Files
solvecraft/docs/render-cpu-progressive.md
T
selune 9e6e0b8df1 Render workspace: complete experimental ribbon, resources and effects integration (#5493) (#5494)
* wip(render): reconcile completed material and image pipeline source with 0.20 (#5493)

Source-only integration reference for staged review, not a merge-ready aggregate.
Donor: 0xlillith/solidcraft@aea20b0115.
Upstream: 858e5995d6.

Retain original CPU/material contracts and explicit GPU donor provenance in
docs/render-materials-lab.md (#2067, #2191, #2780, #3306). Preserve released
format migrations and current navigation/startup/shell behavior. New saved
records use pending descriptors. The physical-card descriptor follows the
maintainer's material-engineering-cards naming request, with a legacy-lab
format-29 regression. Replace raw new panel hosts with shared containers.

Repository checks, complete workspace/all-target Clippy, and the independent
physical/resource authoring checks pass for this source. The fresh full local
CI invocation has reached test compilation and is not yet complete. Do not
transfer donor native output or prior test totals to this integration.

Generated review screenshots, video, package ZIP and machine logs remain at
the immutable personal evidence pin. Uncommitted geometric displacement is
backed up separately, not silently included. No kernel/vendor change, direct
main write, release or independent contributor-branch rewrite.

Co-authored-by: Christian Svensson <blue@cmd.nu>

* test(cli): pin the Compact camera-sample admission diagnostic (#5493)

The bounded request still refuses 1024x1024x64 samples before host initialization and writes no output. Update only the stale expected diagnostic to the existing Compact budget wording; production limits and code are unchanged. Exact-tree CLI regression suite: 14 passed, 0 failed. The full repository gate has been invoked separately and is not yet complete.

* Normalize UI container violation paths across platforms

Report repository-relative paths with forward slashes so diagnostics and the existing nested-package regression assertions agree on Windows and Unix. Keep the container detection rules unchanged.

Validation: official ui-containers gate and all seven focused xtask tests passed; rustfmt and diff checks passed. Only xtask/dependencies were compiled; no application build or heavy tests.
(cherry picked from commit 44b58d8e10)

* fix(render): preserve camera contracts and correct acceptance regressions (#5493)

Preserve Design view.apply semantics with a separate strict Render camera command. Retain existing translations when English-only aliases collide; add material-axis entries and camera-specific pitch labels. Compare complete validated engineering cards and retain job handles before cancellation.

Focused and native acceptance is pinned to tree f8a423b21f. Full CI has been invoked separately and is not claimed green. Landing follows the render integration policy in #5493.

* fix(render): resolve showcase licence and upstream resource hygiene (#5493)

Record SychPL's explicit MIT OR Apache-2.0 grant, retaining source/model bytes and the Ubuntu lettering notice. Remove 80 absent attribution rows, personal fork records, dead delivery attributes and three fork/resource-only workflows. Existing xtask assets now performs bounded offline hash checks and rejects orphan render attribution; no parallel draft CI or package installation. Original authoring reproduction remains explicit. Scoped xtask tests, assets, Clippy, formatting and prepared SELENE validation pass; no application or renderer-wide result is inferred.

* fix(render): align saved bounds and progressive sample resolution (#5493)

Address review r4364211669/r4364211677/r4364211681/r4364211691/r4364211695. Persisted light/frame domains respect the existing 1e9 document bound; related LED/rectangle/direction cases use the same bound. Workstation primary resolve shares the 4096 admission cap and film counters use checked conversions. Exact candidate: 485 CPU render tests, 21 render document tests and strict scoped Clippy pass, including positive boundary round trips and 2048/4096 end-to-end checkpoints. No released format/migration edits, relaxed test limits or full-stack acceptance claim.

* wip(render): isolate optional resources and harden sample handling (#5493)

Review work in progress: opt-in encoded render library with reproducible runtime catalog exports; native-only robust dependency through the workspace; checked bloom slices; per-sample point-light diagnostics; independent area-light spatial/film keys. Resource-module binaries in both configurations, generation, Cargo production feature graphs and attribution checks pass. Full engine/render regression and combined acceptance are pending: do not merge this donor as a constituent or treat scoped resource proof as runtime acceptance.

