--- trunk/xvidcore/src/prediction/mbprediction.c 2002/03/28 20:57:25 78 +++ branches/release-1_0-branch/xvidcore/src/prediction/mbprediction.c 2004/05/03 23:28:29 1446 @@ -1,71 +1,51 @@ - /****************************************************************************** - * * - * This file is part of XviD, a free MPEG-4 video encoder/decoder * - * * - * XviD is an implementation of a part of one or more MPEG-4 Video tools * - * as specified in ISO/IEC 14496-2 standard. Those intending to use this * - * software module in hardware or software products are advised that its * - * use may infringe existing patents or copyrights, and any such use * - * would be at such party's own risk. The original developer of this * - * software module and his/her company, and subsequent editors and their * - * companies, will have no liability for use of this software or * - * modifications or derivatives thereof. * - * * - * XviD 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. * - * * - * XviD 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 * - * * - ******************************************************************************/ - - /****************************************************************************** - * * - * mbprediction.c * - * * - * Copyright (C) 2001 - Michael Militzer * - * Copyright (C) 2001 - Peter Ross * - * * - * For more information visit the XviD homepage: http://www.xvid.org * - * * - ******************************************************************************/ - - /****************************************************************************** - * * - * Revision history: * - * * - * 12.12.2001 improved calc_acdc_prediction; removed need for memcpy * - * 15.12.2001 moved pmv displacement to motion estimation * - * 30.11.2001 mmx cbp support * - * 17.11.2001 initial version * - * * - ******************************************************************************/ +/***************************************************************************** + * + * XVID MPEG-4 VIDEO CODEC + * - Prediction module - + * + * Copyright (C) 2001-2003 Michael Militzer + * 2001-2003 Peter Ross + * + * 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: mbprediction.c,v 1.14.2.1 2004-05-03 23:28:29 edgomez Exp $ + * + ****************************************************************************/ +#include + +#include "../global.h" #include "../encoder.h" #include "mbprediction.h" #include "../utils/mbfunctions.h" #include "../bitstream/cbp.h" +#include "../bitstream/mbcoding.h" +#include "../bitstream/zigzag.h" -#define ABS(X) (((X)>0)?(X):-(X)) -#define DIV_DIV(A,B) ( (A) > 0 ? ((A)+((B)>>1))/(B) : ((A)-((B)>>1))/(B) ) - - -static int __inline rescale(int predict_quant, int current_quant, int coeff) +static int __inline +rescale(int predict_quant, + int current_quant, + int coeff) { - return (coeff != 0) ? DIV_DIV((coeff) * (predict_quant), (current_quant)) : 0; + return (coeff != 0) ? DIV_DIV((coeff) * (predict_quant), + (current_quant)) : 0; } -static const int16_t default_acdc_values[15] = { +static const int16_t default_acdc_values[15] = { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 @@ -77,14 +57,21 @@ */ -void predict_acdc(MACROBLOCK *pMBs, - uint32_t x, uint32_t y, uint32_t mb_width, - uint32_t block, - int16_t qcoeff[64], - uint32_t current_quant, - int32_t iDcScaler, - int16_t predictors[8]) +void +predict_acdc(MACROBLOCK * pMBs, + uint32_t x, + uint32_t y, + uint32_t mb_width, + uint32_t block, + int16_t qcoeff[64], + uint32_t current_quant, + int32_t iDcScaler, + int16_t predictors[8], + const int bound, + const int bsversion) + { + const int mbpos = (y * mb_width) + x; int16_t *left, *top, *diag, *current; int32_t left_quant = current_quant; @@ -94,127 +81,131 @@ const int16_t *pTop = default_acdc_values; const int16_t *pDiag = default_acdc_values; - uint32_t index = x + y * mb_width; // current macroblock - int * acpred_direction = &pMBs[index].acpred_directions[block]; + uint32_t index = x + y * mb_width; /* current macroblock */ + int *acpred_direction = &pMBs[index].acpred_directions[block]; uint32_t i; left = top = diag = current = 0; - // grab left,top and diag macroblocks + /* grab left,top and diag macroblocks */ - // left macroblock + /* left macroblock */ - if(x && (pMBs[index - 1].mode == MODE_INTRA - || pMBs[index - 1].mode == MODE_INTRA_Q)) { + if (x && mbpos >= bound + 1 && + (pMBs[index - 1].mode == MODE_INTRA || + pMBs[index - 1].mode == MODE_INTRA_Q)) { left = pMBs[index - 1].pred_values[0]; left_quant = pMBs[index - 1].quant; - //DEBUGI("LEFT", *(left+MBPRED_SIZE)); } - - // top macroblock - - if(y && (pMBs[index - mb_width].mode == MODE_INTRA - || pMBs[index - mb_width].mode == MODE_INTRA_Q)) { + /* top macroblock */ + + if (mbpos >= bound + (int)mb_width && + (pMBs[index - mb_width].mode == MODE_INTRA || + pMBs[index - mb_width].mode == MODE_INTRA_Q)) { top = pMBs[index - mb_width].pred_values[0]; top_quant = pMBs[index - mb_width].quant; } - - // diag macroblock - - if(x && y && (pMBs[index - 1 - mb_width].mode == MODE_INTRA - || pMBs[index - 1 - mb_width].mode == MODE_INTRA_Q)) { + /* diag macroblock */ + + if (x && mbpos >= bound + (int)mb_width + 1 && + (pMBs[index - 1 - mb_width].mode == MODE_INTRA || + pMBs[index - 1 - mb_width].mode == MODE_INTRA_Q)) { diag = pMBs[index - 1 - mb_width].pred_values[0]; } current = pMBs[index].pred_values[0]; - // now grab pLeft, pTop, pDiag _blocks_ - + /* now grab pLeft, pTop, pDiag _blocks_ */ + switch (block) { - - case 0: - if(left) + + case 0: + if (left) pLeft = left + MBPRED_SIZE; - - if(top) + + if (top) pTop = top + (MBPRED_SIZE << 1); - - if(diag) + + if (diag) pDiag = diag + 3 * MBPRED_SIZE; - + break; - + case 1: pLeft = current; left_quant = current_quant; - - if(top) { + + if (top) { pTop = top + 3 * MBPRED_SIZE; pDiag = top + (MBPRED_SIZE << 1); } break; - + case 2: - if(left) { + if (left) { pLeft = left + 3 * MBPRED_SIZE; pDiag = left + MBPRED_SIZE; } - + pTop = current; top_quant = current_quant; break; - + case 3: pLeft = current + (MBPRED_SIZE << 1); left_quant = current_quant; - + pTop = current + MBPRED_SIZE; top_quant = current_quant; - + pDiag = current; - + break; - + case 4: - if(left) + if (left) pLeft = left + (MBPRED_SIZE << 2); - if(top) + if (top) pTop = top + (MBPRED_SIZE << 2); - if(diag) + if (diag) pDiag = diag + (MBPRED_SIZE << 2); break; - + case 5: - if(left) + if (left) pLeft = left + 5 * MBPRED_SIZE; - if(top) + if (top) pTop = top + 5 * MBPRED_SIZE; - if(diag) + if (diag) pDiag = diag + 5 * MBPRED_SIZE; break; } - // determine ac prediction direction & ac/dc predictor - // place rescaled ac/dc predictions into predictors[] for later use - - if(ABS(pLeft[0] - pDiag[0]) < ABS(pDiag[0] - pTop[0])) { - *acpred_direction = 1; // vertical + /* determine ac prediction direction & ac/dc predictor place rescaled ac/dc + * predictions into predictors[] for later use */ + + /* Workaround: Bitstream versions <= 32 used to have a wrong predictor + * stored as it wasn't clipped to the [-2048, 2047] range. We only + * use the right predictors for bs versions > 32 */ +#define BUGGY_CLIPPING_BS_VERSION 32 + if (abs(pLeft[0] - pDiag[0]) < abs(pDiag[0] - pTop[0])) { + *acpred_direction = 1; /* vertical */ predictors[0] = DIV_DIV(pTop[0], iDcScaler); - for (i = 1; i < 8; i++) - { + if (bsversion == 0 || bsversion > BUGGY_CLIPPING_BS_VERSION) + predictors[0] = CLIP(predictors[0], -2048, 2047); + for (i = 1; i < 8; i++) { predictors[i] = rescale(top_quant, current_quant, pTop[i]); } - } - else - { - *acpred_direction = 2; // horizontal + } else { + *acpred_direction = 2; /* horizontal */ predictors[0] = DIV_DIV(pLeft[0], iDcScaler); - for (i = 1; i < 8; i++) - { + if (bsversion == 0 || bsversion > BUGGY_CLIPPING_BS_VERSION) + predictors[0] = CLIP(predictors[0], -2048, 2047); + for (i = 1; i < 8; i++) { predictors[i] = rescale(left_quant, current_quant, pLeft[i + 7]); } } @@ -222,83 +213,82 @@ /* decoder: add predictors to dct_codes[] and - store current coeffs to pred_values[] for future prediction + store current coeffs to pred_values[] for future prediction */ -void add_acdc(MACROBLOCK *pMB, - uint32_t block, - int16_t dct_codes[64], - uint32_t iDcScaler, - int16_t predictors[8]) +void +add_acdc(MACROBLOCK * pMB, + uint32_t block, + int16_t dct_codes[64], + uint32_t iDcScaler, + int16_t predictors[8]) { uint8_t acpred_direction = pMB->acpred_directions[block]; - int16_t * pCurrent = pMB->pred_values[block]; + int16_t *pCurrent = pMB->pred_values[block]; uint32_t i; - dct_codes[0] += predictors[0]; // dc prediction + DPRINTF(XVID_DEBUG_COEFF,"predictor[0] %i\n", predictors[0]); + + dct_codes[0] += predictors[0]; /* dc prediction */ pCurrent[0] = dct_codes[0] * iDcScaler; - if (acpred_direction == 1) - { - for (i = 1; i < 8; i++) - { + if (acpred_direction == 1) { + for (i = 1; i < 8; i++) { int level = dct_codes[i] + predictors[i]; + + DPRINTF(XVID_DEBUG_COEFF,"predictor[%i] %i\n",i, predictors[i]); + dct_codes[i] = level; pCurrent[i] = level; - pCurrent[i+7] = dct_codes[i*8]; + pCurrent[i + 7] = dct_codes[i * 8]; } - } - else if (acpred_direction == 2) - { - for (i = 1; i < 8; i++) - { - int level = dct_codes[i*8] + predictors[i]; - dct_codes[i*8] = level; - pCurrent[i+7] = level; + } else if (acpred_direction == 2) { + for (i = 1; i < 8; i++) { + int level = dct_codes[i * 8] + predictors[i]; + DPRINTF(XVID_DEBUG_COEFF,"predictor[%i] %i\n",i*8, predictors[i]); + + dct_codes[i * 8] = level; + pCurrent[i + 7] = level; pCurrent[i] = dct_codes[i]; } - } - else - { - for (i = 1; i < 8; i++) - { + } else { + for (i = 1; i < 8; i++) { pCurrent[i] = dct_codes[i]; - pCurrent[i+7] = dct_codes[i*8]; + pCurrent[i + 7] = dct_codes[i * 8]; } } } -// ****************************************************************** -// ****************************************************************** +/***************************************************************************** + ****************************************************************************/ /* encoder: subtract predictors from qcoeff[] and calculate S1/S2 -todo: perform [-127,127] clamping after prediction -clamping must adjust the coeffs, so dequant is done correctly - -S1/S2 are used to determine if its worth predicting for AC +returns sum of coeefficients *saved* if prediction is enabled + S1 = sum of all (qcoeff - prediction) S2 = sum of all qcoeff */ -uint32_t calc_acdc(MACROBLOCK *pMB, - uint32_t block, - int16_t qcoeff[64], - uint32_t iDcScaler, - int16_t predictors[8]) +int +calc_acdc_coeff(MACROBLOCK * pMB, + uint32_t block, + int16_t qcoeff[64], + uint32_t iDcScaler, + int16_t predictors[8]) { - int16_t * pCurrent = pMB->pred_values[block]; + int16_t *pCurrent = pMB->pred_values[block]; uint32_t i; - uint32_t S1 = 0, S2 = 0; + int S1 = 0, S2 = 0; /* store current coeffs to pred_values[] for future prediction */ pCurrent[0] = qcoeff[0] * iDcScaler; - for(i = 1; i < 8; i++) { + for (i = 1; i < 8; i++) { pCurrent[i] = qcoeff[i]; pCurrent[i + 7] = qcoeff[i * 8]; } @@ -307,117 +297,305 @@ qcoeff[0] = qcoeff[0] - predictors[0]; - if (pMB->acpred_directions[block] == 1) - { - for(i = 1; i < 8; i++) { + if (pMB->acpred_directions[block] == 1) { + for (i = 1; i < 8; i++) { int16_t level; level = qcoeff[i]; - S2 += ABS(level); + S2 += abs(level); level -= predictors[i]; - S1 += ABS(level); + S1 += abs(level); predictors[i] = level; } - } - else // acpred_direction == 2 + } else /* acpred_direction == 2 */ { - for(i = 1; i < 8; i++) { + for (i = 1; i < 8; i++) { int16_t level; - level = qcoeff[i*8]; - S2 += ABS(level); + level = qcoeff[i * 8]; + S2 += abs(level); level -= predictors[i]; - S1 += ABS(level); + S1 += abs(level); predictors[i] = level; } } - + return S2 - S1; } -/* apply predictors[] to qcoeff */ -void apply_acdc(MACROBLOCK *pMB, - uint32_t block, - int16_t qcoeff[64], - int16_t predictors[8]) +/* returns the bits *saved* if prediction is enabled */ + +int +calc_acdc_bits(MACROBLOCK * pMB, + uint32_t block, + int16_t qcoeff[64], + uint32_t iDcScaler, + int16_t predictors[8]) { - uint32_t i; + const int direction = pMB->acpred_directions[block]; + int16_t *pCurrent = pMB->pred_values[block]; + int16_t tmp[8]; + unsigned int i; + int Z1, Z2; - if (pMB->acpred_directions[block] == 1) - { - for(i = 1; i < 8; i++) - { - qcoeff[i] = predictors[i]; - } + /* store current coeffs to pred_values[] for future prediction */ + pCurrent[0] = qcoeff[0] * iDcScaler; + for (i = 1; i < 