fhelium.backend.ckks.key_switch

Compose hybrid RNS key switching for CKKS ciphertext components.

lift_key_switch_digit

function View source

def lift_key_switch_digit(source, tensors, attributes, digit_index, span, *, out=None, scratch=None): ...

Lift one supplied digit a modulo D=product(q_i) into QP Montgomery form.

Exact lifting uses the representative in [0,D). For D_i=D/q_i and u_i=D_i^{-1} mod q_i, raw CRT uses sum_i(a_i D_i u_i); normalized CRT uses sum_i((a_i u_i mod q_i) D_i). CRT tables are live Tensor operands. Only exact lifting consumes the optional Garner scratch buffer.

accumulate_key_switch_products

function View source

def accumulate_key_switch_products(source: torch.Tensor, *, tensors: Mapping[str, torch.Tensor], attributes: Mapping[str, object], key: torch.Tensor, accumulator: torch.Tensor | None=None) -> torch.Tensor: ...

Accumulate hybrid digit key products into CT2 NTT/Montgomery QP.

source is one coefficient/standard Q polynomial. The optional accumulator is owned by the caller and is updated in place. Keys and digit tables have the same active QP order for every contribution.

key_switch_corrections

function View source

def key_switch_corrections(source: torch.Tensor, *, tensors: Mapping[str, torch.Tensor], attributes: Mapping[str, object], key: torch.Tensor, output_ntt: bool=False, coefficient_c0: torch.Tensor | None=None) -> torch.Tensor: ...

Stream hybrid digit products into QP accumulators and remove P.

NativeKeySwitchImplementation

class View source

NativeKeySwitchImplementation(supports_in_place: bool = False, name: str = 'native-key-switch-streaming', operation_types: tuple[type[Operation], ...] = (ckks.SwitchKeyOp, ckks.ConjugateOp))

Switch CT2 secret relations with a shared streaming QP accumulator.

For input phase $c_0+c_1{s}_{src}$, the key produces corrections $(d_0,d_1)$ under $s_{dst}$ and the result is $(c_0+d_0,d_1)$. Conjugation first applies $X\mapsto X^{-1}$ to both components. Inputs and outputs use coefficient-domain standard Q residues.

Attributes

Name	Type	Default/value
supports_in_place	bool	False
name	str	'native-key-switch-streaming'
operation_types	tuple[type[Operation], ...]	(ckks.SwitchKeyOp, ckks.ConjugateOp)
tensor_requirements		staticmethod(native_key_switch_requirements)
prepare_operation		prepare_native_key_switch

resource_requirements

method

def resource_requirements(invocation: OperationInvocation) -> tuple[ResourceRequirement, ...]: ...

execute

method

def execute(invocation: OperationInvocation, inputs: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativeRelinearizeImplementation

class View source

NativeRelinearizeImplementation(supports_in_place: bool = False, name: str = 'native-relinearize-streaming', operation_types: tuple[type[Operation], ...] = (ckks.RelinearizeOp,))

Relinearize CT3 through one streaming hybrid key-switch schedule.

Attributes

Name	Type	Default/value
supports_in_place	bool	False
name	str	'native-relinearize-streaming'
operation_types	tuple[type[Operation], ...]	(ckks.RelinearizeOp,)
tensor_requirements		staticmethod(native_key_switch_requirements)
prepare_operation		prepare_native_key_switch

resource_requirements

method

def resource_requirements(invocation: OperationInvocation) -> tuple[ResourceRequirement, ...]: ...

execute

method

def execute(invocation: OperationInvocation, inputs: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...