.2.2
998 :	Isibaar	1.44.2.7	for(i = 0; i < 5; i++)
999 :			Data->currentMV[i].x = Data->currentMV[i].y = 0;
1000 :	chl	1.44.2.1
1001 :	syskin	1.44.2.22	if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1002 :			else Data->predMV = pmv[0];
1003 :	Isibaar	1.44.2.12
1004 :	syskin	1.44.2.36	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
1005 :	syskin	1.44.2.14	Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000;
1006 :			Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100;
1007 :	Isibaar	1.44.2.2	Data->iMinSAD[2] = pMB->sad8[1];
1008 :			Data->iMinSAD[3] = pMB->sad8[2];
1009 :			Data->iMinSAD[4] = pMB->sad8[3];
1010 :	chl	1.44.2.1
1011 :			if ((x == 0) && (y == 0)) threshA = 512;
1012 :			else {
1013 :	edgomez	1.44.2.4	threshA = Data->temp[0]; // that's when we keep this SAD atm
1014 :	chl	1.44.2.1	if (threshA < 512) threshA = 512;
1015 :			if (threshA > 1024) threshA = 1024; }
1016 :
1017 :			PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1018 :	syskin	1.44.2.32	prevMBs + x + y * pParam->mb_width, Data->rrv);
1019 :	chl	1.44.2.1
1020 :	syskin	1.44.2.32	if (Data->rrv) CheckCandidate = CheckCandidate32;
1021 :			else if (inter4v \|\| Data->chroma) CheckCandidate = CheckCandidate16;
1022 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
1023 :	edgomez	1.44.2.4
1024 :	chl	1.44.2.1	/* main loop. checking all predictions */
1025 :
1026 :			for (i = 1; i < 7; i++) {
1027 :			if (!(mask = make_mask(pmv, i)) ) continue;
1028 :	edgomez	1.44.2.4	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1029 :	Isibaar	1.44.2.2	if (Data->iMinSAD[0] <= threshA) break;
1030 :	chl	1.44.2.1	}
1031 :
1032 :	Isibaar	1.44.2.2	if ((Data->iMinSAD[0] <= threshA) \|\|
1033 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1034 :			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1035 :	chl	1.44.2.1	inter4v = 0;
1036 :	Isibaar	1.44.2.2	} else {
1037 :	edgomez	1.20
1038 :	syskin	1.44.2.34	MainSearchFunc * MainSearchPtr;
1039 :	Isibaar	1.44.2.2	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1040 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1041 :			else MainSearchPtr = DiamondSearch;
1042 :	chl	1.35
1043 :	Isibaar	1.44.2.2	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1044 :	chl	1.13
1045 :	chl	1.44.2.1	/* extended search, diamond starting in 0,0 and in prediction.
1046 :			note that this search is/might be done in halfpel positions,
1047 :			which makes it more different than the diamond above */
1048 :	Isibaar	1.1
1049 :	Isibaar	1.44.2.2	if (MotionFlags & PMV_EXTSEARCH16) {
1050 :			int32_t bSAD;
1051 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1052 :	syskin	1.44.2.32	if (Data->rrv) {
1053 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1054 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1055 :			} else
1056 :			if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1057 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1058 :	Isibaar	1.44.2.2	if (!(MVequal(startMV, backupMV))) {
1059 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1060 :
1061 :	Isibaar	1.44.2.12	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1062 :	Isibaar	1.44.2.2	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1063 :			if (bSAD < Data->iMinSAD[0]) {
1064 :			Data->currentMV[0] = backupMV;
1065 :			Data->iMinSAD[0] = bSAD; }
1066 :			}
1067 :	chl	1.13
1068 :	Isibaar	1.44.2.2	backupMV = Data->currentMV[0];
1069 :	syskin	1.44.2.32	if (!MotionFlags & PMV_HALFPELREFINE16 \|\| Data->rrv) startMV.x = startMV.y = 0;
1070 :			else startMV.x = startMV.y = 1;
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	1.44.2.1	}
1081 :			}
1082 :	chl	1.13
1083 :	syskin	1.44.2.22	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1084 :	Isibaar	1.1
1085 :	Isibaar	1.44.2.7	for(i = 0; i < 5; i++) {
1086 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1087 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1088 :			}
1089 :
1090 :	syskin	1.44.2.32	if((!Data->rrv) && (pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1091 :	Isibaar	1.44.2.7
1092 :	syskin	1.44.2.22	Data->qpel_precision = 1;
1093 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1094 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1095 :	syskin	1.44.2.8
1096 :	syskin	1.44.2.22	SubpelRefine(Data);
1097 :	Isibaar	1.44.2.7	}
1098 :
1099 :	syskin	1.44.2.14	if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0;
1100 :	Isibaar	1.44.2.2	if (inter4v) {
1101 :			SearchData Data8;
1102 :			Data8.iFcode = Data->iFcode;
1103 :	Isibaar	1.44.2.12	Data8.lambda8 = Data->lambda8;
1104 :	Isibaar	1.44.2.2	Data8.iEdgedWidth = Data->iEdgedWidth;
1105 :	Isibaar	1.44.2.12	Data8.RefQ = Data->RefQ;
1106 :	syskin	1.44.2.14	Data8.qpel = Data->qpel;
1107 :	syskin	1.44.2.32	Data8.rrv = Data->rrv;
1108 :	Isibaar	1.44.2.2	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1109 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1110 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1111 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1112 :	syskin	1.44.2.14
1113 :			if (Data->chroma) {
1114 :	syskin	1.44.2.22	int sumx, sumy, dx, dy;
1115 :	syskin	1.44.2.14
1116 :			if(pParam->m_quarterpel) {
1117 :	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;
1118 :			sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2;
1119 :			} else {
1120 :			sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1121 :			sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1122 :			}
1123 :			dx = (sumx >> 3) + roundtab_76[sumx & 0xf];
1124 :			dy = (sumy >> 3) + roundtab_76[sumy & 0xf];
1125 :	syskin	1.44.2.14
1126 :			Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1127 :			}
1128 :	Isibaar	1.44.2.2	}
1129 :	chl	1.13
1130 :	syskin	1.44.2.32	if (Data->rrv) {
1131 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1132 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1133 :			}
1134 :	chl	1.44.2.1	if (!(inter4v) \|\|
1135 :	Isibaar	1.44.2.2	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1136 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1137 :	chl	1.44.2.1	// INTER MODE
1138 :			pMB->mode = MODE_INTER;
1139 :	Isibaar	1.44.2.2	pMB->mvs[0] = pMB->mvs[1]
1140 :			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1141 :	edgomez	1.20
1142 :	chl	1.44.2.1	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1143 :	Isibaar	1.44.2.2	pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1144 :	chl	1.13
1145 :	Isibaar	1.44.2.7	if(pParam->m_quarterpel) {
1146 :	syskin	1.44.2.22	pMB->qmvs[0] = pMB->qmvs[1]
1147 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1148 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1149 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1150 :	syskin	1.44.2.14	} else {
1151 :	Isibaar	1.44.2.7	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1152 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1153 :			}
