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

Revision 754 - (view) (download)

1 :	Isibaar	3	/**************************************************************************
2 :			*
3 :	chl	259	* XVID MPEG-4 VIDEO CODEC
4 :	syskin	748	* motion estimation
5 :	chl	259	*
6 :			* This program is an implementation of a part of one or more MPEG-4
7 :			* Video tools as specified in ISO/IEC 14496-2 standard. Those intending
8 :			* to use this software module in hardware or software products are
9 :			* advised that its use may infringe existing patents or copyrights, and
10 :			* any such use would be at such party's own risk. The original
11 :			* developer of this software module and his/her company, and subsequent
12 :			* editors and their companies, will have no liability for use of this
13 :			* software or modifications or derivatives thereof.
14 :			*
15 :			* This program is free software; you can redistribute it and/or modify
16 :			* it under the terms of the GNU General Public License as published by
17 :			* the Free Software Foundation; either version 2 of the License, or
18 :			* (at your option) any later version.
19 :			*
20 :			* This program is distributed in the hope that it will be useful,
21 :			* but WITHOUT ANY WARRANTY; without even the implied warranty of
22 :			* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 :			* GNU General Public License for more details.
24 :			*
25 :			* You should have received a copy of the GNU General Public License
26 :			* along with this program; if not, write to the Free Software
27 :			* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 :			*
29 :			*************************************************************************/
30 :
31 :	Isibaar	3	#include <assert.h>
32 :			#include <stdio.h>
33 :	chl	96	#include <stdlib.h>
34 :	suxen_drol	721	#include <string.h> // memcpy
35 :	Isibaar	3
36 :			#include "../encoder.h"
37 :			#include "../utils/mbfunctions.h"
38 :			#include "../prediction/mbprediction.h"
39 :			#include "../global.h"
40 :			#include "../utils/timer.h"
41 :	Isibaar	580	#include "../image/interpolate8x8.h"
42 :	chl	530	#include "motion_est.h"
43 :	suxen_drol	118	#include "motion.h"
44 :	Isibaar	3	#include "sad.h"
45 :	Isibaar	539	#include "../utils/emms.h"
46 :	Isibaar	3
47 :	chl	530	#define INITIAL_SKIP_THRESH (10)
48 :			#define FINAL_SKIP_THRESH (50)
49 :			#define MAX_SAD00_FOR_SKIP (20)
50 :	Isibaar	539	#define MAX_CHROMA_SAD_FOR_SKIP (22)
51 :	Isibaar	3
52 :	chl	530	#define CHECK_CANDIDATE(X,Y,D) { \
53 :			(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
54 :	Isibaar	3
55 :	chl	530	static __inline int
56 :	syskin	724	d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
57 :	Isibaar	3	{
58 :	chl	530	int xb, yb;
59 :	syskin	712	if (qpel) { x = 2; y = 2;}
60 :			else if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
61 :	syskin	748	x -= pred.x;
62 :			y -= pred.y;
63 :	chl	530
64 :	syskin	748	if (x) {
65 :	chl	530	if (x < 0) x = -x;
66 :			x += (1 << (iFcode - 1)) - 1;
67 :			x >>= (iFcode - 1);
68 :			if (x > 32) x = 32;
69 :			xb = mvtab[x] + iFcode;
70 :	syskin	748	} else xb = 1;
71 :	Isibaar	3
72 :	syskin	748	if (y) {
73 :	chl	530	if (y < 0) y = -y;
74 :			y += (1 << (iFcode - 1)) - 1;
75 :			y >>= (iFcode - 1);
76 :			if (y > 32) y = 32;
77 :			yb = mvtab[y] + iFcode;
78 :	syskin	748	} else yb = 1;
79 :	chl	530	return xb + yb;
80 :	Isibaar	3	}
81 :
82 :	syskin	748	static int32_t
83 :	syskin	628	ChromaSAD(int dx, int dy, const SearchData * const data)
84 :	Isibaar	3	{
85 :	syskin	628	int sad;
86 :	syskin	748	const uint32_t stride = data->iEdgedWidth/2;
87 :	Isibaar	3
88 :	syskin	712	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
89 :	syskin	748	data->temp[5] = dx; data->temp[6] = dy; // backup
90 :	syskin	712
91 :			switch (((dx & 1) << 1) \| (dy & 1)) {
92 :	syskin	628	case 0:
93 :	syskin	748	dx = dx / 2; dy = dy / 2;
94 :			sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride);
95 :			sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride);
96 :	syskin	628	break;
97 :			case 1:
98 :			dx = dx / 2; dy = (dy - 1) / 2;
99 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride);
100 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride);
101 :	syskin	628	break;
102 :			case 2:
103 :			dx = (dx - 1) / 2; dy = dy / 2;
104 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride);
105 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride);
106 :	syskin	628	break;
107 :			default:
108 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
109 :	syskin	748	interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding);
110 :			sad = sad8(data->CurU, data->RefQ, stride);
111 :
112 :			interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding);
113 :			sad += sad8(data->CurV, data->RefQ, stride);
114 :	syskin	628	break;
115 :	chl	530	}
116 :	syskin	748	data->temp[7] = sad; //backup, part 2
117 :	syskin	628	return sad;
118 :			}
119 :	Isibaar	3
120 :	syskin	663	static __inline const uint8_t *
121 :	syskin	748	GetReferenceB(const int x, const int y, const int dir, const SearchData * const data)
122 :	Isibaar	606	{
123 :	syskin	663	// dir : 0 = forward, 1 = backward
124 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
125 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
126 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
127 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
128 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
129 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
130 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
131 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
132 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
133 :	Isibaar	606	}
134 :			}
135 :
136 :	syskin	748	// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate
137 :			static __inline const uint8_t *
138 :			GetReference(const int x, const int y, const SearchData * const data)
139 :			{
140 :			switch ( ((x&1)<<1) \| (y&1) ) {
141 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
142 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
143 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
144 :			default : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
145 :			}
146 :			}
147 :
148 :			static uint8_t *
149 :	syskin	663	Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data)
150 :	chl	530	{
151 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
152 :	syskin	668	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
153 :	syskin	663	const int32_t iEdgedWidth = data->iEdgedWidth;
154 :			const uint32_t rounding = data->rounding;
155 :			const int halfpel_x = x/2;
156 :			const int halfpel_y = y/2;
157 :			const uint8_t ref1, ref2, ref3, ref4;
158 :	chl	326
159 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
160 :	syskin	663	ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
161 :			switch( ((x&1)<<1) + (y&1) ) {
162 :			case 0: // pure halfpel position
163 :	syskin	740	return (uint8_t *) ref1;
164 :	syskin	663	break;
165 :	Isibaar	3
166 :	syskin	663	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
167 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
168 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
169 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
170 :	syskin	663	break;
171 :	Isibaar	3
172 :	syskin	663	case 2: // x qpel, y halfpel - left or right during qpel refinement
173 :	syskin	748	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
174 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
175 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
176 :	syskin	663	break;
177 :	Isibaar	3
178 :	syskin	663	default: // x and y in qpel resolution - the "corners" (top left/right and
179 :			// bottom left/right) during qpel refinement
180 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
181 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
182 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
183 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
184 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
185 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
186 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
187 :			break;
188 :			}
189 :			return Reference;
190 :	Isibaar	3	}
191 :
192 :	syskin	748	static uint8_t *
193 :	syskin	663	Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data)
194 :	Isibaar	580	{
195 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
196 :	syskin	668	uint8_t * Reference = (uint8_t )data->RefQ + 16dir;
197 :	syskin	663	const int32_t iEdgedWidth = data->iEdgedWidth;
198 :			const uint32_t rounding = data->rounding;
199 :			const int halfpel_x = x/2;
200 :			const int halfpel_y = y/2;
201 :	Isibaar	580	const uint8_t ref1, ref2, ref3, ref4;
202 :
203 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
204 :	syskin	663	switch( ((x&1)<<1) + (y&1) ) {
205 :			case 0: // pure halfpel position
206 :	syskin	740	return (uint8_t *) ref1;
207 :	Isibaar	580	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
208 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
209 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
210 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
211 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
212 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
213 :	Isibaar	580	break;
214 :
215 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
216 :	syskin	748	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
217 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
218 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
219 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
220 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
221 :	Isibaar	580	break;
222 :
223 :			default: // x and y in qpel resolution - the "corners" (top left/right and
224 :			// bottom left/right) during qpel refinement
225 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
226 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
227 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
228 :	Isibaar	580	interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
229 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
230 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
231 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
232 :			break;
233 :			}
234 :	syskin	663	return Reference;
235 :			}
236 :
237 :			/* CHECK_CANDIATE FUNCTIONS START */
238 :
239 :	syskin	748	static void
240 :	syskin	663	CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
241 :			{
242 :			int t, xc, yc;
243 :			const uint8_t * Reference;
244 :			VECTOR * current;
245 :
246 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
247 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
248 :
249 :			if (data->qpel_precision) { // x and y are in 1/4 precision
250 :			Reference = Interpolate16x16qpel(x, y, 0, data);
251 :			xc = x/2; yc = y/2; //for chroma sad
252 :			current = data->currentQMV;
253 :			} else {
254 :	syskin	748	Reference = GetReference(x, y, data);
255 :	syskin	663	current = data->currentMV;
256 :			xc = x; yc = y;
257 :			}
258 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
259 :	Isibaar	580
260 :	syskin	663	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
261 :	Isibaar	580
262 :	syskin	628	data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
263 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
264 :	Isibaar	580
265 :	syskin	753	if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
266 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
267 :	syskin	628
268 :	Isibaar	580	if (data->temp[0] < data->iMinSAD[0]) {
