revert: "perf: optimise DeferredComposite (#2150)"#2232
Merged
Conversation
This reverts commit 7f64e55.
Contributor
📝 WalkthroughWalkthroughThis PR converts the deferred composite pipeline from a graphics-based approach (vertex/pixel shaders) to a compute shader architecture, refactors normal encoding from sqrt-based to octahedral schemes, updates dependent systems to reference the new normal-roughness render target, and removes obsolete state management utilities. Changes
Sequence Diagram(s)sequenceDiagram
participant Deferred as Deferred::Render
participant Dispatch as GPU Dispatch
participant ComputeShader as DeferredCompositeCS
participant SRVs as Input SRVs<br/>(Albedo, Normal, Depth, etc.)
participant UAVs as Output UAVs<br/>(Main, Normals, Motion)
Deferred->>Deferred: Bind SRVs (Albedo, NormalRoughness, Masks, Depth, etc.)
Deferred->>Deferred: Bind StereoOptModeTexture (optional)
Deferred->>Deferred: Bind UAVs (Main, NormalMask, MotionVectors)
Deferred->>Dispatch: Dispatch(screen width/8, screen height/8, 1)
Dispatch->>ComputeShader: Launch threads (dispatchID)
ComputeShader->>SRVs: Sample per-pixel data (albedo, normal, depth)
alt Depth == 1.0
ComputeShader->>ComputeShader: Generate sky motion vectors
else Standard
ComputeShader->>ComputeShader: Compute standard lighting
end
ComputeShader->>ComputeShader: Decode octahedral normal
ComputeShader->>ComputeShader: Sample SSGI, reflections, sky
ComputeShader->>UAVs: Write Main (final color)
ComputeShader->>UAVs: Write NormalTAAMaskSpecular
ComputeShader->>UAVs: Write MotionVectors
Deferred->>Deferred: Unbind compute resources
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Review rate limit: 7/8 reviews remaining, refill in 7 minutes and 30 seconds.Comment |
Actionable Suggestions
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@package/Shaders/Tests/TestGBuffer.hlsl`:
- Around line 37-59: The tests currently only assert encoded range and decoded
unit length for the sample normals (testNormals) but miss verifying direction
fidelity; update the loop that uses GBuffer::EncodeNormal and
GBuffer::DecodeNormal to also compute the dot product between original and
decoded (e.g., dot(original, decoded)) and assert it exceeds a relaxed threshold
(choose ~0.90–0.97 depending on half precision tolerance) to catch
mirrored/reflected results; add this cosine/dot assertion alongside the existing
length check for variables original and decoded.
In `@src/Deferred.cpp`:
- Around line 356-373: The composite shader reads DepthTexture unguarded in
DeferredCompositeCS but srvs[16] leaves slot t4 null on builds without
dynamicCubemaps, causing bad depth reads; fix by binding the depth SRV into
srvs[4] unconditionally (use Util::GetCurrentSceneDepthSRV(true) for t4) instead
of only when dynamicCubemaps.loaded || REL::Module::IsVR(), and ensure the srvs
array still gracefully uses nullptr for optional textures (reflectance,
envTexture, skylighting, etc.) so resource lifetimes and DX11 binding remain
safe when features are absent.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 5b1b00ef-be67-4da7-9759-36c316185e2a
📒 Files selected for processing (10)
package/Shaders/Common/GBuffer.hlslipackage/Shaders/DeferredCompositeCS.hlslpackage/Shaders/DeferredCompositeVS.hlslpackage/Shaders/Tests/TestGBuffer.hlslsrc/Deferred.cppsrc/Deferred.hsrc/Features/ScreenSpaceGI.cppsrc/Features/SubsurfaceScattering.cppsrc/Features/VRStereoOptimizations.hsrc/Utils/D3DStateBackup.h
💤 Files with no reviewable changes (2)
- package/Shaders/DeferredCompositeVS.hlsl
- src/Utils/D3DStateBackup.h
Comment on lines
+37
to
59
| // Test behavioral properties of octahedral encoding (not exact numerical accuracy) | ||
| // Half precision + quantization means we check: valid output, normalized, reasonable direction | ||
| half3 testNormals[4] = { | ||
| normalize(half3(1.0h, 1.0h, 1.0h)), | ||
| normalize(half3(-1.0h, 1.0h, 1.0h)), | ||
| normalize(half3(1.0h, -1.0h, 1.0h)), | ||
| normalize(half3(1.0h, 1.0h, -1.0h)) | ||
| }; | ||
|
|
||
| for (int i = 0; i < 4; i++) { | ||
| float3 original = testNormals[i]; | ||
| float2 encoded = GBuffer::EncodeNormal(original); | ||
| float3 decoded = GBuffer::DecodeNormal(encoded); | ||
| half3 original = testNormals[i]; | ||
| half2 encoded = GBuffer::EncodeNormal(original); | ||
| half3 decoded = GBuffer::DecodeNormal(encoded); | ||
|
|
||
| // Check behavioral properties (relaxed for half precision quantization): | ||
| // 1. Encoded values are in valid range [0, 1] | ||
| ASSERT(IsTrue, encoded.x >= 0.0h && encoded.x <= 1.0h); | ||
| ASSERT(IsTrue, encoded.y >= 0.0h && encoded.y <= 1.0h); | ||
|
|
||
| float length = sqrt(decoded.x * decoded.x + decoded.y * decoded.y + decoded.z * decoded.z); | ||
| ASSERT(IsTrue, abs(length - 1.0) < 0.05); | ||
| // 2. Decoded normal is normalized (unit length) | ||
| half length = sqrt(decoded.x * decoded.x + decoded.y * decoded.y + decoded.z * decoded.z); | ||
| ASSERT(IsTrue, abs(length - 1.0h) < 0.02h); // Relaxed tolerance for half precision | ||
| } |
Contributor
There was a problem hiding this comment.
