1 |
On Saturday 10 June 2006 00:28, Hemmann, Volker Armin wrote: |
2 |
> On Friday 09 June 2006 23:42, Vladimir G. Ivanovic wrote: |
3 |
> > On Fri, 2006-06-09 at 16:14 -0500, Barry.SCHWARTZ@×××××××××××××.org |
4 |
> > |
5 |
> > wrote: |
6 |
> > > Enough research has gone into writing code for IEEE floating point |
7 |
> > > that I would not try to bypass it without a good reason. It’s there |
8 |
> > > to be your friend. |
9 |
> > |
10 |
> > Performance is the reason we have hardware FPUs and -ffast-math. |
11 |
> |
12 |
> no, ffast-math is for the case, that you |
13 |
> |
14 |
> a) don't need accurate results |
15 |
> |
16 |
> b) the FPU is not fast enough |
17 |
> |
18 |
> |
19 |
> since you can't say if you need accurate math for an app you did not |
20 |
> wrote or examined, using ffast-math is highly dangerous. And sicne the |
21 |
> hardware gets faster on an almost monthly basis, it is even less |
22 |
> convincing why anybody should use it. |
23 |
|
24 |
I completely agree. The gcc info page puts it this way: |
25 |
This option should never be turned on by any `-O' option since it |
26 |
can result in incorrect output for programs which depend on an |
27 |
exact implementation of IEEE or ISO rules/specifications for math |
28 |
functions. |
29 |
|
30 |
In short there is a very good reason it is not on by default and may never |
31 |
be turned on by default. |
32 |
|
33 |
Paul |
34 |
|
35 |
-- |
36 |
Paul de Vrieze |
37 |
Gentoo Developer |
38 |
Mail: pauldv@g.o |
39 |
Homepage: http://www.devrieze.net |