--- trunk/xvidcore/src/motion/motion_est.c 2002/07/19 15:02:39 318 +++ trunk/xvidcore/src/motion/motion_est.c 2002/09/07 13:41:13 444 @@ -1,69 +1,36 @@ -/************************************************************************** +/***************************************************************************** * - * XVID MPEG-4 VIDEO CODEC - * motion estimation + * XVID MPEG-4 VIDEO CODEC + * - Motion Estimation module - * - * 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. + * Copyright(C) 2002 Christoph Lampert + * Copyright(C) 2002 Michael Militzer + * Copyright(C) 2002 Edouard Gomez * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. + * 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. * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. * - *************************************************************************/ - -/************************************************************************** - * - * Modifications: + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - * 01.05.2002 updated MotionEstimationBVOP - * 25.04.2002 partial prevMB conversion - * 22.04.2002 remove some compile warning by chenm001 - * 14.04.2002 added MotionEstimationBVOP() - * 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between - * EPZS and EPZS^2 - * 08.02.2002 split up PMVfast into three routines: PMVFast, PMVFast_MainLoop - * PMVFast_Refine to support multiple searches with different start points - * 07.01.2002 uv-block-based interpolation - * 06.01.2002 INTER/INTRA-decision is now done before any SEARCH8 (speedup) - * changed INTER_BIAS to 150 (as suggested by suxen_drol) - * removed halfpel refinement step in PMVfastSearch8 + quality=5 - * added new quality mode = 6 which performs halfpel refinement - * filesize difference between quality 5 and 6 is smaller than 1% - * (Isibaar) - * 31.12.2001 PMVfastSearch16 and PMVfastSearch8 (gruel) - * 30.12.2001 get_range/MotionSearchX simplified; blue/green bug fix - * 22.12.2001 commented best_point==99 check - * 19.12.2001 modified get_range (purple bug fix) - * 15.12.2001 moved pmv displacement from mbprediction - * 02.12.2001 motion estimation/compensation split (Isibaar) - * 16.11.2001 rewrote/tweaked search algorithms; pross@cs.rmit.edu.au - * 10.11.2001 support for sad16/sad8 functions - * 28.08.2001 reactivated MODE_INTER4V for EXT_MODE - * 24.08.2001 removed MODE_INTER4V_Q, disabled MODE_INTER4V for EXT_MODE - * 22.08.2001 added MODE_INTER4V_Q - * 20.08.2001 added pragma to get rid of internal compiler error with VC6 - * idea by Cyril. Thanks. - * - * Michael Militzer - * - **************************************************************************/ + *************************************************************************/ #include #include @@ -154,7 +121,7 @@ return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); } - + bool MotionEstimation(MBParam * const pParam, FRAMEINFO * const current, @@ -172,6 +139,7 @@ const IMAGE *const pRef = &reference->image; static const VECTOR zeroMV = { 0, 0 }; + VECTOR predMV; int32_t x, y; int32_t iIntra = 0; @@ -185,9 +153,15 @@ MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; + if (pMB->mode == MODE_NOT_CODED) + continue; + + predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); + pMB->sad16 = - SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, - y, current->motion_flags, current->quant, + SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, + x, y, predMV.x, predMV.y, predMV.x, predMV.y, + current->motion_flags, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, &pMB->pmvs[0]); @@ -219,43 +193,52 @@ pMB->dquant == NO_CHANGE)) { int32_t sad8 = IMV16X16 * current->quant; - if (sad8 < pMB->sad16) - + if (sad8 < pMB->sad16) { sad8 += pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, - pCurrent, 2 * x, 2 * y, pMB->mv16.x, - pMB->mv16.y, current->motion_flags, + pCurrent, 2 * x, 2 * y, + pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, + current->motion_flags, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); + } + if (sad8 < pMB->sad16) { - if (sad8 < pMB->sad16) + predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 1); sad8 += pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, - pCurrent, 2 * x + 1, 2 * y, pMB->mv16.x, - pMB->mv16.y, current->motion_flags, + pCurrent, 2 * x + 1, 2 * y, + pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, + current->motion_flags, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); - - if (sad8 < pMB->sad16) + } + if (sad8 < pMB->sad16) { + predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 2); sad8 += pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, - pCurrent, 2 * x, 2 * y + 1, pMB->mv16.x, - pMB->mv16.y, current->motion_flags, + pCurrent, 2 * x, 2 * y + 1, + pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, + current->motion_flags, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); - - if (sad8 < pMB->sad16) + } + if (sad8 < pMB->sad16) { + predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 3); sad8 += pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, 2 * x + 1, 2 * y + 1, - pMB->mv16.x, pMB->mv16.y, - current->motion_flags, current->quant, - current->fcode, pParam, pMBs, prevMBs, - &pMB->mvs[3], &pMB->pmvs[3]); - + pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, + current->motion_flags, + current->quant, current->fcode, pParam, + pMBs, prevMBs, + &pMB->mvs[3], + &pMB->pmvs[3]); + } + /* decide: MODE_INTER or MODE_INTER4V mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v */ @@ -282,19 +265,20 @@ return 0; } + #define CHECK_MV16_ZERO {\ if ( (0 <= max_dx) && (0 >= min_dx) \ && (0 <= max_dy) && (0 >= min_dy) ) \ { \ iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ - iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_16(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ } #define NOCHECK_MV16_CANDIDATE(X,Y) { \ iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ } @@ -304,7 +288,7 @@ && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ { \ iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ } @@ -314,7 +298,7 @@ && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ { \ iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ } @@ -324,7 +308,7 @@ && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ { \ iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ } @@ -332,7 +316,7 @@ #define CHECK_MV8_ZERO {\ iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ - iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_8(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ } @@ -340,7 +324,7 @@ #define NOCHECK_MV8_CANDIDATE(X,Y) \ { \ iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ - iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ } @@ -350,7 +334,7 @@ && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ { \ iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ - iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ } @@ -360,7 +344,7 @@ && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ { \ iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ - iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ } @@ -370,7 +354,7 @@ && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ { \ iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ - iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ + iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ } @@ -425,11 +409,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, - int32_t iMinSAD, - VECTOR * const currMV, - const VECTOR * const pmv, + const int start_x, + const int start_y, + int iMinSAD, + VECTOR * const currMV, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -443,11 +428,12 @@ /* Do a diamond search around given starting point, return SAD of best */ int32_t iDirection = 0; + int32_t iDirectionBackup; int32_t iSAD; VECTOR backupMV; - backupMV.x = startx; - backupMV.y = starty; + backupMV.x = start_x; + backupMV.y = start_y; /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ @@ -456,26 +442,28 @@ CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); - if (iDirection) + if (iDirection) { while (!iFound) { iFound = 1; backupMV = *currMV; + iDirectionBackup = iDirection; - if (iDirection != 2) + if (iDirectionBackup != 2) CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); - if (iDirection != 1) + if (iDirectionBackup != 1) CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); - if (iDirection != 4) + if (iDirectionBackup != 4) CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); - if (iDirection != 3) + if (iDirectionBackup != 3) CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); + } } else { - currMV->x = startx; - currMV->y = starty; + currMV->x = start_x; + currMV->y = start_y; } return iMinSAD; } @@ -488,11 +476,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, - int32_t iMinSAD, - VECTOR * const currMV, - const VECTOR * const pmv, + const int start_x, + const int start_y, + int iMinSAD, + VECTOR * const currMV, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -509,8 +498,8 @@ int32_t iSAD; VECTOR backupMV; - backupMV.x = startx; - backupMV.y = starty; + backupMV.x = start_x; + backupMV.y = start_y; /* It's one search with full square pattern, and new parts for all following diamonds */ @@ -535,7 +524,7 @@ backupMV.y + iDiamondSize, 8); - if (iDirection) + if (iDirection) { while (!iFound) { iFound = 1; backupMV = *currMV; @@ -544,62 +533,62 @@ case 1: CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); break; case 2: - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 3: - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 4: - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); break; case 5: - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); break; case 6: - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; @@ -607,51 +596,52 @@ case 7: CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 8: - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; default: - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); - CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); - CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); - CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, + CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; } + } } else { - currMV->x = startx; - currMV->y = starty; + currMV->x = start_x; + currMV->y = start_y; } return iMinSAD; } @@ -665,11 +655,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, - int32_t iMinSAD, - VECTOR * const currMV, - const VECTOR * const pmv, + const int start_x, + const int start_y, + int iMinSAD, + VECTOR * const currMV, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -684,8 +675,8 @@ int32_t dx, dy; VECTOR backupMV; - backupMV.x = startx; - backupMV.y = starty; + backupMV.x = start_x; + backupMV.y = start_y; for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) @@ -702,11 +693,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, - int32_t iMinSAD, - VECTOR * const currMV, - const VECTOR * const pmv, + int start_x, + int start_y, + int iMinSAD, + VECTOR * const currMV, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -723,26 +715,26 @@ /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ if (iDirection) { - CHECK_MV16_CANDIDATE(startx - iDiamondSize, starty); - CHECK_MV16_CANDIDATE(startx + iDiamondSize, starty); - CHECK_MV16_CANDIDATE(startx, starty - iDiamondSize); - CHECK_MV16_CANDIDATE(startx, starty + iDiamondSize); + CHECK_MV16_CANDIDATE(start_x - iDiamondSize, start_y); + CHECK_MV16_CANDIDATE(start_x + iDiamondSize, start_y); + CHECK_MV16_CANDIDATE(start_x, start_y - iDiamondSize); + CHECK_MV16_CANDIDATE(start_x, start_y + iDiamondSize); } else { int bDirection = 1 + 2 + 4 + 8; do { iDirection = 0; if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); if (bDirection & 2) - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); if (bDirection & 4) - CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); + CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); if (bDirection & 8) - CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); + CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); /* now we're doing diagonal checks near our candidate */ @@ -750,99 +742,100 @@ { bDirection = iDirection; iDirection = 0; - startx = currMV->x; - starty = currMV->y; + start_x = currMV->x; + start_y = currMV->y; if (bDirection & 3) //our candidate is left or right { - CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); - CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); + CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); + CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); } else // what remains here is up or down { - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); } if (iDirection) { bDirection += iDirection; - startx = currMV->x; - starty = currMV->y; + start_x = currMV->x; + start_y = currMV->y; } } else //about to quit, eh? not so fast.... { switch (bDirection) { case 2: - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; case 1: - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); + + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); break; case 2 + 4: - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; case 4: - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); break; case 8: - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); break; case 1 + 4: - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); break; case 2 + 8: - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; case 1 + 8: - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); break; default: //1+2+4+8 == we didn't find anything at all - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; } if (!iDirection) break; //ok, the end. really else { bDirection = iDirection; - startx = currMV->x; - starty = currMV->y; + start_x = currMV->x; + start_y = currMV->y; } } } @@ -851,6 +844,252 @@ return iMinSAD; } +#define CHECK_MV16_F_INTERPOL(X,Y) { \ + if ( ((X) <= f_max_dx) && ((X) >= f_min_dx) \ + && ((Y) <= f_max_dy) && ((Y) >= f_min_dy) ) \ + { \ + iSAD = sad16bi( cur, \ + get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ + get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, b_currMV->x, b_currMV->y, iEdgedWidth), \ + iEdgedWidth); \ + iSAD += calc_delta_16((X) - f_center_x, (Y) - f_center_y, (uint8_t)f_iFcode, iQuant);\ + iSAD += calc_delta_16(b_currMV->x - b_center_x, b_currMV->y - b_center_y, (uint8_t)b_iFcode, iQuant);\ + if (iSAD < iMinSAD) \ + { iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); } } \ +} + +#define CHECK_MV16_F_INTERPOL_FOUND(X,Y) { \ + if ( ((X) <= f_max_dx) && ((X) >= f_min_dx) \ + && ((Y) <= f_max_dy) && ((Y) >= f_min_dy) ) \ + { \ + iSAD = sad16bi( cur, \ + get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ + get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, b_currMV->x, b_currMV->y, iEdgedWidth), \ + iEdgedWidth); \ + iSAD += calc_delta_16((X) - f_center_x, (Y) - f_center_y, (uint8_t)f_iFcode, iQuant);\ + iSAD += calc_delta_16(b_currMV->x - b_center_x, b_currMV->y - b_center_y, (uint8_t)b_iFcode, iQuant);\ + if (iSAD < iMinSAD) \ + { iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); iFound=0;} } \ +} + +#define CHECK_MV16_B_INTERPOL(X,Y) { \ + if ( ((X) <= b_max_dx) && ((X) >= b_min_dx) \ + && ((Y) <= b_max_dy) && ((Y) >= b_min_dy) ) \ + { \ + iSAD = sad16bi( cur, \ + get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, f_currMV->x, f_currMV->y, iEdgedWidth), \ + get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ + iEdgedWidth); \ + iSAD += calc_delta_16(f_currMV->x - f_center_x, f_currMV->y - f_center_y, (uint8_t)f_iFcode, iQuant);\ + iSAD += calc_delta_16((X) - b_center_x, (Y) - b_center_y, (uint8_t)b_iFcode, iQuant);\ + if (iSAD < iMinSAD) \ + { iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); } } \ +} + +#define CHECK_MV16_B_INTERPOL_FOUND(X,Y) { \ + if ( ((X) <= b_max_dx) && ((X) >= b_min_dx) \ + && ((Y) <= b_max_dy) && ((Y) >= b_min_dy) ) \ + { \ + iSAD = sad16bi( cur, \ + get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, f_currMV->x, f_currMV->y, iEdgedWidth), \ + get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ + iEdgedWidth); \ + iSAD += calc_delta_16(f_currMV->x - f_center_x, f_currMV->y - f_center_y, (uint8_t)f_iFcode, iQuant);\ + iSAD += calc_delta_16((X) - b_center_x, (Y) - b_center_y, (uint8_t)b_iFcode, iQuant);\ + if (iSAD < iMinSAD) \ + { iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); iFound=0;} } \ +} + +int32_t +Diamond16_InterpolMainSearch( + const uint8_t * const f_pRef, + const uint8_t * const f_pRefH, + const uint8_t * const f_pRefV, + const uint8_t * const f_pRefHV, + + const uint8_t * const cur, + + const uint8_t * const b_pRef, + const uint8_t * const b_pRefH, + const uint8_t * const b_pRefV, + const uint8_t * const b_pRefHV, + + const int x, + const int y, + + const int f_start_x, + const int f_start_y, + const int b_start_x, + const int b_start_y, + + int iMinSAD, + VECTOR * const f_currMV, + VECTOR * const b_currMV, + + const int f_center_x, + const int f_center_y, + const int b_center_x, + const int b_center_y, + + const int32_t f_min_dx, + const int32_t f_max_dx, + const int32_t f_min_dy, + const int32_t f_max_dy, + + const int32_t b_min_dx, + const int32_t b_max_dx, + const int32_t b_min_dy, + const int32_t b_max_dy, + + const int32_t iEdgedWidth, + const int32_t iDiamondSize, + + const int32_t f_iFcode, + const int32_t b_iFcode, + + const int32_t iQuant, + int iFound) +{ +/* Do a diamond search around given starting point, return SAD of best */ + + int32_t iSAD; + + VECTOR f_backupMV; + VECTOR b_backupMV; + + f_currMV->x = f_start_x; + f_currMV->y = f_start_y; + b_currMV->x = b_start_x; + b_currMV->y = b_start_y; + + do + { + iFound = 1; + + f_backupMV = *f_currMV; + + CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x - iDiamondSize, f_backupMV.y); + CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x + iDiamondSize, f_backupMV.y); + CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x, f_backupMV.y - iDiamondSize); + CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x, f_backupMV.y + iDiamondSize); + + b_backupMV = *b_currMV; + + CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x - iDiamondSize, b_backupMV.y); + CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x + iDiamondSize, b_backupMV.y); + CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x, b_backupMV.y - iDiamondSize); + CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x, b_backupMV.y + iDiamondSize); + + } while (!iFound); + + return iMinSAD; +} + +/* Sorry, these MACROS really got too large... I'll turn them into function soon! */ + +#define CHECK_MV16_DIRECT_FOUND(X,Y) \ + if ( (X)>=(-32) && (X)<=(31) && ((Y)>=-32) && ((Y)<=31) ) \ + { int k;\ + VECTOR mvs,b_mvs; \ + iSAD = 0;\ + for (k = 0; k < 4; k++) { \ + mvs.x = (int32_t) ((TRB * directmv[k].x) / TRD + (X)); \ + b_mvs.x = (int32_t) (((X) == 0) \ + ? ((TRB - TRD) * directmv[k].x) / TRD \ + : mvs.x - directmv[k].x); \ + \ + mvs.y = (int32_t) ((TRB * directmv[k].y) / TRD + (Y)); \ + b_mvs.y = (int32_t) (((Y) == 0) \ + ? ((TRB - TRD) * directmv[k].y) / TRD \ + : mvs.y - directmv[k].y); \ + \ + if ( (mvs.x <= max_dx) && (mvs.x >= min_dx) \ + && (mvs.y <= max_dy) && (mvs.y >= min_dy) \ + && (b_mvs.x <= max_dx) && (b_mvs.x >= min_dx) \ + && (b_mvs.y <= max_dy) && (b_mvs.y >= min_dy) ) { \ + iSAD += sad8bi( cur + 8*(k&1) + 8*(k>>1)*iEdgedWidth, \ + get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, 2*x+(k&1), 2*y+(k>>1), 8, \ + mvs.x, mvs.y, iEdgedWidth), \ + get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, 2*x+(k&1), 2*y+(k>>1), 8, \ + b_mvs.x, b_mvs.y, iEdgedWidth), \ + iEdgedWidth); \ + } \ + else \ + iSAD = 65535; \ + } \ + iSAD += calc_delta_16((X),(Y), 1, iQuant);\ + if (iSAD < iMinSAD) \ + { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iFound=0; } \ +} + + + +int32_t +Diamond16_DirectMainSearch( + const uint8_t * const f_pRef, + const uint8_t * const f_pRefH, + const uint8_t * const f_pRefV, + const uint8_t * const f_pRefHV, + + const uint8_t * const cur, + + const uint8_t * const b_pRef, + const uint8_t * const b_pRefH, + const uint8_t * const b_pRefV, + const uint8_t * const b_pRefHV, + + const int x, + const int y, + + const int TRB, + const int TRD, + + const int start_x, + const int start_y, + + int iMinSAD, + VECTOR * const currMV, + const VECTOR * const directmv, + + const int32_t min_dx, + const int32_t max_dx, + const int32_t min_dy, + const int32_t max_dy, + + const int32_t iEdgedWidth, + const int32_t iDiamondSize, + + const int32_t iQuant, + int iFound) +{ +/* Do a diamond search around given starting point, return SAD of best */ + + int32_t iSAD; + + VECTOR backupMV; + + currMV->x = start_x; + currMV->y = start_y; + +/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ + + do + { + iFound = 1; + + backupMV = *currMV; + + CHECK_MV16_DIRECT_FOUND(backupMV.x - iDiamondSize, backupMV.y); + CHECK_MV16_DIRECT_FOUND(backupMV.x + iDiamondSize, backupMV.y); + CHECK_MV16_DIRECT_FOUND(backupMV.x, backupMV.y - iDiamondSize); + CHECK_MV16_DIRECT_FOUND(backupMV.x, backupMV.y + iDiamondSize); + + } while (!iFound); + + return iMinSAD; +} + + int32_t AdvDiamond8_MainSearch(const uint8_t * const pRef, const uint8_t * const pRefH, @@ -859,11 +1098,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, - int32_t iMinSAD, + int start_x, + int start_y, + int iMinSAD, VECTOR * const currMV, - const VECTOR * const pmv, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -880,26 +1120,26 @@ /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ if (iDirection) { - CHECK_MV8_CANDIDATE(startx - iDiamondSize, starty); - CHECK_MV8_CANDIDATE(startx + iDiamondSize, starty); - CHECK_MV8_CANDIDATE(startx, starty - iDiamondSize); - CHECK_MV8_CANDIDATE(startx, starty + iDiamondSize); + CHECK_MV8_CANDIDATE(start_x - iDiamondSize, start_y); + CHECK_MV8_CANDIDATE(start_x + iDiamondSize, start_y); + CHECK_MV8_CANDIDATE(start_x, start_y - iDiamondSize); + CHECK_MV8_CANDIDATE(start_x, start_y + iDiamondSize); } else { int bDirection = 1 + 2 + 4 + 8; do { iDirection = 0; if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); if (bDirection & 2) - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); if (bDirection & 4) - CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); + CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); if (bDirection & 8) - CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); + CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); /* now we're doing diagonal checks near our candidate */ @@ -907,99 +1147,99 @@ { bDirection = iDirection; iDirection = 0; - startx = currMV->x; - starty = currMV->y; + start_x = currMV->x; + start_y = currMV->y; if (bDirection & 3) //our candidate is left or right { - CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); - CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); + CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); + CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); } else // what remains here is up or down { - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); } if (iDirection) { bDirection += iDirection; - startx = currMV->x; - starty = currMV->y; + start_x = currMV->x; + start_y = currMV->y; } } else //about to quit, eh? not so fast.... { switch (bDirection) { case 2: - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; case 1: - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); break; case 2 + 4: - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; case 4: - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); break; case 8: - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); break; case 1 + 4: - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); break; case 2 + 8: - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; case 1 + 8: - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); break; default: //1+2+4+8 == we didn't find anything at all - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty - iDiamondSize, 1 + 4); - CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, - starty + iDiamondSize, 1 + 8); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty - iDiamondSize, 2 + 4); - CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, - starty + iDiamondSize, 2 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y - iDiamondSize, 1 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, + start_y + iDiamondSize, 1 + 8); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y - iDiamondSize, 2 + 4); + CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, + start_y + iDiamondSize, 2 + 8); break; } if (!