This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] g95 SEGV building libgfortran
On Friday 05 September 2003 1:36 pm, Diego Novillo wrote:
> On Thu, 2003-09-04 at 17:57, Paul Brook wrote:
> > My vote is to leave LOOP_EXPRs in the interface to the frontends (ie
> > GENERIC). Otherwise we're going to end up with large amounts of
> > lowering code duplicated between the different frontends.
>
> Yes. LOOP_EXPR should remain in GENERIC. However, the problem here is
> that I think tree_rest_of_compilation should be the one calling the
> gimplifier.
>
> Paul, IIUC, the Fortran FE does not call the gimplifier directly, but
> calls t_r_o_c? At the moment, each front end is calling the gimplifier
> on its own. When compiling Fortran, the optimizers are now being fed
> GENERIC trees. That is not right.
Correct.
> I'm OK with removing LOOP_EXPR from GIMPLE, but it should remain in
> GENERIC. The problem now is making sure all the FEs call into the
> gimplifier via a common interface. tree_rest_of_compilation seems a
> good candidate.
>
> Richard, can t_r_o_c be that interface now? If not, the Fortran FE
> should be calling into the gimplifier on its own until we fix the
> missing bits.
I've a big stack of other patches waiting to be applied, so I'll probably
add a call to the gimplifier as a temporary fix.
I haven't checked recently, but last time I looked the tree optimizers still
had issues with nested functions, so I might end up disabling them
altogether. Which reminds me, I had a testcase fo these which I never
checked in. I'll see if I can dig it up (unless someone else already put
one in).
Paul