This is a slightly more substantial programming problem.
The conjugate gradient method provides a general method for the solution of linear systems Ax = b for symmetric matrices. The algorithm is explained here:
https://en.wikipedia.org/wiki/Conjugate_gradient_method
It is not necessary to understand the details, as we are just interested in implementing the algorithm.
There are two main steps involved. The first is to perform a matrix-vector multiplication. This was an earlier exercise for which a solution is provided in the current directory.
The second step is to compute a scalar residual from a vector
residual. If we have a vector r of length n this can be
done in serial with:
residual = 0.0;
for (int i = 0; i < n; i++) {
residual += r[i]*r[i];
}
This you should also recognise as the solution to an earlier exercise (a version is also provided).
The exercise then means using these parts to construct the entire algorithm. Restrict yourself to the use of a single GPU.
Remember that you can always check your answer by multiplying out the solution to see if you recover the original right-hand side. This can be done on the host.