--- mbtransquant.c 2003/05/09 22:03:13 1.21.2.10 +++ mbtransquant.c 2003/05/04 15:15:59 1.22.2.1 @@ -1,32 +1,57 @@ -/***************************************************************************** - * - * XVID MPEG-4 VIDEO CODEC - * - MB Transfert/Quantization functions - - * - * Copyright(C) 2001-2003 Peter Ross - * 2001-2003 Michael Militzer - * 2003 Edouard Gomez - * - * 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: mbtransquant.c,v 1.21.2.10 2003/05/09 22:03:13 chl Exp $ - * - ****************************************************************************/ + /****************************************************************************** + * * + * 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 * + * * + ******************************************************************************/ + + /****************************************************************************** + * * + * mbtransquant.c * + * * + * Copyright (C) 2001 - Peter Ross * + * Copyright (C) 2001 - Michael Militzer * + * * + * For more information visit the XviD homepage: http://www.xvid.org * + * * + ******************************************************************************/ + + /****************************************************************************** + * * + * Revision history: * + * * + * 29.03.2002 interlacing speedup - used transfer strides instead of * + * manual field-to-frame conversion * + * 26.03.2002 interlacing support - moved transfers outside loops * + * 22.12.2001 get_dc_scaler() moved to common.h * + * 19.11.2001 introduced coefficient thresholding (Isibaar) * + * 17.11.2001 initial version * + * * + ******************************************************************************/ #include -#include #include "../portab.h" #include "mbfunctions.h" @@ -34,8 +59,6 @@ #include "../global.h" #include "mem_transfer.h" #include "timer.h" -#include "../bitstream/mbcoding.h" -#include "../bitstream/zigzag.h" #include "../dct/fdct.h" #include "../dct/idct.h" #include "../quant/quant_mpeg4.h" @@ -46,49 +69,11 @@ MBFIELDTEST_PTR MBFieldTest; -/* - * Skip blocks having a coefficient sum below this value. This value will be - * corrected according to the MB quantizer to avoid artifacts for quant==1 - */ -#define PVOP_TOOSMALL_LIMIT 1 -#define BVOP_TOOSMALL_LIMIT 3 - -/***************************************************************************** - * Local functions - ****************************************************************************/ +#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ -/* permute block and return field dct choice */ -static __inline uint32_t -MBDecideFieldDCT(int16_t data[6 * 64]) -{ - uint32_t field = MBFieldTest(data); - - if (field) - MBFrameToField(data); - - return field; -} - -/* Performs Forward DCT on all blocks */ static __inline void -MBfDCT(const MBParam * const pParam, - const FRAMEINFO * const frame, - MACROBLOCK * const pMB, - uint32_t x_pos, - uint32_t y_pos, - int16_t data[6 * 64]) +MBfDCT(int16_t data[6 * 64]) { - /* Handles interlacing */ - start_timer(); - pMB->field_dct = 0; - if ((frame->vol_flags & XVID_VOL_INTERLACING) && - (x_pos>0) && (x_posmb_width-1) && - (y_pos>0) && (y_posmb_height-1)) { - pMB->field_dct = MBDecideFieldDCT(data); - } - stop_interlacing_timer(); - - /* Perform DCT */ start_timer(); fdct(&data[0 * 64]); fdct(&data[1 * 64]); @@ -99,400 +84,306 @@ stop_dct_timer(); } -/* Performs Inverse DCT on all blocks */ -static __inline void -MBiDCT(int16_t data[6 * 64], - const uint8_t cbp) -{ - start_timer(); - if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]); - if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]); - if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]); - if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]); - if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]); - if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]); - stop_idct_timer(); -} -/* Quantize all blocks -- Intra mode */ -static __inline void -MBQuantIntra(const MBParam * pParam, - const FRAMEINFO * const frame, - const MACROBLOCK * pMB, - int16_t qcoeff[6 * 64], - int16_t data[6*64]) +static __inline uint32_t +QuantizeInterBlock( int16_t qcoeff[64], + const int16_t data[64], + const uint32_t iQuant, + const uint32_t quant_type) { - int i; + uint32_t sum; - for (i = 0; i < 6; i++) { - uint32_t iDcScaler = get_dc_scaler(pMB->quant, i < 4); + start_timer(); + if (quant_type == H263_QUANT) + sum = quant_inter(qcoeff, data, iQuant); + else + sum = quant4_inter(qcoeff, data, iQuant); - /* Quantize the block */ - start_timer(); - if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) { - quant_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); - } else { - quant4_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); - } - stop_quant_timer(); - } + stop_quant_timer(); + return sum; } -/* DeQuantize all blocks -- Intra