00:00 <weilewei> btw, just get an interesting question from the lab, they said, is it possible for hpx to control 1 system threads while many hpx threads running on top of that system threads?

00:01 <weilewei> One guy suspect some resources are thread independant

00:04 <weilewei> simbergm shall I use your commit to test?

00:08 <weilewei> hmmm. not sure why spin_lock is invoked, though it is not my intention

00:13 <weilewei> Is hpx stable tag gone? https://github.com/STEllAR-GROUP/hpx/tree/stable is not accessable

00:14 <weilewei> sorry if I throw too many questions here...

00:22 <weilewei> The latest stable tag is 1db0e523eefb955780fdf8ed9e6572040e507730, on Oct 17

00:43 <weilewei> oops, after checking out simbergm your commit, all cmake build failed... Is it the recent hpx build system change thing?

09:16 <simbergm> hmm, pushing the stable tag keeps failing... I'll look into that

13:58 <weilewei> simbergm thanks!

16:21 <simbergm> weilewei do you have a log of your cmake failure?

16:21 <weilewei> simbergm yes, because lots of executables reply on hpx_setup_target, etc.

16:22 <weilewei> let me pull it out

16:24 <weilewei> need to rebuild hpx.. give me some time

16:51 <weilewei> simbergm https://gist.github.com/weilewei/6f78da8f25e5f68ccd6e5e9f8dc561fa

16:51 <weilewei> please find here

16:58 <simbergm> weilewei please try latest master

16:58 <weilewei> btw, just curious, is there any hpx settings that will force only 1 system thread running, and then multiple hpx threads can run on top of that system thread?

16:58 <weilewei> simbergm ok, let me try

17:00 <simbergm> weilewei not sure what you mean other than - - hpx:threads=1?

17:01 <weilewei> hmmm, so using --hpx:threads=1 actually allows only 1 hpx thread to run at a time?

17:01 <weilewei> ok, I see what it means

17:02 <zao> A number of system threads can be turned into something that can process HPX work.

17:03 <weilewei> ok

17:04 <zao> I don't know if you can explicitly make system threads you start yourself into HPX workers.

17:05 <zao> HPX normally spins up a thread pool with workers.

17:05 <zao> Which I'm assuming that command line option controls.

17:11 <weilewei> -- Found CUDA: /sw/summit/cuda/10.1.168 (found version "10.1") CMake Error at cmake/HPX_SetupCUDA.cmake:29 (target_link_directories): Cannot specify link directories for target "hpx::vc" which is not built by this project.Call Stack (most recent call first): CMakeLists.txt:1621 (include)

17:12 <weilewei> what is vc?

17:13 <weilewei> so I need to install vc first to enable HPX_WITH_CUDA...

17:16 <zao> Vc is a library for SIMD operations.

17:17 <weilewei> Right, before that Vc is not required for HPX_WITH_CUDA

17:28 <simbergm> weilewei are you specifying HPX_WITH_DATAPAR_VC? if not it's probably my fault... You should also not need (or even be able to use) on summit

17:30 <simbergm> roti did a hpx::vc maybe sneak into HPX_SETUPCUDA?

17:32 <weilewei> simbergm I am not specifying HPX_WITH_DATAPAR_VC on my build

17:32 <simbergm> Ah yes, hpx::vc should be hpx::cuda in that file...

17:33 <weilewei> oh, i see, thanks for that response..

17:33 <simbergm> weilewei could I ask you to do that change and maybe even open a pr?

17:33 <simbergm> Line 29 in cmake/HPX_SetupCUDA

17:34 <simbergm> .cmake

17:34 <weilewei> yes, sue, I would like to, I will do it this afternoon

17:35 <simbergm> Thanks!

17:35 <weilewei> np, I will create that PR after my lunch : )

17:41 <simbergm> weilewei there's a couple of other hpx::vcs in that file that should be changed as well if you don't mind :)

18:36 <weilewei> simbergm noted and I just created PR to HPX

18:44 <hkaiser> weilewei: thanks a lot!

18:45 <weilewei> hkaiser welcome

18:45 <weilewei> I try to build dca with latest hpx branch + my cuda cmake change, then I got this error:

18:46 <weilewei> CMake Error at cmake/dca_testing.cmake:85 (add_executable): Target "tp_accumulator_gpu_test" links to target "hpx::cuda" but the target was not found. Perhaps a find_package() call is missing for an IMPORTED target, or an ALIAS target is missing?Call Stack (most recent call first):

18:46 <weilewei> test/unit/phys/dca_step/cluster_solver/shared_tools/accumulation/tp/CMakeLists.txt:30 (dca_add_gtest)

18:46 <hkaiser> nod

18:46 <weilewei> What should I do with this?

18:50 <weilewei> Also, it is not relevant to cuda thing, but when I build dca with latest HPX, I have to pass JEMALLOC_LIBRARY JEMALLOC_INCLUDE_DIR to my cmake, otherwise it complains

18:51 <weilewei> -- Could NOT find Jemalloc (missing: JEMALLOC_LIBRARY JEMALLOC_INCLUDE_DIR) CMake Error at /gpfs/alpine/proj-shared/cph102/weile/dev/install/hpx_Debug/lib64/cmake/HPX/HPX_Message.cmake:48 (message): HPX_WITH_MALLOC was set to JEMALLOC, but JEMALLOC could not be found. Valid options for HPX_WITH_MALLOC are: system, tcmalloc, jemalloc, mimalloc,

18:51 <weilewei> tbbmalloc, and customCall Stack (most recent call first): /gpfs/alpine/proj-shared/cph102/weile/dev/install/hpx_Debug/lib64/cmake/HPX/HPX_SetupAllocator.cmake:63 (hpx_error) /gpfs/alpine/proj-shared/cph102/weile/dev/install/hpx_Debug/lib64/cmake/HPX/HPXConfig.cmake:52 (include) src/parallel/hpx/CMakeLists.txt:12 (find_package)-- Configuring

18:51 <weilewei> incomplete, errors occurred!

18:52 <weilewei> Although hpx itself builds successfully, and can find jemalloc. Before hpx changes, I do not need to manually pass JEMALLOC_LIBRARY JEMALLOC_INCLUDE_DIR to my dca cmake

18:52 <hkaiser> simbergm: ^^

18:59 <simbergm> hkaiser weilewei, urgh, sorry about that

19:00 <hkaiser> simbergm: np at all

19:01 <simbergm> I think easiest right now would be to go back to the commit right before we merged the cmake branch and set HPX_HAVE_VERIFY_LOCKS=OFF

19:01 <simbergm> That way we're not blocking you with these problems

19:01 <simbergm> Or just go back to 1.3.0 even

19:01 <weilewei> simbergm thanks for that, I will try it out

19:02 <simbergm> Daint is back online again so we can test and fix the Cuda configuration (again...)

19:02 <simbergm> Tomorrow