Content originally posted in LPCWare by domen on Mon Nov 08 02:24:21 MST 2010
It's not that simple.
Flash transfers go through ICode bus... wait states are for that bus, they don't necessary stall the CPU. Also, yes, 32-bit fetches mean 2 thumb instructions can be fetched at once. And then there's 64-bit cache.
To sum it up: Flash interface on cortex-m3 is quite smart. IIRC you don't even get speed increases when running code from RAM.
Or you could just benchmark it, to see it really makes a difference.