OK. Thanks again. I'll go digging.
Also I'm using a Beagleboard Rev. B5 with the "slow NEON" unit in it, right? Will Rev. C boards have the later silicon and make a big difference in this area as well (I just read up on that issue)?
I know that compilers have hell with the vector units, but the OMAP 3 does *so* well at all the other stuff! 8> I'll keep digging and post back as I have time.