269 :			data->iMinSAD[0] = data->temp[0];
270 :	syskin	663	current[0].x = x; current[0].y = y;
271 :			*dir = Direction; }
272 :	Isibaar	580
273 :			if (data->temp[1] < data->iMinSAD[1]) {
274 :	syskin	663	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; }
275 :	Isibaar	580	if (data->temp[2] < data->iMinSAD[2]) {
276 :	syskin	663	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
277 :	Isibaar	580	if (data->temp[3] < data->iMinSAD[3]) {
278 :	syskin	663	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
279 :	Isibaar	580	if (data->temp[4] < data->iMinSAD[4]) {
280 :	syskin	663	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
281 :
282 :	Isibaar	580	}
283 :
284 :	syskin	748	static void
285 :	syskin	704	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
286 :			{
287 :			int t;
288 :			const uint8_t * Reference;
289 :
290 :			if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) \|\| //non-zero integer value
291 :			( x > data->max_dx) \|\| ( x < data->min_dx)
292 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
293 :
294 :	syskin	748	Reference = GetReference(x, y, data);
295 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
296 :	syskin	704
297 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
298 :
299 :			data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
300 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
301 :
302 :			if (data->temp[0] < data->iMinSAD[0]) {
303 :			data->iMinSAD[0] = data->temp[0];
304 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
305 :			*dir = Direction; }
306 :
307 :			if (data->temp[1] < data->iMinSAD[1]) {
308 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
309 :			if (data->temp[2] < data->iMinSAD[2]) {
310 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
311 :			if (data->temp[3] < data->iMinSAD[3]) {
312 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
313 :			if (data->temp[4] < data->iMinSAD[4]) {
314 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
315 :			}
316 :
317 :	syskin	748	static void
318 :	syskin	663	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
319 :	syskin	658	{
320 :			int32_t sad;
321 :	syskin	663	const uint8_t * Reference;
322 :			int t;
323 :			VECTOR * current;
324 :	syskin	658
325 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
326 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
327 :
328 :	syskin	712	if (data->rrv)
329 :	syskin	704	if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) ) return; //non-zero integer value
330 :
331 :	syskin	663	if (data->qpel_precision) { // x and y are in 1/4 precision
332 :			Reference = Interpolate16x16qpel(x, y, 0, data);
333 :			current = data->currentQMV;
334 :			} else {
335 :	syskin	748	Reference = GetReference(x, y, data);
336 :	syskin	663	current = data->currentMV;
337 :	syskin	658	}
338 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode,
339 :	syskin	748	data->qpel && !data->qpel_precision, data->rrv);
340 :	syskin	658
341 :			sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
342 :	syskin	663	sad += (data->lambda16 * t * sad)/1000;
343 :	syskin	658
344 :	syskin	663	if (sad < *(data->iMinSAD)) {
345 :			*(data->iMinSAD) = sad;
346 :			current->x = x; current->y = y;
347 :			*dir = Direction; }
348 :	syskin	658	}
349 :
350 :	syskin	748	static void
351 :	syskin	739	CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
352 :	Isibaar	539	{
353 :	syskin	658	// maximum speed - for P/B/I decision
354 :	Isibaar	539
355 :			if (( x > data->max_dx) \|\| ( x < data->min_dx)
356 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
357 :
358 :	syskin	739	data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
359 :			data->iEdgedWidth, data->temp+1);
360 :			if (data->temp[0] < *(data->iMinSAD)) {
361 :			*(data->iMinSAD) = data->temp[0];
362 :	Isibaar	539	data->currentMV[0].x = x; data->currentMV[0].y = y;
363 :			*dir = Direction; }
364 :	syskin	739	if (data->temp[1] < data->iMinSAD[1]) {
365 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
366 :			if (data->temp[2] < data->iMinSAD[2]) {
367 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
368 :			if (data->temp[3] < data->iMinSAD[3]) {
369 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
370 :			if (data->temp[4] < data->iMinSAD[4]) {
371 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
372 :
373 :	Isibaar	539	}
374 :
375 :	syskin	748	static void
376 :	chl	530	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
377 :			{
378 :			int32_t sad;
379 :	syskin	663	int xb, yb, t;
380 :	chl	530	const uint8_t ReferenceF, ReferenceB;
381 :	syskin	663	VECTOR *current;
382 :	Isibaar	3
383 :	chl	530	if (( xf > data->max_dx) \|\| ( xf < data->min_dx)
384 :			\|\| ( yf > data->max_dy) \|\| (yf < data->min_dy)) return;
385 :	Isibaar	3
386 :	syskin	663	if (data->qpel_precision) {
387 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
388 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
389 :			current = data->currentQMV;
390 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
391 :			} else {
392 :	syskin	748	ReferenceF = GetReference(xf, yf, data);
393 :	syskin	663	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
394 :	syskin	748	ReferenceB = GetReferenceB(xb, yb, 1, data);
395 :	syskin	663	current = data->currentMV;
396 :	chl	530	}
397 :	edgomez	195
398 :	syskin	724	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0)
399 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
400 :	syskin	712
401 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
402 :	syskin	663	sad += (data->lambda16 * t * sad)/1000;
403 :	syskin	628
404 :	chl	530	if (sad < *(data->iMinSAD)) {
405 :			*(data->iMinSAD) = sad;
406 :	syskin	663	current->x = xf; current->y = yf;
407 :	chl	530	*dir = Direction; }
408 :	Isibaar	3	}
409 :
410 :	syskin	748	static void
411 :	chl	530	CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
412 :	Isibaar	3	{
413 :	syskin	628	int32_t sad = 0;
414 :	chl	530	int k;
415 :			const uint8_t *ReferenceF;
416 :			const uint8_t *ReferenceB;
417 :			VECTOR mvs, b_mvs;
418 :	syskin	724	const VECTOR zeroMV={0,0};
419 :	Isibaar	3
420 :	chl	530	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
421 :	Isibaar	3
422 :	chl	530	for (k = 0; k < 4; k++) {
423 :			mvs.x = data->directmvF[k].x + x;
424 :			b_mvs.x = ((x == 0) ?
425 :			data->directmvB[k].x
426 :			: mvs.x - data->referencemv[k].x);
427 :	Isibaar	3
428 :	chl	530	mvs.y = data->directmvF[k].y + y;
429 :			b_mvs.y = ((y == 0) ?
430 :			data->directmvB[k].y
431 :			: mvs.y - data->referencemv[k].y);
432 :
433 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
434 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
435 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
436 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
437 :	Isibaar	3
438 :	syskin	748	if (!data->qpel) {
439 :			mvs.x = 2; mvs.y = 2;
440 :	syskin	663	b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
441 :	chl	530	}
442 :	syskin	663	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
443 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
444 :	chl	530
445 :			sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
446 :	syskin	663	ReferenceF, ReferenceB,
447 :	chl	530	data->iEdgedWidth);
448 :			if (sad > *(data->iMinSAD)) return;
449 :	edgomez	78	}
450 :	chl	530
451 :	syskin	724	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000;
452 :	syskin	628
453 :	chl	530	if (sad < *(data->iMinSAD)) {
454 :			*(data->iMinSAD) = sad;
455 :			data->currentMV->x = x; data->currentMV->y = y;
456 :			*dir = Direction; }
457 :	Isibaar	3	}
458 :
459 :	syskin	748	static void
460 :	chl	530	CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
461 :	chl	96	{
462 :	chl	530	int32_t sad;
463 :			const uint8_t *ReferenceF;
464 :			const uint8_t *ReferenceB;
465 :			VECTOR mvs, b_mvs;
466 :	syskin	724	const VECTOR zeroMV = {0,0};
467 :	chl	96
468 :	syskin	576	if (( x > 31) \|\| ( x < -32) \|\| ( y > 31) \|\| (y < -32)) return;
469 :	edgomez	195
470 :	chl	530	mvs.x = data->directmvF[0].x + x;
471 :			b_mvs.x = ((x == 0) ?
472 :			data->directmvB[0].x
473 :			: mvs.x - data->referencemv[0].x);
474 :	edgomez	195
475 :	chl	530	mvs.y = data->directmvF[0].y + y;
476 :			b_mvs.y = ((y == 0) ?
477 :			data->directmvB[0].y
478 :			: mvs.y - data->referencemv[0].y);
479 :
480 :			if (( mvs.x > data->max_dx ) \|\| ( mvs.x < data->min_dx )
481 :			\|\| ( mvs.y > data->max_dy ) \|\| ( mvs.y < data->min_dy )
482 :			\|\| ( b_mvs.x > data->max_dx ) \|\| ( b_mvs.x < data->min_dx )
483 :			\|\| ( b_mvs.y > data->max_dy ) \|\| ( b_mvs.y < data->min_dy )) return;
484 :	chl	96
485 :	syskin	748	if (!data->qpel) {
486 :			mvs.x = 2; mvs.y = 2;
487 :	syskin	663	b_mvs.x = 2; b_mvs.y = 2; //we move to qpel precision anyway
488 :	syskin	712	}
489 :	syskin	663	ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
490 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
491 :	chl	96
492 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
493 :	syskin	724	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000;
494 :	chl	96
495 :	chl	530	if (sad < *(data->iMinSAD)) {
496 :			*(data->iMinSAD) = sad;
497 :			data->currentMV->x = x; data->currentMV->y = y;
498 :			*dir = Direction; }
499 :			}
500 :	chl	96
501 :	chl	530	static void
502 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
503 :			{
504 :	syskin	628	int32_t sad; int t;
505 :	chl	530	const uint8_t * Reference;
506 :	edgomez	195
507 :	chl	530	if (( x > data->max_dx) \|\| ( x < data->min_dx)
508 :			\|\| ( y > data->max_dy) \|\| (y < data->min_dy)) return;
509 :	chl	96
510 :	syskin	663	if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data);
511 :	syskin	748	else Reference = GetReference(x, y, data);
512 :	edgomez	195
513 :	chl	530	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
514 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0);
515 :	edgomez	195
516 :	syskin	628	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100;
517 :	chl	96
518 :	Isibaar	606	if (sad < *(data->iMinSAD)) {
519 :			*(data->iMinSAD) = sad;
520 :			data->currentMV->x = x; data->currentMV->y = y;
521 :			*dir = Direction; }
522 :			}
523 :
524 :	Isibaar	539	/* CHECK_CANDIATE FUNCTIONS END */
525 :	chl	96
526 :	chl	530	/* MAINSEARCH FUNCTIONS START */
527 :	edgomez	195
528 :	chl	530	static void
529 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
530 :	chl	181	{
531 :
532 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
533 :
534 :	syskin	712	int iDirection;
535 :	edgomez	195
536 :	syskin	712	for(;;) { //forever
537 :			iDirection = 0;
538 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
539 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
540 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
541 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
542 :	chl	181
543 :	syskin	712	/* now we're doing diagonal checks near our candidate */
544 :	chl	181
545 :	syskin	712	if (iDirection) { //checking if anything found
546 :			bDirection = iDirection;
547 :			iDirection = 0;
548 :			x = data->currentMV->x; y = data->currentMV->y;
549 :			if (bDirection & 3) { //our candidate is left or right
550 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
551 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
552 :			} else { // what remains here is up or down
553 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
554 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
555 :	syskin	712	}
556 :	chl	181
557 :	syskin	712	if (iDirection) {
558 :			bDirection += iDirection;
559 :	syskin	748	x = data->currentMV->x; y = data->currentMV->y;
560 :	chl	181	}
561 :	syskin	712	} else { //about to quit, eh? not so fast....
