[Users] "signbit" not defined during compilation of ETK on XSEDE:Kraken?
Erik Schnetter
schnetter at cct.lsu.edu
Tue Feb 7 11:03:17 CST 2012
On Tue, Feb 7, 2012 at 11:50 AM, Kelly, Bernard J.
(GSFC-660.0)[UNIVERSITY OF MARYLAND BALTIMORE COUNTY]
<bernard.j.kelly at nasa.gov> wrote:
> That might be an idea. Do you have a feeling for the relative efficiency
> of PGI vs Intel on this machine?
There is a table near the bottom of
<https://docs.einsteintoolkit.org/et-docs/Vectorisation>. Hopper and
Kraken had very similar configurations when this table was made.
At this point, vectorisation did not work with PGI, leading to
439.572s (PGI) vs. 272.993s (Intel) for the run time for this
benchmark.
Of course, this is a benchmark, and your numbers will vary.
-erik
> I sent the NICS people a ticket about the later compilation issues (who
> knows; perhaps they're coming from the same source as the signbit problem).
>
> Bernard
>
> P.S. Any idea why my ETK TRAC ticket wasn't accepted? I don't have a login
> (I think ...), so I was submitting as "bernard.j.kelly at nasa.gov".
>
>
> On 2/7/12 11:44 AM, "Erik Schnetter" <schnetter at cct.lsu.edu> wrote:
>
>>Bernard
>>
>>The PGI compiler is often picky about what C++/OpenMP constructs it
>>accepts. Do you want to try the Intel compiler instead? I've been
>>using it successfully on Kraken.
>>
>>-erik
>>
>>On Tue, Feb 7, 2012 at 10:54 AM, Kelly, Bernard J.
>>(GSFC-660.0)[UNIVERSITY OF MARYLAND BALTIMORE COUNTY]
>><bernard.j.kelly at nasa.gov> wrote:
>>> Hi Erik.
>>>
>>> Thanks for the advice. I commented out all references to "signbit", and
>>>to
>>> "kifpos" lower down, and it now gets through the Vectors thorn. I'm now
>>> hitting an issue with an omp pragma in CarpetLib, but that's a separate
>>> problem.
>>>
>>> BTW, I just tried to create a ticket (about the signbit issue) on
>>> trac.einsteintoolkit.org, but ...
>>>
>>> "Submission rejected as potential spam"
>>>
>>> I thought it was the list of modules I put in, but even without them it
>>> gets blocked.
>>> Also, the TRAC system doesn't list the Maxwell release (ET_2011_10) as a
>>> possible "version", which makes it hard to give an accurate report.
>>>
>>> Bernard
>>>
>>>
>>>
>>>-------------------------------------------------------------------------
>>>--
>>> --
>>> Bernard Kelly -- CRESST Research Associate, NASA/GSFC
>>>
>>> Phone: +1 (301) 286-7243
>>> E-Mail: bernard.j.kelly at nasa.gov
>>> Web:
>>>
>>>http://science.gsfc.nasa.gov/sed/index.cfm?fuseAction=people.jumpBio&ipho
>>>ne
>>> bookid=13052
>>>
>>>-------------------------------------------------------------------------
>>>--
>>> --
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> On 2/6/12 11:17 PM, "Erik Schnetter" <schnetter at cct.lsu.edu> wrote:
>>>
>>>>Bernard
>>>>
>>>>test.cc is just a self-test. You can try commenting out the respective
>>>>code (or the whole routine), and see where this gets you.
>>>>
>>>>Carpet (which may also use signbit) has a different, more complex
>>>>logic to handle various levels of C++ compliance, and may actually
>>>>still do the right thing.
>>>>
>>>>-erik
>>>>
>>>>On Mon, Feb 6, 2012 at 10:53 PM, Kelly, Bernard J.
>>>>(GSFC-660.0)[UNIVERSITY OF MARYLAND BALTIMORE COUNTY]
>>>><bernard.j.kelly at nasa.gov> wrote:
>>>>> Hi all.
