[Haskell-cafe] Performance with do notation, mwc-random and unboxed vector

Roman Leshchinskiy rl at cse.unsw.edu.au
Wed Jun 13 09:56:07 CEST 2012


On 12 Jun 2012, at 12:52, Dmitry Dzhus <dima at dzhus.org> wrote:

> 12.06.2012, 01:08, "Roman Leshchinskiy" <rl at cse.unsw.edu.au>:
> 
>> perhaps the state hack is getting in the way.
> 
> I don't quite understand the internals of this yet, but `-fno-state-hack` leads to great performance in both cases!
> How safe is that?

It doesn't change the semantics of your program but it can make it significantly slower (or faster, as in this case). The various state hack related tickets on trac might give you an idea of what is happening here.

We really need some proper arity analysis!

Roman





More information about the Haskell-Cafe mailing list