Skip to Content.
Sympa Menu

charm - Re: [charm] cuould -DCMK_OPTIMIZE be --without-trace ?

charm AT lists.cs.illinois.edu

Subject: Charm++ parallel programming system

List archive

Re: [charm] cuould -DCMK_OPTIMIZE be --without-trace ?


Chronological Thread 
  • From: Phil Miller <mille121 AT illinois.edu>
  • To: Jim Phillips <jim AT ks.uiuc.edu>
  • Cc: charm AT cs.uiuc.edu
  • Subject: Re: [charm] cuould -DCMK_OPTIMIZE be --without-trace ?
  • Date: Tue, 9 Feb 2010 14:26:33 -0600
  • List-archive: <http://lists.cs.uiuc.edu/pipermail/charm>
  • List-id: CHARM parallel programming system <charm.cs.uiuc.edu>

On Tue, Feb 9, 2010 at 14:15, Jim Phillips
<jim AT ks.uiuc.edu>
wrote:
> In order to build an "optimized" Charm++ application, I need to specify
> -DCMK_OPTIMIZE on both the build command line and (I assume) in the
> application compilation options.  Is there a reason the build script
> couldn't just add it to conv-mach-opt.h where it would automatically get
> picked up by any application compiling against that build?  That seems to
> be what happens when building Tau.

We should probably have a more encompassing --production flag to the
build script, that turns on the set of options believed to produce
optimal performance on the system in question. This would encompass
-DCMK_OPTIMIZE (which does more than disable tracing), compiler
optimization (which the user shouldn't have to specifically ask of the
runtime), and any build options that make the runtime suitable for
maximum-performance execution.

People doing development of charm itself would still use individual
options, but application users shouldn't have to. I'm not sure exactly
what set of options is best suited to people developing applications
on charm but not modifying charm itself. That's something we can
discuss in the next core group meeting.

Phil





Archive powered by MHonArc 2.6.16.

Top of Page