NVIDIA To Focus On LLVM Upstream For Further Fortran/Flang Development

NVIDIA by way of their GPU compute / CUDA Fortran interests and having acquired the PGI compiler company nearly a decade ago has been active contributors to the LLVM Fortran scene. NVIDIA spearheaded the work on the modern LLVM Fortran compiler support and worked with other vendors and the open-source ecosystem on the since-upstreamed FLANG compiler. NVIDIA had been maintaining a "fir-dev" downstream for their latest Fortran compiler patches while now moving forward they will be focused on upstream LLVM contributions.

NVIDIA engineers have been active with the fir-dev branch of the f18-llvm-project repository for Flang even after the Flang Fortran front-end was upstreamed in 2020, but moving forward they are going to be focused on LLVM upstream rather than this code fork. (The "FIR" naming in this context is for Flang's Fortran IR, a dialect of the MLIR IR.)

It was announced today that their fir-dev branch of 18-llvm-project will be ending. As of today, NVIDIA will transition to contributing directly to the LLVM project.

The fir-dev code will remain publicly available and they intend to work with the open-source community on upstreaming the relevant patches into the LLVM project.

From today's announcement, it's great seeing NVIDIA contributing more to an upstream-first approach for their Fortran compiler work to LLVM. NVIDIA's Steve Scalpone commented in the announcement, "NVIDIA will continue to contribute to llvm-project/flang with the same dedication that we have in the past. Our commitment to flang is unwavering."