If I understand your first question correctly you want to know how an, as yet unimplemented, function is implemented ? Specifically, you want to know the performance characteristics of findloc compared with linear and binary search as you might have implemented them yourself ? There is certainly nothing in the language standard which mandates how the function is to be implemented, so the answer to your question is entirely compiler specific.
As to your second question, I would expect most compilers to cause the creation of a temporary array in response to your expression array-value. The creation of such a temporary is likely to be a relatively time-consuming operation and will be added to the execution time of the call to minloc. How the Intel implementation of minloc works I don't know but I expect that it is a linear scan through the array. There's no way for the intrinsic function to know that an array is sorted and that binary search might be faster.
If your arrays are small and unsorted I would expect linear search to be the fastest option. If they are small and sorted you might be able to write a binary search (or similar) to outperform linear search. I expect the performance graphs of the two approaches to have a crossover, where that crossover sits wrt your idea of small size I haven't a clue.
However, as always with performance matters, what I (or anybody else) think is useless, data is what you need, why don't you go ahead and make some measurements ?
