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/ | GSoC2018: https://wp.me/p4pxJf-k1
<nikunj>
oh ok
<nikunj>
hkaiser: no change, it still gives the erro
<nikunj>
*error ^^
<hkaiser>
then I don't know
<nikunj>
hkaiser: I will look into it after I sleep. I should be able to fix things by the end of the day.
<M-ms>
well, you could chain the thens together into one (not a must though, readability etc...)
<M-ms>
but there's no need to call async inside the lambda you pass to then
<M-ms>
the lambda is already a new task
<M-ms>
so you can either directly pass the whatever function you want to call next as the argument to then (i.e. no lambda), but then you need the function to take a future or use hpx::util::unwrap(ping)
<M-ms>
or you use the lambda as now and call the function you want next directly without async
<M-ms>
jakub_golinowski: ^&
<jakub_golinowski>
M-ms, thanks - I think the issue was having nested asyncs -> I was getting the segmentation fault
<M-ms>
jakub_golinowski: hmm, okay, it should not directly lead to segfaults though, it's just unnecessary to call async there
<M-ms>
remember that if you capture by reference in the lambda you have to make sure whatever is captured lives long enough
<M-ms>
if the opencv Mat class is reference counted and does only a shallow copy on assignment you could just capture them by value (don't know if this is the case)
<jakub_golinowski>
M-ms, seems reasonable
jakub_golinowski has quit [Ping timeout: 240 seconds]
jakub_golinowski has joined #ste||ar
anushi has quit [Ping timeout: 244 seconds]
anushi has joined #ste||ar
<jakub_golinowski>
M-ms, so generally I see that wrapping pool_executors in std::ref makes it pass compilation as I am enforcing passing through reference but in the runtime it segfaults
mcopik has joined #ste||ar
<M-ms>
check your lifetimes again, if you return from hpx_main before you use the executors you'll have problems
<M-ms>
you can also pass the executors by value (or move them if they're used only once)
<M-ms>
do you have an example of something that fails?
jbjnr has quit [Ping timeout: 240 seconds]
jbjnr has joined #ste||ar
K-ballo has joined #ste||ar
<jakub_golinowski>
M-ms, I have a working but probably not the best version in my last commit to the pr in the opencv_hpx_backend repo
<jakub_golinowski>
I copy everythin once on the call to the start_webcam_capture
<jakub_golinowski>
But use reference passing in the async to detect_face which is useful and should actually improve performance because before the bind was making a copy in each loop
jakub_golinowski has quit [Ping timeout: 276 seconds]
hkaiser has joined #ste||ar
<heller___>
jbjnr: sorry about earilier, I am in meetings all week. I told ahmed though that i might not make it
<nikunj>
hkaiser: yt?
<hkaiser>
here
<nikunj>
did you read my mail?
<hkaiser>
nikunj: I saw your email, I agree with what you suggest
<nikunj>
ok, then I'll implement it and send a pr
<hkaiser>
tks
<nikunj>
hkaiser: btw yesterday the inspect build failed because we have specified the NOLIB target to it
<hkaiser>
ok, so NOLIB needs to prevent the linker flag to be added as well
<nikunj>
and we added the CMAKE_EXE_LINKER_FLAGS is global
<hkaiser>
right, which now will change
<nikunj>
so it did not add the link file but added the linker flag anyway
<nikunj>
hkaiser: we need to prevent addition of linker flag as well when NOLIB target is specified
<hkaiser>
yes
<nikunj>
hkaiser: quick question. If we use set_target_properties to add LINK_FLAGS does it also append the link flags globally?
<hkaiser>
no
<hkaiser>
just for the target
<nikunj>
for some reason CMAKE_EXE_LINKER_FLAGS contain my linker flag
<nikunj>
even though I'm adding it after a few checks for a target
<hkaiser>
no idea, shouldn't happen
<hkaiser>
have youd eleted you cmake cache?
<nikunj>
ya, it's unexpected behavior
<nikunj>
yes I think so. Let me try it again
<nikunj>
my bad I didn't change my CMAKE_EXE_LINKER_FLAGS
<nikunj>
works perfectly now!
<nikunj>
hkaiser: I've made the changes, refining it before adding the pr
<hkaiser>
nikunj: please verify all kind of combinations, i.e. enable/disable the setting etc.
<nikunj>
yes I'm checking that right now
<zao>
^_^
anushi has quit [Ping timeout: 265 seconds]
<diehlpk_work>
To all our GSoC students: three evaluations are missing. Please do it by today!
<diehlpk_work>
jbjnr, hkaiser M-ms mcopik Could you please finish your evaluation by Wed?
<hkaiser>
yes
<jbjnr>
yes
<mcopik>
ja
<jbjnr>
mcopik: did you make any sense of that logging stuff I sent?
<mcopik>
jbjnr: I have some questins but I also talked with Tal and I have few things to share about the scope of possible integration
<jbjnr>
ok
<mcopik>
but I have paper deadline tonight and it slipped my mind to respond
<mcopik>
jbjnr: I'll try to write a detailed response tonight
<jbjnr>
no worries. just curious at my end. I wasn't planning on doing much else about it for now
<mcopik>
I think we have an idea how both sides could benefit from such cooperation
<jbjnr>
ok
<M-ms>
diehlpk_work: will do
hkaiser has quit [Read error: Connection reset by peer]
hkaiser has joined #ste||ar
hkaiser has quit [Quit: bye]
<jbjnr>
K-ballo: ever seen this one before hpx/runtime/threads/thread_id_type.hpp:97:41: error: constexpr function never produces a constant expression [-Winvalid-constexpr]
<K-ballo>
that can't be gcc/clang, can it? is it intel?
