* Make sure you have buses with signed short immediate fields.
Small negative numbers suck as -1, -4 and -8 are quite common immediates,
And using long immediate often causes bottlenecks.z
Connectivity: (see 2.3.3 of manual)
* Full connectivity leads to very slow achievable clock speeds.
* Only Reachable connectivity leads to addition of extra
register-to-register copies and cause considerable slowdows
* Best compromise between IPC and clock speed is _often_ "Fully RF Connected".
This is however not always true; If the clock speed is limited by some other
component than the interconnect, Directly reachable might give better performance,
and sometimes using only Reachable connectivity might give so big clock speed advantage
that the overall performance is still better, but currently this is quite rare.
After some improvements to the compiler the performance of reachable architectures
may have definite improvements, which will also lead updating this guide.
* Make sure you have enough registers, and register read ports.
But using more than 2 read ports in one register file may make it big and power-hungry.
Consider using multiple register files.