Matrix Storage#

The oneMKL BLAS and LAPACK routines for DPC++ use several matrix and vector storage formats. These are the same formats used in traditional Fortran BLAS/LAPACK. LAPACK routines require column major layout.

General Matrix

A general matrix A of m rows and n columns with leading dimension lda is represented as a one dimensional array a of size of at least lda * n if column major layout is used and at least lda * m if row major layout is used. Before entry in any BLAS function using a general matrix, the leading m by n part of the array a must contain the matrix A. For column (respectively row) major layout, the elements of each column (respectively row) are contiguous in memory while the elements of each row (respectively column) are at distance lda from the element in the same row (respectively column) and the previous column (respectively row).

Visually, the matrix

\[\begin{split}A = \begin{bmatrix} A_{11} & A_{12} & A_{13} & \ldots & A_{1n}\\ A_{21} & A_{22} & A_{23} & \ldots & A_{2n}\\ A_{31} & A_{32} & A_{33} & \ldots & A_{3n}\\ \vdots & \vdots & \vdots & \ddots & \vdots\\ A_{m1} & A_{m2} & A_{m3} & \ldots & A_{mn} \end{bmatrix}\end{split}\]

is stored in memory as an array

  • For column major layout,

\[\scriptstyle a = [\underbrace{\underbrace{A_{11},A_{21},A_{31},...,A_{m1},*,...,*}_\text{lda}, \underbrace{A_{12},A_{22},A_{32},...,A_{m2},*,...,*}_\text{lda}, ..., \underbrace{A_{1n},A_{2n},A_{3n},...,A_{mn},*,...,*}_\text{lda}} _\text{lda x n}]\]
  • For row major layout,

\[\scriptstyle a = [\underbrace{\underbrace{A_{11},A_{12},A_{13},...,A_{1n},*,...,*}_\text{lda}, \underbrace{A_{21},A_{22},A_{23},...,A_{2n},*,...,*}_\text{lda}, ..., \underbrace{A_{m1},A_{m2},A_{m3},...,A_{mn},*,...,*}_\text{lda}} _\text{m x lda}]\]

Triangular Matrix

A triangular matrix A of n rows and n columns with leading dimension lda is represented as a one dimensional array a, of a size of at least lda * n. When column (respectively row) major layout is used, the elements of each column (respectively row) are contiguous in memory while the elements of each row (respectively column) are at distance lda from the element in the same row (respectively column) and the previous column (respectively row).

Before entry in any BLAS function using a triangular matrix,

  • If upper_lower = uplo::upper, the leading n by n upper triangular part of the array a must contain the upper triangular part of the matrix A. The strictly lower triangular part of the array a is not referenced. In other words, the matrix

    \[\begin{split}A = \begin{bmatrix} A_{11} & A_{12} & A_{13} & \ldots & A_{1n}\\ * & A_{22} & A_{23} & \ldots & A_{2n}\\ * & * & A_{33} & \ldots & A_{3n}\\ \vdots & \vdots & \vdots & \ddots & \vdots\\ * & * & * & \ldots & A_{nn} \end{bmatrix}\end{split}\]

    is stored in memory as the array

    • For column major layout,

    \[\scriptstyle a = [\underbrace{\underbrace{A_{11},*,...,*}_\text{lda}, \underbrace{A_{12},A_{22},*,...,*}_\text{lda}, ..., \underbrace{A_{1n},A_{2n},A_{3n},...,A_{nn},*,...,*}_\text{lda}} _\text{lda x n}]\]
    • For row major layout,

    \[\scriptstyle a = [\underbrace{\underbrace{A_{11},A_{12},A_{13},...,A_{1n},*,...,*}_\text{lda}, \underbrace{*,A_{22},A_{23},...,A_{2n},*,...,*}_\text{lda}, ..., \underbrace{*,...,*,A_{nn},*,...,*}_\text{lda}} _\text{lda x n}]\]
  • If upper_lower = uplo::lower, the leading n by n lower triangular part of the array a must contain the lower triangular part of the matrix A. The strictly upper triangular part of the array a is not referenced. That is, the matrix

