[QE-developers] Problem building QE 6.6a2

[Louis Stuber]

Hi Passant,

Would you mind using 6.6a1 instead ? I’ve heard about issues with 6.6a2, and managed to validate 6.6a1 myself.

I validated it with PGI 19.10 + CUDA 10 and 11.

Going forward I’d recommend to build with NVHPC 20.9<https://developer.nvidia.com/nvidia-hpc-sdk-downloads> (contains PGI+CUDA) but this requires changes in the configure script because the compiler was renamed from pgcc to nvc and that causes parsing issue.


Alternatively, if you’re OK with containers, there is a public one for qe-gpu, super easy to run with docker:

docker run --gpus all --shm-size 4GB --privileged -it nvcr.io/hpc/quantum_espresso:v6.6a1

This container has a few known perf issues (10-15%) but should be fixed in future updates.



Best regards,
Louis Stuber
Nvidia – HPC devtech

From: developers <developers-bounces at lists.quantum-espresso.org> On Behalf Of Passant A. Hafez
Sent: Monday, November 23, 2020 11:51 AM
To: developers at lists.quantum-espresso.org
Subject: [QE-developers] Problem building QE 6.6a2

External email: Use caution opening links or attachments

Hello,

I have a problem building QE 6.6a2

At a stage I'm getting:


./q-e-gpu-qe-gpu-6.6a2_2/PW/src/add_vuspsi_gpu.f90:228: undefined reference to `tb_dev__tb_dev_t___td_'
./q-e-gpu-qe-gpu-6.6a2_2/PW/src/add_vuspsi_gpu.f90:228: undefined reference to `_gbuffers_0_'

A lot of these.


The configure command:
./install/configure --prefix=$PREFIX --enable-parallel --enable-openmp --wit
h-cuda=/sw/csgv/cuda/10.2.89 --with-cuda-cc=60 --with-cuda-runtime=10.2 --with-scalapack=no

Using PGI 18.10

Also I tried these and all failed:

pgi20.7_cuda11.0.194
pgi20.1_cuda10.2
pgi18.10_cuda9.2
pgi18.10_cuda10.2

Let me know if further details are needed. If you can advise which versions of CUDA and PGI and other dependencies if any to use that's tested with this version of QE (6.6a2), that'd be great.

--
Passant
​
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20201123/77f5a820/attachment.html>