Extending CUDAQ with a new Simulator: nvqir_add_backend Error

[Maxime-pixel]

Hi!

I'm trying to run the MySimulator example to add a new simulator Backend to CUDA-Q.
But when running:

cmake .. -G Ninja -DNVQIR_DIR="$CUDA_QUANTUM_PATH/lib/cmake/nvqir"

I run into the following error:

CMake Error at /opt/nvidia/cudaq/lib/cmake/nvqir/NVQIRConfig.cmake:39 (add_library):
  No SOURCES given to target: nvqir-MySimulator
Call Stack (most recent call first):
  CMakeLists.txt:4 (nvqir_add_backend)

As I'm unaware of the behavior of nvqir_add_backend I don't know if I can fix it myself.
I saw someone else seems to have another issue while building but the error seemed completely different.
Any help would be appreciated.

Some details on what I've done:
I did an exact copy of the CMakelists.txt in the tutorial.
and replaced all "..." by some prints in the MySimulator.cpp file since I just want to build for now.

Thank you for your help!

[1tnguyen]

@Maxime-pixel Thank you for the question.

Indeed, the CMakeLists.txt snippet is not up to date. nvqir_add_backend requires a GPURequirements boolean argument after the backend name.

A correct usage should be: nvqir_add_backend(MySimulator true MySimulator.cpp "") (if the backend is GPU-accelerated or false otherwise)

Additionally, the code snippet for MySimulator.cpp requires some modifications to build successfully.
I've added this ticket for us to update this page.

In the meantime, the code snippet below should compile

#include "CircuitSimulator.h"

namespace {

class MySimulator : public nvqir::CircuitSimulatorBase<double> {

protected:
  /// @brief Grow the state vector by one qubit.
  void addQubitToState() override {}

  /// @brief Apply the given gate
  void applyGate(const GateApplicationTask &task) override {}

public:
  MySimulator() = default;
  virtual ~MySimulator() = default;

  bool measureQubit(std::size_t qubitIdx) override {}

  void resetQubit(const std::size_t index) override {}

  cudaq::ExecutionResult sample(const std::vector<std::size_t> &qubits,
                                const int shots) override {}

  std::string name() const override { return "MySimulator"; }
  void deallocateStateImpl() override {}
  void setToZeroState() override {}

  NVQIR_SIMULATOR_CLONE_IMPL(MySimulator)
};

} // namespace

/// Register this Simulator with NVQIR.
NVQIR_REGISTER_SIMULATOR(MySimulator, MySimulator)

You could also refer to CUDA-Q simulator implementations (https://github.com/NVIDIA/cuda-quantum/tree/main/runtime/nvqir) for reference.

[Maxime-pixel]

@Maxime-pixel Thank you for the question.

Indeed, the CMakeLists.txt snippet is not up to date. nvqir_add_backend requires a GPURequirements boolean argument after the backend name.

A correct usage should be: nvqir_add_backend(MySimulator true MySimulator.cpp "") (if the backend is GPU-accelerated or false otherwise)

Additionally, the code snippet for MySimulator.cpp requires some modifications to build successfully. I've added this ticket for us to update this page.

In the meantime, the code snippet below should compile

#include "CircuitSimulator.h"

namespace {

class MySimulator : public nvqir::CircuitSimulatorBase<double> {

protected:
  /// @brief Grow the state vector by one qubit.
  void addQubitToState() override {}

  /// @brief Apply the given gate
  void applyGate(const GateApplicationTask &task) override {}

public:
  MySimulator() = default;
  virtual ~MySimulator() = default;

  bool measureQubit(std::size_t qubitIdx) override {}

  void resetQubit(const std::size_t index) override {}

  cudaq::ExecutionResult sample(const std::vector<std::size_t> &qubits,
                                const int shots) override {}

  std::string name() const override { return "MySimulator"; }
  void deallocateStateImpl() override {}
  void setToZeroState() override {}

  NVQIR_SIMULATOR_CLONE_IMPL(MySimulator)
};

} // namespace

/// Register this Simulator with NVQIR.
NVQIR_REGISTER_SIMULATOR(MySimulator, MySimulator)

You could also refer to CUDA-Q simulator implementations (https://github.com/NVIDIA/cuda-quantum/tree/main/runtime/nvqir) for reference.

Thank you! (and sorry for the late reply) I was able to run the ninja install with this.
Running this line did add my libnvqir-MySimulator.so in /lib and Mysimulator.config in /targets (The documentation says /platforms btw) but I still don't see MySimulator as an available backend when running:

$CUDA_QUANTUM_PATH/bin/nvq++ --list-targets

And I get the following when trying to invoke it with --target MySimulator

$CUDA_QUANTUM_PATH/bin/nvq++ test.cpp --target MySimulator
nvq++:646: Invalid Target: (MySimulator)

Is there another step I nee to take before I can invoke it?

[1tnguyen]

There was a bug in the nvqir_add_backend CMake function, which is fixed in #3164.
Could you please try it again with the latest CUDA-Q main or the 'nightly' Docker image?