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

Revision 801 - (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 :	syskin	801	#include <math.h> // lrint
36 :	Isibaar	3
37 :			#include "../encoder.h"
38 :			#include "../utils/mbfunctions.h"
39 :			#include "../prediction/mbprediction.h"
40 :			#include "../global.h"
41 :			#include "../utils/timer.h"
42 :	Isibaar	580	#include "../image/interpolate8x8.h"
43 :	chl	530	#include "motion_est.h"
44 :	suxen_drol	118	#include "motion.h"
45 :	Isibaar	3	#include "sad.h"
46 :	Isibaar	539	#include "../utils/emms.h"
47 :	Isibaar	3
48 :	chl	530	#define INITIAL_SKIP_THRESH (10)
49 :			#define FINAL_SKIP_THRESH (50)
50 :			#define MAX_SAD00_FOR_SKIP (20)
51 :	Isibaar	539	#define MAX_CHROMA_SAD_FOR_SKIP (22)
52 :	Isibaar	3
53 :	chl	530	#define CHECK_CANDIDATE(X,Y,D) { \
54 :			(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
55 :	Isibaar	3
56 :	syskin	768	static __inline uint32_t
57 :	syskin	724	d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
58 :	Isibaar	3	{
59 :	chl	530	int xb, yb;
60 :	syskin	768	x += x * qpel; y += y * qpel;
61 :			if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
62 :	syskin	748	x -= pred.x;
63 :			y -= pred.y;
64 :	syskin	768
65 :	syskin	748	if (x) {
66 :	syskin	768	x = ABS(x);
67 :	chl	530	x += (1 << (iFcode - 1)) - 1;
68 :			x >>= (iFcode - 1);
69 :			if (x > 32) x = 32;
70 :			xb = mvtab[x] + iFcode;
71 :	syskin	748	} else xb = 1;
72 :	Isibaar	3
73 :	syskin	748	if (y) {
74 :	syskin	768	y = ABS(y);
75 :	chl	530	y += (1 << (iFcode - 1)) - 1;
76 :			y >>= (iFcode - 1);
77 :			if (y > 32) y = 32;
78 :			yb = mvtab[y] + iFcode;
79 :	syskin	748	} else yb = 1;
80 :	chl	530	return xb + yb;
81 :	Isibaar	3	}
82 :
83 :	syskin	801	static int32_t ChromaSAD2(int fx, int fy, int bx, int by, const SearchData * const data)
84 :			{
85 :			int sad;
86 :			const uint32_t stride = data->iEdgedWidth/2;
87 :			uint8_t * f_refu = data->RefQ,
88 :			* f_refv = data->RefQ + 8,
89 :			* b_refu = data->RefQ + 16,
90 :			* b_refv = data->RefQ + 24;
91 :
92 :			switch (((fx & 1) << 1) \| (fy & 1)) {
93 :			case 0:
94 :			fx = fx / 2; fy = fy / 2;
95 :			f_refu = (uint8_t)data->RefCU + fy stride + fx, stride;
96 :			f_refv = (uint8_t)data->RefCV + fy stride + fx, stride;
97 :			break;
98 :			case 1:
99 :			fx = fx / 2; fy = (fy - 1) / 2;
100 :			interpolate8x8_halfpel_v(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
101 :			interpolate8x8_halfpel_v(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
102 :			break;
103 :			case 2:
104 :			fx = (fx - 1) / 2; fy = fy / 2;
105 :			interpolate8x8_halfpel_h(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
106 :			interpolate8x8_halfpel_h(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
107 :			break;
108 :			default:
109 :			fx = (fx - 1) / 2; fy = (fy - 1) / 2;
110 :			interpolate8x8_halfpel_hv(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
111 :			interpolate8x8_halfpel_hv(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
112 :			break;
113 :			}
114 :
115 :			switch (((bx & 1) << 1) \| (by & 1)) {
116 :			case 0:
117 :			bx = bx / 2; by = by / 2;
118 :			b_refu = (uint8_t)data->b_RefCU + by stride + bx, stride;
119 :			b_refv = (uint8_t)data->b_RefCV + by stride + bx, stride;
120 :			break;
121 :			case 1:
122 :			bx = bx / 2; by = (by - 1) / 2;
123 :			interpolate8x8_halfpel_v(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
124 :			interpolate8x8_halfpel_v(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
125 :			break;
126 :			case 2:
127 :			bx = (bx - 1) / 2; by = by / 2;
128 :			interpolate8x8_halfpel_h(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
129 :			interpolate8x8_halfpel_h(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
130 :			break;
131 :			default:
132 :			bx = (bx - 1) / 2; by = (by - 1) / 2;
133 :			interpolate8x8_halfpel_hv(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
134 :			interpolate8x8_halfpel_hv(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
135 :			break;
136 :			}
137 :
138 :			sad = sad8bi(data->CurU, b_refu, f_refu, stride);
139 :			sad += sad8bi(data->CurV, b_refv, f_refv, stride);
140 :
141 :			return sad;
142 :			}
143 :
144 :
145 :	syskin	748	static int32_t
146 :	syskin	628	ChromaSAD(int dx, int dy, const SearchData * const data)
147 :	Isibaar	3	{
148 :	syskin	628	int sad;
149 :	syskin	748	const uint32_t stride = data->iEdgedWidth/2;
150 :	Isibaar	3
151 :	syskin	712	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
152 :	syskin	768	data->temp[5] = dx; data->temp[6] = dy; // backup
153 :	syskin	712
154 :			switch (((dx & 1) << 1) \| (dy & 1)) {
155 :	syskin	628	case 0:
156 :	syskin	748	dx = dx / 2; dy = dy / 2;
157 :			sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride);
158 :			sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride);
159 :	syskin	628	break;
160 :			case 1:
161 :			dx = dx / 2; dy = (dy - 1) / 2;
162 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride);
163 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride);
164 :	syskin	628	break;
165 :			case 2:
166 :			dx = (dx - 1) / 2; dy = dy / 2;
167 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride);
168 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride);
169 :	syskin	628	break;
170 :			default:
171 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
172 :	syskin	748	interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding);
173 :			sad = sad8(data->CurU, data->RefQ, stride);
174 :
175 :			interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding);
176 :			sad += sad8(data->CurV, data->RefQ, stride);
177 :	syskin	628	break;
178 :	chl	530	}
179 :	syskin	748	data->temp[7] = sad; //backup, part 2
180 :	syskin	628	return sad;
181 :			}
182 :	Isibaar	3
183 :	syskin	663	static __inline const uint8_t *
184 :	syskin	768	GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data)
185 :	Isibaar	606	{
186 :	syskin	663	// dir : 0 = forward, 1 = backward
187 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
188 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
189 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
190 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
191 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
192 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
193 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
194 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
195 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
196 :	Isibaar	606	}
197 :			}
198 :
199 :	syskin	748	// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate
200 :			static __inline const uint8_t *
201 :			GetReference(const int x, const int y, const SearchData * const data)
202 :			{
203 :			switch ( ((x&1)<<1) \| (y&1) ) {
204 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
205 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
206 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
207 :			default : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
208 :			}
209 :			}
210 :
211 :			static uint8_t *
212 :	syskin	768	Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data)
213 :	chl	530	{
214 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
215 :	syskin	768	uint8_t * Reference = data->RefQ + 16*dir;
216 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
217 :	syskin	663	const uint32_t rounding = data->rounding;
218 :			const int halfpel_x = x/2;
219 :			const int halfpel_y = y/2;
220 :			const uint8_t ref1, ref2, ref3, ref4;
221 :	chl	326
222 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
223 :	syskin	663	ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
224 :			switch( ((x&1)<<1) + (y&1) ) {
225 :			case 0: // pure halfpel position
226 :	syskin	740	return (uint8_t *) ref1;
227 :	syskin	663	break;
228 :	Isibaar	3
229 :	syskin	663	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
230 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
231 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
232 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
233 :	syskin	663	break;
234 :	Isibaar	3
235 :	syskin	663	case 2: // x qpel, y halfpel - left or right during qpel refinement
236 :	syskin	748	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
237 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
238 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
239 :	syskin	663	break;
240 :	Isibaar	3
241 :	syskin	663	default: // x and y in qpel resolution - the "corners" (top left/right and
242 :			// bottom left/right) during qpel refinement
243 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
244 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
245 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
246 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
247 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
248 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
249 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
250 :			break;
251 :			}
252 :			return Reference;
253 :	Isibaar	3	}
254 :
255 :	syskin	748	static uint8_t *
256 :	syskin	768	Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data)
257 :	Isibaar	580	{
258 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
259 :	syskin	768	uint8_t * Reference = data->RefQ + 16*dir;
260 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
261 :	syskin	663	const uint32_t rounding = data->rounding;
262 :			const int halfpel_x = x/2;
263 :			const int halfpel_y = y/2;
264 :	Isibaar	580	const uint8_t ref1, ref2, ref3, ref4;
265 :
266 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
267 :	syskin	663	switch( ((x&1)<<1) + (y&1) ) {
268 :			case 0: // pure halfpel position
269 :	syskin	740	return (uint8_t *) ref1;
270 :	Isibaar	580	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
271 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
272 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
273 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
274 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
275 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
276 :	Isibaar	580	break;
277 :
278 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
279 :	syskin	768	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
280 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
281 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
282 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
283 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
284 :	Isibaar	580	break;
285 :
286 :			default: // x and y in qpel resolution - the "corners" (top left/right and
287 :			// bottom left/right) during qpel refinement
288 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
289 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
290 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
291 :	Isibaar	580	interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
292 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
