estimation_rd_based_bvop.c

从FFMPEG转换而来的H264解码程序,VC下编译..

/*****************************************************************************
 *
 *  XVID MPEG-4 VIDEO CODEC
 *  - Rate-Distortion Based Motion Estimation for B- VOPs  -
 *
 *  Copyright(C) 2004 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_rd_based_bvop.c,v 1.10 2005/12/09 04:45:35 syskin Exp $
 *
 ****************************************************************************/

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

#include "../encoder.h"
#include "../bitstream/mbcoding.h"
#include "../prediction/mbprediction.h"
#include "../global.h"
#include "../image/interpolate8x8.h"
#include "estimation.h"
#include "motion.h"
#include "sad.h"
#include "../bitstream/zigzag.h"
#include "../quant/quant.h"
#include "../bitstream/vlc_codes.h"
#include "../dct/fdct.h"
#include "motion_inlines.h"

/* rd = BITS_MULT*bits + LAMBDA*distortion */
#define LAMBDA		( (int)(BITS_MULT*1.0) )


static __inline unsigned int
Block_CalcBits_BVOP(int16_t * const coeff,
				int16_t * const data,
				int16_t * const dqcoeff,
				const uint32_t quant, const int quant_type,
				uint32_t * cbp,
				const int block,
				const uint16_t * scan_table,
				const unsigned int lambda,
				const uint16_t * mpeg_quant_matrices,
				const unsigned int quant_sq,
				int * const cbpcost)
{
	int sum;
	int bits;
	int distortion = 0;

	fdct((short * const)data);

	if (quant_type) sum = quant_h263_inter(coeff, data, quant, mpeg_quant_matrices);
	else sum = quant_mpeg_inter(coeff, data, quant, mpeg_quant_matrices);

	if ((sum >= 3) || (coeff[1] != 0) || (coeff[8] != 0) || (coeff[0] != 0)) {
		*cbp |= 1 << (5 - block);
		bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table);
		bits += *cbpcost;
		*cbpcost = 0; /* don't add cbp cost second time */

		if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);
		else dequant_mpeg_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);

		distortion = sse8_16bit(data, dqcoeff, 8*sizeof(int16_t));
	} else {
		const static int16_t zero_block[64] =
			{
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
			};
		bits = 0;
		distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t));
	}

	return bits + (lambda*distortion)/quant_sq;
}


static __inline unsigned int
Block_CalcBits_BVOP_direct(int16_t * const coeff,
				int16_t * const data,
				int16_t * const dqcoeff,
				const uint32_t quant, const int quant_type,
				uint32_t * cbp,
				const int block,
				const uint16_t * scan_table,
				const unsigned int lambda,
				const uint16_t * mpeg_quant_matrices,
				const unsigned int quant_sq,
				int * const cbpcost)
{
	int sum;
	int bits;
	int distortion = 0;

	fdct((short * const)data);

	if (quant_type) sum = quant_h263_inter(coeff, data, quant, mpeg_quant_matrices);
	else sum = quant_mpeg_inter(coeff, data, quant, mpeg_quant_matrices);

	if ((sum >= 3) || (coeff[1] != 0) || (coeff[8] != 0) || (coeff[0] > 0) || (coeff[0] < -1)) {
		*cbp |= 1 << (5 - block);
		bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table);
		bits += *cbpcost;
		*cbpcost = 0;

		if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);
		else dequant_mpeg_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);

		distortion = sse8_16bit(data, dqcoeff, 8*sizeof(int16_t));
	} else {
		const static int16_t zero_block[64] =
			{
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
				0, 0, 0, 0, 0, 0, 0, 0,
			};
		bits = 0;
		distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t));
	}

	return bits + (lambda*distortion)/quant_sq;
}

static void
CheckCandidateRDBF(const int x, const int y, SearchData * const data, const unsigned int Direction)
{

	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
	int32_t rd = (3+2)*BITS_MULT; /* 3 bits for mode + 2 for vector (minimum) */
	VECTOR * current;
	const uint8_t * ptr;
	int i, xc, yc;
	unsigned cbp = 0;
	int cbpcost = 7*BITS_MULT; /* how much to add if cbp turns out to be non-zero */

