iohann.s.titov@gmail.com Ivan S. Titov AMD's LLVM fork, as shipped with ROCm. Built from the same ROCm/llvm-project source archive that dev-libs/rocm-device-libs, dev-libs/rocm-comgr and dev-util/hipcc already consume, so it adds no new distfile. This exists because the ROCm math libraries cannot be compiled by a vanilla LLVM: AMD's fork carries device-code changes that upstream LLVM does not have, notably bf16 signatures for the WMMA intrinsics and AMDGPU assembler support for source modifiers that vanilla rejects. Installed to a private prefix so it never competes with llvm-core/llvm. ROCm/llvm-project https://github.com/ROCm/llvm-project/issues