Ghost qubit in library mode

[boschmitt]

Required prerequisites

• 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 sampling of a circuit with a single qubit returns a result with a bitstring with two bits.

Steps to reproduce the bug

Compile the following circuit in library mode:

#include <cudaq.h>
#include <iostream>

void bar() __qpu__ {
  cudaq::qubit a;
  cudaq::x(a);
  auto a0 = cudaq::mz(a);
  cudaq::reset(a);
  auto a1 = cudaq::mz(a);
}

int main() {
  std::cout << "Bar:\n";
  auto result = cudaq::sample(bar);
  result.dump();

  return 0;
}

It's execution returns:

Bar:
{ 00:1000 }

Note: this problem seems to not happen when we add a quantum operator after the reset.

Expected behavior

It should return:

Bar:
{ 0:1000 }

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

Not a regression

Environment

• CUDA Quantum version: nvq++ proto-0.6.0 (main branch)
• C++ compiler: g++ 13.2.0
• Operating system: Ubuntu 23.10 (Docker container)

Suggestions

No response

[bmhowe23]

It seems to me that two separate measurements should return two separate answers, so I would think that the correct answer should be:

Bar:
{ 10:1000 }

[boschmitt]

It seems to me that two separate measurements should return two separate answers, so I would think that the correct answer should be:

Bar:
{ 10:1000 }

This is not the semantics of cudaq::sample. In the bitstrings of its result, each bit correspond to the measure state of a qubit, not a sequence of measurements of the same qubit.

[bmhowe23]

This is not the semantics of cudaq::sample. In the bitstrings of its result, each bit correspond to the measure state of a qubit, not a sequence of measurements of the same qubit.

This would mean that cudaq::sample would be unable to reuse a qubit and save the measurement from earlier usages. Maybe it's just me, but that seems undesirable to me. It's something that the QIR spec permits, so prohibiting it within a core CUDA Quantum feature (like cudaq::sample) seems suboptimal.

Stepping back a bit, will this issue still be relevant if we consider changes to cudaq::sample that allow the user to more clearly specify the return values (i.e. use something other than void)?