Parent Directory
| 
Revision Log
Revision 1273 - (view) (download)
| 1 : | edgomez | 1142 | /***************************************************************************** |
| 2 : | * | ||
| 3 : | * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | * - Motion Estimation for B-VOPs - | ||
| 5 : | * | ||
| 6 : | * Copyright(C) 2002 Christoph Lampert <gruel@web.de> | ||
| 7 : | * 2002 Michael Militzer <michael@xvid.org> | ||
| 8 : | * 2002-2003 Radoslaw Czyz <xvid@syskin.cjb.net> | ||
| 9 : | * | ||
| 10 : | * This program is free software ; you can redistribute it and/or modify | ||
| 11 : | * it under the terms of the GNU General Public License as published by | ||
| 12 : | * the Free Software Foundation ; either version 2 of the License, or | ||
| 13 : | * (at your option) any later version. | ||
| 14 : | * | ||
| 15 : | * This program is distributed in the hope that it will be useful, | ||
| 16 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 17 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 18 : | * GNU General Public License for more details. | ||
| 19 : | * | ||
| 20 : | * You should have received a copy of the GNU General Public License | ||
| 21 : | * along with this program ; if not, write to the Free Software | ||
| 22 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 23 : | * | ||
| 24 : | Isibaar | 1273 | * $Id: estimation_bvop.c,v 1.1.2.8 2003-12-18 02:02:08 Isibaar Exp $ |
| 25 : | edgomez | 1142 | * |
| 26 : | ****************************************************************************/ | ||
| 27 : | |||
| 28 : | |||
| 29 : | #include <assert.h> | ||
| 30 : | #include <stdio.h> | ||
| 31 : | #include <stdlib.h> | ||
| 32 : | #include <string.h> /* memcpy */ | ||
| 33 : | |||
| 34 : | #include "../encoder.h" | ||
| 35 : | #include "../global.h" | ||
| 36 : | #include "../image/interpolate8x8.h" | ||
| 37 : | #include "estimation.h" | ||
| 38 : | #include "motion.h" | ||
| 39 : | #include "sad.h" | ||
| 40 : | #include "motion_inlines.h" | ||
| 41 : | |||
| 42 : | |||
| 43 : | static int32_t | ||
| 44 : | ChromaSAD2(const int fx, const int fy, const int bx, const int by, | ||
| 45 : | syskin | 1215 | SearchData * const data) |
| 46 : | edgomez | 1142 | { |
| 47 : | int sad; | ||
| 48 : | const uint32_t stride = data->iEdgedWidth/2; | ||
| 49 : | uint8_t *f_refu, *f_refv, *b_refu, *b_refv; | ||
| 50 : | |||
| 51 : | const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = { | ||
| 52 : | NULL, | ||
| 53 : | interpolate8x8_halfpel_v, | ||
| 54 : | interpolate8x8_halfpel_h, | ||
| 55 : | interpolate8x8_halfpel_hv | ||
| 56 : | }; | ||
| 57 : | |||
| 58 : | int offset = (fx>>1) + (fy>>1)*stride; | ||
| 59 : | int filter = ((fx & 1) << 1) | (fy & 1); | ||
| 60 : | |||
| 61 : | if (filter != 0) { | ||
| 62 : | f_refu = data->RefQ; | ||
| 63 : | f_refv = data->RefQ + 8; | ||
| 64 : | interpolate8x8_halfpel[filter](f_refu, data->RefP[4] + offset, stride, data->rounding); | ||
| 65 : | interpolate8x8_halfpel[filter](f_refv, data->RefP[5] + offset, stride, data->rounding); | ||
| 66 : | } else { | ||
| 67 : | f_refu = (uint8_t*)data->RefP[4] + offset; | ||
| 68 : | f_refv = (uint8_t*)data->RefP[5] + offset; | ||
| 69 : | } | ||
| 70 : | |||
| 71 : | offset = (bx>>1) + (by>>1)*stride; | ||
| 72 : | filter = ((bx & 1) << 1) | (by & 1); | ||
| 73 : | |||
| 74 : | if (filter != 0) { | ||
| 75 : | b_refu = data->RefQ + 16; | ||
| 76 : | b_refv = data->RefQ + 24; | ||
| 77 : | interpolate8x8_halfpel[filter](b_refu, data->b_RefP[4] + offset, stride, data->rounding); | ||
| 78 : | interpolate8x8_halfpel[filter](b_refv, data->b_RefP[5] + offset, stride, data->rounding); | ||
| 79 : | } else { | ||
| 80 : | b_refu = (uint8_t*)data->b_RefP[4] + offset; | ||
| 81 : | b_refv = (uint8_t*)data->b_RefP[5] + offset; | ||
| 82 : | } | ||
| 83 : | |||
| 84 : | sad = sad8bi(data->CurU, b_refu, f_refu, stride); | ||
| 85 : | sad += sad8bi(data->CurV, b_refv, f_refv, stride); | ||
| 86 : | |||
| 87 : | return sad; | ||
| 88 : | } | ||
| 89 : | |||
| 90 : | static void | ||
| 91 : | syskin | 1215 | CheckCandidateInt(const int x, const int y, SearchData * const data, const unsigned int Direction) |
| 92 : | edgomez | 1142 | { |
| 93 : | syskin | 1215 | int32_t sad, xf, yf, xb, yb, xcf, ycf, xcb, ycb; |
| 94 : | edgomez | 1142 | uint32_t t; |
| 95 : | syskin | 1215 | |
| 96 : | edgomez | 1142 | const uint8_t *ReferenceF, *ReferenceB; |
| 97 : | VECTOR *current; | ||
| 98 : | |||
| 99 : | syskin | 1215 | if ((x > data->max_dx) || (x < data->min_dx) || |
| 100 : | (y > data->max_dy) || (y < data->min_dy)) | ||
| 101 : | edgomez | 1142 | return; |
| 102 : | |||
| 103 : | syskin | 1215 | if (Direction == 1) { /* x and y mean forward vector */ |
| 104 : | VECTOR backward = data->qpel_precision ? data->currentQMV[1] : data->currentMV[1]; | ||
| 105 : | xb = backward.x; | ||
| 106 : | yb = backward.y; | ||
| 107 : | xf = x; yf = y; | ||
| 108 : | } else { /* x and y mean backward vector */ | ||
| 109 : | VECTOR forward = data->qpel_precision ? data->currentQMV[0] : data->currentMV[0]; | ||
| 110 : | xf = forward.x; | ||
| 111 : | yf = forward.y; | ||
| 112 : | xb = x; yb = y; | ||
| 113 : | } | ||
| 114 : | |||
| 115 : | edgomez | 1142 | if (!data->qpel_precision) { |
| 116 : | ReferenceF = GetReference(xf, yf, data); | ||
| 117 : | ReferenceB = GetReferenceB(xb, yb, 1, data); | ||
| 118 : | syskin | 1215 | current = data->currentMV + Direction - 1; |
| 119 : | edgomez | 1142 | xcf = xf; ycf = yf; |