293 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
294 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
295 :			break;
296 :			}
297 :	syskin	663	return Reference;
298 :			}
299 :
300 :			/* CHECK_CANDIATE FUNCTIONS START */
301 :
302 :	syskin	748	static void
303 :	syskin	663	CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
304 :			{
305 :			int t, xc, yc;
306 :			const uint8_t * Reference;
307 :			VECTOR * current;
308 :
309 :	syskin	768	if ( (x > data->max_dx) \| (x < data->min_dx)
310 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
311 :	syskin	663
312 :			if (data->qpel_precision) { // x and y are in 1/4 precision
313 :			Reference = Interpolate16x16qpel(x, y, 0, data);
314 :			xc = x/2; yc = y/2; //for chroma sad
315 :			current = data->currentQMV;
316 :			} else {
317 :	syskin	748	Reference = GetReference(x, y, data);
318 :	syskin	663	current = data->currentMV;
319 :			xc = x; yc = y;
320 :			}
321 :	syskin	768	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
322 :
323 :	syskin	663	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
324 :	Isibaar	580
325 :	syskin	768	data->temp[0] += (data->lambda16 * t * data->temp[0])>>10;
326 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
327 :	Isibaar	580
328 :	syskin	753	if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
329 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
330 :	syskin	628
331 :	Isibaar	580	if (data->temp[0] < data->iMinSAD[0]) {
332 :			data->iMinSAD[0] = data->temp[0];
333 :	syskin	663	current[0].x = x; current[0].y = y;
334 :			*dir = Direction; }
335 :	Isibaar	580
336 :			if (data->temp[1] < data->iMinSAD[1]) {
337 :	syskin	663	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; }
338 :	Isibaar	580	if (data->temp[2] < data->iMinSAD[2]) {
339 :	syskin	663	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
340 :	Isibaar	580	if (data->temp[3] < data->iMinSAD[3]) {
341 :	syskin	663	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
342 :	Isibaar	580	if (data->temp[4] < data->iMinSAD[4]) {
343 :	syskin	663	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
344 :
345 :	Isibaar	580	}
346 :
347 :	syskin	748	static void
348 :	syskin	704	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
349 :			{
350 :	syskin	768	uint32_t t;
351 :	syskin	704	const uint8_t * Reference;
352 :
353 :	syskin	768	if ( (!(x&1) && x !=0) \| (!(y&1) && y !=0) \|\| //non-zero integer value
354 :			(x > data->max_dx) \| (x < data->min_dx)
355 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
356 :	syskin	704
357 :	syskin	748	Reference = GetReference(x, y, data);
358 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
359 :	syskin	704
360 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
361 :
362 :	syskin	768	data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10;
363 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
364 :	syskin	704
365 :			if (data->temp[0] < data->iMinSAD[0]) {
366 :			data->iMinSAD[0] = data->temp[0];
367 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
368 :			*dir = Direction; }
369 :
370 :			if (data->temp[1] < data->iMinSAD[1]) {
371 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
372 :			if (data->temp[2] < data->iMinSAD[2]) {
373 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
374 :			if (data->temp[3] < data->iMinSAD[3]) {
375 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
376 :			if (data->temp[4] < data->iMinSAD[4]) {
377 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
378 :			}
379 :
380 :	syskin	748	static void
381 :	syskin	663	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
382 :	syskin	658	{
383 :	syskin	801	int32_t sad, xc, yc;
384 :	syskin	663	const uint8_t * Reference;
385 :	syskin	768	uint32_t t;
386 :	syskin	663	VECTOR * current;
387 :	syskin	658
388 :	syskin	768	if ( (x > data->max_dx) \| ( x < data->min_dx)
389 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
390 :	syskin	658
391 :	syskin	768	if (data->rrv && (!(x&1) && x !=0) \| (!(y&1) && y !=0) ) return; //non-zero even value
392 :	syskin	704
393 :	syskin	663	if (data->qpel_precision) { // x and y are in 1/4 precision
394 :			Reference = Interpolate16x16qpel(x, y, 0, data);
395 :			current = data->currentQMV;
396 :	syskin	801	xc = x/2; yc = y/2;
397 :	syskin	663	} else {
398 :	syskin	748	Reference = GetReference(x, y, data);
399 :	syskin	663	current = data->currentMV;
400 :	syskin	801	xc = x; yc = y;
401 :	syskin	658	}
402 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode,
403 :	syskin	768	data->qpel^data->qpel_precision, data->rrv);
404 :
405 :	syskin	658	sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
406 :	syskin	768	sad += (data->lambda16 * t * sad)>>10;
407 :	syskin	658
408 :	syskin	801	if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
409 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
410 :
411 :
412 :	syskin	663	if (sad < *(data->iMinSAD)) {
413 :			*(data->iMinSAD) = sad;
414 :			current->x = x; current->y = y;
415 :	syskin	801	*dir = Direction;
416 :			}
417 :	syskin	658	}
418 :
419 :	syskin	748	static void
420 :	syskin	739	CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
421 :	Isibaar	539	{
422 :	syskin	658	// maximum speed - for P/B/I decision
423 :	Isibaar	539
424 :	syskin	768	if ( (x > data->max_dx) \| (x < data->min_dx)
425 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
426 :	Isibaar	539
427 :	syskin	739	data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
428 :			data->iEdgedWidth, data->temp+1);
429 :	syskin	768
430 :	syskin	739	if (data->temp[0] < *(data->iMinSAD)) {
431 :			*(data->iMinSAD) = data->temp[0];
432 :	Isibaar	539	data->currentMV[0].x = x; data->currentMV[0].y = y;
433 :			*dir = Direction; }
434 :	syskin	739	if (data->temp[1] < data->iMinSAD[1]) {
435 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
436 :			if (data->temp[2] < data->iMinSAD[2]) {
437 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
438 :			if (data->temp[3] < data->iMinSAD[3]) {
439 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
440 :			if (data->temp[4] < data->iMinSAD[4]) {
441 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
442 :
443 :	Isibaar	539	}
444 :
445 :	syskin	748	static void
446 :	chl	530	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
447 :			{
448 :	syskin	801	int32_t sad, xb, yb, xcf, ycf, xcb, ycb;
449 :	syskin	768	uint32_t t;
450 :	chl	530	const uint8_t ReferenceF, ReferenceB;
451 :	syskin	663	VECTOR *current;
452 :	Isibaar	3
453 :	syskin	768	if ( (xf > data->max_dx) \| (xf < data->min_dx)
454 :			\| (yf > data->max_dy) \| (yf < data->min_dy) ) return;
455 :	Isibaar	3
456 :	syskin	768	if (!data->qpel_precision) {
457 :	syskin	748	ReferenceF = GetReference(xf, yf, data);
458 :	syskin	663	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
459 :	syskin	748	ReferenceB = GetReferenceB(xb, yb, 1, data);
460 :	syskin	663	current = data->currentMV;
461 :	syskin	801	xcf = xf; ycf = yf;
462 :			xcb = xb; ycb = yb;
463 :	syskin	768	} else {
464 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
465 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
466 :			current = data->currentQMV;
467 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
468 :	syskin	801	xcf = xf/2; ycf = yf/2;
469 :			xcb = xb/2; ycb = yb/2;
470 :	chl	530	}
471 :	edgomez	195
472 :	syskin	768	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0)
473 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0);
474 :
475 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
476 :	syskin	768	sad += (data->lambda16 * t * sad)>>10;
477 :	syskin	628
478 :	syskin	801	if (data->chroma) sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3],
479 :			(ycf >> 1) + roundtab_79[ycf & 0x3],
480 :			(xcb >> 1) + roundtab_79[xcb & 0x3],
481 :			(ycb >> 1) + roundtab_79[ycb & 0x3], data);
482 :
483 :	chl	530	if (sad < *(data->iMinSAD)) {
484 :			*(data->iMinSAD) = sad;
485 :	syskin	663	current->x = xf; current->y = yf;
486 :	syskin	801	*dir = Direction;
487 :			}
488 :	Isibaar	3	}
489 :
490 :	syskin	748	static void
491 :	chl	530	CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
492 :	Isibaar	3	{
493 :	syskin	801	int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0;
494 :	syskin	768	uint32_t k;
495 :	chl	530	const uint8_t *ReferenceF;
496 :			const uint8_t *ReferenceB;
497 :			VECTOR mvs, b_mvs;
498 :	Isibaar	3
499 :	syskin	768	if (( x > 31) \| ( x < -32) \| ( y > 31) \| (y < -32)) return;
500 :	Isibaar	3
501 :	chl	530	for (k = 0; k < 4; k++) {
502 :			mvs.x = data->directmvF[k].x + x;
503 :			b_mvs.x = ((x == 0) ?
504 :			data->directmvB[k].x
505 :			: mvs.x - data->referencemv[k].x);
506 :	Isibaar	3
507 :	chl	530	mvs.y = data->directmvF[k].y + y;
508 :			b_mvs.y = ((y == 0) ?
509 :			data->directmvB[k].y
510 :			: mvs.y - data->referencemv[k].y);
511 :	syskin	768
512 :	syskin	801	if ( (mvs.x > data->max_dx) \| (mvs.x < data->min_dx)
513 :			\| (mvs.y > data->max_dy) \| (mvs.y < data->min_dy)
514 :			\| (b_mvs.x > data->max_dx) \| (b_mvs.x < data->min_dx)
515 :			\| (b_mvs.y > data->max_dy) \| (b_mvs.y < data->min_dy) ) return;
516 :	Isibaar	3
517 :	syskin	801	if (data->qpel) {
518 :			xcf += mvs.x/2; ycf += mvs.y/2;
519 :			xcb += b_mvs.x/2; ycb += b_mvs.y/2;
520 :			} else {
521 :			xcf += mvs.x; ycf += mvs.y;
522 :			xcb += b_mvs.x; ycb += b_mvs.y;
523 :			mvs.x = 2; mvs.y = 2; //we move to qpel precision anyway
524 :			b_mvs.x = 2; b_mvs.y = 2;
525 :			}
526 :
527 :	syskin	663	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
528 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
529 :	syskin	768
530 :	chl	530	sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
531 :	syskin	768	ReferenceF, ReferenceB, data->iEdgedWidth);
532 :	chl	530	if (sad > *(data->iMinSAD)) return;
533 :	edgomez	78	}
534 :	chl	530
535 :	syskin	768	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
536 :	syskin	628
537 :	syskin	801	if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
538 :			(ycf >> 3) + roundtab_76[ycf & 0xf],
539 :			(xcb >> 3) + roundtab_76[xcb & 0xf],
540 :			(ycb >> 3) + roundtab_76[ycb & 0xf], data);
541 :
542 :	chl	530	if (sad < *(data->iMinSAD)) {
543 :			*(data->iMinSAD) = sad;
544 :			data->currentMV->x = x; data->currentMV->y = y;
545 :	syskin	801	*dir = Direction;
546 :			}
547 :	Isibaar	3	}
548 :
549 :	syskin	748	static void
550 :	chl	530	CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
551 :	chl	96	{
552 :	syskin	801	int32_t sad, xcf, ycf, xcb, ycb;
553 :	chl	530	const uint8_t *ReferenceF;
554 :			const uint8_t *ReferenceB;
555 :			VECTOR mvs, b_mvs;
556 :	chl	96
557 :	syskin	768	if (( x > 31) \| ( x < -32) \| ( y > 31) \| (y < -32)) return;
558 :	edgomez	195
559 :	chl	530	mvs.x = data->directmvF[0].x + x;
560 :			b_mvs.x = ((x == 0) ?
561 :			data->directmvB[0].x
562 :			: mvs.x - data->referencemv[0].x);
563 :	edgomez	195
564 :	chl	530	mvs.y = data->directmvF[0].y + y;
565 :			b_mvs.y = ((y == 0) ?
