CFD Poisson Solver Code Migration from CUDA* to SYCL*: Developer Spotlight

The Poisson equation is a standard test solver governing several physical processes, such as diffusion and heat transfer. The premiere Indian academic institute named IIT-Goa (Indian Institute of Technology, Goa) migrated its 2D Poisson Equation Solver from CUDA* to SYCL* using the Intel® DPC++ Compatibility Tool, available in the Intel® oneAPI Base Toolkit (Base Kit). Unlike the original hardware vendor-locked Poisson solver, the SYCL-migrated solver comprises a unified codebase that can be seamlessly executed on heterogeneous architectures, including Intel® CPUs, Intel GPUs, and NVIDIA* GPUs. The SYCL version, when compiled with the Intel® oneAPI DPC++/C++ Compiler, performed ~1.9x faster [1] on Intel® Data Center GPU Max 1550 than the CUDA binary on NVIDIA* A100 GPU. After fixing some performance degradation of the SYCL code on NVIDIA*A100, it could perform as well as the CUDA binary on NVIDIA*A100. Codeplay*’s oneAPI plugin allows the Base Kit to be supported on NVIDIA* GPUs.

Check out the complete customer success story in this article.

Explore Intel® DPC++ Compatibility Tool

Intel DPC++ Compatibility Tool and its open-source counterpart SYCLomatic are automated tools for easy CUDA-to-SYCL code migration. They automatically migrate the majority (90%-95%)[2] of CUDA kernels and library API calls to SYCL. They also provide inline comments that help you repair the SYCL output to ensure functional correctness.

For more information, visit the Intel DPC++ Compatibility Tool’s documentation.

What’s Next?

Read IIT-Goa's success story – learn how oneAPI helped them improve performance with a cross-architecture supported SYCL-migrated version of the Poisson solver. Get started with the Intel DPC++ Compatibility Tool today!

We encourage you to learn about other AI, HPC, and Rendering tools in Intel’s oneAPI-powered software portfolio.

Get the Software

Install the Intel DPC++ Compatibility Tool as a part of the Base Kit or download its stand-alone version.

Additional Resources

• Migrate from CUDA to C++ with the SYCL portal
• Easy CUDA to SYCL Migration
• oneAPI SYCL C++ code samples
• Intel® Data Center GPU Max Series

[1] see the “Performance Results on Intel® Data Center GPU Max 1550: Significant Improvement” section of the original article here

[2] Intel estimates as of September 2021. Based on measurements on a set of 70 HPC benchmarks and samples, with examples like Rodinia, SHOC, and PENNANT. Results may vary.

Performance varies by use, configuration, and other factors. Learn more at www.Intel.com/PerformanceIndex.