[polly] Add diagnostic remark for ReportVariantBasePtr

[Tobias Grosser]

On 26/06/2014 16:14, Tobias Grosser wrote:
> On 26/06/2014 16:08, Andreas Simbuerger wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>>
>>
>> On 06/26/2014 04:00 PM, Tobias Grosser wrote:
>>> On 26/06/2014 15:38, Andreas Simbuerger wrote:
>>>> -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
>>>>
>>>> At the moment the DebugLoc is derived from the BaseValue. I'm not
>>>> 100% sure if the resulting DebugLoc corresponds to the A or the
>>>> (A->b) BaseValue.
>>>
>>> Possibly A->b.
>>
>> Yep, I meant 100%-sure from the output of the DebugLoc ;-).
>>
>>>
>>>> As far as I can tell, both would map to the same source location
>>>> in the debug-metadata.
>>>
>>> The caret diagnostic points for me at the beginning of the line.
>>> This is good. If we ever figure out how to get the precise column
>>> location, that would be amazing.
>>
>> That's what I would love to have. But well, so far it should be
>> helpful enough. ;-)
>
> Its extremely useful already. No complains from my side. Just putting
> this on the 'nice-to-have-if-someone-has-an-idea-and-time' list.

Diego's thread just pointed me to -gcolumn-info. After enabling it we 
already get the caret at the right location. So we only need to wait for 
it being enabled.

I also tested the following SCoP with -polly-detect-keep-going:

  void a(struct b *A) {
    for (int i=0; i<32; i++)
      A->b[i] = A->b[i+1];
  }

Only the first location comes with a source location. I wonder if we 
could fix this by looking at the location of the load instruction, as it 
is less likely to be removed by gvn?

Cheers,
Tobias