[releng_6 tinderbox] failure on sparc64/sparc64

Marius Strobl marius at alchemy.franken.de
Fri Feb 3 04:09:10 PST 2006

On Fri, Feb 03, 2006 at 09:09:43AM +0100, Harti Brandt wrote:
> On Thu, 2 Feb 2006, Dag-Erling Sm?rgrav wrote:
> DS>Scott Long <scottl at samsco.org> writes:
> DS>> I've been trying to reproduce this on my local hardware, but I can't
> DS>> trigger it.
> DS>
> DS>The ISP driver abuses the inline keyword.  As I told mjacob earlier,
> DS>the extensive inlining not only breaks the build, but probably hurts
> DS>performance as well.
> DS>
> DS>(what gcc is complaining about, specifically, is that expanding calls
> DS>to inlined functions causes isp_target_notify() to grow by more than
> DS>100%)
> The interesting point is: why does it build on my real sparc (2-UII CPUs, 
> 512MByte memory), but not on the tinderbox. Is there something about the 
> crosscompiler that is different?

GCC apparently does different intermediate optimizations when built
as a cross compiler than the native one; in the past we've e.g. seen
a GCC bug in the machine code generation that was only triggered
when GCC was built as cross compiler and fed with differently
optimized intermediate code due to that. Interestingly the resulting
object files generated by the cross compiler and the native one for
the source file where this triggered where the same once the bug
in the machine code generation was fixed.


This mail was scanned by AntiVir Milter.
This product is licensed for non-commercial use.
See www.antivir.de for details.

More information about the freebsd-stable mailing list