Should "Motion vector storing process for geometric partitioning mode" store HpelIfIdx?

[peterderivaz]

In the General clause of "Decoding process for inter blocks", the array HpelIfIdx is written at the same time as other motion vector information is stored:

MvL0[ xSb + x ][ ySb + y ] = mvL0[ xSbIdx ][ ySbIdx ] (887)
MvL1[ xSb + x ][ ySb + y ] = mvL1[ xSbIdx ][ ySbIdx ] (888)
MvDmvrL0[ xSb + x ][ ySb + y ] = refMvL0[ xSbIdx ][ ySbIdx ] (889)
MvDmvrL1[ xSb + x ][ ySb + y ] = refMvL1[ xSbIdx ][ ySbIdx ] (890)
RefIdxL0[ xSb + x ][ ySb + y ] = refIdxL0 (891)
RefIdxL1[ xSb + x ][ ySb + y ] = refIdxL1 (892)
PredFlagL0[ xSb + x ][ ySb + y ] = predFlagL0[ xSbIdx ][ ySbIdx ] (893)
PredFlagL1[ xSb + x ][ ySb + y ] = predFlagL1[ xSbIdx ][ ySbIdx ] (894)
HpelIfIdx[ xSb + x ][ ySb + y ] = hpelIfIdx (895)
BcwIdx[ xSb + x ][ ySb + y ] = bcwIdx (896)

However, in the "Motion vector storing process for geometric partitioning mode", HpelIfIdx is not written:

MvL0[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = mvL0 (1045)
MvL1[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = mvL1 (1046)
MvDmvrL0[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = mvL0 (1047)
MvDmvrL1[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = mvL1 (1048)
RefIdxL0[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = refIdxL0 (1049)
RefIdxL1[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = refIdxL1 (1050)
PredFlagL0[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = predFlagL0 (1051)
PredFlagL1[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = predFlagL1 (1052)
BcwIdx[ ( xSbIdx << 2 ) + x ][ ( ySbIdx << 2 ) + y ] = 0 (1053)

As far as I can understand, I think it is possible for these unspecified HpelIfIdx values to affect the decoding process?

[bbross]

Good catch, thanks! HpekIfIdx should always be written since it might be accessed later so it is definitely missing in "Motion vector storing process for geometric partitioning mode" and shall be added for v4.

[yk]

Fixed in JVET-AN2001-v1.