------- Additional Comments From law at redhat dot com 2005-02-14 20:01 -------
The new jump thread selection code handles this properly.
I will note that your comments about the code we ought to be able to generate
are wrong. The correctly optimized code should look like this: (note the
difference in where we test c == 3.
foo (a, b, c)
{
# BLOCK 0
# PRED: ENTRY [100.0%] (fallthru,exec)
if (a_1 == 1) goto <L3>; else goto <L0>;
# SUCC: 3 [14.9%] (true,exec) 1 [85.1%] (false,exec)
# BLOCK 1
# PRED: 0 [85.1%] (false,exec)
<L0>:;
if (b_3 == 2) goto <L2>; else goto <L5>;
# SUCC: 2 [45.3%] (true,exec) 5 [54.7%] (false,exec)
# BLOCK 2
# PRED: 1 [45.3%] (true,exec)
<L2>:;
if (c_2 == 3) goto <L3>; else goto <L4>;
# SUCC: 3 [48.8%] (true,exec) 4 [51.2%] (false,exec)
# BLOCK 3
# PRED: 0 [14.9%] (true,exec) 2 [48.8%] (true,exec)
<L3>:;
f1 ();
goto <bb 5> (<L5>);
# SUCC: 5 [100.0%] (fallthru,exec)
# BLOCK 4
# PRED: 2 [51.2%] (false,exec)
<L4>:;
f2 ();
# SUCC: 5 [100.0%] (fallthru,exec)
# BLOCK 5
# PRED: 1 [54.7%] (false,exec) 3 [100.0%] (fallthru,exec) 4 [100.0%]
(fallthru,exec)
<L5>:;
return;
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15352
