About the plan for CUDA Fortran support in Flang

I’ve seen that @klausler has done great work on supporting CUDA Fortran in Flang: rGf674ddc19fee. These patches add the initial support for CUDA’s extension to Fortran, including parsing and semantics analysis. I’m familiar with Fortran and have some experience with compilers for GPU so I would like to contribute to this project. But I don’t know the roadmap for this goal. I wonder what I can do to help. So I have these questions to

  1. What are the next steps? Could I help with any open issues for CUDA Fortran support in Flang?
  2. Will the CUDA Fortran in Flang also support other vendors’ GPUs like AMD, Intel, and others…
  3. How will the IR be like? It’s natural that Flang maps CUDA Fortran’s GPU related operations to MLIR’s gpu dialect.
  4. Is the backend work waiting for Flang runtime for offloading to device?
1 Like