562 :			switch (bDirection) {
563 :			case 2:
564 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
565 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
566 :			break;
567 :			case 1:
568 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
569 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
570 :			break;
571 :			case 2 + 4:
572 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
573 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
574 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
575 :			break;
576 :			case 4:
577 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
578 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
579 :			break;
580 :			case 8:
581 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
582 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
583 :			break;
584 :			case 1 + 4:
585 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
586 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
587 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
588 :			break;
589 :			case 2 + 8:
590 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
591 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
592 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
593 :			break;
594 :			case 1 + 8:
595 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
596 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
597 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
598 :			break;
599 :			default: //1+2+4+8 == we didn't find anything at all
600 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
601 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
602 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
603 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
604 :			break;
605 :			}
606 :			if (!iDirection) break; //ok, the end. really
607 :			bDirection = iDirection;
608 :			x = data->currentMV->x; y = data->currentMV->y;
609 :	chl	181	}
610 :	syskin	712	}
611 :	chl	181	}
612 :
613 :	chl	530	static void
614 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
615 :	chl	326	{
616 :	chl	530	int iDirection;
617 :	chl	326
618 :	chl	530	do {
619 :			iDirection = 0;
620 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
621 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
622 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
623 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
624 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
625 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
626 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
627 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
628 :	chl	326
629 :	chl	530	bDirection = iDirection;
630 :			x = data->currentMV->x; y = data->currentMV->y;
631 :			} while (iDirection);
632 :	chl	346	}
633 :
634 :	chl	530	static void
635 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
636 :	chl	346	{
637 :
638 :	chl	181	/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
639 :
640 :	syskin	712	int iDirection;
641 :	edgomez	195
642 :	syskin	712	do {
643 :			iDirection = 0;
644 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
645 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
646 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
647 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
648 :	chl	181
649 :	syskin	712	/* now we're doing diagonal checks near our candidate */
650 :	chl	181
651 :	syskin	712	if (iDirection) { //checking if anything found
652 :			bDirection = iDirection;
653 :			iDirection = 0;
654 :			x = data->currentMV->x; y = data->currentMV->y;
655 :			if (bDirection & 3) { //our candidate is left or right
656 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
657 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
658 :			} else { // what remains here is up or down
659 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
660 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
661 :	chl	181	}
662 :	syskin	712	bDirection += iDirection;
663 :			x = data->currentMV->x; y = data->currentMV->y;
664 :	chl	181	}
665 :	syskin	712	}
666 :			while (iDirection);
667 :	chl	181	}
668 :
669 :	chl	530	/* MAINSEARCH FUNCTIONS END */
670 :	chl	181
671 :	chl	530	/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
672 :
673 :			static void
674 :	syskin	663	SubpelRefine(const SearchData * const data)
675 :	chl	96	{
676 :	syskin	663	/* Do a half-pel or q-pel refinement */
677 :			VECTOR backupMV;
678 :	chl	530	int iDirection; //not needed
679 :	chl	96
680 :	syskin	663	if (data->qpel_precision)
681 :			backupMV = *(data->currentQMV);
682 :			else backupMV = *(data->currentMV);
683 :	edgomez	195
684 :	syskin	748	CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
685 :	Isibaar	580	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
686 :	syskin	748	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
687 :			CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
688 :			CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
689 :	Isibaar	580	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
690 :			CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
691 :	syskin	748	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
692 :	Isibaar	580	}
693 :
694 :	chl	530	static __inline int
695 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
696 :			const int x, const int y,
697 :	syskin	704	const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv)
698 :	chl	96
699 :	Isibaar	3	{
700 :	chl	530	/* keep repeating checks for all b-frames before this P frame,
701 :			to make sure that SKIP is possible (todo)
702 :			how: if skip is not possible set sad00 to a very high value */
703 :	syskin	704	if(rrv) {
704 :			uint32_t sadC = sad16(current->u + x16 + y(iEdgedWidth/2)*16,
705 :			reference->u + x16 + y(iEdgedWidth/2)16, iEdgedWidth/2, 2564096);
706 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
707 :			sadC += sad16(current->v + (x + y(iEdgedWidth/2))16,
708 :			reference->v + (x + y(iEdgedWidth/2))16, iEdgedWidth/2, 256*4096);
709 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
710 :			return 1;
711 :			} else {
712 :			uint32_t sadC = sad8(current->u + x8 + y(iEdgedWidth/2)*8,
713 :			reference->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2);
714 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
715 :			sadC += sad8(current->v + (x + y(iEdgedWidth/2))8,
716 :			reference->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
717 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
718 :			return 1;
719 :			}
720 :	chl	530	}
721 :	edgomez	195
722 :	chl	530	static __inline void
723 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
724 :			{
725 :			pMB->mode = MODE_NOT_CODED;
726 :	Isibaar	539	pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
727 :			pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
728 :	syskin	628
729 :	Isibaar	580	pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
730 :			pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
731 :
732 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
733 :	Isibaar	3	}
734 :
735 :	syskin	628	bool
736 :	chl	530	MotionEstimation(MBParam * const pParam,
737 :			FRAMEINFO * const current,
738 :			FRAMEINFO * const reference,
739 :			const IMAGE * const pRefH,
740 :			const IMAGE * const pRefV,
741 :			const IMAGE * const pRefHV,
742 :			const uint32_t iLimit)
743 :	Isibaar	3	{
744 :	chl	530	MACROBLOCK *const pMBs = current->mbs;
745 :			const IMAGE *const pCurrent = &current->image;
746 :			const IMAGE *const pRef = &reference->image;
747 :	Isibaar	3
748 :	chl	530	const VECTOR zeroMV = { 0, 0 };
749 :	Isibaar	3
750 :	syskin	704	uint32_t mb_width = pParam->mb_width;
751 :			uint32_t mb_height = pParam->mb_height;
752 :	suxen_drol	702
753 :	chl	530	uint32_t x, y;
754 :			uint32_t iIntra = 0;
755 :	syskin	628	int32_t InterBias, quant = current->quant, sad00;
756 :	edgomez	195
757 :	Isibaar	539	// some pre-initialized thingies for SearchP
758 :	syskin	712	int32_t temp[8];
759 :	Isibaar	539	VECTOR currentMV[5];
760 :	Isibaar	580	VECTOR currentQMV[5];
761 :	Isibaar	539	int32_t iMinSAD[5];
762 :			SearchData Data;
763 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
764 :	Isibaar	539	Data.iEdgedWidth = pParam->edged_width;
765 :			Data.currentMV = currentMV;
766 :	Isibaar	580	Data.currentQMV = currentQMV;
767 :	Isibaar	539	Data.iMinSAD = iMinSAD;
768 :			Data.temp = temp;
769 :			Data.iFcode = current->fcode;
770 :	Isibaar	580	Data.rounding = pParam->m_rounding_type;
771 :	syskin	628	Data.qpel = pParam->m_quarterpel;
772 :			Data.chroma = current->global_flags & XVID_ME_COLOUR;
773 :	syskin	704	Data.rrv = current->global_flags & XVID_REDUCED;
774 :	Isibaar	539
775 :	syskin	704	if ((current->global_flags & XVID_REDUCED)) {
776 :	suxen_drol	702	mb_width = (pParam->width + 31) / 32;
777 :			mb_height = (pParam->height + 31) / 32;
778 :	syskin	704	Data.qpel = Data.chroma = 0;
779 :	suxen_drol	702	}
780 :
781 :	syskin	754	Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose)
782 :	chl	530	if (sadInit) (*sadInit) ();
783 :	edgomez	195
784 :	suxen_drol	702	for (y = 0; y < mb_height; y++) {
785 :			for (x = 0; x < mb_width; x++) {
786 :	chl	530	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
787 :	syskin	628
788 :	syskin	748	if (Data.rrv) pMB->sad16 =
789 :	syskin	704	sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32,
790 :			pRef->y + (x + y * pParam->edged_width) * 32,
791 :			pParam->edged_width, pMB->sad8 );
792 :
793 :	syskin	748	else pMB->sad16 =
794 :	syskin	704	sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
795 :	chl	530	pRef->y + (x + y * pParam->edged_width) * 16,
796 :			pParam->edged_width, pMB->sad8 );
797 :	edgomez	195
798 :	syskin	628	if (Data.chroma) {
799 :			pMB->sad16 += sad8(pCurrent->u + x8 + y(pParam->edged_width/2)*8,
800 :			pRef->u + x8 + y(pParam->edged_width/2)*8, pParam->edged_width/2);
801 :
802 :			pMB->sad16 += sad8(pCurrent->v + (x + y(pParam->edged_width/2))8,
803 :			pRef->v + (x + y(pParam->edged_width/2))8, pParam->edged_width/2);
804 :			}
805 :
806 :			sad00 = pMB->sad16; //if no gmc; else sad00 = (..)
807 :
808 :	chl	530	if (!(current->global_flags & XVID_LUMIMASKING)) {
809 :			pMB->dquant = NO_CHANGE;
810 :	syskin	748	pMB->quant = current->quant;
811 :	syskin	601	} else {
812 :	syskin	574	if (pMB->dquant != NO_CHANGE) {
813 :			quant += DQtab[pMB->dquant];
814 :			if (quant > 31) quant = 31;
815 :			else if (quant < 1) quant = 1;
816 :			}
817 :	syskin	601	pMB->quant = quant;
818 :			}
819 :	chl	181
820 :	syskin	628	//initial skip decision
821 :	chl	619	/* no early skip for GMC (global vector = skip vector is unknown!) */
822 :			if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
823 :	syskin	704	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
824 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
825 :	chl	619	SkipMacroblockP(pMB, sad00);
826 :			continue;
827 :			}
828 :			}
829 :	syskin	628
830 :			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
831 :	chl	530	y, current->motion_flags, pMB->quant,
832 :	Isibaar	539	&Data, pParam, pMBs, reference->mbs,
833 :	chl	530	current->global_flags & XVID_INTER4V, pMB);
834 :	edgomez	195
835 :	chl	530	/* final skip decision, a.k.a. "the vector you found, really that good?" */
836 :	syskin	748	if (current->coding_type == P_VOP) {
837 :	chl	619	if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
838 :	syskin	704	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
839 :			if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) {
840 :	chl	619	SkipMacroblockP(pMB, sad00);
841 :			continue;
842 :			}
843 :			}
844 :	syskin	628
845 :	chl	530	/* finally, intra decision */
846 :	Isibaar	3
847 :	chl	530	InterBias = MV16_INTER_BIAS;
848 :	syskin	639	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
849 :	chl	530	if (y != 0)
850 :	syskin	639	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
851 :	chl	530	if (x != 0)
852 :	syskin	639	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
853 :	syskin	628
854 :			if (Data.chroma) InterBias += 50; // to compensate bigger SAD
855 :	syskin	704	if (Data.rrv) InterBias *= 4; //??
