Serious performance regression -- some tree optimizer questions

Daniel Berlin dberlin@dberlin.org
Mon Dec 20 00:26:00 GMT 2004



On Mon, 20 Dec 2004, Ulrich Weigand wrote:

>
>
>
>
> Daniel Berlin <dberlin@dberlin.org> wrote on 12/18/2004 02:10:38 AM:
>
>> I've attached a patch that will do the optimization you request.
>
> Thanks for looking into this!
>
>> It should, however, work for your testcase.
>
> Indeed it does, the combinations I was requesting are performed now.
> Unfortunately the different set of temporaries now generated appear
> to trigger some other suboptimal behaviour of ivopts, so the end result
> is still quite bad (even with Zdenek's patch) -- I'll have to
> investigate this further ...


If you find it helps other testcases, or whatever, i'm happy to try to 
explore making that patch usable.
It may actually be worthwile to try one level of backsubstitution per 
expression or something, assuming the number of possible combinations is 
small or the type of expression is expensive.

--Dan



More information about the Gcc mailing list