[3304] in Athena Bugs
compiler differences
daemon@ATHENA.MIT.EDU (geer@ATHENA.MIT.EDU)
Wed Sep 27 09:57:25 1989
From: geer@ATHENA.MIT.EDU
To: bugs@ATHENA.MIT.EDU
Date: Wed, 27 Sep 89 09:57:07 EDT
------- Forwarded Message
Received: by E40-PO.MIT.EDU (5.45/4.7) id AA15850; Tue, 26 Sep 89 13:23:10 EDT
Received: from MOZART.MIT.EDU by ATHENA.MIT.EDU with SMTP
id AA24836; Tue, 26 Sep 89 13:24:47 EDT
From: nschmidt@ATHENA.MIT.EDU
Received: by E40-343A-2.MIT.EDU (5.61/4.7) id AA10681; Tue, 26 Sep 89 13:22:59 -0400
Message-Id: <8909261722.AA10681@E40-343A-2.MIT.EDU>
To: geer@ATHENA.MIT.EDU
Subject: Matlab vax vs rt versions
Date: Tue, 26 Sep 89 13:22:55 EDT
In tracing down the bugs in the new Matlab, I have discovered that the
versions compiled for the vax and the rt give different answers in
calculating a hairy floating point function (kaiser function, which uses
bessel functions). The rt version gives the same answer as did the
previous version of Matlab, which ran only on the vax. The new vax
version gives a divide by zero error. When the Matlab folks run under
ultrix, they get the correct answer. They are looking into it at
their end, but Anne suggested that I let you know about it, since
there is a history of some floating point problems in our kernel.
Naomi
------- End of Forwarded Message
