ParallelStencil.jl 0.11.1

Release notes

• Use extension for Enzyme dependency (#139, #140)
• Add support for CellArrays 0.2 (#141, #142)

ParallelStencil.jl 0.11.0

Release notes

• Change scope of ParallelStencil initialization parameters to module scope: @init_parallel_stencil must newly be called once per module - right after using ParallelStencil is recommended (CHECK COMPATIBILITY!) (#130)
• Enable dimension-agnostic kernel generation thanks to new ndims tuple expansion and type parameter substitution keyword N (#134, #136)
• Enable architecture-agnostic creation of indices arrays with the allocator macros (#137)
• Fix type instability when using type aliases directly (#135)

ParallelStencil.jl 0.10.1

Release notes

• Add @d2_xa and @d2_ya macros for 2D (#131)

ParallelStencil.jl 0.10.0

Release notes

• Enable automatic ranges detection for any bitstype variables and nested (named) tuples containing any supported type (#121)
• Add tuple type aliases and NamedTuple constructors in the Data module (#121)
• Allow complex arguments for compute macros (expanding arguments) (#121)
• Add support for AMDGPU 0.8 (#123)
• Add support for CUDA 5 (#129)

ParallelStencil.jl 0.9.0

Release notes

• Add support for automatic addition of @inbounds in kernels by keyword (inbounds) and make global default settable in @init_parallel_stencil (#119)
• Enable usage of Structs, NamedTuples and other expressions in @parallel ∇ calls (#114).
• Enable writing dimension-agnostic @parallel_indices kernels by supporting parallel indices generation as a tuple (syntax: @parallel_indices (I...) <kernel>) (#116)
• Add support for direct usage of FiniteDifferences{1|2|3}D macros and any other macros programmed with ParallelStencil.INDICES in @parallel_indices kernels (#111)
• Allow for return statements in nested functions within @parallel_indices kernels (#117, #118)
• Enable multiple computation calls in @hide_communication and explicit ranges for computation calls (#115)
• Remove initialization-related warnings in interactive usage (#113)
• Fix 2D thermo-mechanics miniapp (#74)

ParallelStencil.jl 0.8.2

Release notes

• Fix stream synchronization for AMDGPU backend (#109)

ParallelStencil.jl 0.8.1

Release notes

• Add AMDGPU v0.5 support (#107)

ParallelStencil.jl 0.8.0

Release notes

• Add high-level support for architecture-agnostic Enzyme-powered automatic differentiation (#101)
• Make generic Enzyme-based automatic differentiation of ParallelStencil kernels more convenient (#99)

ParallelStencil.jl 0.7.1

Release notes

• Make shared memory allocation robust for compilation throughout all CUDA/AMDGPU versions (#98)

ParallelStencil.jl 0.7.0

Release notes

• Add keyword memopt to @parallel and @parallel_indices, exposing generalized optimization of fast memory usage (of registers and shared memory) (#81, #94 )
• Add support for AMDGPU (#69, #81 , #93, #95 )
• Add support for arrays of small arrays/structs leveraging CellArrays via keywords in the hardware-agnostic allocators (#54, #95)
• Add @fill, @falses and trues allocators (#54)
• Enable allocation with enums using @fill and @rand (#62)
• Support numbertype omission in ParallelStencil initialization (#47)
• Add macro to compute harmonic averages (#57)
• Add documentation for memopt optimization, CellArrays and AMDGPU (#97)
• Add support for CUDA v4 (#81)
• Add support for Julia 1.9 (#81)