[info-performer] MISSES stats

Date view	Thread view	Subject view	Author view

From: Arthur Lorenz (ifw99140++at++informatik.fh-muenchen.de)
Date: 05/14/2004 05:07:58

Next message: Gordon Tomlinson: "RE: [info-performer] MultiTexture with flt"
Previous message: Tiziano Diamanti: "[info-performer] MultiTexture with flt"

Hello Performers,

I'm running some kind of benchmark on my system (Xeon 2.4 GHz, nVidia FX 5900).
There are 2213 waypoints with each waypoint rendered.
When using multi-processing mode PFMP_APPCULLDRAW, PFPHASE_LIMIT=120
as frame synchronization and culling disabled the MISSES stats yield
(summed up for 2213 waypoints):

PFFSTATSVAL_PFTIMES_MISSES_TOTAL: 0
PFFSTATSVAL_PFTIMES_MISSES_APP: 2
PFFSTATSVAL_PFTIMES_MISSES_CULL: 0
PFFSTATSVAL_PFTIMES_MISSES_DRAW: 1493
PFFSTATSVAL_PFTIMES_MISSES_ISECT: 0
PFFSTATSVAL_PFTIMES_MISSES_DBASE: 0
PFFSTATSVAL_PFTIMES_MISSES_LPOINT: 0

<code snippet>
/usr/include/Performer/prstats.h explains:

     * misses - number of frames each stage missed goal frame rate
     * here, total refers to a dectection of an overall miss of a
     * frame -- even if no singe stage detected a missed frame.
     */
    pfFStatsValProc misses; /* PFFSTATSVAL_PFTIMES_MISSES */
</code snippet>

Shouldn't PFFSTATSVAL_PFTIMES_MISSES_TOTAL be (1493+2) = 1495
or a greater number of missed frames?

PFMP_APPCULL_DL_DRAW shows a similar behavior, whereas other
multi-processing modes, eg. PFMP_APPCULLDRAW, PFPHASE_LIMIT=120
as frame synchronization and culling disabled the MISSES stats tell me
(again, summed up for 2213 waypoints):

PFFSTATSVAL_PFTIMES_MISSES_TOTAL: 1636
PFFSTATSVAL_PFTIMES_MISSES_APP: 3
PFFSTATSVAL_PFTIMES_MISSES_CULL: 0
PFFSTATSVAL_PFTIMES_MISSES_DRAW: 1492
PFFSTATSVAL_PFTIMES_MISSES_ISECT: 0
PFFSTATSVAL_PFTIMES_MISSES_DBASE: 0
PFFSTATSVAL_PFTIMES_MISSES_LPOINT: 0

PFFSTATSVAL_PFTIMES_MISSES_TOTAL > 1495
which seems ok to me because a frame might not be rendered in time within
a frame boundary (although no stage missed the goal frame rate) so it will be
counted as a TOTAL missed frame.
Is the behavior with PFMP_APPCULLDRAW and PFMP_APPCULL_DL_DRAW
a bug??

Regards,

Arthur

--
Arthur Lorenz
Munich University Of Applied Sciences
Department Of Computer Science
ifw99140++at++cs.fhm.edu

Next message: Gordon Tomlinson: "RE: [info-performer] MultiTexture with flt"
Previous message: Tiziano Diamanti: "[info-performer] MultiTexture with flt"

Date view	Thread view	Subject view	Author view

This archive was generated by hypermail 2b29 : Fri May 14 2004 - 05:07:30 PDT