1154 :	chl	1.44.2.1	} else {
1155 :			// INTER4V MODE; all other things are already set in Search8
1156 :			pMB->mode = MODE_INTER4V;
1157 :	Isibaar	1.44.2.2	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1158 :			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1159 :	chl	1.44.2.1	}
1160 :			}
1161 :	edgomez	1.20
1162 :	chl	1.44.2.1	static void
1163 :			Search8(const SearchData * const OldData,
1164 :			const int x, const int y,
1165 :			const uint32_t MotionFlags,
1166 :			const MBParam * const pParam,
1167 :			MACROBLOCK * const pMB,
1168 :			const MACROBLOCK * const pMBs,
1169 :	Isibaar	1.44.2.2	const int block,
1170 :			SearchData * const Data)
1171 :	chl	1.44.2.1	{
1172 :	syskin	1.44.2.32	int i = 0;
1173 :	Isibaar	1.44.2.2	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1174 :			Data->currentMV = OldData->currentMV + 1 + block;
1175 :	Isibaar	1.44.2.7	Data->currentQMV = OldData->currentQMV + 1 + block;
1176 :
1177 :	syskin	1.44.2.11	if(pParam->m_quarterpel) {
1178 :	syskin	1.44.2.32	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1179 :	syskin	1.44.2.36	if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1180 :			Data->predMV, Data->iFcode, 0, 0);
1181 :	syskin	1.44.2.34
1182 :	Isibaar	1.44.2.12	} else {
1183 :	syskin	1.44.2.32	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1184 :			if (block != 0) {
1185 :	syskin	1.44.2.36	if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1186 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1187 :	syskin	1.44.2.32	}
1188 :	Isibaar	1.44.2.12	}
1189 :	Isibaar	1.1
1190 :	syskin	1.44.2.32	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1191 :
1192 :	chl	1.44.2.1	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8)) {
1193 :	syskin	1.44.2.32	if (Data->rrv) i = 2; else i = 1;
1194 :	edgomez	1.20
1195 :	syskin	1.44.2.32	Data->Ref = OldData->Ref + i8 ((block&1) + pParam->edged_width*(block>>1));
1196 :			Data->RefH = OldData->RefH + i8 ((block&1) + pParam->edged_width*(block>>1));
1197 :			Data->RefV = OldData->RefV + i8 ((block&1) + pParam->edged_width*(block>>1));
1198 :			Data->RefHV = OldData->RefHV + i8 ((block&1) + pParam->edged_width*(block>>1));
1199 :	Isibaar	1.1
1200 :	syskin	1.44.2.32	Data->Cur = OldData->Cur + i8 ((block&1) + pParam->edged_width*(block>>1));
1201 :	syskin	1.44.2.22	Data->qpel_precision = 0;
1202 :	syskin	1.44.2.34
1203 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1204 :			pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv);
1205 :
1206 :			if (Data->rrv) CheckCandidate = CheckCandidate16no4v;
1207 :			else CheckCandidate = CheckCandidate8;
1208 :	Isibaar	1.1
1209 :	chl	1.44.2.1	if (MotionFlags & PMV_EXTSEARCH8) {
1210 :	Isibaar	1.44.2.7	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1211 :
1212 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1213 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1214 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1215 :			else MainSearchPtr = DiamondSearch;
1216 :	Isibaar	1.1
1217 :	Isibaar	1.44.2.7	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1218 :	Isibaar	1.1
1219 :	syskin	1.44.2.11	if(*(Data->iMinSAD) < temp_sad) {
1220 :	Isibaar	1.44.2.7	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1221 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1222 :			}
1223 :			}
1224 :
1225 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1226 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1227 :
1228 :	syskin	1.44.2.22	SubpelRefine(Data); // perform halfpel refine of current best vector
1229 :	Isibaar	1.44.2.7
1230 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1231 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1232 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1233 :			}
1234 :			}
1235 :
1236 :	syskin	1.44.2.34	if(!Data->rrv && Data->qpel) {
1237 :	syskin	1.44.2.11	if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1238 :			(MotionFlags & PMV_QUARTERPELREFINE8)) {
1239 :	syskin	1.44.2.22	Data->qpel_precision = 1;
1240 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1241 :			pParam->width, pParam->height, OldData->iFcode, 1, 0);
1242 :	syskin	1.44.2.22	SubpelRefine(Data);
1243 :	syskin	1.44.2.11	}
1244 :	Isibaar	1.44.2.7	}
1245 :			}
1246 :
1247 :	syskin	1.44.2.32	if (Data->rrv) {
1248 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1249 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1250 :			}
1251 :
1252 :	syskin	1.44.2.34	if(Data->qpel) {
1253 :	syskin	1.44.2.22	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1254 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1255 :			pMB->qmvs[block] = *(Data->currentQMV);
1256 :	syskin	1.44.2.32	} else {
1257 :	Isibaar	1.44.2.7	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1258 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1259 :	Isibaar	1.1	}
1260 :
1261 :	Isibaar	1.44.2.2	pMB->mvs[block] = *(Data->currentMV);
1262 :	syskin	1.44.2.11	pMB->sad8[block] = 4 * (*Data->iMinSAD);
1263 :	Isibaar	1.1	}
1264 :
1265 :	chl	1.44.2.1	/* B-frames code starts here */
1266 :	Isibaar	1.1
1267 :	chl	1.44.2.1	static __inline VECTOR
1268 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1269 :			{
1270 :			/* the stupidiest function ever */
1271 :			if (mode == MODE_FORWARD) return pMB->mvs[0];
1272 :			else return pMB->b_mvs[0];
1273 :			}
1274 :	Isibaar	1.1
1275 :	chl	1.44.2.1	static void __inline
1276 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1277 :			const uint32_t iWcount,
1278 :			const MACROBLOCK * const pMB,
1279 :			const uint32_t mode_curr)
1280 :			{
1281 :	Isibaar	1.1
1282 :	chl	1.44.2.1	// [0] is prediction
1283 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1284 :	Isibaar	1.1
1285 :	chl	1.44.2.1	pmv[1].x = pmv[1].y = 0; // [1] is zero
1286 :	Isibaar	1.1
1287 :	chl	1.44.2.1	pmv[2] = ChoosePred(pMB, mode_curr);
1288 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1289 :	Isibaar	1.1
1290 :	chl	1.44.2.1	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1291 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1292 :	Isibaar	1.44.2.2	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1293 :			} else pmv[3].x = pmv[3].y = 0;
1294 :	chl	1.39
1295 :	chl	1.44.2.1	if (y != 0) {
1296 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1297 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1298 :			} else pmv[4].x = pmv[4].y = 0;
1299 :	chl	1.39
1300 :	chl	1.44.2.1	if (x != 0) {
1301 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1302 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1303 :			} else pmv[5].x = pmv[5].y = 0;
1304 :	chl	1.39
1305 :	chl	1.44.2.1	if ((x != 0)&&(y != 0)) {
1306 :			pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1307 :			pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1308 :			} else pmv[6].x = pmv[6].y = 0;
1309 :	chl	1.39
1310 :	chl	1.44.2.1	// more?
1311 :	chl	1.39	}
1312 :
1313 :
1314 :	chl	1.44.2.1	/* search backward or forward, for b-frames */
1315 :			static void
1316 :			SearchBF( const uint8_t * const pRef,
1317 :			const uint8_t * const pRefH,
1318 :			const uint8_t * const pRefV,
1319 :			const uint8_t * const pRefHV,
1320 :			const IMAGE * const pCur,
1321 :			const int x, const int y,
1322 :			const uint32_t MotionFlags,
1323 :			const uint32_t iFcode,
1324 :			const MBParam * const pParam,
1325 :			MACROBLOCK * const pMB,
1326 :			const VECTOR * const predMV,
1327 :			int32_t * const best_sad,
1328 :	h	1.44.2.3	const int32_t mode_current,
1329 :			SearchData * const Data)
1330 :	Isibaar	1.1	{
1331 :
1332 :	chl	1.44.2.1	const int32_t iEdgedWidth = pParam->edged_width;
1333 :
1334 :			int i, iDirection, mask;
1335 :	h	1.44.2.3	VECTOR pmv[7];
1336 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1337 :	h	1.44.2.3	*Data->iMinSAD = MV_MAX_ERROR;
1338 :			Data->iFcode = iFcode;
1339 :	syskin	1.44.2.22	Data->qpel_precision = 0;
1340 :	chl	1.44.2.1
1341 :	h	1.44.2.3	Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1342 :			Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1343 :			Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1344 :			Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1345 :	chl	1.44.2.1
1346 :	h	1.44.2.3	Data->predMV = *predMV;
1347 :	edgomez	1.20
1348 :	h	1.44.2.3	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1349 :	syskin	1.44.2.34	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1350 :	Isibaar	1.1
1351 :	h	1.44.2.3	pmv[0] = Data->predMV;
1352 :	syskin	1.44.2.19	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1353 :	h	1.44.2.3	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1354 :	Isibaar	1.1
1355 :	h	1.44.2.3	Data->currentMV->x = Data->currentMV->y = 0;
1356 :	chl	1.44.2.1	CheckCandidate = CheckCandidate16no4v;
1357 :	edgomez	1.20
1358 :	chl	1.44.2.1	// main loop. checking all predictions
1359 :			for (i = 0; i < 8; i++) {
1360 :			if (!(mask = make_mask(pmv, i)) ) continue;
1361 :	h	1.44.2.3	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1362 :	edgomez	1.20	}
1363 :
1364 :	chl	1.44.2.1	if (MotionFlags & PMV_USESQUARES16)
1365 :			MainSearchPtr = SquareSearch;
1366 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1367 :			MainSearchPtr = AdvDiamondSearch;
1368 :			else MainSearchPtr = DiamondSearch;
1369 :	Isibaar	1.1
1370 :	h	1.44.2.3	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1371 :	Isibaar	1.1
1372 :	syskin	1.44.2.22	SubpelRefine(Data);
1373 :	syskin	1.44.2.19
1374 :			if (Data->qpel) {
1375 :			Data->currentQMV->x = 2*Data->currentMV->x;
1376 :			Data->currentQMV->y = 2*Data->currentMV->y;
1377 :	syskin	1.44.2.22	Data->qpel_precision = 1;
1378 :	syskin	1.44.2.34	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1379 :			pParam->width, pParam->height, iFcode, 1, 0);
1380 :	syskin	1.44.2.22	SubpelRefine(Data);
1381 :	syskin	1.44.2.19	}
1382 :	Isibaar	1.1
1383 :	chl	1.44.2.1	// three bits are needed to code backward mode. four for forward
1384 :			// we treat the bits just like they were vector's
1385 :	syskin	1.44.2.14	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1386 :			else Data->iMinSAD += 3 Data->lambda16;
1387 :	chl	1.44.2.1
1388 :	h	1.44.2.3	if (Data->iMinSAD < best_sad) {
1389 :			best_sad = Data->iMinSAD;
1390 :	chl	1.44.2.1	pMB->mode = mode_current;
1391 :	syskin	1.44.2.19	if (Data->qpel) {
1392 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1393 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1394 :			if (mode_current == MODE_FORWARD)
1395 :			pMB->qmvs[0] = *Data->currentQMV;
1396 :			else
1397 :			pMB->b_qmvs[0] = *Data->currentQMV;
1398 :			} else {
1399 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1400 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1401 :			}
1402 :			if (mode_current == MODE_FORWARD)
1403 :			pMB->mvs[0] = (Data->currentMV+2) = Data->currentMV;
1404 :			else
1405 :			pMB->b_mvs[0] = (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1406 :
1407 :	Isibaar	1.1	}
1408 :	chl	1.44.2.1
1409 :			}
1410 :
1411 :	syskin	1.44.2.34	static void
1412 :			SkipDecisionB(const IMAGE * const pCur,
1413 :			const IMAGE * const f_Ref,
1414 :			const IMAGE * const b_Ref,
1415 :			MACROBLOCK * const pMB,
1416 :			const uint32_t quant,
1417 :			const uint32_t x, const uint32_t y,
1418 :			const SearchData * const Data)
1419 :			{
1420 :			int dx, dy, b_dx, b_dy;
1421 :			uint32_t sum;
1422 :			//this is not full chroma compensation, only it's fullpel approximation. should work though
1423 :			if (Data->qpel) {
1424 :			dy = Data->directmvF[0].y/2 + Data->directmvF[1].y/2 +
1425 :			Data->directmvF[2].y/2 + Data->directmvF[3].y/2;
1426 :
1427 :			dx = Data->directmvF[0].x/2 + Data->directmvF[1].x/2 +
1428 :			Data->directmvF[2].x/2 + Data->directmvF[3].x/2;
1429 :
1430 :			b_dy = Data->directmvB[0].y/2 + Data->directmvB[1].y/2 +
1431 :			Data->directmvB[2].y/2 + Data->directmvB[3].y/2;
1432 :
1433 :			b_dx = Data->directmvB[0].x/2 + Data->directmvB[1].x/2 +
1434 :			Data->directmvB[2].x/2 + Data->directmvB[3].x/2;
1435 :
1436 :			} else {
1437 :			dy = Data->directmvF[0].y + Data->directmvF[1].y +
1438 :			Data->directmvF[2].y + Data->directmvF[3].y;
