Re: SIGFPE and -mieee

[Tyson Whitehead <twhitehe@uwo.ca>]

> > I'd also suggest -ffast-math turns on -mno-ieee.
>
> It makes sense.......OTOH, my understanding is that -ffast-math
> is only supposed to turn on options that produce faster but
> slightly incorrect results.  This is something a little
> different--it doesn't produce "slightly incorrect results," it
> causes certain apps to outright crash!  I don't know if this is
> such a good thing to go under -ffast-math.

I also have to confess, that I don't know how to hook it any options but the 
-m ones cleanly (even figuring that out took a couple of days *grin*).

The -m switches are nice, because, as it turned out, they are all handled 
neatly though this special define statement (in the architecture specific 
configuration file).  The define creates an array of structures that 
specifies the switches, a help line (for --target-help), bits to set/unset in 
architecture specific mask, and the default setting.

It made for a very clean patch.  I just had to add an entry that clears the 
IEEE bits, and modified the default entry to included them set.  I don't 
think crossing non-architecture specific flags with architecture specific 
flags would be so clean.

From reading the man pages, it seems that architecture specific options stand 
by themselves (i.e. I couldn't find any other none of the non-architecture 
specific flags that set/unset -m options).

I think this is because there seems to be (I'm not an expert) a very clear 
break in the compiler between the intermediate RTL code (on which 
optimization, etc, are performed) and the final translation into actual 
machine specific assembler.

Later  -T

-- 
 Tyson Whitehead  (-twhitehe@uwo.ca -- WSC-)
 Computer Engineer                        Dept. of Applied Mathematics,
 Graduate Student- Applied Mathematics    University of Western Ontario,
 GnuPG Key ID# 0x8A2AB5D8                 London, Ontario, Canada