Denis Demidov

Not sure why do you get the `simple_cmake_build` failure (could cmake configure go wrong? you could try to clean up the build directory and try to reconfigure), but I think...

Both boost.compute and vexcl vectors are just wrappers around raw opencl buffers, so the conversion is possible both ways.

See examples in ``, which uses Boost.compute algorithms with vexcl vectors: https://github.com/ddemidov/vexcl/blob/ec0cc1fe037f68a9d1c19f56960e92cd820f4f20/vexcl/external/boost_compute.hpp#L60-L102

Probably not, unless someone makes a pull request. CSC is not very effective for matrix vector product, especially on GPUs. And even although vexcl supports CSR as input format, it...

I am sorry, I have little experience with OpenCL to OpenGL interaction. Also, I don't really understand what you are saying here. Could you provide a minimal, single-file example demonstrating...

> ptxas error : Entry function 'merge' uses too much shared data (0xc008 bytes, 0xc000 max) You could try to reduce the block size (and the required shared memory size)...

I am sorry, I can not reproduce this, so I won't be able to help you unless you give me a minimal compilable test-case demonstrating the problem.

I don't remember all the implementation details right away, but it seems it would be possible to do the setup phase in OpenCL. It could even make sense because all...

I think it should be enough to implement the [control lattice](https://github.com/ddemidov/vexcl/blob/c1cebc73128174a4443bf85141cf933808b0ff1c/vexcl/mba.hpp#L320) structure with OpenCL. See the referenced paper for the details of the algorithm. The biggest problem is that it...

I don't like the idea of keeping separate (but very similar) kernels when they all may be generated from a single source.