Re: Kernel FP software completion
On Tue, 26 Jan 1999, James D. Freels wrote:
> I hope this is what I think it is! What has been missing for me to
> use Linux/Alpha instead of DEC-Unix/Alpha is the floating point
> problem. Am I right in assuming that with this kernel correction,
> that DEC-Unix floating point errors (allowed lets say with the
> DEC-Unix FORTRAN compiler -fpe1 switch) will now work correctly under
> Linux/Alpha? Further, will this kernel ability allow one to not
> utilize the -mieee switch on the g77 compiler and, hence, allow code
> to run faster under Linux/Alpha?
I doubt it. If it's what I think it is (not having looked at the code at
this time in the 2.2.0 tree), it's been in 2.1 kernels awhile, but was
often "defaulting" to enabled without a 'make config' option.
Anyway, RTH explained it to me once and, yes, you do need it, but if it's
just a reiteration of what's been in the 2.1 kernels, it's not "the answer
to all things needing IEEE math". I cannot recall (or find in my
archives) the exact explanation of it, but I do know that the code is
fairly documented and will explain much more about what it actually does.
I'm compiling up a 2.2.0 kernel right now and will repost with the details
when it's through....
C
Reply to: