Text archives Help
- From: David E DeMarle <
>
- To:
- Subject: [Manta] Re: Re: Re: Re: Re: Re: two frame lag?
- Date: Fri, 30 Apr 2010 16:34:57 -0400
Thanks!
Normally manta does:
callbacks(f) display(f-1) render(f) //where worker threads skip display(f-1)
So after vtkManta issued callbacks to set up frame f, it had to ask
manta to render twice to get pixels for frame f out.
When the new flag is on, manta does:
callbacks(f) render(f) display(f)
I couldn't figure out how to do it via callbacks as you suggested
before, and I wasn't able to figure out another option, but that
certainly doesn't mean there isn't one.
I suspect there is some good way to overlap the VTK and Manta thread
work as Manta wants, but I don't anticipate getting to that for some
time.
David E DeMarle
Kitware, Inc.
R&D Engineer
28 Corporate Drive
Clifton Park, NY 12065-8662
Phone: 518-371-3971 x109
On Fri, Apr 30, 2010 at 3:55 PM, Carson Brownlee
<
>
wrote:
>
Hi David sorry for getting back to you so late,
>
I believe that will be fine, adding one if statement shouldn't kill
>
performance.
>
I believe the intention was to have a modular pluggable pipeline but
>
that it was never implemented so you're right that changes to RTRT would
>
have to be made.
>
>
could you explain what your code does? if I understand it correctly you
>
essentially added a second call to display the image which means it
>
renders twice? Or am I way off?
>
>
however could this have fit into the old code by waiting for an update
>
through a transaction instead of a custom image display? The problem
>
you were having was that you were adding transactions in the image
>
display, then it would render, then update those transactions, then
>
render and then finally display correct? Is there something preventing
>
you from putting a lock in a transaction itself and then unlocking it
>
when you want to submit scene updates much as you do now? That way you
>
could run the scene updates as soon as you unlock the transaction and
>
not have to wait for an extra render cycle. I'm not sure since I
>
haven't done this myself but was thinking of trying this out.
>
Carson
>
>
On Tue, 2010-04-27 at 18:45 -0400, David E DeMarle wrote:
>
> So I'm getting back to this now.
>
>
>
> I could not think of a way to get the image I ask for out of manta's
>
> two stage pipeline without calling render twice. The caption here and
>
> in the paper imply that Manta's two stage pipeline is configurable,
>
>
>
> http://mantawiki.sci.utah.edu/manta/index.php/Manta_architecture
>
>
>
> but I couldn't figure out how to reconfigure it other than by changing
>
> RTRT.cc.
>
>
>
> Attached is my proposed change. It doesn't speed up vtkManta by two as
>
> I hoped (likely because the render threads are idle while Display is
>
> happening) but it does speed it up significantly. I don't think it
>
> will appreciably slow anything else down either.
>
>
>
> MantaBenchmark -threads 8 -triangles 1000000 -fuzziness 0.0 -noChanges
>
> prechange rendering rate = 52.6fps
>
> postchange rendering rate = 70.1fps
>
>
>
> Let me know if you've got a better alternative, or I will check mine
>
> in later this week.
>
>
>
> thanks,
>
>
>
> David E DeMarle
>
> Kitware, Inc.
>
> R&D Engineer
>
> 28 Corporate Drive
>
> Clifton Park, NY 12065-8662
>
> Phone: 518-371-3971 x109
>
>
>
Archive powered by MHonArc 2.6.16.