- Increased BlockHeader size, added blockStart, and improved alignment logic in FreeList allocator.
- Changed _MIN_BLOCK_SIZE to 32 and consolidated to a single implementation.
- Updated allocation and free logic for correct pointer alignment and header management.
- MemoryUtility now throws OutOfMemoryException on allocation failure.
- Optimized GGXMipGenerationBenchmark SPMD output with MaskScatter and minor math/cleanup improvements.
- Cleaned up Program.cs and enabled global/test initialization.
- Bumped assembly version to 1.6.19.
- Change IJobSPMD.Execute to (indices, mask, ctx) signature for all arities, enabling proper vectorized/masked SPMD execution.
- Update all SPMD job wrappers, extension methods, and test jobs to use new interface.
- Add AVX2 gather/masked gather support to MathV.GatherVector2/3/4 and related methods; use [ConstantExpected] byte scale.
- Improve gather/select logic, pointer arithmetic, and overloads for ref/int* index access.
- Refactor GGXMipGenerationBenchmark and jobs for SPMD, with per-mip-level vectorized jobs and improved memory access.
- Clean up code, fix naming, update comments, and bump version to 1.3.6.
- Introduce `CombinedDependenciesJob` for efficient dependency handling and memory management
- Add `ScheduleCustom<T>` for user-defined job execution/free logic
- Refactor `JobInfo` and `JobDataPool<T>` for safer resource management and custom function support
- Improve SPMD extension type constraint formatting
- Update SPMD project content path and increment assembly versions
- Add unit tests for combined dependencies and custom jobs
- Remove `[Timeout]` from tests to prevent spurious failures
- Add TODO for future `WideLane` optimizations
- Replace legacy .sln with .slnx for better solution structure
Refactored thread-local stack allocator in AllocationManager to use ThreadLocalStackPool, removing global stack pointer arrays and locks. In FreeList, replaced fixed-size cache array and maxConcurrencyLevel with a dynamic linked-list system using SharedState and CacheReclaimer for thread cache lifecycle management. Block headers now store cache pointers instead of indices. Updated allocation/free logic and tests accordingly. Bumped assembly version to 3.1.3.
Updated Add/Remove/Enqueue/Push/etc. methods in core unsafe collections to accept parameters as 'scoped in T' for improved performance and safety. Bumped assembly versions in both csproj files.
Removed unmanaged struct requirement from SPMD job wrappers and extension methods, allowing managed types. Updated all wrappers and extension methods to require only the interface constraint. Refactored SPMD test jobs to use safe ref-based Store overloads. Improved README and docs with clearer debug/mimalloc instructions and a better SPMD example. Cleaned up Program.cs by removing obsolete experimental code. Enhanced math precision in GGXMipGenerationBenchmark. Updated T4 template to generate new constraints and APIs.
- Add static AllBitsSet property to ISPMDLane and implement in ScalarLane and WideLane
- Refactor WideLane shuffle table pointers and update usages
- Improve pointer safety and mask handling in CompressStore, Gather, and MaskLoad
- Enhance Sin, Cos, SinCos with fast-math and hardware fallback
- Add Newton-Raphson refinement for reciprocal/sqrt when not fast-math
- Optimize MathV.Vector vector loading (struct init, pointer ops)
- Update project file: version 1.3.4, content packaging, AOT settings
- Minor code cleanup and naming consistency fixes
- Require TLane : unmanaged, ISPMDLane for stricter type safety and direct memory ops
- Refactor GatherVectorN and WideLane<T>.Gather to use Unsafe.SkipInit and direct assignment, removing stackalloc and TLane.Load for better SIMD performance
- Use Vector.Sum in WideLane<T>.ReduceAdd
- Add/improve XML docs for ReduceAdd/ReduceMax/ReduceMin
- Update test project for AOT, AVX2, speed optimization, and disable reflection
- Tweak GGXMipGenerationBenchmark and Program.cs for improved benchmarking and output
- Added MaskLoad, Gather, and reduction methods (ReduceAdd, ReduceMax, ReduceMin) to ISPMDLane<TSelf, TNumber> with XML docs
- Changed SinCos to use out parameters instead of tuple return
- Implemented reductions in ScalarLane and WideLane (loop-based, TODO: SIMD)
- Added GetUnsafePtr to ISPMDLane
- Extended MathV to support Sin, Cos, SinCos, Tan, Asin, Acos, Atan, Atan2 for Vector2/3/4
- Improved WideLane.Sequence to use best vector type
- Updated GGX mip generation for new SinCos signature
- Bumped version to 1.3.2
- Enabled PNG dumping in GGX benchmark
Replaced struct with generic SPMD version for SIMD, added type aliases (commented), optimized RadicalInverse_VdC, and adjusted SampleEquirectangularMap for better performance and code separation.
- Replaced SIMD-based Sin/Cos/SinCos in WideLane with generic polynomial approximations for hardware independence.
