From guest Thu Jun 16 12:51:15 1994 From: "Drew Hess" Message-Id: <9406161250.ZM21841@airy.arc.nasa.gov> Date: Thu, 16 Jun 1994 12:50:48 -0700 X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) To: info-performer@sgi.sgi.com Subject: multiple pfChannels vs. multiple pfPipes Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 Status: O I want to render several views of the same scene on a 4-processor Onyx/RE2 (one RE2 pipeline at this point) in a Performer application. Is it generally better to configure Performer to use one pfPipe and multiple pfChannels, using channel groups to share as much state information as possible; or should I use multiple pfPipes and split the number of channels up evenly among the pfPipes? Also, if I get a second RE2 pipeline then obviously I would want multiple pfPipes, but assuming I still need more pfChannels than I have hardware pipelines, then I would assume the same partitioning issue applies. (That is, do I create 2 pfPipes -- one for each RE2 pipeline -- and split up the pfChannels among them or do I create multiple pfPipes per hardware pipeline and reduce the number of pfChannels per software pipe (i.e. pfPipes)?) Thanks -dwh- dhess@vision.arc.nasa.gov dhess@cs.stanford.edu From guest Thu Jun 16 13:27:27 1994 From: aaron@qbert.dseg.ti.com (Aaron Hightower) Subject: Re: multiple pfChannels vs. multiple pfPipes To: iris-performer@sgi.sgi.com Date: Thu, 16 Jun 1994 15:30:21 -0500 (CDT) X-Mailer: ELM [version 2.4 PL23] Content-Type: text Content-Length: 2411 Status: OR "Attribution from author Drew Hess" > From holodeck.asd.sgi.com!guest Thu Jun 16 15:09:57 1994 > From: "Drew Hess" > Message-Id: <9406161250.ZM21841@airy.arc.nasa.gov> > Date: Thu, 16 Jun 1994 12:50:48 -0700 > X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) > To: info-performer@sgi.com > Subject: multiple pfChannels vs. multiple pfPipes > Content-Type: text/plain; charset=us-ascii > Mime-Version: 1.0 > > I want to render several views of the same scene on a 4-processor Onyx/RE2 (one > RE2 pipeline at this point) in a Performer application. > > Is it generally better to configure Performer to use one pfPipe and multiple > pfChannels, using channel groups to share as much state information as > possible; or should I use multiple pfPipes and split the number of channels up > evenly among the pfPipes? From the man page for pfPipe: A pfPipe is tied to a specific hardware pipeline according to the hardware screen that the graphics window is opened on. In initFunc, scrnselect should be used to select a hardware pipeline. If initFunc is NULL, the full-screen window will be opened on the hardware pipeline whose id corresponds to the pfPipe id, e.g.- scrnselect(1); will be used for pfPipe number 1. For best performance only one pfPipe should render ^^^^^^^^^^^^^^^^ to a given hardware pipeline. If multiple views on a single screen are desired, use multiple pfChannels. > Also, if I get a second RE2 pipeline then obviously I would want multiple > pfPipes, but assuming I still need more pfChannels than I have hardware > pipelines, then I would assume the same partitioning issue applies. (That is, > do I create 2 pfPipes -- one for each RE2 pipeline -- and split up the > pfChannels among them or do I create multiple pfPipes per hardware pipeline and > reduce the number of pfChannels per software pipe (i.e. pfPipes)?) Hope this helps, _ _ | | *-------------------------------------------* | | __| |___ | Aaron.Hightower@dseg.ti.com 214.575.6759 | __| |___ \ / | Simulation & Planning Technology | \ / \_ / | 6620 Chase Oaks Blvd M/S 8518 | \_ / \ ( | Plano TX 75023 | \ ( \/ *-------------------------------------------* \/ From guest Tue Jun 21 11:16:36 1994 Message-Id: <9406211816.AA22934@surreal.asd.sgi.com> To: "Ran Yakir" Cc: info-performer@sgi.sgi.com Subject: Re: multi-channel synchronizing In-Reply-To: Your message of "Tue, 21 Jun 94 20:23:37 -0000." <9406212023.ZM11108@amcor.bvr.co.il> Date: Tue, 21 Jun 94 11:16:27 -0700 From: Jim Helman Status: OR I can think of two things that might contribute to this: 1) You don't mention whether your pipes are genlocked. When driving multiple pipes from the same Performer application, it's important that the pipes be genlocked for good synchronization. 2) There's a bug in 1.2 regarding the vertical retrace clock initialization in multiple hardware pipes. We don't initialize them so they can be out of sync. The workaround is to do it yourself by simulataneously (i.e. within the same video field) calling pfInitVClock(0.0). A good place to do this synchronization is during the first invocation of the channel draw callback. Something like this should work: if first invocation of a channel draw callback on this pfPipe { wait for other draw processes to reach this point /* block on pipe until next retrace */ gsync(); cpack(0x000000); /* initialize clocks */ pfInitVClock(0.0f); } rgds, -jim helman jimh@surreal.asd.sgi.com 415/390-1151 From guest Tue Jun 21 10:24:47 1994 From: "Ran Yakir" Message-Id: <9406212023.ZM11108@amcor.bvr.co.il> Date: Tue, 21 Jun 1994 20:23:37 +0000 X-Face: aaaaaaaaaaljasddfjsdlkfjasd X-Mailer: Z-Mail (3.1.0 22feb94 MediaMail) To: info-performer@sgi.sgi.com Subject: multi-channel synchronizing Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 Status: O Hi I'm trying to run a dual piped system, which, upon arrival of hardware, should become triple piped. Right now I have only 4 CPUs on my ONYX, su I use PFMP_APP_CULLDRAW. The weird thing is that the left and right channels don't seem to synchronize. Although their cull-draws run on different processors, they tend not to run concurrently. Moreover, they skip frames in a manner that channel 0 does frame # N, channel 1 does (N+1), channel 0 does (N+2), channel 1 does (N+3), and so on ... . Sometimes they suddenly sync, and run concurrently, but not for long. Just to be sure its not an overload, I asked pfFrameRate to be 10Hz (100 msec), while the whole draw/cull thing takes less then 20 msec. I do plush the pipe at the end of each frame to get accurate timing results. In addition, I used the cpu locking utils from libpfutil, and runed as root, so that they don't switch processors on me. And ofcourse I run with PFPHASE_LOCK. Can anyone understand what that is ? Ran -- __ | Ran Yakir /_) _ __ \ / _ / o __ | Graphics App. Chief Engineer / )_ (_(_) ) \/ (_(_/<_(_)( | BVR Technologies Ltd. _/ | -------------------------------------+-------------------------------- Phone : | E-mail : rany@bvr.co.il Work : 972-3-5715671 | Res. : 972-3-6995364 | Fax : 972-3-5715668 |