Class CudaSolveRefactorization

CudaSolveRefactorization: The cuSolverRF library was designed to accelerate solution of sets of linear systems by fast re-factorization when given new coefficients in the same sparsity pattern A_i x_i = f_i

Inheritance

System.Object

CudaSolveRefactorization

Implements

System.IDisposable

Inherited Members

System.Object.Equals(System.Object)

System.Object.Equals(System.Object, System.Object)

System.Object.GetHashCode()

System.Object.GetType()

System.Object.MemberwiseClone()

System.Object.ReferenceEquals(System.Object, System.Object)

System.Object.ToString()

Namespace: ManagedCuda.CudaSolve

Assembly: CudaSolve.dll

Syntax

public class CudaSolveRefactorization : IDisposable

Constructors

CudaSolveRefactorization()

Create new refactorization solve instance

Declaration

public CudaSolveRefactorization()

Methods

AccessBundledFactorsDevice(out Int32, out CudaDeviceVariable<Int32>, out CudaDeviceVariable<Int32>, out CudaDeviceVariable<Double>)

This routine allows direct access to the lower L and upper U triangular factors stored in the cuSolverRF library handle. The factors are compressed into a single matrix M=(LI)+ U, where the unitary diagonal of L is not stored. It is assumed that a prior call to the cusolverRfRefactor() was done in order to generate these triangular factors.

Declaration

public void AccessBundledFactorsDevice(out int nnzM, out CudaDeviceVariable<int> Mp, out CudaDeviceVariable<int> Mi, out CudaDeviceVariable<double> Mx)

Parameters

Type	Name	Description
System.Int32	nnzM	the number of non-zero elements of matrix M.
CudaDeviceVariable<System.Int32>	Mp	the array of offsets corresponding to the start of each row in the arrays Mi and Mx. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix $M$. The array size is n+1.
CudaDeviceVariable<System.Int32>	Mi	the array of column indices corresponding to the non-zero elements in the matrix M. It is assumed that this array is sorted by row and by column within each row. The array size is nnzM.
CudaDeviceVariable<System.Double>	Mx	the array of values corresponding to the non-zero elements in the matrix M. It is assumed that this array is sorted by row and by column within each row. The array size is nnzM.

Analyze()

This routine performs the appropriate analysis of parallelism available in the LU refactorization depending upon the algorithm chosen by the user.

Declaration

public void Analyze()

BatchAnalyze()

This routine performs the appropriate analysis of parallelism available in the batched LU re-factorization.

Declaration

public void BatchAnalyze()

BatchRefactor()

This routine performs the LU re-factorization

Declaration

public void BatchRefactor()

BatchResetValues(Int32, Int32, Int32, CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, CudaDeviceVariable<Double>[], CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>)

This routine updates internal data structures with the values of the new coefficient matrix. It is assumed that the arrays csrRowPtrA, csrColIndA, P and Q have not changed since the last call to the cusolverRfbatch_setup_host routine.

Declaration

public void BatchResetValues(int batchSize, int n, int nnzA, CudaDeviceVariable<int> csrRowPtrA, CudaDeviceVariable<int> csrColIndA, CudaDeviceVariable<double>[] csrValA_array, CudaDeviceVariable<int> P, CudaDeviceVariable<int> Q)

Parameters

Type	Name	Description
System.Int32	batchSize	the number of matrices in batched mode.
System.Int32	n	the number of rows (and columns) of matrix A.
System.Int32	nnzA	the number of non-zero elements of matrix A.
CudaDeviceVariable<System.Int32>	csrRowPtrA	the array of offsets corresponding to the start of each row in the arrays csrColIndA and csrValA. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix. The array size is n+1.
CudaDeviceVariable<System.Int32>	csrColIndA	the array of column indices corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
CudaDeviceVariable<System.Double>[]	csrValA_array	array of pointers of size batchSize, each pointer points to the array of values corresponding to the non-zero elements in the matrix.
CudaDeviceVariable<System.Int32>	P	the left permutation (often associated with pivoting). The array size in n.
CudaDeviceVariable<System.Int32>	Q	the right permutation (often associated with reordering). The array size in n.