    \[\begin{split}A = \begin{bmatrix} A_{11} & * & * & \ldots & * \\ A_{21} & A_{22} & * & \ldots & * \\ A_{31} & A_{32} & A_{33} & \ldots & * \\ \vdots & \vdots & \vdots & \ddots & \vdots\\ A_{n1} & A_{n2} & A_{n3} & \ldots & A_{nn} \end{bmatrix}\end{split}\]

    is stored in memory as the array

    • For column major layout,

    \[\scriptstyle a = [\underbrace{\underbrace{A_{11},A_{21},A_{31},..,A_{n1},*,...,*}_\text{lda}, \underbrace{*,A_{22},A_{32},...,A_{n2},*,...,*}_\text{lda}, ..., \underbrace{*,...,*,A_{nn},*,...,*}_\text{lda}} _\text{lda x n}]\]
    • For row major layout,

    \[\scriptstyle a = [\underbrace{\underbrace{A_{11},*,...,*}_\text{lda}, \underbrace{A_{21},A_{22},*,...,*}_\text{lda}, ..., \underbrace{A_{n1},A_{n2},A_{n3},...,A_{nn},*,...,*}_\text{lda}} _\text{lda x n}]\]

Band Matrix

A general band matrix A of m rows and n columns with kl sub-diagonals, ku super-diagonals, and leading dimension lda is represented as a one dimensional array a of a size of at least lda * n (respectively lda * m) if column (respectively row) major layout is used.

Before entry in any BLAS function using a general band matrix, the leading (kl + ku + 1) by n (respectively m) part of the array a must contain the matrix A. This matrix must be supplied column-by-column (respectively row-by-row), with the main diagonal of the matrix in row ku (respectively kl) of the array (0-based indexing), the first super-diagonal starting at position 1 (respectively 0) in row (ku - 1) (respectively column (kl + 1)), the first sub-diagonal starting at position 0 (respectively 1) in row (ku + 1) (respectively column (kl - 1)), and so on. Elements in the array a that do not correspond to elements in the band matrix (such as the top left ku by ku triangle) are not referenced.

Visually, the matrix A

\[\begin{split}A = \left[\begin{smallmatrix} A_{11} & A_{12} & A_{13} & \ldots & A_{1,ku+1} & * & \ldots & \ldots & \ldots & \ldots & \ldots & * \\ A_{21} & A_{22} & A_{23} & A_{24} & \ldots & A_{2,ku+2} & * & \ldots & \ldots & \ldots & \ldots & * \\ A_{31} & A_{32} & A_{33} & A_{34} & A_{35} & \ldots & A_{3,ku+3} & * & \ldots & \ldots & \ldots & * \\ \vdots & A_{42} & A_{43} & \ddots & \ddots & \ddots & \ddots & \ddots & * & \ldots & \ldots & \vdots \\ A_{kl+1,1} & \vdots & A_{53} & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & * & \ldots & \vdots \\ * & A_{kl+2,2} & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \vdots \\ \vdots & * & A_{kl+3,3} & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & * \\ \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & A_{n-ku,n}\\ \vdots & \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \vdots \\ \vdots & \vdots & \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & A_{m-2,n} \\ \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & A_{m-1,n} \\ * & * & * & \ldots & \ldots & \ldots & * & A_{m,m-kl} & \ldots & A_{m,n-2} & A_{m,n-1} & A_{m,n} \end{smallmatrix}\right]\end{split}\]

is stored in memory as an array

  • For column major layout,

\[\scriptscriptstyle a = [\underbrace{ \underbrace{\underbrace{*,...,*}_\text{ku},A_{11}, A_{12},...,A_{min(kl+1,m),1},*,...,*}_\text{lda}, \underbrace{\underbrace{*,...,*}_\text{ku-1},A_{max(1,2-ku),2},...,A_{min(kl+2,m),2},*,...*}_\text{lda}, ..., \underbrace{\underbrace{*,...,*}_\text{max(0,ku-n+1)},A_{max(1,n-ku),n},...,A_{min(kl+n,m),n},*,...*}_\text{lda} }_\text{lda x n}]\]
  • For row major layout,

