Here's a simple test program using LibQGLViewer (screen capture with xvidcap). The vertex array and the index array that OpenGL draws (using glDrawElements) are held in a GLData class which holds a mutex. The Viewer class locks the mutex while drawing, and the worker-thread locks the mutex while updating the data. Here the worker task re-positions the original vertex position, signals the Viewer to draw, and then sleeps for 40 ms. When we don't rotate or zoom with the mouse the frame-rate should thus be 25 Hz. Rotating or zooming causes more frequent re-draws and a higher frame-rate.
To gain any real benefit on a multi-core machine I think the worker thread needs to work on a 'dirty' copy of the data, and we only lock the mutex for a minimal time while swapping in the updated data for the real data. Anyone have any good example code for this? Both a case where the worker produces new data at a slow rate (slower than Viewer re-draws), and at a faster rate should be handled.
Here's an UML(ish) diagram drawn with dia:
One thought on “Threading and OpenGL, test 2”
This looks interesting: http://software.intel.com/en-us/articles/multithreading-the-rendering-pipeline-for-3d-model-animation/