856 :	Isibaar	3
857 :	chl	530	if (InterBias < pMB->sad16) {
858 :	syskin	704	int32_t deviation;
859 :			if (Data.rrv) {
860 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32,
861 :			pParam->edged_width)
862 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16,
863 :			pParam->edged_width)
864 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width,
865 :			pParam->edged_width)
866 :			+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1),
867 :			pParam->edged_width);
868 :			} else
869 :			deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
870 :	chl	530	pParam->edged_width);
871 :	Isibaar	3
872 :	syskin	704	if (deviation < (pMB->sad16 - InterBias)) {
873 :	syskin	754	if (++iIntra >= iLimit) return 1;
874 :	chl	530	pMB->mode = MODE_INTRA;
875 :	Isibaar	539	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
876 :			pMB->mvs[3] = zeroMV;
877 :	Isibaar	580	pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
878 :			pMB->qmvs[3] = zeroMV;
879 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
880 :			pMB->sad8[3] = 0;
881 :			}
882 :			}
883 :			}
884 :	Isibaar	3	}
885 :	chl	619
886 :			if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
887 :	syskin	628	current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
888 :	chl	619	else
889 :			current->GMC_MV = zeroMV;
890 :
891 :	syskin	628	return 0;
892 :	chl	530	}
893 :	Isibaar	3
894 :	edgomez	195
895 :	chl	530	#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16\|PMV_HALFPELREFINE16)
896 :	Isibaar	3
897 :	chl	530	static __inline int
898 :			make_mask(const VECTOR * const pmv, const int i)
899 :			{
900 :	Isibaar	539	int mask = 255, j;
901 :	chl	530	for (j = 0; j < i; j++) {
902 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
903 :			if (pmv[i].x == pmv[j].x) {
904 :			if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
905 :			if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
906 :	syskin	748	} else
907 :	chl	530	if (pmv[i].y == pmv[j].y) {
908 :			if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
909 :			if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
910 :			}
911 :	Isibaar	3	}
912 :	chl	530	return mask;
913 :			}
914 :	edgomez	195
915 :	syskin	748	static __inline void
916 :	syskin	704	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
917 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
918 :	chl	530	{
919 :	edgomez	195
920 :	chl	530	//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
921 :	syskin	704	if (rrv) { iWcount /= 2; iHcount /= 2; }
922 :	edgomez	195
923 :	syskin	704	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
924 :	chl	530	pmv[5].x = EVEN(pmv[3].x);
925 :	syskin	748	pmv[5].y = EVEN(pmv[3].y);
926 :	Isibaar	539	} else pmv[5].x = pmv[5].y = 0;
927 :	chl	169
928 :	chl	530	if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
929 :			else pmv[3].x = pmv[3].y = 0;
930 :	Isibaar	3
931 :	chl	530	if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
932 :			else pmv[4].x = pmv[4].y = 0;
933 :	chl	169
934 :	chl	530	// [1] median prediction
935 :	syskin	704	if (rrv) { //median is in halfzero-precision
936 :			pmv[1].x = RRV_MV_SCALEUP(pmv[0].x);
937 :			pmv[1].y = RRV_MV_SCALEUP(pmv[0].y);
938 :			} else { pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); }
939 :	chl	169
940 :	chl	530	pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
941 :	chl	169
942 :	chl	530	pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
943 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
944 :	chl	169
945 :	syskin	704	if ((x < iWcount-1) && (y < iHcount-1)) {
946 :	chl	530	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
947 :	syskin	748	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
948 :	Isibaar	539	} else pmv[6].x = pmv[6].y = 0;
949 :	syskin	704
950 :			if (rrv) {
951 :			int i;
952 :			for (i = 0; i < 7; i++) {
953 :			pmv[i].x = RRV_MV_SCALEDOWN(pmv[i].x);
954 :			pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // a trick
955 :			}
956 :			}
957 :	chl	530	}
958 :	chl	169
959 :	chl	530	static void
960 :	syskin	628	SearchP(const IMAGE * const pRef,
961 :	chl	530	const uint8_t * const pRefH,
962 :			const uint8_t * const pRefV,
963 :			const uint8_t * const pRefHV,
964 :			const IMAGE * const pCur,
965 :			const int x,
966 :			const int y,
967 :			const uint32_t MotionFlags,
968 :			const uint32_t iQuant,
969 :	Isibaar	539	SearchData * const Data,
970 :	chl	530	const MBParam * const pParam,
971 :			const MACROBLOCK * const pMBs,
972 :			const MACROBLOCK * const prevMBs,
973 :			int inter4v,
974 :			MACROBLOCK * const pMB)
975 :			{
976 :	chl	169
977 :	chl	530	int i, iDirection = 255, mask, threshA;
978 :	Isibaar	539	VECTOR pmv[7];
979 :	Isibaar	3
980 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
981 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
982 :	Isibaar	3
983 :	syskin	712	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
984 :	syskin	628
985 :	syskin	712	Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache
986 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
987 :			Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
988 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
989 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
990 :	Isibaar	3
991 :	syskin	704	Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
992 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
993 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
994 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
995 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
996 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
997 :
998 :	syskin	601	Data->lambda16 = lambda_vec16[iQuant];
999 :	Isibaar	606	Data->lambda8 = lambda_vec8[iQuant];
1000 :	syskin	663	Data->qpel_precision = 0;
1001 :	chl	169
1002 :	edgomez	568	if (pMB->dquant != NO_CHANGE) inter4v = 0;
1003 :	edgomez	195
1004 :	Isibaar	580	for(i = 0; i < 5; i++)
1005 :			Data->currentMV[i].x = Data->currentMV[i].y = 0;
1006 :	Isibaar	539
1007 :	syskin	663	if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1008 :			else Data->predMV = pmv[0];
1009 :	Isibaar	606
1010 :	syskin	724	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
1011 :	syskin	628	Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000;
1012 :			Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100;
1013 :	Isibaar	539	Data->iMinSAD[2] = pMB->sad8[1];
1014 :			Data->iMinSAD[3] = pMB->sad8[2];
1015 :			Data->iMinSAD[4] = pMB->sad8[3];
1016 :
1017 :	chl	530	if ((x == 0) && (y == 0)) threshA = 512;
1018 :			else {
1019 :	edgomez	568	threshA = Data->temp[0]; // that's when we keep this SAD atm
1020 :	chl	530	if (threshA < 512) threshA = 512;
1021 :			if (threshA > 1024) threshA = 1024; }
1022 :	Isibaar	3
1023 :	chl	530	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1024 :	syskin	704	prevMBs + x + y * pParam->mb_width, Data->rrv);
1025 :	edgomez	195
1026 :	syskin	704	if (Data->rrv) CheckCandidate = CheckCandidate32;
1027 :			else if (inter4v \|\| Data->chroma) CheckCandidate = CheckCandidate16;
1028 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
1029 :	edgomez	568
1030 :	chl	530	/* main loop. checking all predictions */
1031 :
1032 :			for (i = 1; i < 7; i++) {
1033 :			if (!(mask = make_mask(pmv, i)) ) continue;
1034 :	edgomez	568	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1035 :	Isibaar	539	if (Data->iMinSAD[0] <= threshA) break;
1036 :	chl	530	}
1037 :
1038 :	Isibaar	539	if ((Data->iMinSAD[0] <= threshA) \|\|
1039 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1040 :			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1041 :	chl	530	inter4v = 0;
1042 :	Isibaar	539	} else {
1043 :	Isibaar	3
1044 :	syskin	748	MainSearchFunc * MainSearchPtr;
1045 :	Isibaar	539	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1046 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1047 :			else MainSearchPtr = DiamondSearch;
1048 :	chl	181
1049 :	Isibaar	539	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1050 :	Isibaar	3
1051 :	chl	530	/* extended search, diamond starting in 0,0 and in prediction.
