fix: preserve vanilla water TAA when no upscaler is active

[jturnley]

Fixes #1979

Problem

Two hooks in the Upscaling module unconditionally override the engine's TAA state, disabling water temporal stabilization when no upscaler is active (kNONE):

1. ConfigureTAA() — runs every frame via Main_UpdateJitter before the vanilla jitter function:

*enableWaterTAA = !(upscaleMethod == kNONE || upscaleMethod == kTAA);  // false for kNONE
BSImagespaceShaderISTemporalAA->taaEnabled = upscaleMethod != kNONE;   // false for kNONE

2. Main_PostProcessing::thunk() — runs later, right before the actual post-processing dispatch:

BSImagespaceShaderISTemporalAA->taaEnabled = upscaleMethod == kTAA;    // false for kNONE
func(a_this, a3, a_target, a_4, a_5);                                  // ISWaterBlend skipped
BSImagespaceShaderISTemporalAA->taaEnabled = false;

Both disable ISWaterBlend's 95% temporal history blend that stabilizes water reflections. Without it, every per-frame variation in auto-exposure adaptation is directly visible as water flickering between transparent, white/opaque, and normal reflections (~6 second convergence when stationary).

Fix

ConfigureTAA(): Early return when kNONE — let the vanilla UpdateJitter function (called immediately after) manage TAA state correctly.

Main_PostProcessing::thunk(): Skip all TAA state overrides when kNONE — call func() with vanilla state preserved.

For kTAA, kFSR, and kDLSS, behavior is unchanged from the original code.

Testing

Isolated Upscaling as the sole trigger via binary search across 20 features added since 1.4.7 RC. Verified fix eliminates flickering for Upscaler Disabled and TAA modes. FSR/DLSS water flicker is a separate pre-existing issue (ISWaterBlend was already disabled for those modes in the original code).

Tested on: Skyrim SE v1.6.1170, RX 9060 XT, 5120×1440, Upscaler Disabled and TAA modes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 030c28dd-560c-4c00-bcf1-7cc764453495

📥 Commits

Reviewing files that changed from the base of the PR and between 47e3a79 and f4c0234.

📒 Files selected for processing (1)

• src/Features/Upscaling.cpp

---

📝 Walkthrough

Walkthrough

A targeted bug fix to Upscaling::ConfigureTAA() that adds an early return when no upscaler is active, adjusts water TAA enablement logic, and refines explanatory comments to prevent water surface flickering caused by inadvertent temporal stabilization disablement.

Changes

Cohort / File(s)	Summary
ConfigureTAA Logic Fix src/Features/Upscaling.cpp	Added early return to skip TAA configuration when upscale method is NONE, preserving vanilla behavior. Inverted enableWaterTAA condition to enable water TAA for non-TAA methods. Changed taaEnabled assignment to unconditionally set true for active upscalers. Refined comments clarifying TAA replacement and water TAA behavior across upscaling paths.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Possibly related PRs

• PR #1726: Directly modifies Upscaling::ConfigureTAA's water TAA enablement logic by inverting the conditional for non-TAA upscalers.
• PR #1498: Introduces the Upscaling::ConfigureTAA function that this PR refines, providing relevant historical context.
• PR #1482: Adjusts TAA enable/disable state management in Upscaling.cpp, including explicit post-call modifications to taaEnabled.

Suggested reviewers

• alandtse
• davo0411
• doodlum

Poem

🐰 Hop along, dear water so bright,
No flicker shall mar thy reflections of light,
With jitter preserved and TAA set true,
Temporal blending gives frame-to-frame glue. ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'fix: preserve vanilla water TAA when no upscaler is active' directly matches the main objective: implementing an early return when UpscaleMethod == kNONE to preserve vanilla TAA behavior and fix water flicker.
Linked Issues check	✅ Passed	The PR successfully addresses all coding requirements from issue #1979: implements early return for kNONE case, adjusts water TAA logic for kTAA (disable) and kFSR/kDLSS (enable), and fixes the water flicker bug.
Out of Scope Changes check	✅ Passed	All changes in Upscaling.cpp are directly scoped to fixing the identified bug: early return for kNONE, logic adjustments for water TAA and TAA state, and refined comments explaining the behavior.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

📝 Coding Plan

• Generate coding plan for human review comments

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Tip

You can get early access to new features in CodeRabbit.

Enable the early_access setting to enable early access features such as new models, tools, and more.

[Copilot]

Pull request overview

Fixes a regression in the Upscaling feature where ConfigureTAA() was overriding vanilla TAA/water-temporal state even when no upscaler is selected (UpscaleMethod::kNONE), which could disable the vanilla water temporal stabilization (ISWaterBlend) and cause water flicker.

Changes:

• Add an early-return in Upscaling::ConfigureTAA() for kNONE to preserve vanilla TAA/water state.
• Adjust water TAA toggling so only CS TAA mode (kTAA) disables water TAA, while FSR/DLSS keep it enabled.
• Keep taaEnabled forced on only for non-kNONE modes (behavior preserved via the early return).

[github-actions]

No actionable suggestions for changed features.

[github-actions]

✅ A pre-release build is available for this PR:
Download

[alandtse]

Thanks!

[alandtse]

@jturnley can you check #1988 and see if that reintroduces the issue you tried to fix?

[jturnley]

Sorry, been busy this week - it didn't cause a reversion, but it did surface a different location in the code where the same forced disabling of vanilla TAA was happening. I fixed that and had Claude parse through the upscaling section to make sure I didn't miss more. PR #1986 now fixes all flickering when Upscaling is disabled, or set to TAA.

I don't know enough about how DLSS and FSR work to figure out what's wrong, but I sent the problem through Claude and it says

"The cause is fundamentally different — it's projection jitter. When FSR/DLSS are active, ConfigureUpscaling applies sub-pixel jitter to the projection matrix every frame. Water normals are sampled with these jittered coordinates, which means the Fresnel blend factor, reflection direction, and surface detail shift slightly each frame. The upscaler is supposed to reconstruct stability from these jittered samples via temporal accumulation, but water defeats it because:

• Water has constant animated wave motion that the motion vectors don't capture
• Fresnel reflections change drastically with tiny view angle shifts from jitter
• The depth buffer shows the riverbed (through the water), not the water surface, so the upscaler's depth-based reprojection is wrong for water pixels

This is a known limitation of temporal upscalers with transparent/reflective surfaces. The correct fix is complex — either running ISWaterBlend at render resolution before the upscaler, improving the reactive/transparency mask in EncodeTexturesCS for water, or providing water-specific motion vectors."

That's basically Greek to me so I would not let it try to fix it since I don't have the background knowledge to know if it's fixing something or breaking it worse, but maybe it will help someone who does.