mode */ -static __inline void -MBDeQuantIntra(const MBParam * pParam, - const int iQuant, - int16_t qcoeff[6 * 64], - int16_t data[6*64]) +void +MBTransQuantIntra(const MBParam * const pParam, + FRAMEINFO * const frame, + MACROBLOCK * const pMB, + const uint32_t x_pos, + const uint32_t y_pos, + int16_t data[6 * 64], + int16_t qcoeff[6 * 64]) { - int i; - - for (i = 0; i < 6; i++) { - uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); - - start_timer(); - if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) - dequant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); - else - dequant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); - stop_iquant_timer(); - } -} + uint32_t stride = pParam->edged_width; + const uint32_t stride2 = stride / 2; + uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); + int i; + const uint32_t iQuant = pMB->quant; + uint8_t *pY_Cur, *pU_Cur, *pV_Cur; + const IMAGE * const pCurrent = &frame->image; -static int -dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero); + start_timer(); + if ((frame->global_flags & XVID_REDUCED)) + { + pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); + pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); + pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); -static int -dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero); + filter_18x18_to_8x8(&data[0 * 64], pY_Cur, stride); + filter_18x18_to_8x8(&data[1 * 64], pY_Cur + 16, stride); + filter_18x18_to_8x8(&data[2 * 64], pY_Cur + next_block, stride); + filter_18x18_to_8x8(&data[3 * 64], pY_Cur + next_block + 16, stride); + filter_18x18_to_8x8(&data[4 * 64], pU_Cur, stride2); + filter_18x18_to_8x8(&data[5 * 64], pV_Cur, stride2); + } else { + pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); + pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); + pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); + transfer_8to16copy(&data[0 * 64], pY_Cur, stride); + transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); + transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); + transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); + transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); + transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); + } + stop_transfer_timer(); -/* Quantize all blocks -- Inter mode */ -static __inline uint8_t -MBQuantInter(const MBParam * pParam, - const FRAMEINFO * const frame, - const MACROBLOCK * pMB, - int16_t data[6 * 64], - int16_t qcoeff[6 * 64], - int bvop, - int limit) -{ + /* XXX: rrv+interlacing is buggy */ + start_timer(); + pMB->field_dct = 0; + if ((frame->global_flags & XVID_INTERLACING) && + (x_pos>0) && (x_posmb_width-1) && + (y_pos>0) && (y_posmb_height-1)) { + pMB->field_dct = MBDecideFieldDCT(data); + } + stop_interlacing_timer(); - int i; - uint8_t cbp = 0; - int sum; - int code_block; + MBfDCT(data); for (i = 0; i < 6; i++) { + const uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); - /* Quantize the block */ start_timer(); - if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) { - sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant); - if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) { - sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1; - limit = 1; - } - } else { - sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant); -// if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) -// sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1; - } + if (pParam->m_quant_type == H263_QUANT) + quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); + else + quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); stop_quant_timer(); - /* - * We code the block if the sum is higher than the limit and if the first - * two AC coefficients in zig zag order are not zero. - */ - code_block = 0; - if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { - code_block = 1; - } else { + /* speedup: dont decode when encoding only ivops */ + if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) + { + start_timer(); + if (pParam->m_quant_type == H263_QUANT) + dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); + else + dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); + stop_iquant_timer(); - if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) { - /* dark blocks prevention for direct mode */ - if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0)) - code_block = 1; - } else { - /* not direct mode */ - if (qcoeff[i*64] != 0) - code_block = 1; - } + start_timer(); + idct(&data[i * 64]); + stop_idct_timer(); } - - /* Set the corresponding cbp bit */ - cbp |= code_block << (5 - i); } - return(cbp); -} - -/* DeQuantize all blocks -- Inter mode */ -static __inline void -MBDeQuantInter(const MBParam * pParam, - const int iQuant, - int16_t data[6 * 64], - int16_t qcoeff[6 * 64], - const uint8_t cbp) -{ - int i; + /* speedup: dont decode when encoding only ivops */ + if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) + { + + if (pMB->field_dct) { + next_block = stride; + stride *= 2; + } - for (i = 0; i < 6; i++) { - if (cbp & (1 << (5 - i))) { - start_timer(); - if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) - dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); - else - dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); - stop_iquant_timer(); + start_timer(); + if ((frame->global_flags & XVID_REDUCED)) { + copy_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); + copy_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); + copy_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); + copy_upsampled_8x8_16to8(pY_Cur + next_block + 16, &data[3 * 64], stride); + copy_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); + copy_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); + } else { + transfer_16to8copy(pY_Cur, &data[0 * 64], stride); + transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); + transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); + transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); + transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); + transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); } + stop_transfer_timer(); } -} - -typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS); -typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS); +} -static __inline void -MBTrans8to16(const MBParam * const pParam, - const FRAMEINFO * const frame, - const MACROBLOCK * const pMB, - const uint32_t x_pos, - const uint32_t y_pos, - int16_t data[6 * 64]) +uint8_t +MBTransQuantInter(const MBParam * const pParam, + FRAMEINFO * const frame, + MACROBLOCK * const pMB, + const uint32_t x_pos, + const uint32_t y_pos, + int16_t data[6 * 64], + int16_t qcoeff[6 * 64]) { uint32_t stride = pParam->edged_width; - uint32_t stride2 = stride / 2; - uint32_t next_block = stride * 8; - int32_t cst; + const uint32_t stride2 = stride / 2; + uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); + int i; + const uint32_t iQuant = pMB->quant; uint8_t *pY_Cur, *pU_Cur, *pV_Cur; + int cbp = 0; + uint32_t sum; const IMAGE * const pCurrent = &frame->image; - transfer_operation_8to16_t *transfer_op = NULL; - if ((frame->vop_flags & XVID_VOP_REDUCED)) { - - /* Image pointers */ - pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); + if ((frame->global_flags & XVID_REDUCED)) { + pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); - - /* Block size */ - cst = 16; - - /* Operation function */ - transfer_op = (transfer_operation_8to16_t*)filter_18x18_to_8x8; } else { - - /* Image pointers */ - pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); + pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); - - /* Block size */ - cst = 8; - - /* Operation function */ - transfer_op = (transfer_operation_8to16_t*)transfer_8to16copy; } - /* Do the transfer */ start_timer(); - transfer_op(&data[0 * 64], pY_Cur, stride); - transfer_op(&data[1 * 64], pY_Cur + cst, stride); - transfer_op(&data[2 * 64], pY_Cur + next_block, stride); - transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride); - transfer_op(&data[4 * 64], pU_Cur, stride2); - transfer_op(&data[5 * 64], pV_Cur, stride2); - stop_transfer_timer(); -} + pMB->field_dct = 0; + if ((frame->global_flags & XVID_INTERLACING) && + (x_pos>0) && (x_posmb_width-1) && + (y_pos>0) && (y_posmb_height-1)) { + pMB->field_dct = MBDecideFieldDCT(data); + } + stop_interlacing_timer(); -static __inline void -MBTrans16to8(const MBParam * const pParam, - const FRAMEINFO * const frame, - const MACROBLOCK * const pMB, - const uint32_t x_pos, - const uint32_t y_pos, - int16_t data[6 * 64], - const uint32_t add, - const uint8_t cbp) -{ - uint8_t *pY_Cur, *pU_Cur, *pV_Cur; - uint32_t stride = pParam->edged_width; - uint32_t stride2 = stride / 2; - uint32_t next_block = stride * 8; - uint32_t cst; - const IMAGE * const pCurrent = &frame->image; - transfer_operation_16to8_t *transfer_op = NULL; + MBfDCT(data); - if (pMB->field_dct) { - next_block = stride; - stride *= 2; - } + for (i = 0; i < 6; i++) { + const uint32_t limit = TOOSMALL_LIMIT + ((iQuant == 1) ? 