LLVM Flang Begins Seeing NVIDIA CUDA Fortran Support

The LLVM Fortran compiler "Flang" has begun seeing NVIDIA CUDA support land in the upstream code-base.

NVIDIA engineer Peter Klausler has contributed the initial CUDA Fortran pieces to the codebase for LLVM 17 Git. Merged this week included the pieces:


- [flang] CUDA Fortran - part 1/5: parsing
- [flang] CUDA Fortran - part 2/5: symbols & scopes
- [flang] CUDA Fortran - part 3/5: declarations checking
- [flang] CUDA Fortran - part 4/5: definability and characteristics
- [flang] CUDA Fortran - part 5/5: statement semantics

With that all of the initial NVIDIA CUDA Fortran pieces are in upstream Flang now. Though there is some related code still outstanding like experimental device builds of the Flang runtime for handling the Fortran run-time as a CUDA or OpenMP offload target library.

The Fortran programming language remains a popular choice in some scientific communities and NVIDIA continues investing in CUDA support around Fortran with LLVM playing an important role in their proprietary compute ecosystem.