bug#17049: [PATCH] Make reverse! forego the cost of SCM_VALIDATE_LIST

[David Kastrup <dak@gnu.org>]

[-- Attachment #1: Type: text/plain, Size: 250 bytes --]

Andy Wingo <wingo@pobox.com> writes:

> Hi,
>
> Thanks for the patch.  What is its performance impact for your use case?

Here is an artificial use case (I make care to get my list scattered
over memory, assuming that sort! keeps the cells around).


[-- Attachment #2: zoppo.scm --]
[-- Type: application/octet-stream, Size: 270 bytes --]

(use-modules (srfi srfi-19) (srfi srfi-1))
(define xxx (sort! (map! random (make-list 50000 10000000)) <))
(define start-time (current-time))
(for-each (lambda x (set! xxx (reverse! xxx))) (make-list 20001))
(display (time-difference (current-time) start-time))

      

[-- Attachment #3: Type: text/plain, Size: 597 bytes --]


The output for my version first and the default version afterwards is

#<time type: time-duration nanosecond: 449629000 second: 8>dak@lola:/usr/local/tmp/guile$ meta/guile /tmp/zoppo.scm 
#<time type: time-duration nanosecond: 898802000 second: 15>dak@lola:/usr/local/tmp/guile$ 

So it's a bit better than my 7:4 estimate (rather a factor of 1.88),
probably because I forgot that the CPU does not have to wait for the
write cycle to complete for continuing.

Now that's a somewhat artificial benchmark, but still: almost a factor
of 2 for the operation itself is pretty good.

-- 
David Kastrup