I have been recently engaged in extensive research creating CUDA-enabled Nuke nodes. Most of the time, the most efficient way to process a frame is to send the entire thing to the GPU to be processed on the very first _engine call, and then to hand Nuke the scan lines from the processed buffers as they are requested.
All of this works very well, with one exception -- cleanup. It's rather easy to detect the very first _engine call, but how does one detect the *last* _engine call? The _close method is unreliable, and it's not possible to count scan lines, because one has no idea how many Nuke is going to request. How does one determine/detect when Nuke is *finished* rendering a frame, and to therefore know that the resources required for frame processing can be released?
_______________________________________________ Nuke-dev mailing list [email protected], http://forums.thefoundry.co.uk/ http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
