[RFC] Handle device pointer for state initialization

[sacpis]

Fixes #1788

This PR checks if the pointer is a device pointer and then does memcpy from device to device.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[bmhowe23]

The change LGTM, but it would be good for @1tnguyen to review as well. I don't know if #1788 was documenting the desire to avoid a memory copy entirely (similar to a std::move()) or not.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[sacpis]

@1tnguyen May I please request for you review?

[1tnguyen]

LGTM 👍 Thanks, @sacpis.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[bmhowe23]

Just capturing the offline discussion amongst the 4 of us...

[bmhowe23]

This was discussed offline, but just to capture it here, we think it would be good to add a multi-GPU test somewhere to exercise this new code in this if branch; otherwise, the code in here would never have been tested, and there is some question about whether or not the calls to cudaSetDevice would work as intended here.

[bmhowe23]

Another alternative would be to throw an "unsupported" error message here, but I think we should still add a test for the "same GPU" functionality as well.

[bmhowe23]

Same comment here.

[sacpis]

Here I am blocked on including nvqir/CircuitSimulator in get_state_tester. Seems like the builder is not happy including it in the file as it throws the below error.

In file included from /workspaces/cuda-quantum/runtime/common/Logger.h:14,
                 from /workspaces/cuda-quantum/runtime/nvqir/CircuitSimulator.h:13,
                 from /workspaces/cuda-quantum/unittests/gpu/get_state_tester.cu:12:
/workspaces/cuda-quantum/runtime/common/FmtCore.h:16:10: fatal error: fmt/chrono.h: No such file or directory
   16 | #include <fmt/chrono.h>
      |          ^~~~~~~~~~~~~~
compilation terminated.

[bmhowe23]

Here I am blocked on including nvqir/CircuitSimulator in get_state_tester. Seems like the builder is not happy including it in the file as it throws the below error.

In file included from /workspaces/cuda-quantum/runtime/common/Logger.h:14,
                 from /workspaces/cuda-quantum/runtime/nvqir/CircuitSimulator.h:13,
                 from /workspaces/cuda-quantum/unittests/gpu/get_state_tester.cu:12:
/workspaces/cuda-quantum/runtime/common/FmtCore.h:16:10: fatal error: fmt/chrono.h: No such file or directory
   16 | #include <fmt/chrono.h>
      |          ^~~~~~~~~~~~~~
compilation terminated.

Thanks, Sachin. I'm definitely open to any other ideas for testing here. I just think that we need to have some sort of test to prove that code changes are working as intended. If a unit test isn't working out, perhaps an application-level test (one where the user is providing a pointer to GPU memory) would work better than a unit test?

[sacpis]

Just had a call with @1tnguyen regarding passing the pointer to GPU memory. Putting this PR into the draft mode and adding RFC here.

Seems like it might involve introduction of some new APIs to qvector, something like these

  qubit::qubit(const vector<complex<double>>&);
  qubit::qubit(const initializer_list<complex<double>>&);
  qvector::qvector(const vector<complex<double>>&);
  qvector::qvector(const initializer_list<complex<double>>&);

in order to support passing a pointer to the GPU memory.
This is related to issue #1086.

@1tnguyen @amccaskey Would you please add your thoughts here?

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.