[LLVMdev] unexpectedly loop hanging

[Alexandru Ionut Diaconescu]

As an update, it is a memory problem which I don't know how to fix.

I tried to skip the problematic piece of code when in the case when the
loop hangs. So I did something like :

          if( instr )
          {
            LLVMContext& C = instr->getContext();

              Value* values[cnt];
              errs()<<"\ngy: \n";

          if(!(desters==7)){    // this is the condition I put to skip the
case when it hanged

              for(int gy=0;gy<cnt;gy++){

values[gy]=ConstantInt::getSigned(Type::getInt64Ty(C),cebag[gy]);
                errs()<<" "<<gy;
              }

             SmallVector<Value*, 100> bla;
             for(int gy=0;gy<cnt;gy++){
              bla.push_back(values[gy]);
              }

            instr->setMetadata("path",MDNode::get(C,bla));


                if( (instr->getMetadata("path")) ){
                  for(int gy=0;gy<cnt;gy++){
                      if(instr->getMetadata("path")->getOperand(gy)) {
                    errs()<<"\n
"<<*(is->getMetadata("path")->getOperand(gy))<<"\n";

                  }
                 }
                }

            } // closing bracket for the extra condition that I put

          }


>From the terminal output, I see that the problematic case is skipped, but
then it was printed:

---------------------------PROCEED TO NEXT
BB------------------------------------
opt: malloc.c:3790: _int_malloc: Assertion `(unsigned long)(size) >=
(unsigned long)(nb)' failed.

So I thought that the problem is regarding memory freeing. I was trying to
free the memory for arrays like values or bla, using delete [] name and
even free(), but I am getting segfault.

I think it is some basic stuff that I miss.

Thank you for any suggestion !



On Tue, May 28, 2013 at 10:02 AM, Alexandru Ionut Diaconescu <
alexandruionutdiaconescu at gmail.com> wrote:

> Hello everyone,
>
> I was able to get all the execution paths between 2 points (basic blocks)
> in my program (with the condition to traverse a loop only once). I mapped
> all the basic blocks to integers and created a correspondent directed graph.
>
> I was able to get all the paths (a path is represented by an integer
> identifier). For my target program I have 72 paths, but the program hangs
> unexpectedly at a for loop when I am adding metadata (which is containing
> the paths). This part of code was working perfectly before of changing the
> algorithm to traverse the loop only once. However, the traverse algorithm
> should be totally independent to the part of the code where I add metadata.
> The single influence that I see is that I have to add more metadata
> operands to instructions. I mention that for each metadata operand I add a
> path = an integer identifier. When this was working, I used to add up to 17
> metadata operands, now I have up to 72.
>
> How do I add metadata: Inside a* *loop iterating through basic blocks,
> for each basic block I take a particular instruction on which I want to
> attach the metadata (a path = an integer identifier). I do like this :
>
> if( instruc )
> {
> LLVMContext& C = instruc->getContext();
>
>              Value* values[cnt];
>              errs()<<"\ngy: \n";
>               for(int gy=0;gy<cnt;gy++)
>               {
>
> values[gy]=ConstantInt::getSigned(Type::getInt64Ty(C),myarray[gy]);
>                 errs()<<" "<<gy;
>                }
> }
>
> 1. I checked before this part of the code the values of myarray and they
> are good
> 2. It works well for the first 6 instructions (the maximum number of
> metadata operands they need is 70), but when I get to the 7th instruction
> to add metadata (with 72 operands), I hangs inside the for loop from above,
> having :
>
> gy:
>  0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
>
> and it hangs. Before this part of the code I print myarray[gy] and it it
> prints all the values from 0 to 71 (the basic block is contained in all
> possible execution paths).
>
> What do you think is the problem? Some memory allocation (I have
> sufficient allocated), maybe I cannot add so many metadata operands?
>
> Thank you for any suggestion !
>
>
>
>
>
>


-- 
Best regards,
Alexandru Ionut Diaconescu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130528/4e2583ea/attachment.html>