Add new Example20 with interaction queues instead of SOAData #234
Conversation
|
Can one of the admins verify this patch? |
|
Great to see that this works, thanks. Have you considered creating an example20 out of this, and document the difference to ex19 as was done for the other examples? |
Yes, and I can do so if you want. However, there is no difference in the executed workflow, it's just implemented differently. In the past, we did these "technical improvements" within existing examples... |
hahnjo
force-pushed
the
example19-interaction-queues
branch
2 times, most recently
from
1a2bbfb to
61e33bc
Compare
Yes, please do so. Thanks! |
This revives an old concept from the toy examples 1, 2, and 4 (deleted in commit 9230c4a) to have one queue per discrete interaction. This appears to be equally fast as the quite complex scheduling kernel based on SOAData.
hahnjo
changed the title
hahnjo
force-pushed
the
example19-interaction-queues
branch
from
61e33bc to
5d8bdf8
Compare
Done. Would be interesting to know what you have in mind with |
This revives an old concept from the toy examples 1, 2, and 4 (deleted in commit 9230c4a) to have one queue per discrete interaction. This appears to be equally fast as the quite complex scheduling kernel based on SOAData.
@bernhardmgruber see here for not using
__shared__memory in Example19 as mentioned here: #229 (comment)