Thanks Bob,

Yes, we do use a mask created by 3dAutomask. Our main "problem" is that we are looking at learning over the course of about 100 trials (700 EPI BOLDs) using 20 or more regressors in 3dDeconvolve on data that's been normalized and resampled to +tlrc.

There seem to be several possibilties in terms of compilers. Below is info from a FAQ page for the VAST compiler suggesting that we might be able to transform C/C++ and Fortran loops into "vectorized" code without rewriting or adding anything to source code.

Does the AFNI source code, 3dDeconvolve in particular, contain much in the way of C/C++ and Fortran loops?

Thanks, and we'll let you know how things go.

-jim

From the [www.psrv.com] web page...
VAST is a high-level optimizer that automatically optimizes C/c++ and
Fortran programs. For AltiVec-enhanced processors, it can automatically
transform loops into AltiVec instructions. VAST is normally used with a
compiler driver that combines vectorization with the rest of the compilation
process. VAST vectorizes C code by replacing loops with AltiVec extensions. There are no comparable extensions for Fortran, so VAST vectorizes Fortran code by cutting out vectorizable sections into C routines that contain AltiVec
extensions.