Denis Demidov

This line looks very suspicious for me: ``` :0:0: in function vexcl_vector_kernel void (i64, double addrspace(1)*, double, double addrspace(1)*): unsupported call to function get_global_id ``` Are you able to compile...

Can you share the source of `compiler_bug.cpp` here?

In that case its a known issue (see #92). That issue have never been resolved as far as I know. I would raise a ticket with Apple support, since the...

Thank you for the info! If you are able to find a workaround, I would be glad to either introduce it to vexcl or accept a pull request.

I think Apple's OpenCL implementation does not support workgroups of more than one workitem on CPUs (vexcl uses this restriction for kernels on CPU devices), but in the gist the...

I think it would make more sense to run unit tests distributed with vexcl. You can do this with ``` cd $VEXCL_ROOT mkdir build cd build cmake .. make OCL_DEVICE=i7...

So the failing kernels I can see are: - In vector/multivector arithmetics: ``` .cpp #if defined(cl_khr_fp64) # pragma OPENCL EXTENSION cl_khr_fp64: enable #elif defined(cl_amd_fp64) # pragma OPENCL EXTENSION cl_amd_fp64: enable...

I don't see anything wrong with this kernel, and it does work with any other OpenCL platform I have access to. I believe nothing left here but opening an issue...

Re fft optimization: I don't think it makes sense when fftw is available. On a CPU one can just map the device memory to a host pointer and then use...

Your `result` vector is [100 times less](https://gist.github.com/lajash/1645b473676633b35d9e#file-stdev-cpp-L18) in size than `A`. So you should get an out of boundary error and a segfault [here](https://gist.github.com/lajash/1645b473676633b35d9e#file-stdev-cpp-L55). Stencil operators use slow path when...