| 120 : | xcb = xb; ycb = yb; | ||
| 121 : | } else { | ||
| 122 : | ReferenceF = xvid_me_interpolate16x16qpel(xf, yf, 0, data); | ||
| 123 : | syskin | 1215 | current = data->currentQMV + Direction - 1; |
| 124 : | edgomez | 1142 | ReferenceB = xvid_me_interpolate16x16qpel(xb, yb, 1, data); |
| 125 : | xcf = xf/2; ycf = yf/2; | ||
| 126 : | xcb = xb/2; ycb = yb/2; | ||
| 127 : | } | ||
| 128 : | |||
| 129 : | t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0) | ||
| 130 : | + d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0); | ||
| 131 : | |||
| 132 : | sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); | ||
| 133 : | sad += (data->lambda16 * t * sad)>>10; | ||
| 134 : | |||
| 135 : | if (data->chroma && sad < *data->iMinSAD) | ||
| 136 : | sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3], | ||
| 137 : | (ycf >> 1) + roundtab_79[ycf & 0x3], | ||
| 138 : | (xcb >> 1) + roundtab_79[xcb & 0x3], | ||
| 139 : | (ycb >> 1) + roundtab_79[ycb & 0x3], data); | ||
| 140 : | |||
| 141 : | if (sad < *(data->iMinSAD)) { | ||
| 142 : | syskin | 1215 | *data->iMinSAD = sad; |
| 143 : | current->x = x; current->y = y; | ||
| 144 : | data->dir = Direction; | ||
| 145 : | edgomez | 1142 | } |
| 146 : | } | ||
| 147 : | |||
| 148 : | static void | ||
| 149 : | syskin | 1215 | CheckCandidateDirect(const int x, const int y, SearchData * const data, const unsigned int Direction) |
| 150 : | edgomez | 1142 | { |
| 151 : | int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; | ||
| 152 : | uint32_t k; | ||
| 153 : | const uint8_t *ReferenceF; | ||
| 154 : | const uint8_t *ReferenceB; | ||
| 155 : | VECTOR mvs, b_mvs; | ||
| 156 : | |||
| 157 : | if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; | ||
| 158 : | |||
| 159 : | for (k = 0; k < 4; k++) { | ||
| 160 : | mvs.x = data->directmvF[k].x + x; | ||
| 161 : | b_mvs.x = ((x == 0) ? | ||
| 162 : | data->directmvB[k].x | ||
| 163 : | : mvs.x - data->referencemv[k].x); | ||
| 164 : | |||
| 165 : | mvs.y = data->directmvF[k].y + y; | ||
| 166 : | b_mvs.y = ((y == 0) ? | ||
| 167 : | data->directmvB[k].y | ||
| 168 : | : mvs.y - data->referencemv[k].y); | ||
| 169 : | |||
| 170 : | if ((mvs.x > data->max_dx) || (mvs.x < data->min_dx) || | ||
| 171 : | (mvs.y > data->max_dy) || (mvs.y < data->min_dy) || | ||
| 172 : | (b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) || | ||
| 173 : | (b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) ) | ||
| 174 : | return; | ||
| 175 : | |||
| 176 : | if (data->qpel) { | ||
| 177 : | xcf += mvs.x/2; ycf += mvs.y/2; | ||
| 178 : | xcb += b_mvs.x/2; ycb += b_mvs.y/2; | ||
| 179 : | } else { | ||
| 180 : | xcf += mvs.x; ycf += mvs.y; | ||
| 181 : | xcb += b_mvs.x; ycb += b_mvs.y; | ||
| 182 : | mvs.x *= 2; mvs.y *= 2; /* we move to qpel precision anyway */ | ||
| 183 : | b_mvs.x *= 2; b_mvs.y *= 2; | ||
| 184 : | } | ||
| 185 : | |||
| 186 : | ReferenceF = xvid_me_interpolate8x8qpel(mvs.x, mvs.y, k, 0, data); | ||
| 187 : | ReferenceB = xvid_me_interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data); | ||
| 188 : | |||
| 189 : | sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), | ||
| 190 : | ReferenceF, ReferenceB, data->iEdgedWidth); | ||
| 191 : | if (sad > *(data->iMinSAD)) return; | ||
| 192 : | } | ||
| 193 : | |||
| 194 : | sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; | ||
| 195 : | |||
| 196 : | if (data->chroma && sad < *data->iMinSAD) | ||
| 197 : | sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], | ||
| 198 : | (ycf >> 3) + roundtab_76[ycf & 0xf], | ||
| 199 : | (xcb >> 3) + roundtab_76[xcb & 0xf], | ||
| 200 : | (ycb >> 3) + roundtab_76[ycb & 0xf], data); | ||
| 201 : | |||
| 202 : | if (sad < *(data->iMinSAD)) { | ||
| 203 : | syskin | 1215 | data->iMinSAD[0] = sad; |
| 204 : | edgomez | 1142 | data->currentMV->x = x; data->currentMV->y = y; |
| 205 : | syskin | 1215 | data->dir = Direction; |
| 206 : | edgomez | 1142 | } |
| 207 : | } | ||
| 208 : | |||
| 209 : | static void | ||
| 210 : | syskin | 1215 | CheckCandidateDirectno4v(const int x, const int y, SearchData * const data, const unsigned int Direction) |
| 211 : | edgomez | 1142 | { |
| 212 : | int32_t sad, xcf, ycf, xcb, ycb; | ||
| 213 : | const uint8_t *ReferenceF; | ||
| 214 : | const uint8_t *ReferenceB; | ||
| 215 : | VECTOR mvs, b_mvs; | ||
| 216 : | |||
| 217 : | if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; | ||
| 218 : | |||
| 219 : | mvs.x = data->directmvF[0].x + x; | ||
| 220 : | b_mvs.x = ((x == 0) ? | ||
| 221 : | data->directmvB[0].x | ||
| 222 : | : mvs.x - data->referencemv[0].x); | ||
| 223 : | |||
| 224 : | mvs.y = data->directmvF[0].y + y; | ||
| 225 : | b_mvs.y = ((y == 0) ? | ||
| 226 : | data->directmvB[0].y | ||
| 227 : | : mvs.y - data->referencemv[0].y); | ||
| 228 : | |||
| 229 : | if ( (mvs.x > data->max_dx) || (mvs.x < data->min_dx) | ||
| 230 : | || (mvs.y > data->max_dy) || (mvs.y < data->min_dy) | ||
| 231 : | || (b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) | ||
| 232 : | || (b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) ) return; | ||
| 233 : | |||
| 234 : | if (data->qpel) { | ||
| 235 : | xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2); | ||
| 236 : | xcb = 4*(b_mvs.x/2); ycb = 4*(b_mvs.y/2); | ||
| 237 : | ReferenceF = xvid_me_interpolate16x16qpel(mvs.x, mvs.y, 0, data); | ||
| 238 : | ReferenceB = xvid_me_interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); | ||
| 239 : | } else { | ||
| 240 : | xcf = 4*mvs.x; ycf = 4*mvs.y; | ||
| 241 : | xcb = 4*b_mvs.x; ycb = 4*b_mvs.y; | ||
| 242 : | ReferenceF = GetReference(mvs.x, mvs.y, data); | ||
| 243 : | ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data); | ||
| 244 : | } | ||
| 245 : | |||
| 246 : | sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); | ||
| 247 : | sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; | ||
| 248 : | |||
| 249 : | if (data->chroma && sad < *data->iMinSAD) | ||
| 250 : | sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], | ||
| 251 : | (ycf >> 3) + roundtab_76[ycf & 0xf], | ||
| 252 : | (xcb >> 3) + roundtab_76[xcb & 0xf], | ||
| 253 : | (ycb >> 3) + roundtab_76[ycb & 0xf], data); | ||
| 254 : | |||
| 255 : | if (sad < *(data->iMinSAD)) { | ||
| 256 : | *(data->iMinSAD) = sad; | ||
| 257 : | data->currentMV->x = x; data->currentMV->y = y; | ||
| 258 : | syskin | 1215 | data->dir = Direction; |
| 259 : | edgomez | 1142 | } |
| 260 : | } | ||
| 261 : | |||
| 262 : | void | ||
| 263 : | syskin | 1215 | CheckCandidate16no4v(const int x, const int y, SearchData * const data, const unsigned int Direction) |
| 264 : | edgomez | 1142 | { |
| 265 : | int32_t sad, xc, yc; | ||
| 266 : | const uint8_t * Reference; | ||
| 267 : | uint32_t t; | ||
| 268 : | VECTOR * current; | ||
| 269 : | |||
| 270 : | if ( (x > data->max_dx) || ( x < data->min_dx) | ||
| 271 : | || (y > data->max_dy) || (y < data->min_dy) ) return; | ||
| 272 : | |||
| 273 : | if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; /* non-zero even value */ | ||
| 274 : | |||
| 275 : | if (data->qpel_precision) { /* x and y are in 1/4 precision */ | ||
| 276 : | Reference = xvid_me_interpolate16x16qpel(x, y, 0, data); | ||
| 277 : | current = data->currentQMV; | ||
| 278 : | xc = x/2; yc = y/2; | ||
| 279 : | } else { | ||
| 280 : | Reference = GetReference(x, y, data); | ||
| 281 : | current = data->currentMV; | ||
| 282 : | xc = x; yc = y; | ||
| 283 : | } | ||
| 284 : | t = d_mv_bits(x, y, data->predMV, data->iFcode, | ||
| 285 : | data->qpel^data->qpel_precision, data->rrv); | ||
| 286 : | |||
| 287 : | sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); | ||
| 288 : | sad += (data->lambda16 * t * sad)>>10; | ||
| 289 : | |||
| 290 : | if (data->chroma && sad < *data->iMinSAD) | ||
| 291 : | sad += xvid_me_ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], | ||
| 292 : | (yc >> 1) + roundtab_79[yc & 0x3], data); | ||
| 293 : | |||
| 294 : | if (sad < *(data->iMinSAD)) { | ||
| 295 : | *(data->iMinSAD) = sad; | ||
| 296 : | current->x = x; current->y = y; | ||
| 297 : | syskin | 1215 | data->dir = Direction; |
| 298 : | edgomez | 1142 | } |
| 299 : | } | ||
| 300 : | |||
| 301 : | Isibaar | 1273 | void |
| 302 : | CheckCandidate16no4v_qpel(const int x, const int y, SearchData * const data, const unsigned int Direction) | ||
| 303 : | { | ||
| 304 : | int32_t sad, xc, yc; | ||
| 305 : | const uint8_t * Reference; | ||
| 306 : | uint32_t t; | ||
| 307 : | |||
| 308 : | if ( (x > data->max_dx) || ( x < data->min_dx) | ||
| 309 : | || (y > data->max_dy) || (y < data->min_dy) ) return; | ||
| 310 : | |||
| 311 : | if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; /* non-zero even value */ | ||
| 312 : | |||
| 313 : | Reference = xvid_me_interpolate16x16qpel(x, y, 0, data); | ||
| 314 : | |||
| 315 : | xc = x/2; yc = y/2; | ||
| 316 : | t = d_mv_bits(x, y, data->predMV, data->iFcode, | ||
| 317 : | data->qpel^data->qpel_precision, data->rrv); | ||
| 318 : | |||
| 319 : | sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); | ||
| 320 : | sad += (data->lambda16 * t * sad)>>10; | ||
| 321 : | |||
| 322 : | if (data->chroma && sad < *data->iMinSAD) | ||
| 323 : | sad += xvid_me_ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], | ||
| 324 : | (yc >> 1) + roundtab_79[yc & 0x3], data); | ||
| 325 : | |||
| 326 : | if (sad < *(data->iMinSAD)) { | ||
| 327 : | data->iMinSAD2 = *(data->iMinSAD); | ||
| 328 : | data->currentQMV2.x = data->currentQMV->x; | ||
| 329 : | data->currentQMV2.y = data->currentQMV->y; | ||
| 330 : | |||
| 331 : | data->iMinSAD[0] = sad; | ||
| 332 : | data->currentQMV[0].x = x; data->currentQMV[0].y = y; | ||
| 333 : | } else if (sad < data->iMinSAD2) { | ||
| 334 : | data->iMinSAD2 = sad; | ||
| 335 : | data->currentQMV2.x = x; data->currentQMV2.y = y; | ||
| 336 : | } | ||
| 337 : | } | ||
| 338 : | |||
| 339 : | edgomez | 1142 | static __inline VECTOR |
| 340 : | ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode) | ||
| 341 : | { | ||
| 342 : | /* the stupidiest function ever */ | ||
| 343 : | syskin | 1215 | return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]); |
| 344 : | edgomez | 1142 | } |
| 345 : | |||
| 346 : | static void __inline | ||
| 347 : | PreparePredictionsBF(VECTOR * const pmv, const int x, const int y, | ||
| 348 : | const uint32_t iWcount, | ||
| 349 : | const MACROBLOCK * const pMB, | ||
| 350 : | const uint32_t mode_curr) | ||
| 351 : | { | ||
| 352 : | |||
| 353 : | /* [0] is prediction */ | ||
| 354 : | pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y); | ||
| 355 : | |||
| 356 : | pmv[1].x = pmv[1].y = 0; /* [1] is zero */ | ||
| 357 : | |||
| 358 : | pmv[2] = ChoosePred(pMB, mode_curr); | ||
| 359 : | pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); | ||
| 360 : | |||
| 361 : | if ((y != 0)&&(x != (int)(iWcount+1))) { /* [3] top-right neighbour */ | ||
| 362 : | pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr); | ||