(iDirection)) break; //ok, the end. really else { bDirection = iDirection; - startx = currMV->x; - starty = currMV->y; + start_x = currMV->x; + start_y = currMV->y; } } } @@ -1017,11 +1257,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, - int32_t iMinSAD, - VECTOR * const currMV, - const VECTOR * const pmv, + const int start_x, + const int start_y, + int iMinSAD, + VECTOR * const currMV, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -1036,8 +1277,8 @@ int32_t dx, dy; VECTOR backupMV; - backupMV.x = startx; - backupMV.y = starty; + backupMV.x = start_x; + backupMV.y = start_y; for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) @@ -1058,7 +1299,8 @@ const int y, VECTOR * const currMV, int32_t iMinSAD, - const VECTOR * const pmv, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -1087,6 +1329,7 @@ #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) + int32_t PMVfastSearch16(const uint8_t * const pRef, const uint8_t * const pRefH, @@ -1095,6 +1338,10 @@ const IMAGE * const pCur, const int x, const int y, + const int start_x, /* start is searched first, so it should contain the most */ + const int start_y, /* likely motion vector for this block */ + const int center_x, /* center is from where length of MVs is measured */ + const int center_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, @@ -1148,14 +1395,8 @@ } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ - //bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); -/* fprintf(stderr,"pmv: %d %d / %d --- %d %d %d %d %d %d - %d %d %d\n", - pmv[0].x,pmv[0].y,psad[0], - pmv[1].x,pmv[1].y,pmv[2].x,pmv[2].y,pmv[3].x,pmv[3].y, - psad[1],psad[2],psad[3]); -*/ if ((x == 0) && (y == 0)) { threshA = 512; threshB = 1024; @@ -1179,7 +1420,9 @@ If SAD<=256 goto Step 10. */ - *currMV = pmv[0]; /* current best := prediction */ + currMV->x = start_x; + currMV->y = start_y; + if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ currMV->x = EVEN(currMV->x); currMV->y = EVEN(currMV->y); @@ -1203,7 +1446,7 @@ get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); iMinSAD += - calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, + calc_delta_16(currMV->x - center_x, currMV->y - center_y, (uint8_t) iFcode, iQuant); if ((iMinSAD < 256) || @@ -1340,12 +1583,11 @@ backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ -// fprintf(stderr,"Entering Diamond %d %d (%d):\n",x,y,iMinSAD); - /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ iSAD = - (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, - currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, + (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, + currMV->x, currMV->y, iMinSAD, &newMV, center_x, center_y, + min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -1360,7 +1602,7 @@ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, - pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, + center_x, center_y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -1373,8 +1615,9 @@ if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, - iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, - max_dy, iEdgedWidth, iDiamondSize, iFcode, + iMinSAD, &newMV, center_x, center_y, + min_dx, max_dx, min_dy, max_dy, + iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { @@ -1392,14 +1635,12 @@ if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step iMinSAD = Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, - iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, + iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); -/*fprintf(stderr,"Chosen for %d %d: %d %d - %d %d\n",x,y,currMV->x,currMV->y,pmv[0].x,pmv[0].y); -*/ PMVfast16_Terminate_without_Refine: - currPMV->x = currMV->x - pmv[0].x; - currPMV->y = currMV->y - pmv[0].y; + currPMV->x = currMV->x - center_x; + currPMV->y = currMV->y - center_y; return iMinSAD; } @@ -1416,11 +1657,12 @@ const uint8_t * const cur, const int x, const int y, - int32_t startx, - int32_t starty, + int32_t start_x, + int32_t start_y, int32_t iMinSAD, VECTOR * const currMV, - const VECTOR * const pmv, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -1434,11 +1676,12 @@ /* Do a diamond search around given starting point, return SAD of best */ int32_t iDirection = 0; + int32_t iDirectionBackup; int32_t iSAD; VECTOR backupMV; - backupMV.x = startx; - backupMV.y = starty; + backupMV.x = start_x; + backupMV.y = start_y; /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ @@ -1447,30 +1690,217 @@ CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); - if (iDirection) + if (iDirection) { while (!iFound) { iFound = 1; backupMV = *currMV; // since iDirection!