ref: 82a492262e7d426e844af64b8f8752b4d8617e9e
dir: /codec/encoder/core/src/encoder_ext.cpp/
/*!
* \copy
* Copyright (c) 2009-2013, Cisco Systems
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*
* \file encoder_ext.c
*
* \brief core encoder for SVC
*
* \date 7/24/2009 Created
*
*************************************************************************************
*/
#include <string.h>
#include <stdlib.h>
#include <assert.h>
#include "encoder.h"
#include "extern.h"
#include "encoder_context.h"
#include "typedefs.h"
#include "wels_const.h"
#include "wels_common_basis.h"
#include "codec_def.h"
#include "param_svc.h"
#include "cpu_core.h"
#include "cpu.h"
#include "utils.h"
#include "svc_enc_frame.h"
#include "svc_enc_golomb.h"
#include "svc_enc_slice_segment.h"
#include "au_set.h"
#include "picture_handle.h"
#include "codec_app_def.h"
#include "svc_base_layer_md.h"
#include "svc_encode_slice.h"
#include "decode_mb_aux.h"
#include "deblocking.h"
#include "rc.h"
#include "ref_list_mgr_svc.h"
#include "md.h"
#include "ls_defines.h"
#include "set_mb_syn_cavlc.h"
#include "crt_util_safe_x.h" // Safe CRT routines like utils for cross platforms
#include "array_stack_align.h"
// for MT, 4/22/2010
#include "slice_multi_threading.h"
#if defined(DYNAMIC_SLICE_ASSIGN) || defined(MT_DEBUG)
#include "measure_time.h"
#endif//DYNAMIC_SLICE_ASSIGN
namespace WelsSVCEnc {
int32_t WelsCodeOnePicPartition( sWelsEncCtx *pCtx,
SLayerBSInfo *pLbi,
int32_t *pNalIdxInLayer,
int32_t* pLayerSize,
int32_t iFirstMbInPartition, // first mb inclusive in partition
int32_t iEndMbInPartition, // end mb exclusive in partition
int32_t iStartSliceIdx
);
/*!
* \brief validate checking in parameter configuration
* \pParam pParam SWelsSvcCodingParam*
* \return successful - 0; otherwise none 0 for failed
*/
int32_t ParamValidation( SWelsSvcCodingParam *pCfg )
{
float fMaxFrameRate = 0.0f;
const float fEpsn = 0.000001f;
int32_t i = 0;
int32_t iLastSpatialWidth = 0;
int32_t iLastSpatialHeight = 0;
float fLastFrameRateIn = 0.0f;
float fLastFrameRateOut = 0.0f;
SDLayerParam *pLastSpatialParam = NULL;
assert( pCfg != NULL );
for (i = 0; i < pCfg->iNumDependencyLayer; ++ i)
{
SDLayerParam *fDlp = &pCfg->sDependencyLayers[i];
if ( fDlp->fOutputFrameRate > fDlp->fInputFrameRate || (fDlp->fInputFrameRate >= -fEpsn && fDlp->fInputFrameRate <= fEpsn)
|| (fDlp->fOutputFrameRate >= -fEpsn && fDlp->fOutputFrameRate <= fEpsn) )
{
#if defined (_DEBUG)
fprintf(stderr, "Invalid settings in input frame rate(%.6f) or output frame rate(%.6f) of layer #%d config file..\n",
fDlp->fInputFrameRate, fDlp->fOutputFrameRate, i);
#endif
return 1;
}
if ( UINT_MAX == GetLogFactor(fDlp->fOutputFrameRate, fDlp->fInputFrameRate) )
{
#if defined (_DEBUG)
fprintf(stderr, "Invalid settings in input frame rate(%.6f) and output frame rate(%.6f) of layer #%d config file: iResult of output frame rate divided by input frame rate should be power of 2(i.e,in/pOut=2^n)..\n",
fDlp->fInputFrameRate, fDlp->fOutputFrameRate, i);
#endif
return 1;
}
}
for (i = 0; i < pCfg->iNumDependencyLayer; ++ i)
{
SDLayerParam *fDlp = &pCfg->sDependencyLayers[i];
if ( fDlp->fInputFrameRate > fMaxFrameRate )
fMaxFrameRate = fDlp->fInputFrameRate;
}
if ( fMaxFrameRate > fEpsn && (fMaxFrameRate - pCfg->fMaxFrameRate > fEpsn || fMaxFrameRate - pCfg->fMaxFrameRate < -fEpsn) )
{
pCfg->fMaxFrameRate = fMaxFrameRate;
}
for (i = 0; i < pCfg->iNumDependencyLayer; ++ i)
{
SDLayerParam *fDlp = &pCfg->sDependencyLayers[i];
pLastSpatialParam = fDlp;
iLastSpatialWidth = fDlp->iFrameWidth;
iLastSpatialHeight = fDlp->iFrameHeight;
fLastFrameRateIn = fDlp->fInputFrameRate;
fLastFrameRateOut = fDlp->fOutputFrameRate;
}
return 0;
}
int32_t ParamValidationExt( void *pParam )
{
SWelsSvcCodingParam *pCodingParam = (SWelsSvcCodingParam *)pParam;
int8_t i = 0;
int32_t iIdx = 0;
assert ( pCodingParam != NULL );
if ( NULL == pCodingParam )
return 1;
if ( pCodingParam->iNumDependencyLayer < 1 || pCodingParam->iNumDependencyLayer > MAX_DEPENDENCY_LAYER ){
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), monitor invalid pCodingParam->iNumDependencyLayer: %d!\n", pCodingParam->iNumDependencyLayer);
#endif//#if _DEBUG
return 1;
}
if ( pCodingParam->iNumTemporalLayer < 1 || pCodingParam->iNumTemporalLayer > MAX_TEMPORAL_LEVEL ){
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), monitor invalid pCodingParam->iNumTemporalLayer: %d!\n", pCodingParam->iNumTemporalLayer);
#endif//#if _DEBUG
return 1;
}
if ( pCodingParam->uiGopSize < 1 || pCodingParam->uiGopSize > MAX_GOP_SIZE ){
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), monitor invalid pCodingParam->uiGopSize: %d!\n", pCodingParam->uiGopSize);
#endif//#if _DEBUG
return 1;
}
if ( pCodingParam->uiIntraPeriod && pCodingParam->uiIntraPeriod < pCodingParam->uiGopSize )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), uiIntraPeriod(%d) should be not less than that of uiGopSize(%d) or -1 specified!\n",
pCodingParam->uiIntraPeriod, pCodingParam->uiGopSize);
#endif//#if _DEBUG
return 1;
}
if ( pCodingParam->uiIntraPeriod && (pCodingParam->uiIntraPeriod & (pCodingParam->uiGopSize-1)) != 0 )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), uiIntraPeriod(%d) should be multiple of uiGopSize(%d) or -1 specified!\n",
pCodingParam->uiIntraPeriod, pCodingParam->uiGopSize);
#endif//#if _DEBUG
return 1;
}
#ifdef MT_ENABLED
//about iMultipleThreadIdc, bDeblockingParallelFlag, iLoopFilterDisableIdc, & uiSliceMode
// (1) Single Thread
// if (THREAD==1)//single thread
// no parallel_deblocking: bDeblockingParallelFlag = 0;
// (2) Multi Thread: see uiSliceMode decision
if ( pCodingParam->iMultipleThreadIdc == 1 )
{
//now is single thread. no parallel deblocking, set flag=0
pCodingParam->bDeblockingParallelFlag = false;
}
else
{
pCodingParam->bDeblockingParallelFlag = true;
}
#else
pCodingParam->bDeblockingParallelFlag = false;
#endif//MT_ENABLED
for ( i = 0; i < pCodingParam->iNumDependencyLayer; ++ i ){
SDLayerParam *fDlp = &pCodingParam->sDependencyLayers[i];
const int32_t kiPicWidth = fDlp->iFrameWidth;
const int32_t kiPicHeight= fDlp->iFrameHeight;
int32_t iMbWidth = 0;
int32_t iMbHeight = 0;
int32_t iMbNumInFrame = 0;
int32_t iMaxSliceNum = MAX_SLICES_NUM;
if ( kiPicWidth <= 0 || kiPicHeight <= 0 ){
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid %d x %d in dependency layer settings!\n", kiPicWidth, kiPicHeight);
#endif//#if _DEBUG
return 1;
}
if ( (kiPicWidth & 0x0F) != 0 || (kiPicHeight & 0x0F) != 0 ){
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), in layer #%d iWidth x iHeight(%d x %d) both should be multiple of 16, can not support with arbitrary size currently!\n", i, kiPicWidth, kiPicHeight);
#endif//#if _DEBUG
return 1;
}
if ( fDlp->sMso.uiSliceMode >= SM_RESERVED ){
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceMode (%d) settings!\n", fDlp->sMso.uiSliceMode );
#endif//#if _DEBUG
return 1;
}
//check pSlice settings under multi-pSlice
if ( kiPicWidth<=16 && kiPicHeight<=16 ){
//only have one MB, set to single_slice
fDlp->sMso.uiSliceMode = SM_SINGLE_SLICE;
}
switch ( fDlp->sMso.uiSliceMode )
{
case SM_SINGLE_SLICE:
fDlp->sMso.sSliceArgument.iSliceNum = 1;
fDlp->sMso.sSliceArgument.uiSliceSizeConstraint = 0;
fDlp->sMso.sSliceArgument.iSliceNum = 0;
for (iIdx=0; iIdx<MAX_SLICES_NUM;iIdx++)
{
fDlp->sMso.sSliceArgument.uiSliceMbNum[iIdx] = 0;
}
break;
case SM_FIXEDSLCNUM_SLICE:
{
fDlp->sMso.sSliceArgument.uiSliceSizeConstraint = 0;
iMbWidth = (kiPicWidth+15)>>4;
iMbHeight = (kiPicHeight+15)>>4;
iMbNumInFrame = iMbWidth * iMbHeight;
iMaxSliceNum = MAX_SLICES_NUM;
if ( fDlp->sMso.sSliceArgument.iSliceNum <= 0
|| fDlp->sMso.sSliceArgument.iSliceNum > iMaxSliceNum )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceNum (%d) settings!\n", fDlp->sMso.sSliceArgument.iSliceNum );
#endif//#if _DEBUG
return 1;
}
if (fDlp->sMso.sSliceArgument.iSliceNum == 1)
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), uiSliceNum(%d) you set for SM_FIXEDSLCNUM_SLICE, now turn to SM_SINGLE_SLICE type!\n", fDlp->sMso.sSliceArgument.iSliceNum );
#endif//#if _DEBUG
fDlp->sMso.uiSliceMode = SM_SINGLE_SLICE;
break;
}
if (pCodingParam->bEnableRc) // multiple slices verify with gom
{
//check uiSliceNum
GomValidCheckSliceNum( iMbWidth, iMbHeight, (int32_t*)&fDlp->sMso.sSliceArgument.iSliceNum );
assert(fDlp->sMso.sSliceArgument.iSliceNum > 1);
//set uiSliceMbNum with current uiSliceNum
GomValidCheckSliceMbNum( iMbWidth, iMbHeight, &fDlp->sMso.sSliceArgument );
}
else if ( !CheckFixedSliceNumMultiSliceSetting( iMbNumInFrame, &fDlp->sMso.sSliceArgument ) ) // verify interleave mode settings
{//check uiSliceMbNum with current uiSliceNum
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceMbNum (%d) settings!\n", fDlp->sMso.sSliceArgument.uiSliceMbNum[0] );
#endif//#if _DEBUG
return 1;
}
// considering the coding efficient and performance, iCountMbNum constraint by MIN_NUM_MB_PER_SLICE condition of multi-pSlice mode settting
if ( iMbNumInFrame <= MIN_NUM_MB_PER_SLICE )
{
fDlp->sMso.uiSliceMode = SM_SINGLE_SLICE;
fDlp->sMso.sSliceArgument.iSliceNum = 1;
break;
}
}
break;
case SM_RASTER_SLICE:
{
fDlp->sMso.sSliceArgument.uiSliceSizeConstraint = 0;
iMbWidth = (kiPicWidth+15)>>4;
iMbHeight = (kiPicHeight+15)>>4;
iMbNumInFrame = iMbWidth * iMbHeight;
iMaxSliceNum = MAX_SLICES_NUM;
if ( fDlp->sMso.sSliceArgument.uiSliceMbNum[0] <= 0 )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceMbNum (%d) settings!\n", fDlp->sMso.sSliceArgument.uiSliceMbNum[0] );
#endif//#if _DEBUG
return 1;
}
if ( !CheckRasterMultiSliceSetting( iMbNumInFrame, &fDlp->sMso.sSliceArgument ) ) // verify interleave mode settings
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceMbNum (%d) settings!\n", fDlp->sMso.sSliceArgument.uiSliceMbNum[0] );
#endif//#if _DEBUG
return 1;
}
if ( fDlp->sMso.sSliceArgument.iSliceNum <= 0 || fDlp->sMso.sSliceArgument.iSliceNum > iMaxSliceNum ) // verify interleave mode settings
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceNum (%d) in SM_RASTER_SLICE settings!\n", fDlp->sMso.sSliceArgument.iSliceNum );
#endif//#if _DEBUG
return 1;
}
if (fDlp->sMso.sSliceArgument.iSliceNum == 1)
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), pSlice setting for SM_RASTER_SLICE now turn to SM_SINGLE_SLICE!\n" );
#endif//#if _DEBUG
fDlp->sMso.uiSliceMode = SM_SINGLE_SLICE;
break;
}
#ifdef MT_ENABLED
if (pCodingParam->bEnableRc && fDlp->sMso.sSliceArgument.iSliceNum > 1)
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), WARNING: GOM based RC do not support SM_RASTER_SLICE!\n" );
#endif//#if _DEBUG
}
#endif
// considering the coding efficient and performance, iCountMbNum constraint by MIN_NUM_MB_PER_SLICE condition of multi-pSlice mode settting
if ( iMbNumInFrame <= MIN_NUM_MB_PER_SLICE )
{
fDlp->sMso.uiSliceMode = SM_SINGLE_SLICE;
fDlp->sMso.sSliceArgument.iSliceNum = 1;
break;
}
}
break;
case SM_ROWMB_SLICE:
{
fDlp->sMso.sSliceArgument.uiSliceSizeConstraint = 0;
iMbWidth = (kiPicWidth+15)>>4;
iMbHeight = (kiPicHeight+15)>>4;
iMaxSliceNum = MAX_SLICES_NUM;
if ( iMbHeight > iMaxSliceNum )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceNum (%d) settings more than MAX!\n", iMbHeight );
#endif//#if _DEBUG
return 1;
}
fDlp->sMso.sSliceArgument.iSliceNum = iMbHeight;
if ( fDlp->sMso.sSliceArgument.iSliceNum <= 0 )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceNum (%d) settings!\n", fDlp->sMso.sSliceArgument.iSliceNum );
#endif//#if _DEBUG
return 1;
}
if ( !CheckRowMbMultiSliceSetting( iMbWidth, &fDlp->sMso.sSliceArgument ) ) // verify interleave mode settings
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceMbNum (%d) settings!\n", fDlp->sMso.sSliceArgument.uiSliceMbNum[0] );
#endif//#if _DEBUG
return 1;
}
}
break;
case SM_DYN_SLICE:
{
iMbWidth = (kiPicWidth+15)>>4;
iMbHeight = (kiPicHeight+15)>>4;
if ( fDlp->sMso.sSliceArgument.uiSliceSizeConstraint <= 0 )
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid iSliceSize (%d) settings!\n", fDlp->sMso.sSliceArgument.uiSliceSizeConstraint );
#endif//#if _DEBUG
return 1;
}
// considering the coding efficient and performance, iCountMbNum constraint by MIN_NUM_MB_PER_SLICE condition of multi-pSlice mode settting
if ( iMbWidth * iMbHeight <= MIN_NUM_MB_PER_SLICE )
{
fDlp->sMso.uiSliceMode = SM_SINGLE_SLICE;
fDlp->sMso.sSliceArgument.iSliceNum = 1;
break;
}
}
break;
default:
{
#if defined (_DEBUG)
fprintf(stderr, "ParamValidationExt(), invalid uiSliceMode (%d) settings!\n", pCodingParam->sDependencyLayers[0].sMso.uiSliceMode );
#endif//#if _DEBUG
return 1;
}
break;
}
}
return ParamValidation(pCodingParam);
}
/*!
* \brief acquire count number of layers and NALs based on configurable paramters dependency
* \pParam pCtx sWelsEncCtx*
* \pParam pParam SWelsSvcCodingParam*
* \pParam pCountLayers pointer of count number of layers indeed
* \pParam iCountNals pointer of count number of nals indeed
* \return 0 - successful; otherwise failed
*/
static inline int32_t AcquireLayersNals( sWelsEncCtx **ppCtx, SWelsSvcCodingParam *pParam, int32_t *pCountLayers, int32_t *pCountNals )
{
int32_t iCountNumLayers = 0;
int32_t iCountNumNals = 0;
int32_t iNumDependencyLayers = 0;
int32_t iDIndex = 0;
#if defined(MT_ENABLED) && defined(PACKING_ONE_SLICE_PER_LAYER)
int32_t iNumLayersPack = 0;
#endif//MT_ENABLED && PACKING_ONE_SLICE_PER_LAYER
if ( NULL == pParam || NULL == ppCtx || NULL == *ppCtx )
return 1;
iNumDependencyLayers = pParam->iNumDependencyLayer;
do {
SDLayerParam *pDLayer = &pParam->sDependencyLayers[iDIndex];
// pDLayer->ptr_cfg = pParam;
int32_t iOrgNumNals = iCountNumNals;
//Note: Sep. 2010
//Review this part and suggest no change, since the memory over-use
//(1) counts little to the overall performance
//(2) should not be critial even under mobile case
if ( SM_DYN_SLICE == pDLayer->sMso.uiSliceMode )
{
iCountNumNals += MAX_SLICES_NUM;
// plus prefix NALs
if ( iDIndex == 0 )
iCountNumNals += MAX_SLICES_NUM;
// MAX_SLICES_NUM < MAX_LAYER_NUM_OF_FRAME ensured at svc_enc_slice_segment.h
#if defined(MT_ENABLED) && defined(PACKING_ONE_SLICE_PER_LAYER)
assert(MAX_SLICES_NUM < MAX_LAYER_NUM_OF_FRAME);
// iNumLayersPack += MAX_SLICES_NUM; // do not count it for dynamic slicing mode
#else//!MT_ENABLED || !PACKING_ONE_SLICE_PER_LAYER
assert(iCountNumNals - iOrgNumNals <= MAX_NAL_UNITS_IN_LAYER );
#endif//MT_ENABLED && PACKING_ONE_SLICE_PER_LAYER
}
else /*if ( SM_SINGLE_SLICE != pDLayer->sMso.uiSliceMode )*/
{
const int32_t kiNumOfSlice = GetInitialSliceNum( (pDLayer->iFrameWidth+0x0f)>>4,
(pDLayer->iFrameHeight+0x0f)>>4,
&pDLayer->sMso );
// NEED check iCountNals value in case multiple slices is used
iCountNumNals += kiNumOfSlice; // for pSlice VCL NALs
// plus prefix NALs
if ( iDIndex == 0 )
iCountNumNals += kiNumOfSlice;
#if defined(MT_ENABLED) && defined(PACKING_ONE_SLICE_PER_LAYER)
assert(num_of_slice <= MAX_SLICES_NUM && MAX_SLICES_NUM < MAX_LAYER_NUM_OF_FRAME);
iNumLayersPack += num_of_slice;
#else//!MT_ENABLED || !PACKING_ONE_SLICE_PER_LAYER
assert(iCountNumNals - iOrgNumNals <= MAX_NAL_UNITS_IN_LAYER );
#endif//MT_ENALBED && PACKING_ONE_SLICE_PER_LAYER
if ( kiNumOfSlice > MAX_SLICES_NUM )
{
WelsLog( *ppCtx, WELS_LOG_ERROR, "AcquireLayersNals(), num_of_slice(%d) > MAX_SLICES_NUM(%d) per (iDid= %d, qid= %d) settings!\n",
kiNumOfSlice, MAX_SLICES_NUM, iDIndex, 0 );
return 1;
}
}
#if !defined(MT_ENABLED) || !defined(PACKING_ONE_SLICE_PER_LAYER)
if ( iCountNumNals - iOrgNumNals > MAX_NAL_UNITS_IN_LAYER )
{
WelsLog( *ppCtx, WELS_LOG_ERROR, "AcquireLayersNals(), num_of_nals(%d) > MAX_NAL_UNITS_IN_LAYER(%d) per (iDid= %d, qid= %d) settings!\n",
(iCountNumNals - iOrgNumNals), MAX_NAL_UNITS_IN_LAYER, iDIndex, 0 );
return 1;
}
#endif//!MT_ENABLED) || !PACKING_ONE_SLICE_PER_LAYER
iCountNumLayers ++;
++ iDIndex;
} while(iDIndex < iNumDependencyLayers);
iCountNumNals += 1 + iNumDependencyLayers + (iCountNumLayers<<1) + iCountNumLayers; // plus iCountNumLayers for reserved application
#if defined(MT_ENABLED) && defined(PACKING_ONE_SLICE_PER_LAYER)
iNumLayersPack += 1 + iNumDependencyLayers + (iCountNumLayers<<1);
#endif//MT_ENABLED && PACKING_ONE_SLICE_PER_LAYER
// to check number of layers / nals / slices dependencies, 12/8/2010
#if !defined(MT_ENABLED)
if ( iCountNumLayers > MAX_LAYER_NUM_OF_FRAME )
{
WelsLog( *ppCtx, WELS_LOG_ERROR, "AcquireLayersNals(), iCountNumLayers(%d) > MAX_LAYER_NUM_OF_FRAME(%d)!", iCountNumLayers, MAX_LAYER_NUM_OF_FRAME );
return 1;
}
#else//MT_ENABLED
#if defined(PACKING_ONE_SLICE_PER_LAYER)
if ( iNumLayersPack > MAX_LAYER_NUM_OF_FRAME )
{
WelsLog( *ppCtx, WELS_LOG_ERROR, "AcquireLayersNals(), num_layers_pack_overall(%d) > MAX_LAYER_NUM_OF_FRAME(%d)!", iNumLayersPack, MAX_LAYER_NUM_OF_FRAME );
return 1;
}
#else//!PACKING_ONE_SLICE_PER_LAYER
if ( iCountNumLayers > MAX_LAYER_NUM_OF_FRAME )
{
WelsLog( *ppCtx, WELS_LOG_ERROR, "AcquireLayersNals(), iCountNumLayers(%d) > MAX_LAYER_NUM_OF_FRAME(%d)!", iCountNumLayers, MAX_LAYER_NUM_OF_FRAME );
return 1;
}
#endif//PACKING_ONE_SLICE_PER_LAYER
#endif//!MT_ENABLED
if ( NULL != pCountLayers )
*pCountLayers = iCountNumLayers;
if ( NULL != pCountNals )
*pCountNals = iCountNumNals;
return 0;
}
/*!
