Hello,
I've encountered a bug in the cse (over-)optimization. A patch, testing input
for test-linearize and its outputs before/after patch are attached.
Note there is still something odd which the attached test-case revealed. The
output of test-unssa still contains two (4 without my previous patch applied)
unprocessed phisrc nodes. As time permits I'll try to write another for it if
nobody will be faster ;-)
Kamil
test:
.L0x7fbd14330010
<entry-point>
phisrc.32 %phi1(i) <- %arg1
phisrc.32 %phi4(i) <- %arg1
br .L0x7fbd14330130
.L0x7fbd14330130
copy.32 %r1(i) <- %r5(i)
br %r1(i), .L0x7fbd14330178, .L0x7fbd143300e8
.L0x7fbd14330178
call test, $0
br .L0x7fbd143301c0
.L0x7fbd143301c0
add.32 %r4 <- %r1(i), $1
phisrc.32 %phi2(i) <- %r4
phisrc.32 %phi5(i) <- %r4
br .L0x7fbd14330130
.L0x7fbd143300e8
ret
static void test(int i)
{
while (i) {
if (i)
test(0);
i++;
}
}
test:
.L0x7f43c1f02010
<entry-point>
phisrc.32 %phi1(i) <- %arg1
phisrc.32 %phi4(i) <- %arg1
br .L0x7f43c1f02130
.L0x7f43c1f02130
br %r3, .L0x7f43c1f02178, .L0x7f43c1f020e8
.L0x7f43c1f02178
call test, $0
br .L0x7f43c1f021c0
.L0x7f43c1f021c0
copy.32 %r3 <- %r5
add.32 %r4 <- %r3, $1
phisrc.32 %phi2(i) <- %r4
phisrc.32 %phi5(i) <- %r4
br .L0x7f43c1f02130
.L0x7f43c1f020e8
ret
>From 7658e89de5b2568d1a8663277c306b6f009b7ca7 Mon Sep 17 00:00:00 2001
From: Kamil Dudka <kdudka@xxxxxxxxxx>
Date: Sun, 9 Aug 2009 22:43:17 +0200
Subject: [PATCH] cse: treat PHI-nodes as other instructions
Without this patch test-linearize fails on a simple example:
static void test(int i)
{
while (i) {
if (i)
test(0);
i++;
}
}
It generates a conditional jump depending on an uninitialized value
which is obviously not in the input code.
Signed-off-by: Kamil Dudka <kdudka@xxxxxxxxxx>
---
cse.c | 7 -------
1 files changed, 0 insertions(+), 7 deletions(-)
diff --git a/cse.c b/cse.c
index 2a15745..2aabb65 100644
--- a/cse.c
+++ b/cse.c
@@ -317,13 +317,6 @@ static struct instruction * try_to_cse(struct entrypoint *ep, struct instruction
b2 = i2->bb;
/*
- * PHI-nodes do not care where they are - the only thing that matters
- * are the PHI _sources_.
- */
- if (i1->opcode == OP_PHI)
- return cse_one_instruction(i1, i2);
-
- /*
* Currently we only handle the uninteresting degenerate case where
* the CSE is inside one basic-block.
*/
--
1.6.4
