Class SparseLinearAlgebra

Namespace

NVMathNet.Sparse

Assembly

NVMath.Net.dll

Static helpers for sparse linear algebra operations using cuSPARSE.

public static class SparseLinearAlgebra

Inheritance

object

SparseLinearAlgebra

Inherited Members

object.Equals(object?)

object.Equals(object?, object?)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object?, object?)

object.ToString()

Methods

SddmmAsync<T>(DeviceBuffer<T>, DeviceBuffer<T>, CsrMatrix<T>, long, T?, T?, CudaStream?, CancellationToken)

Sampled dense-dense matrix product: C = alpha · (A · B) ∘ spy(C) + beta · C, where only the entries in the sparsity pattern of C are computed and C's values are updated in place.

public static Task SddmmAsync<T>(DeviceBuffer<T> a, DeviceBuffer<T> b, CsrMatrix<T> c, long k, T? alpha = null, T? beta = null, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

a DeviceBuffer<T>

Dense A (m × k), column-major on device.

b DeviceBuffer<T>

Dense B (k × n), column-major on device.

c CsrMatrix<T>

Sparse CSR matrix C (m × n); its values are updated in place.

k long

Inner dimension.

alpha T?

Scale for the product. Default: 1.

beta T?

Scale for the existing values of C. Default: 0.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpGemmAsync<T>(CsrMatrix<T>, CsrMatrix<T>, T?, CudaStream?, CancellationToken)

Sparse-sparse matrix product: returns a new CSR matrix C = alpha · A · B.

public static Task<CsrMatrix<T>> SpGemmAsync<T>(CsrMatrix<T> a, CsrMatrix<T> b, T? alpha = null, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

a CsrMatrix<T>

Sparse CSR matrix A (m × k).

b CsrMatrix<T>

Sparse CSR matrix B (k × n).

alpha T?

Scale for the product. Default: 1.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task<CsrMatrix<T>>

The product C; the caller owns and must dispose it.

Type Parameters

T

SpMMAsync<T>(CsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, float, float, bool, CudaStream?, CancellationToken)

Sparse matrix-dense matrix product: C = alpha * op(A) * B + beta * C.

public static Task SpMMAsync<T>(CsrMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CsrMatrix<T>

Sparse CSR matrix A (m × k).

b DeviceBuffer<T>

Dense matrix B (k × n), column-major on device.

c DeviceBuffer<T>

Dense output matrix C (m × n), column-major on device; updated in place.

n long

Number of columns in B and C.

alpha float

Scalar multiplier for the sparse-dense product. Default: 1.

beta float

Scalar multiplier for the existing values in C. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMMAsync<T>(CooMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, float, float, bool, CudaStream?, CancellationToken)

Sparse-dense matrix product C = alpha · op(A) · B + beta · C for a COO matrix.

public static Task SpMMAsync<T>(CooMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CooMatrix<T>

Sparse CSR matrix A (m × k).

b DeviceBuffer<T>

Dense matrix B (k × n), column-major on device.

c DeviceBuffer<T>

Dense output matrix C (m × n), column-major on device; updated in place.

n long

Number of columns in B and C.

alpha float

Scalar multiplier for the sparse-dense product. Default: 1.

beta float

Scalar multiplier for the existing values in C. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMMAsync<T>(BsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, float, float, bool, CudaStream?, CancellationToken)

Sparse-dense matrix product C = alpha · op(A) · B + beta · C for a BSR matrix.

public static Task SpMMAsync<T>(BsrMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix BsrMatrix<T>

Sparse CSR matrix A (m × k).

b DeviceBuffer<T>

Dense matrix B (k × n), column-major on device.

c DeviceBuffer<T>

Dense output matrix C (m × n), column-major on device; updated in place.

n long

Number of columns in B and C.

alpha float

Scalar multiplier for the sparse-dense product. Default: 1.

beta float

Scalar multiplier for the existing values in C. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMMAsync<T>(CscMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, float, float, bool, CudaStream?, CancellationToken)

Sparse-dense matrix product C = alpha · op(A) · B + beta · C for a CSC matrix.

public static Task SpMMAsync<T>(CscMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CscMatrix<T>

Sparse CSR matrix A (m × k).

b DeviceBuffer<T>

Dense matrix B (k × n), column-major on device.

c DeviceBuffer<T>

Dense output matrix C (m × n), column-major on device; updated in place.

n long

Number of columns in B and C.

alpha float

Scalar multiplier for the sparse-dense product. Default: 1.

