Parent Directory
|
Revision Log
Global assembler optimization of halfpel8_refine, via inlinig of sad for ia64
/************************************************************************** * * XVID MPEG-4 VIDEO CODEC * motion estimation * * 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 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 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. * * 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. * *************************************************************************/ /************************************************************************** * * Modifications: * * 01.05.2002 updated MotionEstimationBVOP * 25.04.2002 partial prevMB conversion * 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> * 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 <isibaar@videocoding.de> * **************************************************************************/ #include <assert.h> #include <stdio.h> #include <stdlib.h> #include "../encoder.h" #include "../utils/mbfunctions.h" #include "../prediction/mbprediction.h" #include "../global.h" #include "../utils/timer.h" #include "motion.h" #include "sad.h" static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ { 0, (int) (1.00235 + 0.5), (int) (1.15582 + 0.5), (int) (1.31976 + 0.5), (int) (1.49591 + 0.5), (int) (1.68601 + 0.5), (int) (1.89187 + 0.5), (int) (2.11542 + 0.5), (int) (2.35878 + 0.5), (int) (2.62429 + 0.5), (int) (2.91455 + 0.5), (int) (3.23253 + 0.5), (int) (3.58158 + 0.5), (int) (3.96555 + 0.5), (int) (4.38887 + 0.5), (int) (4.85673 + 0.5), (int) (5.37519 + 0.5), (int) (5.95144 + 0.5), (int) (6.59408 + 0.5), (int) (7.31349 + 0.5), (int) (8.12242 + 0.5), (int) (9.03669 + 0.5), (int) (10.0763 + 0.5), (int) (11.2669 + 0.5), (int) (12.6426 + 0.5), (int) (14.2493 + 0.5), (int) (16.1512 + 0.5), (int) (18.442 + 0.5), (int) (21.2656 + 0.5), (int) (24.8580 + 0.5), (int) (29.6436 + 0.5), (int) (36.4949 + 0.5) }; static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now */ // mv.length table static const uint32_t mvtab[33] = { 1, 2, 3, 4, 6, 7, 7, 7, 9, 9, 9, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 11, 11, 11, 11, 11, 11, 12, 12 }; static __inline uint32_t mv_bits(int32_t component, const uint32_t iFcode) { if (component == 0) return 1; if (component < 0) component = -component; if (iFcode == 1) { if (component > 32) component = 32; return mvtab[component] + 1; } component += (1 << (iFcode - 1)) - 1; component >>= (iFcode - 1); if (component > 32) component = 32; return mvtab[component] + 1 + iFcode - 1; } static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) { return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); } static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) { return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); } bool MotionEstimation(MBParam * const pParam, FRAMEINFO * const current, FRAMEINFO * const reference, const IMAGE * const pRefH, const IMAGE * const pRefV, const IMAGE * const pRefHV, const uint32_t iLimit) { const uint32_t iWcount = pParam->mb_width; const uint32_t iHcount = pParam->mb_height; MACROBLOCK *const pMBs = current->mbs; MACROBLOCK *const prevMBs = reference->mbs; const IMAGE *const pCurrent = ¤t->image; const IMAGE *const pRef = &reference->image; const VECTOR zeroMV = { 0, 0 }; int32_t x, y; int32_t iIntra = 0; VECTOR pmv; if (sadInit) (*sadInit) (); for (y = 0; y < iHcount; y++) { for (x = 0; x < iWcount; x ++) { MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; pMB->sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, y, current->motion_flags, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, &pMB->pmvs[0]); if (0 < (pMB->sad16 - MV16_INTER_BIAS)) { int32_t deviation; deviation = dev16(pCurrent->y + x * 16 + y * 16 * pParam->edged_width, pParam->edged_width); if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) { pMB->mode = MODE_INTRA; pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV; pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; iIntra++; if (iIntra >= iLimit) return 1; continue; } } pmv = pMB->pmvs[0]; if (current->global_flags & XVID_INTER4V) if ((!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE)) { int32_t sad8 = IMV16X16 * current->quant; 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, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); if (sad8 < pMB->sad16) 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, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); if (sad8 < pMB->sad16) 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, current->quant, current->fcode, pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); if (sad8 < pMB->sad16) 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]); /* decide: MODE_INTER or MODE_INTER4V mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v */ if (sad8 < pMB->sad16) { pMB->mode = MODE_INTER4V; pMB->sad8[0] *= 4; pMB->sad8[1] *= 4; pMB->sad8[2] *= 4; pMB->sad8[3] *= 4; continue; } } pMB->mode = MODE_INTER; pMB->pmvs[0] = pmv; /* pMB->pmvs[1] = pMB->pmvs[2] = pMB->pmvs[3] are not needed for INTER */ pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = pMB->sad16; } } 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);\ 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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ } #define CHECK_MV16_CANDIDATE(X,Y) { \ if ( ((X) <= max_dx) && ((X) >= min_dx) \ && ((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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ } #define CHECK_MV16_CANDIDATE_DIR(X,Y,D) { \ if ( ((X) <= max_dx) && ((X) >= min_dx) \ && ((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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ } #define CHECK_MV16_CANDIDATE_FOUND(X,Y,D) { \ if ( ((X) <= max_dx) && ((X) >= min_dx) \ && ((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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ } #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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ } #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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ } #define CHECK_MV8_CANDIDATE(X,Y) { \ if ( ((X) <= max_dx) && ((X) >= min_dx) \ && ((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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ } #define CHECK_MV8_CANDIDATE_DIR(X,Y,D) { \ if ( ((X) <= max_dx) && ((X) >= min_dx) \ && ((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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ } #define CHECK_MV8_CANDIDATE_FOUND(X,Y,D) { \ if ( ((X) <= max_dx) && ((X) >= min_dx) \ && ((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);\ if (iSAD < iMinSAD) \ { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ } /* too slow and not fully functional at the moment */ /* int32_t ZeroSearch16( const uint8_t * const pRef, const uint8_t * const pRefH, const uint8_t * const pRefV, const uint8_t * const pRefHV, const IMAGE * const pCur, const int x, const int y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, MBParam * const pParam, const MACROBLOCK * const pMBs, const MACROBLOCK * const prevMBs, VECTOR * const currMV, VECTOR * const currPMV) { const int32_t iEdgedWidth = pParam->edged_width; const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; int32_t iSAD; VECTOR pred; pred = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); if (iSAD <= iQuant * 96) iSAD -= MV16_00_BIAS; currMV->x = 0; currMV->y = 0; currPMV->x = -pred.x; currPMV->y = -pred.y; return iSAD; } */ int32_t Diamond16_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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 diamond search around given starting point, return SAD of best */ int32_t iDirection = 0; int32_t iSAD; VECTOR backupMV; backupMV.x = startx; backupMV.y = starty; /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); if (iDirection) while (!iFound) { iFound = 1; backupMV = *currMV; if (iDirection != 2) CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); if (iDirection != 1) CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); if (iDirection != 4) CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); if (iDirection != 3) CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); } else { currMV->x = startx; currMV->y = starty; } return iMinSAD; } int32_t Square16_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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 = startx; backupMV.y = starty; /* 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_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); if (iDirection) while (!iFound) { iFound = 1; backupMV = *currMV; switch (iDirection) { case 1: CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); break; case 2: CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 3: CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 4: CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); break; case 5: CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); break; case 6: CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 7: CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; case 8: CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; default: CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y - iDiamondSize, 5); CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y + iDiamondSize, 6); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y - iDiamondSize, 7); CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y + iDiamondSize, 8); break; } } else { currMV->x = startx; currMV->y = starty; } return iMinSAD; } int32_t Full16_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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) { int32_t iSAD; int32_t dx, dy; VECTOR backupMV; backupMV.x = startx; backupMV.y = starty; for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) NOCHECK_MV16_CANDIDATE(dx, dy); return iMinSAD; } int32_t AdvDiamond16_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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 iDirection) { int32_t iSAD; /* 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); } 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); if (bDirection & 2) CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); if (bDirection & 4) CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); if (bDirection & 8) CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); /* now we're doing diagonal checks near our candidate */ if (iDirection) //checking if anything found { bDirection = iDirection; iDirection = 0; startx = currMV->x; starty = 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); } else // what remains here is up or down { CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); } if (iDirection) { bDirection += iDirection; startx = currMV->x; starty = 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); break; case 1: CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty - iDiamondSize, 1 + 4); CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty + 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); break; case 4: CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty - iDiamondSize, 2 + 4); CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty - 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); 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); 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); 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); 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); break; } if (!iDirection) break; //ok, the end. really else { bDirection = iDirection; startx = currMV->x; starty = currMV->y; } } } while (1); //forever } return iMinSAD; } int32_t AdvDiamond8_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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 