Fastest way to npoints() inside a dopnet

[WLVL]

I am trying to optimize a heavy pop sims, I would like to switch some operations OFF whenever there are no more particles in the sim. ( all died etc )

First thought is to have a switch DOP with a npoints() function in it. But by checking the performance monitor it seems to have an overhead.

Doesn't DOP already have the number of points being process available somewhere? So that I can grab that info without making the npoints() Hscript Expression work?

Note that I only need to know if empty or not, not the actual point count. 

Any pointers appreciated Thanks!

 

 

 

 

switch_when_popempty_v01_recovered.hip

Edited October 18, 2023 by WLVL
attached example file

[dleonhardt]

Instead of processing the input of your Switch node with npoints(), you can provide the POPObject directly with:

npoints("popobject")

 

But why wouldn't you just paste the expression into the activation parameters and ditch the switch completely?

Edited October 18, 2023 by dleonhardt

[WLVL]

Thanks Dennis!

Seems like the performance still gets a hit even when using npoints() to the popobject.

 

The reason for using a switch is that, in the real scene, I have many nodes to be deactivated when no points are present, and linking the "activation" parm for all of them becomes impractical

[dleonhardt]

The kind of comparison you are doing here is a bit disingenuous, you just have Switches with no other nodes inside the networks. For a better comparison you'd actually need to have POP nodes which do something alongside your Switches and compare it against a network with only POPs and no Switches.

If disabling big parts of your DOP network for optimizations sake is the main concern, then try Enable Solvers.

dl_switch_when_popempty_v01_recovered.hipnc

[WLVL]

The comparison I have performed is to analyze the overhead caused by the "empty simulation check" , which is performed by the npoints() function.

If I am able to skip pop operations for empty sims, but performing that empty check adds up time to the sim, then it might not be worth it, depends on the net difference. 

So I was trying to benchmark the pure overhead caused by the npoints() check.

Maybe I badly designed the experiment and wired nodes incorrectly :) - reposting a more clear scene with some other findings

With the latest tests seems like the npoints() function itself adds very little overhead in both cases ( "popobject" or 0 as argument )

switch_when_popempty_v04.hip