>>>>>
>>>>> I recently checked out ETK Maxwell on the machine XSEDE machine
>>>>>Kraken,
>>>>> and am having trouble compiling and running a simple test case. The
>>>>> problem is with LSUThorns/Vectors/src/test.cc, where the "signbit"
>>>>> function is (apparently) undefined. No doubt the problem recurs with
>>>>>later
>>>>> source files, but I can't get past this one without abandoning Vectors
>>>>>and
>>>>> hence all of Carpet.
>>>>>
>>>>> Now I can compile a "hello world" program in C invoking signbit.cc
>>>>>without
>>>>> a problem, as long as I include <math.h>, so presumably the ETK
>>>>>process
>>>>>is
>>>>> looking in the wrong location for this library or its headers.
>>>>>
>>>>> Since Kraken is a "known" machine in ETK's simfactory, and lots of
>>>>> Numerical Relativity groups must have used it during the recently
>>>>>expired
>>>>> "NR-AR" allocation, I assume the problem's at my end. Any ideas?
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Bernard
>>>>>
>>>>> Some details:
>>>>>
>>>>> (1) I initially used the simfactory "build" mechanism, but I have this
>>>>> problem with a straightforward Cactus "make" as well.
>>>>>
>>>>> (2) Here's a list of my loaded modules on Kraken. I'm using the PGI
>>>>> compilers (item 18 below):
>>>>>
>>>>> 1) modules/3.1.6.5
>>>>> 2) torque/2.4.14
>>>>> 3) moab/5.4.3.s16991
>>>>> 4) /opt/cray/xt-asyncpe/default/modulefiles/xtpe-istanbul
>>>>> 5) tgusage/3.0-r2
>>>>> 6) altd/1.0
>>>>> 7) DefApps
>>>>> 8) xtpe-target-cnl
>>>>> 9) xt-service/2.2.74
>>>>> 10) xt-os/2.2.74
>>>>> 11) xt-boot/2.2.74
>>>>> 12) xt-lustre-ss/2.2.74_1.6.5
>>>>> 13) cray/job/1.5.5-0.1_2.0202.21413.56.7
>>>>> 14) cray/csa/3.0.0-1_2.0202.21426.77.7
>>>>> 15) cray/account/1.0.0-2.0202.19482.49.18
>>>>> 16) cray/projdb/1.0.0-1.0202.19483.52.1
>>>>> 17) Base-opts/2.2.74
>>>>> 18) pgi/11.4.0
>>>>> 19) xt-libsci/10.5.02
>>>>> 20) pmi/2.1.4-1.0000.8596.15.1.ss
>>>>> 21) xt-mpt/5.2.3
>>>>> 22) xt-pe/2.2.74
>>>>> 23) xt-asyncpe/4.9
>>>>> 24) PrgEnv-pgi/2.2.74
>>>>> 25) cray/MySQL/5.0.64-1.0202.2899.21.1
>>>>> 26) git/1.7.4.2
>>>>> 27) python/2.6.4
>>>>> 28) mercurial/1.6.4
>>>>> 29) subversion/1.6.9
>>>>> 30) hdf5-parallel/1.6.10
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>-----------------------------------------------------------------------
>>>>>--
>>>>>--
>>>>> --
>>>>> Bernard Kelly -- CRESST Research Associate, NASA/GSFC
>>>>>
>>>>> Phone: +1 (301) 286-7243
>>>>> E-Mail: bernard.j.kelly at nasa.gov
>>>>> Web:
>>>>>
>>>>>http://science.gsfc.nasa.gov/sed/index.cfm?fuseAction=people.jumpBio&ip
>>>>>ho
>>>>>ne
>>>>> bookid=13052
>>>>>
>>>>>-----------------------------------------------------------------------
>>>>>--
>>>>>--
>>>>> --
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Users mailing list
>>>>> Users at einsteintoolkit.org
>>>>> http://lists.einsteintoolkit.org/mailman/listinfo/users
>>>>
>>>>
>>>>
>>>>--
>>>>Erik Schnetter <schnetter at cct.lsu.edu>
>>>>http://www.cct.lsu.edu/~eschnett/
>>>
>>
>>
>>
>>--
>>Erik Schnetter <schnetter at cct.lsu.edu> http://www.cct.lsu.edu/~eschnett/
>
--
Erik Schnetter <schnetter at cct.lsu.edu> http://www.cct.lsu.edu/~eschnett/
More information about the Users
mailing list