better -ffast-math handling

[rdieter]

issue #488

See also
http://public.kitware.com/pipermail/cmake/2015-April/060479.html

[mayeut]

@rdieter,

Could you update your PR to include the same guards as in commit c5dd603 please ?

[malaterre]

I do not believe we handle this in openjpeg

[mayeut]

We do on master and now on branch openjpeg-1.5 after commit 650df81

[malaterre]

I believe that openjpeg lossless mode currenly only works with -fexcess-precision=standard

[mayeut]

Lossless tests are okay without -fexcess-precision=standard since #591 and #571 were fixed. Am I missing something ?

[malaterre]

That's precisely my point. OpenJPEG lossless code path was actually lossy until a couple of weeks back (project is > 10 yrs old!). So I am suggesting to be a little bit modest and do not pretend we handle those advanced options for now. If distro packager wants to use this flags then it is their responsabilities, that's fine.

[mayeut]

I understand your point. However this is another issue than the one taken care of here which used to mess with floating point configuration for people linking with openjpeg.

[malaterre]

point taken. sorry for the noise.