FastLapackInterface separates workspace allocation and actual running for some Lapack algorithms:

• LU factorization
• QR factorization
• Schur factorization

LU factorization and linear problem solution

• ws = LUWs(A) creates storage for a linear problem with matrix representation A.
• LAPACK.getrf!(A, ws) computes the LU factorization using the preallocated workspace and returns the arguments that can be used to construct LinearAlgebra.LU.

A = [1.2 2.3
     6.2 3.3]
ws = LUWs(A)
LinearAlgebra.LU(LAPACK.getrf!(A, ws)...)

QR factorization ( A = Q*R)

• ws = QRWs(A) allocates workspace for QR factorization of a matrix similar to A
• geqrf!(A, ws) computes QR factorization of matrix A and stores it in A and ws.τ, and returns the arguments for the constructor of LinearAlgebra.QR.
• ormqr!(side, A, C, ws) computes Q*C (side='L') or C*Q (side='R')
• ormqr!(side, transpose(A), C, ws) computes transpose(Q)*C (side='L') or C*transpose(Q) (side='R')
• Workspaces exist for QRCompactWY (QRWYWs) and QRPivoted (QRpWs).

A = [1.2 2.3
     6.2 3.3]
ws = QRWs(A)
LinearAlgebra.QR(LAPACK.geqrf!(A, ws)...)

Schur factorization

• SchurWs(A) allocates workspace for the real Schur decomposition of a matrix similar to A.
• gees!(A, ws) computes the Schur decomposition and returns the arguments to the constructor of LinearAlgebra.Schur.
• A workspace for the generalized Schur decomposition also exists.
• It is possible to use select functions with gees! or gges! to order the eigenvalues.

A = [1.2 2.3
     6.2 3.3]
ws = SchurWs(A)
LinearAlgebra.Schur(LAPACK.gees!('V', A, ws)...)

For more info see the Documentation

Package version

• v0.1.3: works only with Julia >= 1.7