566 :			data->directmvB[0].y
567 :			: mvs.y - data->referencemv[0].y);
568 :	syskin	768
569 :	syskin	801	if ( (mvs.x > data->max_dx) \| (mvs.x < data->min_dx)
570 :			\| (mvs.y > data->max_dy) \| (mvs.y < data->min_dy)
571 :			\| (b_mvs.x > data->max_dx) \| (b_mvs.x < data->min_dx)
572 :			\| (b_mvs.y > data->max_dy) \| (b_mvs.y < data->min_dy) ) return;
573 :	chl	96
574 :	syskin	801	if (data->qpel) {
575 :			xcf = 4(mvs.x/2); ycf = 4(mvs.y/2);
576 :			xcb = 4(b_mvs.x/2); ycb = 4(b_mvs.y/2);
577 :			ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
578 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
579 :			} else {
580 :			xcf = 4mvs.x; ycf = 4mvs.y;
581 :			xcb = 4b_mvs.x; ycb = 4b_mvs.y;
582 :			ReferenceF = GetReference(mvs.x, mvs.y, data);
583 :			ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data);
584 :			}
585 :	syskin	768
586 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
587 :	syskin	768	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
588 :	chl	96
589 :	syskin	801	if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
590 :			(ycf >> 3) + roundtab_76[ycf & 0xf],
591 :			(xcb >> 3) + roundtab_76[xcb & 0xf],
592 :			(ycb >> 3) + roundtab_76[ycb & 0xf], data);
593 :
594 :	chl	530	if (sad < *(data->iMinSAD)) {
595 :			*(data->iMinSAD) = sad;
596 :			data->currentMV->x = x; data->currentMV->y = y;
597 :	syskin	801	*dir = Direction;
598 :			}
599 :	chl	530	}
600 :	chl	96
601 :	chl	530	static void
602 :			CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
603 :			{
604 :	syskin	768	int32_t sad; uint32_t t;
605 :	chl	530	const uint8_t * Reference;
606 :	edgomez	195
607 :	syskin	768	if ( (x > data->max_dx) \| (x < data->min_dx)
608 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
609 :	chl	96
610 :	syskin	663	if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data);
611 :	syskin	768	else Reference = GetReference(x, y, data);
612 :	edgomez	195
613 :	chl	530	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
614 :	syskin	801	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
615 :	edgomez	195
616 :	syskin	768	sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10;
617 :	chl	96
618 :	Isibaar	606	if (sad < *(data->iMinSAD)) {
619 :			*(data->iMinSAD) = sad;
620 :			data->currentMV->x = x; data->currentMV->y = y;
621 :			*dir = Direction; }
622 :			}
623 :
624 :	Isibaar	539	/* CHECK_CANDIATE FUNCTIONS END */
625 :	chl	96
626 :	chl	530	/* MAINSEARCH FUNCTIONS START */
627 :	edgomez	195
628 :	chl	530	static void
629 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
630 :	chl	181	{
631 :
632 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
633 :
634 :	syskin	712	int iDirection;
635 :	edgomez	195
636 :	syskin	712	for(;;) { //forever
637 :			iDirection = 0;
638 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
639 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
640 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
641 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
642 :	chl	181
643 :	syskin	712	/* now we're doing diagonal checks near our candidate */
644 :	chl	181
645 :	syskin	768	if (iDirection) { //if anything found
646 :	syskin	712	bDirection = iDirection;
647 :			iDirection = 0;
648 :			x = data->currentMV->x; y = data->currentMV->y;
649 :			if (bDirection & 3) { //our candidate is left or right
650 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
651 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
652 :			} else { // what remains here is up or down
653 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
654 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
655 :	syskin	712	}
656 :	chl	181
657 :	syskin	712	if (iDirection) {
658 :			bDirection += iDirection;
659 :	syskin	748	x = data->currentMV->x; y = data->currentMV->y;
660 :	chl	181	}
661 :	syskin	712	} else { //about to quit, eh? not so fast....
662 :			switch (bDirection) {
663 :			case 2:
664 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
665 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
666 :			break;
667 :			case 1:
668 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
669 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
670 :			break;
671 :			case 2 + 4:
672 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
673 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
674 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
675 :			break;
676 :			case 4:
677 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
678 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
679 :			break;
680 :			case 8:
681 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
682 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
683 :			break;
684 :			case 1 + 4:
685 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
686 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
687 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
688 :			break;
689 :			case 2 + 8:
690 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
691 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
692 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
693 :			break;
694 :			case 1 + 8:
695 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
696 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
697 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
698 :			break;
699 :			default: //1+2+4+8 == we didn't find anything at all
700 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
701 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
702 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
703 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
704 :			break;
705 :			}
706 :			if (!iDirection) break; //ok, the end. really
707 :			bDirection = iDirection;
708 :			x = data->currentMV->x; y = data->currentMV->y;
709 :	chl	181	}
710 :	syskin	712	}
711 :	chl	181	}
712 :
713 :	chl	530	static void
714 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
715 :	chl	326	{
716 :	chl	530	int iDirection;
717 :	chl	326
718 :	chl	530	do {
719 :			iDirection = 0;
720 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
721 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
722 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
723 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
724 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
725 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
726 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
727 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
728 :	chl	326
729 :	chl	530	bDirection = iDirection;
730 :			x = data->currentMV->x; y = data->currentMV->y;
731 :			} while (iDirection);
732 :	chl	346	}
733 :
734 :	chl	530	static void
735 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
736 :	chl	346	{
737 :
738 :	chl	181	/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
739 :
740 :	syskin	712	int iDirection;
741 :	edgomez	195
742 :	syskin	712	do {
743 :			iDirection = 0;
744 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
745 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
746 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
747 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
748 :	chl	181
749 :	syskin	712	/* now we're doing diagonal checks near our candidate */
750 :	chl	181
751 :	syskin	712	if (iDirection) { //checking if anything found
752 :			bDirection = iDirection;
753 :			iDirection = 0;
754 :			x = data->currentMV->x; y = data->currentMV->y;
755 :			if (bDirection & 3) { //our candidate is left or right
756 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
757 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
758 :			} else { // what remains here is up or down
759 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
760 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
761 :	chl	181	}
762 :	syskin	712	bDirection += iDirection;
763 :			x = data->currentMV->x; y = data->currentMV->y;
764 :	chl	181	}
765 :	syskin	712	}
766 :			while (iDirection);
767 :	chl	181	}
768 :
769 :	chl	530	/* MAINSEARCH FUNCTIONS END */
770 :	chl	181
771 :	chl	530	static void
772 :	syskin	663	SubpelRefine(const SearchData * const data)
773 :	chl	96	{
774 :	syskin	663	/* Do a half-pel or q-pel refinement */
775 :	syskin	768	const VECTOR centerMV = data->qpel_precision ? data->currentQMV : data->currentMV;
776 :			int iDirection; //only needed because macro expects it
777 :	chl	96
778 :	syskin	768	CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0);
779 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0);
780 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, 0);
781 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, 0);
782 :			CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, 0);
783 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, 0);
784 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, 0);
785 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, 0);
786 :	Isibaar	580	}
787 :
788 :	chl	530	static __inline int
789 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
790 :			const int x, const int y,
791 :	syskin	768	const uint32_t stride, const uint32_t iQuant, int rrv)
792 :	chl	96
793 :	Isibaar	3	{
794 :	syskin	768	if(!rrv) {
795 :			uint32_t sadC = sad8(current->u + x8 + ystride*8,
796 :			reference->u + x8 + ystride*8, stride);
797 :	syskin	704	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
798 :	syskin	768	sadC += sad8(current->v + (x + ystride)8,
799 :			reference->v + (x + ystride)8, stride);
800 :	syskin	704	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
801 :			return 1;
802 :	syskin	768
803 :			} else {
804 :			uint32_t sadC = sad16(current->u + x16 + ystride*16,
805 :			reference->u + x16 + ystride16, stride, 2564096);
806 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
807 :			sadC += sad16(current->v + (x + ystride)16,
808 :			reference->v + (x + ystride)16, stride, 256*4096);
809 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
810 :			return 1;
811 :	syskin	704	}
812 :	chl	530	}
813 :	edgomez	195
814 :	chl	530	static __inline void
815 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
816 :			{
817 :			pMB->mode = MODE_NOT_CODED;
818 :	syskin	768	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV;
819 :			pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV;
820 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
821 :	Isibaar	3	}
822 :
823 :	syskin	628	bool
824 :	chl	530	MotionEstimation(MBParam * const pParam,
825 :			FRAMEINFO * const current,
826 :			FRAMEINFO * const reference,
827 :			const IMAGE * const pRefH,
828 :			const IMAGE * const pRefV,
829 :			const IMAGE * const pRefHV,
830 :			const uint32_t iLimit)
831 :	Isibaar	3	{
832 :	chl	530	MACROBLOCK *const pMBs = current->mbs;
833 :			const IMAGE *const pCurrent = &current->image;
834 :			const IMAGE *const pRef = &reference->image;
835 :	Isibaar	3
836 :	syskin	704	uint32_t mb_width = pParam->mb_width;
837 :			uint32_t mb_height = pParam->mb_height;
838 :	syskin	768	const uint32_t iEdgedWidth = pParam->edged_width;
839 :	suxen_drol	702
840 :	chl	530	uint32_t x, y;
841 :			uint32_t iIntra = 0;
842 :	syskin	628	int32_t InterBias, quant = current->quant, sad00;
843 :	edgomez	195
844 :	Isibaar	539	// some pre-initialized thingies for SearchP
845 :	syskin	712	int32_t temp[8];
846 :	Isibaar	539	VECTOR currentMV[5];
847 :	Isibaar	580	VECTOR currentQMV[5];
848 :	Isibaar	539	int32_t iMinSAD[5];
849 :			SearchData Data;
850 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
851 :	syskin	768	Data.iEdgedWidth = iEdgedWidth;
852 :	Isibaar	539	Data.currentMV = currentMV;
