Re: cannot pass variable size array to subroutine



Richard Maine <nospam@xxxxxxxxxxxxx> wrote:
(big snip comparing assumed shape and such)

< I find it a little hard to generalize much, as the NAG and g95 data are
< so different in trends. My g95 data seems consistent with yours. I do
< note the support for my prior comment that the NAG compiler switches
< should make the assumed-shape performance match that of explicit shape.

This reminds me of a problem that should have gone away by
now, but maybe not. For intel processors starting with the
pentium, load/store of REAL*8 data is much faster aligned on
an eight byte boundary. Previous processors had no advantage
for eight byte alignment, and it was common to only supply four
byte alignment.

That should have been fixed by now, but I remember it being
mentioned that it was still there somewhat recently.

It could easily make large timing differences appear, unrelated
to the actual change in the program. Hopefully someone will tell
me that all compilers now generate eight byte alignment, but I
am not so sure yet.

-- glen
.