llvm dev - May 2014 - [LLVMdev] An unexpected behavior in RegionInfo's block_iterator

Hi Fellows,

   I notice an unexpected behavior in RegionInfo's block_iterator. Consider
the following situation: a user creates her own region containing just a
single basic block TheBB, then tries to have the block_iterator to print a
DFS traversal of the region. The expected behavior should be that only the
single basic block TheBB will be printed, but the real behavior is that
block_iterator prints out all the basic blocks of the CFG starting from
TheBB to the end.

    It looks like the issue originates from setting the end iterator to
(BasicBlock *) 0. I understand that the "region" detected by
RegionInfo
should never contain a single basic block or even a sequence of basic
blocks. So maybe the above degenerated case is considered "will never
happen"?

    ... ...

    BasicBlock *BB = Func.getARandomBasicBlock;

    Region *R = new Region(BB, BB, RI, DT);
    for (Region::block_iterator i = R->block_begin(),
           e = R->block_end(); i != e; ++i) {
         errs() << (*i)->getName() << "\n";
      }

      ... ...



Best Regards,
Paul
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140502/aaf19ef1/attachment.html>

On 02/05/2014 23:26, Paul Vario wrote:
> Hi Fellows,
>
>     I notice an unexpected behavior in RegionInfo's block_iterator.
Consider
> the following situation: a user creates her own region containing just a
> single basic block TheBB, then tries to have the block_iterator to print a
> DFS traversal of the region. The expected behavior should be that only the
> single basic block TheBB will be printed, but the real behavior is that
> block_iterator prints out all the basic blocks of the CFG starting from
> TheBB to the end.
>
>      It looks like the issue originates from setting the end iterator to
> (BasicBlock *) 0. I understand that the "region" detected by
RegionInfo
> should never contain a single basic block or even a sequence of basic
> blocks. So maybe the above degenerated case is considered "will never
> happen"?
>
>      ... ...
>
>      BasicBlock *BB = Func.getARandomBasicBlock;
>
>      Region *R = new Region(BB, BB, RI, DT);
>      for (Region::block_iterator i = R->block_begin(),
>             e = R->block_end(); i != e; ++i) {
>           errs() << (*i)->getName() << "\n";
>        }

Hi Paul,

the exit basic block is the block after the region.

  BB0 < Entering Block
   |
   v
  BB1 < Entry and Exiting block
   |
   v
  BB2 < Exit block

So your code should probably be:

new Region(BB, *pred_begin(BB), RI, DT)

Let me know if it works for you. In case it does, we could add an assert 
to the constructor to catch these cases.

Cheers,
Tobias

Hi Tobias,

     Thanks so much for the quick response. Your approach fixes the issue.
     On a bigger context, would it make more sense to make the region exit
part of the region? For example, a while loop gets lowered down to LLVM IR
contains while.cond, while.body and while.end. If one tries to use
RegionInfo as a substitute for structural analysis, she might think
while.end should belong to the region/structure ... is it necessary to
exclude the exit from being part of the region or both ways are equally
correct in terms of uniquely representing regions?

Best,
Paul


[image: Inline image 1]


On Fri, May 2, 2014 at 4:42 PM, Tobias Grosser <tobias at grosser.es>
wrote:
> On 02/05/2014 23:26, Paul Vario wrote:
>
>> Hi Fellows,
>>
>>     I notice an unexpected behavior in RegionInfo's block_iterator.
>> Consider
>> the following situation: a user creates her own region containing just
a
>> single basic block TheBB, then tries to have the block_iterator to
print a
>> DFS traversal of the region. The expected behavior should be that only
the
>> single basic block TheBB will be printed, but the real behavior is that
>> block_iterator prints out all the basic blocks of the CFG starting from
>> TheBB to the end.
>>
>>      It looks like the issue originates from setting the end iterator
to
>> (BasicBlock *) 0. I understand that the "region" detected by
RegionInfo
>> should never contain a single basic block or even a sequence of basic
>> blocks. So maybe the above degenerated case is considered "will
never
>> happen"?
>>
>>      ... ...
>>
>>      BasicBlock *BB = Func.getARandomBasicBlock;
>>
>>      Region *R = new Region(BB, BB, RI, DT);
>>      for (Region::block_iterator i = R->block_begin(),
>>             e = R->block_end(); i != e; ++i) {
>>           errs() << (*i)->getName() << "\n";
>>        }
>>
>
>
> Hi Paul,
>
> the exit basic block is the block after the region.
>
>  BB0 < Entering Block
>   |
>   v
>  BB1 < Entry and Exiting block
>   |
>   v
>  BB2 < Exit block
>
> So your code should probably be:
>
> new Region(BB, *pred_begin(BB), RI, DT)
>
> Let me know if it works for you. In case it does, we could add an assert
> to the constructor to catch these cases.
>
> Cheers,
> Tobias
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140502/29ef8f21/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 204527 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140502/29ef8f21/attachment.png>