Re: Kernel FP software completion
On Monday 25 January 1999, at 23 h 39, the keyboard of John Goerzen
<jgoerzen@complete.org> wrote:
> This option is required for IEEE compliant floating point arithmetic
> on the Alpha.
...
> So, my question is, why does Alpha suddenly require math emulation when it
> didn't before? Isn't FP one of the Alpha's strongest areas?
It is *not* FP emulation and it has nothing to do with i387 emulation on the
PC.
The problem is the following: by default, the Alpha traps when you divide by
zero. In the IEEE standard, this is wrong (yes, you can divide by zero,
according to the Standard). So, it can break some software on the Alpha when
they rely on IEEE behaviour, which is common (see my clustalw package for an
example, I had to recompile it with -mieee for it to work). Hence this
workaround.
Reply to: