Added to MultiPIPE in TMuC 0.7.3 extra information, which makes it possible for the decoder to distribute the task of entropy decoding among several parallel CPUs, such that the computational load on each CPU would be roughly equal. The "BalancedCPU" option in the configuration file sets the expected number of decoding CPUs (see Dresden A120). This option has no effect for serial PIPE.

The load balancing information is added by the encoder, but the decoder silently ignores it in this version.

In the RDOQ function TComTrQuant::xRateDistOptQuant the cost of coding the coded block flag is based on m_pcEstBitsSbac->blockCbpBits which is always 1 bit when RQT is enabled [QuadtreeTUFlag:1 in default config] (it is based on a context initialized to a 50/50 probability and which is not used when RQT is enabled and hence never updated). The cost of the coded block flag is thus not taken into account when deciding whether to encode any coefficient at all in a block (which is not necessarily a bad thing by itself).

When RQT is disabled the cost of the coded block flag is taken into account as m_pcEstBitsSbac->blockCbpBits is updated according to past history. This has an impact on the comparison of RQT on vs off. In some cases (QP=37) the chroma PSNR was seen to drop by about 1dB when RQT is disabled.

The simplest solution may simply be to always ignore the cost of the coded block flag in the RDOQ function. This would not impact the default setting defined in JCT-VC B300.

The two macros DCM_PBIC and HHI_MRG_PU were added in parallel and need to be harmonized with each other