[FEA] Allow RMM to Handle Using Conda Compilers

[aleksficek]

Is your feature request related to a problem? Please describe.
Currently using conda compilers with RMM results in breakages. RAPIDS is posed to take advantage of anaconda compiler versions of gcc and g++ (https://docs.conda.io/projects/conda-build/en/latest/resources/compiler-tools.html) in the future. This is part of an overarching effort in project scout where developers will be able to one day install the scout conda package which includes gcc, g++ and nvcc conda compilers at specific versions with ccache hooked up for fast compilation.

Describe the solution you'd like
This is similar to the issue and fix as part of #315 but based on my own testing in RMM a similar failure has since resurfaced.

Steps to simulate failure
Conda install conda compilers gcc_linux-64 and gxx_linux-64 to conda environment. Then build the repo and see failures. If this does not simulate the error then the error can be reproduced by first pulling the gpuci/rapidsai:0.15-cuda10.2-devel-ubuntu16.04-py3.7 docker image, cloning the rmm repo, installing conda compilers and then building it.

Additional context
After this is solved, the general plan of attack will be to integrate conda compilers into cuDF and then distribute the requests outwards to each other repo.

[jrhemstad]

What are the breakages?

[aleksficek]

It used to be breaking exactly like #315 but as I just tested it, the cmake command worked but the process is now failing at make -j. There's a chance I'm just setting my environment wrong but in that case, all that's needed is a verification that conda compilers are working in rmm.

rmm error.txt

[aleksficek]

Looks like I used an incorrect version of conda's gcc. I'll offload this debugging to one of the members on ops I did a knowledge transfer with and they'll take it from here.

[github-actions]

This issue has been marked rotten due to no recent activity in the past 90d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[github-actions]

This issue has been marked stale due to no recent activity in the past 30d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be marked rotten if there is no activity in the next 60d.

[harrism]

@kkraus14 @aleksficek @mike-wendt is this still relevant?

[kkraus14]

Yes, still relevant

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[mike-wendt]

The current plan is to adopt conda compilers for the RAPIDS libraries in 0.20 when CUDA 10.X support is dropped

[jakirkham]

Thanks for the update Mike! 😄

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[harrism]

@beckernick I saw you closed the cuDF equivalent of this issue. Is this issue still relevant or can it be closed?

[beckernick]

cuDF can build with the conda compilers, but I don't know if RMM can currently. We may want to leave this open until we verify.

[kkraus14]

I can confirm that building RMM CUDA / C++ tests and building the RMM Cython extensions works as expected with the conda compilers with version 9.3.0.

[jakirkham]

This is also how I've built RMM in the past (though haven't done this recently).

Given the C++ side of RMM is header only, suspect this is a non-issue from that side.

[harrism]

Thanks @kkraus14 and @jakirkham . Closing.