HON’s Wiki # Clang/LLVM

Home / Software Engineering

Usage

Options

Platform

Hardware Offloading

OpenMP Offloading

NVIDIA GPUs (CUDA/NVPTX)

LLVM is an extensive compiler platform and toolchain. It’s typically used as the compiler backend, with Clang as the frontend.

Usage

TODO Compilation (C++): g++ ${CPPFLAGS} ${CXXFLAGS} -c -o file1.o file1.cpp
TODO Linking (C++): g++ ${LDFLAGS} ${LDLIBS} -o app file1.o file2.o
Show supported targets

Specify target CPU platform: -target <platform>
- E.g. -target x86_64-unknown-linux-gnu.
- This is generally needed only if cross-compiling, the default will be a target fitting the local computer.
- Use llc --version to show supported platforms.

Specify target OpenMP GPU platform/arch: -fopenmp -fopenmp-targets=<platform> -Xopenmp-targets=<platform> -march <arch>
- Requires an LLVM-based compiler with proper OpenMP offloading support, like AMD ROCm’s LLVM.
- E.g. -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx1030 (AMD RX 6800 XT).
- E.g. -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda -Xopenmp-target=nvptx64-nvidia-cuda -march=sm_70 (NVIDIA V100)
Emit optimization diagnostics during compilation:
- -Rpass=...: When a pass makes a transformation.
- -Rpass-missed=...: When a pass fails a transformation.
- -Rpass-analysis=...: More info about whether or not to make a transformation.
- The options take a regex to match which optimization pass to show (use '.*' to match all).
Show runtime debug info: LIBOMPTARGET_INFO=1