1 |
/***************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Motion Compensation module - |
* - Motion Compensation related code - |
5 |
* |
* |
6 |
* Copyright(C) 2002 Michael Militzer <michael@xvid.org> |
* Copyright(C) 2002 Peter Ross <pross@xvid.org> |
7 |
* Copyright(C) 2002 Edouard Gomez <ed.gomez@wanadoo.fr> |
* 2003 Christoph Lampert <gruel@web.de> |
|
* Copyright(C) 2002 Christoph Lampert <gruel@web.de> |
|
|
* |
|
|
* This program is an implementation of a part of one or more MPEG-4 |
|
|
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
|
|
* to use this software module in hardware or software products are |
|
|
* advised that its use may infringe existing patents or copyrights, and |
|
|
* any such use would be at such party's own risk. The original |
|
|
* developer of this software module and his/her company, and subsequent |
|
|
* editors and their companies, will have no liability for use of this |
|
|
* software or modifications or derivatives thereof. |
|
8 |
* |
* |
9 |
* This program is free software; you can redistribute it and/or modify |
* This program is free software; you can redistribute it and/or modify |
10 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
20 |
* along with this program; if not, write to the Free Software |
* along with this program; if not, write to the Free Software |
21 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
22 |
* |
* |
23 |
*************************************************************************/ |
* $Id: motion_comp.c,v 1.18.2.7 2003-06-28 15:53:07 chl Exp $ |
24 |
|
* |
25 |
|
****************************************************************************/ |
26 |
|
|
27 |
|
#include <stdio.h> |
28 |
|
|
29 |
#include "../encoder.h" |
#include "../encoder.h" |
30 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
31 |
#include "../image/interpolate8x8.h" |
#include "../image/interpolate8x8.h" |
32 |
|
#include "../image/reduced.h" |
33 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
34 |
#include "motion.h" |
#include "motion.h" |
35 |
|
|
36 |
#define ABS(X) (((X)>0)?(X):-(X)) |
#ifndef RSHIFT |
37 |
#define SIGN(X) (((X)>0)?1:-1) |
#define RSHIFT(a,b) ((a) > 0 ? ((a) + (1<<((b)-1)))>>(b) : ((a) + (1<<((b)-1))-1)>>(b)) |
38 |
|
#endif |
39 |
|
|
40 |
|
/* assume b>0 */ |
41 |
|
#ifndef RDIV |
42 |
|
#define RDIV(a,b) (((a)>0 ? (a) + ((b)>>1) : (a) - ((b)>>1))/(b)) |
43 |
|
#endif |
44 |
|
|
45 |
|
|
46 |
|
/* This is borrowed from bitstream.c until we find a common solution */ |
47 |
|
|
48 |
|
static uint32_t __inline |
49 |
|
log2bin(uint32_t value) |
50 |
|
{ |
51 |
|
/* Changed by Chenm001 */ |
52 |
|
#if !defined(_MSC_VER) |
53 |
|
int n = 0; |
54 |
|
|
55 |
|
while (value) { |
56 |
|
value >>= 1; |
57 |
|
n++; |
58 |
|
} |
59 |
|
return n; |
60 |
|
#else |
61 |
|
__asm { |
62 |
|
bsr eax, value |
63 |
|
inc eax |
64 |
|
} |
65 |
|
#endif |
66 |
|
} |
67 |
|
|
68 |
|
|
69 |
static __inline void |
static __inline void |
70 |
compensate8x8_halfpel(int16_t * const dct_codes, |
compensate16x16_interpolate(int16_t * const dct_codes, |
71 |
uint8_t * const cur, |
uint8_t * const cur, |
72 |
const uint8_t * const ref, |
const uint8_t * const ref, |
73 |
const uint8_t * const refh, |
const uint8_t * const refh, |
74 |
const uint8_t * const refv, |
const uint8_t * const refv, |
75 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
76 |
const uint32_t x, |
uint8_t * const tmp, |