BatchSetupHost(Int32, Int32, Int32, Int32[], Int32[], Double[][], Int32, Int32[], Int32[], Double[], Int32, Int32[], Int32[], Double[], Int32[], Int32[], cusolverRfHandle)

This routine assembles the internal data structures of the cuSolverRF library for batched operation. It is called after the call to the cusolverRfCreate() routine, and before any other batched routines.

Declaration

public void BatchSetupHost(int batchSize, int n, int nnzA, int[] h_csrRowPtrA, int[] h_csrColIndA, double[][] h_csrValA_array, int nnzL, int[] h_csrRowPtrL, int[] h_csrColIndL, double[] h_csrValL, int nnzU, int[] h_csrRowPtrU, int[] h_csrColIndU, double[] h_csrValU, int[] h_P, int[] h_Q, cusolverRfHandle handle)

Parameters

Type	Name	Description
System.Int32	batchSize	the number of matrices in the batched mode.
System.Int32	n	the number of rows (and columns) of matrix A.
System.Int32	nnzA	the number of non-zero elements of matrix A.
System.Int32[]	h_csrRowPtrA	the array of offsets corresponding to the start of each row in the arrays h_csrColIndA and h_csrValA. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix. The array size is n+1.
System.Int32[]	h_csrColIndA	the array of column indices corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
System.Double[][]	h_csrValA_array	array of pointers of size batchSize, each pointer points to the array of values corresponding to the non-zero elements in the matrix.
System.Int32	nnzL	the number of non-zero elements of matrix L.
System.Int32[]	h_csrRowPtrL	the array of offsets corresponding to the start of each row in the arrays h_csrColIndL and h_csrValL. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix L. The array size is n+1.
System.Int32[]	h_csrColIndL	the array of column indices corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is nnzL.
System.Double[]	h_csrValL	the array of values corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is nnzL.
System.Int32	nnzU	the number of non-zero elements of matrix U.
System.Int32[]	h_csrRowPtrU	the array of offsets corresponding to the start of each row in the arrays h_csrColIndU and h_csrValU. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix U. The array size is n+1.
System.Int32[]	h_csrColIndU	the array of column indices corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is nnzU.
System.Double[]	h_csrValU	the array of values corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is nnzU.
System.Int32[]	h_P	the left permutation (often associated with pivoting). The array size in n.
System.Int32[]	h_Q	the right permutation (often associated with reordering). The array size in n.
cusolverRfHandle	handle	the pointer to the cuSolverRF library handle.

BatchZeroPivot(Int32[])

The user can query which matrix failed LU refactorization by checking corresponding value in position array. The input parameter position is an integer array of size batchSize.

Declaration

public void BatchZeroPivot(int[] position)

Parameters

Type	Name	Description
System.Int32[]	position	integer array of size batchSize. The value of position(j) reports singularity of matrix Aj, -1 if no structural / numerical zero, k >= 0 if Aj(k,k) is either structural zero or numerical zero.

cusolverRfBatchSolve(CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, Int32, Double[], Int32, Double[][], Int32)

To solve A_j * x_j = b_j, first we reform the equation by M_j * Q * x_j = P * b_j. Then do refactorization by cusolverRfBatch_Refactor(). Further cusolverRfBatch_Solve() takes over the remaining steps.

