[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Gcl-devel] Re: gcl-2.6.5hi
From: |
Camm Maguire |
Subject: |
[Gcl-devel] Re: gcl-2.6.5hi |
Date: |
17 Sep 2004 16:53:13 -0400 |
User-agent: |
Gnus/5.09 (Gnus v5.9.0) Emacs/21.2 |
Greetings!
Robert Boyer <address@hidden> writes:
> Camm,
>
> Thanks so very much for the in-lining improvements to gethash!
>
> Just to test gcl-2.6.5hi a bit, I have built ACL2 on top of it and run some
> very gethash-intensive tests, getting overall speeds ups of 2.5, 8, and 14
> percent on various tests. No troubles encountered. So I think you've done a
> wonderful thing.
>
Great!
1) Are these the same tests on which you've previously reported gains
of 30%? Current implementation should be no worse, but correct
:-).
2) The real speedups are to be had through a better hashing
algorithm. If I recall, Warren had a proposed patch in this
regard. Please feel free to send me such if you still feel hashing
is too slow.
> Thanks,
>
> Bob
>
> P. S. Now for what is undoubtedly a really dumb question. In the last
> message you sent giving various examples such as:
>
This is now perhaps a no-op in gcl. Older code had this in several
places to trigger various inlining behavior, perhaps due to
since-fixed bugs in the type system. I haven't verified down its
vacuousness in all cases, and so habitually throw it into tests like
these.
Take care,
> (defun food (x h)
> (declare (object x h))
> (multiple-value-bind (x h) (gethash x h 'bar) (declare (fixnum x h)))
> 'foo)
>
> what is the point of (declare (object x h))? I have no idea what the type
> "object" is. Looking up "object" in the index of the ANSI spec gets me no
> further than the idea than an object is a "real thing".
>
>
>
>
>
>
--
Camm Maguire address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens." -- Baha'u'llah
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [Gcl-devel] Re: gcl-2.6.5hi,
Camm Maguire <=