Re: [PATCH] Take some lowhanging fruit to speed up R6RS fixnum operations

[Andreas Rottmann]

address@hidden (Ludovic Courtès) writes:

> Hello,
>
> Andreas Rottmann <address@hidden> writes:
>
>> So, that's around 5% improvment (on the ZIP benchmark) for an IMHO
>> significantly more hackish implementation.  I'm not sure that's worth
>> it. WDYT?
>
> Was it with ‘fixnum?’ inline, or with the ‘fixnum?’ instruction?
>
The numbers from my previous mail were with the "clever" `fixnum?'
(i.e. the one that uses object-address and bit-twiddling), defined with
`define-inline'.  I have yet to produce numbers with `fixnum?' as VM
primitive and the more complicated, partially-inlinable fixnum
operators.  I'll keep you posted.

> It’s ironic that while R6RS fixnums are a performance hack, they end up
> being less efficient than unbounded integers in Guile.
>
Indeed.

> Do you know how other implementations deal with that?
>
I will do some research on that topic, investigating Racket, Ikarus and
Ypsilon -- these are the R6RS implementations, which I'm most familiar
with (besides Guile, obviously).

Regards, Rotty
-- 
Andreas Rottmann -- <http://rotty.yi.org/>