| 363 : | pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); | ||
| 364 : | } else pmv[3].x = pmv[3].y = 0; | ||
| 365 : | |||
| 366 : | if (y != 0) { | ||
| 367 : | pmv[4] = ChoosePred(pMB-iWcount, mode_curr); | ||
| 368 : | pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y); | ||
| 369 : | } else pmv[4].x = pmv[4].y = 0; | ||
| 370 : | |||
| 371 : | if (x != 0) { | ||
| 372 : | pmv[5] = ChoosePred(pMB-1, mode_curr); | ||
| 373 : | pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y); | ||
| 374 : | } else pmv[5].x = pmv[5].y = 0; | ||
| 375 : | |||
| 376 : | if (x != 0 && y != 0) { | ||
| 377 : | pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr); | ||
| 378 : | pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y); | ||
| 379 : | } else pmv[6].x = pmv[6].y = 0; | ||
| 380 : | } | ||
| 381 : | |||
| 382 : | |||
| 383 : | /* search backward or forward */ | ||
| 384 : | static void | ||
| 385 : | SearchBF( const IMAGE * const pRef, | ||
| 386 : | const uint8_t * const pRefH, | ||
| 387 : | const uint8_t * const pRefV, | ||
| 388 : | const uint8_t * const pRefHV, | ||
| 389 : | const int x, const int y, | ||
| 390 : | const uint32_t MotionFlags, | ||
| 391 : | const uint32_t iFcode, | ||
| 392 : | const MBParam * const pParam, | ||
| 393 : | MACROBLOCK * const pMB, | ||
| 394 : | const VECTOR * const predMV, | ||
| 395 : | int32_t * const best_sad, | ||
| 396 : | const int32_t mode_current, | ||
| 397 : | SearchData * const Data) | ||
| 398 : | { | ||
| 399 : | |||
| 400 : | int i; | ||
| 401 : | VECTOR pmv[7]; | ||
| 402 : | *Data->iMinSAD = MV_MAX_ERROR; | ||
| 403 : | Data->iFcode = iFcode; | ||
| 404 : | Data->qpel_precision = 0; | ||
| 405 : | syskin | 1237 | Data->chromaX = Data->chromaY = Data->chromaSAD = 256*4096; /* reset chroma-sad cache */ |
| 406 : | edgomez | 1142 | |
| 407 : | Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; | ||
| 408 : | Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; | ||
| 409 : | Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16; | ||
| 410 : | Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16; | ||
| 411 : | Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; | ||
| 412 : | Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; | ||
| 413 : | |||
| 414 : | Data->predMV = *predMV; | ||
| 415 : | |||
| 416 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, | ||
| 417 : | pParam->width, pParam->height, iFcode - Data->qpel, 1, 0); | ||
| 418 : | |||
| 419 : | pmv[0] = Data->predMV; | ||
| 420 : | if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } | ||
| 421 : | |||
| 422 : | PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); | ||
| 423 : | |||
| 424 : | Data->currentMV->x = Data->currentMV->y = 0; | ||
| 425 : | |||
| 426 : | /* main loop. checking all predictions */ | ||
| 427 : | for (i = 0; i < 7; i++) | ||
| 428 : | if (!vector_repeats(pmv, i) ) | ||
| 429 : | CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); | ||
| 430 : | |||
| 431 : | if (*Data->iMinSAD > 512) { | ||
| 432 : | syskin | 1215 | unsigned int mask = make_mask(pmv, 7, Data->dir); |
| 433 : | syskin | 1237 | |
| 434 : | MainSearchFunc *MainSearchPtr; | ||
| 435 : | if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; | ||
| 436 : | else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; | ||
| 437 : | else MainSearchPtr = xvid_me_DiamondSearch; | ||
| 438 : | |||
| 439 : | edgomez | 1142 | MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
| 440 : | } | ||
| 441 : | |||
| 442 : | xvid_me_SubpelRefine(Data, CheckCandidate16no4v); | ||
| 443 : | |||
| 444 : | if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { | ||
| 445 : | Data->currentQMV->x = 2*Data->currentMV->x; | ||
| 446 : | Data->currentQMV->y = 2*Data->currentMV->y; | ||
| 447 : | Data->qpel_precision = 1; | ||
| 448 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, | ||
| 449 : | pParam->width, pParam->height, iFcode, 2, 0); | ||
| 450 : | Isibaar | 1273 | |
| 451 : | if (MotionFlags & XVID_ME_QUARTERPELREFINE16) { | ||
| 452 : | if(MotionFlags & XVID_ME_FASTREFINE16) | ||
| 453 : | SubpelRefine_Fast(Data, CheckCandidate16no4v_qpel); | ||
| 454 : | else | ||
| 455 : | xvid_me_SubpelRefine(Data, CheckCandidate16no4v); | ||
| 456 : | } | ||
| 457 : | edgomez | 1142 | } |
| 458 : | |||
| 459 : | /* three bits are needed to code backward mode. four for forward */ | ||
| 460 : | |||
| 461 : | if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16; | ||
| 462 : | else *Data->iMinSAD += 3 * Data->lambda16; | ||
| 463 : | |||
| 464 : | if (*Data->iMinSAD < *best_sad) { | ||
| 465 : | *best_sad = *Data->iMinSAD; | ||
| 466 : | pMB->mode = mode_current; | ||
| 467 : | if (Data->qpel) { | ||
| 468 : | pMB->pmvs[0].x = Data->currentQMV->x - predMV->x; | ||
| 469 : | pMB->pmvs[0].y = Data->currentQMV->y - predMV->y; | ||
| 470 : | if (mode_current == MODE_FORWARD) | ||
| 471 : | pMB->qmvs[0] = *Data->currentQMV; | ||
| 472 : | else | ||
| 473 : | pMB->b_qmvs[0] = *Data->currentQMV; | ||
| 474 : | } else { | ||
| 475 : | pMB->pmvs[0].x = Data->currentMV->x - predMV->x; | ||
| 476 : | pMB->pmvs[0].y = Data->currentMV->y - predMV->y; | ||
| 477 : | } | ||
| 478 : | if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; | ||
| 479 : | else pMB->b_mvs[0] = *Data->currentMV; | ||
| 480 : | } | ||
| 481 : | |||
| 482 : | if (mode_current == MODE_FORWARD) *(Data->currentMV+2) = *Data->currentMV; | ||
| 483 : | else *(Data->currentMV+1) = *Data->currentMV; /* we store currmv for interpolate search */ | ||