iDirection) { int32_t iSAD; /* 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); } 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); if (bDirection & 2) CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); if (bDirection & 4) CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); if (bDirection & 8) CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); /* now we're doing diagonal checks near our candidate */ if (iDirection) //checking if anything found { bDirection = iDirection; iDirection = 0; startx = currMV->x; starty = 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); } else // what remains here is up or down { CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); } if (iDirection) { bDirection += iDirection; startx = currMV->x; starty = 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); break; case 1: CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty - iDiamondSize, 1 + 4); CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty + 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); break; case 4: CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty - iDiamondSize, 2 + 4); CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty - 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); 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); 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); 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); 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); break; } if (!(iDirection)) break; //ok, the end. really else { bDirection = iDirection; startx = currMV->x; starty = currMV->y; } } } while (1); //forever } return iMinSAD; } int32_t Full8_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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) { int32_t iSAD; int32_t dx, dy; VECTOR backupMV; backupMV.x = startx; backupMV.y = starty; for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) NOCHECK_MV8_CANDIDATE(dx, dy); return iMinSAD; } Halfpel8_RefineFuncPtr Halfpel8_Refine; int32_t Halfpel16_Refine(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, VECTOR * const currMV, int32_t iMinSAD, const VECTOR * const pmv, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, const int32_t max_dy, const int32_t iFcode, const int32_t iQuant, const int32_t iEdgedWidth) { /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ int32_t iSAD; VECTOR backupMV = *currMV; CHECK_MV16_CANDIDATE(backupMV.x - 1, backupMV.y - 1); CHECK_MV16_CANDIDATE(backupMV.x, backupMV.y - 1); CHECK_MV16_CANDIDATE(backupMV.x + 1, backupMV.y - 1); CHECK_MV16_CANDIDATE(backupMV.x - 1, backupMV.y); CHECK_MV16_CANDIDATE(backupMV.x + 1, backupMV.y); CHECK_MV16_CANDIDATE(backupMV.x - 1, backupMV.y + 1); CHECK_MV16_CANDIDATE(backupMV.x, backupMV.y + 1); CHECK_MV16_CANDIDATE(backupMV.x + 1, backupMV.y + 1); return iMinSAD; } #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) int32_t PMVfastSearch16(const uint8_t * const pRef, const uint8_t * const pRefH, const uint8_t * const pRefV, const uint8_t * const pRefHV, const IMAGE * const pCur, const int x, const int y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, const MBParam * const pParam, const MACROBLOCK * const pMBs, const MACROBLOCK * const prevMBs, VECTOR * const currMV, VECTOR * const currPMV) { const uint32_t iWcount = pParam->mb_width; const int32_t iWidth = pParam->width; const int32_t iHeight = pParam->height; const int32_t iEdgedWidth = pParam->edged_width; const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; int32_t iDiamondSize; int32_t min_dx; int32_t max_dx; int32_t min_dy; int32_t max_dy; int32_t iFound; VECTOR newMV; VECTOR backupMV; /* just for PMVFAST */ VECTOR pmv[4]; int32_t psad[4]; MainSearch16FuncPtr MainSearchPtr; const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; 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, iFcode); /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ if (!(MotionFlags & PMV_HALFPEL16)) { min_dx = EVEN(min_dx); max_dx = EVEN(max_dx); min_dy = EVEN(min_dy); 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); /* 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; } else { threshA = psad[0]; threshB = threshA + 256; if (threshA < 512) threshA = 512; if (threshA > 1024) threshA = 1024; if (threshB > 1792) threshB = 1792; } iFound = 0; /* Step 4: Calculate SAD around the Median prediction. MinSAD=SAD If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. If SAD<=256 goto Step 10. */ *currMV = pmv[0]; /* current best := prediction */ if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ currMV->x = EVEN(currMV->x); currMV->y = EVEN(currMV->y); } if (currMV->x > max_dx) { currMV->x = max_dx; } if (currMV->x < min_dx) { currMV->x = min_dx; } if (currMV->y > max_dy) { currMV->y = max_dy; } if (currMV->y < min_dy) { currMV->y = min_dy; } iMinSAD = sad16(cur, 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, (uint8_t) iFcode, iQuant); if ((iMinSAD < 256) || ((MVequal(*currMV, prevMB->mvs[0])) && ((int32_t) iMinSAD < prevMB->sad16))) { if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode { if (!MVzero(*currMV)) { iMinSAD += MV16_00_BIAS; CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures iMinSAD -= MV16_00_BIAS; } } if (MotionFlags & PMV_QUICKSTOP16) goto PMVfast16_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto PMVfast16_Terminate_with_Refine; } /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion vector of the median. If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 */ if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[0]))) iFound = 2; /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. Otherwise select large Diamond Search. */ if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) iDiamondSize = 1; // halfpel! else iDiamondSize = 2; // halfpel! if (!