Closed ikitayama closed 4 years ago
@ikitayama : that’s interesting because we are deploying neuron with gcc 8.3.0 on our red hat system. It seems like you are using gold linker.
Could you provide the output of cmake .. command and also make VERBOSE=1?
I'm using configure. Is this still supported, right?
Yes, that’s still supported! In that case output of configure and make VERBOSE=1 will be helpful. (And additionally config.log file will be great!)
@pramodk while we are at it, NEURON does not support pthreads nor OpenMP, am I correct?
I will take a look at the logs later today. Thanks!
NEURON does support threading via Pthread.
What about OpenMP?
On Fri, Jun 19, 2020 at 17:29 Pramod Kumbhar notifications@github.com wrote:
I will take a look at the logs later today. Thanks!
NEURON does support threading via Pthread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/neuronsimulator/nrn/issues/605#issuecomment-646512120, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAHLBZ7JCSNOMXI2KQGYMMTRXMOWVANCNFSM4OCN77UQ .
Only the CoreNEURON plugin supports OpenMP
In the source directory, I executed build.sh again and did run configure. Build was ok.
Do I do it every time I did git pull
?
What about OpenMP?
Only the CoreNEURON plugin supports OpenMP
@ikitayama : I am curious about your question / usage about OpenMP. Any specific model or use case that you targeting with OpenMP? I am asking because CoreNEURON doesn't support all features of NEURON and hence if you could describe your use case, we can guide in right direction.
@pramodk I have models that are inherited from a former Ph.D. student, but I am not sure if they are all thread-safe. I thought CoreNEURON is a successor to NEURON, but I was wrong.
I am afraid I can not disclose the model files, sorry.
@pramodk I have models that are inherited from a former Ph.D. student, but I am not sure if they are all thread-safe. I thought CoreNEURON is a successor to NEURON, but I was wrong. I am afraid I can not disclose the model files, sorry.
Ok understood. As Michael said, CoreNEURON is more like plugin in NEURON that we are developing to improve performance on new computer architecture. The details are described in this manuscript.
Hi again @pramodk I have access to our FX 700 system thus I am wondering if the CoreNEURON can fully pull our of power of an SVE Arm core. This should be discussed a bit later though in the CoreNEURON community.
Yeah, that would be very interesting for us as well and I believe CoreNEURON can bring significant performance improvement.
In the past I had access to K system at RIKEN (through Toshiyuki Maeda) but no longer have any active account.
Let us know if we can help in some way (and that will also help us to prepare NEURON/CoreNEURON for A64fx).
Would you provide me a script that does simulate a real life neuronal network so that I can test it on A64FX with SVE enabled?
On Wed, Jun 24, 2020 at 17:37 Pramod Kumbhar notifications@github.com wrote:
Yeah, that would be very interesting for us as well and I believe CoreNEURON can bring significant performance improvement.
In the past I had access to K system at RIKEN (through Toshiyuki Maeda) but no longer have any active account.
Let us know if we can help in some way (and that will also help us to prepare NEURON/CoreNEURON for A64fx).
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/neuronsimulator/nrn/issues/605#issuecomment-648681637, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAHLBZ35CARMDH65HCKH3JLRYG3NLANCNFSM4OCN77UQ .
@ikitayama : Sure. I think it would be also helpful if we test simple kernel of NEURON to see if vectorisation is enabled. From my past conversation with ARM engineers, ARM GCC/LLVM compilers had issue with auto-vectorisation. I will follow with you on this in coming days.
@pramodk during the summer recess, I'd like to test some test kernels on SVE capable cores that I have access to.
@ikitayama : Sorry for delay on this, was busy with multiple other tasks. I will make sure to provide you micro-benchmark in coming week.
@pramodk I fully understand that; I just wanted to touch base with you.
The current master can not be built on x86 with GCC 8.3.0.