[PATCH] Add warning capabilities in LLVM (backend part), Take 2

Quentin Colombet qcolombet at apple.com
Wed Dec 11 10:42:55 PST 2013

On Dec 11, 2013, at 10:33 AM, Renato Golin <renato.golin at linaro.org> wrote:

> On 11 December 2013 18:20, Quentin Colombet <qcolombet at apple.com> wrote:
>> However, beyond the plugin case, I thought it would have been a good class to experiment things (like vectorizer information on why something has not been optimized using a DS_Note) or to report a richer message than InlineAsm without creating a specific subclass for each pattern.
> I want to have a special kind for the vectorizer because this will
> allow us to communicate warnings to the user that a #pragma vectrorise
> enable could not be fulfilled because the target has no vectors, or
> there is a memory conflict. This is important to go beyond the debug
> level into warning level.
Agree, this was just an example :).
That said, I think the vectorizer guys are better suited to define the subclass they need, thus it should be fine as a follow-up step.

>> For now, I think Hal’s suggestion is a good start. That said, I was not looking in addressing the problem adding new diagnostics at runtime. Thus, we may need another thread to discuss that.
>> I do not think that was block the current proposal.
> I may not have understood Hal's proposal too well, then. Ignore my
> comments... ;)
Sure :).

Thanks for your comments.


> cheers,
> --renato

More information about the llvm-commits mailing list