Struct TPUQuantumSimulator 

pub struct TPUQuantumSimulator { /* private fields */ }

TPU-accelerated quantum simulator

Implementations

impl TPUQuantumSimulator

pub fn new(config: TPUConfig) -> Result<Self>

Create a new TPU quantum simulator.

HONEST AVAILABILITY GATE: this build links no TPU runtime (no JAX/XLA/libtpu), so it cannot place tensors on, or dispatch work to, a physical Cloud/Edge TPU. Requesting a real device type (TPUv2..TPUv5p) therefore fails loudly rather than fabricating that the silicon is present.

TPUDeviceType::Simulated is explicitly supported: it is a CPU-side numerical simulation of the device math (the gate applications below compute the exact state-vector linear algebra on the CPU). It never claims a TPU executed anything.

pub fn execute_batch_circuit( &mut self, circuits: &[InterfaceCircuit], initial_states: &[Array1<Complex64>], ) -> Result<Vec<Array1<Complex64>>>

Execute batched quantum circuit

pub fn compute_expectation_values_tpu( &mut self, states: &[Array1<Complex64>], observables: &[String], ) -> Result<Array2<f64>>

Compute Pauli-observable expectation values (CPU numerical simulation).

Each observable is a compact single-qubit Pauli string of the form "<P><qubit>", e.g. "Z0", "X1", "Y2" (identity on every other qubit). The result <psi|P|psi> is computed exactly from the amplitudes; no value is fabricated.

pub const fn get_device_info(&self) -> &TPUDeviceInfo

Get device information

pub const fn get_stats(&self) -> &TPUStats

Get performance statistics

pub fn reset_stats(&mut self)

Reset performance statistics

pub const fn is_tpu_available(&self) -> bool

Whether a real TPU runtime is available.

HONEST: this build links no TPU runtime, so a physical TPU is never available. This returns false even though a CPU Simulated device is in use — that is a numerical model, not real silicon. Use Self::is_simulated to check for the CPU-simulation device.

pub fn is_simulated(&self) -> bool

Whether this simulator is the CPU-side numerical Simulated device.

pub const fn get_memory_usage(&self) -> (usize, usize)

Get memory usage

pub fn garbage_collect(&mut self) -> Result<usize>

Reclaim memory held by tensor buffers that are no longer device-resident.

HONEST: this frees exactly the bytes of buffers whose on_device flag is false (i.e. genuinely releasable in the model) and updates the accounting accordingly. It does not fabricate a fixed “freed 10%” figure. Returns the number of bytes actually reclaimed.