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

[Pete Cooper via llvm-commits]

> On Jan 11, 2016, at 2:17 PM, Keno Fischer <kfischer at college.harvard.edu> wrote:
> 
> 
> Running the verifier was my solution too.
> 
> I wrote my patch over a year ago, well before Duncan had improved the verifier to the point where it is now.  
> 
> I found that reducing the debug info metadata for all of llc in a single bc was enough to crash the backend because of mismatches between the asserts in DwarfDebug.cpp and the verifier.  I took a quick look and the verifier seems much more comprehensive now so I’d expect any fallout from this to be easy enough to improve the verifier.
> 
> I ran some fairly big bc files through this and it worked fine, but I would be happy to run some more test cases. How were you running it on the llc .bc, i.e. what crash were you looking for?
Unfortunately it was too long ago to remember what I was testing.  I may even have been testing a non-debug info failure and just got frustrated to see debug info in the test case.  Just can’t remember.

Thats great you’ve done some large binaries though.  I’d expect you’d have triggered some bad asserts by now if there was anything critically missing from the debug info verifier.
>  
> If you’re curious, my solution did involve knowing more about the format of the debug info its.  I reduced each of the fields of the compile unit (types, enums, imports, etc) all the way to empty arrays if possible.  Then once the CU entries were 0, I could also remove references to the debug info from instructions, and try to get to a small number of references from instructions or the remaining entries in the compile unit.  It seemed to work pretty reasonably.
> 
> I believe Adrian’s approach was to look for multiple compile units and remove whole compile units first.  That is a huge win on LTO where you may be able to remove all but the bad compile unit.
> 
> Do note that I already wrote some code a while back to drop the arguments of named metadata, which does a pretty good job of getting rid of whole compile units which are not needed to reproduce the crash. That pass runs before getting to to this.
I didn’t know that.  Thats excellent.  This should great help reduce what metadata is remaining.

Thanks,
Pete
>  
> My code is well out of date, so i wouldn’t suggest we revisit it, but its interesting to see the alternative approach.
> 
> As for the patch itself, i’ll take a look in phab.  Don’t want to confuse things with email comments on the code itself.
> 
> Thanks,
> Pete
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160111/24502cc3/attachment.html>