77 |
const uint32_t y, |
uint32_t x, |
78 |
|
uint32_t y, |
79 |
const int32_t dx, |
const int32_t dx, |
80 |
const int dy, |
const int32_t dy, |
81 |
const uint32_t stride) |
const int32_t stride, |
82 |
|
const int quarterpel, |
83 |
|
const int reduced_resolution, |
84 |
|
const int32_t rounding) |
85 |
{ |
{ |
86 |
int32_t ddx, ddy; |
const uint8_t * ptr; |
87 |
|
|
88 |
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
if (!reduced_resolution) { |
|
{ |
|
|
case 0: |
|
|
ddx = dx / 2; |
|
|
ddy = dy / 2; |
|
|
transfer_8to16sub(dct_codes, cur + y * stride + x, |
|
|
ref + (int) ((y + ddy) * stride + x + ddx), stride); |
|
|
break; |
|
89 |
|
|
90 |
case 1: |
if(quarterpel) { |
91 |
ddx = dx / 2; |
if ((dx&3) | (dy&3)) { |
92 |
ddy = (dy - 1) / 2; |
interpolate16x16_quarterpel(tmp - y * stride - x, |
93 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
(uint8_t *) ref, tmp + 32, |
94 |
refv + (int) ((y + ddy) * stride + x + ddx), stride); |
tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); |
95 |
break; |
ptr = tmp; |
96 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; /* fullpixel position */ |
97 |
|
|
98 |
case 2: |
} else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
|
ddx = (dx - 1) / 2; |
|
|
ddy = dy / 2; |
|
|
transfer_8to16sub(dct_codes, cur + y * stride + x, |
|
|
refh + (int) ((y + ddy) * stride + x + ddx), stride); |
|
|
break; |
|
99 |
|
|
|
default: // case 3: |
|
|
ddx = (dx - 1) / 2; |
|
|
ddy = (dy - 1) / 2; |
|
100 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
101 |
refhv + (int) ((y + ddy) * stride + x + ddx), stride); |
ptr, stride); |
102 |
break; |
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
103 |
|
ptr + 8, stride); |
104 |
|
transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
105 |
|
ptr + 8*stride, stride); |
106 |
|
transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
107 |
|
ptr + 8*stride + 8, stride); |
108 |
|
|
109 |
|
} else { /* reduced_resolution */ |
110 |
|
|
111 |
|
x *= 2; y *= 2; |
112 |
|
|
113 |
|
ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
114 |
|
|
115 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
116 |
|
filter_diff_18x18_to_8x8(dct_codes, ptr, stride); |
117 |
|
|
118 |
|
filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
119 |
|
filter_diff_18x18_to_8x8(dct_codes+64, ptr + 16, stride); |
120 |
|
|
121 |
|
filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
122 |
|
filter_diff_18x18_to_8x8(dct_codes+128, ptr + 16*stride, stride); |
123 |
|
|
124 |
|
filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
125 |
|
filter_diff_18x18_to_8x8(dct_codes+192, ptr + 16*stride + 16, stride); |
126 |
|
|
127 |
|
transfer32x32_copy(cur + y*stride + x, ptr, stride); |
128 |
} |
} |
129 |
} |
} |
130 |
|
|
131 |
|
static __inline void |
132 |
|
compensate8x8_interpolate( int16_t * const dct_codes, |
133 |
|
uint8_t * const cur, |
134 |
|
const uint8_t * const ref, |
135 |
|
const uint8_t * const refh, |
136 |
|
const uint8_t * const refv, |
137 |
|
const uint8_t * const refhv, |
138 |
|
uint8_t * const tmp, |
139 |
|
uint32_t x, |
140 |
|
uint32_t y, |
141 |
|
const int32_t dx, |
142 |
|
const int32_t dy, |
143 |
|
const int32_t stride, |
144 |
|
const int32_t quarterpel, |
145 |
|
const int reduced_resolution, |
146 |
|
const int32_t rounding) |
147 |
|
{ |
148 |
|
const uint8_t * ptr; |
149 |
|
|
150 |
|
if (!reduced_resolution) { |
151 |
