[llvm-commits] [PATCH] Stack Coloring optimization

[Sergei Larin]

Nadav,

> Thanks for checking the patch and for the feedback.

My pleasure, if I can help any further with analysis just let me know.

> This optimization reduces the used stack size. It is not supposed to
> change the execution time of your workloads.

I understand that, but was really hoping for the second order effect :)

> Performing slot-merging in the codegen will actually give you more
> control. If you find that slot-merging hurts MI scheduling then we can
> add code to allow you to control some of the thresholds.

Do not get me wrong - I like the idea a lot.
I am just a bit paranoid about things we depend on a lot, and I do not yet
fully understand all possible implications of...

Thanks.

Sergei

---
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by
The Linux Foundation


> -----Original Message-----
> From: Nadav Rotem [mailto:nrotem at apple.com]
> Sent: Wednesday, September 05, 2012 7:24 AM
> To: Sergei Larin
> Cc: llvm-commits at cs.uiuc.edu; 'Jakob Stoklund Olesen'
> Subject: Re: [llvm-commits] [PATCH] Stack Coloring optimization
> 
> Hi Sergei,
> 
> Thanks for checking the patch and for the feedback.
> 
> > I echo the concern expressed by Jakob, but after porting this patch
> to
> > Hexagon and trying it out, in fairness, I see no performance impact
> > (or real
> > benefit) on our internal test suite,
> 
> This optimization reduces the used stack size. It is not supposed to
> change the execution time of your workloads.
> 
> > which has the scenario you are trying
> > to address here... (with the obvious caveat that it really operates
> as
> > intended)...
> > But if in the future, it will reduce in any way aliasing resolution
> > during MI scheduling, this might become a serious issue for Hexagon.
> 
> This optimization pass comes to replace the inline-and-merge-hack which
> merges allocas in the IR.
> 
> Performing slot-merging in the codegen will actually give you more
> control. If you find that slot-merging hurts MI scheduling then we can
> add code to allow you to control some of the thresholds.
> 
> Thanks,
> Nadav