When looking deeper into the problem I discovered that very sometimes
try_combine is called with the same insn for I1 and I2. This is quite
bad since try_combine does not expect that at all. I expect this is
caused by my patches adding a LOG_LINK per register, which can mean
a pair of insns has more than one LOG_LINK between them.
This patch fixes it by making try_combine refuse identical insns early.
Bootstrapped on powerpc64-linux and powerpc-linux; the fails are gone.
Also bootstrapped on x86_64-linux.
Is this okay for mainline? (regtest on powerpc64-linux in process).
Segher
2014-12-15 Segher Boessenkool <seg...@kernel.crashing.org>
gcc/
PR target/64268
* combine.c (try_combine): Immediately return if any of I0,I1,I2
are the same insn.
---
gcc/combine.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/gcc/combine.c b/gcc/combine.c
index de2e49f..8e5d1f7 100644
--- a/gcc/combine.c
+++ b/gcc/combine.c
@@ -2620,6 +2620,11 @@ try_combine (rtx_insn *i3, rtx_insn *i2, rtx_insn *i1,
rtx_insn *i0,
rtx new_other_notes;
int i;
+ /* Immediately return if any of I0,I1,I2 are the same insn (I3 can
+ never be). */
+ if (i1 == i2 || i0 == i2 || (i0 && i0 == i1))
+ return 0;
+
/* Only try four-insn combinations when there's high likelihood of
success. Look for simple insns, such as loads of constants or
binary operations involving a constant. */
--
1.8.1.4
