search

ddemidov / vexcl

VexCL is a C++ vector expression template library for OpenCL/CUDA/OpenMP
http://vexcl.readthedocs.org
MIT License
701 stars 81 forks source link

vector_arithmetics, multivector_arithmetics tests fail on MacOSX #92

Closed ddemidov closed 6 years ago

ddemidov commented 10 years ago

See #90, #91, https://github.com/ddemidov/vexcl/commit/a64811d21c8d429e96268777990d6012b0332b6c#commitcomment-5205979

ddemidov commented 10 years ago

@sebsjoberg, does https://gist.github.com/ddemidov/8681608 compiles and runs successfully? It uses plain C++ bindings and should only output the name of your compute device. In case kernel compilation fails, there will be errors.

In case it does not work, I would use it for a bug report with Apple support.

sebsjoberg commented 10 years ago

@ddemidov, it doesn't run successfully. I tested changing the min expression to just assign zero and then it worked. I guess it's Apple support next.

I tried marking the testcases as expected failures but it didn't seem to work, probably because it's not the BOOST_CHECK* macros that fails but rather a exception before that.

ddemidov commented 10 years ago

I think there are several people that successfully use VexCL on MacOSX (I don't have a Mac to test this), so the problem could be just with your system. So I don't want to remove the test for all Apple systems yet.

ddemidov commented 10 years ago

But you could just enclose the failing tests with 

#ifndef __APPLE__
...
#endif

in your local copy (or just remove the tests altogether).

ddemidov commented 10 years ago

Better yet, cce38c6 gets rid of min altogether, so all tests should pass for you now.

sebsjoberg commented 10 years ago

@ddemidov, actually that fails too. I'll just assume that my computer is at fault and get back to you regarding this issue when I've done a clean reinstall.

ddemidov commented 10 years ago

Do the same tests fail? I have also updated https://gist.github.com/ddemidov/8681608. Does that fail as well?

sebsjoberg commented 10 years ago

Yes the same tests fail and the code in the gist does fail too.

ddemidov commented 10 years ago

@sebsjoberg , have you been able to try this with another machine?

sebsjoberg commented 10 years ago

@ddemidov, no not yet. I did however update to OS X Mavericks but the error is still present. I also tested changing the cl.hpp header version from 1.2 to 1.1 but there was no difference.

ddemidov commented 10 years ago

That is strange, because this is apparently the only kernel that fails for you (and the rest of the kernels use the same pattern). Please inform me if you are able to rephrase the kernel source to prevent compiler from crashing.