Josh

It's working fine here. If you have the create tool selected you won't be able to select objects.

The biggest difficulty with getting consistent shadow map offsets is the fact that depth textures are stored in non-linear space. This is good for camera depth testing but terrible for shadows. You can see in this image the area near the light source has higher precision than the area further away from the camera. If the precision was equal for all distances it would be much easier to set or calculate an offset that would prevent the "shadow acne" artifact. My idea is to modify the gl_FragPosition.z value in the output of the vertex shader, in order to reverse the linear-to-depth transformation before it is stored in the shadow map, which will equal a linear value that can then be compared to a linear position in the shadow sampling step in the lighting shaders. I am having no luck getting this to work so far.

If you edit a map file by hand you will destroy its contents, because there is some binary data at the end.

Can't profile a video!

I wanted to talk to you guys about this topic. In the Vulkan renderer we had a transfer and render hook. I don't see any need for these two hooks anymore. Do you? What information needs to be passed to the hook? I think everything can be retrieved with functions: wglGetCurrentContext(); glGetIntegerv(GL_FRAMEBUFFER_BINDING, &fbo); At what point in the renderer should this hook be called? I am thinking after all objects are rendered, before post-processing effects. Should it be set per-camera instead of per-world perhaps?

The editor does not use a log file. Usually these don't help with crashes because once the crash occurs program execution stops. Are you having trouble with something?

I also want to note that the animations themselves seem to be coming through perfectly. The skeleton looks correct. So in the future I will be looking more closely at the skin bind matrices and initial orientation of the model, which seem wrong.

0.9.5 Update improves performance in debug build.

Okay, this was being caused by all the Assert statements I had in the source. Constructing a WString from a const char apparently is pretty slow in debug builds. Once I removed that the change in framerate was more like 4000 -> 3000 in the debug build.

That does seem to be a big reduction in performance just based on a single panel being visible. I'm going from 4000 to 400 FPS when the space key is hit, in debug mode. I don't see the same change in release mode. #include "UltraEngine.h" using namespace UltraEngine; shared_ptr<Window> window; shared_ptr<Framebuffer> framebuffer; shared_ptr<World> menuWold; shared_ptr<Interface> ui; shared_ptr<Camera> uiCamera; shared_ptr<Widget> panel; shared_ptr<Icon> icon1; shared_ptr<Icon> icon2; bool isFirst = true; int main(int argc, const char* argv[]) { //Get the displays auto displays = GetDisplays(); //Create a window window = CreateWindow("Ultra Engine", 0, 0, 1280, 720, displays[0], WINDOW_DEFAULT); //Create a world menuWold = CreateWorld(); menuWold->RecordStats(); //Create a framebuffer framebuffer = CreateFramebuffer(window); //Create light auto light = CreateBoxLight(menuWold); light->SetRange(-10, 10); light->SetRotation(15, 15, 0); light->SetColor(2); //Create camera auto camera = CreateCamera(menuWold); camera->SetClearColor(0.125); camera->SetPosition(0, 0, -3); camera->SetFov(70); uiCamera = CreateCamera(menuWold, PROJECTION_ORTHOGRAPHIC); uiCamera->SetPosition((float)framebuffer->GetSize().x * 0.5f, (float)framebuffer->GetSize().y * 0.5f, 0); uiCamera->SetRenderLayers(2); uiCamera->SetClearMode(CLEAR_DEPTH); uiCamera->SetLighting(false); //Main loop while (window->Closed() == false and window->KeyDown(KEY_ESCAPE) == false) { if (window->KeyHit(KEY_SPACE)) { if (not ui) { auto default_font = LoadFont("Fonts\\arial.ttf"); ui = CreateInterface(menuWold, default_font, framebuffer->GetSize()); ui->SetRenderLayers(2); ui->root->SetColor(0.0f, 0.0f, 0.0f, 0.0f); panel = CreatePanel(10, 50, 64, 64, ui->root, PANEL_DEFAULT); icon1 = LoadIcon("https://raw.githubusercontent.com/UltraEngine/Documentation/master/Assets/Icons/help.svg"); panel->SetIcon(icon1); } } window->SetText("FPS: " + String(menuWold->renderstats.framerate)); menuWold->Update(); menuWold->Render(framebuffer, false); } return 0; }

I believe this is fixed now.

Would also disable your directional light for testing.

0.9.5 Prefabs are back in, fixed reported bug in these.

I would also be looking at the rest of the render statistic values.

Adding the prefab works now, but still skipping if the prefab is in a map file...

At the moment this isn't really actionable information. What is different about your demo and the performance benchmark apps? Are you constantly streaming geometry in?

Can confirm, here is a test file that shows the problem test.zip

I've poked around in the file but I can't find anything that's unique about it that would be causing a problem when many other animated glTF files load correctly. There is a problem in my loader, because both Windows Object Viewer and the Babylon.js glTF viewer both load it correctly, but I don't know how to solve it. I recommend running this through Blender or whatever other method you were using that makes it export a model Ultra can load correctly. I am attaching the model here for future investigating, but I don't plan to continue working on this for now. Human_2.zip

If I drag him into a viewport, then scale and rotate him he actually looks fine. However, the moment animation is applied, he turns into The Thing:

Vulkan validation layers were quite slow, and there's also more code being executed in the driver in OpenGL. So the results you are posting right now are not surprising.

Spiderpig game model.

Typically your file format should start with some sort of identifier like "SPGM" for the first four bytes or so. This is usually how binary file formats are identified.

Preview.exe has been rebuilt and may work better now.

This may be fixed in 0.9.5 now.

Problem is resolved in 0.9.5, on the beta branch.