PennyLaneAI/pennylane v0.46.0b1

Release 0.46.0 Beta 1

on GitHub

Functionality for heterogeneous compilation and low-latency execution of QNodes across GPUs, CPUs, FPGAs, and QPUs has been added to PennyLane. qp.Backline builds a device from a qp.Controller, zero or more qp.Coprocessor, and a transport, describing where each part of a workload runs and how the parts communicate. The resulting device binds to a QNode, and requires the Catalyst compiler to execute. (#9772) (#9970) (#9975)

import pennylane as qp


qdev = qp.device("lightning.qubit", wires=4)

CPU = qp.Controller(
    device=qdev,
    remote=True,
    executor_options={"host": "192.168.3.15", "port": 7810},
)

GPU = qp.Coprocessor(
    coprocessor_fn="decoder",
    hardware="gpu",
    endpoint=qp.Endpoint("192.168.1.3", 18590),
)

dev = qp.Backline(controller=CPU, coprocessors=[GPU], transport="rdma")

@qp.qjit(capture=True)
@qp.qnode(dev)
def circuit(x):
    qp.RX(x, wires=0)
    return qp.expval(qp.Z(0))

When using a backline to execute a quantum program, there are multiple ways to incorporate quantum error correction (QEC) encoding and decoding:

• Implicit QEC: provide the qec_code argument to qp.Backline, and encoding (as an MLIR compilation pass) and decoding are automatically applied.

• Explicit QEC: if the qec_code argument is not provided, no automatic encoding or decoding will occur. It will be assumed that the circuit defined in the PennyLane frontend represents a physical circuit; encoding and decoding (via qp.backline.decode) should be manually defined in the frontend.

  @qp.qjit(capture=True, autograph=True)
  @qp.qnode(dev)
  def circuit():
      # encoded logical circuit
      ...
  
      # measure syndromes, decode, and correct
      syndrome = extract_syndromes() # via qp.measure
  
      correction = qp.backline.decode(syndrome)
  
      if correction[0] == 1:
          qp.X(0)
  
      return qp.expval(qp.Z(0))

The decoding function to be executed on the coprocessor can be specified in multiple ways:

• A Triton kernel: The provided function qp.backline.triton_decoder compiles a Python Triton decoder function into a shared library that can be used as a coprocessing function. Alternatively, qp.backline.css_bp_decoder is a convenience function to compile a CSS Tanner graph to a belief propagation decoder using Triton.

• A precompiled library: CoprocessorFunction registers a precompiled library symbol and (optionally) the library path.

For more details, see the backline documentation and the Backline demo.

The ability for a compiled program to call a runtime entry point directly via its C symbol name has been added. A symbol's signature is declared once with qp.runtime_declare and called with qp.runtime_call from inside a qjit program. (#9970)

import pennylane as qp

qp.runtime_declare("example_local_rounds", "(ptr, u32) -> u64", library="/path/librounds.so")

def program(session):
    return qp.runtime_call("example_local_rounds", session, 100000)