ImplicitGlobalGrid.jl 0.16.0

Release notes

• Widen array support to CellArrays and any bitstype dense array (CPU, CUDA, AMDGPU), including SubArrays (#102) (@omlins)

ImplicitGlobalGrid.jl 0.15.3

Release notes

• Remove usage of deprecated CUDA.Mem (#99) (@omlins)

ImplicitGlobalGrid.jl 0.15.2

Release notes

• Update AMDGPU compat (#93)

ImplicitGlobalGrid.jl 0.15.1

Release notes

• Replace LoopVectorization extension with Polyester extension as an alternative to Threads for efficient CPU halo updates (#87)

ImplicitGlobalGrid.jl 0.15.0

Release notes

• Use extensions for GPU dependencies: CUDA/AMDGPU must newly be imported before ImplicitGlobalGrid to load the corresponding extension (CHECK COMPATIBILITY!) (#84)
• Use extension for LoopVectorization dependency (#83)
• Enable gather! also for ndims > 3 (#85)

ImplicitGlobalGrid.jl 0.14.0

Release notes

• Generalize halo updates to support halos of any width enabling higher order methods in space (#69)
• Enable device_type to be "none" to make the use case of using CPUs only on a system having also GPUs more clear
  (#79)
• Add support for AMDGPU v0.8 (#74)

ImplicitGlobalGrid.jl 0.13.0

Release notes

• Refactor to be compatible with AMDGPU v0.5 (#55)
• Refactor to be compatible with MPI v0.20 (#55)
• Add support for CUDA v4 (#51)

ImplicitGlobalGrid.jl 0.12.0

Release notes

• Replace traditional CUDA dependent conditional (pre-)compilation with runtime detection whether CUDA is functional (#29)
• Add keyword argument for automatically selecting device in init_global_grid: the device is newly automatically selected by default in order to remove the necessity of calling select_device for running on multi-device nodes (CHECK COMPATIBILITY!) (#33)
• Add support for CUDA.jl 3.7 and 3.8 (#32, #33; resolves #25)
• Give error in init_global_grid if the global grid was already initialized (#31)

ImplicitGlobalGrid.jl 0.11.1

Release notes

• Improve CPU buffer copies for small arrays and add initial LoopVectorization support (#27)
• Allow gather! to take nothing as the global array on non-root procs (#24)
• Fix compat for CUDA.jl and remove ExprTools dependency workaround (#26)

ImplicitGlobalGrid.jl 0.11.0

Release notes

• Add support for complex numbers (#18)
• Adjust gather! to work with all current MPI.jl versions (#18)
• Workaround CUDA.jl/GPUCompiler.jl incompatibility of indirect dependency ExprTools.jl v0.1.4 (#19)