Clock rate on SGI's

New Message Reply Date view Thread view Subject view Author view

John W. Barrus (barrus++at++merl.com)
Wed, 3 Jul 96 11:57:04 EDT

This may not seem like it is directly applicable to Performer, but people on
this list probably care about this issue and the Performer team is always
very responsive to questions relating to SGI performance.

I have been using the cycle counter on several different types of sgi
machines. This is explained in 'man syssgi' under SGI_QUERY_CYCLECNTR.

The second argument in the syssgi call is a pointer to a place in memory in
which the SGI can put a clockrate or cyclerate which is in picoseconds
(10^-12 seconds) per cycle.

However, after a little bit of fooling around, I found the following
discrepancies:

               reported measured calc'ed
Machine cyclerate ticks/second cyclerate clockspeed

extreme 32,000 25,000,000 40,000 250 MHz
indy 45,454 22,222,222 45,000 132 Mhz
onyx 21,000 47,629,629 21,000 150 MHz
other onyx 21,000 47,629,629 21,000 200 MHz

So: I have been writing code to find out the actual cyclerate inside my
program when I realized that it would be much better if I could just test
each machine and put the right value in the right place on the machine once
and not have to run the test everytime my program ran.

Anyone know how to do this? How should I go about complaining to SGI to fix
their cyclerate values?

Please e-mail me directly in addition to posting your responses to the
mailing list and if there is an interesting outcome, I will post a summary
to the list.

Thanks,

John B.
John W. Barrus Research Scientist
Mitsubishi Electric Research Labs Voice: (617) 621-7535
201 Broadway Fax: (617) 621-7550
Cambridge, MA 02139 E-mail: barrus++at++merl.com
                        http://www.merl.com/

=======================================================================
List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/ <--new!
            Submissions: info-performer++at++sgi.com
        Admin. requests: info-performer-request++at++sgi.com

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:53:08 PDT

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.