Epic: replace the hardcoded forward renderer with a graph-driven hybrid renderer #781
Labels
No labels
Context: Bug
Context: Enhancements
Platform: Android
Platform: Linux
Platform: Web
Platform: Windows
Platform: iOS
Platform: macOS
Release
Target: CI
Target: CLI
Target: Dependency
Target: Engine
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
SakulFlee/Orbital#781
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Workstreams:
(API break allowed; all examples/templates updated in the same change)
Acceptance:
Implementation status (2026-10-29): All workstreams are implemented and merged. Headless pipeline/layout validation is covered by the noop-backend tests (#965). Remaining before closing: real GPU visual and performance validation.
Refractor main rendererto Rework Renderer / Introduce Hybrid Rendering approachRework Renderer / Introduce Hybrid Rendering approachto Epic: replace the hardcoded forward renderer with a graph-driven hybrid rendererNo shader graph. Node-based shader composition stays; the missing piece is a surface/material contract.
Phase 1 work is split into three PRs:
refactor(renderer): extract render scene and own frame resources— ECS→renderer snapshot, renderer-owned fallbacks and world bind group, no raw model pointers.perf(renderer): resolve timestamps in the scene submission— one scene submit instead of render + timestamp submit. Stacked on #933.perf(renderer): reuse instance and cull buffers— in-place instance updates, cached bounds, persistent cull buffers. Stacked on #933.#934 and #935 will be retargeted to
mainonce #933 lands. Local fmt/clippy/tests pass for the affected crates; the Forgejo runners are currently failing before cargo starts (no space left on device / missing actiondist/index.js).Additional phase-1 PR: #936
perf(renderer): cache the world bind group— keyed by camera/light/environment/shadow identities, built after the shadow pass, with the keyed Arcs kept alive to avoid address-reuse false hits. Also stacked on #933.Phase 2 is split into three stacked PRs:
feat(renderer): add frame graph and persistent texture pool—orbital_render_graph, dependency-ordered passes, pass timings/debug dump/DOT, descriptor-keyed texture pool, pooled depth target. Base:main.feat(renderer): render the scene into HDR and tonemap in post— HDR scene target, ACES tonemap pass, built-in PBR/sky shaders now output linear HDR, 3D pipelines targetRgba16Float. Stacked on #937.feat(renderer): add FXAA post pass—AntiAliasing::{Off, Fxaa}(FXAA default), pooled LDR intermediate, fullscreen FXAA. Stacked on #938.SMAA remains the optional follow-up for #585. The 2D/UI/text/overlay paths still render after tonemapping, so UI stays sharp.
Phase 3 PRs:
refactor(material): split shader programs from material instances—ShaderProgram+ pass-awarePipelineCache, layout/variable split inShaderDescriptor, shared pipeline cache resource, material render state (AlphaMode, blend, depth write/compare, render queue, shadow flags), and a fix for the eager material-cache path. Base:main.feat(renderer): add sorted transparency and PBR alpha modes— PBR alpha from albedo + Opaque/Mask/Blend, glTF alpha modes, sorted transparent pass, camera/model position data for sorting, shadow filtering for non-casting models. Stacked on #941.G-buffer/deferred pass variants and the
Surfacecontract are next.Implementation complete across the renderer PR series (#937, #939, #942, #948, #953, #954, #955, #956, #958, #960, #962, #963, #964, #965). Workstreams are checked off. The epic stays open only for real GPU visual and performance validation; close it once that passes.
All workstreams are implemented and merged. Real-GPU visual validation has passed, and the remaining performance acceptance is covered by the live frame/stage timings in the debug UI. Closing the epic; any future renderer tuning should be tracked as a separate issue.