| 1 |
|
/***************************************************************************** |
| 2 |
|
* |
| 3 |
|
* XVID MPEG-4 VIDEO CODEC |
| 4 |
|
* - Motion Compensation related code - |
| 5 |
|
* |
| 6 |
|
* Copyright(C) 2002 Peter Ross <pross@xvid.org> |
| 7 |
|
* 2003 Christoph Lampert <gruel@web.de> |
| 8 |
|
* |
| 9 |
|
* 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 |
| 11 |
|
* the Free Software Foundation ; either version 2 of the License, or |
| 12 |
|
* (at your option) any later version. |
| 13 |
|
* |
| 14 |
|
* This program is distributed in the hope that it will be useful, |
| 15 |
|
* but WITHOUT ANY WARRANTY ; without even the implied warranty of |
| 16 |
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 |
|
* GNU General Public License for more details. |
| 18 |
|
* |
| 19 |
|
* You should have received a copy of the GNU General Public License |
| 20 |
|
* along with this program ; if not, write to the Free Software |
| 21 |
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 22 |
|
* |
| 23 |
|
* $Id: motion_comp.c,v 1.20 2004-03-22 22:36:24 edgomez 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/qpel.h" |
| 33 |
|
#include "../image/reduced.h" |
| 34 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
| 35 |
|
#include "motion.h" |
| 36 |
|
|
| 37 |
|
#ifndef RSHIFT |
| 38 |
|
#define RSHIFT(a,b) ((a) > 0 ? ((a) + (1<<((b)-1)))>>(b) : ((a) + (1<<((b)-1))-1)>>(b)) |
| 39 |
|
#endif |
| 40 |
|
|
| 41 |
|
/* assume b>0 */ |
| 42 |
|
#ifndef RDIV |
| 43 |
|
#define RDIV(a,b) (((a)>0 ? (a) + ((b)>>1) : (a) - ((b)>>1))/(b)) |
| 44 |
|
#endif |
| 45 |
|
|
| 46 |
|
|
| 47 |
|
/* This is borrowed from bitstream.c until we find a common solution */ |
| 48 |
|
|
| 49 |
|
static uint32_t __inline |
| 50 |
|
log2bin(uint32_t value) |
| 51 |
|
{ |
| 52 |
|
/* Changed by Chenm001 */ |
| 53 |
|
#if !defined(_MSC_VER) |
| 54 |
|
int n = 0; |
| 55 |
|
|
| 56 |
|
while (value) { |
| 57 |
|
value >>= 1; |
| 58 |
|
n++; |
| 59 |
|
} |
| 60 |
|
return n; |
| 61 |
|
#else |
| 62 |
|
__asm { |
| 63 |
|
bsr eax, value |
| 64 |
|
inc eax |
| 65 |
|
} |
| 66 |
|
#endif |
| 67 |
|
} |
| 68 |
|
|
| 69 |
#define ABS(X) (((X)>0)?(X):-(X)) |
/* |
| 70 |
#define SIGN(X) (((X)>0)?1:-1) |
* getref: calculate reference image pointer |
| 71 |
|
* the decision to use interpolation h/v/hv or the normal image is |
| 72 |
|
* based on dx & dy. |
| 73 |
|
*/ |
| 74 |
|
|
| 75 |
static __inline void compensate8x8_halfpel( |
static __inline const uint8_t * |
| 76 |
int16_t * const dct_codes, |
get_ref(const uint8_t * const refn, |
| 77 |
|
const uint8_t * const refh, |
| 78 |
|
const uint8_t * const refv, |
| 79 |
|
const uint8_t * const refhv, |
| 80 |
|
const uint32_t x, |
| 81 |
|
const uint32_t y, |
| 82 |
|
const uint32_t block, |
| 83 |
|
const int32_t dx, |
| 84 |
|
const int32_t dy, |
| 85 |
|
const int32_t stride) |
| 86 |
|
{ |
| 87 |
|
switch (((dx & 1) << 1) + (dy & 1)) { |
| 88 |
|
case 0: |
| 89 |
|
return refn + (int) ((x * block + dx / 2) + (y * block + dy / 2) * stride); |
| 90 |
|
case 1: |
| 91 |
|
return refv + (int) ((x * block + dx / 2) + (y * block + (dy - 1) / 2) * stride); |
| 92 |
|
case 2: |
| 93 |
|
return refh + (int) ((x * block + (dx - 1) / 2) + (y * block + dy / 2) * stride); |
| 94 |
|
default: |
| 95 |
|
