API reference · 10

Matrix

Stores value needed to assemble a block-sparse matrix on the GPU.

matrix

matrix(rows=0, cols=0, symmetric_storage=False)

Create an empty block-sparse matrix. Derivative matrices normally populate this representation automatically.

Parameter	Type	Default	Description
`rows`	`int`	`0`	Number of matrix rows.
`cols`	`int`	`0`	Number of matrix columns.
`symmetric_storage`	`bool`	`False`	Treat each stored off-diagonal block as also representing its transpose.

Example

empty_matrix = matrix(rows=12, cols=12)

Returns: new matrix.

Raises: ValueError when either dimension is negative.

spmv(x, transpose=False)

Multiply the numeric static and dynamic sparse blocks by a device-resident YASPS vector.

Parameter	Type	Default	Description
`x`	compatible YASPS vector	required	Input vector matching `cols`, or `rows` when transposed.
`transpose`	`bool`	`False`	Multiply by the matrix transpose when `True`.

Example

previous_position_contribution = previous_position_jacobian.spmv(auxiliary, transpose=True)

Returns: new compatible YASPS vector with the multiplication result.

Raises: TypeError for an incompatible vector or transpose flag; ValueError when dimensions or sparse block metadata do not match.

The matrix * x and matrix @ x operators are equivalent to matrix.spmv(x).

Property	Type	Description
`rows`	`int`	Matrix row count.
`cols`	`int`	Matrix column count.
`symmetric_storage`	`bool`	Whether stored off-diagonal blocks also represent their transposes.
`block_dimensions`	`list[int]`	Static block dimensions stored as row/column pairs.
`blocks_flattened`	`pycuda.gpuarray.GPUArray`	Flattened static block values.
`blocks_start_indices`	`list[int]`	Value-buffer start for each static block category.
`block_positions`	`pycuda.gpuarray.GPUArray`	Flattened row/column coordinates for static blocks.
`block_counts`	`list[int]`	Static block count per dimension category.
`block_dimensions_dynamic`	`list[int]`	Dynamic block dimensions stored as row/column pairs.
`blocks_flattened_dynamic`	`pycuda.gpuarray.GPUArray`	Flattened dynamic block values.
`blocks_start_indices_dynamic`	`list[int]`	Value-buffer start for each dynamic block category.
`block_positions_dynamic`	`pycuda.gpuarray.GPUArray`	Flattened row/column coordinates for dynamic blocks.
`block_counts_dynamic`	`list[int]`	Dynamic block count per dimension category.