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

Revision 817 - (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	817	x = qpel ? x<<1 : x;
61 :			y = qpel ? y<<1 : y;
62 :	syskin	768	if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
63 :	syskin	817
64 :	syskin	748	x -= pred.x;
65 :			y -= pred.y;
66 :	syskin	768
67 :	syskin	748	if (x) {
68 :	syskin	768	x = ABS(x);
69 :	chl	530	x += (1 << (iFcode - 1)) - 1;
70 :			x >>= (iFcode - 1);
71 :			if (x > 32) x = 32;
72 :			xb = mvtab[x] + iFcode;
73 :	syskin	748	} else xb = 1;
74 :	Isibaar	3
75 :	syskin	748	if (y) {
76 :	syskin	768	y = ABS(y);
77 :	chl	530	y += (1 << (iFcode - 1)) - 1;
78 :			y >>= (iFcode - 1);
79 :			if (y > 32) y = 32;
80 :			yb = mvtab[y] + iFcode;
81 :	syskin	748	} else yb = 1;
82 :	chl	530	return xb + yb;
83 :	Isibaar	3	}
84 :
85 :	syskin	801	static int32_t ChromaSAD2(int fx, int fy, int bx, int by, const SearchData * const data)
86 :			{
87 :			int sad;
88 :			const uint32_t stride = data->iEdgedWidth/2;
89 :			uint8_t * f_refu = data->RefQ,
90 :			* f_refv = data->RefQ + 8,
91 :			* b_refu = data->RefQ + 16,
92 :			* b_refv = data->RefQ + 24;
93 :
94 :			switch (((fx & 1) << 1) \| (fy & 1)) {
95 :			case 0:
96 :			fx = fx / 2; fy = fy / 2;
97 :			f_refu = (uint8_t)data->RefCU + fy stride + fx, stride;
98 :			f_refv = (uint8_t)data->RefCV + fy stride + fx, stride;
99 :			break;
100 :			case 1:
101 :			fx = fx / 2; fy = (fy - 1) / 2;
102 :			interpolate8x8_halfpel_v(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
103 :			interpolate8x8_halfpel_v(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
104 :			break;
105 :			case 2:
106 :			fx = (fx - 1) / 2; fy = fy / 2;
107 :			interpolate8x8_halfpel_h(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
108 :			interpolate8x8_halfpel_h(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
109 :			break;
110 :			default:
111 :			fx = (fx - 1) / 2; fy = (fy - 1) / 2;
112 :			interpolate8x8_halfpel_hv(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
113 :			interpolate8x8_halfpel_hv(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
114 :			break;
115 :			}
116 :
117 :			switch (((bx & 1) << 1) \| (by & 1)) {
118 :			case 0:
119 :			bx = bx / 2; by = by / 2;
120 :			b_refu = (uint8_t)data->b_RefCU + by stride + bx, stride;
121 :			b_refv = (uint8_t)data->b_RefCV + by stride + bx, stride;
122 :			break;
123 :			case 1:
124 :			bx = bx / 2; by = (by - 1) / 2;
125 :			interpolate8x8_halfpel_v(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
126 :			interpolate8x8_halfpel_v(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
127 :			break;
128 :			case 2:
129 :			bx = (bx - 1) / 2; by = by / 2;
130 :			interpolate8x8_halfpel_h(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
131 :			interpolate8x8_halfpel_h(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
132 :			break;
133 :			default:
134 :			bx = (bx - 1) / 2; by = (by - 1) / 2;
135 :			interpolate8x8_halfpel_hv(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
136 :			interpolate8x8_halfpel_hv(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
137 :			break;
138 :			}
139 :
140 :			sad = sad8bi(data->CurU, b_refu, f_refu, stride);
141 :			sad += sad8bi(data->CurV, b_refv, f_refv, stride);
142 :
143 :			return sad;
144 :			}
145 :
146 :
147 :	syskin	748	static int32_t
148 :	syskin	628	ChromaSAD(int dx, int dy, const SearchData * const data)
149 :	Isibaar	3	{
150 :	syskin	628	int sad;
151 :	syskin	748	const uint32_t stride = data->iEdgedWidth/2;
152 :	Isibaar	3
153 :	syskin	712	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
154 :	syskin	768	data->temp[5] = dx; data->temp[6] = dy; // backup
155 :	syskin	712
156 :			switch (((dx & 1) << 1) \| (dy & 1)) {
157 :	syskin	628	case 0:
158 :	syskin	748	dx = dx / 2; dy = dy / 2;
159 :			sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride);
160 :			sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride);
161 :	syskin	628	break;
162 :			case 1:
163 :			dx = dx / 2; dy = (dy - 1) / 2;
164 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride);
165 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride);
166 :	syskin	628	break;
167 :			case 2:
168 :			dx = (dx - 1) / 2; dy = dy / 2;
169 :	syskin	748	sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride);
170 :			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride);
171 :	syskin	628	break;
172 :			default:
173 :			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
174 :	syskin	748	interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding);
175 :			sad = sad8(data->CurU, data->RefQ, stride);
176 :
177 :			interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding);
178 :			sad += sad8(data->CurV, data->RefQ, stride);
179 :	syskin	628	break;
180 :	chl	530	}
181 :	syskin	748	data->temp[7] = sad; //backup, part 2
182 :	syskin	628	return sad;
183 :			}
184 :	Isibaar	3
185 :	syskin	663	static __inline const uint8_t *
186 :	syskin	768	GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data)
187 :	Isibaar	606	{
188 :	syskin	663	// dir : 0 = forward, 1 = backward
189 :			switch ( (dir << 2) \| ((x&1)<<1) \| (y&1) ) {
190 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
191 :			case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
192 :			case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
193 :			case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
194 :			case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
195 :			case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
196 :			case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
197 :			default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
198 :	Isibaar	606	}
199 :			}
200 :
201 :	syskin	748	// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate
202 :			static __inline const uint8_t *
203 :			GetReference(const int x, const int y, const SearchData * const data)
204 :			{
205 :			switch ( ((x&1)<<1) \| (y&1) ) {
206 :			case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
207 :	syskin	817	case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
208 :	syskin	748	case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
209 :	syskin	817	default : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); //case 2
210 :	syskin	748	}
211 :			}
212 :
213 :			static uint8_t *
214 :	syskin	768	Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data)
215 :	chl	530	{
216 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
217 :	syskin	768	uint8_t * Reference = data->RefQ + 16*dir;
218 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
219 :	syskin	663	const uint32_t rounding = data->rounding;
220 :			const int halfpel_x = x/2;
221 :			const int halfpel_y = y/2;
222 :			const uint8_t ref1, ref2, ref3, ref4;
223 :	chl	326
224 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
225 :	syskin	663	ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
226 :			switch( ((x&1)<<1) + (y&1) ) {
227 :			case 0: // pure halfpel position
228 :	syskin	740	return (uint8_t *) ref1;
229 :	syskin	663	break;
230 :	Isibaar	3
231 :	syskin	663	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
232 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
233 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
234 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
235 :	syskin	663	break;
236 :	Isibaar	3
237 :	syskin	663	case 2: // x qpel, y halfpel - left or right during qpel refinement
238 :	syskin	748	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
239 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
240 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
241 :	syskin	663	break;
242 :	Isibaar	3
243 :	syskin	663	default: // x and y in qpel resolution - the "corners" (top left/right and
244 :			// bottom left/right) during qpel refinement
245 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
246 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
247 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
248 :	syskin	663	ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
249 :			ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
250 :			ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
251 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
252 :			break;
253 :			}
254 :			return Reference;
255 :	Isibaar	3	}
256 :
257 :	syskin	748	static uint8_t *
258 :	syskin	768	Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data)
259 :	Isibaar	580	{
260 :	syskin	663	// create or find a qpel-precision reference picture; return pointer to it
261 :	syskin	768	uint8_t * Reference = data->RefQ + 16*dir;
262 :			const uint32_t iEdgedWidth = data->iEdgedWidth;
263 :	syskin	663	const uint32_t rounding = data->rounding;
264 :			const int halfpel_x = x/2;
265 :			const int halfpel_y = y/2;
266 :	Isibaar	580	const uint8_t ref1, ref2, ref3, ref4;
267 :
268 :	syskin	748	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
269 :	syskin	663	switch( ((x&1)<<1) + (y&1) ) {
270 :	syskin	817	case 3: // x and y in qpel resolution - the "corners" (top left/right and
271 :			// bottom left/right) during qpel refinement
272 :			ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
273 :			ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
274 :			ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
275 :			interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
276 :			interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
277 :			interpolate8x8_avg4(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8iEdgedWidth, ref3+8iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
278 :			interpolate8x8_avg4(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8iEdgedWidth+8, ref3+8iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
279 :			break;
280 :
281 :	Isibaar	580	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
282 :	syskin	748	ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
283 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
284 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
285 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
286 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
287 :	Isibaar	580	break;
288 :
289 :			case 2: // x qpel, y halfpel - left or right during qpel refinement
290 :	syskin	768	ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
291 :	Isibaar	665	interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
292 :			interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8);
293 :			interpolate8x8_avg2(Reference+8iEdgedWidth, ref1+8iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8);
294 :			interpolate8x8_avg2(Reference+8iEdgedWidth+8, ref1+8iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
295 :	Isibaar	580	break;
296 :
297 :	syskin	817	case 0: // pure halfpel position
298 :			return (uint8_t *) ref1;
299 :	Isibaar	580	}
300 :	syskin	663	return Reference;
301 :			}
302 :
303 :			/* CHECK_CANDIATE FUNCTIONS START */
304 :
305 :	syskin	748	static void
306 :	syskin	663	CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
307 :			{
308 :	syskin	817	int xc, yc;
309 :	syskin	663	const uint8_t * Reference;
310 :			VECTOR * current;
311 :	syskin	817	int32_t sad; uint32_t t;
312 :	syskin	663
313 :	syskin	817	if ( (x > data->max_dx) \|\| (x < data->min_dx)
314 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