return refhv + (int) ((x * block + (dx - 1) / 2) + (y * block + (dy - 1) / 2) * stride); |
| 96 |
|
} |
| 97 |
|
} |
| 98 |
|
|
| 99 |
|
static __inline void |
| 100 |
|
compensate16x16_interpolate(int16_t * const dct_codes, |
| 101 |
uint8_t * const cur, |
uint8_t * const cur, |
| 102 |
const uint8_t * const ref, |
const uint8_t * const ref, |
| 103 |
const uint8_t * const refh, |
const uint8_t * const refh, |
| 104 |
const uint8_t * const refv, |
const uint8_t * const refv, |
| 105 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
| 106 |
const uint32_t x, const uint32_t y, |
uint8_t * const tmp, |
| 107 |
const int32_t dx, const int dy, |
uint32_t x, |
| 108 |
const uint32_t stride) |
uint32_t y, |
| 109 |
|
const int32_t dx, |
| 110 |
|
const int32_t dy, |
| 111 |
|
const int32_t stride, |
| 112 |
|
const int quarterpel, |
| 113 |
|
const int reduced_resolution, |
| 114 |
|
const int32_t rounding) |
| 115 |
{ |
{ |
| 116 |
int32_t ddx,ddy; |
const uint8_t * ptr; |
| 117 |
|
|
| 118 |
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 + (y+ddy)*stride + x+ddx, stride); |
|
|
break; |
|
| 119 |
|
|
| 120 |
case 1 : |
if(quarterpel) { |
| 121 |
ddx = dx/2; |
if ((dx&3) | (dy&3)) { |
| 122 |
ddy = (dy-1)/2; |
#if defined(ARCH_IS_IA32) /* new_interpolate is only faster on x86 (MMX) machines */ |
| 123 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
new_interpolate16x16_quarterpel(tmp - y * stride - x, |
| 124 |
refv + (y+ddy)*stride + x+ddx, stride); |
(uint8_t *) ref, tmp + 32, |
| 125 |
break; |
tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); |
| 126 |
|
#else |
| 127 |
|
interpolate16x16_quarterpel(tmp - y * stride - x, |
| 128 |
|
(uint8_t *) ref, tmp + 32, |
| 129 |
|
tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); |
| 130 |
|
#endif |
| 131 |
|
ptr = tmp; |
| 132 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; /* fullpixel position */ |
| 133 |
|
|
| 134 |
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 + (y+ddy)*stride + x+ddx, stride); |
|
|
break; |
|
| 135 |
|
|
|
default : // case 3: |
|
|
ddx = (dx-1)/2; |
|
|
ddy = (dy-1)/2; |
|
| 136 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
| 137 |
refhv + (y+ddy)*stride + x+ddx, stride); |
ptr, stride); |
| 138 |
break; |
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
| 139 |
|
ptr + 8, stride); |
| 140 |
|
transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
| 141 |
|
ptr + 8*stride, stride); |
| 142 |
|
transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
| 143 |
|
ptr + 8*stride + 8, stride); |
| 144 |
|
|
| 145 |
|
} else { /* reduced_resolution */ |
| 146 |
|
|
| 147 |
|
x *= 2; y *= 2; |
| 148 |
|
|
| 149 |
|
ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
| 150 |
|
|
| 151 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
| 152 |
|
filter_diff_18x18_to_8x8(dct_codes, ptr, stride); |
| 153 |
|
|
| 154 |
|
filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
| 155 |
|
filter_diff_18x18_to_8x8(dct_codes+64, ptr + 16, stride); |
| 156 |
|
|
| 157 |
|
filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
| 158 |
|
filter_diff_18x18_to_8x8(dct_codes+128, ptr + 16*stride, stride); |
| 159 |
|
|
| 160 |
|
filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
| 161 |
|
filter_diff_18x18_to_8x8(dct_codes+192, ptr + 16*stride + 16, stride); |