(MotionFlags & PMV_HALFPELDIAMOND16)) iDiamondSize *= 2; /* Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. Also calculate (0,0) but do not subtract offset. Let MinSAD be the smallest SAD up to this point. If MV is (0,0) subtract offset. */ // (0,0) is always possible if (!MVzero(pmv[0])) CHECK_MV16_ZERO; // previous frame MV is always possible if (!MVzero(prevMB->mvs[0])) if (!MVequal(prevMB->mvs[0], pmv[0])) CHECK_MV16_CANDIDATE(prevMB->mvs[0].x, prevMB->mvs[0].y); // left neighbour, if allowed if (!MVzero(pmv[1])) if (!MVequal(pmv[1], prevMB->mvs[0])) if (!MVequal(pmv[1], pmv[0])) { if (!(MotionFlags & PMV_HALFPEL16)) { pmv[1].x = EVEN(pmv[1].x); pmv[1].y = EVEN(pmv[1].y); } CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); } // top neighbour, if allowed if (!MVzero(pmv[2])) if (!MVequal(pmv[2], prevMB->mvs[0])) if (!MVequal(pmv[2], pmv[0])) if (!MVequal(pmv[2], pmv[1])) { if (!(MotionFlags & PMV_HALFPEL16)) { pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); } CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); // top right neighbour, if allowed if (!MVzero(pmv[3])) if (!MVequal(pmv[3], prevMB->mvs[0])) if (!MVequal(pmv[3], pmv[0])) if (!MVequal(pmv[3], pmv[1])) if (!MVequal(pmv[3], pmv[2])) { if (!(MotionFlags & PMV_HALFPEL16)) { pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); } CHECK_MV16_CANDIDATE(pmv[3].x, pmv[3].y); } } if ((MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) iMinSAD -= MV16_00_BIAS; /* Step 6: If MinSAD <= thresa goto Step 10. If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. */ if ((iMinSAD <= threshA) || (MVequal(*currMV, prevMB->mvs[0]) && ((int32_t) iMinSAD < prevMB->sad16))) { if (MotionFlags & PMV_QUICKSTOP16) goto PMVfast16_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto PMVfast16_Terminate_with_Refine; } /************ (Diamond Search) **************/ /* Step 7: Perform Diamond search, with either the small or large diamond. If Found=2 only examine one Diamond pattern, and afterwards goto step 10 Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. If center then goto step 10. Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. Refine by using small diamond and goto step 10. */ if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = Square16_MainSearch; else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamond16_MainSearch; else MainSearchPtr = Diamond16_MainSearch; 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, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } if (MotionFlags & PMV_EXTSEARCH16) { /* extended: search (up to) two more times: orignal prediction and (0,0) */ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } 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, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } } /* Step 10: The motion vector is chosen according to the block corresponding to MinSAD. */ PMVfast16_Terminate_with_Refine: 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, 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; return iMinSAD; } int32_t Diamond8_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 startx, int32_t starty, int32_t iMinSAD, VECTOR * const currMV, const VECTOR * const pmv, 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 diamond search around given starting point, return SAD of best */ int32_t iDirection = 0; int32_t iSAD; VECTOR backupMV; backupMV.x = startx; backupMV.y = starty; /* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ 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); if (iDirection) while (!iFound) { iFound = 1; backupMV = *currMV; // since iDirection!=0, this is well defined! if (iDirection != 2) CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, 1); if (iDirection != 1) CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, 2); if (iDirection != 4) CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, 3); if (iDirection != 3) CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, 4); } else { currMV->x = startx; currMV->y = starty; } return iMinSAD; } int32_t Halfpel8_Refine_c(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, VECTOR * const currMV, int32_t iMinSAD, const VECTOR * const pmv, const int32_t min_dx, const int32_t max_dx, const int32_t min_dy, const int32_t max_dy, const int32_t iFcode, const int32_t iQuant, const int32_t iEdgedWidth) { /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ int32_t iSAD; VECTOR backupMV = *currMV; CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y - 1); CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y - 1); CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y - 1); CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y); CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y); CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y + 1); CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y + 1); CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y + 1); return iMinSAD; } #define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) int32_t PMVfastSearch8(const uint8_t * const pRef, const uint8_t * const pRefH, const uint8_t * const pRefV, const uint8_t * const pRefHV, const IMAGE * const pCur, const int x, const int y, const int start_x, const int start_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, const MBParam * const pParam, const MACROBLOCK * const pMBs, const MACROBLOCK * const prevMBs, VECTOR * const currMV, VECTOR * const currPMV) { const uint32_t iWcount = pParam->mb_width; const int32_t iWidth = pParam->width; const int32_t iHeight = pParam->height; const int32_t iEdgedWidth = pParam->edged_width; const uint8_t *cur = pCur->y + x * 8 + y * 8 * iEdgedWidth; int32_t iDiamondSize; int32_t min_dx; int32_t max_dx; int32_t min_dy; int32_t max_dy; VECTOR pmv[4]; int32_t psad[4]; VECTOR newMV; VECTOR backupMV; VECTOR startMV; // const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; const MACROBLOCK *const prevMB = prevMBs + (x >> 1) + (y >> 1) * iWcount; int32_t threshA, threshB; int32_t iFound, bPredEq; int32_t iMinSAD, iSAD; int32_t iSubBlock = (y & 1) + (y & 1) + (x & 1); MainSearch8FuncPtr MainSearchPtr; /* Init variables */ startMV.x = start_x; startMV.y = start_y; /* Get maximum range */ get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, iFcode); if (!(MotionFlags & PMV_HALFPELDIAMOND8)) { min_dx = EVEN(min_dx); max_dx = EVEN(max_dx); min_dy = EVEN(min_dy); max_dy = EVEN(max_dy); } /* 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)) { threshA = 512 / 4; threshB = 1024 / 4; } else { threshA = psad[0] / 4; /* good estimate */ threshB = threshA + 256 / 4; if (threshA < 512 / 4) threshA = 512 / 4; if (threshA > 1024 / 4) threshA = 1024 / 4; if (threshB > 1792 / 4) threshB = 1792 / 4; } iFound = 0; /* Step 4: Calculate SAD around the Median prediction. MinSAD=SAD If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. If SAD<=256 goto Step 10. */ // Prepare for main loop // if (MotionFlags & PMV_USESQUARES8) // MainSearchPtr = Square8_MainSearch; // else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamond8_MainSearch; else MainSearchPtr = Diamond8_MainSearch; *currMV = startMV; iMinSAD = sad8(cur, 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, (uint8_t) iFcode, iQuant); if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) && ((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { if (MotionFlags & PMV_QUICKSTOP16) goto PMVfast8_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto PMVfast8_Terminate_with_Refine; } /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion vector of the median. If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 */ if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[iSubBlock]))) iFound = 2; /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. Otherwise select large Diamond Search. */ if ((!MVzero(pmv[0])) || (threshB < 1536 / 4) || (bPredEq)) iDiamondSize = 1; // 1 halfpel! else iDiamondSize = 2; // 2 halfpel = 1 full pixel! if (!(MotionFlags & PMV_HALFPELDIAMOND8)) iDiamondSize *= 2; /* Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. Also calculate (0,0) but do not subtract offset. Let MinSAD be the smallest SAD up to this point. If MV is (0,0) subtract offset. */ // the median prediction might be even better than mv16 if (!MVequal(pmv[0], startMV)) CHECK_MV8_CANDIDATE(pmv[0].x, pmv[0].y); // (0,0) if needed if (!MVzero(pmv[0])) if (!MVzero(startMV)) CHECK_MV8_ZERO; // previous frame MV if needed if (!MVzero(prevMB->mvs[iSubBlock])) if (!MVequal(prevMB->mvs[iSubBlock], startMV)) if (!MVequal(prevMB->mvs[iSubBlock], pmv[0])) CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x, prevMB->mvs[iSubBlock].y); if ((iMinSAD <= threshA) || (MVequal(*currMV, prevMB->mvs[iSubBlock]) && ((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { if (MotionFlags & PMV_QUICKSTOP16) goto PMVfast8_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto PMVfast8_Terminate_with_Refine; } // left neighbour, if allowed and needed if (!MVzero(pmv[1])) if (!MVequal(pmv[1], startMV)) if (!MVequal(pmv[1], prevMB->mvs[iSubBlock])) if (!MVequal(pmv[1], pmv[0])) { if (!(MotionFlags & PMV_HALFPEL8)) { pmv[1].x = EVEN(pmv[1].x); pmv[1].y = EVEN(pmv[1].y); } CHECK_MV8_CANDIDATE(pmv[1].x, pmv[1].y); } // top neighbour, if allowed and needed if (!MVzero(pmv[2])) if (!MVequal(pmv[2], startMV)) if (!MVequal(pmv[2], prevMB->mvs[iSubBlock])) if (!MVequal(pmv[2], pmv[0])) if (!MVequal(pmv[2], pmv[1])) { if (!(MotionFlags & PMV_HALFPEL8)) { pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); } CHECK_MV8_CANDIDATE(pmv[2].x, pmv[2].y); // top right neighbour, if allowed and needed if (!MVzero(pmv[3])) if (!MVequal(pmv[3], startMV)) if (!MVequal(pmv[3], prevMB->mvs[iSubBlock])) if (!MVequal(pmv[3], pmv[0])) if (!MVequal(pmv[3], pmv[1])) if (!MVequal(pmv[3], pmv[2])) { if (! (MotionFlags & PMV_HALFPEL8)) { pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); } CHECK_MV8_CANDIDATE(pmv[3].x, pmv[3].y); } } if ((MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) iMinSAD -= MV8_00_BIAS; /* Step 6: If MinSAD <= thresa goto Step 10. If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. */ if ((iMinSAD <= threshA) || (MVequal(*currMV, prevMB->mvs[iSubBlock]) && ((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { if (MotionFlags & PMV_QUICKSTOP16) goto PMVfast8_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto PMVfast8_Terminate_with_Refine; } /************ (Diamond Search) **************/ /* Step 7: Perform Diamond search, with either the small or large diamond. If Found=2 only examine one Diamond pattern, and afterwards goto step 10 Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. If center then goto step 10. Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. Refine by using small diamond and goto step 10. */ backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ /* 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, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } if (MotionFlags & PMV_EXTSEARCH8) { /* extended: search (up to) two more times: orignal prediction and (0,0) */ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } 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, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } } /* Step 10: The motion vector is chosen according to the block corresponding to MinSAD. By performing an optional local half-pixel search, we can refine this result even further. */ PMVfast8_Terminate_with_Refine: 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, iFcode, iQuant, iEdgedWidth); PMVfast8_Terminate_without_Refine: currPMV->x = currMV->x - pmv[0].x; currPMV->y = currMV->y - pmv[0].y; return iMinSAD; } int32_t EPZSSearch16(const uint8_t * const pRef, const uint8_t * const pRefH, const uint8_t * const pRefV, const uint8_t * const pRefHV, const IMAGE * const pCur, const int x, const int