This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Some Haifa scheduler bugs
- To: Paul Koning <pkoning at xedia dot com>
- Subject: Re: Some Haifa scheduler bugs
- From: David Edelsohn <dje at watson dot ibm dot com>
- Date: Wed, 03 Sep 97 14:49:18 -0400
- Cc: law at cygnus dot com, crux at Pool dot Informatik dot RWTH-Aachen dot DE, egcs at cygnus dot com
>>>>> Paul Koning writes:
Paul> "law@cygnus.com" wrote:
>> However, for execution tests that fail, I wonder how many (if any) are
>> caused by -fsched-spec-load-dangerous. If I remember right that
>> relies on the OS to support non-faulting loads from zero.
Paul> If that switch indeed depends on not faulting references to zero, it
Paul> probably isn't worth having. Or if it's kept it needs a warning that
Paul> in many OS's it will fail. The reason is that faulting on references
Paul> to zero is universally considered a Good Thing and many OS's go out of
Paul> their way to provide that feature.
I disagree with your justification that faulting on dereferncing
address zero is considered a Good Thing. This depends on your priorities
and the correctness of your program. I do not want to get into a general
discussion about programming styles and kernel design, but speculatively
dereferencing zero is very important for good performance on some systems;
some operating systems go out of their way specifically to make page zero
valid, as opposed to making it a VM error. This needs to be set in the
configuation file on a system-by-system basis and not universally enabled
or disabled.
David