| 162 |
|
|
| 163 |
|
transfer32x32_copy(cur + y*stride + x, ptr, stride); |
| 164 |
} |
} |
| 165 |
} |
} |
| 166 |
|
|
| 167 |
|
static __inline void |
| 168 |
|
compensate8x8_interpolate( int16_t * const dct_codes, |
| 169 |
|
uint8_t * const cur, |
| 170 |
|
const uint8_t * const ref, |
| 171 |
|
const uint8_t * const refh, |
| 172 |
|
const uint8_t * const refv, |
| 173 |
|
const uint8_t * const refhv, |
| 174 |
|
uint8_t * const tmp, |
| 175 |
|
uint32_t x, |
| 176 |
|
uint32_t y, |
| 177 |
|
const int32_t dx, |
| 178 |
|
const int32_t dy, |
| 179 |
|
const int32_t stride, |
| 180 |
|
const int32_t quarterpel, |
| 181 |
|
const int reduced_resolution, |
| 182 |
|
const int32_t rounding) |
| 183 |
|
{ |
| 184 |
|
const uint8_t * ptr; |
| 185 |
|
|
| 186 |
|
if (!reduced_resolution) { |
| 187 |
|
|
| 188 |
void MBMotionCompensation( |
if(quarterpel) { |
| 189 |
MACROBLOCK * const mb, |
if ((dx&3) | (dy&3)) { |
| 190 |
|
#if defined(ARCH_IS_IA32) /* new_interpolate is only faster on x86 (MMX) machines */ |
| 191 |
|
new_interpolate8x8_quarterpel(tmp - y*stride - x, |
| 192 |
|
(uint8_t *) ref, tmp + 32, |
| 193 |
|
tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); |
| 194 |
|
#else |
| 195 |
|
interpolate8x8_quarterpel(tmp - y*stride - x, |
| 196 |
|
(uint8_t *) ref, tmp + 32, |
| 197 |
|
tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding); |
| 198 |
|
#endif |
| 199 |
|
ptr = tmp; |
| 200 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; /* fullpixel position */ |
| 201 |
|
} else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
| 202 |
|
|
| 203 |
|
transfer_8to16sub(dct_codes, cur + y * stride + x, ptr, stride); |
| 204 |
|
|
| 205 |
|
} else { /* reduced_resolution */ |
| 206 |
|
|
| 207 |
|
x *= 2; y *= 2; |
| 208 |
|
|
| 209 |
|
ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
| 210 |
|
|
| 211 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
| 212 |
|
filter_diff_18x18_to_8x8(dct_codes, ptr, stride); |
| 213 |
|
|
| 214 |
|
transfer16x16_copy(cur + y*stride + x, ptr, stride); |
| 215 |
|
} |
| 216 |
|
} |
| 217 |
|
|
| 218 |
|
/* XXX: slow, inelegant... */ |
| 219 |
|
static void |
| 220 |
|
interpolate18x18_switch(uint8_t * const cur, |
| 221 |
|
const uint8_t * const refn, |
| 222 |
|
const uint32_t x, |
| 223 |
|
const uint32_t y, |
| 224 |
|
const int32_t dx, |
| 225 |
|
const int dy, |
| 226 |
|
const int32_t stride, |
| 227 |
|
const int32_t rounding) |
| 228 |
|
{ |
| 229 |
|
interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); |
| 230 |
|
interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); |
| 231 |
|
interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); |
| 232 |
|
|
| 233 |
|
interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); |
| 234 |
|
interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); |
| 235 |
|
interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); |
| 236 |
|
|
| 237 |
|
interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); |
| 238 |
|
interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); |
| 239 |
|
interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); |
| 240 |
|
} |
| 241 |
|
|
| 242 |
|
static void |
| 243 |
|
CompensateChroma( int dx, int dy, |
| 244 |
|
const int i, const int j, |