315 :	syskin	663
316 :	syskin	817	if (!data->qpel_precision) {
317 :			Reference = GetReference(x, y, data);
318 :			current = data->currentMV;
319 :			xc = x; yc = y;
320 :			} else { // x and y are in 1/4 precision
321 :	syskin	663	Reference = Interpolate16x16qpel(x, y, 0, data);
322 :			xc = x/2; yc = y/2; //for chroma sad
323 :			current = data->currentQMV;
324 :			}
325 :	syskin	817
326 :			sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
327 :	syskin	768	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
328 :
329 :	syskin	817	sad += (data->lambda16 * t * sad)>>10;
330 :	syskin	768	data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
331 :	Isibaar	580
332 :	syskin	817	if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
333 :	syskin	753	(yc >> 1) + roundtab_79[yc & 0x3], data);
334 :	syskin	628
335 :	syskin	817	if (sad < data->iMinSAD[0]) {
336 :			data->iMinSAD[0] = sad;
337 :	syskin	663	current[0].x = x; current[0].y = y;
338 :	syskin	817	*dir = Direction;
339 :			}
340 :	Isibaar	580
341 :			if (data->temp[1] < data->iMinSAD[1]) {
342 :	syskin	817	data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y = y; }
343 :	Isibaar	580	if (data->temp[2] < data->iMinSAD[2]) {
344 :	syskin	663	data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
345 :	Isibaar	580	if (data->temp[3] < data->iMinSAD[3]) {
346 :	syskin	663	data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
347 :	Isibaar	580	if (data->temp[4] < data->iMinSAD[4]) {
348 :	syskin	663	data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
349 :
350 :	Isibaar	580	}
351 :
352 :	syskin	748	static void
353 :	syskin	817	CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
354 :			{
355 :			int32_t sad; uint32_t t;
356 :			const uint8_t * Reference;
357 :
358 :			if ( (x > data->max_dx) \|\| (x < data->min_dx)
359 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
360 :
361 :			if (!data->qpel_precision) Reference = GetReference(x, y, data);
362 :			else Reference = Interpolate16x16qpel(x, y, 0, data);
363 :
364 :			sad = sad8(data->Cur, Reference, data->iEdgedWidth);
365 :			t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
366 :
367 :			sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10;
368 :
369 :			if (sad < *(data->iMinSAD)) {
370 :			*(data->iMinSAD) = sad;
371 :			data->currentMV->x = x; data->currentMV->y = y;
372 :			*dir = Direction;
373 :			}
374 :			}
375 :
376 :
377 :			static void
378 :	syskin	704	CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
379 :			{
380 :	syskin	768	uint32_t t;
381 :	syskin	704	const uint8_t * Reference;
382 :
383 :	syskin	817	if ( (!(x&1) && x !=0) \|\| (!(y&1) && y !=0) \|\| //non-zero integer value
384 :			(x > data->max_dx) \|\| (x < data->min_dx)
385 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
386 :	syskin	704
387 :	syskin	748	Reference = GetReference(x, y, data);
388 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
389 :	syskin	704
390 :			data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
391 :
392 :	syskin	768	data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10;
393 :			data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
394 :	syskin	704
395 :			if (data->temp[0] < data->iMinSAD[0]) {
396 :			data->iMinSAD[0] = data->temp[0];
397 :			data->currentMV[0].x = x; data->currentMV[0].y = y;
398 :			*dir = Direction; }
399 :
400 :			if (data->temp[1] < data->iMinSAD[1]) {
401 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
402 :			if (data->temp[2] < data->iMinSAD[2]) {
403 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
404 :			if (data->temp[3] < data->iMinSAD[3]) {
405 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
406 :			if (data->temp[4] < data->iMinSAD[4]) {
407 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
408 :			}
409 :
410 :	syskin	748	static void
411 :	syskin	663	CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
412 :	syskin	658	{
413 :	syskin	801	int32_t sad, xc, yc;
414 :	syskin	663	const uint8_t * Reference;
415 :	syskin	768	uint32_t t;
416 :	syskin	663	VECTOR * current;
417 :	syskin	658
418 :	syskin	768	if ( (x > data->max_dx) \| ( x < data->min_dx)
419 :			\| (y > data->max_dy) \| (y < data->min_dy) ) return;
420 :	syskin	658
421 :	syskin	768	if (data->rrv && (!(x&1) && x !=0) \| (!(y&1) && y !=0) ) return; //non-zero even value
422 :	syskin	704
423 :	syskin	663	if (data->qpel_precision) { // x and y are in 1/4 precision
424 :			Reference = Interpolate16x16qpel(x, y, 0, data);
425 :			current = data->currentQMV;
426 :	syskin	801	xc = x/2; yc = y/2;
427 :	syskin	663	} else {
428 :	syskin	748	Reference = GetReference(x, y, data);
429 :	syskin	663	current = data->currentMV;
430 :	syskin	801	xc = x; yc = y;
431 :	syskin	658	}
432 :	syskin	724	t = d_mv_bits(x, y, data->predMV, data->iFcode,
433 :	syskin	768	data->qpel^data->qpel_precision, data->rrv);
434 :
435 :	syskin	658	sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
436 :	syskin	768	sad += (data->lambda16 * t * sad)>>10;
437 :	syskin	658
438 :	syskin	801	if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
439 :			(yc >> 1) + roundtab_79[yc & 0x3], data);
440 :
441 :
442 :	syskin	663	if (sad < *(data->iMinSAD)) {
443 :			*(data->iMinSAD) = sad;
444 :			current->x = x; current->y = y;
445 :	syskin	801	*dir = Direction;
446 :			}
447 :	syskin	658	}
448 :
449 :	syskin	748	static void
450 :	syskin	739	CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
451 :	Isibaar	539	{
452 :	syskin	658	// maximum speed - for P/B/I decision
453 :	syskin	817	int32_t sad;
454 :	Isibaar	539
455 :	syskin	817	if ( (x > data->max_dx) \|\| (x < data->min_dx)
456 :			\|\| (y > data->max_dy) \|\| (y < data->min_dy) ) return;
457 :	Isibaar	539
458 :	syskin	817	sad = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
459 :	syskin	739	data->iEdgedWidth, data->temp+1);
460 :	syskin	768
461 :	syskin	817	if (sad < *(data->iMinSAD)) {
462 :			*(data->iMinSAD) = sad;
463 :	Isibaar	539	data->currentMV[0].x = x; data->currentMV[0].y = y;
464 :	syskin	817	*dir = Direction;
465 :			}
466 :	syskin	739	if (data->temp[1] < data->iMinSAD[1]) {
467 :			data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
468 :			if (data->temp[2] < data->iMinSAD[2]) {
469 :			data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
470 :			if (data->temp[3] < data->iMinSAD[3]) {
471 :			data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
472 :			if (data->temp[4] < data->iMinSAD[4]) {
473 :			data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
474 :
475 :	Isibaar	539	}
476 :
477 :	syskin	748	static void
478 :	chl	530	CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
479 :			{
480 :	syskin	801	int32_t sad, xb, yb, xcf, ycf, xcb, ycb;
481 :	syskin	768	uint32_t t;
482 :	chl	530	const uint8_t ReferenceF, ReferenceB;
483 :	syskin	663	VECTOR *current;
484 :	Isibaar	3
485 :	syskin	768	if ( (xf > data->max_dx) \| (xf < data->min_dx)
486 :			\| (yf > data->max_dy) \| (yf < data->min_dy) ) return;
487 :	Isibaar	3
488 :	syskin	768	if (!data->qpel_precision) {
489 :	syskin	748	ReferenceF = GetReference(xf, yf, data);
490 :	syskin	663	xb = data->currentMV[1].x; yb = data->currentMV[1].y;
491 :	syskin	748	ReferenceB = GetReferenceB(xb, yb, 1, data);
492 :	syskin	663	current = data->currentMV;
493 :	syskin	801	xcf = xf; ycf = yf;
494 :			xcb = xb; ycb = yb;
495 :	syskin	768	} else {
496 :			ReferenceF = Interpolate16x16qpel(xf, yf, 0, data);
497 :			xb = data->currentQMV[1].x; yb = data->currentQMV[1].y;
498 :			current = data->currentQMV;
499 :			ReferenceB = Interpolate16x16qpel(xb, yb, 1, data);
500 :	syskin	801	xcf = xf/2; ycf = yf/2;
501 :			xcb = xb/2; ycb = yb/2;
502 :	chl	530	}
503 :	edgomez	195
504 :	syskin	768	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0)
505 :			+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0);
506 :
507 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
508 :	syskin	768	sad += (data->lambda16 * t * sad)>>10;
509 :	syskin	628
510 :	syskin	801	if (data->chroma) sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3],
511 :			(ycf >> 1) + roundtab_79[ycf & 0x3],
512 :			(xcb >> 1) + roundtab_79[xcb & 0x3],
513 :			(ycb >> 1) + roundtab_79[ycb & 0x3], data);
514 :
515 :	chl	530	if (sad < *(data->iMinSAD)) {
516 :			*(data->iMinSAD) = sad;
517 :	syskin	663	current->x = xf; current->y = yf;
518 :	syskin	801	*dir = Direction;
519 :			}
520 :	Isibaar	3	}
521 :
522 :	syskin	748	static void
523 :	chl	530	CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
524 :	Isibaar	3	{
525 :	syskin	801	int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0;
526 :	syskin	768	uint32_t k;
527 :	chl	530	const uint8_t *ReferenceF;
528 :			const uint8_t *ReferenceB;
529 :			VECTOR mvs, b_mvs;
530 :	Isibaar	3
531 :	syskin	768	if (( x > 31) \| ( x < -32) \| ( y > 31) \| (y < -32)) return;
532 :	Isibaar	3
533 :	chl	530	for (k = 0; k < 4; k++) {
534 :			mvs.x = data->directmvF[k].x + x;
535 :			b_mvs.x = ((x == 0) ?
536 :			data->directmvB[k].x
537 :			: mvs.x - data->referencemv[k].x);
538 :	Isibaar	3
539 :	chl	530	mvs.y = data->directmvF[k].y + y;
540 :			b_mvs.y = ((y == 0) ?
541 :			data->directmvB[k].y
542 :			: mvs.y - data->referencemv[k].y);
543 :	syskin	768
544 :	syskin	801	if ( (mvs.x > data->max_dx) \| (mvs.x < data->min_dx)
545 :			\| (mvs.y > data->max_dy) \| (mvs.y < data->min_dy)
546 :			\| (b_mvs.x > data->max_dx) \| (b_mvs.x < data->min_dx)
547 :			\| (b_mvs.y > data->max_dy) \| (b_mvs.y < data->min_dy) ) return;
548 :	Isibaar	3
549 :	syskin	801	if (data->qpel) {
550 :			xcf += mvs.x/2; ycf += mvs.y/2;
551 :			xcb += b_mvs.x/2; ycb += b_mvs.y/2;
552 :			} else {
553 :			xcf += mvs.x; ycf += mvs.y;
554 :			xcb += b_mvs.x; ycb += b_mvs.y;
555 :			mvs.x = 2; mvs.y = 2; //we move to qpel precision anyway
556 :			b_mvs.x = 2; b_mvs.y = 2;
557 :			}
558 :
559 :	syskin	663	ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
560 :			ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
561 :	syskin	768
562 :	chl	530	sad += sad8bi(data->Cur + 8(k&1) + 8(k>>1)*(data->iEdgedWidth),
563 :	syskin	768	ReferenceF, ReferenceB, data->iEdgedWidth);
564 :	chl	530	if (sad > *(data->iMinSAD)) return;
565 :	edgomez	78	}
566 :	chl	530
567 :	syskin	768	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
568 :	syskin	628
569 :	syskin	801	if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
570 :			(ycf >> 3) + roundtab_76[ycf & 0xf],
571 :			(xcb >> 3) + roundtab_76[xcb & 0xf],
572 :			(ycb >> 3) + roundtab_76[ycb & 0xf], data);
573 :
574 :	chl	530	if (sad < *(data->iMinSAD)) {
575 :			*(data->iMinSAD) = sad;
576 :			data->currentMV->x = x; data->currentMV->y = y;
577 :	syskin	801	*dir = Direction;
578 :			}
579 :	Isibaar	3	}
580 :
581 :	syskin	748	static void
582 :	chl	530	CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
583 :	chl	96	{
584 :	syskin	801	int32_t sad, xcf, ycf, xcb, ycb;
585 :	chl	530	const uint8_t *ReferenceF;
586 :			const uint8_t *ReferenceB;
587 :			VECTOR mvs, b_mvs;
588 :	chl	96
589 :	syskin	768	if (( x > 31) \| ( x < -32) \| ( y > 31) \| (y < -32)) return;
590 :	edgomez	195
591 :	chl	530	mvs.x = data->directmvF[0].x + x;
592 :			b_mvs.x = ((x == 0) ?
593 :			data->directmvB[0].x
594 :			: mvs.x - data->referencemv[0].x);
595 :	edgomez	195
596 :	chl	530	mvs.y = data->directmvF[0].y + y;
597 :			b_mvs.y = ((y == 0) ?