| 484 : | } | ||
| 485 : | |||
| 486 : | static void | ||
| 487 : | SkipDecisionB(const IMAGE * const pCur, | ||
| 488 : | const IMAGE * const f_Ref, | ||
| 489 : | const IMAGE * const b_Ref, | ||
| 490 : | MACROBLOCK * const pMB, | ||
| 491 : | const uint32_t x, const uint32_t y, | ||
| 492 : | const SearchData * const Data) | ||
| 493 : | { | ||
| 494 : | int k; | ||
| 495 : | |||
| 496 : | syskin | 1237 | if (!Data->chroma) { |
| 497 : | int dx = 0, dy = 0, b_dx = 0, b_dy = 0; | ||
| 498 : | int32_t sum; | ||
| 499 : | const uint32_t stride = Data->iEdgedWidth/2; | ||
| 500 : | /* this is not full chroma compensation, only it's fullpel approximation. should work though */ | ||
| 501 : | edgomez | 1142 | |
| 502 : | syskin | 1237 | for (k = 0; k < 4; k++) { |
| 503 : | dy += Data->directmvF[k].y >> Data->qpel; | ||
| 504 : | dx += Data->directmvF[k].x >> Data->qpel; | ||
| 505 : | b_dy += Data->directmvB[k].y >> Data->qpel; | ||
| 506 : | b_dx += Data->directmvB[k].x >> Data->qpel; | ||
| 507 : | } | ||
| 508 : | edgomez | 1142 | |
| 509 : | syskin | 1237 | dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
| 510 : | dx = (dx >> 3) + roundtab_76[dx & 0xf]; | ||
| 511 : | b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf]; | ||
| 512 : | b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; | ||
| 513 : | edgomez | 1142 | |
| 514 : | syskin | 1237 | sum = sad8bi(pCur->u + 8 * x + 8 * y * stride, |
| 515 : | f_Ref->u + (y*8 + dy/2) * stride + x*8 + dx/2, | ||
| 516 : | b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, | ||
| 517 : | stride); | ||
| 518 : | edgomez | 1142 | |
| 519 : | syskin | 1237 | if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
| 520 : | edgomez | 1142 | |
| 521 : | syskin | 1237 | sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
| 522 : | f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, | ||
| 523 : | b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, | ||
| 524 : | stride); | ||
| 525 : | |||
| 526 : | if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ | ||
| 527 : | edgomez | 1142 | } |
| 528 : | syskin | 1237 | |
| 529 : | /* skip */ | ||
| 530 : | pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ | ||
| 531 : | for (k = 0; k < 4; k++) { | ||
| 532 : | pMB->qmvs[k] = pMB->mvs[k] = Data->directmvF[k]; | ||
| 533 : | pMB->b_qmvs[k] = pMB->b_mvs[k] = Data->directmvB[k]; | ||
| 534 : | } | ||
| 535 : | edgomez | 1142 | } |
| 536 : | |||
| 537 : | static uint32_t | ||
| 538 : | SearchDirect(const IMAGE * const f_Ref, | ||
| 539 : | const uint8_t * const f_RefH, | ||
| 540 : | const uint8_t * const f_RefV, | ||
| 541 : | const uint8_t * const f_RefHV, | ||
| 542 : | const IMAGE * const b_Ref, | ||
| 543 : | const uint8_t * const b_RefH, | ||
| 544 : | const uint8_t * const b_RefV, | ||
| 545 : | const uint8_t * const b_RefHV, | ||
| 546 : | const IMAGE * const pCur, | ||
| 547 : | const int x, const int y, | ||
| 548 : | const uint32_t MotionFlags, | ||
| 549 : | const int32_t TRB, const int32_t TRD, | ||
| 550 : | const MBParam * const pParam, | ||
| 551 : | MACROBLOCK * const pMB, | ||
| 552 : | const MACROBLOCK * const b_mb, | ||
| 553 : | int32_t * const best_sad, | ||
| 554 : | SearchData * const Data) | ||
| 555 : | |||
| 556 : | { | ||
| 557 : | int32_t skip_sad; | ||
| 558 : | int k = (x + Data->iEdgedWidth*y) * 16; | ||
| 559 : | MainSearchFunc *MainSearchPtr; | ||
| 560 : | CheckFunc * CheckCandidate; | ||
| 561 : | |||
| 562 : | *Data->iMinSAD = 256*4096; | ||
| 563 : | Data->RefP[0] = f_Ref->y + k; | ||
| 564 : | Data->RefP[2] = f_RefH + k; | ||
| 565 : | Data->RefP[1] = f_RefV + k; | ||
| 566 : | Data->RefP[3] = f_RefHV + k; | ||
| 567 : | Data->b_RefP[0] = b_Ref->y + k; | ||
| 568 : | Data->b_RefP[2] = b_RefH + k; | ||
| 569 : | Data->b_RefP[1] = b_RefV + k; | ||
| 570 : | Data->b_RefP[3] = b_RefHV + k; | ||
| 571 : | Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 572 : | Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 573 : | Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 574 : | Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 575 : | |||
| 576 : | k = Data->qpel ? 4 : 2; | ||
| 577 : | Data->max_dx = k * (pParam->width - x * 16); | ||
| 578 : | Data->max_dy = k * (pParam->height - y * 16); | ||
| 579 : | Data->min_dx = -k * (16 + x * 16); | ||
| 580 : | Data->min_dy = -k * (16 + y * 16); | ||
| 581 : | |||
| 582 : | Data->referencemv = Data->qpel ? b_mb->qmvs : b_mb->mvs; | ||
| 583 : | Data->qpel_precision = 0; | ||
| 584 : | |||
| 585 : | for (k = 0; k < 4; k++) { | ||
| 586 : | pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); | ||
| 587 : | pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD; | ||
| 588 : | pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD); | ||
| 589 : | pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD; | ||
| 590 : | |||
| 591 : | if ( (pMB->b_mvs[k].x > Data->max_dx) | (pMB->b_mvs[k].x < Data->min_dx) | ||
| 592 : | | (pMB->b_mvs[k].y > Data->max_dy) | (pMB->b_mvs[k].y < Data->min_dy) ) { | ||
| 593 : | |||
| 594 : | *best_sad = 256*4096; /* in that case, we won't use direct mode */ | ||
| 595 : | pMB->mode = MODE_DIRECT; /* just to make sure it doesn't say "MODE_DIRECT_NONE_MV" */ | ||
| 596 : | pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; | ||
| 597 : | return 256*4096; | ||
| 598 : | } | ||
| 599 : | if (b_mb->mode != MODE_INTER4V) { | ||
| 600 : | pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; | ||
