LLM for Compiler - chenglong92/LLM4Compiler GitHub Wiki

1. Triton History

The original Triton architecture is depicted in Figure 1. It is firstly proposed by Phiiippe Tillet in his PHD thesis - "Blocked Algorithms for Neural Networks: Design and Implementation on GPUs".

image{#fig:triton_ori}

图片描述 Triton https://www.kapilsharma.dev/posts/deep-dive-into-triton-internals/

2. Installation Tutorial on Mac(CPU)

Step 1: download triton-cpu

git clone https://github.com/triton-lang/triton-cpu.git

Step 2: install LLVM-project

git clone https://github.com/llvm/llvm-project.git
cd llvm-project
mkdir build
mkdir install
export LLVM_HOME=$PWD
cd build
cmake -G Ninja  -DCMAKE_C_COMPILER=/usr/bin/clang -DCMAKE_CXX_COMPILER=/usr/bin/clang++ -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi" -DLIBCXXABI_USE_LLVM_UNWINDER=OFF -DLLVM_ENABLE_RTTI=ON -DLLVM_ENABLE_EH=ON -DCMAKE_BUILD_TYPE=Release -DCLANG_DEFAULT_CXX_STDLIB=libc++ -DDEFAULT_SYSROOT=$(xcrun --show-sdk-path)  -DCMAKE_INSTALL_PREFIX=/Users/chenglong/workspace/Deps/llvm-project/install ../llvm

3. Datasets

[KernelBench]: triton-based, used in KernelLLM, https://github.com/ScalingIntelligence/KernelBench/tree/main [TritonBench]: https://github.com/thunlp/TritonBench

⚠️ **GitHub.com Fallback** ⚠️