[
afni.nimh.nih.gov] contains some information on how I turned 3dDeconvolve.c into a parallel program without mangling it hopelessly, and also contains Tom Ross's results from running on an 8 CPU system.
How much interest is there in parallelizing 3dNLfim (nonlinear time series regression)? A quick look at the code indicates that similar code mods as were done to 3dDeconvolve could be inserted using only arthroscopic surgery.
bob cox