On Thu, Feb 23, 2012 at 5:12 AM, Alexander Thomas
> On the one hand this surprises me because I would expect this kind of
> CPU to be able to handle this, but on the other hand I have once run a
> simple DSP-like benchmark where the Overo scored worst of all the
> machines I had ever run that benchmark on (including a Pentium II).
Are any of the other machines (to which you are comparing the Gumstix)
> I found reports of a 10x speed difference between compiling Speex with
> or without ARM optimizations. I would assume that the libspeex that is
> included in the console image is built with these optimizations, or am
> I wrong?
Without seeing the compile line(s) for the library specifically and
without examining the generated assembly it would be impossible to
The omap35xx includes a separate TMS320C64x+ DSP Core Accelerator
Subsystem as well as a NEON SIMD Coprocessor. If neither of these
units are being used and all the processing is going through the
Cortex-A8 then it wouldn't be surprising to see such poor results. If
I'm not mistaken this processor doesn't support division natively
either. So if the library is trying to do, say, floating point
division by emulation on the Cortex-A8 then yes, I would expect the
performance to be terrible.
Maybe the arm.com website has a tech note or library that might help?
I should point out, though, that this is all hypothetical on my part.