y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, const MBParam * const pParam, const MACROBLOCK * const pMBs, const MACROBLOCK * const prevMBs, VECTOR * const currMV, VECTOR * const currPMV) { const uint32_t iWcount = pParam->mb_width; const uint32_t iHcount = pParam->mb_height; const int32_t iWidth = pParam->width; const int32_t iHeight = pParam->height; const int32_t iEdgedWidth = pParam->edged_width; const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; int32_t min_dx; int32_t max_dx; int32_t min_dy; int32_t max_dy; VECTOR newMV; VECTOR backupMV; VECTOR pmv[4]; int32_t psad[8]; static MACROBLOCK *oldMBs = NULL; // const MACROBLOCK * const pMB = pMBs + x + y * iWcount; const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; MACROBLOCK *oldMB = NULL; int32_t thresh2; int32_t bPredEq; int32_t iMinSAD, iSAD = 9999; MainSearch16FuncPtr MainSearchPtr; if (oldMBs == NULL) { oldMBs = (MACROBLOCK *) calloc(iWcount * iHcount, sizeof(MACROBLOCK)); // fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); } oldMB = oldMBs + x + y * iWcount; /* Get maximum range */ get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, iFcode); if (!(MotionFlags & PMV_HALFPEL16)) { min_dx = EVEN(min_dx); max_dx = EVEN(max_dx); min_dy = EVEN(min_dy); 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. MinSAD=SAD If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. If SAD<=256 goto Step 10. */ // Prepare for main loop *currMV = pmv[0]; /* current best := median prediction */ if (!(MotionFlags & PMV_HALFPEL16)) { currMV->x = EVEN(currMV->x); currMV->y = EVEN(currMV->y); } if (currMV->x > max_dx) currMV->x = max_dx; if (currMV->x < min_dx) currMV->x = min_dx; if (currMV->y > max_dy) currMV->y = max_dy; if (currMV->y < min_dy) currMV->y = min_dy; /***************** This is predictor SET A: only median prediction ******************/ iMinSAD = sad16(cur, 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, (uint8_t) iFcode, iQuant); // thresh1 is fixed to 256 if ((iMinSAD < 256) || ((MVequal(*currMV, prevMB->mvs[0])) && ((int32_t) iMinSAD < prevMB->sad16))) { if (MotionFlags & PMV_QUICKSTOP16) goto EPZS16_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto EPZS16_Terminate_with_Refine; } /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ // previous frame MV CHECK_MV16_CANDIDATE(prevMB->mvs[0].x, prevMB->mvs[0].y); // set threshhold based on Min of Prediction and SAD of collocated block // CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want if ((x == 0) && (y == 0)) { thresh2 = 512; } else { /* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ thresh2 = MIN(psad[0], iSAD) * 6 / 5 + 128; } // MV=(0,0) is often a good choice CHECK_MV16_ZERO; // left neighbour, if allowed if (x != 0) { if (!(MotionFlags & PMV_HALFPEL16)) { pmv[1].x = EVEN(pmv[1].x); pmv[1].y = EVEN(pmv[1].y); } CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); } // top neighbour, if allowed if (y != 0) { if (!(MotionFlags & PMV_HALFPEL16)) { pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); } CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); // top right neighbour, if allowed if ((uint32_t) x != (iWcount - 1)) { if (!(MotionFlags & PMV_HALFPEL16)) { pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); } CHECK_MV16_CANDIDATE(pmv[3].x, pmv[3].y); } } /* Terminate if MinSAD <= T_2 Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] */ if ((iMinSAD <= thresh2) || (MVequal(*currMV, prevMB->mvs[0]) && ((int32_t) iMinSAD <= prevMB->sad16))) { if (MotionFlags & PMV_QUICKSTOP16) goto EPZS16_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto EPZS16_Terminate_with_Refine; } /***** predictor SET C: acceleration MV (new!), neighbours in prev. frame(new!) ****/ backupMV = prevMB->mvs[0]; // collocated MV backupMV.x += (prevMB->mvs[0].x - oldMB->mvs[0].x); // acceleration X backupMV.y += (prevMB->mvs[0].y - oldMB->mvs[0].y); // acceleration Y CHECK_MV16_CANDIDATE(backupMV.x, backupMV.y); // left neighbour if (x != 0) CHECK_MV16_CANDIDATE((prevMB - 1)->mvs[0].x, (prevMB - 1)->mvs[0].y); // top neighbour if (y != 0) CHECK_MV16_CANDIDATE((prevMB - iWcount)->mvs[0].x, (prevMB - iWcount)->mvs[0].y); // right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs if ((uint32_t) x != iWcount - 1) CHECK_MV16_CANDIDATE((prevMB + 1)->mvs[0].x, (prevMB + 1)->mvs[0].y); // bottom neighbour, dito if ((uint32_t) y != iHcount - 1) CHECK_MV16_CANDIDATE((prevMB + iWcount)->mvs[0].x, (prevMB + iWcount)->mvs[0].y); /* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ if (iMinSAD <= thresh2) { if (MotionFlags & PMV_QUICKSTOP16) goto EPZS16_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP16) goto EPZS16_Terminate_with_Refine; } /************ (if Diamond Search) **************/ backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = Square16_MainSearch; else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamond16_MainSearch; else MainSearchPtr = Diamond16_MainSearch; /* 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, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } if (MotionFlags & PMV_EXTSEARCH16) { /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); } if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } 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, 2, iFcode, iQuant, 0); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } } /*************** Choose best MV found **************/ EPZS16_Terminate_with_Refine: 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, iFcode, iQuant, iEdgedWidth); EPZS16_Terminate_without_Refine: *oldMB = *prevMB; currPMV->x = currMV->x - pmv[0].x; currPMV->y = currMV->y - pmv[0].y; return iMinSAD; } int32_t EPZSSearch8(const uint8_t * const pRef, const uint8_t * const pRefH, const uint8_t * const pRefV, const uint8_t * const pRefHV, const IMAGE * const pCur, const int x, const int y, const int start_x, const int start_y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, const