| 601 : | pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0]; | ||
| 602 : | Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0]; | ||
| 603 : | Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0]; | ||
| 604 : | break; | ||
| 605 : | } | ||
| 606 : | } | ||
| 607 : | |||
| 608 : | CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; | ||
| 609 : | |||
| 610 : | CheckCandidate(0, 0, Data, 255); | ||
| 611 : | |||
| 612 : | /* initial (fast) skip decision */ | ||
| 613 : | edgomez | 1160 | if (*Data->iMinSAD < (int)Data->iQuant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
| 614 : | edgomez | 1142 | /* possible skip */ |
| 615 : | syskin | 1237 | SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
| 616 : | if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; /* skipped */ | ||
| 617 : | edgomez | 1142 | } |
| 618 : | |||
| 619 : | *Data->iMinSAD += Data->lambda16; | ||
| 620 : | skip_sad = *Data->iMinSAD; | ||
| 621 : | |||
| 622 : | Isibaar | 1273 | if (!(MotionFlags & XVID_ME_SKIP_DELTASEARCH)) { |
| 623 : | if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; | ||
| 624 : | else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; | ||
| 625 : | else MainSearchPtr = xvid_me_DiamondSearch; | ||
| 626 : | edgomez | 1142 | |
| 627 : | Isibaar | 1273 | MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
| 628 : | edgomez | 1142 | |
| 629 : | Isibaar | 1273 | xvid_me_SubpelRefine(Data, CheckCandidate); |
| 630 : | } | ||
| 631 : | edgomez | 1142 | |
| 632 : | *best_sad = *Data->iMinSAD; | ||
| 633 : | |||
| 634 : | if (Data->qpel || b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT; | ||
| 635 : | else pMB->mode = MODE_DIRECT_NO4V; /* for faster compensation */ | ||
| 636 : | |||
| 637 : | pMB->pmvs[3] = *Data->currentMV; | ||
| 638 : | |||
| 639 : | for (k = 0; k < 4; k++) { | ||
| 640 : | pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x; | ||
| 641 : | pMB->b_mvs[k].x = ( (Data->currentMV->x == 0) | ||
| 642 : | ? Data->directmvB[k].x | ||
| 643 : | :pMB->mvs[k].x - Data->referencemv[k].x); | ||
| 644 : | pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y); | ||
| 645 : | pMB->b_mvs[k].y = ((Data->currentMV->y == 0) | ||
| 646 : | ? Data->directmvB[k].y | ||
| 647 : | : pMB->mvs[k].y - Data->referencemv[k].y); | ||
| 648 : | if (Data->qpel) { | ||
| 649 : | pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2; | ||
| 650 : | pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2; | ||
| 651 : | pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2; | ||
| 652 : | pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2; | ||
| 653 : | } | ||
| 654 : | |||
| 655 : | if (b_mb->mode != MODE_INTER4V) { | ||
| 656 : | pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; | ||
| 657 : | pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0]; | ||
| 658 : | pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0]; | ||
| 659 : | pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0]; | ||
| 660 : | break; | ||
| 661 : | } | ||
| 662 : | } | ||
| 663 : | return skip_sad; | ||
| 664 : | } | ||
| 665 : | |||
| 666 : | syskin | 1215 | |
| 667 : | static void set_range(int * range, SearchData * Data) | ||
| 668 : | { | ||
| 669 : | Data->min_dx = range[0]; | ||
| 670 : | Data->max_dx = range[1]; | ||
| 671 : | Data->min_dy = range[2]; | ||
| 672 : | Data->max_dy = range[3]; | ||
| 673 : | } | ||
| 674 : | |||
| 675 : | edgomez | 1142 | static void |
| 676 : | syskin | 1215 | SubpelRefine_dir(SearchData * const data, CheckFunc * const CheckCandidate, const int dir) |
| 677 : | { | ||
| 678 : | /* Do a half-pel or q-pel refinement */ | ||
| 679 : | const VECTOR centerMV = data->qpel_precision ? | ||
| 680 : | data->currentQMV[dir-1] : data->currentMV[dir-1]; | ||
| 681 : | |||
| 682 : | CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, dir); | ||
| 683 : | CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, dir); | ||
| 684 : | CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, dir); | ||
| 685 : | CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, dir); | ||
| 686 : | CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, dir); | ||
| 687 : | CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, dir); | ||
| 688 : | CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, dir); | ||
| 689 : | CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, dir); | ||
| 690 : | } | ||
| 691 : | |||
| 692 : | static void | ||
| 693 : | edgomez | 1142 | SearchInterpolate(const IMAGE * const f_Ref, |
| 694 : | const uint8_t * const f_RefH, | ||
| 695 : | const uint8_t * const f_RefV, | ||
| 696 : | const uint8_t * const f_RefHV, | ||
| 697 : | const IMAGE * const b_Ref, | ||
| 698 : | const uint8_t * const b_RefH, | ||
| 699 : | const uint8_t * const b_RefV, | ||
| 700 : | const uint8_t * const b_RefHV, | ||
| 701 : | const int x, const int y, | ||
| 702 : | const uint32_t fcode, | ||
| 703 : | const uint32_t bcode, | ||
| 704 : | const uint32_t MotionFlags, | ||
| 705 : | const MBParam * const pParam, | ||
| 706 : | const VECTOR * const f_predMV, | ||
| 707 : | const VECTOR * const b_predMV, | ||
| 708 : | MACROBLOCK * const pMB, | ||
| 709 : | int32_t * const best_sad, | ||
| 710 : | syskin | 1215 | SearchData * const Data) |
| 711 : | edgomez | 1142 | |
| 712 : | { | ||
| 713 : | int i, j; | ||
| 714 : | syskin | 1215 | int b_range[4], f_range[4]; |
| 715 : | edgomez | 1142 | |
| 716 : | syskin | 1215 | Data->qpel_precision = 0; |
