On Saturday 27 August 2011 17:53:23 Linus Torvalds wrote:
> phi2/phi3 are both dead, but theor 'phisrc' instructions haven't been
> killed. Ugly.
The attached patch solves the dangling PHI nodes and does not seem to break
anything at first glance. It is probably not the most efficient solution,
but it might at least show where to look for the problem.
Kamil
From 7024fcdfba9fce20569f86e2a279c76cb94cbfd8 Mon Sep 17 00:00:00 2001
From: Kamil Dudka <kdudka@xxxxxxxxxx>
Date: Sun, 28 Aug 2011 01:29:13 +0200
Subject: [PATCH] cse: update PHI users when throwing away an instruction
Signed-off-by: Kamil Dudka <kdudka@xxxxxxxxxx>
---
cse.c | 13 +++++++++++++
1 files changed, 13 insertions(+), 0 deletions(-)
diff --git a/cse.c b/cse.c
index 2a15745..1b63c0e 100644
--- a/cse.c
+++ b/cse.c
@@ -250,6 +250,19 @@ static void sort_instruction_list(struct instruction_list **list)
static struct instruction * cse_one_instruction(struct instruction *insn, struct instruction *def)
{
convert_instruction_target(insn, def->target);
+
+ if (insn->opcode == OP_PHI) {
+ /* Remove the instruction from PHI users */
+ pseudo_t phi;
+ FOR_EACH_PTR(insn->phi_list, phi) {
+ struct pseudo_user *pu;
+ FOR_EACH_PTR(phi->users, pu) {
+ if (pu->insn == insn)
+ DELETE_CURRENT_PTR(pu);
+ } END_FOR_EACH_PTR(pu);
+ } END_FOR_EACH_PTR(phi);
+ }
+
insn->opcode = OP_NOP;
insn->bb = NULL;
repeat_phase |= REPEAT_CSE;
--
1.7.6
