(feat): Metal.jl (Apple Silicon GPU) backend support#34
Merged
Conversation
There was a problem hiding this comment.
Pull request overview
Adds Metal.jl (Apple Silicon GPU) backend support for AdaptiveArrayPools, mirroring the existing CUDA extension. Also extracts shared GPU pooling algorithms into a common layer (gpu_backend_common.jl) used by both CUDA and Metal extensions, and gates both GPU extensions on Julia 1.11+.
Changes:
- New Metal extension (
ext/AdaptiveArrayPoolsMetalExt/) with types, acquire, state management, safety, display, and convenience functions forMtlArraypooling - Extracted shared GPU backend common layer (
src/gpu_backend_common.jl) with_gpu_claim_slot!,_gpu_get_cached_array!,_gpu_cached_reshape!,_gpu_invalidate_released_slots!, refactoring CUDA to use the same shared code - Added Julia 1.11+ version gate to both CUDA and Metal GPU extensions
Reviewed changes
Copilot reviewed 35 out of 36 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| Project.toml | Add Metal weak dependency and compat |
| src/AdaptiveArrayPools.jl | Include gpu_backend_common.jl, export Metal stubs |
| src/gpu_backend_common.jl | New shared GPU pooling algorithms (hooks + implementations) |
| src/task_local_pool.jl | Add Metal task-local pool function stubs |
| src/utils.jl | Add _show_wrapper_cache, loop over GPU backends in pool_stats() |
| ext/AdaptiveArrayPoolsMetalExt/*.jl | Full Metal extension (types, acquire, state, debug, macros, convenience, utils, task_local_pool) |
| ext/AdaptiveArrayPoolsCUDAExt/*.jl | Refactored to use shared GPU backend hooks, added Julia 1.11 gate |
| test/metal/*.jl | Comprehensive Metal test suite |
| test/runtests.jl, test/cuda/runtests.jl | Version gates and Metal test inclusion |
| docs/ | Metal documentation and navigation updates |
| README.md | Add Metal to supported backends |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
You can also share your feedback on Copilot code review. Take the survey.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #34 +/- ##
=======================================
Coverage 96.53% 96.53%
=======================================
Files 14 14
Lines 2566 2568 +2
=======================================
+ Hits 2477 2479 +2
Misses 89 89
🚀 New features to boost your workflow:
|
Standalone Metal extension with full parity to CUDA: checkpoint/rewind,
lazy mode, typed-lazy mode, safety (poisoning/escape detection/borrow
tracking), @with_pool :metal, DisabledPool{:metal}, and task-local
multi-device pools. No shared GPU common layer — each backend is
self-contained to preserve type stability and zero-alloc guarantees.
CUDA extension imports modern-path-only functions (_store_arr_wrapper!, _reshape_impl!, _check_pool_growth) that don't exist on Julia 1.10 LTS. Wrap all includes in @static if VERSION >= v"1.11-" to prevent precompilation failure on older Julia.
Replace `return`-based early exit (unreliable inside @static if on 1.10) with if-else branching to properly skip GPU tests on Julia LTS.
mgyoo86
force-pushed
the
feat/Metal_support
branch
from
974ec5e to
733a41f
Compare
- Add docs/src/features/metal-support.md with full Metal API reference - Fix cuda-support.md: replace incorrect N-way cache description with actual arr_wrappers-based direct-index caching mechanism - Add Metal backend to README supported backends and doc links - Add Metal page to docs/make.jl navigation and path mappings
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
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
Adds Metal.jl backend support for AdaptiveArrayPools, enabling zero-allocation GPU array pooling on Apple Silicon. Same checkpoint/rewind semantics and
@with_poolmacro support as CUDA.What's new
@with_pool :metal— pool GPU arrays on Apple Silicon with the same API as:cudasrc/gpu_backend_common.jl) — extracted common algorithms (_gpu_claim_slot!,_gpu_get_cached_array!,_gpu_cached_reshape!,_gpu_invalidate_released_slots!) used by both CUDA and Metal extensionsVERSION >= v"1.11-"(requiressetfield!-based Array internals)Usage
Known limitations
MtlArrayinternal fields (:data,:maxsize,:offset,:dims,.data.rc), guarded with@staticchecks