No OpenCL Support

[mgregoro]

Why are you guys all using Nvidia's proprietary API to do all this cool stuff?!

[asuhan]

I cannot promise anything, but this might change soon. We don't depend on CUDA that much, we have room to add support for other graphic cards. The code generation should work with minor changes, also memory allocation and data transfer should be easy to port to OpenCL. We also rely on Thrust for radix sort, which means we need to roll our own or use the HIP transpiler (https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP).

[m1mc]

This is not as good and handy as thrust::radix_sort yet: https://github.com/KhronosGroup/SyclParallelSTL/blob/master/include/sycl/algorithm/sort.hpp

[gstoner]

We are going to bring MAPD to ROCm platform. We have lot more to talk about soon. We are very excited by the Open Sourcing of MAPD it fits with our philosophy of bringing GPU computing forward.

[asuhan]

@gstoner We're looking forward to this and happy to help with it.

[gstoner]

I look forward to the help

[randyzwitch]

Closing this issue for now, but please feel free to open another issue if this is still a problem.