On Tue, 11 Apr 2023, Jakub Jelinek wrote:
> On Wed, Apr 05, 2023 at 02:11:08PM +0200, Richard Biener wrote:
> > Ok, but maybe there?s a gimple_build overload that meanwhile implements
> > the desired semantics? It would probably need to specify the valueization
> > hook to follow SSA edges beyond the sequence we?re currently building.
>
> Jeff has apparently committed the patch in the meantime.
> For gimple_build, did you mean to use it just for this fallback case,
> or instead of the initial resimplification as well?
I was originally thinking of the initial resimplification.
But yes, using gimple_build would have simplified it to
if (!res)
res = gimple_build (&stmts, BIT_FIELD_REF, type, t, bitsize, bitpos);
With the gsi overloads the whole function should now be able to
be turned into
return gimple_build (gsi, true, GSI_SAME_STMT, UNKNOWN_LOCATION,
BIT_FIELD_REF, type, t, bitsize, bitpos);
I think. Note that's not 100% equivalent to what we do now, so it's
probably better to defer to stage1. What it doesn't do is
/* We're using the resimplify API and maybe_push_res_to_seq to
simplify the BIT_FIELD_REF but restrict the simplification to
a single stmt while at the same time following SSA edges for
simplification with already emitted CTORs. */
which is achieved by passing NULL as the gimple_seq * argument to
.resimplify. The above replacement would instead do what a later
forwprop pass would do when folding. ISTR I wanted to do minimal
changes when rewriting this.
Richard.
> > > 2023-04-05 Jakub Jelinek <ja...@redhat.com>
> > >
> > > PR tree-optimization/109392
> > > * tree-vect-generic.cc (tree_vec_extract): If maybe_push_res_to_seq
> > > fails, build BIT_FIELD_REF insn without trying to simplify it.
> > >
> > > * gcc.dg/pr109392.c: New test.
> > >
> > > --- gcc/tree-vect-generic.cc.jj 2023-03-23 10:02:18.997935620 +0100
> > > +++ gcc/tree-vect-generic.cc 2023-04-04 14:28:32.977729134 +0200
> > > @@ -174,7 +174,16 @@ tree_vec_extract (gimple_stmt_iterator *
> > > opr.resimplify (NULL, follow_all_ssa_edges);
> > > gimple_seq stmts = NULL;
> > > tree res = maybe_push_res_to_seq (&opr, &stmts);
> > > - gcc_assert (res);
> > > + if (!res)
> > > + {
> > > + /* This can happen if SSA_NAME_OCCURS_IN_ABNORMAL_PHI are
> > > + used. Build BIT_FIELD_REF manually otherwise. */
> > > + t = build3 (BIT_FIELD_REF, type, t, bitsize, bitpos);
> > > + res = make_ssa_name (type);
> > > + gimple *g = gimple_build_assign (res, t);
> > > + gsi_insert_before (gsi, g, GSI_SAME_STMT);
> > > + return res;
> > > + }
> > > gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
> > > return res;
> > > }
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)