598 :			data->directmvB[0].y
599 :			: mvs.y - data->referencemv[0].y);
600 :	syskin	768
601 :	syskin	801	if ( (mvs.x > data->max_dx) \| (mvs.x < data->min_dx)
602 :			\| (mvs.y > data->max_dy) \| (mvs.y < data->min_dy)
603 :			\| (b_mvs.x > data->max_dx) \| (b_mvs.x < data->min_dx)
604 :			\| (b_mvs.y > data->max_dy) \| (b_mvs.y < data->min_dy) ) return;
605 :	chl	96
606 :	syskin	801	if (data->qpel) {
607 :			xcf = 4(mvs.x/2); ycf = 4(mvs.y/2);
608 :			xcb = 4(b_mvs.x/2); ycb = 4(b_mvs.y/2);
609 :			ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data);
610 :			ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
611 :			} else {
612 :			xcf = 4mvs.x; ycf = 4mvs.y;
613 :			xcb = 4b_mvs.x; ycb = 4b_mvs.y;
614 :			ReferenceF = GetReference(mvs.x, mvs.y, data);
615 :			ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data);
616 :			}
617 :	syskin	768
618 :	syskin	628	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
619 :	syskin	768	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10;
620 :	chl	96
621 :	syskin	801	if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
622 :			(ycf >> 3) + roundtab_76[ycf & 0xf],
623 :			(xcb >> 3) + roundtab_76[xcb & 0xf],
624 :			(ycb >> 3) + roundtab_76[ycb & 0xf], data);
625 :
626 :	chl	530	if (sad < *(data->iMinSAD)) {
627 :			*(data->iMinSAD) = sad;
628 :			data->currentMV->x = x; data->currentMV->y = y;
629 :	syskin	801	*dir = Direction;
630 :			}
631 :	chl	530	}
632 :	chl	96
633 :	Isibaar	539	/* CHECK_CANDIATE FUNCTIONS END */
634 :	chl	96
635 :	chl	530	/* MAINSEARCH FUNCTIONS START */
636 :	edgomez	195
637 :	chl	530	static void
638 :			AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
639 :	chl	181	{
640 :
641 :			/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
642 :
643 :	syskin	712	int iDirection;
644 :	edgomez	195
645 :	syskin	712	for(;;) { //forever
646 :			iDirection = 0;
647 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
648 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
649 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
650 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
651 :	chl	181
652 :	syskin	712	/* now we're doing diagonal checks near our candidate */
653 :	chl	181
654 :	syskin	768	if (iDirection) { //if anything found
655 :	syskin	712	bDirection = iDirection;
656 :			iDirection = 0;
657 :			x = data->currentMV->x; y = data->currentMV->y;
658 :			if (bDirection & 3) { //our candidate is left or right
659 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
660 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
661 :			} else { // what remains here is up or down
662 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
663 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
664 :	syskin	712	}
665 :	chl	181
666 :	syskin	712	if (iDirection) {
667 :			bDirection += iDirection;
668 :	syskin	748	x = data->currentMV->x; y = data->currentMV->y;
669 :	chl	181	}
670 :	syskin	712	} else { //about to quit, eh? not so fast....
671 :			switch (bDirection) {
672 :			case 2:
673 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
674 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
675 :			break;
676 :			case 1:
677 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
678 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
679 :			break;
680 :			case 2 + 4:
681 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
682 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
683 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
684 :			break;
685 :			case 4:
686 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
687 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
688 :			break;
689 :			case 8:
690 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
691 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
692 :			break;
693 :			case 1 + 4:
694 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
695 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
696 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
697 :			break;
698 :			case 2 + 8:
699 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
700 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
701 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
702 :			break;
703 :			case 1 + 8:
704 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
705 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
706 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
707 :			break;
708 :			default: //1+2+4+8 == we didn't find anything at all
709 :			CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
710 :			CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
711 :			CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
712 :			CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
713 :			break;
714 :			}
715 :			if (!iDirection) break; //ok, the end. really
716 :			bDirection = iDirection;
717 :			x = data->currentMV->x; y = data->currentMV->y;
718 :	chl	181	}
719 :	syskin	712	}
720 :	chl	181	}
721 :
722 :	chl	530	static void
723 :			SquareSearch(int x, int y, const SearchData * const data, int bDirection)
724 :	chl	326	{
725 :	chl	530	int iDirection;
726 :	chl	326
727 :	chl	530	do {
728 :			iDirection = 0;
729 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
730 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
731 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
732 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
733 :			if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
734 :			if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
735 :			if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
736 :			if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
737 :	chl	326
738 :	chl	530	bDirection = iDirection;
739 :			x = data->currentMV->x; y = data->currentMV->y;
740 :			} while (iDirection);
741 :	chl	346	}
742 :
743 :	chl	530	static void
744 :			DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
745 :	chl	346	{
746 :
747 :	chl	181	/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
748 :
749 :	syskin	712	int iDirection;
750 :	edgomez	195
751 :	syskin	712	do {
752 :			iDirection = 0;
753 :			if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
754 :			if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
755 :			if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
756 :			if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
757 :	chl	181
758 :	syskin	712	/* now we're doing diagonal checks near our candidate */
759 :	chl	181
760 :	syskin	712	if (iDirection) { //checking if anything found
761 :			bDirection = iDirection;
762 :			iDirection = 0;
763 :			x = data->currentMV->x; y = data->currentMV->y;
764 :			if (bDirection & 3) { //our candidate is left or right
765 :			CHECK_CANDIDATE(x, y + iDiamondSize, 8);
766 :			CHECK_CANDIDATE(x, y - iDiamondSize, 4);
767 :			} else { // what remains here is up or down
768 :			CHECK_CANDIDATE(x + iDiamondSize, y, 2);
769 :	syskin	748	CHECK_CANDIDATE(x - iDiamondSize, y, 1);
770 :	chl	181	}
771 :	syskin	712	bDirection += iDirection;
772 :			x = data->currentMV->x; y = data->currentMV->y;
773 :	chl	181	}
774 :	syskin	712	}
775 :			while (iDirection);
776 :	chl	181	}
777 :
778 :	chl	530	/* MAINSEARCH FUNCTIONS END */
779 :	chl	181
780 :	chl	530	static void
781 :	syskin	663	SubpelRefine(const SearchData * const data)
782 :	chl	96	{
783 :	syskin	663	/* Do a half-pel or q-pel refinement */
784 :	syskin	768	const VECTOR centerMV = data->qpel_precision ? data->currentQMV : data->currentMV;
785 :			int iDirection; //only needed because macro expects it
786 :	chl	96
787 :	syskin	768	CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0);
788 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0);
789 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, 0);
790 :			CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, 0);
791 :			CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, 0);
792 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, 0);
793 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, 0);
794 :			CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, 0);
795 :	Isibaar	580	}
796 :
797 :	chl	530	static __inline int
798 :			SkipDecisionP(const IMAGE * current, const IMAGE * reference,
799 :			const int x, const int y,
800 :	syskin	768	const uint32_t stride, const uint32_t iQuant, int rrv)
801 :	chl	96
802 :	Isibaar	3	{
803 :	syskin	768	if(!rrv) {
804 :			uint32_t sadC = sad8(current->u + x8 + ystride*8,
805 :			reference->u + x8 + ystride*8, stride);
806 :	syskin	704	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
807 :	syskin	768	sadC += sad8(current->v + (x + ystride)8,
808 :			reference->v + (x + ystride)8, stride);
809 :	syskin	704	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
810 :			return 1;
811 :	syskin	768
812 :			} else {
813 :			uint32_t sadC = sad16(current->u + x16 + ystride*16,
814 :			reference->u + x16 + ystride16, stride, 2564096);
815 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
816 :			sadC += sad16(current->v + (x + ystride)16,
817 :			reference->v + (x + ystride)16, stride, 256*4096);
818 :			if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
819 :			return 1;
820 :	syskin	704	}
821 :	chl	530	}
822 :	edgomez	195
823 :	chl	530	static __inline void
824 :			SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
825 :			{
826 :			pMB->mode = MODE_NOT_CODED;
827 :	syskin	768	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV;
828 :			pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV;
829 :	chl	530	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
830 :	Isibaar	3	}
831 :
832 :	syskin	628	bool
833 :	chl	530	MotionEstimation(MBParam * const pParam,
834 :			FRAMEINFO * const current,
835 :			FRAMEINFO * const reference,
836 :			const IMAGE * const pRefH,
837 :			const IMAGE * const pRefV,
838 :			const IMAGE * const pRefHV,
839 :			const uint32_t iLimit)
840 :	Isibaar	3	{
841 :	chl	530	MACROBLOCK *const pMBs = current->mbs;
842 :			const IMAGE *const pCurrent = &current->image;
843 :			const IMAGE *const pRef = &reference->image;
844 :	Isibaar	3
845 :	syskin	704	uint32_t mb_width = pParam->mb_width;
846 :			uint32_t mb_height = pParam->mb_height;
847 :	syskin	768	const uint32_t iEdgedWidth = pParam->edged_width;
848 :	suxen_drol	702
849 :	chl	530	uint32_t x, y;
850 :			uint32_t iIntra = 0;
851 :	syskin	628	int32_t InterBias, quant = current->quant, sad00;
852 :	edgomez	195
853 :	Isibaar	539	// some pre-initialized thingies for SearchP
854 :	syskin	712	int32_t temp[8];
855 :	Isibaar	539	VECTOR currentMV[5];
856 :	Isibaar	580	VECTOR currentQMV[5];
857 :	Isibaar	539	int32_t iMinSAD[5];
858 :			SearchData Data;
859 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
860 :	syskin	768	Data.iEdgedWidth = iEdgedWidth;
861 :	Isibaar	539	Data.currentMV = currentMV;
862 :	Isibaar	580	Data.currentQMV = currentQMV;
863 :	Isibaar	539	Data.iMinSAD = iMinSAD;
864 :			Data.temp = temp;
865 :			Data.iFcode = current->fcode;
866 :	Isibaar	580	Data.rounding = pParam->m_rounding_type;
867 :	syskin	628	Data.qpel = pParam->m_quarterpel;
868 :	syskin	801	Data.chroma = current->motion_flags & PMV_CHROMA16;
869 :	syskin	704	Data.rrv = current->global_flags & XVID_REDUCED;
870 :	Isibaar	539
