I’ve converted some code from SSE to ARM NEON and I tested the execution time of my code on both Ubuntu and Angstrom.
The results are quite different. On Ubuntu it takes much longer and the equivalent C function of my code takes less time, which is very strange.
Any idea why?
Double check your compiler settings on Ubuntu, as by default Ubuntu
only uses the first 16 vfpv3 registers, such that they could support
'lesser' cortex-a8 implementations.. Unlike Angstrom, which is tuned
to fully utilizing the Core...
Regards,
Could you please give me some more details on how to do that? Thank you
Ok I’ll try thank you.