Add nvbench environment class for initializing RMM in benchmarks

[davidwendt]

Description

Adds an environment class to initialize the RMM pool memory resource manager before running benchmarks through nvbench. This removes the need to initialize RMM on every benchmark call which improves benchmark run performance and fixes some GPU metrics gathering when run under nsys.

This requires a patch to the nvbench source to enable this feature and is part of the following pull request: NVIDIA/nvbench#123
The patch can be removed once the PR is merged and source available to the libcudf build.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-23.04@0cab19a). Click here to learn what that means.
Patch has no changes to coverable lines.

❗ Current head 3f47f2d differs from pull request most recent head b0335e3. Consider uploading reports for the commit b0335e3 to get more accurate results

Additional details and impacted files

@@               Coverage Diff               @@
##             branch-23.04   #12728   +/-   ##
===============================================
  Coverage                ?   85.85%           
===============================================
  Files                   ?      158           
  Lines                   ?    25204           
  Branches                ?        0           
===============================================
  Hits                    ?    21638           
  Misses                  ?     3566           
  Partials                ?        0           

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[vyasr]

@davidwendt did you want to merge this PR and revert the patch once the nvbench PR is merged, or wait for the nvbench PR? I'm fine either way; I know that the dev cycle on nvbench PRs is slower than ours.

[davidwendt]

@davidwendt did you want to merge this PR and revert the patch once the nvbench PR is merged, or wait for the nvbench PR? I'm fine either way; I know that the dev cycle on nvbench PRs is slower than ours.

I did not want to wait for the nvbench PR to be merged first.

[ttnghia]

So does this mean this PR depends on NVIDIA/nvbench#123? Look like that PR is ready to merge.

[bdice]

Approved. Let’s try to note a “fixed in” version if we know what we might expect.

[robertmaynard]

CMake changes lgtm

[robertmaynard]

CMake changes are now great 👍

[karthikeyann]

LGTM 👍

[davidwendt]

/merge