| 245 |
|
IMAGE * const Cur, |
| 246 |
|
const IMAGE * const Ref, |
| 247 |
|
uint8_t * const temp, |
| 248 |
|
int16_t * const coeff, |
| 249 |
|
const int32_t stride, |
| 250 |
|
const int rounding, |
| 251 |
|
const int rrv) |
| 252 |
|
{ /* uv-block-based compensation */ |
| 253 |
|
|
| 254 |
|
if (!rrv) { |
| 255 |
|
transfer_8to16sub(coeff, Cur->u + 8 * j * stride + 8 * i, |
| 256 |
|
interpolate8x8_switch2(temp, Ref->u, 8 * i, 8 * j, |
| 257 |
|
dx, dy, stride, rounding), |
| 258 |
|
stride); |
| 259 |
|
transfer_8to16sub(coeff + 64, Cur->v + 8 * j * stride + 8 * i, |
| 260 |
|
interpolate8x8_switch2(temp, Ref->v, 8 * i, 8 * j, |
| 261 |
|
dx, dy, stride, rounding), |
| 262 |
|
stride); |
| 263 |
|
} else { |
| 264 |
|
uint8_t * current, * reference; |
| 265 |
|
|
| 266 |
|
current = Cur->u + 16*j*stride + 16*i; |
| 267 |
|
reference = temp - 16*j*stride - 16*i; |
| 268 |
|
interpolate18x18_switch(reference, Ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
| 269 |
|
filter_18x18_to_8x8(coeff, current, stride); |
| 270 |
|
filter_diff_18x18_to_8x8(coeff, temp, stride); |
| 271 |
|
transfer16x16_copy(current, temp, stride); |
| 272 |
|
|
| 273 |
|
current = Cur->v + 16*j*stride + 16*i; |
| 274 |
|
interpolate18x18_switch(reference, Ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
| 275 |
|
filter_18x18_to_8x8(coeff + 64, current, stride); |
| 276 |
|
filter_diff_18x18_to_8x8(coeff + 64, temp, stride); |
| 277 |
|
transfer16x16_copy(current, temp, stride); |
| 278 |
|
} |
| 279 |
|
} |
| 280 |
|
|
| 281 |
|
void |
| 282 |
|
MBMotionCompensation(MACROBLOCK * const mb, |
| 283 |
const uint32_t i, |
const uint32_t i, |
| 284 |
const uint32_t j, |
const uint32_t j, |
| 285 |
const IMAGE * const ref, |
const IMAGE * const ref, |
| 286 |
const IMAGE * const refh, |
const IMAGE * const refh, |
| 287 |
const IMAGE * const refv, |
const IMAGE * const refv, |
| 288 |
const IMAGE * const refhv, |
const IMAGE * const refhv, |
| 289 |
|
const IMAGE * const refGMC, |
| 290 |
IMAGE * const cur, |
IMAGE * const cur, |
| 291 |
int16_t dct_codes[][64], |
int16_t * dct_codes, |
| 292 |
const uint32_t width, |
const uint32_t width, |
| 293 |
const uint32_t height, |
const uint32_t height, |
| 294 |
const uint32_t edged_width, |
const uint32_t edged_width, |
| 295 |
const uint32_t rounding) |
const int32_t quarterpel, |
| 296 |
|
const int reduced_resolution, |
| 297 |
|
const int32_t rounding) |
| 298 |
{ |
{ |
| 299 |
static const uint32_t roundtab[16] = |
int32_t dx; |
| 300 |
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
int32_t dy; |
| 301 |
|
|
| 302 |
|
uint8_t * const tmp = refv->u; |
| 303 |
|
|
| 304 |
|
if ( (!reduced_resolution) && (mb->mode == MODE_NOT_CODED) ) { /* quick copy for early SKIP */ |
| 305 |
|
/* early SKIP is only activated in P-VOPs, not in S-VOPs, so mcsel can never be 1 */ |
| 306 |
|
|
| 307 |
|
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
| 308 |
|
ref->y + 16 * (i + j * edged_width), |
| 309 |
|
edged_width); |
| 310 |
|
|
| 311 |
|
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
| 312 |
|
ref->u + 8 * (i + j * edged_width/2), |
| 313 |
|
edged_width / 2); |
| 314 |
|
transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), |
| 315 |
|
ref->v + 8 * (i + j * edged_width/2), |
| 316 |
|
edged_width / 2); |
