Coexistence of OpenCL and XRT APIs and long-term support

[qberthet]

In the context of the ATLAS experiment at CERN, we are exploring the potential utilization of AMD FPGA accelerators in a forthcoming computing farm. Presently, we have multiple teams engaged in independent work on kernel algorithms, with varying preferences regarding the use of OpenCL versus the Native XRT API. We aim to initiate the integration of these efforts into the existing software framework.

Question 1: Is it feasible to employ both the Native XRT and OpenCL APIs within the same software solution? In other words, can a device programmed with the Native XRT API be utilized alongside the OpenCL API (Without reloading the XCLBIN with OpenCL)? If so, it would be great to have some pointers on how to do achieve this.

Question 2: Does AMD provide an official recommendation regarding the preferred API choice? We seek guidance on factors such as features, performance, and long-term support, as the decision between the two APIs is currently challenging based on the available online information.