Charm++ parallel programming system

[Phil Miller <mille121 AT illinois.edu>]

On Fri, Sep 23, 2011 at 14:01, Pritish Jetley 
<pjetley2 AT uiuc.edu>
 wrote:
> ---------- Forwarded message ----------
> From: Casey W. Stark 
> <caseywstark AT gmail.com>
> Date: Fri, Sep 23, 2011 at 1:55 PM
> Subject: Re: [charm] build arch: mpi-darwin
> To: Pritish Jetley 
> <pjetley2 AT uiuc.edu>
> Cc: charm 
> <charm AT cs.uiuc.edu>
>
>
> Hi Pritish.
> Thanks for the suggestion. I know it's a bit silly to use the mpi build on a
> single node. I was just trying this because I know that's what you would
> build for larger machines. Eventually, I want to use charm on Hopper. Maybe
> it doesn't matter, but I thought it would be good to use the same parallel
> scheme in dev and production. It's a long way off anyway.
> Does this mean you don't support mpi versions on Macs? I understand if it's
> a pain supporting the custom apple stuff, just want to be sure.
> I tried multicore-darwin-x86_64 and it aborted when I used gcc 4.6.0. I
> noticed that charm depends on the Apple-only -arch option, so it fails with
> newer (not apple built) versions of gcc. I might want gfortran for some of
> my code also, so I'm not sure how to reconcile that.
> When I try multicore-darwin-x86_64 with the system gcc,
> "i686-apple-darwin10-gcc-4.2.1 (GCC) 4.2.1 (Apple Inc. build 5666) (dot 3)",
> it also aborts. I checked multicore-darwin-x86_64/tmp/charmconfig.out:
>>
>> g++ -arch x86_64 -Wno-long-double -fPIC -dynamic -fno-common -D_REENTRANT
>> -I../include -I. -c test.cpp -o test.o
>> cc1plus: error: unrecognized command line option "-Wno-long-double"
>
> It appears that the charm build system is using the deprecated
> -Wno-long-double option. Is there an easy way to get rid of this flag?

Casey,

The particular compiler flags used on a given target platform are
defined in src/arch/<target>/conv-mach.sh.  You can edit
multicore-darwin-x86_64/conv-mach.sh to add or remove flags as needed
for your particular environment.

The -Wno-long-double flag was removed from our development tree a few
months ago. Getting releases out the door has been a bit slack in that
time, though.

There's no real reason to use the same kind of Charm++ target for
development on your workstation and on a target supercomputer. They
should all work the same. Where they don't, that's generally either
because you're explicitly using machine-specific features, or it's a
bug.

Phil