Tile Branch is in Trunk

As Jeroen Bakker post in mailing list, the tile branch is complete.
All nodes are converted to use multicore cpu.

Revision: 46092
Author: jbakker
Date: 2012-04-29 21:47:41 +0000 (Sun, 29 Apr 2012)
Log Message:


  • Added the crop node
    • still needs some work with the crop image size
  • Added the Double Edge Mask
  • Changed the max tile size of the compositor

These were the last two nodes. now let’s start with the clean up and bug fixing…

  • At Mind -
    Jeroen & Monique.

There is a bugtracker for tile branch as there are a few small bugs to report.
But you can post bugs here to verify and somebody put in tracker (Me?).

A very new for windows and some older for linux and OSX but I think newer arrived soon.


First bug to verify: Bright/Contrast node turns all B-White

Cheers, mib.

Awesome news! Is this in trunk yet?

@Fahr, ofcourse not. Jeroen will (as stated in the message above) now start bug hunting, and cleaning up the code.
it is planned to get it into 2.64 though :smiley:

I forgot, if you do building by your self, here is the tile branch:


And it is fast as hell :slight_smile:

if this is openCl do we have to have opencl selected for compute device in the user preferences
for this to work or will it automatically find the card and work if you have one?

I’m really impressed. The compositor is ridiculously fast.

Great news :slight_smile: A 2.64 project so early finished.
But what about OpenCL? Does it work already? I can’t see any speed difference when I check OpenCL.

OpenCL is not working yet. I think the plan is to have 1 or 2 heavy nodes (like Defocus) to make use of OpenCL at the beginning.

@Jeroen Bakker: You are amazing! Thus far problems = 0!

But the tile rendering is GPU accelerated, right? Isn’t that the whole idea or is it ‘just’ an optimized rewrite of the code? I mean, if it’s not GPU accelerated, howcome it’s so fast? #lostinspace

well… the thing is that the old compositor was/is horrible in the way it handled memory. tile branch apparently handles the computers resources in a better way

It is both.
Very much optimized and multithreaded code (which makes it much faster) and is going to support GPU in due time (which will make it even faster). = Awesome!
So I believe anyway. Anyone to correct me if I am wrong? :slight_smile:

@ktysdal & ejnaren

Aah, I thought the whole thing was about GPU acceleration as the demo specifies CPU vs Nvidia 5xx something… But either way, very cool. And very interesting with OpenCL support, especially if Brecht & co. get Cycles running better with AMD & OpenCL, seeing Nvidia seems to have scrapped computation performance in their never cards…

…and just amazing to see how Blender developed these last years and it doesn’t seem to slow down either… Wow!

im getting nothing but crash bang with the tile branch…
im using the tungerz build.

I believe that opencl will require further development currently it is multi-threaded

i’d love to see some Linux 64-bit builds on graphicall, if anyone would be so kind! i tried to build it, but i’m getting some errors with scons that i don’t understand.

edit: never mind, i got it working with cmake. SNAP, that IS fast! niiiiiiiice.

tried tungerz build and demohero build both crash…
i think this needs more work before it goes primetime…

i’m getting an error message “unknown conversion detected”, does anyone know what that might pertain to?

@holyenigma74, don´t work with old node setups atm. sometimes they don´t work sometimes they crash.
The sense of this thread is to find bugs, if you got one make a simple file and post a bug report or post the file here.
I play all day yesterday and never got a crash.
There is a fund at the doku page I posted and it is not finished.
About OpenCL, read DingTo´s post.

Cheers, mib.
EDIT: I post a bug in my first post somebody could verify?

Wow, really great news. I hope we can import old node groups in the future.