Re: Making Guile slower?

[Greg Harvey <Greg dot Harvey at thezone dot net>]

Clark McGrew <mcgrew@ale.physics.sunysb.edu> writes:

> Hi,
> 
>     Mikael> The idea I wanted to express here is that if you
>     Mikael> repeatedly make small changes increasing evaluation time
>     Mikael> by 5%, say, it won't be long before you have a big effect
>     Mikael> on the total.  Therefore, it is my personal opinion that
>     Mikael> every change influencing performance should have a really
>     Mikael> strong motivation, and be made with knowledge of the
>     Mikael> effects.
> 
> This one change isn't worth a major battle, but I hope that the gurus
> don't forget that speed is *extremely* important.  Loosing a percent
> here and there may not seem like a big deal, but it addes up
> quickly.

I wouldn't claim otherwise, but in this particular case it isn't like
we're adding something so significant (or even adding anything, since
the intent all along was to add a sloppy version for the critical
places... there are only a very small number of places that don't tie
NIMP with another test) that it's worth a big round of public
debate. Sure, the `and' and the `branch' enjoy it; it's been a long
time since they've had so much attention, everybody nowadays is
concerned with all those newfangled instructions, but we could
accomplish much more by improving the way the code does what it does
than by giving so much attention to microscopic issues (there are
already places where it overchecks this; nobody seems to notice
;). Improving the readability does make it easier to gain real
improvements in speed.

> Guile development is not democracy, but I have to vote with Mikael.  I
> hope that major structural changes like this are thought about very
> carefully and are only made if they fix a major problem.  One of
> guile's strengths has been that it's quite fast for an interpreter.  I
> know that people often say "Get a faster computer", but that's a cop
> out.

I don't think anybody'd be so thickheaded as to suggest this (within
reason: I would suggest that you won't likely ever be able to run
guile on a c64 [maybe >;')]). I won't be totally happy until I can run
guile comfortably and quickly on my old 486 (maybe I should pick up a
cheap 386 for testing :). 

-- 
Greg