On 18.01.2021 04:41, Clemens Eisserer wrote:
So sure, for GPU limited cases this won't help a lot - however, typical java2d is usually CPU limited with tons of very small primitives and many state changes in between.
It does not affect so much the xrender which do the same CPU related steps, but this speeds up so !much! OGL if executed in parallel. I think the benchmark is cheating here, it does not call XSync nor glflush for each operation, so it just checks the speed of pushing the data to the buffer, and not measured the actual rendering for both pipelines, unlike the software-based rendering. -- Best regards, Sergey.