Joachim Meyer

Hi @aaronmondal, thanks for tracking this. As a reference, with a "typical" setup, it seems to work quite fine: https://github.com/fodinabor/hipSYCL/runs/7296288804?check_suite_focus=true https://github.com/fodinabor/hipSYCL/runs/7296289502?check_suite_focus=true I'm not too familiar with bazel and thus don't...

A slightly older, but still active project for mapping OpenCL SPIR-V to Vulkan SPIR-V is https://github.com/google/clspv which is e.g. used by https://github.com/kpet/clvk or [pocl](http://portablecl.org/docs/html/vulkan.html). Iirc, clspv also supports ingesting OpenCL...

Thanks for reporting all those issues, I tried to fix all that were easily testable without ROCm available. Re 1, see #507, Re 3 it seems somewhat difficult to get...

Cheers, you are currently trying to invoke cufft functions from the GPU (the single task may be executed on the GPU). The cufft functions are afaict host functions though. You'll...

Hi, I have the feeling, you might be building hipSYCL on Isambard? When using Isambard with hipSYCL, I ended up building everything manually, including boost and LLVM, not sure that's...

Hey, as commented in the newly created issue on ROCm side (https://github.com/RadeonOpenCompute/ROCm/issues/1680), the issue here is that the LLVM version ROCm uses is not really equal to the LLVM 14...

Hi, using nvc++ is just another way of using Nvidia devices with hipSYCL, where hipSYCL mostly acts as a runtime library. See [this blogpost](https://hipsycl.github.io/hipsycl/release/cpu/extension/nvc++/hipsycl-0.9.2/#immediate-support-for-new-nvidia-hardware-nvc-backend) for more information. The default still...

I don't really know how the ROCm releases work (i.e. which version of Clang they package), but in upstream Clang this [commit](https://github.com/llvm/llvm-project/commit/61d065e21ff37fb9040aed711c97daddac2f7577) should lead to the support of this feature,...

Also.. Why is it, that `.xxx()` is only possible on a `vec` and not on a `vec`?

Hmm... I see your point, but that's not necessarily true, that permutation may only include same size vecs.. other implementations as triSYCL and ComputeCpp at least do support swizzling to...