* [PATCH] Fix PR83518 for aarch64
@ 2019-07-23 10:09 Richard Biener
0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2019-07-23 10:09 UTC (permalink / raw)
To: gcc-patches
This is the last piece to fix the testcase for aarch64 and it's
gimplification of int a[] = { 1,2,3....} to a = *.LC0;
It also fixes having calls in GIMPLE FE IL when emitting a CFG.
Bootstrapped / tested on x86_64-unknown-linux-gnu, applied.
Richard.
2019-07-23 Richard Biener <rguenther@suse.de>
PR tree-optimization/83518
* tree-ssa-sccvn.c (vn_reference_lookup_3): Handle aggregate
init from a constant even when partial defs are already recorded.
c/
* gimple-parser.c (c_parser_parse_gimple_body): When we have
a CFG also rebuild cgraph edges.
* gcc.dg/tree-ssa/ssa-fre-79.c: New testcase.
Index: gcc/testsuite/gcc.dg/tree-ssa/ssa-fre-79.c
===================================================================
--- gcc/testsuite/gcc.dg/tree-ssa/ssa-fre-79.c (nonexistent)
+++ gcc/testsuite/gcc.dg/tree-ssa/ssa-fre-79.c (working copy)
@@ -0,0 +1,29 @@
+/* { dg-do run } */
+/* { dg-options "-O -fgimple -fdump-tree-fre1" } */
+
+struct S { char a[4]; };
+const struct S cs = { 1, 2, 3, 4 };
+
+int __GIMPLE(ssa,startwith("fre"))
+main ()
+{
+ struct S s;
+ short _1;
+
+ __BB(2):
+ s = cs;
+ s.a[1] = _Literal (char) 3;
+ _1 = __MEM <short, 1> (&s + 1);
+ if (_1 != _Literal (short) 0x303)
+ goto __BB3;
+ else
+ goto __BB4;
+
+ __BB(3):
+ __builtin_abort ();
+
+ __BB(4):
+ return 0;
+}
+
+/* { dg-final { scan-tree-dump-not "abort" "fre1" } } */
Index: gcc/c/gimple-parser.c
===================================================================
--- gcc/c/gimple-parser.c (revision 273657)
+++ gcc/c/gimple-parser.c (working copy)
@@ -356,6 +356,8 @@ c_parser_parse_gimple_body (c_parser *cp
gcov_type t = PARAM_VALUE (PARAM_GIMPLE_FE_COMPUTED_HOT_BB_THRESHOLD);
set_hot_bb_threshold (t);
update_max_bb_count ();
+ cgraph_node::get_create (cfun->decl);
+ cgraph_edge::rebuild_edges ();
}
dump_function (TDI_gimple, current_function_decl);
}
Index: gcc/tree-ssa-sccvn.c
===================================================================
--- gcc/tree-ssa-sccvn.c (revision 273667)
+++ gcc/tree-ssa-sccvn.c (working copy)
@@ -2702,9 +2702,7 @@ vn_reference_lookup_3 (ao_ref *ref, tree
&& gimple_assign_single_p (def_stmt)
&& (DECL_P (gimple_assign_rhs1 (def_stmt))
|| TREE_CODE (gimple_assign_rhs1 (def_stmt)) == MEM_REF
- || handled_component_p (gimple_assign_rhs1 (def_stmt)))
- /* Handling this is more complicated, give up for now. */
- && data->partial_defs.is_empty ())
+ || handled_component_p (gimple_assign_rhs1 (def_stmt))))
{
tree base2;
int i, j, k;
@@ -2808,8 +2806,30 @@ vn_reference_lookup_3 (ao_ref *ref, tree
/* Try folding the new reference to a constant. */
tree val = fully_constant_vn_reference_p (vr);
if (val)
- return vn_reference_lookup_or_insert_for_pieces
- (vuse, vr->set, vr->type, vr->operands, val);
+ {
+ if (data->partial_defs.is_empty ())
+ return vn_reference_lookup_or_insert_for_pieces
+ (vuse, vr->set, vr->type, vr->operands, val);
+ /* This is the only interesting case for partial-def handling
+ coming from targets that like to gimplify init-ctors as
+ aggregate copies from constant data like aarch64 for
+ PR83518. */
+ if (maxsize.is_constant (&maxsizei)
+ && known_eq (ref->size, maxsize))
+ {
+ pd_data pd;
+ pd.rhs = val;
+ pd.offset = 0;
+ pd.size = maxsizei / BITS_PER_UNIT;
+ return data->push_partial_def (pd, vuse, maxsizei);
+ }
+ }
+
+ /* Continuing with partial defs isn't easily possible here, we
+ have to find a full def from further lookups from here. Probably
+ not worth the special-casing everywhere. */
+ if (!data->partial_defs.is_empty ())
+ return (void *)-1;
/* Adjust *ref from the new operands. */
if (!ao_ref_init_from_vn_reference (&r, vr->set, vr->type, vr->operands))
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2019-07-23 9:59 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-07-23 10:09 [PATCH] Fix PR83518 for aarch64 Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).