* \brief alloc spatial layers pictures (I420 based source pictures)
*/
int32_t AllocSpatialPictures( sWelsEncCtx **ppCtx, SWelsSvcCodingParam *pParam )
{
CMemoryAlign *pMa = (*ppCtx)->pMemAlign;
const int32_t kiDlayerCount = pParam->iNumDependencyLayer;
int32_t iDlayerIndex = 0;
// spatial pictures
iDlayerIndex = 0;
do {
const int32_t kiPicWidth = pParam->sDependencyLayers[iDlayerIndex].iFrameWidth;
const int32_t kiPicHeight = pParam->sDependencyLayers[iDlayerIndex].iFrameHeight;
const uint8_t kuiLayerInTemporal = 2 + WELS_MAX(pParam->sDependencyLayers[iDlayerIndex].iHighestTemporalId, 1);
const uint8_t kuiRefNumInTemporal = kuiLayerInTemporal + pParam->iLTRRefNum;
uint8_t i = 0;
do {
SPicture *pPic = AllocPicture( pMa, kiPicWidth, kiPicHeight, false );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pPic), FreeMemorySvc(ppCtx); *ppCtx = NULL )
(*ppCtx)->pSpatialPic[iDlayerIndex][i] = pPic;
++ i;
} while( i < kuiRefNumInTemporal);
(*ppCtx)->uiSpatialLayersInTemporal[iDlayerIndex] = kuiLayerInTemporal;
(*ppCtx)->uiSpatialPicNum[iDlayerIndex] = kuiRefNumInTemporal;
++ iDlayerIndex;
} while( iDlayerIndex < kiDlayerCount );
return 0;
}
void FreeSpatialPictures( sWelsEncCtx *pCtx )
{
CMemoryAlign *pMa = pCtx->pMemAlign;
int32_t j = 0;
while( j < pCtx->pSvcParam->iNumDependencyLayer )
{
uint8_t i = 0;
uint8_t uiRefNumInTemporal = pCtx->uiSpatialPicNum[j];
while( i < uiRefNumInTemporal )
{
if ( NULL != pCtx->pSpatialPic[j][i] )
{
FreePicture( pMa, &pCtx->pSpatialPic[j][i] );
}
++ i;
}
pCtx->uiSpatialLayersInTemporal[j] = 0;
++ j;
}
}
static void InitMbInfo(sWelsEncCtx * pEnc, SMB * pList, SDqLayer * pLayer, const int32_t kiDlayerId, const int32_t kiMaxMbNum )
{
int32_t iMbWidth = pLayer->iMbWidth;
int32_t iMbHeight = pLayer->iMbHeight;
int32_t iIdx;
int32_t iMbNum = iMbWidth * iMbHeight;
SSliceCtx *pSliceCtx= pLayer->pSliceEncCtx;
uint32_t uiNeighborAvail;
const int32_t kiOffset = (kiDlayerId & 0x01) * kiMaxMbNum;
SMVUnitXY (*pLayerMvUnitBlock4x4)[MB_BLOCK4x4_NUM] = (SMVUnitXY(*)[MB_BLOCK4x4_NUM])(&pEnc->pMvUnitBlock4x4[MB_BLOCK4x4_NUM*kiOffset]);
int8_t (*pLayerRefIndexBlock8x8)[MB_BLOCK8x8_NUM] = (int8_t(*)[MB_BLOCK8x8_NUM])(&pEnc->pRefIndexBlock4x4[MB_BLOCK8x8_NUM*kiOffset]);
for( iIdx = 0; iIdx<iMbNum; iIdx++ ){
BOOL_T bLeft;
BOOL_T bTop;
BOOL_T bLeftTop;
BOOL_T bRightTop;
int32_t iLeftXY, iTopXY, iLeftTopXY, iRightTopXY;
uint8_t uiSliceIdc;
pList[iIdx].iMbX = pEnc->pStrideTab->pMbIndexX[kiDlayerId][iIdx];
pList[iIdx].iMbY = pEnc->pStrideTab->pMbIndexY[kiDlayerId][iIdx];
pList[iIdx].iMbXY = iIdx;
uiSliceIdc = WelsMbToSliceIdc(pSliceCtx, iIdx);
iLeftXY = iIdx - 1;
iTopXY = iIdx - iMbWidth;
iLeftTopXY = iTopXY - 1;
iRightTopXY = iTopXY + 1;
bLeft = (pList[iIdx].iMbX > 0) && (uiSliceIdc == WelsMbToSliceIdc(pSliceCtx, iLeftXY));
bTop = (pList[iIdx].iMbY > 0) && (uiSliceIdc == WelsMbToSliceIdc(pSliceCtx, iTopXY));
bLeftTop = (pList[iIdx].iMbX > 0) && (pList[iIdx].iMbY > 0) && (uiSliceIdc ==
WelsMbToSliceIdc(pSliceCtx, iLeftTopXY));
bRightTop = (pList[iIdx].iMbX < (iMbWidth-1)) && (pList[iIdx].iMbY > 0) && (uiSliceIdc ==
WelsMbToSliceIdc(pSliceCtx, iRightTopXY));
uiNeighborAvail = 0;
if( bLeft ){
uiNeighborAvail |= LEFT_MB_POS;
}
if( bTop ){
uiNeighborAvail |= TOP_MB_POS;
}
if( bLeftTop ){
uiNeighborAvail |= TOPLEFT_MB_POS;
}
if( bRightTop ){
uiNeighborAvail |= TOPRIGHT_MB_POS;
}
pList[iIdx].uiSliceIdc = uiSliceIdc; // merge from svc_hd_opt_b for multiple slices coding
pList[iIdx].uiNeighborAvail = uiNeighborAvail;
uiNeighborAvail = 0;
if(pList[iIdx].iMbX >= BASE_MV_MB_NMB)
uiNeighborAvail |= LEFT_MB_POS;
if(pList[iIdx].iMbX <= (iMbWidth-1-BASE_MV_MB_NMB))
uiNeighborAvail |= RIGHT_MB_POS;
if (pList[iIdx].iMbY >= BASE_MV_MB_NMB)
uiNeighborAvail |= TOP_MB_POS;
if(pList[iIdx].iMbY <= (iMbHeight-1-BASE_MV_MB_NMB))
uiNeighborAvail |= BOTTOM_MB_POS;
pList[iIdx].sMv = pLayerMvUnitBlock4x4[iIdx];
pList[iIdx].pRefIndex = pLayerRefIndexBlock8x8[iIdx];
pList[iIdx].pSadCost = &pEnc->pSadCostMb[iIdx];
pList[iIdx].pIntra4x4PredMode = &pEnc->pIntra4x4PredModeBlocks[iIdx*INTRA_4x4_MODE_NUM];
pList[iIdx].pNonZeroCount = &pEnc->pNonZeroCountBlocks[iIdx*MB_LUMA_CHROMA_BLOCK4x4_NUM];
}
}
int32_t InitMbListD( sWelsEncCtx ** ppCtx)
{
int32_t iNumDlayer = (*ppCtx)->pSvcParam->iNumDependencyLayer;
int32_t iMbSize[MAX_DEPENDENCY_LAYER] = { 0 };
int32_t iOverallMbNum = 0;
int32_t iMbWidth = 0;
int32_t iMbHeight= 0;
int32_t i;
if ( iNumDlayer > MAX_DEPENDENCY_LAYER )
return 1;
for( i=0;i<iNumDlayer;i++ ){
iMbWidth = ((*ppCtx)->pSvcParam->sDependencyLayers[i].iFrameWidth + 15)>>4;
iMbHeight = ((*ppCtx)->pSvcParam->sDependencyLayers[i].iFrameHeight + 15)>>4;
iMbSize[i] = iMbWidth * iMbHeight;
iOverallMbNum += iMbSize[i];
}
(*ppCtx)->ppMbListD = static_cast<SMB **>((*ppCtx)->pMemAlign->WelsMalloc(iNumDlayer * sizeof(SMB *), "ppMbListD"));
(*ppCtx)->ppMbListD[0] = NULL;
WELS_VERIFY_RETURN_PROC_IF(1, (*ppCtx)->ppMbListD==NULL, FreeMemorySvc(ppCtx));
(*ppCtx)->ppMbListD[0] = static_cast<SMB*>((*ppCtx)->pMemAlign->WelsMallocz(iOverallMbNum * sizeof(SMB), "ppMbListD[0]"));
WELS_VERIFY_RETURN_PROC_IF(1, (*ppCtx)->ppMbListD[0]==NULL, FreeMemorySvc(ppCtx));
(*ppCtx)->ppDqLayerList[0]->sMbDataP = (*ppCtx)->ppMbListD[0];
InitMbInfo(*ppCtx, (*ppCtx)->ppMbListD[0], (*ppCtx)->ppDqLayerList[0], 0, iMbSize[iNumDlayer-1]);
for( i=1;i<iNumDlayer;i++ ){
(*ppCtx)->ppMbListD[i] = (*ppCtx)->ppMbListD[i-1] + iMbSize[i-1];
(*ppCtx)->ppDqLayerList[i]->sMbDataP = (*ppCtx)->ppMbListD[i];
InitMbInfo(*ppCtx, (*ppCtx)->ppMbListD[i], (*ppCtx)->ppDqLayerList[i], i, iMbSize[iNumDlayer-1]);
}
return 0;
}
int32_t AllocMbCacheAligned( SMbCache *pMbCache, CMemoryAlign *pMa )
{
pMbCache->pCoeffLevel = (int16_t *)pMa->WelsMalloc(MB_COEFF_LIST_SIZE*sizeof(int16_t), "pMbCache->pCoeffLevel");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pCoeffLevel));
pMbCache->pMemPredMb = (uint8_t *)pMa->WelsMalloc(2*256*sizeof(uint8_t), "pMbCache->pMemPredMb");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pMemPredMb));
pMbCache->pSkipMb = (uint8_t *)pMa->WelsMalloc(384*sizeof(uint8_t), "pMbCache->pSkipMb");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pSkipMb));
pMbCache->pMemPredBlk4 = (uint8_t *)pMa->WelsMalloc(2*16*sizeof(uint8_t), "pMbCache->pMemPredBlk4");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pMemPredBlk4));
pMbCache->pBufferInterPredMe = (uint8_t *)pMa->WelsMalloc(4*640*sizeof(uint8_t), "pMbCache->pBufferInterPredMe");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pBufferInterPredMe));
pMbCache->pPrevIntra4x4PredModeFlag = (bool_t *)pMa->WelsMalloc(16*sizeof(bool_t), "pMbCache->pPrevIntra4x4PredModeFlag");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pPrevIntra4x4PredModeFlag));
pMbCache->pRemIntra4x4PredModeFlag = (int8_t *)pMa->WelsMalloc(16*sizeof(int8_t), "pMbCache->pRemIntra4x4PredModeFlag");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pRemIntra4x4PredModeFlag));
pMbCache->pDct = (SDCTCoeff *)pMa->WelsMalloc(sizeof(SDCTCoeff), "pMbCache->pDct");
WELS_VERIFY_RETURN_IF(1, (NULL==pMbCache->pDct));
return 0;
}
void FreeMbCache( SMbCache *pMbCache, CMemoryAlign *pMa )
{
if ( NULL != pMbCache->pCoeffLevel )
{
pMa->WelsFree( pMbCache->pCoeffLevel, "pMbCache->pCoeffLevel" );
pMbCache->pCoeffLevel = NULL;
}
if ( NULL != pMbCache->pMemPredMb )
{
pMa->WelsFree( pMbCache->pMemPredMb, "pMbCache->pMemPredMb" );
pMbCache->pMemPredMb = NULL;
}
if ( NULL != pMbCache->pSkipMb )
{
pMa->WelsFree( pMbCache->pSkipMb, "pMbCache->pSkipMb" );
pMbCache->pSkipMb = NULL;
}
if ( NULL != pMbCache->pMemPredBlk4 )
{
pMa->WelsFree( pMbCache->pMemPredBlk4, "pMbCache->pMemPredBlk4" );
pMbCache->pMemPredBlk4 = NULL;
}
if ( NULL != pMbCache->pBufferInterPredMe )
{
pMa->WelsFree( pMbCache->pBufferInterPredMe, "pMbCache->pBufferInterPredMe" );
pMbCache->pBufferInterPredMe = NULL;
}
if ( NULL != pMbCache->pPrevIntra4x4PredModeFlag )
{
pMa->WelsFree( pMbCache->pPrevIntra4x4PredModeFlag, "pMbCache->pPrevIntra4x4PredModeFlag" );
pMbCache->pPrevIntra4x4PredModeFlag = NULL;
}
if ( NULL != pMbCache->pRemIntra4x4PredModeFlag )
{
pMa->WelsFree( pMbCache->pRemIntra4x4PredModeFlag, "pMbCache->pRemIntra4x4PredModeFlag" );
pMbCache->pRemIntra4x4PredModeFlag = NULL;
}
if ( NULL != pMbCache->pDct )
{
pMa->WelsFree( pMbCache->pDct, "pMbCache->pDct" );
pMbCache->pDct = NULL;
}
}
/*!
* \brief initialize ppDqLayerList and slicepEncCtx_list due to count number of layers available
* \pParam pCtx sWelsEncCtx*
* \return 0 - successful; otherwise failed
*/
static inline int32_t InitDqLayers( sWelsEncCtx **ppCtx )
{
SWelsSvcCodingParam *pParam = NULL;
SWelsSPS *pSps = NULL;
SSubsetSps *pSubsetSps = NULL;
SWelsPPS *pPps = NULL;
CMemoryAlign *pMa = NULL;
SStrideTables *pStrideTab = NULL;
int32_t iDlayerCount = 0;
int32_t iDlayerIndex = 0;
uint32_t iSpsId = 0;
uint32_t iPpsId = 0;
uint32_t iNumRef = 0;
int32_t iResult = 0;
if ( NULL == ppCtx || NULL == *ppCtx )
return 1;
pMa = (*ppCtx)->pMemAlign;
pParam = (*ppCtx)->pSvcParam;
iDlayerCount = pParam->iNumDependencyLayer;
iNumRef = pParam->iNumRefFrame;
// highest_layers_in_temporal = 1 + WELS_MAX(pParam->iDecompStages, 1);
pStrideTab = (*ppCtx)->pStrideTab;
iDlayerIndex = 0;
while (iDlayerIndex < iDlayerCount)
{
SRefList *pRefList = NULL;
uint32_t i = 0;
const int32_t kiWidth = pParam->sDependencyLayers[iDlayerIndex].iFrameWidth;
const int32_t kiHeight = pParam->sDependencyLayers[iDlayerIndex].iFrameHeight;
int32_t iPicWidth = WELS_ALIGN(kiWidth, MB_WIDTH_LUMA) + (PADDING_LENGTH<<1); // with iWidth of horizon
int32_t iPicChromaWidth = iPicWidth >> 1;
iPicWidth = WELS_ALIGN( iPicWidth, 32 ); // 32(or 16 for chroma below) to match original imp. here instead of iCacheLineSize
iPicChromaWidth = WELS_ALIGN( iPicChromaWidth, 16 );
WelsGetEncBlockStrideOffset( (*ppCtx)->pStrideTab->pStrideEncBlockOffset[iDlayerIndex], iPicWidth, iPicChromaWidth);
// pRef list
pRefList = (SRefList *)pMa->WelsMallocz( sizeof(SRefList), "pRefList" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pRefList), FreeMemorySvc(ppCtx) )
do {
pRefList->pRef[i] = AllocPicture( pMa, kiWidth, kiHeight, true ); // to use actual size of current layer
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pRefList->pRef[i]), FreeMemorySvc(ppCtx) )
++ i;
} while(i < 1 + iNumRef);
pRefList->pNextBuffer = pRefList->pRef[0];
(*ppCtx)->ppRefPicListExt[iDlayerIndex] = pRefList;
++ iDlayerIndex;
}
// for I420 based source spatial pictures
if ( AllocSpatialPictures( ppCtx, pParam ) )
{
FreeMemorySvc( ppCtx );
return 1;
}
iDlayerIndex = 0;
while (iDlayerIndex < iDlayerCount) {
SDqLayer *pDqLayer = NULL;
SDLayerParam *pDlayer = &pParam->sDependencyLayers[iDlayerIndex];
const int32_t kiMbW = (pDlayer->iFrameWidth + 0x0f) >> 4;
const int32_t kiMbH = (pDlayer->iFrameHeight + 0x0f) >> 4;
int32_t iMaxSliceNum = 1;
const int32_t kiSliceNum = GetInitialSliceNum( kiMbW, kiMbH, &pDlayer->sMso );
if ( iMaxSliceNum < kiSliceNum )
iMaxSliceNum = kiSliceNum;
// pDq layers list
pDqLayer = (SDqLayer *)pMa->WelsMallocz( sizeof(SDqLayer), "pDqLayer" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pDqLayer), FreeMemorySvc(ppCtx) )
// for dynamic slicing mode
if ( SM_DYN_SLICE == pDlayer->sMso.uiSliceMode )
{
const int32_t iSize = pParam->iCountThreadsNum * sizeof(int32_t);
pDqLayer->pNumSliceCodedOfPartition = (int32_t *)pMa->WelsMallocz( iSize, "pNumSliceCodedOfPartition" );
pDqLayer->pLastCodedMbIdxOfPartition = (int32_t *)pMa->WelsMallocz( iSize, "pLastCodedMbIdxOfPartition" );
pDqLayer->pLastMbIdxOfPartition = (int32_t *)pMa->WelsMallocz( iSize, "pLastMbIdxOfPartition" );
WELS_VERIFY_RETURN_PROC_IF( 1,
(NULL == pDqLayer->pNumSliceCodedOfPartition ||
NULL == pDqLayer->pLastCodedMbIdxOfPartition ||
NULL == pDqLayer->pLastMbIdxOfPartition),
FreeMemorySvc(ppCtx) )
}
pDqLayer->iMbWidth = kiMbW;
pDqLayer->iMbHeight = kiMbH;
#ifndef MT_ENABLED
if ( SM_DYN_SLICE == pDlayer->sMso.uiSliceMode )//wmalloc pSliceInLayer
{
SSlice *pSlice = NULL;
int32_t iSliceIdx = 0;
//wmalloc AVERSLICENUM_CONSTANT of pDqLayer->sLayerInfo.pSliceInLayer,
//wmalloc AVERSLICENUM_CONSTANT num of pSlice as initialization
//only set value for the first pSlice
pDqLayer->sLayerInfo.pSliceInLayer = (SSlice *)pMa->WelsMallocz( sizeof(SSlice) * iMaxSliceNum, "pSliceInLayer" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pDqLayer->sLayerInfo.pSliceInLayer), FreeMemorySvc(ppCtx) )
{
pSlice = &pDqLayer->sLayerInfo.pSliceInLayer[0];
pSlice->uiSliceIdx = 0;
pSlice->pSliceBsa = &(*ppCtx)->pOut->sBsWrite;
}
while(iSliceIdx < iMaxSliceNum)
{
pSlice = &pDqLayer->sLayerInfo.pSliceInLayer[iSliceIdx];
if ( AllocMbCacheAligned(&pSlice->sMbCacheInfo, pMa) )
{
FreeMemorySvc(ppCtx);
return 1;
}
++ iSliceIdx;
}
}
else
#endif//!MT_ENABLED
{
int32_t iSliceIdx = 0;
pDqLayer->sLayerInfo.pSliceInLayer = (SSlice *)pMa->WelsMallocz( sizeof(SSlice) * iMaxSliceNum, "pSliceInLayer" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pDqLayer->sLayerInfo.pSliceInLayer), FreeMemorySvc(ppCtx) )
if ( iMaxSliceNum > 1 )
{
while (iSliceIdx < iMaxSliceNum) {
SSlice *pSlice = &pDqLayer->sLayerInfo.pSliceInLayer[iSliceIdx];
pSlice->uiSliceIdx = iSliceIdx;
#ifdef MT_ENABLED
if ( pParam->iMultipleThreadIdc > 1 )
pSlice->pSliceBsa = &(*ppCtx)->pSliceBs[iSliceIdx].sBsWrite;
else
pSlice->pSliceBsa = &(*ppCtx)->pOut->sBsWrite;
#else
pSlice->pSliceBsa = &(*ppCtx)->pOut->sBsWrite;
#endif//MT_ENABLED
if ( AllocMbCacheAligned(&pSlice->sMbCacheInfo, pMa) )
{
FreeMemorySvc(ppCtx);
return 1;
}
++ iSliceIdx;
}
}
// fix issue in case single pSlice coding might be inclusive exist in variant spatial layer setting, also introducing multi-pSlice modes
else // only one pSlice
{
SSlice *pSlice = &pDqLayer->sLayerInfo.pSliceInLayer[0];
pSlice->uiSliceIdx = 0;
pSlice->pSliceBsa = &(*ppCtx)->pOut->sBsWrite;
if ( AllocMbCacheAligned(&pSlice->sMbCacheInfo, pMa) )
{
FreeMemorySvc(ppCtx);
return 1;
}
}
}
//deblocking parameters initialization
//target-layer deblocking
pDqLayer->iLoopFilterDisableIdc = pParam->iLoopFilterDisableIdc;
pDqLayer->iLoopFilterAlphaC0Offset = (pParam->iLoopFilterAlphaC0Offset)<<1;
pDqLayer->iLoopFilterBetaOffset = (pParam->iLoopFilterBetaOffset)<<1;
//inter-layer deblocking
pDqLayer->uiDisableInterLayerDeblockingFilterIdc = pParam->iInterLayerLoopFilterDisableIdc;
pDqLayer->iInterLayerSliceAlphaC0Offset = (pParam->iInterLayerLoopFilterAlphaC0Offset)<<1;
pDqLayer->iInterLayerSliceBetaOffset = (pParam->iInterLayerLoopFilterBetaOffset)<<1;
//parallel deblocking
pDqLayer->bDeblockingParallelFlag = pParam->bDeblockingParallelFlag;
//deblocking parameter adjustment
if ( SM_SINGLE_SLICE == pDlayer->sMso.uiSliceMode )
{
//iLoopFilterDisableIdc: will be 0 or 1 under single_slice
if ( 2 == pParam->iLoopFilterDisableIdc )
{
pDqLayer->iLoopFilterDisableIdc = 0;
}
//bDeblockingParallelFlag
pDqLayer->bDeblockingParallelFlag = false;
}
else
{//multi-pSlice
#ifdef MT_ENABLED
if ( 0 == pDqLayer->iLoopFilterDisableIdc )
{
pDqLayer->bDeblockingParallelFlag = false;
}
#endif
}
(*ppCtx)->ppDqLayerList[iDlayerIndex] = pDqLayer;
++ iDlayerIndex;
}
// for dynamically malloc for parameter sets memory instead of maximal items for standard to reduce size, 3/18/2010
if ( &(*ppCtx)->pSvcParam->bMgsT0OnlyStrategy )
{
(*ppCtx)->pPPSArray = (SWelsPPS *)pMa->WelsMalloc( (1+iDlayerCount) * sizeof(SWelsPPS), "pPPSArray" );
}
else
{
(*ppCtx)->pPPSArray = (SWelsPPS *)pMa->WelsMalloc( iDlayerCount * sizeof(SWelsPPS), "pPPSArray" );
}
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pPPSArray), FreeMemorySvc(ppCtx) )
(*ppCtx)->pSpsArray = (SWelsSPS *)pMa->WelsMalloc( sizeof(SWelsSPS), "pSpsArray" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pSpsArray), FreeMemorySvc(ppCtx) )
if ( iDlayerCount > 1 )
{
(*ppCtx)->pSubsetArray = (SSubsetSps *)pMa->WelsMalloc( (iDlayerCount-1) * sizeof(SSubsetSps), "pSubsetArray" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pSubsetArray), FreeMemorySvc(ppCtx) )
}
(*ppCtx)->pDqIdcMap = (SDqIdc *)pMa->WelsMallocz( iDlayerCount * sizeof(SDqIdc), "pDqIdcMap" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pDqIdcMap), FreeMemorySvc(ppCtx) )
iDlayerIndex = 0;
while (iDlayerIndex < iDlayerCount) {
SDqIdc *pDqIdc = &(*ppCtx)->pDqIdcMap[iDlayerIndex];
const bool_t bUseSubsetSps = (iDlayerIndex > BASE_DEPENDENCY_ID);
SDLayerParam *pDlayerParam = &pParam->sDependencyLayers[iDlayerIndex];
pDqIdc->uiSpatialId = iDlayerIndex;
pPps = &(*ppCtx)->pPPSArray[iPpsId];
if ( !bUseSubsetSps )
{
pSps = &(*ppCtx)->pSpsArray[iSpsId];
}
else
{
pSubsetSps = &(*ppCtx)->pSubsetArray[iSpsId];
pSps = &pSubsetSps->pSps;
}
// Need port pSps/pPps initialization due to spatial scalability changed
if ( !bUseSubsetSps )
{
WelsInitSps( pSps, pDlayerParam, pParam->uiIntraPeriod, pParam->iNumRefFrame, iSpsId,
pParam->bEnableFrameCroppingFlag, pParam->bEnableRc );
if( iDlayerCount > 1 )
{
pSps->bConstraintSet0Flag = true;
pSps->bConstraintSet1Flag = true;
pSps->bConstraintSet2Flag = true;
}
}
else
{
WelsInitSubsetSps( pSubsetSps, pDlayerParam, pParam->uiIntraPeriod, pParam->iNumRefFrame, iSpsId,
pParam->bEnableFrameCroppingFlag, pParam->bEnableRc );
}
// initialize pPps
WelsInitPps( pPps, pSps, pSubsetSps, iPpsId, true, bUseSubsetSps );
// Not using FMO in SVC coding so far, come back if need FMO
{
iResult = InitSlicePEncCtx( &(*ppCtx)->pSliceCtxList[iDlayerIndex],
(*ppCtx)->pMemAlign,
false,
pSps->iMbWidth,
pSps->iMbHeight,
&(pDlayerParam->sMso),
pPps );
if ( iResult )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "InitDqLayers(), InitSlicePEncCtx failed(%d)!", iResult );
FreeMemorySvc( ppCtx );
return 1;
}
(*ppCtx)->ppDqLayerList[iDlayerIndex]->pSliceEncCtx = &(*ppCtx)->pSliceCtxList[iDlayerIndex];
}
pDqIdc->iSpsId = iSpsId;
pDqIdc->iPpsId = iPpsId;
(*ppCtx)->sPSOVector.bPpsIdMappingIntoSubsetsps[iPpsId] = bUseSubsetSps;
if ( bUseSubsetSps )
++ iSpsId;
++ iPpsId;
++ (*ppCtx)->iSpsNum;
++ (*ppCtx)->iPpsNum;
