Essentially I want a fast way of doing: c = (A^k)*b0. But I want the result for multiple values of k (I don't need it for all values of k, just some).
At the moment, I am just doing this in a normal for loop (b1 = A*b, b2 = A*b1, b3 = A*b2, etc.) for all k. But I am wondering if there is a faster way (maybe using GPUs).
Doing loops in GPUs doesn't seem like the way forward. I was thinking I could just request c = (A^k)*b0 (which is very fast on the GPU) for only the k that I want, but if I want many (for example for k = [1:5:1000]) this still ends up being slower than just doing it on a loop on the CPU.
Any suggestions? Thanks -
N = 301;
k = 1000; A = randn(N)/17; b = rand(N,1);
f = @() r(A,b,k);
t = timeit(f);disp(t)
function b = r(A,b,k)
for ix = 1:k
b = A*b;