=0, this is well defined! + iDirectionBackup = iDirection; - if (iDirection != 2) + if (iDirectionBackup != 2) CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); - if (iDirection != 1) + if (iDirectionBackup != 1) CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); - if (iDirection != 4) + if (iDirectionBackup != 4) CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); - if (iDirection != 3) + if (iDirectionBackup != 3) CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); + } + } else { + currMV->x = start_x; + currMV->y = start_y; + } + return iMinSAD; +} + + + + +int32_t +Square8_MainSearch(const uint8_t * const pRef, + const uint8_t * const pRefH, + const uint8_t * const pRefV, + const uint8_t * const pRefHV, + const uint8_t * const cur, + const int x, + const int y, + int32_t start_x, + int32_t start_y, + int32_t iMinSAD, + VECTOR * const currMV, + const int center_x, + const int center_y, + const int32_t min_dx, + const int32_t max_dx, + const int32_t min_dy, + const int32_t max_dy, + const int32_t iEdgedWidth, + const int32_t iDiamondSize, + const int32_t iFcode, + const int32_t iQuant, + int iFound) +{ +/* Do a square search around given starting point, return SAD of best */ + + int32_t iDirection = 0; + int32_t iSAD; + VECTOR backupMV; + + backupMV.x = start_x; + backupMV.y = start_y; + +/* It's one search with full square pattern, and new parts for all following diamonds */ + +/* new direction are extra, so 1-4 is normal diamond + 537 + 1*2 + 648 +*/ + + CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); + CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); + CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); + CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); + + CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + + + if (iDirection) { + while (!iFound) { + iFound = 1; + backupMV = *currMV; + + switch (iDirection) { + case 1: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y, 1); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + break; + case 2: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, + 2); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + break; + + case 3: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, + 4); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + break; + + case 4: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, + 3); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + break; + + case 5: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, + 1); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, + 3); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + break; + + case 6: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, + 2); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, + 3); + + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + + break; + + case 7: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y, 1); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, + 4); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + break; + + case 8: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, + 2); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, + 4); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + break; + default: + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, + 1); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, + 2); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, + 3); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, + 4); + + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y - iDiamondSize, 5); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, + backupMV.y + iDiamondSize, 6); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y - iDiamondSize, 7); + CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, + backupMV.y + iDiamondSize, 8); + break; + } + } } else { - currMV->x = startx; - currMV->y = starty; + currMV->x = start_x; + currMV->y = start_y; } return iMinSAD; } + + + + int32_t Halfpel8_Refine_c(const uint8_t * const pRef, const uint8_t * const pRefH, @@ -1481,7 +1911,8 @@ const int y, VECTOR * const currMV, int32_t iMinSAD, - const VECTOR * const pmv, + const int center_x, + const int center_y, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, @@ -1520,6 +1951,8 @@ const int y, const int start_x, const int start_y, + const int center_x, + const int center_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, @@ -1576,7 +2009,6 @@ } /* because we might use IF (dx>max_dx) THEN dx=max_dx; */ - //bPredEq = get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); bPredEq = get_pmvdata2(pMBs, iWcount, 0, (x >> 1), (y >> 1), iSubBlock, pmv, psad); if ((x == 0) && (y == 0)) { @@ -1584,7 +2016,7 @@ threshB = 1024 / 4; } else { - threshA = psad[0] / 4; /* good estimate */ + threshA = psad[0] / 4; /* good estimate? */ threshB = threshA + 256 / 4; if (threshA < 512 / 4) threshA = 512 / 4; @@ -1606,9 +2038,9 @@ // Prepare for main loop -// if (MotionFlags & PMV_USESQUARES8) -// MainSearchPtr = Square8_MainSearch; -// else + if (MotionFlags & PMV_USESQUARES8) + MainSearchPtr = Square8_MainSearch; + else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamond8_MainSearch; @@ -1623,7 +2055,7 @@ get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), iEdgedWidth); iMinSAD += - calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, + calc_delta_8(currMV->x - center_x, currMV->y - center_y, (uint8_t) iFcode, iQuant); if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) @@ -1666,7 +2098,7 @@ // the median prediction might be even better than mv16 if (!MVequal(pmv[0], startMV)) - CHECK_MV8_CANDIDATE(pmv[0].x, pmv[0].y); + CHECK_MV8_CANDIDATE(center_x, center_y); // (0,0) if needed if (!MVzero(pmv[0])) @@ -1763,7 +2195,7 @@ /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, - currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, + currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -1778,7 +2210,7 @@ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, - pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, + pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -1791,7 +2223,7 @@ if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, - iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, + iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -1810,13 +2242,13 @@ if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step iMinSAD = Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, - iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, + iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); PMVfast8_Terminate_without_Refine: - currPMV->x = currMV->x - pmv[0].x; - currPMV->y = currMV->y - pmv[0].y; + currPMV->x = currMV->x - center_x; + currPMV->y = currMV->y - center_y; return iMinSAD; } @@ -1829,6 +2261,10 @@ const IMAGE * const pCur, const int x, const int y, + const int start_x, + const int start_y, + const int center_x, + const int center_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, @@ -1887,7 +2323,6 @@ max_dy = EVEN(max_dy); } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ - //bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); /* Step 4: Calculate SAD around the Median prediction. @@ -1899,7 +2334,9 @@ // Prepare for main loop - *currMV = pmv[0]; /* current best := median prediction */ + currMV->x = start_x; + currMV->y = start_y; + if (!(MotionFlags & PMV_HALFPEL16)) { currMV->x = EVEN(currMV->x); currMV->y = EVEN(currMV->y); @@ -1921,7 +2358,7 @@ get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); iMinSAD += - calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, + calc_delta_16(currMV->x - center_x, currMV->y - center_y, (uint8_t) iFcode, iQuant); // thresh1 is fixed to 256 @@ -2045,7 +2482,7 @@ iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, - currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, + currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); if (iSAD < iMinSAD) { @@ -2060,7 +2497,7 @@ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, - pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, + pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); } @@ -2073,7 +2510,7 @@ if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, - iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, + iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); if (iSAD < iMinSAD) { @@ -2089,15 +2526,15 @@ if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step iMinSAD = Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, - iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, + iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); EPZS16_Terminate_without_Refine: *oldMB = *prevMB; - currPMV->x = currMV->x - pmv[0].x; - currPMV->y = currMV->y - pmv[0].y; + currPMV->x = currMV->x - center_x; + currPMV->y = currMV->y - center_y; return iMinSAD; } @@ -2112,6 +2549,8 @@ const int y, const int start_x, const int start_y, + const int center_x, + const int center_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, @@ -2166,7 +2605,6 @@ max_dy = EVEN(max_dy); } /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ - //bPredEq = get_pmvdata(pMBs, x >> 1, y >> 1, iWcount, iSubBlock, pmv, psad); bPredEq = get_pmvdata2(pMBs, iWcount, 0, x >> 1, y >> 1, iSubBlock, pmv, psad); @@ -2202,7 +2640,7 @@ get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), iEdgedWidth); iMinSAD += - calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, + calc_delta_8(currMV->x - center_x, currMV->y - center_y, (uint8_t) iFcode, iQuant); @@ -2278,9 +2716,9 @@ // there is no EPZS^2 for inter4v at the moment -// if (MotionFlags & PMV_USESQUARES8) -// MainSearchPtr = Square8_MainSearch; -// else + if (MotionFlags & PMV_USESQUARES8) + MainSearchPtr = Square8_MainSearch; + else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamond8_MainSearch; @@ -2289,7 +2727,7 @@ iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, - currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, + currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); @@ -2305,7 +2743,7 @@ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, - pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, + pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); @@ -2318,7 +2756,7 @@ if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, - iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, + iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); @@ -2335,13 +2773,13 @@ if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step iMinSAD = Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, - iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, + iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); EPZS8_Terminate_without_Refine: - currPMV->x = currMV->x - pmv[0].x; - currPMV->y = currMV->y - pmv[0].y; + currPMV->x = currMV->x - center_x; + currPMV->y = currMV->y - center_y; return iMinSAD; } @@ -2355,6 +2793,10 @@ const IMAGE * const pCur, const int x, const int y, + const int start_x, /* start should be most likely vector */ + const int start_y, + const int center_x, /* center is from where length of MVs is measured */ + const int center_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, @@ -2382,7 +2824,7 @@ int32_t iFound; VECTOR newMV; - VECTOR backupMV; /* just for PMVFAST */ + VECTOR backupMV; VECTOR pmv[4]; int32_t psad[4]; @@ -2395,6 +2837,7 @@ int32_t threshA, threshB; int32_t bPredEq; int32_t iMinSAD, iSAD; + /* Get maximum range */ get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, @@ -2411,6 +2854,9 @@ *currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; } else { + + bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); + threshA = psad[0]; threshB = threshA + 256; if (threshA < 512) @@ -2420,7 +2866,6 @@ if (threshB > 1792) threshB = 1792; - bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); *currMV = pmv[0]; /* current best := prediction */ } @@ -2451,7 +2896,7 @@ get_iref_mv(pRef, x, y, 16, currMV, iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); iMinSAD += - calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, + calc_delta_16(currMV->x - center_x, currMV->y - center_y, (uint8_t) iFcode, iQuant); if ((iMinSAD < 256) || @@ -2570,7 +3015,7 @@ /* default: use best prediction as starting point for one call of PMVfast_MainSearch */ iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, - currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, + currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -2585,7 +3030,7 @@ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, - pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, + pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -2598,7 +3043,7 @@ if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, - iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, + iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); @@ -2621,193 +3066,14 @@ if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step iMinSAD = Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, - iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, + iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); pmv[0] = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); // get _REAL_ prediction (halfpel possible) PMVfastInt16_Terminate_without_Refine: - currPMV->x = currMV->x - pmv[0].x; - currPMV->y = currMV->y - pmv[0].y; + currPMV->x = currMV->x - center_x; + currPMV->y = currMV->y - center_y; return iMinSAD; } - - -/* *********************************************************** - bvop motion estimation -// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) -***************************************************************/ - - -void -MotionEstimationBVOP(MBParam * const pParam, - FRAMEINFO * const frame, - const int32_t time_bp, - const int32_t time_pp, - // forward (past) reference - const MACROBLOCK * const f_mbs, - const IMAGE * const f_ref, - const IMAGE * const f_refH, - const IMAGE * const f_refV, - const IMAGE * const f_refHV, - // backward (future) reference - const MACROBLOCK * const b_mbs, - const IMAGE * const b_ref, - const IMAGE * const b_refH, - const IMAGE * const b_refV, - const IMAGE * const b_refHV) -{ - const int mb_width = pParam->mb_width; - const int mb_height = pParam->mb_height; - const int edged_width = pParam->edged_width; - - int i, j, k; - - static const VECTOR zeroMV={0,0}; - - int f_sad16; /* forward (as usual) search */ - int b_sad16; /* backward (only in b-frames) search */ - int i_sad16; /* interpolated (both direction, b-frames only) */ - int d_sad16; /* direct mode (assume linear motion) */ - - int best_sad; - - VECTOR pmv_dontcare; - - int f_count=0; - int b_count=0; - int i_count=0; - int d_count=0; - int dnv_count=0; - int s_count=0; - const int64_t TRB = (int32_t)time_pp - (int32_t)time_bp; - const int64_t TRD = (int32_t)time_pp; - - // fprintf(stderr,"TRB = %lld TRD = %lld time_bp =%d time_pp =%d\n\n",TRB,TRD,time_bp,time_pp); - // note: i==horizontal, j==vertical - for (j = 0; j < mb_height; j++) { - for (i = 0; i < mb_width; i++) { - MACROBLOCK *mb = &frame->mbs[i + j * mb_width]; - const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; - const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; - - VECTOR directMV; - VECTOR deltaMV=zeroMV; - -/* special case, if collocated block is SKIPed: encoding is forward(0,0) */ - - if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && - b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { - mb->mode = MODE_NOT_CODED; - mb->mvs[0].x = 0; - mb->mvs[0].y = 0; - mb->b_mvs[0].x = 0; - mb->b_mvs[0].y = 0; - continue; - } - - /* same method of scaling as in decoder.c, so we copy 1:1 from there */ - - for (k = 0; k < 4; k++) { - directMV = b_mb->mvs[k]; - - mb->mvs[k].x = (int32_t) ((TRB * directMV.x) / TRD + deltaMV.x); - mb->b_mvs[k].x = (int32_t) ((deltaMV.x == 0) - ? ((TRB - TRD) * directMV.x) / TRD - : mb->mvs[k].x - directMV.x); - mb->mvs[k].y = (int32_t) ((TRB * directMV.y) / TRD + deltaMV.y); - mb->b_mvs[k].y = (int32_t) ((deltaMV.y == 0) - ? ((TRB - TRD) * directMV.y) / TRD - : mb->mvs[k].y - directMV.y); - } - - d_sad16 = - sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, - get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, - i, j, 16, &mb->mvs[0], edged_width), - get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, - i, j, 16, &mb->b_mvs[0], edged_width), - edged_width); - - // forward search - f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, - &frame->image, i, j, frame->motion_flags, - frame->quant, frame->fcode, pParam, - f_mbs, f_mbs, /* todo */ - &mb->mvs[0], &pmv_dontcare); // ignore pmv (why?) - - - // backward search - b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, - &frame->image, i, j, frame->motion_flags, - frame->quant, frame->bcode, pParam, - b_mbs, b_mbs, /* todo */ - &mb->b_mvs[0], &pmv_dontcare); // ignore pmv - - i_sad16 = - sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, - get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, - i, j, 16, &mb->mvs[0], edged_width), - get_ref(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, - i, j, 16, -mb->b_mvs[0].x, -mb->b_mvs[0].y, edged_width), - edged_width); - - // TODO: direct search - // predictor + range of [-32,32] - - DEBUG2("f_MV: ",mb->mvs[0].x,mb->mvs[0].y); - DEBUG2("b_MV: ",mb->b_mvs[0].x,mb->b_mvs[0].y); - -/* fprintf(stderr,"f_sad16 = %d, b_sad16 = %d, i_sad16 = %d, d_sad16 = %d\n", - f_sad16,b_sad16,i_sad16,d_sad16); -*/ - -// d_sad16 -= 50; -// d_sad16 = 65535; -// i_sad16 = 65535; -// b_sad16 = 65535; - - if (f_sad16 < b_sad16) { - best_sad = f_sad16; - mb->mode = MODE_FORWARD; - } else { - best_sad = b_sad16; - mb->mode = MODE_BACKWARD; - } - - if (i_sad16 < best_sad) { - best_sad = i_sad16; - mb->mode = MODE_INTERPOLATE; - } - - if (d_sad16 < best_sad) { - best_sad = d_sad16; - mb->mode = MODE_DIRECT_NONE_MV; - } - - switch (mb->mode) - { - case MODE_FORWARD: - f_count++; break; - case MODE_BACKWARD: - b_count++; break; - case MODE_INTERPOLATE: - i_count++; break; - case MODE_DIRECT: - d_count++; break; - case MODE_DIRECT_NONE_MV: - dnv_count++; break; - default: - s_count++; break; - } - - } - } - -#ifdef _DEBUG_BFRAME_STAT - fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D0: %04d D: %04d S: %04d\n", - f_count,b_count,i_count,dnv_count,d_count,s_count); -#endif - -}