Switch POCL backend to CompilerCaching.jl#728
Merged
Conversation
vchuravy
marked this pull request as ready for review
July 14, 2026 09:05
Contributor
Benchmark ResultsShow table
Benchmark PlotsA plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR. |
Member
|
Rebasing to pick up #731 should fix the buildkite failures |
Adopts the GPUCompiler 2.0 caching API (JuliaGPU/GPUCompiler.jl#794): compilation results are stored via `GPUCompiler.cached_results` on the CodeInstance (Julia's integrated code cache on 1.11+, persisting through precompilation) instead of the legacy per-context `cached_compilation` dictionaries. Follows the same approach as JuliaGPU/OpenCL.jl#431. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Remove the `[sources]` pin to the `tb/compilercaching` branch; the caching API it relied on shipped in the registered GPUCompiler 2.0 (compat is already `"2"`). Adapt `compile_or_lookup` to the released `cached_results` API: it now returns `nothing` until code exists for the job (rather than always materializing an empty results struct), so re-fetch after compiling — which populates Julia's code cache — via `@something`. Matches the final form merged in JuliaGPU/OpenCL.jl#431. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Drop EnzymeCore from weakdeps/extensions so EnzymeExt is not loaded. The Enzyme testset was already commented out and Aqua already ignores Enzyme/EnzymeCore for stale deps/compat checks. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
vchuravy
force-pushed
the
vc/compilercaching
branch
from
July 17, 2026 15:06
7d63e97 to
bff8c5c
Compare
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #728 +/- ##
==========================================
+ Coverage 64.57% 64.76% +0.19%
==========================================
Files 23 23
Lines 2010 2021 +11
==========================================
+ Hits 1298 1309 +11
Misses 712 712 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
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
2 participants
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.
Adopts the new GPUCompiler 2.0 caching API for the POCL backend, x-ref JuliaGPU/GPUCompiler.jl#794, following the same approach as JuliaGPU/OpenCL.jl#431:
device_rngflag) now live in anOpenCLResultsstruct managed byGPUCompiler.cached_results, i.e. attached to theCodeInstancein Julia's integrated code cache on 1.11+ (persisting through precompilation) instead of the legacy per-contextcached_compilationdictionaries.cl.Kernelhandles are session-local and kept in a small per-context linear cache on the results struct; they are not populated while precompiling, so no dangling handles end up in package images.GPUCompiler.reset_runtime()is gone (the runtime library is cached through the same mechanism now).GPUCompiler 2.0 is now registered, so the
[sources]branch pin is dropped and thecompatbound is"2".compile_or_lookupis adapted to the releasedcached_resultsAPI, which returnsnothinguntil code exists for the job and is re-fetched after compiling — matching the final form merged in JuliaGPU/OpenCL.jl#431.🤖 Generated with Claude Code
Closes #729