Feature #873
send messages from non-PE threads
Start date:
10/28/2015
Due date:
% Done:
0%
Description
CUDA has the ability to call back to the CPU when a kernel finishes, but the callback executes in a thread that is launched by the CUDA runtime so messages sends fail. If messages could be sent from this thread then it would not be necessary to poll the GPU for kernel completion. This would only need to be supported in smp/multicore versions. This could also be used for dedicated I/O threads.
History
#1 Updated by Eric Bohm about 2 years ago
- Target version changed from 6.8.0 to 6.9.0
#2 Updated by Sam White about 2 years ago
- Category set to Machine Layers
#3 Updated by Jim Phillips almost 2 years ago
I observe that #1393 seems to have solved this issue for GPUManager.
#4 Updated by Eric Bohm over 1 year ago
- Assignee set to Michael Robson
#5 Updated by Michael Robson about 1 year ago
- Status changed from New to Implemented
See above comments about 1393 solving this issue