[LLVMdev] Selection DAG output as bare DAG, code review

Sean Silva silvas at purdue.edu
Mon Sep 3 21:01:41 PDT 2012

It's not going to be that helpful to just look at the edges, since the
edges "come out of" and "go into" different parts of the nodes, and
that is really important information. I've attached a simple example
of what it looks like. I made that with Mathematica, with:

g = Import["/tmp/llvm_Cnrx9F/dag.setit.dot", {"DOT", "EdgeRules"}];
Graph[Union[g], ImageSize -> {400, 400}]

Note that this won't scale. That BB was trivial (I believe it was one
assignment). Once you get into anything more complicated, it's just
going to become a ball of spaghetti without the node info.

If you realy want to, you take the dot files produced by
`-view-dag-combine1-dags` and friends
and then do some postprocessing to get it into the form that you want.

--Sean Silva

On Mon, Sep 3, 2012 at 6:40 AM, Joseph Reddington <joseph at cs.rhul.ac.uk> wrote:
> Hello all,
> I recently foudn myself wanting to view the basic blocks in the
> selection DAG as pure DAGs - so just as a list of edges, with no other
> information.  I added the below code to the start of the "
> void SelectionDAGISel::CodeGenAndEmitDAG()" function.  It creates a
> separate txt file for each basic block and gives a list of edges
> between nodes.  The segment of code is below - I'm looking for any
> feedback on how I might have done things differently (I'm a very basic
> c++ programmer) or if is a much easier way of getting the same
> information - or if there are any obvious omissions in the code that
> might be affecting the results.
> Thanking you all in advance.
> Joe
> void SelectionDAGISel::CodeGenAndEmitDAG() {
>   std::string GroupName;
>     std::cout<<"-Joe: we have entered a block"<<std::endl;
>     static int numberOfBlocks=0;
>     numberOfBlocks++;
>     std::cout<<"-We are in Block:"<<numberOfBlocks<<std::endl;
>     int in=0;
>     for (SelectionDAG::allnodes_iterator I =
> CurDAG->allnodes_begin(),E = CurDAG->allnodes_end(); I != E; ++I)
>     {
>         I->setNodeId(in++);
>     }
>     std::ofstream myfile;
>     std::string filename="DDGBLOCK"+convertInt(numberOfBlocks)+".txt";
>     myfile.open (filename.c_str());
>     myfile << "{\n";
>     for (SelectionDAG::allnodes_iterator I =
> CurDAG->allnodes_begin(),E = CurDAG->allnodes_end(); I != E; ++I)
>     {
>         std::cout<<"-Operation "<<I->getOperationName(CurDAG)<<"has
> number "<<I->getNodeId()<<std::endl;
>         int operands =I->getNumOperands();
>         std::cout<<"-  has "<<operands<<" operands"<<std::endl;
>         for(int i=0;i<operands;i++)
>         {
>             std::cout<<"-
> "<<I->getOperand(i)->getOperationName(CurDAG)<<"number:
> "<<I->getOperand(i)->getNodeId()<<std::endl;
>             myfile<<"( "<<I->getOperand(i)->getNodeId()<<"
> "<<I->getNodeId()<<" )"<<std::endl;
>         }
>     }
>     myfile<<"}"<<std::endl;
>     myfile.close();
> //rest of the  void SelectionDAGISel::CodeGenAndEmitDAG() method
> continues as normal...
> --
> Dr Joseph Reddington
> Computer Science, Royal Holloway
> For 2012 I have a resolution to write all my email responses within
> 24-hours - no exceptions. If you've not had a response, there's been a
> misunderstanding somewhere, and it's worth reminding me.
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dag-no-nodes.png
Type: image/png
Size: 8991 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20120904/6e77c4ba/attachment.png>

More information about the llvm-dev mailing list