LLVM CUDA: Load/Store operands not captured


I am trying to write a CUDA memory instrumentation code. So far I have made little progress, please see attached file.

In lines 82 and 88 I get error as: ‘LoadOperand’ is not captured and ‘StoreOperand’ is not captured

Unfortunately, I am not able to tell if the error is generic C++ or LLVM related. Earlier I had verified the “syntax” of such code in a dummy C++ implementation, which worked just fine. But I am lost here.

Can someone please point me in the right direction?

Note: the code is still incomplete (profiling instructions insertion part etc.). And might have more bugs. You can ignore those if you want.

thank you.


CUDAMemTrace.cpp (3.79 KB)