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

[Malcolm Wallace]

On 11 Jun 2012, at 10:38, Dmitry Dzhus wrote:

> main = do
>  g <- create
>  e' <- VU.replicateM count $ standard g
>  return ()

In all likelhood, ghc is spotting that the value e' is not used, and that there are no side-effects, so it does not do anything at runtime.  If you expand the action argument to replicateM, such that it uses do-notation instead, perhaps ghc can no longer prove the lack of side-effects, and so actually runs the computation before throwing away its result.

When writing toy benchmarks in a lazy language, it is always important to understand to what extent your program _uses_ the data from a generator, or you are bound to get misleading performance measurements.

Regards,
    Malcolm