Description
The term project is to build on the VeriSimple4 RISC-V pipeline from project 3 to create a more advanced pipeline with a few of the features we are studying in class. Projects will be done in groups of five students (other sizes must be approved by staff). All groups are required to implement certain “base” features and these base features vary by the number of group members. In addition, you will have the opportunity to add more advanced features to improve the performance of your pipeline. A significant portion of your grade will depend on the absolute performance of your pipeline (CPI and clock rate). Your project grade will be the weighted average of scores based on the following areas:
- Implementation of base features: 25%. Did you implement all of the base features listed below?
- Correctness and testing: 20%. Does your pipeline correctly implement the ISA, and did you convince us of that through your testing methodology?
- Performance: 20%. How well does your pipeline perform on the test benchmarks provided (including, but not limited to, the ones used for programming assignment 3)? Performance will be calculated using the CPI derived from the Verilog simulator and the timing reported by the synthesis tool. Note that for every test program that your design does not complete correctly, we factor it into your relative performance as 10% slower than the worst-performing correct design. So correctness is far more important that performance – you can’t get performance points until your design works. Furthermore, we need an accurate cycle time for your design, so you can’t get any performance points if your design does not work correctly after synthesis.
- Optional features: 15%. Extra points for those who attempt ambitious designs. Ideally these points will be in addition to the performance points that these features provide. In the worst case, they are points for trying something bold that didn’t quite work out. There are guidelines of typical optional features we’ve seen in the past (and their approximate point values). However, you are encouraged to propose something new.
- Analysis: 10%. Did you uncover the impact of your features on performance? For example, on a superscalar machine how many instructions do you complete per cycle? What is the prediction accuracy of your branch predictor and/or BTB? How full is your ROB? If you add an interesting “Additional feature” it would be nice to learn how successful it is with respect to performance. Note that your grade won’t suffer from showing us that something is actually a bad idea. What we want to see is that you can measure how good or bad the idea/feature was. This data will show up in your report.
- Documentation: 7%. Did your report describe your design, the motivation for your design decisions, your testing methodology, and your performance evaluation in a readable, concise manner? Although the documentation itself counts for only 7% of the project grade, I will be basing your scores for the other areas on the information you provide in your report. A poorly written report could bring down your scores in all areas.
- Check-points: 3%. Did you meet the requirements of the first check-point? Was the module a reasonable one and did it work? Were you prepared for the second and third check-points?
