Nek5000 / nekRS

our next generation fast and scalable CFD code
https://nek5000.mcs.anl.gov/
Other
284 stars 75 forks source link

Unable to run CUDA mode after Summit RHEL 8 upgrade #369

Closed MalachiTimothyPhillips closed 3 years ago

MalachiTimothyPhillips commented 3 years ago

Describe the bug After the Summit RHEL 8 upgrade, occa has been unable to run CUDA mode in both the master and next branches.

To Reproduce ssh <username>@summit.olcf.ornl.gov

  1. ssh <username>@summit.olcf.ornl.gov
  2. module load gcc cmake cuda
  3. Build, e.g. ./makenrs for master, or ./nrsconfig && cmake --build ./build --target install -j8 for next
  4. Run, e.g., echo "Y" | ./nrsqsub_summit ethier 1 00:05

Pre-compilation output:

                 __    ____  _____
   ____   ___   / /__ / __ \/ ___/
  / __ \ / _ \ / //_// /_/ /\__ \ 
 / / / //  __// ,<  / _, _/___/ / 
/_/ /_/ \___//_/|_|/_/ |_|/____/  v21.1 (ffc02bca)

COPYRIGHT (c) 2019-2021 UCHICAGO ARGONNE, LLC

MPI tasks: 1

using NEKRS_HOME: /ccs/home/malachi/.local/nekrs
using OCCA_CACHE_DIR: /gpfs/alpine/csc262/scratch/malachi/ethier/.cache/occa/

Initializing device 
[CUDA] mode is not enabled, defaulting to [Serial] mode
active occa mode: Serial

active occa mode does not match selected backend!
yslan commented 3 years ago

An easier way to reproduce (master): cd nekrs/3rd_party/occa make -j Then it will show occa info and CUDA is not there as a backend.

stgeke commented 3 years ago

Looks like this was fixed by OLCF.

nandini1610 commented 2 years ago

I have the similar issue let me know the fix please?