Importing an AI Engine Kernel with Run-Time Parameters - 2020.2 English

Model Composer supports importing AI Engine kernels with run-time parameters in kernel functions alongside window and stream types. The following table lists the scalar data types that can be passed as run-time parameters.

Implicit ports are inferred for each parameter (scalar and vector data types) in the function argument. The following table describes the type of port inferred for each function argument.

In the following example, the simple_rtp function has two real-time parameters. Notice the function argument select which is passed by value, and argument weight which is passed by reference.

#ifndef __RTP_KERNEL_H__
#define __RTP_KERNEL_H__
 
void simple_rtp(input_window_cint16 * in,output_window_cint16 * outw, int32 &weight, int32 select);
 
#endif //__RTP_KERNEL_H__

When imported for the above function, the AI Engine kernel block looks as shown in the following figure. In Model Composer, the inout port appears as the output port on the AI Engine kernel block.

Because RTPs are used alongside the window and stream ports, the procedure for importing the kernel function remains the same. When the above kernel function with RTPs are imported, the AI Engine kernel block looks as shown in the following figure.

Figure 1. AI Engine Kernel (RTPs)

Notice that the AI Engine kernel block name (simple_rtp) is same as AI Engine kernel function name.

After a successful import, the Function tab GUI displays automatically. You can quickly review the function definition and run-time parameter ports as shown.

Figure 2. Function Tab

The following table provides details and description on the parameters specific to RTPs.