impr: Use Immediates in examples

[reczkok]

No description provided.

[github-actions]

pkg.pr.new

packages
Ready to be installed by your favorite package manager ⬇️

https://pkg.pr.new/software-mansion/TypeGPU/typegpu@5aa6bea72181c40591725c5f3565920dca10aeee

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/gl@5aa6bea72181c40591725c5f3565920dca10aeee

benchmark
view benchmark

commit
view commit

[github-actions]

Resolution Time Benchmark

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Random Branching (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [1.01, 1.98, 4.61, 6.22, 7.25, 11.50, 22.76, 25.30]
  line [0.95, 1.94, 4.32, 7.06, 7.69, 11.72, 23.52, 26.69]
  line [1.02, 1.94, 4.60, 6.69, 7.27, 10.64, 23.72, 22.93]

Loading

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Linear Recursion (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.35, 0.55, 0.70, 0.81, 1.05, 1.16, 1.37, 1.55]
  line [0.29, 0.48, 0.63, 0.77, 1.03, 1.13, 1.35, 1.49]
  line [0.33, 0.55, 0.68, 0.83, 1.09, 1.16, 1.32, 1.52]

Loading

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Full Tree (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.86, 2.41, 3.51, 7.23, 12.27, 26.36, 55.16, 111.20]
  line [0.83, 2.28, 3.98, 7.03, 12.60, 26.59, 55.56, 112.12]
  line [0.89, 2.05, 3.43, 7.29, 12.43, 26.40, 54.39, 110.90]

Loading

[github-actions]

Bundle size comparison (import * as ... in PR vs import * as ... in target):

🟢 Decreased	➖ Unchanged	🔴 Increased (max 1.77%)	❔ Unknown
0	39	286	0

import * as ... in PR vs import * as ... in target (did bundle size increase?):

Test	tsdown
STATIC_tgpu.ts	277.66 kB ($${\color{red}+1.8\%}$$)
tgpu_initFromDevice.ts	267.04 kB ($${\color{red}+1.6\%}$$)
tgpu_init.ts	267.58 kB ($${\color{red}+1.6\%}$$)
STATIC_allImports.ts	303.91 kB ($${\color{red}+1.6\%}$$)
d_ref.ts	5.03 kB ($${\color{red}+0.9\%}$$)

If you wish to run a comparison for other, slower bundlers, run the 'Tree-shake test' from the GitHub Actions menu.

[pullfrog]

ℹ️ No critical issues — one minor suggestion inline.

Reviewed changes

• Lighting shaders — replaced DSL chains (.mul/.sub/.div) and tgpu.fn with 'use gpu' bodies using infix operators and direct accessor reads; depthToColor is now a plain TS 'use gpu' fn.
• PointLight immediates — dropped the six shadow Cameras in favor of a tgpu.accessor(d.mat4x4f) fed by immediateVar + pass.setImmediates, with a per-face uniform-bind-group fallback when immediate_address_space is unsupported.
• Render loop — moved to an explicit root['~unstable'].createCommandEncoder() with hand-rolled passes; destroyed depthTexture/msaaTexture on resize; scene._instanceBuffer now destroyed before rebuild.
• Camera / types — CameraData struct replaced by a single d.mat4x4f viewProj uniform (dropped unused inverseViewProjectionMatrix).
• box-geometry — model matrix built with wgpu-matrix (quat.fromEuler(…, 'zyx') + mat4.fromQuat/setTranslation/scale); verified equivalent to the old T*Rz*Ry*Rx*S chain.
• Snapshot test — updated WGSL inline snapshot (immediate path).

I verified the two things that looked risky: pointLight.positionUniform.$ needs no explicit bind group (TypeGPU's auto-generated catch-all bind group covers it, see resolutionCtx.ts), and the wgpu-matrix fromEuler('zyx') quaternion is exactly qz⊗qy⊗qx, i.e. the same rotation the old rotationZ().mul(rotationY()).mul(rotationX()) chain produced.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏}

[Copilot]

🟢 Approval recommended

The changes consistently update the example code, shader bindings, and tests to the new immediates/encoder-based approach without introducing any confirmed correctness issues.