\[\scriptscriptstyle a = [\underbrace{ \underbrace{\underbrace{*,...,*}_\text{kl},A_{11}, A_{12},...,A_{1,min(ku+1,n)},*,...,*}_\text{lda}, \underbrace{\underbrace{*,...,*}_\text{kl-1},A_{2,max(1,2-kl)},...,A_{2,min(ku+2,n)},*,...*}_\text{lda}, ..., \underbrace{\underbrace{*,...,*}_\text{max(0,kl-m+1)},A_{m,max(1,m-kl)},...,A_{m,min(ku+m,n)},*,...*}_\text{lda} }_\text{lda x m}]\]

The following program segment transfers a band matrix from conventional full matrix storage (variable matrix, with leading dimension ldm) to band storage (variable a, with leading dimension lda):

  • Using matrices stored with column major layout,

for (j = 0; j < n; j++) {
    k = ku – j;
    for (i = max(0, j – ku); i < min(m, j + kl + 1); i++) {
        a[(k + i) + j * lda] = matrix[i + j * ldm];
    }
}
  • Using matrices stored with row major layout,

for (i = 0; i < m; i++) {
    k = kl – i;
    for (j = max(0, i – kl); j < min(n, i + ku + 1); j++) {
        a[(k + j) + i * lda] = matrix[j + i * ldm];
    }
}

Triangular Band Matrix

A triangular band matrix A of n rows and n columns with k sub/super-diagonals and leading dimension lda is represented as a one dimensional array a of size at least lda * n.

Before entry in any BLAS function using a triangular band matrix,

  • If upper_lower = uplo::upper, the leading (k + 1) by n part of the array a must contain the upper triangular band part of the matrix A. When using column major layout, this matrix must be supplied column-by-column (respectively row-by-row) with the main diagonal of the matrix in row (k) (respectively column 0) of the array, the first super-diagonal starting at position 1 (respectively 0) in row (k - 1) (respectively column 1), and so on. Elements in the array a that do not correspond to elements in the triangular band matrix (such as the top left k by k triangle) are not referenced.

    Visually, the matrix

    \[\begin{split}A = \left[\begin{smallmatrix} A_{11} & A_{12} & A_{13} & \ldots & A_{1,k+1} & * & \ldots & \ldots & \ldots & \ldots & \ldots & * \\ * & A_{22} & A_{23} & A_{24} & \ldots & A_{2,k+2} & * & \ldots & \ldots & \ldots & \ldots & * \\ \vdots & * & A_{33} & A_{34} & A_{35} & \ldots & A_{3,k+3} & * & \ldots & \ldots & \ldots & * \\ \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & * & \ldots & \ldots & \vdots \\ \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & * & \ldots & \vdots \\ \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \vdots \\ \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & * \\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & A_{n-k,n}\\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \vdots \\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & A_{n-2,n} \\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & A_{n-1,n} \\ * & * & * & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & * & A_{n,n} \end{smallmatrix}\right]\end{split}\]

    is stored as an array

  • For column major layout,

    \[\scriptstyle a = [\underbrace{ \underbrace{\underbrace{*,...,*}_\text{ku},A_{11},*,...,*}_\text{lda}, \underbrace{\underbrace{*,...,*}_\text{ku-1},A_{max(1,2-k),2},...,A_{2,2},*,...*}_\text{lda}, ..., \underbrace{\underbrace{*,...,*}_\text{max(0,k-n+1)},A_{max(1,n-k),n},...,A_{n,n},*,...*}_\text{lda} }_\text{lda x n}]\]
  • For row major layout,

    \[\scriptstyle a = [\underbrace{ \underbrace{A_{11},A_{21},...,A_{min(k+1,n),1},*,...,*}_\text{lda}, \underbrace{A_{2,2},...,A_{min(k+2,n),2},*,...,*}_\text{lda}, ..., \underbrace{A_{n,n},*,...*}_\text{lda} }_\text{lda x n}]\]