Declaration

public void cusolverRfBatchSolve(CudaDeviceVariable<int> P, CudaDeviceVariable<int> Q, int nrhs, double[] Temp, int ldt, double[][] XF_array, int ldxf)

Parameters

Type	Name	Description
CudaDeviceVariable<System.Int32>	P	the left permutation (often associated with pivoting). The array size in n.
CudaDeviceVariable<System.Int32>	Q	the right permutation (often associated with reordering). The array size in n.
System.Int32	nrhs	the number right-hand-sides to be solved.
System.Double[]	Temp	the dense matrix that contains temporary workspace (of size ldt*nrhs).
System.Int32	ldt	the leading dimension of dense matrix Temp (ldt >= n).
System.Double[][]	XF_array	array of pointers of size batchSize, each pointer points to the dense matrix that contains the right-hand-sides F and solutions X (of size ldxf*nrhs).
System.Int32	ldxf	the leading dimension of dense matrix XF (ldxf >= n).

cusolverRfSetupDevice(Int32, Int32, CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, CudaDeviceVariable<Double>, Int32, CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, CudaDeviceVariable<Double>, Int32, CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, CudaDeviceVariable<Double>, CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>)

This routine assembles the internal data structures of the cuSolverRF library. It is often the first routine to be called after the call to the cusolverRfCreate() routine.

Declaration

public void cusolverRfSetupDevice(int n, int nnzA, CudaDeviceVariable<int> csrRowPtrA, CudaDeviceVariable<int> csrColIndA, CudaDeviceVariable<double> csrValA, int nnzL, CudaDeviceVariable<int> csrRowPtrL, CudaDeviceVariable<int> csrColIndL, CudaDeviceVariable<double> csrValL, int nnzU, CudaDeviceVariable<int> csrRowPtrU, CudaDeviceVariable<int> csrColIndU, CudaDeviceVariable<double> csrValU, CudaDeviceVariable<int> P, CudaDeviceVariable<int> Q)

Parameters

Type	Name	Description
System.Int32	n	the number of rows (and columns) of matrix A.
System.Int32	nnzA	the number of non-zero elements of matrix A.
CudaDeviceVariable<System.Int32>	csrRowPtrA	the array of offsets corresponding to the start of each row in the arrays h_csrColIndA and h_csrValA. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix. The array size is n+1.
CudaDeviceVariable<System.Int32>	csrColIndA	the array of column indices corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
CudaDeviceVariable<System.Double>	csrValA	the array of values corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
System.Int32	nnzL	the number of non-zero elements of matrix L.
CudaDeviceVariable<System.Int32>	csrRowPtrL	the array of offsets corresponding to the start of each row in the arrays h_csrColIndL and h_csrValL. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix L. The array size is n+1.
CudaDeviceVariable<System.Int32>	csrColIndL	the array of column indices corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is nnzL.
CudaDeviceVariable<System.Double>	csrValL	the array of values corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is nnzL.
System.Int32	nnzU	the number of non-zero elements of matrix U.
CudaDeviceVariable<System.Int32>	csrRowPtrU	the array of offsets corresponding to the start of each row in the arrays h_csrColIndU and h_csrValU. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix U. The array size is n+1.
CudaDeviceVariable<System.Int32>	csrColIndU	the array of column indices corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is nnzU.
CudaDeviceVariable<System.Double>	csrValU	the array of values corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is nnzU.
CudaDeviceVariable<System.Int32>	P	the left permutation (often associated with pivoting). The array size in n.
CudaDeviceVariable<System.Int32>	Q	the right permutation (often associated with reordering). The array size in n.

Dispose()

Dispose

Declaration

public void Dispose()

Dispose(Boolean)

For IDisposable

Declaration

protected virtual void Dispose(bool fDisposing)

Parameters

Type	Name	Description
System.Boolean	fDisposing

ExtractBundledFactorsHost(Int32, out Int32, out Int32[], out Int32[], out Double[])

This routine allows direct access to the lower L and upper U triangular factors stored in the cuSolverRF library handle. The factors are compressed into a single matrix M=(LI)+ U, where the unitary diagonal of L is not stored. It is assumed that a prior call to the cusolverRfRefactor() was done in order to generate these triangular factors.

