LLVM build:
git clone https://github.com/llvm/llvm-project.git
mkdir ./build
cd ./build
cmake -G Ninja -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
-DLLVM_ENABLE_PROJECTS="clang;lld;clang-tools-extra;compiler-rt" \
-DLLVM_INSTALL_UTILS=On -DLLVM_TARGETS_TO_BUILD="AArch64;X86" \
-DCMAKE_INSTALL_PREFIX="/opt/llvm" -DLLVM_DEFAULT_TARGET_TRIPLE="aarch64-linux-gnu" \
-DCMAKE_BUILD_TYPE=RelWithDebInfo -DLLVM_OPTIMIZED_TABLEGEN=On -DLLVM_ENABLE_ASSERTIONS=On \
-DLLVM_PARALLEL_LINK_JOBS=1 -DLLVM_BUILD_LLVM_DYLIB=On -DLLVM_LINK_LLVM_DYLIB=On \
-DLLVM_DYLIB_COMPONENTS=all -DLLVM_ENABLE_LTO=Off -DLLVM_USE_LINKER=/usr/bin/ld.lld \
../llvm
Notes on the build:
- The default way of statically linking all LLVM tools like clang results in disk usage of > 100GB for a debug build.
- Using GNU binutils and GCC and the GLIBC for cross compilation is still much easier than using the equivalent LLVM tools.
- To many parallel linking tasks can consume a lot of main memory.
Also required (for libc and linker etc.): aarch64-linux-gnu-gcc etc.
The ./VBV directory has to be placed in ./llvm-project/llvm/lib/Transforms/ and ./llvm-project/llvm/lib/Transforms/CMakeLists.txt
has to be patched like this:
diff --git a/llvm/lib/Transforms/CMakeLists.txt b/llvm/lib/Transforms/CMakeLists.txt
index dda5f6de11e3..489dd160369e 100644
--- a/llvm/lib/Transforms/CMakeLists.txt
+++ b/llvm/lib/Transforms/CMakeLists.txt
@@ -9,3 +9,4 @@ add_subdirectory(Hello)
add_subdirectory(ObjCARC)
add_subdirectory(Coroutines)
add_subdirectory(CFGuard)
+add_subdirectory(VBV)In order to test the code, fix the path to your LLVM build in the Makefile of the test directory and then run make.
The test program axb_main_opt checks if the vectorization was correct (run it using qemu-aarch64 if you are not
running on a AArch64 system or one that does not support SVE).