estimation_rd_based.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 P- and S- VOPs  -
 *
 *  Copyright(C) 2003 Radoslaw Czyz <[e-mail]>
 *               2003 Michael Militzer <[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.c,v 1.14 2005/12/09 04:45:35 syskin Exp $
 *
 ****************************************************************************/

/* RD mode decision and search */

#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(	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 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 > 0) {
		*cbp |= 1 << (5 - block);
		bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table);

		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_CalcBitsIntra(MACROBLOCK * pMB,
					const unsigned int x,
					const unsigned int y,
					const unsigned int mb_width,
					const uint32_t block,
					int16_t coeff[64],
					int16_t qcoeff[64],
					int16_t dqcoeff[64],
					int16_t predictors[8],
					const uint32_t quant,
					const int quant_type,
					unsigned int bits[2],
					unsigned int cbp[2],
					unsigned int lambda,
					const uint16_t * mpeg_quant_matrices,
					const unsigned int quant_sq)
{
	int direction;
	int16_t *pCurrent;
	unsigned int i, coded;
	unsigned int distortion = 0;
	const uint32_t iDcScaler = get_dc_scaler(quant, block < 4);

	fdct((short * const)coeff);

	if (quant_type) {
		quant_h263_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices);
		dequant_h263_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices);
	} else {
		quant_mpeg_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices);
		dequant_mpeg_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices);
	}

	predict_acdc(pMB-(x+mb_width*y), x, y, mb_width, block, qcoeff,
				quant, iDcScaler, predictors, 0);
	
	direction = pMB->acpred_directions[block];
	pCurrent = (int16_t*)pMB->pred_values[block];

	/* store current coeffs to pred_values[] for future prediction */
	pCurrent[0] = qcoeff[0] * iDcScaler;
	pCurrent[0] = CLIP(pCurrent[0], -2048, 2047);
	for (i = 1; i < 8; i++) {
		pCurrent[i] = qcoeff[i];
		pCurrent[i + 7] = qcoeff[i * 8];
	}

	/* dc prediction */
	qcoeff[0] = qcoeff[0] - predictors[0];

	if (block < 4) bits[1] = bits[0] = dcy_tab[qcoeff[0] + 255].len - 3; /* 3 bits added before (4 times) */
	else bits[1] = bits[0] = dcc_tab[qcoeff[0] + 255].len - 2; /* 2 bits added before (2 times)*/

	/* calc cost before ac prediction */
	bits[0] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]);
	if (coded > 0) cbp[0] |= 1 << (5 - block);

	/* apply ac prediction & calc cost*/
	if (direction == 1) {
		for (i = 1; i < 8; i++) {
			qcoeff[i] -= predictors[i];
			predictors[i] = qcoeff[i];
		}
	} else {						/* acpred_direction == 2 */
		for (i = 1; i < 8; i++) {
			qcoeff[i*8] -= predictors[i];
			predictors[i] = qcoeff[i*8];
		}
	}

	bits[1] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[direction]);
	if (coded > 0) cbp[1] |= 1 << (5 - block);

	distortion = sse8_16bit(coeff, dqcoeff, 8*sizeof(int16_t));

	return (lambda*distortion)/quant_sq;
}



static void
CheckCandidateRD16(const int x, const int y, SearchData * const data, const unsigned int Direction)
{

	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
	/* minimum nuber of bits INTER can take is 1 (mcbpc) + 2 (cby) + 2 (vector) */
	int32_t rd = BITS_MULT * (1+2+2);
	VECTOR * current;
	const uint8_t * ptr;
	int i, t, xc, yc;
	unsigned cbp = 0;

	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;
	}

	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 += data->temp[i] = Block_CalcBits(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);
	}

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

	if (data->temp[0] + t < data->iMinSAD[1]) {
		data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); }
	if (data->temp[1] < data->iMinSAD[2]) {
		data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | (cbp&16); }
	if (data->temp[2] < data->iMinSAD[3]) {
		data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | (cbp&8); }
	if (data->temp[3] < data->iMinSAD[4]) {
		data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); }

	rd += BITS_MULT * (xvid_cbpy_tab[15-(cbp>>2)].len - 2);

	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(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);
	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(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);

	rd += BITS_MULT * (mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len - 1); /* one was added before */

	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
CheckCandidateRD8(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;
	VECTOR * current;
	const uint8_t * ptr;
	unsigned int cbp = 0;

	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;
	} else { /* x and y are in 1/4 precision */
		ptr = xvid_me_interpolate8x8qpel(x, y, 0, 0, data);
		current = data->currentQMV;
	}

	transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth);
	rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, 
						&cbp, 5, data->scan_table, data->lambda[0], 
						data->mpeg_quant_matrices, data->quant_sq);
	/* we took 2 bits into account before */
	rd += BITS_MULT * (d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision) - 2); 


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


static int
findRD_inter(SearchData * const Data,
			const int x, const int y,
			const MBParam * const pParam,
			const uint32_t MotionFlags)
{
	int i;
	int32_t bsad[5];

	if (Data->qpel) {
		for(i = 0; i < 5; i++) {
			Data->currentMV[i].x = Data->currentQMV[i].x/2;
			Data->currentMV[i].y = Data->currentQMV[i].y/2;
		}
		Data->qpel_precision = 1;
		CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255);

		if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */
			for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i];
			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
						pParam->width, pParam->height, Data->iFcode - Data->qpel, 1);
			Data->qpel_precision = 0;
			if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1)
				CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
		}

	} else { /* not qpel */

		CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
	}

	if (MotionFlags&XVID_ME_EXTSEARCH_RD)