K-ballo changed the topic of #ste||ar to: 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/
<hkaiser>
jaafar: there are 2 or 3 tests that are known to occasionally fail
<jaafar>
Oh dear
<jaafar>
Well, it looks like I'm getting inconsistent failures, which is a bummer
<jaafar>
On master
<jaafar>
is_heap, is_heap_until, is_sorted_until
<jaafar>
but only when I run several at once instead of 1 at a time
<jaafar>
or if I run them in a loop
<jaafar>
is_heap_until, for example, ran 19 times and failed on the 20th
<jaafar>
oh it's this thing again: empty function object should not be used: HPX(bad_function_call)
<jaafar>
Last time make clean fixed it, I will try that again :)
<K-ballo>
which function object is it even coming from?
<K-ballo>
I doubt any algorithm is using those in its implementation
<jaafar>
I've seen this before, it's kinda weird.
<jaafar>
It will probably go away after the clean rebuild, and then I'll see where I stand
<srinivasyadav227>
hkaiser: when building with clang with vc, its showing test passed (found flag) for -Wtrampolines and -Wno-sync-nand, but building with clang without vc, its showing test failed (could not find) these flags
<hkaiser>
k
<srinivasyadav227>
hkaiser: so does it inherit cmake of Vc in any means?, i just felt it was bit hard coding ;-)
<hkaiser>
nod, could be a Vc thing
<srinivasyadav227>
ok
<hkaiser>
thanks for working on this!
<hkaiser>
I have a couple of comments, but I will add those tomorrow
<srinivasyadav227>
srinivasyadav227: glad to work here, i just learned many things, especially debugging with circle ci ;-)
<srinivasyadav227>
hkaiser: sure, i will work on foreach_performance branch with datapar till then
<srinivasyadav227>
hkaiser: would you recommend me to start working on proposal along with PR?
<hkaiser>
yes, absolutely
<srinivasyadav227>
ok, thanks ;-)
diehlpk_work has quit [Remote host closed the connection]
hkaiser has quit [Quit: bye]
bita has quit [Ping timeout: 240 seconds]
<jaafar>
yoiks. Clean build... still does it. Randomly. I think I'll file a bug, with the understanding I'm probably doing something wrong :)
<jaafar>
Still it looks weird. The backtrace goes through hpx::components::server::runtime_support::call_startup_functions
<jaafar>
Maybe there's a bug in something else I'm using
<jaafar>
I'll try a debug build overnight
<jaafar>
OK "make clean" doesn't actually clean everything lol
<gonidelis[m]>
hkaiser: if you check the 8. Community section on the HPX paper line 3 (i don't know if I could pm you any link with overleaf), that might be a miss?
<hkaiser>
fixed
<gonidelis[m]>
k
<hkaiser>
thanks for spotting
<hkaiser>
Happy Birthday HPX, btw
<gonidelis[m]>
wow
<gonidelis[m]>
13 years
<gonidelis[m]>
huh?
<hkaiser>
yah
<gonidelis[m]>
happy birthday !
nanmiao has joined #ste||ar
bita has joined #ste||ar
diehlpk_work has joined #ste||ar
<gnikunj[m]>
hkaiser I'm running late for the meeting. Can I update you on the progress in our afternoon meeting? :/
<ms[m]>
hkaiser: btw, what about #5172? would you still like to look through it?
<bita>
K-ballo, thanks for your response. I have added these flags and it still didn't work, so I am not sure what I need to change. Those flags are: "-DHPX_WITH_CXX17=ON -DCMAKE_CXX_STANDARD=17 -DHPX_USE_CMAKE_CXX_STANDARD=ON"
<K-ballo>
bita: you'd have to look at the corresponding cmake source in the context refered to by the error, do you know where that is?
<bita>
Unfortunately I don't
<bita>
The thing is the folder it refers to is empty: "phylanx\cmake-build-debug\CMakeFiles\config_tests\CMakeFiles\CMakeTmp" is empty
<hkaiser>
ms[m]: yah, still planning to have a look
<hkaiser>
pedro_barbosa[m]: I'm not sure that for_loop supports GPU execution to begin with :/
<hkaiser>
I need to have a look, but I'm almost certain, that we have never adapted for_loop
<hkaiser>
in general, I would suggest you don't use hpx.compute as it's unclear how this will evolve
<pedro_barbosa[m]>
Ok, thank you for the help
nanmiao has quit [Quit: Connection closed]
<jaafar>
Similar to jedi18[m] 's comment from yesterday I just discovered the inclusive_scan and exclusive_scan algorithms have slightly different APIs
<jaafar>
exclusive_scan offers a version with "init" defaulted; inclusive_scan offers "op" defaulted and both init and op defaulted, but no "init defaulted"
<jaafar>
That's a bit surprising
<K-ballo>
those changed a couple of times during standardization
<gonidelis[m]>
any ideas how i could allocate a specific node on rostam
<gonidelis[m]>
i use `-p medusa` for example but i want sth like `-p medusa01`
nanmiao has joined #ste||ar
<gnikunj[m]>
hkaiser: could you point at the place where we serialize (non-captured) lambdas?