estimation_bvop.c

从FFMPEG转换而来的H264解码程序,VC下编译..

/*****************************************************************************
 *
 *  XVID MPEG-4 VIDEO CODEC
 *  - Motion Estimation for B-VOPs  -
 *
 *  Copyright(C) 2002 Christoph Lampert <[e-mail]>
 *               2002 Michael Militzer <[e-mail]>
 *               2002-2003 Radoslaw Czyz <[e-mail]>
 *
 *  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., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 *
 * $Id: estimation_bvop.c,v 1.25 2006/02/25 01:20:41 syskin Exp $
 *
 ****************************************************************************/


#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>	/* memcpy */

#include "../encoder.h"
#include "../global.h"
#include "../image/interpolate8x8.h"
#include "estimation.h"
#include "motion.h"
#include "sad.h"
#include "motion_inlines.h"

static int32_t
ChromaSAD2(const int fx, const int fy, const int bx, const int by,
			SearchData * const data)
{
	int sad;
	const uint32_t stride = data->iEdgedWidth/2;
	uint8_t *f_refu, *f_refv, *b_refu, *b_refv;
	int offset, filter;

	const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = {
		NULL,
		interpolate8x8_halfpel_v,
		interpolate8x8_halfpel_h,
		interpolate8x8_halfpel_hv
	};
	
	if (data->chromaX == fx && data->chromaY == fy && 
		data->b_chromaX == bx && data->b_chromaY == by) 
		return data->chromaSAD;

	offset = (fx>>1) + (fy>>1)*stride;
	filter = ((fx & 1) << 1) | (fy & 1);

	if (filter != 0) {
		f_refu = data->RefQ + 64;
		f_refv = data->RefQ + 64 + 8;
		if (data->chromaX != fx || data->chromaY != fy) {
			interpolate8x8_halfpel[filter](f_refu, data->RefP[4] + offset, stride, data->rounding);
			interpolate8x8_halfpel[filter](f_refv, data->RefP[5] + offset, stride, data->rounding);
		}
	} else {
		f_refu = (uint8_t*)data->RefP[4] + offset;
		f_refv = (uint8_t*)data->RefP[5] + offset;
	}
	data->chromaX = fx; data->chromaY = fy;

	offset = (bx>>1) + (by>>1)*stride;
	filter = ((bx & 1) << 1) | (by & 1);

	if (filter != 0) {
		b_refu = data->RefQ + 64 + 16;
		b_refv = data->RefQ + 64 + 24;
		if (data->b_chromaX != bx || data->b_chromaY != by) {
			interpolate8x8_halfpel[filter](b_refu, data->b_RefP[4] + offset, stride, data->rounding);
			interpolate8x8_halfpel[filter](b_refv, data->b_RefP[5] + offset, stride, data->rounding);
		}
	} else {
		b_refu = (uint8_t*)data->b_RefP[4] + offset;
		b_refv = (uint8_t*)data->b_RefP[5] + offset;
	}
	data->b_chromaX = bx; data->b_chromaY = by;

	sad = sad8bi(data->CurU, b_refu, f_refu, stride);
	sad += sad8bi(data->CurV, b_refv, f_refv, stride);

	data->chromaSAD = sad;
	return sad;
}

static void
CheckCandidateInt(const int x, const int y, SearchData * const data, const unsigned int Direction)
{
	int32_t sad, xf, yf, xb, yb, xcf, ycf, xcb, ycb;
	uint32_t t;

	const uint8_t *ReferenceF, *ReferenceB;
	VECTOR *current;

	if ((x > data->max_dx) || (x < data->min_dx) ||
		(y > data->max_dy) || (y < data->min_dy))
		return;

	if (Direction == 1) { /* x and y mean forward vector */
		VECTOR backward = data->qpel_precision ? data->currentQMV[1] : data->currentMV[1];
		xb = backward.x;
		yb = backward.y;
		xf = x; yf = y;
	} else { /* x and y mean backward vector */
		VECTOR forward = data->qpel_precision ? data->currentQMV[0] : data->currentMV[0];
		xf = forward.x;
		yf = forward.y;
		xb = x; yb = y;
	}

