07:29 <Yorlik> I have sortof a style question: We have a header only networking library and I am thinking of separation declarations and definitions into e.g. a file client_api.hpp and client.hpp(impl). Users would have to include client.hpp (which includes client_api.hpp at the top). To look up the api one would have to look at client_api, which would also have the documenting comments for doxygen. Does that make

07:29 <Yorlik> sense? I'd just like to separate implementation and declarations for better readability. What do you think?

14:12 <srinivasyadav227> hkaiser: what is the use of these 2 lines here https://github.com/STEllAR-GROUP/hpx/blob/master/examples/quickstart/executor_with_thread_hooks.cpp#L63-L65 ? They are not used any where in the struct, are those for inheritance purposes ?

14:13 <hkaiser> those are used to expose the category etc. through well defined names

14:13 <hkaiser> this simplifies introspection

14:13 <hkaiser> (i.e. figure out properties of an executor)

14:14 <hkaiser> all executors should have those embedded types defined

14:14 <srinivasyadav227> ok cool, but `execution_category` does not exist for sequenced executor's right ?

14:14 <hkaiser> well, there is a default value used if not explicitly defined

14:15 <hkaiser> let me find it

14:15 <srinivasyadav227> ok, how do I extract for sequenced executor ? 😅

14:16 <hkaiser> give me a sec

14:16 <srinivasyadav227> thanks :)

14:16 <srinivasyadav227> sure

14:18 <hkaiser> srinivasyadav227: here: https://github.com/STEllAR-GROUP/hpx/blob/master/libs/core/execution/include/hpx/execution/traits/executor_traits.hpp#L100-L111

14:19 <srinivasyadav227> I got more question please :), when using parallel executors, by default each core/thread gets 4 chunks ?

14:19 <srinivasyadav227> cool, got it :)

14:51 <hkaiser> srinivasyadav227: yes, we decided that 4 is a good number ;-)

14:52 <hkaiser> srinivasyadav227: by default, this code calculates the number of chunks: https://github.com/STEllAR-GROUP/hpx/blob/master/libs/core/execution/include/hpx/execution/executors/static_chunk_size.hpp#L73-L79

14:53 <srinivasyadav227> cool xD

14:54 <hkaiser> well, it calculates the chunk size, but that implicitly controls their number

14:57 <srinivasyadav227> cool, i think threads are working greatly with likwid 😅

14:57 <srinivasyadav227> thread hooks*

14:58 <srinivasyadav227> with this code https://gist.github.com/srinivasyadav18/1f615733a4b8ca9c20bad461cb79b7f8

14:58 <srinivasyadav227> I get this https://gist.github.com/srinivasyadav18/6d4f19bbb359cb5368a3c996d988d02f output from likwid

14:59 <srinivasyadav227> which means, each thread is properly getting initialised by `likwid_thread_init` by thread hook lambda's

15:23 <hkaiser> srinivasyadav227: nice!

15:25 <hkaiser> srinivasyadav227: note, the LIKWID_MARKER_THREADINIT is now called for each pthread HPX creates to execute tasks, while LIKWID_MARKER_START/STOP are invoked for each chunk of the parallel loop (HPX creates one task for each of those)

15:33 <srinivasyadav227> yeah, got it!