Re: [AMBER] amber14 parallel build problems

From: Ada Sedova <>
Date: Sat, 14 Jan 2017 23:01:57 -0500

Thank you, David.

I will check these scripts. I also have a help ticket in to ORNL OLCF user
assistance but I expect not to hear back from them until after the MLK
holiday. However, I doubt they will be willing to dig too deep into the
code itself.

I agree, it seems to be calling the local yacc.

My colleagues at the NCCS Scientific Computing group also were wondering
why you would HAVE to build in serial first (it is not a common procedure).
Also, it is strange that after advising me to do this, they are also
telling me that there is a common problem with residual files from the
serial build interfering with the parallel build, and I should use <make
distclean>; furthermore there is no rule for distclean, only for clean, in
the top Makefile. Very confusing.

While I have your attention, I was wonder what it would take to get an
upgrade of our site license to amber16/AmberTools16. I would still have to
build it but your group is telling me it has fewer bugs/problems.

Thanks again,

Ada Sedova

-SciComp, NCCS, ORNL

On Sat, Jan 14, 2017 at 3:05 PM, David Case <> wrote:

> On Sat, Jan 14, 2017, Ada Sedova wrote:
> > I will try, burt honestly, looking at the error, it seems that this is a
> > result of a discrepancy between my redhat version of the yacc parser's
> > syntax and the way it is called in the cifparse script.
> Save the config.h script after the serial build, and compare it to the one
> after the parallel configure step, looking at the YACC variable.
> In all cases, you should be using the byacc code we supply, not your
> operating
> system's yacc.
> Also examine the output of "mpicc -show": it should but calling the same
> gcc
> as you used for the serial build.
> We apologize for these problems: you have stumbled upon something I don't
> recall having run across before.
> ...dac
> [Amber developers: someone should step up and figure out *why* we need a
> serial compile first; and then see if we can't remove that requirement.
> I'd also like to split off the nab parallel build (which is leading to the
> problem here: I doubt that very many people use that, and that we should
> say "run 'make mpinb'" if you want it.]
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Sat Jan 14 2017 - 20:30:02 PST
Custom Search