Re: [MIT-Scheme-devel] string-head!

[Taylor R Campbell]

   Date: Thu, 24 Sep 2009 13:48:34 -0700
   From: Joe Marshall <address@hidden>

   Has anyone measured the performance difference between smashing
   the string head and just taking a substring?  I'd bet it's a pretty minimal
   improvement against the background of the other things that happen during 
I/O.

I conducted a thoroughly unscientific comparison whose source code is
attached: making a string of length 2^(i + 1), taking a prefix of
length 2^i, and then discarding it, repeating the process ten thousand
times.  Using SUBSTRING becomes slower than SET-STRING-MAXIMUM-LENGTH!
when i is about 5.  This is in line with SUBSTRING-MOVE-{LEFT,RIGHT}!,
actually, which defer to the primitive as soon as the string's length
exceeds 2^5.  Winning on just GC time starts to happen when i is about
10.  (Your mileage may vary, of course.  I ran this code compiled, by
the way.)

So how about the following definition?

(define (string-head! string end)
  (guarantee-string string 'STRING-HEAD!)
  (guarantee-substring-end-index end (string-length string) 'STRING-HEAD!)
  (if (and (fix:< end (string-length string))
           (or (fix:< end 32)           ;32 computed by science (ha!).
               (begin
                 ;; Some implementations of the microcode don't
                 ;; support SET-STRING-MAXIMUM-LENGTH!.  In this case,
                 ;; defer to %STRING-HEAD.
                 ((ucode-primitive set-string-maximum-length! 2) string end)
                 (not (fix:= end (string-maximum-length string))))))
      (%string-head string end)
      string))

test.scm
Description: Text Data