imputed data processing

hi,

When a bfile is created directly from a BGEN file, some genotype calls can become NA. To address this, PLINK2 provides the fill-missing-from-dosage option to recover missing values using dosage information. 

However, I would like to confirm the correct workflow: is it better to fill NA values first and then apply filtering(maf, geno, mind, etc), or to perform filtering before filling NA values?