https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124266
Richard Sandiford <rsandifo at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |acoplan at gcc dot gnu.org,
| |rsandifo at gcc dot gnu.org
--- Comment #7 from Richard Sandiford <rsandifo at gcc dot gnu.org> ---
As mentioned in reply to Wilco's patch, I think the problem here is that
pair-fusion is dropping RTX_FRAME_RELATED_P when combining a REG_CFA_RESTORE
load with a non-frame-related load. It is correctly preserving the notes
themselves though.
The following proof-of-concept patch fixes it for me:
diff --git a/gcc/pair-fusion.cc b/gcc/pair-fusion.cc
index 7aaf4fd6372..c5469993531 100644
--- a/gcc/pair-fusion.cc
+++ b/gcc/pair-fusion.cc
@@ -1645,6 +1645,8 @@ pair_fusion_bb_info::fuse_pair (bool load_p,
PATTERN (first->rtl ()),
PATTERN (second->rtl ())
};
+ bool frame_related_p = (RTX_FRAME_RELATED_P (first->rtl ())
+ || RTX_FRAME_RELATED_P (first->rtl ()));
// Make copies of the patterns as we might need to refer to the original RTL
// later, for example when updating debug uses (which is after we've updated
@@ -2027,6 +2029,8 @@ pair_fusion_bb_info::fuse_pair (bool load_p,
for (auto uid : tombstone_uids)
track_tombstone (uid);
+ RTX_FRAME_RELATED_P (pair_change->insn ()->rtl ()) = frame_related_p;
+
return true;
}
but I'm not sure that's the best way.
