[LLVMdev] [Polly] Analysis of the expensive compile-time overhead of Polly Dependence pass
tobias at grosser.es
Mon Jul 29 09:27:27 PDT 2013
On 07/29/2013 09:15 AM, Sven Verdoolaege wrote:
> On Mon, Jul 29, 2013 at 07:37:14AM -0700, Tobias Grosser wrote:
>> On 07/29/2013 03:18 AM, Sven Verdoolaege wrote:
>>> On Sun, Jul 28, 2013 at 04:42:25PM -0700, Tobias Grosser wrote:
>>>> Sven: In terms of making the behaviour of isl easier to understand,
>>>> it may make sense to fail/assert in case operands have parameters that
>>>> are named identical, but that refer to different pointer values.
>>> No, you are allowed to have different identifiers with the same name.
>>> I could optionally print the pointer values, but then I'd have
>>> to think about what to do with them when reading a textual
>>> representation of a set with such pointer values in them.
>> Yes, this is how it is today.
> No, the pointer values are currently not printed.
I was referring to the first sentence. I do not think printing pointer
values is what we want. It would make the output unpredictable not only
when address space randomisation is involved.
>> I wondered if there is actually a need to
>> allow the use of different identifiers with the same name (except all being
>> unnamed?). I personally do not see such a need and would prefer isl to
>> assert/fail in case someone tries to do so. This may avoid confusions as
>> happened here. Do you see a reason why isl should allow this?
> Removing this feature would break existing users.
Even if it would, the benefits for future users may outweigh this.
Also, are you aware of a user that actually breaks?
Anyway, on the Polly side we know the behaviour and can handle it. So
this is nothing I am very strong about. I just mentioned it as it seemed
to be a good idea.
More information about the llvm-dev