Struct RowFoldTables 

pub struct RowFoldTables<F, const N_TABLES: usize> { /* private fields */ }

Subset-sum tables for folding a matrix of single-bit rows against one weight per row.

The fold contracts the row axis, leaving one field element per column:

    out[b] = sum_r weight[r] * bit_b(row[r])

Taking eight rows at a time turns that inner sum into a single table lookup. Table g covers rows 8g through 8g + 7 and holds every subset sum of their weights.

Implementations

impl<F: BinaryField, const N_TABLES: usize> RowFoldTables<F, N_TABLES>

pub fn new(weights: &[F]) -> Self

Builds the tables from one weight per row, from the first row onwards.

Weights past the end of the slice read as zero. Those weight rows past the end of the matrix, which read as zero as well. So one table layout serves a chunk the row list does not fill.

pub fn fold_into<PB>( &self, groups: impl IntoIterator<Item = RowGroup<PB>>, sums: &mut ColumnSums<F, N_TABLES>, )

where PB: PackedField<Scalar = B1> + UnderlierView, PB::Underlier: Divisible<u8>,

Folds each group of eight rows into the column sums.

Groups pair with tables in order, so the gth group is weighted by rows 8g onwards. An iterator yielding fewer groups leaves the remaining rows out, which reads them as zero.

Preconditions

• A row must be one byte wide per table, so the groups cover every column exactly once.

Trait Implementations

impl<F: Clone, const N_TABLES: usize> Clone for RowFoldTables<F, N_TABLES>

fn clone(&self) -> RowFoldTables<F, N_TABLES>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<F: Debug, const N_TABLES: usize> Debug for RowFoldTables<F, N_TABLES>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.