hkaiser changed the topic of #ste||ar to: The topic is 'STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar.cct.lsu.edu | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | Buildbot: http://rostam.cct.lsu.edu/ | Log: http://irclog.cct.lsu.edu/
primef4 has joined #ste||ar
primef4 has quit [Ping timeout: 252 seconds]
primef4 has joined #ste||ar
primef4 has quit [Ping timeout: 260 seconds]
primef4 has joined #ste||ar
primef4 has quit [Ping timeout: 272 seconds]
hkaiser has quit [Quit: bye]
primef4 has joined #ste||ar
primef4 has quit [Ping timeout: 246 seconds]
maxwellr96 has quit [Ping timeout: 260 seconds]
nikunj has joined #ste||ar
nikunj has quit [Ping timeout: 265 seconds]
rori has joined #ste||ar
<simbergm>
can I just not find it or does our hpx::compute::cuda::target not actually allow setting a stream? or do we have something else which lets us set a stream and then get a future from the stream?
<simbergm>
now I see... it creates a new one when you access the stream
nikunj has joined #ste||ar
primef4 has joined #ste||ar
nikunj has quit [Ping timeout: 246 seconds]
primef4 has quit [Ping timeout: 260 seconds]
nikunj has joined #ste||ar
nikunj has quit [Ping timeout: 268 seconds]
rori has quit [Ping timeout: 246 seconds]
gregordaiss[m] has joined #ste||ar
gregordaiss[m] is now known as gdaiss[m]
primef4 has joined #ste||ar
hkaiser has joined #ste||ar
hkaiser has quit [Quit: bye]
primef4 has quit [Ping timeout: 245 seconds]
primef4 has joined #ste||ar
primef4 has quit [Ping timeout: 272 seconds]
nikunj has joined #ste||ar
primef4 has joined #ste||ar
nikunj97 has joined #ste||ar
nikunj has quit [Ping timeout: 260 seconds]
<jbjnr>
simbergm: you need to look at the cuda stram helper
<jbjnr>
^stream
<diehlpk_work>
simbergm, jbjnr heller2 This morning is the last possibility to edit the GSoC applicaiton
<jbjnr>
does it need anything
<diehlpk_work>
From my side it is fine
<simbergm>
all good from my side as well
<jbjnr>
that's good enough for me
<simbergm>
jbjnr: I'll look at that
<simbergm>
thanks
<diehlpk_work>
jbjnr, Which libfabric branch do you use?
<nikunj97>
so stellar is applying for gsoc!
<diehlpk_work>
nikunj97, Yes
<jbjnr>
diehlpk_work: rdma_jan_2020
<diehlpk_work>
Ok, I will compile it on Cori this evening
<diehlpk_work>
Can I use the options from the last SC paper?
<jbjnr>
diehlpk_work: apart from the fact that it doesn't work, nothing much
<diehlpk_work>
Ok, so it makes no sense to run the problem on two to four nodes yet?
<jbjnr>
you can try. The main problem I had was my libfabric build was broken, but since I fixed it, I've not had time to run tests properly yet
<jbjnr>
I had a PMI boot problem when I ran a larger job, but it might be different on cori
<diehlpk_work>
Ok, let me run some test this evening and I will send around some primarily results tomorrow
<diehlpk_work>
Whoch hpx commit should I use?
<jbjnr>
the branch I said before
<diehlpk_work>
Ok, sorry too less sleep last night
<jbjnr>
I've got a ton of uncommitted stuff that I'll try to clean up tonight and push
<jbjnr>
actually, I think you need that stuff, it probably won't compile on my main branch
<jbjnr>
give me another day or two
<diehlpk_work>
Ok, sounds good
<jbjnr>
simbergm: the stuff in that cuda future example is what I want to move into a new module called cuda_support - it can be used by projects that need cuda, without having to compile hpx with full cuda::compute - only the stream stuff needs to be setup
<jbjnr>
so I will have cuda_support with cuda futures and mpi_support with the mpi ones
<simbergm>
jbjnr: yep, sounds good
<simbergm>
the other cuda compute stuff is not much though, but having them separate is nice anyway
<simbergm>
I actually only needed (at least for now) `future<void> get_future(cudaStream_t)` without the target wrapper thing
<simbergm>
i.e. I wanted to provide the stream myself
<jbjnr>
Isn't there a target constructor that takes a stream handle? (if not, you can make one I expect)
<simbergm>
there's not but I might add one
<simbergm>
I just added the free function to get things running but I'll have to make a PR with it eventually
<simbergm>
(I'll be away thursday, friday)
<jbjnr>
feel free to create a new module with the cuda future stuff in whilst you're at it - it should only require moving a few files around (I hope)
<jbjnr>
(and making the CMake glue do the right thing)
<jbjnr>
(and changing the namespace to hpx::cuda)
<jbjnr>
(and I'll have a Black tea with milk please, no sugar thanks)