[Gcl-devel] Re: Hhmmm

[Camm Maguire]

Greetings!  If your program spends 100sec collecting garbage, then you
won't see the large percentage but small absolute speedups of say 5
sec -> .3 sec from fast-linking, prototyping, and variable
declarations. 

I take it you can't post your file?

Take care,

Eric <address@hidden> writes:

> Dear Camm,
> 
> I'm skeptical. None of this (below) explains why I
> don't see any difference in the performance of GCL
> with/without function prototypes, so that the compiler
> can pass function parameters on the C stack vs. the
> heap. This is the central question at the moment.
> 
> Your response to my very first email emphatically
> states that you have done it, i.e. achieved a
> difference in the performance of GCL with/without
> function prototypes. Yet, you can't explain to me the
> procedure for accomplishing this. I'm confused.
> 
> Best regards,
> 
> Eric
> 
> --- Camm Maguire <address@hidden> wrote:
> > 1) Can you send me your file?  Pretty hard to help
> > with so little
> >    information. 
> > 
> > 2) Speed differences of this magnitude are almost
> > always due to
> >    garbage collection.  (setq si::*notify-gbc* t) to
> > find out.  Also
> >    look at (si::gbc-time) and its info page.  If so,
> > make sure you
> >    declare the types of your variables where
> > possible.
> > 
> > 3) gcl-2.6.1 and above comes with a profiling
> > option.  Compile with
> >    --enable-gprof, then bracket your code with
> > (si::gprof-start) and
> >    (si::gprof-quit). 
> > 
> > 4) I don't know what your lisp is, but I'm
> > reasonably confident that
> >    your performance issues are far more basic than
> > those related to
> >    fine tuning compiler switches.
> 
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah