This tool is used to modify and transform CUDA source codes. This README file provides instructions on how to install, configure, and use the tool.
-
To use this tool, first download and compile the LLVM project.
-
While compiling the LLVM project, clone a repo under the
clang-tools-extra
directory. -
Add the following line to the
CMakeLists.txt
file under theclang-tools-extra
directory:add_subdirectory(CUDAIntegratedTransformerTool)
-
Navigate to the build directory of the LLVM project and create make files with cmake. Enable clang-tools-extra option. You dont't need for this every changes in the code. You can just do step 7 for any source code changes after the first installation.
sudo cmake -DLLVM_ENABLE_PROJECTS="clang-tools-extra;clang" -DCMAKE_BUILD_TYPE=Release -G "Unix Makefiles" ../llvm
-
Finally, navigate to the build directory of the LLVM project and compile the tool by running the following commands. "-j 16 " for building with 16 core. :
sudo make -j 16 install CUDAIntegratedTransformerTool
Samples created for PolyBench-ACC benchmark programs. Samples link https://drive.google.com/drive/folders/1aiotg3w3qJnpWaZLr2So1B73umVrQbQR