1489 2024-01-22 02:44:50 +0100 GraphUI Group: Resolve Performance Regression in Scene.pickShape(): Drop invisible and clipped shapes 2024-02-21 13:15:37 +0100 1 1 3 JogAmp GraphUI Core 2.6.0 All all RESOLVED FIXED P1 critical --- 805 1 sgothel sgothel DEFECT 5f9fb7159fa33bc979e5050d384b6939658049bd 19fac36ae64ffb219fb40449b537219d74a1f000 8bb2f6dec8ab731b07387b947715fa1959c680e4 --- oldest_to_newest 7050 0 sgothel 2024-01-22 02:44:50 +0100 After implementing Bug 1487 (Frustum Clipping/Culling) and using thousands of shapes within one Group mostly culled (outside of frustum), overall mouse Scene.pickShape() caused tremendous lagging. This is caused by Scene.pickShape() traversing through _all_ shapes, rendered ones, invisible ones and culled ones. +++ Solution is to have Scene and Group provide a pre-sorted list of actually rendered shapes, i.e. isVisible() and not culled. Scene.pickShape() can now use this reduced and pre-sorted list reducing the load considerably. 7056 1 sgothel 2024-01-22 06:36:28 +0100 commit 5f9fb7159fa33bc979e5050d384b6939658049bd Bug 1489 - GraphUI Group: Resolve Performance Regression in Scene.pickShape(): Drop invisible and clipped shapes implemented as described. +++ Further - cleanup TreeTool - rename Container methods: -- setFrustumCullingEnabled() -> setPMvCullingEnabled() -- isFrustumCullingEnabled() -> isPMvCullingEnabled() - supply Container with -- isCullingEnabled() -- List<Shape> getRenderedShapes() -- isOutside() -- isOutside2() -- forAllRendered() 7063 2 sgothel 2024-01-25 08:55:56 +0100 19fac36ae64ffb219fb40449b537219d74a1f000 Scene.dispatchMouseEventPickShape(): Reuse PMVMatrix4f, Ray and Vec3f within EDT, reducing temp objects on mouse moves 7147 3 sgothel 2024-02-21 13:15:37 +0100 commit 8bb2f6dec8ab731b07387b947715fa1959c680e4 Bug 1489: Lock-Free Double-Buffered 'renderedShapes' causes data-race between rendering & input-edt, use synchronized tripple-buffering Tripple-buffering _almost_ produces zero data-race collisions, however .. it still does rarely -> hence synchronize on the used ArrayList<>. This adds a minimal chance for blocking the input-EDT, but gives correct code & results. Double-buffered 'renderedShapes' was introduced to resolve Bug 1489 in commit 5f9fb7159fa33bc979e5050d384b6939658049bd This solution is tested by passing '-swapInterval 0' via CommandlineOptions for FontView01, UIMediaGrid01 .., i.e. rendering faster than picking and hence provoking the data-race condition.