Re: g95 timings

Thanks to all who replied. I have tried all the options
suggested, with no more than a few percent gain on the Intel
processor (and no obvious difference on the AMD).

S. Bosscher pointed out that -O4 is invalid. That was a typo in
my post only (not in the makefile, which uses -O3).

R. Baer asked how I was measuring elapsed time. I use a Python
script that records wall time, which seems good enough for these
rough calculations.

T. Prince asked, "Does g95 not support sse code for 32-bit
platforms (-mfpmath=sse)? Nor vectorization?" I was able to
use SSE math on the Intel processor, and it made little or no
difference. I have tried vectorizing the code previously, but
the calculations are not particularly amenable to it, so I
haven't bothered in a while.

Probably because I am blissfully ignorant of compiler design,
it's still puzzling to me why the time penalty (of g95 vs. LF95)
is so much greater on one architecture than on another.

Still, g95 is quite an achievement.

MHP


--
Mike Prager, NOAA, Beaufort, NC
Address spam-trapped; remove color to reply.
* Opinions expressed are personal and not represented otherwise.
* Any use of tradenames does not constitute a NOAA endorsement.
.