Tags: astrozot/Metal.jl
Tags
[Diff since v1.5.1](JuliaGPU/Metal.jl@v1.5.1...v1.6.0) - Metal, MPS, and MPSGraph frameworks’ enums and objects are now automatically wrapped with Clang.jl - Initial MPSGraph support. Currently used to replace MPS matrix multiplication on configurations where the previous method could fail (JuliaGPU#381) - Many more improvements and bug fixes **Merged pull requests:** - Add error checking to command buffer completion handler (JuliaGPU#521) (@vovw) - Enable `add_functions!` test under shader validation (JuliaGPU#522) (@christiangnrd) - Update wrapping readme (JuliaGPU#525) (@christiangnrd) - Automatically wrap Metal and MPS headers (feat. Properties) (JuliaGPU#526) (@christiangnrd) - Objective-C Availability support (JuliaGPU#527) (@christiangnrd) - Small followup to JuliaGPU#526 (JuliaGPU#528) (@christiangnrd) - Add `nextafter` intrinsic (JuliaGPU#529) (@christiangnrd) - Improvements to float intrinsics (JuliaGPU#531) (@christiangnrd) - Fix Float16 sincos intrinsic (JuliaGPU#533) (@christiangnrd) - Fix format suggestion formatting when diff contains "```" (JuliaGPU#534) (@christiangnrd) - Fix rewriter for version-gated expressions (JuliaGPU#535) (@christiangnrd) - Move BFloat16 code out of extension (JuliaGPU#536) (@christiangnrd) - Version-related fixes to MPSNDArray (JuliaGPU#537) (@christiangnrd) - Use GPUToolbox.jl (JuliaGPU#538) (@christiangnrd) - [tests] Assume compatible system if xcode not installed (JuliaGPU#539) (@christiangnrd) - Add .git-blame-ignore-revs and a few other fixes (JuliaGPU#541) (@christiangnrd) - Integer & atomic Intrinsics improvements (JuliaGPU#544) (@christiangnrd) - [NFC] Indentation consistency (JuliaGPU#545) (@christiangnrd) - Update .git-blame-ignore-revs (JuliaGPU#546) (@christiangnrd) - Link to issue (JuliaGPU#548) (@christiangnrd) - Test both simd shuffle intrinsics. (JuliaGPU#553) (@christiangnrd) - Enabling previously disabled test and mark broken (JuliaGPU#554) (@christiangnrd) - Support pow with Int exponent (JuliaGPU#557) (@christiangnrd) - Fixes and more tests for `unsafe_wrap` (JuliaGPU#558) (@christiangnrd) - `Pi` and `e` to `Float32` and `Float16` (JuliaGPU#559) (@christiangnrd) - Update GPUToolbox compat (JuliaGPU#560) (@christiangnrd) - Split up intrinsics tests (JuliaGPU#561) (@christiangnrd) - Remove reference to 'dir' in profiling docs (JuliaGPU#563) (@christiangnrd) - Ensure synchronization before unsafe_wrapping of shared gpu array (JuliaGPU#564) (@christiangnrd) - [NFC] Move linear algebra wrappers out of `MPS` lib (JuliaGPU#565) (@christiangnrd) - Initial MPSGraph support (JuliaGPU#566) (@christiangnrd) - Remove copy when possible from cpu rnad using GPU RNG (JuliaGPU#568) (@christiangnrd) - Code coverage and misc fixes (JuliaGPU#569) (@christiangnrd) - Remove obsolete `MtlLargerDeviceArray` (JuliaGPU#574) (@christiangnrd) - Silence warning on 1.12+ (JuliaGPU#575) (@christiangnrd) - 15.4 SDK changes (JuliaGPU#579) (@christiangnrd) - Faster matmul sometimes (JuliaGPU#580) (@christiangnrd) - Fix `erf` and a few other improvements (JuliaGPU#582) (@christiangnrd) - Use an appropriate amount of threads in unified memory example (JuliaGPU#583) (@christiangnrd) - Clean up test imports (JuliaGPU#584) (@christiangnrd) - Fix some type ambiguities (JuliaGPU#585) (@christiangnrd) - Fix findall output type (JuliaGPU#587) (@christiangnrd) - Fix tests for macOS 13 (JuliaGPU#591) (@christiangnrd) - Minor `findall` and `accumulate` tests improvements (JuliaGPU#592) (@christiangnrd) - Bump version (JuliaGPU#595) (@christiangnrd) **Closed issues:** - API Validation failures (JuliaGPU#467) - Handle MTLCommandBuffer Error Logs (JuliaGPU#510) - `sincos` intrinsic fails to compile with Float16 (JuliaGPU#530) - Can't compare `Float32` with `pi` on Metal (JuliaGPU#551) - `^(::Float32, ::Integer)` uses double precision (JuliaGPU#552) - Code coverage broken (JuliaGPU#556) - Remove `MtlLargerDeviceArray`? (JuliaGPU#573) - Installation of Metal does not work when DiffEqGPU is also installed. (JuliaGPU#577) - `findall` output always uses default storage mode instead of matching input storage mode (JuliaGPU#578) - Qualify definitions of BroadcastStyle (JuliaGPU#586) - New Dynamic Dispatch when using ColorTypes (JuliaGPU#588) - Release new version of Metal.jl (JuliaGPU#593)
[Diff since v1.5.0](JuliaGPU/Metal.jl@v1.5.0...v1.5.1) **Merged pull requests:** - Adapt to `minver` in ObjectiveC.jl (JuliaGPU#513) (@christiangnrd) - Add Runic action to suggest formatting changes. (JuliaGPU#517) (@maleadt) - Increase polling interval for benchmark action (JuliaGPU#518) (@christiangnrd) - Adapt to GPUArrays.jl changes. (JuliaGPU#519) (@maleadt) **Closed issues:** - Benchmark CI failures due to too many requests (JuliaGPU#516)
[Diff since v1.4.2](JuliaGPU/Metal.jl@v1.4.2...v1.5.0) Metal.jl 1.5 is a relatively minor release, which the most important change being behind the scenes: [GPUArrays.jl v11 has switched to KernelAbstractions.jl](https://juliagpu.org/post/2025-01-07-gpuarrays-11/) (JuliaGPU#461). There is also one (technically) breaking change: `code_agx` and `@device_code_agx` have been removed (JuliaGPU#512) because of the heavy Python dependency, and conflicts with PythonCall.jl. This functionality did not support recent M GPUs anyway, so it is unlikely to affect many users. - Improve performance of shared storage copies: JuliaGPU#445 - Add an `is_m4` function: JuliaGPU#498 - JuliaGPU#499 - Fix `fill`: JuliaGPU#496 **Merged pull requests:** - Add more tests to api validation testing (JuliaGPU#447) (@christiangnrd) - Adapt to GPUArrays.jl transition to KernelAbstractions.jl. (JuliaGPU#461) (@maleadt) - Switch CI to 1.11. (JuliaGPU#462) (@maleadt) - Remove old code and test cleanup (JuliaGPU#464) (@christiangnrd) - Adapt to JuliaGPU/GPUArrays.jl#567. (JuliaGPU#475) (@maleadt) - Bump LLVM downgrader (JuliaGPU#479) (@maleadt) - Store more debug files when encountering compilation errors. (JuliaGPU#482) (@maleadt) - Use `OncePerProcess` in 1.12+ (JuliaGPU#483) (@christiangnrd) - Don't run benchmarks from fork (JuliaGPU#485) (@christiangnrd) - Still run GH Action when merged (JuliaGPU#486) (@christiangnrd) - Bump IR downgrader (JuliaGPU#489) (@maleadt) - Move MTL tests and add a few (JuliaGPU#491) (@christiangnrd) - Generate MTL and MPS structs and enums with Clang.jl (JuliaGPU#492) (@christiangnrd) - Fix copy tests (JuliaGPU#493) (@christiangnrd) - Simplify benchmark runner and pipelines (JuliaGPU#494) (@maleadt) - Fix global linear indexing (`fill!