On 19.06.2006 18:49, Peter Amstutz wrote:
> For what it is worth, I believe that ptmalloc() has been the standard 
> allocator for glibc on Linux for quite a while 
> (http://www.malloc.de/en/index.html), so it doesn't really make sense to 
> have CS provide its own allocator there (and indeed might even be 
> counterproductive by interfering with other debugging tools).

Though, the glibc allocator is ptmalloc2, the CS one is ptmalloc3.
(According to the 2nd chart of the right side of
http://www.nedprod.com/programs/portable/nedmalloc/index.html, ptmalloc3
is a good deal quicker than ptmalloc2).

If it interferes with utilities, use of ptmalloc can be disabled via
configure.

> The Windows memory allocator is terrible, so that will be a welcome 
> performance improvement.

Absolutely.

> Incidentally, for multithreaded/multiprocessor applications the Hoard 
> allocator (http://www.cs.umass.edu/~emery/hoard/) is supposed to be 
> superior by employing strategies to avoid lock contention of the heap 
> between CPUs.  This is probably not necessary for the mostly 
> single-threaded CS, though.

ptmalloc3 also does stuff like using different heaps for different
threads to make it faster in concurrent situations. I can't really say
how it compares with Hoard technically, but I would as well think that
for CS' it's more than sufficient. (Also, since we try to avoid doing
lots of alloc/deallocs during a frame anyway, the allocator speed is
mostly relevant to level loading.)

-f.r.

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
Crystal-main mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/crystal-main
Unsubscribe: mailto:[EMAIL PROTECTED]

Reply via email to