The following program segment transfers a band matrix from conventional full matrix storage (variable matrix, with leading dimension ldm) to band storage (variable a, with leading dimension lda):

  • Using matrices stored with column major layout,

for (j = 0; j < n; j++) {
    m = k – j;
    for (i = max(0, j – k); i <= j; i++) {
        a[(m + i) + j * lda] = matrix[i + j * ldm];
    }
}
  • Using matrices stored with row major layout,

for (i = 0; i < n; i++) {
    m = –i;
    for (j = i; j < min(n, i + k + 1); j++) {
        a[(m + j) + i * lda] = matrix[j + i * ldm];
    }
}
  • If upper_lower = uplo::lower, the leading (k + 1) by n part of the array a must contain the upper triangular band part of the matrix A. This matrix must be supplied column-by-column with the main diagonal of the matrix in row 0 of the array, the first sub-diagonal starting at position 0 in row 1, and so on. Elements in the array a that do not correspond to elements in the triangular band matrix (such as the bottom right k by k triangle) are not referenced.

    That is, the matrix

    \[\begin{split}A = \left[\begin{smallmatrix} A_{11} & * & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & * \\ A_{21} & A_{22} & * & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & * \\ A_{31} & A_{32} & A_{33} & * & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & * \\ \vdots & A_{42} & A_{43} & \ddots & \ddots & \ldots & \ldots & \ldots & \ldots & \ldots & \ldots & \vdots \\ A_{k+1,1} & \vdots & A_{53} & \ddots & \ddots & \ddots & \ldots & \ldots & \ldots & \ldots & \ldots & \vdots \\ * & A_{k+2,2} & \vdots & \ddots & \ddots & \ddots & \ddots & \ldots & \ldots & \ldots & \ldots & \vdots \\ \vdots & * & A_{k+3,3} & \ddots & \ddots & \ddots & \ddots & \ddots & \ldots & \ldots & \ldots & \vdots \\ \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ldots & \ldots & \vdots \\ \vdots & \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ldots & \vdots \\ \vdots & \vdots & \vdots & \vdots & * & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \vdots \\ \vdots & \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & * \\ * & * & * & \ldots & \ldots & \ldots & * & A_{n,n-k} & \ldots & A_{n,n-2} & A_{n,n-1} & A_{n,n} \end{smallmatrix}\right]\end{split}\]

    is stored as the array

  • For column major layout,

    \[\scriptstyle a = [\underbrace{ \underbrace{A_{11},A_{21},...,A_{min(k+1,n),1},*,...,*}_\text{lda}, \underbrace{A_{2,2},...,A_{min(k+2,n),2},*,...,*}_\text{lda}, ..., \underbrace{A_{n,n},*,...*}_\text{lda} }_\text{lda x n}]\]
  • For row major layout,

    \[\scriptstyle a = [\underbrace{ \underbrace{\underbrace{*,...,*}_\text{k},A_{11},*,...,*}_\text{lda}, \underbrace{\underbrace{*,...,*}_\text{k-1},A_{max(1,2-k),2},...,A_{2,2},*,...*}_\text{lda}, ..., \underbrace{\underbrace{*,...,*}_\text{max(0,k-n+1)},A_{max(1,n-k),n},...,A_{n,n},*,...*}_\text{lda} }_\text{lda x n}]\]

The following program segment transfers a band matrix from conventional full matrix storage (variable matrix, with leading dimension ldm) to band storage (variable a, with leading dimension lda):

  • Using matrices stored with column major layout,

for (j = 0; j < n; j++) {
    m = –j;
    for (i = j; i < min(n, j + k + 1); i++) {
        a[(m + i) + j * lda] = matrix[i + j * ldm];
    }
}
  • Using matrices stored with row major layout,

for (i = 0; i < n; i++) {
    m = k – i;
    for (j = max(0, i – k); j <= i; j++) {
        a[(m + j) + i * lda] = matrix[j + i * ldm];
    }
}

Packed Triangular Matrix

A triangular matrix A of n rows and n columns is represented in packed format as a one dimensional array a of size at least (n*(n + 1))/2. All elements in the upper or lower part of the matrix A are stored contiguously in the array a.

