Migrate CUDA Code to SYCL … Pretty Quickly

Karl Qi, Intel

Get performance portability without vendor lock-in via a one-time-migration tool that transforms 90% to 95% of CUDA code to multi-platform DPC++/SYCL code.

In the realm of high-performance, heterogeneous computing, GPUs have been among the most efficient and effective accelerators for real-time, high-res, 3D graphics compute-intensive tasks. And CUDA—launched in 2007—has been the go-to platform for developing these types of applications.

The challenge: CUDA code has not been easy to port to other vendor architectures.

But that’s no longer the case.

In this session, learn how the Intel® DPC++ Compatibility Tool can port CUDA to DPC++/SYCL code in a one-time migration activity, including kernels and library API calls.

Key takeaways:

  • An overview of standards-based C++ with SYCL—its origins and benefits to developers
  • Hands-on demonstrations with real-world examples of the migration concept, process, and expectations
  • Discussion of common tasks required to complete the migration, including scenarios where manual porting is required
  • An exploration of post-migration performance tuning that may be needed due to differences between CUDA and DPC++/SYCL

Sign up today.