[llvm-commits] [llvm] r166157 - /llvm/trunk/lib/VMCore/Verifier.cpp

[Bill Wendling]

On Oct 18, 2012, at 3:39 PM, Eli Friedman <eli.friedman at gmail.com> wrote:

> On Thu, Oct 18, 2012 at 3:27 PM, Bill Wendling <isanbard at gmail.com> wrote:
>> On Oct 18, 2012, at 3:21 PM, Chris Lattner <clattner at apple.com> wrote:
>> 
>>> On Oct 18, 2012, at 12:39 PM, Bill Wendling <wendling at apple.com> wrote:
>>> 
>>>> On Oct 18, 2012, at 11:41 AM, Chris Lattner <clattner at apple.com> wrote:
>>>> 
>>>>> On Oct 18, 2012, at 8:04 AM, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>>>>>>> Oh. I didn't know that LLVM defines things in a non-standard way.
>>>>>> 
>>>>>> It is very standard actually. A dominates B if all execution paths
>>>>>> that reach B pass through A. In this case, all the 0 paths that reach
>>>>>> B pass through A.
>>>>> 
>>>>> This is also the only reasonable way to define dominance and SSA.  If not, things like trivial copy propagation are potentially really complicated and expensive.
>>>>> 
>>>> Okay. Good to know. Because one pass is creating this for reachable code, which is making GVN blow up.
>>> 
>>> The verifier will reject that, GVN shouldn't have to handle it.
>>> 
>> The verifier didn't reject it as far as I could tell.
> 
> The verifier is supposed to check dominance.
> 
> Random guess, but have you tried -verify-dom-info?
> 
Not yet. I'm able to reproduce it only during an actual build so far. I'm still getting a .ll file.

-bw