Mojo GPU Syntax or specific GPU stdlib data types?
I'm looking forward to using MAX, I'm curious though outside of MAX how will the GPU syntax look? Will we need to use different data types other than SIMD that will be specific for GPU's like what is kind of done with cuda in C++? I'm hoping to be able to test this outside of AI paradigms that I couldn't use MAX for, like maybe improving the speed of opencv-like smoothing kernels or ray tracing.
3 Replies
not yet out I belive a similar question "Tutorial / Example of targetting GPU with llvm_instrinsic or __mlir_op etc?" was asked and the answer was that nothing is really out yet for the public
Whatever Mojo ends up exposing syntatically, it's likely going to be a veneer over top of https://mlir.llvm.org/docs/Dialects/GPU/
'gpu' Dialect - MLIR
Multi-Level IR Compiler Framework
i’ve also seen some example code / talks about llvm.nvvm functions, which you can call right now with llvm_instrinsic