jkelling

Btw.: I can reproduce this issue locally with clang main (14) and in the CI with clang 13. I can *not* reproduce it locally with clang 11 for some reason...

Checking `omp_get_num_threads()` would make the check never trigger because this would always return `1` outside of a parallel region, so it would become pointless. I guess the intention of the...

Btw.: `throw`ing in a `single` region is only allowed if the exception is also caught in the same region. As is exception does not appear to be intended to be...

Spawning a non-OpenMP thread that does not itself use any OpenMP (directives or runtime API functions) may still be UB (not where to look for an explicit statement about this,...

There is, btw. a big caveat to adding a queue bases on OpenMP tasks: An non-blockig queue would require OpenMP 5.0's tasks. OpenMP 4.0 already has `depend` which would be...

While you are at it: Could you please change the build type of check with enabled sanitizer which are currently `Release` to `RelWithDebInfo` to make the printed backtraces more useful?...

> However, it is not yet clear to me how this can improve things. Yes, we could simply execute multiple blocks in parallel. However, this restricts the block size to...

Btw. OS threads sharing data in a block are probably just as likely to slow each other down due to false sharing (causing each others caches to be flushed) as...

> > Btw. OS threads sharing data in a block are probably just as likely to slow each other down due to false sharing (causing each others caches to be...

It is good to know, that a VEX_FUNCTION can use a view. Maybe I could use this in some places. * in one place I need to use a kernel,...