Working vector instructions and some luck with optimizations
The last few weeks were a confusing mix of both some decent work and procrastination with way too many blockers faced. It didn’t feel productive at all despite some data proving the opposite. One thing I can say for sure is I didn’t make any progress on the blockchain itself this time, just wasn’t in the mood for it.
CoreMark score and more failed optimizations
Another two weeks or so wrestling with compilers. Not always successful, but numerous lessons were learned in the process, which I’ll take as a (small) win.
On the bright side, there were new releases of RISC-V crates with more improvements, and there is now a CoreMark runner, which allows somewhat objective comparison of interpreter performance against alternatives.
How to not make interpreter faster
I like posting updates with substantial performance improvements, but sometimes that is just not happening despite plausible approaches and substantial effort. It is still a useful learning experience to share, though. And I have non-performance improvements to share too, so let’s get to it.