MBParam * const pParam, const MACROBLOCK * const pMBs, const MACROBLOCK * const prevMBs, VECTOR * const currMV, VECTOR * const currPMV) { /* Please not that EPZS might not be a good choice for 8x8-block motion search ! */ const uint32_t iWcount = pParam->mb_width; const int32_t iWidth = pParam->width; const int32_t iHeight = pParam->height; const int32_t iEdgedWidth = pParam->edged_width; const uint8_t *cur = pCur->y + x * 8 + y * 8 * iEdgedWidth; int32_t iDiamondSize = 1; int32_t min_dx; int32_t max_dx; int32_t min_dy; int32_t max_dy; VECTOR newMV; VECTOR backupMV; VECTOR pmv[4]; int32_t psad[8]; const int32_t iSubBlock = ((y & 1) << 1) + (x & 1); // const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; const MACROBLOCK *const prevMB = prevMBs + (x >> 1) + (y >> 1) * iWcount; int32_t bPredEq; int32_t iMinSAD, iSAD = 9999; MainSearch8FuncPtr MainSearchPtr; /* Get maximum range */ get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, iFcode); /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ if (!(MotionFlags & PMV_HALFPEL8)) { min_dx = EVEN(min_dx); max_dx = EVEN(max_dx); min_dy = EVEN(min_dy); 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); /* Step 4: Calculate SAD around the Median prediction. MinSAD=SAD If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. If SAD<=256 goto Step 10. */ // Prepare for main loop if (!(MotionFlags & PMV_HALFPEL8)) { currMV->x = EVEN(currMV->x); currMV->y = EVEN(currMV->y); } if (currMV->x > max_dx) currMV->x = max_dx; if (currMV->x < min_dx) currMV->x = min_dx; if (currMV->y > max_dy) currMV->y = max_dy; if (currMV->y < min_dy) currMV->y = min_dy; /***************** This is predictor SET A: only median prediction ******************/ iMinSAD = sad8(cur, 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, (uint8_t) iFcode, iQuant); // thresh1 is fixed to 256 if (iMinSAD < 256 / 4) { if (MotionFlags & PMV_QUICKSTOP8) goto EPZS8_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP8) goto EPZS8_Terminate_with_Refine; } /************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ // MV=(0,0) is often a good choice CHECK_MV8_ZERO; // previous frame MV CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x, prevMB->mvs[iSubBlock].y); // left neighbour, if allowed if (psad[1] != MV_MAX_ERROR) { if (!(MotionFlags & PMV_HALFPEL8)) { pmv[1].x = EVEN(pmv[1].x); pmv[1].y = EVEN(pmv[1].y); } CHECK_MV8_CANDIDATE(pmv[1].x, pmv[1].y); } // top neighbour, if allowed if (psad[2] != MV_MAX_ERROR) { if (!(MotionFlags & PMV_HALFPEL8)) { pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); } CHECK_MV8_CANDIDATE(pmv[2].x, pmv[2].y); // top right neighbour, if allowed if (psad[3] != MV_MAX_ERROR) { if (!(MotionFlags & PMV_HALFPEL8)) { pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); } CHECK_MV8_CANDIDATE(pmv[3].x, pmv[3].y); } } /* // this bias is zero anyway, at the moment! if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) ) // && (iMinSAD <= iQuant * 96) iMinSAD -= MV8_00_BIAS; */ /* Terminate if MinSAD <= T_2 Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] */ if (iMinSAD < 512 / 4) { /* T_2 == 512/4 hardcoded */ if (MotionFlags & PMV_QUICKSTOP8) goto EPZS8_Terminate_without_Refine; if (MotionFlags & PMV_EARLYSTOP8) goto EPZS8_Terminate_with_Refine; } /************ (Diamond Search) **************/ backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ if (!(MotionFlags & PMV_HALFPELDIAMOND8)) iDiamondSize *= 2; /* default: use best prediction as starting point for one call of EPZS_MainSearch */ // there is no EPZS^2 for inter4v at the moment // if (MotionFlags & PMV_USESQUARES8) // MainSearchPtr = Square8_MainSearch; // else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamond8_MainSearch; else MainSearchPtr = Diamond8_MainSearch; iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } if (MotionFlags & PMV_EXTSEARCH8) { /* extended mode: search (up to) two more times: orignal prediction and (0,0) */ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } 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, iQuant, 0); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } } /*************** Choose best MV found **************/ EPZS8_Terminate_with_Refine: 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, iFcode, iQuant, iEdgedWidth); EPZS8_Terminate_without_Refine: currPMV->x = currMV->x - pmv[0].x; currPMV->y = currMV->y - pmv[0].y; return iMinSAD; } int32_t PMVfastIntSearch16(const uint8_t * const pRef, const uint8_t * const pRefH, const uint8_t * const pRefV, const uint8_t * const pRefHV, const IMAGE * const pCur, const int x, const int y, const uint32_t MotionFlags, const uint32_t iQuant, const uint32_t iFcode, const MBParam * const pParam, const MACROBLOCK * const pMBs, const MACROBLOCK * const prevMBs, VECTOR * const currMV, VECTOR * const currPMV) { const uint32_t iWcount = pParam->mb_width; const int32_t iWidth = pParam->width; const int32_t iHeight = pParam->height; const int32_t iEdgedWidth = pParam->edged_width; const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; const VECTOR zeroMV = { 0, 0 }; int32_t iDiamondSize; int32_t min_dx; int32_t max_dx; int32_t min_dy; int32_t max_dy; int32_t iFound; VECTOR newMV; VECTOR backupMV; /* just for PMVFAST */ VECTOR pmv[4]; int32_t psad[4]; MainSearch16FuncPtr MainSearchPtr; const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; MACROBLOCK *const pMB = pMBs + x + y * iWcount; 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, iFcode); /* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ if ((x == 0) && (y == 0)) { threshA = 512; threshB = 1024; bPredEq = 0; psad[0] = psad[1] = psad[2] = psad[3] = 0; *currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; } else { threshA = psad[0]; threshB = threshA + 256; if (threshA < 512) threshA = 512; if (threshA > 1024) threshA = 1024; if (threshB > 1792) threshB = 1792; bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); *currMV = pmv[0]; /* current best := prediction */ } iFound = 0; /* Step 4: Calculate SAD around the Median prediction. MinSAD=SAD If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. If SAD<=256 goto Step 10. */ if (currMV->x > max_dx) { currMV->x = EVEN(max_dx); } if (currMV->x < min_dx) { currMV->x = EVEN(min_dx); } if (currMV->y > max_dy) { currMV->y = EVEN(max_dy); } if (currMV->y < min_dy) { currMV->y = EVEN(min_dy); } iMinSAD = sad16(cur, 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, (uint8_t) iFcode, iQuant); if ((iMinSAD < 256) || ((MVequal(*currMV, prevMB->i_mvs[0])) && ((int32_t) iMinSAD < prevMB->i_sad16))) { if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode { if (!MVzero(*currMV)) { iMinSAD += MV16_00_BIAS; CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures iMinSAD -= MV16_00_BIAS; } } if (MotionFlags & PMV_EARLYSTOP16) goto PMVfastInt16_Terminate_with_Refine; } /* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion vector of the median. If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 */ if ((bPredEq) && (MVequal(pmv[0], prevMB->i_mvs[0]))) iFound = 2; /* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. Otherwise select large Diamond Search. */ if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) iDiamondSize = 2; // halfpel units! else iDiamondSize = 4; // halfpel units! /* Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. Also calculate (0,0) but do not subtract offset. Let MinSAD be the smallest SAD up to this point. If MV is (0,0) subtract offset. */ // (0,0) is often a good choice if (!MVzero(pmv[0])) CHECK_MV16_ZERO; // previous frame MV is always possible if (!MVzero(prevMB->i_mvs[0])) if (!MVequal(prevMB->i_mvs[0], pmv[0])) CHECK_MV16_CANDIDATE(prevMB->i_mvs[0].x, prevMB->i_mvs[0].y); // left neighbour, if allowed if (!MVzero(pmv[1])) if (!MVequal(pmv[1], prevMB->i_mvs[0])) if (!MVequal(pmv[1], pmv[0])) CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); // top neighbour, if allowed if (!MVzero(pmv[2])) if (!MVequal(pmv[2], prevMB->i_mvs[0])) if (!MVequal(pmv[2], pmv[0])) if (!MVequal(pmv[2], pmv[1])) CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); // top right neighbour, if allowed if (!MVzero(pmv[3])) if (!MVequal(pmv[3], prevMB->i_mvs[0])) if (!MVequal(pmv[3], pmv[0])) if (!MVequal(pmv[3], pmv[1])) if (!MVequal(pmv[3], pmv[2])) CHECK_MV16_CANDIDATE(pmv[3].x, pmv[3].y); if ((MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) iMinSAD -= MV16_00_BIAS; /* Step 6: If MinSAD <= thresa goto Step 10. If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. */ if ((iMinSAD <= threshA) || (MVequal(*currMV, prevMB->i_mvs[0]) && ((int32_t) iMinSAD < prevMB->i_sad16))) { if (MotionFlags & PMV_EARLYSTOP16) goto PMVfastInt16_Terminate_with_Refine; } /************ (Diamond Search) **************/ /* Step 7: Perform Diamond search, with either the small or large diamond. If Found=2 only examine one Diamond pattern, and afterwards goto step 10 Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. If center then goto step 10. Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. Refine by using small diamond and goto step 10. */ if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = Square16_MainSearch; else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamond16_MainSearch; else MainSearchPtr = Diamond16_MainSearch; backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ /* 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, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } if (MotionFlags & PMV_EXTSEARCH16) { /* extended: search (up to) two more times: orignal prediction and (0,0) */ if (!(MVequal(pmv[0], backupMV))) { iSAD = (*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } 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, iQuant, iFound); if (iSAD < iMinSAD) { *currMV = newMV; iMinSAD = iSAD; } } } /* Step 10: The motion vector is chosen according to the block corresponding to MinSAD. */ PMVfastInt16_Terminate_with_Refine: pMB->i_mvs[0] = pMB->i_mvs[1] = pMB->i_mvs[2] = pMB->i_mvs[3] = pMB->i_mv16 = *currMV; pMB->i_sad8[0] = pMB->i_sad8[1] = pMB->i_sad8[2] = pMB->i_sad8[3] = pMB->i_sad16 = iMinSAD; 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, 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; return iMinSAD; } /* *********************************************************** bvop motion estimation // TODO: need to incorporate prediction here (eg. sad += calc_delta_16) ***************************************************************/ void MotionEstimationBVOP(MBParam * const pParam, FRAMEINFO * const frame, // 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 uint32_t mb_width = pParam->mb_width; const uint32_t mb_height = pParam->mb_height; const int32_t edged_width = pParam->edged_width; uint32_t i, j; int32_t f_sad16; int32_t b_sad16; int32_t i_sad16; int32_t d_sad16; int32_t best_sad; VECTOR pmv_dontcare; // 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]; 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; } /* force F_SAD16 f_sad16 = 100; b_sad16 = 65535; mb->mode = MODE_FORWARD; mb->mvs[0].x = 1; mb->mvs[0].y = 1; mb->b_mvs[0].x = 1; mb->b_mvs[0].y = 1; continue; ^^ force F_SAD16 */ // 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 // 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 // interpolate search (simple, but effective) i_sad16 = 65535; /* x/y range somewhat buggy i_sad16 = sad16bi_c(frame->image.y + i * 16 + j * 16 * edged_width, get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, i, j, 16, mb->mvs[0].x, mb->mvs[0].y, 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].x, edged_width), edged_width); */ // TODO: direct search // predictor + range of [-32,32] d_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; } } } }
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |