Re: LuaJIT performance

[Mike Pall <mikelu-0908@...>]

Alexander Gladysh wrote:
> Do you consider to release a test suite along with LJ2?

Probably not. Because mine is a real big mess, thrown together
from snippets of various sources with various copyrights and I
haven't kept track of their origins.

There's of course the standard Lua test suite, but it's of limited
use with a JIT compiler. Most of it is never compiled and other
tests fail, because they test for implementation characteristics
and not just language characteristics.

E.g. the test suite checks whether 'for i=1,100 do local t={} end'
allocates memory and drives the GC forward. But LJ2 is able to
eliminate the table allocation and thus doesn't need to run the GC.
[Yes, I'm of the strong opinion that side-effects of the GC are
implementation-defined.]

But I may document, package and release my Lua benchmark
collection whenever I get around to it.

> > But I have to say it was an expensive decision: I've considerably
> > underestimated the amount of research and trial-and-error which
> > was needed to convert a research toy into a production compiler.
> > There are some important implementation details which the few
> > papers about trace compilers completely fail to mention ... :-|
> 
> Perhaps that is the reason to write one yourself someday? ;-)

Well, Roberto has urged me to write some papers, too. But I'm a
lot better at writing code than writing papers. And, not being at
a university anymore, there's no urgent need to write papers ...

--Mike