Flex, 45290, Big Performance Hit using actor feedback

Not sure if this is a bug, or just a fact of life.

When utilizing the Actor COMP Flex feedback TOPs I see a pretty big hit to my whole system. This seems to scale with how strapped the rest of my network is for resources, which makes me wonder if the performance monitor’s line for the flex solver is catching other things. When i look at Perf. Monitor, I am seeing the flex solver’s supposed cooktime increase greatly when I have TOPs in the feedback position and velocity pars, as well as a general performance drop.

I am utilizing the work I am doing to feed those TOPs in other parts of the system, so that stuff is all still cooking even if not being fed back in.

So main question is, is it a “bug” or an unavoidable consequence of using the Flex feedback feature?

Will try to prep a file to demonstrate, but wanted to float the question in case Eric perhaps can shed light generally.

Thanks
P

It’s doing a CPU copy to feedback the values back into the simulation. I imagine that’s what’s causing the performance hit. Out of curiosity, how many particles are you feedbacking?

Hmm, gotcha, is there no way around the CPU for this under the hood, my knowledge of flex is limited? I love the ability to add my own simulation stage, but the hit I’m getting is heavy enough to reconsider my approach unfortunately.

I’ve been doing a 64x64 texture, as well as a 128x128 texture in a simpler test.

Sorry for the delay. I did find a way to avoid the CPU copy, which should provide improved performance when feedbacking. Just working out some issues but it should be ready for the next experimental build.

Oh that’s exciting, thanks eric, happy to test when its available!

Cheers
P

Here’s a build with the changes:

Let me know how that works for you.