Is a DO-constructed array calculation faster than using Fortran 90 array intrinsics?

[R. T. (Fortran Man)]

Is a DO-constructed array calculation faster than using Fortran 90 array intrinsics?

[Lahey Support]

Here are a few rules of thumb regarding the speed of most of the array intrinsics. We would be surprised if any comparison between the intrinsic procedures and user code showed a difference of more than 2:1 (or less than 1:2).

* if whole arrays (rather than non-contiguous sections) are passed as arguments the intrinsics will be about as fast as a nested do loop construct in the calling routine.

* if the arrays are higher order than rank 3 then the intrinsic function will probably be faster the than user code.

* passing array sections to array intrinsics gets a performance hit, and a do construct in the caller will likely be faster.

* the intrinsic will almost always be faster than a *routine* that a user writes to do the same function. (routine is emphasized above to differentiate from code that the user inserts into the calling routine.)