RE: problems with pfFrame and pfSync on multipipe

New Message Reply Date view Thread view Subject view Author view

From: Dick Rous (dick++at++sgi.com)
Date: 08/31/2000 03:13:15


Hi,

what is your pfPhase set to? If set to FLOAT or LOCK, pfSync wait for the
next FRAME boundary,
which, at 20 Hz, is at 50 ms intervals.
When you don't do any processing, it is not neccessary to call pfSync
anyway, it is implied in
pfFrame when omitted.
Strictly speaking, the sequence should be: non critical updates, pfSync,
latency critical updates,
pfFrame, back to start.

Cheers,

Dick.
__________________________________________________________
Dick Rous SGI - European Technical
Support
email: dick++at++sgi.com
phone: +31-35-6423160 fax: +31-35-6423162 VNET: 955-6868
ETS Central Phone: +44 8700 716 300
__________________________________________________________

-----Original Message-----
From: eylon [mailto:eylon++at++bvr.co.il]
Sent: 31 August 2000 12:45
To: performer
Subject: problems with pfFrame and pfSync on multipipe

Hi everybody

Our system has 8 performer processes (APP + 3xDRAW + 3xCULL + COMPUTE),
running on a 3 pipes and 10 cpu's (each process runs on it's own locked
cpu).
The basic stages are :
        call pfFrame;
        update dcs's;
        call pfSync;

The problem is that when the scene become complicated (a lot of dcs's)
the time it takes pfFrame to finish become extremely large and pfSync
misses the next frame.
I checked the time it takes each stage on a complicated scene and i got:
pfFrame - 16 msec
update dcs's - 29 msec
pfSync - 55 msec
so we miss a frame and we are running 10 HZ instead of 20 HZ

The question is Why does pfFrame take so much time and is that what
cause pfSync to miss the next frame?

Greetings

Eylon
-----------------------------------------------------------------------
List Archives, FAQ, FTP: http://www.sgi.com/software/performer/
            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 2b29 : Thu Aug 31 2000 - 03:13:25 PDT

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