About MV++

MV++ is a small, efficient, set of concrete vector and matrix classes specifically designed for high performance numerical computing.

The MV++ package includes interfaces to the computational kernels found in the Basic Linear Algebra Subprograms (BLAS), such as scalar updates, vector sums, and dot products. The idea is to utilize vendor-supplied, or optimized BLAS routines that are fine-tuned for particular platforms.

The various MV++ classes form the building blocks of larger user-level libraries such as SparseLib++[2] and LAPACK++[1]. The MV++ library was put together with the following principles in mind:

• a simple numerical vector and column-oriented dense matrix class.

• minimal overhead in constructing, assigning, and copying vectors and matrices

• performance competitive with optimized Fortran kernels

• data structure compatibility with Fortran libraries and subroutines

• concrete classes, no complicated inheritance hierarchy, but

• support generic element types through templated paramters

• support operations with contiguous subvectors and submatrices (e.g. zeroing out a section of a vector.)

• optional runtime support for array-bounds checking.