| 717 : | *Data->iMinSAD = 4096*256; | ||
| 718 : | Data->iFcode = fcode; Data->bFcode = bcode; | ||
| 719 : | edgomez | 1142 | |
| 720 : | syskin | 1215 | i = (x + y * Data->iEdgedWidth) * 16; |
| 721 : | edgomez | 1142 | |
| 722 : | syskin | 1215 | Data->RefP[0] = f_Ref->y + i; |
| 723 : | Data->RefP[2] = f_RefH + i; | ||
| 724 : | Data->RefP[1] = f_RefV + i; | ||
| 725 : | Data->RefP[3] = f_RefHV + i; | ||
| 726 : | Data->b_RefP[0] = b_Ref->y + i; | ||
| 727 : | Data->b_RefP[2] = b_RefH + i; | ||
| 728 : | Data->b_RefP[1] = b_RefV + i; | ||
| 729 : | Data->b_RefP[3] = b_RefHV + i; | ||
| 730 : | Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 731 : | Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 732 : | Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 733 : | Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; | ||
| 734 : | edgomez | 1142 | |
| 735 : | syskin | 1215 | Data->predMV = *f_predMV; |
| 736 : | Data->bpredMV = *b_predMV; | ||
| 737 : | edgomez | 1142 | |
| 738 : | syskin | 1215 | Data->currentMV[0] = Data->currentMV[2]; /* forward search left its vector here */ |
| 739 : | edgomez | 1142 | |
| 740 : | syskin | 1215 | get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, fcode - Data->qpel, 1, 0); |
| 741 : | get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, bcode - Data->qpel, 1, 0); | ||
| 742 : | edgomez | 1142 | |
| 743 : | syskin | 1215 | if (Data->currentMV[0].x > f_range[1]) Data->currentMV[0].x = f_range[1]; |
| 744 : | if (Data->currentMV[0].x < f_range[0]) Data->currentMV[0].x = f_range[0]; | ||
| 745 : | if (Data->currentMV[0].y > f_range[3]) Data->currentMV[0].y = f_range[3]; | ||
| 746 : | if (Data->currentMV[0].y < f_range[2]) Data->currentMV[0].y = f_range[2]; | ||
| 747 : | edgomez | 1142 | |
| 748 : | syskin | 1215 | if (Data->currentMV[1].x > b_range[1]) Data->currentMV[1].x = b_range[1]; |
| 749 : | if (Data->currentMV[1].x < b_range[0]) Data->currentMV[1].x = b_range[0]; | ||
| 750 : | if (Data->currentMV[1].y > b_range[3]) Data->currentMV[1].y = b_range[3]; | ||
| 751 : | if (Data->currentMV[1].y < b_range[2]) Data->currentMV[1].y = b_range[2]; | ||
| 752 : | edgomez | 1142 | |
| 753 : | syskin | 1215 | set_range(f_range, Data); |
| 754 : | |||
| 755 : | CheckCandidateInt(Data->currentMV[0].x, Data->currentMV[0].y, Data, 1); | ||
| 756 : | |||
| 757 : | edgomez | 1142 | /* diamond */ |
| 758 : | do { | ||
| 759 : | syskin | 1215 | Data->dir = 0; |
| 760 : | edgomez | 1142 | /* forward MV moves */ |
| 761 : | syskin | 1215 | i = Data->currentMV[0].x; j = Data->currentMV[0].y; |
| 762 : | edgomez | 1142 | |
| 763 : | syskin | 1215 | CheckCandidateInt(i + 1, j, Data, 1); |
| 764 : | CheckCandidateInt(i, j + 1, Data, 1); | ||
| 765 : | CheckCandidateInt(i - 1, j, Data, 1); | ||
| 766 : | CheckCandidateInt(i, j - 1, Data, 1); | ||
| 767 : | edgomez | 1142 | |
| 768 : | /* backward MV moves */ | ||
| 769 : | syskin | 1215 | set_range(b_range, Data); |
| 770 : | i = Data->currentMV[1].x; j = Data->currentMV[1].y; | ||
| 771 : | edgomez | 1142 | |
| 772 : | syskin | 1215 | CheckCandidateInt(i + 1, j, Data, 2); |
| 773 : | CheckCandidateInt(i, j + 1, Data, 2); | ||
| 774 : | CheckCandidateInt(i - 1, j, Data, 2); | ||
| 775 : | CheckCandidateInt(i, j - 1, Data, 2); | ||
| 776 : | edgomez | 1142 | |
| 777 : | syskin | 1215 | set_range(f_range, Data); |
| 778 : | |||
| 779 : | } while (Data->dir != 0); | ||
| 780 : | |||
| 781 : | edgomez | 1142 | /* qpel refinement */ |
| 782 : | syskin | 1215 | if (Data->qpel) { |
| 783 : | if (*Data->iMinSAD > *best_sad + 500) return; | ||
| 784 : | Data->qpel_precision = 1; | ||
| 785 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0); | ||
| 786 : | |||
| 787 : | Data->currentQMV[0].x = 2 * Data->currentMV[0].x; | ||
| 788 : | Data->currentQMV[0].y = 2 * Data->currentMV[0].y; | ||
| 789 : | Data->currentQMV[1].x = 2 * Data->currentMV[1].x; | ||
| 790 : | Data->currentQMV[1].y = 2 * Data->currentMV[1].y; | ||
| 791 : | SubpelRefine_dir(Data, CheckCandidateInt, 1); | ||
| 792 : | if (*Data->iMinSAD > *best_sad + 300) return; | ||
| 793 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); | ||
| 794 : | SubpelRefine_dir(Data, CheckCandidateInt, 2); | ||
| 795 : | edgomez | 1142 | } |
| 796 : | |||
| 797 : | syskin | 1237 | *Data->iMinSAD += 2 * Data->lambda16; /* two bits are needed to code interpolate mode. */ |
| 798 : | edgomez | 1142 | |
| 799 : | syskin | 1215 | if (*Data->iMinSAD < *best_sad) { |
| 800 : | *best_sad = *Data->iMinSAD; | ||
| 801 : | pMB->mvs[0] = Data->currentMV[0]; | ||
| 802 : | pMB->b_mvs[0] = Data->currentMV[1]; | ||
| 803 : | edgomez | 1142 | pMB->mode = MODE_INTERPOLATE; |
| 804 : | syskin | 1215 | if (Data->qpel) { |
| 805 : | pMB->qmvs[0] = Data->currentQMV[0]; | ||
| 806 : | pMB->b_qmvs[0] = Data->currentQMV[1]; | ||
| 807 : | edgomez | 1142 | pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x; |
| 808 : | pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y; | ||
| 809 : | pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x; | ||
| 810 : | pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y; | ||
| 811 : | } else { | ||
| 812 : | pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; | ||
| 813 : | pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y; | ||
| 814 : | pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x; | ||
| 815 : | pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y; | ||
| 816 : | } | ||
| 817 : | } | ||
