[llvm-commits] [llvm] r115323 - in /llvm/trunk: lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp test/DebugInfo/2010-10-01-crash.ll

[Devang Patel]

On Oct 1, 2010, at 2:58 PM, Chris Lattner wrote:

> 
> On Oct 1, 2010, at 2:52 PM, Devang Patel wrote:
> 
>> 
>> On Oct 1, 2010, at 2:30 PM, Chris Lattner wrote:
>> 
>>> 
>>> On Oct 1, 2010, at 12:00 PM, Devang Patel wrote:
>>> 
>>>> Author: dpatel
>>>> Date: Fri Oct  1 14:00:44 2010
>>>> New Revision: 115323
>>>> 
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=115323&view=rev
>>>> Log:
>>>> Fix code gen crash reported in PR 8235. We still lose debug info for the unused argument here. This is a known limitation recorded debuginfo-tests/trunk/dbg-declare2.ll function 'f6' test case.
>>> 
>>> Thanks for fixing the crash Devang.  How hard is it to get the FI# here?  It would be "really great" to be able to get debug info for these.  Losing it at -O0 seems bad.
>> 
>> We need to find a way to use SDNode created during LowerArugments() in such cases. This happens only when A) argument is not used at -O0 and B) we do not use FastISel. (In normal course, at -O0 we expect FastISel to handle entry block dbg.declares.)  We had a scheme to save nodes for unused argument. IIRC, recent changes in the way we do bottom up fastisel, we broke that delicate link. Now, FuncInfo is cleared before dbg.declare from entry block is processed.  We need to find another way. 
> 
> Why isn't there a frameindex assigned to it?  Even if it is dead, there is no harm in creating the FI#

That's what EmitFuncArgumentDbgValue() was expecting. That is a real bug.
-
Devang

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20101001/5498c8b7/attachment.html>