Bruce Evans <[EMAIL PROTECTED]> wrote:
> On Sat, 10 Nov 2001, John Baldwin wrote:
> > On 08-Nov-01 Bruce Evans wrote:
> > > The i386 <machine/atomic.h> still uses archaic constraints for some
> > > input-output operands ("0" for the first operand). These never worked
> > > right and if fact don't actually work for compiling this file without
> > > optimization.
> > Hmm, would you prefer this diff then, I've had it floating around for a while
> > now but wasn't sure it was right:
> Yes, it is right provided all the operand renumbering is right. I suppose
> it can't be checked simply by comparing all objects, because it sometimes
> changes the register allocation?
> There are a couple more "0"s in atomic_cmpset_int(), and many more in other
> files (even in cpufunc.h).
Actually there are only 14 files that make use of "0" .. "9".
% find ./sys -type f | xargs grep -l ':[[:space:]]*"[0-9]"'
./sys/gnu/i386/fpemul/fpu_entry.c & ./sys/i386/i386/math_emulate.c
were quite easy to change. I'm not using math emulation, but the
changes to these two were almost trivial and I already mailed jhb a
hopefully correct diff. It's also online at
For the rest of the i386 files, it'll take the rest of my Sunday
afternoon and a buildworld or two, I guess. I can't make any sparc64
or alpha changes though.
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message