Overview
Examples
Screenshots
Comparisons
Applications
Download
Documentation
Tutorials
Bazaar
Status & Roadmap
FAQ
Authors & License
Forums
Funding Ultimate++
Search on this site
Search in forums












SourceForge.net Logo
Home » Community » U++ community news and announcements » Updated and new benchmarks
Updated and new benchmarks [message #14774] Fri, 14 March 2008 12:58 Go to next message
mirek is currently offline  mirek
Messages: 13975
Registered: November 2005
Ultimate Member
I have adjusted numbers in U++ vs C++ std:: benchmarks

http://www.ultimatepp.org/www$uppweb$vsstd$en-us.html

and also created a new one, this time comparing C++ with D:

http://www.ultimatepp.org/www$uppweb$vsd$en-us.html

Almost 5 times faster than GCC's std:: implementation in supposedly I/O limited benchmark and 70% faster than D... not bad Smile

I guess we are now finaly approaching the "physical limits of silicon" Smile

Mirek
Re: Updated and new benchmarks [message #14781 is a reply to message #14774] Fri, 14 March 2008 15:52 Go to previous messageGo to next message
cbpporter is currently offline  cbpporter
Messages: 1401
Registered: September 2007
Ultimate Contributor
luzr wrote on Fri, 14 March 2008 13:58

I have adjusted numbers in U++ vs C++ std:: benchmarks

http://www.ultimatepp.org/www$uppweb$vsstd$en-us.html

and also created a new one, this time comparing C++ with D:

http://www.ultimatepp.org/www$uppweb$vsd$en-us.html

Almost 5 times faster than GCC's std:: implementation in supposedly I/O limited benchmark and 70% faster than D... not bad Smile

I guess we are now finaly approaching the "physical limits of silicon" Smile

Mirek


Interesting results. What has been done to obtain an almost double performance since last time?

edit:
luzr wrote on Fri, 14 March 2008 13:56

Inequality comparison of small strings are now heavily optimized, Sorting Vector<String> is now almost twice as fast (if most strings are < 15 characters).

Mirek


I've read this after my post.

[Updated on: Fri, 14 March 2008 15:54]

Report message to a moderator

Re: Updated and new benchmarks [message #14783 is a reply to message #14774] Fri, 14 March 2008 17:02 Go to previous messageGo to next message
zsolt is currently offline  zsolt
Messages: 693
Registered: December 2005
Location: Budapest, Hungary
Contributor
The title of third column should be "D language / U++".
Re: Updated and new benchmarks [message #14786 is a reply to message #14783] Fri, 14 March 2008 18:44 Go to previous messageGo to next message
mirek is currently offline  mirek
Messages: 13975
Registered: November 2005
Ultimate Member
zsolt wrote on Fri, 14 March 2008 12:02

The title of third column should be "D language / U++".


Well, depends. Strictly said, yes. But maybe it is confusing that way too.

Mirek
Re: Updated and new benchmarks [message #14787 is a reply to message #14781] Fri, 14 March 2008 18:45 Go to previous messageGo to next message
mirek is currently offline  mirek
Messages: 13975
Registered: November 2005
Ultimate Member
cbpporter wrote on Fri, 14 March 2008 10:52

luzr wrote on Fri, 14 March 2008 13:58

I have adjusted numbers in U++ vs C++ std:: benchmarks

http://www.ultimatepp.org/www$uppweb$vsstd$en-us.html

and also created a new one, this time comparing C++ with D:

http://www.ultimatepp.org/www$uppweb$vsd$en-us.html

Almost 5 times faster than GCC's std:: implementation in supposedly I/O limited benchmark and 70% faster than D... not bad Smile

I guess we are now finaly approaching the "physical limits of silicon" Smile

Mirek


Interesting results. What has been done to obtain an almost double performance since last time?

edit:
luzr wrote on Fri, 14 March 2008 13:56

Inequality comparison of small strings are now heavily optimized, Sorting Vector<String> is now almost twice as fast (if most strings are < 15 characters).

Mirek


I've read this after my post.


Actually, this last optimization was only the last 2-7% improvement. The majority is from new String implementation, using SSO with very fast equality comparison and hash value generation.

Mirek
Re: Updated and new benchmarks [message #15226 is a reply to message #14774] Thu, 10 April 2008 15:07 Go to previous messageGo to next message
jmansion is currently offline  jmansion
Messages: 15
Registered: April 2006
Location: London
Promising Member
These numbers are interesting enough to renew my interest in ntl. And lo, my password reset info comes from cxl@ntllib.org.

But www.ntllib.org returns 403 (Forbidden).

Oh well. Crying or Very Sad

How can I access the ntllib files, and can they be used without 'all' of UPP?

Also - do you know how your heap implementation stacks up against nedmalloc or jemalloc?
Re: Updated and new benchmarks [message #15229 is a reply to message #15226] Thu, 10 April 2008 17:33 Go to previous messageGo to next message
mirek is currently offline  mirek
Messages: 13975
Registered: November 2005
Ultimate Member
jmansion wrote on Thu, 10 April 2008 09:07

These numbers are interesting enough to renew my interest in ntl. And lo, my password reset info comes from cxl@ntllib.org.

But www.ntllib.org returns 403 (Forbidden).



Site is basically abandoned. NTL was sort of "U++ preview"...

Quote:


How can I access the ntllib files, and can they be used without 'all' of UPP?



Not quite possible now. Sorry.

We are sort of looking for volunteer that would take on this task ("extracting" NTL from U++ sources).

Quote:


Also - do you know how your heap implementation stacks up against nedmalloc or jemalloc?



First time I hear those names Smile It would be nice to test - if you have time Smile

From the brief excursion via google, there was too little info to grasp nedmalloc or jemalloc.

It is also true that U++ allocator is still quite concerned about single-threaded performance; there is some solution to cache-line contention problem, but hard to say how it works in practice. New allocator is very specifically tuned for U++ framework too.

However, single-threaded performance is now so effective, that checking for NULL (C++ standard requirement) is now visible as a couple of percents performance degradation Smile

Mirek
Re: Updated and new benchmarks [message #15382 is a reply to message #15229] Fri, 18 April 2008 14:30 Go to previous messageGo to next message
jmansion is currently offline  jmansion
Messages: 15
Registered: April 2006
Location: London
Promising Member
Just had a look at the code and was surprised to find no namespace.

Is there any plan to add one?

I'll definitely need one in my local build.
Re: Updated and new benchmarks [message #15384 is a reply to message #15382] Fri, 18 April 2008 17:15 Go to previous messageGo to next message
mirek is currently offline  mirek
Messages: 13975
Registered: November 2005
Ultimate Member
jmansion wrote on Fri, 18 April 2008 08:30

Just had a look at the code and was surprised to find no namespace.

Is there any plan to add one?

I'll definitely need one in my local build.



You must have been looking into wrong places. U++ is in Upp namespace.

Mirek
Re: Updated and new benchmarks [message #15392 is a reply to message #14774] Sun, 20 April 2008 16:00 Go to previous message
tvanriper is currently offline  tvanriper
Messages: 85
Registered: September 2007
Location: Germantown, MD, USA
Member
I think the namespace is hidden in a macro, so it's easy to miss.
Previous Topic: Upp 2008.2beta released
Next Topic: Scripts to build debian packages added to SVN repository
Goto Forum:
  


Current Time: Fri Mar 29 09:34:03 CET 2024

Total time taken to generate the page: 0.01738 seconds