Phoxis

In a previous post “Allocating multidimentional array at runtime in C” I have explained a technique to allocate multidimensional arrays on runtime. While playing around with OpenMPI, I came to know that while sending/receiving a buffer, it requires the elements of a 2d matrix (or any dimension) to be in adjacent. Basically it does not care what you send, or receive, what it cares is the number of elements to be send should be adjacent to one another. In the previous post, the process will allocate the 2d or n-d matrix, but the rows of the matrix may not be adjacent to each other, as each row was allocated separately with malloc and then inserted into another array of pointers, each of which location points to the base addresses of these memory block. Read the post for details.

In C language the 2d array/matrix are stored in a row-major order, that is the elements of the matrix are stored adjacent to each other in the memory row wise. The first row comes first then just after the first row the second row starts, and so on. In the previous method the rows of the matrix may be scattered throughout the memory, as they are allocated with seperate malloc calls, but each of these returned addresses to the memory blocks (used as rows) are assigned to another array of pointers, which holds the rows together, and allows the mat[i][j] syntax to work.

For the applications in which, we might need to allocate the matrix dynamically at runtime, also have the rows of the matrix requires to be adjacent in the memory, and also make the mat[i][j] syntax work can be fulfilled by the following approach.
Continue reading →