fhelium.backend.ntt.tables

Typed, family-specific NTT table materialization.

IndexedRadix2Tables

class View source

IndexedRadix2Tables(forward_even_indices: torch.Tensor, forward_odd_indices: torch.Tensor, forward_twiddles: torch.Tensor, inverse_even_indices: torch.Tensor, inverse_odd_indices: torch.Tensor, inverse_twiddles: torch.Tensor)

Schedules and stage-expanded twiddles for indexed execution.

Index tensors are torch.int32 on the backend device. Preparation validates their per-stage coefficient partitions. The index structure stays fixed throughout execution; changing it requires preparing the schedule again. Twiddle tensors are integral [limb, stage, butterfly] Montgomery rows aligned with QP parameter columns.

Attributes

Name	Type	Default/value
forward_even_indices	torch.Tensor
forward_odd_indices	torch.Tensor
forward_twiddles	torch.Tensor
inverse_even_indices	torch.Tensor
inverse_odd_indices	torch.Tensor
inverse_twiddles	torch.Tensor

CompactRadix2Tables

class View source

CompactRadix2Tables(forward_twiddles: torch.Tensor, inverse_twiddles: torch.Tensor)

Compact execution tables aligned with QP parameter columns.

Montgomery tables have shape [limb,N]. Shoup32 tables have shape [limb,N,2] and contain standard-root/quotient-word pairs.

Attributes

Name	Type	Default/value
forward_twiddles	torch.Tensor
inverse_twiddles	torch.Tensor

CompactPowerOfTwoRadixTables

class View source

CompactPowerOfTwoRadixTables(forward_outer_twiddles: torch.Tensor, inverse_outer_twiddles: torch.Tensor, forward_radix_root_powers: torch.Tensor, inverse_radix_root_powers: torch.Tensor)

Packed outer twists and roots for strict fixed-radix digits.

Outer shapes are [limb, N - 1] and root shapes are [limb, radix] in QP row order, with Montgomery residues.

Attributes

Name	Type	Default/value
forward_outer_twiddles	torch.Tensor
inverse_outer_twiddles	torch.Tensor
forward_radix_root_powers	torch.Tensor
inverse_radix_root_powers	torch.Tensor

NttTables

data View source

NttTables = IndexedRadix2Tables | CompactRadix2Tables | CompactPowerOfTwoRadixTables

MaterializeParameterRows

data View source

MaterializeParameterRows = Callable[[torch.Tensor], torch.Tensor]

prepare_ntt_tables

function View source

def prepare_ntt_tables(policy: NttBackendPolicy, ckks_config: CkksConfig, *, materialize_parameter_rows: MaterializeParameterRows, device: torch.device, rns_params: torch.Tensor, dtype: torch.dtype) -> NttTables: ...

Build execution-ready Tensor tables in the selected numerical layout.

Prime rows follow depth-zero QP order. Montgomery tables are converted during preparation using the supplied RNS parameters. Shoup32 pairs retain their standard roots and quotient words without Montgomery conversion.