Does GTX 1050 ti support grid synchronization? I got syntax error “identifier “grid_group” is undefined”, even though included #include<cooperative_groups.h> and set rpc on true.
Look at a CUDA grid sync sample code such as conjugateGradientMultiBlockCG
you need to do something like:
cg::grid_group
not just:
grid_group
(you need to indicate the correct namespace, and its not cg unless your code makes it so – study the sample code)
also, grid sync is only available if the property indicates so.
[url]CUDA Runtime API :: CUDA Toolkit Documentation
cooperativeLaunch
It’s good practice to query the property. Why does this matter?
A GTX 1050 Ti on linux will support grid sync (i.e. cooperative kernel launch). A GTX 1050 Ti on windows will not. The property will be set correctly in either case.