Re: [AMBER-Developers] NAB and PBSA broken in CVS?

From: case <case.biomaps.rutgers.edu>
Date: Thu, 29 Oct 2009 22:59:10 -0400

On Thu, Oct 29, 2009, Ross Walker wrote:
>
> Has anyone else noticed problems running the test cases for AMBER Tools?
>
> I just tried the following with ifort l_fc_p_10.1.018 and mkl 10.1.1.019

This is promising...we've had reports before about segfaults in some of
the nab tests, but it always seemed to be on MacOSX 10.6, which I didn't
have any access to. If you can get a failure (I assume this is Linux?),
then we have a better chance of tracking it down.

FWIW, here are the machines I usually test on:

1. Mac OSX 10.5, gcc 4.4.0 (20081219), often but not always with the Goto
BLAS libraries.

2. Linux/intel with icc 11.1.056 and mkl 11.1.056

3. Linux/intel with icc 10.1.018 and mkl 10.0.5.025 (Note that this is pretty
close to what Ross reports above -- same compiler anyway.) I've also tested
with the same icc version and no MKL.

>
> and I get a segfault on the very first test case:
>
> ( cd nab; make test )
> make[1]: Entering directory
> `/server-home/rcw/cvs_checkouts/amber11/test/nab'
> =====================================================
> Running test to make dna duplex:
>
> /bin/sh: line 1: 12593 Segmentation fault ./duplex <duplex.in

This is bug 112 on bugzilla (aside from compiler/OS differences). But my
understanding(?) was that Ben had been able to use -m32 everywhere, and to get
things to work. [Ben: is that correct? if so, please update bug 112.]

Does it fail with AmberTools.07oct09.tar.bz2 (the latest release candidate)?
     In general: are these likely to be new problems, or have you just not
     tested on this particular computer or compiler recently?
Can you catalog which nab tests work and which ones fail? [See bug 113 in
      bugzilla, for example.]
Can you run without MKL?
Can you link without -static?
Then the usual runs without optimization, with gdb, etc.

I don't want to dump all this on your shoulders, but I have never seen the
behavior you report, even with what looks like the same compiler.

[Wild-ass guess: the code for duplex.nab (and what it calls) hasn't changed
for a long time, as far as I can see. But we *have* been playing around a lot
in the last couple of weeks with compiler optimization flags, p4, sse, etc.
My OCFLAGS now has "-O3 -ip -axSTPW", and it might depend not only on what
compiler you have, but on what sort of chipset you are using, AMD vs Intel,
vs who knows?? I have no idea what all these flags mean, or whether or not
one might somehow be creating illegal code on certain types of processors.
Anyway, if this is a recent bug, compiler flags are worth looking at.]

[Also note: for faster testing, you can do these steps:

    cd amber11/src;
    make -f Makefile_at clean
    ./configure....options here
    make -f Makefile_at nabonly (just compile nab)
    cd ../test/nab
    make test (just test nab) ]

...thx...dac


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Oct 29 2009 - 20:00:03 PDT
Custom Search