Re: Pitfalls in interference graph ?

Peter Bergner <bergner@vnet.ibm.com>
Mon, 01 Oct 2007 11:57:07 -0500

          From comp.compilers

Related articles
Pitfalls in interference graph ? shafitvm@gmail.com (Mohamed Shafi) (2007-09-28)
Re: Pitfalls in interference graph ? SidTouati@inria.fr (ST) (2007-10-01)
Re: Pitfalls in interference graph ? torbenm@app-1.diku.dk (2007-10-01)
Re: Pitfalls in interference graph ? johnl@iecc.com (2007-10-01)
Re: Pitfalls in interference graph ? bergner@vnet.ibm.com (Peter Bergner) (2007-10-01)
Re: Pitfalls in interference graph ? marcov@stack.nl (Marco van de Voort) (2007-10-01)
Re: Pitfalls in interference graph ? rayiner@gmail.com (Rayiner Hashem) (2007-10-01)
Re: Pitfalls in interference graph ? rayiner@gmail.com (Rayiner Hashem) (2007-10-01)
Re: Pitfalls in interference graph ? jeremy.wright@microfocus.com (Jeremy Wright) (2007-10-02)
Re: Pitfalls in interference graph ? shafitvm@gmail.com (shafi) (2007-10-15)
Re: Pitfalls in interference graph ? torbenm@app-6.diku.dk (2007-10-17)
[4 later articles]
| List of all articles for this month |

From: Peter Bergner <bergner@vnet.ibm.com>
Newsgroups: comp.compilers
Date: Mon, 01 Oct 2007 11:57:07 -0500
Organization: Compilers Central
References: 07-09-104 07-10-007 07-10-010
Keywords: optimize, registers

On Mon, 2007-10-01 at 15:46 +0100, Jeremy Wright wrote:
> FPR vs GPR is an interesting issue. Muchnick, in "Advanced Compiler
> Design and Implementation" recommends combining FPR and GPR
> allocation. There are cases, such as copying large well aligned blocks
> of memory, where one can use an FPR or GPR. Cooper, Harvey, and
> Torczon (all from Rice, as Preston Briggs) in "How to Build an
> Interference Graph" prefer to treat them separately.


What Muchnick recommends and what Cooper et al. recommends are
solutions to two different problems. Muchnick is concerned with
whether we should allocate registers of different types separately or
at the same time. What Cooper et al. are concerned with is how we
represent the interference information in the interference graph (ie,
do we store all the interferences for all of the register/pseudo types
in one big interference graph or do we partition them into separate
interference graphs, one for each register type).


So you can still allocate FPR and GPR registers at the same time as
recommended by Muchnick while using separate interference graphs for
FPRs and GPRs as recommended by Cooper et al. On the other hand, if
you want, you can also allocate FPRs and GPRs separately while using a
combined interference graph. Both issues are just two of the
implementation details up to the compiler designer.


<disclaimer>
I'm not actually arguing for or against either Muchnick's or Cooper et al.'s
recommendations. I'm just trying to show they're not mutually exclusive.
</disclaimer>




Peter
--
Peter Bergner
Linux on Power Toolchain
IBM Linux Technology Center


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.