Declaration

public void ExtractBundledFactorsHost(int n, out int h_nnzM, out int[] h_Mp, out int[] h_Mi, out double[] h_Mx)

Parameters

Type	Name	Description
System.Int32	n	Size of Matrix M (n x n)
System.Int32	h_nnzM	the number of non-zero elements of matrix M.
System.Int32[]	h_Mp	the array of offsets corresponding to the start of each row in the arrays Mi and Mx. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix $M$. The array size is n+1.
System.Int32[]	h_Mi	the array of column indices corresponding to the non-zero elements in the matrix M. It is assumed that this array is sorted by row and by column within each row. The array size is nnzM.
System.Double[]	h_Mx	the array of values corresponding to the non-zero elements in the matrix M. It is assumed that this array is sorted by row and by column within each row. The array size is nnzM.

ExtractSplitFactorsHost(Int32, out Int32, out Int32[], out Int32[], out Double[], out Int32, out Int32[], out Int32[], out Double[])

This routine extracts lower (L) and upper (U) triangular factors from the cuSolverRF library handle into the host memory. It is assumed that a prior call to the cusolverRfRefactor() was done in order to generate these triangular factors.

Declaration

public void ExtractSplitFactorsHost(int n, out int h_nnzL, out int[] h_csrRowPtrL, out int[] h_csrColIndL, out double[] h_csrValL, out int h_nnzU, out int[] h_csrRowPtrU, out int[] h_csrColIndU, out double[] h_csrValU)

Parameters

Type	Name	Description
System.Int32	n	Size of Matrix M (n x n)
System.Int32	h_nnzL	the number of non-zero elements of matrix L.
System.Int32[]	h_csrRowPtrL	the array of offsets corresponding to the start of each row in the arrays h_Li and h_Lx. This array has also an extra entry at the end that stores the number of nonzero elements in the matrix L. The array size is n+1.
System.Int32[]	h_csrColIndL	the array of column indices corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is h_nnzL.
System.Double[]	h_csrValL	the array of values corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is h_nnzL.
System.Int32	h_nnzU	the number of non-zero elements of matrix U.
System.Int32[]	h_csrRowPtrU	the array of offsets corresponding to the start of each row in the arrays h_Ui and h_Ux. This array has also an extra entry at the end that stores the number of nonzero elements in the matrix U. The array size is n+1.
System.Int32[]	h_csrColIndU	the array of column indices corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is h_nnzU.
System.Double[]	h_csrValU	the array of values corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is h_nnzU.

Finalize()

For dispose

Declaration

protected void Finalize()

GetAlgs(ref Factorization, ref TriangularSolve)

This routine gets the algorithm used for the refactorization in cusolverRfRefactor() and the triangular solve in cusolverRfSolve(). It may be called once prior to cusolverRfAnalyze() routine.

Declaration

public void GetAlgs(ref Factorization factAlg, ref TriangularSolve solveAlg)

Parameters

Type	Name	Description
Factorization	factAlg	the enumerated algorithm type.
TriangularSolve	solveAlg	the enumerated algorithm type.

GetMatrixFormat(ref MatrixFormat, ref UnitDiagonal)

This routine gets the matrix format used in the cusolverRfSetup(), cusolverRfSetupHost(), cusolverRfResetValues(), cusolverRfExtractBundledFactorsHost() and cusolverRfExtractSplitFactorsHost() routines.

Declaration

public void GetMatrixFormat(ref MatrixFormat format, ref UnitDiagonal diag)

Parameters

Type	Name	Description
MatrixFormat	format	the enumerated matrix format type.
UnitDiagonal	diag	the enumerated unit diagonal type.

GetNumericBoostReport(ref NumericBoostReport)

This routine gets the report whether numeric boosting was used in the cusolverRfRefactor() and cusolverRfSolve() routines.

Declaration

public void GetNumericBoostReport(ref NumericBoostReport report)

Parameters

Type	Name	Description
NumericBoostReport	report	the enumerated boosting report type.

GetNumericProperties(ref Double, ref Double)

This routine gets the numeric values used for checking for "zero" pivot and for boosting it in the cusolverRfRefactor() and cusolverRfSolve() routines. It may be called multiple times prior to cusolverRfRefactor() and cusolverRfSolve() routines. The numeric boosting will be used only if boost > 0.0.

