The array m_maxTSLayerMinus1 is declared in TComSlice.h as

UInt m_maxTSLayerMinus1[MAX_VPS_LAYER_ID_PLUS1 - 1];

It should be

UInt m_maxTSLayerMinus1[MAX_VPS_LAYER_ID_PLUS1];

because the #sub-layers is signalled for each layer; hence "-1" should not be there.

In TDecCavlc::parsePPS(), we need to initially set pcPPS->setCGSFlag( 0 ) just in case colour_mapping_enabled_flag is not present in the bitstream.

In a 4 layered scenario where layer 1 depends on layer 0, layer 2 is independent and layer 3 depends on layer 2 so that a decoder can skip layers 0 and 1 and start decoding from layer 3, encoder does not work properly. When I checked the code, I realized that layer 3 does not refer to layer 2 in mode decision due to a bug in the predInterSearchILRUni function. The bugfix may be to change the line number 6013 in TEncSearch.cpp in SHM-6.0 as follows:

if( pcCU->getSlice()->getRefPic(eRefPicList, refIdx)->isILR(pcCU->getLayerId())) && pcCU->getSlice()->getRefPic(eRefPicList, refIdx)->getLayerId() == refLayerId )

I removed the last part where the reference layer Id is compared to refLayerId variable which is obtained from pcSlice->getInterLayerPredLayerIdc(refLayer). (reference layer Id is 2 but the InterLayerPredLayerIdc is 0)

I attached a configuration file where you can duplicate the problem and two encoder log files for SHM-6.0 and bugfix.