[petsc-dev] Seeking OLCF users complaining about poor build times

[Barry Smith]

  configure on MPICH is now the most time consuming part of my build cycle now; on both Mac and Linux

   Barry

> On Mar 2, 2015, at 12:06 PM, Matthew Knepley <knepley at gmail.com> wrote:
> 
> On Mon, Mar 2, 2015 at 12:03 PM, Satish Balay <balay at mcs.anl.gov> wrote:
> On Mon, 2 Mar 2015, Matthew Knepley wrote:
> 
> > On Mon, Mar 2, 2015 at 11:00 AM, Aron Ahmadia <aron at ahmadia.net> wrote:
> >
> > > Another side-note.
> > >
> > > We have made some progress in pulling out some of the major external
> > > dependencies (SuperLU, hypre, Metis) into their own packages in HashDist,
> > > which has been helpful for speeding up rebuilds.  As others have noted,
> > > configure is the real killer for PETSc builds.
> > >
> >
> > I just pushed an update for parallel configure (using only 2 threads right
> > now) to next. We will see if the
> > builds blow up.
> >
> 
> cool.. configure time on my laptop for a simple build reduced from:
> 
> <master>
> real   1m4.681s
> user   0m27.292s
> sys    0m22.514s
> 
> to:
> 
> <next>
> real    0m39.509s
> user    0m24.133s
> sys     0m19.685s
> 
> 
> 
> 'real' time is still longer than 'user' time - so presumably this will
> also benefit 'single' core machines [if any]..
> 
> Excellent. I will put in the flag for setting the number of threads. What should
> it be called? I was going to overload --useThreads=<num> to take a number.
> 
>   Thanks,
> 
>      Matt
>  
> 
> Satish
> 
> 
> 
> 
> 
> -- 
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener