topic assignment efficiency in Intel® Fortran Compiler

assignment efficiency

lklawrie — Tue, 09 Mar 2010 16:16:14 GMT

Is it more efficient to assign via a Do loop:

do i=1,n

x(i)=someotherarray(i,1)

enddo

x(1:n)=someotherarray(1:n,1)

or does it matter?

Linda

assignment efficiency

TimP — Tue, 09 Mar 2010 16:27:54 GMT

For a rank 1 assignment, it's not likely to make a difference.

assignment efficiency

abhimodak — Tue, 09 Mar 2010 17:32:09 GMT

Hi Linda

If you are using Win64 and have large arrays you may want to take a quick look at : http://software.intel.com/en-us/forums/showthread.php?t=61077

The last post by Mr. Green indicates that Intel may need do a little bit of further analysis, but based on the response received in Premier Support, I think Intel has pulled the plug on it. (The outer "dead" loop pointed out by Mr. Green is not optimized-out in even in the latest release 11.1.060. But as far as I know this happens only on Win64.)

Also I recall that Steve Lionel had mentioned somewhere to use the whole array syntax instead of specifying the complete partition, if it is possible i.e. avoid A(1:N) = B(1:N) and use A = B if the array size is N. (I interprete this as compiler may think it has to do something extra when the partition is mentioned.)

Abhi

assignment efficiency

abhimodak — Tue, 09 Mar 2010 17:35:04 GMT

Hi Tim

Does this mean for rank n with n>1 would have a bigger impact? Can you please illustrate this point with an example? It may be super-elementary but I think it is worth to "put it on the cap"; your advice deeply appreciated.

Sincerely

Abhi

assignment efficiency

Steven_L_Intel1 — Tue, 09 Mar 2010 18:38:56 GMT

Is n the upper bound of array x? If so, then remove the (1:n).