1 : 0); + /* + * no need to transfer 8->16-bit + * (this is performed already in motion compensation) + */ - if ((frame->vop_flags & XVID_VOP_REDUCED)) { + sum = QuantizeInterBlock(&qcoeff[i * 64], &data[i * 64], iQuant, pParam->m_quant_type); - /* Image pointers */ - pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); - pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); - pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); + if(frame->global_flags & XVID_CARTOON_MODE) { + limit *= 3; + } - /* Block size */ - cst = 16; + if (sum >= limit) { - /* Operation function */ - if(add) - transfer_op = (transfer_operation_16to8_t*)add_upsampled_8x8_16to8; - else - transfer_op = (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8; - } else { + start_timer(); + if (pParam->m_quant_type == H263_QUANT) + dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); + else + dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); + stop_iquant_timer(); - /* Image pointers */ - pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); - pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); - pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); + cbp |= 1 << (5 - i); - /* Block size */ - cst = 8; + start_timer(); + idct(&data[i * 64]); + stop_idct_timer(); + } + } - /* Operation function */ - if(add) - transfer_op = (transfer_operation_16to8_t*)transfer_16to8add; - else - transfer_op = (transfer_operation_16to8_t*)transfer_16to8copy; + if (pMB->field_dct) { + next_block = stride; + stride *= 2; } - /* Do the operation */ start_timer(); - if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); - if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride); - if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); - if (cbp& 4) transfer_op(pY_Cur + next_block + cst, &data[3 * 64], stride); - if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); - if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); + if ((frame->global_flags & XVID_REDUCED)) { + if (cbp & 32) + add_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); + if (cbp & 16) + add_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); + if (cbp & 8) + add_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); + if (cbp & 4) + add_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride); + if (cbp & 2) + add_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); + if (cbp & 1) + add_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); + } else { + if (cbp & 32) + transfer_16to8add(pY_Cur, &data[0 * 64], stride); + if (cbp & 16) + transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride); + if (cbp & 8) + transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride); + if (cbp & 4) + transfer_16to8add(pY_Cur + next_block + 8, &data[3 * 64], stride); + if (cbp & 2) + transfer_16to8add(pU_Cur, &data[4 * 64], stride2); + if (cbp & 1) + transfer_16to8add(pV_Cur, &data[5 * 64], stride2); + } stop_transfer_timer(); -} - -/***************************************************************************** - * Module functions - ****************************************************************************/ - -void -MBTransQuantIntra(const MBParam * const pParam, - const FRAMEINFO * const frame, - MACROBLOCK * const pMB, - const uint32_t x_pos, - const uint32_t y_pos, - int16_t data[6 * 64], - int16_t qcoeff[6 * 64]) -{ - - /* Transfer data */ - MBTrans8to16(pParam, frame, pMB, x_pos, y_pos, data); - - /* Perform DCT (and field decision) */ - MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); - - /* Quantize the block */ - MBQuantIntra(pParam, frame, pMB, data, qcoeff); - - /* DeQuantize the block */ - MBDeQuantIntra(pParam, pMB->quant, data, qcoeff); - - /* Perform inverse DCT*/ - MBiDCT(data, 0x3F); - /* Transfer back the data -- Don't add data */ - MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 0, 0x3F); + return (uint8_t) cbp; } - uint8_t -MBTransQuantInter(const MBParam * const pParam, - const FRAMEINFO * const frame, - MACROBLOCK * const pMB, - const uint32_t x_pos, - const uint32_t y_pos, +MBTransQuantInterBVOP(const MBParam * pParam, + FRAMEINFO * frame, + MACROBLOCK * pMB, int16_t data[6 * 64], int16_t qcoeff[6 * 64]) { - uint8_t cbp; - uint32_t limit; + int cbp = 0; + int i; - /* - * There is no MBTrans8to16 for Inter block, that's done in motion compensation - * already - */ +/* there is no MBTrans for Inter block, that's done in motion compensation already */ - /* Perform DCT (and field decision) */ - MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); + start_timer(); + pMB->field_dct = 0; + if ((frame->global_flags & XVID_INTERLACING)) { + pMB->field_dct = MBDecideFieldDCT(data); + } + stop_interlacing_timer(); - /* Set the limit threshold */ - limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0); + MBfDCT(data); - /* Quantize the block */ - cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit); + for (i = 0; i < 6; i++) { + int codedecision = 0; + + int sum = QuantizeInterBlock(&qcoeff[i * 64], &data[i * 64], pMB->quant, pParam->m_quant_type); - /* DeQuantize the block */ - MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp); + if(frame->global_flags & XVID_CARTOON_MODE) { + limit *= 2; + } - /* Perform inverse DCT*/ - MBiDCT(data, cbp); + if ((sum > 2) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0) ) codedecision = 1; + else { + if (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V) { + // dark blocks prevention for direct mode + if ( (qcoeff[i*64] < -1) || (qcoeff[i*64] > 