Sorry, my fault. I missed your FA business at the beginning. All these automatic alignment programs require somewhat similar data. @auto_tlrc and auto_warp.py want the input and base template datasets to be the same modality. There are several ways to go about the alignment.
1.One popular approach is align the raw DWI datasets to a T2 template that is already in alignment with a T1 template and atlas or align the T2 template to a T1-based template, like TT_N27.
2. It is possible to align an FA image to a T1 template. The data is in fact similar in that the white matter is brighter than the gray matter in both types of datasets (except for infants). This approach works reasonably well with @auto_tlrc, and better with align_epi_anat.py with lpa or nmi cost functions and giant_move.
3. Manually Talairach to Talairach space using the stereotaxic procedure in afni. The dataset will be in generic Talairach space, and you may use one of the MNI-TLRC approximations or the TT_N27 template and atlases.
Note if you were doing the DWI tensor calculation in the T1 or T2 template space, you would also want to rotate the gradients, but because you want to put everything back in the original space, that's not an issue; the FA maps are somewhat insensitive to these transformations. Our sister/brother group at NIH that produce the NIH Tortoise package have worked out the details on all sorts of preprocessing steps.