Abstract:

This live stream details the ongoing development of a new renderer for the Hazel Engine, specifically focusing on the complex transition from a Vulkan-based backend to NVIDIA's Rendering Hardware Interface (NVHI). The developer, Cherno, outlines his strategy of "Vertical Slice Driven Development" (VSDD), prioritizing getting the core editor UI (ImGUI) functional with NVHI before tackling more complex 3D scene rendering. The stream documents extensive debugging and troubleshooting, including resolving shader compilation errors (HLSL pre-processing, entry points) and addressing numerous runtime issues related to uninitialized components and legacy code paths. Despite encountering significant technical hurdles, the video concludes with progress toward a basic functional rendering loop for the UI, setting the stage for future incremental integration of the engine's full rendering capabilities, with the ultimate goal of releasing Hazel Engine for free.

Working on New Renderer // Hazel Engine Stream (19.05.2025)

• 0:00 Introduction & Setup Challenges: Cherno begins a live stream, acknowledging initial "rustiness" with his streaming setup, including reconfiguring cables and addressing audience concerns about audio and video resolution (720p vs. 1440p).
• 1:50 Project Roadmap & Stream Logistics: He informs viewers that past streams will be exported to his "Cherno Unplugged" YouTube channel (though currently failing) and directs them to hazelen.com/roadmap for the "Render 2025" project, which aims to release Hazel Engine for free. He also humorously notes his recent 30th birthday.
• 4:15 Renderer Migration Strategy - Vertical Slice Driven Development (VSDD): Cherno recaps the challenge of porting the renderer backend from Vulkan to NVHI and introduces his "Vertical Slice Driven Development" (VSDD) philosophy. This approach, which he previously discussed in a conference talk, emphasizes getting a complete, though simplified, pipeline working end-to-end first, then iteratively refining and expanding functionality.
• 10:15 Initial Implementation Focus: ImGUI with NVHI: The chosen strategy is to initially strip back the rendering to focus solely on getting the Hazel editor's UI (built using ImGUI) working with NVHI. This involves using NVIDIA's "Donut" framework's ImGui_NVRHI implementation, which is simpler as it doesn't require immediate support for advanced features like ImGUI viewports (dragging windows out).
• 11:45 NVHI as a Low-Level Abstraction: Cherno clarifies that NVHI (Rendering Hardware Interface) is an abstraction layer over low-level graphics APIs like DirectX and Vulkan, providing some higher-level features (e.g., automatic barriers) but still offering fine-grained control, unlike a high-level renderer.
• 1:11:45 Debugging Shader Compilation: He begins integrating the ImGui_NVRHI code, encountering compilation and linking errors. A major hurdle involves the HLSL shader compilation process, specifically issues with "missing entry point definition" due to how pragma stage directives were pre-processed and how entry points (main) were expected. After extensive debugging, he resolves the shader compilation by adjusting the shader pre-processing logic.
• 1:57:03 Addressing Runtime Issues: With shaders compiling, the stream shifts to tackling numerous runtime errors. These include issues with uninitialized components (e.g., render 2D init, thumbnail generator), null pointer exceptions, and conflicts with legacy Vulkan-specific code. He temporarily disables non-critical engine components (like the thumbnail generator and scene renderer) to isolate the core UI rendering loop.
• 2:22:14 Resolving Window Context & User Pointer Issues: A persistent crash related to get window context is traced back to the window's user pointer being incorrectly reset, which he fixes by ensuring the device manager's context is correctly maintained.
• 2:28:50 Current Status & Next Steps: The stream concludes with the application successfully launching a window, though 3D scene rendering is still disabled. The immediate next step is to ensure ImGUI can render frames reliably within this new NVHI-based setup. Subsequent work will involve incrementally re-enabling and porting Hazel's full 3D renderer components, aligning with the VSDD approach.