--- encoder.c 2002/12/04 12:31:18 1.76.2.24 +++ encoder.c 2002/09/04 21:16:02 1.77 @@ -3,6 +3,8 @@ * XVID MPEG-4 VIDEO CODEC * - Encoder main module - * + * Copyright(C) 2002 Michael Militzer + * * This program 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 @@ -28,22 +30,6 @@ * ****************************************************************************/ -/***************************************************************************** - * - * History - * - * 10.07.2002 added BFRAMES_DEC_DEBUG support - * MinChen - * 20.06.2002 bframe patch - * 08.05.2002 fix some problem in DEBUG mode; - * MinChen - * 14.04.2002 added FrameCodeB() - * - * $Id: encoder.c,v 1.76.2.24 2002/12/04 12:31:18 suxen_drol Exp $ - * - ****************************************************************************/ - - #include #include #include @@ -54,8 +40,10 @@ #include "global.h" #include "utils/timer.h" #include "image/image.h" +#ifdef BFRAMES #include "image/font.h" #include "motion/sad.h" +#endif #include "motion/motion.h" #include "bitstream/cbp.h" #include "utils/mbfunctions.h" @@ -68,6 +56,9 @@ #include "quant/quant_matrix.h" #include "utils/mem_align.h" +#ifdef _SMP +#include "motion/smp_motion_est.h" +#endif /***************************************************************************** * Local macros ****************************************************************************/ @@ -89,10 +80,12 @@ bool force_inter, bool vol_header); +#ifdef BFRAMES static void FrameCodeB(Encoder * pEnc, FRAMEINFO * frame, Bitstream * bs, uint32_t * pBits); +#endif /***************************************************************************** * Local data @@ -107,6 +100,13 @@ }; +static void __inline +image_null(IMAGE * image) +{ + image->y = image->u = image->v = NULL; +} + + /***************************************************************************** * Encoder creation * @@ -224,7 +224,11 @@ pEnc->mbParam.m_quant_type = H263_QUANT; - pEnc->fMvPrevSigma = -1; +#ifdef _SMP + pEnc->mbParam.num_threads = MIN(pParam->num_threads, MAXNUMTHREADS); +#endif + + pEnc->sStat.fMvPrevSigma = -1; /* Fill rate control parameters */ @@ -258,11 +262,11 @@ #ifdef _DEBUG_PSNR image_null(&pEnc->sOriginal); #endif - +#ifdef BFRAMES image_null(&pEnc->f_refh); image_null(&pEnc->f_refv); image_null(&pEnc->f_refhv); - +#endif image_null(&pEnc->current->image); image_null(&pEnc->reference->image); image_null(&pEnc->vInterH); @@ -277,7 +281,7 @@ pEnc->mbParam.edged_height) < 0) goto xvid_err_memory3; #endif - +#ifdef BFRAMES if (image_create (&pEnc->f_refh, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height) < 0) @@ -290,7 +294,7 @@ (&pEnc->f_refhv, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height) < 0) goto xvid_err_memory3; - +#endif if (image_create (&pEnc->current->image, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height) < 0) @@ -323,11 +327,11 @@ /* B Frames specific init */ +#ifdef BFRAMES pEnc->global = pParam->global; pEnc->mbParam.max_bframes = pParam->max_bframes; pEnc->bquant_ratio = pParam->bquant_ratio; - pEnc->bquant_offset = pParam->bquant_offset; pEnc->frame_drop_ratio = pParam->frame_drop_ratio; pEnc->bframes = NULL; @@ -402,11 +406,13 @@ pEnc->queue_tail = 0; pEnc->queue_size = 0; - pEnc->mbParam.m_stamp = 0; + pEnc->mbParam.m_seconds = 0; + pEnc->mbParam.m_ticks = 0; pEnc->m_framenum = 0; - pEnc->current->stamp = 0; - pEnc->reference->stamp = 0; + pEnc->last_pframe = 0; + pEnc->last_sync = 0; +#endif pParam->handle = (void *) pEnc; @@ -425,7 +431,7 @@ /* * We handle all XVID_ERR_MEMORY here, this makes the code lighter */ - +#ifdef BFRAMES xvid_err_memory5: @@ -459,18 +465,22 @@ xvid_free(pEnc->bframes); } +#endif + xvid_err_memory3: #ifdef _DEBUG_PSNR image_destroy(&pEnc->sOriginal, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); #endif +#ifdef BFRAMES image_destroy(&pEnc->f_refh, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); image_destroy(&pEnc->f_refv, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); image_destroy(&pEnc->f_refhv, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); +#endif image_destroy(&pEnc->current->image, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); @@ -515,11 +525,14 @@ int encoder_destroy(Encoder * pEnc) { +#ifdef BFRAMES int i; +#endif ENC_CHECK(pEnc); /* B Frames specific */ +#ifdef BFRAMES if (pEnc->mbParam.max_bframes > 0) { for (i = 0; i < pEnc->mbParam.max_bframes; i++) { @@ -549,6 +562,7 @@ xvid_free(pEnc->bframes); } +#endif /* All images, reference, current etc ... */ @@ -566,14 +580,14 @@ pEnc->mbParam.edged_height); image_destroy(&pEnc->vInterHVf, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); - +#ifdef BFRAMES image_destroy(&pEnc->f_refh, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); image_destroy(&pEnc->f_refv, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); image_destroy(&pEnc->f_refhv, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); - +#endif #ifdef _DEBUG_PSNR image_destroy(&pEnc->sOriginal, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height); @@ -593,15 +607,31 @@ } -static __inline void inc_frame_num(Encoder * pEnc) +void inc_frame_num(Encoder * pEnc) { - pEnc->current->stamp = pEnc->mbParam.m_stamp; // first frame is zero - pEnc->mbParam.m_stamp += pEnc->mbParam.fincr; + pEnc->mbParam.m_ticks += pEnc->mbParam.fincr; + +#ifdef BFRAMES + pEnc->mbParam.m_ticks = pEnc->mbParam.m_ticks % pEnc->mbParam.fbase; + if (pEnc->mbParam.m_ticks < pEnc->last_sync) + pEnc->mbParam.m_seconds = 1; // more than 1 second since last I or P is not supported. + else + pEnc->mbParam.m_seconds = 0; + + if (pEnc->current->coding_type != B_VOP) + pEnc->last_sync = pEnc->mbParam.m_ticks; +#else + + pEnc->mbParam.m_seconds = pEnc->mbParam.m_ticks / pEnc->mbParam.fbase; + pEnc->mbParam.m_ticks = pEnc->mbParam.m_ticks % pEnc->mbParam.fbase; + +#endif + } -static __inline void -queue_image(Encoder * pEnc, XVID_ENC_FRAME * pFrame) +#ifdef BFRAMES +void queue_image(Encoder * pEnc, XVID_ENC_FRAME * pFrame) { if (pEnc->queue_size >= pEnc->mbParam.max_bframes) { @@ -617,37 +647,17 @@ start_timer(); if (image_input (&pEnc->queue[pEnc->queue_tail], pEnc->mbParam.width, pEnc->mbParam.height, - pEnc->mbParam.edged_width, pFrame->image, pFrame->stride, pFrame->colorspace, pFrame->general & XVID_INTERLACING)) + pEnc->mbParam.edged_width, pFrame->image, pFrame->colorspace)) return; stop_conv_timer(); pEnc->queue_size++; pEnc->queue_tail = (pEnc->queue_tail + 1) % pEnc->mbParam.max_bframes; } - -static __inline void -set_timecodes(FRAMEINFO* pCur,FRAMEINFO *pRef, int32_t time_base) -{ - - pCur->ticks = (int32_t)pCur->stamp % time_base; - pCur->seconds = ((int32_t)pCur->stamp / time_base) - ((int32_t)pRef->stamp / time_base) ; - - //HEAVY DEBUG OUTPUT remove when timecodes prove to be stable - -/* fprintf(stderr,"WriteVop: %d - %d \n", - ((int32_t)pCur->stamp / time_base), ((int32_t)pRef->stamp / time_base)); - fprintf(stderr,"set_timecodes: VOP %1d stamp=%lld ref_stamp=%lld base=%d\n", - pCur->coding_type, pCur->stamp, pRef->stamp, time_base); - fprintf(stderr,"set_timecodes: VOP %1d seconds=%d ticks=%d (ref-sec=%d ref-tick=%d)\n", - pCur->coding_type, pCur->seconds, pCur->ticks, pRef->seconds, pRef->ticks); - -*/ -} - - - +#endif +#ifdef BFRAMES /***************************************************************************** * IPB frame encoder entry point * @@ -664,10 +674,9 @@ { uint16_t x, y; Bitstream bs; - uint32_t bits, mode; + uint32_t bits; int input_valid = 1; - int bframes_count = 0; #ifdef _DEBUG_PSNR float psnr; @@ -708,14 +717,11 @@ SWAP(pEnc->current, pEnc->bframes[pEnc->bframenum_tail]); FrameCodeP(pEnc, &bs, &bits, 1, 0); - bframes_count = 0; BitstreamPad(&bs); pFrame->length = BitstreamLength(&bs); pFrame->intra = 0; - emms(); - return XVID_ERR_OK; } @@ -729,48 +735,33 @@ BitstreamPad(&bs); pFrame->length = BitstreamLength(&bs); - pFrame->intra = 2; + pFrame->intra = 0; if (input_valid) queue_image(pEnc, pFrame); - emms(); - return XVID_ERR_OK; } if (pEnc->bframenum_head > 0) { pEnc->bframenum_head = pEnc->bframenum_tail = 0; - /* write an empty marker to the bitstream. - - for divx5 decoder compatibility, this marker must consist - of a not-coded p-vop, with a time_base of zero, and time_increment - indentical to the future-referece frame. - */ - if ((pEnc->global & XVID_GLOBAL_PACKED)) { - int tmp; DPRINTF(DPRINTF_DEBUG,"*** EMPTY bf: head=%i tail=%i queue: head=%i tail=%i size=%i", pEnc->bframenum_head, pEnc->bframenum_tail, pEnc->queue_head, pEnc->queue_tail, pEnc->queue_size); - BitstreamPad(&bs); - - tmp = pEnc->current->seconds; - pEnc->current->seconds = 0; /* force time_base = 0 */ BitstreamWriteVopHeader(&bs, &pEnc->mbParam, pEnc->current, 0); - pEnc->current->seconds = tmp; + BitstreamPad(&bs); + BitstreamPutBits(&bs, 0x7f, 8); pFrame->length = BitstreamLength(&bs); - pFrame->intra = 4; + pFrame->intra = 0; if (input_valid) queue_image(pEnc, pFrame); - emms(); - return XVID_ERR_OK; } } @@ -801,11 +792,8 @@ start_timer(); if (image_input (&pEnc->current->image, pEnc->mbParam.width, pEnc->mbParam.height, - pEnc->mbParam.edged_width, pFrame->image, pFrame->stride, pFrame->colorspace, pFrame->general & XVID_INTERLACING)) - { - emms(); + pEnc->mbParam.edged_width, pFrame->image, pFrame->colorspace)) return XVID_ERR_FORMAT; - } stop_conv_timer(); // queue input frame, and dequue next image @@ -828,28 +816,33 @@ pEnc->queue_head = (pEnc->queue_head + 1) % pEnc->mbParam.max_bframes; pEnc->queue_size--; - } else { - - /* if nothing was encoded, write an 'ignore this frame' flag - to the bitstream */ - - if (BitstreamPos(&bs) == 0) { + } else if (BitstreamPos(&bs) == 0) { - DPRINTF(DPRINTF_DEBUG,"*** SKIP bf: head=%i tail=%i queue: head=%i tail=%i size=%i", + DPRINTF(DPRINTF_DEBUG,"*** SKIP bf: head=%i tail=%i queue: head=%i tail=%i size=%i", pEnc->bframenum_head, pEnc->bframenum_tail, pEnc->queue_head, pEnc->queue_tail, pEnc->queue_size); - BitstreamPutBits(&bs, 0x7f, 8); - pFrame->intra = 5; - } + pFrame->intra = 0; + + BitstreamWriteVopHeader(&bs, &pEnc->mbParam, pEnc->current, 0); // write N_VOP + BitstreamPad(&bs); + pFrame->length = BitstreamLength(&bs); + + return XVID_ERR_OK; + + } else { pFrame->length = BitstreamLength(&bs); - emms(); return XVID_ERR_OK; } pEnc->flush_bframes = 0; + /* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% + * Well there was a separation here so i put it in ANSI C + * comment style :-) + * %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% */ + emms(); // only inc frame num, adapt quant, etc. if we havent seen it before @@ -873,6 +866,9 @@ pEnc->current->fcode = pEnc->mbParam.m_fcode; pEnc->current->bcode = pEnc->mbParam.m_fcode; + pEnc->current->seconds = pEnc->mbParam.m_seconds; + pEnc->current->ticks = pEnc->mbParam.m_ticks; + inc_frame_num(pEnc); #ifdef _DEBUG_PSNR @@ -884,7 +880,7 @@ if ((pEnc->global & XVID_GLOBAL_DEBUG)) { image_printf(&pEnc->current->image, pEnc->mbParam.edged_width, pEnc->mbParam.height, 5, 5, - "%i if:%i st:%i", pEnc->m_framenum++, pEnc->iFrameNum, pEnc->current->stamp); + "%i if:%i st:%i:%i", pEnc->m_framenum++, pEnc->iFrameNum, pEnc->current->seconds, pEnc->current->ticks); } /* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% @@ -926,33 +922,17 @@ /* %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% * ivop/pvop/bvop selection * %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% */ - pEnc->iFrameNum++; + if (pEnc->iFrameNum == 0 || pFrame->intra == 1 || pEnc->bframenum_dx50bvop >= 0 || (pFrame->intra < 0 && pEnc->iMaxKeyInterval > 0 && pEnc->iFrameNum >= pEnc->iMaxKeyInterval) - || 2 == (mode = MEanalysis(&pEnc->reference->image, pEnc->current, - &pEnc->mbParam, pEnc->iMaxKeyInterval, - (pFrame->intra < 0) ? pEnc->iFrameNum : 0, - bframes_count++))) { - + || image_mad(&pEnc->reference->image, &pEnc->current->image, + pEnc->mbParam.edged_width, pEnc->mbParam.width, + pEnc->mbParam.height) > 30) { /* * This will be coded as an Intra Frame */ - if ((pEnc->current->global_flags & XVID_QUARTERPEL)) - pEnc->mbParam.m_quarterpel = 1; - else - pEnc->mbParam.m_quarterpel = 0; - - if (pEnc->current->global_flags & XVID_MPEGQUANT) pEnc->mbParam.m_quant_type = MPEG4_QUANT; - - if ((pEnc->current->global_flags & XVID_CUSTOM_QMATRIX) > 0) { - if (pFrame->quant_intra_matrix != NULL) - set_intra_matrix(pFrame->quant_intra_matrix); - if (pFrame->quant_inter_matrix != NULL) - set_inter_matrix(pFrame->quant_inter_matrix); - } - DPRINTF(DPRINTF_DEBUG,"*** IFRAME bf: head=%i tail=%i queue: head=%i tail=%i size=%i", pEnc->bframenum_head, pEnc->bframenum_tail, @@ -974,14 +954,12 @@ image_printf(&pEnc->current->image, pEnc->mbParam.edged_width, pEnc->mbParam.height, 5, 100, "DX50 BVOP->PVOP"); } FrameCodeP(pEnc, &bs, &bits, 1, 0); - bframes_count = 0; pFrame->intra = 0; } else { FrameCodeI(pEnc, &bs, &bits); - bframes_count = 0; pFrame->intra = 1; pEnc->bframenum_dx50bvop = -1; @@ -990,7 +968,7 @@ pEnc->flush_bframes = 1; if ((pEnc->global & XVID_GLOBAL_PACKED) && pEnc->bframenum_tail > 0) { - BitstreamPadAlways(&bs); + BitstreamPad(&bs); input_valid = 0; goto ipvop_loop; } @@ -999,8 +977,7 @@ * NB : sequences like "IIBB" decode fine with msfdam but, * go screwy with divx 5.00 */ - } else if (pEnc->bframenum_tail >= pEnc->mbParam.max_bframes || mode != 0) { -// } else if (pFrame->intra == 0 || pEnc->bframenum_tail >= pEnc->mbParam.max_bframes || mode != 0) { + } else if (pEnc->bframenum_tail >= pEnc->mbParam.max_bframes) { /* * This will be coded as a Predicted Frame */ @@ -1014,12 +991,11 @@ } FrameCodeP(pEnc, &bs, &bits, 1, 0); - bframes_count = 0; pFrame->intra = 0; pEnc->flush_bframes = 1; - if ((pEnc->global & XVID_GLOBAL_PACKED) && (pEnc->bframenum_tail > 0)) { - BitstreamPadAlways(&bs); + if ((pEnc->global & XVID_GLOBAL_PACKED)) { + BitstreamPad(&bs); input_valid = 0; goto ipvop_loop; } @@ -1034,21 +1010,24 @@ } if (pFrame->bquant < 1) { - pEnc->current->quant = ((((pEnc->reference->quant + pEnc->current->quant) * - pEnc->bquant_ratio) / 2) + pEnc->bquant_offset)/100; - + pEnc->current->quant = + ((pEnc->reference->quant + + pEnc->current->quant) * pEnc->bquant_ratio) / 200; } else { pEnc->current->quant = pFrame->bquant; } + if (pEnc->current->quant < 1) + pEnc->current->quant = 1; - if (pEnc->current->quant < 1) - pEnc->current->quant = 1; - else if (pEnc->current->quant > 31) - pEnc->current->quant = 31; + if (pEnc->current->quant > 31) + pEnc->current->quant = 31; - DPRINTF(DPRINTF_DEBUG,"*** BFRAME (store) bf: head=%i tail=%i queue: head=%i tail=%i size=%i quant=%i\n", - pEnc->bframenum_head, pEnc->bframenum_tail, - pEnc->queue_head, pEnc->queue_tail, pEnc->queue_size,pEnc->current->quant); + + DPRINTF(DPRINTF_DEBUG,"*** BFRAME (store) bf: head=%i tail=%i queue: head=%i tail=%i size=%i quant=%i\n", + pEnc->bframenum_head, pEnc->bframenum_tail, + pEnc->queue_head, pEnc->queue_tail, pEnc->queue_size,pEnc->current->quant); + + /* store frame into bframe buffer & swap ref back to current */ SWAP(pEnc->current, pEnc->bframes[pEnc->bframenum_tail]); @@ -1056,23 +1035,24 @@ pEnc->bframenum_tail++; -// bframe report by koepi - pFrame->intra = 2; + pFrame->intra = 0; pFrame->length = 0; input_valid = 0; goto bvop_loop; } + pEnc->iFrameNum++; + BitstreamPad(&bs); pFrame->length = BitstreamLength(&bs); if (pResult) { pResult->quant = pEnc->current->quant; - pResult->hlength = pFrame->length - (pEnc->current->sStat.iTextBits / 8); - pResult->kblks = pEnc->current->sStat.kblks; - pResult->mblks = pEnc->current->sStat.mblks; - pResult->ublks = pEnc->current->sStat.ublks; + pResult->hlength = pFrame->length - (pEnc->sStat.iTextBits / 8); + pResult->kblks = pEnc->sStat.kblks; + pResult->mblks = pEnc->sStat.mblks; + pResult->ublks = pEnc->sStat.ublks; } emms(); @@ -1092,13 +1072,15 @@ pFrame->length, pFrame->intra); } + stop_global_timer(); write_timer(); - emms(); return XVID_ERR_OK; } +#endif + /***************************************************************************** @@ -1136,21 +1118,14 @@ pEnc->current->global_flags = pFrame->general; pEnc->current->motion_flags = pFrame->motion; + pEnc->current->seconds = pEnc->mbParam.m_seconds; + pEnc->current->ticks = pEnc->mbParam.m_ticks; pEnc->mbParam.hint = &pFrame->hint; - inc_frame_num(pEnc); - - /* disable alternate scan flag if interlacing is not enabled */ - if ((pEnc->current->global_flags & XVID_ALTERNATESCAN) && - !(pEnc->current->global_flags & XVID_INTERLACING)) - { - pEnc->current->global_flags -= XVID_ALTERNATESCAN; - } - start_timer(); if (image_input (&pEnc->current->image, pEnc->mbParam.width, pEnc->mbParam.height, - pEnc->mbParam.edged_width, pFrame->image, pFrame->stride, pFrame->colorspace, pFrame->general & XVID_INTERLACING) < 0) + pEnc->mbParam.edged_width, pFrame->image, pFrame->colorspace) < 0) return XVID_ERR_FORMAT; stop_conv_timer(); @@ -1169,11 +1144,6 @@ pEnc->current->quant = pFrame->quant; } - if ((pEnc->current->global_flags & XVID_QUARTERPEL)) - pEnc->mbParam.m_quarterpel = 1; - else - pEnc->mbParam.m_quarterpel = 0; - if ((pEnc->current->global_flags & XVID_LUMIMASKING)) { int *temp_dquants = (int *) xvid_malloc(pEnc->mbParam.mb_width * @@ -1257,10 +1227,10 @@ if (pResult) { pResult->quant = pEnc->current->quant; - pResult->hlength = pFrame->length - (pEnc->current->sStat.iTextBits / 8); - pResult->kblks = pEnc->current->sStat.kblks; - pResult->mblks = pEnc->current->sStat.mblks; - pResult->ublks = pEnc->current->sStat.ublks; + pResult->hlength = pFrame->length - (pEnc->sStat.iTextBits / 8); + pResult->kblks = pEnc->sStat.kblks; + pResult->mblks = pEnc->sStat.mblks; + pResult->ublks = pEnc->sStat.ublks; } emms(); @@ -1279,6 +1249,7 @@ DEBUG(temp); #endif + inc_frame_num(pEnc); pEnc->iFrameNum++; stop_global_timer(); @@ -1527,27 +1498,22 @@ pEnc->iFrameNum = 0; pEnc->mbParam.m_rounding_type = 1; pEnc->current->rounding_type = pEnc->mbParam.m_rounding_type; - pEnc->current->quarterpel = pEnc->mbParam.m_quarterpel; pEnc->current->coding_type = I_VOP; BitstreamWriteVolHeader(bs, &pEnc->mbParam, pEnc->current); - +#ifdef BFRAMES #define DIVX501B481P "DivX501b481p" if ((pEnc->global & XVID_GLOBAL_PACKED)) { BitstreamWriteUserData(bs, DIVX501B481P, strlen(DIVX501B481P)); } - -#define XVID_ID "XviD" XVID_BS_VERSION - BitstreamWriteUserData(bs, XVID_ID, strlen(XVID_ID)); - - set_timecodes(pEnc->current,pEnc->reference,pEnc->mbParam.fbase); +#endif BitstreamWriteVopHeader(bs, &pEnc->mbParam, pEnc->current, 1); *pBits = BitstreamPos(bs); - pEnc->current->sStat.iTextBits = 0; - pEnc->current->sStat.kblks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; - pEnc->current->sStat.mblks = pEnc->current->sStat.ublks = 0; + pEnc->sStat.iTextBits = 0; + pEnc->sStat.kblks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; + pEnc->sStat.mblks = pEnc->sStat.ublks = 0; for (y = 0; y < pEnc->mbParam.mb_height; y++) for (x = 0; x < pEnc->mbParam.mb_width; x++) { @@ -1569,14 +1535,16 @@ qcoeff[4*64+0]=0; /* zero, because for INTRA MBs DC value is saved */ qcoeff[5*64+0]=0; } - MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->current->sStat); + MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->sStat); stop_coding_timer(); } emms(); *pBits = BitstreamPos(bs) - *pBits; - pEnc->fMvPrevSigma = -1; + pEnc->sStat.fMvPrevSigma = -1; + pEnc->sStat.iMvSum = 0; + pEnc->sStat.iMvCount = 0; pEnc->mbParam.m_fcode = 2; if (pEnc->current->global_flags & XVID_HINTEDME_GET) { @@ -1588,7 +1556,7 @@ #define INTRA_THRESHOLD 0.5 -#define BFRAME_SKIP_THRESHHOLD 30 +#define BFRAME_SKIP_THRESHHOLD 16 static int FrameCodeP(Encoder * pEnc, @@ -1605,19 +1573,19 @@ int iLimit; int x, y, k; int iSearchRange; - int bIntra, skip_possible; + int bIntra; /* IMAGE *pCurrent = &pEnc->current->image; */ IMAGE *pRef = &pEnc->reference->image; start_timer(); image_setedges(pRef, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, - pEnc->mbParam.width, pEnc->mbParam.height); + pEnc->mbParam.width, pEnc->mbParam.height, + pEnc->current->global_flags & XVID_INTERLACING); stop_edges_timer(); pEnc->mbParam.m_rounding_type = 1 - pEnc->mbParam.m_rounding_type; pEnc->current->rounding_type = pEnc->mbParam.m_rounding_type; - pEnc->current->quarterpel = pEnc->mbParam.m_quarterpel; pEnc->current->fcode = pEnc->mbParam.m_fcode; if (!force_inter) @@ -1632,53 +1600,48 @@ image_interpolate(pRef, &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, - pEnc->mbParam.m_quarterpel, pEnc->current->rounding_type); stop_inter_timer(); } - if (pEnc->current->global_flags & XVID_GMC) { -// printf("Global Motion = %d %d quarterpel=%d\n", pEnc->current->GMC_MV.x, pEnc->current->GMC_MV.y,pEnc->current->quarterpel); - DPRINTF(DPRINTF_HEADER, "Global Motion = %d %d quarterpel=%d\n", pEnc->current->GMC_MV.x, pEnc->current->GMC_MV.y,pEnc->current->quarterpel); - pEnc->current->coding_type = S_VOP; - } else - pEnc->current->coding_type = P_VOP; - start_timer(); if (pEnc->current->global_flags & XVID_HINTEDME_SET) { HintedMESet(pEnc, &bIntra); - if (bIntra == 0) { - pEnc->current->fcode = FindFcode(&pEnc->mbParam, pEnc->current); - MotionEstimationHinted(&pEnc->mbParam, pEnc->current, pEnc->reference, - &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV); - } - } else { +#ifdef _SMP + if (pEnc->mbParam.num_threads > 1) + bIntra = + SMP_MotionEstimation(&pEnc->mbParam, pEnc->current, pEnc->reference, + &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, + iLimit); + else +#endif bIntra = MotionEstimation(&pEnc->mbParam, pEnc->current, pEnc->reference, &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, iLimit); + } stop_motion_timer(); - if (bIntra == 1) return FrameCodeI(pEnc, bs, pBits); - - if ( (pEnc->current->GMC_MV.x == 0) && (pEnc->current->GMC_MV.y == 0) ) - pEnc->current->coding_type = P_VOP; /* no global motion -> no GMC */ + if (bIntra == 1) { + return FrameCodeI(pEnc, bs, pBits); + } + pEnc->current->coding_type = P_VOP; if (vol_header) BitstreamWriteVolHeader(bs, &pEnc->mbParam, pEnc->current); - - set_timecodes(pEnc->current,pEnc->reference,pEnc->mbParam.fbase); BitstreamWriteVopHeader(bs, &pEnc->mbParam, pEnc->current, 1); *pBits = BitstreamPos(bs); - pEnc->current->sStat.iTextBits = pEnc->current->sStat.iMvSum = pEnc->current->sStat.iMvCount = - pEnc->current->sStat.kblks = pEnc->current->sStat.mblks = pEnc->current->sStat.ublks = 0; + pEnc->sStat.iTextBits = 0; + pEnc->sStat.iMvSum = 0; + pEnc->sStat.iMvCount = 0; + pEnc->sStat.kblks = pEnc->sStat.mblks = pEnc->sStat.ublks = 0; for (y = 0; y < pEnc->mbParam.mb_height; y++) { for (x = 0; x < pEnc->mbParam.mb_width; x++) { @@ -1695,7 +1658,6 @@ dct_codes, pEnc->mbParam.width, pEnc->mbParam.height, pEnc->mbParam.edged_width, - pEnc->mbParam.m_quarterpel, pEnc->current->rounding_type); stop_comp_timer(); @@ -1713,10 +1675,9 @@ pMB->field_pred = 0; - if (pMB->mode != MODE_NOT_CODED) - pMB->cbp = - MBTransQuantInter(&pEnc->mbParam, pEnc->current, pMB, x, y, - dct_codes, qcoeff); + pMB->cbp = + MBTransQuantInter(&pEnc->mbParam, pEnc->current, pMB, x, y, + dct_codes, qcoeff); } else { CodeIntraMB(pEnc, pMB); MBTransQuantIntra(&pEnc->mbParam, pEnc->current, pMB, x, y, @@ -1728,77 +1689,63 @@ stop_prediction_timer(); if (pMB->mode == MODE_INTRA || pMB->mode == MODE_INTRA_Q) { - pEnc->current->sStat.kblks++; + pEnc->sStat.kblks++; } else if (pMB->cbp || pMB->mvs[0].x || pMB->mvs[0].y || pMB->mvs[1].x || pMB->mvs[1].y || pMB->mvs[2].x || pMB->mvs[2].y || pMB->mvs[3].x || pMB->mvs[3].y) { - pEnc->current->sStat.mblks++; + pEnc->sStat.mblks++; } else { - pEnc->current->sStat.ublks++; + pEnc->sStat.ublks++; } start_timer(); /* Finished processing the MB, now check if to CODE or SKIP */ - skip_possible = (pMB->cbp == 0) & (pMB->mode == MODE_INTER) & - (pMB->dquant == NO_CHANGE); + if (pMB->cbp == 0 && pMB->mode == MODE_INTER && pMB->mvs[0].x == 0 && + pMB->mvs[0].y == 0) { - if(pEnc->mbParam.m_quarterpel) - { skip_possible &= (pMB->qmvs[0].x == pEnc->current->GMC_MV.x) & (pMB->qmvs[0].y == pEnc->current->GMC_MV.y); - } - else - { skip_possible &= (pMB->mvs[0].x == pEnc->current->GMC_MV.x) & (pMB->mvs[0].y == pEnc->current->GMC_MV.y); - } - - if ( (pMB->mode == MODE_NOT_CODED) || (skip_possible)) { +/* This is a candidate for SKIPping, but check intermediate B-frames first */ -/* This is a candidate for SKIPping, but for P-VOPs check intermediate B-frames first */ - int bSkip = 1; - - if (pEnc->current->coding_type == P_VOP) /* special rule for P-VOP's SKIP */ - for (k=pEnc->bframenum_head; k< pEnc->bframenum_tail; k++) - { - int iSAD; - iSAD = sad16(pEnc->reference->image.y + 16*y*pEnc->mbParam.edged_width + 16*x, - pEnc->bframes[k]->image.y + 16*y*pEnc->mbParam.edged_width + 16*x, +#ifdef BFRAMES + int iSAD=BFRAME_SKIP_THRESHHOLD; + int bSkip=1; + + for (k=pEnc->bframenum_head; k< pEnc->bframenum_tail; k++) + { + iSAD = sad16(pEnc->reference->image.y + 16*y*pEnc->mbParam.edged_width + 16*x, + pEnc->bframes[k]->image.y + 16*y*pEnc->mbParam.edged_width + 16*x, pEnc->mbParam.edged_width,BFRAME_SKIP_THRESHHOLD); - if (iSAD >= BFRAME_SKIP_THRESHHOLD * pMB->quant) - { bSkip = 0; - break; - } - } - - if (!bSkip) - { - VECTOR predMV; - if(pEnc->mbParam.m_quarterpel) { - predMV = get_qpmv2(pEnc->current->mbs, pEnc->mbParam.mb_width, 0, x, y, 0); - pMB->pmvs[0].x = pMB->qmvs[0].x - predMV.x; /* with GMC, qmvs doesn't have to be (0,0)! */ - pMB->pmvs[0].y = pMB->qmvs[0].y - predMV.y; + if (iSAD >= BFRAME_SKIP_THRESHHOLD) + { bSkip = 0; + break; } - else { - predMV = get_pmv2(pEnc->current->mbs, pEnc->mbParam.mb_width, 0, x, y, 0); - pMB->pmvs[0].x = pMB->mvs[0].x - predMV.x; /* with GMC, mvs doesn't have to be (0,0)! */ - pMB->pmvs[0].y = pMB->mvs[0].y - predMV.y; + } + if (!bSkip) + { + if (pEnc->current->global_flags & XVID_GREYSCALE) + { pMB->cbp &= 0x3C; /* keep only bits 5-2 */ + qcoeff[4*64+0]=0; /* zero, because DC for INTRA MBs DC value is saved */ + qcoeff[5*64+0]=0; } - pMB->mode = MODE_INTER; - pMB->cbp = 0; - MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->current->sStat); + MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->sStat); + pMB->cbp = 0x80; /* trick! so cbp!