`denormals_are_zeros` setting has unexpected side effects

[richrines1]

qsim updates global flush-to-zero and denormals-are-zero control flags, but never resets them to their original values. This can lead to unexpected changes in the behavior of downstream code, e.g.:

import cirq
import ctypes

from qsimcirq import QSimOptions, QSimSimulator


orig_value = ctypes.c_float(1e-40).value

_ = QSimSimulator(qsim_options=QSimOptions(denormals_are_zeros=False)).simulate(cirq.Circuit())

assert ctypes.c_float(1e-40).value == orig_value  # passes

_ = QSimSimulator(qsim_options=QSimOptions(denormals_are_zeros=True)).simulate(cirq.Circuit())

assert ctypes.c_float(1e-40).value == orig_value  # fails

_ = QSimSimulator().simulate(cirq.Circuit())

assert ctypes.c_float(1e-40).value == orig_value  # passes

it would be nice if these flags were reset after the simulation (ideally even if the simulation itself fails)

[msoud2509]

Hi @mhucka, I am interested in helping out here, I saw you had a draft PR of an implementation, were you looking to take this?

Thanks!

[mhucka]

Hi @mhucka, I am interested in helping out here, I saw you had a draft PR of an implementation, were you looking to take this?

Thank you for your interest! I did start, but I'm not convinced it's great. You're welcome to try it, perhaps with another approach. (Or start with the draft I had and see if it still works and can be verified to do the right thing, and fix it or improve it as needed.)