variables used in section 8.5.6.4 are confusing

[zhou]

suggest making the following editorial changes to improve the readability. Specifically:

– one (sbWidth + borderExtension) * (sbHeight + borderExtension) prediction sample array predSamplesLX,
– one (sbWidth * sbHeight) motion vector difference array diffMvLX.
Output of this process is the (sbWidth)x(sbHeight) array pbSamplesLX of prediction sample values.
Variable shift1 is set equal to Max( 6, BitDepthY − 6 ).
For x =0..sbWidth − 1, y =0..sbHeight − 1, the following ordered steps apply:
– The variables gradientH[ x ][ y ] and gradientV[ x ][ y ] are derived as follows:
gradientHLX[ x ][ y ] = ( predSamplesLX[ x + 2 ][ y ] >> shift1 ) − ( predSamplesLX[ x ][ y ] >> shift1 ) (8 796)
gradientVLX[ x ][ y ] = ( predSamplesLX[ x ][ y + 2 ] >> shift1 ) − ( predSamplesLX[ x ][ y ] >> shift1 ) (8 797)
– The variable dI is derived as follows:
dI = gradientHLX[ x ][ y ] * diffMvLX[ x ][ y ][ 0 ] + gradientVLX[ x ][ y ] * diffMvLX[ x ][ y ][ 1 ] (8 798)
– Prediction sample value at location ( x, y ) in the subblock is derived as follows:
pbSamplesLX[ x ][ y ] = predSamplesLX[ x + 1 ][ y + 1 ] + ( ( dI + 1 ) >> 1 ) (8 799)

[bbross]

Thanks for spotting this.
Besides that editorial change, there are some more issues related to that, which I fixed in a document we are preparing to be submitted as an editorial input to the 16th JVET meeting:

• borderExtension not defined in the inputs, should be replaced by 2 for PROF input
• clarified in the invocation of the PROF process 8.5.6.4 that the input array is 2 samples larger in width and height than the output sample array.