[LLVMdev] dense maps

[Jakob Stoklund Olesen]

On Jan 25, 2012, at 9:44 PM, Preston Briggs wrote:

> Reading the LLVM Programmer's Manual, the description of DenseSet mentions:
> 
> Note that DenseSet has the same requirements for the value type that DenseMap has.
> 
> But when I read about DenseMap, I don't really see any requirements for the values, just a warning about space.

DenseSet<T> is implemented as a DenseMap<T,char>, so requirements for DenseMap keys apply. It seems the documentation could be clearer.

If your elements are pointers, SmallPtrSet<> is a better choice.

>  On the other hand, the keys have special requirements, which aren't entirely clear to me.  It says
> 
> Finally, you must implement a partial specialization of DenseMapInfo for the key that you want, if it isn't already supported. This is required to tell DenseMap about two special marker values (which can never be inserted into the map) that it needs internally.
> 
> Will my code fail to compile if the required specialization of DenseMapInfo is not present?

Yes. The non-specialized DenseMapInfo template doesn't provide any of the required methods.

> I wonder all these things because I've tripped across a problem where a method hangs repeatably, and I don't see the problem.
> 
> typedef SmallVector<BasicBlock *, 4> Frontier;
> DenseMap<BasicBlock *, Frontier> frontier(blocks*3/2);

Pointer keys work fine, you don't need to provide any specializations for that.

> 
> BasicBlock *Xblock = ...
> 
> errs() << "start\n";
> Frontier &f = frontier[Xblock];    // Xblock is not yet represented in frontier
> errs() << "finish\n";
> 
> It compiles fine (as part of a function-level pass), but the call frontier[Xblock] fails (hangs) after several successful calls.  I tried replacing the declaration of frontier with 
> 
> DenseMap<BasicBlock *, Frontier *> frontier(blocks*3/2);
> 
> but trip over the same problem in the same place.
> 
> I've used the DenseMap successfully in several other places, so I'm not sure that's gone wrong here.
> 
> Any ideas?

DenseMap iterators are invalidated by any container mutation. That is different behavior from std::map. Same thing for pointers and references to elements in the container. DenseMap invalidation works more like std::vector. It is not clear from your code if that is our problem.

I don't recall seeing a DenseMap::operator[] hanging. Most likely, the backing memory was corrupted somehow.

/jakob

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20120125/935e1f89/attachment.html>