* fix(gpu): preserve donor allocations through the scope-safety repair (#5722)

Port the native-validated #5723 scope-cleanup correction into the experimental donor, retaining its adaptive, volume, aging, photographic and bloom allocations. Same reverse-drain helper and regression cases; the donor combination still needs its own compile/test gate and is not covered retroactively by render-base results.

* claim: #1369

* Reject unknown pending concepts after format folding (#1369)

* test(render): close bounded sampling and bloom review regressions (#5493)

Validate the optional-resource/lighting/dependency donor at its exact tree. Required Vulkan GPU 127/0/3; CPU renderer 487/0; engine Render 72/0; document render 21/0; lean library 7/0; strict scoped Clippy, assets/layers/lists/containers and lean-engine WASM pass. Preserve original failed controller/typing attempts. This is the draft donor/disposition source, not a replacement for constituent render PRs or final main acceptance.

* wip(render): preserve stable aging target validation candidate

Exact staged source from prior review work on donor 3b892eab5e. This is a recovery/review checkpoint, NOT accepted delivery. An actual command-path regression currently fails because the selected executor refuses ordered surface aging. No assertion, capability or admission limit is weakened. Native/UI/full combination acceptance remains pending. Saved/runtime split adds one pending descriptor and retains original file scope; no main/render branch is changed.

* test(render): preserve numeric setup intent at the admitted coordinate bound

Follow up jyviko review on #5494 without weakening the corrected saved-coordinate contract. The positive clicked fixture uses 1e9 instead of inadmissible 2e9; explicit out-of-range cases still refuse, and the full saved setup round-trips through JSON and the native container. Exact-source Render workspace 95/0 and strict all-target UI Clippy pass. This does not incorporate the separately failing stable-aging candidate or claim full current-render acceptance.

* test(render): use the real GPU executor for saved aging

Move successful native cold-reopen Check/Start coverage from the CPU-backed protocol fixture into the production GPU executor tests. Verify capture-index changes preserve the stable target and that enabling aging changes both PNG and raw EXR output. Retain explicit, allocation-free refusal coverage for executors without aging support. Production capabilities and admission checks are unchanged.

* fix(render): persist stable identities for targeted aging (#5493)

Separate saved RenderBodyTarget identities from captured mesh ordinals. Reject legacy ambiguous targets, resolve Check/Start after visibility changes and preserve native reopening, copies and undo. Unsupported executors still refuse; real AMD Vulkan regression completes saved aging after reordered capture.

Source tree 6b84c8ad passes document 23, engine Render 75, GPU effects 5 and Render UI 95 with no failures; scoped strict Clippy, WASM and repository gates pass. Fresh CLI/native reopen and clean/aged GPU exports pass with identical physical totals and inspected actual images. Whole current-render integration is separate.

* fix(render): reconcile combined CPU/GPU contracts on current upstream

Unify captured resources/status and CPU-vs-GPU request schemas after the real two-parent integration; remove obsolete duplicate Vulkan selector and retain combined platform-filter regression. Deduplicate only identical or superseded test definitions, retaining the stronger card metadata and reserved-executor assertions. No assertions, admission limits or feature refusals are weakened.

All targets of render dependencies, GPU, engine, UI, CLI and native app pass cargo check with raytrace-cpu and bundled-render-library at tree c70d00d1. Native executable tests and full current-source acceptance remain pending. This is a private integration rehearsal for subsequent constituent PRs, not an aggregate official render/main merge.

* test(render): retain one complete host lifetime regression set

Deduplicate byte-identical DropProbe/task-lifetime tests and retain the adapted environment/texture ownership regressions with every assertion and complete memory estimate. Keep the registered-geometry admission regression once; only struct field order differed. Replace a one-element diagnostic loop with its equivalent block, preserving all wrap/width assertions.

Test-only integration repair. Formatting and diff checks pass; refreshed executable validation is required. No production behavior, limits, assertions or ignores changed.

