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.
Provides '(i, j)' element access and general forms of inner and outer
products, transpose etc. for square and rectangular VectorSpaces.
VectorSpaces default to be column-vectors as before whereas row-vectors
may be represented as 1xn MatrixSpaces. In the future it may be
preferable to create a specializations of VectorSpace for column- and
maybe row-vectors but it would add complexity to MatrixSpace to handle
all the type combinations.
Tensor is now a 3x3 specialization of MatrixSpace.
Sub-block const and non-const access is provided via the
'.block<SubTensor, RowStart, ColStart>()' member functions. Consistent
sub-block access is also provide for VectorSpace so that columns of
MatrixSpaces may be accessed and substituted.
These new classes will be used to create a more extensive set of
primitive vector and tensor types over the next few weeks.
Henry G. Weller
CFD Direct
