[llvm-commits] SelectionDAG Loosing debug information

[Devang Patel]

On May 9, 2011, at 4:38 PM, Villmow, Micah wrote:

> 
>> -----Original Message-----
>> From: Devang Patel [mailto:dpatel at apple.com]
>> Sent: Monday, May 09, 2011 4:28 PM
>> To: Villmow, Micah
>> Cc: llvm-commits
>> Subject: Re: [llvm-commits] SelectionDAG Loosing debug information
>> 
>> DanglingDebugInfoMap maps llvm::Value to a node. So, the mapping is
>> invalid whenever NodeMap is cleared. Are you sure this patch is the
>> right approach here ? 
> [Villmow, Micah] Maybe, I'm not 100% familiar with the code, but it 
> seems to fix the problem we are having. The problem we are having is this:
> The bitcode attached to the bug has a case where after running inlineAll
> and then mem2reg, the resulting debug information is attached to a 
> phi node. The debug information references an instruction in a previous
> basic block but has no references in the current basic block. By my
> understanding after looking at the code, no nodes are created for the
> phi node and the debug information is added to the map. However, since
> no nodes in the current block reference the debug information, there
> can be no resolving of the dangling reference by the end of the
> block, which the map then gets cleared.
> 
> What this change does is allows for all of the dangling references to
> be resolved by the current or previous block at some point in the future
> and after all the basic blocks have been processed and there is no
> possibility of resolving the dangling reference anymore, then it is cleared.
> 
> I don't see this as being a problem when NodeMap is cleared simply because
> if the node existed in the current basic block, then the dangling reference
> wouldn't exist in the DanglingDebugInfoMap as it is replaced once it is resolved.

OK this makes. Go ahead and apply the patch, please include a test case also.
Thanks,
-
Devang

> Micah
>> -
>> Devang
>> 
>> On May 9, 2011, at 4:04 PM, Villmow, Micah wrote:
>> 
>>> Patch attached.
>>> 
>>>> -----Original Message-----
>>>> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
>>>> bounces at cs.uiuc.edu] On Behalf Of Eric Christopher
>>>> Sent: Monday, May 09, 2011 12:50 PM
>>>> To: Villmow, Micah
>>>> Cc: llvm-commits
>>>> Subject: Re: [llvm-commits] SelectionDAG Loosing debug information
>>>> 
>>>> 
>>>> On May 9, 2011, at 11:57 AM, Villmow, Micah wrote:
>>>> 
>>>>> I've attached a patch to bug9879 that stops SelectionDAG from
>>>> dropping debug information. Can someone review to see if this
>> approach
>>>> is correct or if a better one is preferred?
>>>>> 
>>>> No patch attached.
>>>> 
>>>> -eric
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>> 
>>> <bug9869.patch>_______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20110510/d40914ff/attachment.html>