871 :	syskin	704	if ((current->global_flags & XVID_REDUCED)) {
872 :	suxen_drol	702	mb_width = (pParam->width + 31) / 32;
873 :			mb_height = (pParam->height + 31) / 32;
874 :	syskin	704	Data.qpel = Data.chroma = 0;
875 :	suxen_drol	702	}
876 :
877 :	syskin	754	Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose)
878 :	chl	530	if (sadInit) (*sadInit) ();
879 :	edgomez	195
880 :	suxen_drol	702	for (y = 0; y < mb_height; y++) {
881 :			for (x = 0; x < mb_width; x++) {
882 :	chl	530	MACROBLOCK pMB = &pMBs[x + y pParam->mb_width];
883 :	syskin	768
884 :			if (!Data.rrv) pMB->sad16 =
885 :			sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16,
886 :			pRef->y + (x + y * iEdgedWidth) * 16,
887 :	syskin	704	pParam->edged_width, pMB->sad8 );
888 :	syskin	768
889 :	syskin	748	else pMB->sad16 =
890 :	syskin	768	sad32v_c(pCurrent->y + (x + y * iEdgedWidth) * 32,
891 :			pRef->y + (x + y * iEdgedWidth) * 32,
892 :	chl	530	pParam->edged_width, pMB->sad8 );
893 :	edgomez	195
894 :	syskin	628	if (Data.chroma) {
895 :	syskin	768	Data.temp[7] = sad8(pCurrent->u + x8 + y(iEdgedWidth/2)*8,
896 :			pRef->u + x8 + y(iEdgedWidth/2)*8, iEdgedWidth/2)
897 :			+ sad8(pCurrent->v + (x + y(iEdgedWidth/2))8,
898 :			pRef->v + (x + y(iEdgedWidth/2))8, iEdgedWidth/2);
899 :			pMB->sad16 += Data.temp[7];
900 :	syskin	628	}
901 :
902 :	syskin	768	sad00 = pMB->sad16;
903 :	syskin	628
904 :	chl	530	if (!(current->global_flags & XVID_LUMIMASKING)) {
905 :			pMB->dquant = NO_CHANGE;
906 :	syskin	601	} else {
907 :	syskin	574	if (pMB->dquant != NO_CHANGE) {
908 :			quant += DQtab[pMB->dquant];
909 :			if (quant > 31) quant = 31;
910 :			else if (quant < 1) quant = 1;
911 :			}
912 :	syskin	601	}
913 :	syskin	768	pMB->quant = current->quant;
914 :	chl	181
915 :	syskin	628	//initial skip decision
916 :	chl	769	/* no early skip for GMC (global vector = skip vector is unknown!) */
917 :			if (!(current->global_flags & XVID_GMC)) { /* no fast SKIP for S(GMC)-VOPs */
918 :	syskin	801	if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
919 :	syskin	768	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
920 :	chl	619	SkipMacroblockP(pMB, sad00);
921 :			continue;
922 :			}
923 :			}
924 :	syskin	768
925 :	syskin	628	SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
926 :	chl	530	y, current->motion_flags, pMB->quant,
927 :	Isibaar	539	&Data, pParam, pMBs, reference->mbs,
928 :	chl	530	current->global_flags & XVID_INTER4V, pMB);
929 :	edgomez	195
930 :	chl	530	/* final skip decision, a.k.a. "the vector you found, really that good?" */
931 :	chl	769	if (!(current->global_flags & XVID_GMC)) {
932 :	chl	619	if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
933 :	syskin	704	&& ((100pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH (Data.rrv ? 4:1)) )
934 :	syskin	768	if (Data.chroma \|\| SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
935 :	chl	619	SkipMacroblockP(pMB, sad00);
936 :			continue;
937 :			}
938 :			}
939 :	syskin	768
940 :	chl	530	/* finally, intra decision */
941 :	Isibaar	3
942 :	chl	530	InterBias = MV16_INTER_BIAS;
943 :	syskin	768	if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work
944 :	chl	530	if (y != 0)
945 :	syskin	639	if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
946 :	chl	530	if (x != 0)
947 :	syskin	639	if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
948 :	syskin	768
949 :	syskin	628	if (Data.chroma) InterBias += 50; // to compensate bigger SAD
950 :	syskin	768	if (Data.rrv) InterBias *= 4;
951 :	Isibaar	3
952 :	syskin	768	if (InterBias < pMB->sad16) {
953 :	syskin	704	int32_t deviation;
954 :	syskin	768	if (!Data.rrv)
955 :			deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 16, iEdgedWidth);
956 :			else {
957 :			deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 32, iEdgedWidth)
958 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16, iEdgedWidth)
959 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * iEdgedWidth, iEdgedWidth)
960 :			+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * (iEdgedWidth+1), iEdgedWidth);
961 :			}
962 :			if (deviation < (pMB->sad16 - InterBias)) {
963 :	syskin	754	if (++iIntra >= iLimit) return 1;
964 :	syskin	768	SkipMacroblockP(pMB, 0); //same thing
965 :	chl	530	pMB->mode = MODE_INTRA;
966 :			}
967 :			}
968 :			}
969 :	Isibaar	3	}
970 :	chl	619
971 :	chl	770	if (current->global_flags & XVID_GMC ) /* GMC only for S(GMC)-VOPs */
972 :	chl	769	{
973 :			current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV);
974 :			}
975 :
976 :	syskin	628	return 0;
977 :	chl	530	}
978 :	Isibaar	3
979 :	edgomez	195
980 :	chl	530	static __inline int
981 :			make_mask(const VECTOR * const pmv, const int i)
982 :			{
983 :	Isibaar	539	int mask = 255, j;
984 :	chl	530	for (j = 0; j < i; j++) {
985 :			if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
986 :			if (pmv[i].x == pmv[j].x) {
987 :	syskin	768	if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4;
988 :			else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8;
989 :	syskin	748	} else
990 :	chl	530	if (pmv[i].y == pmv[j].y) {
991 :	syskin	768	if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1;
992 :			else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2;
993 :	chl	530	}
994 :	Isibaar	3	}
995 :	chl	530	return mask;
996 :			}
997 :	edgomez	195
998 :	syskin	748	static __inline void
999 :	syskin	704	PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
1000 :			int iHcount, const MACROBLOCK * const prevMB, int rrv)
1001 :	chl	530	{
1002 :	edgomez	195
1003 :	chl	530	//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
1004 :	syskin	704	if (rrv) { iWcount /= 2; iHcount /= 2; }
1005 :	edgomez	195
1006 :	syskin	704	if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour
1007 :	chl	530	pmv[5].x = EVEN(pmv[3].x);
1008 :	syskin	748	pmv[5].y = EVEN(pmv[3].y);
1009 :	Isibaar	539	} else pmv[5].x = pmv[5].y = 0;
1010 :	chl	169
1011 :	chl	530	if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
1012 :			else pmv[3].x = pmv[3].y = 0;
1013 :	Isibaar	3
1014 :	chl	530	if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
1015 :	syskin	768	else pmv[4].x = pmv[4].y = 0;
1016 :	chl	169
1017 :	chl	530	// [1] median prediction
1018 :	syskin	767	pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
1019 :	chl	169
1020 :	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
1021 :	chl	169
1022 :	chl	530	pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
1023 :			pmv[2].y = EVEN(prevMB->mvs[0].y);
1024 :	chl	169
1025 :	syskin	704	if ((x < iWcount-1) && (y < iHcount-1)) {
1026 :	chl	530	pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
1027 :	syskin	748	pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
1028 :	Isibaar	539	} else pmv[6].x = pmv[6].y = 0;
1029 :	syskin	704
1030 :			if (rrv) {
1031 :			int i;
1032 :			for (i = 0; i < 7; i++) {
1033 :	syskin	768	pmv[i].x = RRV_MV_SCALEUP(pmv[i].x);
1034 :	syskin	767	pmv[i].y = RRV_MV_SCALEUP(pmv[i].y);
1035 :	syskin	704	}
1036 :			}
1037 :	chl	530	}
1038 :	chl	169
1039 :	chl	530	static void
1040 :	syskin	628	SearchP(const IMAGE * const pRef,
1041 :	chl	530	const uint8_t * const pRefH,
1042 :			const uint8_t * const pRefV,
1043 :			const uint8_t * const pRefHV,
1044 :			const IMAGE * const pCur,
1045 :			const int x,
1046 :			const int y,
1047 :			const uint32_t MotionFlags,
1048 :			const uint32_t iQuant,
1049 :	Isibaar	539	SearchData * const Data,
1050 :	chl	530	const MBParam * const pParam,
1051 :			const MACROBLOCK * const pMBs,
1052 :			const MACROBLOCK * const prevMBs,
1053 :			int inter4v,
1054 :			MACROBLOCK * const pMB)
1055 :			{
1056 :	chl	169
1057 :	chl	530	int i, iDirection = 255, mask, threshA;
1058 :	Isibaar	539	VECTOR pmv[7];
1059 :	Isibaar	3
1060 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1061 :			pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
1062 :	Isibaar	3
1063 :	syskin	768	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp);
1064 :	syskin	628
1065 :	syskin	768	Data->temp[5] = Data->temp[6] = 0; // chroma-sad cache
1066 :			i = Data->rrv ? 2 : 1;
1067 :	syskin	704	Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i;
1068 :			Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1069 :			Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1070 :	Isibaar	3
1071 :	syskin	704	Data->Ref = pRef->y + (x + Data->iEdgedWidthy) 16*i;
1072 :			Data->RefH = pRefH + (x + Data->iEdgedWidthy) 16*i;
1073 :			Data->RefV = pRefV + (x + Data->iEdgedWidthy) 16*i;
1074 :			Data->RefHV = pRefHV + (x + Data->iEdgedWidthy) 16*i;
1075 :			Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1076 :			Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
1077 :
1078 :	syskin	601	Data->lambda16 = lambda_vec16[iQuant];
1079 :	Isibaar	606	Data->lambda8 = lambda_vec8[iQuant];
1080 :	syskin	663	Data->qpel_precision = 0;
1081 :	chl	169
1082 :	edgomez	568	if (pMB->dquant != NO_CHANGE) inter4v = 0;
1083 :	edgomez	195
1084 :	syskin	768	for(i = 0; i < 5; i++)
1085 :	Isibaar	580	Data->currentMV[i].x = Data->currentMV[i].y = 0;
1086 :	Isibaar	539
1087 :	syskin	768	if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1088 :	syskin	663	else Data->predMV = pmv[0];
1089 :	Isibaar	606
1090 :	syskin	724	i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0);
1091 :	syskin	768	Data->iMinSAD[0] = pMB->sad16 + ((Data->lambda16 * i * pMB->sad16)>>10);
1092 :			Data->iMinSAD[1] = pMB->sad8[0] + ((Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS)) >> 10);
1093 :	Isibaar	539	Data->iMinSAD[2] = pMB->sad8[1];
1094 :			Data->iMinSAD[3] = pMB->sad8[2];
1095 :			Data->iMinSAD[4] = pMB->sad8[3];
1096 :
1097 :	syskin	768	if (x \| y) {
1098 :	edgomez	568	threshA = Data->temp[0]; // that's when we keep this SAD atm
1099 :	chl	530	if (threshA < 512) threshA = 512;
1100 :	syskin	768	else if (threshA > 1024) threshA = 1024;
1101 :			} else threshA = 512;
1102 :	Isibaar	3
1103 :	chl	530	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1104 :	syskin	704	prevMBs + x + y * pParam->mb_width, Data->rrv);
1105 :	edgomez	195
1106 :	syskin	768	if (!Data->rrv) {
1107 :			if (inter4v \| Data->chroma) CheckCandidate = CheckCandidate16;
1108 :			else CheckCandidate = CheckCandidate16no4v; //for extra speed
1109 :			} else CheckCandidate = CheckCandidate32;
1110 :	edgomez	568
1111 :	syskin	768	/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/
1112 :	chl	530
1113 :			for (i = 1; i < 7; i++) {
1114 :			if (!(mask = make_mask(pmv, i)) ) continue;
1115 :	edgomez	568	(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1116 :	Isibaar	539	if (Data->iMinSAD[0] <= threshA) break;
1117 :	chl	530	}
1118 :
1119 :	Isibaar	539	if ((Data->iMinSAD[0] <= threshA) \|\|
1120 :			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1121 :	syskin	768	(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16)))
1122 :	chl	530	inter4v = 0;
1123 :	syskin	768	else {
1124 :	Isibaar	3
1125 :	syskin	748	MainSearchFunc * MainSearchPtr;
1126 :	Isibaar	539	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1127 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1128 :			else MainSearchPtr = DiamondSearch;
1129 :	chl	181
1130 :	Isibaar	539	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1131 :	Isibaar	3
1132 :	chl	530	/* extended search, diamond starting in 0,0 and in prediction.
