Indexing of predSamples in Section 8.5.6.1

[bheng]

In Section 8.5.6.1, the predSamples output is supposed to be a (cbWidth)x(cbHeight) array.

However, the derivation produces samples for the following range of index:

predSamples[ xL + xSb ][ yL + ySb ] with xL = 0..sbWidth − 1 and yL = 0..sbHeight − 1

Here, the location ( xSb, ySb ) is the top-left sample of the current subblock relative to the top left luma sample of the current picture.

Since they are relative to the picture, not the coding block, in general xSb and ySb will be completely outside the predSamples array.

[jlchen]

Really good catch, Brian.

To solve this issue, I would like to define ( xSbInCb, ySbInCb ) as follows, and use it instead of xSb as the array index of predSamples.

– The sample location ( xSbInCb, ySbInCb ) is set equal to ( xSbIdx * sbWidth, ySbIdx * sbHeight ).

[bheng]

Thanks for the fix!

Should probably use the new ( xSbInCb, ySbInCb ) location in this sentence as well:

"If cIdx is equal to 0, the prediction samples inside the current luma subblock, predSamples[ xL + xSb ][ yL + ySb ] with xL = 0..sbWidth − 1 and yL = 0..sbHeight − 1, are derived as follows"

[bbross]

Thansk for checking and reopening. The array indices will be fixed as follows in this sentence in JVET-S2001-vH:
predSamples[ xL + xSbInCb ][ yL + ySbInCb ]