Fig. 2

Being the most computational intensive part of the evaluation, all the threads are divided up to rapidly compute the FOI, λ. For illustrative purposes, there are 32 threads (in total) active in this example. There are four lambdas calculated in parallel; for each lambda there are eight threads in charge of executing the actual matrix-vector calculation. This window of eight threads per row slides over the elements of the row until the lambda is completely calculated