--- trunk/xvidcore/src/motion/motion_est.c 2002/07/21 23:34:08 326 +++ trunk/xvidcore/src/motion/motion_est.c 2002/07/28 02:55:41 346 @@ -458,6 +458,7 @@ /* Do a diamond search around given starting point, return SAD of best */ int32_t iDirection = 0; + int32_t iDirectionBackup; int32_t iSAD; VECTOR backupMV; @@ -471,23 +472,25 @@ 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 = start_x; currMV->y = start_y; @@ -551,7 +554,7 @@ backupMV.y + iDiamondSize, 8); - if (iDirection) + if (iDirection) { while (!iFound) { iFound = 1; backupMV = *currMV; @@ -560,62 +563,62 @@ 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; @@ -623,48 +626,49 @@ 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 = start_x; currMV->y = start_y; @@ -870,78 +874,69 @@ return iMinSAD; } - -#define CHECK_MV16_F_INTERPOL(X,Y,BX,BY) { \ - 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) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ - if (iSAD < iMinSAD) \ - { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ -} - -#define CHECK_MV16_F_INTERPOL_DIR(X,Y,BX,BY,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) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ - if (iSAD < iMinSAD) \ - { iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ -} - -#define CHECK_MV16_F_INTERPOL_FOUND(X,Y,BX,BY,D) { \ - if ( ((X) <= max_dx) && ((X) >= min_dx) \ - && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ +#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 = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ + 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; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ + { iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); } } \ } - -#define CHECK_MV16_B_INTERPOL(FX,FY,X,Y) { \ - if ( ((X) <= max_dx) && ((X) >= min_dx) \ - && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ +#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 = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ + 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; currMV->x=(X); currMV->y=(Y); } } \ + { iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); iFound=0;} } \ } - -#define CHECK_MV16_B_INTERPOL_DIR(FX,FY,X,Y,D) { \ - if ( ((X) <= max_dx) && ((X) >= min_dx) \ - && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ +#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 = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ + 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; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ + { iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); } } \ } - -#define CHECK_MV16_B_INTERPOL_FOUND(FX,FY,X,Y,D) { \ - if ( ((X) <= max_dx) && ((X) >= min_dx) \ - && ((Y) <= max_dy) && ((Y) >= min_dy) ) \ +#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 = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ - iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ + 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; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ + { iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); iFound=0;} } \ } - -#if (0==1) 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, @@ -966,64 +961,163 @@ const int b_center_x, const int b_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 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 f_iFcode, - const int32_t b_iFcode, + const int32_t iEdgedWidth, + const int32_t iDiamondSize, + + const int32_t f_iFcode, + const int32_t b_iFcode, - const int32_t iQuant, - int iFound) + const int32_t iQuant, + int iFound) { /* Do a diamond search around given starting point, return SAD of best */ - int32_t f_iDirection = 0; - int32_t b_iDirection = 0; int32_t iSAD; VECTOR f_backupMV; VECTOR b_backupMV; - f_backupMV.x = start_x; - f_backupMV.y = start_y; - b_backupMV.x = start_x; - b_backupMV.y = start_y; + 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 */ - 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); + 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); - if (iDirection) - while (!iFound) { - iFound = 1; - backupMV = *currMV; + } while (!iFound); - 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 = start_x; - currMV->y = start_y; - } return iMinSAD; } -#endif int32_t @@ -1274,9 +1368,9 @@ const IMAGE * const pCur, const int x, const int y, - const int start_x, - const int start_y, - const int center_x, + 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, @@ -1613,6 +1707,7 @@ /* Do a diamond search around given starting point, return SAD of best */ int32_t iDirection = 0; + int32_t iDirectionBackup; int32_t iSAD; VECTOR backupMV; @@ -1626,23 +1721,206 @@ 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 = start_x; currMV->y = start_y; @@ -1650,6 +1928,10 @@ return iMinSAD; } + + + + int32_t Halfpel8_Refine_c(const uint8_t * const pRef, const uint8_t * const pRefH, @@ -1788,9 +2070,9 @@ // 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; @@ -2468,9 +2750,9 @@ // 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; @@ -2545,10 +2827,10 @@ 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 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, @@ -2576,7 +2858,7 @@ int32_t iFound; VECTOR newMV; - VECTOR backupMV; /* just for PMVFAST */ + VECTOR backupMV; VECTOR pmv[4]; int32_t psad[4]; @@ -2831,13 +3113,8 @@ /* *********************************************************** bvop motion estimation -// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) ***************************************************************/ - -#define DIRECT_PENALTY 0 -#define DIRECT_UPPERLIMIT 256 // never use direct mode if SAD is larger than this - void MotionEstimationBVOP(MBParam * const pParam, FRAMEINFO * const frame, @@ -2860,6 +3137,9 @@ const int mb_height = pParam->mb_height; const int edged_width = pParam->edged_width; + const int32_t iWidth = pParam->width; + const int32_t iHeight = pParam->height; + int i, j, k; static const VECTOR zeroMV={0,0}; @@ -2867,20 +3147,23 @@ 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 dnv_sad16; /* direct mode (assume linear motion) without correction vector */ + int d_sad16; /* direct mode (assume almost linear motion) */ int best_sad; - VECTOR f_predMV, b_predMV; /* there is no direct prediction */ + VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ + VECTOR f_interpolMV, b_interpolMV; VECTOR pmv_dontcare; + int min_dx, max_dx, min_dy, max_dy; + int f_min_dx, f_max_dx, f_min_dy, f_max_dy; + int b_min_dx, b_max_dx, b_min_dy, b_max_dy; + 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; @@ -2896,11 +3179,11 @@ 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; + mb->deltamv=zeroMV; -/* special case, if collocated block is SKIPed: encoding is forward(0,0) */ +/* special case, if collocated block is SKIPed: encoding is forward (0,0), cpb=0 without further ado */ +#ifndef _DISABLE_SKIP 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; @@ -2910,27 +3193,27 @@ mb->b_mvs[0].y = 0; continue; } - - dnv_sad16 = DIRECT_PENALTY; +#endif if (b_mb->mode == MODE_INTER4V) { - + d_sad16 = 0; /* same method of scaling as in decoder.c, so we copy from there */ for (k = 0; k < 4; k++) { - directMV = b_mb->mvs[k]; + mb->directmv[k] = 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); + mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); + mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) + ? ((TRB - TRD) * mb->directmv[k].x) / TRD + : mb->mvs[k].x - mb->directmv[k].x); + + mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); + mb->b_mvs[k].y = (int32_t) ((mb->deltamv.y == 0) + ? ((TRB - TRD) * mb->directmv[k].y) / TRD + : mb->mvs[k].y - mb->directmv[k].y); - dnv_sad16 += + d_sad16 += sad8bi(frame->image.y + 2*(i+(k&1))*8 + 2*(j+(k>>1))*8*edged_width, get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, 2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->mvs[k], edged_width), @@ -2941,34 +3224,35 @@ } else { - directMV = b_mb->mvs[0]; + mb->directmv[3] = mb->directmv[2] = mb->directmv[1] = + mb->directmv[0] = b_mb->mvs[0]; - mb->mvs[0].x = (int32_t) ((TRB * directMV.x) / TRD + deltaMV.x); - mb->b_mvs[0].x = (int32_t) ((deltaMV.x == 0) - ? ((TRB - TRD) * directMV.x) / TRD - : mb->mvs[0].x - directMV.x); - mb->mvs[0].y = (int32_t) ((TRB * directMV.y) / TRD + deltaMV.y); - mb->b_mvs[0].y = (int32_t) ((deltaMV.y == 0) - ? ((TRB - TRD) * directMV.y) / TRD - : mb->mvs[0].y - directMV.y); + mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); + mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) + ? ((TRB - TRD) * mb->directmv[0].x) / TRD + : mb->mvs[0].x - mb->directmv[0].x); + + mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); + mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) + ? ((TRB - TRD) * mb->directmv[0].y) / TRD + : mb->mvs[0].y - mb->directmv[0].y); - dnv_sad16 = DIRECT_PENALTY + - sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, + 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); - } + d_sad16 += calc_delta_16(mb->deltamv.x, mb->deltamv.y, 1, frame->quant); // forward search f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, &frame->image, i, j, mb->mvs[0].x, mb->mvs[0].y, /* start point f_directMV */ f_predMV.x, f_predMV.y, /* center is f-prediction */ - frame->motion_flags & (~(PMV_EARLYSTOP16|PMV_QUICKSTOP16)), + frame->motion_flags, frame->quant, frame->fcode, pParam, f_mbs, f_mbs, &mb->mvs[0], &pmv_dontcare); @@ -2979,7 +3263,7 @@ &frame->image, i, j, mb->b_mvs[0].x, mb->b_mvs[0].y, /* start point b_directMV */ b_predMV.x, b_predMV.y, /* center is b-prediction */ - frame->motion_flags & (~(PMV_EARLYSTOP16|PMV_QUICKSTOP16)), + frame->motion_flags, frame->quant, frame->bcode, pParam, b_mbs, b_mbs, &mb->b_mvs[0], &pmv_dontcare); @@ -2993,15 +3277,60 @@ edged_width); i_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, frame->fcode, frame->quant); - i_sad16 += calc_delta_16(mb->b_mvs[0].y-b_predMV.y, mb->b_mvs[0].y-b_predMV.y, + i_sad16 += calc_delta_16(mb->b_mvs[0].x-b_predMV.x, mb->b_mvs[0].y-b_predMV.y, frame->bcode, frame->quant); - // TODO: direct search - // predictor + delta vector in range [-32,32] (fcode=1) - -// i_sad16 = 65535; + get_range(&f_min_dx, &f_max_dx, &f_min_dy, &f_max_dy, i, j, 16, iWidth, iHeight, + frame->fcode); + get_range(&b_min_dx, &b_max_dx, &b_min_dy, &b_max_dy, i, j, 16, iWidth, iHeight, + frame->bcode); + +/* Interpolated MC motion vector search, this is tedious and more complicated because there are + two values for everything, always one for backward and one for forward ME. Still, we don't gain + much from this search, maybe it should simply be skipped and simply current i_sad16 value used + as "optimal". */ + + i_sad16 = Diamond16_InterpolMainSearch( + f_ref->y, f_refH->y, f_refV->y, f_refHV->y, + frame->image.y + i * 16 + j * 16 * edged_width, + b_ref->y, b_refH->y, b_refV->y, b_refHV->y, + i, j, + mb->mvs[0].x, mb->mvs[0].y, + mb->b_mvs[0].x, mb->b_mvs[0].y, + i_sad16, + &f_interpolMV, &b_interpolMV, + f_predMV.x, f_predMV.y, b_predMV.x, b_predMV.y, + f_min_dx, f_max_dx, f_min_dy, f_max_dy, + b_min_dx, b_max_dx, b_min_dy, b_max_dy, + edged_width, 1, + frame->fcode, frame->bcode,frame->quant,0); + + +/* DIRECT MODE DELTA VECTOR SEARCH. + This has to be made more effective, but at the moment I'm happy it's running at all */ + +/* range is taken without fcode restriction, just a