Nevermind. I misread the documentation.
On Sat, Apr 30, 2011 at 8:12 PM, Thomas Jablin <tjablin at gmail.com>
wrote:> Hi,
>
> Is it valid to query ScalarEvolution::getSCEVAtScope with a Value
> which is an Instruction inside the provided Loop? If so, I believe
> there is a bug in computeSCEVAtScope. Specifically:
>
> 04716 // If the scope is outside the addrec's loop, evaluate it by
using the
> 04717 // loop exit value of the addrec.
> 04718 if (!AddRec->getLoop()->contains(L)) {
> 04719 // To evaluate this recurrence, we need to know how many
> times the AddRec
> 04720 // loop iterates. Compute this now.
> 04721 const SCEV *BackedgeTakenCount >
getBackedgeTakenCount(AddRec->getLoop());
> 04722 if (BackedgeTakenCount == getCouldNotCompute()) return AddRec;
> 04723
> 04724 // Then, evaluate the AddRec.
> 04725 return AddRec->evaluateAtIteration(BackedgeTakenCount,
*this);
> 04726 }
>
> The guard checks that !AddRec->getLoop()->contains(L), but not
> !L->contains(AddRec->getLoop()).
>
> Consequently, if getSCEVAtScope is called with GetElementPtrInst and a
> Loop that contains it indirectly, the SCEV will be evaluated with
> respect to the end of the inner Loop. For example:
>
> L1 for(i to N)
> L2 for(j to N)
> arrayidx = gep A, 0, j, i
>
> getSCEVAtScope(arrayIdx, L1) will evaluate j as N.
>
> If it is valid to call getSCEVAtScope with an outer loop, I believe
> the guard should be:
> 04718 if (!AddRec->getLoop()->contains(L) &&
> !L->contains(AddRec->getLoop())) {
>
> If it is not valid, an assertion might be helpful.
>
> Tom
>