|
|
152 |
|
if(quarterpel) { |
153 |
|
if ((dx&3) | (dy&3)) { |
154 |
|
interpolate8x8_quarterpel(tmp - y*stride - x, |
155 |
|
(uint8_t *) ref, tmp + 32, |
156 |
|
tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); |
157 |
|
ptr = tmp; |
158 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; /* fullpixel position */ |
159 |
|
} else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
160 |
|
|
161 |
|
transfer_8to16sub(dct_codes, cur + y * stride + x, ptr, stride); |
162 |
|
|
163 |
|
} else { /* reduced_resolution */ |
164 |
|
|
165 |
|
x *= 2; y *= 2; |
166 |
|
|
167 |
|
ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
168 |
|
|
169 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
170 |
|
filter_diff_18x18_to_8x8(dct_codes, ptr, stride); |
171 |
|
|
172 |
|
transfer16x16_copy(cur + y*stride + x, ptr, stride); |
173 |
|
} |
174 |
|
} |
175 |
|
|
176 |
|
/* XXX: slow, inelegant... */ |
177 |
|
static void |
178 |
|
interpolate18x18_switch(uint8_t * const cur, |
179 |
|
const uint8_t * const refn, |
180 |
|
const uint32_t x, |
181 |
|
const uint32_t y, |
182 |
|
const int32_t dx, |
183 |
|
const int dy, |
184 |
|
const int32_t stride, |
185 |
|
const int32_t rounding) |
186 |
|
{ |
187 |
|
interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); |
188 |
|
interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); |
189 |
|
interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); |
190 |
|
|
191 |
|
interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); |
192 |
|
interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); |
193 |
|
interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); |
194 |
|
|
195 |
|
interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); |
196 |
|
interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); |
197 |
|
interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); |
198 |
|
} |
199 |
|
|
200 |
|
static void |
201 |
|
CompensateChroma( int dx, int dy, |
202 |
|
const int i, const int j, |
203 |
|
IMAGE * const Cur, |
204 |
|
const IMAGE * const Ref, |
205 |
|
uint8_t * const temp, |
206 |
|
int16_t * const coeff, |
207 |
|
const int32_t stride, |
208 |
|
const int rounding, |
209 |
|
const int rrv) |
210 |
|
{ /* uv-block-based compensation */ |
211 |
|
|
212 |
|
if (!rrv) { |
213 |
|
transfer_8to16sub(coeff, Cur->u + 8 * j * stride + 8 * i, |
214 |
|
interpolate8x8_switch2(temp, Ref->u, 8 * i, 8 * j, |
215 |
|
dx, dy, stride, rounding), |
216 |
|
stride); |
217 |
|
transfer_8to16sub(coeff + 64, Cur->v + 8 * j * stride + 8 * i, |
218 |
|
interpolate8x8_switch2(temp, Ref->v, 8 * i, 8 * j, |
219 |
|
dx, dy, stride, rounding), |
220 |
|
stride); |
221 |
|
} else { |
222 |
|
uint8_t * current, * reference; |
223 |
|
|
224 |
|
current = Cur->u + 16*j*stride + 16*i; |
225 |
|
reference = temp - 16*j*stride - 16*i; |
226 |
|
interpolate18x18_switch(reference, Ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
227 |
|
filter_18x18_to_8x8(coeff, current, stride); |
228 |
|
filter_diff_18x18_to_8x8(coeff, temp, stride); |
229 |
|
transfer16x16_copy(current, temp, stride); |
230 |
|
|
231 |
|
current = Cur->v + 16*j*stride + 16*i; |
232 |
|
interpolate18x18_switch(reference, Ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
233 |
|
filter_18x18_to_8x8(coeff + 64, current, stride); |
234 |
|
filter_diff_18x18_to_8x8(coeff + 64, temp, stride); |
235 |
|
transfer16x16_copy(current, temp, stride); |
