The kernels/functions needed to accelerate are programmed using C++ with AI Engine APIs that target the VLIW scalar and vector processors of the AI Engine. The AI Engine kernel code is compiled using the AI Engine compiler (aiecompiler) that is included in the AMD Vitis™ development kit. The AI Engine compiler compiles the kernels to produce executable and linkable format (ELF) files that are run on the AI Engine processors. The kernels can be mapped to the independent AI Engines based on the algorithm to perform the parallel computation.
The data flow concept should be applied to the design specification to obtain the benefit of parallel execution whenever possible. It is also important to make sure the AI Engine kernels are running faster to obtain the benefit of the data flow programming. See Data Flow Programming.