0) ) codedecision = 1; + } else + if (qcoeff[i*64] != 0) codedecision = 1; // not direct mode + } - /* Transfer back the data -- Add the data */ - MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp); + if (codedecision) cbp |= 1 << (5 - i); + } - return(cbp); +/* we don't have to DeQuant, iDCT and Transfer back data for B-frames if we don't reconstruct this frame */ +/* warning: reconstruction not supported yet */ + return (uint8_t) cbp; } -uint8_t -MBTransQuantInterBVOP(const MBParam * pParam, - FRAMEINFO * frame, - MACROBLOCK * pMB, - const uint32_t x_pos, - const uint32_t y_pos, - int16_t data[6 * 64], - int16_t qcoeff[6 * 64]) -{ - uint8_t cbp; - uint32_t limit; - - /* - * There is no MBTrans8to16 for Inter block, that's done in motion compensation - * already - */ - - /* Perform DCT (and field decision) */ - MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); - - /* Set the limit threshold */ - limit = BVOP_TOOSMALL_LIMIT; - - /* Quantize the block */ - cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit); - - /* - * History comment: - * We don't have to DeQuant, iDCT and Transfer back data for B-frames. - * - * BUT some plugins require the original frame to be passed so we have - * to take care of that here - */ - if((pParam->plugin_flags & XVID_REQORIGINAL)) { - - /* DeQuantize the block */ - MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp); - - /* Perform inverse DCT*/ - MBiDCT(data, cbp); +/* permute block and return field dct choice */ - /* Transfer back the data -- Add the data */ - MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp); - } +static uint32_t +MBDecideFieldDCT(int16_t data[6 * 64]) +{ + const uint32_t field = MBFieldTest(data); + if (field) MBFrameToField(data); - return(cbp); + return field; } /* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ + uint32_t MBFieldTest_c(int16_t data[6 * 64]) { @@ -506,25 +397,25 @@ for (i = 0; i < 7; ++i) { for (j = 0; j < 8; ++j) { frame += - abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]); + ABS(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]); frame += - abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]); + ABS(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]); frame += - abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]); + ABS(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]); frame += - abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]); + ABS(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]); field += - abs(data[blocks[i + 1] + lines[i + 1] + j] - + ABS(data[blocks[i + 1] + lines[i + 1] + j] - data[blocks[i] + lines[i] + j]); field += - abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] - + ABS(data[blocks[i + 1] + lines[i + 1] + 8 + j] - data[blocks[i] + lines[i] + 8 + j]); field += - abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] - + ABS(data[blocks[i + 1] + 64 + lines[i + 1] + j] - data[blocks[i] + 64 + lines[i] + j]); field += - abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] - + ABS(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] - data[blocks[i] + 64 + lines[i] + 8 + j]); } } @@ -536,7 +427,7 @@ /* deinterlace Y blocks vertically */ #define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp)) -#define LINE(X,Y) &data[X*64 + Y*8] +#define LINE(X,Y) &data[X*64 + Y*8] void MBFrameToField(int16_t data[6 * 64]) @@ -599,441 +490,3 @@ MOVLINE(LINE(3, 5), LINE(3, 3)); MOVLINE(LINE(3, 3), tmp); } - - - - - -/************************************************************************ - * Trellis based R-D optimal quantization * - * * - * Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net * - * * - ************************************************************************/ - - -static int -dct_quantize_trellis_inter_mpeg_c (int16_t *qcoeff, const int16_t *data, int quant) -{ return 63; } - - -////////////////////////////////////////////////////////// -// -// Trellis-Based quantization -// -// So far I understand this paper: -// -// "Trellis-Based R-D Optimal Quantization in H.263+" -// J.Wen, M.Luttrell, J.Villasenor -// IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000. -// -// we are at stake with a simplified Bellmand-Ford / Dijkstra Single -// Source Shorted Path algo. But due to the underlying graph structure -// ("Trellis"), it can be turned into a dynamic programming algo, -// partially saving the explicit graph's nodes representation. And -// without using a heap, since the open frontier of the DAG is always -// known, and of fixed sized. -// -////////////////////////////////////////////////////////// - - -////////////////////////////////////////////////////////// -// Codes lengths for relevant levels. - - // let's factorize: -static const uint8_t Code_Len0[64] = { - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len1[64] = { - 20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len2[64] = { - 