`) (JuliaGPU#496) (@christiangnrd) - Couple typos and `is_m4` function (JuliaGPU#498) (@christiangnrd) - Initial support for MPSNDArray (JuliaGPU#499) (@christiangnrd) - Tweak benchmark CI job (JuliaGPU#501) (@maleadt) - Fix `MPSNDArrayDescriptor` wrapper (JuliaGPU#502) (@christiangnrd) - Metal library parsing: using CodecBzip2 feature to ignore padding. (JuliaGPU#504) (@maleadt) - Followup to JuliaGPU#492: Enable C function wrapping (JuliaGPU#505) (@christiangnrd) - Rerun random tests with chance of false negative once. (JuliaGPU#506) (@christiangnrd) - Bump LLVM downgrader (JuliaGPU#507) (@maleadt) - Test loading of package on unsupported platforms (JuliaGPU#509) (@christiangnrd) - Remove `device_code_agx` (JuliaGPU#512) (@christiangnrd) - Fix typo in random tests (JuliaGPU#514) (@christiangnrd) - Fix Documenter failures (JuliaGPU#515) (@christiangnrd) **Closed issues:** - KernelAbstractions: add Atomix back-end (JuliaGPU#218) - `@device_code_agx` errors when Metal Shader Validation is enabled (JuliaGPU#463) - `fill` broken after KA integration (JuliaGPU#466) - Compilation to native code failed: NSError: Undefined symbols (JuliaGPU#480) - `ObjectiveC.Foundation.NSErrorInstance(ObjectiveC.id{ObjectiveC.Foundation.NSError}(0x000000014cb8bd90))` (JuliaGPU#487) - phi-related IR downgrade issue (JuliaGPU#488) - Circular dependency when precompiling (JuliaGPU#495) - Bad interaction between PyCall and Metal (JuliaGPU#500) - Add github actions CI for linux, windows and non-functional macOS to ensure that precompilation and loading works (JuliaGPU#508)
[Diff since v1.4.1](JuliaGPU/Metal.jl@v1.4.1...v1.4.2) **Merged pull requests:** - Fix loading on unsupported platforms (JuliaGPU#459) (@christiangnrd) **Closed issues:** - Relax package requirements (JuliaGPU#22) - [windows:] Metal does not precompile anymore when installation not functional (JuliaGPU#457) - [MacOS:] `Metal.functional()` wrongly returns true despite no GPUs available (JuliaGPU#458)
[Diff since v1.4.0](JuliaGPU/Metal.jl@v1.4.0...v1.4.1) **Merged pull requests:** - Update Readme (JuliaGPU#444) (@christiangnrd) - Use CPU copy with SharedStorage (JuliaGPU#445) (@christiangnrd) - Disable nightly CI and fix invalid Metal API usage (JuliaGPU#448) (@christiangnrd) - Don't report benchmarks on main branch commits (JuliaGPU#450) (@christiangnrd) - Fix JuliaGPU#451 and a couple other fixes (JuliaGPU#452) (@christiangnrd) - Only load BFloat16s extension on Apple systems (JuliaGPU#454) (@christiangnrd) - CompatHelper: bump compat for GPUCompiler to 1, (keep existing compat) (JuliaGPU#455) (@github-actions[bot]) **Closed issues:** - Don't run benchmarks on the master branch? (JuliaGPU#449) - `unsafe_wrap(Array, ...)` of a view does not preserve offset information (JuliaGPU#451) - Metal does not load any more without error when installation not functional (JuliaGPU#453)
[Diff since v1.3.0](JuliaGPU/Metal.jl@v1.3.0...v1.4.0) **Merged pull requests:** - Use unified memory for scalar indexing of permutation matrices (JuliaGPU#313) (@tgymnich) - Add `MPSMatrixRandom` (JuliaGPU#321) (@christiangnrd) - [.gitignore] Also ignore versioned Manifests (JuliaGPU#410) (@christiangnrd) - Remove broken link in Docs (JuliaGPU#413) (@christiangnrd) - Remove unused [extras] section in Project.toml (JuliaGPU#415) (@christiangnrd) - Small fix and typos (JuliaGPU#417) (@christiangnrd) - Add Benchmarking CI (JuliaGPU#420) (@christiangnrd) - [NFC] Fix warning in `topk` docstrings (JuliaGPU#421) (@christiangnrd) - Allow initialisation of MTLSize with tuples of different integer types (JuliaGPU#425) (@tgymnich) - Add CI for macOS 15 (JuliaGPU#426) (@christiangnrd) - Simplify versioninfo() and report more packages. (JuliaGPU#429) (@maleadt) - Allow controlling compilation target versions. (JuliaGPU#430) (@maleadt) - Add a missing memory fence to a SIMD test. (JuliaGPU#432) (@maleadt) - Fix `MPS.synchronize_state` (JuliaGPU#434) (@christiangnrd) - Make `lu` results have