Multi-GPU via threading instead of processes#50

Draft

marius311 wants to merge 6 commits intomasterfrom

Commits on Feb 26, 2021

WIP one-GPU-per-thread
marius311
committed

Commits on Feb 27, 2021

add function to adapt to unified GPU memory
marius311
committed

Commits on Apr 15, 2021

Merge branch 'cuda3' into threadgpu
marius311
committed

Commits on Apr 26, 2021

Merge branch 'origin/master' into threadgpu
marius311
committed

Commits on Apr 27, 2021

leave CUDA's unsafe_execute be
marius311
committed

Commits on Apr 28, 2021

turn off TimerOutputs when using threads
marius311
committed