++ iDlayerIndex;
}
return 0;
}
int32_t AllocStrideTables( sWelsEncCtx **ppCtx, const int32_t kiNumSpatialLayers )
{
CMemoryAlign *pMa = (*ppCtx)->pMemAlign;
SWelsSvcCodingParam *pParam = (*ppCtx)->pSvcParam;
SStrideTables *pPtr = NULL;
int16_t *pTmpRow = NULL, *pRowX = NULL, *pRowY = NULL, *p = NULL;
uint8_t *pBase = NULL;
uint8_t *pBaseDec = NULL, *pBaseEnc = NULL, *pBaseMbX = NULL, *pBaseMbY = NULL;
struct {
int32_t iMbWidth;
int32_t iCountMbNum; // count number of SMB in each spatial
int32_t iSizeAllMbAlignCache; // cache line size aligned in each spatial
} sMbSizeMap[MAX_DEPENDENCY_LAYER] = {0};
int32_t iLineSizeY[MAX_DEPENDENCY_LAYER][2] = {0};
int32_t iLineSizeUV[MAX_DEPENDENCY_LAYER][2]= {0};
int32_t iMapSpatialIdx[MAX_DEPENDENCY_LAYER][2] = {0};
int32_t iSizeDec = 0;
int32_t iSizeEnc = 0;
int32_t iCountLayersNeedCs[2] = {0};
const int32_t kiUnit1Size = 24 * sizeof(int32_t);
int32_t iUnit2Size = 0;
int32_t iNeedAllocSize = 0;
int32_t iRowSize = 0;
int16_t iMaxMbWidth = 0;
int16_t iMaxMbHeight = 0;
int32_t i = 0;
int32_t iSpatialIdx = 0;
int32_t iTemporalIdx = 0;
int32_t iCntTid = 0;
if ( kiNumSpatialLayers <= 0 || kiNumSpatialLayers > MAX_DEPENDENCY_LAYER)
return 1;
pPtr = (SStrideTables *)pMa->WelsMalloc(sizeof(SStrideTables), "SStrideTables");
if (NULL == pPtr)
return 1;
(*ppCtx)->pStrideTab = pPtr;
iCntTid = pParam->iNumTemporalLayer > 1 ? 2 : 1;
iSpatialIdx = 0;
while (iSpatialIdx < kiNumSpatialLayers) {
const int32_t kiTmpWidth = (pParam->sDependencyLayers[iSpatialIdx].iFrameWidth + 15) >> 4;
const int32_t kiTmpHeight= (pParam->sDependencyLayers[iSpatialIdx].iFrameHeight + 15) >> 4;
int32_t iNumMb = kiTmpWidth * kiTmpHeight;
sMbSizeMap[iSpatialIdx].iMbWidth = kiTmpWidth;
sMbSizeMap[iSpatialIdx].iCountMbNum = iNumMb;
iNumMb *= sizeof(int16_t);
sMbSizeMap[iSpatialIdx].iSizeAllMbAlignCache = iNumMb;
iUnit2Size += iNumMb;
++ iSpatialIdx;
}
// Adaptive size_cs, size_fdec by implementation dependency
iTemporalIdx= 0;
while ( iTemporalIdx < iCntTid )
{
const bool_t kbBaseTemporalFlag = (iTemporalIdx == 0);
iSpatialIdx = 0;
while ( iSpatialIdx < kiNumSpatialLayers )
{
SDLayerParam *fDlp = &pParam->sDependencyLayers[iSpatialIdx];
const int32_t kiWidthPad = WELS_ALIGN( fDlp->iFrameWidth, 16 ) + (PADDING_LENGTH<<1);
iLineSizeY[iSpatialIdx][kbBaseTemporalFlag] = WELS_ALIGN( kiWidthPad, 32 );
iLineSizeUV[iSpatialIdx][kbBaseTemporalFlag]= WELS_ALIGN( (kiWidthPad>>1), 16 );
iMapSpatialIdx[iCountLayersNeedCs[kbBaseTemporalFlag]][kbBaseTemporalFlag] = iSpatialIdx;
++ iCountLayersNeedCs[kbBaseTemporalFlag];
++ iSpatialIdx;
}
++ iTemporalIdx;
}
iSizeDec= kiUnit1Size * (iCountLayersNeedCs[0] + iCountLayersNeedCs[1]);
iSizeEnc= kiUnit1Size * kiNumSpatialLayers;
iNeedAllocSize = iSizeDec + iSizeEnc + (iUnit2Size << 1);
pBase = (uint8_t *)pMa->WelsMalloc( iNeedAllocSize, "pBase" );
if ( NULL == pBase )
{
return 1;
}
pBaseDec= pBase; // iCountLayersNeedCs
pBaseEnc= pBaseDec + iSizeDec; // iNumSpatialLayers
pBaseMbX = pBaseEnc + iSizeEnc; // iNumSpatialLayers
pBaseMbY = pBaseMbX + iUnit2Size; // iNumSpatialLayers
iTemporalIdx= 0;
while ( iTemporalIdx < iCntTid )
{
const bool_t kbBaseTemporalFlag = (iTemporalIdx == 0);
iSpatialIdx = 0;
while ( iSpatialIdx < iCountLayersNeedCs[kbBaseTemporalFlag] )
{
const int32_t kiActualSpatialIdx = iMapSpatialIdx[iSpatialIdx][kbBaseTemporalFlag];
const int32_t kiLumaWidth = iLineSizeY[kiActualSpatialIdx][kbBaseTemporalFlag];
const int32_t kiChromaWidth = iLineSizeUV[kiActualSpatialIdx][kbBaseTemporalFlag];
WelsGetEncBlockStrideOffset( (int32_t *)pBaseDec, kiLumaWidth, kiChromaWidth );
pPtr->pStrideDecBlockOffset[kiActualSpatialIdx][kbBaseTemporalFlag] = (int32_t *)pBaseDec;
pBaseDec+= kiUnit1Size;
++ iSpatialIdx;
}
++ iTemporalIdx;
}
iTemporalIdx= 0;
while ( iTemporalIdx < iCntTid )
{
const bool_t kbBaseTemporalFlag = (iTemporalIdx == 0);
iSpatialIdx = 0;
while (iSpatialIdx < kiNumSpatialLayers)
{
int32_t iMatchIndex = 0;
bool_t bInMap = false;
bool_t bMatchFlag = false;
i = 0;
while ( i < iCountLayersNeedCs[kbBaseTemporalFlag] )
{
const int32_t kiActualIdx = iMapSpatialIdx[i][kbBaseTemporalFlag];
if ( kiActualIdx == iSpatialIdx )
{
bInMap = true;
break;
}
if ( !bMatchFlag )
{
iMatchIndex = kiActualIdx;
bMatchFlag = true;
}
++ i;
}
if ( bInMap )
{
++ iSpatialIdx;
continue;
}
// not in spatial map and assign match one to it
pPtr->pStrideDecBlockOffset[iSpatialIdx][kbBaseTemporalFlag] = pPtr->pStrideDecBlockOffset[iMatchIndex][kbBaseTemporalFlag];
++ iSpatialIdx;
}
++ iTemporalIdx;
}
iSpatialIdx = 0;
while ( iSpatialIdx < kiNumSpatialLayers )
{
const int32_t kiAllocMbSize = sMbSizeMap[iSpatialIdx].iSizeAllMbAlignCache;
pPtr->pStrideEncBlockOffset[iSpatialIdx] = (int32_t *)pBaseEnc;
pPtr->pMbIndexX[iSpatialIdx] = (int16_t *)pBaseMbX;
pPtr->pMbIndexY[iSpatialIdx] = (int16_t *)pBaseMbY;
pBaseEnc += kiUnit1Size;
pBaseMbX += kiAllocMbSize;
pBaseMbY += kiAllocMbSize;
++ iSpatialIdx;
}
while ( iSpatialIdx < MAX_DEPENDENCY_LAYER )
{
pPtr->pStrideDecBlockOffset[iSpatialIdx][0] = NULL;
pPtr->pStrideDecBlockOffset[iSpatialIdx][1] = NULL;
pPtr->pStrideEncBlockOffset[iSpatialIdx] = NULL;
pPtr->pMbIndexX[iSpatialIdx] = NULL;
pPtr->pMbIndexY[iSpatialIdx] = NULL;
++ iSpatialIdx;
}
// initialize pMbIndexX and pMbIndexY tables as below
iMaxMbWidth = sMbSizeMap[kiNumSpatialLayers-1].iMbWidth;
iMaxMbWidth = WELS_ALIGN(iMaxMbWidth, 4); // 4 loops for int16_t required introduced as below
iRowSize = iMaxMbWidth * sizeof(int16_t);
pTmpRow = (int16_t*)pMa->WelsMalloc( iRowSize, "pTmpRow" );
if ( NULL == pTmpRow )
{
return 1;
}
pRowX = pTmpRow;
pRowY = pRowX;
// initialize pRowX & pRowY
i = 0;
p = pRowX;
while ( i < iMaxMbWidth )
{
*p = i;
*(p+1) = 1+i;
*(p+2) = 2+i;
*(p+3) = 3+i;
p += 4;
i += 4;
}
iSpatialIdx = kiNumSpatialLayers;
while ( --iSpatialIdx >= 0 )
{
int16_t *pMbIndexX = pPtr->pMbIndexX[iSpatialIdx];
const int32_t kiMbWidth = sMbSizeMap[iSpatialIdx].iMbWidth;
const int32_t kiMbHeight = sMbSizeMap[iSpatialIdx].iCountMbNum / kiMbWidth;
const int32_t kiLineSize = kiMbWidth * sizeof(int16_t);
i = 0;
while ( i < kiMbHeight )
{
memcpy( pMbIndexX, pRowX, kiLineSize ); // confirmed_safe_unsafe_usage
pMbIndexX += kiMbWidth;
++ i;
}
}
memset(pRowY, 0, iRowSize);
iMaxMbHeight = sMbSizeMap[kiNumSpatialLayers-1].iCountMbNum / sMbSizeMap[kiNumSpatialLayers-1].iMbWidth;
i = 0;
for (;;)
{
ENFORCE_STACK_ALIGN_1D(int16_t, t, 4, 16)
int32_t t32 = 0;
int16_t j = 0;
for ( iSpatialIdx = kiNumSpatialLayers-1; iSpatialIdx >= 0; -- iSpatialIdx )
{
const int32_t kiMbWidth = sMbSizeMap[iSpatialIdx].iMbWidth;
const int32_t kiMbHeight = sMbSizeMap[iSpatialIdx].iCountMbNum / kiMbWidth;
const int32_t kiLineSize = kiMbWidth * sizeof(int16_t);
int16_t *pMbIndexY = pPtr->pMbIndexY[iSpatialIdx] + i * kiMbWidth;
if ( i < kiMbHeight )
{
memcpy( pMbIndexY, pRowY, kiLineSize ); // confirmed_safe_unsafe_usage
}
}
++ i;
if (i >= iMaxMbHeight)
break;
t32 = i | (i << 16);
ST32( t , t32 );
ST32( t+2, t32 );
p = pRowY;
while ( j < iMaxMbWidth )
{
ST64( p, LD64(t) );
p += 4;
j += 4;
}
}
pMa->WelsFree( pTmpRow, "pTmpRow" );
pTmpRow = NULL;
return 0;
}
/*!
* \brief request specific memory for SVC
* \pParam pEncCtx sWelsEncCtx*
* \return successful - 0; otherwise none 0 for failed
*/
int32_t RequestMemorySvc( sWelsEncCtx **ppCtx )
{
SWelsSvcCodingParam *pParam = (*ppCtx)->pSvcParam;
CMemoryAlign *pMa = (*ppCtx)->pMemAlign;
SDLayerParam *pFinalSpatial = NULL;
int32_t iCountBsLen = 0;
int32_t iCountNals = 0;
int32_t iMaxPicWidth = 0;
int32_t iMaxPicHeight = 0;
int32_t iCountMaxMbNum = 0;
int32_t iIndex = 0;
int32_t iCountLayers = 0;
int32_t iResult = 0;
float fCompressRatioThr = .5f;
const int32_t kiNumDependencyLayers = pParam->iNumDependencyLayer;
const uint32_t kuiMvdInterTableSize = ( kiNumDependencyLayers == 1 ? (1 + (648 << 1)) : (1 + (972 << 1)) );
const uint32_t kuiMvdCacheAlginedSize = kuiMvdInterTableSize * sizeof(uint16_t);
int32_t iVclLayersBsSizeCount = 0;
int32_t iNonVclLayersBsSizeCount = 0;
#if defined(MT_ENABLED)
int32_t iTargetSpatialBsSize = 0;
#endif//MT_ENABLED
if ( kiNumDependencyLayers < 1 || kiNumDependencyLayers > MAX_DEPENDENCY_LAYER )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc() failed due to invalid iNumDependencyLayers(%d)!\n", kiNumDependencyLayers);
FreeMemorySvc( ppCtx );
return 1;
}
if ( pParam->uiGopSize == 0 || ( pParam->uiIntraPeriod && ((pParam->uiIntraPeriod % pParam->uiGopSize) != 0)) )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc() failed due to invalid uiIntraPeriod(%d) (=multipler of uiGopSize(%d)!",
pParam->uiIntraPeriod, pParam->uiGopSize);
FreeMemorySvc( ppCtx );
return 1;
}
pFinalSpatial = &pParam->sDependencyLayers[kiNumDependencyLayers - 1];
iMaxPicWidth = pFinalSpatial->iFrameWidth;
iMaxPicHeight = pFinalSpatial->iFrameHeight;
iCountMaxMbNum= ((15+iMaxPicWidth)>>4) * ((15+iMaxPicHeight)>>4);
iResult = AcquireLayersNals( ppCtx, pParam, &iCountLayers, &iCountNals );
if ( iResult )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc(), AcquireLayersNals failed(%d)!", iResult);
FreeMemorySvc( ppCtx );
return 1;
}
iNonVclLayersBsSizeCount = SSEI_BUFFER_SIZE + pParam->iNumDependencyLayer * SPS_BUFFER_SIZE + (1+pParam->iNumDependencyLayer) * PPS_BUFFER_SIZE;
int32_t iLayerBsSize = 0;
iIndex = 0;
while(iIndex < pParam->iNumDependencyLayer)
{
SDLayerParam *fDlp = &pParam->sDependencyLayers[iIndex];
fCompressRatioThr = COMPRESS_RATIO_DECIDED_BY_RESOLUTION(fDlp->iFrameWidth, fDlp->iFrameHeight);
iLayerBsSize = WELS_ROUND( ( (3 * fDlp->iFrameWidth * fDlp->iFrameHeight)>>1) * fCompressRatioThr);
iLayerBsSize = WELS_ALIGN(iLayerBsSize, 4); // 4 bytes alinged
iVclLayersBsSizeCount += iLayerBsSize;
++ iIndex;
}
#if defined(MT_ENABLED)
iTargetSpatialBsSize = iLayerBsSize;
#endif//MT_ENABLED
iCountBsLen = iNonVclLayersBsSizeCount + iVclLayersBsSizeCount;
pParam->iNumRefFrame = WELS_CLIP3(pParam->iNumRefFrame, MIN_REF_PIC_COUNT, MAX_REFERENCE_PICTURE_COUNT_NUM);
// Output
(*ppCtx)->pOut = (SWelsEncoderOutput *)pMa->WelsMalloc( sizeof(SWelsEncoderOutput), "SWelsEncoderOutput" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pOut), FreeMemorySvc(ppCtx) )
(*ppCtx)->pOut->pBsBuffer = (uint8_t *)pMa->WelsMalloc( iCountBsLen, "pOut->pBsBuffer" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pOut->pBsBuffer), FreeMemorySvc(ppCtx) )
(*ppCtx)->pOut->uiSize = iCountBsLen;
(*ppCtx)->pOut->sNalList = (SWelsNalRaw *)pMa->WelsMalloc( iCountNals * sizeof(SWelsNalRaw), "pOut->sNalList" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pOut->sNalList), FreeMemorySvc(ppCtx) )
(*ppCtx)->pOut->iCountNals = iCountNals;
(*ppCtx)->pOut->iNalIndex = 0;
#ifdef MT_ENABLED
if ( pParam->iMultipleThreadIdc > 1 )
{
(*ppCtx)->pFrameBs = (uint8_t *)pMa->WelsMalloc( iCountBsLen + (iTargetSpatialBsSize * ((*ppCtx)->iMaxSliceCount-1)), "pFrameBs" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pFrameBs), FreeMemorySvc(ppCtx) )
(*ppCtx)->iFrameBsSize = iCountBsLen * (*ppCtx)->iMaxSliceCount;
}
else
#endif//MT_ENABLED
{
(*ppCtx)->pFrameBs = (uint8_t *)pMa->WelsMalloc( iCountBsLen, "pFrameBs" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pFrameBs), FreeMemorySvc(ppCtx) )
(*ppCtx)->iFrameBsSize = iCountBsLen;
}
(*ppCtx)->iPosBsBuffer = 0;
#ifdef MT_ENABLED
// for pSlice bs buffers
if ( pParam->iMultipleThreadIdc > 1 && RequestMtResource( ppCtx, pParam, iCountBsLen, iTargetSpatialBsSize ) )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc(), RequestMtResource failed!");
FreeMemorySvc( ppCtx );
return 1;
}
#endif
(*ppCtx)->pIntra4x4PredModeBlocks = static_cast<int8_t*>
(pMa->WelsMallocz( iCountMaxMbNum * INTRA_4x4_MODE_NUM, "pIntra4x4PredModeBlocks" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pIntra4x4PredModeBlocks), FreeMemorySvc(ppCtx) )
(*ppCtx)->pNonZeroCountBlocks = static_cast<int8_t*>
(pMa->WelsMallocz( iCountMaxMbNum * MB_LUMA_CHROMA_BLOCK4x4_NUM, "pNonZeroCountBlocks" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pNonZeroCountBlocks), FreeMemorySvc(ppCtx) )
(*ppCtx)->pMvUnitBlock4x4 = static_cast<SMVUnitXY*>
(pMa->WelsMallocz( iCountMaxMbNum * 2 * MB_BLOCK4x4_NUM * sizeof(SMVUnitXY), "pMvUnitBlock4x4" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pMvUnitBlock4x4), FreeMemorySvc(ppCtx) )
(*ppCtx)->pRefIndexBlock4x4 = static_cast<int8_t*>
(pMa->WelsMallocz( iCountMaxMbNum * 2 * MB_BLOCK8x8_NUM * sizeof(int8_t), "pRefIndexBlock4x4" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pRefIndexBlock4x4), FreeMemorySvc(ppCtx) )
(*ppCtx)->pSadCostMb = static_cast<int32_t*>
(pMa->WelsMallocz( iCountMaxMbNum * sizeof(int32_t), "pSadCostMb" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pSadCostMb), FreeMemorySvc(ppCtx))
(*ppCtx)->bEncCurFrmAsIdrFlag = true; // make sure first frame is IDR
(*ppCtx)->iGlobalQp = 26; // global qp in default
(*ppCtx)->pLtr = (SLTRState *)pMa->WelsMalloc( kiNumDependencyLayers*sizeof(SLTRState), "SLTRState" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pLtr), FreeMemorySvc(ppCtx) )
int32_t i = 0;
for( i = 0; i < kiNumDependencyLayers; i++ )
{
ResetLtrState( &(*ppCtx)->pLtr[i] );
}
(*ppCtx)->ppRefPicListExt = (SRefList**)pMa->WelsMalloc( kiNumDependencyLayers * sizeof(SRefList *), "ppRefPicListExt" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->ppRefPicListExt), FreeMemorySvc(ppCtx) )
// pSlice context list
(*ppCtx)->pSliceCtxList = (SSliceCtx *)pMa->WelsMallocz( kiNumDependencyLayers * sizeof(SSliceCtx), "pSliceCtxList" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pSliceCtxList), FreeMemorySvc(ppCtx) )
(*ppCtx)->ppDqLayerList = (SDqLayer **)pMa->WelsMalloc( kiNumDependencyLayers * sizeof(SDqLayer *), "ppDqLayerList" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->ppDqLayerList), FreeMemorySvc(ppCtx) )
// stride tables
if ( AllocStrideTables( ppCtx, kiNumDependencyLayers ) )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc(), AllocStrideTables failed!");
FreeMemorySvc( ppCtx );
return 1;
}
//Rate control module memory allocation
// only malloc once for RC pData, 12/14/2009
(*ppCtx)->pWelsSvcRc = (SWelsSvcRc *)pMa->WelsMallocz( kiNumDependencyLayers * sizeof(SWelsSvcRc), "pWelsSvcRc" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pWelsSvcRc), FreeMemorySvc(ppCtx) )
//End of Rate control module memory allocation
//pVaa memory allocation
(*ppCtx)->pVaa = (SVAAFrameInfo *)pMa->WelsMallocz( sizeof(SVAAFrameInfo), "pVaa" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa), FreeMemorySvc(ppCtx) )
if((*ppCtx)->pSvcParam->bEnableAdaptiveQuant)//malloc mem
{
(*ppCtx)->pVaa->sAdaptiveQuantParam.pMotionTextureUnit = static_cast<SMotionTextureUnit*>
(pMa->WelsMallocz( iCountMaxMbNum * sizeof(SMotionTextureUnit), "pVaa->sAdaptiveQuantParam.pMotionTextureUnit" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sAdaptiveQuantParam.pMotionTextureUnit), FreeMemorySvc(ppCtx) )
(*ppCtx)->pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp = static_cast<int8_t*>
(pMa->WelsMallocz( iCountMaxMbNum * sizeof(int8_t), "pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp), FreeMemorySvc(ppCtx) )
}
(*ppCtx)->pVaa->pVaaBackgroundMbFlag = (int8_t *)pMa->WelsMallocz( iCountMaxMbNum * sizeof(int8_t), "pVaa->vaa_skip_mb_flag" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->pVaaBackgroundMbFlag), FreeMemorySvc(ppCtx) )
(*ppCtx)->pVaa->sVaaCalcInfo.pSad8x8 = static_cast<int32_t(*)[4]>
(pMa->WelsMallocz( iCountMaxMbNum * 4 * sizeof(int32_t), "pVaa->sVaaCalcInfo.sad8x8" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sVaaCalcInfo.pSad8x8), FreeMemorySvc(ppCtx) )
(*ppCtx)->pVaa->sVaaCalcInfo.pSsd16x16 = static_cast<int32_t*>
(pMa->WelsMallocz( iCountMaxMbNum * sizeof(int32_t), "pVaa->sVaaCalcInfo.pSsd16x16" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sVaaCalcInfo.pSsd16x16), FreeMemorySvc(ppCtx) )
(*ppCtx)->pVaa->sVaaCalcInfo.pSum16x16 = static_cast<int32_t*>
(pMa->WelsMallocz( iCountMaxMbNum * sizeof(int32_t), "pVaa->sVaaCalcInfo.pSum16x16" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sVaaCalcInfo.pSum16x16), FreeMemorySvc(ppCtx) )
(*ppCtx)->pVaa->sVaaCalcInfo.pSumOfSquare16x16 = static_cast<int32_t*>
(pMa->WelsMallocz( iCountMaxMbNum * sizeof(int32_t), "pVaa->sVaaCalcInfo.pSumOfSquare16x16" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sVaaCalcInfo.pSumOfSquare16x16), FreeMemorySvc(ppCtx) )
if ((*ppCtx)->pSvcParam->bEnableBackgroundDetection) //BGD control
{
(*ppCtx)->pVaa->sVaaCalcInfo.pSumOfDiff8x8 = static_cast<int32_t(*)[4]>
(pMa->WelsMallocz( iCountMaxMbNum * 4 * sizeof(int32_t), "pVaa->sVaaCalcInfo.sd_16x16" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sVaaCalcInfo.pSumOfDiff8x8), FreeMemorySvc(ppCtx) )
(*ppCtx)->pVaa->sVaaCalcInfo.pMad8x8 = static_cast<uint8_t(*)[4]>
(pMa->WelsMallocz( iCountMaxMbNum * 4 * sizeof(uint8_t), "pVaa->sVaaCalcInfo.mad_16x16" ));
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pVaa->sVaaCalcInfo.pMad8x8), FreeMemorySvc(ppCtx) )
}
//End of pVaa memory allocation
iResult = InitDqLayers( ppCtx );
if ( iResult )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc(), InitDqLayers failed(%d)!", iResult );
FreeMemorySvc( ppCtx );
return iResult;
}
if( InitMbListD( ppCtx ) )
{
WelsLog( *ppCtx, WELS_LOG_WARNING, "RequestMemorySvc(), InitMbListD failed!" );
FreeMemorySvc( ppCtx );
return 1;
}
(*ppCtx)->pMvdCostTableInter = (uint16_t *)pMa->WelsMallocz( 52 * kuiMvdCacheAlginedSize, "pMvdCostTableInter" );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pMvdCostTableInter), FreeMemorySvc(ppCtx) )
MvdCostInit( (*ppCtx)->pMvdCostTableInter, kuiMvdInterTableSize ); //should put to a better place?
if ( (*ppCtx)->ppRefPicListExt[0] != NULL && (*ppCtx)->ppRefPicListExt[0]->pRef[0] != NULL )
(*ppCtx)->pDecPic = (*ppCtx)->ppRefPicListExt[0]->pRef[0];
else
(*ppCtx)->pDecPic = NULL; // error here
(*ppCtx)->pSps = &(*ppCtx)->pSpsArray[0];
(*ppCtx)->pPps = &(*ppCtx)->pPPSArray[0];
return 0;
}
/*!
