Vortex-Optimized Light-weight Toolchain (VOLT)

VOLT is an end-to-end, open-source compiler framework for the Vortex open-source GPU and its variants. It supports multiple front-end languages, including CUDA and OpenCL, by extending two open-source projects, PoCL and CuPBoP. VOLT also extends LLVM to enable SIMT-aware IR-level optimizations for portability across GPU variants, and augments LLVM-RISC-V to provide full support for the Vortex ISA.

The figure shows an overview of VOLT, with the modules extended or introduced by VOLT highlighted in blue.

Citation and Presentation Materials

Vortex Tutorials

• https://vortex.cc.gatech.edu/
• https://github.com/vortexgpgpu/vortex_tutorials/tree/main

VOLT paper

The VOLT paper was accepted to CC 2026 and will appear in January 2026.

Arixv Paper (link)

@misc{jeong2025insidevoltdesigningopensource,
      title={Inside VOLT: Designing an Open-Source GPU Compiler}, 
      author={Shinnung Jeong and Chihyo Ahn and Huanzhi Pu and Jisheng Zhao and Hyesoon Kim and Blaise Tine},
      year={2025},
      eprint={2511.13751},
      archivePrefix={arXiv},
      primaryClass={cs.DC},
      url={https://arxiv.org/abs/2511.13751}, 
}

Build Toolchain (Vortex v2, 64-bit)

1. Folder structure

<VOLT>
├── tools # location of libraries and bin files
│   ├── libc32
│   ├── libc64
│   ├── libcrt32
│   ├── libcrt64
│   ├── llvm-vortex
│   ├── pocl
│   ├── riscv32-gnu-toolchain
│   ├── riscv64-gnu-toolchain
│   ├── sv2v
│   ├── verilator
│   ├── vortex
│   └── yosys
├── pocl 
├── llvm-vortex
└── vortex

2. Set up folders

cd Volt
mkdir tools 

3. Make a config file (add to the bash file)

  export PRJ=<path-to-volt-folder>
  export TOOL_DIR=$PRJ/tools
  export VORTEX_PREFIX=$TOOL_DIR/vortex
  export LLVM_PREFIX=$TOOL_DIR/llvm-vortex
  export POCL_PREFIX=$TOOL_DIR/pocl
  source $PRJ/vortex/build/ci/toolchain_env.sh

4. Install vortex

cd $PRJ/vortex
sudo ./ci/install_dependencies.sh
mkdir build && cd build

../configure --xlen=64 --tooldir=$TOOL_DIR --prefix=$VORTEX_PREFIX #setup 32bit
./ci/toolchain_install.sh --all # install prebuilt toolchain
source ./ci/toolchain_env.sh
make -s
make install
./ci/blackbox.sh --cores=2 --app=opencl/vecadd # test

5. Install llvm-vortex

cd $PRJ
mv $LLVM_PREFIX $TOOL_DIR/llvm-vortex-prebuilt 
mkdir $TOOL_DIR/llvm-vortex

cd llvm
mkdir build && cd build

export RISCV_TOOLCHAIN_ELF_PATH=$PRJ/tools/riscv64-gnu-toolchain/riscv64-unknown-elf

cmake -G "Unix Makefiles" -DLLVM_ABI_BREAKING_CHECKS=FORCE_OFF -DLLVM_INCLUDE_EXAMPLES=OFF -DLLVM_INCLUDE_TESTS=OFF -DBUILD_SHARED_LIBS=True -DCMAKE_BUILD_TYPE=RelWithDebInfo -DLLVM_ENABLE_ASSERTIONS=ON -DLLVM_ENABLE_PROJECTS="clang" -DCMAKE_INSTALL_PREFIX=$LLVM_PREFIX -DDEFAULT_SYSROOT=$RISCV_TOOLCHAIN_ELF_PATH -DLLVM_DEFAULT_TARGET_TRIPLE="riscv64-unknown-elf" -DLLVM_TARGETS_TO_BUILD="X86;RISCV;NVPTX" ../llvm
make -j`nproc`
make install

