id	summary	reporter	owner	description	type	status	priority	milestone	component	version	resolution	keywords	cc
1184	Mismatch between spec and SW in MV rounding for affine AMVR with 1/16 precision	chhuanb		"The MV rounding process 8.5.2.14 is invoked during affine AMVP list construction (8.5.5.7 Derivation process for luma affine control point motion vector predictors) as following:

–	The rounding process for motion vectors as specified in clause 8.5.2.14 is invoked with mvX set equal to cpMvpLX[ cpIdx ], '''rightShift set equal to AmvrShift''', and '''leftShift set equal to AmvrShift''' as inputs and the rounded cpMvpLX[ cpIdx ] with cpIdx = 0 .. numCpMv − 1 as output.

In case the CPMV precision for affine is 1/16, AmvrShift is set to 0 as '''Table 16'''. 
Then the input rightShift and leftShift to 8.5.2.14 are both 0.
According to 8.5.2.14, in such case if mv >=0, the output mv will become '''mv-1''' :

offset = ( rightShift  = =  0 ) ? 0 : ( 1  <<  ( rightShift − 1 ) )	(599)
mvX[ 0 ] = ( ( mvX[ 0 ] + offset − '''( mvX[ 0 ]  >=  0 )''' )  >>  rightShift )  <<  leftShift	(600)
mvX[ 1 ] = ( ( mvX[ 1 ] + offset − '''( mvX[ 1 ]  >=  0 )''' )  >>  rightShift )  <<  leftShift	(601)

However, the mv is unchanged in SW.

Suggest align spec to SW as following:

8.5.2.14	Rounding process for motion vectors
Inputs to this process are
–	the motion vector mvX,
–	the right shift parameter rightShift for rounding,
–	the left shift parameter leftShift for resolution increase.
Output of this process is the rounded motion vector mvX.
For the rounding of mvX, the following applies:

If rightShift == 0:
  mvX[ 0 ] = mvX[ 0 ] <<  leftShift
  mvX[ 1 ] = mvX[ 1 ] <<  leftShift
else:
  offset = ( 1  <<  ( rightShift − 1 ) )	(599)
  mvX[ 0 ] = ( ( mvX[ 0 ] + offset − ( mvX[ 0 ]  >=  0 ) )  >>  rightShift )  <<  leftShift	(600)
  mvX[ 1 ] = ( ( mvX[ 1 ] + offset − ( mvX[ 1 ]  >=  0 ) )  >>  rightShift )  <<  leftShift	(601)


"	defect	closed	minor	VVC D10	spec		fixed		ksuehring bbross XiangLi fbossen jvet@…