* \brief free memory in SVC core encoder
* \pParam pEncCtx sWelsEncCtx*
* \return none
*/
void FreeMemorySvc( sWelsEncCtx **ppCtx )
{
if ( NULL != *ppCtx )
{
sWelsEncCtx *pCtx = *ppCtx;
CMemoryAlign *pMa = pCtx->pMemAlign;
SWelsSvcCodingParam *pParam= pCtx->pSvcParam;
int32_t ilayer = 0;
// SStrideTables
if ( NULL != pCtx->pStrideTab )
{
if ( NULL != pCtx->pStrideTab->pStrideDecBlockOffset[0][1] )
{
pMa->WelsFree( pCtx->pStrideTab->pStrideDecBlockOffset[0][1], "pBase" );
pCtx->pStrideTab->pStrideDecBlockOffset[0][1] = NULL;
}
pMa->WelsFree(pCtx->pStrideTab, "SStrideTables");
pCtx->pStrideTab = NULL;
}
// pDq idc map
if ( NULL != pCtx->pDqIdcMap )
{
pMa->WelsFree( pCtx->pDqIdcMap, "pDqIdcMap" );
pCtx->pDqIdcMap = NULL;
}
if ( NULL != pCtx->pOut )
{
// bs pBuffer
if ( NULL != pCtx->pOut->pBsBuffer )
{
pMa->WelsFree( pCtx->pOut->pBsBuffer, "pOut->pBsBuffer" );
pCtx->pOut->pBsBuffer = NULL;
}
// NALs list
if ( NULL != pCtx->pOut->sNalList )
{
pMa->WelsFree( pCtx->pOut->sNalList, "pOut->sNalList" );
pCtx->pOut->sNalList = NULL;
}
pMa->WelsFree( pCtx->pOut, "SWelsEncoderOutput" );
pCtx->pOut = NULL;
}
#ifdef MT_ENABLED
if ( pParam != NULL && pParam->iMultipleThreadIdc > 1 )
ReleaseMtResource( ppCtx );
#endif//MT_ENABLED
// frame bitstream pBuffer
if ( NULL != pCtx->pFrameBs )
{
pMa->WelsFree( pCtx->pFrameBs, "pFrameBs" );
pCtx->pFrameBs = NULL;
}
// pSpsArray
if ( NULL != pCtx->pSpsArray )
{
pMa->WelsFree( pCtx->pSpsArray, "pSpsArray" );
pCtx->pSpsArray = NULL;
}
// pPPSArray
if ( NULL != pCtx->pPPSArray )
{
pMa->WelsFree( pCtx->pPPSArray, "pPPSArray" );
pCtx->pPPSArray = NULL;
}
// subset_sps_array
if ( NULL != pCtx->pSubsetArray )
{
pMa->WelsFree( pCtx->pSubsetArray, "pSubsetArray" );
pCtx->pSubsetArray = NULL;
}
if( NULL != pCtx->pIntra4x4PredModeBlocks ){
pMa->WelsFree( pCtx->pIntra4x4PredModeBlocks, "pIntra4x4PredModeBlocks" );
pCtx->pIntra4x4PredModeBlocks = NULL;
}
if( NULL != pCtx->pNonZeroCountBlocks ){
pMa->WelsFree( pCtx->pNonZeroCountBlocks, "pNonZeroCountBlocks" );
pCtx->pNonZeroCountBlocks = NULL;
}
if ( NULL != pCtx->pMvUnitBlock4x4)
{
pMa->WelsFree( pCtx->pMvUnitBlock4x4, "pMvUnitBlock4x4" );
pCtx->pMvUnitBlock4x4 = NULL;
}
if ( NULL != pCtx->pRefIndexBlock4x4)
{
pMa->WelsFree( pCtx->pRefIndexBlock4x4, "pRefIndexBlock4x4" );
pCtx->pRefIndexBlock4x4 = NULL;
}
if ( NULL != pCtx->ppMbListD )
{
if( NULL != pCtx->ppMbListD[0] ){
pMa->WelsFree( pCtx->ppMbListD[0], "ppMbListD[0]" );
(*ppCtx)->ppMbListD[0] = NULL;
}
pMa->WelsFree( pCtx->ppMbListD, "ppMbListD" );
pCtx->ppMbListD = NULL;
}
if ( NULL != pCtx->pSadCostMb)
{
pMa->WelsFree( pCtx->pSadCostMb, "pSadCostMb" );
pCtx->pSadCostMb = NULL;
}
// SLTRState
if ( NULL != pCtx->pLtr )
{
pMa->WelsFree( pCtx->pLtr, "SLTRState" );
pCtx->pLtr = NULL;
}
// pDq layers list
ilayer = 0;
if ( NULL != pCtx->ppDqLayerList && pParam != NULL )
{
while (ilayer < pParam->iNumDependencyLayer) {
SDqLayer *pDq = pCtx->ppDqLayerList[ilayer];
SDLayerParam *pDlp = &pCtx->pSvcParam->sDependencyLayers[ilayer];
const BOOL_T kbIsDynamicSlicing = (SM_DYN_SLICE == pDlp->sMso.uiSliceMode);
// pDq layers
if ( NULL != pDq )
{
if ( NULL != pDq->sLayerInfo.pSliceInLayer )
{
int32_t iSliceIdx = 0;
int32_t iSliceNum = GetInitialSliceNum( pDq->iMbWidth, pDq->iMbHeight, &pDlp->sMso );
if (iSliceNum < 1)
iSliceNum = 1;
while(iSliceIdx < iSliceNum)
{
SSlice *pSlice = &pDq->sLayerInfo.pSliceInLayer[iSliceIdx];
FreeMbCache(&pSlice->sMbCacheInfo, pMa);
++ iSliceIdx;
}
pMa->WelsFree( pDq->sLayerInfo.pSliceInLayer, "pSliceInLayer" );
pDq->sLayerInfo.pSliceInLayer = NULL;
}
if ( kbIsDynamicSlicing )
{
pMa->WelsFree( pDq->pNumSliceCodedOfPartition, "pNumSliceCodedOfPartition" );
pDq->pNumSliceCodedOfPartition = NULL;
pMa->WelsFree( pDq->pLastCodedMbIdxOfPartition, "pLastCodedMbIdxOfPartition" );
pDq->pLastCodedMbIdxOfPartition = NULL;
pMa->WelsFree( pDq->pLastMbIdxOfPartition, "pLastMbIdxOfPartition" );
pDq->pLastMbIdxOfPartition = NULL;
}
pMa->WelsFree( pDq, "pDq" );
pDq = NULL;
pCtx->ppDqLayerList[ilayer] = NULL;
}
++ ilayer;
}
pMa->WelsFree( pCtx->ppDqLayerList, "ppDqLayerList" );
pCtx->ppDqLayerList = NULL;
}
FreeSpatialPictures( pCtx );
// reference picture list extension
if ( NULL != pCtx->ppRefPicListExt && pParam != NULL )
{
ilayer = 0;
while (ilayer < pParam->iNumDependencyLayer) {
SRefList *pRefList = pCtx->ppRefPicListExt[ilayer];
if ( NULL != pRefList )
{
int32_t iRef = 0;
do {
if ( pRefList->pRef[iRef] != NULL )
{
FreePicture( pMa, &pRefList->pRef[iRef] );
}
++ iRef;
} while(iRef < 1 + pParam->iNumRefFrame);
pMa->WelsFree( pCtx->ppRefPicListExt[ilayer], "ppRefPicListExt[]" );
pCtx->ppRefPicListExt[ilayer] = NULL;
}
++ ilayer;
}
pMa->WelsFree( pCtx->ppRefPicListExt, "ppRefPicListExt" );
pCtx->ppRefPicListExt = NULL;
}
// pSlice context list
if ( NULL != pCtx->pSliceCtxList && pParam != NULL )
{
ilayer = 0;
while (ilayer < pParam->iNumDependencyLayer) {
SSliceCtx *pSliceCtx = &pCtx->pSliceCtxList[ilayer];
if ( NULL != pSliceCtx )
UninitSlicePEncCtx( pSliceCtx, pMa );
++ ilayer;
}
pMa->WelsFree( pCtx->pSliceCtxList, "pSliceCtxList" );
pCtx->pSliceCtxList = NULL;
}
// VAA
if ( NULL != pCtx->pVaa )
{
if(pCtx->pSvcParam->bEnableAdaptiveQuant)//free mem
{
pMa->WelsFree( pCtx->pVaa->sAdaptiveQuantParam.pMotionTextureUnit, "pVaa->sAdaptiveQuantParam.pMotionTextureUnit" );
pCtx->pVaa->sAdaptiveQuantParam.pMotionTextureUnit = NULL;
pMa->WelsFree( pCtx->pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp, "pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp" );
pCtx->pVaa->sAdaptiveQuantParam.pMotionTextureIndexToDeltaQp = NULL;
}
pMa->WelsFree( pCtx->pVaa->pVaaBackgroundMbFlag, "pVaa->pVaaBackgroundMbFlag");
pCtx->pVaa->pVaaBackgroundMbFlag = NULL;
pMa->WelsFree( pCtx->pVaa->sVaaCalcInfo.pSad8x8, "pVaa->sVaaCalcInfo.sad8x8" );
pCtx->pVaa->sVaaCalcInfo.pSad8x8 = NULL;
pMa->WelsFree( pCtx->pVaa->sVaaCalcInfo.pSsd16x16, "pVaa->sVaaCalcInfo.pSsd16x16" );
pCtx->pVaa->sVaaCalcInfo.pSsd16x16 = NULL;
pMa->WelsFree( pCtx->pVaa->sVaaCalcInfo.pSum16x16, "pVaa->sVaaCalcInfo.pSum16x16" );
pCtx->pVaa->sVaaCalcInfo.pSum16x16 = NULL;
pMa->WelsFree( pCtx->pVaa->sVaaCalcInfo.pSumOfSquare16x16, "pVaa->sVaaCalcInfo.pSumOfSquare16x16" );
pCtx->pVaa->sVaaCalcInfo.pSumOfSquare16x16 = NULL;
if (pCtx->pSvcParam->bEnableBackgroundDetection) //BGD control
{
pMa->WelsFree( pCtx->pVaa->sVaaCalcInfo.pSumOfDiff8x8, "pVaa->sVaaCalcInfo.pSumOfDiff8x8" );
pCtx->pVaa->sVaaCalcInfo.pSumOfDiff8x8 = NULL;
pMa->WelsFree( pCtx->pVaa->sVaaCalcInfo.pMad8x8, "pVaa->sVaaCalcInfo.pMad8x8" );
pCtx->pVaa->sVaaCalcInfo.pMad8x8 = NULL;
}
pMa->WelsFree( pCtx->pVaa, "pVaa" );
pCtx->pVaa = NULL;
}
WelsRcFreeMemory(pCtx);
// rate control module memory free
if ( NULL != pCtx->pWelsSvcRc )
{
pMa->WelsFree( pCtx->pWelsSvcRc, "pWelsSvcRc" );
pCtx->pWelsSvcRc = NULL;
}
/* MVD cost tables for Inter */
if ( NULL != pCtx->pMvdCostTableInter )
{
pMa->WelsFree( pCtx->pMvdCostTableInter, "pMvdCostTableInter" );
pCtx->pMvdCostTableInter = NULL;
}
#ifdef ENABLE_TRACE_FILE
if ( NULL != pCtx->pFileLog )
{
fclose( pCtx->pFileLog );
pCtx->pFileLog = NULL;
}
pCtx->uiSizeLog = 0;
#endif//ENABLE_TRACE_FILE
FreeCodingParam( &pCtx->pSvcParam, pMa );
if ( NULL != pCtx->pFuncList )
{
pMa->WelsFree(pCtx->pFuncList, "SWelsFuncPtrList");
pCtx->pFuncList = NULL;
}
#if defined(MEMORY_MONITOR)
assert(pMa->WelsGetMemoryUsage() == 0); // ensure all memory free well
#endif//MEMORY_MONITOR
if ( (*ppCtx)->pMemAlign != NULL )
{
WelsLog( NULL, WELS_LOG_INFO, "FreeMemorySvc(), verify memory usage (%d bytes) after free..\n", (*ppCtx)->pMemAlign->WelsGetMemoryUsage() );
delete (*ppCtx)->pMemAlign;
(*ppCtx)->pMemAlign = NULL;
}
free(*ppCtx);
*ppCtx = NULL;
}
}
int32_t InitSliceSettings( SWelsSvcCodingParam *pCodingParam, const int32_t kiCpuCores, int16_t *pMaxSliceCount )
{
int32_t iSpatialIdx = 0, iSpatialNum = pCodingParam->iNumDependencyLayer;
int16_t iMaxSliceCount = 0;
do {
SDLayerParam *pDlp = &pCodingParam->sDependencyLayers[iSpatialIdx];
SMulSliceOption *pMso = &pDlp->sMso;
SSliceArgument *pSlcArg = &pMso->sSliceArgument;
const int32_t kiMbWidth = (pDlp->iFrameWidth+15)>>4;
const int32_t kiMbHeight = (pDlp->iFrameHeight+15)>>4;
const int32_t kiMbNumInFrame = kiMbWidth * kiMbHeight;
#if defined(MT_ENABLED)
#if defined(DYNAMIC_SLICE_ASSIGN)
int32_t iSliceNum = (SM_FIXEDSLCNUM_SLICE == pMso->uiSliceMode || SM_DYN_SLICE == pMso->uiSliceMode) ? kiCpuCores : pSlcArg->iSliceNum; // uiSliceNum per input has been validated at ParamValidationExt()
#else//!DYNAMIC_SLICE_ASSIGN
int32_t iSliceNum = (SM_DYN_SLICE == pMso->uiSliceMode) ? kiCpuCores : pSlcArg->uiSliceNum; // uiSliceNum per input has been validated at ParamValidationExt()
#endif//DYNAMIC_SLICE_ASSIGN
#else//!MT_ENABLED
int16_t iSliceNum = pSlcArg->iSliceNum; // uiSliceNum per input has been validated at ParamValidationExt()
#endif//MT_ENABLED
// NOTE: Per design, in case MT/DYNAMIC_SLICE_ASSIGN enabled, for SM_FIXEDSLCNUM_SLICE mode,
// uiSliceNum of current spatial layer settings equals to uiCpuCores number; SM_DYN_SLICE mode,
// uiSliceNum intials as uiCpuCores also, stay tuned dynamically slicing in future
pSlcArg->iSliceNum = iSliceNum; // used fixed one
switch(pMso->uiSliceMode)
{
case SM_DYN_SLICE:
iMaxSliceCount = AVERSLICENUM_CONSTRAINT;
//#ifndef MT_ENABLED
break; // go through for MT_ENABLED & SM_DYN_SLICE?
//#endif//MT_ENABLED
case SM_FIXEDSLCNUM_SLICE:
if ( iSliceNum > iMaxSliceCount )
iMaxSliceCount = iSliceNum;
// need perform check due uiSliceNum might change, although has been initialized somewhere outside
if (pCodingParam->bEnableRc)
{
GomValidCheckSliceMbNum( kiMbWidth, kiMbHeight, pSlcArg );
}
else
{
CheckFixedSliceNumMultiSliceSetting( kiMbNumInFrame, pSlcArg );
}
break;
case SM_SINGLE_SLICE:
if ( iSliceNum > iMaxSliceCount )
iMaxSliceCount = iSliceNum;
break;
case SM_RASTER_SLICE:
if ( iSliceNum > iMaxSliceCount )
iMaxSliceCount = iSliceNum;
break;
case SM_ROWMB_SLICE:
if ( iSliceNum > iMaxSliceCount )
iMaxSliceCount = iSliceNum;
break;
default:
break;
}
++ iSpatialIdx;
} while(iSpatialIdx < iSpatialNum);
#ifdef MT_ENABLED
pCodingParam->iCountThreadsNum = WELS_MIN(kiCpuCores, iMaxSliceCount);
pCodingParam->iMultipleThreadIdc = pCodingParam->iCountThreadsNum;
#else
pCodingParam->iMultipleThreadIdc = 1;
pCodingParam->iCountThreadsNum = 1;
#endif//MT_ENABLED
#ifndef WELS_TESTBED // for product release and non-SGE testing
if ( kiCpuCores < 2 ) // single CPU core, make no sense for MT parallelization
{
pCodingParam->iMultipleThreadIdc = 1;
pCodingParam->iCountThreadsNum = 1;
}
#endif
*pMaxSliceCount = iMaxSliceCount;
return 0;
}
/*!
* \brief log output for cpu features/capabilities
*/
void OutputCpuFeaturesLog( uint32_t uiCpuFeatureFlags, uint32_t uiCpuCores, int32_t iCacheLineSize )
{
// welstracer output
WelsLog(NULL, WELS_LOG_INFO, "WELS CPU features/capacities (0x%x) detected: \t" \
"HTT: %c, " \
"MMX: %c, " \
"MMXEX: %c, " \
"SSE: %c, " \
"SSE2: %c, " \
"SSE3: %c, " \
"SSSE3: %c, " \
"SSE4.1: %c, " \
"SSE4.2: %c, " \
"AVX: %c, " \
"FMA: %c, " \
"X87-FPU: %c, " \
"3DNOW: %c, " \
"3DNOWEX: %c, " \
"ALTIVEC: %c, " \
"CMOV: %c, " \
"MOVBE: %c, " \
"AES: %c, " \
"NUMBER OF LOGIC PROCESSORS ON CHIP: %d, " \
"CPU CACHE LINE SIZE (BYTES): %d\n",
uiCpuFeatureFlags,
(uiCpuFeatureFlags & WELS_CPU_HTT) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_MMX) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_MMXEXT) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE2) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE3) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSSE3) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE41) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE42) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_AVX) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_FMA) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_FPU) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_3DNOW) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_3DNOWEXT) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_ALTIVEC) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_CMOV) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_MOVBE) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_AES) ? 'Y' : 'N',
uiCpuCores,
iCacheLineSize );
#ifdef _DEBUG // output at console & _debug
fprintf( stderr, "WELS CPU features/capacities (0x%x) detected: \n" \
"HTT: %c, " \
"MMX: %c, " \
"MMXEX: %c, " \
"SSE: %c, " \
"SSE2: %c, " \
"SSE3: %c, " \
"SSSE3: %c, " \
"SSE4.1: %c, " \
"SSE4.2: %c, " \
"AVX: %c, " \
"FMA: %c, " \
"X87-FPU: %c, " \
"3DNOW: %c, " \
"3DNOWEX: %c, " \
"ALTIVEC: %c, " \
"CMOV: %c, " \
"MOVBE: %c, " \
"AES: %c, " \
"NUMBER OF LOGIC PROCESSORS ON CHIP: %d, " \
"CPU CACHE LINE SIZE (BYTES): %d\n",
uiCpuFeatureFlags,
(uiCpuFeatureFlags & WELS_CPU_HTT) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_MMX) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_MMXEXT) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE2) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE3) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSSE3) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE41) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_SSE42) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_AVX) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_FMA) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_FPU) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_3DNOW) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_3DNOWEXT) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_ALTIVEC) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_CMOV) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_MOVBE) ? 'Y' : 'N',
(uiCpuFeatureFlags & WELS_CPU_AES) ? 'Y' : 'N',
uiCpuCores,
iCacheLineSize );
#endif//_DEBUG
}
/*!
* \brief initialize Wels avc encoder core library
* \pParam ppCtx sWelsEncCtx**
* \pParam pParam SWelsSvcCodingParam*
* \return successful - 0; otherwise none 0 for failed
*/
int32_t WelsInitEncoderExt( sWelsEncCtx **ppCtx, SWelsSvcCodingParam *pCodingParam )
{
sWelsEncCtx *pCtx = NULL;
int32_t iRet = 0;
uint32_t uiCpuFeatureFlags = 0; // CPU features
int32_t uiCpuCores = 1; // number of logic processors on physical processor package, one logic processor means HTT not supported
int32_t iCacheLineSize = 16; // on chip cache line size in byte
int16_t iSliceNum = 1; // number of slices used
if ( NULL == ppCtx || NULL == pCodingParam )
{
WelsLog(NULL, WELS_LOG_ERROR, "WelsInitEncoderExt(), NULL == ppCtx(0x%p) or NULL == pCodingParam(0x%p).\n", (void *)ppCtx, (void *)pCodingParam);
return 1;
}
iRet = ParamValidationExt( pCodingParam );
if ( iRet != 0 )
{
WelsLog(NULL, WELS_LOG_ERROR, "WelsInitEncoderExt(), ParamValidationExt failed return %d.\n", iRet);
return iRet;
}
// for cpu features detection, Only detect once??
