Skip to content

LLVM VE RV

Simon Moll edited this page Apr 6, 2020 · 2 revisions

Clang(++) code generation flags

-mno-vevpu disables VPU code generation. Vector code generation is enabled by default.

-mvelintrin enables support for VEL Intrinsics.

OpenMP target offloading flags

-fopenmp-targets=aurora-nec-veort enables VH to VE OpenMP target offloading through sotoc (prototype).

-fopenmp-nec-compiler=<COMPILER> selects the VE compiler to use. COMPILER can be either of clang|rvclang|ncc. In case (rv)clang are used, note that the clang binary needs to be in the PATH.

Using the region vectorizer

Either use rvclang or rvclang++ or invoke clang with -fplugin=libRV.so -mllvm -rv. RV requires loop vectorization pragmas to operate. Please refer to the Main RV wiki for further documentation.

Clone this wiki locally