Jan Stephan

The way we handle CUDA has changed over the past months. Is this still an issue?

I think this may also be related to #966.

Reviving this ancient issue. > There should not be direct access to memory buffers. I agree! SYCL introduced memory accessors which hide the pointers quite nicely and provide a clean...

I've been giving this some thought since this is (kind of...) a requirement for the SYCL backend. Here are some ideas I'd like your opinions on: 1. Replace `alpaka::allocBuf` with...

> What is the rational for renaming the function? I think `allocBuf` is a good name. Because there is not necessarily an allocation taking place: Both sub-buffer creation and taking...

@bernhardmgruber and I just had a VC where we also addressed this issue. A short summary: * We need to be careful that we do not introduce more code bloat....

Yesterday we had a longer video conference about this issue. A few points from my notes: ### Current shortcomings * We agreed on pointers being passed to kernels is something...

We will implement this for alpaka 0.7.0. Assigned to @j-stephan.

> What about FPGAs? > @j-stephan Is there anything in the SYCL standard about math functions? SYCL uses OpenCL's math functions and places them into the `sycl` namespace. I don't...

I played around with some SYCL implementations. Test program: ```c++ #include #include #include #include class kernel; auto main() -> int { try { auto queue = cl::sycl::queue{cl::sycl::default_selector{}}; auto buf =...