[llvm-commits] Intel Atom optimization - use LEA to adjust stack pointer

[Zhang, Andy]

Hi Anton,

We are planning on implementing a number of optimizations that apply only to
the this family of processors.

I agree that if the optimization could apply to multiple processors, we should
create one subtarget feature per optimization. But in this case, the feature is
really "optimizations specific to Atom". Maybe it could be named something else.

Hope that clarifies things a bit. Do you still strongly oppose such a subtarget
feature in this case?

Regards,
Andy

> -----Original Message-----
> From: Anton Korobeynikov [mailto:anton at korobeynikov.info]
> Sent: Monday, January 16, 2012 5:26 PM
> To: Zhang, Andy
> Cc: llvm-commits at cs.uiuc.edu
> Subject: Re: [llvm-commits] Intel Atom optimization - use LEA to adjust
> stack pointer
> 
> Hi Andy,
> 
> > Do you mean that this optimization should be enabled by a separate
> subtarget feature?
> > Since this optimization really only applies to the Atom, we added a
> blanket 'Atom'
> > subtarget feature, and enable Atom-specific optimizations when the
> Atom subtarget
> > feature is enabled.
> Usually the features are named based on the "contents" of the feature.
> E.g. the name of the instruction set, or what the feature actually
> means.
> For me naming as "atom" is definitely wrong idea. Consider, there will
> be some other CPU which will require the same optimization?
> 
> --
> With best regards, Anton Korobeynikov
> Faculty of Mathematics and Mechanics, Saint Petersburg State University