[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

20040520: adding some compiler options to McIDAS

>From: Robert Mullenax <address@hidden>
>Organization: Universal Weather
>Keywords: 200405201932.i4KJWltK000072 McIDAS build

Hi Robert,

>I would like to wring a little more performance out of McIDAS by using
>-xtarget=pentium4
>under Solaris x86 using the Solaris x86 Early Access compilers.
>I believe I should add that in either makefile or mccomp.sh but I am not
>sure where.

You would add this to mccomps.h.  Here is an example:

 ...

        # use the SunOS vendor compilers
        case ${vendor:-$MCIDAS_CC} in
        ?*)
            case `uname -r` in
            5.*)
                cc=cc
                ccopts="-xtarget=pentium4"
                ;;

            4.*)
                cc=acc
                ;;
            esac
            ;;
        esac
        case ${vendor:-$MCIDAS_F77} in
        ?*)
            case `uname -r` in
            5.*)
                fc=f77
                fcopts="-xtarget=pentium4"
                flopts=
                ;;

            4.*)
                fc=f77
                ;;
            esac
            ;;
        esac
 ...


Please let us know if this shows significant performance improvement
on your SolarisX86 machine.

Cheers,

Tom
--
NOTE: All email exchanges with Unidata User Support are recorded in the
Unidata inquiry tracking system and then made publically available
through the web.  If you do not want to have your interactions made
available in this way, you must let us know in each email you send to us.

>From address@hidden  Thu May 20 14:07:04 2004

Thanks. I'll compile with gcc/g77 as well and let you know the difference.
On GEMPAK, other than nex2gini not working at all (strangely enough the same
kind of problem as I see under RH Linux 9) I have seen 40-70% speed
improvements over the Unidata binary compiled with WS5.2 

>From address@hidden  Sun May 23 09:19:16 2004

At least for my purposes (trying to do a whole lot of things and needing
to do them as quickly as possible) it's a decent improvement, although
not as dramatic of an improvement that I saw with GEMPAK.

Imgremap is one of the most time consuming tasks, so I focused on that.

In doing several different remaps, including remapping a 540MB hires topo
file, the version compiled with the x86 Beta Studio 9 compilers (using -O
-xtarget=pentium4 -xlic_lib=sunperf....Sun now has performance libs for x86)
averages 13-20% faster than using straight gcc/g77.  Remapping that
540MB file went from 42 seconds to 34 seconds.   

I haven't tested everything yet to see if it works..but so far eveything
I normally use in my scripts works fine.