I have noticed what looks like a discrepancy between the draft spec and HM-5.1. It is regarding the context increment derivation of coeff_abs_level_greater1_flag.

The spec says the following:

• If n is equal to 15 or all previous syntax elements coeff_abs_level_greater1_flag[pos] with pos greater than n are derived to be equal to 0 instead of being explicitly parsed, the following applies.
  • ....
  • When the subset i is not the first one to be processed in this subclause, the following applies.
    • The variable numGreater1 is set equal to the variable numGreater1 that has been derived during the last invocation of subclause 9.3.3.1.1.6 for the syntax element coeff_abs_level_greater2_flag for the subset i + 1.
    • When (numGreater1 >> 1) is greater than 0, ctxSet is incremented by one as follows. ctxSet = ctxSet + 1

In HM-5.1, the comparison above is done before the downshift (TDecSbac.cpp, line 1937):

UInt uiCtxSet = (iSubSet > 0 && eTType==TEXT_LUMA) ? 3 : 0;
if( uiNumOne > 0 )
{

uiCtxSet++;
if(eTType==TEXT_LUMA && uiNumOne > 3)
{

uiCtxSet++;

}

}

uiNumOne >>= 1;

I think this can be fixed in the working draft by simply removing the downshift in the comparison