6. Install POCL-Vortex

cd $PRJ
mv $POCL_PREFIX $TOOL_DIR/pocl-vortex-prebuilt 
mkdir $TOOL_DIR/pocl-vortex

cd $PRJ/pocl
mkdir build && cd build

cmake -G "Unix Makefiles" \
  -DCMAKE_BUILD_TYPE=Debug \
  -DCMAKE_INSTALL_PREFIX=$POCL_PREFIX \
  -DWITH_LLVM_CONFIG=$LLVM_PREFIX/bin/llvm-config \
  -DVORTEX_HOME=$VORTEX_HOME \
  -DVORTEX_BUILD=$VORTEX_PREFIX \
  -DENABLE_VORTEX=ON \
  -DKERNEL_CACHE_DEFAULT=OFF \
  -DENABLE_HOST_CPU_DEVICES=OFF \
  -DENABLE_TESTS=OFF \
  -DPOCL_DEBUG_MESSAGES=ON \
  -DENABLE_ICD=OFF ..

make -j`nproc`
make install

7. Install CuPBoP-Vortex

cd $PRJ/cupbop
export CuPBoP_PATH=pwd
export LD_LIBRARY_PATH=$CuPBoP_PATH/build/runtime:$CuPBoP_PATH/build/runtime/threadPool:$LD_LIBRARY_PATH

wget "https://dl.dropboxusercontent.com/scl/fi/m9ap1tiybau4zk720t2z7/cuda-header.tar.gz?rlkey=zmdpst5l66t48ywrbtkj426nu&st=luao6zy7" -O cuda-header.tar.gz
tar -xzf 'cuda-header.tar.gz'
cp -r include/* runtime/threadPool/include/

wget https://developer.download.nvidia.com/compute/cuda/12.1.1/local_installers/cuda_12.1.1_530.30.02_linux.run
mkdir -p cuda-12.1
sh cuda_12.1.1_530.30.02_linux.run --silent --toolkit --toolkitpath=$CuPBoP_PATH/cuda-12.1

$CuPBoP_PATH/CuPBoP_env_setup.sh
export VORTEX_ARCHITECTUE = 64                  # 64 if using 64-bit vortex
mkdir build && cd build
cmake .. -DLLVM_CONFIG_PATH=`which llvm-config` # need path to llvm-config
make 

Extend the config file (add to the bash file)

export RISCV_TOOLCHAIN=${TOOL_DIR}/riscv64-gnu-toolchain # Assume Vortex is setup with 64-bit RISC-V
export VORTEX_HOME= # VORTEX repo path
export CuPBoP_PATH= # CUPBOP repo path
export VORTEX_PATH=$VORTEX_HOME/build
export PATH=$LLVM_PREFIX/bin:$PATH
export LD_LIBRARY_PATH=$CuPBoP_PATH/build/runtime:$CuPBoP_PATH/build/runtime/threadPool:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$CuPBoP_PATH/cuda-12.1/lib64:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=${LLVM_PREFIX}/lib:${LD_LIBRARY_PATH}
export LD_LIBRARY_PATH=${VORTEX_PATH}/runtime/:${LD_LIBRARY_PATH}

CuPBoP Running Test

cd ./exampless/cfd
./kjrun_llvm18.sh

Test Toolchain

0. Test Benchmark Location for Each Pipeline

• OpenCL to Vortex pipeline is in vortex/tests/opencl
• CUDA to Vortex pipeline is in cupbop/examples

1. Test OpenCL Benchmarks with Thread Divergence Optimizations

cd $PRJ/tests/test_opencl_w_divergence_opt
./run_evaluation.sh

2. Test CuPBoP with ISA extensions

cd $PRJ/tests/test_cupbop
./run_isa_extensions.sh

3. Test CuPBoP with Memory Hierarchy Support and API Support

cd $PRJ/tests/test_cupbop
./run_mem_hierarchy.sh