Matrix Kernels | NumKong for JavaScript

API

ClassPackedMatrixOpaque packed matrix container.9

class PackedMatrix

Packed layout is not indexable — this is a data container for packed GEMM kernels.

N-API path: buffer is a V8-managed ArrayBuffer, auto-freed by GC.
WASM path: stores a heap pointer, dispose() calls Module._free().

PackedMatrix

types.ts:387 ↗

new PackedMatrix(
    buffer: ArrayBuffer,
    columns: number,
    depth: number,
    dtype: DType,
    byteLength: number,
)

dispose

types.ts:395 ↗

dispose(): void

toString

types.ts:398 ↗

toString(): string

columns

types.ts:380 ↗

readonly columns: number

depth

types.ts:381 ↗

readonly depth: number

dtype

types.ts:382 ↗

readonly dtype: DType

byteLength

types.ts:383 ↗

readonly byteLength: number

buffer

types.ts:384 ↗

readonly buffer: ArrayBuffer

disposed

types.ts:396 ↗

readonly disposed: boolean

Functions

dotsPackedSizeQueries the packed buffer's byte count for a matrix with the given [columns,depth] and dtype, matching what dotsPack would allocate.

numkong.ts:440 ↗

function dotsPackedSize(
    columns: number,
    depth: number,
    dtype: DType,
): number

Parameters

`columnsnumber`		Number of vectors being packed, the matrix rows.
`depthnumber`		Dimensionality per vector, the matrix columns.
`dtypeDType`		Element dtype of the source matrix.

Returns

The byte count of the packed buffer.

dotsPackPacks a Matrix for use with packed GEMM-like operations, such as dotsPacked.

numkong.ts:449 ↗

function dotsPack(matrix: Matrix): PackedMatrix

Parameters

`matrixMatrix`		The matrix to pack, [rows,columns] shaped.

Returns

The packed matrix, sized via dotsPackedSize.

dotsPackedComputes the dot products between every row of a and every packed vector in packed.

numkong.ts:499 ↗

function dotsPacked(
    a: Matrix,
    packed: PackedMatrix,
    out?: Matrix,
): Matrix

Parameters

`aMatrix`		The query matrix, [rows,columns] shaped, columns matching packed's depth.
`packedPackedMatrix`		The packed matrix produced by `dotsPack`.
`outMatrix`		Optional output matrix to write into, [a.rows,packed.columns] shaped.

Returns

The distance matrix: out when given, otherwise a newly allocated Matrix.

angularsPackedComputes the angular distances between every row of a and every packed vector in packed.

numkong.ts:510 ↗

function angularsPacked(
    a: Matrix,
    packed: PackedMatrix,
    out?: Matrix,
): Matrix

Parameters

`aMatrix`		The query matrix, [rows,columns] shaped, columns matching packed's depth.
`packedPackedMatrix`		The packed matrix produced by `dotsPack`.
`outMatrix`		Optional output matrix to write into, [a.rows,packed.columns] shaped.

Returns

The distance matrix: out when given, otherwise a newly allocated Matrix.

euclideansPackedComputes the Euclidean distances between every row of a and every packed vector in packed.

numkong.ts:521 ↗

function euclideansPacked(
    a: Matrix,
    packed: PackedMatrix,
    out?: Matrix,
): Matrix

Parameters

`aMatrix`		The query matrix, [rows,columns] shaped, columns matching packed's depth.
`packedPackedMatrix`		The packed matrix produced by `dotsPack`.
`outMatrix`		Optional output matrix to write into, [a.rows,packed.columns] shaped.

Returns

The distance matrix: out when given, otherwise a newly allocated Matrix.

dotsSymmetricComputes the all-pairs dot products between every row of vectors and every other row.

numkong.ts:533 ↗

function dotsSymmetric(
    vectors: Matrix,
    out?: Matrix,
    options?: { rowsBegin?: number; rowsEnd?: number },
): Matrix

Parameters

`vectorsMatrix`		The matrix of vectors, [rows,columns] shaped.
`outMatrix`		Optional output matrix to write into, [vectors.rows,vectors.rows] shaped.
`options{ rowsBegin?: number; rowsEnd?: number }`		Optional row range: `rowsBegin` and `rowsEnd` restrict which rows of the symmetric matrix are computed, defaulting to all rows.

Returns

The distance matrix: out when given, otherwise a newly allocated Matrix.

angularsSymmetricComputes the all-pairs angular distances between every row of vectors and every other row.

numkong.ts:545 ↗

function angularsSymmetric(
    vectors: Matrix,
    out?: Matrix,
    options?: { rowsBegin?: number; rowsEnd?: number },
): Matrix

Parameters

`vectorsMatrix`		The matrix of vectors, [rows,columns] shaped.
`outMatrix`		Optional output matrix to write into, [vectors.rows,vectors.rows] shaped.
`options{ rowsBegin?: number; rowsEnd?: number }`		Optional row range: `rowsBegin` and `rowsEnd` restrict which rows of the symmetric matrix are computed, defaulting to all rows.

Returns

The distance matrix: out when given, otherwise a newly allocated Matrix.

euclideansSymmetricComputes the all-pairs Euclidean distances between every row of vectors and every other row.

numkong.ts:557 ↗

function euclideansSymmetric(
    vectors: Matrix,
    out?: Matrix,
    options?: { rowsBegin?: number; rowsEnd?: number },
): Matrix

Parameters

`vectorsMatrix`		The matrix of vectors, [rows,columns] shaped.
`outMatrix`		Optional output matrix to write into, [vectors.rows,vectors.rows] shaped.
`options{ rowsBegin?: number; rowsEnd?: number }`		Optional row range: `rowsBegin` and `rowsEnd` restrict which rows of the symmetric matrix are computed, defaulting to all rows.

Returns

The distance matrix: out when given, otherwise a newly allocated Matrix.