Yet Another Opportunity to Optimize by the Front End.

[Steve Kargl]

On Sat, Oct 28, 2006 at 02:14:23PM +0200, Toon Moene wrote:
> 
> One of the peculiarities of this set up is that the "correct" entry in 
> the linked list is found by performing the following test:
> 
>       SUBROUTINE AAP(M,N,...,WIM,....)
>       ...
>       REAL WIM(M,N)
>       ...
>       IF ( ALL(CURRENT % WIM == WIM) ) THEN
>       ...
> 
> IOW, the check for matching is ALL("The stored array equals the array 
> passed as an argument").
> 
> Now this idiom would be greatly sped up if the front end didn't compile 
> this the obvious way:
> 
>       allocate (temp (m,n))
>       loop over n
>          loop over m
>             temp (...) = x(...) == y(...)
>       loop over n
>          loop over m
>             all = all .and. temp(...)
> 
> and evaluated the reduction operator "all" directly into a register.
> 
> Is this already considered by some on their to-do list ?
> 

There is a famous qoute about premature optimization, which
I'm too lazy to look up at the moment.  There is a 4.3 projects
list on the gfortran wiki.  Micro-optimizations isn't listed.

-- 
Steve