Regenerate the
xclbin
file. You need to domake clean
andmake
to delete the existing xclbin with one CU and to create a new xclbin with three kernel CUs.make clean make run TARGET=hw_emu
View the new
xrt.run_summary
in the Vitis analyzer.
You can now see that the application takes advantage of the three CUs, and that the kernel executions overlaps and executes in parallel, speeding up the overall application.