[llvm-commits] [llvm] r94480 - in /llvm/trunk: lib/CodeGen/SelectionDAG/FastISel.cpp lib/Target/X86/X86FastISel.cpp test/DebugInfo/2010-01-18-DbgValue.ll

[Chris Lattner]

On Jan 29, 2010, at 11:40 AM, Dale Johannesen wrote:

>
> On Jan 26, 2010, at 5:58 PMPST, Chris Lattner wrote:
>> On Jan 25, 2010, at 4:09 PM, Dale Johannesen wrote:
>>
>>> Author: johannes
>>> Date: Mon Jan 25 18:09:58 2010
>>> New Revision: 94480
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=94480&view=rev
>>> Log:
>>> Generate DEBUG_VALUE comments on x86.  The (limited)
>>> dbg.declare's we currently generate go through both
>>> register allocators without perturbing the results.
>>
>> Hi Dale,
>>
>>> +++ llvm/trunk/lib/Target/X86/X86FastISel.cpp Mon Jan 25 18:09:58  
>>> 2010
>>> @@ -1153,6 +1153,16 @@
>>> // FIXME: Handle more intrinsics.
>>> switch (I.getIntrinsicID()) {
>>> default: return false;
>>> +  case Intrinsic::dbg_declare: {
>>> +    DbgDeclareInst *DI = cast<DbgDeclareInst>(&I);
>>> +    X86AddressMode AM;
>>> +    if (!X86SelectAddress(DI->getAddress(), AM))
>>
>> DI->getAddress() can return null here, I think the dbg.declare  
>> should silently be dropped in that case.
>
> No it can't, we detected the null case earlier (in FastISel) and  
> never reached here.

Ah, ok.  Please add a comment or assert to that effect as  
documentation, thanks!

-Chris