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

Re: Poor Sun timings, as competition...

    Date: Tue, 31 Jul 90 12:47:15 -0400
    From: kanderso@DINO.BBN.COM

    The way i read your timing results is that compiling and loading are

True.  Genera is supposed to be one of the best development
environments, and developers do lots of compiling and loading.

	   My guess is that his is mostly due to software than hardware.

I'm not sure what "mostly" means in this context.  If the hardware were
twice as fast wouldn't the compiler speed up by approximately this
amount?  If the software is inefficient, then the hardware needs to be
faster to make up for it.  Supposedly, that's one reason specialized
Lisp hardware is being used.

    Compiling and loading on a LISPM could be slow for at least the following

    o I/O is slow.

Often suggested.

    o Compiler is slow.

I first mentioned this to the person who asked me to use compilation as
the benchmark.  I said that compilation speed is not something Symbolics
has worked on significantly.  His response is that he doesn't care *why*
compiling his code takes a long time in Genera.  If compilation speed is
his concern, then he'll prefer the system with the fastest compiler.

    o Loading and compiling do more on LISPM than on Suns, such as
      who-calls and meta-dot support.

I don't think this is as much a factor as it once was.  Lucid provides
meta-dot support, and Genera defers most of the work of who-calls
support either in a background process or at the end of a Load System
(and it doesn't take very long).  And compiling Lisp for a Lisp Machine
should be easier than compiling for a RISC machine; the Lispm compiler
shouldn't need to be as clever.