[Top][All Lists]

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

RE: [Axiom-developer] Re: lisp

From: Bill Page
Subject: RE: [Axiom-developer] Re: lisp
Date: Tue, 6 Apr 2004 16:04:42 -0400


On Tuesday, April 06, 2004 1:53 PM you wrote:
> ...  With the new autotune of the maximum pages by type 
> (which is not in your tree yet but will be in gcl 2.6.2) plus 
> those AXIOM patches, the compile time for EXPEXPAN.spad is 
> down by about a factor of ten on my box.  Full build took 
> just over an hour (2.4Ghz).

> I know such questions are of quite lesser significance, but 
> someone did state on the list that they wanted a faster 
> compile, which kicked me off in this direction.

I think it is VERY important. It is my understanding that because of the
dependency loops in Axiom's algebra definition (which necessitates the
current "bootstrap from LISP" method) it is, at least in prinicple necessary
to iterate the compilation of the algebra at least twice - once to
re-generate the bootstrap LISP code from the SPAD sources, replacing the
bootstrap code with the new LSIP. And then a second time  to make sure that
the generated bootstrap LISP code does not change (i.e. is a fixed point).

This is true, isn't it Tim?

On the other hand, I used the words "in principle" above because I think
(but haven't checked fully yet) that due to the currently rather simple
nature of the dependecy loops in Axiom's algebra code, one iteration is
sufficient. I think there is a need for some automatic means to check for
the fixed point. I have seen that the internally generated LISP names do not
stay the same from the original bootstrap to the first iteration LISP
output. So a comparison is not so trivial as a simple diff.

Bill Page.

reply via email to

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