Targeting different architectures with Triton

[AgamemnonasKyriazis]

Hello,

I'm considering the development of a compilation pipeline for Triton to target parallel architectures like FPGAs.
However, Triton's current design seems tightly coupled with GPUs and relies heavily on CUDA-related structures, which could be an issue for adapting it for FPGA targets or even invalidate the whole idea.

• Are there any existing discussions around extending Triton to support other parallel architectures that I am not aware of?
• What are your views and thoughts around the idea of targeting other parallel architectures ?