[LLVMdev] Calling a function with bad signature, possible bug.
Iaroslav Markov
ymarkov at cs.stonybrook.edu
Thu Jul 26 10:30:10 PDT 2012
Hello,
I'm having troubles with writing a pass. In my pass I've have created a function that has two parameters - both of type i8*. Initially I wrote this function in C, that I translated it into IR, and then by using llc -march=cpp I got it's implementation in cpp code that actually inserts IR instructions. Then, I inserted this cpp code in my pass.
And in some places of a program my pass inserts instrumentation that calls this function.
The problem I'm experiencing is very strange. I'm getting "Calling a function with bad signature", when my pass inserts a call to this function. But I don't have any problems with my pass if I run clang -emit-llvm -S, then llc and then gcc. With this approach I get a binary with my instrumentation inserted.
This line causes a problem - assert((i >= FTy->getNumParams() ||
FTy->getParamType(i) == Args[i]->getType()) &&
"Calling a function with a bad signature!");
I tried to get a dump of FTy->getParamType(i) and Args[i]->getType(), but I wasn't able to do it in gdb - everytime I create a breakpoint it just misses it. So I made the following:
FTy->getParamType(i)->dump();
errs() << "\nparamtype end----------- \n";
Args[i]->getType()->dump();
errs() << "\ngettype end ------------\n";
errs() << i << "\ni------------------\n";
errs() << FTy->getNumParams() << "\ngetnumparams end ------------\n";
errs() << "assert " << (FTy->getParamType(i) == Args[i]->getType()) << "\n";
assert((i >= FTy->getNumParams() ||
FTy->getParamType(i) == Args[i]->getType()) &&
"Calling a function with a bad signature!");
I know, this is a bizzare, but still. The output I've got:
doTaint
i8*
paramtype end-----------
i8*
gettype end ------------
0
i------------------
2
getnumparams end ------------
assert 0
You see - totally unexplainable: FTy->getParamType(i) and Args[i]->getType() are the same, but assertion still fails.
Can anybody comment on this.
Best,
Yaroslav.
More information about the llvm-dev
mailing list