[LLVMdev] RFC: Should we have (something like) -extra-vectorizer-passes in -O2?

[Hal Finkel]

----- Original Message -----
> From: "Chandler Carruth" <chandlerc at google.com>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: "Andrew Trick" <atrick at apple.com>, "LLVM Developers Mailing List" <llvmdev at cs.uiuc.edu>, "James Molloy"
> <james at jamesmolloy.co.uk>
> Sent: Tuesday, October 14, 2014 12:41:38 PM
> Subject: Re: [LLVMdev] RFC: Should we have (something like) -extra-vectorizer-passes in -O2?
> 
> 
> 
> 
> 
> On Tue, Oct 14, 2014 at 10:28 AM, Hal Finkel < hfinkel at anl.gov >
> wrote:
> 
> 
> > I think it’s generally useful to have an “extreme” level of
> > optimization without much regard for compile time, and in that
> > scenario this pipeline makes sense. But this is hardly something
> > that should happen at -O2/-Os, unless real data shows otherwise.
> 
> Doing all this only at >= -O3 does not seem unreasonable to me.
> FWIW, I think we're being overly conservative if we're relegating
> these to -O3 when the total cost is 2%. That doesn't seem like the
> right tradeoff.

While I think it is not unreasonable not to do it, I also think it would be reasonable to do it. Late cleanup is currently a place where we could use some improvement.

> 
> 
> I actually agree that the set I proposed is on the aggressive end --

I did say 'all' ;)

> that was the point -- but we have more than 2% fluctuations in the
> optimizers' runtime from month to month. If we want to rip stuff out
> it should be because of a principled reason that it isn't going to
> help the code in that phase.

I completely agree.

 -Hal


-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory