Denis Demidov

Could you provide a working openmp prototype for your idea? Just for the main loop over data points on page 4 with some random input.

I was not saying that you are wrong, I merely asked a working prototype of your parallelization idea to see what effect on performance and memory footprint it would make....

An openmp implementation should be enough. Also, you don't need to implement the full BA algorithm. A parallelization of the loop on p.4 is enough.

So you will need to run the strided loop in (b) `4^ndim` times where `ndim` is the number of dimensions (2 in your case). I suspect that each run would...

Also, could you please enclose the code snippets with code markers as in ~~~{.cpp} int foo = 1; ~~~ This would increase readability and look like this: ``` .cpp int...

Yes, the docs are supposed to be built from full git tree. The docs are built on readthedocs (https://vexcl.readthedocs.io/en/latest/), so may be it is enough to point there? It should...

This looks good. I've added sphinx.ext.autodoc to docs/conf.py to save you one patch: 626aab93d5931cac2dd8887e625ea1bddf7611d9.

I've been thinking about this for some time and I am not sure its possible to provide a useful abstraction over constant memory that would work both for OpenCL and...

It should be still possible to use custom kernel with constant memory in CUDA.

I think I have VS 2013 installed on my Windows test machine. Will look into this tomorrow.