1439 :
1440 :			dx = Data->directmvF[0].x + Data->directmvF[1].x +
1441 :			Data->directmvF[2].x + Data->directmvF[3].x;
1442 :
1443 :			b_dy = Data->directmvB[0].y + Data->directmvB[1].y +
1444 :			Data->directmvB[2].y + Data->directmvB[3].y;
1445 :
1446 :			b_dx = Data->directmvB[0].x + Data->directmvB[1].x +
1447 :			Data->directmvB[2].x + Data->directmvB[3].x;
1448 :			}
1449 :
1450 :
1451 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1452 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1453 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1454 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1455 :
1456 :			sum = sad8bi(pCur->u + 8x + 8y*(Data->iEdgedWidth/2),
1457 :			f_Ref->u + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1458 :			b_Ref->u + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1459 :			Data->iEdgedWidth/2);
1460 :			sum += sad8bi(pCur->v + 8x + 8y*(Data->iEdgedWidth/2),
1461 :			f_Ref->v + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1462 :			b_Ref->v + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1463 :			Data->iEdgedWidth/2);
1464 :
1465 :			if (sum < 2MAX_CHROMA_SAD_FOR_SKIP quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1466 :			}
1467 :
1468 :
1469 :
1470 :			static __inline uint32_t
1471 :	Isibaar	1.44.2.2	SearchDirect(const IMAGE * const f_Ref,
1472 :	chl	1.44.2.1	const uint8_t * const f_RefH,
1473 :			const uint8_t * const f_RefV,
1474 :			const uint8_t * const f_RefHV,
1475 :	Isibaar	1.44.2.2	const IMAGE * const b_Ref,
1476 :	chl	1.44.2.1	const uint8_t * const b_RefH,
1477 :			const uint8_t * const b_RefV,
1478 :			const uint8_t * const b_RefHV,
1479 :			const IMAGE * const pCur,
1480 :			const int x, const int y,
1481 :			const uint32_t MotionFlags,
1482 :			const int32_t TRB, const int32_t TRD,
1483 :			const MBParam * const pParam,
1484 :			MACROBLOCK * const pMB,
1485 :			const MACROBLOCK * const b_mb,
1486 :	h	1.44.2.3	int32_t * const best_sad,
1487 :			SearchData * const Data)
1488 :	chl	1.44.2.1
1489 :			{
1490 :	h	1.44.2.3	int32_t skip_sad;
1491 :	chl	1.44.2.1	int k;
1492 :	h	1.44.2.3
1493 :	chl	1.44.2.1	MainSearchFunc *MainSearchPtr;
1494 :
1495 :	h	1.44.2.3	Data->iMinSAD = 2564096;
1496 :	chl	1.44.2.1
1497 :	h	1.44.2.3	Data->Ref = f_Ref->y + (x + Data->iEdgedWidthy) 16;
1498 :			Data->RefH = f_RefH + (x + Data->iEdgedWidthy) 16;
1499 :			Data->RefV = f_RefV + (x + Data->iEdgedWidthy) 16;
1500 :			Data->RefHV = f_RefHV + (x + Data->iEdgedWidthy) 16;
1501 :			Data->bRef = b_Ref->y + (x + Data->iEdgedWidthy) 16;
1502 :			Data->bRefH = b_RefH + (x + Data->iEdgedWidthy) 16;
1503 :			Data->bRefV = b_RefV + (x + Data->iEdgedWidthy) 16;
1504 :			Data->bRefHV = b_RefHV + (x + Data->iEdgedWidthy) 16;
1505 :
1506 :			Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1507 :			Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1508 :			Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1509 :			Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1510 :	syskin	1.44.2.19	if (Data->qpel) { //we measure in qpixels
1511 :			Data->max_dx *= 2;
1512 :			Data->max_dy *= 2;
1513 :			Data->min_dx *= 2;
1514 :			Data->min_dy *= 2;
1515 :			Data->referencemv = b_mb->qmvs;
1516 :			} else Data->referencemv = b_mb->mvs;
1517 :	syskin	1.44.2.26	Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here
1518 :	chl	1.44.2.1
1519 :			for (k = 0; k < 4; k++) {
1520 :	h	1.44.2.3	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1521 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1522 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1523 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1524 :
1525 :			if ( ( pMB->b_mvs[k].x > Data->max_dx ) \|\| ( pMB->b_mvs[k].x < Data->min_dx )
1526 :			\|\| ( pMB->b_mvs[k].y > Data->max_dy ) \|\| ( pMB->b_mvs[k].y < Data->min_dy )) {
1527 :	chl	1.44.2.1
1528 :	Isibaar	1.44.2.2	best_sad = 2564096; // in that case, we won't use direct mode
1529 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1530 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1531 :	syskin	1.44.2.34	return 256*4096;
1532 :	Isibaar	1.44.2.2	}
1533 :			if (b_mb->mode != MODE_INTER4V) {
1534 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1535 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1536 :	h	1.44.2.3	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1537 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1538 :	Isibaar	1.44.2.2	break;
1539 :			}
1540 :			}
1541 :	chl	1.44.2.1
1542 :	syskin	1.44.2.22
1543 :			if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect;
1544 :			else CheckCandidate = CheckCandidateDirectno4v;
1545 :	syskin	1.44.2.19
1546 :	h	1.44.2.3	(*CheckCandidate)(0, 0, 255, &k, Data);
1547 :	Isibaar	1.44.2.2
1548 :	syskin	1.44.2.34	// initial (fast) skip decision
1549 :			if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH*2) {
1550 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data->chroma, Data); //possible skip - checking chroma
1551 :			if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1552 :	Isibaar	1.44.2.2	}
1553 :
1554 :	h	1.44.2.3	skip_sad = *Data->iMinSAD;
1555 :	Isibaar	1.44.2.2
1556 :	chl	1.44.2.1	// DIRECT MODE DELTA VECTOR SEARCH.
