Skip to main content

xgetrf

singe_cusolver::dense

Function xgetrf 

Source 
pub fn xgetrf<TA: DataTypeLike>(
    ctx: &Context,
    params: &Params,
    m: usize,
    n: usize,
    a: MatrixMut<'_, TA>,
    pivots: Option<&mut DeviceMemory<i64>>,
    compute_type: DataType,
    workspace: ByteWorkspaceMut<'_>,
    dev_info: &mut DeviceMemory<i32>,
) -> Result<()>
Expand description

Computes the LU factorization of an $m \times n$ matrix

where A is an $m \times n$ matrix, P is a permutation matrix, L is a lower triangular matrix with unit diagonal, and U is an upper triangular matrix.

If LU factorization failed, that is, matrix A (U) is singular, dev_info = i indicates U(i,i) = 0.

If the reported info value is -i, the ith parameter is invalid.

If pivots is None, no pivoting is performed. The factorization is A=L*U, which is not numerically stable.

Whether LU factorization succeeds or fails, pivots contains the pivoting sequence. Row i is interchanged with row pivots[i].

Provide device and host workspace through workspace. Use xgetrf_buffer_size to determine the required sizes for workspace.device and workspace.host.

Callers can combine xgetrf and xgetrs to complete a linear solver.

Currently, xgetrf supports two algorithms. To select the legacy implementation, call Params::set_adv_options.

Algorithms supported by xgetrf

AlgorithmNotes
AlgorithmMode::DefaultFastest algorithm; requires a large workspace of m*n elements.
AlgorithmMode::Algorithm1Legacy implementation.

List of input arguments for xgetrf_buffer_size and xgetrf:

The generic cuSOLVER routine has two data types: data_type_a is the data type of matrix A, and compute_type is the operation’s compute type. xgetrf only supports the following four combinations.

Valid combination of data type and compute type

§Errors

Returns an error if cuSOLVER has not been initialized, if the matrix dimensions or leading dimension are invalid, or if cuSOLVER reports an internal failure.