Hello,
On 04/04/2017 02:57 AM, David Case wrote:
> On Mon, Apr 03, 2017, Gerald Monard wrote:
>>
>> However, I've just quickly try to compile AmberTools17 on two different
>> machines: my laptop (pgi 2016 community edition) and at a HPC center (pgi
>> 16). On both, it fails to compile for different reasons (see attachements),
>> but I'd be interested to test the speed of pmemd and cpptraj :-).
>
> The laptop errors are all in /usr/include/c++/6.3.1/exception, which is a
> part of GCC. Is PGI supposed to use this file? Is the community version
> supposed to be compatible with verison gcc 6.3.1?
I dug into it.
PGI seems to rely on gcc internal installation. My laptop uses Fedora 25
which uses gcc 6.3.1 by default. That's a problem and Fedora 25 does not
seem to be really supported by PGI.
I tried to trick PGI by making it aware of my gcc 4.9.4 installation
(editing of the localrc file), but it didn't work. I'll try again later
but let's say that it's a PGI install problem, not Amber's.
>
> The hpc error is in configure: looks like the way pgi is reporting version
> information is not understood by configure2. What is the exact (dot) version
> of PGI you are using? Dan reports sucess at this step for PGI 16.9.
>
> Can you report the output of "pgcc -V" on the hpc machine?
$ pgcc -V
pgcc 16.3-0 64-bit target on x86-64 Linux -tp sandybridge
The Portland Group - PGI Compilers and Tools
Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
I found the problem. It's an install problem in my HPC center.
When I load the PGI module, it creates many variables: CC, FC, CXX, etc.
$ echo $CC
/apps/pgi/linux86-64/16.3/bin/pgcc
Thus configure2 is confused because it runs "$cc -V" which is
"/apps/pgi/linux86-64/16.3/bin/pgcc -V" instead of "pgcc -V".
I unset CC, FC, etc. and then configure2 runs properly.
Now, I can build the serial version with PGI (not tested yet), but the
OpenMP build fails:
make[2]: Entering directory
`/home/gmonard/ReleaseCandidate/build04/amber16/AmberTools/src/saxs'
pgc++ -c -fPIC -O2 -mp -DOPENMP -fPIC
-I/home/gmonard/ReleaseCandidate/build04/amber16/include -o saxs_rism.o
saxs_rism.cpp
PGCC-S-0155-omp atomic pragmas bind only to assignment statements
(saxs_rism.cpp: 645)
PGCC-S-0155-omp atomic pragmas bind only to assignment statements
(saxs_rism.cpp: 672)
PGCC-S-0155-omp atomic pragmas bind only to assignment statements
(saxs_rism.cpp: 679)
PGCC/x86 Linux 16.3-0: compilation completed with severe errors
make[2]: *** [saxs_rism.o] Error 2
make[2]: Leaving directory
`/home/gmonard/ReleaseCandidate/build04/amber16/AmberTools/src/saxs'
make[1]: *** [openmp] Error 2
make[1]: Leaving directory
`/home/gmonard/ReleaseCandidate/build04/amber16/AmberTools/src'
make: *** [install] Error 2
Anybody else can repeat this problem or am I the only one? (then I would
guess that this PGI install is pretty buggy).
Gerald.
>
> We have 120 lines of fairly fragile code in configure2, just to check versions
> from 9 through 16.5 of PGI. I have the feeling that the only time anyone
> uses PGI compilers with Amber is in the weeks right before a release. I
> don't remember any complaints from actual users making it to the mailing list.
>
> ....dac
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
____________________________________________________________________________
Prof. Gerald MONARD
Directeur du mésocentre EXPLOR
Université de Lorraine
Boulevard des Aiguillettes B.P. 70239
F-54506 Vandoeuvre-les-Nancy, FRANCE
e-mail : Gerald.Monard.univ-lorraine.fr
phone : +33 (0)372.745.279
web : http://www.monard.info
____________________________________________________________________________
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Apr 04 2017 - 07:00:02 PDT