1557 :			// This has to be made more effective, but at the moment I'm happy it's running at all
1558 :
1559 :			if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1560 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1561 :			else MainSearchPtr = DiamondSearch;
1562 :
1563 :	h	1.44.2.3	(*MainSearchPtr)(0, 0, Data, 255);
1564 :	chl	1.44.2.1
1565 :	syskin	1.44.2.22	SubpelRefine(Data);
1566 :	chl	1.44.2.1
1567 :	h	1.44.2.3	best_sad = Data->iMinSAD;
1568 :	chl	1.44.2.1
1569 :	syskin	1.44.2.26	if (b_mb->mode == MODE_INTER4V)
1570 :	chl	1.44.2.1	pMB->mode = MODE_DIRECT;
1571 :	syskin	1.44.2.26	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1572 :	chl	1.44.2.1
1573 :	h	1.44.2.3	pMB->pmvs[3] = *Data->currentMV;
1574 :	chl	1.44.2.1
1575 :			for (k = 0; k < 4; k++) {
1576 :	h	1.44.2.3	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1577 :	syskin	1.44.2.19	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1578 :	h	1.44.2.3	? Data->directmvB[k].x
1579 :	syskin	1.44.2.19	:pMB->mvs[k].x - Data->referencemv[k].x);
1580 :	h	1.44.2.3	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1581 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1582 :			? Data->directmvB[k].y
1583 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1584 :	syskin	1.44.2.19	if (Data->qpel) {
1585 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1586 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1587 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1588 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1589 :			}
1590 :
1591 :	chl	1.44.2.1	if (b_mb->mode != MODE_INTER4V) {
1592 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1593 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1594 :	syskin	1.44.2.19	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1595 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1596 :	chl	1.44.2.1	break;
1597 :			}
1598 :			}
1599 :	h	1.44.2.3	return skip_sad;
1600 :	chl	1.44.2.1	}
1601 :
1602 :	h	1.44.2.3
1603 :	syskin	1.44.2.34	static void
1604 :	chl	1.44.2.1	SearchInterpolate(const uint8_t * const f_Ref,
1605 :			const uint8_t * const f_RefH,
1606 :			const uint8_t * const f_RefV,
1607 :			const uint8_t * const f_RefHV,
1608 :			const uint8_t * const b_Ref,
1609 :			const uint8_t * const b_RefH,
1610 :			const uint8_t * const b_RefV,
1611 :			const uint8_t * const b_RefHV,
1612 :			const IMAGE * const pCur,
1613 :			const int x, const int y,
1614 :			const uint32_t fcode,
1615 :			const uint32_t bcode,
1616 :			const uint32_t MotionFlags,
1617 :			const MBParam * const pParam,
1618 :			const VECTOR * const f_predMV,
1619 :			const VECTOR * const b_predMV,
1620 :			MACROBLOCK * const pMB,
1621 :	h	1.44.2.3	int32_t * const best_sad,
1622 :			SearchData * const fData)
1623 :	edgomez	1.20
1624 :	chl	1.3	{
1625 :
1626 :	chl	1.44.2.1	int iDirection, i, j;
1627 :	h	1.44.2.3	SearchData bData;
1628 :	edgomez	1.20
1629 :	syskin	1.44.2.34	fData->qpel_precision = 0;
1630 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1631 :			fData->iMinSAD = 4096256;
1632 :			bData.currentMV ++; bData.currentQMV ++;
1633 :	h	1.44.2.3	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1634 :
1635 :	syskin	1.44.2.34	i = (x + y * fData->iEdgedWidth) * 16;
1636 :			bData.bRef = fData->Ref = f_Ref + i;
1637 :			bData.bRefH = fData->RefH = f_RefH + i;
1638 :			bData.bRefV = fData->RefV = f_RefV + i;
1639 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1640 :			bData.Ref = fData->bRef = b_Ref + i;
1641 :			bData.RefH = fData->bRefH = b_RefH + i;
1642 :			bData.RefV = fData->bRefV = b_RefV + i;
1643 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1644 :	h	1.44.2.3
1645 :			bData.bpredMV = fData->predMV = *f_predMV;
1646 :			fData->bpredMV = bData.predMV = *b_predMV;
1647 :	syskin	1.44.2.20	fData->currentMV[0] = fData->currentMV[2];
1648 :	syskin	1.44.2.34
1649 :			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);
1650 :			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);
1651 :	chl	1.3
1652 :	h	1.44.2.3	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1653 :	syskin	1.44.2.27	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1654 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1655 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1656 :	h	1.44.2.3
1657 :			if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1658 :	syskin	1.44.2.27	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1659 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1660 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1661 :	Isibaar	1.44.2.2
1662 :	h	1.44.2.3	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1663 :	chl	1.3
1664 :	chl	1.44.2.1	//diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1665 :	chl	1.3
1666 :	chl	1.44.2.1	do {
1667 :			iDirection = 255;
1668 :			// forward MV moves
1669 :	h	1.44.2.3	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1670 :	chl	1.35
1671 :	h	1.44.2.3	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1672 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1673 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1674 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1675 :	edgomez	1.20
1676 :	chl	1.44.2.1	// backward MV moves
1677 :	h	1.44.2.3	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1678 :			fData->currentMV[2] = fData->currentMV[0];
1679 :	Isibaar	1.44.2.2	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1680 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1681 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1682 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1683 :	chl	1.3
1684 :	chl	1.44.2.1	} while (!(iDirection));
1685 :	chl	1.3
1686 :	syskin	1.44.2.19	if (fData->qpel) {
1687 :	syskin	1.44.2.27	CheckCandidate = CheckCandidateInt;
1688 :	syskin	1.44.2.22	fData->qpel_precision = bData.qpel_precision = 1;
1689 :	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);
1690 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1691 :	syskin	1.44.2.19	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1692 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1693 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1694 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1695 :	syskin	1.44.2.22	SubpelRefine(fData);
1696 :	syskin	1.44.2.19	fData->currentQMV[2] = fData->currentQMV[0];
1697 :	syskin	1.44.2.22	SubpelRefine(&bData);
1698 :	syskin	1.44.2.19	}
1699 :	syskin	1.44.2.28
1700 :	syskin	1.44.2.34	fData->iMinSAD += (2+2) fData->lambda16; // two bits are needed to code interpolate mode.