19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len3[64] = { - 18,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len4[64] = { - 17,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len5[64] = { - 16,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len6[64] = { - 15,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len7[64] = { - 13,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len8[64] = { - 11,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len9[64] = { - 12,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len10[64] = { - 12,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len11[64] = { - 12,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len12[64] = { - 11,17,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len13[64] = { - 11,15,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len14[64] = { - 10,12,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len15[64] = { - 10,13,17,19,21,21,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len16[64] = { - 9,12,13,18,18,19,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; -static const uint8_t Code_Len17[64] = { - 8,11,13,14,14,14,15,19,19,19,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len18[64] = { - 7, 9,11,11,13,13,13,15,15,15,16,22,22,22,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len19[64] = { - 5, 7, 9,10,10,11,11,11,11,11,13,14,16,17,17,18,18,18,18,18,18,18,18,20,20,21,21,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; -static const uint8_t Code_Len20[64] = { - 3, 4, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9,10,10,10,10,10,10,10,10,12,12,13,13,12,13,14,15,15, - 15,16,16,16,16,17,17,17,18,18,19,19,19,19,19,19,19,19,21,21,22,22,30,30,30,30,30,30,30,30,30,30 }; - - // a few more table for LAST table: -static const uint8_t Code_Len21[64] = { - 13,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; -static const uint8_t Code_Len22[64] = { - 12,15,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, - 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; -static const uint8_t Code_Len23[64] = { - 10,12,15,15,15,16,16,16,16,17,17,17,17,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,20,20,20, - 20,21,21,21,21,21,21,21,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; -static const uint8_t Code_Len24[64] = { - 5, 7, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,10,10,10,10,10,10,10,10,11,11,11,11,12,12,12, - 12,13,13,13,13,13,13,13,13,14,16,16,16,16,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,19,19}; - - -static const uint8_t * const B16_17_Code_Len[24] = { // levels [1..24] - Code_Len20,Code_Len19,Code_Len18,Code_Len17, - Code_Len16,Code_Len15,Code_Len14,Code_Len13, - Code_Len12,Code_Len11,Code_Len10,Code_Len9, - Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5, - Code_Len4, Code_Len3, Code_Len3 ,Code_Len2, - Code_Len2, Code_Len1, Code_Len1, Code_Len1, -}; - -static const uint8_t * const B16_17_Code_Len_Last[6] = { // levels [1..6] - Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1, -}; - -#define TL(q) 0xfe00/(q*q) - -static const int Trellis_Lambda_Tabs[31] = { - TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7), - TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15), - TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23), - TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31) -}; -#undef TL - -static inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i) -{ - while(i>=0) - if (C[Zigzag[i]]) - return i; - else i--; - return -1; -} - -////////////////////////////////////////////////////////// - -#define DBG 0 - -static uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias, - const uint16_t * Zigzag, int Max, int Lambda) -{ -#if (DBG>0) - const int16_t * const Ref = C + 6*64; - int Last = Max; - while(Last>=0 && C[Zigzag[Last]]==0) Last--; - int Bits = 0; - if (Last>=0) { - Bits = 2; // CBP - int j=0, j0=0; - int Run, Level; - while(j=-24 && Level<=24) Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run]; - else Bits += 30; - } - Level = C[Zigzag[Last]]; - Run = j - j0; - if (Level>=-6 && Level<=6) Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run]; - else Bits += 30; - } - - int Dist = 0; - int i; - for(i=0; i<=Last; ++i) { - int V = C[Zigzag[i]]*Mult; - if (V>0) V += Bias; - else if (V<0) V -= Bias; - V -= Ref[Zigzag[i]]; - Dist += V*V; - } - uint32_t Cost = Lambda*Dist + (Bits<<16); - if (DBG==1) - printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 ); - return Cost; - -#else - return 0; -#endif -} - - -static int -dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero) -{ - - // Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]), - // not quantized one (Out[]). However, it only improves the result *very* - // slightly (~0.01dB), whereas speed drops to crawling level :) - // Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps, - - typedef struct { int16_t Run, Level; } NODE; - - NODE Nodes[65], Last; - uint32_t Run_Costs0[64+1], * const Run_Costs = Run_Costs0 + 1; - - const int Mult = 2*Q; - const int Bias = (Q-1) | 1; - const int Lev0 = Mult + Bias; - const int Lambda = Trellis_Lambda_Tabs[Q-1]; // it's 1/lambda, actually - - int Run_Start = -1; - Run_Costs[-1] = 2<<16; // source (w/ CBP penalty) - uint32_t Min_Cost = 2<<16; - - int Last_Node = -1; - uint32_t Last_Cost = 0; - -#if (DBG>0) - Last.Level = 0; Last.Run = -1; // just initialize to smthg -#endif - - int i, j; - - Non_Zero = Find_Last(Out, Zigzag, Non_Zero); - if (Non_Zero<0) - return -1; - - for(i=0; i<=Non_Zero; i++) - { - const int AC = In[Zigzag[i]]; - const int Level1 = Out[Zigzag[i]]; - const int Dist0 = Lambda* AC*AC; - uint32_t Best_Cost = 0xf0000000; - Last_Cost += Dist0; - - if ((uint32_t)(Level1+1)<3) // very specialized loop for -1,0,+1 - { - int dQ; - int Run; - - if (AC<0) { - Nodes[i].Level = -1; - dQ = Lev0 + AC; - } else { - Nodes[i].Level = 1; - dQ = Lev0 - AC; - } - const uint32_t Cost0 = Lambda*dQ*dQ; - - Nodes[i].Run = 1; - Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0; - for(Run=i-Run_Start; Run>0; --Run) - { - const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; - const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16); - // TODO: what about tie-breaks? Should we favor short runs or - // long runs? Although the error is the same, it would not be - // spread the same way along high and low frequencies... - if (Cost>12 ); - else if (j>Run_Start && j>12 ); - else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 ); - else printf( " - |" ); - } - printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run ); - printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run ); - printf( " AC:%3.0d Dist0:%3d Dist(%d)=%d", AC, Dist0>>12, Nodes[i].Level, Cost0>>12 ); - printf( "\n" ); - } - } - else // "big" levels - { - const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last; - int Level2; - int dQ1, dQ2; - int Run; - - if (Level1>1) { - dQ1 = Level1*Mult-AC + Bias; - dQ2 = dQ1 - Mult; - Level2 = Level1-1; - Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0; - Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0; - Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0; - Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0; - } - else { // Level1<-1 - dQ1 = Level1*Mult-AC - Bias; - dQ2 = dQ1 + Mult; - Level2 = Level1 + 1; - Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0; - Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0; - Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0; - Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0; - } - const uint32_t Dist1 = Lambda*dQ1*dQ1; - const uint32_t Dist2 = Lambda*dQ2*dQ2; - const int dDist21 = Dist2-Dist1; - - for(Run=i-Run_Start; Run>0; --Run) - { - const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run]; - -// for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following: -// if (Cost_Base>=Best_Cost) continue; - - uint32_t Cost1, Cost2; - int bLevel; - - Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16); - Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21; - - if (Cost2>12 ); - else if (j>Run_Start && j>12 ); - else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 ); - else printf( " - |" ); - } - printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run ); - printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run ); - printf( " AC:%3.0d Dist0:%3d Dist(%2d):%3d Dist(%2d):%3d", AC, Dist0>>12, Level1, Dist1>>12, Level2, Dist2>>12 ); - printf( "\n" ); - } - } - - Run_Costs[i] = Best_Cost; - - if (Best_Cost < Min_Cost + Dist0) { - Min_Cost = Best_Cost; - Run_Start = i; - } - else - { - // as noticed by Michael Niedermayer (michaelni at gmx.at), there's - // a code shorter by 1 bit for a larger run (!), same level. We give - // it a chance by not moving the left barrier too much. - while( Run_Costs[Run_Start]>Min_Cost+(1<<16) ) - Run_Start++; - - // spread on preceding coeffs the cost incurred by skipping this one - for(j=Run_Start; j " ); - for(i=0; i<=Non_Zero; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] ); - printf( "\n" ); - } - } - - if (Last_Node<0) - return -1; - - // reconstruct optimal sequence backward with surviving paths - bzero(Out, 64*sizeof(*Out)); - Out[Zigzag[Last_Node]] = Last.Level; - i = Last_Node - Last.Run; - while(i>=0) { - Out[Zigzag[i]] = Nodes[i].Level; - i -= Nodes[i].Run; - } - - if (DBG) { - uint32_t Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda); - if (DBG==1) { - printf( "<= " ); - for(i=0; i<=Last_Node; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] ); - printf( "\n--------------------------------\n" ); - } - if (Cost>Last_Cost) printf( "!!! %u > %u\n", Cost, Last_Cost ); - } - return Last_Node; -} - -#undef DBG