cvs commit: src/lib/msun/alpha fenv.h src/lib/msun/amd64 fenv.h src/lib/msun/arm fenv.h src/lib/msun/i387 fenv.h src/lib/msun/ia64 fenv.h src/lib/msun/powerpc fenv.h src/lib/msun/sparc64 fenv.h

[Kris Kennaway]

On Fri, Jan 14, 2005 at 10:09:39AM -0800, David O'Brien wrote:
> On Fri, Jan 14, 2005 at 05:28:29AM -0800, Kris Kennaway wrote:
> > On Fri, Jan 14, 2005 at 08:24:33AM -0500, David Schultz wrote:
> > > On Fri, Jan 14, 2005, Kris Kennaway wrote:
> > > > On Fri, Jan 14, 2005 at 07:09:23AM +0000, David Schultz wrote:
> > > > > das         2005-01-14 07:09:23 UTC
> ...
> > > > >   Log:
> > > > >   Mark all inline asms that read the floating-point control or status
> > > > >   registers as volatile.  Instructions that *wrote* to FP state were
> > > > >   already marked volatile, but apparently gcc has license to move
> > > > >   non-volatile asms past volatile asms.  This broke amd64's feupdateenv
> > > > >   at -O2 due to a WAR conflict between fnstsw and fldenv there.
> > > > 
> > > > Do you think this is likely to the cause of the build failures
> > > > reported with a number of ports when world is compiled with -O2
> > > > (i.e. does the feupdateenv failure cascade to other commonly-used
> > > > parts of the code)?
> > > 
> > > No, this would lead to a runtime failure, and I would be surprised
> > > if very many ports use fenv.h anyway.
> > 
> > OK :-(
> > 
> > Someone needs to revert amd64 back to using -O in the meantime, then.
> 
> I can't tell if you're talking about a need to move back to -O is for
> /usr/ports or /usr/src?

src.  This was discussed on current@ in December; some amd64 ports do
not build when world is compiled with -O2 (even if the port itself is
built with -O), but build when world is recompiled with -O.

There was some analysis of the problem cause in the mailing list
thread (which you were CC'ed on), but no conclusion.

>  If for /usr/ports, we should just change ports/Mk/bsd.port.mk to
> s/-O2/-O/ on AMD64?  If we move back to -O in /usr/src we'll never
> get all the issues found and ironed out.

The issues have been reported, so someone either needs to fix them or
apply the workaround of reverting src to -O until the former can take
place.

Kris
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/cvs-src/attachments/20050114/236d59d5/attachment.bin