1701 :	syskin	1.44.2.19
1702 :	h	1.44.2.3	if (fData->iMinSAD < best_sad) {
1703 :			best_sad = fData->iMinSAD;
1704 :			pMB->mvs[0] = fData->currentMV[0];
1705 :			pMB->b_mvs[0] = fData->currentMV[1];
1706 :	chl	1.44.2.1	pMB->mode = MODE_INTERPOLATE;
1707 :	syskin	1.44.2.19	if (fData->qpel) {
1708 :			pMB->qmvs[0] = fData->currentQMV[0];
1709 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1710 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1711 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1712 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1713 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1714 :			} else {
1715 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1716 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1717 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1718 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1719 :			}
1720 :	edgomez	1.20	}
1721 :	chl	1.3	}
1722 :
1723 :	chl	1.44.2.1	void
1724 :			MotionEstimationBVOP(MBParam * const pParam,
1725 :			FRAMEINFO * const frame,
1726 :			const int32_t time_bp,
1727 :			const int32_t time_pp,
1728 :			// forward (past) reference
1729 :			const MACROBLOCK * const f_mbs,
1730 :			const IMAGE * const f_ref,
1731 :			const IMAGE * const f_refH,
1732 :			const IMAGE * const f_refV,
1733 :			const IMAGE * const f_refHV,
1734 :			// backward (future) reference
1735 :	syskin	1.44.2.18	const FRAMEINFO * const b_reference,
1736 :	chl	1.44.2.1	const IMAGE * const b_ref,
1737 :			const IMAGE * const b_refH,
1738 :			const IMAGE * const b_refV,
1739 :			const IMAGE * const b_refHV)
1740 :			{
1741 :			uint32_t i, j;
1742 :	syskin	1.44.2.34	int32_t best_sad;
1743 :			uint32_t skip_sad;
1744 :	chl	1.44.2.1	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1745 :			static const VECTOR zeroMV={0,0};
1746 :	syskin	1.44.2.18	const MACROBLOCK * const b_mbs = b_reference->mbs;
1747 :	chl	1.3
1748 :	chl	1.44.2.1	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1749 :	chl	1.3
1750 :	chl	1.44.2.1	const int32_t TRB = time_pp - time_bp;
1751 :			const int32_t TRD = time_pp;
1752 :	syskin	1.44.2.19	uint8_t * qimage;
1753 :	chl	1.3
1754 :	h	1.44.2.3	// some pre-inintialized data for the rest of the search
1755 :
1756 :			SearchData Data;
1757 :			int32_t iMinSAD;
1758 :			VECTOR currentMV[3];
1759 :	syskin	1.44.2.19	VECTOR currentQMV[3];
1760 :	syskin	1.44.2.32	memset(&Data, 0, sizeof(SearchData));
1761 :	h	1.44.2.3	Data.iEdgedWidth = pParam->edged_width;
1762 :	syskin	1.44.2.19	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1763 :	h	1.44.2.3	Data.iMinSAD = &iMinSAD;
1764 :	syskin	1.44.2.34	Data.lambda16 = lambda_vec16[frame->quant];
1765 :			Data.chroma = frame->quant;
1766 :	syskin	1.44.2.19	Data.qpel = pParam->m_quarterpel;
1767 :	syskin	1.44.2.22	Data.rounding = 0;
1768 :	h	1.44.2.3
1769 :	syskin	1.44.2.19	if((qimage = (uint8_t ) malloc(32 pParam->edged_width)) == NULL)
1770 :			return; // allocate some mem for qpel interpolated blocks
1771 :			// somehow this is dirty since I think we shouldn't use malloc outside
1772 :			// encoder_create() - so please fix me!
1773 :			Data.RefQ = qimage;
1774 :	edgomez	1.20
1775 :	syskin	1.44.2.19	// note: i==horizontal, j==vertical
1776 :	chl	1.44.2.1	for (j = 0; j < pParam->mb_height; j++) {
1777 :	chl	1.3
1778 :	chl	1.44.2.1	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1779 :	edgomez	1.20
1780 :	chl	1.44.2.1	for (i = 0; i < pParam->mb_width; i++) {
1781 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1782 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1783 :	edgomez	1.20
1784 :	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 */
1785 :	syskin	1.44.2.18	if (b_reference->coding_type != S_VOP)
1786 :			if (b_mb->mode == MODE_NOT_CODED) {
1787 :			pMB->mode = MODE_NOT_CODED;
1788 :			continue;
1789 :			}
1790 :	edgomez	1.20
1791 :	h	1.44.2.3	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1792 :	syskin	1.44.2.11	pMB->quant = frame->quant;
1793 :	syskin	1.44.2.14
1794 :	chl	1.44.2.1	/* direct search comes first, because it (1) checks for SKIP-mode
1795 :			and (2) sets very good predictions for forward and backward search */
1796 :	Isibaar	1.44.2.2	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1797 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1798 :	chl	1.44.2.1	&frame->image,
1799 :			i, j,
1800 :			frame->motion_flags,
1801 :			TRB, TRD,
1802 :			pParam,
1803 :			pMB, b_mb,
1804 :	h	1.44.2.3	&best_sad,
1805 :			&Data);
1806 :	edgomez	1.20
1807 :	Isibaar	1.44.2.2	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1808 :	syskin	1.44.2.19
1809 :	chl	1.44.2.1	// forward search
1810 :			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1811 :			&frame->image, i, j,
1812 :			frame->motion_flags,
1813 :	h	1.44.2.3	frame->fcode, pParam,
1814 :	chl	1.44.2.1	pMB, &f_predMV, &best_sad,
1815 :	h	1.44.2.3	MODE_FORWARD, &Data);
1816 :	chl	1.10
1817 :	chl	1.44.2.1	// backward search
1818 :			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1819 :			&frame->image, i, j,
1820 :			frame->motion_flags,
1821 :	h	1.44.2.3	frame->bcode, pParam,
1822 :	chl	1.44.2.1	pMB, &b_predMV, &best_sad,
1823 :	h	1.44.2.3	MODE_BACKWARD, &Data);
