Add cuQuantum-based TFQ native ops and layers

[QuantumJaeYoo]

This has large changes and requires some deps error, rewrite documentations, and fixed internal bugs inside random ops, and so forth.

Please look at the PR descriptions in my repo to understand the procedures : https://github.com/jaeyoo/quantum/pulls?q=is%3Apr+is%3Aclosed

Breaking Changes

• Build, compilation, and packaging:
  • The TensorFlow dependency has been upgraded from 2.7.0 to 2.11.0:
  • TensorFlow Quantum is now compiled with _GLIBCXX_USE_CXX11_ABI=1. Downstream projects that encounter std::__cxx11 or [abi:cxx11] linker errors will need to adopt this compiler option. See the GNU C++ Library docs on Dual ABI.
  • TensorFlow Quantum is now compiled with -std=c++17, see install.md for build instructions.
• Cirq dependency has been upgraded from 0.13.1 to ~=1.0
  • cirq_google.XMON was deprecated : [cirq-google] Proposal: Remove XmonDevice, Bristlecone, Foxtail and rename optimize_for_xmon. quantumlib/Cirq#4856
  • QuantumEngineSampler was deprecated : Deprecate QuantumEngineSampler quantumlib/Cirq#5371
    • So, we need ProcessorSampler() for testing
  • cirq.CNOT interface was changed.
    • https://quantumai.google/reference/python/cirq/CNOT
    • No more control, target argument.
  • cirq.SingleQubitGate was deprecated.
    • For testing, use cirq.testing.SingleQubitGate : https://github.com/quantumlib/Cirq/pull/5272/files
    • For implementation, use cirq.Gate.

Major Features and Improvements

• Significant performance improvements by introducing cuQuantum support for circuit execution on Nvidia GPUs:
  • TensorFlow Quantum Keras layers can now be executed on GPU by setting the optional arguement use_cuquantum=True at layer instantiation. Examples:
    • tfq.layers.Expectation(use_cuquantum=True)
    • tfq.layers.SampledExpectation(use_cuquantum=True) (note that cuQuantum runtime is unsupported for any noisy circuit operations
    • tfq.layers.State(use_cuquantum=True)
    • tfq.layers.Sample(use_cuquantum=True)
    • tfq.layers.PQC(model_circuit, operators, use_cuquantum=True)
    • tfq.layers.ControlledPQC(model_circuit, operators, use_cuquantum=True)
  • Important notes:
    • CuQuantum execution is currently only supported for source distributions meaning that the user must build TensorFlow Quantum & tensorFlow-cpu from source following the instructions in install.md.
      • Ensure that the first entry is "N" in the configure.sh script at this step of building. This ensures that you build upon tensorflow-cpu as tensorflow-gpu is unnecessary for CuQuantum support in TensorFlow Quantum.
    • The cuQuantum SDK must be installed locally. See installation instructions for details. As part of the installation process, ensure that the CUQUANTUM_ROOT environment variable is set (referred to in the installation instructions). If not set, bazel will attempt to automatically locate the folder containing the cuQuantum installation upon running configure.sh at this step.
      • Tested on Titan, Ampere and Volta Nvidia GPU architectures. Note that Pascal GPU architectures are not supported, see documentation to check whether your GPU is compatible with cuQuantum
    • Quantum concurrency (global context option) should be turned off when use_cuquantum=True. This can be done by running: tfq.python.quantum_context.set_quantum_concurrent_op_mode(False)

Source: https://github.com/Sinestro38/quantum/blob/master/tensorflow_quantum/release.md
cc: @MichaelBroughton @QuantumJaeYoo

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.