-ffast-math option live up to its name?

[Tim Prince]

On 12/14/2010 12:28 PM, kevin diggs wrote:
> Does the -ffast-math option speed up operations enough to make
> fighting with mozilla/firefox source worth the trouble? I think the
> main problem is identifying NaNs.
>
If you want consistent treatment of NaNs, -ffast-math is not your 
choice.  The biggest speedups come in the auto-vectorization of sum and 
dot product reductions, and in the complex arithmetic shortcuts.  Those 
seem unlikely to be relevant to firefox, but I don't know why firefox 
would depend on correct treatment of NaNs either.

-- 
Tim Prince