Inliner behaviour - tiny changes lead to huge performance differences

[Bulat Ziganshin]

Hello Simon,

Friday, November 13, 2009, 4:43:34 PM, you wrote:

> I think I can explain this one, at least partially.  When you mark a
> function INLINE, GHC does not optimise the body of the function itself,
> on the grounds that it will be inlined at the call site anyway and get
> optimised there.

yes! i've seen this problem in 6.4. in particular, if function to
inline had a cycle (i.e. self-recursion), it wasn't optimized nor
inlined

-- 
Best regards,
 Bulat                            mailto:Bulat.Ziganshin at gmail.com