236 |
|
} |
237 |
|
} |
238 |
|
|
239 |
void |
void |
240 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
244 |
const IMAGE * const refh, |
const IMAGE * const refh, |
245 |
const IMAGE * const refv, |
const IMAGE * const refv, |
246 |
const IMAGE * const refhv, |
const IMAGE * const refhv, |
247 |
|
const IMAGE * const refGMC, |
248 |
IMAGE * const cur, |
IMAGE * const cur, |
249 |
int16_t * dct_codes, |
int16_t * dct_codes, |
250 |
const uint32_t width, |
const uint32_t width, |
251 |
const uint32_t height, |
const uint32_t height, |
252 |
const uint32_t edged_width, |
const uint32_t edged_width, |
253 |
const uint32_t rounding) |
const int32_t quarterpel, |
254 |
|
const int reduced_resolution, |
255 |
|
const int32_t rounding) |
256 |
{ |
{ |
257 |
static const uint32_t roundtab[16] = |
int32_t dx; |
258 |
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
int32_t dy; |
259 |
|
|
260 |
|
uint8_t * const tmp = refv->u; |
261 |
|
|
262 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
if ( (!reduced_resolution) && (mb->mode == MODE_NOT_CODED) ) { /* quick copy for early SKIP */ |
263 |
int32_t dx = mb->mvs[0].x; |
/* early SKIP is only activated in P-VOPs, not in S-VOPs, so mcsel can never be 1 */ |
|
int32_t dy = mb->mvs[0].y; |
|
264 |
|
|
265 |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
266 |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
ref->y + 16 * (i + j * edged_width), |
|
edged_width); |
|
|
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i + 8, 16 * j, dx, dy, |
|
267 |
edged_width); |
edged_width); |
|
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i, 16 * j + 8, dx, dy, |
|
|
edged_width); |
|
|
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, dx, |
|
|
dy, edged_width); |
|
|
|
|
|
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
|
|
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
|
|
|
|
/* uv-image-based compensation */ |
|
|
|
|
|
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, rounding); |
|
|
transfer_8to16sub(&dct_codes[4 * 64], |
|
|
cur->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
edged_width / 2); |
|
268 |
|
|
269 |
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
270 |
edged_width / 2, rounding); |
ref->u + 8 * (i + j * edged_width/2), |
271 |
transfer_8to16sub(&dct_codes[5 * 64], |
edged_width / 2); |
272 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), |
273 |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
ref->v + 8 * (i + j * edged_width/2), |
274 |
edged_width / 2); |
edged_width / 2); |
275 |
|
return; |
276 |
|
} |
277 |
|
|
278 |
|
if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER |
279 |
|
|| mb->mode == MODE_INTER_Q)) { |
280 |
|
|
281 |
|
/* reduced resolution + GMC: not possible */ |
282 |
|
|
283 |
|
if (mb->mcsel) { |
284 |
|
|
285 |
|
/* call normal routine once, easier than "if (mcsel)"ing all the time */ |
286 |
|
|
287 |
|
transfer_8to16sub(&dct_codes[0*64], cur->y + 16*j*edged_width + 16*i, |
288 |
|
refGMC->y + 16*j*edged_width + 16*i, edged_width); |
289 |
|
transfer_8to16sub(&dct_codes[1*64], cur->y + 16*j*edged_width + 16*i+8, |
290 |
|
refGMC->y + 16*j*edged_width + 16*i+8, edged_width); |
291 |
|
transfer_8to16sub(&dct_codes[2*64], cur->y + (16*j+8)*edged_width + 16*i, |
292 |
|
refGMC->y + (16*j+8)*edged_width + 16*i, edged_width); |
293 |
|
transfer_8to16sub(&dct_codes[3*64], cur->y + (16*j+8)*edged_width + 16*i+8, |
