CuPBoP is a framework designed to execute unmodified CUDA source code on non-NVIDIA devices. It currently supports several CPU backends, including x86 and AArch64, as well as GPU backends such as AMD GPUs and the open-source RISC-V (Vortex). This repository specifically focuses on enabling CUDA execution on Vortex GitHub link.
Please look at the following files to understand the basics of CuPBoP:
Basic Definitions and Concepts
- Linux: Verified on Ubuntu 18.04 / 20.04
- LLVM 18.1.7 (Vortex extended LLVM)
- NVIDIA CUDA Toolkit: Verified on CUDA 12.1.1
- x86 CPU
- GCC: Verified with 7.5.0, 10.5.0
- CMake (>= 3.10)
- System libraries (Ubuntu/Debian):
libtinfo-devzlib1g-dev
- Vortex 2.0: Verified with commit on Aug 2, 2025
If you have access to the Kia or Volvo (server that needs GT credentials to access), follow the shared Notion setup guide.
If not, follow this guide instead: Non Kia/Volvo Setup
Please look at File Directory Setup to understand how your codespace should look like in the end and how to set up environment variables.
Test benchmarks (table updated on May 14, 2025)
CuPBoP includes a subset of Rodinia/Hetero-Mark benchmarks to test functionality. Benchmarks supported with Vortex 2.0 are available in the /example folder, each with a kjrun_llvm18.sh file.
While you can run a benchmark by running cd/Cupbop/examples/[benchmark name] & ./kjrun_llvm18.sh, you will generate a lot of files and directories in the process. It would take some time to remove all newly generated files and directories and there is a chance one can accidentally delete the wrong file(s). Please use the makefile in the \examples folder and you can refer to this guide: Makefile Guide
Take a look at this list to identify which benchmarks are and aren't supported.
If you want to add more RODINIA benchmarks to the examples/ folder, please refer to the following guide to help you: Benchmark Creation
Refer to this guide to figure out some issues you might be facing: Common Issues
Feel free to add to this guide if you fixed any issues so it can be helpful for anyone in the future!
Any kinds of contributions are welcome.
Please refer to Contribution.md for more detail.