#ifdef X86_ASM
uiCpuFeatureFlags = WelsCPUFeatureDetect( &uiCpuCores ); // detect cpu capacity features
if ( uiCpuFeatureFlags & WELS_CPU_CACHELINE_128 )
iCacheLineSize = 128;
else if ( uiCpuFeatureFlags & WELS_CPU_CACHELINE_64 )
iCacheLineSize = 64;
else if ( uiCpuFeatureFlags & WELS_CPU_CACHELINE_32 )
iCacheLineSize = 32;
else if ( uiCpuFeatureFlags & WELS_CPU_CACHELINE_16 )
iCacheLineSize = 16;
OutputCpuFeaturesLog( uiCpuFeatureFlags, uiCpuCores, iCacheLineSize );
#else
iCacheLineSize = 16; // 16 bytes aligned in default
#endif//X86_ASM
#ifndef WELS_TESTBED
#if defined(MT_ENABLED) && defined(DYNAMIC_DETECT_CPU_CORES)
if ( pCodingParam->iMultipleThreadIdc > 0 )
uiCpuCores = pCodingParam->iMultipleThreadIdc;
else
{
if ( uiCpuFeatureFlags == 0 ) // cpuid not supported, use high level system API as followed to detect number of pysical/logic processor
uiCpuCores = DynamicDetectCpuCores();
// So far so many cpu cores up to MAX_THREADS_NUM mean for server platforms,
// for client application here it is constrained by maximal to MAX_THREADS_NUM
if ( uiCpuCores > MAX_THREADS_NUM ) // MAX_THREADS_NUM
uiCpuCores = MAX_THREADS_NUM; // MAX_THREADS_NUM
else if ( uiCpuCores < 1 ) // just for safe
uiCpuCores = 1;
}
#endif//MT_ENABLED && DYNAMIC_DETECT_CPU_CORES
#else//WELS_TESTBED
uiCpuCores = pCodingParam->iMultipleThreadIdc; // assigned uiCpuCores from iMultipleThreadIdc from SGE testing
#endif//WELS_TESTBED
uiCpuCores = WELS_CLIP3(uiCpuCores, 1, MAX_THREADS_NUM);
if ( InitSliceSettings(pCodingParam, uiCpuCores, &iSliceNum ) )
{
WelsLog(NULL, WELS_LOG_ERROR, "WelsInitEncoderExt(), InitSliceSettings failed.\n");
return 1;
}
*ppCtx = NULL;
pCtx = static_cast<sWelsEncCtx*>(malloc( sizeof(sWelsEncCtx) ));
WELS_VERIFY_RETURN_IF(1, (NULL == pCtx))
memset( pCtx, 0, sizeof(sWelsEncCtx) );
pCtx->pMemAlign = new CMemoryAlign( iCacheLineSize );
WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pCtx->pMemAlign), FreeMemorySvc(&pCtx) )
// for logs
#ifdef ENABLE_TRACE_FILE
if (wlog == WelsLogDefault)
{
str_t fname[MAX_FNAME_LEN] = {0};
#if defined (_MSC_VER)
#if _MSC_VER>=1500
SNPRINTF(fname, MAX_FNAME_LEN, MAX_FNAME_LEN, "%swels_svc_encoder_trace.txt", pCodingParam->sTracePath ); // confirmed_safe_unsafe_usage
#else
SNPRINTF(fname, MAX_FNAME_LEN, "%swels_svc_encoder_trace.txt", pCodingParam->sTracePath ); // confirmed_safe_unsafe_usage
#endif//_MSC_VER>=1500
#else
//GNUC/
SNPRINTF(fname, MAX_FNAME_LEN, "%swels_svc_encoder_trace.txt", pCodingParam->sTracePath ); // confirmed_safe_unsafe_usage
#endif//_MSC_VER
#if defined(__GNUC__)
pCtx->pFileLog = FOPEN(fname, "wt+");
#else//WIN32
#if defined(WIN32) && defined(_MSC_VER)
#if _MSC_VER >= 1500
FOPEN(&pCtx->pFileLog,fname, "wt+");
#else
pCtx->pFileLog = FOPEN(fname, "wt+");
#endif//_MSC_VER>=1500
#endif//WIN32 && _MSC_VER
#endif//__GNUC__
pCtx->uiSizeLog = 0;
}
#endif//ENABLE_TRACE_FILE
pCodingParam->DetermineTemporalSettings();
iRet = AllocCodingParam( &pCtx->pSvcParam, pCtx->pMemAlign, pCodingParam->iNumDependencyLayer );
if ( iRet != 0 )
{
FreeMemorySvc( &pCtx );
return iRet;
}
memcpy( pCtx->pSvcParam, pCodingParam, sizeof(SWelsSvcCodingParam) ); // confirmed_safe_unsafe_usage
pCtx->pFuncList = (SWelsFuncPtrList *)pCtx->pMemAlign->WelsMalloc(sizeof(SWelsFuncPtrList), "SWelsFuncPtrList");
if ( NULL == pCtx->pFuncList )
{
FreeMemorySvc( &pCtx );
return 1;
}
InitFunctionPointers( pCtx->pFuncList, pCtx->pSvcParam, uiCpuFeatureFlags );
pCtx->iActiveThreadsNum = pCodingParam->iCountThreadsNum;
pCtx->iMaxSliceCount = iSliceNum;
iRet = RequestMemorySvc( &pCtx );
if ( iRet != 0 )
{
WelsLog(pCtx, WELS_LOG_ERROR, "WelsInitEncoderExt(), RequestMemorySvc failed return %d.\n", iRet);
FreeMemorySvc( &pCtx );
return iRet;
}
#ifdef MT_ENABLED
if ( pCodingParam->iMultipleThreadIdc > 1 )
iRet = CreateSliceThreads( pCtx);
#endif
WelsRcInitModule( pCtx, pCtx->pSvcParam->bEnableRc ? WELS_RC_GOM : WELS_RC_DISABLE);
pCtx->pVpp = new CWelsPreProcess((void *)pCtx);
if ( pCtx->pVpp == NULL )
{
WelsLog(pCtx, WELS_LOG_ERROR, "WelsInitEncoderExt(), pOut of memory in case new CWelsPreProcess().\n");
FreeMemorySvc( &pCtx );
return iRet;
}
#if defined(MEMORY_MONITOR)
WelsLog(pCtx, WELS_LOG_INFO, "WelsInitEncoderExt() exit, overall memory usage: %lu bytes\n", sizeof(sWelsEncCtx) /* requested size from malloc() or new operator */
+ pCtx->pMemAlign->WelsGetMemoryUsage() /* requested size from CMemoryAlign::WelsMalloc() */
);
#endif//MEMORY_MONITOR
*ppCtx = pCtx;
WelsLog(pCtx, WELS_LOG_DEBUG, "WelsInitEncoderExt(), pCtx= 0x%p.\n", (void *)pCtx);
return 0;
}
/*
*
* status information output
*/
#if defined(STAT_OUTPUT)
void StatOverallEncodingExt(sWelsEncCtx *pCtx)
{
int8_t i = 0;
int8_t j = 0;
for (i = 0;i<pCtx->pSvcParam->iNumDependencyLayer;i++)
{
fprintf( stdout,"\nDependency layer : %d\n",i);
fprintf( stdout,"Quality layer : %d\n",j);
{
const int32_t iCount = pCtx->sStatData[i][j].sSliceData.iSliceCount[I_SLICE] +
pCtx->sStatData[i][j].sSliceData.iSliceCount[P_SLICE] +
pCtx->sStatData[i][j].sSliceData.iSliceCount[B_SLICE];
#if defined(MB_TYPES_CHECK)
if (iCount > 0){
int32_t iCountNumIMb = pCtx->sStatData[i][j].sSliceData.iMbCount[I_SLICE][Intra4x4] + pCtx->sStatData[i][j].sSliceData.iMbCount[I_SLICE][Intra16x16]+ pCtx->sStatData[i][j].sSliceData.iMbCount[I_SLICE][7];
int32_t iCountNumPMb = pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Intra4x4] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Intra16x16] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][7] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter16x16] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter16x8] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter8x16] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter8x8] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][10] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][PSkip];
int32_t count_p_mbL0 = pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter16x16] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter16x8] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter8x16] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter8x8] +
pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][10];
int32_t iMbCount = iCountNumIMb + iCountNumPMb;
if ( iMbCount > 0 ){
fprintf( stderr,
"SVC: overall Slices MBs: %d Avg\nI4x4: %.3f%% I16x16: %.3f%% IBL: %.3f%%\nP16x16: %.3f%% P16x8: %.3f%% P8x16: %.3f%% P8x8: %.3f%% SUBP8x8: %.3f%% PSKIP: %.3f%%\nILP(All): %.3f%% ILP(PL0): %.3f%% BLSKIP(PL0): %.3f%% RP(PL0): %.3f%%\n",
iMbCount,
(100.0f * (pCtx->sStatData[i][j].sSliceData.iMbCount[I_SLICE][Intra4x4] + pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Intra4x4]) / iMbCount),
(100.0f * (pCtx->sStatData[i][j].sSliceData.iMbCount[I_SLICE][Intra16x16] + pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Intra16x16]) / iMbCount),
(100.0f * (pCtx->sStatData[i][j].sSliceData.iMbCount[I_SLICE][7] + pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][7]) / iMbCount),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter16x16] / iMbCount ),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter16x8] / iMbCount ),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter8x16] / iMbCount ),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][Inter8x8] / iMbCount),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][10] / iMbCount),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][PSkip] / iMbCount),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][11] / iMbCount),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][11] / count_p_mbL0),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][8] / count_p_mbL0),
(100.0f * pCtx->sStatData[i][j].sSliceData.iMbCount[P_SLICE][9] / count_p_mbL0)
);
}
}
#endif //#if defined(MB_TYPES_CHECK)
if (iCount > 0){
fprintf( stdout, "SVC: overall PSNR Y: %2.3f U: %2.3f V: %2.3f kb/s: %.1f fps: %.3f\n\n",
(pCtx->sStatData[i][j].sQualityStat.rYPsnr[I_SLICE]+pCtx->sStatData[i][j].sQualityStat.rYPsnr[P_SLICE]+pCtx->sStatData[i][j].sQualityStat.rYPsnr[B_SLICE]) / (float)(iCount),
(pCtx->sStatData[i][j].sQualityStat.rUPsnr[I_SLICE]+pCtx->sStatData[i][j].sQualityStat.rUPsnr[P_SLICE]+pCtx->sStatData[i][j].sQualityStat.rUPsnr[B_SLICE]) / (float)(iCount),
(pCtx->sStatData[i][j].sQualityStat.rVPsnr[I_SLICE]+pCtx->sStatData[i][j].sQualityStat.rVPsnr[P_SLICE]+pCtx->sStatData[i][j].sQualityStat.rVPsnr[B_SLICE]) / (float)(iCount),
1.0f * pCtx->pSvcParam->sDependencyLayers[i].fOutputFrameRate *(pCtx->sStatData[i][j].sSliceData.iSliceSize[I_SLICE] +pCtx->sStatData[i][j].sSliceData.iSliceSize[P_SLICE] +pCtx->sStatData[i][j].sSliceData.iSliceSize[B_SLICE] ) / (float)(iCount+pCtx->pWelsSvcRc[i].iSkipFrameNum)/1000,
1.0f * pCtx->pSvcParam->sDependencyLayers[i].fOutputFrameRate );
}
}
}
}
#endif
/*!
* \brief uninitialize Wels encoder core library
* \pParam pEncCtx sWelsEncCtx*
* \return none
*/
void WelsUninitEncoderExt( sWelsEncCtx **ppCtx )
{
if ( NULL == ppCtx || NULL == *ppCtx )
return;
WelsLog( *ppCtx, WELS_LOG_INFO, "WelsUninitEncoderExt(), pCtx= %p, iThreadCount= %d, iMultipleThreadIdc= %d.\n", (void *)(*ppCtx), (*ppCtx)->pSvcParam->iCountThreadsNum, (*ppCtx)->pSvcParam->iMultipleThreadIdc );
#if defined(STAT_OUTPUT)
StatOverallEncodingExt( *ppCtx );
#endif
#if defined(MT_ENABLED)
if ( (*ppCtx)->pSvcParam->iMultipleThreadIdc > 1 && (*ppCtx)->pSliceThreading != NULL )
{
const int32_t iThreadCount = (*ppCtx)->pSvcParam->iCountThreadsNum;
int32_t iThreadIdx = 0;
#if defined(WIN32)
if ( (*ppCtx)->pSliceThreading->pExitEncodeEvent != NULL )
{
do {
if ( (*ppCtx)->pSliceThreading->pThreadHandles[iThreadIdx] != NULL ) // iThreadIdx is already created successfully
WelsEventSignal( &(*ppCtx)->pSliceThreading->pExitEncodeEvent[iThreadIdx] );
++ iThreadIdx;
} while(iThreadIdx < iThreadCount);
WelsMultipleEventsWaitAllBlocking( iThreadCount, &(*ppCtx)->pSliceThreading->pFinSliceCodingEvent[0] );
}
#elif defined(__GNUC__)
while ( iThreadIdx < iThreadCount )
{
int res = 0;
if ( (*ppCtx)->pSliceThreading->pThreadHandles[iThreadIdx] )
{
res = WelsThreadCancel( (*ppCtx)->pSliceThreading->pThreadHandles[iThreadIdx] );
WelsLog( *ppCtx, WELS_LOG_INFO, "WelsUninitEncoderExt(), WelsThreadCancel(pThreadHandles%d) return %d..\n", iThreadIdx, res);
res = WelsThreadJoin( (*ppCtx)->pSliceThreading->pThreadHandles[iThreadIdx] ); // waiting thread exit
WelsLog( *ppCtx, WELS_LOG_INFO, "WelsUninitEncoderExt(), pthread_join(pThreadHandles%d) return %d..\n", iThreadIdx, res);
(*ppCtx)->pSliceThreading->pThreadHandles[iThreadIdx] = 0;
}
#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
if ( (*ppCtx)->pSliceThreading->pUpdateMbListThrdHandles[iThreadIdx] )
{
res = WelsThreadCancel( (*ppCtx)->pSliceThreading->pUpdateMbListThrdHandles[iThreadIdx] );
WelsLog( *ppCtx, WELS_LOG_INFO, "WelsUninitEncoderExt(), WelsThreadCancel(pUpdateMbListThrdHandles%d) return %d..\n", iThreadIdx, res);
res = WelsThreadJoin( (*ppCtx)->pSliceThreading->pUpdateMbListThrdHandles[iThreadIdx] ); // waiting thread exit
WelsLog( *ppCtx, WELS_LOG_INFO, "WelsUninitEncoderExt(), pthread_join(pUpdateMbListThrdHandles%d) return %d..\n", iThreadIdx, res);
(*ppCtx)->pSliceThreading->pUpdateMbListThrdHandles[iThreadIdx] = 0;
}
#endif//DYNAMIC_SLICE_ASSIGN && TRY_SLICING_BALANCE
++ iThreadIdx;
}
#endif//WIN32
}
#endif//MT_ENABLED
if ((*ppCtx)->pVpp)
{
delete (*ppCtx)->pVpp;
(*ppCtx)->pVpp = NULL;
}
FreeMemorySvc( ppCtx );
*ppCtx = NULL;
}
/*!
* \brief get temporal level due to configuration and coding context
*/
static inline int32_t GetTemporalLevel( SDLayerParam *fDlp, const int32_t kiFrameNum, const int32_t kiGopSize )
{
const int32_t kiCodingIdx = kiFrameNum & (kiGopSize-1);
return fDlp->uiCodingIdx2TemporalId[kiCodingIdx];
}
void DynslcUpdateMbNeighbourInfoListForAllSlices( SSliceCtx *pSliceCtx, SMB *pMbList )
{
const int32_t kiMbWidth = pSliceCtx->iMbWidth;
const int32_t kiEndMbInSlice = pSliceCtx->iMbNumInFrame - 1;
int32_t iIdx = 0;
do {
SMB *pMb = &pMbList[iIdx];
uint32_t uiNeighborAvailFlag = 0;
const int32_t kiMbXY = pMb->iMbXY;
const int32_t kiMbX = pMb->iMbX;
const int32_t kiMbY = pMb->iMbY;
BOOL_T bLeft;
BOOL_T bTop;
BOOL_T bLeftTop;
BOOL_T bRightTop;
int32_t uiSliceIdc;
int32_t iLeftXY, iTopXY, iLeftTopXY, iRightTopXY;
uiSliceIdc = WelsMbToSliceIdc(pSliceCtx, kiMbXY);
pMb->uiSliceIdc = uiSliceIdc;
iLeftXY = kiMbXY - 1;
iTopXY = kiMbXY - kiMbWidth;
iLeftTopXY = iTopXY - 1;
iRightTopXY = iTopXY + 1;
bLeft = (kiMbX > 0) && (uiSliceIdc == WelsMbToSliceIdc(pSliceCtx, iLeftXY));
bTop = (kiMbY > 0) && (uiSliceIdc == WelsMbToSliceIdc(pSliceCtx, iTopXY));
bLeftTop = (kiMbX > 0) && (kiMbY > 0) && (uiSliceIdc == WelsMbToSliceIdc(pSliceCtx, iLeftTopXY));
bRightTop = (kiMbX < (kiMbWidth-1)) && (kiMbY > 0) && (uiSliceIdc == WelsMbToSliceIdc(pSliceCtx, iRightTopXY));
if( bLeft ){
uiNeighborAvailFlag |= LEFT_MB_POS;
}
if( bTop ){
uiNeighborAvailFlag |= TOP_MB_POS;
}
if( bLeftTop ){
uiNeighborAvailFlag |= TOPLEFT_MB_POS;
}
if( bRightTop ){
uiNeighborAvailFlag |= TOPRIGHT_MB_POS;
}
pMb->uiNeighborAvail = (uint8_t)uiNeighborAvailFlag;
++ iIdx;
} while(iIdx <= kiEndMbInSlice);
}
/*
* TUNE back if number of picture partition decision algorithm based on past if available
*/
int32_t PicPartitionNumDecision( sWelsEncCtx *pCtx )
{
int32_t iPartitionNum = 1;
#ifdef MT_ENABLED
if ( pCtx->pSvcParam->iMultipleThreadIdc > 1 )
{
iPartitionNum = pCtx->pSvcParam->iCountThreadsNum;
#if !defined(FIXED_PARTITION_ASSIGN)
if ( P_SLICE == pCtx->eSliceType )
iPartitionNum = 1;
#endif//!FIXED_PARTITION_ASSIGN
}
return iPartitionNum;
#else
return iPartitionNum;
#endif//MT_ENABLED
}
#if defined(MT_ENABLED)
void WelsInitCurrentQBLayerMltslc( sWelsEncCtx *pCtx )
{
//pData init
SDqLayer* pCurDq = pCtx->pCurDqLayer;
SSliceCtx* pSliceCtx = (pCurDq->pSliceEncCtx);
//mb_neighbor
DynslcUpdateMbNeighbourInfoListForAllSlices( pSliceCtx, pCurDq->sMbDataP );
}
void UpdateSlicepEncCtxWithPartition( SSliceCtx *pSliceCtx, int32_t iPartitionNum )
{
const int32_t kiMbNumInFrame = pSliceCtx->iMbNumInFrame;
int32_t iCountMbNumPerPartition = kiMbNumInFrame;
int32_t iAssignableMbLeft = kiMbNumInFrame;
int32_t iFirstMbIdx = 0;
int32_t i/*, j*/;
if ( iPartitionNum <= 0 )
iPartitionNum = 1;
else if ( iPartitionNum > AVERSLICENUM_CONSTRAINT )
iPartitionNum = AVERSLICENUM_CONSTRAINT; // AVERSLICENUM_CONSTRAINT might be variable, however not fixed by MACRO
iCountMbNumPerPartition /= iPartitionNum;
pSliceCtx->iSliceNumInFrame = iPartitionNum;
i = 0;
while( i < iPartitionNum )
{
if ( i + 1 == iPartitionNum )
{
pSliceCtx->pCountMbNumInSlice[i] = iAssignableMbLeft;
}
else
{
pSliceCtx->pCountMbNumInSlice[i] = iCountMbNumPerPartition;
}
pSliceCtx->pFirstMbInSlice[i] = iFirstMbIdx;
memset( pSliceCtx->pOverallMbMap+iFirstMbIdx, (uint8_t)i, pSliceCtx->pCountMbNumInSlice[i]*sizeof(uint8_t) );
// for next partition(or pSlice)
iFirstMbIdx += pSliceCtx->pCountMbNumInSlice[i];
iAssignableMbLeft -= pSliceCtx->pCountMbNumInSlice[i];
++ i;
}
}
void WelsInitCurrentDlayerMltslc( sWelsEncCtx *pCtx, int32_t iPartitionNum )
{
SDqLayer* pCurDq = pCtx->pCurDqLayer;
SSliceCtx* pSliceCtx = pCurDq->pSliceEncCtx;
UpdateSlicepEncCtxWithPartition( pSliceCtx, iPartitionNum );
if ( I_SLICE == pCtx->eSliceType )//check if uiSliceSizeConstraint too small
{
#define byte_complexIMBat26 (60)
uint8_t iCurDid = pCtx->uiDependencyId;
uint32_t uiFrmByte = 0;
if ( pCtx->pSvcParam->bEnableRc )
{//RC case
uiFrmByte = (
( (uint32_t)(pCtx->pSvcParam->sDependencyLayers[iCurDid].iSpatialBitrate)
/(uint32_t)(pCtx->pSvcParam->sDependencyLayers[iCurDid].fInputFrameRate) ) >> 3 );
}
else
{//fixed QP case
const int32_t iTtlMbNumInFrame = pSliceCtx->iMbNumInFrame;
int32_t iQDeltaTo26 = ( 26 - pCtx->pSvcParam->sDependencyLayers[iCurDid].iDLayerQp );
uiFrmByte = (iTtlMbNumInFrame * byte_complexIMBat26);
if ( iQDeltaTo26 > 0 )
{
//smaller QP than 26
uiFrmByte = (uint32_t)( uiFrmByte * ( (float)iQDeltaTo26 / 4 ) );
}
else if ( iQDeltaTo26 < 0 )
{
//larger QP than 26
iQDeltaTo26 = ( (-iQDeltaTo26) >> 2 ); //delta mod 4
uiFrmByte = ( uiFrmByte >> (iQDeltaTo26) ); //if delta 4, byte /2
}
}
//MINPACKETSIZE_CONSTRAINT
if ( pSliceCtx->uiSliceSizeConstraint
<
(uint32_t)( uiFrmByte//suppose 16 byte per mb at average
/ ( pSliceCtx->iMaxSliceNumConstraint ) )
)
{
WelsLog( pCtx,
WELS_LOG_WARNING,
"Set-SliceConstraint(%d) too small for current resolution (MB# %d) under QP/BR!\n",
pSliceCtx->uiSliceSizeConstraint,
pSliceCtx->iMbNumInFrame
);
}
}
WelsInitCurrentQBLayerMltslc( pCtx );
}
#else
void WelsInitCurrentQBLayerMltslc( sWelsEncCtx *pCtx )
{
//pData init
SDqLayer* pCurDq = pCtx->pCurDqLayer;
SSliceCtx* pSliceCtx = (pCurDq->pSliceEncCtx);
SSlice * pSlice = &pCurDq->sLayerInfo.pSliceInLayer[0];
int32_t iTtlMbNumInFrame = pSliceCtx->iMbNumInFrame;
//pSliceCtx
memset( pSliceCtx->pOverallMbMap, 0, iTtlMbNumInFrame * sizeof(uint8_t) );
memset( pSliceCtx->pCountMbNumInSlice, 0, pSliceCtx->iSliceNumInFrame * sizeof(int32_t) );
memset( pSliceCtx->pFirstMbInSlice, 0, pSliceCtx->iSliceNumInFrame * sizeof(int16_t) );
pSliceCtx->iSliceNumInFrame = 1;//
pSliceCtx->pCountMbNumInSlice[0] = iTtlMbNumInFrame;
//mb_neighbor
DynslcUpdateMbNeighbourInfoListForAllSlices( pSliceCtx, pCurDq->sMbDataP );
//pSlice init
pSlice->uiSliceIdx = 0;
pSlice->pSliceBsa = &pCtx->pOut->sBsWrite;
pSlice->bDynamicSlicingSliceSizeCtrlFlag = false;
pSlice->uiAssumeLog2BytePerMb = ( pCtx->eSliceType == P_SLICE ) ? 0 : 1;
}
void WelsInitCurrentDlayerMltslc( sWelsEncCtx *pCtx, int32_t iPartitionNum )
{
SDqLayer* pCurDq = pCtx->pCurDqLayer;
SSliceCtx* pSliceCtx = ( pCurDq->pSliceEncCtx );
int32_t iTtlMbNumInFrame = pCurDq->iMbHeight*pCurDq->iMbWidth;
pSliceCtx->iMbNumInFrame
= pSliceCtx->pCountMbNumInSlice[0] = iTtlMbNumInFrame;
if ( I_SLICE == pCtx->eSliceType )//check if uiSliceSizeConstraint too small
{
#define byte_complexIMBat26 (60)
uint8_t iCurDid = pCtx->uiDependencyId;
uint32_t uiFrmByte = 0;
if ( pCtx->pSvcParam->bEnableRc )
{//RC case
uiFrmByte = (
( (uint32_t)(pCtx->pSvcParam->sDependencyLayers[iCurDid].iSpatialBitrate)
/(uint32_t)(pCtx->pSvcParam->sDependencyLayers[iCurDid].fInputFrameRate) ) >> 3 );
}
else
{//fixed QP case
int32_t iQDeltaTo26 = ( 26 - pCtx->pSvcParam->sDependencyLayers[iCurDid].iDLayerQp );
uiFrmByte = (iTtlMbNumInFrame * byte_complexIMBat26);
if ( iQDeltaTo26 > 0 )
{
//smaller QP than 26
uiFrmByte = (uint32_t)( uiFrmByte * ( (float)iQDeltaTo26 / 4 ) );
}
else if ( iQDeltaTo26 < 0 )
{
//larger QP than 26
iQDeltaTo26 = ( (-iQDeltaTo26) >> 2 ); //delta mod 4
uiFrmByte = ( uiFrmByte >> (iQDeltaTo26) ); //if delta 4, byte /2
}
}
//MINPACKETSIZE_CONSTRAINT
if ( pSliceCtx->uiSliceSizeConstraint
<
(uint32_t)( uiFrmByte//suppose 16 byte per mb at average
/ ( pSliceCtx->iMaxSliceNumConstraint ) )
)
{
WelsLog( pCtx,
WELS_LOG_WARNING,
"Set-SliceConstraint(%d) too small for current resolution (MB# %d) under QP/BR!\n",
pSliceCtx->uiSliceSizeConstraint,
pSliceCtx->iMbNumInFrame
);
}
}
WelsInitCurrentQBLayerMltslc( pCtx );
}
#endif
/*!
