WIP: Use non-atomic scatterview in hist

[pgrete]

PR Summary

PR Checklist

• Code passes cpplint
• New features are documented.
• Adds a test for any bugs fixed. Adds tests for new features.
• Code is formatted
• Changes are summarized in CHANGELOG.md
• Change is breaking (API, behavior, ...)
  • Change is additionally added to CHANGELOG.md in the breaking section
  • PR is marked as breaking
  • Short summary API changes at the top of the PR (plus optionally with an automated update/fix script)
• CI has been triggered on Darwin for performance regression tests.
• Docs build
• (@lanl.gov employees) Update copyright on changed files

[BenWibking]

Are you compiling with -munsafe-fp-atomics? It should be safe to use this flag, since (I think) we don't depend on coherent memory while GPU kernels are running (https://www.exascaleproject.org/wp-content/uploads/2022/02/crusher_tips_and_tricks-final.pdf)

[pgrete]

Oh, good point(er). I'll test that.

[BenWibking]

Did that help?

[pgrete]

Indeed it did, so I'm now wondering if the performance issue we saw with Ascent on Frontier are also related to this.

[BenWibking]

Indeed it did, so I'm now wondering if the performance issue we saw with Ascent on Frontier are also related to this.

Ascent histograms are computed on the host, using unified memory support. I can't think of an obvious place it would otherwise use device atomics. Or are you thinking of a different performance issue?