llvm dev - Apr 2010 - [LLVMdev] SIGFPE generation

Just a quick question...

I've been attempting to trap floating point badness by setting up a  
SIGFPE handler. However, I'm not seeing it trigger, even for simple  
test code that forces division by zero. I'm generating bitcode with  
llvm-gcc and then executing it with a (modified) version of the  
jitter, for reference.

Is there any arcane magic necessary to turn on SIGFPE generation?

Thank you in advance,
Sarah

On Apr 9, 2010, at 2:52 PM, Sarah Thompson wrote:
> Just a quick question...
> 
> I've been attempting to trap floating point badness by setting up a  
> SIGFPE handler. However, I'm not seeing it trigger, even for simple  
> test code that forces division by zero. I'm generating bitcode with  
> llvm-gcc and then executing it with a (modified) version of the  
> jitter, for reference.
> 
> Is there any arcane magic necessary to turn on SIGFPE generation?

Most platforms default to "non-stop" mode, where floating-point
exceptions do not trap. The interfaces defined in <fenv.h> can be
used to query floating-point exceptions. Trapping on exceptions
can usually be enabled, though it requires platform-specific code.
See your platform's documentation for details.

Note that enabling traps can cause degraded performance, and it
can cause incompatibilities with library code that doesn't expect
it.

Dan