In reply to @bluewhale_unkown_x "question  did you": No HIP or Metal because,
I  develop on an Nvidia machine on Linux mostly.

I have Mac as well but no time yet to test Metal + I need to setup my dev env
for GPU.

And for HIP, I can only try on my iGPU on a laptop/mini PC because no AMD
discrete GPU at home. But it shouldn't be too bad. hipcc can reuse cuda code
almost verbatim and I did more complex stuff via LLVM and in-memory linking
here:https://github.com/mratsim/constantine/blob/master/constantine%2Fmath_comp
iler%2Fcodegen_amdgpu.nim