Core Scientific Library Routines

The Sun S3L core routines consist of:

• Dense matrix operations

  • 2-Norm - Compute the global 2-norm of a parallel array.

  • Inner product - Compute the global inner product over all axes of two source parallel arrays. The inner product is added to the destination. A routine that takes the conjugate of the second operand is provided for complex data.

  • Outer product - Compute one or more instances of an outer product of two vectors. The result is added to the destination. For complex data, a routine that takes the conjugate of the second operand is provided.

  • Matrix-vector multiplication - Compute one or more instances of a matrix-vector product. The result is added to the destination, or is added to a second parallel array. For complex data, a routine that takes the conjugate of the matrix is provided.

  • Matrix multiplication - Compute one or more matrix products. Each routine add the result to the destination. Routines that take the transpose of either or both operand matrices (or, for complex data, the Hermitian of either matrix) are provided.

• LU-factorization and LU-solve routines

  • LU-factorization routine - For each m x n coefficient matrix A of a, computes LU factorization using partial pivoting with row interchanges.

  • LU-solve routine - Uses the L and U factors produced by the LU-factorization routine to produce solutions to the system AX=B. B may represent one or more right-hand sides for each instance of the systems of equations.

  • LU-invert routine - For each m x m (square) instance of matrix A, computes the inverse of A using the LU-factorization results of the S3L_lu_factor routine.

• Parallel 1D, 2D, and 3D FFTs

  • Setup and deallocation of FFT handles - Initialize and deallocate FFT handles for both complex and real data types. Separate routines are used for the two data types.

  • Simple complex-to-complex, mixed-radix, forward and inverse FFT routines - Performs forward or inverse Fast Fourier Transform of a parallel array of type complex or double complex. Supports both power-of-two and arbitrary radix parameters.

  • Detailed complex-to-complex FFT routine - Allows independent specification along each data axis of the transform direction in a complex-to-complex FFT. Can improve performance over the simple FFT in some cases.

  • Simple real-to-complex and complex-to-real FFT routines - Perform the forward (real-to-complex) and inverse (complex-to-real) FFT operations on 1-, 2-, or 3-dimensional arrays.

• Structured solver

  • Tridiagonal solver - Solves collections of tridiagonal linear systems of equations using Gaussian elimination with pivoting.

  • Banded solver - Solves collections of banded linear systems of equations using Gaussian elimination with pivoting.

• Dense symmetric eigenvalue solver - Computes selected eigenvalues and, optionally, engenvectors of hermitian matrices.

• Dense Singular Value Decomposition (SVD) - Computes the singular value decomposition of an M x N matrix and, optionally, the left and right singular vectors.

• Sparse routines

  • Declare array handle for a sparse matrix.

  • Read data from a file into a distributed matrix, with support for both COO and CSR sparse storage formats.

  • Compute the product of a sparse matrix with a dense vector.

• Iterative solver - Solves a general sparse linear system of equations using iterative methods, with or without preconditioning.

• Convolution/Deconvolution

  • Convolve - Computes 1D or 2D convolution of one array with another.

  • Deconvolve - Deconvolves an array into a vector.

• Iterative eigensolver - Computes selected eigenpairs of dense or sparse matrices, with optional specification of eigenpair properties.

• Autocorrelation - Computes 1D or 2D autocorrelation of a signal.

• Sort and grade - Sort and grade arrays.

• Parallel random number generators

  • Fibonacci RNG setup and deallocation - Initializes and deallocates the state table of a lagged Fibonacci random number generator (LFG).

  • Fibonacci RNG - Uses an LFG to initialize a parallel array.

  • LCG RNG setup - Defines the parameters used in the Sun S3L linear congruential random number generator (LCG).

  • LCG RNG - Uses a parallel LCG to produce random numbers that are independent of the array distribution.

• Parallel sort - Sorts a 1D parallel array.

• Parallel transpose - Performs a generalized transposition of a parallel array.

• Copy array routine - Copies the elements of one array onto another.

• Zero array elements - Replaces all elements in an array with zero.