Choose timezone
Your profile timezone:
Madgraph5 GPU development
→
Europe/Zurich
31/S-028 (CERN)
Madgraph5 GPU development
# Madgraph4gpu dev meeting Tue 20 Feb 2024
https://indico.cern.ch/event/1355146/
Present/remote: SR, AV (must leave 4:45), CV, ZW, TC, OM, NN
## Round table
ZW: was back in Milano in the last two weeks, working with Marco Zaro to kickstart the development for NLO, quite good progress, looking at two issues to solve.
OM: working on the three steps, survey, refine and combine. As we saw at the hackathon, the combine step can be very slow. There are two changes in the code, first the combination of events is done on a subset of events, the second change is that these steps can be done in parallel in several threads.
SR: is the remainder I/O bound? OM: not clear. SR: maybe binary format for lhe files would improve here.
NN: created a PR to copy to epochX/sycl the stuff that was in my fork.
NN: also working on including sycl support in cudacpp; for the moment it seems that this needs cuda streams and hip streams as an additional interface, but this is not yet clear. It seems that sycl queues uses cuda streams-like objects as an interface.
OM: if there is also a need to split kernels this is better done in a second step.
TC: ntr
SR: making progress on the channelid, working on gpu and cpu, also added a few things on mg5amcnlo gpucpp_wrap. Not a lot of changes in the end. One minor issue to fix is integer arrays, where there is an ugly static cast that I would like to remove, this has a clash with an `auto` that the compiler does not figure out.
CV: ntr
AV: shows the slides attached to the agenda
## AOB
SR: there is one position open for a quest fellow in the NextGenTrigger, up to maximum 3 years after phd. Will circulate it in the mailing list or minutes.
Next meeting in two weeks, March 5th.
AV and OM will not be able to join (6th would be good for both of us). Decide to skip it, next meeting on March 19th.
There are minutes attached to this event.
Show them.
