chore(benchmark): head-to-head diffusion benchmark against Chmy.jl - #68
Merged
Conversation
Head-to-head against Chmy.jl v0.1.26's published examples/diffusion_2d.jl on an identical problem, isolating the one thing that differs: Chmy materializes a staggered flux field across two kernel launches, MFO fuses the 5-point stencil into a single pass. Single-threaded, MFO leads at every size — 16.9× at the published 126², settling to a ~2.6× asymptote once DRAM-resident (5 array passes per step against Chmy's 7). Threaded, Chmy's KernelAbstractions CPU backend scales while MFO's CPU path is serial, so Chmy takes a middle band. Recorded as measured rather than tuned around; a threaded MFO (#51) plus step fusion (#52) is the answer. Final fields agree to machine epsilon at every size, which is what confirms the two are the same discretization and that the timing gap is design, not setup.
An empty npm lockfile (zero packages) left in the repo root by tooling. There is no package.json here and this is a Julia package, so it is detritus rather than a dependency record — ignore it instead of letting it sit untracked in status.
The [sources] path and the .gitignore results/ rule follow the new depth; everything else resolves through @__DIR__ and moved unchanged.
kylebeggs
changed the title
|
Benchmark ResultsTime
Memory and allocations
Benchmark PlotsA plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
benchmark/comparisons/chmy_diffusion_2d/— a head-to-head CPU benchmark against Chmy.jl v0.1.26's publishedexamples/diffusion_2d.jl, solving an identical problem so the only variable is how each package applies the discrete operator.benchmark/alongside the AirspeedVelocity suite, in its owncomparisons/subtree — cross-package measurement, kept separate from the in-repo regression benchmarks thatbenchpkgruns. ItsProject.tomlis self-contained, sobenchmarks.jlis untouched.results/is gitignored; field dumps reach 2.1 GB each at 16382² and go to a temp dir.package-lock.json(nopackage.jsonin this Julia repo).Nothing under
src/changes — this is measurement infrastructure only.