Michal Babej

I've added a regression test with the kernel from the description. Unless i made a mistake in the test, this issue seems fixed in PoCL's main branch.

I agree that it should not abort, however IMO the proposed solution is too over-complicated with very little payoff.... Should be sufficient to return a failure from inside the build-program...

Yep the builtin library code for PoCL-CUDA is different from the builtin library of the CPU drivers, and it's very likely both incomplete and non-compliant. Should be possible to check...

> A redesign with import targets would make this error impossible Indeed, most of the CMake code is old and doesn't use targets. However any redesign must also keep in...

Thanks for the effort, however this won't work as it is. Let me try to explain. As you have noticed, commit 4500774f4af2a134e027169 removed these from `lib/kernel/host`. In short, the problem...

Actually after re-reading the bugreport, ignore my suggestion about triple-underscore names, they only appear in the .so object files, and with the CPU driver this should never happen. Seems like...

I'm guessing this has more to do with QEMU rather then Ubuntu 18.04. Hypervisors tend to mess with the CPUID flags, and both PoCL & LLVM use these to detect...

Tested on two different PCs with current `release_4_0` branch, with PoCL's CTS fork, cannot reproduce. Can you reliably reproduce this failure ? If yes, which hardware / software (LLVM version...

Suggestion: instead of trying to assign a new "meaning" to the acronym .. stop using it as an acronym, start using it as a name. SYCL does the same thing...

This seems like a different issue than #1196. Do you have a backtrace ? Screenshots aren't too useful, but i see one problem: `dev->dispatch` is `0x1` - not a valid...