Cannot reuse a std::vector<cudaq::measure_result> variable in __qpu__ kernels

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

The compiler reports a fatal casting error when trying to save a measurement to a pre-existing variable in a kernel. This operation (and other similar operations) are needed to implement quantum error correction algorithms in CUDA-Q, and these operations don't appear to be supported when --enable-mlir is selected.

Steps to reproduce the bug

The following file demonstrates the problem:

#include "cudaq.h"

void kernel() __qpu__ {
  cudaq::qvector q(2);
  std::vector<cudaq::measure_result> mz_res;
  mz_res = mz(q);
  // std::vector<cudaq::measure_result> mz_res = mz(q);   // This works but the above does not.
}

int main(int argc, char *argv[]) {
  auto counts = cudaq::sample(kernel);
  counts.dump();
  return 0;
}

Compiling produces this error:

$ nvq++ --enable-mlir issue.cpp
cudaq-quake: /opt/llvm/include/llvm/Support/Casting.h:567: decltype(auto) llvm::cast(const From&) [with To = mlir::FunctionType; From = mlir::Type]: Assertion `isa<To>(Val) && "cast<Ty>() argument of incompatible type!"' failed.
PLEASE submit a bug report to https://github.com/NVIDIA/cuda-quantum and include the crash backtrace.
Stack dump:
<snip>

We ultimately want to get this level of support working:

void kernel() __qpu__ {
  cudaq::qvector q(2);
  std::vector<bool> syndrome;
  std::vector<cudaq::measure_result> last_mz = mz(q);
  for (std::size_t iter = 0; iter < 10; iter++) {
    // Perform some operations

    // Then measure some qubits and compare them to the last time we measured them.
    auto new_mz = mz(q);

    // We want to either perform this:
    syndrome = last_mz ^ new_mz;  // error: invalid operands to binary expression 
    last_mz = new_mz;             // error: not yet implemented: unhandled std::vector member function

    // Or this:
    for (std::size_t i = 0; i < new_mz.size(); i++) {
      syndrome[i] = last_mz[i] ^ new_mz[i]; // error: vector dereference is not yet supported
      last_mz[i] = new_mz[i];               // This appears to compile without issues
    }

    // But neither of those paths work yet, due to the aforementioned errors.
  }
}

int main(int argc, char *argv[]) {
  auto counts = cudaq::sample(kernel);
  counts.dump();
  return 0;
}

Expected behavior

We need all of the aforementioned examples to compile and run without errors.

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• CUDA Quantum version: latest (f7983754b)
• Python version:
• C++ compiler:
• Operating system:

Suggestions

No response

Start with the reproducer in issue.cpp and run nvq++ --enable-mlir to observe the fatal casting error. Trace the compiler path for assigning std::vectorcudaq::measure_result values in qpu kernels; done means the listed measurement, assignment, indexing, and loop examples compile and run without errors.