Hello,
If the following is not documented somewhere, I would be interested in learning about the following:
- Why was LVM chosen has the JIT backend?
- Are there any performance drawbacks? If yes, anything worth knowing?
It would help me (at least) if I could correlate with my _javascript_ ARM/JIT implementation. We did it the “right way”, and it ran faster than Google’s V8 using their own benchmarks. Then came Apple and delivered an implementation that was x86-oriented,
and then we needed to change our implementation. At least with _javascript_, there are two issues with JIT:
- The heuristics to JIT are based on benchmarks => if your typical case is not in the benchmarks, the JIT won’t work for you as well as it could. Is this also the case with Lua+LVM?
- If the HW model is biased, then the JIT will work better for one HW but not as well as it could for another HW. I guess this would be an LVM specific issue and common to JIT in any language, and since no one has implemented ARM/JIT in LVM, no one knows
if LVM works well for that architecture. This is more a statement than a question.
Thank you.
André Dolenc