* \brief initialize current layer
*/
void WelsInitCurrentLayer( sWelsEncCtx *pCtx,
const int32_t kiWidth,
const int32_t kiHeight )
{
SWelsSvcCodingParam *pParam = pCtx->pSvcParam;
SPicture *pEncPic = pCtx->pEncPic;
SPicture *pDecPic = pCtx->pDecPic;
SDqLayer *pCurDq = pCtx->pCurDqLayer;
SSlice *pBaseSlice = &pCurDq->sLayerInfo.pSliceInLayer[0];
SSlice *pSlice = NULL;
const uint8_t kiCurDid = pCtx->uiDependencyId;
const bool_t kbUseSubsetSpsFlag= (kiCurDid > BASE_DEPENDENCY_ID);
SDLayerParam *fDlp = &pParam->sDependencyLayers[kiCurDid];
SNalUnitHeaderExt *pNalHdExt = &pCurDq->sLayerInfo.sNalHeaderExt;
SNalUnitHeader *pNalHd = &pNalHdExt->sNalHeader;
SDqIdc *pDqIdc = &pCtx->pDqIdcMap[kiCurDid];
int32_t iIdx = 0;
int32_t iSliceCount = 0;
if ( NULL == pCurDq )
return;
pCurDq->pDecPic = pDecPic;
if ( fDlp->sMso.uiSliceMode == SM_DYN_SLICE ) // need get extra slices for update
iSliceCount = GetInitialSliceNum( pCurDq->iMbWidth, pCurDq->iMbHeight, &fDlp->sMso );
else
iSliceCount = GetCurrentSliceNum( pCurDq->pSliceEncCtx );
assert( iSliceCount > 0 );
pBaseSlice->sSliceHeaderExt.sSliceHeader.iPpsId = pDqIdc->iPpsId;
pCurDq->sLayerInfo.pPpsP =
pBaseSlice->sSliceHeaderExt.sSliceHeader.pPps = &pCtx->pPPSArray[pBaseSlice->sSliceHeaderExt.sSliceHeader.iPpsId];
pBaseSlice->sSliceHeaderExt.sSliceHeader.iSpsId = pDqIdc->iSpsId;
if ( kbUseSubsetSpsFlag )
{
pCurDq->sLayerInfo.pSubsetSpsP = &pCtx->pSubsetArray[pDqIdc->iSpsId];
pCurDq->sLayerInfo.pSpsP =
pBaseSlice->sSliceHeaderExt.sSliceHeader.pSps = &pCurDq->sLayerInfo.pSubsetSpsP->pSps;
}
else
{
pCurDq->sLayerInfo.pSubsetSpsP = NULL;
pCurDq->sLayerInfo.pSpsP =
pBaseSlice->sSliceHeaderExt.sSliceHeader.pSps = &pCtx->pSpsArray[pBaseSlice->sSliceHeaderExt.sSliceHeader.iSpsId];
}
pSlice = pBaseSlice;
iIdx = 1;
while ( iIdx < iSliceCount ) {
++ pSlice;
pSlice->sSliceHeaderExt.sSliceHeader.iPpsId = pBaseSlice->sSliceHeaderExt.sSliceHeader.iPpsId;
pSlice->sSliceHeaderExt.sSliceHeader.pPps = pBaseSlice->sSliceHeaderExt.sSliceHeader.pPps;
pSlice->sSliceHeaderExt.sSliceHeader.iSpsId = pBaseSlice->sSliceHeaderExt.sSliceHeader.iSpsId;
pSlice->sSliceHeaderExt.sSliceHeader.pSps = pBaseSlice->sSliceHeaderExt.sSliceHeader.pSps;
++ iIdx;
}
memset( pNalHdExt, 0, sizeof(SNalUnitHeaderExt) );
pNalHd->uiNalRefIdc = pCtx->eNalPriority;
pNalHd->eNalUnitType = pCtx->eNalType;
pNalHdExt->uiDependencyId = kiCurDid;
pNalHdExt->bDiscardableFlag = (pCtx->bNeedPrefixNalFlag) ? (pNalHd->uiNalRefIdc == NRI_PRI_LOWEST) : false;
pNalHdExt->bIdrFlag = (pCtx->iFrameNum == 0) && ((pCtx->eNalType == NAL_UNIT_CODED_SLICE_IDR) || (pCtx->eSliceType == I_SLICE));
pNalHdExt->uiTemporalId = pCtx->uiTemporalId;
pBaseSlice->bSliceHeaderExtFlag = (NAL_UNIT_CODED_SLICE_EXT == pNalHd->eNalUnitType);
pSlice = pBaseSlice;
iIdx = 1;
while (iIdx < iSliceCount) {
++ pSlice;
pSlice->bSliceHeaderExtFlag = pBaseSlice->bSliceHeaderExtFlag;
++ iIdx;
}
// pEncPic pData
pCurDq->pEncData[0] = pEncPic->pData[0];
pCurDq->pEncData[1] = pEncPic->pData[1];
pCurDq->pEncData[2] = pEncPic->pData[2];
pCurDq->iEncStride[0] = pEncPic->iLineSize[0];
pCurDq->iEncStride[1] = pEncPic->iLineSize[1];
pCurDq->iEncStride[2] = pEncPic->iLineSize[2];
// cs pData
pCurDq->pCsData[0] = pDecPic->pData[0];
pCurDq->pCsData[1] = pDecPic->pData[1];
pCurDq->pCsData[2] = pDecPic->pData[2];
pCurDq->iCsStride[0] = pDecPic->iLineSize[0];
pCurDq->iCsStride[1] = pDecPic->iLineSize[1];
pCurDq->iCsStride[2] = pDecPic->iLineSize[2];
if ( pCurDq->pRefLayer != NULL )
{
pCurDq->bBaseLayerAvailableFlag = true;
}
else
{
pCurDq->bBaseLayerAvailableFlag = false;
}
}
void PreprocessSliceCoding( sWelsEncCtx *pCtx )
{
SDqLayer *pCurLayer = pCtx->pCurDqLayer;
const bool_t kbBaseAvail = pCurLayer->bBaseLayerAvailableFlag;
/* function pointers conditional assignment under sWelsEncCtx, layer_mb_enc_rec (in stack) is exclusive */
if ( P_SLICE == pCtx->eSliceType )
{
if ( kbBaseAvail )
{
if ( pCtx->pSvcParam->iNumDependencyLayer == (pCurLayer->sLayerInfo.sNalHeaderExt.uiDependencyId + 1) ) //
{
pCtx->pFuncList->pfMotionSearch = WelsMotionEstimateSearchSad;
pCtx->pFuncList->pfFirstIntraMode = WelsMdFirstIntraMode;
pCtx->pFuncList->pfIntraFineMd = WelsMdIntraFinePartitionVaa;
pCtx->pFuncList->pfInterFineMd = WelsMdInterFinePartitionVaa;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3Sad;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3Sad;
pCtx->pFuncList->sSampleDealingFuncs.pfMdCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSad;
}
else
{
pCtx->pFuncList->pfMotionSearch = WelsMotionEstimateSearchSatd;
pCtx->pFuncList->pfFirstIntraMode = WelsMdFirstIntraMode;
pCtx->pFuncList->pfIntraFineMd = WelsMdIntraFinePartition;
pCtx->pFuncList->pfInterFineMd = WelsMdInterFinePartition;
pCtx->pFuncList->sSampleDealingFuncs.pfMdCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSatd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3Satd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3Satd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra4x4Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra4x4Combined3Satd;
}
pCtx->pFuncList->sSampleDealingFuncs.pfMeCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSatd;
}
else
{
//case 3: pBase layer MD + encoding
if ( pCurLayer->sLayerInfo.sNalHeaderExt.uiDependencyId+1 == pCtx->pSvcParam->iNumDependencyLayer )
{
pCtx->pFuncList->pfMotionSearch = WelsMotionEstimateSearchSad;
pCtx->pFuncList->pfFirstIntraMode = WelsMdFirstIntraMode;
pCtx->pFuncList->pfIntraFineMd = WelsMdIntraFinePartitionVaa;
pCtx->pFuncList->pfInterFineMd = WelsMdInterFinePartitionVaa;
pCtx->pFuncList->sSampleDealingFuncs.pfMdCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSad;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3Sad;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3Sad;
}
else
{
pCtx->pFuncList->pfMotionSearch = WelsMotionEstimateSearchSatd;
pCtx->pFuncList->pfFirstIntraMode = WelsMdFirstIntraMode;
pCtx->pFuncList->pfIntraFineMd = WelsMdIntraFinePartition;
pCtx->pFuncList->pfInterFineMd = WelsMdInterFinePartition;
pCtx->pFuncList->sSampleDealingFuncs.pfMdCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSatd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3Satd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3Satd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra4x4Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra4x4Combined3Satd;
}
pCtx->pFuncList->sSampleDealingFuncs.pfMeCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSatd;
}
}
else if ( I_SLICE == pCtx->eSliceType )
{
if ( pCurLayer->sLayerInfo.sNalHeaderExt.uiDependencyId+1 == pCtx->pSvcParam->iNumDependencyLayer )
{
pCtx->pFuncList->sSampleDealingFuncs.pfMdCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSad;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3Sad;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3Sad;
pCtx->pFuncList->pfIntraFineMd = WelsMdIntraFinePartitionVaa;
}
else
{
pCtx->pFuncList->sSampleDealingFuncs.pfMdCost = pCtx->pFuncList->sSampleDealingFuncs.pfSampleSatd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra16x16Combined3Satd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra8x8Combined3Satd;
pCtx->pFuncList->sSampleDealingFuncs.pfIntra4x4Combined3 = pCtx->pFuncList->sSampleDealingFuncs.pfIntra4x4Combined3Satd;
pCtx->pFuncList->pfIntraFineMd = WelsMdIntraFinePartition;
}
}
}
/*!
* \brief swap pDq layers between current pDq layer and reference pDq layer
*/
static inline void WelsSwapDqLayers( sWelsEncCtx *pCtx )
{
// swap and assign reference
const int32_t kiDid = pCtx->uiDependencyId;
const int32_t kiNextDqIdx = 1 + kiDid;
SDqLayer *pTmpLayer = pCtx->ppDqLayerList[kiNextDqIdx];
SDqLayer *pRefLayer = pCtx->pCurDqLayer;
pCtx->pCurDqLayer = pTmpLayer;
pCtx->pCurDqLayer->pRefLayer = pRefLayer;
}
/*!
* \brief prefetch reference picture after WelsBuildRefList
*/
static inline void PrefetchReferencePicture( sWelsEncCtx *pCtx, const EFrameType keFrameType )
{
SSlice *pSliceBase = &pCtx->pCurDqLayer->sLayerInfo.pSliceInLayer[0];
const int32_t kiSliceCount = GetCurrentSliceNum( pCtx->pCurDqLayer->pSliceEncCtx );
int32_t iIdx = 0;
uint8_t uiRefIdx = -1;
assert( kiSliceCount > 0 );
if ( keFrameType != WELS_FRAME_TYPE_IDR )
{
assert( pCtx->iNumRef0 > 0 );
pCtx->pRefPic = pCtx->pRefList0[0]; // always get item 0 due to reordering done
pCtx->pCurDqLayer->pRefPic = pCtx->pRefPic;
uiRefIdx = 0; // reordered reference iIndex
}
else // safe for IDR coding
{
pCtx->pRefPic = NULL;
pCtx->pCurDqLayer->pRefPic = NULL;
}
iIdx = 0;
while (iIdx < kiSliceCount) {
pSliceBase->sSliceHeaderExt.sSliceHeader.uiRefIndex = uiRefIdx;
++ pSliceBase;
++ iIdx;
}
}
void ParasetIdAdditionIdAdjust( SParaSetOffsetVariable *sParaSetOffsetVariable, const int32_t kiCurEncoderParaSetId, const uint32_t kuiMaxIdInBs )//paraset_type = 0: SPS; =1: PPS
{
//SPS_ID in avc_sps and pSubsetSps will be different using this
//SPS_ID case example:
//1st enter: next_spsid_in_bs == 0; spsid == 0; delta==0; //actual spsid_in_bs == 0
//1st finish: next_spsid_in_bs == 1;
//2nd enter: next_spsid_in_bs == 1; spsid == 0; delta==1; //actual spsid_in_bs == 1
//2nd finish: next_spsid_in_bs == 2;
//31st enter: next_spsid_in_bs == 31; spsid == 0~2; delta==31~29; //actual spsid_in_bs == 31
//31st finish: next_spsid_in_bs == 0;
//31st enter: next_spsid_in_bs == 0; spsid == 0~2; delta==-2~0; //actual spsid_in_bs == 0
//31st finish: next_spsid_in_bs == 1;
const int32_t kiEncId = kiCurEncoderParaSetId;
const uint32_t kuiPrevIdInBs = sParaSetOffsetVariable->iParaSetIdDelta[kiEncId] + kiEncId;//mark current_id
const bool_t *kpUsedIdPointer = &sParaSetOffsetVariable->bUsedParaSetIdInBs[0];
uint32_t uiNextIdInBs = sParaSetOffsetVariable->uiNextParaSetIdToUseInBs;
#if _DEBUG
if ( 0 != sParaSetOffsetVariable->iParaSetIdDelta[kiEncId] )
assert ( sParaSetOffsetVariable->bUsedParaSetIdInBs[kuiPrevIdInBs] ); //sure the prev-used one was marked activated correctly
#endif
//update current layer's pCodingParam
sParaSetOffsetVariable->iParaSetIdDelta[kiEncId] = uiNextIdInBs - kiEncId; //for current parameter set, change its id_delta
//write pso pData for next update:
sParaSetOffsetVariable->bUsedParaSetIdInBs[kuiPrevIdInBs] = false; //
sParaSetOffsetVariable->bUsedParaSetIdInBs[uiNextIdInBs] = true; // update current used_id
//prepare for next update:
// find the next avaibable iId
do
{
++uiNextIdInBs;
if (uiNextIdInBs >= kuiMaxIdInBs )
{
uiNextIdInBs = 0;//ensure the SPS_ID wound not exceed MAX_SPS_COUNT
}
}while ( kpUsedIdPointer[uiNextIdInBs] );
// update next_id
sParaSetOffsetVariable->uiNextParaSetIdToUseInBs = uiNextIdInBs;
#if _DEBUG
assert ( !sParaSetOffsetVariable->bUsedParaSetIdInBs[uiNextIdInBs] ); //sure the next-to-use one is marked activated correctly
#endif
}
/*!
* \brief write all parameter sets introduced in SVC extension
* \return size in bytes of bitstream wrote
*/
int32_t WelsWriteParameterSets( sWelsEncCtx *pCtx, int32_t *pNalLen, int32_t *pNumNal )
{
int32_t iSize = 0;
int32_t iNal = 0;
int32_t iIdx = 0;
int32_t iId = 0;
int32_t iCountNal = 0;
if ( NULL == pCtx || NULL == pNalLen || NULL == pNumNal )
return 0;
/* write all SPS */
iIdx = 0;
while (iIdx < pCtx->iSpsNum) {
SDqIdc *pDqIdc = &pCtx->pDqIdcMap[iIdx];
const int32_t kiDid = pDqIdc->uiSpatialId;
const bool_t kbUsingSubsetSps = (kiDid > BASE_DEPENDENCY_ID);
iNal = pCtx->pOut->iNalIndex;
if ( pCtx->pSvcParam->bEnableSpsPpsIdAddition )
{
#if _DEBUG
pCtx->sPSOVector.bEnableSpsPpsIdAddition = 1;
assert(kiDid < MAX_DEPENDENCY_LAYER);
assert(iIdx < MAX_DQ_LAYER_NUM);
#endif
ParasetIdAdditionIdAdjust( &(pCtx->sPSOVector.sParaSetOffsetVariable[kbUsingSubsetSps ? PARA_SET_TYPE_SUBSETSPS : PARA_SET_TYPE_AVCSPS]),
(kbUsingSubsetSps)?(pCtx->pSubsetArray[iIdx - 1].pSps.uiSpsId):(pCtx->pSpsArray[0].uiSpsId ),
MAX_SPS_COUNT );
}
else
{
memset(&(pCtx->sPSOVector), 0, sizeof(pCtx->sPSOVector) );
}
if ( kbUsingSubsetSps ){
iId = iIdx - 1;
/* generate Subset SPS */
WelsLoadNal( pCtx->pOut, NAL_UNIT_SUBSET_SPS, NRI_PRI_HIGHEST );
WelsWriteSubsetSpsSyntax( &pCtx->pSubsetArray[iId], &pCtx->pOut->sBsWrite, &(pCtx->sPSOVector.sParaSetOffsetVariable[PARA_SET_TYPE_SUBSETSPS].iParaSetIdDelta[0]) );
WelsUnloadNal( pCtx->pOut );
}
else{
iId = 0;
/* generate sequence parameters set */
WelsLoadNal( pCtx->pOut, NAL_UNIT_SPS, NRI_PRI_HIGHEST );
WelsWriteSpsNal( &pCtx->pSpsArray[0], &pCtx->pOut->sBsWrite, &(pCtx->sPSOVector.sParaSetOffsetVariable[PARA_SET_TYPE_AVCSPS].iParaSetIdDelta[0]) );
WelsUnloadNal( pCtx->pOut );
}
pNalLen[iCountNal] = WelsEncodeNal( &pCtx->pOut->sNalList[iNal], pCtx->pFrameBs + pCtx->iPosBsBuffer, &pNalLen[iCountNal] );
pCtx->iPosBsBuffer += pNalLen[iCountNal];
iSize += pNalLen[iCountNal];
++ iIdx;
++ iCountNal;
}
/* write all PPS */
iIdx = 0;
while (iIdx < pCtx->iPpsNum) {
if ( pCtx->pSvcParam->bEnableSpsPpsIdAddition )
{
//para_set_type = 2: PPS, use MAX_PPS_COUNT
ParasetIdAdditionIdAdjust( &pCtx->sPSOVector.sParaSetOffsetVariable[PARA_SET_TYPE_PPS], pCtx->pPPSArray[iIdx].iPpsId, MAX_PPS_COUNT );
}
iNal = pCtx->pOut->iNalIndex;
/* generate picture parameter set */
WelsLoadNal( pCtx->pOut, NAL_UNIT_PPS, NRI_PRI_HIGHEST );
WelsWritePpsSyntax( &pCtx->pPPSArray[iIdx], &pCtx->pOut->sBsWrite, &(pCtx->sPSOVector) );
WelsUnloadNal( pCtx->pOut );
pNalLen[iCountNal] = WelsEncodeNal( &pCtx->pOut->sNalList[iNal], pCtx->pFrameBs + pCtx->iPosBsBuffer, &pNalLen[iCountNal] );
pCtx->iPosBsBuffer += pNalLen[iCountNal];
iSize += pNalLen[iCountNal];
++ iIdx;
++ iCountNal;
}
*pNumNal = iCountNal;
return iSize;
}
static inline int32_t AddPrefixNal( sWelsEncCtx *pCtx,
SLayerBSInfo *pLayerBsInfo,
int32_t *pNalLen,
int32_t *pNalIdxInLayer,
const EWelsNalUnitType keNalType,
const EWelsNalRefIdc keNalRefIdc )
{
int32_t iPayloadSize = 0;
if ( keNalRefIdc != NRI_PRI_LOWEST )
{
WelsLoadNal( pCtx->pOut, NAL_UNIT_PREFIX, keNalRefIdc );
WelsWriteSVCPrefixNal( &pCtx->pOut->sBsWrite, keNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == keNalType) );
WelsUnloadNal( pCtx->pOut );
iPayloadSize = WelsEncodeNalExt( &pCtx->pOut->sNalList[pCtx->pOut->iNalIndex-1],
&pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt,
pCtx->pFrameBs + pCtx->iPosBsBuffer,
&pNalLen[*pNalIdxInLayer] );
pCtx->iPosBsBuffer += iPayloadSize;
pLayerBsInfo->iNalLengthInByte[*pNalIdxInLayer] = iPayloadSize;
(*pNalIdxInLayer) ++;
}
else // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
{
WelsLoadNal( pCtx->pOut, NAL_UNIT_PREFIX, keNalRefIdc );
// No need write any syntax of prefix NAL Unit RBSP here
WelsUnloadNal( pCtx->pOut );
iPayloadSize = WelsEncodeNalExt( &pCtx->pOut->sNalList[pCtx->pOut->iNalIndex-1],
&pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt,
pCtx->pFrameBs + pCtx->iPosBsBuffer,
&pNalLen[*pNalIdxInLayer] );
pCtx->iPosBsBuffer += iPayloadSize;
pLayerBsInfo->iNalLengthInByte[*pNalIdxInLayer] = iPayloadSize;
(*pNalIdxInLayer) ++;
}
return iPayloadSize;
}
int32_t WritePadding(sWelsEncCtx *pCtx, int32_t iLen)
{
int32_t i=0;
int32_t iNal = 0;
SBitStringAux *pBs = NULL;
int32_t iNalLen;
int32_t iSize=0;
iNal = pCtx->pOut->iNalIndex;
pBs = &pCtx->pOut->sBsWrite; // SBitStringAux instance for non VCL NALs decoding
if((pBs->pBufEnd - pBs->pBufPtr) < iLen || iNal >= pCtx->pOut->iCountNals)
{
#if GOM_TRACE_FLAG
WelsLog( pCtx, WELS_LOG_ERROR,"[RC] paddingcal pBuffer overflow, bufferlen=%d, paddinglen=%d, iNalIdx= %d, iCountNals= %d\n",
(pBs->pBufEnd-pBs->pBufPtr), iLen, iNal, pCtx->pOut->iCountNals);
#endif
return 0;
}
WelsLoadNal( pCtx->pOut, NAL_UNIT_FILLER_DATA, NRI_PRI_LOWEST );
for(i=0;i<iLen;i++)
{
BsWriteBits( pBs, 8, 0xff);
}
BsRbspTrailingBits( pBs );
BsFlush( pBs );
WelsUnloadNal( pCtx->pOut );
iNalLen = WelsEncodeNal( &pCtx->pOut->sNalList[iNal], pCtx->pFrameBs + pCtx->iPosBsBuffer, &iNalLen );
pCtx->iPosBsBuffer += iNalLen;
iSize += iNalLen;
return iSize;
}
/*
* post process of dynamic slicing bs writing in case PACKING_ONE_SLICE_PER_LAYER
* include: count bs size of over all the slices in layer,
* return: count number of slices in layer
*/
#if defined(MT_ENABLED) && defined(PACKING_ONE_SLICE_PER_LAYER)
int32_t PostProcDynamicSlicingBsWriting( sWelsEncCtx *pCtx, SLayerBSInfo *pLayerBsInfo, int32_t *pLayerSize, const int32_t kiPartitionCnt )
{
SDqLayer *pCurDq = pCtx->pCurDqLayer;
int32_t iPartitionIdx = 0;
int32_t iCheckingIdx = 0;
int32_t iSwappingIdx = -1;
int32_t iSliceCount = 0;
int32_t iLayerSize = 0;
// count number of slices in layer and layer size
while(iPartitionIdx < kiPartitionCnt)
{
const int32_t coded_slice_cnt = pCurDq->pNumSliceCodedOfPartition[iPartitionIdx];
iLayerSize += pCtx->pSliceThreading->pCountBsSizeInPartition[iPartitionIdx];
iSliceCount += coded_slice_cnt;
++ iPartitionIdx;
}
*pLayerSize = iLayerSize;
// reordering pLayerBs pointers, but do not ensure raster scan order of picture
// just maintain discontinuous items,i.e,
// input:
// partition 1: uiSliceIdx: 0 2 4 6
// partition 2: uiSliceIdx: 1 3 5 7 9 11 13
// output:
// uiSliceIdx: 0 1 2 3 4 5 6 7 8 9 10
iCheckingIdx = 0;
while(true)
{
bool_t bMatchFlag = false;
iPartitionIdx = 0;
while(iPartitionIdx < kiPartitionCnt)
{
const int32_t coded_slice_cnt = pCurDq->pNumSliceCodedOfPartition[iPartitionIdx];
// iCheckingIdx need convert to iIndex of iPartitionIdx based to avoid linear searching
// belong this partition and not exceed the number of slices coded in partition
if ( iPartitionIdx == (iCheckingIdx % kiPartitionCnt)
&& iCheckingIdx / kiPartitionCnt < coded_slice_cnt )
{
if ( iSwappingIdx >= 0 )
{
// memory swapping
memmove(pLayerBsInfo+iSwappingIdx, LayerBsInfo+iCheckingIdx, sizeof(SLayerBSInfo)); // confirmed_safe_unsafe_usage
++ iSwappingIdx; // record iSwappingIdx
}
++ iCheckingIdx;
bMatchFlag = true;
break;
}
++ iPartitionIdx;
}
if ( !bMatchFlag )
{
if ( iSwappingIdx < 0 )
iSwappingIdx = iCheckingIdx;
++ iCheckingIdx;
}
if ( iSwappingIdx >= iSliceCount )
break;
}
return iSliceCount;
}
#endif//MT_ENABLED && PACKING_ONE_SLICE_PER_LAYER
/*
* Force coding IDR as follows
*/
int32_t ForceCodingIDR( sWelsEncCtx *pCtx )
{
if ( NULL == pCtx )
return 1;
pCtx->bEncCurFrmAsIdrFlag = true;
pCtx->iCodingIndex = 0;
return 0;
}
/*!