Declaration

public void GetNumericProperties(ref double zero, ref double boost)

Parameters

Type	Name	Description
System.Double	zero	the value below which zero pivot is flagged.
System.Double	boost	the value which is substituted for zero pivot (if the later is flagged).

GetResetValuesFastMode(ref ResetValuesFastMode)

This routine gets the mode used in the cusolverRfResetValues routine.

Declaration

public void GetResetValuesFastMode(ref ResetValuesFastMode fastMode)

Parameters

Type	Name	Description
ResetValuesFastMode	fastMode	the enumerated mode type.

Refactor(cusolverRfHandle)

This routine performs the LU re-factorization

Declaration

public void Refactor(cusolverRfHandle handle)

Parameters

Type	Name	Description
cusolverRfHandle	handle

ResetValues(Int32, Int32, CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, CudaDeviceVariable<Double>, CudaDeviceVariable<Double>, CudaDeviceVariable<Double>)

This routine updates internal data structures with the values of the new coefficient matrix. It is assumed that the arrays csrRowPtrA, csrColIndA, P and Q have not changed since the last call to the cusolverRfSetup[Host] routine. This assumption reflects the fact that the sparsity pattern of coefficient matrices as well as reordering to minimize fill-in and pivoting remain the same in the set of linear systems

Declaration

public void ResetValues(int n, int nnzA, CudaDeviceVariable<int> csrRowPtrA, CudaDeviceVariable<int> csrColIndA, CudaDeviceVariable<double> csrValA, CudaDeviceVariable<double> P, CudaDeviceVariable<double> Q)

Parameters

Type	Name	Description
System.Int32	n	the number of rows (and columns) of matrix A.
System.Int32	nnzA	the number of non-zero elements of matrix A.
CudaDeviceVariable<System.Int32>	csrRowPtrA	the array of offsets corresponding to the start of each row in the arrays csrColIndA and csrValA. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix. The array size is n+1.
CudaDeviceVariable<System.Int32>	csrColIndA	the array of column indices corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
CudaDeviceVariable<System.Double>	csrValA	the array of values corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
CudaDeviceVariable<System.Double>	P	the left permutation (often associated with pivoting). The array size in n.
CudaDeviceVariable<System.Double>	Q	the right permutation (often associated with reordering). The array size in n.

SetAlgs(Factorization, TriangularSolve)

This routine sets the algorithm used for the refactorization in cusolverRfRefactor() and the triangular solve in cusolverRfSolve(). It may be called once prior to cusolverRfAnalyze() routine.

Declaration

public void SetAlgs(Factorization factAlg, TriangularSolve solveAlg)

Parameters

Type	Name	Description
Factorization	factAlg	the enumerated algorithm type.
TriangularSolve	solveAlg	the enumerated algorithm type.

SetMatrixFormat(MatrixFormat, UnitDiagonal)

This routine sets the matrix format used in the cusolverRfSetup(), cusolverRfSetupHost(), cusolverRfResetValues(), cusolverRfExtractBundledFactorsHost() and cusolverRfExtractSplitFactorsHost() routines.

Declaration

public void SetMatrixFormat(MatrixFormat format, UnitDiagonal diag)

Parameters

Type	Name	Description
MatrixFormat	format	the enumerated matrix format type.
UnitDiagonal	diag	the enumerated unit diagonal type.

SetNumericProperties(Double, Double)

This routine sets the numeric values used for checking for "zero" pivot and for boosting it in the cusolverRfRefactor() and cusolverRfSolve() routines. It may be called multiple times prior to cusolverRfRefactor() and cusolverRfSolve() routines. The numeric boosting will be used only if boost > 0.0.

Declaration

public void SetNumericProperties(double zero, double boost)

Parameters

Type	Name	Description
System.Double	zero	the value below which zero pivot is flagged.
System.Double	boost	the value which is substituted for zero pivot (if the later is flagged).