	if (!data->qpel_precision) {
		ReferenceF = GetReference(xf, yf, data);
		ReferenceB = GetReferenceB(xb, yb, 1, data);
		current = data->currentMV + Direction - 1;
		xcf = xf; ycf = yf;
		xcb = xb; ycb = yb;
	} else {
		ReferenceF = xvid_me_interpolate16x16qpel(xf, yf, 0, data);
		current = data->currentQMV + Direction - 1;
		ReferenceB = xvid_me_interpolate16x16qpel(xb, yb, 1, data);
		xcf = xf/2; ycf = yf/2;
		xcb = xb/2; ycb = yb/2;
	}

	t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision)
		 + d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision);

	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
	sad += (data->lambda16 * t);

	if (data->chroma && sad < *data->iMinSAD)
		sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3],
							(ycf >> 1) + roundtab_79[ycf & 0x3],
							(xcb >> 1) + roundtab_79[xcb & 0x3],
							(ycb >> 1) + roundtab_79[ycb & 0x3], data);

	if (sad < *(data->iMinSAD)) {
		*data->iMinSAD = sad;
		current->x = x; current->y = y;
		data->dir = Direction;
	}
}

static void
CheckCandidateDirect(const int x, const int y, SearchData * const data, const unsigned int Direction)
{
	int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0;
	uint32_t k;
	const uint8_t *ReferenceF;
	const uint8_t *ReferenceB;
	VECTOR mvs, b_mvs;
	const int blocks[4] = {0, 8, 8*data->iEdgedWidth, 8*data->iEdgedWidth+8};

	if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;

	for (k = 0; k < 4; k++) {
		mvs.x = data->directmvF[k].x + x;
		b_mvs.x = ((x == 0) ?
			data->directmvB[k].x
			: mvs.x - data->referencemv[k].x);

		mvs.y = data->directmvF[k].y + y;
		b_mvs.y = ((y == 0) ?
			data->directmvB[k].y
			: mvs.y - data->referencemv[k].y);

		if ((mvs.x > data->max_dx)   || (mvs.x < data->min_dx)   ||
			(mvs.y > data->max_dy)   || (mvs.y < data->min_dy)   ||
			(b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) ||
			(b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) )
			return;

		if (data->qpel) {
			xcf += mvs.x/2; ycf += mvs.y/2;
			xcb += b_mvs.x/2; ycb += b_mvs.y/2;
			if (data->qpel_precision) {
				ReferenceF = xvid_me_interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
				ReferenceB = xvid_me_interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
				goto done;
			}
			mvs.x >>=1; mvs.y >>=1; b_mvs.x >>=1; b_mvs.y >>=1; // qpel->hpel
		} else {
			xcf += mvs.x; ycf += mvs.y;
			xcb += b_mvs.x; ycb += b_mvs.y;
		}
		ReferenceF = GetReference(mvs.x, mvs.y, data) + blocks[k];
		ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data) + blocks[k];
done:
		sad += data->iMinSAD[k+1] = 
			sad8bi(data->Cur + blocks[k],
					ReferenceF, ReferenceB, data->iEdgedWidth);
		if (sad > *(data->iMinSAD)) return;
	}