<jbjnr>
I'm using clang 7 on powerpc
<jbjnr>
non-constexpr function 'operator()' cannot be used in a constant expression
<K-ballo>
that one I suggest commenting out /*constexpr*/
<jbjnr>
it doesn't like the less than operator for thread_id
<K-ballo>
I have an open issue against the standard library for that one
<jbjnr>
ok
<jbjnr>
maybe less operator should be consexpr
<K-ballo>
it doesn't look like `std::less` will ever be constexpr for arbitrary pointers, I'm told it is not implementable
<jbjnr>
compiling now
<jbjnr>
thanks
<jbjnr>
crap. there are more
<K-ballo>
yeah, the 4 relational ones
<jbjnr>
lots
anushi has joined #ste||ar
<github>
[hpx] NK-Nikunj opened pull request #3374: Adds hpx_wrap to HPX_LINK_LIBRARIES which links only when specified. (master...add_to_hpx_library) https://git.io/fNUFz
anushi has quit [Remote host closed the connection]
anushi has joined #ste||ar
eschnett has joined #ste||ar
<nikunj>
zao: yt?
anushi has quit [Ping timeout: 240 seconds]
anushi has joined #ste||ar
<zao>
nikunj: Not really.
<nikunj>
no worries, I'll talk when you're free
<zao>
About to go home from work, as nothing is working here anyway <_<
anushi has quit [Remote host closed the connection]
anushi has joined #ste||ar
galabc has joined #ste||ar
david_pfander has quit [Ping timeout: 260 seconds]
hkaiser has joined #ste||ar
<github>
[hpx] hkaiser closed pull request #3365: Fix order of hpx libs in HPX_CONF_LIBRARIES. (master...fix_lib_order) https://git.io/fbhmA
<zao>
Not sure if inspect target is available at all when using Ninja.
<zao>
Bleh, need -DHPX_WITH_TOOLS.
<nikunj>
oh yes, zao: -DPHYLANX_WITH_TOOLS=On
<nikunj>
try with this
<zao>
0 problems reported for phylanx, assuming you just need to invoke the thing.
<nikunj>
that's great. So everything is working properly now
<nikunj>
seems like my implementation worked!
<nikunj>
Let's wait for circle ci tests to run now.
jakub_golinowski has joined #ste||ar
anushi has quit [Ping timeout: 260 seconds]
<M-ms>
jakub_golinowski: yt?
mbremer has quit [Ping timeout: 252 seconds]
<M-ms>
trying to run some tests with the hpx_main backend and it's failing on most tests
<M-ms>
had any problems yourself?
<M-ms>
(except for the ones we know of)
<M-ms>
for example test_dnn has a third of the tests failing and I seem to be setting OPENCV_TEST_DATA_PATH...
<jakub_golinowski>
M-ms, here
<jakub_golinowski>
you mean perf tests?
anushi has joined #ste||ar
<M-ms>
Either, perf also fails
<M-ms>
Are you able to run them?
<jakub_golinowski>
I was not running perf in a orginised manner yet
<M-ms>
Well, test_dnn also fails, I would assume others as well, but trying to figure out if I've misconfigured something or if they're failing for you too
<M-ms>
This is on the main hpx_backend branch btw, hpx_backend_stable doesn't include hpx_main if I saw correctly
<jakub_golinowski>
M-ms, so yes for me a lot of tests fail but they also fail for the other backends (but build from the same HPX repo)
<M-ms>
jakub_golinowski: all right, I'll give it another try later, it's failing only with the hpx backend for me
<hkaiser>
M-ms, jakub_golinowski: we have that issue on master currently that could affect building dependent projects
<hkaiser>
try #3374, that may fix things
jakub_golinowski has quit [Ping timeout: 256 seconds]
jakub_golinowski has joined #ste||ar
K-ballo has quit [Quit: K-ballo]
<jakub_golinowski>
hkaiser, thanks will have a look
<jakub_golinowski>
M-ms, I observe changes to the previous runs - now I get the segfault for imgprocs accuracy test for hpx backend
<jakub_golinowski>
but at the same time the opencv *master* from 5th of July has 242 failed tests in imgproc (however no segfault)
<hkaiser>
jakub_golinowski: alternatively, you could try setting HPX_WITH_DYNAMIC_HPX_MAIN=OFF, which should revert things to the old behavior
nikunj has quit [Quit: goodnight]
K-ballo has joined #ste||ar
galabc has quit [Quit: Leaving]
hkaiser has quit [Quit: bye]
eschnett has joined #ste||ar
<jakub_golinowski>
M-ms, when I am comparing the hpx backend (as in hpx_backend PR) with the master of opencv I have 39 failed tests out of 232 run tests on both of them
<jakub_golinowski>
and these are exactly the same tests
<jakub_golinowski>
M-ms, but I am also skipping Layer_Test_BatchNorm.Accuracy as it was causing segfaults
jakub_golinowski has quit [Ping timeout: 256 seconds]
jakub_golinowski has joined #ste||ar
<jakub_golinowski>
M-ms, that is strange - when I rerun the tests new tests on opencv with hpx backned get segfault - the ones that previously passed :/
mcopik_ has joined #ste||ar
mcopik has quit [Ping timeout: 256 seconds]
mcopik_ has quit [Ping timeout: 264 seconds]
hkaiser has joined #ste||ar
jakub_golinowski has quit [Ping timeout: 256 seconds]