1824 :	chl	1.10
1825 :	chl	1.44.2.1	// interpolate search comes last, because it uses data from forward and backward as prediction
1826 :	syskin	1.44.2.21
1827 :	chl	1.44.2.1	SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1828 :			b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1829 :			&frame->image,
1830 :			i, j,
1831 :			frame->fcode, frame->bcode,
1832 :			frame->motion_flags,
1833 :	h	1.44.2.3	pParam,
1834 :	chl	1.44.2.1	&f_predMV, &b_predMV,
1835 :	h	1.44.2.3	pMB, &best_sad,
1836 :			&Data);
1837 :	syskin	1.44.2.21
1838 :	syskin	1.44.2.34	// final skip decision
1839 :			if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2)
1840 :			&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1841 :			SkipDecisionB(&frame->image, f_ref, b_ref, pMB,frame->quant, i, j, &Data);
1842 :
1843 :	chl	1.44.2.1	switch (pMB->mode) {
1844 :			case MODE_FORWARD:
1845 :			f_count++;
1846 :	syskin	1.44.2.34	if (Data.qpel) f_predMV = pMB->qmvs[0];
1847 :	syskin	1.44.2.19	else f_predMV = pMB->mvs[0];
1848 :	chl	1.44.2.1	break;
1849 :			case MODE_BACKWARD:
1850 :			b_count++;
1851 :	syskin	1.44.2.34	if (Data.qpel) b_predMV = pMB->b_qmvs[0];
1852 :	syskin	1.44.2.19	else b_predMV = pMB->b_mvs[0];
1853 :	chl	1.44.2.1	break;
1854 :			case MODE_INTERPOLATE:
1855 :			i_count++;
1856 :	syskin	1.44.2.34	if (Data.qpel) {
1857 :	syskin	1.44.2.19	f_predMV = pMB->qmvs[0];
1858 :			b_predMV = pMB->b_qmvs[0];
1859 :			} else {
1860 :			f_predMV = pMB->mvs[0];
1861 :			b_predMV = pMB->b_mvs[0];
1862 :			}
1863 :	chl	1.44.2.1	break;
1864 :			case MODE_DIRECT:
1865 :			case MODE_DIRECT_NO4V:
1866 :			d_count++;
1867 :			default:
1868 :			break;
1869 :	chl	1.10	}
1870 :			}
1871 :			}
1872 :	syskin	1.44.2.19	free(qimage);
1873 :	chl	1.44.2.1	}
1874 :	chl	1.3
1875 :	Isibaar	1.44.2.2	static __inline int
1876 :			MEanalyzeMB ( const uint8_t * const pRef,
1877 :			const uint8_t * const pCur,
1878 :			const int x,
1879 :			const int y,
1880 :			const MBParam * const pParam,
1881 :			const MACROBLOCK * const pMBs,
1882 :	h	1.44.2.3	MACROBLOCK * const pMB,
1883 :			SearchData * const Data)
1884 :	Isibaar	1.44.2.2	{
1885 :	chl	1.32
1886 :	Isibaar	1.44.2.12	int i = 255, mask;
1887 :	h	1.44.2.3	VECTOR pmv[3];
1888 :			*(Data->iMinSAD) = MV_MAX_ERROR;
1889 :	syskin	1.44.2.17
1890 :			//median is only used as prediction. it doesn't have to be real
1891 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1892 :			else
1893 :			if (x == 1) //left macroblock does not have any vector now
1894 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1895 :			else if (y == 1) // top macroblock don't have it's vector
1896 :			Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1897 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1898 :
1899 :	h	1.44.2.3	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1900 :	syskin	1.44.2.34	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, 0);
1901 :	chl	1.35
1902 :	h	1.44.2.3	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1903 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1904 :	Isibaar	1.44.2.2
1905 :			pmv[1].x = EVEN(pMB->mvs[0].x);
1906 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1907 :	syskin	1.44.2.17	pmv[2].x = EVEN(Data->predMV.x);
1908 :			pmv[2].y = EVEN(Data->predMV.y);
1909 :			pmv[0].x = pmv[0].y = 0;
1910 :
1911 :	syskin	1.44.2.34	CheckCandidate16no4vI(0, 0, 255, &i, Data);
1912 :	syskin	1.44.2.17
1913 :			//early skip for 0,0
1914 :			if (Data->iMinSAD < MAX_SAD00_FOR_SKIP 4) {
1915 :			pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1916 :			pMB->mode = MODE_NOT_CODED;
1917 :			return 0;
1918 :			}
1919 :	Isibaar	1.44.2.2
1920 :			if (!(mask = make_mask(pmv, 1)))
1921 :	syskin	1.44.2.34	CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
1922 :	Isibaar	1.44.2.2	if (!(mask = make_mask(pmv, 2)))
1923 :	syskin	1.44.2.34	CheckCandidate16no4vI(pmv[2].x, pmv[2].y, mask, &i, Data);
1924 :	Isibaar	1.44.2.2
1925 :	syskin	1.44.2.34	if (Data->iMinSAD > MAX_SAD00_FOR_SKIP 6) // diamond only if needed
1926 :	syskin	1.44.2.17	DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1927 :	Isibaar	1.44.2.2
1928 :	syskin	1.44.2.17	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1929 :	syskin	1.44.2.14	pMB->mode = MODE_INTER;
1930 :	h	1.44.2.3	return *(Data->iMinSAD);
1931 :	Isibaar	1.44.2.2	}
1932 :
1933 :			#define INTRA_THRESH 1350
1934 :	syskin	1.44.2.26	#define INTER_THRESH 1200
1935 :	Isibaar	1.44.2.2
1936 :	syskin	1.44.2.18
1937 :	Isibaar	1.44.2.2	int
1938 :			MEanalysis( const IMAGE * const pRef,
1939 :	syskin	1.44.2.18	FRAMEINFO * const Current,
1940 :	Isibaar	1.44.2.2	MBParam * const pParam,
1941 :	syskin	1.44.2.18	int maxIntra, //maximum number if non-I frames
1942 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1943 :			int bCount) // number if B frames in a row
1944 :	Isibaar	1.44.2.2	{
1945 :			uint32_t x, y, intra = 0;
1946 :			int sSAD = 0;
1947 :	syskin	1.44.2.18	MACROBLOCK * const pMBs = Current->mbs;
1948 :			const IMAGE * const pCurrent = &Current->image;
1949 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1950 :	h	1.44.2.3
1951 :			VECTOR currentMV;
1952 :			int32_t iMinSAD;