1052 :			note that this search is/might be done in halfpel positions,
1053 :			which makes it more different than the diamond above */
1054 :	chl	259
1055 :	Isibaar	539	if (MotionFlags & PMV_EXTSEARCH16) {
1056 :			int32_t bSAD;
1057 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1058 :	syskin	704	if (Data->rrv) {
1059 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1060 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1061 :	syskin	748	} else
1062 :	syskin	704	if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1063 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1064 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1065 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1066 :	Isibaar	3
1067 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1068 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1069 :			if (bSAD < Data->iMinSAD[0]) {
1070 :			Data->currentMV[0] = backupMV;
1071 :			Data->iMinSAD[0] = bSAD; }
1072 :			}
1073 :	Isibaar	3
1074 :	Isibaar	539	backupMV = Data->currentMV[0];
1075 :	syskin	704	if (!MotionFlags & PMV_HALFPELREFINE16 \|\| Data->rrv) startMV.x = startMV.y = 0;
1076 :			else startMV.x = startMV.y = 1;
1077 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1078 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1079 :	edgomez	195
1080 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1081 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1082 :			if (bSAD < Data->iMinSAD[0]) {
1083 :			Data->currentMV[0] = backupMV;
1084 :			Data->iMinSAD[0] = bSAD; }
1085 :			}
1086 :	Isibaar	3	}
1087 :			}
1088 :
1089 :	syskin	663	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1090 :	Isibaar	3
1091 :	Isibaar	580	for(i = 0; i < 5; i++) {
1092 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1093 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1094 :			}
1095 :
1096 :	syskin	748	if((Data->qpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1097 :	Isibaar	580
1098 :	syskin	663	Data->qpel_precision = 1;
1099 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1100 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1101 :	syskin	601
1102 :	syskin	663	SubpelRefine(Data);
1103 :	Isibaar	580	}
1104 :
1105 :	syskin	628	if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0;
1106 :	Isibaar	539	if (inter4v) {
1107 :			SearchData Data8;
1108 :			Data8.iFcode = Data->iFcode;
1109 :	Isibaar	606	Data8.lambda8 = Data->lambda8;
1110 :	Isibaar	539	Data8.iEdgedWidth = Data->iEdgedWidth;
1111 :	Isibaar	606	Data8.RefQ = Data->RefQ;
1112 :	syskin	628	Data8.qpel = Data->qpel;
1113 :	syskin	704	Data8.rrv = Data->rrv;
1114 :	Isibaar	539	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1115 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1116 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1117 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1118 :	syskin	628
1119 :			if (Data->chroma) {
1120 :	syskin	753	int sumx, sumy;
1121 :	syskin	628
1122 :			if(pParam->m_quarterpel) {
1123 :	syskin	753	sumx = pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2;
1124 :	syskin	663	sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2;
1125 :			} else {
1126 :			sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1127 :			sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1128 :			}
1129 :	syskin	628
1130 :	syskin	753	Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf],
1131 :			(sumy >> 3) + roundtab_76[sumy & 0xf], Data);
1132 :	syskin	628	}
1133 :	Isibaar	539	}
1134 :	Isibaar	3
1135 :	syskin	704	if (Data->rrv) {
1136 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1137 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1138 :			}
1139 :	syskin	740
1140 :	chl	530	if (!(inter4v) \|\|
1141 :	syskin	748	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1142 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1143 :	chl	530	// INTER MODE
1144 :			pMB->mode = MODE_INTER;
1145 :	Isibaar	539	pMB->mvs[0] = pMB->mvs[1]
1146 :			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1147 :	Isibaar	3
1148 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1149 :	Isibaar	539	pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1150 :	Isibaar	3
1151 :	Isibaar	580	if(pParam->m_quarterpel) {
1152 :	syskin	663	pMB->qmvs[0] = pMB->qmvs[1]
1153 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1154 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1155 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1156 :	syskin	628	} else {
1157 :	Isibaar	580	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1158 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1159 :			}
1160 :	edgomez	195	} else {
1161 :	syskin	748	// INTER4V MODE; all other things are already set in Search8
1162 :	chl	530	pMB->mode = MODE_INTER4V;
1163 :	syskin	748	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1164 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1165 :	edgomez	78	}
1166 :	Isibaar	3	}
1167 :
1168 :	chl	530	static void
1169 :			Search8(const SearchData * const OldData,
1170 :			const int x, const int y,
1171 :			const uint32_t MotionFlags,
1172 :			const MBParam * const pParam,
1173 :			MACROBLOCK * const pMB,
1174 :			const MACROBLOCK * const pMBs,
1175 :	Isibaar	539	const int block,
1176 :			SearchData * const Data)
1177 :	chl	345	{
1178 :	syskin	704	int i = 0;
1179 :	Isibaar	539	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1180 :			Data->currentMV = OldData->currentMV + 1 + block;
1181 :	Isibaar	580	Data->currentQMV = OldData->currentQMV + 1 + block;
1182 :	chl	345
1183 :	syskin	601	if(pParam->m_quarterpel) {
1184 :	syskin	704	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1185 :	syskin	724	if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1186 :			Data->predMV, Data->iFcode, 0, 0);
1187 :	syskin	712
1188 :	Isibaar	606	} else {
1189 :	syskin	704	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1190 :			if (block != 0) {
1191 :	syskin	724	if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1192 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1193 :	syskin	704	}
1194 :	Isibaar	606	}
1195 :	chl	345
1196 :	syskin	704	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1197 :
1198 :	chl	530	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8)) {
1199 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
1200 :	chl	345
1201 :	syskin	704	Data->Ref = OldData->Ref + i8 ((block&1) + pParam->edged_width*(block>>1));
1202 :			Data->RefH = OldData->RefH + i8 ((block&1) + pParam->edged_width*(block>>1));
1203 :			Data->RefV = OldData->RefV + i8 ((block&1) + pParam->edged_width*(block>>1));
1204 :			Data->RefHV = OldData->RefHV + i8 ((block&1) + pParam->edged_width*(block>>1));
1205 :	chl	345
1206 :	syskin	704	Data->Cur = OldData->Cur + i8 ((block&1) + pParam->edged_width*(block>>1));
1207 :	syskin	663	Data->qpel_precision = 0;
1208 :	chl	345
1209 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1210 :			pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv);
1211 :
1212 :			if (Data->rrv) CheckCandidate = CheckCandidate16no4v;
1213 :			else CheckCandidate = CheckCandidate8;
1214 :
1215 :	chl	530	if (MotionFlags & PMV_EXTSEARCH8) {
1216 :	Isibaar	580	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1217 :
1218 :	chl	530	MainSearchFunc *MainSearchPtr;
1219 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1220 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1221 :			else MainSearchPtr = DiamondSearch;
1222 :	chl	345
1223 :	Isibaar	580	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1224 :	chl	345
1225 :	syskin	601	if(*(Data->iMinSAD) < temp_sad) {
1226 :	Isibaar	580	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1227 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1228 :			}
1229 :			}
1230 :
1231 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1232 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1233 :
1234 :	syskin	663	SubpelRefine(Data); // perform halfpel refine of current best vector
1235 :	Isibaar	580
1236 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1237 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1238 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1239 :			}
1240 :			}
1241 :
1242 :	syskin	748	if(Data->qpel) {
1243 :	syskin	601	if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1244 :			(MotionFlags & PMV_QUARTERPELREFINE8)) {
1245 :	syskin	663	Data->qpel_precision = 1;
1246 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1247 :			pParam->width, pParam->height, OldData->iFcode, 1, 0);
1248 :	syskin	663	SubpelRefine(Data);
1249 :	syskin	601	}
1250 :	Isibaar	580	}
1251 :	chl	345	}
1252 :	chl	530
1253 :	syskin	704	if (Data->rrv) {
1254 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1255 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1256 :			}
1257 :
1258 :	syskin	712	if(Data->qpel) {
1259 :	syskin	663	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1260 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1261 :			pMB->qmvs[block] = *(Data->currentQMV);
1262 :	syskin	704	} else {
1263 :	Isibaar	580	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1264 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1265 :			}
1266 :
1267 :	Isibaar	539	pMB->mvs[block] = *(Data->currentMV);
1268 :	syskin	601	pMB->sad8[block] = 4 * (*Data->iMinSAD);
1269 :	chl	345	}
1270 :
1271 :	chl	530	/* B-frames code starts here */
1272 :	chl	345
1273 :	chl	530	static __inline VECTOR
1274 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1275 :	Isibaar	3	{
1276 :	chl	530	/* the stupidiest function ever */
1277 :			if (mode == MODE_FORWARD) return pMB->mvs[0];
1278 :			else return pMB->b_mvs[0];
1279 :	Isibaar	3	}
1280 :
1281 :	chl	530	static void __inline
1282 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1283 :			const uint32_t iWcount,
1284 :			const MACROBLOCK * const pMB,
1285 :			const uint32_t mode_curr)
1286 :	Isibaar	3	{
1287 :
1288 :	chl	530	// [0] is prediction
1289 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1290 :	Isibaar	3
1291 :	chl	530	pmv[1].x = pmv[1].y = 0; // [1] is zero
1292 :	Isibaar	3
1293 :	chl	530	pmv[2] = ChoosePred(pMB, mode_curr);
1294 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1295 :	edgomez	195
1296 :	chl	530	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1297 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1298 :	syskin	748	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1299 :	Isibaar	539	} else pmv[3].x = pmv[3].y = 0;
1300 :	Isibaar	3
1301 :	chl	530	if (y != 0) {
1302 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1303 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1304 :			} else pmv[4].x = pmv[4].y = 0;
1305 :	Isibaar	3
1306 :	chl	530	if (x != 0) {
1307 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1308 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1309 :			} else pmv[5].x = pmv[5].y = 0;
1310 :	Isibaar	3
1311 :	chl	530	if ((x != 0)&&(y != 0)) {
1312 :			pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1313 :			pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1314 :			} else pmv[6].x = pmv[6].y = 0;
1315 :	edgomez	195
1316 :	chl	530	// more?