same storage mode as input (JuliaGPU#435) (@christiangnrd) - Fix benchmarking CI and benchmark Shared and Private storage modes (JuliaGPU#437) (@christiangnrd) - NFC tweak to MPSMatrixCopy tests (JuliaGPU#439) (@christiangnrd) - Get more descriptive errors from flaky test (JuliaGPU#440) (@christiangnrd) **Closed issues:** - Port the opportunistic synchronization from CUDA.jl (JuliaGPU#317) - Control flow-related miscompilation: (JuliaGPU#401) - More sporadic 1.11 hangs (JuliaGPU#412) - Support for `LinearAlgebra.kron` (JuliaGPU#422) - Can't use gemm! methods with Metal (JuliaGPU#423) - Error for thread/group size with different integer types (JuliaGPU#424) - README example broken (JuliaGPU#427) - Intermittent load_store_tg test failure (JuliaGPU#428)
[Diff since v1.2.0](JuliaGPU/Metal.jl@v1.2.0...v1.3.0) **Merged pull requests:** - Fix typo in docs (JuliaGPU#384) (@christiangnrd) - Bump minimal Julia requirement to v1.10. (JuliaGPU#385) (@maleadt) - Remove Requires dependency (JuliaGPU#386) (@christiangnrd) - Reflection: Figure out kernel names by looking at metallib section. (JuliaGPU#390) (@maleadt) - Add tests for broadcasting minimum and maximum (JuliaGPU#391) (@tgymnich) - Don't export `MTL` (JuliaGPU#392) (@christiangnrd) - Add erfinv (JuliaGPU#394) (@tgymnich) - Add expm1 (JuliaGPU#395) (@tgymnich) - Cleanup some imports (JuliaGPU#398) (@christiangnrd) - Remove type-pirated function (JuliaGPU#399) (@christiangnrd) - Unexport some high-level MPS functionality from `MPS` (JuliaGPU#400) (@christiangnrd) - Adapt to new REPL precompile changes (JuliaLang/julia#55210) (JuliaGPU#403) (@christiangnrd) - Bump GPUCompiler. (JuliaGPU#404) (@maleadt) - Bump LLVM compat (JuliaGPU#407) (@maleadt) - Make 1.11 CI success mandatory. (JuliaGPU#408) (@maleadt) **Closed issues:** - Audit exports/public symbols (JuliaGPU#359) - Compilation failure on 1.11 (JuliaGPU#370) - `MTLBinaryArchive` (JuliaGPU#387) - `Metal.code_agx()` failing in MacOS 15 Beta 3 (JuliaGPU#388) - Test for min / max broadcasting issue (JuliaGPU#389) - Type piracy (JuliaGPU#396) - Potentially unused code in gpuarrays.jl (JuliaGPU#397) - `Shared` vs `SharedStorage` in examples/unified_memory (JuliaGPU#405) - Unsuported call to an unknown function when calling `Distributions` (JuliaGPU#406)
## Metal v1.2.0 [Diff since v1.1.0](JuliaGPU/Metal.jl@v1.1.0...v1.2.0) **Merged pull requests:** - Avoid constructing `MulAddMul`s on Julia v1.12+ (JuliaGPU#295) (@dkarrasch) - Trigger the runtime profiler when a test times out. (JuliaGPU#330) (@maleadt) - Add MPSMatrixSoftMax (JuliaGPU#333) (@christiangnrd) - Reorganize and add some MPS tests (JuliaGPU#335) (@christiangnrd) - Typo fix (JuliaGPU#336) (JuliaGPU#337) (@101001000) - Add error message for running Metal.jl under Rosetta (JuliaGPU#339) (@tgymnich) - Add `MPSCommandBuffer` (JuliaGPU#340) (@christiangnrd) - Bump julia-actions/setup-julia from 1 to 2 (JuliaGPU#341) (@dependabot[bot]) - Revert error message for Rosetta (JuliaGPU#342) (@tgymnich) - Update to ObjectiveC.jl v3. (JuliaGPU#343) (@maleadt) - Add autoreleasepools to MPS interface methods. (JuliaGPU#344) (@maleadt) - Don't redundantly return the cmdbuf from commit methods. (JuliaGPU#345) (@maleadt) - Whitespace fixes (JuliaGPU#346) (@christiangnrd) - CompatHelper: bump compat for LLVM to 7, (keep existing compat) (JuliaGPU#347) (@github-actions[bot]) - CompatHelper: add new compat entry for SpecialFunctions in [weakdeps] at version 2, (keep existing