Chapter 9. OpenACC and Interoperability
Jiri Kraus, NVIDIA
One of the major strengths of OpenACC is that it is interoperable with native or low-level APIs to program the accelerator. This enables two important use cases:
• Calling native device code from OpenACC. This allows you to easily call into high-performance libraries, such as BLAS or FFT, or custom kernels using the native device API from an OpenACC program without introducing any unnecessary data staging.
• Calling OpenACC code from native device code. This also avoids unnecessary data staging.
This chapter explains how OpenACC interoperability works for both cases, covering the following:
• How OpenACC-managed accelerator memory can be used by kernels written in a native device API ...