Currently, HostThreadTeamData assumes that jobs should be scheduled over all the threads available. For resilient execution spaces we need a bit more control to ensure that duplicate kernels run in parallel. One possibility is to add an extra parameter to control how many threads to use (defaulted to league size). This should be brought up in discussion.
Morning @nmm0 -- would you please comment on this issue (based on design discussions with Christian), and close it once you think the way forward is documented?
Currently,
HostThreadTeamData
assumes that jobs should be scheduled over all the threads available. For resilient execution spaces we need a bit more control to ensure that duplicate kernels run in parallel. One possibility is to add an extra parameter to control how many threads to use (defaulted to league size). This should be brought up in discussion.