Failed paranoia test on leon2?

[mike]

Hi all,

Did anybody run paranoia on a leon2 with FPU? If not, do you use the function 'pow' on any hardware with FPU (not 
software floating point)?

- I ran  the 'paranoia' test in the RTEMS testsuites and ran it on a board with Atmel leon2 (AT697E) and saw that the 
function 'pow(x,y)' failed. The
test was compiled with 'hard-float' (I use RTEMS 4.10, gcc 4.4.6 built from distribution. I have also tried Gaisler's 
RTEMS 4.10 binaries with the same
failed results).

  -----------------------------------

Testing powers Z^i for small Integers Z and i.

ERROR: Severity: DEFECT:computing

(2.00000000000000000e+00) ^ (3.00000000000000000e+00)

yielded 3.44895534024799920e+16;

which compared unequal to correct 8.00000000000000000e+00 ;

they differ by 3.44895534024799840e+16 .

Errors like this may invalidate financial calculations

involving interest rates.

Similar discrepancies have occurred 260 times.

-------------------------------------

  Then I just compiled a simple app just calling pow(x,y) in a loop and the function failed with y > 2 (with x being any 
value, as long as y>2 then it will give incorrect result).

Any comments/suggestions that I can try?

  Thank you.

-mike-

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/users/attachments/20110721/a0186342/attachment-0001.html>