Test report by @Flamefire
SUCCESS
Build succeeded for 4 out of 4 (4 easyconfigs in total)
taurusml25 - Linux RHEL 7.6, POWER, 8335-GTX (power9le), 6 x NVIDIA Tesla V100-SXM2-32GB, 440.64.00, Python 2.7.5
See https://gist.github.com/3e0e61935afee1f938a1c520ff5a4109 for a full test report.

Tested with an installed PyTorch/1.6.0-fosscuda-2019b-Python-3.7.4 via:

• eb PyTorch-1.6.0-fosscuda-2019b-Python-3.7.4.eb --rebuild --installpath=/tmp --stop=patch
• ml PyTorch/1.6.0-fosscuda-2019b-Python-3.7.4
• cd /dev/shm/s3248973-EasyBuild/PyTorch/1.6.0/fosscuda-2019b-Python-3.7.4/pytorch-1.6.0/test
• for i in OpenBLAS/0.3.7-GCC-8.3.0 OpenBLAS/0.3.8-GCC-9.2.0 OpenBLAS/0.3.9-GCC-9.3.0 OpenBLAS/0.3.12-GCC-10.2.0; do echo $i; ml $i && python test_nn.py TestNNDeviceTypeCPU.test_conv_large_cpu; done

Prior to this PR it failed, after the PR it succeeded.

Note that the patch differs slightly from upstream as I needed to backport it:

Until 0.3.11 there was no #ifdef branch for POWER10 (introduced with OpenMathLib/OpenBLAS#2686) so up until 0.3.10 the only required patch is for POWER9, i.e. https://github.com/xianyi/OpenBLAS/blob/v0.3.10/param.h#L2263

For 0.3.12 POWER9 and POWER10 used the same values, so there is also only 1 change required: https://github.com/xianyi/OpenBLAS/blob/v0.3.12/param.h#L2391

In 0.3.14 and above the issue is fixed upstream by the linked PR.

Going in, thanks @Flamefire!

@boegelbot please test @ generoso

@boegel: Request for testing this PR well received on login1

PR test command 'EB_PR=16199 EB_ARGS= /opt/software/slurm/bin/sbatch --job-name test_PR_16199 --ntasks=4 ~/boegelbot/eb_from_pr_upload_generoso.sh' executed!

• exit code: 0
• output:

Submitted batch job 9174

Test results coming soon (I hope)...

Test report by @boegelbot
SUCCESS
Build succeeded for 4 out of 4 (4 easyconfigs in total)
cns1 - Linux Rocky Linux 8.5, x86_64, Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz (haswell), Python 3.6.8
See https://gist.github.com/9f15deace183de3d2a688b5178046190 for a full test report.