compiler operations research

compiler operations research

[roy rosen]

Hi,

Are there any known methodologies/tools/flows that enable operations
research on the compiler generated assembly?

The reasoning behind the question is that compiler heuristics
complexity are restricted by compilation time, while test environment
can run for a long time taking into account both intermediate outputs
of the different compilation stages and profiling of execution on real
targets.

The question does not relate to traditional profiler feedback
techniques which are usually restricted for the positioning of code
blocks/functions to minimize change of flow penalty but rather to the
ISA utilization (whether the correct instruction is utilized in the
correct places).

Thanks, Roy.

Re: compiler operations research

[Dave Korn]

On 07/04/2010 12:29, roy rosen wrote:
> Hi,
> 
> Are there any known methodologies/tools/flows that enable operations
> research on the compiler generated assembly?

  Something like MILEPOST+ICI?  An automated machine learning version of gcc:

http://ctuning.org/wiki/index.php/CTools:MilepostGCC
http://unidapt.org/index.php/Developments:ICI

  Might also be some interesting links at

http://www.hipeac.net/tools

    cheers,
      DaveK

Re: compiler operations research

[roy rosen]

Thanks Dave,

I'll have a look at these.

Roy.

2010/4/7, Dave Korn <dave.korn.cygwin@googlemail.com>:
> On 07/04/2010 12:29, roy rosen wrote:
> > Hi,
> >
> > Are there any known methodologies/tools/flows that enable operations
> > research on the compiler generated assembly?
>
>  Something like MILEPOST+ICI?  An automated machine learning version of gcc:
>
> http://ctuning.org/wiki/index.php/CTools:MilepostGCC
> http://unidapt.org/index.php/Developments:ICI
>
>  Might also be some interesting links at
>
> http://www.hipeac.net/tools
>
>    cheers,
>      DaveK
>