Posted on behalf of Paras:
Hi,
I am hoping that the c gurus can help me understand how OpenMx matrices
are stored, updated and passed around.
Here is the situation:
We are trying design data-structures for the block-sparse BLAS routines
and are struggling with the issue of ownership of "double" matrices. Our
key data-structure is called the "blockSparseMatrix". For example, there
may be a huge "factor-loading" matrix where each dense sub-block would
map on to one or more OpenMx parameter matrices. In the ordinary
factor-model, each dense sub-block would point to the single
factor-loading matrix. Alternatively in the growth-curve case with
definition variables, each dense-block would point to person specific
factor-loading matrix.
The struct blockSparseMatrix has a field **dns or a "pointer to an array
of doubles representing the dense matrix". From a memory and efficiency
perspective, it makes sense for OpenMx to create, update and destroy the
memory for dense-matrices. Pointer to the dense matrix would be passed
to the backend at the beginning and would be retained throughout the
estimation process as a member of the blockSparseMatrix structure.
>From an OO perspective, having the main application write directly to a
struct-member seems to be problematic. However, I see no alternative to
doing so -- without copying huge amounts of data at each iteration.
>From OpenMx's design perspective and how you envision development of
other plug-ins, what approach would you recommend?
How does OpenMx deal with R? If I understand correctly, R insists that
all data be "copied" between R and external-packages, unless the access
is read-only. Is this what happens at each iteration?