[Python-checkins] python/dist/src/Modules arraymodule.c,2.83,2.84

[Neal Norwitz]

On Mon, Feb 24, 2003 at 07:38:01AM -0500, Guido van Rossum wrote:
> > This improves speed by about 5.6% for me.

        ["This" refers to adding roundupsize() to bump up
         the size of an array by increasing large multiples 
         (same as a list).]

> I don't understand this comment.  If the fix works, it should change
> quadratic time to something vaguely N-log-N-ish, which cannot be
> expressed as a percentage without knowing N.

By adding roundupsize(), the difference was calling PyMem_RESIZE()
fewer times and also doing a memmove less.  I was surprised at
how little benefit there was, that was why I added the comment.
I would like to know if this affects other platforms more.
I noticed the original problem was on windows.

With 10,000,000 elements, I tested before and after the patch:

        before= 42.5821830034 seconds
        after=  40.1672360897 seconds

I'm not sure what the deal is.  The code looks like it is the same as
in listobject, except that array uses a memmove while list uses a for
loop.  Can someone test on windows to see if the timing is drastically
improved?  Maybe it's just a realloc issue?

Neal