The aligned anatomical data was transformed to match the epi data, not any mask dataset. To check the alignment, you can use the "-AddEdge" option to align_epi_anat.py or call the @AddEdge script. See more discussion of visualization strategies here:
[
afni.nimh.nih.gov]
The choice of which direction to do the alignment is mostly dependent on your processing pipeline as to which is more convenient. Both epi2anat and anat2epi alignments resample the output dataset to match the grid of the original dataset. This is based on the assumption that motion will be relatively small, so slices are not rotated out of plane too badly. The tlrc output using the tlrc_apar option together with epi2anat uses a minimum voxel dimension for cubic voxels. This can be overridden with the various master options though. Because volume registration is normally applied anyway, the EPI data will be interpolated anyway, so the combined interpolation of alignment to anatomical data and motion correction volume registration does not smooth data further.
Also, as you noticed, the tlrc_apar is not used if you are doing anat2epi only alignment. It is used for the epi2anat alignment.