nBLAS+

Node >=4.0 C++ bindings for all single- and double-precision CBLAS (Basic Linear Algebra Subprograms) and SPBLAS (Sparse BLAS) routines. Also LAPACK routines (for now only solve general system of linear equations AX=B).

Preinstall

LINUX:

$ cd preinstall
$ sh debian.sh

Tested on Ubuntu 16

OSX:

$ cd preinstall
$ sh osx.sh

Tested on OSX ElCapitan

WIN:

Not tested.

https://icl.cs.utk.edu/lapack-for-windows/

Install

$ npm install nblas-plus
$ npm test

API

• BLAS Level 1 Routines and Functions

  • ?asum (x)
  • ?axpy (x, y, [alpha = 1.0])
  • ?copy (x, y)
  • ?dot (x, y)
  • ?nrm2 (x)
  • ?rot (x, y, c, s)
  • ?rotg (x, y, c, s)
  • ?rotm (x, y, param)
  • ?rotmg (d1, d2, x1, y1, param)
  • ?scal (x, alpha)
  • ?swap (x, y)
  • i?amax (x)

• BLAS Level 2 Routines

  • ?gbmv (a, x, y, [kl = 0], [ku = 0], [alpha = 1.0], [beta = 0], [trans = nblas.NoTrans])
  • ?gemv (a, x, y, [alpha = 1.0], [beta = 0], [trans = nblas.NoTrans])
  • ?ger (a, x, y, [alpha = 1.0])
  • ?sbmv (a, x, y, [uplo = nblas.Upper], [alpha = 1.0], [beta = 0])
  • ?spmv (ap, x, y, [uplo = nblas.Upper], [alpha = 1.0], [beta = 0])
  • ?spr (ap, x, [uplo = nblas.Upper], [alpha = 1.0])
  • ?spr2 (ap, x, y, [uplo = nblas.Upper], [alpha = 1.0])
  • ?symv (a, x, y, [uplo = nblas.Upper], [alpha = 1.0], [beta = 0])
  • ?syr (a, x, [uplo = nblas.Upper], [alpha = 1.0])
  • ?syr2 (a, x, y, [uplo = nblas.Upper], [alpha = 1.0], [beta = 0])
  • ?tbmv (a, x, y, [uplo = nblas.Upper], [trans = nblas.NoTrans], [diag = nblas.NonUnit])
  • ?tbsv (a, x, [uplo = nblas.Upper], [diag = nblas.NonUnit])
  • ?tpmv (ap, x, [uplo = nblas.Upper], [trans = nblas.NoTrans], [diag = nblas.NonUnit])
  • ?tpsv (ap, x, [uplo = nblas.Upper], [trans = nblas.NoTrans], [diag = nblas.NonUnit])
  • ?trmv (a, x, [uplo = nblas.Upper], [trans = nblas.NoTrans], [diag = nblas.NonUnit])
  • ?trsv (a, x, [uplo = nblas.Upper], [trans = nblas.NoTrans], [diag = nblas.NonUnit])

• BLAS Level 3 Routines

  • ?gemm (a, b, c, m, n, k, [transa = 111], [transb = 111], [alpha = 1.0], [beta = 0])
  • ?symm (a, b, c, m, n, [side = nblas.Left], [uplo = nblas.Upper], [alpha = 1.0], [beta = 0])
  • ?syrk (a, c, n, k, [uplo = nblas.Upper], [trans = nblas.NoTrans], [alpha = 1.0], [beta = 0])
  • ?syr2k (a, b, c, n, k, [uplo = nblas.Upper], [trans = nblas.NoTrans], [alpha = 1.0], [beta = 0])
  • ?trmm (a, b, m, n, [side = nblas.Left], [uplo = nblas.Upper], [transa = 111], [diag = nblas.NonUnit], [alpha = 1.0])
  • ?trsm (a, b, m, n, [diag = nblas.NonUnit], [uplo = nblas.Upper], [transa = 111], [diag = nblas.NonUnit], [alpha = 1.0])

• LAPACK Routines (list of routines)

  • [?gesv (A, B, m, n, [ipiv])] (https://software.intel.com/ru-ru/node/468876)
  • [?gesvx (A, B, X, m, n, [AF], [IPIV], [fact = nblas.Lapack.Fact.N], [trans = nblas.Lapack.Trans.N], [equed = nblas.Lapack.Equed.N], [R], [C])] (https://software.intel.com/ru-ru/node/468878)
  • [?getrf (A, IPIV, m, n)] (https://software.intel.com/ru-ru/node/468682)
  • others ... todo?

