Re: Enormous benchmark speedup

[Juhani Viheräkoski]

> > With recent changes in guile vm there are lots on improvements on the
> > Gambit benchmarks.
>
> Improvements compared to what?

I should have been more concise.. I have run these benchmarks against 
guile HEAD for some time (after I reported a few bugs, performance 
regressions included,  affecting the benchmark suite I have done this 
every once in a while to get some additional test coverage) and now I had 
something to report. Guile I compared to is a git version before vector 
instructions were added to VM. Guile is now a little bit slower 
generally, but significantly faster in many real-world benchmarks using 
vectors.

> > but there are huge wins in testcases involving vectors offsetting
> > this.
>
> Oh, that's cool, but it feels like we're slightly cheating on these.
> ;-)

Well it seems to be a good cheat then :)

> This is not tail-recursive (because of the nested `ack' call in the
> `else' clause), which is why it can lead to a stack overflow.

I understand that, but I reported this because interpreter (both 1.8.6 and 
1.9.0) run this test successfully but VM does not. Probably real-world 
programs don't have so deep a recursion, although they might when run with 
complex data..

Thanks for your reply,

Juhani Viheräoski