Katja:

The below code took 6.85 secs on babyblue and .06 secs
on the Origin ute. If I replace the merge with if/then/else, 
it runs just as fast on babyblue as on the SGI, and using 
the conditional does not appear to hurt the time on the SGI.

So I will begin experimenting w/ replacing merge with
conditionals and maybe which-constructs in my code, but 
in the meantime I wonder if it would be worth asking IBM's 
compiler people to look at this -- maybe they could speed 
it up somehow..

--Ben

!
      program f90merge
      implicit none
      real :: tsource,fsource,x
      integer :: i,n
!
      n = 1000000
      tsource = 1.
      fsource = 0.
      x = 0.
      do i=1,n
        x = merge(tsource,fsource, x < 200.) 
!       if (x < 200.) then
!         x = tsource
!       else
!         x = fsource
!       endif
      enddo
      end program f90merge


> From katjas@meeker.ucar.edu Tue Apr  4 06:22 MDT 2000

> I talked to John Levesque about it, and he says that yes, merge is very
> slow on ibm. It's up to you, but the code would be faster. Maybe it would
> be worth it to change the simple cases and see what kind of improvement
> you get.
> Katja