Each uplink AI Engine graph has eight input data ports, 32 input coefficient ports, and four output ports. A ulbf_data PL kernel has eight AXI4-Stream output ports that plug into the eight input data ports of the uplink AI Engine graph. A ulbf_coeffs PL kernel has eight AXI4-Stream output ports. So, four ulbf_coeffs PL kernels are required to plug into the 8*4=32 input coefficient ports. A ulbf_slave PL kernel has one AXI4-Stream input port. So, four ulbf_slave PL kernels are required to plug into the four output ports.
Remember, our system instantiates three uplink graphs, so the total number of ULBF PL kernels becomes: three ulbf_data, 12 ulbf_coeffs, and 12 ulbf_slaves.