For Part 3, you are going to increase the performance of the design by adjusting the TP_CASC_LEN parameter on the chan_FIR filter.
In system_settings.h
, the following parameter has been changed:
#define AIES_CHAN 5
informing the library element to allocate 5 AI Engines to the FIR filter kernel.
In fir_graph.h
, the only change here is using a for loop to lock the location constraints of the chan_FIR AI Engine tiles.
for (int i=0; i < AIES_CHAN; i++) {
location<kernel>(chan_FIR.m_firKernels[i]) = tile(18+i, 0);
}
location<kernel>(hb_FIR.m_firKernels[0]) = tile(23,0);
The testbench file, test.cpp
, is unchanged.