* \brief core svc encoding process
*
* \pParam pCtx sWelsEncCtx*, encoder context
* \pParam pDst FrameBSInfo*
* \pParam pSrc SSourcePicture* for need_ds = true or SSourcePicture** for need_ds = false
* \pParam iConfiguredLayerNum =1 in case need_ds = true or >1 in case need_ds = false
* \pParam need_ds Indicate whether need down sampling desired
* [NO in picture list case, YES in console aplication based]
* \return EFrameType (WELS_FRAME_TYPE_IDR/WELS_FRAME_TYPE_I/WELS_FRAME_TYPE_P)
*/
int32_t WelsEncoderEncodeExt( sWelsEncCtx *pCtx, void *pDst, const SSourcePicture **ppSrcList, const int32_t iConfiguredLayerNum )
{
SFrameBSInfo *pFbi = (SFrameBSInfo *)pDst;
SLayerBSInfo *pLayerBsInfo = &pFbi->sLayerInfo[0];
SWelsSvcCodingParam *pSvcParam = pCtx->pSvcParam;
SSpatialPicIndex *pSpatialIndexMap= &pCtx->sSpatialIndexMap[0];
#if defined(ENABLE_FRAME_DUMP) || defined(ENABLE_PSNR_CALC)
SPicture *fsnr = NULL;
#endif//ENABLE_FRAME_DUMP || ENABLE_PSNR_CALC
SPicture *pEncPic = NULL; // to be decided later
#if defined(MT_ENABLED) && (defined(DYNAMIC_SLICE_ASSIGN) || defined(MT_DEBUG))
int32_t did_list[MAX_DEPENDENCY_LAYER] = {0};
#endif//MT_ENABLED && DYNAMIC_SLICE_ASSIGN
int32_t iLayerNum = 0;
int32_t iLayerSize = 0;
int32_t iSpatialNum = 0; // available count number of spatial layers due to frame size changed in this given frame
int32_t iSpatialIdx = 0; // iIndex of spatial layers due to frame size changed in this given frame
int32_t iFrameSize = 0;
int32_t iNalLen[128] = {0};
int32_t iNalIdxInLayer = 0;
int32_t iCountNal = 0;
EFrameType eFrameType = WELS_FRAME_TYPE_AUTO;
int32_t iCurWidth = 0;
int32_t iCurHeight = 0;
EWelsNalUnitType eNalType = NAL_UNIT_UNSPEC_0;
EWelsNalRefIdc eNalRefIdc = NRI_PRI_LOWEST;
int8_t iCurDid = 0;
int8_t iCurTid = 0;
bool_t bAvcBased = false;
#if defined(ENABLE_PSNR_CALC)
real32_t snr_y = .0f, snr_u = .0f, snr_v = .0f;
#endif//ENABLE_PSNR_CALC
#if defined(_DEBUG)
int32_t i = 0, j = 0, k = 0;
#endif//_DEBUG
pFbi->iLayerNum = 0; // for initialization
// perform csc/denoise/downsample/padding, generate spatial layers
iSpatialNum = pCtx->pVpp->WelsPreprocessStep1(pCtx, ppSrcList, iConfiguredLayerNum);
if ( iSpatialNum < 1 ) // skip due to temporal layer settings (different frame rate)
{
++ pCtx->iCodingIndex;
return WELS_FRAME_TYPE_SKIP;
}
eFrameType = DecideFrameType( pCtx, iSpatialNum );
if (eFrameType == WELS_FRAME_TYPE_SKIP)
return eFrameType;
InitFrameCoding( pCtx, eFrameType );
iCurTid = GetTemporalLevel( &pSvcParam->sDependencyLayers[pSpatialIndexMap->iDid], pCtx->iCodingIndex, pSvcParam->uiGopSize );
pCtx->uiTemporalId = iCurTid;
pLayerBsInfo->pBsBuf = pCtx->pFrameBs ;
if ( eFrameType == WELS_FRAME_TYPE_IDR )
{
++ pCtx->sPSOVector.uiIdrPicId;
//if ( pSvcParam->bEnableSSEI )
// write parameter sets bitstream here
WelsWriteParameterSets( pCtx, &iNalLen[0], &iCountNal );
pLayerBsInfo->uiPriorityId = 0;
pLayerBsInfo->uiSpatialId = 0;
pLayerBsInfo->uiTemporalId = 0;
pLayerBsInfo->uiQualityId = 0;
pLayerBsInfo->uiLayerType = NON_VIDEO_CODING_LAYER;
pLayerBsInfo->iNalCount = iCountNal;
for (int32_t iNalIndex = 0; iNalIndex < iCountNal; ++ iNalIndex)
{
pLayerBsInfo->iNalLengthInByte[iNalIndex] = iNalLen[iNalIndex];
}
++ pLayerBsInfo;
pLayerBsInfo->pBsBuf = pCtx->pFrameBs + pCtx->iPosBsBuffer;
++ iLayerNum;
}
pCtx->pCurDqLayer = pCtx->ppDqLayerList[pSpatialIndexMap->iDid];
pCtx->pCurDqLayer->pRefLayer = NULL;
while ( iSpatialIdx < iSpatialNum )
{
const int32_t d_idx = (pSpatialIndexMap+iSpatialIdx)->iDid; // get iDid
SDLayerParam *param_d = &pSvcParam->sDependencyLayers[d_idx];
pCtx->uiDependencyId = iCurDid = (int8_t)d_idx;
pCtx->pVpp->WelsPreprocessStep3(pCtx, d_idx);
pCtx->pEncPic = pEncPic = (pSpatialIndexMap+iSpatialIdx)->pSrc;
pCtx->pEncPic->iPictureType = pCtx->eSliceType;
pCtx->pEncPic->iFramePoc = pCtx->iPOC;
iCurWidth = param_d->iFrameWidth;
iCurHeight = param_d->iFrameHeight;
#if defined(MT_ENABLED) && (defined(DYNAMIC_SLICE_ASSIGN) || defined(MT_DEBUG))
did_list[iSpatialIdx] = iCurDid;
#endif//MT_ENABLED && DYNAMIC_SLICE_ASSIGN
// Encoding this picture might mulitiple sQualityStat layers potentially be encoded as followed
switch ( param_d->sMso.uiSliceMode )
{
case SM_FIXEDSLCNUM_SLICE:
{
#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN)
if ( (iCurDid > 0) && (pSvcParam->iMultipleThreadIdc > 1) &&
(pSvcParam->sDependencyLayers[iCurDid].sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE && pSvcParam->iMultipleThreadIdc >= pSvcParam->sDependencyLayers[iCurDid].sMso.sSliceArgument.iSliceNum )
)
AdjustEnhanceLayer( pCtx, iCurDid );
#endif//MT_ENABLED && DYNAMIC_SLICE_ASSIGN
break;
}
case SM_DYN_SLICE:
{
int32_t iPicIPartitionNum = PicPartitionNumDecision( pCtx );
// MT compatibility
pCtx->iActiveThreadsNum = iPicIPartitionNum; // we try to active number of threads, equal to number of picture partitions
WelsInitCurrentDlayerMltslc( pCtx, iPicIPartitionNum );
break;
}
default:
{
break;
}
}
/* coding each spatial layer, only one sQualityStat layer within spatial support */
int32_t iSliceCount = 1;
if ( iLayerNum >= MAX_LAYER_NUM_OF_FRAME ) // check available layer_bs_info writing as follows
{
WelsLog( pCtx, WELS_LOG_ERROR, "WelsEncoderEncodeExt(), iLayerNum(%d) overflow(max:%d)!", iLayerNum, MAX_LAYER_NUM_OF_FRAME);
return -1;
}
iNalIdxInLayer = 0;
bAvcBased = (iCurDid == BASE_DEPENDENCY_ID);
pCtx->bNeedPrefixNalFlag = (bAvcBased &&
(pSvcParam->bPrefixNalAddingCtrl ||
(pSvcParam->iNumDependencyLayer > 1) ));
if ( eFrameType == WELS_FRAME_TYPE_P )
{
eNalType = bAvcBased ? NAL_UNIT_CODED_SLICE : NAL_UNIT_CODED_SLICE_EXT;
}
else if ( eFrameType == WELS_FRAME_TYPE_IDR )
{
eNalType = bAvcBased ? NAL_UNIT_CODED_SLICE_IDR : NAL_UNIT_CODED_SLICE_EXT;
}
if ( iCurTid == 0 || pCtx->eSliceType == I_SLICE )
eNalRefIdc = NRI_PRI_HIGHEST;
else if ( iCurTid == pSvcParam->iDecompStages )
eNalRefIdc = NRI_PRI_LOWEST;
else if ( 1 + iCurTid == pSvcParam->iDecompStages )
eNalRefIdc = NRI_PRI_LOW;
else // more details for other temporal layers?
eNalRefIdc = NRI_PRI_HIGHEST;
pCtx->eNalType = eNalType;
pCtx->eNalPriority = eNalRefIdc;
pCtx->pDecPic = pCtx->ppRefPicListExt[iCurDid]->pNextBuffer;
#if defined(ENABLE_FRAME_DUMP) || defined(ENABLE_PSNR_CALC)
fsnr = pCtx->pDecPic;
#endif//#if defined(ENABLE_FRAME_DUMP) || defined(ENABLE_PSNR_CALC)
pCtx->pDecPic->iPictureType = pCtx->eSliceType;
pCtx->pDecPic->iFramePoc = pCtx->iPOC;
WelsInitCurrentLayer( pCtx, iCurWidth, iCurHeight );
WelsMarkPic(pCtx);
if ( !WelsBuildRefList( pCtx, pCtx->iPOC ) )
{
// Force coding IDR as followed
ForceCodingIDR( pCtx );
WelsLog(pCtx, WELS_LOG_WARNING, "WelsEncoderEncodeExt(), WelsBuildRefList failed for P frames, pCtx->iNumRef0= %d.\n", pCtx->iNumRef0);
return -1;
}
#ifdef LONG_TERM_REF_DUMP
dump_ref(pCtx);
#endif
WelsUpdateRefSyntax(pCtx, pCtx->iPOC, eFrameType); //get reordering syntax used for writing slice header and transmit to encoder.
PrefetchReferencePicture( pCtx, eFrameType ); // update reference picture for current pDq layer
pCtx->pFuncList->pfRc.pfWelsRcPictureInit(pCtx);
PreprocessSliceCoding( pCtx ); // MUST be called after pfWelsRcPictureInit() and WelsInitCurrentLayer()
iLayerSize = 0;
if ( SM_SINGLE_SLICE == param_d->sMso.uiSliceMode ) // only one slice within a sQualityStat layer
{
int32_t iSliceSize = 0;
if ( pCtx->bNeedPrefixNalFlag )
{
iLayerSize += AddPrefixNal( pCtx, pLayerBsInfo, &iNalLen[0], &iNalIdxInLayer, eNalType, eNalRefIdc );
}
WelsLoadNal( pCtx->pOut, eNalType, eNalRefIdc );
WelsCodeOneSlice( pCtx, 0, eNalType );
WelsUnloadNal( pCtx->pOut );
iSliceSize = WelsEncodeNalExt( &pCtx->pOut->sNalList[pCtx->pOut->iNalIndex-1],
&pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt,
pCtx->pFrameBs + pCtx->iPosBsBuffer,
&iNalLen[iNalIdxInLayer] );
iLayerSize += iSliceSize;
pCtx->iPosBsBuffer += iSliceSize;
pLayerBsInfo->uiLayerType = VIDEO_CODING_LAYER;
pLayerBsInfo->uiSpatialId = iCurDid;
pLayerBsInfo->uiTemporalId = iCurTid;
pLayerBsInfo->uiQualityId = 0;
pLayerBsInfo->uiPriorityId = 0;
pLayerBsInfo->iNalLengthInByte[iNalIdxInLayer] = iSliceSize;
pLayerBsInfo->iNalCount = ++ iNalIdxInLayer;
}
// for dynamic slicing single threading..
#ifndef MT_ENABLED
else if ( SM_DYN_SLICE == param_d->sMso.uiSliceMode )
#else // MT_ENABLED
else if ( (SM_DYN_SLICE == param_d->sMso.uiSliceMode) && (pSvcParam->iMultipleThreadIdc <= 1) )
#endif//MT_ENABLED
{
const int32_t kiLastMbInFrame = pCtx->pCurDqLayer->pSliceEncCtx->iMbNumInFrame;
WelsCodeOnePicPartition( pCtx, pLayerBsInfo, &iNalIdxInLayer, &iLayerSize, 0, kiLastMbInFrame, 0 );
}
else
{//other multi-slice uiSliceMode
#if defined(MT_ENABLED)
int err = 0;
// THREAD_FULLY_FIRE_MODE/THREAD_PICK_UP_MODE for any mode of non-SM_DYN_SLICE
if ( (SM_DYN_SLICE != param_d->sMso.uiSliceMode) && (pSvcParam->iMultipleThreadIdc > 1) )
{
iSliceCount = GetCurrentSliceNum( pCtx->pCurDqLayer->pSliceEncCtx );
if ( iLayerNum +
#if defined(PACKING_ONE_SLICE_PER_LAYER)
iSliceCount
#else
1
#endif//PACKING_ONE_SLICE_PER_LAYER
>= MAX_LAYER_NUM_OF_FRAME ) // check available layer_bs_info for further writing as followed
{
WelsLog( pCtx, WELS_LOG_ERROR, "WelsEncoderEncodeExt(), iLayerNum(%d) overflow(max:%d) at iDid= %d uiSliceMode= %d, iSliceCount= %d!",
iLayerNum, MAX_LAYER_NUM_OF_FRAME, iCurDid, param_d->sMso.uiSliceMode, iSliceCount );
return -1;
}
if ( iSliceCount <= 1 )
{
WelsLog( pCtx, WELS_LOG_ERROR, "WelsEncoderEncodeExt(), iSliceCount(%d) from GetCurrentSliceNum() is untrusted due stack/heap crupted!\n", iSliceCount );
return -1;
}
if ( pSvcParam->iCountThreadsNum >= iSliceCount ) //THREAD_FULLY_FIRE_MODE
{
#if defined(PACKING_ONE_SLICE_PER_LAYER)
int32_t iSliceIdx = 1;
int32_t iOrgSlicePos[MAX_SLICES_NUM] = {0};
iOrgSlicePos[0] = pCtx->iPosBsBuffer;
while (uiSliceIdx < iSliceCount)
{
iOrgSlicePos[uiSliceIdx] = pCtx->pSliceBs[uiSliceIdx].uiBsPos;
++ uiSliceIdx;
}
#elif defined(MT_DEBUG)
int64_t t_bs_append = 0;
#endif//PACKING_ONE_SLICE_PER_LAYER
pCtx->iActiveThreadsNum = iSliceCount;
// to fire slice coding threads
err = FiredSliceThreads( &pCtx->pSliceThreading->pThreadPEncCtx[0], &pCtx->pSliceThreading->pReadySliceCodingEvent[0], pLayerBsInfo, iSliceCount, pCtx->pCurDqLayer->pSliceEncCtx, FALSE );
if ( err )
{
WelsLog( pCtx, WELS_LOG_ERROR, "[MT] WelsEncoderEncodeExt(), FiredSliceThreads return(%d) failed and exit encoding frame, iCountThreadsNum= %d, iSliceCount= %d, uiSliceMode= %d, iMultipleThreadIdc= %d!!\n",
err, pSvcParam->iCountThreadsNum, iSliceCount, param_d->sMso.uiSliceMode, pSvcParam->iMultipleThreadIdc );
return -1;
}
WelsMultipleEventsWaitAllBlocking( iSliceCount, &pCtx->pSliceThreading->pSliceCodedEvent[0] );
// all slices are finished coding here
// append exclusive slice 0 bs to pFrameBs
#if defined(PACKING_ONE_SLICE_PER_LAYER)
iLayerSize = pCtx->iPosBsBuffer - iOrgSlicePos[0];
uiSliceIdx = 1;
while (uiSliceIdx < iSliceCount)
{
iLayerSize += pCtx->pSliceBs[uiSliceIdx].uiBsPos - iOrgSlicePos[uiSliceIdx];
++ uiSliceIdx;
}
iLayerNum += iSliceCount; // each slice stickly output as layer info for performance improvement directly
pLayerBsInfo += iSliceCount;
#else
#if defined(MT_DEBUG)
t_bs_append = WelsTime();
#endif//MT_DEBUG
iLayerSize = AppendSliceToFrameBs( pCtx, pLayerBsInfo, iSliceCount );
#if defined(MT_DEBUG)
t_bs_append = WelsTime() - t_bs_append;
if ( pCtx->pSliceThreading->pFSliceDiff )
{
fprintf(pCtx->pSliceThreading->pFSliceDiff,
#if defined(WIN32)
"%6I64d us consumed at AppendSliceToFrameBs() for coding_idx: %d iDid: %d qid: %d\n",
#else
"%6lld us consumed at AppendSliceToFrameBs() for coding_idx: %d iDid: %d qid: %d\n",
#endif//WIN32
t_bs_append, pCtx->iCodingIndex, iCurDid, 0 );
}
#endif//MT_DEBUG
#endif//PACKING_ONE_SLICE_PER_LAYER
}
else //THREAD_PICK_UP_MODE
{
int32_t iNumThreadsRunning = 0;
int32_t iNumThreadsScheduled = 0;
int32_t iIndexOfSliceToBeCoded = 0;
#if defined(PACKING_ONE_SLICE_PER_LAYER)
int32_t iSliceIdx = 1;
int32_t iOrgSlicePos[MAX_SLICES_NUM] = {0};
iOrgSlicePos[0] = pCtx->iPosBsBuffer;
while (uiSliceIdx < iSliceCount)
{
iOrgSlicePos[uiSliceIdx] = pCtx->pSliceBs[uiSliceIdx].uiBsPos;
++ uiSliceIdx;
}
#endif//PACKING_ONE_SLICE_PER_LAYER
pCtx->iActiveThreadsNum = pSvcParam->iCountThreadsNum;
iNumThreadsScheduled = pCtx->iActiveThreadsNum;
iNumThreadsRunning = iNumThreadsScheduled;
// to fire slice coding threads
err = FiredSliceThreads( &pCtx->pSliceThreading->pThreadPEncCtx[0], &pCtx->pSliceThreading->pReadySliceCodingEvent[0], pLayerBsInfo, iNumThreadsRunning, pCtx->pCurDqLayer->pSliceEncCtx, FALSE );
if ( err )
{
WelsLog( pCtx, WELS_LOG_ERROR, "[MT] WelsEncoderEncodeExt(), FiredSliceThreads return(%d) failed and exit encoding frame, iCountThreadsNum= %d, iSliceCount= %d, uiSliceMode= %d, iMultipleThreadIdc= %d!!\n",
err, pSvcParam->iCountThreadsNum, iSliceCount, param_d->sMso.uiSliceMode, pSvcParam->iMultipleThreadIdc );
return -1;
}
iIndexOfSliceToBeCoded = iNumThreadsRunning;
while (1)
{
if ( iIndexOfSliceToBeCoded >= iSliceCount && iNumThreadsRunning <= 0 )
break;
#ifdef WIN32
WELS_THREAD_ERROR_CODE lwait = 0;
int32_t iEventId = -1;
lwait = WelsMultipleEventsWaitSingleBlocking( iNumThreadsScheduled,
&pCtx->pSliceThreading->pSliceCodedEvent[0],
2 ); // 2 ms for one tick
iEventId = (int32_t)(lwait - WELS_THREAD_ERROR_WAIT_OBJECT_0);
if ( iEventId >= 0 && iEventId < iNumThreadsScheduled )
{
if ( iIndexOfSliceToBeCoded < iSliceCount )
{
// pick up succeeding slice for threading
// thread_id equal to iEventId per implementation here
pCtx->pSliceThreading->pThreadPEncCtx[iEventId].iSliceIndex = iIndexOfSliceToBeCoded;
#ifdef PACKING_ONE_SLICE_PER_LAYER
pCtx->pSliceThreading->pThreadPEncCtx[iEventId].pLayerBs = pLayerBsInfo+iIndexOfSliceToBeCoded;
#endif//PACKING_ONE_SLICE_PER_LAYER
WelsEventSignal( &pCtx->pSliceThreading->pReadySliceCodingEvent[iEventId] );
++ iIndexOfSliceToBeCoded;
}
else // no other slices left for coding
{
-- iNumThreadsRunning;
}
}
else
{
WelsSleep(1);
}
#else//__GNUC__
// TODO for pthread platforms
// alternate implementation using blocking due non-blocking with timeout mode not support at wels thread lib, tune back if available
WelsMultipleEventsWaitAllBlocking( iNumThreadsRunning, &pCtx->pSliceThreading->pSliceCodedEvent[0] );
if ( iIndexOfSliceToBeCoded < iSliceCount )
{
int32_t iThreadIdx = 0;
// pick up succeeding slices for threading if left
while ( iThreadIdx < iNumThreadsScheduled )
{
if ( iIndexOfSliceToBeCoded >= iSliceCount )
break;
pCtx->pSliceThreading->pThreadPEncCtx[iThreadIdx].iSliceIndex = iIndexOfSliceToBeCoded;
#ifdef PACKING_ONE_SLICE_PER_LAYER
pCtx->pSliceThreading->pThreadPEncCtx[iThreadIdx].pLayerBs = pLayerBsInfo+iIndexOfSliceToBeCoded;
#endif//PACKING_ONE_SLICE_PER_LAYER
WelsEventSignal( pCtx->pSliceThreading->pReadySliceCodingEvent[iThreadIdx] );
++ iIndexOfSliceToBeCoded;
++ iThreadIdx;
}
// update iNumThreadsRunning
iNumThreadsRunning = iThreadIdx;
}
else
{
iNumThreadsRunning = 0;
}
#endif//WIN32
}//while(1)
// all slices are finished coding here
// append exclusive slice 0 bs to pFrameBs
#if defined(PACKING_ONE_SLICE_PER_LAYER)
iLayerSize = pCtx->iPosBsBuffer - iOrgSlicePos[0];
uiSliceIdx = 1;
while (uiSliceIdx < iSliceCount)
{
iLayerSize += pCtx->pSliceBs[uiSliceIdx].uiBsPos - iOrgSlicePos[uiSliceIdx];
++ uiSliceIdx;
}
iLayerNum += iSliceCount; // each slice stickly output as layer info for performance improvement directly
pLayerBsInfo += iSliceCount;
#else
iLayerSize = AppendSliceToFrameBs( pCtx, pLayerBsInfo, iSliceCount );
#endif//PACKING_ONE_SLICE_PER_LAYER
}
}
// THREAD_FULLY_FIRE_MODE && SM_DYN_SLICE
else if ( (SM_DYN_SLICE == param_d->sMso.uiSliceMode) && (pSvcParam->iMultipleThreadIdc > 1) )
{
const int32_t kiPartitionCnt = pCtx->iActiveThreadsNum; //pSvcParam->iCountThreadsNum;
#if defined(PACKING_ONE_SLICE_PER_LAYER)
ResetCountBsSizeInPartitions( pCtx->pSliceThreading->pCountBsSizeInPartition, kiPartitionCnt );
pCtx->pCurDqLayer->pSliceEncCtx->iMaxSliceNumConstraint = WELS_MIN ( MAX_SLICES_NUM, DynamicMaxSliceNumConstraint( MAX_LAYER_NUM_OF_FRAME, iLayerNum, 1 + /*( num_qlayer - 1) +*/ ( ( (iCurDid==0) && ( pSvcParam->uiGopSize>1 ) ) ? 1: 0 ) ) );
#endif//PACKING_ONE_SLICE_PER_LAYER
// to fire slice coding threads
err = FiredSliceThreads( &pCtx->pSliceThreading->pThreadPEncCtx[0], &pCtx->pSliceThreading->pReadySliceCodingEvent[0], pLayerBsInfo, kiPartitionCnt, pCtx->pCurDqLayer->pSliceEncCtx, TRUE );
if ( err )
{
WelsLog( pCtx, WELS_LOG_ERROR, "[MT] WelsEncoderEncodeExt(), FiredSliceThreads return(%d) failed and exit encoding frame, iCountThreadsNum= %d, iSliceCount= %d, uiSliceMode= %d, iMultipleThreadIdc= %d!!\n",
err, pSvcParam->iCountThreadsNum, iSliceCount, param_d->sMso.uiSliceMode, pSvcParam->iMultipleThreadIdc );
return -1;
}
WelsMultipleEventsWaitAllBlocking( kiPartitionCnt, &pCtx->pSliceThreading->pSliceCodedEvent[0] );
#if defined(PACKING_ONE_SLICE_PER_LAYER)
iSliceCount = PostProcDynamicSlicingBsWriting( pCtx, pLayerBsInfo, &iLayerSize, kiPartitionCnt );
assert(iLayerNum + iSliceCount < MAX_LAYER_NUM_OF_FRAME);
pLayerBsInfo += iSliceCount;
iLayerNum += iSliceCount;
#else
iLayerSize = AppendSliceToFrameBs( pCtx, pLayerBsInfo, kiPartitionCnt );
#endif//PACKING_ONE_SLICE_PER_LAYER
}
else // for non-dynamic-slicing mode single threading branch..