853 :	Isibaar	580	Data.currentQMV = currentQMV;
854 :	Isibaar	539	Data.iMinSAD = iMinSAD;
855 :			Data.temp = temp;
856 :			Data.iFcode = current->fcode;
857 :	Isibaar	580	Data.rounding = pParam->m_rounding_type;
858 :	syskin	628	Data.qpel = pParam->m_quarterpel;
859 :	syskin	801	Data.chroma = current->motion_flags & PMV_CHROMA16;
860 :	syskin	704	Data.rrv = current->global_flags & XVID_REDUCED;
861 :	Isibaar	539
862 :	syskin	704	if ((current->global_flags & XVID_REDUCED)) {
863 :	suxen_drol	702	mb_width = (pParam->width + 31) / 32;
864 :			mb_height = (pParam->height + 31) / 32;
865 :	syskin	704	Data.qpel = Data.chroma = 0;
866 :	suxen_drol	702	}
867 :
868 :	syskin	754	Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose)
869 :	chl	530	if (sadInit) (*sadInit) ();
870 :	edgomez	195
871 :	suxen_drol	702	for (y = 0; y < mb_height; y++) {
872 :			for (x = 0; x < mb_width; x++) {
873 :	chl	530	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
874 :	syskin	768
875 :			if (!Data.rrv) pMB->sad16 =
876 :			sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16,
877 :			pRef->y + (x + y * iEdgedWidth) * 16,
878 :	syskin	704	pParam->edged_width, pMB->sad8 );
879 :	syskin	768
880 :	syskin	748	else pMB->sad16 =
881 :	syskin	768	sad32v_c(pCurrent->y + (x + y * iEdgedWidth) * 32,
882 :			pRef->y + (x + y * iEdgedWidth) * 32,
883 :	chl	530	pParam->edged_width, pMB->sad8 );
884 :	edgomez	195
885 :	syskin	628	if (Data.chroma) {
886 :	syskin	768	Data.temp[7] = sad8(pCurrent->u + x8 + y(iEdgedWidth/2)*8,
887 :			pRef->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2)
888 :			+ sad8(pCurrent->v + (x + y(iEdgedWidth/2))8,
889 :			pRef->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
890 :			pMB->sad16 += Data.temp[7];
891 :	syskin	628	}
892 :
893 :	syskin	768	sad00 = pMB->sad16;
894 :	syskin	628
895 :	chl	530	if (!(current->global_flags & XVID_LUMIMASKING)) {
896 :			pMB->dquant = NO_CHANGE;
897 :	syskin	601	} else {
898 :	syskin	574	if (pMB->dquant != NO_CHANGE) {
899 :			quant += DQtab[pMB->dquant];
900 :			if (quant > 31) quant = 31;
901 :			else if (quant < 1) quant = 1;
902 :			}
903 :	syskin	601	}
904 :	syskin	768	pMB->quant = current->quant;
905 :	chl	181
906 :	syskin	628	//initial skip decision
907 :	chl	769	/* no early skip for GMC (global vector = skip vector is unknown!) */
908 :			if (!(current->global_flags & XVID_GMC)) { /* no fast SKIP for S(GMC)-VOPs */
909 :	syskin	801	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
910 :	syskin	768	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
911 :	chl	619	SkipMacroblockP(pMB, sad00);
912 :			continue;
913 :			}
914 :			}
915 :	syskin	768
916 :	syskin	628	SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
917 :	chl	530	y, current->motion_flags, pMB->quant,
918 :	Isibaar	539	&Data, pParam, pMBs, reference->mbs,
919 :	chl	530	current->global_flags & XVID_INTER4V, pMB);
920 :	edgomez	195
921 :	chl	530	/* final skip decision, a.k.a. "the vector you found, really that good?" */
922 :	chl	769	if (!(current->global_flags & XVID_GMC)) {
923 :	chl	619	if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
924 :	syskin	704	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
925 :	syskin	768	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
926 :	chl	619	SkipMacroblockP(pMB, sad00);
927 :			continue;
928 :			}
929 :			}
930 :	syskin	768
931 :	chl	530	/* finally, intra decision */
932 :	Isibaar	3
933 :	chl	530	InterBias = MV16_INTER_BIAS;
934 :	syskin	768	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
935 :	chl	530	if (y != 0)
936 :	syskin	639	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
937 :	chl	530	if (x != 0)
938 :	syskin	639	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
939 :	syskin	768
940 :	syskin	628	if (Data.chroma) InterBias += 50; // to compensate bigger SAD
941 :	syskin	768	if (Data.rrv) InterBias *= 4;
942 :	Isibaar	3
943 :	syskin	768	if (InterBias < pMB->sad16) {
944 :	syskin	704	int32_t deviation;
945 :	syskin	768	if (!Data.rrv)
946 :			deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 16, iEdgedWidth);
947 :			else {
948 :			deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 32, iEdgedWidth)
949 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16, iEdgedWidth)
950 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * iEdgedWidth, iEdgedWidth)
951 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * (iEdgedWidth+1), iEdgedWidth);
952 :			}
953 :			if (deviation < (pMB->sad16 - InterBias)) {
954 :	syskin	754	if (++iIntra >= iLimit) return 1;
955 :	syskin	768	SkipMacroblockP(pMB, 0); //same thing
956 :	chl	530	pMB->mode = MODE_INTRA;
957 :			}
958 :			}
959 :			}
960 :	Isibaar	3	}
961 :	chl	619
962 :	chl	770	if (current->global_flags & XVID_GMC ) /* GMC only for S(GMC)-VOPs */
963 :	chl	769	{
964 :			current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV);
965 :			}
966 :
967 :	syskin	628	return 0;
968 :	chl	530	}
969 :	Isibaar	3
970 :	edgomez	195
971 :	chl	530	static __inline int
972 :			make_mask(const VECTOR * const pmv, const int i)
973 :			{
974 :	Isibaar	539	int mask = 255, j;
975 :	chl	530	for (j = 0; j < i; j++) {
976 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
977 :			if (pmv[i].x == pmv[j].x) {
978 :	syskin	768	if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4;
979 :			else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8;
980 :	syskin	748	} else
981 :	chl	530	if (pmv[i].y == pmv[j].y) {
982 :	syskin	768	if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1;
983 :			else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2;
984 :	chl	530	}
985 :	Isibaar	3	}
986 :	chl	530	return mask;
987 :			}
988 :	edgomez	195
989 :	syskin	748	static __inline void
990 :	syskin	704	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
991 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
992 :	chl	530	{
993 :	edgomez	195
994 :	chl	530	//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
995 :	syskin	704	if (rrv) { iWcount /= 2; iHcount /= 2; }
996 :	edgomez	195
997 :	syskin	704	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
998 :	chl	530	pmv[5].x = EVEN(pmv[3].x);
999 :	syskin	748	pmv[5].y = EVEN(pmv[3].y);
1000 :	Isibaar	539	} else pmv[5].x = pmv[5].y = 0;
1001 :	chl	169
1002 :	chl	530	if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
1003 :			else pmv[3].x = pmv[3].y = 0;
1004 :	Isibaar	3
1005 :	chl	530	if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
1006 :	syskin	768	else pmv[4].x = pmv[4].y = 0;
1007 :	chl	169
1008 :	chl	530	// [1] median prediction
1009 :	syskin	767	pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
1010 :	chl	169
1011 :	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
1012 :	chl	169
1013 :	chl	530	pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
1014 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
1015 :	chl	169
1016 :	syskin	704	if ((x < iWcount-1) && (y < iHcount-1)) {
1017 :	chl	530	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
1018 :	syskin	748	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
1019 :	Isibaar	539	} else pmv[6].x = pmv[6].y = 0;
1020 :	syskin	704
1021 :			if (rrv) {
1022 :			int i;
1023 :			for (i = 0; i < 7; i++) {
1024 :	syskin	768	pmv[i].x = RRV_MV_SCALEUP(pmv[i].x);
1025 :	syskin	767	pmv[i].y = RRV_MV_SCALEUP(pmv[i].y);
1026 :	syskin	704	}
1027 :			}
1028 :	chl	530	}
1029 :	chl	169
1030 :	chl	530	static void
1031 :	syskin	628	SearchP(const IMAGE * const pRef,
1032 :	chl	530	const uint8_t * const pRefH,
1033 :			const uint8_t * const pRefV,
1034 :			const uint8_t * const pRefHV,
1035 :			const IMAGE * const pCur,
1036 :			const int x,
1037 :			const int y,
1038 :			const uint32_t MotionFlags,
1039 :			const uint32_t iQuant,
1040 :	Isibaar	539	SearchData * const Data,
1041 :	chl	530	const MBParam * const pParam,
1042 :			const MACROBLOCK * const pMBs,
1043 :			const MACROBLOCK * const prevMBs,
1044 :			int inter4v,
1045 :			MACROBLOCK * const pMB)
1046 :			{
1047 :	chl	169
1048 :	chl	530	int i, iDirection = 255, mask, threshA;
1049 :	Isibaar	539	VECTOR pmv[7];
1050 :	Isibaar	3
1051 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1052 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
1053 :	Isibaar	3
1054 :	syskin	768	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp);
1055 :	syskin	628
1056 :	syskin	768	Data->temp[5] = Data->temp[6] = 0; // chroma-sad cache
1057 :			i = Data->rrv ? 2 : 1;
1058 :	syskin	704	Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
1059 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1060 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1061 :	Isibaar	3
1062 :	syskin	704	Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
1063 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
1064 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
1065 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
1066 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1067 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1068 :
1069 :	syskin	601	Data->lambda16 = lambda_vec16[iQuant];
1070 :	Isibaar	606	Data->lambda8 = lambda_vec8[iQuant];
1071 :	syskin	663	Data->qpel_precision = 0;
1072 :	chl	169
1073 :	edgomez	568	if (pMB->dquant != NO_CHANGE) inter4v = 0;
1074 :	edgomez	195
1075 :	syskin	768	for(i = 0; i < 5; i++)
1076 :	Isibaar	580	Data->currentMV[i].x = Data->currentMV[i].y = 0;
1077 :	Isibaar	539
1078 :	syskin	768	if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1079 :	syskin	663	else Data->predMV = pmv[0];
1080 :	Isibaar	606
1081 :	syskin	724	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
1082 :	syskin	768	Data->iMinSAD[0] = pMB->sad16 + ((Data->lambda16 * i * pMB->sad16)>>10);
1083 :			Data->iMinSAD[1] = pMB->sad8[0] + ((Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS)) >> 10);
1084 :	Isibaar	539	Data->iMinSAD[2] = pMB->sad8[1];
1085 :			Data->iMinSAD[3] = pMB->sad8[2];
1086 :			Data->iMinSAD[4] = pMB->sad8[3];
1087 :
1088 :	syskin	768	if (x \| y) {
1089 :	edgomez	568	threshA = Data->temp[0]; // that's when we keep this SAD atm
1090 :	chl	530	if (threshA < 512) threshA = 512;
1091 :	syskin	768	else if (threshA > 1024) threshA = 1024;
1092 :			} else threshA = 512;
1093 :	Isibaar	3
1094 :	chl	530	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1095 :	syskin	704	prevMBs + x + y * pParam->mb_width, Data->rrv);
1096 :	edgomez	195
1097 :	syskin	768	if (!Data->rrv) {
1098 :			if (inter4v \| Data->chroma) CheckCandidate = CheckCandidate16;
1099 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
1100 :			} else CheckCandidate = CheckCandidate32;
1101 :	edgomez	568
1102 :	syskin	768	/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/
1103 :	chl	530
1104 :			for (i = 1; i < 7; i++) {
1105 :			if (!(mask = make_mask(pmv, i)) ) continue;
1106 :	edgomez	568	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1107 :	Isibaar	539	if (Data->iMinSAD[0] <= threshA) break;
1108 :	chl	530	}
1109 :
1110 :	Isibaar	539	if ((Data->iMinSAD[0] <= threshA) \|\|
1111 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1112 :	syskin	768	(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16)))
1113 :	chl	530	inter4v = 0;
1114 :	syskin	768	else {
1115 :	Isibaar	3
1116 :	syskin	748	MainSearchFunc * MainSearchPtr;
1117 :	Isibaar	539	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1118 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1119 :			else MainSearchPtr = DiamondSearch;
1120 :	chl	181
1121 :	Isibaar	539	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1122 :	Isibaar	3
1123 :	chl	530	/* extended search, diamond starting in 0,0 and in prediction.
