|
From: Thomas T. <to...@ts...> - 2013-02-18 00:33:28
|
On Sun, Feb 17, 2013 at 6:59 PM, David Burgess <da...@ra...> wrote: > We have managed to (barely) run a combination 5 beacon plus 1 TCH/F on a 500 > MHz Gumstix Overo. So I would expect that you see better performance. Not a > lot better, but better. The Overo (Cortex-A8) has a NEON unit and horribly slow (non-pipelined) scalar floating point unit. In other words, it's an absolute mess for the compiler. But, it's a modern, capable processor and improvements from SIMD optimization can be huge; 10x speedups are not unheard of. On the other hand, the ARMv6 based processor on the Pi is from a past generation. Some code changes are possible (the divide on the sinc filter is brutal), but without a true floating point vector unit, it's just not well suited for signal processing. There's a case here for fixed point, but that's a painful route to go down... Thomas |