	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0));

	if (data->chroma && sad < *data->iMinSAD)
		sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
							(ycf >> 3) + roundtab_76[ycf & 0xf],
							(xcb >> 3) + roundtab_76[xcb & 0xf],
							(ycb >> 3) + roundtab_76[ycb & 0xf], data);

	if (sad < *(data->iMinSAD)) {
		data->iMinSAD[0] = sad;
		data->currentMV->x = x; data->currentMV->y = y;
		data->dir = Direction;
	}
}

static void
CheckCandidateDirectno4v(const int x, const int y, SearchData * const data, const unsigned int Direction)
{
	int32_t sad, xcf, ycf, xcb, ycb;
	const uint8_t *ReferenceF;
	const uint8_t *ReferenceB;
	VECTOR mvs, b_mvs;

	if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;

	mvs.x = data->directmvF[0].x + x;
	b_mvs.x = ((x == 0) ?
		data->directmvB[0].x
		: mvs.x - data->referencemv[0].x);

	mvs.y = data->directmvF[0].y + y;
	b_mvs.y = ((y == 0) ?
		data->directmvB[0].y
		: mvs.y - data->referencemv[0].y);

	if ( (mvs.x > data->max_dx) || (mvs.x < data->min_dx)
		|| (mvs.y > data->max_dy) || (mvs.y < data->min_dy)
		|| (b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx)
		|| (b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) ) return;

	if (data->qpel) {
		xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2);
		xcb = 4*(b_mvs.x/2); ycb = 4*(b_mvs.y/2);
		if (data->qpel_precision) {
			ReferenceF = xvid_me_interpolate16x16qpel(mvs.x, mvs.y, 0, data);
			ReferenceB = xvid_me_interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data);
			goto done;
		}
		mvs.x >>=1; mvs.y >>=1; b_mvs.x >>=1; b_mvs.y >>=1; // qpel->hpel
	} else {
		xcf = 4*mvs.x; ycf = 4*mvs.y;
		xcb = 4*b_mvs.x; ycb = 4*b_mvs.y;
	}
	ReferenceF = GetReference(mvs.x, mvs.y, data);
	ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data);

done:
	sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
	sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0));

	if (data->chroma && sad < *data->iMinSAD)
		sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf],
							(ycf >> 3) + roundtab_76[ycf & 0xf],
							(xcb >> 3) + roundtab_76[xcb & 0xf],
							(ycb >> 3) + roundtab_76[ycb & 0xf], data);

	if (sad < *(data->iMinSAD)) {
		*(data->iMinSAD) = sad;
		data->currentMV->x = x; data->currentMV->y = y;
		data->dir = Direction;
	}
}

void
CheckCandidate16no4v(const int x, const int y, SearchData * const data, const unsigned int Direction)
{
	int32_t sad, xc, yc;
	const uint8_t * Reference;
	uint32_t t;
	VECTOR * current;

	if ( (x > data->max_dx) || ( x < data->min_dx)
		|| (y > data->max_dy) || (y < data->min_dy) ) return;

	if (data->qpel_precision) { /* x and y are in 1/4 precision */
		Reference = xvid_me_interpolate16x16qpel(x, y, 0, data);
		current = data->currentQMV;
		xc = x/2; yc = y/2;
	} else {
		Reference = GetReference(x, y, data);
		current = data->currentMV;
		xc = x; yc = y;
	}
	t = d_mv_bits(x, y, data->predMV, data->iFcode,
					data->qpel^data->qpel_precision);

	sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
	sad += (data->lambda16 * t);

	if (data->chroma && sad < *data->iMinSAD)
		sad += xvid_me_ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
								(yc >> 1) + roundtab_79[yc & 0x3], data);

	if (sad < *(data->iMinSAD)) {
		*(data->iMinSAD) = sad;
		current->x = x; current->y = y;
		data->dir = Direction;
	}
}


static void 
initialize_searchData(SearchData * Data_d,
					  SearchData * Data_f,
					  SearchData * Data_b,
					  SearchData * Data_i,
					  int x, int y,
					  const IMAGE * const f_Ref,
					  const uint8_t * const f_RefH,
					  const uint8_t * const f_RefV,
					  const uint8_t * const f_RefHV,
					  const IMAGE * const b_Ref,
					  const uint8_t * const b_RefH,
					  const uint8_t * const b_RefV,
					  const uint8_t * const b_RefHV,
					  const IMAGE * const pCur,
					  const MACROBLOCK * const b_mb)
{

