On Wed, 27 Mar 2002, George Greer wrote:
> http://www.hpl.hp.com/personal/Hans_Boehm/gc/
The collector, BTW, is very good. A simple test allocating 1 million
lightweight objects (wrappers around a native int) and printing their
contents (to avoid GCC seeing the assignment as deadcode and whacking it)
gives:
C++ (GCC 3.0.3) using Hans-Boehm collector (and std::printf()[1]):
real 0m2.421s
user 0m2.400s
sys 0m0.000s
memory 1.09MB
C (GCC 3.0.3) using Hans-Boehm collector:
real 0m2.241s
user 0m2.230s
sys 0m0.000s
memory 1.09MB
The fastest JVM I have around (Sun's HotSpot Server [J2SE] 1.4.0) is about
13 times slower than C/C++, but that's certainly not a good comparison of
the collectors. (GIJ, which uses the Hans-Boehm collector, is dog slow
and using GCJ to compile Java natively produced code about 14 times slower
than C/C++, although that it worked at all [let alone so well] was quite
wonderful, if you ask me.)
-dak
[1] In case you're wondering why I didn't use std::cout:
C++ (using IO streams (std::cout)):
real 0m34.517s
user 0m30.370s
sys 0m4.130s
Just bloody awful.
--
+---------------------------------------------------------------+
| FAQ: http://qsilver.queensu.ca/~fletchra/Circle/list-faq.html |
| Archives: http://post.queensu.ca/listserv/wwwarch/circle.html |
| Newbie List: http://groups.yahoo.com/group/circle-newbies/ |
+---------------------------------------------------------------+
This archive was generated by hypermail 2b30 : 06/25/03 PDT