Take a qmlbench case like benchmarks\auto\changes\changing_over_isolated_with_clip_rotated.qml. Here the clipping disables merging so we end up with ca. 4000 merged/unmerged batches.
The d3d11 backend performs 10x (compared to gl) and 20x (compared to vulkan) better. This seems to be caused by calls to QRhiPassResourceTracker::trackedRegisterBuffer()
- is required for
-
QTBUG-78587
Identify and possibly fix performance issues with qmlbench on the QRhi based rendering path in Qt Quick
-
- Closed
-
- relates to
-
QTBUG-78792
Varied performance in scenes with > 1000 unbatched geometry nodes
-
- Closed
-
- mentioned in
-
Page Loading...