On 09/23/2016 02:21 AM, Segher Boessenkool wrote:
A restore is supposed to restore some certain register. Restoring it
into some other register will not work. Don't.
2016-09-23 Segher Boessenkool <seg...@kernel.crashing.org>
* regrename.c (build_def_use): Invalidate chains that have a
REG_CFA_RESTORE on some instruction.
---
gcc/regrename.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/gcc/regrename.c b/gcc/regrename.c
index 54c7768..00a5d02 100644
--- a/gcc/regrename.c
+++ b/gcc/regrename.c
@@ -1867,6 +1867,12 @@ build_def_use (basic_block bb)
scan_rtx (insn, &XEXP (note, 0), NO_REGS, terminate_dead,
OP_IN);
}
+
+ /* Step 8: Kill the chains involving register restores. Those
+ should restore _that_ register. */
+ for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
+ if (REG_NOTE_KIND (note) == REG_CFA_RESTORE)
+ scan_rtx (insn, &XEXP (note, 0), NO_REGS, mark_all_read, OP_IN);
}
else if (DEBUG_INSN_P (insn)
&& !VAR_LOC_UNKNOWN_P (INSN_VAR_LOCATION_LOC (insn)))
Seems like a good thing regardless of the shrink-wrapping changes.
There's a comment about 200 lines earlier (egad) which outlines the
steps. Can you please add a comment there too.
It would probably be a good idea to refactor build_def_use a bit, but
I'd understand if you don't want to tackle that. I don't think that
desire should block this patch.
As I've said before, I'm not sure we're getting CFA notes right,
particularly on the older ports, but I don't think that should block
this patch.
With the earlier comment update change this is OK.
jeff
