[parisc-linux] [2.5] next issues ...

[John David Anglin]

> no PA 2.0 here.

If you use a PA 2.0 machine and don't care if your code is compatible
with older machines, then using PA 2.0 code provides a number of
advantages.  Probably, the most important is the change in the maximum
branch distance for pc-relative calls (22-bit vs. 17-bit relocation).
In large programs, it easy to exceed the maximum distance of the 17-bit
relocation.

> > was picking, I would go with the current mainline 3.3 or the debian 3.0.
> 
> ok, how and where to get 3.3 from ?
> 	- subversions.gnu.org cvs ?

This is a mirror of the main gcc source.  I use it except for submissions.

> 	- cvs.parisc-linux.org cvs ? (looks older)
> 
> I am going to dig into this.
> 
> What about binutils 2.13.90.0.10-2 ... do you know what changes where
> there from previous versions ?

I use the main binutils source.  Don't know anything about the debian
changes.  They should be fed back for review if the have any merit.
However, I think Alan Modra did that awhile ago when he was working
on the PA toolchain.

> Ok, first thing I am going to do is build with gcc 3.0.4 and see if it
> works.

Be aware that debian 3.0.4 suffers from a problem where calls that have
a branch distance exceeding the max for a 17-bit relocation are not
correctly changed to an indirect call in all cases.  This will cause a
problem building expr.c in stage1 if you don't include "-O2" in STAGE1_CFLAGS.
This is fixed in 3.1 and latter.  The 3.1/3.2 fix didn't include millicode
calls or PA 2.0 adjustments.  That's now in 3.3.

I'll be glad when debian 3.0.4 is gone.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)