294 |
|
refGMC->y + (16*j+8)*edged_width + 16*i+8, edged_width); |
295 |
|
|
296 |
|
/* lumi is needed earlier for mode decision, but chroma should be done block-based, but it isn't, yet. */ |
297 |
|
|
298 |
|
transfer_8to16sub(&dct_codes[4 * 64], cur->u + 8 *j*edged_width/2 + 8*i, |
299 |
|
refGMC->u + 8 *j*edged_width/2 + 8*i, edged_width/2); |
300 |
|
|
301 |
|
transfer_8to16sub(&dct_codes[5 * 64], cur->v + 8*j* edged_width/2 + 8*i, |
302 |
|
refGMC->v + 8*j* edged_width/2 + 8*i, edged_width/2); |
303 |
|
|
304 |
|
return; |
305 |
|
} |
306 |
|
|
307 |
|
/* ordinary compensation */ |
308 |
|
|
309 |
|
dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
310 |
|
dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
311 |
|
|
312 |
|
if (reduced_resolution) { |
313 |
|
dx = RRV_MV_SCALEUP(dx); |
314 |
|
dy = RRV_MV_SCALEUP(dy); |
315 |
|
} |
316 |
|
|
317 |
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
318 |
|
refv->y, refhv->y, tmp, 16 * i, 16 * j, dx, dy, |
319 |
|
edged_width, quarterpel, reduced_resolution, rounding); |
320 |
|
|
321 |
|
if (quarterpel) { dx /= 2; dy /= 2; } |
322 |
|
|
323 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
324 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
325 |
|
|
326 |
|
} else { /* mode == MODE_INTER4V */ |
327 |
|
int k, sumx = 0, sumy = 0; |
328 |
|
const VECTOR * const mvs = (quarterpel ? mb->qmvs : mb->mvs); |
329 |
|
|
330 |
|
for (k = 0; k < 4; k++) { |
331 |
|
dx = mvs[k].x; |
332 |
|
dy = mvs[k].y; |
333 |
|
sumx += quarterpel ? dx/2 : dx; |
334 |
|
sumy += quarterpel ? dy/2 : dy; |
335 |
|
|
336 |
|
if (reduced_resolution){ |
337 |
|
dx = RRV_MV_SCALEUP(dx); |
338 |
|
dy = RRV_MV_SCALEUP(dy); |
339 |
|
} |
340 |
|
|
341 |
|
compensate8x8_interpolate(&dct_codes[k * 64], cur->y, ref->y, refh->y, |
342 |
|
refv->y, refhv->y, tmp, 16 * i + 8*(k&1), 16 * j + 8*(k>>1), dx, |
343 |
|
dy, edged_width, quarterpel, reduced_resolution, rounding); |
344 |
|
} |
345 |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
346 |
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
347 |
|
} |
348 |
|
|
349 |
} else // mode == MODE_INTER4V |
CompensateChroma(dx, dy, i, j, cur, ref, tmp, |
350 |
|
&dct_codes[4 * 64], edged_width / 2, rounding, reduced_resolution); |
351 |
|
} |
352 |
|
|
353 |
|
|
354 |
|
void |
355 |
|
MBMotionCompensationBVOP(MBParam * pParam, |
356 |
|
MACROBLOCK * const mb, |
357 |
|
const uint32_t i, |
358 |
|
const uint32_t j, |
359 |
|
IMAGE * const cur, |
360 |
|
const IMAGE * const f_ref, |
361 |
|
const IMAGE * const f_refh, |
362 |
|
const IMAGE * const f_refv, |
363 |
|
const IMAGE * const f_refhv, |
364 |
|
const IMAGE * const b_ref, |
365 |
|
const IMAGE * const b_refh, |
366 |
|
const IMAGE * const b_refv, |
367 |
|
const IMAGE * const b_refhv, |
368 |
|
int16_t * dct_codes) |
369 |
{ |
{ |
370 |
int32_t sum, dx, dy; |
const uint32_t edged_width = pParam->edged_width; |
371 |
|
int32_t dx, dy, b_dx, b_dy, sumx, sumy, b_sumx, b_sumy; |
372 |
|
int k; |
373 |
|
const int quarterpel = pParam->vol_flags & XVID_VOL_QUARTERPEL; |
374 |
|
const uint8_t * ptr1, * ptr2; |
375 |
|
uint8_t * const tmp = f_refv->u; |
376 |
|
const VECTOR * const fmvs = (quarterpel ? mb->qmvs : mb->mvs); |
377 |
|
const VECTOR * const bmvs = (quarterpel ? mb->b_qmvs : mb->b_mvs); |
378 |
|
|
379 |
|
switch (mb->mode) { |
380 |
|
case MODE_FORWARD: |
381 |
|
dx = fmvs->x; dy = fmvs->y; |
382 |
|
|