1133 :			note that this search is/might be done in halfpel positions,
1134 :			which makes it more different than the diamond above */
1135 :	chl	259
1136 :	Isibaar	539	if (MotionFlags & PMV_EXTSEARCH16) {
1137 :			int32_t bSAD;
1138 :			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1139 :	syskin	704	if (Data->rrv) {
1140 :			startMV.x = RRV_MV_SCALEUP(startMV.x);
1141 :			startMV.y = RRV_MV_SCALEUP(startMV.y);
1142 :	syskin	748	} else
1143 :	syskin	704	if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1144 :			startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1145 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1146 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1147 :	Isibaar	3
1148 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1149 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1150 :			if (bSAD < Data->iMinSAD[0]) {
1151 :			Data->currentMV[0] = backupMV;
1152 :			Data->iMinSAD[0] = bSAD; }
1153 :			}
1154 :	Isibaar	3
1155 :	Isibaar	539	backupMV = Data->currentMV[0];
1156 :	syskin	768	if (MotionFlags & PMV_HALFPELREFINE16 && !Data->rrv) startMV.x = startMV.y = 1;
1157 :			else startMV.x = startMV.y = 0;
1158 :	Isibaar	539	if (!(MVequal(startMV, backupMV))) {
1159 :			bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1160 :	edgomez	195
1161 :	Isibaar	606	(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1162 :	Isibaar	539	(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1163 :			if (bSAD < Data->iMinSAD[0]) {
1164 :			Data->currentMV[0] = backupMV;
1165 :			Data->iMinSAD[0] = bSAD; }
1166 :			}
1167 :	Isibaar	3	}
1168 :			}
1169 :
1170 :	syskin	663	if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data);
1171 :	Isibaar	3
1172 :	Isibaar	580	for(i = 0; i < 5; i++) {
1173 :			Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1174 :			Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1175 :			}
1176 :
1177 :	syskin	768	if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE16) {
1178 :	syskin	663	Data->qpel_precision = 1;
1179 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1180 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1181 :	syskin	601
1182 :	syskin	663	SubpelRefine(Data);
1183 :	Isibaar	580	}
1184 :
1185 :	syskin	768	if (Data->iMinSAD[0] < (int32_t)iQuant * 30) inter4v = 0;
1186 :	Isibaar	539	if (inter4v) {
1187 :			SearchData Data8;
1188 :	syskin	768	memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data
1189 :
1190 :	Isibaar	539	Search8(Data, 2x, 2y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1191 :			Search8(Data, 2x + 1, 2y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1192 :			Search8(Data, 2x, 2y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1193 :			Search8(Data, 2x + 1, 2y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1194 :	syskin	768
1195 :	syskin	628	if (Data->chroma) {
1196 :	syskin	768	int sumx = 0, sumy = 0;
1197 :			const int div = 1 + Data->qpel;
1198 :			const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs;
1199 :	syskin	628
1200 :	syskin	768	for (i = 0; i < 4; i++) {
1201 :			sumx += mv[i].x / div;
1202 :			sumy += mv[i].y / div;
1203 :	syskin	663	}
1204 :	syskin	768
1205 :	syskin	753	Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf],
1206 :			(sumy >> 3) + roundtab_76[sumy & 0xf], Data);
1207 :	syskin	628	}
1208 :	Isibaar	539	}
1209 :	Isibaar	3
1210 :	syskin	704	if (Data->rrv) {
1211 :			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
1212 :			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
1213 :			}
1214 :	syskin	740
1215 :	chl	530	if (!(inter4v) \|\|
1216 :	syskin	748	(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1217 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1218 :	chl	530	// INTER MODE
1219 :			pMB->mode = MODE_INTER;
1220 :	syskin	768	pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1221 :			pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1222 :	Isibaar	3
1223 :	syskin	768	if(Data->qpel) {
1224 :	syskin	663	pMB->qmvs[0] = pMB->qmvs[1]
1225 :			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1226 :			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
1227 :			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
1228 :	syskin	628	} else {
1229 :	Isibaar	580	pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1230 :			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1231 :			}
1232 :	edgomez	195	} else {
1233 :	syskin	748	// INTER4V MODE; all other things are already set in Search8
1234 :	chl	530	pMB->mode = MODE_INTER4V;
1235 :	syskin	748	pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1236 :	Isibaar	539	Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1237 :	edgomez	78	}
1238 :	Isibaar	3	}
1239 :
1240 :	chl	530	static void
1241 :			Search8(const SearchData * const OldData,
1242 :			const int x, const int y,
1243 :			const uint32_t MotionFlags,
1244 :			const MBParam * const pParam,
1245 :			MACROBLOCK * const pMB,
1246 :			const MACROBLOCK * const pMBs,
1247 :	Isibaar	539	const int block,
1248 :			SearchData * const Data)
1249 :	chl	345	{
1250 :	syskin	704	int i = 0;
1251 :	Isibaar	539	Data->iMinSAD = OldData->iMinSAD + 1 + block;
1252 :			Data->currentMV = OldData->currentMV + 1 + block;
1253 :	Isibaar	580	Data->currentQMV = OldData->currentQMV + 1 + block;
1254 :	chl	345
1255 :	syskin	768	if(Data->qpel) {
1256 :	syskin	704	Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1257 :	syskin	724	if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
1258 :			Data->predMV, Data->iFcode, 0, 0);
1259 :	Isibaar	606	} else {
1260 :	syskin	704	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
1261 :	syskin	768	if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
1262 :			Data->predMV, Data->iFcode, 0, Data->rrv);
1263 :	Isibaar	606	}
1264 :	chl	345
1265 :	syskin	768	(Data->iMinSAD) += (Data->lambda8 i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
1266 :	syskin	704
1267 :	syskin	768	if (MotionFlags & (PMV_EXTSEARCH8\|PMV_HALFPELREFINE8\|PMV_QUARTERPELREFINE8)) {
1268 :	syskin	704	if (Data->rrv) i = 2; else i = 1;
1269 :	chl	345
1270 :	syskin	768	Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1271 :			Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1272 :			Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1273 :			Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1274 :	chl	345
1275 :	syskin	768	Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
1276 :	syskin	663	Data->qpel_precision = 0;
1277 :	chl	345
1278 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1279 :	syskin	768	pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
1280 :	syskin	712
1281 :	syskin	768	if (!Data->rrv) CheckCandidate = CheckCandidate8;
1282 :			else CheckCandidate = CheckCandidate16no4v;
1283 :	syskin	712
1284 :	chl	530	if (MotionFlags & PMV_EXTSEARCH8) {
1285 :	Isibaar	580	int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1286 :	syskin	768
1287 :	chl	530	MainSearchFunc *MainSearchPtr;
1288 :			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1289 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1290 :			else MainSearchPtr = DiamondSearch;
1291 :	chl	345
1292 :	Isibaar	580	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1293 :	chl	345
1294 :	syskin	601	if(*(Data->iMinSAD) < temp_sad) {
1295 :	Isibaar	580	Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1296 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1297 :			}
1298 :			}
1299 :
1300 :			if (MotionFlags & PMV_HALFPELREFINE8) {
1301 :			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1302 :
1303 :	syskin	663	SubpelRefine(Data); // perform halfpel refine of current best vector
1304 :	Isibaar	580
1305 :			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1306 :			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1307 :			Data->currentQMV->y = 2 * Data->currentMV->y;
1308 :			}
1309 :			}
1310 :
1311 :	syskin	768	if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE8) {
1312 :			Data->qpel_precision = 1;
1313 :			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1314 :			pParam->width, pParam->height, Data->iFcode, 1, 0);
1315 :			SubpelRefine(Data);
1316 :	Isibaar	580	}
1317 :	chl	345	}
1318 :	chl	530
1319 :	syskin	704	if (Data->rrv) {
1320 :			Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x);
1321 :			Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y);
1322 :			}
1323 :	syskin	768
1324 :	syskin	712	if(Data->qpel) {
1325 :	syskin	663	pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
1326 :			pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
1327 :	syskin	768	pMB->qmvs[block] = *Data->currentQMV;
1328 :	syskin	704	} else {
1329 :	Isibaar	580	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1330 :			pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1331 :			}
1332 :
1333 :	syskin	768	pMB->mvs[block] = *Data->currentMV;
1334 :			pMB->sad8[block] = 4 * *Data->iMinSAD;
1335 :	chl	345	}
1336 :
1337 :	syskin	768	/* motion estimation for B-frames */
1338 :	chl	345
1339 :	chl	530	static __inline VECTOR
1340 :			ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1341 :	Isibaar	3	{
1342 :	chl	530	/* the stupidiest function ever */
1343 :	syskin	768	return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]);
1344 :	Isibaar	3	}
1345 :
1346 :	chl	530	static void __inline
1347 :			PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1348 :			const uint32_t iWcount,
1349 :			const MACROBLOCK * const pMB,
1350 :			const uint32_t mode_curr)
1351 :	Isibaar	3	{
1352 :
1353 :	chl	530	// [0] is prediction
1354 :			pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1355 :	Isibaar	3
1356 :	chl	530	pmv[1].x = pmv[1].y = 0; // [1] is zero
1357 :	Isibaar	3
1358 :	chl	530	pmv[2] = ChoosePred(pMB, mode_curr);
1359 :			pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1360 :	edgomez	195
1361 :	chl	530	if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1362 :			pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1363 :	syskin	748	pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1364 :	Isibaar	539	} else pmv[3].x = pmv[3].y = 0;
1365 :	Isibaar	3
1366 :	chl	530	if (y != 0) {
1367 :			pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1368 :			pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1369 :			} else pmv[4].x = pmv[4].y = 0;
1370 :	Isibaar	3
1371 :	chl	530	if (x != 0) {
1372 :			pmv[5] = ChoosePred(pMB-1, mode_curr);
1373 :			pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1374 :			} else pmv[5].x = pmv[5].y = 0;
1375 :	Isibaar	3
1376 :	syskin	768	if (x != 0 && y != 0) {
1377 :	chl	530	pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1378 :	syskin	767	pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y);
1379 :	chl	530	} else pmv[6].x = pmv[6].y = 0;
1380 :			}
1381 :	chl	181
1382 :	edgomez	170
1383 :	syskin	768	/* search backward or forward */
1384 :	chl	530	static void
1385 :	syskin	801	SearchBF( const IMAGE * const pRef,
1386 :	chl	530	const uint8_t * const pRefH,
1387 :			const uint8_t * const pRefV,
1388 :			const uint8_t * const pRefHV,
1389 :			const IMAGE * const pCur,
1390 :			const int x, const int y,
1391 :			const uint32_t MotionFlags,
1392 :			const uint32_t iFcode,
1393 :			const MBParam * const pParam,
1394 :			MACROBLOCK * const pMB,
1395 :			const VECTOR * const predMV,
1396 :			int32_t * const best_sad,
1397 :	h	545	const int32_t mode_current,
1398 :			SearchData * const Data)
1399 :	chl	530	{
1400 :	Isibaar	3
1401 :	syskin	740	int i, iDirection = 255, mask;
1402 :	h	545	VECTOR pmv[7];
1403 :	chl	530	MainSearchFunc *MainSearchPtr;
1404 :	h	545	*Data->iMinSAD = MV_MAX_ERROR;
1405 :			Data->iFcode = iFcode;
1406 :	syskin	663	Data->qpel_precision = 0;
1407 :	syskin	801	Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache
1408 :	Isibaar	3
1409 :	syskin	801	Data->Ref = pRef->y + (x + y * Data->iEdgedWidth) * 16;
1410 :	syskin	768	Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16;
1411 :			Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16;
1412 :			Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16;
1413 :	syskin	801	Data->RefCU = pRef->u + (x + y * Data->iEdgedWidth/2) * 8;
1414 :			Data->RefCV = pRef->v + (x + y * Data->iEdgedWidth/2) * 8;
1415 :	Isibaar	3
1416 :	h	545	Data->predMV = *predMV;
1417 :	edgomez	195
1418 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1419 :	syskin	712	pParam->width, pParam->height, iFcode - Data->qpel, 0, 0);
1420 :	Isibaar	3
1421 :	h	545	pmv[0] = Data->predMV;
1422 :	syskin	658	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
1423 :	syskin	768
1424 :	h	545	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1425 :	edgomez	195
1426 :	h	545	Data->currentMV->x = Data->currentMV->y = 0;
1427 :	chl	530	CheckCandidate = CheckCandidate16no4v;
1428 :	chl	181
1429 :	chl	530	// main loop. checking all predictions
1430 :	syskin	740	for (i = 0; i < 7; i++) {
1431 :	chl	530	if (!(mask = make_mask(pmv, i)) ) continue;
1432 :	h	545	CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1433 :	edgomez	78	}
1434 :	Isibaar	3
1435 :	syskin	768	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1436 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1437 :	chl	530	else MainSearchPtr = DiamondSearch;
1438 :	chl	96
1439 :	syskin	740	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1440 :	chl	169
1441 :	syskin	663	SubpelRefine(Data);
1442 :	syskin	768
1443 :	syskin	740	if (Data->qpel && Data->iMinSAD < best_sad + 300) {
1444 :	syskin	658	Data->currentQMV->x = 2*Data->currentMV->x;
1445 :			Data->currentQMV->y = 2*Data->currentMV->y;
1446 :	syskin	663	Data->qpel_precision = 1;
1447 :	syskin	712	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1448 :			pParam->width, pParam->height, iFcode, 1, 0);
1449 :	syskin	663	SubpelRefine(Data);
1450 :	syskin	658	}
1451 :	chl	169
