[Bug 500] Shark on ARM/x86_64 Math.log incosistent results

[bugzilla-daemon at icedtea.classpath.org]

http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=500


xerxes at zafena.se changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gbenson at redhat.com
           Platform|arm                         |all
            Summary|Shark on ARM Math.log       |Shark on ARM/x86_64 Math.log
                   |incosistent results         |incosistent results




------- Comment #2 from xerxes at zafena.se  2010-05-25 08:06 -------
Tom Rodriguez have given some good background info on this bug.
http://mail.openjdk.java.net/pipermail/jdk6-dev/2010-May/001619.html
"The test applies to a bug which was fixed in hs17 so testing with hs14 is sure
to fail on x86_64 which is where the bug was."

To sum it up:

In order to fix Shark on various architectures  we must make sure that the zero
interpreter and the shark jit uses the same intrinsic for log calculations and
other math intrinsics that uses some sort of aproximation in order to calculate
the result.

Its ok for the result to be within +- 1ulps from matematically correct answer
using the aproximation math intrinsics.

Its not ok for the calculated result to change when the code are moved from the
interpreter to the jit.


-- 
Configure bugmail: http://icedtea.classpath.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.