Rodinia Benchmark Suite

This repository now builds with CMake. Legacy Makefile-based builds and configs have been removed.

Requirements

• CMake 3.20+
• C/C++ compiler (GCC, Clang, or Intel oneAPI)
• OpenMP runtime for OpenMP and OpenCL benchmarks
• OpenCL SDK (headers + libs) for OpenCL benchmarks
• CUDA toolkit for CUDA benchmarks
• OpenGL (GL/GLU) for the mummergpu benchmarks

Configure and Build

cmake -S . -B build
cmake --build build

Binaries are placed under build/bin by default, grouped per backend:

• build/bin/omp
• build/bin/opencl
• build/bin/cuda

You can override the output root:

cmake -S . -B build -DRODINIA_OUTPUT_ROOT=/path/to/bin

Backend Options

Enable or disable backends as needed:

cmake -S . -B build \
  -DRODINIA_ENABLE_OPENMP=ON \
  -DRODINIA_ENABLE_OPENCL=ON \
  -DRODINIA_ENABLE_CUDA=OFF

CUDA-specific configuration:

• CMAKE_CUDA_ARCHITECTURES controls GPU architectures (recommended).
• RODINIA_CUDA_SAMPLES_INCLUDE_DIR sets the CUDA samples include directory (needed for cuda/cfd and cuda/hybridsort which include helper_cuda.h).
• RODINIA_ENABLE_CUDA_BACKEND=OFF skips building cuda/ benchmarks while still enabling CUDA for OpenMP mummergpu.

Example:

cmake -S . -B build \
  -DRODINIA_ENABLE_CUDA=ON \
  -DCMAKE_CUDA_ARCHITECTURES=80 \
  -DRODINIA_CUDA_SAMPLES_INCLUDE_DIR=/usr/local/cuda/samples/common/inc

Data

Input datasets are downloaded on demand into data/ (ignored by git). Run:

scripts/generate_data.sh

The script fetches the full data bundle from the upstream Rodinia data repository; override with RODINIA_DATA_URL or --url. See data/README.md for details.

Notes

• OpenCL benchmarks include omp.h in several places; OpenMP must be available.
• If a dependency is missing, CMake will skip the corresponding backend or target and print a warning during configuration.
• CUDA 12+ toolkits are supported; CUDA benchmarks no longer rely on legacy texture references.