void copy(matrix_slice<matrix<NumericT, column_major, 1> > const & gpu_matrix_slice,
CPUMatrixT & cpu_matrix)
{
assert( (cpu_matrix.size1() == gpu_matrix_slice.size1())
&& (cpu_matrix.size2() == gpu_matrix_slice.size2())
&& bool("Matrix size mismatch!"));
if ( (gpu_matrix_slice.size1() > 0) && (gpu_matrix_slice.size1() > 0) )
{
vcl_size_t num_entries = gpu_matrix_slice.size1() * gpu_matrix_slice.stride1(); //no. of entries per stride
std::vector<NumericT> entries(num_entries);
//copy each column stride separately:
for (vcl_size_t j=0; j < gpu_matrix_slice.size2(); ++j)
{
vcl_size_t start_offset = gpu_matrix_slice.start1() + (gpu_matrix_slice.start2() + j * gpu_matrix_slice.stride2()) * gpu_matrix_slice.internal_size1();
viennacl::backend::memory_read(gpu_matrix_slice.handle(), sizeof(NumericT)*start_offset, sizeof(NumericT)*num_entries, &(entries[0]));
for (vcl_size_t i=0; i < gpu_matrix_slice.size1(); ++i)
cpu_matrix(i,j) = entries[i * gpu_matrix_slice.stride1()];
}
}
}
matrix_slice(matrix_slice<MatrixType> const & A,
slice const & row_slice,
slice const & col_slice) : base_type(const_cast<handle_type &>(A.handle()),
row_slice.size(), row_slice.start() * A.stride1() + A.start1(), row_slice.stride() * A.stride1(), A.internal_size1(),
col_slice.size(), col_slice.start() * A.stride2() + A.start2(), col_slice.stride() * A.stride2(), A.internal_size2(),
A.row_major()) {}