Use AcceleratedKernels.mapreduce in max_scaled_speed and integrate_via_indices

[vchuravy]

Demonstrate how to use AcceleratedKernels.

#2590 (comment)

There are several more places where we would need to do this,
but this is the crux of it.

fixes: #2823

[github-actions]

Review checklist

This checklist is meant to assist creators of PRs (to let them know what reviewers will typically look for) and reviewers (to guide them in a structured review process). Items do not need to be checked explicitly for a PR to be eligible for merging.

Purpose and scope

• The PR has a single goal that is clear from the PR title and/or description.
• All code changes represent a single set of modifications that logically belong together.
• No more than 500 lines of code are changed or there is no obvious way to split the PR into multiple PRs.

Code quality

• The code can be understood easily.
• Newly introduced names for variables etc. are self-descriptive and consistent with existing naming conventions.
• There are no redundancies that can be removed by simple modularization/refactoring.
• There are no leftover debug statements or commented code sections.
• The code adheres to our conventions and style guide, and to the Julia guidelines.

Documentation

• New functions and types are documented with a docstring or top-level comment.
• Relevant publications are referenced in docstrings (see example for formatting).
• Inline comments are used to document longer or unusual code sections.
• Comments describe intent ("why?") and not just functionality ("what?").
• If the PR introduces a significant change or new feature, it is documented in NEWS.md with its PR number.

Testing

• The PR passes all tests.
• New or modified lines of code are covered by tests.
• New or modified tests run in less then 10 seconds.

Performance

• There are no type instabilities or memory allocations in performance-critical parts.
• If the PR intent is to improve performance, before/after time measurements are posted in the PR.

Verification

• The correctness of the code was verified using appropriate tests.
• If new equations/methods are added, a convergence test has been run and the results
  are posted in the PR.

Created with ❤️ by the Trixi.jl community.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.08%. Comparing base (ce719f3) to head (5a13214).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2882      +/-   ##
==========================================
+ Coverage   96.76%   97.08%   +0.32%     
==========================================
  Files         610      610              
  Lines       47500    47515      +15     
==========================================
+ Hits        45960    46128     +168     
+ Misses       1540     1387     -153     

Flag	Coverage Δ
unittests	97.08% <100.00%> (+0.32%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sloede]

Thanks for this PR with a proof of concept. I'd hold off with merging this to #2590 and do this at a later stage, so as not to further delay the merge of #2590 to main.

[ranocha]

Thanks! Do you have some benchmark results showing the impact of this?

[vchuravy]

One of the "annoying" things here is that mapreduce implies a contraction to a scalar; thus, at the end of the computation, we have to move data from the device to the host. There has been some discussion before to use an AsyncNumber type, but that wouldn't help here since we immediately use the values.

[ranocha]

What is the status of this PR? Is it basically ready except for formatting issues and failing tests due to the OrdinaryDiffEqSDIRK problem?

[vchuravy]

Is it basically ready

Yeah, ready from myside.

[ranocha]

Can you please show some benchmarks comparing this to the current version on main?

[vchuravy]

Using the profiler to look at how time is being spent, the previous implementation used a KernelAbstraction kernel (2.244ms) + GPUArrays mapreduce (5us). The AcceleratedKernel implementation is 2.274ms (only one kernel launch)

Using the NVTX ranges from #2908
Before:

│    1.16% │   10.71 ms │     5 │   2.14 ms ± 0.01   (  2.13 ‥ 2.16)    │ Trixi.calculate dt       │

After:

│    1.09% │   10.74 ms │     5 │   2.15 ms ± 0.04   (  2.09 ‥ 2.2)     │ Trixi.calculate dt       │

I think I mistakenly said during our weekly meeting that the previous calulcate_dt was host based, but that is only true for the other code I rewrote as part of this PR.

[ranocha]

So it's within the error bars of having the same performance, right? Is it worth the additional dependency and expected to be better in the log term?

[vchuravy]

Is it worth the additional dependency and expected to be better in the log term?

It allows us to use one consistent code-pattern for reductions, so I would say the additional dependency is worth it.

[ranocha]

Thanks!