8; i++) { + pCurrent[i] = qcoeff[i]; + pCurrent[i + 7] = qcoeff[i * 8]; } - else - { - for(i = 1; i < 8; i++) - { - qcoeff[i*8] = predictors[i]; + + + /* dc prediction */ + qcoeff[0] = qcoeff[0] - predictors[0]; + + /* calc cost before ac prediction */ + Z2 = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]); + + /* apply ac prediction & calc cost*/ + if (direction == 1) { + for (i = 1; i < 8; i++) { + tmp[i] = qcoeff[i]; + qcoeff[i] -= predictors[i]; + predictors[i] = qcoeff[i]; + } + }else{ /* acpred_direction == 2 */ + for (i = 1; i < 8; i++) { + tmp[i] = qcoeff[i*8]; + qcoeff[i*8] -= predictors[i]; + predictors[i] = qcoeff[i*8]; } } + + Z1 = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[direction]); + + /* undo prediction */ + if (direction == 1) { + for (i = 1; i < 8; i++) + qcoeff[i] = tmp[i]; + }else{ /* acpred_direction == 2 */ + for (i = 1; i < 8; i++) + qcoeff[i*8] = tmp[i]; + } + + return Z2-Z1; +} + +/* apply predictors[] to qcoeff */ + +void +apply_acdc(MACROBLOCK * pMB, + uint32_t block, + int16_t qcoeff[64], + int16_t predictors[8]) +{ + unsigned int i; + + if (pMB->acpred_directions[block] == 1) { + for (i = 1; i < 8; i++) + qcoeff[i] = predictors[i]; + } else { + for (i = 1; i < 8; i++) + qcoeff[i * 8] = predictors[i]; + } } -void MBPrediction(MBParam *pParam, - uint32_t x, - uint32_t y, - uint32_t mb_width, - int16_t qcoeff[6*64], - MACROBLOCK *mbs) +void +MBPrediction(FRAMEINFO * frame, + uint32_t x, + uint32_t y, + uint32_t mb_width, + int16_t qcoeff[6 * 64]) { int32_t j; - int32_t iDcScaler, iQuant = pParam->quant; - int32_t S = 0; + int32_t iDcScaler, iQuant; + int S = 0; int16_t predictors[6][8]; - MACROBLOCK *pMB = &mbs[x + y * mb_width]; + MACROBLOCK *pMB = &frame->mbs[x + y * mb_width]; + iQuant = pMB->quant; if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_INTRA_Q)) { - - for(j = 0; j < 6; j++) - { - iDcScaler = get_dc_scaler(iQuant, (j < 4) ? 1 : 0); - - predict_acdc(mbs, - x, - y, - mb_width, - j, - &qcoeff[j*64], - iQuant, - iDcScaler, - predictors[j]); - - S += calc_acdc(pMB, - j, - &qcoeff[j*64], - iDcScaler, - predictors[j]); + + for (j = 0; j < 6; j++) { + iDcScaler = get_dc_scaler(iQuant, j<4); + + predict_acdc(frame->mbs, x, y, mb_width, j, &qcoeff[j * 64], + iQuant, iDcScaler, predictors[j], 0, 0); + + if ((frame->vop_flags & XVID_VOP_HQACPRED)) + S += calc_acdc_bits(pMB, j, &qcoeff[j * 64], iDcScaler, predictors[j]); + else + S += calc_acdc_coeff(pMB, j, &qcoeff[j * 64], iDcScaler, predictors[j]); } - if (S < 0) // dont predict - { - for(j = 0; j < 6; j++) - { + if (S<=0) { /* dont predict */ + for (j = 0; j < 6; j++) pMB->acpred_directions[j] = 0; - } - } - else - { - for(j = 0; j < 6; j++) - { - apply_acdc(pMB, j, &qcoeff[j*64], predictors[j]); - } + }else{ + for (j = 0; j < 6; j++) + apply_acdc(pMB, j, &qcoeff[j * 64], predictors[j]); } + pMB->cbp = calc_cbp(qcoeff); } +} + +static const VECTOR zeroMV = { 0, 0 }; + +VECTOR +get_pmv2(const MACROBLOCK * const mbs, + const int mb_width, + const int bound, + const int x, + const int y, + const int block) +{ + int lx, ly, lz; /* left */ + int tx, ty, tz; /* top */ + int rx, ry, rz; /* top-right */ + int lpos, tpos, rpos; + int num_cand = 0, last_cand = 1; + + VECTOR pmv[4]; /* left neighbour, top neighbour, top-right neighbour */ + + switch (block) { + case 0: + lx = x - 1; ly = y; lz = 1; + tx = x; ty = y - 1; tz = 2; + rx = x + 1; ry = y - 1; rz = 2; + break; + case 1: + lx = x; ly = y; lz = 0; + tx = x; ty = y - 1; tz = 3; + rx = x + 1; ry = y - 1; rz = 2; + break; + case 2: + lx = x - 1; ly = y; lz = 3; + tx = x; ty = y; tz = 0; + rx = x; ry = y; rz = 1; + break; + default: + lx = x; ly = y; lz = 2; + tx = x; ty = y; tz = 0; + rx = x; ry = y; rz = 1; + } + + lpos = lx + ly * mb_width; + rpos = rx + ry * mb_width; + tpos = tx + ty * mb_width; + + if (lpos >= bound && lx >= 0) { + num_cand++; + pmv[1] = mbs[lpos].mvs[lz]; + } else pmv[1] = zeroMV; + + if (tpos >= bound) { + num_cand++; + last_cand = 2; + pmv[2] = mbs[tpos].mvs[tz]; + } else pmv[2] = zeroMV; + + if (rpos >= bound && rx < mb_width) { + num_cand++; + last_cand = 3; + pmv[3] = mbs[rpos].mvs[rz]; + } else pmv[3] = zeroMV; + + /* If there're more than one candidate, we return the median vector */ + + if (num_cand > 1) { + /* set median */ + pmv[0].x = + MIN(MAX(pmv[1].x, pmv[2].x), + MIN(MAX(pmv[2].x, pmv[3].x), MAX(pmv[1].x, pmv[3].x))); + pmv[0].y = + MIN(MAX(pmv[1].y, pmv[2].y), + MIN(MAX(pmv[2].y, pmv[3].y), MAX(pmv[1].y, pmv[3].y))); + return pmv[0]; + } + + return pmv[last_cand]; /* no point calculating median mv */ +} + +VECTOR +get_qpmv2(const MACROBLOCK * const mbs, + const int mb_width, + const int bound, + const int x, + const int y, + const int block) +{ + int lx, ly, lz; /* left */ + int tx, ty, tz; /* top */ + int rx, ry, rz; /* top-right */ + int lpos, tpos, rpos; + int num_cand = 0, last_cand = 1; + + VECTOR pmv[4]; /* left neighbour, top neighbour, top-right neighbour */ + + switch (block) { + case 0: + lx = x - 1; ly = y; lz = 1; + tx = x; ty = y - 1; tz = 2; + rx = x + 1; ry = y - 1; rz = 2; + break; + case 1: + lx = x; ly = y; lz = 0; + tx = x; ty = y - 1; tz = 3; + rx = x + 1; ry = y - 1; rz = 2; + break; + case 2: + lx = x - 1; ly = y; lz = 3; + tx = x; ty = y; tz = 0; + rx = x; ry = y; rz = 1; + break; + default: + lx = x; ly = y; lz = 2; + tx = x; ty = y; tz = 0; + rx = x; ry = y; rz = 1; + } + + lpos = lx + ly * mb_width; + rpos = rx + ry * mb_width; + tpos = tx + ty * mb_width; + + if (lpos >= bound && lx >= 0) { + num_cand++; + pmv[1] = mbs[lpos].qmvs[lz]; + } else pmv[1] = zeroMV; + + if (tpos >= bound) { + num_cand++; + last_cand = 2; + pmv[2] = mbs[tpos].qmvs[tz]; + } else pmv[2] = zeroMV; + + if (rpos >= bound && rx < mb_width) { + num_cand++; + last_cand = 3; + pmv[3] = mbs[rpos].qmvs[rz]; + } else pmv[3] = zeroMV; + + /* If there're more than one candidate, we return the median vector */ + + if (num_cand > 1) { + /* set median */ + pmv[0].x = + MIN(MAX(pmv[1].x, pmv[2].x), + MIN(MAX(pmv[2].x, pmv[3].x), MAX(pmv[1].x, pmv[3].x))); + pmv[0].y = + MIN(MAX(pmv[1].y, pmv[2].y), + MIN(MAX(pmv[2].y, pmv[3].y), MAX(pmv[1].y, pmv[3].y))); + return pmv[0]; + } + return pmv[last_cand]; /* no point calculating median mv */ }