383 |
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
384 |
|
f_refv->y, f_refhv->y, tmp, 16 * i, 16 * j, dx, |
385 |
|
dy, edged_width, quarterpel, 0, 0); |
386 |
|
|
387 |
|
if (quarterpel) { dx /= 2; dy /= 2; } |
388 |
|
|
389 |
|
CompensateChroma( (dx >> 1) + roundtab_79[dx & 0x3], |
390 |
|
(dy >> 1) + roundtab_79[dy & 0x3], |
391 |
|
i, j, cur, f_ref, tmp, |
392 |
|
&dct_codes[4 * 64], edged_width / 2, 0, 0); |
393 |
|
|
394 |
|
return; |
395 |
|
|
396 |
|
case MODE_BACKWARD: |
397 |
|
b_dx = bmvs->x; b_dy = bmvs->y; |
398 |
|
|
399 |
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
400 |
|
b_refv->y, b_refhv->y, tmp, 16 * i, 16 * j, b_dx, |
401 |
|
b_dy, edged_width, quarterpel, 0, 0); |
402 |
|
|
403 |
|
if (quarterpel) { b_dx /= 2; b_dy /= 2; } |
404 |
|
|
405 |
|
CompensateChroma( (b_dx >> 1) + roundtab_79[b_dx & 0x3], |
406 |
|
(b_dy >> 1) + roundtab_79[b_dy & 0x3], |
407 |
|
i, j, cur, b_ref, tmp, |
408 |
|
&dct_codes[4 * 64], edged_width / 2, 0, 0); |
409 |
|
|
410 |
|
return; |
411 |
|
|
412 |
|
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
413 |
|
case MODE_DIRECT_NO4V: |
414 |
|
dx = fmvs->x; dy = fmvs->y; |
415 |
|
b_dx = bmvs->x; b_dy = bmvs->y; |
416 |
|
|
417 |
|
if (quarterpel) { |
418 |
|
|
419 |
|
if ((dx&3) | (dy&3)) { |
420 |
|
interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width, |
421 |
|
(uint8_t *) f_ref->y, tmp + 32, |
422 |
|
tmp + 64, tmp + 96, 16*i, 16*j, dx, dy, edged_width, 0); |
423 |
|
ptr1 = tmp; |
424 |
|
} else ptr1 = f_ref->y + (16*j + dy/4)*edged_width + 16*i + dx/4; /* fullpixel position */ |
425 |
|
|
426 |
|
if ((b_dx&3) | (b_dy&3)) { |
427 |
|
interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width + 16, |
428 |
|
(uint8_t *) b_ref->y, tmp + 32, |
429 |
|
tmp + 64, tmp + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
430 |
|
ptr2 = tmp + 16; |
431 |
|
} else ptr2 = b_ref->y + (16*j + b_dy/4)*edged_width + 16*i + b_dx/4; /* fullpixel position */ |
432 |
|
|
433 |
|
b_dx /= 2; |
434 |
|
b_dy /= 2; |
435 |
|
dx /= 2; |
436 |
|
dy /= 2; |
437 |
|
|
438 |
|
} else { |
439 |
|
ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
440 |
|
i, j, 16, dx, dy, edged_width); |
441 |
|
|
442 |
|
ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
443 |
|
i, j, 16, b_dx, b_dy, edged_width); |
444 |
|
} |
445 |
|
for (k = 0; k < 4; k++) |
446 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
447 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
448 |
|
ptr1 + (k&1)*8 + (k>>1)*8*edged_width, |
449 |
|
ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width); |
450 |
|
|
451 |
|
|
452 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
453 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
454 |
|
|
455 |
|
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
456 |
|
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
457 |
|
|
458 |
|
break; |
459 |
|
|
460 |
|
default: /* MODE_DIRECT (or MODE_DIRECT_NONE_MV in case of bframes decoding) */ |
461 |
|
sumx = sumy = b_sumx = b_sumy = 0; |
462 |
|
|
463 |
|
for (k = 0; k < 4; k++) { |
464 |
|
|
465 |
|
dx = fmvs[k].x; dy = fmvs[k].y; |
466 |
|
b_dx = bmvs[k].x; b_dy = bmvs[k].y; |
467 |
|
|
468 |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
if (quarterpel) { |
469 |
refv->y, refhv->y, 16 * i, 16 * j, mb->mvs[0].x, |
sumx += dx/2; sumy += dy/2; |
470 |
mb->mvs[0].y, edged_width); |
b_sumx += b_dx/2; b_sumy += b_dy/2; |
471 |
