Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!swbatl!texbell!texsun!pollux!ti-csl!islington-terrace.csc.ti.com
From: pf@islington-terrace.csc.ti.com (Paul Fuqua)
Newsgroups: comp.lang.lisp
Subject: Re: CLOS speed
Message-ID: <96030@ti-csl.csc.ti.com>
Date: 27 Oct 89 17:09:12 GMT
References: <31037@news.Think.COM>
Sender: news@ti-csl.csc.ti.com
Organization: TI Computer Science Center, Dallas, Texas
Lines: 52
Moon:    Waning Crescent (3% of full)


    Date: Wednesday, October 25, 1989  1:41am (CDT)
    From: barmar at kulla (Barry Margolin)
    Subject: Re: CLOS speed
    Newsgroups: comp.lang.lisp
    
    Symbolics announced in the Summer that their next release, which should be
    released this winter, will include CLOS.  I suspect Lucid and Franz will be
    including CLOS by the summer.

TI includes CLOS in their (our) Release 6.0, which I believe started
shipping last week.

    Note that you shouldn't expect generic function calls to be as efficient as
    normal function calls.  Remember that CLOS provides extra functionality per
    line of code.  Naturally, there are some tradeoffs.

Since the sequence of events will be something like (1) call the generic
function, (2) look up the method, (3) call the method, a generic
function call will take about 2x the time for a regular function call,
plus method-lookup overhead.  On my Explorer, I'm seeing 3x or 4x for
simple one-argument dispatches.

Method-lookup overhead can be reduced by caching methods, and some of
the function-call overhead can be reduced by avoiding duplicate work for
the second call.  A really hot compiler might do the method-lookup at
compile time if it had enough type information, but I don't think anyone
does that yet.
    
    Date: Wednesday, October 25, 1989  10:03pm (CDT)
    From: barmar at kulla (Barry Margolin)
    Subject: Re: CLOS speed
    Newsgroups: comp.lang.lisp
    
    Accessing a local variable is probably the fastest single operation in any
    Lisp implementation.  CLOS slots should be compared with DEFSTRUCT slots,
    not with local variables.
    
    In a high performance implementation, with type declarations and block
    compilation, the offset can be determined at compile time, and most of the
    above can be compiled away or inlined.

Another off-the-cuff Explorer example:  SLOT-VALUE with no type
information takes about 50x a DEFSTRUCT slot reference, SLOT-VALUE with
type information takes 25x, an accessor method takes 13x, and WITH-SLOTS
within a method takes 4x or 5x.  Sometimes I've found it useful to make
a function into a method just to improve the slot-access performance.

Paul Fuqua                     pf@csc.ti.com
                               {smu,texsun,cs.utexas.edu,rice}!ti-csl!pf
Texas Instruments Computer Science Center
PO Box 655474 MS 238, Dallas, Texas 75265