Restore defaults

Restore "hybrid CPU/GPU" and "paging" defaults.

sha1sum results:
a5598a3e99c313bc1079e027bb8693a3969e6b63 cirqqrack-0.8.4-py3-none-any.whl
0a9bf4834c8117575a6f28c22050be90c5cfa733 cirqqrack-0.8.4.tar.gz

Disable "hybrid stabilizer" by default

This release (only) disables "hybrid stabilizer" mode option by default.

sha1sum results:
f7cdd8248ee07afa88c691c69bb1790bafb99665 cirqqrack-0.8.3-py3-none-any.whl
6c6c455fe94a79072b6e8a76aa2c819a8d27cfcd cirqqrack-0.8.3.tar.gz

Turn CPU/GPU "hybrid" option off by default

This release (only) turns CPU/GPU "hybrid" simulation mode off by default. (The option can still be turned back on by users.)

sha1sum results:
450199a9eb40e89683cdad758e1d0917bc10503d cirqqrack-0.8.2-py3-none-any.whl
f57d49ba85e5b609c747e07391df7c00d9f617d3 cirqqrack-0.8.2.tar.gz

"noise" option

This release enables noisy simulation with a depolarization parameter set by the "noise" option, for every single gate in a circuit.

Cases of circuits with mid-circuit measurement have also been debugged.

Full Changelog: v0.7.0...v0.8.1

sha1sum results:
9d3983c758c6039bed7fd996fa44a83eca77cc20 cirqqrack-0.8.1-py3-none-any.whl
dc15c83b1331d7a6abf2048d95fed25dd06baebf cirqqrack-0.8.1.tar.gz

"tensor_network" flag

Post v1.21.0, PyQrack offers a new optimizing "QTensorNetwork" "layer": this is now controlled by the Cirq simulator flag option "is_tensor_network."

Update for v1.1

This supports breaking changes in Cirq v1.1.

Aggressively try_separate() FSim gate

If we more aggressively try_separate(a, b) in the FSim gate, by calling try_separate(a, b) both before and after the gate, then SDRP might become more effective.

Add ISwap and inverse

This release adds the ISwap gate and its inverse to the API.

Optimize single-shot measurement

The case of simulating for a single measurement shot has been optimized.

Set up for quantum supremacy tests

QUnitMulti can be turned on upon demand, but the convention for some time in the other Qrack plugins is to turn it off by default, as it tends to hurt performance on Optimus laptops, unless the user intentionally configures a multi-device environment.

The FSim() gate has been modified to "aggressively" try to Schmidt decompose its targets before dispatching the gate payload, as with Qrack's C++11 implementation of the 2019 Sycamore ("quantum supremacy") experiment. This is necessary to see maximum "Schmidt decomposition rounding parameter" ("SDRP") actually manage to (vacuously) simulate Sycamore circuits without leading to out-of-memory exceptions, at a baseline. (If SDRP is at its default setting, this should only impose a relatively slight additional execution time cost on the FSim() gate, without affecting fidelity.)