beta float

Scalar multiplier for the existing values in C. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMMAsync<T>(CsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, T, T, bool, CudaStream?, CancellationToken)

C = alpha · op(A) · B + beta · C for a CSR matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMMAsync<T>(CsrMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CsrMatrix<T>

b DeviceBuffer<T>

c DeviceBuffer<T>

n long

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMMAsync<T>(CooMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, T, T, bool, CudaStream?, CancellationToken)

C = alpha · op(A) · B + beta · C for a COO matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMMAsync<T>(CooMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CooMatrix<T>

b DeviceBuffer<T>

c DeviceBuffer<T>

n long

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMMAsync<T>(BsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, T, T, bool, CudaStream?, CancellationToken)

C = alpha · op(A) · B + beta · C for a BSR matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMMAsync<T>(BsrMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix BsrMatrix<T>

b DeviceBuffer<T>

c DeviceBuffer<T>

n long

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMMAsync<T>(CscMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, T, T, bool, CudaStream?, CancellationToken)

C = alpha · op(A) · B + beta · C for a CSC matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMMAsync<T>(CscMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CscMatrix<T>

b DeviceBuffer<T>

c DeviceBuffer<T>

n long

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMVAsync<T>(CsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, float, float, bool, CudaStream?, CancellationToken)

Sparse matrix-vector product: y = alpha * op(A) * x + beta * y.

public static Task SpMVAsync<T>(CsrMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CsrMatrix<T>

Sparse CSR matrix A.

x DeviceBuffer<T>

Dense vector x on device.

y DeviceBuffer<T>

Dense vector y on device; updated in place.

alpha float

Scalar multiplier for the product. Default: 1.

beta float

Scalar multiplier for existing y. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMVAsync<T>(CooMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, float, float, bool, CudaStream?, CancellationToken)

Sparse matrix-vector product y = alpha · op(A) · x + beta · y for a COO matrix.

public static Task SpMVAsync<T>(CooMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CooMatrix<T>

Sparse CSR matrix A.

x DeviceBuffer<T>

Dense vector x on device.

y DeviceBuffer<T>

Dense vector y on device; updated in place.

alpha float

Scalar multiplier for the product. Default: 1.

beta float

Scalar multiplier for existing y. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMVAsync<T>(BsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, float, float, bool, CudaStream?, CancellationToken)

Sparse matrix-vector product y = alpha · op(A) · x + beta · y for a BSR matrix.

public static Task SpMVAsync<T>(BsrMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix BsrMatrix<T>

Sparse CSR matrix A.

x DeviceBuffer<T>

Dense vector x on device.

y DeviceBuffer<T>

Dense vector y on device; updated in place.

alpha float

Scalar multiplier for the product. Default: 1.

beta float

Scalar multiplier for existing y. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMVAsync<T>(CscMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, float, float, bool, CudaStream?, CancellationToken)

Sparse matrix-vector product y = alpha · op(A) · x + beta · y for a CSC matrix.

public static Task SpMVAsync<T>(CscMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, float alpha = 1, float beta = 0, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CscMatrix<T>

Sparse CSR matrix A.

x DeviceBuffer<T>

Dense vector x on device.

y DeviceBuffer<T>

Dense vector y on device; updated in place.

alpha float

Scalar multiplier for the product. Default: 1.

beta float

Scalar multiplier for existing y. Default: 0.

transpose bool

If true, uses the transpose of the sparse matrix.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpMVAsync<T>(CsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, T, T, bool, CudaStream?, CancellationToken)

y = alpha · op(A) · x + beta · y for a CSR matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMVAsync<T>(CsrMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CsrMatrix<T>

x DeviceBuffer<T>

y DeviceBuffer<T>

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMVAsync<T>(CooMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, T, T, bool, CudaStream?, CancellationToken)

y = alpha · op(A) · x + beta · y for a COO matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMVAsync<T>(CooMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CooMatrix<T>

x DeviceBuffer<T>

y DeviceBuffer<T>

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMVAsync<T>(BsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, T, T, bool, CudaStream?, CancellationToken)

y = alpha · op(A) · x + beta · y for a BSR matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMVAsync<T>(BsrMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix BsrMatrix<T>

x DeviceBuffer<T>

y DeviceBuffer<T>

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpMVAsync<T>(CscMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, T, T, bool, CudaStream?, CancellationToken)

y = alpha · op(A) · x + beta · y for a CSC matrix with scalars of the element type, which allows complex scalars for System.Numerics.Complex matrices.

