Scott,
I am not building in a separate directory anymore. I was told that was a
problem a while ago. The parallel build IS rebuilding yacc. I traced it.
The logfiles are, as I said in a previous thread that was not responded to,
and it WAS addressed to the amber email.
Ada
On Wed, Jan 18, 2017 at 2:06 PM, Scott Brozell <sbrozell.rci.rutgers.edu>
wrote:
> Hi,
>
> A note on practical approaches - if you follow Hai's advice to
> define YACC then you need to use make -e
>
> A note on the 'amber side bigger picture' - of course we are not
> rebuilding yacc -- you have no yacc since you are building parallel
> in a separate directory without having built serial...
>
> A debugging comment - these parallely built yaccs work fine for building
> cifparse :
> mpicc -show -v
> mpicc for MVAPICH2 version 2.1
> gcc -v -I/usr/local/mvapich2/gnu/5.2/2.1/include
> -L/usr/local/mvapich2/gnu/5.2/2.1/lib -Wl,-rpath
> -Wl,/usr/local/mvapich2/gnu/5.2/2.1/lib -Wl,--enable-new-dtags -lmpi
> and
> mpicc for MVAPICH2 version 2.1
> icc -v -I/usr/local/mvapich2/intel/15/2.1/include
> -L/usr/local/mvapich2/intel/15/2.1/lib -Wl,-rpath
> -Wl,/usr/local/mvapich2/intel/15/2.1/lib -Wl,--enable-new-dtags -lmpi
>
> IMHO take a practical approach, document it for your future use, move on.
>
> scott
>
> On Wed, Jan 18, 2017 at 01:44:41PM -0500, Scott Brozell wrote:
> >
> > As a practical matter you could produce the serial yacc and change its
> > permission to owner not writable and then build in parallel perhaps
> > with make -i if a non writable amberhome/bin/yacc produces an error.
> >
> > FWIW i just checked this thread :
> > [AMBER] amber14 parallel build problems
> > and i do not see the verbose and verbatim details such as logfiles etc.
> > Perhaps these did not make it into the list.
> > Also the message below appears to have been sent to an individual not to
> > the list.
> >
> > If we really really want to find the local problem then you may need to
> > run the parallelly produced yacc in a debugger. I'll take a look at the
> > amber side bigger picture - why rebuild yacc if a serial build already
> > made it.
> >
> > scott
> >
> > On Wed, Jan 18, 2017 at 01:19:24PM -0500, Ada Sedova wrote:
> > > Hi, I have not received any responses about this problem since my last
> > > reply. There is also another reply from me in a previous thread which
> > > provided the requested logfiles and other info.
> > >
> > > I am still trying (unsuccessfully) to solve this problem. I think I
> can be
> > > pretty confident that the mpi make is rebuilding the byacc binary,
> however,
> > > and that is what is causing errors.
> > >
> > >
> > > Thanks,
> > >
> > > Ada
> > >
> > >
> > > ---------- Forwarded message ----------
> > > From: Ada Sedova <ada.a.sedova.gmail.com>
> > > Date: Tue, Jan 17, 2017 at 2:42 PM
> > > Subject: some clues about yacc errors on OLCF rhea
> > > To: david.case.rutgers.edu
> > >
> > >
> > > Hi David,
> > >
> > > I'm guessing you guys are building yacc as part of the amber build.
> When I
> > > run
> > >
> > > /ccs/home/adaa/AMBER/amber14/bin/yacc -v -d cifparse.y
> > >
> > > from the serial-build folder,
> > >
> > > I get no error.
> > >
> > >
> > > However, when I run the same command from the parallel build folder,
> > >
> > > I get the same error that crashes the make install:
> > >
> > >
> > > bash-4.1$ /ccs/home/adaa/AMBER/AMBER_PARALLEL/amber14/bin/yacc -v -d
> > > cifparse.y
> > >
> > > yacc:39196 terminated with signal 11 at PC=40abac SP=7ffed5cb7530.
> > > Backtrace:
> > >
> > > /ccs/home/adaa/AMBER/AMBER_PARALLEL/amber14/bin/yacc[0x40abac]
> > >
> > > /ccs/home/adaa/AMBER/AMBER_PARALLEL/amber14/bin/yacc[0x40ce6b]
> > >
> > > /ccs/home/adaa/AMBER/AMBER_PARALLEL/amber14/bin/yacc[0x404dfd]
> > >
> > > /lib64/libc.so.6(__libc_start_main+0xfd)[0x7fcb1871ed1d]
> > >
> > > /ccs/home/adaa/AMBER/AMBER_PARALLEL/amber14/bin/yacc[0x400ea9]
> > >
> > >
> > > Also. when I try it using the system yacc in the parallel folder, I
> get no
> > > error:
> > >
> > > -bash-4.1$ which yacc
> > >
> > > /usr/bin/yacc
> > >
> > > -bash-4.1$ /usr/bin/yacc -v -d cifparse.y
> > >
> > > (no error)
> > >
> > >
> > > I'm guessing that building byacc with mpicc is incompatible with our
> > > system, and I should try to get the cifparse workflow to use the system
> > > yacc, if that would work.
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jan 18 2017 - 12:00:03 PST
