[polly] r271128 - Determination of statements that contain matrix multiplication

[Mehdi Amini via llvm-commits]

> On Jun 3, 2016, at 9:24 AM, Roman Gareev <gareevroman at gmail.com> wrote:
> 
> 2016-06-02 22:06 GMT+05:00 Michael Kruse <llvm-commits at meinersbur.de>:
>> 2016-06-02 17:20 GMT+02:00 Mehdi Amini <mehdi.amini at apple.com>:
>>>> I find a different possible underflow with the expression "DimNum -
>>>> 1", as DimNum is unsigned. Might happen if the map has 0 range
>>>> dimensions which probably never happens.
>>> 
>>> Yes it seems that it is what Coverity is reporting, feel free to ignore if you think it's fine.
>> 
>> An assert() or "if (DimNum == 0) return IslMap" could still be useful.
>> Because I don't know the context in which the function is used, I
>> leave it to Roman if he wants to change it.
> 
> Yes, I think that it’s a potential issue and we should fix it.
> 
> Thank you for the solution! I've implemented it in
> http://reviews.llvm.org/D20969 . Unfortunately, I can't check it using
> Coverity.

Coverity runs on LLVM every week or so: https://scan.coverity.com/projects/llvm
We'll know post-commit...

I got the report because it was a new failure detected, but if you (Polly devs) are interested, there are plenty of issues reported with Polly that you can view online (accounts are free).

-- 
Mehdi