-ffast-math issues

[Orion Poplawski]

On 04/28/2015 11:35 AM, Orion Poplawski wrote:
> Re: https://bugzilla.redhat.com/show_bug.cgi?id=1127544
> 
> It appears that shared libraries compiled with -ffast-math (as libopenjpeg.so
> is) will call set_fast_math() in the library init code:
> 
> Watchpoint 1: $mxcsr
> 
> Old value = [ IM DM ZM OM UM PM ]
> New value = [ DAZ IM DM ZM OM UM PM FZ ]
> 0x00007fffe619d112 in set_fast_math () from /lib64/libopenjpeg.so.1
> (gdb) bt
> #0  0x00007fffe619d112 in set_fast_math () from /lib64/libopenjpeg.so.1
> #1  0x00007ffff7deaf2a in call_init (l=<optimized out>, argc=argc at entry=1,
>     argv=argv at entry=0x7fffffffd6a8, env=env at entry=0x7fffffffd6b8) at dl-init.c:76
> 
> This will then affect the floating point operations of all code run after the
> library is loaded, regardless of its use of -ffast-math or not, which can lead
> to some unexpected behavior.
> 
> This seems like it's a pretty broad and generic problem and I'm wondering how
> best to handle it.
> 

So, the solution appears to be that libopenjpeg (and any other library) should
not link with -ffast-math, but only compile with it.

https://bugzilla.redhat.com/show_bug.cgi?id=881489

This appears to be the gcc bug mentioned in that report:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55522


-- 
Orion Poplawski
Technical Manager                     303-415-9701 x222
NWRA, Boulder/CoRA Office             FAX: 303-415-9702
3380 Mitchell Lane                       orion at nwra.com
Boulder, CO 80301                   http://www.nwra.com