Valentin Churavy

I don't think we need to recurse into that just yet. @jpsamaroo are these tasks that are both local and remote?

I am not sure I understand this issue. You are running on a SGE cluster and you are trying to add processes with `addprocs_qrsh`? That won't work. > For some...

Yeah so either you get stuck in allocating forever or ~the time-out doesn't trigger.~ Looks like qsub doesn't have a time-out. You may want to instrument this code. https://github.com/JuliaParallel/ClusterManagers.jl/blob/e375f50f2c4eab3d3f4cefcea3465c82734cfb71/src/qsub.jl#L83

This has become a pile of various changes I needed for a project, I will separate them out as well as cleaning them up.

Not that old :) and I would keep it open as a reminder to myself that this work needs to get cleaned up and finished or migrated over to Dagger.

Oh that is rather interesting! I see you are running it on travis.

Yeah just came across this the whole vector testsuite is not being run. We wanted to do `push!(CuArray([1]), 1)`.

That is an excellent find! I think the short-term solution is for GPUArrays to use a custom `combine_eltypes` that uses the device interpreter. I don't think something like `with_interpeter` (besides...

> I had expected this when running with a single thread, because the main task isn't preemtible, but even with multiple threads the main task getting stuck apparently blocks the...

I discussed this with @trws yesterday and he saw good or even better performance using LTO with openmp offloading and cudadevrt. x-ref: https://github.com/JuliaGPU/CUDA.jl/blob/de004245e51e4f27b24d6952cc6dba989bc1ba98/src/compiler/execution.jl#L437-L438 Do we need to use nvlink or...