MT19937 (Was Re: [Sbcl-devel] Web-based SBCL internals documentation effort)

[Raymond T. <to...@rt...>]

>>>>> "Daniel" == Daniel Barlow <da...@te...> writes:

    Daniel> As threatened, I have created a CLiki-based web site for SBCL
    Daniel> internals docs - on the grounds that even if nobody else uses it, I
    Daniel> find it helpful.

Cool!  I find it helpful.  CMUCL needs this kind of stuff too.

However, while poking around I noticed the MT19937 page which says:

        On other platforms, there is an implementation in pure Common
        Lisp, which is of the order of 10 times slower, and conses

Did you measure this?  Way back when I helped Douglas do the Lisp
version, I'm pretty sure this wasn't true, at least on Sparc.  The
very first version did cons, but that was fixed very quickly.  At that
time, I looked at the generated assembly and it was pretty close to
what I would have done by hand.

Ray