My MLIR Track #1 - Installation

**Disclaimer** : This is a documentation of my learning process only. Following these steps does not guarantee identical results. In the beginning, LLVM was developed to investigate dynamic compilation techniques for both static and dynamic programming languages. However, now day LLVM severs as portable, high-level assembly language, called language-independent intermediate representation (IR) to optimize the computation [1]. MLIR (Multi-Level Intermediate Representation ) was developed in 2018 by Google and released as a sub-project of LLVM in 2019. It is developed to manage the complexity of heterogeneous (hybrid) architectures, enabling precise orchestration of computation loading and memory usage across diverse hardware targets[2]. In order to delicately manage heterogeneous architectures, working on the compiler becomes now an essential pain to truly orchestrate data movement and workload balance. ## Installing MLIR It needs some dependencies to install the MLIR. ```bash # Ubuntu sudo apt update sudo apt-get install cmake ninja-build ccache sudo apt install clang lld ``` Next, It has to clone the LLVM repository. MLIR is built directly within it, and the download may take a while due to the repository’s size. ```bash git clone https://github.com/llvm/llvm-project mkdir llvm-project/build cd llvm-project/build ``` ## CMake Configuration Flags ```bash cmake -G Ninja ../llvm \ -DLLVM_ENABLE_PROJECTS=mlir \ -DLLVM_BUILD_EXAMPLES=ON \ -DLLVM_TARGETS_TO_BUILD="Native;ARM;X86" \ -DCMAKE_BUILD_TYPE=Release \ -DLLVM_ENABLE_ASSERTIONS=ON \ -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \ -DLLVM_CCACHE_BUILD=ON -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON ``` - ```-G Ninja``` : Faster build system selection. Chooses the Ninja build system for faster compilation. - ```-DLLVM_ENABLE_PROJECTS=mlir``` : Compiles MLIR along with LLVM. - ```-DLLVM_BUILD_EXAMPLES=ON``` : Builds official MLIR/LLVM examples for reference. - ```-DLLVM_TARGETS_TO_BUILD="Native;ARM;X86"``` : Target CPU support. Specifies target architectures: ‘Native’ for the host CPU, plus ARM and X86 support. - ```-DCMAKE_BUILD_TYPE=Release``` : Release optimized mode for performance and much smaller in size - ```-DLLVM_ENABLE_ASSERTIONS=ON``` : If any error while running, it will print a error massage. - ```-DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++``` : Choose Clang to bild the entire system - ```-DLLVM_CCACHE_BUILD=ON``` : With ccache enabled, LLVM/MLIR will remember the already-finished parts, making it much faster if you ever need to recompile. - ```-DCMAKE_BUILD_WITH_INSTALL_RPATH=ON``` : Ninja builds the porjects and save to the output files to the current directory, llvm-project/build ## Verify the Build. It take hours depending on the native CPU. ```bash cmake --build . --target check-mlir sudo cmake --build . --target install ``` - ```cmake — build . — target check-mlir``` : Verifying the Build and it take 5–10 minutes depending on the native CPU. - ```cmake — build . — target install ``` : It copies the finished executable files, headers, and libraries from the local build directory to ```/usr/local/binor``` the path specified by ```CMAKE_INSTALL_PREFIX```. ## Check the installation of MLIR ```bash mlir-opt --version ``` Then. it shows the version of LLVM. ```bash LLVM (http://llvm.org/): LLVM version 23.0.0git Optimized build with assertions. ``` Here the LLVM and MLIR installation is done and it is ready to go programming with MKIR. ## Reference [1] [LLVM Language Reference Manual](https://llvm.org/docs/LangRef.html) [2] Lattner, C., Pienaar, J., et al. (2020). “MLIR: A Compiler Infrastructure for the End of Moore’s Law.” Published in the 2021 IEEE/ACM International Symposium on Code Generation and Optimization (CGO).[arXiv:2002.11054](https://arxiv.org/abs/2002.11054) [3] [Introduction to MLIR]( https://www.stephendiehl.com/posts/mlir_introduction/) ## 純粹抱怨、碎碎念 >因為老黃不想遵守大家一起訂出來的工業標準，所以現在大家為了GPU通解得降到更底層的編譯器 / 組合語言去工作。這不是我們愛找麻煩，是環境逼我們的。