llvm dev - Nov 2011 - [LLVMdev] How to make Polly ignore some non-affine memory accesses

Sorry for the noobish question, but what kind of subscripts generate a
SCEVCouldNotCompute  from the SCEV engine?
I tried for a while but I wasn't able to trigger that.

2011/11/20 Tobias Grosser <tobias at grosser.es>:
> On 11/20/2011 03:01 AM, Marcello Maggioni wrote:
>>
>> 2011/11/19 Tobias Grosser<tobias at grosser.es>:
>>>
>>> On 11/18/2011 01:34 PM, Marcello Maggioni wrote:
>>>>
>>>> Ok , this is what I believe is the final patch that adds the
>>>> non-affine accept functionality to Polly, this should have no
issues.
>>>>
>>>> I added three tests, two in ScopInfo (two simple tests, one
expected
>>>> fail and one success based on the same source) and one in
CodeGen that
>>>> verifies that the code is generated.
>>>>
>>>> The patch is attached.
>>>
>>> The patch includes only one test case. I also have some tiny last
>>> comments
>>> inline. Otherwise the patch looks good.
>>>
>>>> +++ test/ScopInfo/simple_nonaffine_loop_exfail.ll     (revision
0)
>>>> @@ -0,0 +1,42 @@
>>>> +; RUN: opt %loadPolly %defaultOpts -polly-scops -analyze %s |
>>>> FileCheck %s
>>>> +; XFAIL: *
>>>
>>> Why is this one still XFAIL?
>>
>> I wanted to add a test case that in case of a "missing the
flag" at
>> command line confirmed the functionality was actually disabled (and
>> this is the reason of the XFAIL test).he
>
> XFAIL it not the way to test this. You can just add another command line
> to the passing test, remove the flag to enable non-affine access functions
> and replace '| FileCheck' with '| not FileCheck'.
>
>> Originally you said I had to add a minimst al test case to the patch,
>> should I add some more tests now withore complex scops?
>
> I do not see that a more complex scope would test more of the feature. You
> could add a test case where the access subscript is not a valid
> SCEV, but COULD_NOT_COMPUTE.
>
> Cheers
> Tobi
>

On 11/20/2011 04:36 PM, Marcello Maggioni wrote:
> Sorry for the noobish question, but what kind of subscripts generate a
> SCEVCouldNotCompute  from the SCEV engine?
> I tried for a while but I wasn't able to trigger that
Hi Marcello,

the SCEV returns SCEVCouldNotCompute in case it cannot analyze an 
expression or if the analysis would be to complicated. I am currently
not sure if this may actually happen when calling getSCEV(), because
getSCEV() could just return a SCEVUnknown referencing the Value itself.
Maybe SCEVCouldNotCompute is just generated by functions like
SE->getBackedgeTakenCount()?

In case you cannot generate a test case that yields to this, I don't 
think there is a need to try more. Even without such a test case the 
patch should be OK.

Cheers
Tobi

2011/11/21 Tobias Grosser <tobias at grosser.es>:
> On 11/20/2011 04:36 PM, Marcello Maggioni wrote:
>>
>> Sorry for the noobish question, but what kind of subscripts generate a
>> SCEVCouldNotCompute  from the SCEV engine?
>> I tried for a while but I wasn't able to trigger that
>
> Hi Marcello,
>
> the SCEV returns SCEVCouldNotCompute in case it cannot analyze an
expression
> or if the analysis would be to complicated. I am currently
> not sure if this may actually happen when calling getSCEV(), because
> getSCEV() could just return a SCEVUnknown referencing the Value itself.
> Maybe SCEVCouldNotCompute is just generated by functions like
> SE->getBackedgeTakenCount()?
>
> In case you cannot generate a test case that yields to this, I don't
think
> there is a need to try more. Even without such a test case the patch should
> be OK.
>
> Cheers
> Tobi
>
Ok, thanks, I'll try other test cases to check if everything is ok,
better being sure.

About SCOPs with a "bitcast instruction" inside them I see that
ScopDetection discards those.
What problems can cause bitcasts to the Scop? Is it related to aliasing?

Marcello