This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/23821] [4.0/4.1/4.2 Regression] DOM and VRP creating harder to optimize code
- From: "sebastian dot pop at cri dot ensmp dot fr" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 13 Feb 2006 08:45:15 -0000
- Subject: [Bug tree-optimization/23821] [4.0/4.1/4.2 Regression] DOM and VRP creating harder to optimize code
- References: <bug-23821-5077@http.gcc.gnu.org/bugzilla/>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Comment #8 from sebastian dot pop at cri dot ensmp dot fr 2006-02-13 08:45 -------
Subject: Re: [4.0/4.1/4.2 Regression] DOM and VRP creating harder to optimize
code
This case reminds me the peeled chrec unification that I had to
disable on autovect branch (I probably have to run the transformation
as a stand alone pass outside the analyzer for not disturbing the user
passes). In that case we're looking at a code like
loop
x = phi (0, a)
a = phi (1, a + 1)
endloop
such that a simple transformation can make "x" a simple iv. This case
is also quite important, as it occurs about 300 times during a bootstrap.
Now for the current problem, we could run a pass just after loop_init
for cleaning all these constructs. As suggested, we would have to
build an equivalence relation either like VRP, or on demand.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23821