1124 :			note that this search is/might be done in halfpel positions,
1125 :			which makes it more different than the diamond above */
1126 :	chl	259
1127 :	Isibaar	539	if (MotionFlags & PMV_EXTSEARCH16) {
1128 :			int32_t bSAD;
1129 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1130 :	syskin	704	if (Data->rrv) {
1131 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1132 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1133 :	syskin	748	} else
1134 :	syskin	704	if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1135 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1136 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1137 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1138 :	Isibaar	3
1139 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1140 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1141 :			if (bSAD < Data->iMinSAD[0]) {
1142 :			Data->currentMV[0] = backupMV;
1143 :			Data->iMinSAD[0] = bSAD; }
1144 :			}
1145 :	Isibaar	3
1146 :	Isibaar	539	backupMV = Data->currentMV[0];
1147 :	syskin	768	if (MotionFlags & PMV_HALFPELREFINE16 && !Data->rrv) startMV.x = startMV.y = 1;
1148 :			else startMV.x = startMV.y = 0;
1149 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1150 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1151 :	edgomez	195
1152 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1153 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1154 :			if (bSAD < Data->iMinSAD[0]) {
1155 :			Data->currentMV[0] = backupMV;
1156 :			Data->iMinSAD[0] = bSAD; }
1157 :			}
1158 :	Isibaar	3	}
1159 :			}
1160 :
1161 :	syskin	663	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1162 :	Isibaar	3
1163 :	Isibaar	580	for(i = 0; i < 5; i++) {
1164 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1165 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1166 :			}
1167 :
1168 :	syskin	768	if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE16) {
1169 :	syskin	663	Data->qpel_precision = 1;
1170 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1171 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1172 :	syskin	601
1173 :	syskin	663	SubpelRefine(Data);
1174 :	Isibaar	580	}
1175 :
1176 :	syskin	768	if (Data->iMinSAD[0] < (int32_t)iQuant * 30) inter4v = 0;
1177 :	Isibaar	539	if (inter4v) {
1178 :			SearchData Data8;
1179 :	syskin	768	memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data
1180 :
1181 :	Isibaar	539	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1182 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1183 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1184 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1185 :	syskin	768
1186 :	syskin	628	if (Data->chroma) {
1187 :	syskin	768	int sumx = 0, sumy = 0;
1188 :			const int div = 1 + Data->qpel;
1189 :			const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs;
1190 :	syskin	628
1191 :	syskin	768	for (i = 0; i < 4; i++) {
1192 :			sumx += mv[i].x / div;
1193 :			sumy += mv[i].y / div;
1194 :	syskin	663	}
1195 :	syskin	768
1196 :	syskin	753	Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf],
1197 :			(sumy >> 3) + roundtab_76[sumy & 0xf], Data);
1198 :	syskin	628	}
1199 :	Isibaar	539	}
1200 :	Isibaar	3
1201 :	syskin	704	if (Data->rrv) {
1202 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1203 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1204 :			}
1205 :	syskin	740
1206 :	chl	530	if (!(inter4v) \|\|
1207 :	syskin	748	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1208 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1209 :	chl	530	// INTER MODE
1210 :			pMB->mode = MODE_INTER;
1211 :	syskin	768	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1212 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1213 :	Isibaar	3
1214 :	syskin	768	if(Data->qpel) {
1215 :	syskin	663	pMB->qmvs[0] = pMB->qmvs[1]
1216 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1217 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1218 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1219 :	syskin	628	} else {
1220 :	Isibaar	580	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1221 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1222 :			}
1223 :	edgomez	195	} else {
1224 :	syskin	748	// INTER4V MODE; all other things are already set in Search8
1225 :	chl	530	pMB->mode = MODE_INTER4V;
1226 :	syskin	748	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1227 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1228 :	edgomez	78	}
1229 :	Isibaar	3	}
1230 :
1231 :	chl	530	static void
1232 :			Search8(const SearchData * const OldData,
1233 :			const int x, const int y,
1234 :			const uint32_t MotionFlags,
1235 :			const MBParam * const pParam,
1236 :			MACROBLOCK * const pMB,
1237 :			const MACROBLOCK * const pMBs,
1238 :	Isibaar	539	const int block,
1239 :			SearchData * const Data)
1240 :	chl	345	{
1241 :	syskin	704	int i = 0;
1242 :	Isibaar	539	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1243 :			Data->currentMV = OldData->currentMV + 1 + block;
1244 :	Isibaar	580	Data->currentQMV = OldData->currentQMV + 1 + block;
1245 :	chl	345
1246 :	syskin	768	if(Data->qpel) {
1247 :	syskin	704	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1248 :	syskin	724	if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1249 :			Data->predMV, Data->iFcode, 0, 0);
1250 :	Isibaar	606	} else {
1251 :	syskin	704	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1252 :	syskin	768	if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1253 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1254 :	Isibaar	606	}
1255 :	chl	345
1256 :	syskin	768	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
1257 :	syskin	704
1258 :	syskin	768	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8\|PMV_QUARTERPELREFINE8)) {
1259 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
1260 :	chl	345
1261 :	syskin	768	Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1262 :			Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1263 :			Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1264 :			Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1265 :	chl	345
1266 :	syskin	768	Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1267 :	syskin	663	Data->qpel_precision = 0;
1268 :	chl	345
1269 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1270 :	syskin	768	pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
1271 :	syskin	712
1272 :	syskin	768	if (!Data->rrv) CheckCandidate = CheckCandidate8;
1273 :			else CheckCandidate = CheckCandidate16no4v;
1274 :	syskin	712
1275 :	chl	530	if (MotionFlags & PMV_EXTSEARCH8) {
1276 :	Isibaar	580	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1277 :	syskin	768
1278 :	chl	530	MainSearchFunc *MainSearchPtr;
1279 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1280 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1281 :			else MainSearchPtr = DiamondSearch;
1282 :	chl	345
1283 :	Isibaar	580	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1284 :	chl	345
1285 :	syskin	601	if(*(Data->iMinSAD) < temp_sad) {
1286 :	Isibaar	580	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1287 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1288 :			}
1289 :			}
1290 :
1291 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1292 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1293 :
1294 :	syskin	663	SubpelRefine(Data); // perform halfpel refine of current best vector
1295 :	Isibaar	580
1296 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1297 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1298 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1299 :			}
1300 :			}
1301 :
1302 :	syskin	768	if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE8) {
1303 :			Data->qpel_precision = 1;
1304 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1305 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1306 :			SubpelRefine(Data);
1307 :	Isibaar	580	}
1308 :	chl	345	}
1309 :	chl	530
1310 :	syskin	704	if (Data->rrv) {
1311 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1312 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1313 :			}
1314 :	syskin	768
1315 :	syskin	712	if(Data->qpel) {
1316 :	syskin	663	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1317 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1318 :	syskin	768	pMB->qmvs[block] = *Data->currentQMV;
1319 :	syskin	704	} else {
1320 :	Isibaar	580	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1321 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1322 :			}
1323 :
1324 :	syskin	768	pMB->mvs[block] = *Data->currentMV;
1325 :			pMB->sad8[block] = 4 * *Data->iMinSAD;
1326 :	chl	345	}
1327 :
1328 :	syskin	768	/* motion estimation for B-frames */
1329 :	chl	345
1330 :	chl	530	static __inline VECTOR
1331 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1332 :	Isibaar	3	{
1333 :	chl	530	/* the stupidiest function ever */
1334 :	syskin	768	return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]);
1335 :	Isibaar	3	}
1336 :
1337 :	chl	530	static void __inline
1338 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1339 :			const uint32_t iWcount,
1340 :			const MACROBLOCK * const pMB,
1341 :			const uint32_t mode_curr)
1342 :	Isibaar	3	{
1343 :
1344 :	chl	530	// [0] is prediction
1345 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1346 :	Isibaar	3
1347 :	chl	530	pmv[1].x = pmv[1].y = 0; // [1] is zero
1348 :	Isibaar	3
1349 :	chl	530	pmv[2] = ChoosePred(pMB, mode_curr);
1350 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1351 :	edgomez	195
1352 :	chl	530	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1353 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1354 :	syskin	748	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1355 :	Isibaar	539	} else pmv[3].x = pmv[3].y = 0;
1356 :	Isibaar	3
1357 :	chl	530	if (y != 0) {
1358 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1359 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1360 :			} else pmv[4].x = pmv[4].y = 0;
1361 :	Isibaar	3
1362 :	chl	530	if (x != 0) {
1363 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1364 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1365 :			} else pmv[5].x = pmv[5].y = 0;
1366 :	Isibaar	3
1367 :	syskin	768	if (x != 0 && y != 0) {
1368 :	chl	530	pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1369 :	syskin	767	pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y);
1370 :	chl	530	} else pmv[6].x = pmv[6].y = 0;
1371 :			}
1372 :	chl	181
1373 :	edgomez	170
1374 :	syskin	768	/* search backward or forward */
1375 :	chl	530	static void
1376 :	syskin	801	SearchBF( const IMAGE * const pRef,
1377 :	chl	530	const uint8_t * const pRefH,
1378 :			const uint8_t * const pRefV,
1379 :			const uint8_t * const pRefHV,
1380 :			const IMAGE * const pCur,
1381 :			const int x, const int y,
1382 :			const uint32_t MotionFlags,
1383 :			const uint32_t iFcode,
1384 :			const MBParam * const pParam,
1385 :			MACROBLOCK * const pMB,
1386 :			const VECTOR * const predMV,
1387 :			int32_t * const best_sad,
1388 :	h	545	const int32_t mode_current,
1389 :			SearchData * const Data)
1390 :	chl	530	{
1391 :	Isibaar	3
1392 :	syskin	740	int i, iDirection = 255, mask;
1393 :	h	545	VECTOR pmv[7];
1394 :	chl	530	MainSearchFunc *MainSearchPtr;
1395 :	h	545	*Data->iMinSAD = MV_MAX_ERROR;
1396 :			Data->iFcode = iFcode;
1397 :	syskin	663	Data->qpel_precision = 0;
1398 :	syskin	801	Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache
1399 :	Isibaar	3
1400 :	syskin	801	Data->Ref = pRef->y + (x + y * Data->iEdgedWidth) * 16;
1401 :	syskin	768	Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16;
1402 :			Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16;
1403 :			Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16;
1404 :	syskin	801	Data->RefCU = pRef->u + (x + y * Data->iEdgedWidth/2) * 8;
1405 :			Data->RefCV = pRef->v + (x + y * Data->iEdgedWidth/2) * 8;
1406 :	Isibaar	3
1407 :	h	545	Data->predMV = *predMV;
1408 :	edgomez	195
1409 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1410 :	syskin	712	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1411 :	Isibaar	3
1412 :	h	545	pmv[0] = Data->predMV;
1413 :	syskin	658	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1414 :	syskin	768
1415 :	h	545	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1416 :	edgomez	195
1417 :	h	545	Data->currentMV->x = Data->currentMV->y = 0;
1418 :	chl	530	CheckCandidate = CheckCandidate16no4v;
1419 :	chl	181
1420 :	chl	530	// main loop. checking all predictions
1421 :	syskin	740	for (i = 0; i < 7; i++) {
1422 :	chl	530	if (!(mask = make_mask(pmv, i)) ) continue;
1423 :	h	545	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1424 :	edgomez	78	}
1425 :	Isibaar	3
1426 :	syskin	768	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1427 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1428 :	chl	530	else MainSearchPtr = DiamondSearch;
1429 :	chl	96
1430 :	syskin	740	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1431 :	chl	169
1432 :	syskin	663	SubpelRefine(Data);
1433 :	syskin	768
1434 :	syskin	740	if (Data->qpel && Data->iMinSAD < best_sad + 300) {
1435 :	syskin	658	Data->currentQMV->x = 2*Data->currentMV->x;
1436 :			Data->currentQMV->y = 2*Data->currentMV->y;
1437 :	syskin	663	Data->qpel_precision = 1;
1438 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1439 :			pParam->width, pParam->height, iFcode, 1, 0);
1440 :	syskin	663	SubpelRefine(Data);
1441 :	syskin	658	}
1442 :	chl	169
