On Tue, Apr 4, 2023 at 7:45 PM Mark Sibly <marksi...@gmail.com> wrote:
> I'm building a 3rd party c++ lib in emscripten that uses pthreads, and I'm > finding that I have to use -sPTHREAD_POOL_SIZE=23 for it to work, otherwise > it hangs. > > I did think leaving out -sPTHREAD_POOL_SIZE (or setting it to 0) would > create threads on demand, but I just found this on the emscripten pthreads > docs page and suspect this is what's happening as the lib's API is just a > single function that appears to execute synchronously: > > "you cannot call pthread_create and then keep running code synchronously > that expects the worker to start running" > > So OK, I'll need to manually specify a pool size, but what value should I > use? The lib bases it's thread usage on std::thread::hardware_concurrency() > (and seems to be a bit off as I have 16 but need a pool size of 23) but I > can't know that at link time so what should I use? I'll also need to hard > code this as a maximum into the lib of course but that's OK. Is there a way > to determine PTHREAD_POOL_SIZE at runtime? > You can specify it during startup, so you can tell it to use the browser's thread count (which is reflected to C++ in thread::hardware_concurrency()). See this note: https://github.com/emscripten-core/emscripten/blob/b2f3b50b0291ed6b0e3058194b1b6db927ab4c33/src/settings.js#L1547-L1553 > > Can I just use 256 or something crazy, or would that break on some low > specced computers, ie: how 'heavyweight' are these WebWorkers? I'm sort of > feeling an '8' here for some reason... > Each worker is a full JS context, so it does take several MB at least. It might make sense to tell the application to use a maximum of say 8 or so, yeah, unless it can really benefit from more... > > One other thing, what exactly does -sPTHREAD_POOL_SIZE_STRICT do? I'm > finding setting it to '0' causes my app to hang if I haven't set thread > pool size high enough, and setting it to '2' causes an exception to be > thrown instead. > > So I've just set it to '2', but am I missing something with '0'? Are there > any other useful values? Note that there's nothing about > -sPTHREAD_POOL_SIZE_STRICT on the emscripten pthreads docs page. > > The lower-level docs have that info I believe: https://github.com/emscripten-core/emscripten/blob/b2f3b50b0291ed6b0e3058194b1b6db927ab4c33/src/settings.js#L1557-L1572 Finally, the lib makes quite complex use of threads - there are > std::futures and std::promises all over the place and I initally thought > I'd never get it going in emscripten, but once I found the > PTHREAD_POOL_SIZE fix everything just worked! This is a very impressive > achievement for emscripten IMO! > > Great! :) Yeah, there are quite a lot of weird things with how the Web does threads, and so we have to use a lot of hacks and things, but often a lot of code ends up just working with one or two settings adjustments, thanks to a great deal of work that's gone into this... - Alon > Bye! > Mark > > -- > You received this message because you are subscribed to the Google Groups > "emscripten-discuss" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to emscripten-discuss+unsubscr...@googlegroups.com. > To view this discussion on the web visit > https://groups.google.com/d/msgid/emscripten-discuss/db07f3cc-0080-4c1f-946c-d57a3fef4cd5n%40googlegroups.com > <https://groups.google.com/d/msgid/emscripten-discuss/db07f3cc-0080-4c1f-946c-d57a3fef4cd5n%40googlegroups.com?utm_medium=email&utm_source=footer> > . > -- You received this message because you are subscribed to the Google Groups "emscripten-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to emscripten-discuss+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/emscripten-discuss/CAEX4NpQUr0v27%3Dc71BMi%3D6D9R1Q9g1uREKpTXA4iTLSX-wCsKA%40mail.gmail.com.
