xtensor
xtensor copied to clipboard
reducing functions on non-contiguous memory
I have a question regarding the internals of reducing functions that I can't find an answer for in the documentation. When applying a reducing function like xt::mean to an axis of a multidimensional array it can happen that operations on non-contiguous memory have to be performed. Are there cases where the array is first rearranged so that only operations on contiguous memory have to be performed (for access time and vectorization), or are the operations always directly performed on the different memory parts using the strides to "jump" between them?