[petsc-dev] worthless information
Barry Smith
bsmith at mcs.anl.gov
Tue Aug 30 14:29:27 CDT 2011
On Aug 30, 2011, at 2:25 PM, Aron Ahmadia wrote:
> Knowing what compiler 'script' is getting used by PETSc is extremely useful for recreating the build as Matt suggested. Many newbie PETSc errors I deal with arise from mismatching system MPI libraries with a PETSc-installed MPI and this is a quick way to detect the choice of MPI.
Again, I need to make clear, sorry I forgot my sarcasm banner in my previous email: I have no objection to having that line printed but with it must also be printed the true compiler used under the covers.
BTW: yes I still despise MPI compiler wrappers, their existence demonstrates the shear computing incompetence of most people doing HPC.
Barry
>
> A
>
> On Tue, Aug 30, 2011 at 9:33 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>
> On Aug 30, 2011, at 1:27 PM, Satish Balay wrote:
>
> > On Tue, 30 Aug 2011, Barry Smith wrote:
> >
> >>
> >> On Aug 30, 2011, at 1:19 PM, Satish Balay wrote:
> >>
> >>> On Tue, 30 Aug 2011, Barry Smith wrote:
> >>>
> >>>>
> >>>> Compilers:
> >>>> C Compiler: /home/balay/petsc-dev/arch-cuda-double/bin/mpicc
> >>>>
> >>>>
> >>>> It should also print the underlying compiler that is used by mpicc.
> >>>
> >>> Right now - configure.log is the location for details. [and wrt
> >>> --download stuff - make.log also has the configure command which has
> >>> the details]
> >>>
> >>> Sure - we don't have everything in summary printed by configure
> >>
> >> Well if room is so tight in the "summary printed by configure" then why do we waste room printing
> >>
> >>>> C Compiler: /home/balay/petsc-dev/arch-cuda-double/bin/mpicc
> >>
> >> since that conveys no information. Why not take it out and then there will be room for actually providing useful information.
> >
> > There is verious levels of usefullness. Right now its says its using
> > mpicc installed by user - instead of some sysmte mpicc or somthing
> > else.
> >
> > C compiler: /home/balay/petsc-dev/arch-cuda-double/bin/mpicc
> > Fortran compiler: fortran
> >
> > There is still some useful info here.
> >
> > Its just that it doesn't have the info you are currently looking for..
> >
> > Perhaps Matt can update configure to have this info detected in a
> > useable format to be printed here.. [whats in configure.log is some
> > verbose info - not suitable for summary]
>
> Exactly. Whenever it prints the mpicc/f90 etc it should also print in a simple form the compiler backing it up. For example icc or whatever. -show is not good because it prints too much crap
>
> >
> > Satish
>
>
More information about the petsc-dev
mailing list