| 317 |
|
return; |
| 318 |
|
} |
| 319 |
|
|
| 320 |
|
if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER |
| 321 |
|
|| mb->mode == MODE_INTER_Q)) { |
| 322 |
|
|
| 323 |
|
/* reduced resolution + GMC: not possible */ |
| 324 |
|
|
| 325 |
|
if (mb->mcsel) { |
| 326 |
|
|
| 327 |
|
/* call normal routine once, easier than "if (mcsel)"ing all the time */ |
| 328 |
|
|
| 329 |
|
transfer_8to16sub(&dct_codes[0*64], cur->y + 16*j*edged_width + 16*i, |
| 330 |
|
refGMC->y + 16*j*edged_width + 16*i, edged_width); |
| 331 |
|
transfer_8to16sub(&dct_codes[1*64], cur->y + 16*j*edged_width + 16*i+8, |
| 332 |
|
refGMC->y + 16*j*edged_width + 16*i+8, edged_width); |
| 333 |
|
transfer_8to16sub(&dct_codes[2*64], cur->y + (16*j+8)*edged_width + 16*i, |
| 334 |
|
refGMC->y + (16*j+8)*edged_width + 16*i, edged_width); |
| 335 |
|
transfer_8to16sub(&dct_codes[3*64], cur->y + (16*j+8)*edged_width + 16*i+8, |
| 336 |
|
refGMC->y + (16*j+8)*edged_width + 16*i+8, edged_width); |
| 337 |
|
|
| 338 |
|
/* lumi is needed earlier for mode decision, but chroma should be done block-based, but it isn't, yet. */ |
| 339 |
|
|
| 340 |
|
transfer_8to16sub(&dct_codes[4 * 64], cur->u + 8 *j*edged_width/2 + 8*i, |
| 341 |
|
refGMC->u + 8 *j*edged_width/2 + 8*i, edged_width/2); |
| 342 |
|
|
| 343 |
|
transfer_8to16sub(&dct_codes[5 * 64], cur->v + 8*j* edged_width/2 + 8*i, |
| 344 |
|
refGMC->v + 8*j* edged_width/2 + 8*i, edged_width/2); |
| 345 |
|
|
| 346 |
|
return; |
| 347 |
|
} |
| 348 |
|
|
| 349 |
|
/* ordinary compensation */ |
| 350 |
|
|
| 351 |
|
dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
| 352 |
|
dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
| 353 |
|
|
| 354 |
|
if (reduced_resolution) { |
| 355 |
|
dx = RRV_MV_SCALEUP(dx); |
| 356 |
|
dy = RRV_MV_SCALEUP(dy); |
| 357 |
|
} |
| 358 |
|
|
| 359 |
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
| 360 |
|
refv->y, refhv->y, tmp, 16 * i, 16 * j, dx, dy, |
| 361 |
|
edged_width, quarterpel, reduced_resolution, rounding); |
| 362 |
|
|
| 363 |
|
if (quarterpel) { dx /= 2; dy /= 2; } |
| 364 |
|
|
| 365 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
| 366 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
| 367 |
|
|
| 368 |
|
} else { /* mode == MODE_INTER4V */ |
| 369 |
|
int k, sumx = 0, sumy = 0; |
| 370 |
|
const VECTOR * const mvs = (quarterpel ? mb->qmvs : mb->mvs); |
| 371 |
|
|
| 372 |
|
for (k = 0; k < 4; k++) { |
| 373 |
|
dx = mvs[k].x; |
| 374 |
|
dy = mvs[k].y; |
| 375 |
|
sumx += quarterpel ? dx/2 : dx; |
| 376 |
|
sumy += quarterpel ? dy/2 : dy; |
| 377 |
|
|
| 378 |
|
if (reduced_resolution){ |
| 379 |
|
dx = RRV_MV_SCALEUP(dx); |
| 380 |
|
dy = RRV_MV_SCALEUP(dy); |
| 381 |
|
} |
| 382 |
|
|
| 383 |
|
compensate8x8_interpolate(&dct_codes[k * 64], cur->y, ref->y, refh->y, |
| 384 |
|
refv->y, refhv->y, tmp, 16 * i + 8*(k&1), 16 * j + 8*(k>>1), dx, |
| 385 |
|
dy, edged_width, quarterpel, reduced_resolution, rounding); |
| 386 |
|
} |
| 387 |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
| 388 |
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
| 389 |
|
} |
| 390 |
|
|
| 391 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) |
CompensateChroma(dx, dy, i, j, cur, ref, tmp, |
| 392 |
|
&dct_codes[4 * 64], edged_width / 2, rounding, reduced_resolution); |
