Basic Features

MV++ provides two basic classes: a numerical vector (1-d array), and column (Fortran) oriented dense matrix. Indexing is performed via the operator(), as in A(i,j).

Subvectors and submatrices can be accessed through MV_VecIndex classes. In other words, if B is the vector

then B(MV_VecIndex(1,5)) = 9.9 sets B to

and is equivalent to

       for (i=1; i<=5; B[i++] = 9.9 )

Other considerations:

• efficient indexing as fast as native C arrays
• supports only unit strides (for efficient indexing)
• copy-by-value semantics
• optional "share" semantics allows vectors to be constructed as "views", or "references" of an existing memory, using MV_Vector_::ref modifier in the constructor. (see note below.)
• vector views can assign and references sections of vector, but cannot modify their size.
• block-range indexing via MV_VecIndex class (e.g. A(I) = B; ) (note for the above to work, A(I) must return a vector view.)
• optional range checking (compile switch)
• support for both [] and () style indexing ([] not available for matrices.)

• Unrolling of loops (depth=4) for copying and assigning vectors. Thus on some machines, it may be faster to execute A=scalar, than to manually assign a native C array using an explicit for loop:

      
         for (i=0; i<N; d[i++] = scalar);
  

• function code for the () and [] operators has been inlined into the class declaration, for compilers that refuse to inline otherwise.

• optinal share semnatics. To create a view of (or reference to) an existing MV++ matrix or vector, use

              MV_Vector_double A( &d[0], n, MV_Vector_::ref );
  

  This allows one to construct vectors as views of any contiguous C array. It will not release the memory space when the vector is destroyed or goes out of scope.