Potential issue in the range of MaxNumGpmMergeCand

[chunchic]

According to the semantics of max_num_merge_cand_minus_max_num_gpm_cand, the value of MaxNumGpmMergeCand can be either 0 or anyone between 2 and MaxNumMergeCand, inclusive.
However, the last sentence of the semantics (as shown below) of max_num_merge_cand_minus_max_num_gpm_cand states that "The value of MaxNumGpmMergeCand shall be in the range of 2 to MaxNumMergeCand, inclusive.", which conflicts with that MaxNumGpmMergeCand may be 0 in some case when max_num_merge_cand_minus_max_num_gpm_cand is not present.

max_num_merge_cand_minus_max_num_gpm_cand specifies the maximum number of geometric partitioning merge mode candidates supported in the SPS subtracted from MaxNumMergeCand.
The maximum number of geometric partitioning merge mode candidates, MaxNumGpmMergeCand, is derived as follows:
if( sps_gpm_enabled_flag  &&  MaxNumMergeCand  >=  3 )
	MaxNumGpmMergeCand = MaxNumMergeCand −
			max_num_merge_cand_minus_max_num_gpm_cand	(66)
else if( sps_gpm_enabled_flag  &&  MaxNumMergeCand  = =  2 )
	MaxNumGpmMergeCand = 2
else
	MaxNumGpmMergeCand = 0
The value of MaxNumGpmMergeCand shall be in the range of 2 to MaxNumMergeCand, inclusive.

Below is the suggested fix:

When the max_num_merge_cand_minus_max_num_gpm_cand is present, the value of MaxNumGpmMergeCand shall be in the range of 2 to MaxNumMergeCand, inclusive.

[yk]

Good catch! Thanks! It seems better to fix this specifying that the value of max_num_merge_cand_minus_max_num_gpm_cand shall be in the range of 0 to MaxNumMergeCand − 2, inclusive. This way is more consistent with other similar places in the spec. Fixed in Q2001-vE.