[CUDAKernels] add an implicit sync to kernels with no dependencies (!222) · Merge requests · Churavy, Valentin / KernelAbstractions.jl

Churavy, Valentin requested to merge vc/cuda_event into master Mar 17, 2021

Created by: vchuravy

Fixes #221 (closed), by placing synchronizing against the task local stream for kernels that are not launched with dependencies. Maybe we should generally synchronize KA launches against the task-local stream?

Also uses the task-local stream instead of the CuDefaultStream.

[CUDAKernels] add an implicit sync to kernels with no dependencies

Merge request reports