CuPBoP: Cuda for Parallelized and Broad-range Processors

Introduction

CuPBoP is a framework which support executing unmodified CUDA source code on non-NVIDIA devices. Currently, CuPBoP support serveral CPU backends, including x86, AArch64, and RISC-V. Supporting the RISC-V GPU Vortex is working in progress.

Install

Prerequisites

• Linux system
• LLVM 14.0.1

Installation

1. Clone from github

   git clone --recursive https://github.com/drcut/CuPBoP
   cd CuPBoP
   export CuPBoP_PATH=`pwd`
   export LD_LIBRARY_PATH=$CuPBoP_PATH/build/runtime:$CuPBoP_PATH/build/runtime/threadPool:$LD_LIBRARY_PATH
   

2. As CuPBoP relies on CUDA structures, we need to download the CUDA header file

   wget https://www.dropbox.com/s/r18io0zu3idke5p/cuda-header.tar.gz?dl=1
   tar -xzf 'cuda-header.tar.gz?dl=1'
   cp -r include/* runtime/threadPool/include/
   

3. Other CUDA files are also required for compiling CUDA source code to LLVM IR

   wget https://www.dropbox.com/s/4pckqsjnl920gpn/cuda-10.1.tar.gz?dl=1
   tar -xzf 'cuda-10.1.tar.gz?dl=1'
   

4. Build CuPBoP

   mkdir build && cd build
   cmake .. -DLLVM_CONFIG_PATH=`which llvm-config` # need path to llvm-config
   make
   

Run Vector Addition example

cd examples/vecadd
# Compile CUDA source code (both host and kernel) to bitcode files
clang++ -std=c++11 vecadd.cu \
      -I../.. --cuda-path=$CuPBoP_PATH/cuda-10.1 \
      --cuda-gpu-arch=sm_50 -L$CuPBoP_PATH/cuda-10.1/lib64 \
      -lcudart_static -ldl -lrt -pthread -save-temps -v  || true
# Apply compilation transformations on the kernel bitcode file
$CuPBoP_PATH/build/compilation/kernelTranslator \
      vecadd-cuda-nvptx64-nvidia-cuda-sm_50.bc kernel.bc
# Apply compilation transformations on the host bitcode file
$CuPBoP_PATH/build/compilation/hostTranslator \
      vecadd-host-x86_64-unknown-linux-gnu.bc host.bc
# Generate object files
llc --relocation-model=pic --filetype=obj  kernel.bc
llc --relocation-model=pic --filetype=obj  host.bc
# Link with runtime libraries and generate the executable file
g++ -o vecadd -fPIC -no-pie \
      -I$CuPBoP_PATH/runtime/threadPool/include \
      -L$CuPBoP_PATH/build/runtime  \
      -L$CuPBoP_PATH/build/runtime/threadPool \
      host.o kernel.o \
      -I../.. -lpthread -lc -lx86Runtime -lthreadPool
# Execute
./vecadd

How to contribute?

Any kinds of contributions are welcome. Please refer to Contribution.md for more detail.

Related publications

• COX: Exposing CUDA Warp-Level Functions to CPUs ACM Transactions on Architecture and Code Optimization link
• CuPBoP: CUDA for Parallelized and Broad-range Processors arxiv preprint link

Contributors

• Ruobing Han
• Jun Chen
• Bhanu Garg
• Xule Zhou
• John Lu
• Chihyo Ahn
• Haotian Sheng
• Blaise Tine
• Hyesoon Kim