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

	if (!data->qpel_precision) {
		ptr = GetReference(x, y, data);
		current = data->currentMV;
		xc = x; yc = y;
	} else { /* x and y are in 1/4 precision */
		ptr = xvid_me_interpolate16x16qpel(x, y, 0, data);
		current = data->currentQMV;
		xc = x/2; yc = y/2;
	}

	rd += BITS_MULT*(d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision)-2);

	for(i = 0; i < 4; i++) {
		int s = 8*((i&1) + (i>>1)*data->iEdgedWidth);
		transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth);
		rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
								&cbp, i, data->scan_table, data->lambda[i], data->mpeg_quant_matrices, 
								data->quant_sq, &cbpcost);
		if (rd >= data->iMinSAD[0]) return;
	}

	/* chroma */
	xc = (xc >> 1) + roundtab_79[xc & 0x3];
	yc = (yc >> 1) + roundtab_79[yc & 0x3];

	/* chroma U */
	ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
	transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2);
	rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
								&cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices, 
								data->quant_sq, &cbpcost);
	if (rd >= data->iMinSAD[0]) return;

	/* chroma V */
	ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
	transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2);
	rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, 
								&cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices, 
								data->quant_sq, &cbpcost);

	if (rd < data->iMinSAD[0]) {
		data->iMinSAD[0] = rd;
		current[0].x = x; current[0].y = y;
		data->dir = Direction;
		*data->cbp = cbp;
	}
}

static void
CheckCandidateRDDirect(const int x, const int y, SearchData * const data, const unsigned int Direction)
{
	int32_t xcf = 0, ycf = 0, xcb = 0, ycb = 0;
	int32_t rd = 1*BITS_MULT;
	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
	unsigned int cbp = 0;
	unsigned int k;
	VECTOR mvs, b_mvs;
	int cbpcost = 6*BITS_MULT; /* how much to add if cbp turns out to be non-zero */

	const uint8_t *ReferenceF, *ReferenceB;

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

	for (k = 0; k < 4; k++) {
		int s = 8*((k&1) + (k>>1)*data->iEdgedWidth);

		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;
			ReferenceF = xvid_me_interpolate8x8qpel(mvs.x, mvs.y, k, 0, data);
			ReferenceB = xvid_me_interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data);
		} else {
			xcf += mvs.x; ycf += mvs.y;
			xcb += b_mvs.x; ycb += b_mvs.y;
			ReferenceF = GetReference(mvs.x, mvs.y, data) + s;
			ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data) + s;
		}

		transfer_8to16sub2ro(in, data->Cur + s, ReferenceF, ReferenceB, data->iEdgedWidth);
		rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, 
										&cbp, k, data->scan_table, data->lambda[k], data->mpeg_quant_matrices, 
										data->quant_sq, &cbpcost);
		if (rd > *(data->iMinSAD)) return;
	}
	
	/* chroma */
	xcf = (xcf >> 3) + roundtab_76[xcf & 0xf];
	ycf = (ycf >> 3) + roundtab_76[ycf & 0xf];
	xcb = (xcb >> 3) + roundtab_76[xcb & 0xf];
	ycb = (ycb >> 3) + roundtab_76[ycb & 0xf];

	/* chroma U */
	ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding);
	ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[4], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding);
	transfer_8to16sub2ro(in, data->CurU, ReferenceF, ReferenceB, data->iEdgedWidth/2);
	rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
									&cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices,
									data->quant_sq, &cbpcost);
	if (rd >= data->iMinSAD[0]) return;

	/* chroma V */
	ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding);
	ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[5], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding);
	transfer_8to16sub2ro(in, data->CurV, ReferenceF, ReferenceB, data->iEdgedWidth/2);
	rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
									&cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices,
									data->quant_sq, &cbpcost);

	if (cbp || x != 0 || y != 0)
		rd += BITS_MULT * d_mv_bits(x, y, zeroMV, 1, 0);

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

static void
CheckCandidateRDInt(const int x, const int y, SearchData * const data, const unsigned int Direction)
{
	int32_t xf, yf, xb, yb, xcf, ycf, xcb, ycb;
	int32_t rd = 2*BITS_MULT;
	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
	unsigned int cbp = 0;
	unsigned int i;
	int cbpcost = 7*BITS_MULT; /* how much to add if cbp turns out to be non-zero */

	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);