=0, but still nothing is written to bs */ } else - { - pMB->mode = MODE_NOT_CODED; MBSkip(bs); - } - + +#else + MBSkip(bs); /* without B-frames, no precautions are needed */ + +#endif + } else { if (pEnc->current->global_flags & XVID_GREYSCALE) { pMB->cbp &= 0x3C; /* keep only bits 5-2 */ qcoeff[4*64+0]=0; /* zero, because DC for INTRA MBs DC value is saved */ qcoeff[5*64+0]=0; } - MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->current->sStat); + MBCoding(pEnc->current, pMB, qcoeff, bs, &pEnc->sStat); } stop_coding_timer(); @@ -1811,61 +1758,67 @@ HintedMEGet(pEnc, 0); } - if (pEnc->current->sStat.iMvCount == 0) - pEnc->current->sStat.iMvCount = 1; + if (pEnc->sStat.iMvCount == 0) + pEnc->sStat.iMvCount = 1; - fSigma = (float) sqrt((float) pEnc->current->sStat.iMvSum / pEnc->current->sStat.iMvCount); + fSigma = (float) sqrt((float) pEnc->sStat.iMvSum / pEnc->sStat.iMvCount); iSearchRange = 1 << (3 + pEnc->mbParam.m_fcode); if ((fSigma > iSearchRange / 3) - && (pEnc->mbParam.m_fcode <= (3 + pEnc->mbParam.m_quarterpel))) // maximum search range 128 + && (pEnc->mbParam.m_fcode <= 3)) // maximum search range 128 { pEnc->mbParam.m_fcode++; iSearchRange *= 2; } else if ((fSigma < iSearchRange / 6) - && (pEnc->fMvPrevSigma >= 0) - && (pEnc->fMvPrevSigma < iSearchRange / 6) - && (pEnc->mbParam.m_fcode >= (2 + pEnc->mbParam.m_quarterpel))) // minimum search range 16 + && (pEnc->sStat.fMvPrevSigma >= 0) + && (pEnc->sStat.fMvPrevSigma < iSearchRange / 6) + && (pEnc->mbParam.m_fcode >= 2)) // minimum search range 16 { pEnc->mbParam.m_fcode--; iSearchRange /= 2; } - pEnc->fMvPrevSigma = fSigma; + pEnc->sStat.fMvPrevSigma = fSigma; +#ifdef BFRAMES /* frame drop code */ - // DPRINTF(DPRINTF_DEBUG, "kmu %i %i %i", pEnc->current->sStat.kblks, pEnc->current->sStat.mblks, pEnc->current->sStat.ublks); - if (pEnc->current->sStat.kblks + pEnc->current->sStat.mblks < + // DPRINTF(DPRINTF_DEBUG, "kmu %i %i %i", pEnc->sStat.kblks, pEnc->sStat.mblks, pEnc->sStat.ublks); + if (pEnc->sStat.kblks + pEnc->sStat.mblks < (pEnc->frame_drop_ratio * pEnc->mbParam.mb_width * pEnc->mbParam.mb_height) / 100) { - pEnc->current->sStat.kblks = pEnc->current->sStat.mblks = 0; - pEnc->current->sStat.ublks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; + pEnc->sStat.kblks = pEnc->sStat.mblks = 0; + pEnc->sStat.ublks = pEnc->mbParam.mb_width * pEnc->mbParam.mb_height; BitstreamReset(bs); - - set_timecodes(pEnc->current,pEnc->reference,pEnc->mbParam.fbase); BitstreamWriteVopHeader(bs, &pEnc->mbParam, pEnc->current, 0); // copy reference frame details into the current frame pEnc->current->quant = pEnc->reference->quant; pEnc->current->motion_flags = pEnc->reference->motion_flags; pEnc->current->rounding_type = pEnc->reference->rounding_type; - pEnc->current->quarterpel = pEnc->reference->quarterpel; pEnc->current->fcode = pEnc->reference->fcode; pEnc->current->bcode = pEnc->reference->bcode; image_copy(&pEnc->current->image, &pEnc->reference->image, pEnc->mbParam.edged_width, pEnc->mbParam.height); memcpy(pEnc->current->mbs, pEnc->reference->mbs, sizeof(MACROBLOCK) * pEnc->mbParam.mb_width * pEnc->mbParam.mb_height); } +#endif *pBits = BitstreamPos(bs) - *pBits; +#ifdef BFRAMES + pEnc->time_pp = ((int32_t)pEnc->mbParam.fbase - (int32_t)pEnc->last_pframe + (int32_t)pEnc->current->ticks) % + (int32_t)pEnc->mbParam.fbase; + pEnc->last_pframe = pEnc->current->ticks; +#endif + return 0; // inter } -static __inline void +#ifdef BFRAMES +static void FrameCodeB(Encoder * pEnc, FRAMEINFO * frame, Bitstream * bs, @@ -1874,6 +1827,8 @@ int16_t dct_codes[6 * 64]; int16_t qcoeff[6 * 64]; uint32_t x, y; + VECTOR forward; + VECTOR backward; IMAGE *f_ref = &pEnc->reference->image; IMAGE *b_ref = &pEnc->current->image; @@ -1890,36 +1845,35 @@ } #endif - frame->quarterpel = pEnc->mbParam.m_quarterpel; - // forward image_setedges(f_ref, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, pEnc->mbParam.width, - pEnc->mbParam.height); + pEnc->mbParam.height, + frame->global_flags & XVID_INTERLACING); start_timer(); image_interpolate(f_ref, &pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, - pEnc->mbParam.m_quarterpel, 