[cfe-dev] Question about LLVM dependence graph

[John Criswell]

On 10/31/13 8:42 AM, Feng Lu wrote:
> Hi, all,
>  I am new to LLVM and Clang,
>  I am interested in the following problem:
>    Given a function with parameters, obtain the part of parameters 
> which are used in branch conditions within the function or sub-functions.
> For the following example, I hope to get v.c1
>   struct p {
>       int c1,
>       int c2
>   };
>   void foo (p v) {
>       if (v.c1)
>           ...
>       else
>          ...
>   }
> Does LLVM or Clang contains the code that could easily achieve this 
> efficiently?

For scalar function arguments, you can just follow the explict def-use 
chains in the LLVM IR.  See the Value::use_begin() and Value::use_end() 
methods in the doxygen documentation: 
http://llvm.org/doxygen/classllvm_1_1Value.html.

For values that live in or escape into memory, the problem is trickier: 
you need to do classical Kam-Ullman data flow analysis (specifically, 
reaching definitions analysis), and you'll need to deal with aliasing of 
pointers.  LLVM does not currently have an analysis for this, as far as 
I know.

Of course, all of this gets even more complicated if you're doing it 
inter-procedurally or want to take into account external library code.

-- John T.



> Thanks,
> Feng Lu
>
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20131031/abc480f6/attachment.html>