hkaiser 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/ | GSoD: https://developers.google.com/season-of-docs/
<jaafar> This doesn't exist anymore, does it?
<jaafar> executor_parameter_traits<my_parameter_t>::get_chunk_size
<jaafar> I can't find executor_parameter_traits anywhere in the code
nikunj has quit [Remote host closed the connection]
<jaafar> How can I supply a fixed chunk size to an algorithm?
<jaafar> I am trying:
<jaafar> hpx::parallel::execution::static_chunk_size fixed(50000);
<jaafar> auto ex = execution::par.with(fixed);
<jaafar> and then using "ex"
<jaafar> but get_bulk_iteration_shape calculates the chunk size anyway
<jaafar> that is, the overload for has_variable_chunk_size == false_type calculates a chunk size regardless
<hkaiser> jaafar: do you have a small example we could look at?
<hkaiser> the code you showed above should work properly
K-ballo has quit [Quit: K-ballo]
<jaafar> Thanks hkaiser if you say so I'll try making an MWE
rori has joined #ste||ar
rori has quit [Ping timeout: 276 seconds]
<simbergm> heller: yt?
K-ballo has joined #ste||ar
<heller> simbergm: what's up?
simbergm has quit [Write error: Connection reset by peer]
simbergm has joined #ste||ar
<simbergm> I'm trying to compile hpx with cuda clang, but I'm getting errors about placement new operators missing on the device
<simbergm> I'm starting to think my clang isn't actually built with cuda support (even though it has the cuda command line options)
<simbergm> do you remember if I need to do anything special to enable cuda support when building clang?
<heller> no, building clang doesn't require anythin
<heller> it's when invoking the clang cuda build
<heller> you need to have the right (tm) cuda sdk stuff in path
<heller> but that should be handled by cmake, IIRC
<heller> placement new operator missing on device: try to `#include <new>` on the problematic source files
<heller> placement new only works with that ;)
<heller> simbergm: ^^
<simbergm> heller: hmm, done that
<simbergm> grepping my llvm include directory gives me no device functions, but the device operator new overloads are there in my source directory... somethings off
<heller> strange
<simbergm> do you happen to have an llvm install around to check for yourself? or are you busy?
<simbergm> ok, I take that back about the files missing from my install directory but it's still not happy
<simbergm> ok, specifying question: have you used a custom cuda clang with libc++ or libstdc++?
<heller> simbergm: don't remember sorry
<heller> simbergm: yeah, pretty busy right now
<simbergm> heller: no worries, I'll keep looking
mdiers_ has quit [Quit: mdiers_]
rori has joined #ste||ar
<simbergm> hkaiser: my clang-format is consistent with what circleci says, I have no clue why yours isn't though
<K-ballo> different binary versions?
<heller> hkaiser: https://godbolt.org/z/dnNh7V now with dynamically extendable extra data. I think that's the key to solve issues. Let me give it a shot later
<hkaiser> simbergm: thanks for checking
<hkaiser> heller: thanks!
aserio has joined #ste||ar
hkaiser has quit [Ping timeout: 245 seconds]
<simbergm> hkaiser: you want me to push the changed lines?
aserio has quit [Ping timeout: 250 seconds]
<heller> oh oh, the serialization branch doesn't build with libstdc++ :/
<zao> :D
<heller> well, clang9 + libstdc++
<heller> no idea about others...
<zao> What branch is this?
<zao> module_serialization_2?
<heller> yes
<heller> using g++ 7.something out of ubuntu...
<zao> Definitely fails with my 8.3 too.
<simbergm> whatever happened to get_default_target? we have a declaration and usage, but no definition
david_pfander has joined #ste||ar
<heller> zao: *nod* same for me
david_pfander has quit [Ping timeout: 264 seconds]
<heller> seems to be a issue with C++17
<zao> For completeness sake, master builds on this same setup.
<zao> (oh, gonna build the tests too, but the default targets build at least)
<heller> yeah, same here
<heller> simbergm: no idea, really
<heller> simbergm: more context needed ;)
<simbergm> heller: got undefined reference to get_default_target with my cuda clang build
<simbergm> it was removed a couple of years ago, but I don't get how the nvcc build works...
<heller> simbergm: ifdef'ed out?
<heller> hum
<simbergm> (I switched from my own clang to cray clang btw, got past my operator new problems)
<simbergm> no, the definition is not there at all
<simbergm> it's quite a mystery tbh
<simbergm> anyway, I've added it back for now
lsl88 has joined #ste||ar
<heller> ok
<heller> good to know
rori has quit [Quit: WeeChat 1.9.1]
aserio has joined #ste||ar
hkaiser has joined #ste||ar
<hkaiser> heller: I have no idea why gcc and clang are failing for the serialization branch
nikunj has joined #ste||ar
<K-ballo> looks like recursive instantiation
<K-ballo> try splitting each enable_if in two, one to disable basic_any first, then the constructible chekc
<hkaiser> yah, that's what I was thinking, but I felt dirty doing it ;-)
<hkaiser> it's clang after all...
<K-ballo> why dirty? it is the correct way
<K-ballo> the current way is bogus, lacks short circuiting
<hkaiser> ok
<hkaiser> I thought it would do shortcutting
<K-ballo> it does evaluation short-circuiting, which here is meaningless, but not instantiation short-circuiting
<K-ballo> enable_if<A && B> needs to instantiate both A and B regardless of the value of A
hkaiser has quit [Ping timeout: 240 seconds]
aserio has quit [Ping timeout: 264 seconds]
aserio has joined #ste||ar
hkaiser has joined #ste||ar
<hkaiser> K-ballo: ok
aserio has quit [Quit: aserio]