compat) (JuliaGPU#352) (@github-actions[bot]) - [NFC] Fix indentation (JuliaGPU#353) (@christiangnrd) - Bump LLVM downgrader (JuliaGPU#354) (@maleadt) - Don't export non-existent `contents` (JuliaGPU#356) (@christiangnrd) - Remove/fix unused exports (JuliaGPU#357) (@christiangnrd) - Unexport `SimpleVersion` and `AS` (JuliaGPU#360) (@christiangnrd) - Add support for opaque pointers (JuliaGPU#361) (@maleadt) - Docstrings (JuliaGPU#362) (@christiangnrd) - Initial MacOS 15 support (JuliaGPU#365) (@christiangnrd) - Replace `current_device()` with `device()` (JuliaGPU#366) (@christiangnrd) - Support reading metallib v1.2.8 files from macOS 15. (JuliaGPU#367) (@maleadt) - Add metallib (dis)assembly helper scripts. (JuliaGPU#368) (@maleadt) - Simplify testing of examples. (JuliaGPU#369) (@maleadt) - Temporarily allow 1.11 to fail. (JuliaGPU#371) (@maleadt) - CompatHelper: add new compat entry for PrecompileTools at version 1, (keep existing compat) (JuliaGPU#372) (@github-actions[bot]) - Define complex sqrt (JuliaGPU#374) (@mtfishman) - Check the macOS version during initialization. (JuliaGPU#375) (@maleadt) - CompatHelper: bump compat for LLVM to 8, (keep existing compat) (JuliaGPU#376) (@github-actions[bot]) - Add `accumulate` implementation (JuliaGPU#377) (@chengchingwen) - fix derived device array (JuliaGPU#378) (@chengchingwen) - avoid ReshapedArray using Int128 in metal kernel (JuliaGPU#379) (@chengchingwen) - improve type stability of derived array (JuliaGPU#380) (@chengchingwen) - add `findall` implementation (JuliaGPU#382) (@zhenwu0728) - Bump version (JuliaGPU#383) (@christiangnrd) **Closed issues:** - Tests sporadically timing out on 1.11 (JuliaGPU#329) - ReshapedArray indexing broken because of Int128 operation (JuliaGPU#332) - KernelAbstractions copyto! typo (JuliaGPU#336) - Segmentation Faults (JuliaGPU#338) - Port `accmulate!` and `findall` from CUDA.jl (JuliaGPU#348) - Tests failing with `GPUCompiler` v0.26.5 and `LLVM` v7.1 (JuliaGPU#350) - downgrades LLVM (JuliaGPU#355) - sqrt(::Complex) unsupported due to conversion exceptions (JuliaGPU#364)
## Metal v1.1.0 [Diff since v1.0.0](JuliaGPU/Metal.jl@v1.0.0...v1.1.0) **Merged pull requests:** - Add `resize!` (JuliaGPU#279) (@mtfishman) - Initial MTLTexture support (JuliaGPU#280) (@christiangnrd) - Avoid redundant pointer conversions for threadgroup memory. (JuliaGPU#283) (@maleadt) - Re-implement metallib generation in Julia. (JuliaGPU#284) (@maleadt) - CompatHelper: add new compat entry for SHA at version 0.7, (keep existing compat) (JuliaGPU#286) (@github-actions[bot]) - Support more of the metallib format (JuliaGPU#288) (@maleadt) - Address potentiallly buggy `mtl` behaviour. (JuliaGPU#290) (@christiangnrd) - CompatHelper: add new compat entry for CodecBzip2 at version 0.8, (keep existing compat) (JuliaGPU#292) (@github-actions[bot]) - Remove an unneeded pointer method. (JuliaGPU#293) (@maleadt) - Use NSAutoreleasePool to clean up memory. (JuliaGPU#294) (@maleadt) - `adapt_storage`-related improvements (JuliaGPU#296) (@christiangnrd) - CompatHelper: bump compat for ObjectiveC to 2, (keep existing compat) (JuliaGPU#297) (@github-actions[bot]) - Add support for signposts (JuliaGPU#300) (@maleadt) - Retain NSError we rethrow to avoid an UAF. (JuliaGPU#302) (@maleadt) - Minor mapreduce improvements (JuliaGPU#303) (@maleadt) - Specialize broadcast to avoid integer divisions. (JuliaGPU#304) (@maleadt) - Better Support for Unified Memory (JuliaGPU#305) (@tgymnich) - Add 1.11 CI (JuliaGPU#306) (@christiangnrd) - Remove unused files (JuliaGPU#307) (@tgymnich) - Skip profiling tests on macOS 14.4/M1. (JuliaGPU#310) (@maleadt) - Increase test timeout limit to accomodate 1.8 (JuliaGPU#311) (@christiangnrd) - Test all storage modes (JuliaGPU#314) (@christiangnrd) - Fix doctests (JuliaGPU#315) (@christiangnrd) - Fix KernelAbstractions for Unified Memory (JuliaGPU#316) (@tgymnich) - CompatHelper: add new compat entry for Preferences at version 1, (keep existing compat) (JuliaGPU#318) (@github-actions[bot]) - Minor cleanup (JuliaGPU#319) (@christiangnrd) - Create MtlArray using memory allocated by Array (JuliaGPU#320) (@christiangnrd) - Re-enable profiling tests on M1/14.4 when using Xcode 15.3. (JuliaGPU#322) (@maleadt) - Small typo and doc fixup (JuliaGPU#325) (@christiangnrd) - BFloat16s.jl extension and related improvements (JuliaGPU#326) (@christiangnrd) - Support for Julia 1.11 (JuliaGPU#327) (@maleadt) **Closed issues:** - Validation-related back-end crash on macOS Ventura (JuliaGPU#34) - slow broadcast copy in 2D (JuliaGPU#41) - Poor performance of mapreduce (JuliaGPU#46) - Multiplication with SubArrays (JuliaGPU#47) - Add support to creating MtlArray using a memory allocated by Array (JuliaGPU#62) - Improve use of unified memory (JuliaGPU#86) - Use Autoreleasepools with Metal (JuliaGPU#103) - Unknown RFLT tag generated by macOS 13 Metal compiler (JuliaGPU#167) - mapreduce allocates a lot on the CPU (JuliaGPU#211) - Legalization errors with vectorized code (JuliaGPU#257) - Compilation Failure due to undefined symbols (JuliaGPU#276) - `resize!`, `append!` not defined (JuliaGPU#277) - tag new version (JuliaGPU#278) - Panic during profiling tests on 14.4 beta (JuliaGPU#281) - M3 backend cannot handle atomics with complicated pointer conversions (JuliaGPU#282) - Int128 does not compile (JuliaGPU#287) - Two suspicious `mtl`-related behaviours (JuliaGPU#289) - LU factorization: add allowsingular keyword argument (JuliaGPU#299) - Autorelease changes lead to use after free with errors (JuliaGPU#301) - Reductions don't work on Shared Arrays (JuliaGPU#312)
## Metal v1.0.0 [Diff since v0.5.1](JuliaGPU/Metal.jl@v0.5.1...v1.0.0) **Merged pull requests:** - Matrix batches (JuliaGPU#158) (@tgymnich) - Add 1.10 CI. (JuliaGPU#256) (@maleadt) - Update manifest (JuliaGPU#258) (@github-actions[bot]) - CompatHelper: bump compat for GPUCompiler to 0.25, (keep existing compat) (JuliaGPU#259) (@github-actions[bot]) - Bump actions/checkout from 3 to 4 (JuliaGPU#260) (@dependabot[bot]) - Update manifest (JuliaGPU#261) (@github-actions[bot]) - CompatHelper: bump compat for CEnum to 0.5, (keep existing compat) (JuliaGPU#262) (@github-actions[bot]) - Update manifest (JuliaGPU#263) (@github-actions[bot]) - CompatHelper: add new compat entry for Artifacts at version 1, (keep existing compat) (JuliaGPU#264) (@github-actions[bot]) - Reduce launch overhead by generating code to encode arguments. (JuliaGPU#265) (@maleadt) - Remove unused function argument (JuliaGPU#266) (@tgymnich) - Introduce application tracing profiler (JuliaGPU#267) (@maleadt) - Remove content(::MTLBuffer), use convert intead. (JuliaGPU#268) (@maleadt) - Allow more kwargs syntax with kernel launches (JuliaGPU#269) (@maleadt) - Don't re-use the IO object when shelling out to Python. (JuliaGPU#271) (@maleadt) - Preserve storage mode when broadcasting. (JuliaGPU#273) (@maleadt) **Closed issues:** - Support for macOS Sonoma (JuliaGPU#201) - Error with Julia 1.10 (JuliaGPU#274)
PreviousNext