Before entry in any BLAS function using a triangular packed matrix,

  • If upper_lower = uplo::upper, if column (respectively row) major layout is used, the first (n*(n + 1))/2 elements in the array a must contain the upper triangular part of the matrix A packed sequentially, column by column (respectively row by row) so that a[0] contains A11, a[1] and a[2] contain A12 and A22 (respectively A13) respectively, and so on. Hence, the matrix

    \[\begin{split}A = \begin{bmatrix} A_{11} & A_{12} & A_{13} & \ldots & A_{1n}\\ * & A_{22} & A_{23} & \ldots & A_{2n}\\ * & * & A_{33} & \ldots & A_{3n}\\ \vdots & \vdots & \vdots & \ddots & \vdots\\ * & * & * & \ldots & A_{nn} \end{bmatrix}\end{split}\]

    is stored as the array

    • For column major layout,

      \[\scriptstyle a = [A_{11},A_{12},A_{22},A_{13},A_{23},A_{33},...,A_{(n-1),n},A_{nn}]\]
    • For row major layout,

      \[\scriptstyle a = [A_{11},A_{12},A_{13},...,A_{1n}, A_{22},A_{23},...,A_{2n},..., A_{(n-1),(n-1)},A_{(n-1),n},A_{nn}]\]
  • If upper_lower = uplo::lower, if column (respectively row) major layout is used, the first (n*(n + 1))/2 elements in the array a must contain the lower triangular part of the matrix A packed sequentially, column by column (row by row) so that a[0] contains A11, a[1] and a[2] contain A21 and A31 (respectively A22) respectively, and so on. The matrix

    \[\begin{split}A = \begin{bmatrix} A_{11} & * & * & \ldots & * \\ A_{21} & A_{22} & * & \ldots & * \\ A_{31} & A_{32} & A_{33} & \ldots & * \\ \vdots & \vdots & \vdots & \ddots & \vdots\\ A_{n1} & A_{n2} & A_{n3} & \ldots & A_{nn} \end{bmatrix}\end{split}\]

    is stored as the array

    • For column major layout,

    \[\scriptstyle a = [A_{11},A_{21},A_{31},...,A_{n1}, A_{22},A_{32},...,A_{n2},..., A_{(n-1),(n-1)},A_{n,(n-1)},A_{nn}]\]
    • For row major layout,

    \[\scriptstyle a = [A_{11},A_{21},A_{22},A_{31},A_{32},A_{33},...,A_{n,(n-1)},A_{nn}]\]

Vector

A vector X of n elements with increment incx is represented as a one dimensional array x of size at least (1 + (n - 1) * abs(incx)).

Visually, the vector

\[X = (X_{1},X_{2}, X_{3},...,X_{n})\]

is stored in memory as an array

\[\scriptstyle x = [\underbrace{ \underbrace{X_{1},*,...,*}_\text{incx}, \underbrace{X_{2},*,...,*}_\text{incx}, ..., \underbrace{X_{n-1},*,...,*}_\text{incx},X_{n} }_\text{1 + (n-1) x incx}] \quad if \:incx \:> \:0\]
\[\scriptstyle x = [\underbrace{ \underbrace{X_{n},*,...,*}_\text{|incx|}, \underbrace{X_{n-1},*,...,*}_\text{|incx|}, ..., \underbrace{X_{2},*,...,*}_\text{|incx|},X_{1} }_\text{1 + (1-n) x incx}] \quad if \:incx \:< \:0\]

Parent topic: Dense Linear Algebra