Using a variable QP, a CTU size of 64 and class D images (ie, the bottom CTU row does not fully intersect the image being coded), there is a QP inheritance problem on the last row if a dependent slice starts later than the left-most CTU. There is a decoder mismatch during decoding, as a wrong QP may be used.

This is due to the fact that the data in CTUs corresponding to an area outside of the image are not correctly initialized in the decoder when a slice (dependent or independent) is terminated. These data are later used, however, when searching for a 'last QP coded' from a dependent-slice CTU directly to the right.

The attached patch prevents a 'break' occuring during traversal of the CTU when 'isLast' is seen, and correctly completes CTU initialisation when 'isLast' is true.

Possibly related to ticket #671, but this issue was not seen with HM 9.0.