Hello Gordon,
On 13. Jan 2008, at 1:35, Gordon_Erlebacher wrote:
> Here is the configuration.
Looks good. I have a couple of comments below which should improve
performance significantly. I'ld love to test the config myself,
but I don't have a hw setup for it right now.
>
> It works except for the fact that the colors on my mac are wrong.
I suspect that this is an issue of the glX indirect rendering. Is your
Mac a ppc machine?
>
>
> I have a mac (at home), we'll name it mac.
> I have a HP (linux) at work, remotely connected, called "spock",
> with two GPUS.
> I have modified the configuration file 2-pipe.DB.eqc so that the
> image is assembled on channel1 of
> spock, and then sent to channel3 on my mac with a color buffer.
> Perhaps there is a mistake.
> Can you please try this out? It does not run any faster (although it
> should) than a configuration
> where I take the results of channel1 and channel2 and assemble that
> on my mac (channel3).
>
> # two-pipe sort-last config for dual-GPU machines
> # not faster than compositing on my home machine
> #
> # channel1 (frame.channel1) + channel2 (frame.channel1) -> channel1
> (frame.out12)
> # channel1 (frame.out12) -> home mac (frame.out12)
> # colors are wrong, although each piece has a single color
>
> server
> {
> config
> {
> appNode
> {
[...]
>
> }
> compound
> {
> channel "channel3"
> # without the next line, the screen does not refresh when
> I rotate the
> # image. I do not understand why
> task [ CLEAR ASSEMBLE ]
CLEAR is by default only executed on the leaf compound channels. If
you don't
clear, the results accumulate over time. The normal case is that the
destination
channel is also used as a source channel, which does the clear.
>
> buffer [ COLOR ] # adding DEPTH creates problems
Not only that - it is also bad for performance.
>
> wall
> {
> bottom_left [ -.32 -.20 -.75 ]
> bottom_right [ .32 -.20 -.75 ]
> top_left [ -.32 .20 -.75 ]
> }
>
> compound
> {
> channel "channel1"
> buffer [ COLOR DEPTH ]
>
> compound
> {
> range [ 0 .5 ]
> outputframe {}
This output frame, and the corresponding input frame below, are not
needed since the assembly happens on channel1 and the data is already
'in place'.
>
> }
> compound
> {
> channel "channel2"
> range [ .5 1 ]
> outputframe {}
> }
> inputframe { name "frame.channel1" }
Remove this frame - see above.
>
> inputframe { name "frame.channel2" }
> outputframe { name "frame.out12" }
This output frame contains color+depth, since this is the setting
for this compound. Equalizer does not yet detect that it only
needs color on the destination input frame, and therefore reads
back and transfer the depth as well. Use this line instead:
outputframe { buffer [ COLOR ] name "frame.out12" }
> }
> inputframe { name "frame.out12" }
> }
> }
> }
>
_______________________________________________
eq-dev mailing list
[email protected]
https://in-zueri.ch/cgi-bin/mailman/listinfo/eq-dev
http://www.equalizergraphics.com
