PATCH: libjava FreeBSD configuration
Loren James Rittle
rittle@latour.rsch.comm.mot.com
Thu Apr 4 20:30:00 GMT 2002
> Yes, this looks good, please check it in.
OK, done.
> I suspect/hope that we will soon be able to get rid of
> -fcheck-references on more platforms once we have the
> sigcontext->unwind context support, this should make it easier to do
> that for specific platforms.
Excellent, I am a fish out of water at that level.
>> FYI, in case anyone wonders, I am 100% positive that we don't want
>> slow_pthread_self=yes on FreeBSD since it is merely a libc function
>> that returns a global variable.
> Hmm, note that slow_pthread_self avoids the call overhead into libc for
> the common case also.
True, but, comparing the generated asm, the code path in
_Jv_ThreadSelf() [when SLOW_PTHREAD_SELF] is longer and contains an
extra conditional branch than the branch-free pthread_self()
implemented as...
> BTW how can a thread ID be in a global variable?
...at *every* task switch, the global variable that pthread_self()
will return is updated (it is used for other internal purposes as
well). This simple implementation does not allow concurrent threads
of a process to span multiple CPUs (as I'm fairly sure Solaris and
modern Linux implementations do).
BTW, I only played with the tuning knob as was given to me... ;-)
Regards,
Loren
More information about the Java-patches
mailing list