Keep a direction check for the diagonal roundtrip.
This now only proves “valid range + unit length.” A broken octahedral fold/sign path can still return a normalized but mirrored vector, so these angled cases would pass while reflections drift. Add a cosine/dot threshold against original.
🧪 Suggested assertion
for (int i = 0; i < 4; i++) {
half3 original = testNormals[i];
half2 encoded = GBuffer::EncodeNormal(original);
half3 decoded = GBuffer::DecodeNormal(encoded);
// Check behavioral properties (relaxed for half precision quantization):
// 1. Encoded values are in valid range [0, 1]
ASSERT(IsTrue, encoded.x >= 0.0h && encoded.x <= 1.0h);
ASSERT(IsTrue, encoded.y >= 0.0h && encoded.y <= 1.0h);
+
+ // 1b. Decoded direction should still match the source normal closely.
+ ASSERT(IsTrue, dot(decoded, original) > 0.98h);
// 2. Decoded normal is normalized (unit length)
half length = sqrt(decoded.x * decoded.x + decoded.y * decoded.y + decoded.z * decoded.z);
ASSERT(IsTrue, abs(length - 1.0h) < 0.02h); // Relaxed tolerance for half precision
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| // Test behavioral properties of octahedral encoding (not exact numerical accuracy) | |
| // Half precision + quantization means we check: valid output, normalized, reasonable direction | |
| half3 testNormals[4] = { | |
| normalize(half3(1.0h, 1.0h, 1.0h)), | |
| normalize(half3(-1.0h, 1.0h, 1.0h)), | |
| normalize(half3(1.0h, -1.0h, 1.0h)), | |
| normalize(half3(1.0h, 1.0h, -1.0h)) | |
| }; | |
| for (int i = 0; i < 4; i++) { | |
| float3 original = testNormals[i]; | |
| float2 encoded = GBuffer::EncodeNormal(original); | |
| float3 decoded = GBuffer::DecodeNormal(encoded); | |
| half3 original = testNormals[i]; | |
| half2 encoded = GBuffer::EncodeNormal(original); | |
| half3 decoded = GBuffer::DecodeNormal(encoded); | |
| // Check behavioral properties (relaxed for half precision quantization): | |
| // 1. Encoded values are in valid range [0, 1] | |
| ASSERT(IsTrue, encoded.x >= 0.0h && encoded.x <= 1.0h); | |
| ASSERT(IsTrue, encoded.y >= 0.0h && encoded.y <= 1.0h); | |
| float length = sqrt(decoded.x * decoded.x + decoded.y * decoded.y + decoded.z * decoded.z); | |
| ASSERT(IsTrue, abs(length - 1.0) < 0.05); | |
| // 2. Decoded normal is normalized (unit length) | |
| half length = sqrt(decoded.x * decoded.x + decoded.y * decoded.y + decoded.z * decoded.z); | |
| ASSERT(IsTrue, abs(length - 1.0h) < 0.02h); // Relaxed tolerance for half precision | |
| } | |
| for (int i = 0; i < 4; i++) { | |
| half3 original = testNormals[i]; | |
| half2 encoded = GBuffer::EncodeNormal(original); | |
| half3 decoded = GBuffer::DecodeNormal(encoded); | |
| // Check behavioral properties (relaxed for half precision quantization): | |
| // 1. Encoded values are in valid range [0, 1] | |
| ASSERT(IsTrue, encoded.x >= 0.0h && encoded.x <= 1.0h); | |
| ASSERT(IsTrue, encoded.y >= 0.0h && encoded.y <= 1.0h); | |
| // 1b. Decoded direction should still match the source normal closely. | |
| ASSERT(IsTrue, dot(decoded, original) > 0.98h); | |
| // 2. Decoded normal is normalized (unit length) | |
| half length = sqrt(decoded.x * decoded.x + decoded.y * decoded.y + decoded.z * decoded.z); | |
| ASSERT(IsTrue, abs(length - 1.0h) < 0.02h); // Relaxed tolerance for half precision | |
| } |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@package/Shaders/Tests/TestGBuffer.hlsl` around lines 37 - 59, The tests
currently only assert encoded range and decoded unit length for the sample
normals (testNormals) but miss verifying direction fidelity; update the loop
that uses GBuffer::EncodeNormal and GBuffer::DecodeNormal to also compute the
dot product between original and decoded (e.g., dot(original, decoded)) and
assert it exceeds a relaxed threshold (choose ~0.90–0.97 depending on half
precision tolerance) to catch mirrored/reflected results; add this cosine/dot
assertion alongside the existing length check for variables original and
decoded.
SkrubbySkrubInAShrub
approved these changes
Apr 29, 2026
|
✅ A pre-release build is available for this PR: |
Collaborator
Author
|
the feature audit is spurious. Need to teach it about reverts. |
IgorAlanAlbuquerque
pushed a commit
to IgorAlanAlbuquerque/skyrim-community-shaders
that referenced
this pull request
May 29, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This reverts commit 7f64e55.
closes #2223
Summary by CodeRabbit
Performance & Optimization
Refactor