Support for writing loop-based code that executes both on CPU and GPU
GPUifyLoops.jl is no longer under development and has been replaced by KernelAbstractions.jl.
GPUifyLoops is a registered package, and can be installed using the Julia package manager.
julia>]
(v1.1) pkg> add GPUifyLoopsNote: The current version of this package requires Julia 1.1.
Debugging failures to transforma a function for the GPU requires the use of Cthulhu.jl.
using Cthulhu
using GPUifyLoops
# @launch CUDA() f(args...)
descend(GPUifyLoops.signature(f, args...)...)
In order to test this package locally you need to do:
julia --project=test/gpuenv
julia> ]
(gpuenv) pkg> resolve
(gpuenv) pkg> instantiate
This will resolve the GPU environment, please do not checking changes to test/gpuenv/.
Then you can run the tests with julia --project=test/gpuenv test/runtests.jl
GPUifyLoops.jl is licensed under MIT license.