SetResetValuesFastMode(ResetValuesFastMode)

This routine sets the mode used in the cusolverRfResetValues routine.

Declaration

public void SetResetValuesFastMode(ResetValuesFastMode fastMode)

Parameters

Type	Name	Description
ResetValuesFastMode	fastMode	the enumerated mode type.

SetupHost(Int32, Int32, Int32[], Int32[], Double[], Int32, Int32[], Int32[], Double[], Int32, Int32[], Int32[], Double[], Int32[], Int32[])

This routine assembles the internal data structures of the cuSolverRF library. It is often the first routine to be called after the call to the cusolverRfCreate() routine.

Declaration

public void SetupHost(int n, int nnzA, int[] h_csrRowPtrA, int[] h_csrColIndA, double[] h_csrValA, int nnzL, int[] h_csrRowPtrL, int[] h_csrColIndL, double[] h_csrValL, int nnzU, int[] h_csrRowPtrU, int[] h_csrColIndU, double[] h_csrValU, int[] h_P, int[] h_Q)

Parameters

Type	Name	Description
System.Int32	n	the number of rows (and columns) of matrix A.
System.Int32	nnzA	the number of non-zero elements of matrix A.
System.Int32[]	h_csrRowPtrA	the array of offsets corresponding to the start of each row in the arrays h_csrColIndA and h_csrValA. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix. The array size is n+1.
System.Int32[]	h_csrColIndA	the array of column indices corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
System.Double[]	h_csrValA	the array of values corresponding to the non-zero elements in the matrix. It is assumed that this array is sorted by row and by column within each row. The array size is nnzA.
System.Int32	nnzL	the number of non-zero elements of matrix L.
System.Int32[]	h_csrRowPtrL	the array of offsets corresponding to the start of each row in the arrays h_csrColIndL and h_csrValL. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix L. The array size is n+1.
System.Int32[]	h_csrColIndL	the array of column indices corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is nnzL.
System.Double[]	h_csrValL	the array of values corresponding to the non-zero elements in the matrix L. It is assumed that this array is sorted by row and by column within each row. The array size is nnzL.
System.Int32	nnzU	the number of non-zero elements of matrix U.
System.Int32[]	h_csrRowPtrU	the array of offsets corresponding to the start of each row in the arrays h_csrColIndU and h_csrValU. This array has also an extra entry at the end that stores the number of non-zero elements in the matrix U. The array size is n+1.
System.Int32[]	h_csrColIndU	the array of column indices corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is nnzU.
System.Double[]	h_csrValU	the array of values corresponding to the non-zero elements in the matrix U. It is assumed that this array is sorted by row and by column within each row. The array size is nnzU.
System.Int32[]	h_P	the left permutation (often associated with pivoting). The array size in n.
System.Int32[]	h_Q	the right permutation (often associated with reordering). The array size in n.

Solve(CudaDeviceVariable<Int32>, CudaDeviceVariable<Int32>, Int32, Double[], Int32, Double[], Int32)

This routine performs the forward and backward solve with the lower and upper triangular factors resulting from the LU re-factorization

Declaration

public void Solve(CudaDeviceVariable<int> P, CudaDeviceVariable<int> Q, int nrhs, double[] Temp, int ldt, double[] XF, int ldxf)

Parameters

Type	Name	Description
CudaDeviceVariable<System.Int32>	P	the left permutation (often associated with pivoting). The array size in n.
CudaDeviceVariable<System.Int32>	Q	the right permutation (often associated with reordering). The array size in n.
System.Int32	nrhs	the number right-hand-sides to be solved.
System.Double[]	Temp	the dense matrix that contains temporary workspace (of size ldt*nrhs).
System.Int32	ldt	the leading dimension of dense matrix Temp (ldt >= n).
System.Double[]	XF	the dense matrix that contains the righthand-sides F and solutions X (of size ldxf*nrhs).
System.Int32	ldxf	the leading dimension of dense matrix XF (ldxf >= n).

Implements

System.IDisposable