From: Allan Schaffer (allan++at++sgi.com)
Date: 02/27/2000 11:19:52
Performers,
IRIX 6.5.7, including Performer 2.2.7, is now available via the usual
distribution channels and for download from http://support.sgi.com
We've made some internal changes in performer_eoe for version 2.2.7
that could affect your application; please read on.
Feature Improvements:
- Multi-pipe scalability:
In previous releases each additional pfPipe increased APP to
CULL data transfer overhead, by the (number of pipes) x (scene
graph changes per frame). Performer 2.2.7 reduces this
overhead; multi-pipe applications should see a dramatic
improvement in APP process performance.
Known problem (to be fixed for 2.2.8): Using pfASD query points
to place objects on a pfASD surface, the query results arrive
one frame away from the pfASD geometry so alignment is
incorrect.
- Shared memory default allocation scheme:
The default memory allocation parameters for the shared arena
have been modified to reduce memory fragmentation in the shared
arena.
Upon creation of the shared arena, M_FREEHD is set to 1 and
M_MXCHK is set to 2000. Refer to amallopt(3P) for details on
how these parameters affect memory allocation behavior.
You may revert to the previous defaults by setting M_FREEHD to 0
and M_MXCHK to 100 after creation of the shared arena; place
such calls between pfInit() and pfConfig().
Other amallopt parameters such as M_MXFAST may have to be
changed before a call to pfInit() if you decide to modify them
forcing you to separately and explicitly initialize the shared
arena before the pfInit() call. Changing back to the old values
is not recommended; you may wish to experiment by increasing
M_MXCHK even further if you have a forked database process or
would like to trade performance for memory efficiency.
Bug Fixes:
- Triangle fans were not being correctly reported by the stats.
This has been fixed (SCR 769104).
- An array bounds underrun in _pfCuller::pf_applyXform could lead
to memory corruption. This has been fixed (SCR 767432).
- The prototype for pfDTRApplyImageCache was incorrectly
documented. This has been fixed (SCR 752840).
- pfNode::bufferClone could cause a segmentation violation. This
has been fixed (SCR 695725).
- The channel frustum type was being reset by
pfChannel::pf_cullShadows. This has been fixed (SCR 679085).
- A memory leak caused by multi-pipe pfImageCache invalidation has
been fixed.
Enjoy,
Allan
---- Allan Schaffer allan++at++sgi.com Silicon Graphics http://reality.sgi.com/allan
This archive was generated by hypermail 2b29 : Sun Feb 27 2000 - 11:19:58 PST