| 393 |
|
} |
| 394 |
|
|
| 395 |
|
|
| 396 |
|
void |
| 397 |
|
MBMotionCompensationBVOP(MBParam * pParam, |
| 398 |
|
MACROBLOCK * const mb, |
| 399 |
|
const uint32_t i, |
| 400 |
|
const uint32_t j, |
| 401 |
|
IMAGE * const cur, |
| 402 |
|
const IMAGE * const f_ref, |
| 403 |
|
const IMAGE * const f_refh, |
| 404 |
|
const IMAGE * const f_refv, |
| 405 |
|
const IMAGE * const f_refhv, |
| 406 |
|
const IMAGE * const b_ref, |
| 407 |
|
const IMAGE * const b_refh, |
| 408 |
|
const IMAGE * const b_refv, |
| 409 |
|
const IMAGE * const b_refhv, |
| 410 |
|
int16_t * dct_codes) |
| 411 |
{ |
{ |
| 412 |
int32_t dx = mb->mvs[0].x; |
const uint32_t edged_width = pParam->edged_width; |
| 413 |
int32_t dy = mb->mvs[0].y; |
int32_t dx, dy, b_dx, b_dy, sumx, sumy, b_sumx, b_sumy; |
| 414 |
|
int k; |
| 415 |
compensate8x8_halfpel(dct_codes[0], cur->y, ref->y, refh->y, refv->y, refhv->y, |
const int quarterpel = pParam->vol_flags & XVID_VOL_QUARTERPEL; |
| 416 |
16*i, 16*j, dx, dy, edged_width); |
const uint8_t * ptr1, * ptr2; |
| 417 |
compensate8x8_halfpel(dct_codes[1], cur->y, ref->y, refh->y, refv->y, refhv->y, |
uint8_t * const tmp = f_refv->u; |
| 418 |
16*i + 8, 16*j, dx, dy, edged_width); |
const VECTOR * const fmvs = (quarterpel ? mb->qmvs : mb->mvs); |
| 419 |
compensate8x8_halfpel(dct_codes[2], cur->y, ref->y, refh->y, refv->y, refhv->y, |
const VECTOR * const bmvs = (quarterpel ? mb->b_qmvs : mb->b_mvs); |
| 420 |
16*i, 16*j + 8, dx, dy, edged_width); |
|
| 421 |
compensate8x8_halfpel(dct_codes[3], cur->y, ref->y, refh->y, refv->y, refhv->y, |
switch (mb->mode) { |
| 422 |
16*i + 8, 16*j + 8, dx, dy, edged_width); |
case MODE_FORWARD: |
| 423 |
|
dx = fmvs->x; dy = fmvs->y; |
| 424 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
|
| 425 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
| 426 |
|
f_refv->y, f_refhv->y, tmp, 16 * i, 16 * j, dx, |
| 427 |
/* uv-image-based compensation |
dy, edged_width, quarterpel, 0, 0); |
| 428 |
compensate8x8_halfpel(dct_codes[4], cur->u, ref->u, refh->u, refv->u, refhv->u, |
|
| 429 |
8*i, 8*j, dx, dy, edged_width/2); |
if (quarterpel) { dx /= 2; dy /= 2; } |
| 430 |
compensate8x8_halfpel(dct_codes[5], cur->v, ref->v, refh->v, refv->v, refhv->v, |
|
| 431 |
8*i, 8*j, dx, dy, edged_width/2); */ |
CompensateChroma( (dx >> 1) + roundtab_79[dx & 0x3], |
| 432 |
|
(dy >> 1) + roundtab_79[dy & 0x3], |
| 433 |
/* uv-block-based compensation */ |
i, j, cur, f_ref, tmp, |
| 434 |
interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
&dct_codes[4 * 64], edged_width / 2, 0, 0); |
| 435 |
transfer_8to16sub(dct_codes[4], |
|
| 436 |
cur->u + 8*j*edged_width/2 + 8*i, |
return; |
| 437 |
refv->u + 8*j*edged_width/2 + 8*i, edged_width/2); |
|
| 438 |
|
case MODE_BACKWARD: |
| 439 |
interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
b_dx = bmvs->x; b_dy = bmvs->y; |
| 440 |
transfer_8to16sub(dct_codes[5], |
|
| 441 |
cur->v + 8*j*edged_width/2 + 8*i, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
| 442 |
refv->v + 8*j*edged_width/2 + 8*i, edged_width/2); |
b_refv->y, b_refhv->y, tmp, 16 * i, 16 * j, b_dx, |
| 443 |
|
b_dy, edged_width, quarterpel, 0, 0); |
| 444 |
} |
|
| 445 |
else // mode == MODE_INTER4V |
if (quarterpel) { b_dx /= 2; b_dy /= 2; } |
| 446 |
{ |
|
| 447 |
