FLANG: NVIDIA Brings Fortran To LLVM
Flang is to Fortran as Clang is to C/C++.
No announcement has yet to hit the wire, but NVIDIA has made public their code repository for Flang, a Fortran front-end to LLVM. Developers have been looking forward to Fortran support for LLVM and now it's available in the form of Flang. Up to now, LLVM fans wanting Fortran have mostly been using the unmaintained DragonEgg plug-in to pair LLVM with GCC's Fortran support.
This Flang support has been a long time coming: back in 2013 was a GSoC proposal for Flang as an LLVM Fortran compiler while in 2015 is when it was announced that NVIDIA would develop a Fortran LLVM compiler in conjunction with the US Department of Energy's National Nuclear Security Administration.
Last we heard of Flang was a NVIDIA/PGI talk last November where they talked about its development while this week the code has finally surfaced, at least unofficially as we have yet to see any official NVIDIA announcement hit the wire.
The code to NVIDIA's Flang can be found via flang-compiler on GitHub. When ready, hopefully they will pursue making it being an official LLVM sub-project. Flang is interesting not just for compiling to target CPUs but paired with LLVM's NVPTX back-end could be another option for those wanting to write Fortran for NVIDIA GPUs, as an alternative to the proprietary PGI/NVIDIA Fortran compiler and more advanced than the existing GCC Fortran offloading capabilities.
No announcement has yet to hit the wire, but NVIDIA has made public their code repository for Flang, a Fortran front-end to LLVM. Developers have been looking forward to Fortran support for LLVM and now it's available in the form of Flang. Up to now, LLVM fans wanting Fortran have mostly been using the unmaintained DragonEgg plug-in to pair LLVM with GCC's Fortran support.
This Flang support has been a long time coming: back in 2013 was a GSoC proposal for Flang as an LLVM Fortran compiler while in 2015 is when it was announced that NVIDIA would develop a Fortran LLVM compiler in conjunction with the US Department of Energy's National Nuclear Security Administration.
Last we heard of Flang was a NVIDIA/PGI talk last November where they talked about its development while this week the code has finally surfaced, at least unofficially as we have yet to see any official NVIDIA announcement hit the wire.
The code to NVIDIA's Flang can be found via flang-compiler on GitHub. When ready, hopefully they will pursue making it being an official LLVM sub-project. Flang is interesting not just for compiling to target CPUs but paired with LLVM's NVPTX back-end could be another option for those wanting to write Fortran for NVIDIA GPUs, as an alternative to the proprietary PGI/NVIDIA Fortran compiler and more advanced than the existing GCC Fortran offloading capabilities.
26 Comments