[azakai]

I see, thanks.

I ask because performance of small benchmarks has been quite good, except for the uint32 issue mentioned above, often around 3x slower than native code - but on very large codebases it is often much slower, and I do not know why.

[mraleph]

Unfortunately (as you probably know yourself) I don't have a magical answer that would speed everything up. As for V8 there is quite a number of limits that you might be hitting with generated code (e.g. number of locals, size of the function etc) and there can be some bugs or non-done-yet thingies affecting performance. Profiling and looking at the generated code (and filing bugs) is the only suggesting I can give here.

[azakai]

Ok, yeah, I understand there is no simple answer here. Thanks for your clarifications earlier.