5.5 Optimalisations
The software-only implementation that has been made will serve as a reference. In the second phase of the project, this implementation should be improved, using HW/SW codesign techniques. Whichever technique you want to use is fine.
The goal is boost the throughput. A human eye needs between 20 and 30 frames to perceive succesive images as video. However, for many people it is possible to perceive up to 200 frames per second.
- calculate the encoding speed of your reference implementation. Note that you need to find the maximum clock frequency that your implementation can handle;
- pull out some tricks to improve the througput, with the constraint that the overal resource usage can not be more than double;
- make a detailed comparison between the reference implementation and the optimalisation.
To obtain the resource usage of a design, an implementation has to be done. The Vivado tool might run into issues with this as there might be too many input and/or outputs. To solve this, the automatic addition of IObuffers needs to be disabled. This can be done by adjusting a parameter in the synthesis tool (aka the one that adds these IObuf’s).
This can be done through this change, in the syntheses settings:
-mode out_of_context
To be able to finish implementation, an update to clock_and_reset_pynq.vhd is required. This new file can be found here.