int32_t sum, dx, dy; |
CompensateChroma( (b_dx >> 1) + roundtab_79[b_dx & 0x3], |
| 448 |
|
(b_dy >> 1) + roundtab_79[b_dy & 0x3], |
| 449 |
compensate8x8_halfpel(dct_codes[0], cur->y, ref->y, refh->y, refv->y, refhv->y, |
i, j, cur, b_ref, tmp, |
| 450 |
16*i, 16*j, mb->mvs[0].x, mb->mvs[0].y, edged_width); |
&dct_codes[4 * 64], edged_width / 2, 0, 0); |
| 451 |
compensate8x8_halfpel(dct_codes[1], cur->y, ref->y, refh->y, refv->y, refhv->y, |
|
| 452 |
16*i + 8, 16*j, mb->mvs[1].x, mb->mvs[1].y, edged_width); |
return; |
| 453 |
compensate8x8_halfpel(dct_codes[2], cur->y, ref->y, refh->y, refv->y, refhv->y, |
|
| 454 |
16*i, 16*j + 8, mb->mvs[2].x, mb->mvs[2].y, edged_width); |
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
| 455 |
compensate8x8_halfpel(dct_codes[3], cur->y, ref->y, refh->y, refv->y, refhv->y, |
case MODE_DIRECT_NO4V: |
| 456 |
16*i + 8, 16*j + 8, mb->mvs[3].x, mb->mvs[3].y, edged_width); |
dx = fmvs->x; dy = fmvs->y; |
| 457 |
|
b_dx = bmvs->x; b_dy = bmvs->y; |
| 458 |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
|
| 459 |
dx = (sum ? SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
if (quarterpel) { |
| 460 |
|
|
| 461 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
if ((dx&3) | (dy&3)) { |
| 462 |
dy = (sum ? SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width, |
| 463 |
|
(uint8_t *) f_ref->y, tmp + 32, |
| 464 |
/* uv-image-based compensation |
tmp + 64, tmp + 96, 16*i, 16*j, dx, dy, edged_width, 0); |
| 465 |
compensate8x8_halfpel(dct_codes[4], cur->u, ref->u, refh->u, refv->u, refhv->u, |
ptr1 = tmp; |
| 466 |
8*i, 8*j, dx, dy, edged_width/2); |
} else ptr1 = f_ref->y + (16*j + dy/4)*edged_width + 16*i + dx/4; /* fullpixel position */ |
| 467 |
compensate8x8_halfpel(dct_codes[5], cur->v, ref->v, refh->v, refv->v, refhv->v, |
|
| 468 |
8*i, 8*j, dx, dy, edged_width/2); */ |
if ((b_dx&3) | (b_dy&3)) { |
| 469 |
|
interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width + 16, |
| 470 |
/* uv-block-based compensation */ |
(uint8_t *) b_ref->y, tmp + 32, |
| 471 |
interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
tmp + 64, tmp + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
| 472 |
transfer_8to16sub(dct_codes[4], |
ptr2 = tmp + 16; |
| 473 |
cur->u + 8*j*edged_width/2 + 8*i, |
} else ptr2 = b_ref->y + (16*j + b_dy/4)*edged_width + 16*i + b_dx/4; /* fullpixel position */ |
| 474 |
refv->u + 8*j*edged_width/2 + 8*i, edged_width/2); |
|
| 475 |
|
b_dx /= 2; |
| 476 |
interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding); |
b_dy /= 2; |
| 477 |
transfer_8to16sub(dct_codes[5], |
dx /= 2; |
| 478 |
cur->v + 8*j*edged_width/2 + 8*i, |
dy /= 2; |
| 479 |
refv->v + 8*j*edged_width/2 + 8*i, edged_width/2); |
|
| 480 |
|
} else { |
| 481 |
|
ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
| 482 |
|
i, j, 16, dx, dy, edged_width); |
| 483 |
|
|
| 484 |
|
ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
| 485 |
|
i, j, 16, b_dx, b_dy, edged_width); |
| 486 |
} |
} |
| 487 |
|
for (k = 0; k < 4; k++) |
| 488 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
| 489 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
| 490 |
|
ptr1 + (k&1)*8 + (k>>1)*8*edged_width, |
| 491 |
|
ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width); |
| 492 |
|
|
| 493 |
