| 1 |
On Friday 23 February 2007, Chris Gianelloni wrote: |
| 2 |
> On Fri, 2007-02-23 at 01:40 +0100, Carsten Lohrke wrote: |
| 3 |
> > And I'd be fond of having all the -ffast-math filtering ripped out of the |
| 4 |
> > tree as well. |
| 5 |
> |
| 6 |
> Except some things really do not compile with it enabled. Now, if |
| 7 |
> you're meaning you'd prefer patch every compilation failure using |
| 8 |
> -ffast-math instead, then I'd say go for it. Patches are always a |
| 9 |
> better solution than workarounds. |
| 10 |
|
| 11 |
filtering encourages users to continue to put -ffast-math into their global |
| 12 |
CFLAGS ... this is just plain wrong |
| 13 |
|
| 14 |
ignoring compile issues, fast math can break runtime code that actually |
| 15 |
requires the IEEE compliant behavior ... the only people who know |
| 16 |
whether -ffast-math is safe are the upstream maintainers ... leave using the |
| 17 |
flag to them |
| 18 |
-mike |
Archives