[LLVMdev] Stupid '-load-vn -licm' question (LLVM 1.6)
Chris Lattner
sabre at nondot.org
Sun Mar 19 11:46:29 PST 2006
On Sat, 18 Mar 2006, Eric Kidd wrote:
> On Mar 17, 2006, at 12:19 PM, Chris Lattner wrote:
>> Great. Note that -cee is a beta pass... it has some known bugs and isn't
>> planned to be worked on in the immediate future. :(
>
> After a bit of digging, I found one open bug and two failing test cases, all
> of which appear to be feature requests:
>
> #217: LLVM needs a generic dominator update mechanism
> nullpointer.ll: "a load or store of a pointer indicates that the pointer is
> not null."
> looptest.ll: "Note that this is a "feature" test, not a correctness test."
Yup.
> All the other test cases pass, and none of the other unassigned bugs mention
> CEE. (I checked Bugzilla by hand--since searching for "correlated" doesn't
> even find bug #217--so I might have missed something.) Google doesn't seem to
> turn up anything, either.
>
> Does -cee actually produce incorrect code for some inputs? Or is it merely
> incomplete?
As I mentioned on IRC, I don't really remember. I thought it did
miscompile some cases, but it has been too long for me to remember.
Running it through llvm-test would be a good way to find out :)
> I'm looking for an optimizer pass to tweak (as a learning exercise), and if
> -cee isn't too hairy, I'd be happy to add support for 'switch' instructions
> and fix a few minor bugs. :-)
Sounds great!
-Chris
--
http://nondot.org/sabre/
http://llvm.org/
More information about the llvm-dev
mailing list