times() always returns 1 in NIOS II

Dunno, the altera timer block is horrendously over-complicated and needs lots of clock cycoles to do anything at all (not least because it is a 16bit slave!).

I used a simple 32bit up-counter connected to an avalon slave for performance cycle counting (one avalon cycle to read), the same slave also has a millisecond up-counter we use for actual timers.

I was then able to determine/verify the actual cycle counts for most instructions - although signaltap did show one or two things I hadn't noticed!