compiler bug in gcc version 2.8.1 (Tornado1) for PPC corrupts floationg point constants?

[Dirk Zimoch <dirk.zimoch@psi.ch>]

Hi all,

I have some driver software that contains

double scan_period;
scan_period = 0.05;

The function is executed from startup script, i.e. the executing task is 
a floating point task.

However, sometimes when booting R3.13.2, the ioc crashes with "floating 
point underflow" while executing

printf ("init_record_generic: scan_period=%f\n", scan_period);

I couldn't even look at the registers, because ti also crashes when it 
tries to print the register containing scan_period.

According to the bitpattern, scan_period is not 0.05, but some 
denormalized value.

Even more astonishing is that the error only happens sometimes! In rare 
cases, everything is OK. When I boot R3.13.5, also everything is OK. I 
have compared the driver library compiled for both EPICS versions. It is 
exactly the same object code.

We load the driver library in the startup script. Maybe there is 
something wrong with the ld command, corrupting constants?

But how can this depend on the EPICS release? Maybe the code is loaded 
to another location, but does the location matter?

Why does this only happen with one driver?

Any ideas or solutions?


Dirk

--
Dr. Dirk Zimoch
Swiss Light Source
Paul Scherrer Institut
Computing and Controls
phone +41 56 310 5182
fax   +41 56 310 4413