#endif//MT_ENABLED
{
const bool_t bNeedPrefix = pCtx->bNeedPrefixNalFlag;
int32_t iSliceIdx = 0;
iSliceCount = GetCurrentSliceNum( pCtx->pCurDqLayer->pSliceEncCtx );
while (iSliceIdx < iSliceCount)
{
int32_t iSliceSize = 0;
if ( bNeedPrefix )
{
iLayerSize += AddPrefixNal( pCtx, pLayerBsInfo, &iNalLen[0], &iNalIdxInLayer, eNalType, eNalRefIdc );
}
WelsLoadNal( pCtx->pOut, eNalType, eNalRefIdc );
WelsCodeOneSlice( pCtx, iSliceIdx, eNalType );
WelsUnloadNal( pCtx->pOut );
iSliceSize = WelsEncodeNalExt( &pCtx->pOut->sNalList[pCtx->pOut->iNalIndex-1],
&pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt,
pCtx->pFrameBs + pCtx->iPosBsBuffer,
&iNalLen[iNalIdxInLayer] );
pCtx->iPosBsBuffer += iSliceSize;
iLayerSize += iSliceSize;
pLayerBsInfo->iNalLengthInByte[iNalIdxInLayer] = iSliceSize;
#if defined(SLICE_INFO_OUTPUT)
fprintf( stderr,
"@slice=%-6d sliceType:%c idc:%d size:%-6d\n",
iSliceIdx,
(pCtx->eSliceType == P_SLICE ? 'P' : 'I'),
eNalRefIdc,
iSliceSize );
#endif//SLICE_INFO_OUTPUT
++ iNalIdxInLayer;
++ iSliceIdx;
}
pLayerBsInfo->uiLayerType = VIDEO_CODING_LAYER;
pLayerBsInfo->uiSpatialId = iCurDid;
pLayerBsInfo->uiTemporalId = iCurTid;
pLayerBsInfo->uiQualityId = 0;
pLayerBsInfo->uiPriorityId = 0;
pLayerBsInfo->iNalCount = iNalIdxInLayer;
}
}
// deblocking filter
if (
#if defined(MT_ENABLED)
(!pCtx->pCurDqLayer->bDeblockingParallelFlag) &&
#endif//MT_ENABLED
#if !defined(ENABLE_FRAME_DUMP)
( (eNalRefIdc != NRI_PRI_LOWEST) && (param_d->iHighestTemporalId == 0 || iCurTid < param_d->iHighestTemporalId) ) &&
#endif//!ENABLE_FRAME_DUMP
true
)
{
PerformDeblockingFilter( pCtx );
}
// reference picture list update
if ( eNalRefIdc != NRI_PRI_LOWEST )
{
if ( !WelsUpdateRefList( pCtx ) )
{
// Force coding IDR as followed
ForceCodingIDR( pCtx );
WelsLog(pCtx, WELS_LOG_WARNING, "WelsEncoderEncodeExt(), WelsUpdateRefList failed.\n");
return -1;
}
}
iFrameSize += iLayerSize;
pCtx->pFuncList->pfRc.pfWelsRcPictureInfoUpdate(pCtx, iLayerSize);
#ifdef ENABLE_FRAME_DUMP
// Dump reconstruction picture for each sQualityStat layer
if ( iCurDid+1 < pSvcParam->iNumDependencyLayer )
DumpDependencyRec( fsnr, ¶m_d->sRecFileName[0], iCurDid );
#endif//ENABLE_FRAME_DUMP
#if defined(ENABLE_PSNR_CALC)
snr_y = WelsCalcPsnr( fsnr->pData[0],
fsnr->iLineSize[0],
pEncPic->pData[0],
pEncPic->iLineSize[0],
iCurWidth,
iCurHeight );
snr_u = WelsCalcPsnr( fsnr->pData[1],
fsnr->iLineSize[1],
pEncPic->pData[1],
pEncPic->iLineSize[1],
(iCurWidth>>1),
(iCurHeight>>1) );
snr_v = WelsCalcPsnr( fsnr->pData[2],
fsnr->iLineSize[2],
pEncPic->pData[2],
pEncPic->iLineSize[2],
(iCurWidth>>1),
(iCurHeight>>1) );
#endif//ENABLE_PSNR_CALC
#if defined(LAYER_INFO_OUTPUT)
fprintf( stderr, "%2s %5d: %-5d %2s T%1d D%1d Q%-2d QP%3d Y%2.2f U%2.2f V%2.2f %8d bits\n",
(iSpatialIdx == 0) ? "#AU" : " ",
pCtx->iPOC,
pCtx->iFrameNum,
(uiFrameType == WELS_FRAME_TYPE_I || uiFrameType == WELS_FRAME_TYPE_IDR) ? "I": "P",
iCurTid,
iCurDid,
0,
pCtx->pWelsSvcRc[pCtx->uiDependencyId].iAverageFrameQp,
snr_y,
snr_u,
snr_v,
(iLayerSize<<3) );
#endif//LAYER_INFO_OUTPUT
#if defined(STAT_OUTPUT)
#if defined(ENABLE_PSNR_CALC)
{
pCtx->sStatData[iCurDid][0].sQualityStat.rYPsnr[pCtx->eSliceType] += snr_y;
pCtx->sStatData[iCurDid][0].sQualityStat.rUPsnr[pCtx->eSliceType] += snr_u;
pCtx->sStatData[iCurDid][0].sQualityStat.rVPsnr[pCtx->eSliceType] += snr_v;
}
#endif//ENABLE_PSNR_CALC
#if defined(MB_TYPES_CHECK) //091025, frame output
if (pCtx->eSliceType == P_SLICE)
{
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][Intra4x4] += pCtx->sPerInfo.iMbCount[P_SLICE][Intra4x4];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][Intra16x16] += pCtx->sPerInfo.iMbCount[P_SLICE][Intra16x16];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][Inter16x16] += pCtx->sPerInfo.iMbCount[P_SLICE][Inter16x16];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][Inter16x8] += pCtx->sPerInfo.iMbCount[P_SLICE][Inter16x8];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][Inter8x16] += pCtx->sPerInfo.iMbCount[P_SLICE][Inter8x16];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][Inter8x8] += pCtx->sPerInfo.iMbCount[P_SLICE][Inter8x8];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][PSkip] += pCtx->sPerInfo.iMbCount[P_SLICE][PSkip];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][8] += pCtx->sPerInfo.iMbCount[P_SLICE][8];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][9] += pCtx->sPerInfo.iMbCount[P_SLICE][9];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][10] += pCtx->sPerInfo.iMbCount[P_SLICE][10];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[P_SLICE][11] += pCtx->sPerInfo.iMbCount[P_SLICE][11];
}
else if (pCtx->eSliceType == I_SLICE)
{
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[I_SLICE][Intra4x4] += pCtx->sPerInfo.iMbCount[I_SLICE][Intra4x4];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[I_SLICE][Intra16x16] += pCtx->sPerInfo.iMbCount[I_SLICE][Intra16x16];
pCtx->sStatData[iCurDid][0].sSliceData.iMbCount[I_SLICE][7] += pCtx->sPerInfo.iMbCount[I_SLICE][7];
}
memset(pCtx->sPerInfo.iMbCount[P_SLICE], 0, 18*sizeof( int32_t ));
memset(pCtx->sPerInfo.iMbCount[I_SLICE], 0, 18*sizeof( int32_t ));
#endif//MB_TYPES_CHECK
{
//no pCtx->pSvcParam->bMgsT0OnlyStrategy
++ pCtx->sStatData[iCurDid][0].sSliceData.iSliceCount[pCtx->eSliceType]; // for multiple slices coding
pCtx->sStatData[iCurDid][0].sSliceData.iSliceSize[pCtx->eSliceType] += (iLayerSize<<3); // bits
}
#endif//STAT_OUTPUT
#if defined(MT_ENABLED) && defined(PACKING_ONE_SLICE_PER_LAYER)
if ( pSvcParam->iMultipleThreadIdc <= 1 || SM_SINGLE_SLICE == param_d->sMso.uiSliceMode ) // sigle thread actually used
#else
if ( 1 )
#endif//MT_ENABLED && PACKING_ONE_SLICE_PER_LAYER
{
++ iLayerNum;
++ pLayerBsInfo;
}
pLayerBsInfo->pBsBuf = pCtx->pFrameBs + pCtx->iPosBsBuffer;
if( pSvcParam->iPaddingFlag && pCtx->pWelsSvcRc[pCtx->uiDependencyId].iPaddingSize > 0 )
{
const int32_t kiPaddingNalSize = WritePadding(pCtx, pCtx->pWelsSvcRc[pCtx->uiDependencyId].iPaddingSize);
#if GOM_TRACE_FLAG
WelsLog( pCtx, WELS_LOG_INFO,"[RC] encoding_qp%d Padding: %d\n",pCtx->uiDependencyId, pCtx->pWelsSvcRc[pCtx->uiDependencyId].iPaddingSize);
#endif
if ( kiPaddingNalSize <= 0 )
return -1;
pCtx->pWelsSvcRc[pCtx->uiDependencyId].iPaddingBitrateStat += pCtx->pWelsSvcRc[pCtx->uiDependencyId].iPaddingSize;
pCtx->pWelsSvcRc[pCtx->uiDependencyId].iPaddingSize=0;
pLayerBsInfo->uiPriorityId = 0;
pLayerBsInfo->uiSpatialId = 0;
pLayerBsInfo->uiTemporalId = 0;
pLayerBsInfo->uiQualityId = 0;
pLayerBsInfo->uiLayerType = NON_VIDEO_CODING_LAYER;
pLayerBsInfo->iNalCount = 1;
pLayerBsInfo->iNalLengthInByte[0] = kiPaddingNalSize;
++ pLayerBsInfo;
pLayerBsInfo->pBsBuf = pCtx->pFrameBs + pCtx->iPosBsBuffer;
++ iLayerNum;
}
#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
if ( param_d->sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE && pSvcParam->iMultipleThreadIdc > 1 &&
pSvcParam->iMultipleThreadIdc >= param_d->sMso.sSliceArgument.iSliceNum )
{
CalcSliceComplexRatio( pCtx->pSliceThreading->pSliceComplexRatio[iCurDid], pCtx->pCurDqLayer->pSliceEncCtx, pCtx->pSliceThreading->pSliceConsumeTime[iCurDid] );
#if defined(MT_DEBUG)
TrackSliceComplexities( pCtx, iCurDid );
#endif//#if defined(MT_DEBUG)
}
#endif//MT_ENABLED && DYNAMIC_SLICE_ASSIGN && TRY_SLICING_BALANCE
++ iSpatialIdx;
if ( iCurDid+1 < pSvcParam->iNumDependencyLayer )
{
WelsSwapDqLayers( pCtx );
}
if ( pSvcParam->bEnableLongTermReference && (pCtx->pLtr[pCtx->uiDependencyId].bLTRMarkingFlag && (pCtx->pLtr[pCtx->uiDependencyId].iLTRMarkMode == LTR_DELAY_MARK)))
{
pCtx->bLongTermRefFlag[d_idx][0] = true;
}
if ( iCurTid < pCtx->uiSpatialLayersInTemporal[d_idx] - 1 || pSvcParam->iDecompStages == 0 )
{
if ( (iCurTid >= MAX_TEMPORAL_LEVEL)||(pCtx->uiSpatialLayersInTemporal[d_idx]-1>= MAX_TEMPORAL_LEVEL))
{
ForceCodingIDR( pCtx ); // some logic error
return -1;
}
if ( pSvcParam->bEnableLongTermReference && pCtx->bLongTermRefFlag[d_idx][iCurTid] )
{
SPicture *tmp = pCtx->pSpatialPic[d_idx][pCtx->uiSpatialLayersInTemporal[d_idx]+pCtx->pVaa->uiMarkLongTermPicIdx];
pCtx->pSpatialPic[d_idx][pCtx->uiSpatialLayersInTemporal[d_idx]+pCtx->pVaa->uiMarkLongTermPicIdx] = pCtx->pSpatialPic[d_idx][iCurTid];
pCtx->pSpatialPic[d_idx][iCurTid] = pCtx->pSpatialPic[d_idx][pCtx->uiSpatialLayersInTemporal[d_idx]-1];
pCtx->pSpatialPic[d_idx][pCtx->uiSpatialLayersInTemporal[d_idx]-1] = tmp;
pCtx->bLongTermRefFlag[d_idx][iCurTid] = false;
}
else
{
WelsExchangeSpatialPictures( &pCtx->pSpatialPic[d_idx][pCtx->uiSpatialLayersInTemporal[d_idx]-1], &pCtx->pSpatialPic[d_idx][iCurTid] );
}
}
if ( pSvcParam->bEnableLongTermReference && ((pCtx->pLtr[pCtx->uiDependencyId].bLTRMarkingFlag && (pCtx->pLtr[pCtx->uiDependencyId].iLTRMarkMode == LTR_DIRECT_MARK)) || eFrameType == WELS_FRAME_TYPE_IDR))
{
pCtx->bLongTermRefFlag[d_idx][iCurTid] = true;
}
}
#if defined(MT_ENABLED) && defined(MT_DEBUG)
TrackSliceConsumeTime( pCtx, did_list, iSpatialNum );
#endif//MT_ENABLED && MT_DEBUG
#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN)
if ( pSvcParam->iMultipleThreadIdc > 1 && did_list[0] == BASE_DEPENDENCY_ID
&& pSvcParam->sDependencyLayers[0].sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE && pSvcParam->iMultipleThreadIdc >= pSvcParam->sDependencyLayers[0].sMso.sSliceArgument.iSliceNum
&& pSvcParam->sDependencyLayers[did_list[iSpatialNum-1]].sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE && pSvcParam->iMultipleThreadIdc >= pSvcParam->sDependencyLayers[did_list[iSpatialNum-1]].sMso.sSliceArgument.iSliceNum )
{
AdjustBaseLayer( pCtx );
}
#endif//DYNAMIC_SLICE_ASSIGN
#ifdef ENABLE_FRAME_DUMP
DumpRecFrame( fsnr, &pSvcParam->sDependencyLayers[pSvcParam->iNumDependencyLayer-1].sRecFileName[0] ); // pDecPic: final reconstruction output
#endif//ENABLE_FRAME_DUMP
++ pCtx->iCodingIndex;
pCtx->eLastNalPriority = eNalRefIdc;
pFbi->iLayerNum = iLayerNum;
#if defined(X86_ASM)
WelsEmms();
#endif //X86_ASM
return eFrameType;
}
/*!
* \brief Wels SVC encoder parameters adjustment
* SVC adjustment results in new requirement in memory blocks adjustment
*/
int32_t WelsEncoderParamAdjust( sWelsEncCtx **ppCtx, SWelsSvcCodingParam *pNewParam )
{
SWelsSvcCodingParam *pOldParam = NULL;
int32_t iReturn = 0;
int8_t iIndexD= 0;
bool_t bNeedReset = false;
if ( NULL == ppCtx || NULL == *ppCtx || NULL == pNewParam ) return 1;
/* Check validation in new parameters */
iReturn = ParamValidationExt( pNewParam );
if ( iReturn != 0 ) return iReturn;
pOldParam = (*ppCtx)->pSvcParam;
/* Decide whether need reset for IDR frame based on adjusting prarameters changed */
/* Temporal levels, spatial settings and/ or quality settings changed need update parameter sets related. */
bNeedReset = (pOldParam == NULL ) ||
(pOldParam->iNumTemporalLayer != pNewParam->iNumTemporalLayer) ||
(pOldParam->uiGopSize != pNewParam->uiGopSize) ||
(pOldParam->iNumDependencyLayer != pNewParam->iNumDependencyLayer) ||
(pOldParam->iDecompStages != pNewParam->iDecompStages) ||
(pOldParam->iActualPicWidth != pNewParam->iActualPicWidth || pOldParam->iActualPicHeight != pNewParam->iActualPicHeight) ||
(pOldParam->SUsedPicRect.iWidth != pNewParam->SUsedPicRect.iWidth || pOldParam->SUsedPicRect.iHeight != pNewParam->SUsedPicRect.iHeight) ||
(pOldParam->bEnableLongTermReference != pNewParam->bEnableLongTermReference);
if ( !bNeedReset ){ // Check its picture resolutions/quality settings respectively in each dependency layer
iIndexD = 0;
assert( pOldParam->iNumDependencyLayer == pNewParam->iNumDependencyLayer );
do
{
const SDLayerParam *kpOldDlp = &pOldParam->sDependencyLayers[iIndexD];
const SDLayerParam *kpNewDlp = &pNewParam->sDependencyLayers[iIndexD];
float fT1 = .0f;
float fT2 = .0f;
// check frame size settings
if ( kpOldDlp->iFrameWidth != kpNewDlp->iFrameWidth ||
kpOldDlp->iFrameHeight != kpNewDlp->iFrameHeight ||
kpOldDlp->iActualWidth != kpNewDlp->iActualWidth ||
kpOldDlp->iActualHeight != kpNewDlp->iActualHeight ){
bNeedReset = true;
break;
}
if ( kpOldDlp->sMso.uiSliceMode != kpNewDlp->sMso.uiSliceMode ||
kpOldDlp->sMso.sSliceArgument.iSliceNum != kpNewDlp->sMso.sSliceArgument.iSliceNum )
{
bNeedReset = true;
break;
}
// check frame rate
// we can not check whether corresponding fFrameRate is equal or not,
// only need to check d_max/d_min and max_fr/d_max whether it is equal or not
if ( kpNewDlp->fInputFrameRate > EPSN && kpOldDlp->fInputFrameRate > EPSN )
fT1 = kpNewDlp->fOutputFrameRate/kpNewDlp->fInputFrameRate - kpOldDlp->fOutputFrameRate/kpOldDlp->fInputFrameRate;
if ( kpNewDlp->fOutputFrameRate > EPSN && kpOldDlp->fOutputFrameRate > EPSN )
fT2 = pNewParam->fMaxFrameRate/kpNewDlp->fOutputFrameRate - pOldParam->fMaxFrameRate/kpOldDlp->fOutputFrameRate;
if ( fT1 > EPSN || fT1 < -EPSN || fT2 > EPSN || fT2 < -EPSN )
{
bNeedReset = true;
break;
}
if ( kpOldDlp->iHighestTemporalId != kpNewDlp->iHighestTemporalId )
{
bNeedReset = true;
break;
}
++ iIndexD;
} while (iIndexD < pOldParam->iNumDependencyLayer);
}
if ( bNeedReset ){
SParaSetOffsetVariable sTmpPsoVariable[PARA_SET_TYPE];
uint16_t uiTmpIdrPicId;//this is for LTR!
memcpy( sTmpPsoVariable, (*ppCtx)->sPSOVector.sParaSetOffsetVariable, (PARA_SET_TYPE)*sizeof(SParaSetOffsetVariable) );// confirmed_safe_unsafe_usage
uiTmpIdrPicId = (*ppCtx)->sPSOVector.uiIdrPicId;
WelsUninitEncoderExt( ppCtx );
/* Update new parameters */
if ( WelsInitEncoderExt( ppCtx, pNewParam ) )
return 1;
// reset the scaled spatial picture size
(*ppCtx)->pVpp->WelsPreprocessReset(*ppCtx);
//if WelsInitEncoderExt succeed
//for FLEXIBLE_PARASET_ID
memcpy( (*ppCtx)->sPSOVector.sParaSetOffsetVariable, sTmpPsoVariable, (PARA_SET_TYPE)*sizeof(SParaSetOffsetVariable) );// confirmed_safe_unsafe_usage
(*ppCtx)->sPSOVector.uiIdrPicId = uiTmpIdrPicId;
}
else{
/* maybe adjustment introduced in bitrate or little settings adjustment and so on.. */
pNewParam->iNumRefFrame = WELS_CLIP3(pNewParam->iNumRefFrame, MIN_REF_PIC_COUNT, MAX_REFERENCE_PICTURE_COUNT_NUM);
pNewParam->iLoopFilterDisableIdc = WELS_CLIP3(pNewParam->iLoopFilterDisableIdc, 0, 6);
pNewParam->iLoopFilterAlphaC0Offset = WELS_CLIP3(pNewParam->iLoopFilterAlphaC0Offset, -6, 6);
pNewParam->iLoopFilterBetaOffset = WELS_CLIP3(pNewParam->iLoopFilterBetaOffset, -6, 6);
pNewParam->iInterLayerLoopFilterDisableIdc = WELS_CLIP3(pNewParam->iInterLayerLoopFilterDisableIdc, 0, 6);
pNewParam->iInterLayerLoopFilterAlphaC0Offset = WELS_CLIP3(pNewParam->iInterLayerLoopFilterAlphaC0Offset, -6, 6);
pNewParam->iInterLayerLoopFilterBetaOffset = WELS_CLIP3(pNewParam->iInterLayerLoopFilterBetaOffset, -6, 6);
pNewParam->fMaxFrameRate = WELS_CLIP3(pNewParam->fMaxFrameRate, MIN_FRAME_RATE, MAX_FRAME_RATE);
// we can not use direct struct based memcpy due some fields need keep unchanged as before
pOldParam->fMaxFrameRate = pNewParam->fMaxFrameRate; // maximal frame rate [Hz / fps]
pOldParam->iInputCsp = pNewParam->iInputCsp; // color space of input sequence
pOldParam->uiIntraPeriod = pNewParam->uiIntraPeriod; // intra period (multiple of GOP size as desired)
pOldParam->bEnableSpsPpsIdAddition = pNewParam->bEnableSpsPpsIdAddition;
pOldParam->bPrefixNalAddingCtrl = pNewParam->bPrefixNalAddingCtrl;
pOldParam->iNumRefFrame = pNewParam->iNumRefFrame; // number of reference frame used
/* denoise control */
pOldParam->bEnableDenoise = pNewParam->bEnableDenoise;
/* background detection control */
pOldParam->bEnableBackgroundDetection = pNewParam->bEnableBackgroundDetection;
/* adaptive quantization control */
pOldParam->bEnableAdaptiveQuant = pNewParam->bEnableAdaptiveQuant;
/* int32_t term reference control */
pOldParam->bEnableLongTermReference = pNewParam->bEnableLongTermReference;
pOldParam->uiLtrMarkPeriod = pNewParam->uiLtrMarkPeriod;
// keep below values unchanged as before
pOldParam->bEnableSSEI = pNewParam->bEnableSSEI;
pOldParam->bEnableFrameCroppingFlag = pNewParam->bEnableFrameCroppingFlag; // enable frame cropping flag
/* Motion search */
/* Deblocking loop filter */
pOldParam->iLoopFilterDisableIdc = pNewParam->iLoopFilterDisableIdc; // 0: on, 1: off, 2: on except for slice boundaries
pOldParam->iLoopFilterAlphaC0Offset = pNewParam->iLoopFilterAlphaC0Offset;// AlphaOffset: valid range [-6, 6], default 0
pOldParam->iLoopFilterBetaOffset = pNewParam->iLoopFilterBetaOffset; // BetaOffset: valid range [-6, 6], default 0
pOldParam->iInterLayerLoopFilterDisableIdc = pNewParam->iInterLayerLoopFilterDisableIdc; // Employed based upon inter-layer, same comment as above
pOldParam->iInterLayerLoopFilterAlphaC0Offset = pNewParam->iInterLayerLoopFilterAlphaC0Offset; // InterLayerLoopFilterAlphaC0Offset
pOldParam->iInterLayerLoopFilterBetaOffset = pNewParam->iInterLayerLoopFilterBetaOffset; // InterLayerLoopFilterBetaOffset
/* Rate Control */
pOldParam->bEnableRc = pNewParam->bEnableRc;
pOldParam->iRCMode = pNewParam->iRCMode;
pOldParam->iTargetBitrate = pNewParam->iTargetBitrate; // overall target bitrate introduced in RC module
pOldParam->iPaddingFlag = pNewParam->iPaddingFlag;
/* Layer definition */
pOldParam->bPrefixNalAddingCtrl = pNewParam->bPrefixNalAddingCtrl;
// d
iIndexD = 0;
do
{
SDLayerParam *pOldDlp = &pOldParam->sDependencyLayers[iIndexD];
SDLayerParam *pNewDlp = &pNewParam->sDependencyLayers[iIndexD];
pOldDlp->fInputFrameRate = pNewDlp->fInputFrameRate; // input frame rate
pOldDlp->fOutputFrameRate = pNewDlp->fOutputFrameRate; // output frame rate
pOldDlp->iSpatialBitrate = pNewDlp->iSpatialBitrate;
pOldDlp->uiProfileIdc = pNewDlp->uiProfileIdc; // value of profile IDC (0 for auto-detection)
/* Derived variants below */
pOldDlp->iTemporalResolution = pNewDlp->iTemporalResolution;
pOldDlp->iDecompositionStages = pNewDlp->iDecompositionStages;
memcpy(pOldDlp->uiCodingIdx2TemporalId, pNewDlp->uiCodingIdx2TemporalId, sizeof(pOldDlp->uiCodingIdx2TemporalId)); // confirmed_safe_unsafe_usage
++ iIndexD;
} while (iIndexD < pOldParam->iNumDependencyLayer);
}
/* Any else initialization/reset for rate control here? */
return 0;
}
int32_t WelsCodeOnePicPartition( sWelsEncCtx *pCtx,
SLayerBSInfo *pLayerBsInfo,
int32_t *pNalIdxInLayer,
int32_t* pLayerSize,
int32_t iFirstMbInPartition, // first mb inclusive in partition
int32_t iEndMbInPartition, // end mb exclusive in partition
int32_t iStartSliceIdx
)
{
SDqLayer * pCurLayer = pCtx->pCurDqLayer;
SSliceCtx * pSliceCtx = pCurLayer->pSliceEncCtx;
int32_t iNalLen[MAX_NAL_UNITS_IN_LAYER] = {0};
int32_t iNalIdxInLayer = *pNalIdxInLayer;
int32_t iSliceIdx = iStartSliceIdx;
const int32_t kiSliceStep = pCtx->iActiveThreadsNum;
const int32_t kiPartitionId = iStartSliceIdx % kiSliceStep;
int32_t iPartitionBsSize = 0;
int32_t iAnyMbLeftInPartition= iEndMbInPartition - iFirstMbInPartition;
const EWelsNalUnitType keNalType = pCtx->eNalType;
const EWelsNalRefIdc keNalRefIdc = pCtx->eNalPriority;
const bool_t kbNeedPrefix = pCtx->bNeedPrefixNalFlag;
//init
{
pSliceCtx->pFirstMbInSlice[iSliceIdx] = iFirstMbInPartition;
pCurLayer->pNumSliceCodedOfPartition[kiPartitionId] = 1; // one slice per partition intialized, dynamic slicing inside
pCurLayer->pLastMbIdxOfPartition[kiPartitionId] = iEndMbInPartition-1;
}
pCurLayer->pLastCodedMbIdxOfPartition[kiPartitionId] = 0;
while ( iAnyMbLeftInPartition > 0 )
{
int32_t iSliceSize = 0;
if ( iSliceIdx >= pSliceCtx->iMaxSliceNumConstraint ) // insufficient memory in pSliceInLayer[]
{
// TODO: need exception handler for not large enough of MAX_SLICES_NUM related memory usage
// No idea about its solution due MAX_SLICES_NUM is fixed lenght in relevent pData structure
return 1;
}
if ( kbNeedPrefix )
{
iPartitionBsSize += AddPrefixNal( pCtx, pLayerBsInfo, &iNalLen[0], &iNalIdxInLayer, keNalType, keNalRefIdc );
}
WelsLoadNal( pCtx->pOut, keNalType, keNalRefIdc );
WelsCodeOneSlice( pCtx, iSliceIdx, keNalType );
WelsUnloadNal( pCtx->pOut );
iSliceSize = WelsEncodeNalExt( &pCtx->pOut->sNalList[pCtx->pOut->iNalIndex-1],
&pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt,
pCtx->pFrameBs + pCtx->iPosBsBuffer,
&iNalLen[iNalIdxInLayer] );
pCtx->iPosBsBuffer += iSliceSize;
iPartitionBsSize += iSliceSize;
pLayerBsInfo->iNalLengthInByte[iNalIdxInLayer] = iSliceSize;
#if defined(SLICE_INFO_OUTPUT)
fprintf( stderr,
"@slice=%-6d sliceType:%c idc:%d size:%-6d\n",
iSliceIdx,
(pCtx->eSliceType == P_SLICE ? 'P' : 'I'),
eNalRefIdc,
iSliceSize );
#endif//SLICE_INFO_OUTPUT
++ iNalIdxInLayer;
iSliceIdx += kiSliceStep; //if uiSliceIdx is not continuous
iAnyMbLeftInPartition = iEndMbInPartition - (1 + pCurLayer->pLastCodedMbIdxOfPartition[kiPartitionId]);
}
*pLayerSize = iPartitionBsSize;
*pNalIdxInLayer = iNalIdxInLayer;
// slice based packing???
pLayerBsInfo->uiLayerType = VIDEO_CODING_LAYER;
pLayerBsInfo->uiSpatialId = pCtx->uiDependencyId;
pLayerBsInfo->uiTemporalId = pCtx->uiTemporalId;
pLayerBsInfo->uiQualityId = 0;
pLayerBsInfo->uiPriorityId = 0;
pLayerBsInfo->iNalCount = iNalIdxInLayer;
return 0;
}
} // namespace WelsSVCEnc