<jbjnr>
I'll send you inor about travel. You'l need to liaise with heidi or someone again as before
<jbjnr>
heller: ^
<simbergm>
jbjnr: heller: call today to redistribute work for the course?
<jbjnr>
ok
<simbergm>
with the risk of making things even more complicated I've started an hpx channel on matrix.org: https://riot.im/app/#/room/#hpx:matrix.org
<heller>
simbergm: tomorrow would be better for me
<jbjnr>
simbergm: why?
<jbjnr>
I can't even see anything without logging in
<jbjnr>
why not discourse?
<jbjnr>
at least people have heard of it
<simbergm>
heller: jbjnr (I think) and I are away tomorrow
<simbergm>
friday?
<simbergm>
jbjnr: I expect noone to join it
<jbjnr>
I don't think I'll go. Too much to do here
<simbergm>
you should be able to see history without signing in
<heller>
Friday is good as well
<jbjnr>
simbergm: ok, then carry on.
<simbergm>
open source etc
<simbergm>
but yes, why not discourse
<simbergm>
or any other app
<heller>
simbergm: have you seen my execution_context work?
<simbergm>
heller: yes, very much looking forward to it!
<heller>
The first iteration is almost done now
<jbjnr>
everything is going to be so clean soon!
<heller>
This should help you moving forward much much quicker
<jbjnr>
the modularization will make like so much nicer
<jbjnr>
^life
<heller>
I hope so...
<heller>
simbergm: regarding the execution_context stuff, I think we need to reorganize the modules a tiny bit then
<simbergm>
heller: what would you need to do?
<heller>
I think futures, mutex, spinlock and executors should go there as well
<heller>
and condition_variable and friends
<heller>
and then, the coroutine module can build on top of it
<heller>
as well as any other thread pool implementation
<simbergm>
jbjnr: do we need #4086? what's wrong with CXXFLAGS? I know it's more work for you but it's not that much...
<heller>
this should then give you exactly what you want ;)
<jbjnr>
CXXFLAGS you have to make an effort to set by hand, this just does it automatically. Why do we have release and debug builds, you could just ask everyone to add -O3 and do it by hand. Same reason
<heller>
simbergm: an openmp or kokkos integration should then be more than trivial...
<heller>
that is, an openmp based thread pool
rori has joined #ste||ar
<simbergm>
heller: s/more than trivial/easier
<simbergm>
but yes
<simbergm>
jbjnr: fine (cmake supports build types natively)
<heller>
simbergm: well, we just use the kokkos thread pool, implement an execution context (note: execution_context should also expose a executor eventually), and call it a day
<heller>
making it possible to use any HPX facilities inside kokkos
<heller>
and kokkos can more or less remain unchanged
<simbergm>
why do we reuse the data from a previous task when rebinding?
<simbergm>
jbjnr: heller any idea?
<jbjnr>
only khuck knows what's going on in there. I had a long discussion with him about this when I was trying to debug the apex name problem. It uses the wrong name for tasks and I couldn't understand why it uses 'the wrong' task to get name info etc etc