The Clang front-end is now able to target the AMDGPU (formerly known as R600) toolchain with its LLVM back-end that has long been in development. The small commit sets the AMDGPU target, sets the default integrated assembler, and uses lld for linking.
Targeting AMDGPU from Clang would come down to using these switches "-target amdgcn--amdhsa -x assembler -mcpu=kaveri." Of course, there's still more work going on by AMD's team for making the open-source HSA/OpenCL support fully useful to end-users/developers.
The code was pushed with this Git commit and will appear with LLVM/Clang 3.7.
8 Comments
