I’m really quite happy about the Mint talk I just gave at Purdue University. As a recurrent theme, I could have used a bit more practice, but it went a lot smoother than I would have expected, after just having presented at SIGCSE.
The best part were really the questions and comments I got. There were questions about reducing code duplication, whether we mostly see a kind of
fold behavior for staged programs (this occurs in accumulators, but we can do much more), and much more.
Luke also had three crucial pieces of advice:
- The introductory slide comparing a general
powerfunction to a specialized
power17function and giving performance figures like 41 ns and 9 ns was weak. This ends up defining MSP and our work, and if we lose the audience’s interest here, we probably won’t get it back. We really need to open with a bang. A short opening statement about what MSP is would have been good too.
- At PLDI, we need to have a short explanation of how we benchmarked, and using what JVMs. I was able to answer this in the Q&A period, but it should briefly preface the benchmark table.
- An animation of scope extrusion showing the objects, from a memory/stack/heap point of view, could be useful in helping explain code extrusion, and it would break up the text monotony.
Thanks for the opportunity, Purdue CS, and thanks for organizing this for me, Jan and Luke.