gcl-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Gcl-devel] RE: new benchmark results for 8 CL implementations in cliki.


From: Mike Thomas
Subject: [Gcl-devel] RE: new benchmark results for 8 CL implementations in cliki.net
Date: Tue, 3 Aug 2004 11:36:12 +1000

Hi Christian.

As mentioned saparately I've been on leave so am behind on my email.

You wrote:

| * On Fri, Jul 30, 2004 at 12:41:09AM MEST, Mike Thomas wrote:
| > so it looks like this bug will be a good one to stomp on.  I'll
| let you know
| > if/when we manage to find it.
|
| it's good to see that my just-for-fun benchmarking may help to improve
| GCL. I'm not in a hurry though.

It has certainly been useful.  As it happens, on Friday I tried the
benchmarks on Windows and got substantially different results to yourself,
most notably I ran out of space on the large PI calculation after GCL VM
usage reached roughly half a Gigabyte.  I intend to do some comparisons with
one of the Unix platforms at some stage to see what is happening
differently.

Interestingly, compiling and running the CLOS benchmarks at the command
prompt did not lead to any errors, although I may have done something
different to the way in which the makefile works (which is presumably what
happened with your tests).


| Peter Graves who seems to develop the compiler for Armed Bear Common
| Lisp has also been able to identify a bug after he saw that ABCL
| failed on PI-DECIMAL/BIG and PI-ATAN. He writes me that he already
| fixed the corresponding rational-to-float conversion bug this morning.
|
| If I may ask you for advice:

You may but, ironically, despite my position as a GCL developer I know very
little about Common Lisp!!  Camm Maguire and Paul Dietz are better bets for
this type of question, or anything to do with Unix platforms.

| I am maintaining a big Lisp project
| (using mk-defsystem 3.20) that is running with Allegro CL 4.3 on
| Solaris. I've already moved it to Allegro Cl 5.0 on Linux without any
| need of porting. Now I'd like to port it to a Free Lisp. As the
| project was written by many people throughout the nineties, it
| contains a mixture of CLtL1, CLtL2 and perhaps ANSI CL (no CLOS
| though). The code is conditionalized with :SYMBOLICS :LUCID and
| :ALLEGRO features. Do you think that GCL will be able to run it?

My experience with GCL is that it won't compile any major Common Lisp
project which has not been traditionally built with KCL/AKCL/GCL over the
years.  I expect this to change with the move towards ANSI compatibility
upon which we are embarking with 2.7.0.

We will be discussing our plan for this work shortly on this mailing list.


| I am
| also looking at CLISP and CMUCL/SBCL but have not made any experiments
| yet.

I would expect CMUCL/SBCL to be the better bet for broad compatibility, but
that is largely a guess based on what I have read as, of course, neither
CMUCL nor SBCL has been properly ported to Windows yet - although that is
changing rapidly I believe.  Off the top of my head, you could expect
trouble from items such as GUI, socket and foreign function interfaces (that
is, all the practical application bits) in a port to/from any one of the
major Common Lisp compilers, quite apart from issues of ANSI compatibility.

In any event, please keep us posted if you try GCL on your project.

Cheers

Mike Thomas.






reply via email to

[Prev in Thread] Current Thread [Next in Thread]