v0.9.4

KernelAbstractions v0.9.4

Diff since v0.9.3

Merged pull requests:

• Add CPU(static=true) (#388) (@vchuravy)
• Update index.md (#390) (@Ruibin-Liu)

v0.9.3

KernelAbstractions v0.9.3

Diff since v0.9.2

Merged pull requests:

• Migrate from SnoopPrecompile to PrecompileTools (#386) (@timholy)

v0.9.2

KernelAbstractions v0.9.2

Diff since v0.9.1

Closed issues:

• Use occupancy API for autotuning (#19)
• Allow user to turn off contract (#20)
• Assigning ::ROCDevice to ::KA.GPU (#321)
• ROCKernels: using queue pool causes performance regression (#344)
• KernelAbstractions.jl is blocked to v0.8.6 by CUDAKernels (#380)

Merged pull requests:

• CI improvements (#378) (@maleadt)
• Add unsafe_free! (#381) (@pxl-th)

v0.9.1

KernelAbstractions v0.9.1

Diff since v0.9.0

Closed issues:

• Can't run the example in quickstart (#371)

Merged pull requests:

• Add Metal to list of excluded backends (#368) (@maxwindiff)
• Add queries for atomics and float64 support (#369) (@maxwindiff)
• Fix typos (#370) (@tomchor)
• Add reverse CI for Metal PR (#372) (@vchuravy)
• Update reverse CI for CUDA (#373) (@vchuravy)
• Make unit tests skippable (#374) (@maxwindiff)
• Update CUDA to master (#375) (@vchuravy)

v0.9.0

KernelAbstractions v0.9.0

Diff since v0.8.6

Closed issues:

• No speedup on CPU (#322)
• Add Metal support (#326)

Merged pull requests:

• Start removing event system (#317) (@vchuravy)
• Add Metal support (#337) (@tgymnich)
• Prefer blocks over threads (#341) (@vchuravy)
• ROCKernels: Add occupancy API (#342) (@pxl-th)
• [CUDAKernels] add always_inline as device parameter (#343) (@vchuravy)
• [CUDAKernels] Update compat (#345) (@vchuravy)
• Update CI (#346) (@vchuravy)
• ROCKernels: Adapt to AMDGPU changes (#348) (@jpsamaroo)
• [ROCKernels] Fix addrspacecast (#349) (@vchuravy)
• [ROCKernels] Import LLVM (#352) (@pxl-th)
• Update compat for oneAPIKernels.jl (#355) (@utkarsh530)
• Bump oneAPI to 1.0 (#356) (@michel2323)
• Rename device to backend (#359) (@vchuravy)
• Let Event(MtlDevice) actually be a barrier (#360) (@vchuravy)
• Fix Metal workgroup size (#361) (@tgymnich)
• Update docs (#362) (@vchuravy)
• Add optional priority feature (#363) (@vchuravy)
• Backends are adaptors (#364) (@vchuravy)
• Only skip histogram tests on CPU (#365) (@vchuravy)

v0.8.6

KernelAbstractions v0.8.6

Diff since v0.8.5

Closed issues:

• Support for single-threaded kernels even when Threads.nthreads() != 1? (#328)
• Render issue with Docs admonition? (#332)

Merged pull requests:

• Minor fixes in the Docs (#334) (@navidcy)
• Fix quickstart (#335) (@vchuravy)
• setting ndrange macro to output a Tuple (#336) (@leios)

v0.8.5

KernelAbstractions v0.8.5

Diff since v0.8.4

Closed issues:

• Add backend lookup function based on input arguments (#229)
• Update for CUDA.jl 3.0 (#241)

Merged pull requests:

• Fix docstrings rendering (#323) (@navidcy)
• fix typo in docs (#327) (@bjarthur)
• Add @ndrange (#329) (@vchuravy)
• Fix stmt form of at_private (#331) (@vchuravy)

v0.7.3

KernelAbstractions v0.7.3

Diff since v0.7.2

Closed issues:

• Support atomics (#7)
• Add backend lookup function based on input arguments (#229)
• Separate Cassette context from CompilerMetadata (#231)
• Update for CUDA.jl 3.0 (#241)
• Adding a function to get device from array (type)? (#268)
• Support for atomics (#276)
• CUDA 3.6.3 broke KernelAbstactions. (#280)
• Enzyme fails on GPU kernel (#307)

Merged pull requests:

• Add function get_device (#269) (@oschulz)
• Synchronize SpecialFunctions compat (#279) (@charleskawczynski)
• [CUDAKernels] Avoid Cassette looking at device_functions (#281) (@vchuravy)
• Simplify CI and drop bors (#283) (@vchuravy)
• Excise Cassette (#288) (@vchuravy)
• Use Timer to avoid hangs due to error on the device (#291) (@vchuravy)
• don't test KernelGradients on nightly for now (#292) (@vchuravy)
• WIP: Make CUDA stream from cache match CUDA context (#294) (@christophernhill)
• adding atomic support with atomix (#299) (@leios)
• using CPU as default for Scratchpad and SharedMemory (#300) (@leios)
• CompatHelper: add new compat entry for Atomix at version 0.1, (keep existing compat) (#301) (@github-actions[bot])
• Enzyme v0.10 supports GPU compilation again (#303) (@vchuravy)
• Add atomics support for ROCKernels (#304) (@jpsamaroo)
• Enable CPU atomics (#305) (@jpsamaroo)
• Add 'return nothing' to autodiff (#309) (@pxl-th)
• Make examples work by copy-paste into REPL (#310) (@jwscook)
• bounding UnsafeAtomics and UnsafeAtomicsLLVM (#311) (@leios)
• Add oneAPI back-end. (#312) (@maleadt)
• [doc] Some revamping (#314) (@michel2323)
• Fix depwarn (#315) (@ChrisRackauckas)
• ROCKernels: Update to AMDGPU 0.4 (#316) (@jpsamaroo)
• Fix docstrings rendering (#323) (@navidcy)
• fix typo in docs (#327) (@bjarthur)
• Add @ndrange (#329) (@vchuravy)
• Fix stmt form of at_private (#331) (@vchuravy)

v0.8.4

KernelAbstractions v0.8.4

Diff since v0.8.3

Merged pull requests:

• Make examples work by copy-paste into REPL (#310) (@jwscook)
• Add oneAPI back-end. (#312) (@maleadt)
• [doc] Some revamping (#314) (@michel2323)
• Fix depwarn (#315) (@ChrisRackauckas)
• ROCKernels: Update to AMDGPU 0.4 (#316) (@jpsamaroo)

v0.8.3

KernelAbstractions v0.8.3

Diff since v0.8.2

Closed issues:

• Enzyme fails on GPU kernel (#307)

Merged pull requests:

• Add 'return nothing' to autodiff (#309) (@pxl-th)
• bounding UnsafeAtomics and UnsafeAtomicsLLVM (#311) (@leios)