[PATCH][SelectionDAG] Increase coverage for checkForCycles

Adam Nemet anemet at apple.com
Sat May 31 09:34:11 PDT 2014


Hi Andy,

Sounds good, thanks very much.  Checked in as r209975, r209976 and r209977.

Adam

On May 30, 2014, at 6:11 PM, Andrew Trick <atrick at apple.com> wrote:

> Hi Adam,
> 
> This looks great to me. For an internal diagnostic feature, I think it’s safe to commit and it’s easy for other developers to follow up post-commit if they want to.
> 
> -Andy
> 
> On May 30, 2014, at 12:06 PM, Adam Nemet <anemet at apple.com> wrote:
> 
>> Ping?
>> 
>> On May 21, 2014, at 11:02 AM, Adam Nemet <anemet at apple.com> wrote:
>> 
>>> This is a small patchset to provide better by-default diagnostics for DAG cycles.  Currently the check is only performed with ENABLE_EXPENSIVE_CHECKS=1.  After these changes we will also run it in Debug+Asserts.  To avoid slowing things down, we only perform the check when we would abort anyway due to some other inconsistency in the DAG detected in AssignTopologicalOrder.
>>> 
>>> So now, in addition to the cryptic "Overran sorted position”, we also get the node printed where the cycle starts.  Hopefully this will provide more actionable bug report.  This is motivated by my recent work on r209242.
>>> 
>>> OK to commit?
>>> 
>>> Adam
>>> 
>>> 
>>> 
>> <0001-SelectionDAG-Remove-duplicate-declaration.patch><0002-SelectionDAG-Pass-DAG-to-checkForCycles.patch><0003-SelectionDAG-Force-cycle-detection-in-AssignTopologi.patch>
>>> 
>>> _______________________________________________
>>> 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
> 





More information about the llvm-commits mailing list