1452 :	chl	530	// three bits are needed to code backward mode. four for forward
1453 :	syskin	740
1454 :	syskin	768	if (mode_current == MODE_FORWARD) Data->iMinSAD += 4 Data->lambda16;
1455 :			else Data->iMinSAD += 3 Data->lambda16;
1456 :	chl	169
1457 :	h	545	if (Data->iMinSAD < best_sad) {
1458 :			best_sad = Data->iMinSAD;
1459 :	chl	530	pMB->mode = mode_current;
1460 :	syskin	658	if (Data->qpel) {
1461 :			pMB->pmvs[0].x = Data->currentQMV->x - predMV->x;
1462 :			pMB->pmvs[0].y = Data->currentQMV->y - predMV->y;
1463 :	syskin	748	if (mode_current == MODE_FORWARD)
1464 :	syskin	658	pMB->qmvs[0] = *Data->currentQMV;
1465 :	syskin	748	else
1466 :	syskin	658	pMB->b_qmvs[0] = *Data->currentQMV;
1467 :			} else {
1468 :			pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1469 :			pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1470 :			}
1471 :	syskin	739	if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1472 :			else pMB->b_mvs[0] = *Data->currentMV;
1473 :			}
1474 :	syskin	658
1475 :	syskin	768	if (mode_current == MODE_FORWARD) (Data->currentMV+2) = Data->currentMV;
1476 :			else (Data->currentMV+1) = Data->currentMV; //we store currmv for interpolate search
1477 :	chl	530	}
1478 :	chl	169
1479 :	syskin	748	static void
1480 :	syskin	712	SkipDecisionB(const IMAGE * const pCur,
1481 :	syskin	767	const IMAGE * const f_Ref,
1482 :			const IMAGE * const b_Ref,
1483 :			MACROBLOCK * const pMB,
1484 :			const uint32_t x, const uint32_t y,
1485 :			const SearchData * const Data)
1486 :	syskin	712	{
1487 :	syskin	767	int dx = 0, dy = 0, b_dx = 0, b_dy = 0;
1488 :	syskin	768	int32_t sum;
1489 :	syskin	767	const int div = 1 + Data->qpel;
1490 :			int k;
1491 :	syskin	768	const uint32_t stride = Data->iEdgedWidth/2;
1492 :	syskin	712	//this is not full chroma compensation, only it's fullpel approximation. should work though
1493 :
1494 :	syskin	767	for (k = 0; k < 4; k++) {
1495 :			dy += Data->directmvF[k].y / div;
1496 :			dx += Data->directmvF[0].x / div;
1497 :			b_dy += Data->directmvB[0].y / div;
1498 :			b_dx += Data->directmvB[0].x / div;
1499 :	syskin	712	}
1500 :
1501 :			dy = (dy >> 3) + roundtab_76[dy & 0xf];
1502 :			dx = (dx >> 3) + roundtab_76[dx & 0xf];
1503 :			b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf];
1504 :			b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf];
1505 :
1506 :	syskin	768	sum = sad8bi(pCur->u + 8 * x + 8 * y * stride,
1507 :			f_Ref->u + (y8 + dy/2) stride + x*8 + dx/2,
1508 :			b_Ref->u + (y8 + b_dy/2) stride + x*8 + b_dx/2,
1509 :			stride);
1510 :	syskin	767
1511 :	syskin	768	if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; //no skip
1512 :	syskin	767
1513 :	syskin	768	sum += sad8bi(pCur->v + 8x + 8 y * stride,
1514 :			f_Ref->v + (y8 + dy/2) stride + x*8 + dx/2,
1515 :			b_Ref->v + (y8 + b_dy/2) stride + x*8 + b_dx/2,
1516 :			stride);
1517 :	syskin	712
1518 :	syskin	768	if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
1519 :	syskin	712	}
1520 :
1521 :			static __inline uint32_t
1522 :	Isibaar	539	SearchDirect(const IMAGE * const f_Ref,
1523 :	chl	530	const uint8_t * const f_RefH,
1524 :			const uint8_t * const f_RefV,
1525 :			const uint8_t * const f_RefHV,
1526 :	Isibaar	539	const IMAGE * const b_Ref,
1527 :	chl	530	const uint8_t * const b_RefH,
1528 :			const uint8_t * const b_RefV,
1529 :			const uint8_t * const b_RefHV,
1530 :			const IMAGE * const pCur,
1531 :			const int x, const int y,
1532 :			const uint32_t MotionFlags,
1533 :			const int32_t TRB, const int32_t TRD,
1534 :			const MBParam * const pParam,
1535 :			MACROBLOCK * const pMB,
1536 :			const MACROBLOCK * const b_mb,
1537 :	h	545	int32_t * const best_sad,
1538 :			SearchData * const Data)
1539 :	edgomez	195
1540 :	chl	530	{
1541 :	h	545	int32_t skip_sad;
1542 :	syskin	768	int k = (x + Data->iEdgedWidthy) 16;
1543 :	chl	530	MainSearchFunc *MainSearchPtr;
1544 :	Isibaar	3
1545 :	h	545	Data->iMinSAD = 2564096;
1546 :	syskin	768	Data->Ref = f_Ref->y + k;
1547 :			Data->RefH = f_RefH + k;
1548 :			Data->RefV = f_RefV + k;
1549 :			Data->RefHV = f_RefHV + k;
1550 :			Data->bRef = b_Ref->y + k;
1551 :			Data->bRefH = b_RefH + k;
1552 :			Data->bRefV = b_RefV + k;
1553 :			Data->bRefHV = b_RefHV + k;
1554 :	syskin	801	Data->RefCU = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
1555 :			Data->RefCV = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
1556 :			Data->b_RefCU = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
1557 :			Data->b_RefCV = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
1558 :	chl	140
1559 :	syskin	768	k = Data->qpel ? 4 : 2;
1560 :			Data->max_dx = k * (pParam->width - x * 16);
1561 :			Data->max_dy = k * (pParam->height - y * 16);
1562 :			Data->min_dx = -k * (16 + x * 16);
1563 :			Data->min_dy = -k * (16 + y * 16);
1564 :	chl	140
1565 :	syskin	768	Data->referencemv = Data->qpel ? b_mb->qmvs : b_mb->mvs;
1566 :			Data->qpel_precision = 0;
1567 :	Isibaar	3
1568 :	chl	530	for (k = 0; k < 4; k++) {
1569 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1570 :			pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1571 :			pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1572 :			pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1573 :	Isibaar	3
1574 :	syskin	768	if ( (pMB->b_mvs[k].x > Data->max_dx) \| (pMB->b_mvs[k].x < Data->min_dx)
1575 :			\| (pMB->b_mvs[k].y > Data->max_dy) \| (pMB->b_mvs[k].y < Data->min_dy) ) {
1576 :	Isibaar	3
1577 :	Isibaar	539	best_sad = 2564096; // in that case, we won't use direct mode
1578 :			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1579 :			pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1580 :	syskin	712	return 256*4096;
1581 :	Isibaar	539	}
1582 :			if (b_mb->mode != MODE_INTER4V) {
1583 :			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1584 :			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1585 :	h	545	Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1586 :			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1587 :	Isibaar	539	break;
1588 :			}
1589 :			}
1590 :	Isibaar	3
1591 :	syskin	768	CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v;
1592 :
1593 :	h	545	(*CheckCandidate)(0, 0, 255, &k, Data);
1594 :	Isibaar	3
1595 :	syskin	712	// initial (fast) skip decision
1596 :	syskin	801	if (Data->iMinSAD < pMB->quant INITIAL_SKIP_THRESH * (2 + Data->chroma?1:0)) {
1597 :			//possible skip
1598 :			if (Data->chroma) {
1599 :			pMB->mode = MODE_DIRECT_NONE_MV;
1600 :			return *Data->iMinSAD; // skip.
1601 :			} else {
1602 :			SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data);
1603 :			if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip.
1604 :			}
1605 :	Isibaar	539	}
1606 :
1607 :	h	545	skip_sad = *Data->iMinSAD;
1608 :	edgomez	195
1609 :	syskin	768	// DIRECT MODE DELTA VECTOR SEARCH.
1610 :	chl	530	// This has to be made more effective, but at the moment I'm happy it's running at all
1611 :	Isibaar	3
1612 :	chl	530	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1613 :			else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1614 :			else MainSearchPtr = DiamondSearch;
1615 :	edgomez	195
1616 :	h	545	(*MainSearchPtr)(0, 0, Data, 255);
1617 :	Isibaar	3
1618 :	syskin	663	SubpelRefine(Data);
1619 :	Isibaar	3
1620 :	h	545	best_sad = Data->iMinSAD;
1621 :	Isibaar	3
1622 :	syskin	768	if (Data->qpel \|\| b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT;
1623 :	syskin	672	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1624 :	edgomez	195
1625 :	h	545	pMB->pmvs[3] = *Data->currentMV;
1626 :	edgomez	195
1627 :	chl	530	for (k = 0; k < 4; k++) {
1628 :	h	545	pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1629 :	syskin	658	pMB->b_mvs[k].x = ( (Data->currentMV->x == 0)
1630 :	h	545	? Data->directmvB[k].x
1631 :	syskin	658	:pMB->mvs[k].x - Data->referencemv[k].x);
1632 :	h	545	pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1633 :			pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1634 :			? Data->directmvB[k].y
1635 :			: pMB->mvs[k].y - Data->referencemv[k].y);
1636 :	syskin	658	if (Data->qpel) {
1637 :			pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2;
1638 :			pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2;
1639 :			pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2;
1640 :			pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2;
1641 :			}
1642 :
1643 :	chl	530	if (b_mb->mode != MODE_INTER4V) {
1644 :			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1645 :			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1646 :	syskin	658	pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0];
1647 :			pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0];
1648 :	chl	530	break;
1649 :			}
1650 :			}
1651 :	h	545	return skip_sad;
1652 :	Isibaar	3	}
1653 :	chl	96
1654 :	syskin	712	static void
1655 :	syskin	801	SearchInterpolate(const IMAGE * const f_Ref,
1656 :	chl	530	const uint8_t * const f_RefH,
1657 :			const uint8_t * const f_RefV,
1658 :			const uint8_t * const f_RefHV,
1659 :	syskin	801	const IMAGE * const b_Ref,
1660 :	chl	530	const uint8_t * const b_RefH,
1661 :			const uint8_t * const b_RefV,
1662 :			const uint8_t * const b_RefHV,
1663 :			const IMAGE * const pCur,
1664 :			const int x, const int y,
1665 :			const uint32_t fcode,
1666 :			const uint32_t bcode,
1667 :			const uint32_t MotionFlags,
1668 :			const MBParam * const pParam,
1669 :			const VECTOR * const f_predMV,
1670 :			const VECTOR * const b_predMV,
1671 :			MACROBLOCK * const pMB,
1672 :	h	545	int32_t * const best_sad,
1673 :			SearchData * const fData)
1674 :	chl	530
1675 :	chl	96	{
1676 :
1677 :	chl	530	int iDirection, i, j;
1678 :	h	545	SearchData bData;
1679 :	chl	96
1680 :	syskin	712	fData->qpel_precision = 0;
1681 :			memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data
1682 :			fData->iMinSAD = 4096256;
1683 :	syskin	739	bData.currentMV++; bData.currentQMV++;
1684 :	h	545	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1685 :	edgomez	195
1686 :	syskin	712	i = (x + y * fData->iEdgedWidth) * 16;
1687 :	syskin	801	bData.bRef = fData->Ref = f_Ref->y + i;
1688 :	syskin	712	bData.bRefH = fData->RefH = f_RefH + i;
1689 :			bData.bRefV = fData->RefV = f_RefV + i;
1690 :			bData.bRefHV = fData->RefHV = f_RefHV + i;
1691 :	syskin	801	bData.Ref = fData->bRef = b_Ref->y + i;
1692 :	syskin	712	bData.RefH = fData->bRefH = b_RefH + i;
1693 :			bData.RefV = fData->bRefV = b_RefV + i;
1694 :			bData.RefHV = fData->bRefHV = b_RefHV + i;
1695 :	syskin	801	bData.b_RefCU = fData->RefCU = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
1696 :			bData.b_RefCV = fData->RefCV = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
1697 :			bData.RefCU = fData->b_RefCU = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
1698 :			bData.RefCV = fData->b_RefCV = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
1699 :	chl	96
1700 :	syskin	801
1701 :	h	545	bData.bpredMV = fData->predMV = *f_predMV;
1702 :			fData->bpredMV = bData.predMV = *b_predMV;
1703 :	syskin	661	fData->currentMV[0] = fData->currentMV[2];
1704 :	chl	96
1705 :	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);
1706 :			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);
1707 :
1708 :	h	545	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1709 :	syskin	675	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx;
1710 :			if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy;
1711 :			if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1712 :	Isibaar	539
1713 :	h	545	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1714 :	syskin	675	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx;
1715 :			if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
1716 :			if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1717 :	Isibaar	539
1718 :	h	545	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1719 :	chl	96
1720 :	syskin	768	//diamond
1721 :	chl	530	do {
1722 :			iDirection = 255;
1723 :			// forward MV moves
1724 :	h	545	i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1725 :	chl	96
1726 :	h	545	CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1727 :			CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1728 :			CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1729 :			CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1730 :	chl	96
1731 :	chl	530	// backward MV moves
1732 :	h	545	i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1733 :			fData->currentMV[2] = fData->currentMV[0];
1734 :	Isibaar	539	CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1735 :			CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1736 :			CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1737 :			CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1738 :	chl	96
1739 :	chl	530	} while (!(iDirection));
1740 :	edgomez	195
1741 :	syskin	768	//qpel refinement
1742 :	syskin	658	if (fData->qpel) {
1743 :	syskin	740	if (fData->iMinSAD > best_sad + 500) return;
1744 :	syskin	675	CheckCandidate = CheckCandidateInt;
1745 :	syskin	663	fData->qpel_precision = bData.qpel_precision = 1;
1746 :	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);
1747 :			get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
1748 :	syskin	658	fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
1749 :			fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
1750 :			fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
1751 :			fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
1752 :	syskin	663	SubpelRefine(fData);
1753 :	syskin	740	if (fData->iMinSAD > best_sad + 300) return;
1754 :	syskin	658	fData->currentQMV[2] = fData->currentQMV[0];
1755 :	syskin	663	SubpelRefine(&bData);
1756 :	syskin	658	}
1757 :
1758 :	syskin	768	fData->iMinSAD += (2+3) fData->lambda16; // two bits are needed to code interpolate mode.
1759 :
1760 :	h	545	if (fData->iMinSAD < best_sad) {
1761 :			best_sad = fData->iMinSAD;
1762 :			pMB->mvs[0] = fData->currentMV[0];
1763 :			pMB->b_mvs[0] = fData->currentMV[1];
1764 :	chl	530	pMB->mode = MODE_INTERPOLATE;
1765 :	syskin	658	if (fData->qpel) {
1766 :			pMB->qmvs[0] = fData->currentQMV[0];
1767 :			pMB->b_qmvs[0] = fData->currentQMV[1];
1768 :			pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x;
1769 :			pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y;
1770 :			pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x;
1771 :			pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y;
1772 :			} else {
1773 :			pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1774 :			pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1775 :			pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1776 :			pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1777 :			}
1778 :	chl	530	}
1779 :			}
1780 :	chl	96
1781 :	chl	530	void
1782 :			MotionEstimationBVOP(MBParam * const pParam,
1783 :			FRAMEINFO * const frame,
1784 :			const int32_t time_bp,
1785 :			const int32_t time_pp,
1786 :			// forward (past) reference
1787 :			const MACROBLOCK * const f_mbs,
1788 :			const IMAGE * const f_ref,
1789 :			const IMAGE * const f_refH,
1790 :			const IMAGE * const f_refV,
1791 :			const IMAGE * const f_refHV,
1792 :			// backward (future) reference
1793 :	syskin	644	const FRAMEINFO * const b_reference,
1794 :	chl	530	const IMAGE * const b_ref,
1795 :			const IMAGE * const b_refH,
1796 :			const IMAGE * const b_refV,
1797 :			const IMAGE * const b_refHV)
1798 :			{
1799 :			uint32_t i, j;
1800 :	syskin	712	int32_t best_sad;
1801 :			uint32_t skip_sad;
1802 :	chl	530	int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1803 :	syskin	644	const MACROBLOCK * const b_mbs = b_reference->mbs;
1804 :	chl	96
1805 :	chl	530	VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1806 :	chl	96
1807 :	chl	530	const int32_t TRB = time_pp - time_bp;
1808 :			const int32_t TRD = time_pp;
1809 :	chl	96
1810 :	h	545	// some pre-inintialized data for the rest of the search
1811 :
1812 :			SearchData Data;
1813 :			int32_t iMinSAD;
1814 :			VECTOR currentMV[3];
1815 :	syskin	658	VECTOR currentQMV[3];
1816 :	syskin	801	int32_t temp[8];
1817 :	syskin	704	memset(&Data, 0, sizeof(SearchData));
1818 :	h	545	Data.iEdgedWidth = pParam->edged_width;
1819 :	syskin	658	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
1820 :	h	545	Data.iMinSAD = &iMinSAD;
1821 :	syskin	712	Data.lambda16 = lambda_vec16[frame->quant];
1822 :	syskin	658	Data.qpel = pParam->m_quarterpel;
1823 :	syskin	663	Data.rounding = 0;
1824 :	syskin	801	Data.chroma = frame->motion_flags & PMV_CHROMA8;
1825 :			Data.temp = temp;
1826 :	h	545
1827 :	syskin	754	Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose)
1828 :	chl	530	// note: i==horizontal, j==vertical
1829 :			for (j = 0; j < pParam->mb_height; j++) {
1830 :	chl	96
1831 :	chl	530	f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1832 :	edgomez	195
1833 :	chl	530	for (i = 0; i < pParam->mb_width; i++) {
1834 :			MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1835 :			const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1836 :	edgomez	195
1837 :	chl	619	/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1838 :	syskin	644	if (b_reference->coding_type != S_VOP)
1839 :			if (b_mb->mode == MODE_NOT_CODED) {
1840 :			pMB->mode = MODE_NOT_CODED;
1841 :			continue;
1842 :			}
1843 :	chl	96
1844 :	h	545	Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1845 :	syskin	801	Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8;
1846 :			Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8;
1847 :	syskin	601	pMB->quant = frame->quant;
1848 :	syskin	628
1849 :	chl	530	/* direct search comes first, because it (1) checks for SKIP-mode
1850 :			and (2) sets very good predictions for forward and backward search */
1851 :	Isibaar	539	skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1852 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1853 :	chl	530	&frame->image,
1854 :			i, j,
1855 :			frame->motion_flags,
1856 :			TRB, TRD,
1857 :			pParam,
1858 :			pMB, b_mb,
1859 :	h	545	&best_sad,
1860 :			&Data);
1861 :	chl	96
1862 :	Isibaar	539	if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1863 :	syskin	739
1864 :	chl	530	// forward search
1865 :	syskin	801	SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1866 :	chl	530	&frame->image, i, j,
1867 :			frame->motion_flags,
1868 :	h	545	frame->fcode, pParam,
1869 :	chl	530	pMB, &f_predMV, &best_sad,
1870 :	h	545	MODE_FORWARD, &Data);
1871 :	chl	96
1872 :	chl	530	// backward search
1873 :	syskin	801	SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y,
1874 :	chl	530	&frame->image, i, j,
1875 :			frame->motion_flags,
1876 :	h	545	frame->bcode, pParam,
1877 :	chl	530	pMB, &b_predMV, &best_sad,
1878 :	h	545	MODE_BACKWARD, &Data);
1879 :	chl	96
1880 :	chl	530	// interpolate search comes last, because it uses data from forward and backward as prediction
1881 :	syskin	801	SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1882 :			b_ref, b_refH->y, b_refV->y, b_refHV->y,
1883 :	chl	530	&frame->image,
1884 :			i, j,
1885 :			frame->fcode, frame->bcode,
1886 :			frame->motion_flags,
1887 :	h	545	pParam,
1888 :	chl	530	&f_predMV, &b_predMV,
1889 :	h	545	pMB, &best_sad,
1890 :			&Data);
1891 :	syskin	662
1892 :	syskin	712	// final skip decision
1893 :	syskin	801	if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP * 2)
1894 :	syskin	712	&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) )
1895 :	syskin	767	SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data);
1896 :	syskin	712
1897 :	chl	530	switch (pMB->mode) {
1898 :			case MODE_FORWARD:
1899 :			f_count++;
1900 :	syskin	768	f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
1901 :	chl	530	break;
1902 :			case MODE_BACKWARD:
1903 :			b_count++;
1904 :	syskin	768	b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
1905 :	chl	530	break;
1906 :			case MODE_INTERPOLATE:
1907 :			i_count++;
1908 :	syskin	768	f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0];
1909 :			b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0];
1910 :	chl	530	break;
1911 :			case MODE_DIRECT:
1912 :			case MODE_DIRECT_NO4V:
1913 :			d_count++;
1914 :			default:
1915 :			break;
1916 :	chl	96	}
1917 :			}
1918 :			}
1919 :			}
1920 :
1921 :	syskin	739	static __inline void
1922 :	Isibaar	539	MEanalyzeMB ( const uint8_t * const pRef,
1923 :			const uint8_t * const pCur,
1924 :			const int x,
1925 :			const int y,
1926 :			const MBParam * const pParam,
1927 :	syskin	739	MACROBLOCK * const pMBs,
1928 :	h	545	SearchData * const Data)
1929 :	Isibaar	539	{
1930 :	chl	317
1931 :	syskin	739	int i, mask;
1932 :	h	545	VECTOR pmv[3];
1933 :	syskin	739	MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width];
1934 :	syskin	768
1935 :	syskin	739	for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1936 :	suxen_drol	118
1937 :	syskin	640	//median is only used as prediction. it doesn't have to be real
1938 :			if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0;
1939 :			else
1940 :			if (x == 1) //left macroblock does not have any vector now
1941 :			Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median
1942 :	syskin	748	else if (y == 1) // top macroblock doesn't have it's vector
1943 :	syskin	640	Data->predMV = (pMB - 1)->mvs[0]; // left instead of median
1944 :			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
1945 :
1946 :	h	545	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1947 :	syskin	739	pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv);
1948 :	chl	326
1949 :	h	545	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1950 :			Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1951 :	syskin	739
1952 :	Isibaar	539	pmv[1].x = EVEN(pMB->mvs[0].x);
1953 :			pmv[1].y = EVEN(pMB->mvs[0].y);
1954 :	syskin	640	pmv[2].x = EVEN(Data->predMV.x);
1955 :			pmv[2].y = EVEN(Data->predMV.y);
1956 :			pmv[0].x = pmv[0].y = 0;
1957 :	chl	326
1958 :	syskin	739	CheckCandidate32I(0, 0, 255, &i, Data);
1959 :	syskin	640
1960 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) {
1961 :	syskin	640
1962 :	syskin	739	if (!(mask = make_mask(pmv, 1)))
1963 :			CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data);
1964 :			if (!(mask = make_mask(pmv, 2)))
1965 :			CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data);
1966 :	Isibaar	539
1967 :	syskin	739	if (Data->iMinSAD > 4 MAX_SAD00_FOR_SKIP * 4) // diamond only if needed
1968 :			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1969 :	Isibaar	539
1970 :	syskin	739	for (i = 0; i < 4; i++) {
1971 :	syskin	768	MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width];
1972 :	syskin	739	MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
1973 :			MB->mode = MODE_INTER;
1974 :			MB->sad16 = Data->iMinSAD[i+1];
1975 :			}
1976 :			}
1977 :	Isibaar	539	}
1978 :
1979 :	syskin	739	#define INTRA_BIAS 2500
1980 :			#define INTRA_THRESH 1500
1981 :			#define INTER_THRESH 1400
