cleanup remaining uses of old control specialization interface #1490
Conversation
anurudhp
force-pushed
the
2024/11/06-refactor-ctrl-sys
branch
from
f0b0dc5 to
56544d6
Compare
| @@ -142,6 +142,9 @@ def get_ctrl_system(self, ctrl_spec: 'CtrlSpec') -> Tuple['Bloq', 'AddControlled | |||
| def adjoint(self) -> 'Bloq': | |||
| return self | |||
|
|
|||
| def _has_unitary_(self): | |||
There was a problem hiding this comment.
Why is this necessary in this PR?
There was a problem hiding this comment.
cirq.has_unitary fails to realize that this is unitary because when it decomposes, it gets And^dagger gates which aren't unitary (#1488)
| @@ -63,7 +62,7 @@ class BernoulliEncoder(SelectOracle, SpecializedSingleQubitControlledExtension): | |||
|
|
|||
| @cached_property | |||
| def control_registers(self) -> Tuple[Register, ...]: | |||
| return () if self.control_val is None else (Register('control', QBit()),) | |||
| return () | |||
There was a problem hiding this comment.
Do we need to keep around this property if it's just empty? (I missed the older PRs so probably need to read those)
There was a problem hiding this comment.
SelectOracle has this as an abstract property, but as there's no special controlled version I just return ().
Ideally the select/prepare interfaces should be updated to the newer BlockEncoding style interfaces (would be a major refactor), but as of now as callees use .control_registers, I'm keeping it as-is.
Removes the finally uses of the old
SpecializedSingleQubitControlledExtensioninterface in favour of the new interface from #1451.For controlled versions of
LCUBlockEncoding,SelectBlockEncodingetc, do not update the subbloqs to controlled. Instead, use the appropriate subbloq/controlled-subbloq when decomposing/building call graph.