0); + 0); stop_inter_timer(); // backward image_setedges(b_ref, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, pEnc->mbParam.width, - pEnc->mbParam.height); + pEnc->mbParam.height, + frame->global_flags & XVID_INTERLACING); start_timer(); image_interpolate(b_ref, &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, pEnc->mbParam.edged_width, pEnc->mbParam.edged_height, - pEnc->mbParam.m_quarterpel, 0); + 0); stop_inter_timer(); start_timer(); - MotionEstimationBVOP(&pEnc->mbParam, frame, - ((int32_t)(pEnc->current->stamp - frame->stamp)), // time_bp - ((int32_t)(pEnc->current->stamp - pEnc->reference->stamp)), // time_pp - pEnc->reference->mbs, f_ref, + ((int32_t)pEnc->mbParam.fbase + pEnc->last_pframe - frame->ticks) % pEnc->mbParam.fbase, + pEnc->time_pp, + pEnc->reference->mbs, f_ref, &pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, - pEnc->current, b_ref, &pEnc->vInterH, + pEnc->current->mbs, b_ref, &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV); @@ -1931,54 +1885,82 @@ } */ frame->coding_type = B_VOP; - - set_timecodes(frame, pEnc->reference,pEnc->mbParam.fbase); BitstreamWriteVopHeader(bs, &pEnc->mbParam, frame, 1); *pBits = BitstreamPos(bs); - frame->sStat.iTextBits = 0; - frame->sStat.iMvSum = 0; - frame->sStat.iMvCount = 0; - frame->sStat.kblks = frame->sStat.mblks = frame->sStat.ublks = 0; + pEnc->sStat.iTextBits = 0; + pEnc->sStat.iMvSum = 0; + pEnc->sStat.iMvCount = 0; + pEnc->sStat.kblks = pEnc->sStat.mblks = pEnc->sStat.ublks = 0; for (y = 0; y < pEnc->mbParam.mb_height; y++) { + // reset prediction + + forward.x = 0; + forward.y = 0; + backward.x = 0; + backward.y = 0; + for (x = 0; x < pEnc->mbParam.mb_width; x++) { - MACROBLOCK * const mb = &frame->mbs[x + y * pEnc->mbParam.mb_width]; - int direction = pEnc->global & XVID_ALTERNATESCAN ? 2 : 0; + MACROBLOCK *f_mb = + &pEnc->reference->mbs[x + y * pEnc->mbParam.mb_width]; + MACROBLOCK *b_mb = + &pEnc->current->mbs[x + y * pEnc->mbParam.mb_width]; + MACROBLOCK *mb = &frame->mbs[x + y * pEnc->mbParam.mb_width]; // decoder ignores mb when refence block is INTER(0,0), CBP=0 if (mb->mode == MODE_NOT_CODED) { - //mb->mvs[0].x = mb->mvs[0].y = mb->cbp = 0; + mb->mvs[0].x = 0; + mb->mvs[0].y = 0; + + mb->cbp = 0; +#ifdef BFRAMES_DEC_DEBUG + BFRAME_DEBUG +#endif continue; } - if (mb->mode != MODE_DIRECT_NONE_MV) { - MBMotionCompensationBVOP(&pEnc->mbParam, mb, x, y, &frame->image, + MBMotionCompensationBVOP(&pEnc->mbParam, mb, x, y, &frame->image, f_ref, &pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, b_ref, &pEnc->vInterH, &pEnc->vInterV, &pEnc->vInterHV, dct_codes); - if (mb->mode == MODE_DIRECT_NO4V) mb->mode = MODE_DIRECT; - mb->quant = frame->quant; - - mb->cbp = - MBTransQuantInterBVOP(&pEnc->mbParam, frame, mb, dct_codes, qcoeff); + mb->quant = frame->quant; + mb->cbp = + MBTransQuantInterBVOP(&pEnc->mbParam, frame, mb, dct_codes, qcoeff); + //mb->cbp = MBTransQuantBVOP(&pEnc->mbParam, x, y, dct_codes, qcoeff, &frame->image, frame->quant); + + if ( (mb->mode == MODE_DIRECT) && (mb->cbp == 0) + && (mb->deltamv.x == 0) && (mb->deltamv.y == 0) ) { + mb->mode = MODE_DIRECT_NONE_MV; // skipped + } - if ( (mb->mode == MODE_DIRECT) && (mb->cbp == 0) - && (mb->pmvs[3].x == 0) && (mb->pmvs[3].y == 0) ) { - mb->mode = MODE_DIRECT_NONE_MV; // skipped - } +/* update predictors for forward and backward vectors */ + if (mb->mode == MODE_INTERPOLATE || mb->mode == MODE_FORWARD) { + mb->pmvs[0].x = mb->mvs[0].x - forward.x; + mb->pmvs[0].y = mb->mvs[0].y - forward.y; + forward.x = mb->mvs[0].x; + forward.y = mb->mvs[0].y; } + if (mb->mode == MODE_INTERPOLATE || mb->mode == MODE_BACKWARD) { + mb->b_pmvs[0].x = mb->b_mvs[0].x - backward.x; + mb->b_pmvs[0].y = mb->b_mvs[0].y - backward.y; + backward.x = mb->b_mvs[0].x; + backward.y = mb->b_mvs[0].y; + } + +// DPRINTF("%05i : [%i %i] M=%i CBP=%i MVS=%i,%i forward=%i,%i", pEnc->m_framenum, x, y, mb->mode, mb->cbp, mb->mvs[0].x, mb->mvs[0].y, forward.x, forward.y); + #ifdef BFRAMES_DEC_DEBUG BFRAME_DEBUG #endif start_timer(); MBCodingBVOP(mb, qcoeff, frame->fcode, frame->bcode, bs, - &frame->sStat, direction); + &pEnc->sStat); stop_coding_timer(); } } @@ -1997,6 +1979,7 @@ } #endif } +#endif /* in case internal output is needed somewhere... */