{lib}[foss/2021b] TensorFlow v2.8.4 w/ CUDA-11.4.1 and fix patches + extensions in easyconfig for TensorFlow 2.8.4 w/ foss/2021b

[Flamefire]

This adds the new TF 2.8.4 EC with CUDA support and some changes for the non-CUDA version:

• Add missing astor package (SYSTEM_LIB, i.e. will be downloaded when missing which we want to avoid)

• Add the profile plugin for tensorboard as we did in previous ECs

• Reorder sources, patches, checksums so they are next to each other for easier manual verification

• Replace TensorFlow-2.1.0_fix-cuda-build.patch by the upstream patch from resolve gcc_host_compiler_path in a symlink directory tensorflow/tensorflow#56360 so we can drop it in the future

• Fix build on PPC (that was some major work)

• the CUDA build requires the updated easyblock from fix TensorFlow easyblock for new versions of Bazel & TensorFlow easybuild-easyblocks#2854 to work due to changes in TensorFlow and Bazel.

Once this is merged we can finish TF 2.9 in #16620 and #16008 as they suffer from the same issues as the ones fixed here.

[Flamefire]

Test report by @Flamefire
Using easyblocks from PR(s) easybuilders/easybuild-easyblocks#2854
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in total)
taurusml5 - 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/760de78429e2c0d951a3901327fdd7a0 for a full test report.

[jfgrimm]

Test report by @jfgrimm
FAILED
Build succeeded for 10 out of 11 (2 easyconfigs in total)
gpu02.pri.viking.alces.network - Linux CentOS Linux 7.9.2009, x86_64, Intel(R) Xeon(R) Gold 6138 CPU @ 2.00GHz (skylake_avx512), 1 x NVIDIA Tesla V100-SXM2-32GB, 510.47.03, Python 3.6.8
See https://gist.github.com/753931b5a275c7cec40e69a5687a90d1 for a full test report.

[Flamefire]

@jfgrimm The CUDA build requires the new easyblock from easybuilders/easybuild-easyblocks#2854

I made that more clear in the description but your test verifies that indeed (only) the CUDA build requires the update (but doesn't hurt the CPU variant)

[jfgrimm]

yeah I read that, then forgot to add the --include-easyblocks-from-pr bit anyway before submitting 🤦

[jfgrimm]

Test report by @jfgrimm
Using easyblocks from PR(s) easybuilders/easybuild-easyblocks#2854
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in total)
gpu01.pri.viking.alces.network - Linux CentOS Linux 7.9.2009, x86_64, Intel(R) Xeon(R) Gold 6138 CPU @ 2.00GHz (skylake_avx512), 2 x NVIDIA Tesla V100-SXM2-32GB, 510.47.03, Python 3.6.8
See https://gist.github.com/629db23d8a62faa3187616bd114d57c5 for a full test report.

[akesandgren]

Test report by @akesandgren
FAILED
Build succeeded for 0 out of 2 (2 easyconfigs in total)
b-cn1501.hpc2n.umu.se - Linux Ubuntu 20.04, x86_64, Intel(R) Xeon(R) Gold 6132 CPU @ 2.60GHz, 2 x NVIDIA Tesla V100-PCIE-16GB, 470.182.03, Python 3.8.10
See https://gist.github.com/akesandgren/7f5e9427ee0057f0eb74abe40d8887a6 for a full test report.

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in total)
bear-pg0105u03a.bear.cluster - Linux RHEL 8.6, x86_64, Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz (icelake), Python 3.6.8
See https://gist.github.com/branfosj/ca4813f4055b017956b4fd91b6a8df7f for a full test report.

[Flamefire]

After #17892 I'm not sure the replacement of the fix-cuda-build patch by the upstream patch is a good idea as it can cause regressions when using rpath wrappers or ccache, see #17892 (comment) for how fixing the rpath issue is possible from easybuild-framework (but not the ccache use case)

Given the good state of this PR (the only failing report is due to lack of space) I'd suggest to merge this anyway and then do a bulk-update of the TensorFlow ECs (re-)adding the fix-cuda-build patch (potentially additionally to the upstream patch) and doing a single test report with --stop=patch as that patch is proven to work already so we just need to ensure that it applies.

[branfosj]

Going in, thanks @Flamefire!

[branfosj]

Test report by @branfosj
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in total)
bear-pg0203u29a.bear.cluster - Linux RHEL 8.6, x86_64, Intel(R) Xeon(R) Platinum 8360Y CPU @ 2.40GHz (icelake), 1 x NVIDIA NVIDIA A100-SXM4-80GB, 520.61.05, Python 3.6.8
See https://gist.github.com/branfosj/9e04416d6a2fc59dd271a5a6f3928797 for a full test report.

[Flamefire]

@branfosj Opened #18235 for the CUDA patch