* fix(doc): recognize the proven format-29 card descriptor alias

Normalize only the historical format-29 physical-material-cards spelling, and only when material-engineering-cards is registered. Admission and migration de-duplication use the same identity. Unknown concepts, wrong versions and incapable readers still refuse before mutation. Six controlled-history Rust regressions and actual CLI migration/no-clobber acceptance pass. No version, released migration or CI changes. Refs #5793, #1369, #5493.

* test(render): preserve grounding refusal before explicit instance moves

Update the two Render binding fixtures for upstream first-occurrence grounding. Verify grounded movement refuses without changing document/history, then explicitly unground through the public command before the intended move. Retain every target, resource, copy and persistence assertion. Refs #5493.

---------

Co-authored-by: Christian Svensson <blue@cmd.nu>
Co-authored-by: Victor Santos Oliveira <56417388+Vittuu@users.noreply.github.com>
Co-authored-by: cfaulkingham <colin.faulkingham@gmail.com>
2026-10-24 14:21:59 +00:00

8.6 KiB
Raw Blame History

CPU progressive camera sampling

The limits below describe the compatibility Compact profile. Explicit Studio and Workstation envelopes are documented in CPU resource profiles; their implementation batch still requires consolidated validation.

CPU progressive rendering repeatedly traces fresh camera samples into a bounded linear-light film. Checkpoints are complete image passes, at 1, 2, 4, 8, 16 and 32 samples per pixel when below the requested final sample count. The exact requested count produces the final PNG, including non-powers of two such as 3 or 5. A one-sample job has no provisional checkpoint.

This mode uses the native cpu-raytrace executor and its existing direct-light and selected-reflection transport. It does not add diffuse global illumination, caustics, adaptive convergence, a denoiser, or general glass path tracing.

Request and saved intent

The CPU option is nested inside raytrace. The top-level progressive: true request continues to select the registered GPU path executor.

{
  "integrator": "raytrace",
  "settings": { "width": 512, "height": 512 },
  "raytrace": {
    "progressive": { "samples_per_pixel": 32, "seed": 81 },
    "primary_samples": 1,
    "exposure_ev": 0,
    "shadows": true
  }
}

Both render.preflight and render.start use this request. Missing or null raytrace.progressive preserves the existing fixed 1/4/16 spatial AA route. An explicit empty progressive object defaults to 16 samples and seed 0. The seed is an unsigned 32-bit integer. Unknown fields are rejected.

The corresponding portable setup is:

{
  "name": "CPU film",
  "set": {
    "width": 512,
    "height": 512,
    "exposure_ev": 1,
    "job": {
      "method": "cpu_raytrace",
      "options": {
        "progressive": { "samples_per_pixel": 32, "seed": 81 },
        "primary_samples": 1
      }
    }
  }
}

This is a render.settings.set request. Exposure is stored once in the setup; normalization transfers it into the CPU request. Film storage, checkpoint PNGs, job identities, cancellation state and host reservations are never serialized into the design. Saving validates pixel and camera-sample bounds even when no render host is installed.

Supported combinations and limits

Control Initial CPU progressive contract
Dimensions Each side 1..4096; at most 1,048,576 pixels
Samples per pixel 1..64
Total camera samples At most 16,777,216 (width × height × samples_per_pixel)
Fixed spatial AA primary_samples must be 1
Selected mirror / isotropic rough reflection Existing bounded depth; reflections.samples must be 1 when bodies are selected
Directional / point / spot sources Existing CPU light contracts and limits
Constant / captured HDR environment Existing CPU contracts; HDR plus rough reflections still requires certified fully metallic selected bodies and effective faces
Surface emission / isotropic surface effects Existing CPU appearance contracts; emission is visible radiance
Area / LED sources Explicitly rejected in this initial progressive mode
Dielectric / anisotropic face overrides Explicitly rejected in this initial progressive mode
Output Opaque PNG with the existing display background

An incompatible request fails; it does not remove settings or switch methods. The fixed CPU route retains its existing area, dielectric and anisotropy support. At 1024 × 1024 the camera-sample cap allows at most 16 samples per pixel; at 512 × 512 it allows 64. Requests are also subject to the existing geometry, 384 MiB CPU job and 512 MiB shared host-pool admission limits. Traversal, shading, film updates, resolve, PNG encoding and hashing share one cumulative 1,000,000,000-unit work ceiling. The count is not reset at a pass boundary. The camera-sample limit is a cold upper bound, not a promise that every scene will finish below the cumulative runtime work ceiling.

