RE: Gcj build instructions

RE: Gcj build instructions

[Boehm, Hans]

Tom -

I finally got a little bit of time to investigate.  Apparently it currently
only works on X86 and IA64.  (And HP/UX, but AFAIK the rest of gcj still
doesn't work there.)  I suspect we just need to point the GC at the right
compare-and-swap and memory barrier definitions to make it work on other
platforms.  But that clearly won't happen for 3.1.

I think the right description for --enable-parallel-mark is something like:

The enables parallel garbage collection in the Boehm GC.  It causes the
collector to create additional threads which will wake up during garbage
collection to help trace live data.  This may lead to substantial
performance improvements on multiprocessors.  It typically leads to a mild
slowdown on uniprocessors.  Currently only supported on Linux/X86 and
Linux/IA64.  Should normally be enabled if the expected target is a
multiprocessor platform of one of the two preceding types. 

I haven't entirely made sense out of what documentation is where, or where
such things should be described.  It should probably be described in roughly
the same place as --disable-jvmpi.  But I couldn't find a description of
that either.  Should there be a list of Java runtime configuration options
and their descriptions hanfing from
http://gcc.gnu.org/java/build-snapshot.html ?

Hans

> -----Original Message-----
> From: Tom Tromey [mailto:tromey@redhat.com]
> Sent: Friday, April 19, 2002 5:07 PM
> To: Boehm, Hans
> Cc: 'java@gcc.gnu.org'
> Subject: Re: Gcj build instructions
> 
> 
> >>>>> "Hans" == Boehm, Hans <hans_boehm@hp.com> writes:
> 
> Hans> Should the gcj build instructions also mention
> Hans> --enable-parallel-mark?
> 
> Yes, definitely.  Which instructions in particular?
> gcc/doc/install.texi?  Are there otehrs?
> 
> Hans> I haven't done a systematic study with gcj recently, but my
> Hans> impression is that it's a substantial win for multiprocessor
> Hans> servers, and a mild loss for uniprocessors.  I believe it works
> Hans> reliably on at least X86 and IA64 Linux.  I thought others had
> Hans> also used it on other platforms?
> 
> If you can dig up the information I'll be happy to write the patch.
> 
> Tom
> 

Re: Gcj build instructions

[Tom Tromey]

>>>>> "Hans" == Boehm, Hans <hans_boehm@hp.com> writes:

Hans> Should the gcj build instructions also mention
Hans> --enable-parallel-mark?

Yes, definitely.  Which instructions in particular?
gcc/doc/install.texi?  Are there otehrs?

Hans> I haven't done a systematic study with gcj recently, but my
Hans> impression is that it's a substantial win for multiprocessor
Hans> servers, and a mild loss for uniprocessors.  I believe it works
Hans> reliably on at least X86 and IA64 Linux.  I thought others had
Hans> also used it on other platforms?

If you can dig up the information I'll be happy to write the patch.

Tom

Gcj build instructions

[Boehm, Hans]

Should the gcj build instructions also mention --enable-parallel-mark?  I
haven't done a systematic study with gcj recently, but my impression is that
it's a substantial win for multiprocessor servers, and a mild loss for
uniprocessors.  I believe it works reliably on at least X86 and IA64 Linux.
I thought others had also used it on other platforms?  (It doesn't work on
Solaris, since it's incompatible with the thread-stopping code there.  We
would have to switch to using linux_threads.c there to make it work.)

Hans