Dear LLVM,
Recently I incorporated code into my project such that function pointers
were supported, however it seems that the TarjanSCC_iterator no longer
works on the call graph... main() is no longer reached while the function
pointers are... I can provide code, but I have a feeling there is a
simple fix... Has anyone gone through this?
Thanks,
Dave
Recently I incorporated code into my project such that function pointers
were supported, however it seems that the TarjanSCC_iterator no longer
works on the call graph... main() is no longer reached while the function
pointers are... I can provide code, but I have a feeling there is a
simple fix... Has anyone gone through this?
I'll need some more details before I can help. I assume this has to do
with data structure analysis? Can you provide a testcase?
-Chris
Test Cases:
(attached)
Iteration code:
(...)
typedef TarjanSCC_iterator<CallGraph*> MyTarjan;
CallGraph& callGraph = getAnalysis<CallGraph>();
MyTarjan iter = tarj_begin(&callGraph);
MyTarjan end = tarj_end(&callGraph);
while(iter!=end)
iter++;
(...)
if you take the time to print out the function each non-looping node iter
traverses, it never reaches main...
Thanks,
Dave
test2.c (541 Bytes)
if you take the time to print out the function each non-looping node iter
traverses, it never reaches main...
Ok, first note that the CallGraph class does really stupid (but correct)
things for function pointers, so you're not going to see anything
remarkably nice here... Make sure to read the CallGraph.h comments to
understand exactly what is going on.
Despite that, I can verify that the SCC iterator seems to not be doing the
right thing. On the testcase (below), and with the pass below, I get the
following:
$ opt -load /localhome/lattner/llvm/lib/Debug/libtest.so -test < test.o > /dev/null
New SCC
Node: __main
New SCC
Node: func_one
New SCC
Node: func_two
New SCC
Node: func_three
Which doesn't visit main. Vikram, could you look into this if you get a
chance?
-Chris
------------------- Pass code ----------------------------
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Function.h"
#include "llvm/Pass.h"
#include "Support/TarjanSCCIterator.h"
struct Test : public Pass {
virtual bool run(Module &F) {
typedef TarjanSCC_iterator<CallGraph*> MyTarjan;
CallGraph& callGraph = getAnalysis<CallGraph>();
for (MyTarjan I = tarj_begin(&callGraph), E = tarj_end(&callGraph); I != E; ++I) {
SCC<CallGraph*> *S = *I;
assert(S);
std::cerr << "New SCC\n";
for (unsigned i = 0, e = S->size(); i != e; ++i)
if (Function *F = (*S)[i]->getFunction())
std::cerr << " Node: " << F->getName() << "\n";
else
std::cerr << " Indirect node\n";
}
return false;
}
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
AU.addRequired<CallGraph>();
};
};
RegisterOpt<Test> Z("test", "test");
-------------------- Testcase ------------------------
#include <stdlib.h>
#include <stdio.h>
//#define WORKS
int * a_global;
int a_predicate = 1;
int another_pred =0;
int func_one()
{
return 1;
}
int func_two()
{
return 2;
}
int func_three()
{
return 3;
}
int main()
{
int (*func_ptr)();
if(a_predicate)
func_ptr = func_one;
else if(another_pred)
func_ptr = func_two;
else
func_ptr = func_three;
#ifdef WORKS
return func_one();
#else
return func_ptr();
#endif
}
Thanks,
I've been through the documentation, and if I visit main, i think
everything will turn out correctly... Printing out the Call graph reveals
that main is calling this external node i think you are making reference
to. Could this be the problem?
Dave
I've been through the documentation, and if I visit main, i think
everything will turn out correctly... Printing out the Call graph reveals
that main is calling this external node i think you are making reference
to. Could this be the problem?
The tarjan SCC iterator should visit main, so I think it's broken. I'm
just refering to the fact that the indirect call node is wierd and needs
to be known about by clients of the CallGraph class.
For visualization you can use the 'analyze -print-callgraph' to turn a
call graph into a dot file. You'll probably want to commit this patch
first though:
http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20021202/001850.html
-Chris