hack instead of writing down the dimensions, of course */ + + get_range(&min_dx, &max_dx, &min_dy, &max_dy, i, j, 16, iWidth, iHeight, 19); + + d_sad16 = Diamond16_DirectMainSearch( + f_ref->y, f_refH->y, f_refV->y, f_refHV->y, + frame->image.y + i*16 + j*16*edged_width, + b_ref->y, b_refH->y, b_refV->y, b_refHV->y, + i, j, + TRB,TRD, + 0,0, + d_sad16, + &mb->deltamv, + mb->directmv, // this has to be pre-initialized with b_mb->mvs[} + min_dx, max_dx, min_dy, max_dy, + edged_width, 1, frame->quant, 0); + + +// i_sad16 = 65535; /* remove the comment to disable any of the MODEs */ // f_sad16 = 65535; // b_sad16 = 65535; +// d_sad16 = 65535; if (f_sad16 < b_sad16) { best_sad = f_sad16; @@ -3016,56 +3345,47 @@ mb->mode = MODE_INTERPOLATE; } - if (dnv_sad16 < best_sad) { + if (d_sad16 < best_sad) { - if (dnv_sad16 > DIRECT_UPPERLIMIT) - { - /* if SAD value is too large, try same vector with MODE_INTERPOLATE - instead (interpolate has residue encoding, direct mode without MV - doesn't) - - This has to be replaced later by "real" direct mode, including delta - vector and (if needed) residue encoding - - */ - - directMV = b_mb->mvs[0]; - - mb->mvs[0].x = (int32_t) ((TRB * directMV.x) / TRD + deltaMV.x); - mb->b_mvs[0].x = (int32_t) ((deltaMV.x == 0) - ? ((TRB - TRD) * directMV.x) / TRD - : mb->mvs[0].x - directMV.x); - mb->mvs[0].y = (int32_t) ((TRB * directMV.y) / TRD + deltaMV.y); - mb->b_mvs[0].y = (int32_t) ((deltaMV.y == 0) - ? ((TRB - TRD) * directMV.y) / TRD - : mb->mvs[0].y - directMV.y); + if (b_mb->mode == MODE_INTER4V) + { - dnv_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); - dnv_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, - frame->fcode, frame->quant); - dnv_sad16 += calc_delta_16(mb->b_mvs[0].y-b_predMV.y, mb->b_mvs[0].y-b_predMV.y, - frame->bcode, frame->quant); + /* how to calc vectors is defined in standard. mvs[] and b_mvs[] are only for motion compensation */ + /* for the bitstream, the value mb->deltamv is read directly */ + + for (k = 0; k < 4; k++) { - if (dnv_sad16 < best_sad) - { - best_sad = dnv_sad16; - mb->mode = MODE_INTERPOLATE; - -/* fprintf(stderr,"f_sad16 = %d, b_sad16 = %d, i_sad16 = %d, dnv_sad16 = %d\n", - f_sad16,b_sad16,i_sad16,dnv_sad16); -*/ } + mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); + mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) + ? ((TRB - TRD) * mb->directmv[k].x) / TRD + : mb->mvs[k].x - mb->directmv[k].x); + + mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); + mb->b_mvs[k].y = (int32_t) ((mb->deltamv.y == 0) + ? ((TRB - TRD) * mb->directmv[k].y) / TRD + : mb->mvs[k].y - mb->directmv[k].y); + } } else { - best_sad = dnv_sad16; - mb->mode = MODE_DIRECT_NONE_MV; - } + mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); + + mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) + ? ((TRB - TRD) * mb->directmv[0].x) / TRD + : mb->mvs[0].x - mb->directmv[0].x); + + mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); + + mb->b_mvs[0].y = (int32_t) ((mb->deltamv.y == 0) + ? ((TRB - TRD) * mb->directmv[0].y) / TRD + : mb->mvs[0].y - mb->directmv[0].y); + + mb->mvs[3] = mb->mvs[2] = mb->mvs[1] = mb->mvs[0]; + mb->b_mvs[3] = mb->b_mvs[2] = mb->b_mvs[1] = mb->b_mvs[0]; + } + + best_sad = d_sad16; + mb->mode = MODE_DIRECT; } switch (mb->mode) @@ -3081,23 +3401,24 @@ break; case MODE_INTERPOLATE: i_count++; + mb->mvs[0] = f_interpolMV; + mb->b_mvs[0] = b_interpolMV; f_predMV = mb->mvs[0]; b_predMV = mb->b_mvs[0]; break; case MODE_DIRECT: - d_count++; break; - case MODE_DIRECT_NONE_MV: - dnv_count++; break; + d_count++; + break; default: - s_count++; break; + 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); + fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D: %04d\n", + f_count,b_count,i_count,d_count); #endif }