When considering system partitioning, tasks such as retrieving data from lookup tables and extracting integer and fractional parts of floating-point numbers are better suited for programmable logic. AI Engines are most efficient if you program them for continual vector processing on a steady stream of input data.
Data necessary to process a single pixel comprises four reference pixels and fractional parts of the \(x_q\) and \(y_q\) coordinates. Assume each of the six data values uses a 32-bit, single-precision, floating-point format. Figure 6 shows a conceptual illustration of how input data is derived in programmable logic for each pixel. This example design does not include a programmable logic component but it assumes such a component has generates test input data for AI Engine processing.
Figure 6 - Derivation of Input Pixel Data