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
|
472 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
if ((dx&3) | (dy&3)) { |
473 |
mb->mvs[1].x, mb->mvs[1].y, edged_width); |
interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width, |
474 |
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
(uint8_t *) f_ref->y, |
475 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
tmp + 32, tmp + 64, tmp + 96, |
476 |
mb->mvs[2].x, mb->mvs[2].y, edged_width); |
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
477 |
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
ptr1 = tmp; |
478 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
} else ptr1 = f_ref->y + (16*j + (k>>1)*8 + dy/4)*edged_width + 16*i + (k&1)*8 + dx/4; |
479 |
mb->mvs[3].x, mb->mvs[3].y, edged_width); |
|
480 |
|
if ((b_dx&3) | (b_dy&3)) { |
481 |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width + 16, |
482 |
dx = (sum ? SIGN(sum) * |
(uint8_t *) b_ref->y, |
483 |
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
tmp + 16, tmp + 32, tmp + 48, |
484 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
485 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
ptr2 = tmp + 16; |
486 |
dy = (sum ? SIGN(sum) * |
} else ptr2 = b_ref->y + (16*j + (k>>1)*8 + b_dy/4)*edged_width + 16*i + (k&1)*8 + b_dx/4; |
487 |
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
} else { |
488 |
|
sumx += dx; sumy += dy; |
489 |
/* uv-block-based compensation */ |
b_sumx += b_dx; b_sumy += b_dy; |
490 |
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
|
491 |
edged_width / 2, rounding); |
ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
492 |
transfer_8to16sub(&dct_codes[4 * 64], |
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, edged_width); |
493 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
494 |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy, edged_width); |
495 |
|
} |
496 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
497 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
498 |
|
ptr1, ptr2, edged_width); |
499 |
|
|
500 |
|
} |
501 |
|
|
502 |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
503 |
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
504 |
|
b_dx = (b_sumx >> 3) + roundtab_76[b_sumx & 0xf]; |
505 |
|
b_dy = (b_sumy >> 3) + roundtab_76[b_sumy & 0xf]; |
506 |
|
|
507 |
|
break; |
508 |
|
} |
509 |
|
|
510 |
|
/* v block-based chroma interpolation for direct and interpolate modes */ |
511 |
|
transfer_8to16sub2(&dct_codes[4 * 64], |
512 |
|
cur->u + (j * 8) * edged_width / 2 + (i * 8), |
513 |
|
interpolate8x8_switch2(tmp, b_ref->u, 8 * i, 8 * j, |
514 |
|
b_dx, b_dy, edged_width / 2, 0), |
515 |
|
interpolate8x8_switch2(tmp + 8, f_ref->u, 8 * i, 8 * j, |
516 |
|
dx, dy, edged_width / 2, 0), |
517 |
edged_width / 2); |
edged_width / 2); |
518 |
|
|
519 |
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
transfer_8to16sub2(&dct_codes[5 * 64], |
520 |
edged_width / 2, rounding); |
cur->v + (j * 8) * edged_width / 2 + (i * 8), |
521 |
transfer_8to16sub(&dct_codes[5 * 64], |
interpolate8x8_switch2(tmp, b_ref->v, 8 * i, 8 * j, |
522 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
b_dx, b_dy, edged_width / 2, 0), |
523 |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(tmp + 8, f_ref->v, 8 * i, 8 * j, |
524 |
|
dx, dy, edged_width / 2, 0), |
525 |
edged_width / 2); |
edged_width / 2); |
526 |
} |
} |
|
} |
|