I'm finding a significant error for certain input values when I use a particular offset to add values. Due to the nature of the algorithm I'm considering (not relevant to the question), I need to start a sum value with the first element of an array. I was able to extract a minimal working example, which is appended below.
Even though there are less than 50 floating point additions involved, I'm seeing error far beyond that order of magnitude of ulps, and I don't understand what I am doing wrong. See the following code snippet, along with the invocation and what my results are: