darcs patch: Don't charge for int2Word# and word2Int# -- theydon't...

[Simon Marlow]

Bulat Ziganshin wrote:
> Hello Simon,
> 
> Monday, November 6, 2006, 1:54:44 PM, you wrote:
> 
> 
>>I find the instruction counts and memory read/write counts from cachegrind are
>>good things to compare, as a slightly more sensitive, but still deterministic,
>>alternative to allocations.  Caveats apply of course - these figures are not
>>always a good indicator for real world performance, but they are still a useful
>>tool if you bear this in mind.
> 
> 
> if this tool can do CPU cache modelling while counting memory
> accesses, it may become rather exact measure. it should just check
> that memory access is to same memory row (16/32) bytes that is among,
> say, last 256kb accessed

Unfortunately, cachegrind turns out not to be as useful as it sounds for cache 
simulation.  The problem is that to do a good job of cache simulation you have 
to do a pretty sophisticated simulation of the CPU itself, knowing when the CPU 
will speculatively execute code, for example.  I believe there are lots of 
interesting details in Nick Nethercote's thesis (he's the author of cachegrind 
and did some investigation of GHC's cache behaviour a while back).

Cheers,
	Simon