Initial OpenACC Code Begins Landing In The LLVM Clang Compiler

As noted a few weeks back, NVIDIA is working to add OpenACC support to the upstream LLVM Clang compiler for this parallel computing standard. Today that work began landing in LLVM/Clang's development codebase.

NVIDIA is big on the OpenACC parallel computing standard for targeting GPUs with C / C++ / Fortran code particularly for science domains. While there has been the out-of-tree Clang-based CLACC for supporting OpenACC, NVIDIA is working to add an upstream OpenACC 3.3 implementation to Clang.

Merged this morning are the initial Clang commits for handling OpenACC. With this Clang now picks up on the "-fopenacc" argument and other elements of the OpenACC spec have begun to be implemented. This isn't a complete OpenACC implementation yet but it's great seeing the very first bits beginning to land and just a matter of time now before all of it will be built up.

These initial bits are in mainline for what will be LLVM 18.0 while it remains to be seen how far along the OpenACC implementation will be by the time LLVM/Clang 18 debuts in the early months of 2024.