1443 :	chl	530	// three bits are needed to code backward mode. four for forward
1444 :	syskin	740
1445 :	syskin	768	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1446 :			else Data->iMinSAD += 3 Data->lambda16;
1447 :	chl	169
1448 :	h	545	if (Data->iMinSAD < best_sad) {
1449 :			best_sad = Data->iMinSAD;
1450 :	chl	530	pMB->mode = mode_current;
1451 :	syskin	658	if (Data->qpel) {
1452 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1453 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1454 :	syskin	748	if (mode_current == MODE_FORWARD)
1455 :	syskin	658	pMB->qmvs[0] = *Data->currentQMV;
1456 :	syskin	748	else
1457 :	syskin	658	pMB->b_qmvs[0] = *Data->currentQMV;
1458 :			} else {
1459 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1460 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1461 :			}
1462 :	syskin	739	if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1463 :			else pMB->b_mvs[0] = *Data->currentMV;
1464 :			}
1465 :	syskin	658
1466 :	syskin	768	if (mode_current == MODE_FORWARD) (Data->currentMV+2) = Data->currentMV;
1467 :			else (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1468 :	chl	530	}
1469 :	chl	169
1470 :	syskin	748	static void
1471 :	syskin	712	SkipDecisionB(const IMAGE * const pCur,
1472 :	syskin	767	const IMAGE * const f_Ref,
1473 :			const IMAGE * const b_Ref,
1474 :			MACROBLOCK * const pMB,
1475 :			const uint32_t x, const uint32_t y,
1476 :			const SearchData * const Data)
1477 :	syskin	712	{
1478 :	syskin	767	int dx = 0, dy = 0, b_dx = 0, b_dy = 0;
1479 :	syskin	768	int32_t sum;
1480 :	syskin	767	const int div = 1 + Data->qpel;
1481 :			int k;
1482 :	syskin	768	const uint32_t stride = Data->iEdgedWidth/2;
1483 :	syskin	712	//this is not full chroma compensation, only it's fullpel approximation. should work though
1484 :
1485 :	syskin	767	for (k = 0; k < 4; k++) {
1486 :			dy += Data->directmvF[k].y / div;
1487 :			dx += Data->directmvF[0].x / div;
1488 :			b_dy += Data->directmvB[0].y / div;
1489 :			b_dx += Data->directmvB[0].x / div;
1490 :	syskin	712	}
1491 :
1492 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1493 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1494 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1495 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1496 :
1497 :	syskin	768	sum = sad8bi(pCur->u + 8 * x + 8 * y * stride,
1498 :			f_Ref->u + (y8 + dy/2) stride + x*8 + dx/2,
1499 :			b_Ref->u + (y8 + b_dy/2) stride + x*8 + b_dx/2,
1500 :			stride);
1501 :	syskin	767
1502 :	syskin	768	if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; //no skip
1503 :	syskin	767
1504 :	syskin	768	sum += sad8bi(pCur->v + 8x + 8 y * stride,
1505 :			f_Ref->v + (y8 + dy/2) stride + x*8 + dx/2,
1506 :			b_Ref->v + (y8 + b_dy/2) stride + x*8 + b_dx/2,
1507 :			stride);
1508 :	syskin	712
1509 :	syskin	768	if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1510 :	syskin	712	}
1511 :
1512 :			static __inline uint32_t
1513 :	Isibaar	539	SearchDirect(const IMAGE * const f_Ref,
1514 :	chl	530	const uint8_t * const f_RefH,
1515 :			const uint8_t * const f_RefV,
1516 :			const uint8_t * const f_RefHV,
1517 :	Isibaar	539	const IMAGE * const b_Ref,
1518 :	chl	530	const uint8_t * const b_RefH,
1519 :			const uint8_t * const b_RefV,
1520 :			const uint8_t * const b_RefHV,
1521 :			const IMAGE * const pCur,
1522 :			const int x, const int y,
1523 :			const uint32_t MotionFlags,
1524 :			const int32_t TRB, const int32_t TRD,
1525 :			const MBParam * const pParam,
1526 :			MACROBLOCK * const pMB,
1527 :			const MACROBLOCK * const b_mb,
1528 :	h	545	int32_t * const best_sad,
1529 :			SearchData * const Data)
1530 :	edgomez	195
1531 :	chl	530	{
1532 :	h	545	int32_t skip_sad;
1533 :	syskin	768	int k = (x + Data->iEdgedWidthy) 16;
1534 :	chl	530	MainSearchFunc *MainSearchPtr;
1535 :	Isibaar	3
1536 :	h	545	Data->iMinSAD = 2564096;
1537 :	syskin	768	Data->Ref = f_Ref->y + k;
1538 :			Data->RefH = f_RefH + k;
1539 :			Data->RefV = f_RefV + k;
1540 :			Data->RefHV = f_RefHV + k;
1541 :			Data->bRef = b_Ref->y + k;
1542 :			Data->bRefH = b_RefH + k;
1543 :			Data->bRefV = b_RefV + k;
1544 :			Data->bRefHV = b_RefHV + k;
1545 :	syskin	801	Data->RefCU = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
1546 :			Data->RefCV = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
1547 :			Data->b_RefCU = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
1548 :			Data->b_RefCV = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
1549 :	chl	140
1550 :	syskin	768	k = Data->qpel ? 4 : 2;
1551 :			Data->max_dx = k * (pParam->width - x * 16);
1552 :			Data->max_dy = k * (pParam->height - y * 16);
1553 :			Data->min_dx = -k * (16 + x * 16);
1554 :			Data->min_dy = -k * (16 + y * 16);
1555 :	chl	140
1556 :	syskin	768	Data->referencemv = Data->qpel ? b_mb->qmvs : b_mb->mvs;
1557 :			Data->qpel_precision = 0;
1558 :	Isibaar	3
1559 :	chl	530	for (k = 0; k < 4; k++) {
1560 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1561 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1562 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1563 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1564 :	Isibaar	3
1565 :	syskin	768	if ( (pMB->b_mvs[k].x > Data->max_dx) \| (pMB->b_mvs[k].x < Data->min_dx)
1566 :			\| (pMB->b_mvs[k].y > Data->max_dy) \| (pMB->b_mvs[k].y < Data->min_dy) ) {
1567 :	Isibaar	3
1568 :	Isibaar	539	best_sad = 2564096; // in that case, we won't use direct mode
1569 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1570 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1571 :	syskin	712	return 256*4096;
1572 :	Isibaar	539	}
1573 :			if (b_mb->mode != MODE_INTER4V) {
1574 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1575 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1576 :	h	545	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1577 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1578 :	Isibaar	539	break;
1579 :			}
1580 :			}
1581 :	Isibaar	3
1582 :	syskin	768	CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v;
1583 :
1584 :	h	545	(*CheckCandidate)(0, 0, 255, &k, Data);
1585 :	Isibaar	3
1586 :	syskin	712	// initial (fast) skip decision
1587 :	syskin	801	if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH * (2 + Data->chroma?1:0)) {
1588 :			//possible skip
1589 :			if (Data->chroma) {
1590 :			pMB->mode = MODE_DIRECT_NONE_MV;
1591 :			return *Data->iMinSAD; // skip.
1592 :			} else {
1593 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data);
1594 :			if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1595 :			}
1596 :	Isibaar	539	}
1597 :
1598 :	h	545	skip_sad = *Data->iMinSAD;
1599 :	edgomez	195
1600 :	syskin	768	// DIRECT MODE DELTA VECTOR SEARCH.
1601 :	chl	530	// This has to be made more effective, but at the moment I'm happy it's running at all
1602 :	Isibaar	3
1603 :	chl	530	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1604 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1605 :			else MainSearchPtr = DiamondSearch;
1606 :	edgomez	195
1607 :	h	545	(*MainSearchPtr)(0, 0, Data, 255);
1608 :	Isibaar	3
1609 :	syskin	663	SubpelRefine(Data);
1610 :	Isibaar	3
1611 :	h	545	best_sad = Data->iMinSAD;
1612 :	Isibaar	3
1613 :	syskin	768	if (Data->qpel \|\| b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT;
1614 :	syskin	672	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1615 :	edgomez	195
1616 :	h	545	pMB->pmvs[3] = *Data->currentMV;
1617 :	edgomez	195
1618 :	chl	530	for (k = 0; k < 4; k++) {
1619 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1620 :	syskin	658	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1621 :	h	545	? Data->directmvB[k].x
1622 :	syskin	658	:pMB->mvs[k].x - Data->referencemv[k].x);
1623 :	h	545	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1624 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1625 :			? Data->directmvB[k].y
1626 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1627 :	syskin	658	if (Data->qpel) {
1628 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1629 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1630 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1631 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1632 :			}
1633 :
1634 :	chl	530	if (b_mb->mode != MODE_INTER4V) {
1635 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1636 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1637 :	syskin	658	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1638 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1639 :	chl	530	break;
1640 :			}
1641 :			}
1642 :	h	545	return skip_sad;
1643 :	Isibaar	3	}
1644 :	chl	96
1645 :	syskin	712	static void
1646 :	syskin	801	SearchInterpolate(const IMAGE * const f_Ref,
1647 :	chl	530	const uint8_t * const f_RefH,
1648 :			const uint8_t * const f_RefV,
1649 :			const uint8_t * const f_RefHV,
1650 :	syskin	801	const IMAGE * const b_Ref,
1651 :	chl	530	const uint8_t * const b_RefH,
1652 :			const uint8_t * const b_RefV,
1653 :			const uint8_t * const b_RefHV,
1654 :			const IMAGE * const pCur,
1655 :			const int x, const int y,
1656 :			const uint32_t fcode,
1657 :			const uint32_t bcode,
1658 :			const uint32_t MotionFlags,
1659 :			const MBParam * const pParam,
1660 :			const VECTOR * const f_predMV,
1661 :			const VECTOR * const b_predMV,
1662 :			MACROBLOCK * const pMB,
1663 :	h	545	int32_t * const best_sad,
1664 :			SearchData * const fData)
1665 :	chl	530
1666 :	chl	96	{
1667 :
1668 :	chl	530	int iDirection, i, j;
1669 :	h	545	SearchData bData;
1670 :	chl	96
1671 :	syskin	712	fData->qpel_precision = 0;
1672 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1673 :			fData->iMinSAD = 4096256;
1674 :	syskin	739	bData.currentMV++; bData.currentQMV++;
1675 :	h	545	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1676 :	edgomez	195
1677 :	syskin	712	i = (x + y * fData->iEdgedWidth) * 16;
1678 :	syskin	801	bData.bRef = fData->Ref = f_Ref->y + i;
1679 :	syskin	712	bData.bRefH = fData->RefH = f_RefH + i;
1680 :			bData.bRefV = fData->RefV = f_RefV + i;
1681 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1682 :	syskin	801	bData.Ref = fData->bRef = b_Ref->y + i;
1683 :	syskin	712	bData.RefH = fData->bRefH = b_RefH + i;
1684 :			bData.RefV = fData->bRefV = b_RefV + i;
1685 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1686 :	syskin	801	bData.b_RefCU = fData->RefCU = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
1687 :			bData.b_RefCV = fData->RefCV = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
1688 :			bData.RefCU = fData->b_RefCU = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
1689 :			bData.RefCV = fData->b_RefCV = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
1690 :	chl	96
1691 :	syskin	801
1692 :	h	545	bData.bpredMV = fData->predMV = *f_predMV;
1693 :			fData->bpredMV = bData.predMV = *b_predMV;
1694 :	syskin	661	fData->currentMV[0] = fData->currentMV[2];
1695 :	chl	96
1696 :	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);
1697 :			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);
1698 :
1699 :	h	545	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1700 :	syskin	675	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1701 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1702 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1703 :	Isibaar	539
1704 :	h	545	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1705 :	syskin	675	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1706 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1707 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1708 :	Isibaar	539
1709 :	h	545	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1710 :	chl	96
1711 :	syskin	768	//diamond
1712 :	chl	530	do {
1713 :			iDirection = 255;
1714 :			// forward MV moves
1715 :	h	545	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1716 :	chl	96
1717 :	h	545	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1718 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1719 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1720 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1721 :	chl	96
1722 :	chl	530	// backward MV moves
1723 :	h	545	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1724 :			fData->currentMV[2] = fData->currentMV[0];
1725 :	Isibaar	539	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1726 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1727 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1728 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1729 :	chl	96
1730 :	chl	530	} while (!(iDirection));
1731 :	edgomez	195
1732 :	syskin	768	//qpel refinement
1733 :	syskin	658	if (fData->qpel) {
1734 :	syskin	740	if (fData->iMinSAD > best_sad + 500) return;
1735 :	syskin	675	CheckCandidate = CheckCandidateInt;
1736 :	syskin	663	fData->qpel_precision = bData.qpel_precision = 1;
1737 :	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);
1738 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1739 :	syskin	658	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1740 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1741 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1742 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1743 :	syskin	663	SubpelRefine(fData);
1744 :	syskin	740	if (fData->iMinSAD > best_sad + 300) return;
1745 :	syskin	658	fData->currentQMV[2] = fData->currentQMV[0];
1746 :	syskin	663	SubpelRefine(&bData);
1747 :	syskin	658	}
1748 :
1749 :	syskin	768	fData->iMinSAD += (2+3) fData->lambda16; // two bits are needed to code interpolate mode.