Pull request overview

Updates the point-light-shadow docs example to use TypeGPU “Immediates” (with a uniform fallback) and the newer command-encoder/pass-style rendering flow, simplifying several data structures along the way.

Changes:

• Switch shadow-face view-projection data to an accessor backed by an immediate var when supported, with a per-face uniform/bind-group fallback otherwise.
• Simplify camera/view-projection handling (remove CameraData struct usage) and update shaders + bindings accordingly.
• Move rendering to explicit CommandEncoder render passes; add explicit destruction for rebuilt buffers and resized textures.

File summaries

File	Description
apps/typegpu-docs/tests/individual-example-tests/point-light-shadow.test.ts	Updates the WGSL snapshot to reflect immediate-backed view-projection and new binding names.
apps/typegpu-docs/src/examples/rendering/point-light-shadow/types.ts	Removes CameraData from the example’s shared types.
apps/typegpu-docs/src/examples/rendering/point-light-shadow/scene.ts	Destroys/recreates the instance buffer when rebuilding the instance array.
apps/typegpu-docs/src/examples/rendering/point-light-shadow/point-light.ts	Implements per-face shadow rendering using immediates (or per-face uniform fallback) and encoder-based passes.
apps/typegpu-docs/src/examples/rendering/point-light-shadow/index.ts	Refactors the example to use encoder passes, binds the new view-projection inputs, and wires up shadow rendering accordingly.
apps/typegpu-docs/src/examples/rendering/point-light-shadow/camera.ts	Uses a single mat4x4f uniform for view-projection (no inverse matrix).
apps/typegpu-docs/src/examples/rendering/point-light-shadow/box-geometry.ts	Changes model-matrix construction to use wgpu-matrix quaternion/Euler utilities.

Review details

• Files reviewed: 7/7 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[pullfrog]

ℹ️ No critical issues — one minor suggestion inline.

Reviewed changes

• Lighting shaders — replaced .mul/.sub/.div DSL chains and tgpu.fn with 'use gpu' bodies using infix operators and direct accessor reads; depthToColor is now a plain TS 'use gpu' arrow fn inlined into two fragment bodies.
• PointLight immediates — dropped the six shadow Cameras for a tgpu.accessor(d.mat4x4f) fed by immediateVar + pass.setImmediates, with a per-face uniform bind-group (faceLayout) fallback when immediate_address_space is unsupported.
• Render loop — explicit root['~unstable'].createCommandEncoder() with hand-rolled passes; depthTexture/msaaTexture destroyed on resize; scene.#instanceBuffer destroyed before rebuild.
• Camera / types — CameraData struct replaced by a single d.mat4x4f viewProj uniform (dropped the unused inverseViewProjectionMatrix).
• box-geometry — model matrix built with wgpu-matrix (quat.fromEuler(…, 'zyx') + mat4.fromQuat/setTranslation/scale).
• Snapshot test — WGSL inline snapshot updated (immediate path).

I re-verified the previously-reviewed surface against the current head (5aa6bea): the wgpu-matrix fromEuler('zyx') rotation is numerically equivalent to the old T*Rz*Ry*Rx*S chain (max diff ~1e-7, float rounding only); the new pass API defaults depthLoadOp: 'clear' / depthClearValue: 1 for depth attachments, matching the old explicit values; and the non-immediate fallback compiles to valid WGSL (verified with a scratch test that removes immediate_address_space from the mock device's feature set).

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

The uniform fallback branch (#faceUniforms/#faceBindGroups) is only exercised when immediate_address_space isn't supported, but the committed snapshot test always runs with the mock device's feature set containing it, so only the immediate path is covered. I confirmed the fallback compiles to valid WGSL (removing the feature yields @group(1) @binding(0) var<uniform> viewProj in the depth shader, no var<immediate> item) — but since this is the only code path on devices without the extension, a test variant that disables the feature would pin it against regressions.

[reczkok]

Moved to #3034, stacked on #3032.