1953 :			SearchData Data;
1954 :			Data.iEdgedWidth = pParam->edged_width;
1955 :			Data.currentMV = &currentMV;
1956 :			Data.iMinSAD = &iMinSAD;
1957 :	syskin	1.44.2.18	Data.iFcode = Current->fcode;
1958 :	syskin	1.44.2.17	CheckCandidate = CheckCandidate16no4vI;
1959 :	h	1.44.2.3
1960 :	syskin	1.44.2.19	if (intraCount < 10) // we're right after an I frame
1961 :			IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
1962 :	syskin	1.44.2.18	else
1963 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
1964 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
1965 :
1966 :
1967 :	syskin	1.44.2.26	InterThresh += 400 * (1 - bCount);
1968 :	syskin	1.44.2.18	if (InterThresh < 200) InterThresh = 200;
1969 :
1970 :	Isibaar	1.44.2.2	if (sadInit) (*sadInit) ();
1971 :
1972 :	syskin	1.44.2.32	for (y = 1; y < pParam->mb_height-1; y++) {
1973 :			for (x = 1; x < pParam->mb_width-1; x++) {
1974 :	Isibaar	1.44.2.2	int sad, dev;
1975 :			MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
1976 :
1977 :			sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
1978 :	h	1.44.2.3	pParam, pMBs, pMB, &Data);
1979 :	Isibaar	1.44.2.2
1980 :	syskin	1.44.2.18	if (sad > IntraThresh) {
1981 :	syskin	1.44.2.11	dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
1982 :			pParam->edged_width);
1983 :	syskin	1.44.2.18	if (dev + IntraThresh < sad) {
1984 :			pMB->mode = MODE_INTRA;
1985 :	suxen_drol	1.44.2.33	if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
1986 :	Isibaar	1.44.2.2	}
1987 :	syskin	1.44.2.18	}
1988 :	syskin	1.44.2.11	sSAD += sad;
1989 :	suxen_drol	1.6	}
1990 :			}
1991 :	syskin	1.44.2.32	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
1992 :	suxen_drol	1.44.2.33	if (sSAD > InterThresh ) return P_VOP;
1993 :	Isibaar	1.44.2.2	emms();
1994 :	suxen_drol	1.44.2.33	return B_VOP;
1995 :	syskin	1.44.2.34
1996 :	chl	1.44.2.13	}
1997 :
1998 :	syskin	1.44.2.14	static void
1999 :			CheckGMC(int x, int y, const int dir, int * iDirection,
2000 :			const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
2001 :			const MBParam * const pParam)
2002 :			{
2003 :			uint32_t mx, my, a, count = 0;
2004 :
2005 :			for (my = 1; my < pParam->mb_height-1; my++)
2006 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2007 :			VECTOR mv;
2008 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2009 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED) continue;
2010 :			mv = pMB->mvs[0];
2011 :			a = ABS(mv.x - x) + ABS(mv.y - y);
2012 :			if (a < 6) count += 6 - a;
2013 :			}
2014 :
2015 :			if (count > *bestcount) {
2016 :			*bestcount = count;
2017 :			*iDirection = dir;
2018 :			GMC->x = x; GMC->y = y;
2019 :			}
2020 :			}
2021 :	chl	1.44.2.13
2022 :
2023 :	suxen_drol	1.44.2.35	static VECTOR
2024 :	chl	1.44.2.13	GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2025 :			{
2026 :
2027 :	syskin	1.44.2.14	uint32_t count, bestcount = 0;
2028 :	chl	1.44.2.13	int x, y;
2029 :	syskin	1.44.2.14	VECTOR gmc = {0,0};
2030 :	chl	1.44.2.13	int step, min_x, max_x, min_y, max_y;
2031 :			uint32_t mx, my;
2032 :	syskin	1.44.2.14	int iDirection, bDirection;
2033 :	chl	1.44.2.13
2034 :			min_x = min_y = -32<<iFcode;
2035 :			max_x = max_y = 32<<iFcode;
2036 :
2037 :	syskin	1.44.2.14	//step1: let's find a rough camera panning
2038 :	chl	1.44.2.13	for (step = 32; step >= 2; step /= 2) {
2039 :			bestcount = 0;
2040 :			for (y = min_y; y <= max_y; y += step)
2041 :			for (x = min_x ; x <= max_x; x += step) {
2042 :			count = 0;
2043 :			//for all macroblocks
2044 :			for (my = 1; my < pParam->mb_height-1; my++)
2045 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2046 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2047 :			VECTOR mv;
2048 :
2049 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2050 :			continue;
2051 :
2052 :			mv = pMB->mvs[0];
2053 :			if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2054 :			count++;
2055 :			}
2056 :			if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2057 :			}
2058 :			min_x = gmc.x - step;
2059 :			max_x = gmc.x + step;
2060 :			min_y = gmc.y - step;
2061 :			max_y = gmc.y + step;
2062 :
2063 :			}
2064 :	syskin	1.44.2.14
2065 :			if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2066 :			gmc.x = gmc.y = 0; //no camara pan, no GMC
2067 :
2068 :			// step2: let's refine camera panning using gradiend-descent approach.
2069 :			// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
2070 :			bestcount = 0;
2071 :			CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
2072 :			do {
2073 :			x = gmc.x; y = gmc.y;
2074 :			bDirection = iDirection; iDirection = 0;
2075 :			if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2076 :			if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2077 :			if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
2078 :			if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2079 :
2080 :			} while (iDirection);
2081 :
2082 :	chl	1.44.2.13	if (pParam->m_quarterpel) {
2083 :			gmc.x *= 2;
2084 :			gmc.y = 2; / we store the halfpel value as pseudo-qpel to make comparison easier */
2085 :			}
2086 :
2087 :			return gmc;
2088 :	Isibaar	1.44.2.2	}

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