1750 :
1751 :	h	545	if (fData->iMinSAD < best_sad) {
1752 :			best_sad = fData->iMinSAD;
1753 :			pMB->mvs[0] = fData->currentMV[0];
1754 :			pMB->b_mvs[0] = fData->currentMV[1];
1755 :	chl	530	pMB->mode = MODE_INTERPOLATE;
1756 :	syskin	658	if (fData->qpel) {
1757 :			pMB->qmvs[0] = fData->currentQMV[0];
1758 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1759 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1760 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1761 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1762 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1763 :			} else {
1764 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1765 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1766 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1767 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1768 :			}
1769 :	chl	530	}
1770 :			}
1771 :	chl	96
1772 :	chl	530	void
1773 :			MotionEstimationBVOP(MBParam * const pParam,
1774 :			FRAMEINFO * const frame,
1775 :			const int32_t time_bp,
1776 :			const int32_t time_pp,
1777 :			// forward (past) reference
1778 :			const MACROBLOCK * const f_mbs,
1779 :			const IMAGE * const f_ref,
1780 :			const IMAGE * const f_refH,
1781 :			const IMAGE * const f_refV,
1782 :			const IMAGE * const f_refHV,
1783 :			// backward (future) reference
1784 :	syskin	644	const FRAMEINFO * const b_reference,
1785 :	chl	530	const IMAGE * const b_ref,
1786 :			const IMAGE * const b_refH,
1787 :			const IMAGE * const b_refV,
1788 :			const IMAGE * const b_refHV)
1789 :			{
1790 :			uint32_t i, j;
1791 :	syskin	712	int32_t best_sad;
1792 :			uint32_t skip_sad;
1793 :	chl	530	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1794 :	syskin	644	const MACROBLOCK * const b_mbs = b_reference->mbs;
1795 :	chl	96
1796 :	chl	530	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1797 :	chl	96
1798 :	chl	530	const int32_t TRB = time_pp - time_bp;
1799 :			const int32_t TRD = time_pp;
1800 :	chl	96
1801 :	h	545	// some pre-inintialized data for the rest of the search
1802 :
1803 :			SearchData Data;
1804 :			int32_t iMinSAD;
1805 :			VECTOR currentMV[3];
1806 :	syskin	658	VECTOR currentQMV[3];
1807 :	syskin	801	int32_t temp[8];
1808 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
1809 :	h	545	Data.iEdgedWidth = pParam->edged_width;
1810 :	syskin	658	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1811 :	h	545	Data.iMinSAD = &iMinSAD;
1812 :	syskin	712	Data.lambda16 = lambda_vec16[frame->quant];
1813 :	syskin	658	Data.qpel = pParam->m_quarterpel;
1814 :	syskin	663	Data.rounding = 0;
1815 :	syskin	801	Data.chroma = frame->motion_flags & PMV_CHROMA8;
1816 :			Data.temp = temp;
1817 :	h	545
1818 :	syskin	754	Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose)
1819 :	chl	530	// note: i==horizontal, j==vertical
1820 :			for (j = 0; j < pParam->mb_height; j++) {
1821 :	chl	96
1822 :	chl	530	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1823 :	edgomez	195
1824 :	chl	530	for (i = 0; i < pParam->mb_width; i++) {
1825 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1826 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1827 :	edgomez	195
1828 :	chl	619	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1829 :	syskin	644	if (b_reference->coding_type != S_VOP)
1830 :			if (b_mb->mode == MODE_NOT_CODED) {
1831 :			pMB->mode = MODE_NOT_CODED;
1832 :			continue;
1833 :			}
1834 :	chl	96
1835 :	h	545	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1836 :	syskin	801	Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8;
1837 :			Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8;
1838 :	syskin	601	pMB->quant = frame->quant;
1839 :	syskin	628
1840 :	chl	530	/* direct search comes first, because it (1) checks for SKIP-mode
1841 :			and (2) sets very good predictions for forward and backward search */
1842 :	Isibaar	539	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1843 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1844 :	chl	530	&frame->image,
1845 :			i, j,
1846 :			frame->motion_flags,
1847 :			TRB, TRD,
1848 :			pParam,
1849 :			pMB, b_mb,
1850 :	h	545	&best_sad,
1851 :			&Data);
1852 :	chl	96
1853 :	Isibaar	539	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1854 :	syskin	739
1855 :	chl	530	// forward search
1856 :	syskin	801	SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1857 :	chl	530	&frame->image, i, j,
1858 :			frame->motion_flags,
1859 :	h	545	frame->fcode, pParam,
1860 :	chl	530	pMB, &f_predMV, &best_sad,
1861 :	h	545	MODE_FORWARD, &Data);
1862 :	chl	96
1863 :	chl	530	// backward search
1864 :	syskin	801	SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y,
1865 :	chl	530	&frame->image, i, j,
1866 :			frame->motion_flags,
1867 :	h	545	frame->bcode, pParam,
1868 :	chl	530	pMB, &b_predMV, &best_sad,
1869 :	h	545	MODE_BACKWARD, &Data);
1870 :	chl	96
1871 :	chl	530	// interpolate search comes last, because it uses data from forward and backward as prediction
1872 :	syskin	801	SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1873 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1874 :	chl	530	&frame->image,
1875 :			i, j,
1876 :			frame->fcode, frame->bcode,
1877 :			frame->motion_flags,
1878 :	h	545	pParam,
1879 :	chl	530	&f_predMV, &b_predMV,
1880 :	h	545	pMB, &best_sad,
1881 :			&Data);
1882 :	syskin	662
1883 :	syskin	712	// final skip decision
1884 :	syskin	801	if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP * 2)
1885 :	syskin	712	&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1886 :	syskin	767	SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data);
1887 :	syskin	712
1888 :	chl	530	switch (pMB->mode) {
1889 :			case MODE_FORWARD:
1890 :			f_count++;
1891 :	syskin	768	f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
1892 :	chl	530	break;
1893 :			case MODE_BACKWARD:
1894 :			b_count++;
1895 :	syskin	768	b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
1896 :	chl	530	break;
1897 :			case MODE_INTERPOLATE:
1898 :			i_count++;
1899 :	syskin	768	f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
1900 :			b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
1901 :	chl	530	break;
1902 :			case MODE_DIRECT:
1903 :			case MODE_DIRECT_NO4V:
1904 :			d_count++;
1905 :			default:
1906 :			break;
1907 :	chl	96	}
1908 :			}
1909 :			}
1910 :			}
1911 :
1912 :	syskin	739	static __inline void
1913 :	Isibaar	539	MEanalyzeMB ( const uint8_t * const pRef,
1914 :			const uint8_t * const pCur,
1915 :			const int x,
1916 :			const int y,
1917 :			const MBParam * const pParam,
1918 :	syskin	739	MACROBLOCK * const pMBs,
1919 :	h	545	SearchData * const Data)
1920 :	Isibaar	539	{
1921 :	chl	317
1922 :	syskin	739	int i, mask;
1923 :	h	545	VECTOR pmv[3];
1924 :	syskin	739	MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width];
1925 :	syskin	768
1926 :	syskin	739	for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1927 :	suxen_drol	118
1928 :	syskin	640	//median is only used as prediction. it doesn't have to be real
1929 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1930 :			else
1931 :			if (x == 1) //left macroblock does not have any vector now
1932 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1933 :	syskin	748	else if (y == 1) // top macroblock doesn't have it's vector
1934 :	syskin	640	Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1935 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1936 :
1937 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1938 :	syskin	739	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv);
1939 :	chl	326
1940 :	h	545	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1941 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1942 :	syskin	739
1943 :	Isibaar	539	pmv[1].x = EVEN(pMB->mvs[0].x);
1944 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1945 :	syskin	640	pmv[2].x = EVEN(Data->predMV.x);
1946 :			pmv[2].y = EVEN(Data->predMV.y);
1947 :			pmv[0].x = pmv[0].y = 0;
1948 :	chl	326
1949 :	syskin	739	CheckCandidate32I(0, 0, 255, &i, Data);
1950 :	syskin	640
1951 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) {
1952 :	syskin	640
1953 :	syskin	739	if (!(mask = make_mask(pmv, 1)))
1954 :			CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data);
1955 :			if (!(mask = make_mask(pmv, 2)))
1956 :			CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data);
1957 :	Isibaar	539
1958 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) // diamond only if needed
1959 :			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1960 :	Isibaar	539
1961 :	syskin	739	for (i = 0; i < 4; i++) {
1962 :	syskin	768	MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width];
1963 :	syskin	739	MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
1964 :			MB->mode = MODE_INTER;
1965 :			MB->sad16 = Data->iMinSAD[i+1];
1966 :			}
1967 :			}
1968 :	Isibaar	539	}
1969 :
1970 :	syskin	739	#define INTRA_BIAS 2500
1971 :			#define INTRA_THRESH 1500
1972 :			#define INTER_THRESH 1400
1973 :	Isibaar	539
1974 :			int
1975 :	syskin	768	MEanalysis( const IMAGE * const pRef,