| 818 : | } | ||
| 819 : | |||
| 820 : | void | ||
| 821 : | MotionEstimationBVOP(MBParam * const pParam, | ||
| 822 : | FRAMEINFO * const frame, | ||
| 823 : | const int32_t time_bp, | ||
| 824 : | const int32_t time_pp, | ||
| 825 : | /* forward (past) reference */ | ||
| 826 : | const MACROBLOCK * const f_mbs, | ||
| 827 : | const IMAGE * const f_ref, | ||
| 828 : | const IMAGE * const f_refH, | ||
| 829 : | const IMAGE * const f_refV, | ||
| 830 : | const IMAGE * const f_refHV, | ||
| 831 : | /* backward (future) reference */ | ||
| 832 : | const FRAMEINFO * const b_reference, | ||
| 833 : | const IMAGE * const b_ref, | ||
| 834 : | const IMAGE * const b_refH, | ||
| 835 : | const IMAGE * const b_refV, | ||
| 836 : | const IMAGE * const b_refHV) | ||
| 837 : | { | ||
| 838 : | uint32_t i, j; | ||
| 839 : | int32_t best_sad; | ||
| 840 : | uint32_t skip_sad; | ||
| 841 : | |||
| 842 : | const MACROBLOCK * const b_mbs = b_reference->mbs; | ||
| 843 : | |||
| 844 : | VECTOR f_predMV, b_predMV; | ||
| 845 : | |||
| 846 : | const int32_t TRB = time_pp - time_bp; | ||
| 847 : | const int32_t TRD = time_pp; | ||
| 848 : | |||
| 849 : | /* some pre-inintialized data for the rest of the search */ | ||
| 850 : | |||
| 851 : | SearchData Data; | ||
| 852 : | memset(&Data, 0, sizeof(SearchData)); | ||
| 853 : | syskin | 1215 | |
| 854 : | edgomez | 1142 | Data.iEdgedWidth = pParam->edged_width; |
| 855 : | Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; | ||
| 856 : | Data.rounding = 0; | ||
| 857 : | Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; | ||
| 858 : | Data.iQuant = frame->quant; | ||
| 859 : | |||
| 860 : | Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ | ||
| 861 : | |||
| 862 : | /* note: i==horizontal, j==vertical */ | ||
| 863 : | for (j = 0; j < pParam->mb_height; j++) { | ||
| 864 : | |||
| 865 : | f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */ | ||
| 866 : | |||
| 867 : | for (i = 0; i < pParam->mb_width; i++) { | ||
| 868 : | MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; | ||
| 869 : | const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; | ||
| 870 : | |||
| 871 : | /* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */ | ||
| 872 : | if (b_reference->coding_type != S_VOP) | ||
| 873 : | if (b_mb->mode == MODE_NOT_CODED) { | ||
| 874 : | pMB->mode = MODE_NOT_CODED; | ||
| 875 : | syskin | 1215 | pMB->mvs[0] = pMB->b_mvs[0] = zeroMV; |
| 876 : | edgomez | 1142 | continue; |
| 877 : | } | ||
| 878 : | syskin | 1237 | |
| 879 : | Data.lambda16 = xvid_me_lambda_vec16[b_mb->quant]; | ||
| 880 : | edgomez | 1142 | |
| 881 : | Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; | ||
| 882 : | Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; | ||
| 883 : | Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; | ||
| 884 : | |||
| 885 : | /* direct search comes first, because it (1) checks for SKIP-mode | ||
| 886 : | and (2) sets very good predictions for forward and backward search */ | ||
| 887 : | skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y, | ||
| 888 : | b_ref, b_refH->y, b_refV->y, b_refHV->y, | ||
| 889 : | &frame->image, | ||
| 890 : | i, j, | ||
| 891 : | frame->motion_flags, | ||
| 892 : | TRB, TRD, | ||
| 893 : | pParam, | ||
| 894 : | pMB, b_mb, | ||
| 895 : | &best_sad, | ||
| 896 : | &Data); | ||
| 897 : | |||
| 898 : | if (pMB->mode == MODE_DIRECT_NONE_MV) continue; | ||
| 899 : | |||
| 900 : | /* forward search */ | ||
| 901 : | SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, | ||
| 902 : | i, j, | ||
| 903 : | frame->motion_flags, | ||
| 904 : | frame->fcode, pParam, | ||
| 905 : | pMB, &f_predMV, &best_sad, | ||
| 906 : | MODE_FORWARD, &Data); | ||
| 907 : | |||
| 908 : | /* backward search */ | ||
| 909 : | SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y, | ||
| 910 : | i, j, | ||
| 911 : | frame->motion_flags, | ||
| 912 : | frame->bcode, pParam, | ||
| 913 : | pMB, &b_predMV, &best_sad, | ||
| 914 : | MODE_BACKWARD, &Data); | ||
| 915 : | |||
| 916 : | /* interpolate search comes last, because it uses data from forward and backward as prediction */ | ||
| 917 : | SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, | ||
| 918 : | b_ref, b_refH->y, b_refV->y, b_refHV->y, | ||
| 919 : | i, j, | ||
| 920 : | frame->fcode, frame->bcode, | ||
| 921 : | frame->motion_flags, | ||
| 922 : | pParam, | ||
| 923 : | &f_predMV, &b_predMV, | ||
| 924 : | pMB, &best_sad, | ||
| 925 : | &Data); | ||
| 926 : | |||
| 927 : | /* final skip decision */ | ||
| 928 : | syskin | 1237 | if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * (Data.chroma ? 3:2) ) |
| 929 : | && ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) | ||
| 930 : | |||
| 931 : | edgomez | 1142 | SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
| 932 : | |||
| 933 : | switch (pMB->mode) { | ||
| 934 : | case MODE_FORWARD: | ||
| 935 : | f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; | ||
| 936 : | break; | ||
| 937 : | case MODE_BACKWARD: | ||
| 938 : | b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; | ||
| 939 : | break; | ||
| 940 : | case MODE_INTERPOLATE: | ||
| 941 : | f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; | ||
| 942 : | b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; | ||
| 943 : | break; | ||
| 944 : | default: | ||
| 945 : | break; | ||
| 946 : | } | ||
| 947 : | } | ||
| 948 : | } | ||
| 949 : | } | ||
| 950 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |