The implementation of block-GMRES is temporary in ExaPF.jl because we need a kernel to copy the upper triangular part of a dense matrix on GPU.
I don't want to add KernelAbstractions.jl as a dependency here for just one kernel.
I opened a PR in the JuliaLang/Julia repository to add a generic copytrito! such that we can add the kernel directly in GPUArrays.jl in the future.
The implementation of block-GMRES is temporary in ExaPF.jl because we need a kernel to copy the upper triangular part of a dense matrix on GPU. I don't want to add
KernelAbstractions.jlas a dependency here for just one kernel. I opened a PR in theJuliaLang/Juliarepository to add a genericcopytrito!such that we can add the kernel directly inGPUArrays.jlin the future.