1317 :			}
1318 :	chl	181
1319 :	edgomez	170
1320 :	chl	530	/* search backward or forward, for b-frames */
1321 :			static void
1322 :			SearchBF( const uint8_t * const pRef,
1323 :			const uint8_t * const pRefH,
1324 :			const uint8_t * const pRefV,
1325 :			const uint8_t * const pRefHV,
1326 :			const IMAGE * const pCur,
1327 :			const int x, const int y,
1328 :			const uint32_t MotionFlags,
1329 :			const uint32_t iFcode,
1330 :			const MBParam * const pParam,
1331 :			MACROBLOCK * const pMB,
1332 :			const VECTOR * const predMV,
1333 :			int32_t * const best_sad,
1334 :	h	545	const int32_t mode_current,
1335 :			SearchData * const Data)
1336 :	chl	530	{
1337 :	Isibaar	3
1338 :	chl	530	const int32_t iEdgedWidth = pParam->edged_width;
1339 :
1340 :	syskin	740	int i, iDirection = 255, mask;
1341 :	h	545	VECTOR pmv[7];
1342 :	chl	530	MainSearchFunc *MainSearchPtr;
1343 :	h	545	*Data->iMinSAD = MV_MAX_ERROR;
1344 :			Data->iFcode = iFcode;
1345 :	syskin	663	Data->qpel_precision = 0;
1346 :	Isibaar	3
1347 :	h	545	Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1348 :			Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1349 :			Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1350 :			Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1351 :	Isibaar	3
1352 :	h	545	Data->predMV = *predMV;
1353 :	edgomez	195
1354 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1355 :	syskin	712	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1356 :	Isibaar	3
1357 :	h	545	pmv[0] = Data->predMV;
1358 :	syskin	658	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1359 :	h	545	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1360 :	edgomez	195
1361 :	h	545	Data->currentMV->x = Data->currentMV->y = 0;
1362 :	chl	530	CheckCandidate = CheckCandidate16no4v;
1363 :	chl	181
1364 :	chl	530	// main loop. checking all predictions
1365 :	syskin	740	for (i = 0; i < 7; i++) {
1366 :	chl	530	if (!(mask = make_mask(pmv, i)) ) continue;
1367 :	h	545	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1368 :	edgomez	78	}
1369 :	Isibaar	3
1370 :	chl	530	if (MotionFlags & PMV_USESQUARES16)
1371 :			MainSearchPtr = SquareSearch;
1372 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1373 :			MainSearchPtr = AdvDiamondSearch;
1374 :			else MainSearchPtr = DiamondSearch;
1375 :	chl	96
1376 :	syskin	740	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1377 :	chl	169
1378 :	syskin	663	SubpelRefine(Data);
1379 :	syskin	658
1380 :	syskin	740	if (Data->qpel && Data->iMinSAD < best_sad + 300) {
1381 :	syskin	658	Data->currentQMV->x = 2*Data->currentMV->x;
1382 :			Data->currentQMV->y = 2*Data->currentMV->y;
1383 :	syskin	663	Data->qpel_precision = 1;
1384 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1385 :			pParam->width, pParam->height, iFcode, 1, 0);
1386 :	syskin	663	SubpelRefine(Data);
1387 :	syskin	658	}
1388 :	chl	169
1389 :	chl	530	// three bits are needed to code backward mode. four for forward
1390 :	syskin	740
1391 :	syskin	628	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1392 :			else Data->iMinSAD += 3 Data->lambda16;
1393 :	chl	169
1394 :	h	545	if (Data->iMinSAD < best_sad) {
1395 :			best_sad = Data->iMinSAD;
1396 :	chl	530	pMB->mode = mode_current;
1397 :	syskin	658	if (Data->qpel) {
1398 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1399 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1400 :	syskin	748	if (mode_current == MODE_FORWARD)
1401 :	syskin	658	pMB->qmvs[0] = *Data->currentQMV;
1402 :	syskin	748	else
1403 :	syskin	658	pMB->b_qmvs[0] = *Data->currentQMV;
1404 :			} else {
1405 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1406 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1407 :			}
1408 :	syskin	739	if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1409 :			else pMB->b_mvs[0] = *Data->currentMV;
1410 :			}
1411 :			if (mode_current == MODE_FORWARD) (Data->currentMV+2) = Data->currentMV;
1412 :			else (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1413 :	syskin	658
1414 :	chl	530	}
1415 :	chl	169
1416 :	syskin	748	static void
1417 :	syskin	712	SkipDecisionB(const IMAGE * const pCur,
1418 :			const IMAGE * const f_Ref,
1419 :	syskin	748	const IMAGE * const b_Ref,
1420 :	syskin	712	MACROBLOCK * const pMB,
1421 :			const uint32_t quant,
1422 :			const uint32_t x, const uint32_t y,
1423 :			const SearchData * const Data)
1424 :			{
1425 :			int dx, dy, b_dx, b_dy;
1426 :			uint32_t sum;
1427 :			//this is not full chroma compensation, only it's fullpel approximation. should work though
1428 :			if (Data->qpel) {
1429 :			dy = Data->directmvF[0].y/2 + Data->directmvF[1].y/2 +
1430 :			Data->directmvF[2].y/2 + Data->directmvF[3].y/2;
1431 :
1432 :			dx = Data->directmvF[0].x/2 + Data->directmvF[1].x/2 +
1433 :			Data->directmvF[2].x/2 + Data->directmvF[3].x/2;
1434 :
1435 :			b_dy = Data->directmvB[0].y/2 + Data->directmvB[1].y/2 +
1436 :			Data->directmvB[2].y/2 + Data->directmvB[3].y/2;
1437 :
1438 :			b_dx = Data->directmvB[0].x/2 + Data->directmvB[1].x/2 +
1439 :			Data->directmvB[2].x/2 + Data->directmvB[3].x/2;
1440 :
1441 :			} else {
1442 :			dy = Data->directmvF[0].y + Data->directmvF[1].y +
1443 :			Data->directmvF[2].y + Data->directmvF[3].y;
1444 :
1445 :			dx = Data->directmvF[0].x + Data->directmvF[1].x +
1446 :			Data->directmvF[2].x + Data->directmvF[3].x;
1447 :
1448 :			b_dy = Data->directmvB[0].y + Data->directmvB[1].y +
1449 :			Data->directmvB[2].y + Data->directmvB[3].y;
1450 :
1451 :			b_dx = Data->directmvB[0].x + Data->directmvB[1].x +
1452 :			Data->directmvB[2].x + Data->directmvB[3].x;
1453 :			}
1454 :
1455 :
1456 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1457 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1458 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1459 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1460 :
1461 :			sum = sad8bi(pCur->u + 8x + 8y*(Data->iEdgedWidth/2),
1462 :			f_Ref->u + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1463 :			b_Ref->u + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1464 :			Data->iEdgedWidth/2);
1465 :			sum += sad8bi(pCur->v + 8x + 8y*(Data->iEdgedWidth/2),
1466 :			f_Ref->v + (y8 + dy/2) (Data->iEdgedWidth/2) + x*8 + dx/2,
1467 :			b_Ref->v + (y8 + b_dy/2) (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1468 :			Data->iEdgedWidth/2);
1469 :
1470 :			if (sum < 2MAX_CHROMA_SAD_FOR_SKIP quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1471 :			}
1472 :
1473 :
1474 :
1475 :			static __inline uint32_t
1476 :	Isibaar	539	SearchDirect(const IMAGE * const f_Ref,
1477 :	chl	530	const uint8_t * const f_RefH,
1478 :			const uint8_t * const f_RefV,
1479 :			const uint8_t * const f_RefHV,
1480 :	Isibaar	539	const IMAGE * const b_Ref,
1481 :	chl	530	const uint8_t * const b_RefH,
1482 :			const uint8_t * const b_RefV,
1483 :			const uint8_t * const b_RefHV,
1484 :			const IMAGE * const pCur,
1485 :			const int x, const int y,
1486 :			const uint32_t MotionFlags,
1487 :			const int32_t TRB, const int32_t TRD,
1488 :			const MBParam * const pParam,
1489 :			MACROBLOCK * const pMB,
1490 :			const MACROBLOCK * const b_mb,
1491 :	h	545	int32_t * const best_sad,
1492 :			SearchData * const Data)
1493 :	edgomez	195
1494 :	chl	530	{
1495 :	h	545	int32_t skip_sad;
1496 :	chl	530	int k;
1497 :	h	545
1498 :	chl	530	MainSearchFunc *MainSearchPtr;
1499 :	Isibaar	3
1500 :	h	545	Data->iMinSAD = 2564096;
1501 :	chl	140
1502 :	h	545	Data->Ref = f_Ref->y + (x + Data->iEdgedWidthy) 16;
1503 :			Data->RefH = f_RefH + (x + Data->iEdgedWidthy) 16;
1504 :			Data->RefV = f_RefV + (x + Data->iEdgedWidthy) 16;
1505 :			Data->RefHV = f_RefHV + (x + Data->iEdgedWidthy) 16;
1506 :			Data->bRef = b_Ref->y + (x + Data->iEdgedWidthy) 16;
1507 :			Data->bRefH = b_RefH + (x + Data->iEdgedWidthy) 16;
1508 :			Data->bRefV = b_RefV + (x + Data->iEdgedWidthy) 16;
1509 :			Data->bRefHV = b_RefHV + (x + Data->iEdgedWidthy) 16;
1510 :	chl	140
1511 :	h	545	Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1512 :			Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1513 :			Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1514 :			Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1515 :	syskin	658	if (Data->qpel) { //we measure in qpixels
1516 :			Data->max_dx *= 2;
1517 :			Data->max_dy *= 2;
1518 :			Data->min_dx *= 2;
1519 :			Data->min_dy *= 2;
1520 :			Data->referencemv = b_mb->qmvs;
1521 :			} else Data->referencemv = b_mb->mvs;
1522 :	syskin	672	Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here
1523 :	Isibaar	3
1524 :	chl	530	for (k = 0; k < 4; k++) {
1525 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1526 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1527 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1528 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1529 :	Isibaar	3
1530 :	h	545	if ( ( pMB->b_mvs[k].x > Data->max_dx ) \|\| ( pMB->b_mvs[k].x < Data->min_dx )
1531 :			\|\| ( pMB->b_mvs[k].y > Data->max_dy ) \|\| ( pMB->b_mvs[k].y < Data->min_dy )) {
1532 :	Isibaar	3
1533 :	Isibaar	539	best_sad = 2564096; // in that case, we won't use direct mode
1534 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1535 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1536 :	syskin	712	return 256*4096;
1537 :	Isibaar	539	}
1538 :			if (b_mb->mode != MODE_INTER4V) {
1539 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1540 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1541 :	h	545	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1542 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1543 :	Isibaar	539	break;
1544 :			}
1545 :			}
1546 :	Isibaar	3
1547 :	syskin	663	if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect;
1548 :			else CheckCandidate = CheckCandidateDirectno4v;
1549 :	syskin	658
1550 :	h	545	(*CheckCandidate)(0, 0, 255, &k, Data);
1551 :	Isibaar	3
1552 :	syskin	712	// initial (fast) skip decision
1553 :			if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH*2) {
1554 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data->chroma, Data); //possible skip - checking chroma
1555 :			if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1556 :	Isibaar	539	}
1557 :
1558 :	h	545	skip_sad = *Data->iMinSAD;
1559 :	edgomez	195
1560 :	chl	530	// DIRECT MODE DELTA VECTOR SEARCH.