- Updated ScalarLane Cast to use CreateTruncating.
- Applied AggressiveOptimization to key GGX methods; improved luma calculation and radical inverse LUT handling.
- Enhanced GGX benchmark setup, cleanup, and timing logic.
- Bumped project version to 1.3.1.
Updated AssemblyVersion in Jobs and LowLevel projects. Cleaned up Jobs csproj by removing unused properties, project reference, and content file inclusion. Deleted the Wrapper<T> disposable struct wrapper from Ptr.cs.
Relaxed generic constraints for job scheduling/execution to allow reference types (removed struct requirement). Updated IJob, IJobParallelFor, and IJobParallel extension methods to support both value and reference types, introducing RunRef for struct-specific overloads. Adjusted JobExecutor and JobScheduler to match new constraints. Bumped assembly version to 3.1.1. Added Value property to Wrapper<T> for ref access and inlined Get(). Changed GGXMipGenerationJob sample count to linear roughness. Removed unused usings in JobInfo.cs.
- Rename ISPMD interfaces to ISPMDLane for clarity
- Add gather and mask load methods to ISPMDLane, implement for ScalarLane and WideLane
- Add GetUnsafePtr() for direct pointer access
- Update MathV and vector types to use new interface and gather methods
- Update SPMD job interfaces and implementations to ISPMDLane
- Improve hash codes, range checks, and safety checks in vector types
- Update codegen templates for new interface/methods
- Refactor SPMD jobs to use gather methods for efficient vectorized access
Major refactor of job system and memory allocator:
- Replaced threadIndex with preferLocal for scheduling
- Switched local queues to SPMCQueue for better performance
- Introduced lock-free JobEdge pool for dependencies
- Removed remainingBatches; use ref counting for completion
- Updated all scheduling APIs and tests to new model
- Optimized FreeList struct sizes and block management
- Added allocation benchmarks
- Disabled OwnershipTransferAnalyzer temporarily
- Bumped assembly versions
Replaces unsafe pointer-based job data with JobDataPool<T> and ConcurrentSlotMap<T> for safer, type-safe management. JobInfo now references job data by (dataID, dataGeneration). JobExecutor and JobScheduler updated to use the new pool-based approach, requiring T : struct. Removed FreeList and pointer logic. WorkerThread now uses reference counting to prevent use-after-free. Updated all scheduling APIs and benchmarks to match new signatures. Improved documentation and inlining. Bumped assembly version to 3.0.0 due to breaking changes.
- Removed IJobScheduler interface; merged logic into JobScheduler
- Changed scheduling APIs to accept multiple dependencies (ReadOnlySpan)
- Moved WaitItem classes to JobScheduler.cs and updated types
- Updated JobExecutionContext to use JobScheduler and added docs
- Renamed AllocationManagerInitOpts to AllocationManagerDesc (required props)
- Added thread-safe TotalAllocatedMemory property to AllocationManager
- Refactored Integer to Data in benchmarks; updated usage
- Updated tests and improved documentation throughout
Major refactor:
- Add job priority tiers and async wait APIs to IJobScheduler
- Implement priority-based job queues and scheduling logic
- Introduce UnsafeParallelHashMap and refactor UnsafeParallelQueue
- Refactor UnsafeSlotMap to chunked storage for scalability
- Update SlotMap/ConcurrentSlotMap for consistency and perf
- Add new benchmarks and unit tests for parallel collections
- Misc: add MemoryUtility.AlignUp, version bumps, test improvements, bug fixes
- Refactored JobExecutionContext to use init-only properties and added State and SelfHandle for richer context.
- Updated JobInfo field layout for clarity and memory alignment.
- JobScheduler now accepts and exposes an optional state object, passed to each job context.
- Improved memory management by using _freeList for allocations and disposal.
- WorkerThread and benchmarks updated to use new JobExecutionContext pattern.
- Bumped version to 1.6.1 and performed minor code cleanup.
The `AllocationHandle` struct has been updated to replace
`public static readonly` fields with `public static` properties
using lambda expressions. The `Temp`, `FreeList`, and `Persistent`
fields are now properties, dynamically retrieving values from
their respective allocators. This change improves flexibility
and ensures the properties reflect the current state of the
underlying allocators.
Updated `JobScheduler` to enhance dependency tracking by counting valid dependencies upfront and dynamically adjusting counts using `Interlocked` operations. Improved job enqueueing logic to ensure jobs are only enqueued when all dependencies are met.
Replaced `Interlocked.Increment` with `Interlocked.Add` for batch updates to `_totalJobCount`, improving performance. Adjusted `VirtualStack` cleanup to use the correct size variable for memory deallocation.
Simplified `JobDispatchingJob` API by removing `ctx.ThreadIndex` parameter. Updated `TestJobSystem` to pass job handles as dependencies for proper execution order.
Incremented assembly version to 1.5.9 to reflect these changes.