	/* per-macroblock SearchData initialization - too many things would be repeated 4 times */
	const uint8_t * RefP[6], * b_RefP[6], * Cur[3];
	const uint32_t iEdgedWidth = Data_d->iEdgedWidth;
	unsigned int lambda;
	int i;

	/* luma */
	int offset = (x + iEdgedWidth*y) * 16;
	RefP[0] = f_Ref->y + offset;
	RefP[2] = f_RefH + offset;
	RefP[1] = f_RefV + offset;
	RefP[3] = f_RefHV + offset;
	b_RefP[0] = b_Ref->y + offset;
	b_RefP[2] = b_RefH + offset;
	b_RefP[1] = b_RefV + offset;
	b_RefP[3] = b_RefHV + offset;
	Cur[0] = pCur->y + offset;

	/* chroma */
	offset = (x + (iEdgedWidth/2)*y) * 8;
	RefP[4] = f_Ref->u + offset;
	RefP[5] = f_Ref->v + offset;
	b_RefP[4] = b_Ref->u + offset;
	b_RefP[5] = b_Ref->v + offset;
	Cur[1] = pCur->u + offset;
	Cur[2] = pCur->v + offset;

	lambda = xvid_me_lambda_vec16[b_mb->quant];

	for (i = 0; i < 6; i++) {
		Data_d->RefP[i] = Data_f->RefP[i] = Data_i->RefP[i] = RefP[i];
		Data_d->b_RefP[i] = Data_b->RefP[i] = Data_i->b_RefP[i] = b_RefP[i];
	}
	Data_d->Cur = Data_f->Cur = Data_b->Cur = Data_i->Cur = Cur[0];
	Data_d->CurU = Data_f->CurU = Data_b->CurU = Data_i->CurU = Cur[1];
	Data_d->CurV = Data_f->CurV = Data_b->CurV = Data_i->CurV = Cur[2];

	Data_d->lambda16 = Data_f->lambda16 = Data_b->lambda16 = Data_i->lambda16 = lambda;

	/* reset chroma-sad cache */
	Data_d->b_chromaX = Data_d->b_chromaY = Data_d->chromaX = Data_d->chromaY = Data_d->chromaSAD = 256*4096; 
	Data_i->b_chromaX = Data_i->b_chromaY = Data_i->chromaX = Data_i->chromaY = Data_i->chromaSAD = 256*4096; 
	Data_f->chromaX = Data_f->chromaY = Data_f->chromaSAD = 256*4096; 
	Data_b->chromaX = Data_b->chromaY = Data_b->chromaSAD = 256*4096; 

	*Data_d->iMinSAD = *Data_b->iMinSAD = *Data_f->iMinSAD = *Data_i->iMinSAD = 4096*256;
}

static __inline VECTOR
ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
{
/* the stupidiest function ever */
	return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]);
}

static void __inline
PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
							const uint32_t iWcount,
							const MACROBLOCK * const pMB,
							const uint32_t mode_curr,
							const VECTOR hint)
{
	/* [0] is prediction */
	/* [1] is zero */
	pmv[1].x = pmv[1].y = 0; 

	pmv[2].x = hint.x; pmv[2].y = hint.y;

	if ((y != 0)&&(x != (int)(iWcount+1))) {			/* [3] top-right neighbour */
		pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
	} else pmv[3].x = pmv[3].y = 0;

	if (y != 0) {
		pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
	} else pmv[4].x = pmv[4].y = 0;

	if (x != 0) {
		pmv[5] = ChoosePred(pMB-1, mode_curr);
	} else pmv[5].x = pmv[5].y = 0;

	if (x != 0 && y != 0) {
		pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
	} else pmv[6].x = pmv[6].y = 0;
}

/* search backward or forward */
static void
SearchBF_initial(const int x, const int y,
			const uint32_t MotionFlags,
			const uint32_t iFcode,
			const MBParam * const pParam,
			MACROBLOCK * const pMB,
			const VECTOR * const predMV,