1. The gradient calculation path is slower than the Laplacian calculation, so a frame buffer is needed. However the size of the FIFO can be estimated from the latency of the 2DFIR and should be much smaller than one frame. I used a FIFO that can buffer the entire frame, which is not optimal.
2. The Laplacian path and Gradient path access source data at different throughput due to AST backpressure. The Gradient path sees more backpressure and pauses more. I guess in HW there has to be another frame buffer at the source for these two different paths. To simplify in the simulation, I just used two different sources.