1982 :	Isibaar	539
1983 :			int
1984 :	syskin	768	MEanalysis( const IMAGE * const pRef,
1985 :	syskin	644	FRAMEINFO * const Current,
1986 :	Isibaar	539	MBParam * const pParam,
1987 :	syskin	644	int maxIntra, //maximum number if non-I frames
1988 :			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
1989 :	syskin	768	int bCount) // number of B frames in a row
1990 :	Isibaar	539	{
1991 :			uint32_t x, y, intra = 0;
1992 :			int sSAD = 0;
1993 :	syskin	644	MACROBLOCK * const pMBs = Current->mbs;
1994 :			const IMAGE * const pCurrent = &Current->image;
1995 :			int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
1996 :	h	545
1997 :	syskin	739	int32_t iMinSAD[5], temp[5];
1998 :			VECTOR currentMV[5];
1999 :	h	545	SearchData Data;
2000 :			Data.iEdgedWidth = pParam->edged_width;
2001 :	syskin	739	Data.currentMV = currentMV;
2002 :			Data.iMinSAD = iMinSAD;
2003 :	syskin	644	Data.iFcode = Current->fcode;
2004 :	syskin	739	Data.rrv = Current->global_flags & XVID_REDUCED;
2005 :			Data.temp = temp;
2006 :			CheckCandidate = CheckCandidate32I;
2007 :	h	545
2008 :	syskin	768	if (intraCount != 0 && intraCount < 10) // we're right after an I frame
2009 :	syskin	658	IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
2010 :	syskin	644	else
2011 :			if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
2012 :			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
2013 :
2014 :	syskin	672	InterThresh += 400 * (1 - bCount);
2015 :	syskin	739	if (InterThresh < 300) InterThresh = 300;
2016 :	syskin	644
2017 :	Isibaar	539	if (sadInit) (*sadInit) ();
2018 :
2019 :	syskin	768	for (y = 1; y < pParam->mb_height-1; y += 2) {
2020 :			for (x = 1; x < pParam->mb_width-1; x += 2) {
2021 :	syskin	739	int i;
2022 :	Isibaar	539
2023 :	syskin	739	if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
2024 :
2025 :			MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
2026 :	syskin	768
2027 :	syskin	739	for (i = 0; i < 4; i++) {
2028 :			int dev;
2029 :	syskin	768	MACROBLOCK pMB = &pMBs[x+(i&1) + (y+(i>>1)) pParam->mb_width];
2030 :	syskin	739	if (pMB->sad16 > IntraThresh) {
2031 :	syskin	768	dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16,
2032 :			pParam->edged_width);
2033 :	syskin	739	if (dev + IntraThresh < pMB->sad16) {
2034 :			pMB->mode = MODE_INTRA;
2035 :			if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
2036 :			}
2037 :	Isibaar	539	}
2038 :	syskin	739	sSAD += pMB->sad16;
2039 :	syskin	644	}
2040 :	suxen_drol	118	}
2041 :			}
2042 :	syskin	704	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2043 :	syskin	793	// if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP;
2044 :	suxen_drol	709	if (sSAD > InterThresh ) return P_VOP;
2045 :	Isibaar	539	emms();
2046 :	suxen_drol	709	return B_VOP;
2047 :	syskin	712
2048 :	suxen_drol	118	}
2049 :	syskin	576
2050 :	chl	769
2051 :			static WARPPOINTS
2052 :	syskin	801	GlobalMotionEst(const MACROBLOCK * const pMBs,
2053 :	chl	769	const MBParam * const pParam,
2054 :			const FRAMEINFO * const current,
2055 :			const FRAMEINFO * const reference,
2056 :			const IMAGE * const pRefH,
2057 :			const IMAGE * const pRefV,
2058 :			const IMAGE * const pRefHV )
2059 :	syskin	628	{
2060 :	chl	619
2061 :	chl	769	const int deltax=8; // upper bound for difference between a MV and it's neighbour MVs
2062 :			const int deltay=8;
2063 :	syskin	801	const int grad=512; // lower bound for deviation in MB
2064 :	chl	769
2065 :			WARPPOINTS gmc;
2066 :
2067 :			uint32_t mx, my;
2068 :	chl	619
2069 :	chl	769	int MBh = pParam->mb_height;
2070 :			int MBw = pParam->mb_width;
2071 :
2072 :			int MBmask= calloc(MBhMBw,sizeof(int));
2073 :			double DtimesF[4] = { 0.,0., 0., 0. };
2074 :			double sol[4] = { 0., 0., 0., 0. };
2075 :			double a,b,c,n,denom;
2076 :			double meanx,meany;
2077 :			int num,oldnum;
2078 :
2079 :			if (!MBmask) { fprintf(stderr,"Mem error\n"); return gmc;}
2080 :
2081 :	syskin	801	// filter mask of all blocks
2082 :	chl	769
2083 :			for (my = 1; my < MBh-1; my++)
2084 :			for (mx = 1; mx < MBw-1; mx++)
2085 :			{
2086 :			const int mbnum = mx + my * MBw;
2087 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2088 :			const VECTOR mv = pMB->mvs[0];
2089 :
2090 :			if (pMB->mode == MODE_INTRA \|\| pMB->mode == MODE_NOT_CODED)
2091 :			continue;
2092 :
2093 :			if ( ( (ABS(mv.x - (pMB-1)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-1)->mvs[0].y) < deltay) )
2094 :			&& ( (ABS(mv.x - (pMB+1)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+1)->mvs[0].y) < deltay) )
2095 :			&& ( (ABS(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-MBw)->mvs[0].y) < deltay) )
2096 :			&& ( (ABS(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) )
2097 :			MBmask[mbnum]=1;
2098 :	syskin	628	}
2099 :	chl	769
2100 :			for (my = 1; my < MBh-1; my++)
2101 :			for (mx = 1; mx < MBw-1; mx++)
2102 :			{
2103 :			const uint8_t const pCur = current->image.y + 16mypParam->edged_width + 16mx;
2104 :
2105 :			const int mbnum = mx + my * MBw;
2106 :			if (!MBmask[mbnum])
2107 :			continue;
2108 :	syskin	628
2109 :	chl	769	if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= grad )
2110 :			MBmask[mbnum] = 0;
2111 :			if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= grad )
2112 :			MBmask[mbnum] = 0;
2113 :
2114 :			}
2115 :	syskin	628
2116 :	chl	769	emms();
2117 :	chl	619
2118 :	chl	769	do { /* until convergence */
2119 :	chl	619
2120 :	chl	769	a = b = c = n = 0;
2121 :			DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.;
2122 :			for (my = 0; my < MBh; my++)
2123 :			for (mx = 0; mx < MBw; mx++)
2124 :			{
2125 :			const int mbnum = mx + my * MBw;
2126 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2127 :			const VECTOR mv = pMB->mvs[0];
2128 :	chl	619
2129 :	chl	769	if (!MBmask[mbnum])
2130 :			continue;
2131 :
2132 :			n++;
2133 :			a += 16*mx+8;
2134 :			b += 16*my+8;
2135 :			c += (16mx+8)(16mx+8)+(16my+8)(16my+8);
2136 :	syskin	768
2137 :	chl	769	DtimesF[0] += (double)mv.x;
2138 :			DtimesF[1] += (double)mv.x(16mx+8) + (double)mv.y(16my+8);
2139 :			DtimesF[2] += (double)mv.x(16my+8) - (double)mv.y(16mx+8);
2140 :			DtimesF[3] += (double)mv.y;
2141 :			}
2142 :	syskin	768
2143 :	chl	769	denom = aa+bb-c*n;
2144 :
2145 :			/* Solve the system: sol = (D'ED)^{-1} D'EF */
2146 :			/* D'EF has been calculated in the same loop as matrix */
2147 :	syskin	768
2148 :	chl	769	sol[0] = -cDtimesF[0] + aDtimesF[1] + b*DtimesF[2];
2149 :			sol[1] = aDtimesF[0] - nDtimesF[1] + b*DtimesF[3];
2150 :			sol[2] = bDtimesF[0] - nDtimesF[2] - a*DtimesF[3];
2151 :			sol[3] = bDtimesF[1] - aDtimesF[2] - c*DtimesF[3];
2152 :	syskin	628
2153 :	chl	769	sol[0] /= denom;
2154 :			sol[1] /= denom;
2155 :			sol[2] /= denom;
2156 :			sol[3] /= denom;
2157 :
2158 :			meanx = meany = 0.;
2159 :			oldnum = 0;
2160 :			for (my = 0; my < MBh; my++)
2161 :			for (mx = 0; mx < MBw; mx++)
2162 :			{
2163 :			const int mbnum = mx + my * MBw;
2164 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2165 :			const VECTOR mv = pMB->mvs[0];
2166 :	syskin	628
2167 :	chl	769	if (!MBmask[mbnum])
2168 :			continue;
2169 :	syskin	628
2170 :	chl	769	oldnum++;
2171 :			meanx += ABS(( sol[0] + (16mx+8)sol[1] + (16my+8)sol[2] ) - mv.x );
2172 :	chl	770	meany += ABS(( sol[3] - (16mx+8)sol[2] + (16my+8)sol[1] ) - mv.y );
2173 :			}
2174 :	chl	769
2175 :	chl	770	if (4meanx > oldnum) / better fit than 0.25 is useless */
2176 :	chl	769	meanx /= oldnum;
2177 :	syskin	801	else
2178 :	chl	770	meanx = 0.25;
2179 :
2180 :			if (4*meany > oldnum)
2181 :	chl	769	meany /= oldnum;
2182 :	syskin	801	else
2183 :	chl	770	meany = 0.25;
2184 :	chl	769
2185 :	chl	770	/* fprintf(stderr,"sol = (%8.5f, %8.5f, %8.5f, %8.5f)\n",sol[0],sol[1],sol[2],sol[3]);
2186 :			fprintf(stderr,"meanx = %8.5f meany = %8.5f %d\n",meanx,meany, oldnum);
2187 :			*/
2188 :	chl	769	num = 0;
2189 :			for (my = 0; my < MBh; my++)
2190 :			for (mx = 0; mx < MBw; mx++)
2191 :			{
2192 :			const int mbnum = mx + my * MBw;
2193 :			const MACROBLOCK *pMB = &pMBs[mbnum];
2194 :			const VECTOR mv = pMB->mvs[0];
2195 :
2196 :			if (!MBmask[mbnum])
2197 :			continue;
2198 :
2199 :			if ( ( ABS(( sol[0] + (16mx+8)sol[1] + (16my+8)sol[2] ) - mv.x ) > meanx )
2200 :	chl	770	\|\| ( ABS(( sol[3] - (16mx+8)sol[2] + (16my+8)sol[1] ) - mv.y ) > meany ) )
2201 :	chl	769	MBmask[mbnum]=0;
2202 :			else
2203 :			num++;
2204 :			}
2205 :
2206 :			} while ( (oldnum != num) && (num>=4) );
2207 :
2208 :			if (num < 4)
2209 :			{
2210 :			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;
2211 :			} else {
2212 :
2213 :			gmc.duv[0].x=(int)(sol[0]+0.5);
2214 :			gmc.duv[0].y=(int)(sol[3]+0.5);
2215 :
2216 :			gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5);
2217 :			gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5);
2218 :
2219 :			gmc.duv[2].x=0;
2220 :			gmc.duv[2].y=0;
2221 :	chl	619	}
2222 :	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);
2223 :	chl	619
2224 :	chl	769	free(MBmask);
2225 :
2226 :	chl	619	return gmc;
2227 :			}

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