Currently input and output cache files were "shared"
because of how the configuration was made.

Also, the input for the loop algorithm, although the same
as the loop_user, was not properly generated as a separate
input.