[CIG-LONG] Compiling Gale from Source Code?

[Walter Landry]

"Neil de Laplante" <neildlp at mit.edu> wrote:
> Hello Walter,
> 
> I got the previous problem figured out; it was actually a PETSc issue hidden
> in not having the correct fortran compiler installed, making for a pretty
> cryptic error message.  Anyway, I have a new problem:
> 
> When I run "make -d" in the Gale home directory, it crashes when it can't find
> a StGermain file deep in the code (I think):

I do not think the "-d" option has any effect, since it is not passed
on to scons.

> ≈ unknown option: -rpath=/Users/oeit/Documents/NdLp_CUDA/PETSc/petsc-2.3.2-p10
> /lib/darwin9.6.0-c-debug
> collect2: ld returned 1 exit status
> scons: *** [build/StGermain/Base/FlattenXML/src/FlattenXML] Error 1
> scons: building terminated because of errors.
> Reaping losing child 0x001095f0 PID 28833
> make: *** [all] Error 2
> Removing child 0x001095f0 PID 28833 from chain.
> 
> Sure enough, there's no /src/FlattenXML directory or file, so I don't know
> what it was looking for or why.  Any suggestions?

It looks like it does not like the "-rpath" option.  I am guessing
that you are using a Mac?  Did you configure petsc with the option
"--with-shared=0"?  Can I see a copy of "configure_vars" in your Gale
directory?

> For a bit of context, I'm trying to recompile it myself to add debugging flags
> so that I can profile the code.  Once I've got that figured out I'm going to
> port the bottlenecks, which I'm guessing are the finite element integrals and
> matrix solvers, to compute on Graphics Processing Units (GPUs) using CUDA.

Actually, I think the biggest bottlenecks are particle calculations
and matrix assembly.  But I have not done a rigorous analysis.

Cheers,
Walter Landry
walter at geodynamics.org