It is possible to disable the ping-pong mechanism, so that the entire available data memory is available to the kernel for computation. However, the single-buffered window can be accessed only by one agent at a time, and it comes with a performance penalty.
This can be achieved by using the single_buffer() constraint that is applied to an input or output port of each kernel.
single_buffer(firGraph.getKernels()[0].in[0]);