[polly] r231594 - Update test cases to work independently of delinearization default

[Tobias Grosser]

On 08.03.2015 17:08, Johannes Doerfert wrote:
> On 03/08, Tobias Grosser wrote:
>> On 08.03.2015 16:33, Johannes Doerfert wrote:
>>> On 03/08, Tobias Grosser wrote:
>>>> Author: grosser
>>>> Date: Sun Mar  8 10:21:15 2015
>>>> New Revision: 231594
>>>>
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=231594&view=rev
>>>> Log:
>>>> Update test cases to work independently of delinearization default
>>>>
>>>> Modified: polly/trunk/test/ScopInfo/NonAffine/non_affine_access_with_range.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/polly/trunk/test/ScopInfo/NonAffine/non_affine_access_with_range.ll?rev=231594&r1=231593&r2=231594&view=diff
>>>> ==============================================================================
>>>> --- polly/trunk/test/ScopInfo/NonAffine/non_affine_access_with_range.ll (original)
>>>> +++ polly/trunk/test/ScopInfo/NonAffine/non_affine_access_with_range.ll Sun Mar  8 10:21:15 2015
>>>> @@ -2,13 +2,16 @@
>>>>   ;
>>>>   ;    void f(int *A, char c) {
>>>>   ;      for (int i = 0; i < 1024; i++)
>>>> -;        A[i * c]++;
>>>> +;        A[i * c - 3]++;
>>>>   ;    }
>>>>   ;
>>>> -; CHECK: ReadAccess := [Reduction Type: +] [Scalar: 0]
>>>> -; CHECK:     { Stmt_bb2[i0] -> MemRef_A[o0] : o0 <= 261115 and o0 >= -3 };
>>>> -; CHECK: MayWriteAccess := [Reduction Type: +] [Scalar: 0]
>>>> -; CHECK:     { Stmt_bb2[i0] -> MemRef_A[o0] : o0 <= 261115 and o0 >= -3 };
>>>> +; CHECK: Assumed Context:
>>>> +; CHECK:  [p_0] -> {  : 1 = 0 }
>>>> +; CHECK: ReadAccess :=       [Reduction Type: +] [Scalar: 0]
>>>> +; CHECK: [p_0] -> { Stmt_bb2[i0] -> MemRef_A[i0, -3] };
>>>> +; CHECK: MustWriteAccess :=  [Reduction Type: +] [Scalar: 0]
>>>> +; CHECK: [p_0] -> { Stmt_bb2[i0] -> MemRef_A[i0, -3] };
>>> This was actually supposed to check the "range feature" for non-affine
>>> accesses. However, now it is a affine delinearized accesses.
>>
>> Wow, I was too fast here then.
>>
>> Do you have any idea how we can add test coverage back or should we better
>> drop this test case then as it is now "successfully" delinearized.
> Drop it or use another operation then multiply that is still understood
> by the range analysis of ScalarEvolution.

I tried something like 'i * c + i * p - 3', but it seemed as if the 
range analysis did not trigger.

I dropped the test case in 231599.

Tobias