[parisc-linux] HPPA NPTL Support committed to libc-ports, help

[John David Anglin]

> > > FAIL: gcc.c-torture/execute/builtin-bitops-1.c execution,  -O3
> > > -fomit-frame-pointer -funroll-loops
> > > FAIL: gcc.c-torture/execute/builtin-bitops-1.c execution,  -O3
> > > -fomit-frame-pointer -funroll-all-loops -finline-functions
> >
> > This is a known regression.  See
> > <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26244>.
> 
> I don't quite understand what you are testing here?

The tests compare the results from various builtin functions
against that generated by rather simple implementations compiled
with various options.

We are failing at least two tests at -O3.  For example, we incorrectly
calculate the parity of long longs when bit 31 is 1.  The bit is
counted twice (the builtin code generates the correct result).  It's
quite hard to follow what's happening due to the loop unrolling.

I've spent some time looking at this but I haven't fully identified
where things go wrong.  It's somewhere in the rtl dumps posted with
the simplified testcase.  The problem was exposed by changes to loop.
However, the test code works correctly when compiled without inlining.

Dave
-- 
J. David Anglin                                  dave.anglin at nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)