GitLab

- generalize identity matrix constructors for non-scalar types

- add constructors using labelPair for the row/column sizing information.
  For a SquareMatrix, this provides an unambiguous parameter resolution.

- reuse assignment operators

STYLE: adjust matrix comments

- additional operators:
  + compound assignment
  + inner product: operator&
  + outer product: operator^

- additional functions:
   - MatrixBlock methods: subColumn, subRow, subMatrix
   - L2 norms for matrix or column
   - trace, diag, round, transpose

- MatrixBlock methods: col(), block() are deprecated since their
  access patterns with (size, offset) are unnatural/unwieldy.

- verifications by test/Matrix/Test-Matrix

- make names consistent with lduMatrix
    A*x  => Matrix::Amul
    AT*x => Matrix::Tmul  (same as x*A)

- the vector-matrix multiplication is treated as a row vector

- add iterators, begin/end, empty() methods for STL behaviour.
  Use standard algorithms where possible
     * std::fill, std::copy
     * std::min_element, std::max_element

- access methods consistent with other OpenFOAM containers:
     * data(), cdata(), uniform()

- Use ListPolicy to impose output line breaks

- Can recover matrix storage for re-use elsewhere.
  For example, to populate values with 2D i-j addressing and later
  release it as flat linear storage.

- construct/assign moveable

- added minMax() function for Matrix

- additional inplace +=, -=, *=, /= operations

- add named methods at() and rowData() to Matrix.
  Allows a better distinction between linear and row-based addressing

- low-level matrix solve on List/UList instead of Field

Requires gcc version 4.7 or higher

Note: this reuses the existing storage rather than costly reallocation
which requires the initial allocation to be sufficient for the largest
size the ODE system might have.  Attempt to set a size larger than the
initial size is a fatal error.

This avoids serious problems with template parameter deduction when
manipulating blocks of different matrix types e.g. Square and
Rectangular.

The blocks may be specified directly in terms of the size and location in the
parent matrix or with the size obtained from a template specified
VectorSpace or MatrixSpace type.

This change brings OpenFOAM into line with the standard matrix
addressing in other C++ libraries for better interoperability.

The row-start pointer array provided performance benefits on old
computers but now that computation is often cache-miss limited the
benefit of avoiding a integer multiply is more than offset by the
addition memory access into a separately allocated array.

With the new addressing scheme LUsolve is 15% faster.

For consistency with current references and other numerics libraries

Avoids the clutter and maintenance effort associated with providing the
function signature string.

This reverts commit c5bea524.

This reverts commit b18f6cc1.

- uniform use of reinterpret_cast<foo*>(0) instead of
  reinterpret_cast<foo*>(NULL)

- make all static null() members inline since they are really only a cast:
      "*reinterpret_cast<foo*>(0)"

the type it will become as a template argument.

Brought everything into line with this change.

Added the quadraticFit interpolation scheme.