public static Task SpMVAsync<T>(CscMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, T alpha, T beta, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CscMatrix<T>

x DeviceBuffer<T>

y DeviceBuffer<T>

alpha T

beta T

transpose bool

stream CudaStream?

ct CancellationToken

Returns

Task

Type Parameters

T

SpSMAsync<T>(CsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, SparseTriangle, SparseDiagonal, T?, bool, CudaStream?, CancellationToken)

Sparse triangular solve with a dense matrix right-hand side: solves op(A) · C = alpha · B for c.

public static Task SpSMAsync<T>(CsrMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, SparseTriangle triangle = SparseTriangle.Lower, SparseDiagonal diagonal = SparseDiagonal.NonUnit, T? alpha = null, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CsrMatrix<T>

Square sparse CSR matrix A (m × m).

b DeviceBuffer<T>

Dense right-hand side (m × n), column-major on device.

c DeviceBuffer<T>

Dense solution (m × n), column-major on device.

n long

Number of right-hand-side columns.

triangle SparseTriangle

Triangle of A to read.

diagonal SparseDiagonal

Whether the diagonal is implicit ones.

alpha T?

Scale applied to B. Default: 1.

transpose bool

Solve with the transpose of A.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpSMAsync<T>(CooMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, long, SparseTriangle, SparseDiagonal, T?, bool, CudaStream?, CancellationToken)

Triangular solve op(A) · C = alpha · B for a COO matrix.

public static Task SpSMAsync<T>(CooMatrix<T> matrix, DeviceBuffer<T> b, DeviceBuffer<T> c, long n, SparseTriangle triangle = SparseTriangle.Lower, SparseDiagonal diagonal = SparseDiagonal.NonUnit, T? alpha = null, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CooMatrix<T>

Square sparse CSR matrix A (m × m).

b DeviceBuffer<T>

Dense right-hand side (m × n), column-major on device.

c DeviceBuffer<T>

Dense solution (m × n), column-major on device.

n long

Number of right-hand-side columns.

triangle SparseTriangle

Triangle of A to read.

diagonal SparseDiagonal

Whether the diagonal is implicit ones.

alpha T?

Scale applied to B. Default: 1.

transpose bool

Solve with the transpose of A.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpSVAsync<T>(CsrMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, SparseTriangle, SparseDiagonal, T?, bool, CudaStream?, CancellationToken)

Sparse triangular solve with a vector right-hand side: solves op(A) · y = alpha · x for y.

public static Task SpSVAsync<T>(CsrMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, SparseTriangle triangle = SparseTriangle.Lower, SparseDiagonal diagonal = SparseDiagonal.NonUnit, T? alpha = null, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CsrMatrix<T>

Square sparse CSR matrix A whose triangle is referenced.

x DeviceBuffer<T>

Right-hand side vector (device).

y DeviceBuffer<T>

Solution vector (device); must not alias x.

triangle SparseTriangle

Triangle of A to read.

diagonal SparseDiagonal

Whether the diagonal is implicit ones.

alpha T?

Scale applied to x. Default: 1.

transpose bool

Solve with the transpose of A.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T

SpSVAsync<T>(CooMatrix<T>, DeviceBuffer<T>, DeviceBuffer<T>, SparseTriangle, SparseDiagonal, T?, bool, CudaStream?, CancellationToken)

Triangular solve op(A) · y = alpha · x for a COO matrix (row-major sorted entries).

public static Task SpSVAsync<T>(CooMatrix<T> matrix, DeviceBuffer<T> x, DeviceBuffer<T> y, SparseTriangle triangle = SparseTriangle.Lower, SparseDiagonal diagonal = SparseDiagonal.NonUnit, T? alpha = null, bool transpose = false, CudaStream? stream = null, CancellationToken ct = default) where T : unmanaged, INumberBase<T>

Parameters

matrix CooMatrix<T>

Square sparse CSR matrix A whose triangle is referenced.

x DeviceBuffer<T>

Right-hand side vector (device).

y DeviceBuffer<T>

Solution vector (device); must not alias x.

triangle SparseTriangle

Triangle of A to read.

diagonal SparseDiagonal

Whether the diagonal is implicit ones.

alpha T?

Scale applied to x. Default: 1.

transpose bool

Solve with the transpose of A.

stream CudaStream?

Optional CUDA stream. null creates a private stream.

ct CancellationToken

Cancellation token.

Returns

Task

Type Parameters

T