1561 :			// This has to be made more effective, but at the moment I'm happy it's running at all
1562 :	Isibaar	3
1563 :	chl	530	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1564 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1565 :			else MainSearchPtr = DiamondSearch;
1566 :	edgomez	195
1567 :	h	545	(*MainSearchPtr)(0, 0, Data, 255);
1568 :	Isibaar	3
1569 :	syskin	663	SubpelRefine(Data);
1570 :	Isibaar	3
1571 :	h	545	best_sad = Data->iMinSAD;
1572 :	Isibaar	3
1573 :	syskin	743	if (b_mb->mode == MODE_INTER4V \|\| Data->qpel) pMB->mode = MODE_DIRECT;
1574 :	syskin	672	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1575 :	edgomez	195
1576 :	h	545	pMB->pmvs[3] = *Data->currentMV;
1577 :	edgomez	195
1578 :	chl	530	for (k = 0; k < 4; k++) {
1579 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1580 :	syskin	658	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1581 :	h	545	? Data->directmvB[k].x
1582 :	syskin	658	:pMB->mvs[k].x - Data->referencemv[k].x);
1583 :	h	545	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1584 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1585 :			? Data->directmvB[k].y
1586 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1587 :	syskin	658	if (Data->qpel) {
1588 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1589 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1590 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1591 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1592 :			}
1593 :
1594 :	chl	530	if (b_mb->mode != MODE_INTER4V) {
1595 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1596 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1597 :	syskin	658	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1598 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1599 :	chl	530	break;
1600 :			}
1601 :			}
1602 :	h	545	return skip_sad;
1603 :	Isibaar	3	}
1604 :	chl	96
1605 :	h	545
1606 :	syskin	712	static void
1607 :	chl	530	SearchInterpolate(const uint8_t * const f_Ref,
1608 :			const uint8_t * const f_RefH,
1609 :			const uint8_t * const f_RefV,
1610 :			const uint8_t * const f_RefHV,
1611 :			const uint8_t * const b_Ref,
1612 :			const uint8_t * const b_RefH,
1613 :			const uint8_t * const b_RefV,
1614 :			const uint8_t * const b_RefHV,
1615 :			const IMAGE * const pCur,
1616 :			const int x, const int y,
1617 :			const uint32_t fcode,
1618 :			const uint32_t bcode,
1619 :			const uint32_t MotionFlags,
1620 :			const MBParam * const pParam,
1621 :			const VECTOR * const f_predMV,
1622 :			const VECTOR * const b_predMV,
1623 :			MACROBLOCK * const pMB,
1624 :	h	545	int32_t * const best_sad,
1625 :			SearchData * const fData)
1626 :	chl	530
1627 :	chl	96	{
1628 :
1629 :	chl	530	int iDirection, i, j;
1630 :	h	545	SearchData bData;
1631 :	chl	96
1632 :	syskin	712	fData->qpel_precision = 0;
1633 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1634 :			fData->iMinSAD = 4096256;
1635 :	syskin	739	bData.currentMV++; bData.currentQMV++;
1636 :	h	545	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1637 :	edgomez	195
1638 :	syskin	712	i = (x + y * fData->iEdgedWidth) * 16;
1639 :			bData.bRef = fData->Ref = f_Ref + i;
1640 :			bData.bRefH = fData->RefH = f_RefH + i;
1641 :			bData.bRefV = fData->RefV = f_RefV + i;
1642 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1643 :			bData.Ref = fData->bRef = b_Ref + i;
1644 :			bData.RefH = fData->bRefH = b_RefH + i;
1645 :			bData.RefV = fData->bRefV = b_RefV + i;
1646 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1647 :	chl	96
1648 :	h	545	bData.bpredMV = fData->predMV = *f_predMV;
1649 :			fData->bpredMV = bData.predMV = *b_predMV;
1650 :	syskin	661	fData->currentMV[0] = fData->currentMV[2];
1651 :	chl	96
1652 :	syskin	712	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0);
1653 :			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);
1654 :
1655 :	h	545	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1656 :	syskin	675	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1657 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1658 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1659 :	Isibaar	539
1660 :	h	545	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1661 :	syskin	675	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1662 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1663 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1664 :	Isibaar	539
1665 :	h	545	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1666 :	chl	96
1667 :	chl	530	//diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1668 :	chl	96
1669 :	chl	530	do {
1670 :			iDirection = 255;
1671 :			// forward MV moves
1672 :	h	545	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1673 :	chl	96
1674 :	h	545	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1675 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1676 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1677 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1678 :	chl	96
1679 :	chl	530	// backward MV moves
1680 :	h	545	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1681 :			fData->currentMV[2] = fData->currentMV[0];
1682 :	Isibaar	539	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1683 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1684 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1685 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1686 :	chl	96
1687 :	chl	530	} while (!(iDirection));
1688 :	edgomez	195
1689 :	syskin	658	if (fData->qpel) {
1690 :	syskin	740	if (fData->iMinSAD > best_sad + 500) return;
1691 :	syskin	675	CheckCandidate = CheckCandidateInt;
1692 :	syskin	663	fData->qpel_precision = bData.qpel_precision = 1;
1693 :	syskin	712	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0);
1694 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1695 :	syskin	658	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1696 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1697 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1698 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1699 :	syskin	663	SubpelRefine(fData);
1700 :	syskin	740	if (fData->iMinSAD > best_sad + 300) return;
1701 :	syskin	658	fData->currentQMV[2] = fData->currentQMV[0];
1702 :	syskin	663	SubpelRefine(&bData);
1703 :	syskin	658	}
1704 :	syskin	676
1705 :	syskin	740	fData->iMinSAD += (2+3) fData->lambda16; // two bits are needed to code interpolate mode.
1706 :	syskin	658
1707 :	h	545	if (fData->iMinSAD < best_sad) {
1708 :			best_sad = fData->iMinSAD;
1709 :			pMB->mvs[0] = fData->currentMV[0];
1710 :			pMB->b_mvs[0] = fData->currentMV[1];
1711 :	chl	530	pMB->mode = MODE_INTERPOLATE;
1712 :	syskin	658	if (fData->qpel) {
1713 :			pMB->qmvs[0] = fData->currentQMV[0];
1714 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1715 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1716 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1717 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1718 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1719 :			} else {
1720 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1721 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1722 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1723 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1724 :			}
1725 :	chl	530	}
1726 :			}
1727 :	chl	96
1728 :	chl	530	void
1729 :			MotionEstimationBVOP(MBParam * const pParam,
1730 :			FRAMEINFO * const frame,
1731 :			const int32_t time_bp,
1732 :			const int32_t time_pp,
1733 :			// forward (past) reference
1734 :			const MACROBLOCK * const f_mbs,
1735 :			const IMAGE * const f_ref,
1736 :			const IMAGE * const f_refH,
1737 :			const IMAGE * const f_refV,
1738 :			const IMAGE * const f_refHV,
1739 :			// backward (future) reference
1740 :	syskin	644	const FRAMEINFO * const b_reference,
1741 :	chl	530	const IMAGE * const b_ref,
1742 :			const IMAGE * const b_refH,
1743 :			const IMAGE * const b_refV,
1744 :			const IMAGE * const b_refHV)
1745 :			{
1746 :			uint32_t i, j;
1747 :	syskin	712	int32_t best_sad;
1748 :			uint32_t skip_sad;
1749 :	chl	530	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1750 :			static const VECTOR zeroMV={0,0};
1751 :	syskin	644	const MACROBLOCK * const b_mbs = b_reference->mbs;
1752 :	chl	96
1753 :	chl	530	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1754 :	chl	96
1755 :	chl	530	const int32_t TRB = time_pp - time_bp;
1756 :			const int32_t TRD = time_pp;
1757 :	chl	96
1758 :	h	545	// some pre-inintialized data for the rest of the search
1759 :
1760 :			SearchData Data;
1761 :			int32_t iMinSAD;
1762 :			VECTOR currentMV[3];
1763 :	syskin	658	VECTOR currentQMV[3];
1764 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
1765 :	h	545	Data.iEdgedWidth = pParam->edged_width;
1766 :	syskin	658	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1767 :	h	545	Data.iMinSAD = &iMinSAD;
1768 :	syskin	712	Data.lambda16 = lambda_vec16[frame->quant];
1769 :			Data.chroma = frame->quant;
1770 :	syskin	658	Data.qpel = pParam->m_quarterpel;
1771 :	syskin	663	Data.rounding = 0;
1772 :	h	545
1773 :	syskin	754	Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose)
1774 :	chl	530	// note: i==horizontal, j==vertical
1775 :			for (j = 0; j < pParam->mb_height; j++) {
1776 :	chl	96
1777 :	chl	530	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1778 :	edgomez	195
1779 :	chl	530	for (i = 0; i < pParam->mb_width; i++) {
1780 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1781 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1782 :	edgomez	195
1783 :	chl	619	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1784 :	syskin	644	if (b_reference->coding_type != S_VOP)
1785 :			if (b_mb->mode == MODE_NOT_CODED) {
1786 :			pMB->mode = MODE_NOT_CODED;
1787 :			continue;
1788 :			}
1789 :	chl	96
1790 :	h	545	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1791 :	syskin	601	pMB->quant = frame->quant;
1792 :	syskin	628
1793 :	chl	530	/* direct search comes first, because it (1) checks for SKIP-mode
1794 :			and (2) sets very good predictions for forward and backward search */
1795 :	Isibaar	539	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1796 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1797 :	chl	530	&frame->image,
1798 :			i, j,
1799 :			frame->motion_flags,
1800 :			TRB, TRD,
1801 :			pParam,
1802 :			pMB, b_mb,
1803 :	h	545	&best_sad,
1804 :			&Data);
1805 :	chl	96
1806 :	Isibaar	539	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1807 :	syskin	739
1808 :	chl	530	// forward search
1809 :			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1810 :			&frame->image, i, j,
1811 :			frame->motion_flags,
1812 :	h	545	frame->fcode, pParam,
