--- gmc.c 2003/08/09 14:41:20 1.1.2.3 +++ gmc.c 2004/03/04 00:47:17 1.1.2.7 @@ -1,23 +1,27 @@ -/************************************************************************** +/***************************************************************************** * - * XVID MPEG-4 VIDEO CODEC - * GMC interpolation module + * XVID MPEG-4 VIDEO CODEC + * - GMC interpolation module - * - * 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. + * Copyright(C) 2002-2003 Pascal Massimino * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * This program is free software ; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation ; either version 2 of the License, or + * (at your option) any later version. * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY ; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. * - *************************************************************************/ + * You should have received a copy of the GNU General Public License + * along with this program ; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + * $Id: gmc.c,v 1.1.2.7 2004/03/04 00:47:17 syskin Exp $ + * + ****************************************************************************/ #include "../portab.h" #include "../global.h" @@ -26,14 +30,13 @@ #include -/* These are mainly the new GMC routines by -Skal- (C) 2003 */ - -////////////////////////////////////////////////////////// -// Pts = 2 or 3 - -// Warning! *src is the global frame pointer (that is: adress -// of pixel 0,0), not the macroblock one. -// Conversely, *dst is the macroblock top-left adress. +/* ************************************************************ + * Pts = 2 or 3 + * + * Warning! *src is the global frame pointer (that is: adress + * of pixel 0,0), not the macroblock one. + * Conversely, *dst is the macroblock top-left adress. + */ void Predict_16x16_C(const NEW_GMC_DATA * const This, uint8_t *dst, const uint8_t *src, @@ -67,12 +70,18 @@ U += dUx; V += dVx; if (u > 0 && u <= W) { ri = MTab[u&15]; Offset = u>>4; } - else if (u > W) Offset = W>>4; - else Offset = -1; - + else { + if (u > W) Offset = W>>4; + else Offset = -1; + ri = 0; + } + if (v > 0 && v <= H) { rj = MTab[v&15]; Offset += (v>>4)*srcstride; } - else if (v > H) Offset += (H>>4)*srcstride; - else Offset -= srcstride; + else { + if (v > H) Offset += (H>>4)*srcstride; + else Offset -= srcstride; + rj = 0; + } f0 = src[Offset + 0]; f0 |= src[Offset + 1] << 16; @@ -132,7 +141,7 @@ if (u>W) Offset = W>>4; else Offset = -1; } - + if (v > 0 && v <= H) { rj = MTab[v&15]; Offset += (v>>4)*srcstride; @@ -157,9 +166,9 @@ f0 |= vSrc[Offset + 1] << 16; f1 = vSrc[Offset + srcstride + 0]; f1 |= vSrc[Offset + srcstride + 1] << 16; - f0 = (ri*f0)>>16; + f0 = (ri*f0)>>16; f1 = (ri*f1) & 0x0fff0000; - f0 |= f1; + f0 |= f1; f0 = (rj*f0 + Rounder) >> 24; vDst[i] = (uint8_t)f0; @@ -181,25 +190,26 @@ int32_t U, V; U = uo; uo += Dsp->dU[1]; V = vo; vo += Dsp->dV[1]; - for (i=16; i>0; --i) + for (i=16; i>0; --i) { int32_t u,v; - u = U >> 16; U += Dsp->dU[0]; vx += u; + u = U >> 16; U += Dsp->dU[0]; vx += u; v = V >> 16; V += Dsp->dV[0]; vy += v; } } - vx -= (256*x+120) << (5+Dsp->accuracy); // 120 = 15*16/2 + vx -= (256*x+120) << (5+Dsp->accuracy); /* 120 = 15*16/2 */ vy -= (256*y+120) << (5+Dsp->accuracy); mv->x = RSHIFT( vx, 8+Dsp->accuracy - qpel ); mv->y = RSHIFT( vy, 8+Dsp->accuracy - qpel ); } -////////////////////////////////////////////////////////// -// simplified version for 1 warp point +/* ************************************************************ + * simplified version for 1 warp point + */ void Predict_1pt_16x16_C(const NEW_GMC_DATA * const This, - uint8_t *Dst, const uint8_t *Src, + uint8_t *Dst, const uint8_t *Src, int dststride, int srcstride, int x, int y, int rounding) { const int W = This->sW; @@ -208,19 +218,25 @@ const int32_t Rounder = ( 128 - (rounding<<(2*rho)) ) << 16; - int32_t uo = This->Uo + (x<<8); // ((16*x)<<4) + int32_t uo = This->Uo + (x<<8); /* ((16*x)<<4) */ int32_t vo = This->Vo + (y<<8); - const uint32_t ri = MTab[uo & 15]; - const uint32_t rj = MTab[vo & 15]; + uint32_t ri = MTab[uo & 15]; + uint32_t rj = MTab[vo & 15]; int i, j; int32_t Offset; - if ((uint32_t)vo<=(uint32_t)H) Offset = (vo>>4)*srcstride; - else if (vo>H) Offset = ( H>>4)*srcstride; - else Offset =-16*srcstride; - if ((uint32_t)uo<=(uint32_t)W) Offset += (uo>>4); - else if (uo>W) Offset += ( W>>4); - else Offset -= 16; + if (vo>=(-16*4) && vo<=H) Offset = (vo>>4)*srcstride; + else { + if (vo>H) Offset = ( H>>4)*srcstride; + else Offset =-16*srcstride; + rj = MTab[0]; + } + if (uo>=(-16*4) && uo<=W) Offset += (uo>>4); + else { + if (uo>W) Offset += (W>>4); + else Offset -= 16; + ri = MTab[0]; + } Dst += 16; @@ -235,13 +251,13 @@ f1 |= Src[ Offset+srcstride +1 ] << 16; f0 = (ri*f0)>>16; f1 = (ri*f1) & 0x0fff0000; - f0 |= f1; + f0 |= f1; f0 = ( rj*f0 + Rounder ) >> 24; Dst[i] = (uint8_t)f0; } Dst += dststride; } -} +} void Predict_1pt_8x8_C(const NEW_GMC_DATA * const This, uint8_t *uDst, const uint8_t *uSrc, @@ -255,17 +271,23 @@ int32_t uo = This->Uco + (x<<7); int32_t vo = This->Vco + (y<<7); - const uint32_t rri = MTab[uo & 15]; - const uint32_t rrj = MTab[vo & 15]; + uint32_t rri = MTab[uo & 15]; + uint32_t rrj = MTab[vo & 15]; int i, j; int32_t Offset; - if ((uint32_t)vo<=(uint32_t)H) Offset = (vo>>4)*srcstride; - else if (vo>H) Offset = ( H>>4)*srcstride; - else Offset =-8*srcstride; - if ((uint32_t)uo<=(uint32_t)W) Offset += (uo>>4); - else if (uo>W) Offset += (W>>4); - else Offset -= 8; + if (vo>=(-8*4) && vo<=H) Offset = (vo>>4)*srcstride; + else { + if (vo>H) Offset = ( H>>4)*srcstride; + else Offset =-8*srcstride; + rrj = MTab[0]; + } + if (uo>=(-8*4) && uo<=W) Offset += (uo>>4); + else { + if (uo>W) Offset += ( W>>4); + else Offset -= 8; + rri = MTab[0]; + } uDst += 8; vDst += 8; @@ -274,17 +296,17 @@ for(i=-8; i<0; ++i, Offset++) { uint32_t f0, f1; - f0 = uSrc[ Offset + 0 ]; + f0 = uSrc[ Offset + 0 ]; f0 |= uSrc[ Offset + 1 ] << 16; f1 = uSrc[ Offset + srcstride + 0 ]; f1 |= uSrc[ Offset + srcstride + 1 ] << 16; f0 = (rri*f0)>>16; - f1 = (rri*f1) & 0x0fff0000; - f0 |= f1; + f1 = (rri*f1) & 0x0fff0000; + f0 |= f1; f0 = ( rrj*f0 + Rounder ) >> 24; - uDst[i] = (uint8_t)f0; + uDst[i] = (uint8_t)f0; - f0 = vSrc[ Offset + 0 ]; + f0 = vSrc[ Offset + 0 ]; f0 |= vSrc[ Offset + 1 ] << 16; f1 = vSrc[ Offset + srcstride + 0 ]; f1 |= vSrc[ Offset + srcstride + 1 ] << 16; @@ -306,10 +328,9 @@ mv->y = RSHIFT(Dsp->Vo<accuracy = accuracy; gmc->num_wp = nb_pts; - // reduce the number of points, if possible + /* reduce the number of points, if possible */ if (nb_pts<3 || (pts->duv[2].x==-pts->duv[1].y && pts->duv[2].y==pts->duv[1].x)) { if (nb_pts<2 || (pts->duv[1].x==0 && pts->duv[1].y==0)) { if (nb_pts<1 || (pts->duv[0].x==0 && pts->duv[0].y==0)) { @@ -332,19 +353,19 @@ else nb_pts = 2; } else nb_pts = 3; - - // now, nb_pts stores the actual number of points required for interpolation + + /* now, nb_pts stores the actual number of points required for interpolation */ if (nb_pts<=1) { if (nb_pts==1) { - // store as 4b fixed point + /* store as 4b fixed point */ gmc->Uo = pts->duv[0].x << accuracy; gmc->Vo = pts->duv[0].y << accuracy; - gmc->Uco = ((pts->duv[0].x>>1) | (pts->duv[0].x&1)) << accuracy; // DIV2RND() - gmc->Vco = ((pts->duv[0].y>>1) | (pts->duv[0].y&1)) << accuracy; // DIV2RND() + gmc->Uco = ((pts->duv[0].x>>1) | (pts->duv[0].x&1)) << accuracy; /* DIV2RND() */ + gmc->Vco = ((pts->duv[0].y>>1) | (pts->duv[0].y&1)) << accuracy; /* DIV2RND() */ } - else { // zero points?! + else { /* zero points?! */ gmc->Uo = gmc->Vo = 0; gmc->Uco = gmc->Vco = 0; } @@ -353,26 +374,28 @@ gmc->predict_8x8 = Predict_1pt_8x8_C; gmc->get_average_mv = get_average_mv_1pt_C; } - else { // 2 or 3 points - const int rho = 3 - accuracy; // = {3,2,1,0} for Acc={0,1,2,3} + else { /* 2 or 3 points */ + const int rho = 3 - accuracy; /* = {3,2,1,0} for Acc={0,1,2,3} */ int Alpha = log2bin(width-1); int Ws = 1 << Alpha; - gmc->dU[0] = 16*Ws + RDIV( 8*Ws*pts->duv[1].x, width ); // dU/dx - gmc->dV[0] = RDIV( 8*Ws*pts->duv[1].y, width ); // dV/dx + gmc->dU[0] = 16*Ws + RDIV( 8*Ws*pts->duv[1].x, width ); /* dU/dx */ + gmc->dV[0] = RDIV( 8*Ws*pts->duv[1].y, width ); /* dV/dx */ /* disabled, because possibly buggy? */ -/* if (nb_pts==2) { - gmc->dU[1] = -gmc->dV[0]; // -Sin - gmc->dV[1] = gmc->dU[0] ; // Cos +#if 0 + if (nb_pts==2) { + gmc->dU[1] = -gmc->dV[0]; /* -Sin */ + gmc->dV[1] = gmc->dU[0] ; /* Cos */ } - else */ + else +#endif { const int Beta = log2bin(height-1); const int Hs = 1<dU[1] = RDIV( 8*Hs*pts->duv[2].x, height ); // dU/dy - gmc->dV[1] = 16*Hs + RDIV( 8*Hs*pts->duv[2].y, height ); // dV/dy + gmc->dU[1] = RDIV( 8*Hs*pts->duv[2].x, height ); /* dU/dy */ + gmc->dV[1] = 16*Hs + RDIV( 8*Hs*pts->duv[2].y, height ); /* dV/dy */ if (Beta>Alpha) { gmc->dU[0] <<= (Beta-Alpha); gmc->dV[0] <<= (Beta-Alpha); @@ -384,7 +407,7 @@ gmc->dV[1] <<= Alpha - Beta; } } - // upscale to 16b fixed-point + /* upscale to 16b fixed-point */ gmc->dU[0] <<= (16-Alpha - rho); gmc->dU[1] <<= (16-Alpha - rho); gmc->dV[0] <<= (16-Alpha - rho); @@ -403,24 +426,23 @@ } } -////////////////////////////////////////////////////////// - -/* quick and dirty routine to generate the full warped image (pGMC != NULL) - or just all average Motion Vectors (pGMC == NULL) */ +/* ******************************************************************* + * quick and dirty routine to generate the full warped image + * (pGMC != NULL) or just all average Motion Vectors (pGMC == NULL) */ void -generate_GMCimage( const NEW_GMC_DATA *const gmc_data, // [input] precalculated data - const IMAGE *const pRef, // [input] +generate_GMCimage( const NEW_GMC_DATA *const gmc_data, /* [input] precalculated data */ + const IMAGE *const pRef, /* [input] */ const int mb_width, const int mb_height, const int stride, const int stride2, - const int fcode, // [input] some parameters... - const int32_t quarterpel, // [input] for rounding avgMV - const int reduced_resolution, // [input] ignored - const int32_t rounding, // [input] for rounding image data - MACROBLOCK *const pMBs, // [output] average motion vectors - IMAGE *const pGMC) // [output] full warped image + const int fcode, /* [input] some parameters... */ + const int32_t quarterpel, /* [input] for rounding avgMV */ + const int reduced_resolution, /* [input] ignored */ + const int32_t rounding, /* [input] for rounding image data */ + MACROBLOCK *const pMBs, /* [output] average motion vectors */ + IMAGE *const pGMC) /* [output] full warped image */ { unsigned int mj,mi; @@ -431,8 +453,8 @@ const int mbnum = mj*mb_width+mi; if (pGMC) { - gmc_data->predict_16x16(gmc_data, - pGMC->y + mj*16*stride + mi*16, pRef->y, + gmc_data->predict_16x16(gmc_data, + pGMC->y + mj*16*stride + mi*16, pRef->y, stride, stride, mi, mj, rounding); gmc_data->predict_8x8(gmc_data,