Skip to content

gfx: scene-concept rework (the coupled core of #2109) - #2120

Open
jcelerier wants to merge 10 commits into
planc/gfx-infrafrom
pland/scene
Open

gfx: scene-concept rework (the coupled core of #2109)#2120
jcelerier wants to merge 10 commits into
planc/gfx-infrafrom
pland/scene

Conversation

@jcelerier

@jcelerier jcelerier commented Jul 14, 2026

Copy link
Copy Markdown
Member

The scene-concept rework — the coupled core of #2109. This is the part that couldn't be decomposed further: a scene-aware render pipeline plus the GPU-resident scene state, node-lifecycle, and per-backend plumbing that all change together. Everything separable was already pulled into #2117/#2118/#2119 to make this PR as small as it can be.

Rebased onto current master (auto-merges cleanly with master's newer nodal-selection work). Stacked on #2119; retargets to master as the stack merges.

Commits

  • gfx: add the OffsetAllocator submodule (arena backing for GPU-resident scene data).
  • gfx: rework the render pipeline to be scene-aware and incrementally rebuilt — SceneGPUState, the GpuResourceRegistry arena, incremental RenderList rebuild, and the QRhi pipeline cache.
  • gfx: add the scene preprocessor and scene filter nodes.
  • avnd: make the CPU and GPU node lifecycles scene-aware.
  • js: rework the GPU node lifecycle with deterministic teardown.
  • threedim: adapt ModelDisplay to the new pipeline API.

This wires up the caching/offscreen infrastructure landed in #2119.

Validation

  • Builds cleanly against current master (all plugins).
  • ctest: 17/17.
  • Process + device factory sweeps under ASAN + UBSan (the sweeps construct and round-trip every reworked gfx node model): 0 memory errors, 0 undefined-behavior in score's own code.

🤖 Generated with Claude Code

@jcelerier jcelerier mentioned this pull request Jul 14, 2026
Open
@jcelerier
jcelerier force-pushed the pland/scene branch 2 times, most recently from 6135c95 to 726e579 Compare July 17, 2026 11:58
@jcelerier
jcelerier force-pushed the planc/gfx-infra branch 2 times, most recently from 1b6651e to 08516ad Compare July 17, 2026 14:10
jcelerier added a commit that referenced this pull request Jul 17, 2026
@jcelerier @claude
gfx(#2120): fix 5 libisf importer bugs + render/scene test harnesses
5c79ed7 
FIX-ISF (libisf/src/isf.cpp — applies on top of this branch's ISF/CSF
feature rework, hence #2120 not #2119):
 - write_isf: seekp(0,end)-before-tellp corrupted the emitted descriptor
 - use-after-move on m_sourceVertex -> guard with .empty()
 - parse_shadertoy_json: GLSL45 preludes + iMouse vec4(0,0,0,0)
 - glsl_sandbox: duplicate-TIME uniform guard
 - replace_identifier: word-boundary match

Tests: test_unit_isf_importers (all 5 importer branches); golden-image
render regression (16 pinned JS-corpus cases, llvmpipe, SSIM/PSNR);
timeline-driven scenario ramp; gfx resource/leak soak (DISABLED — guards
a still-unfixed render-clock teardown UAF); live-edit churn corpus.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
jcelerier added a commit that referenced this pull request Jul 17, 2026
@jcelerier @claude
gfx(#2120): fix 5 libisf importer bugs + render/scene test harnesses
3d4ffb7 
FIX-ISF (libisf/src/isf.cpp — applies on top of this branch's ISF/CSF
feature rework, hence #2120 not #2119):
 - write_isf: seekp(0,end)-before-tellp corrupted the emitted descriptor
 - use-after-move on m_sourceVertex -> guard with .empty()
 - parse_shadertoy_json: GLSL45 preludes + iMouse vec4(0,0,0,0)
 - glsl_sandbox: duplicate-TIME uniform guard
 - replace_identifier: word-boundary match

Tests: test_unit_isf_importers (all 5 importer branches); golden-image
render regression (16 pinned JS-corpus cases, llvmpipe, SSIM/PSNR);
timeline-driven scenario ramp; gfx resource/leak soak (DISABLED — guards
a still-unfixed render-clock teardown UAF); live-edit churn corpus.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
jcelerier added a commit that referenced this pull request Jul 17, 2026
@jcelerier @claude
tests(#2120): Text-node render validation harness (P3-text)
019649f 
Headless llvmpipe golden+value harness for the Gfx::Text process: grabs the
untouched-defaults frame (hard off-screen-default visibility assertion — guards
the #2119 default-position fix), then live-edits text/font/size/position/scale/
color over OSC /script and asserts pixel VALUES per case (coverage, bbox
ordering across sizes, centroid movement, channel dominance, blank empty
string, unicode/CJK/tofu/2000-char) + golden refs (compare.py strict).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier
jcelerier force-pushed the pland/scene branch 2 times, most recently from 019649f to cb34e12 Compare July 18, 2026 13:37
@jcelerier
jcelerier force-pushed the planc/gfx-infra branch 2 times, most recently from 2676d67 to 387b649 Compare July 18, 2026 14:17
@jcelerier
jcelerier force-pushed the pland/scene branch 2 times, most recently from bd66e34 to 0da070a Compare July 19, 2026 21:59
@jcelerier
jcelerier force-pushed the pland/scene branch 4 times, most recently from 9e6753d to f4015d0 Compare July 22, 2026 06:22
jcelerier added a commit that referenced this pull request Aug 16, 2026
@jcelerier @claude
tests: sweep the whole tester corpus and say what came out
a75cd5d 
One process per case, so a case that crashes or hangs takes only itself down,
and each tester builds its document from a clean application either way.

Verdicts are separated by what actually happened rather than lumped into
pass/fail: RENDER, BLANK, NORENDER, SCREEN, FAIL. SCREEN exists because a grab
that falls back to capturing the desktop is not a weaker result, it is a
meaningless one, and the sweep exits nonzero if any case produced one.

First run, 130 cases: 81 render on the stack against 60 on master, the
difference being shaders whose ISF features only exist above #2120 -- the
measured form of the claim that no pixel test can run below it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
jcelerier added a commit that referenced this pull request Aug 16, 2026
@jcelerier @claude
tests: sweep the whole tester corpus and say what came out
9a4bf1a 
One process per case, so a case that crashes or hangs takes only itself down,
and each tester builds its document from a clean application either way.

Verdicts are separated by what actually happened rather than lumped into
pass/fail: RENDER, BLANK, NORENDER, SCREEN, FAIL. SCREEN exists because a grab
that falls back to capturing the desktop is not a weaker result, it is a
meaningless one, and the sweep exits nonzero if any case produced one.

First run, 130 cases: 81 render on the stack against 60 on master, the
difference being shaders whose ISF features only exist above #2120 -- the
measured form of the claim that no pixel test can run below it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
jcelerier added a commit that referenced this pull request Aug 16, 2026
@jcelerier @claude
tests: sweep the whole tester corpus and say what came out
f9e4c96 
One process per case, so a case that crashes or hangs takes only itself down,
and each tester builds its document from a clean application either way.

Verdicts are separated by what actually happened rather than lumped into
pass/fail: RENDER, BLANK, NORENDER, SCREEN, FAIL. SCREEN exists because a grab
that falls back to capturing the desktop is not a weaker result, it is a
meaningless one, and the sweep exits nonzero if any case produced one.

First run, 130 cases: 81 render on the stack against 60 on master, the
difference being shaders whose ISF features only exist above #2120 -- the
measured form of the claim that no pixel test can run below it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier
jcelerier force-pushed the pland/scene branch 2 times, most recently from 671136f to 119e5ef Compare August 16, 2026 19:57
@jcelerier
jcelerier force-pushed the pland/scene branch 2 times, most recently from 4e0f9ff to 29bed8a Compare August 17, 2026 13:30
jcelerier and others added 10 commits August 17, 2026 09:50
@jcelerier
gfx: add the OffsetAllocator submodule
532d221 
Bring in OffsetAllocator to back the slab-allocated GPU arenas.

(cherry picked from commit 5b7c7fb)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier @claude
gfx: rework the render pipeline to be scene-aware and incrementally r…
3e38bd4 
…ebuilt

The core scene-GPU rework. Adds the supporting infrastructure — OffsetAllocator-backed
GPU arenas (GpuResourceRegistry), a shared decoded-asset cache (AssetTable +
TextureLoader), scene math and pipeline-state helpers, the flat scene state and packer
(SceneGPUState), an offscreen RHI device and preview widget — then reworks the pipeline
on top of it: the ISF parser and nodes gain 3D samplers, uniform inputs and
geometry/vertex stages; the compute and raster pipelines become scene-aware with multiple
render targets and auxiliary outputs; the render graph updates edges, render targets and
passes in place instead of rebuilding wholesale; and the filter, texture-port, window,
capture and video-output paths follow the new node and renderer interfaces.

(cherry picked from commit b824245)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier
gfx: add the scene preprocessor and scene filter nodes
20aece8 
ScenePreprocessorNode turns a scene spec into flat draw commands and arena uploads;
SceneFilterNode, FlattenedSceneFilterNode and MergeGeometriesNode operate on flattened
scenes.

(cherry picked from commit f7bdb1d)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier @claude
js: rework the GPU node lifecycle with deterministic teardown
78592c7 
(cherry picked from commit 8b2c463)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier
avnd: name the port-callback storage condition
b9ebcc4 
(cherry picked from commit daa9a69)
@jcelerier
js: add Score.cable, path/findByPath, cable removal, setAutoTrigger/s…
be533c5 
…etProcessLoop

(cherry picked from commit b1b0062)
@jcelerier
gfx: skip the draw when a pipeline failed to build
425097e 
buildPipeline returns a null pipeline when QRhiGraphicsPipeline::create()
fails (transient during graph rebuild). InvertYRenderer::finishFrame
dereferenced it via setGraphicsPipeline (Q_ASSERT/null-deref), and
quadRenderPass asserted on the then-missing pass. Both now skip the draw,
matching defaultRenderPass / the if(pip.pipeline) guard at pass creation.

(cherry picked from commit 1290424)
@jcelerier @claude
gfx: fix 5 libisf importer bugs + render/scene test harnesses
d0aef6a 
libisf/src/isf.cpp:
 - write_isf: seekp(0,end)-before-tellp corrupted the emitted descriptor
 - use-after-move on m_sourceVertex -> guard with .empty()
 - parse_shadertoy_json: GLSL45 preludes + iMouse vec4(0,0,0,0)
 - glsl_sandbox: duplicate-TIME uniform guard
 - replace_identifier: word-boundary match

Tests: test_unit_isf_importers (all 5 importer branches); golden-image
render regression (16 pinned JS-corpus cases, llvmpipe, SSIM/PSNR);
timeline-driven scenario ramp; gfx resource/leak soak (DISABLED — guards
a still-unfixed render-clock teardown UAF); live-edit churn corpus.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier
tests: Text-node render validation harness
563a56f 
Headless llvmpipe golden+value harness for the Gfx::Text process: grabs the
untouched-defaults frame (hard off-screen-default visibility assertion — guards
the default-position fix), then live-edits text/font/size/position/scale/
color over OSC /script and asserts pixel VALUES per case (coverage, bbox
ordering across sizes, centroid movement, channel dominance, blank empty
string, unicode/CJK/tofu/2000-char) + golden refs (compare.py strict).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
@jcelerier @claude
gfx: drop an output's edges when tearing it down synchronously
b93916a 
destroyOutput() is the synchronous counterpart of the async REMOVE_NODE path,
which already calls removeNodeAndEdges. Without it, Graph::removeNode leaves
m_edges holding Edges that point at the freed output's Ports, so ~Graph ->
clearEdges() unlinks them from freed memory.

Also finish the render-clock conversion in the destructor: the clocks must be
released before the timer pool they borrow from.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014rZgzE8JjWvHDtaVUhxpLE
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@jcelerier @coveralls