fhelium.backend.ckks.tables

Tensor views used by complete hybrid CKKS key-switch implementations.

encapsulation_configs

function View source

def encapsulation_configs(config: CkksConfig, source_depth: int, low_p_moduli: tuple[int, ...], return_prime: int, return_p_moduli: tuple[int, ...]) -> tuple[CkksConfig, CkksConfig]: ...

Describe the low switch basis and guard-prime return basis.

prepare_encapsulated_raise_operands

function View source

def prepare_encapsulated_raise_operands(rns_context, ntt_context, *, source_depth: int, target_depth: int, low_p_moduli: tuple[int, ...], return_prime: int, return_p_moduli: tuple[int, ...], rounding: str) -> tuple[dict[str, torch.Tensor], dict[str, object]]: ...

Prepare low key-switch, centered extension and return-quotient tables.

The return hint uses [guard, Q, P] rows and one digit. Its relation is H0+H1s = guardP*t+e; quotient row order is [Q, guard, P]. Neither preparation context is retained by the returned numerical operands.

KeySwitchTables

class View source

KeySwitchTables(rns_context: RnsContext, ntt_context: NttContext | None, moddown_p_drop_inverses_montgomery_by_depth: tuple[torch.Tensor, ...] = (), galois_generator: int = 3)

Own the tables used by hybrid CKKS key switching.

Attributes

Name	Type	Default/value
rns_context	RnsContext
ntt_context	NttContext | None
moddown_p_drop_inverses_montgomery_by_depth	tuple[torch.Tensor, ...]	()
galois_generator	int	3

p_inverse_montgomery

property

p_inverse_montgomery: torch.Tensor

Prepare the P-only inverse vector when that quotient is selected.

device

property

device: torch.device

moddown_tables

property

moddown_tables: tuple[torch.Tensor, ...]

Return the depth-indexed P-drop inverse tables read by ModDown.

moddown_tensor_operands

method

def moddown_tensor_operands(q_ids: tuple[int, ...], p_ids: tuple[int, ...], *, rounding: str='nearest', depth: int | None=None) -> dict[str, torch.Tensor]: ...

Prepare quotient data for the represented ordered Q and P rows.

tensor_operands

method

def tensor_operands(depth: int, *, moddown_rounding: str='nearest', modup_method: str='exact', rescale: bool=False, rescale_rounding: str='nearest', output_domain: str='coefficient', key_prime_ids: tuple[int, ...] | None=None, digit_prime_ids: tuple[tuple[int, ...], ...] | None=None) -> tuple[dict[str, torch.Tensor], dict[str, object]]: ...

Expose the selected digit-lift tables and active key digit ranges.

Returned data can populate ordinary Program Tensor placeholders. No execution resource object is needed to consume these table views.

RotationSumTables

class View source

RotationSumTables(tensors: Mapping[str, torch.Tensor], attributes: Mapping[str, object], p_montgomery: torch.Tensor, quotient_row_order: torch.Tensor | None, quotient_parameters: torch.Tensor, quotient_inverses: torch.Tensor, quotient_half: torch.Tensor | None, remaining_q_count: int, dropped_q_count: int, divisor: int, rescale_divisor: int)

Tensor operands for a shared QP key-product sum and its quotient.

divisor is P, or P times the complete leading Q depth group. The quotient row order is surviving Q, dropped Q, P. Its pairwise inverses and optional nearest-rounding offset are prepared before execution.

Attributes

Name	Type	Default/value
tensors	Mapping[str, torch.Tensor]
attributes	Mapping[str, object]
p_montgomery	torch.Tensor
quotient_row_order	torch.Tensor | None
quotient_parameters	torch.Tensor
quotient_inverses	torch.Tensor
quotient_half	torch.Tensor | None
remaining_q_count	int
dropped_q_count	int
divisor	int
rescale_divisor	int

tensor_operands

method

def tensor_operands() -> dict[str, torch.Tensor]: ...

Return numerical operands without retaining a preparation owner.

from_operands

method

def from_operands(values, attributes): ...

Bind supplied numerical operands; never construct missing tables.

prepare_rotation_sum_tables

function View source

def prepare_rotation_sum_tables(tables: KeySwitchTables, depth: int, *, rescale: bool=False, rounding: str='nearest', modup_method: str='exact', key_prime_ids: tuple[int, ...] | None=None, digit_prime_ids: tuple[tuple[int, ...], ...] | None=None) -> RotationSumTables: ...

Prepare ordinary numerical data; key material remains caller supplied.