• SPBLAS (doc1, doc2)

  • SPBLAS Creation Routines
    • Construction
      • ?uscr_begin (double, m, n) Construction
      • ?uscr_block_begin (double, Mb, Nb, k, l)] Block construction (Mb, Nb - blocks count, k, l = blocks size, M = Mb * k, N = Nb * l)
      • ?uscr_variable_block_begin (double, Mb, Nb, K, L) Variable block construction (K - array of size Mb, L - array of size Nb)
    • Insertion
      • ?uscr_insert_entry (A, val, i, j)
      • ?uscr_insert_entries (A, nz, vals, indx, jndx)
      • ?uscr_insert_col (A, j, nz, vals, indx)
      • ?uscr_insert_row (A, i, nz, vals, jndx)
      • ?uscr_insert_clique (A, k, l, vals, row_stride, col_stride, indx, jndx)
      • ?uscr_insert_block (A, vals, row_stride, col_stride, i, j)
    • Completion of Construction Routines
      • ?uscr_end (A)
    • Matrix Property Routines
      • usgp (A) get, see list of options below (Sparse matrix properties)
      • ussp (A) set
    • Destruction Routine
      • usds (A)
  • SPBLAS Level 1 Routines
    • ?usdot (x, indx, y) sparse dot product
    • ?usaxpy (x, indx, y, alpha) sparse vector update
    • ?usga (x, indx, y) sparse gather
    • ?usgz (x, indx, y) sparse gather and zero
    • ?ussc (x, indx, y) sparse scatter
  • SPBLAS Level 2 Routines
    • ?usmv (A, x, y, trans = nblas.Trans, alpha = 1.0) sparse matrix-vector multiply, A [m * n] * x [ n * 1 ] = y [ m * 1 ]
    • ?ussv (A, x, trans = nblas.Trans, alpha = 1.0) sparse triangular solve
  • SPBLAS Level 3 Routines
    • ?usmm (A, B, C, nrhs, trans = nblas.Trans, alpha = 1.0) sparse matrix-matrix multiply
    • ?ussm (A, B, nrhs, trans = nblas.Trans, alpha = 1.0) sparse triangular solve

• Matrix layout enums

  • Matrix transpose (trans)

    • nblas.NoTrans (default)
    • nblas.Trans
    • nblas.ConjTrans

  • Upper/lower matrix (uplo)

    • nblas.Upper (default)
    • nblas.Lower

  • Matrix diagonal (diag)

    • nblas.NonUnit (default)
    • nblas.Unit

  • Matrix side (side)

    • nblas.Left (default)
    • nblas.Right

  • Sparse matrix properties

    • nblas.SymmetryType.
      • .blas_general
      • .blas_symmetric
      • .blas_hermitian
      • .blas_triangular
      • .blas_lower_triangular
      • .blas_upper_triangular
      • .blas_lower_symmetric
      • .blas_upper_symmetric
      • .blas_lower_hermitian
      • .blas_upper_hermitian
    • nblas.FieldType.
      • .blas_double_precision
      • .blas_single_precision
    • nblas.SizeType.
      • .blas_num_rows
      • .blas_num_cols
      • .blas_num_nonzeros
    • nblas.HandleType.
      • .blas_invalid_handle
      • .blas_new_handle
      • .blas_open_handle
      • .blas_valid_handle

  • For LAPACK

    • nblas.Lapack.Fact.
      • .N (default)
      • .F
      • .E
    • nblas.Lapack.Trans.
      • .N (default)
      • .T
      • .C
    • nblas.Lapack.Equed.
      • .N (default)
      • .R
      • .C
      • .B

Double precision functions expect Float64Array vectors, single precision functions expect Float32Array vectors.

Examples

See test/test.js

var nblas = require('nblas');

var f64a = new Float64Array([1, 2, 3]),
    f64b = new Float64Array([4, 5, 6]);

nblas.dot(f64a, f64b); // or
nblas.ddot(3, f64a, 1, f64b, 1); // 32

var f32a = new Float32Array([1, 2, 3]),
    f32b = new Float32Array([4, 5, 6]);

nblas.dot(f32a, f32b); // or
nblas.sdot(3, f32a, 1, f32b, 1); // 32