PDC on folder-tracks disables multi-threading (test file included) (FIXED)

[HighVoltage]

Please if you have 1 minute, test this for me, cause i think i'm going insane here.

The thing basically is:

If you have a plugin that has any amount of PDC on a folder-track, it will send all the processing to a single core, thus easily and unnecessarily bottlenecking your system.
It ONLY applies with record armed tracks (or if you have anticipative fx processing OFF)

Here i have a bunch of ReaXcomps routed to a folder track. CPU is evenly distributed on 4 cores. Everything is paying nice.

http://prntscr.com/9ay2dw

As soon as i enable a plugin with PDC on the folder, in this case a reaInsert - (notice it's only 1 sample delay) All processing jumps to one thread.

http://prntscr.com/9ay32j

This is very very bad on real life scenarios, for example if you set up a multi-output drum kit from kontakt, and you use some plugins on the BUS that has PDC. (again: even 3 samples breaks it)

It's not related to my config either,i have tested it on 3 different machines with the DUMMY audio driver too.


Attached a very basic file to test it, you just have to watch your CPU meter while enabling the ReaInsert on the folder track.
(You might have to ease up on the number of Xcomps if you have a slower CPU)

-----------
I had another thread about this here:
http://forum.cockos.com/showthread.php?t=164801
but now i tracked the bug to it's core with only built-in plugins, so anyone can test if it occours.
-----------

[PitchSlap]

Enabling the folder plugin causes my RT CPU to go from around 70% to 250%.

Though strangely in the resource monitor the CPU usage reported goes down noticeably when enabling the folder plugin.

[dimitris]

Confirmed. CPU usage quadrupled, while threads reduced from 4 to 1.

[Justin]

Duplicated, have a fix but need to save it for a major release cycle. For now make sure you use anticipative FX for this scenario

[HighVoltage]

Ooooh yessssssss!!! Can't wait

[ferropop]

Hahah I've bumped like 5 different versions of this exact post in the past few weeks. This gets even worse with nested folders, and with sidechain send/receives.

SO glad to see this get ironed out, please keep us informed Schwa/Justin. Huge killer of the amazing routing possibilities in Reaper.

[dimitris]

Originally Posted by Justin

Duplicated, have a fix but need to save it for a major release cycle. For now make sure you use anticipative FX for this scenario

That's great news! Many thanks!

[HighVoltage]

I wonder what a major release cycle suggests though. Is it version 6?
Cause that seems to be pretty far away

[ferropop]

Sure hope sooner, it explains so many performance issues I and many people have been having.

Defeats all the benefits of using folders in the first place sadly

[serr]

Jeeze, I wonder if this explains the random behavior with those suspect plugins that seem to crash PDC? This sounds like a real good find!

Is this just an issue with folders or does it also follow standard sends/returns?

[mlprod]

This will be VERY interesting to follow. Hopefully for 5.7.

[Justin]

We'll probably put it in some 5.Xpre builds to test, though it might not get into a release until v6, depending on the feedback from testing and what unintended side effects there are.

[Mercado_Negro]

Count me in for the testing. This problem has given me headaches for years!

[ferropop]

Let's get this in the next Pre ! I will test this for 48,000 hours straight.

[ivansc]

(grin) Lemmeseenowwww... 48,000 divided by 24.... divided by 365.... Allow for meal breaks, sleep breaks, bathroom breaks, sick days....vacations....

How old did you say you were again?

[HighVoltage]

Edit: moved to the right thread

[EvilDragon]

You should post that to the prerelease thread, not here.

[rasmusmarshagen]

I know this is an old thread, but is this fixed? I’m experiencing exactly this, the whole signal chain of tracks with anticipative fx turnt OFF will bottleneck to one thread/core. Increasing the RT cpu badly

Anyone got a fix? Reainsert for example needs to be used without the apfx (on just that track),- it’s going to force ALL the tracks it’s feeding to end up on the same thread causing massive RT cpu load. It’s not a problem if it’s the last track I use in a serial chain of course.

Have I missed the fix for this?

[ferropop]

Also wondering, is this rolled up in all the PDC stuff we've been seeing in recent pres?