|
|
| 494 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
| 495 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
| 496 |
|
|
| 497 |
|
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
| 498 |
|
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
| 499 |
|
|
| 500 |
|
break; |
| 501 |
|
|
| 502 |
|
default: /* MODE_DIRECT (or MODE_DIRECT_NONE_MV in case of bframes decoding) */ |
| 503 |
|
sumx = sumy = b_sumx = b_sumy = 0; |
| 504 |
|
|
| 505 |
|
for (k = 0; k < 4; k++) { |
| 506 |
|
|
| 507 |
|
dx = fmvs[k].x; dy = fmvs[k].y; |
| 508 |
|
b_dx = bmvs[k].x; b_dy = bmvs[k].y; |
| 509 |
|
|
| 510 |
|
if (quarterpel) { |
| 511 |
|
sumx += dx/2; sumy += dy/2; |
| 512 |
|
b_sumx += b_dx/2; b_sumy += b_dy/2; |
| 513 |
|
|
| 514 |
|
if ((dx&3) | (dy&3)) { |
| 515 |
|
interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width, |
| 516 |
|
(uint8_t *) f_ref->y, |
| 517 |
|
tmp + 32, tmp + 64, tmp + 96, |
| 518 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
| 519 |
|
ptr1 = tmp; |
| 520 |
|
} else ptr1 = f_ref->y + (16*j + (k>>1)*8 + dy/4)*edged_width + 16*i + (k&1)*8 + dx/4; |
| 521 |
|
|
| 522 |
|
if ((b_dx&3) | (b_dy&3)) { |
| 523 |
|
interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width + 16, |
| 524 |
|
(uint8_t *) b_ref->y, |
| 525 |
|
tmp + 16, tmp + 32, tmp + 48, |
| 526 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
| 527 |
|
ptr2 = tmp + 16; |
| 528 |
|
} else ptr2 = b_ref->y + (16*j + (k>>1)*8 + b_dy/4)*edged_width + 16*i + (k&1)*8 + b_dx/4; |
| 529 |
|
} else { |
| 530 |
|
sumx += dx; sumy += dy; |
| 531 |
|
b_sumx += b_dx; b_sumy += b_dy; |
| 532 |
|
|
| 533 |
|
ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
| 534 |
|
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, edged_width); |
| 535 |
|
ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
| 536 |
|
2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy, edged_width); |
| 537 |
|
} |
| 538 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
| 539 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
| 540 |
|
ptr1, ptr2, edged_width); |
| 541 |
|
|
| 542 |
|
} |
| 543 |
|
|
| 544 |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
| 545 |
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
| 546 |
|
b_dx = (b_sumx >> 3) + roundtab_76[b_sumx & 0xf]; |
| 547 |
|
b_dy = (b_sumy >> 3) + roundtab_76[b_sumy & 0xf]; |
| 548 |
|
|
| 549 |
|
break; |
| 550 |
|
} |
| 551 |
|
|
| 552 |
|
/* v block-based chroma interpolation for direct and interpolate modes */ |
| 553 |
|
transfer_8to16sub2(&dct_codes[4 * 64], |
| 554 |
|
cur->u + (j * 8) * edged_width / 2 + (i * 8), |
| 555 |
|
interpolate8x8_switch2(tmp, b_ref->u, 8 * i, 8 * j, |
| 556 |
|
b_dx, b_dy, edged_width / 2, 0), |
| 557 |
|
interpolate8x8_switch2(tmp + 8, f_ref->u, 8 * i, 8 * j, |
| 558 |
|
dx, dy, edged_width / 2, 0), |
| 559 |
|
edged_width / 2); |
| 560 |
|
|
| 561 |
|
transfer_8to16sub2(&dct_codes[5 * 64], |
| 562 |
|
cur->v + (j * 8) * edged_width / 2 + (i * 8), |
| 563 |
|
interpolate8x8_switch2(tmp, b_ref->v, 8 * i, 8 * j, |
| 564 |
|
b_dx, b_dy, edged_width / 2, 0), |
| 565 |
|
interpolate8x8_switch2(tmp + 8, f_ref->v, 8 * i, 8 * j, |
| 566 |
|
dx, dy, edged_width / 2, 0), |
| 567 |
|
edged_width / 2); |
| 568 |
} |
} |
Parent Directory
| 
Revision Log
| 
Patch