I just saw a talk that somewhat covered TBB. See slides at
http://www.ncsa.uiuc.edu/UserInfo/Training/Workshops/Multicore/presentations/Intel_Threading_Tools.ppt
It's closer to OpenMP than pthreads. One idiom I recall is to change your
loop body to an operator, possibly adding locks or atomics, and then feed
the operator to a parallel_for template. The runtime has a neat method of
splitting the loops to improve cache performance.
They went open source because programmers were unwilling to adopt a
programming system that wasn't portable to non-Intel/AMD platforms. The
main porting work is writing the assembler to do atomic operations.
-Jim
On Thu, 26 Jul 2007, Bill Broadley wrote:
I saw a bunch of press on intel's new Threading Building Blocks 2.0. Seems
similar to a couple other products promising easier parallelism, even down
to a contest for the best use.
Except:
* GPL v2 - an actual open license, not a fake open license.
* Not OS specific: Windows, mac, and linux (I even heard opensolaris and at
least one of the *BSDs mentioned)
* Not compiler specific: gcc, intel, and microsoft
* AMD, Intel, and even found mentions of it working on a G5
Sounds like a potentially useful technology, especially with the ever
increasing number of threads becoming available inside a node and the
ever increasing difference between onchip communication and outside
node communication.
I've downloaded some examples but they have all been windows specific so far
(sigh).
Anyone look at it closer? Is it much different than say pthreads?
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