Sampling, linear film and display

Camera jitter is a deterministic counter sequence keyed by the u32 seed, pixel index and absolute sample ordinal. Reflection and HDR-direct variates use separate domains of the same counter. Target sample count, worker quantum and checkpoint cadence are not keys. Consequently, for an unchanged captured request and scene, a checkpoint at N samples has exactly the same sample prefix and PNG bytes as a job whose final target is N.

Each pixel stores three f64 sums and a primary coverage count. Only complete, uniform passes can be resolved. Camera misses add zero scene radiance and zero coverage. The resolve computes the mean scene radiance, applies exposure and the existing tone curve once to covered radiance, composites the display-linear backplate by coverage, and encodes sRGB once. The background is not illuminated or exposed. Averaging already tone-mapped sample pixels is not equivalent and is not used.

film::primary::PrimaryFilm is a compact camera-specific accumulator alongside the existing statistical FilmAccumulator. It has incremental initialization and requires the exact next scanline pixel and absolute ordinal on every update. The existing general FilmAccumulator/borrowed FilmPngTask APIs are retained; the owned image job uses the shared PngEncoding pipeline without a self-referential borrowed task or a second independent pool.

Preview publication and final export

JobStatus.preview, when present, contains:

PreviewInfo {
    frame_sequence: u32,
    samples_per_pixel: u32,
    frame: FrameInfo
}

JobHandle::preview() returns an Arc<RenderPreview> whose png() is the immutable encoded snapshot. The first checkpoint has sequence 1. The job's JobId { session, serial, generation } and captured scene_generation remain unchanged across every pass. frame_sequence is a separate monotonic counter. Clients cache previews by (JobId, frame_sequence) and may retain an older snapshot without mutating it.

The preview slot is separate from JobStatus.frame/JobHandle::frame(). A preview cannot be exported through render.export, cannot become the session's last completed image, and does not release the running task or film. The final PNG is published only after its digest is complete; it clears the preview slot and is the only new exportable result. render.export writes those exact PNG bytes and does not render or re-encode them.

Cancellation clears the preview slot immediately. Failure and completion also clear it. Publication checks the cancellation token while holding the same record lock used for cancellation, including after the final digest chunk. The worker cannot publish a late checkpoint into a cancelled or replaced job. An externally retained preview remains an immutable snapshot of its original job, but cannot replace a previous successfully completed final frame.

Allocation and retained snapshots

Admission reserves the working film, existing RGBA/encoder/final-PNG capacity, and the maximum encoded capacity plus metadata for every possible nonfinal checkpoint. This conservative upper bound covers a consumer retaining all checkpoints at once. Actual film and PNG capacities are checked against their admitted capacities after fallible allocation.

A completed checkpoint moves its PNG allocation into the preview object. Reservation::split transfers that allocation's share of the running lease without increasing or decreasing the pool total. Replacing or clearing the host preview drops only the host's Arc; any external Arc keeps its memory billed. At final completion the task releases its working buffers before the remaining lease shrinks to the final PNG. Dropping each last external snapshot releases only that snapshot's reservation. There is no full-film clone per checkpoint.

Regression coverage

The source includes regressions for deterministic prefixes and worker quanta, fresh primary rays, non-power-of-two targets, actual HDR/rough-reflection PNGs, linear averaging before tone mapping, coverage and unexposed backgrounds, bounded/cancelled film mutations, and the single cumulative work budget.

Host tests manually advance the worker to the digest/publication boundary. They hold multiple preview Arcs through replacement, cancellation and completion, check allocation identity and exact pool release, and reject an invalid checkpoint sequence. Document tests cover offline limits and saved round trips. Engine tests cover Check/Start equality, cold rejection before HDR capture, session ownership, exact final export, and retention of the previous completed image after a progressive job is cancelled. Execution evidence belongs to the integrated CI receipts; this source handoff does not claim those gates passed.