[llvm-commits] [LLVM, loop-unswitch] Potential bug in RewriteLoopBodyWithConditionConstant

[Stepan Dyatkovskiy]

ping.

-Stepan.

Stepan Dyatkovskiy wrote:
> Hi Duncan. You are right. Please find the fixed patch.
> It is a pretty difficult to create a test case. By now I can only say
> that after replaceUsesOfWith was invoked we implicitly proceed to
> enumerate Replacement's UseList instead of LIC's. But we try to unswitch
> instructions that uses both LIC and Replacement. That why all looks fine.
> I also attached dumped instructions enumerated before and after fix (for
> first invocation of LoopUnswitch::RewriteLoopBodyWithConditionConstant
> with IsEqual = true).
> .ll file that was proceed is also attached.
>
> -Stepan.
>
> Duncan Sands wrote:
>> Hi Stepan,
>>
>>> It seems that the code inside the
>>> LoopUnswitch::RewriteLoopBodyWithConditionConstant contains potential
>>> bugs.
>>> UseList is changed inside the loops that are goes through its items:
>>>
>>> LoopUnswitch.cpp, string #905: for (Value::use_iterator UI =
>>> LIC->use_begin(), E
>>> = LIC->use_end(); UI != E; ++UI) {
>>>
>>> ...and then inside the loop body UseList is changed implicitly (
>>> LoopUnswitch.cpp, string #910):
>>> U->replaceUsesOfWith(LIC, Replacement);
>>>
>>> It seems that after editing UseList loop may produce unpreditable
>>> results. But
>>> we are lucky by now though :-)
>>>
>>> I propose to collect all to be changed and then do some changes?
>>> If so, please find the patch attached for review.
>>
>> can you just do something like this instead? Also, do you have a
>> testcase?
>>
>> Replacement = ConstantInt::get(Type::getInt1Ty(Val->getContext()),
>> !cast<ConstantInt>(Val)->getZExtValue());
>>
>> for (Value::use_iterator UI = LIC->use_begin(), E = LIC->use_end();
>> - UI != E; ++UI) {
>> - Instruction *U = dyn_cast<Instruction>(*UI);
>> + UI != E; ) {
>> + Instruction *U = dyn_cast<Instruction>(*UI++);
>> if (!U || !L->contains(U))
>> continue;
>> U->replaceUsesOfWith(LIC, Replacement);
>> Worklist.push_back(U);
>> }
>>
>> Ciao, Duncan.
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits