[ELDK] [maybe ot] PPC e500v2 floating-point in eldk 5.4

[Albrecht Dreß]

Hi all,

I use ELDK 5.4 to develop software for a PowerPC e500v2 (QorIQ P2020) system, and have a (hopefully not too off-topic) question regarding the floating-point implementation in glibc.

Although I did not enable any of the C99 floating-point exceptions, one of my applications throws SIGFPE, according to the backtrace in exp() (__exp_finite).  Unfortunately, I do not have a core image, so I cannot check which input value triggered the signal (which is wrong anyway, IMHO, as no exception is enabled).

When I tried to reproduce the effect with a stripped-down version of the code (which didn't work, i.e. I could /not/ provoke the signal), I noticed that exp(NAN) returns 1.79769e+308.  If I understand IEEE-754 correctly, this is wrong, as any operation involving NAN shall result in NAN.  Actually, running exp(NAN) on a PC (Ubuntu 14.04 w/ libc 2.19) correctly returns NAN.

I'm not sure if these two effects are linked, but might this be an indication that the exp() implementation is somehow broken for the e500v2?

Thanks in advance for any insight,
Albrecht.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 190 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/eldk/attachments/20151119/d040a542/attachment.sig>