Thanks, the binary is thumb2 compiled so that's at least working. The binary is stripped (I ran strip on it), so this size is after it's stripped.

I used a few different options to see how they affect the end result. What I think happened, is that in the GCC472 target, the configure script created a slightly different setup, as the libraries each is linked to is slightly different. I'll look into that and try to sort it out.

What options do you suggest using? I did try the options from the wiki page, -mthumb -mfpu=neon -ftree-vectorize -fgraphite and even just -mthumb, they pretty much created the same size executable for me and didn't seem to have much effect on the execution of this app. Oh, and -O3.

Just found this page:
http://boards.openpandora.org/index....ery-toolchain/

What do you think of the suggestions there? The Pandora is similar hardware to the n900 (IIRC) Specifically, this "-mcpu=cortex-a8" which webpages I've looked at seem to suggest might help produce code specifically tuned for our CPU. Also, it seems from the reading I've done, that just passing -mfpu=neon may not be enough to enable neon optimizations due to unsafe math functions (??)

Another page about FP on the A8:
http://pandorawiki.org/Floating_Point_Optimization