Migrating OceanFFT Sample from CUDA* to SYCL*

Overview

The OceanFFT sample demonstrates the usage of cuFFT library to synthesize and render an ocean surface in real time. The CUDA cuFFT library allows the floating-point power and parallelism of the GPU without having to develop a custom GPU-based FFT implementation. The equivalent FFT functions in oneAPI Math Kernel Library supports one, two, or three dimensions with mixed radices. The supported functions include complex-to-complex and real-to-complex transforms of arbitrary length in single-precision and double-precision. The original CUDA* source code is migrated to SYCL for portability across GPUs from multiple vendors.

Key Implementation Details

This sample demonstrates the migration of the following prominent CUDA features:

• CUDA FFT API (CUFFT)

The oceanFFT sample first generates wave heightfield at time based on initial heightfield and dispersion relationship. Then FFT API is executed in inverse order to convert to spatial domain. Finally, height map values and slope values are evaluated and compared with reference values to determine whether it is correct or wrong.

The sample is implemented in four mains steps.

• Generate wave spectrum in frequency domain.
• Execute inverse FFT to convert to spatial domain.
• Update height map values based on output of FFT.
• Calculate slope by partial differences in spatial domain.

Original CUDA source files: OceanFFT

Migrated SYCL source files including step by step instructions: guided_OceanFFT_SYCLmigration.

References

• Data Parallel C++, by James Reinders et al
• oneAPI GPU Optimization Guide
• CUDA Toolkit documentation
• Install oneAPI for NVIDIA GPUs