[PATCH] D16083: [bugpoint] Teach bugpoint to reduce MDNodes

[Sergei Larin via llvm-commits]

Sorry - I told you I am starting cold :)


Maybe I am talking about a different issue here - I refer to this loop in CloneModule.cpp:

158 	  // And named metadata....
159 	  for (Module::const_named_metadata_iterator I = M->named_metadata_begin(),
160 	         E = M->named_metadata_end(); I != E; ++I) {
161 	    const NamedMDNode &NMD = *I;
162 	    NamedMDNode *NewNMD = New->getOrInsertNamedMetadata(NMD.getName());
163 	    for (unsigned i = 0, e = NMD.getNumOperands(); i != e; ++i)
164 	      NewNMD->addOperand(MapMetadata(NMD.getOperand(i), VMap));
165 	  }

It seems to me that it indiscriminately will copy all content of llvm.dbg.cu if one is present - at least this is exactly what it does in my case...

Sorry again if this is unrelated...

---
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation


> -----Original Message-----
> From: Keno Fischer [mailto:kfischer at college.harvard.edu]
> Sent: Friday, January 15, 2016 12:09 PM
> To: kfischer at college.harvard.edu; tejohnson at google.com;
> dexonsmith at apple.com; tom at stellard.net
> Cc: slarin at codeaurora.org; daniel.sanders at imgtec.com;
> aprantl at apple.com; Matthew.Arsenault at amd.com; llvm-
> commits at lists.llvm.org
> Subject: Re: [PATCH] D16083: [bugpoint] Teach bugpoint to reduce MDNodes
> 
> loladiro added a comment.
> 
> Dropping metadata is already implemented in the ValueMaterializer
> interface. However, I don't think the other things I'm doing are doable with
> that interface. I suppose we could extend it? @dexonsmith, thoughts on
> that?
> 
> 
> http://reviews.llvm.org/D16083
> 
>