cp2k on BlueGene/P
daniel... at gmail.com
Tue Jun 19 08:12:21 UTC 2012
I found some problems running the version psmp on BG/P in JUGENE.
The job (256mpi_taskX4thread) crashes after the initialization, just before
the SCF OT calculation, with following error:
int DCMF::DMA::DMAManyToMany<TDesc>::Recv::processPacket(int, int, const
char*, int) [with TDesc = DMA_MemoryFifoDescriptor]: Assertion `to_copy >=
<Jun 13 12:06:45.749095> BE_MPI (ERROR): The error message in the job
record is as follows:
<Jun 13 12:06:45.749214> BE_MPI (ERROR): "killed with signal 15"
For the compilation I followed the advice of published PRACE Million Atom
KS-DFT with CP2K<http://www.google.com/url?sa=t&rct=j&q=cp2k%20on%20jugene&source=web&cd=1&ved=0CE8QFjAA&url=http%3A%2F%2Fwww.training.prace-ri.eu%2Fuploads%2Ftx_pracetmo%2FCP2K.pdf&ei=SDHgT-6AE-nP4QTz5LHADQ&usg=AFQjCNEiMhfPmxXZZkUthV5MIPqpdMY1yw&sig2=-X3ptOVkJxPf7yuFCCnNrQ>
of Iain Bethune et al.
The only exception is the lack of library get_memory.o, because this is not
in standard cp2k release and I have no idea how to make it.
So maybe this is the problem, but I would need help to code get_memory.
Another strange behavior is in the beginning of the output at the System
Memory Details cp2k finds all zeros:
MEMORY| system memory details [Kb]
MEMORY| rank 0 min max
MEMORY| MemTotal 0 0 0
MEMORY| MemFree 0 0 0
MEMORY| Buffers 0 0 0
MEMORY| Cached 0 0 0
MEMORY| Slab 0 0 0
MEMORY| SReclaimable 0 0 0
MEMORY| MemLikelyFree 0 0 0
But it happens for popt version also and in this case the job runs until
the regular end.
Have someone experience in compiling and running cp2k on BG/P?
How can I properly do it and possibly increase the scalability of cp2k?
Thank a lot in advantage,
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the CP2K-user