In addition to the unified high-level programming interface, Vitis AI also offers a comprehensive set of advanced low-level C++/Python programming APIs for deep learning applications’ development over Xilinx® edge DPU. These low-level APIs originate from the Deep Neural Network Development Kit (DNNDK) programming interface, which ensures future compatibility for end users. Legacy DNNDK projects can be easily ported to the Vitis AI environment without any code changes.
The Vitis AI advanced low-level interface provides granular manipulations to DPU control at run-time. They implement the functionalities of DPU kernel loading, task instantiation, and encapsulating the calls to invoke Xilinx XRT or the DPU driver for DPU Task scheduling, monitoring, profiling, and resources management. Using these APIs can flexibly meet the diverse requirements under various edge scenarios across Xilinx® Zynq® UltraScale+™ and Zynq® UltraScale+™ MPSoC devices.