1813 :	chl	530	pMB, &f_predMV, &best_sad,
1814 :	h	545	MODE_FORWARD, &Data);
1815 :	chl	96
1816 :	chl	530	// backward search
1817 :			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1818 :			&frame->image, i, j,
1819 :			frame->motion_flags,
1820 :	h	545	frame->bcode, pParam,
1821 :	chl	530	pMB, &b_predMV, &best_sad,
1822 :	h	545	MODE_BACKWARD, &Data);
1823 :	chl	96
1824 :	chl	530	// interpolate search comes last, because it uses data from forward and backward as prediction
1825 :			SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1826 :			b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1827 :			&frame->image,
1828 :			i, j,
1829 :			frame->fcode, frame->bcode,
1830 :			frame->motion_flags,
1831 :	h	545	pParam,
1832 :	chl	530	&f_predMV, &b_predMV,
1833 :	h	545	pMB, &best_sad,
1834 :			&Data);
1835 :	syskin	662
1836 :	syskin	712	// final skip decision
1837 :			if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2)
1838 :			&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1839 :			SkipDecisionB(&frame->image, f_ref, b_ref, pMB,frame->quant, i, j, &Data);
1840 :
1841 :	chl	530	switch (pMB->mode) {
1842 :			case MODE_FORWARD:
1843 :			f_count++;
1844 :	syskin	712	if (Data.qpel) f_predMV = pMB->qmvs[0];
1845 :	syskin	658	else f_predMV = pMB->mvs[0];
1846 :	chl	530	break;
1847 :			case MODE_BACKWARD:
1848 :			b_count++;
1849 :	syskin	712	if (Data.qpel) b_predMV = pMB->b_qmvs[0];
1850 :	syskin	658	else b_predMV = pMB->b_mvs[0];
1851 :	chl	530	break;
1852 :			case MODE_INTERPOLATE:
1853 :			i_count++;
1854 :	syskin	712	if (Data.qpel) {
1855 :	syskin	658	f_predMV = pMB->qmvs[0];
1856 :			b_predMV = pMB->b_qmvs[0];
1857 :	syskin	748	} else {
1858 :	syskin	658	f_predMV = pMB->mvs[0];
1859 :			b_predMV = pMB->b_mvs[0];
1860 :			}
1861 :	chl	530	break;
1862 :			case MODE_DIRECT:
1863 :			case MODE_DIRECT_NO4V:
1864 :			d_count++;
1865 :			default:
1866 :			break;
1867 :	chl	96	}
1868 :			}
1869 :			}
1870 :			}
1871 :
1872 :	syskin	739	static __inline void
1873 :	Isibaar	539	MEanalyzeMB ( const uint8_t * const pRef,
1874 :			const uint8_t * const pCur,
1875 :			const int x,
1876 :			const int y,
1877 :			const MBParam * const pParam,
1878 :	syskin	739	MACROBLOCK * const pMBs,
1879 :	h	545	SearchData * const Data)
1880 :	Isibaar	539	{
1881 :	chl	317
1882 :	syskin	739	int i, mask;
1883 :	h	545	VECTOR pmv[3];
1884 :	syskin	739	MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width];
1885 :
1886 :			for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1887 :	suxen_drol	118
1888 :	syskin	640	//median is only used as prediction. it doesn't have to be real
1889 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1890 :			else
1891 :			if (x == 1) //left macroblock does not have any vector now
1892 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1893 :	syskin	748	else if (y == 1) // top macroblock doesn't have it's vector
1894 :	syskin	640	Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1895 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1896 :
1897 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1898 :	syskin	739	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv);
1899 :	chl	326
1900 :	h	545	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1901 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1902 :	syskin	739
1903 :	Isibaar	539	pmv[1].x = EVEN(pMB->mvs[0].x);
1904 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1905 :	syskin	640	pmv[2].x = EVEN(Data->predMV.x);
1906 :			pmv[2].y = EVEN(Data->predMV.y);
1907 :			pmv[0].x = pmv[0].y = 0;
1908 :	chl	326
1909 :	syskin	739	CheckCandidate32I(0, 0, 255, &i, Data);
1910 :	syskin	640
1911 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) {
1912 :	syskin	640
1913 :	syskin	739	if (!(mask = make_mask(pmv, 1)))
1914 :			CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data);
1915 :			if (!(mask = make_mask(pmv, 2)))
1916 :			CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data);
1917 :	Isibaar	539
1918 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) // diamond only if needed
1919 :			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1920 :	Isibaar	539
1921 :	syskin	739	for (i = 0; i < 4; i++) {
1922 :			MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1) * pParam->mb_width)];
1923 :			MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
1924 :			MB->mode = MODE_INTER;
1925 :			MB->sad16 = Data->iMinSAD[i+1];
1926 :			}
1927 :			}
1928 :	Isibaar	539	}
1929 :
1930 :	syskin	739	#define INTRA_BIAS 2500
1931 :			#define INTRA_THRESH 1500
1932 :			#define INTER_THRESH 1400
1933 :	Isibaar	539
1934 :	syskin	644
1935 :	Isibaar	539	int
1936 :			MEanalysis( const IMAGE * const pRef,
1937 :	syskin	644	FRAMEINFO * const Current,
1938 :	Isibaar	539	MBParam * const pParam,
1939 :	syskin	644	int maxIntra, //maximum number if non-I frames
1940 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1941 :			int bCount) // number if B frames in a row
1942 :	Isibaar	539	{
1943 :			uint32_t x, y, intra = 0;
1944 :			int sSAD = 0;
1945 :	syskin	644	MACROBLOCK * const pMBs = Current->mbs;
1946 :			const IMAGE * const pCurrent = &Current->image;
1947 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1948 :	syskin	739	const VECTOR zeroMV = {0,0};
1949 :	h	545
1950 :	syskin	739	int32_t iMinSAD[5], temp[5];
1951 :			VECTOR currentMV[5];
1952 :	h	545	SearchData Data;
1953 :			Data.iEdgedWidth = pParam->edged_width;
1954 :	syskin	739	Data.currentMV = currentMV;
1955 :			Data.iMinSAD = iMinSAD;
1956 :	syskin	644	Data.iFcode = Current->fcode;
1957 :	syskin	739	Data.rrv = Current->global_flags & XVID_REDUCED;
1958 :			Data.temp = temp;
1959 :			CheckCandidate = CheckCandidate32I;
1960 :	h	545
1961 :	syskin	658	if (intraCount < 10) // we're right after an I frame
1962 :			IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
1963 :	syskin	644	else
1964 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
1965 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
1966 :
1967 :	syskin	672	InterThresh += 400 * (1 - bCount);
1968 :	syskin	739	if (InterThresh < 300) InterThresh = 300;
1969 :	syskin	644
1970 :	Isibaar	539	if (sadInit) (*sadInit) ();
1971 :
1972 :	syskin	739	for (y = 1; y < pParam->mb_height-1; y+=2) {
1973 :			for (x = 1; x < pParam->mb_width-1; x+=2) {
1974 :			int i;
1975 :	Isibaar	539
1976 :	syskin	739	if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
1977 :
1978 :			MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
1979 :	Isibaar	539
1980 :	syskin	739	for (i = 0; i < 4; i++) {
1981 :			int dev;
1982 :			MACROBLOCK pMB = &pMBs[x+(i&1) + y+(i>>1) pParam->mb_width];
1983 :			if (pMB->sad16 > IntraThresh) {
1984 :			dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1))* pParam->edged_width) * 16,
1985 :			pParam->edged_width);
1986 :			if (dev + IntraThresh < pMB->sad16) {
1987 :			pMB->mode = MODE_INTRA;
1988 :			if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
1989 :			}
1990 :	Isibaar	539	}
1991 :	syskin	739	sSAD += pMB->sad16;
1992 :	syskin	644	}
1993 :	suxen_drol	118	}
1994 :			}
1995 :	syskin	704	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
1996 :	syskin	739	if (sSAD > IntraThresh + INTRA_BIAS ) return I_VOP;
1997 :	suxen_drol	709	if (sSAD > InterThresh ) return P_VOP;
1998 :	Isibaar	539	emms();
1999 :	suxen_drol	709	return B_VOP;
2000 :	syskin	712
2001 :	suxen_drol	118	}
2002 :	syskin	576
2003 :	syskin	628	static void
2004 :	syskin	748	CheckGMC(int x, int y, const int dir, int * iDirection,
2005 :	syskin	628	const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
2006 :			const MBParam * const pParam)
2007 :			{
2008 :			uint32_t mx, my, a, count = 0;
2009 :	chl	619
2010 :	syskin	628	for (my = 1; my < pParam->mb_height-1; my++)
2011 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2012 :			VECTOR mv;
2013 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2014 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED) continue;
2015 :			mv = pMB->mvs[0];
2016 :			a = ABS(mv.x - x) + ABS(mv.y - y);
2017 :			if (a < 6) count += 6 - a;
2018 :			}
2019 :	chl	619
2020 :	syskin	628	if (count > *bestcount) {
2021 :			*bestcount = count;
2022 :			*iDirection = dir;
2023 :	syskin	748	GMC->x = x; GMC->y = y;
2024 :	syskin	628	}
2025 :			}
2026 :
2027 :
2028 :	suxen_drol	721	static VECTOR
2029 :	chl	619	GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2030 :			{
2031 :
2032 :	syskin	628	uint32_t count, bestcount = 0;
2033 :	chl	619	int x, y;
2034 :	syskin	628	VECTOR gmc = {0,0};
2035 :	chl	619	int step, min_x, max_x, min_y, max_y;
2036 :			uint32_t mx, my;
2037 :	syskin	628	int iDirection, bDirection;
2038 :	chl	619
2039 :			min_x = min_y = -32<<iFcode;
2040 :			max_x = max_y = 32<<iFcode;
2041 :
2042 :	syskin	628	//step1: let's find a rough camera panning
2043 :	chl	619	for (step = 32; step >= 2; step /= 2) {
2044 :			bestcount = 0;
2045 :			for (y = min_y; y <= max_y; y += step)
2046 :			for (x = min_x ; x <= max_x; x += step) {
2047 :			count = 0;
2048 :			//for all macroblocks
2049 :			for (my = 1; my < pParam->mb_height-1; my++)
2050 :			for (mx = 1; mx < pParam->mb_width-1; mx++) {
2051 :			const MACROBLOCK pMB = &pMBs[mx + my pParam->mb_width];
2052 :			VECTOR mv;
2053 :
2054 :	syskin	748	if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2055 :	chl	619	continue;
2056 :
2057 :			mv = pMB->mvs[0];
2058 :			if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2059 :			count++;
2060 :			}
2061 :			if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2062 :			}
2063 :			min_x = gmc.x - step;
2064 :			max_x = gmc.x + step;
2065 :			min_y = gmc.y - step;
2066 :			max_y = gmc.y + step;
2067 :			}
2068 :	syskin	628
2069 :	syskin	748	if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2070 :	syskin	628	gmc.x = gmc.y = 0; //no camara pan, no GMC
2071 :
2072 :	syskin	748	// step2: let's refine camera panning using gradiend-descent approach
2073 :	syskin	628	// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
2074 :			bestcount = 0;
2075 :			CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
2076 :			do {
2077 :			x = gmc.x; y = gmc.y;
2078 :			bDirection = iDirection; iDirection = 0;
2079 :			if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2080 :			if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2081 :			if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
2082 :			if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2083 :
2084 :			} while (iDirection);
2085 :
2086 :	chl	619	if (pParam->m_quarterpel) {
2087 :			gmc.x *= 2;
2088 :			gmc.y = 2; / we store the halfpel value as pseudo-qpel to make comparison easier */
2089 :			}
2090 :
2091 :			return gmc;
2092 :			}

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