llvm dev - Aug 2008 - [LLVMdev] Floating point optimizations

Hi All,

The LLVM Developer Meeting was a great success.  Many thanks to all  
participants, speakers and sponsors!

For anyone interested, the slides and videos are now available online:
http://llvm.org/devmtg/2008-08/

Speakers: If you would like me to make any changes to your slides or  
abstract, please let me know off-list.

-Chris

Hi,

I have some code which relies on the floating point rounding mode to  
be set to round-to-negative-infinity.  This means that there are some  
floating point optimizations which are legal with the default round-to- 
nearest mode but not with rounding to negative infinity.  Does LLVM  
reorder or otherwise optimize floating point operations?  I've looked  
at the options for "opt" but don't see anything to do with
floating
point so I assume no.

Brendan Younger

On Aug 5, 2008, at 10:18 AMPDT, Brendan Younger wrote:
> Hi,
>
> I have some code which relies on the floating point rounding mode to
> be set to round-to-negative-infinity.  This means that there are some
> floating point optimizations which are legal with the default round- 
> to-
> nearest mode but not with rounding to negative infinity.  Does LLVM
> reorder or otherwise optimize floating point operations?  I've looked
> at the options for "opt" but don't see anything to do with
floating
> point so I assume no.
The intent is that it does optimizations as permitted by IEEE754   
(i.e. negative zeroes, NaNs, and infinities should be handled  
correctly), and more aggressive ones with --enable-unsafe-fp-math (you  
don't want this).   However we are not modelling the rounding mode  
yet, so it's likely this doesn't fully work as you want.   Rounding  
mode handling is something we want to add, but so far there have  
always been more important things.