1976 :	syskin	644	FRAMEINFO * const Current,
1977 :	Isibaar	539	MBParam * const pParam,
1978 :	syskin	644	int maxIntra, //maximum number if non-I frames
1979 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1980 :	syskin	768	int bCount) // number of B frames in a row
1981 :	Isibaar	539	{
1982 :			uint32_t x, y, intra = 0;
1983 :			int sSAD = 0;
1984 :	syskin	644	MACROBLOCK * const pMBs = Current->mbs;
1985 :			const IMAGE * const pCurrent = &Current->image;
1986 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1987 :	h	545
1988 :	syskin	739	int32_t iMinSAD[5], temp[5];
1989 :			VECTOR currentMV[5];
1990 :	h	545	SearchData Data;
1991 :			Data.iEdgedWidth = pParam->edged_width;
1992 :	syskin	739	Data.currentMV = currentMV;
1993 :			Data.iMinSAD = iMinSAD;
1994 :	syskin	644	Data.iFcode = Current->fcode;
1995 :	syskin	739	Data.rrv = Current->global_flags & XVID_REDUCED;
1996 :			Data.temp = temp;
1997 :			CheckCandidate = CheckCandidate32I;
1998 :	h	545
1999 :	syskin	768	if (intraCount != 0 && intraCount < 10) // we're right after an I frame
2000 :	syskin	658	IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
2001 :	syskin	644	else
2002 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
2003 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
2004 :
2005 :	syskin	672	InterThresh += 400 * (1 - bCount);
2006 :	syskin	739	if (InterThresh < 300) InterThresh = 300;
2007 :	syskin	644
2008 :	Isibaar	539	if (sadInit) (*sadInit) ();
2009 :
2010 :	syskin	768	for (y = 1; y < pParam->mb_height-1; y += 2) {
2011 :			for (x = 1; x < pParam->mb_width-1; x += 2) {
2012 :	syskin	739	int i;
2013 :	Isibaar	539
2014 :	syskin	739	if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
2015 :
2016 :			MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
2017 :	syskin	768
2018 :	syskin	739	for (i = 0; i < 4; i++) {
2019 :			int dev;
2020 :	syskin	768	MACROBLOCK pMB = &pMBs[x+(i&1) + (y+(i>>1)) pParam->mb_width];
2021 :	syskin	739	if (pMB->sad16 > IntraThresh) {
2022 :	syskin	768	dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16,
2023 :			pParam->edged_width);
2024 :	syskin	739	if (dev + IntraThresh < pMB->sad16) {
2025 :			pMB->mode = MODE_INTRA;
2026 :			if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
2027 :			}
2028 :	Isibaar	539	}
2029 :	syskin	739	sSAD += pMB->sad16;
2030 :	syskin	644	}
2031 :	suxen_drol	118	}
2032 :			}
2033 :	syskin	704	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2034 :	syskin	793	// if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP;
2035 :	suxen_drol	709	if (sSAD > InterThresh ) return P_VOP;
2036 :	Isibaar	539	emms();
2037 :	suxen_drol	709	return B_VOP;
2038 :	syskin	712
2039 :	suxen_drol	118	}
2040 :	syskin	576
2041 :	chl	769
2042 :			static WARPPOINTS
2043 :	syskin	801	GlobalMotionEst(const MACROBLOCK * const pMBs,
2044 :	chl	769	const MBParam * const pParam,
2045 :			const FRAMEINFO * const current,
2046 :			const FRAMEINFO * const reference,
2047 :			const IMAGE * const pRefH,
2048 :			const IMAGE * const pRefV,
2049 :			const IMAGE * const pRefHV )
2050 :	syskin	628	{
2051 :	chl	619
2052 :	chl	769	const int deltax=8; // upper bound for difference between a MV and it's neighbour MVs
2053 :			const int deltay=8;
2054 :	syskin	801	const int grad=512; // lower bound for deviation in MB
2055 :	chl	769
2056 :			WARPPOINTS gmc;
2057 :
2058 :			uint32_t mx, my;
2059 :	chl	619
2060 :	chl	769	int MBh = pParam->mb_height;
2061 :			int MBw = pParam->mb_width;
2062 :
2063 :			int MBmask= calloc(MBhMBw,sizeof(int));
2064 :			double DtimesF[4] = { 0.,0., 0., 0. };
2065 :			double sol[4] = { 0., 0., 0., 0. };
2066 :			double a,b,c,n,denom;
2067 :			double meanx,meany;
2068 :			int num,oldnum;
2069 :
2070 :			if (!MBmask) { fprintf(stderr,"Mem error\n"); return gmc;}
2071 :
2072 :	syskin	801	// filter mask of all blocks
2073 :	chl	769
2074 :			for (my = 1; my < MBh-1; my++)
2075 :			for (mx = 1; mx < MBw-1; mx++)
2076 :			{
2077 :			const int mbnum = mx + my * MBw;
2078 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2079 :			const VECTOR mv = pMB->mvs[0];
2080 :
2081 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2082 :			continue;
2083 :
2084 :			if ( ( (ABS(mv.x - (pMB-1)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-1)->mvs[0].y) < deltay) )
2085 :			&& ( (ABS(mv.x - (pMB+1)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+1)->mvs[0].y) < deltay) )
2086 :			&& ( (ABS(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-MBw)->mvs[0].y) < deltay) )
2087 :			&& ( (ABS(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) )
2088 :			MBmask[mbnum]=1;
2089 :	syskin	628	}
2090 :	chl	769
2091 :			for (my = 1; my < MBh-1; my++)
2092 :			for (mx = 1; mx < MBw-1; mx++)
2093 :			{
2094 :			const uint8_t const pCur = current->image.y + 16mypParam->edged_width + 16mx;
2095 :
2096 :			const int mbnum = mx + my * MBw;
2097 :			if (!MBmask[mbnum])
2098 :			continue;
2099 :	syskin	628
2100 :	chl	769	if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= grad )
2101 :			MBmask[mbnum] = 0;
2102 :			if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= grad )
2103 :			MBmask[mbnum] = 0;
2104 :
2105 :			}
2106 :	syskin	628
2107 :	chl	769	emms();
2108 :	chl	619
2109 :	chl	769	do { /* until convergence */
2110 :	chl	619
2111 :	chl	769	a = b = c = n = 0;
2112 :			DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.;
2113 :			for (my = 0; my < MBh; my++)
2114 :			for (mx = 0; mx < MBw; mx++)
2115 :			{
2116 :			const int mbnum = mx + my * MBw;
2117 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2118 :			const VECTOR mv = pMB->mvs[0];
2119 :	chl	619
2120 :	chl	769	if (!MBmask[mbnum])
2121 :			continue;
2122 :
2123 :			n++;
2124 :			a += 16*mx+8;
2125 :			b += 16*my+8;
2126 :			c += (16mx+8)(16mx+8)+(16my+8)(16my+8);
2127 :	syskin	768
2128 :	chl	769	DtimesF[0] += (double)mv.x;
2129 :			DtimesF[1] += (double)mv.x(16mx+8) + (double)mv.y(16my+8);
2130 :			DtimesF[2] += (double)mv.x(16my+8) - (double)mv.y(16mx+8);
2131 :			DtimesF[3] += (double)mv.y;
2132 :			}
2133 :	syskin	768
2134 :	chl	769	denom = aa+bb-c*n;
2135 :
2136 :			/* Solve the system: sol = (D'ED)^{-1} D'EF */
2137 :			/* D'EF has been calculated in the same loop as matrix */
2138 :	syskin	768
2139 :	chl	769	sol[0] = -cDtimesF[0] + aDtimesF[1] + b*DtimesF[2];
2140 :			sol[1] = aDtimesF[0] - nDtimesF[1] + b*DtimesF[3];
2141 :			sol[2] = bDtimesF[0] - nDtimesF[2] - a*DtimesF[3];
2142 :			sol[3] = bDtimesF[1] - aDtimesF[2] - c*DtimesF[3];
2143 :	syskin	628
2144 :	chl	769	sol[0] /= denom;
2145 :			sol[1] /= denom;
2146 :			sol[2] /= denom;
2147 :			sol[3] /= denom;
2148 :
2149 :			meanx = meany = 0.;
2150 :			oldnum = 0;
2151 :			for (my = 0; my < MBh; my++)
2152 :			for (mx = 0; mx < MBw; mx++)
2153 :			{
2154 :			const int mbnum = mx + my * MBw;
2155 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2156 :			const VECTOR mv = pMB->mvs[0];
2157 :	syskin	628
2158 :	chl	769	if (!MBmask[mbnum])
2159 :			continue;
2160 :	syskin	628
2161 :	chl	769	oldnum++;
2162 :			meanx += ABS(( sol[0] + (16mx+8)sol[1] + (16my+8)sol[2] ) - mv.x );
2163 :	chl	770	meany += ABS(( sol[3] - (16mx+8)sol[2] + (16my+8)sol[1] ) - mv.y );
2164 :			}
2165 :	chl	769
2166 :	chl	770	if (4meanx > oldnum) / better fit than 0.25 is useless */
2167 :	chl	769	meanx /= oldnum;
2168 :	syskin	801	else
2169 :	chl	770	meanx = 0.25;
2170 :
2171 :			if (4*meany > oldnum)
2172 :	chl	769	meany /= oldnum;
2173 :	syskin	801	else
2174 :	chl	770	meany = 0.25;
2175 :	chl	769
2176 :	chl	770	/* fprintf(stderr,"sol = (%8.5f, %8.5f, %8.5f, %8.5f)\n",sol[0],sol[1],sol[2],sol[3]);
2177 :			fprintf(stderr,"meanx = %8.5f meany = %8.5f %d\n",meanx,meany, oldnum);
2178 :			*/
2179 :	chl	769	num = 0;
2180 :			for (my = 0; my < MBh; my++)
2181 :			for (mx = 0; mx < MBw; mx++)
2182 :			{
2183 :			const int mbnum = mx + my * MBw;
2184 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2185 :			const VECTOR mv = pMB->mvs[0];
2186 :
2187 :			if (!MBmask[mbnum])
2188 :			continue;
2189 :
2190 :			if ( ( ABS(( sol[0] + (16mx+8)sol[1] + (16my+8)sol[2] ) - mv.x ) > meanx )
2191 :	chl	770	\|\| ( ABS(( sol[3] - (16mx+8)sol[2] + (16my+8)sol[1] ) - mv.y ) > meany ) )
2192 :	chl	769	MBmask[mbnum]=0;
2193 :			else
2194 :			num++;
2195 :			}
2196 :
2197 :			} while ( (oldnum != num) && (num>=4) );
2198 :
2199 :			if (num < 4)
2200 :			{
2201 :			gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0;
2202 :			} else {
2203 :
2204 :			gmc.duv[0].x=(int)(sol[0]+0.5);
2205 :			gmc.duv[0].y=(int)(sol[3]+0.5);
2206 :
2207 :			gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5);
2208 :			gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5);
2209 :
2210 :			gmc.duv[2].x=0;
2211 :			gmc.duv[2].y=0;
2212 :	chl	619	}
2213 :	chl	769	// fprintf(stderr,"wp1 = ( %4d, %4d) wp2 = ( %4d, %4d) \n", gmc.duv[0].x, gmc.duv[0].y, gmc.duv[1].x, gmc.duv[1].y);
2214 :	chl	619
2215 :	chl	769	free(MBmask);
2216 :
2217 :	chl	619	return gmc;
2218 :			}

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