[parsec-users] Blackscholes TBB Version: err_chk uses wrong tolerance

Yungang Bao ybao at CS.Princeton.EDU
Tue Dec 6 20:49:37 EST 2011


Hi Christian,

Thank you very much for the patch. I will use your patch to fix the problem in PARSEC 3.0

Best Regards,
Yungang

----- Original Message -----
From: "Christian Jülg" <c.juelg at googlemail.com>
To: parsec-users at lists.cs.princeton.edu
Sent: Tuesday, December 6, 2011 6:30:35 PM
Subject: [parsec-users] Blackscholes TBB Version: err_chk uses wrong	tolerance


Hi, 


as I noticed here 
https://lists.cs.pr incieton.edu/pipermail/parsec-users/2011-May/001135.html 


The error treshold for "err_chk" in blackscholes has been changed from "1e-5" to "1e-4" in Parsec 2 or 2.1. 


Yet the TBB version of the blackscholes benchmark still uses the old treshold which leads to errors on large datasets using "float". 


In the attachment is a patch that fixes the tolerance, bringing it in line with the other blackscholes variants. 


Best Regards 
Christian 
_______________________________________________
parsec-users mailing list
parsec-users at lists.cs.princeton.edu
https://lists.cs.princeton.edu/mailman/listinfo/parsec-users


More information about the parsec-users mailing list