[taylnos]

Problem

reflectionprobe_gl_test fails on an AMD RX 6800 (Windows, driver 26.8.1). It fails only through sampleReflectionProbesWater, at REFMAP_LEVEL 3 with REF_SAMPLE_COUNT 3: 1,630 ambient taps come out far darker than the list's.

Cause

The bad texels fall in 8×8 blocks, one GPU wave of 64 pixels each. They are exactly the 39 waves whose first pixel has a full list. That pixel skips water's own default-probe tap, if (count < REF_SAMPLE_COUNT) probeTap(0, ...).

Every other pixel of those waves that takes the tap gets none of probe 0's irradiance. Many read exactly amblit * (1 - refParams[0].x), which is what a zero irradiance mixes to.

The pixels' inputs are per-pixel hashes, and the shader has no cross-lane operations. So one pixel's branch changing another's result is the driver's compiler. The automatic probes' own tap of probe 0, under automatic && count < REF_SAMPLE_COUNT, was right in all 343 waves whose first pixel skipped it.

Change

sampleProbeTaps takes probe 0 through one tap for both cases. It taps where the list has room and either the automatic probes are not shut out or the point is water's: (automatic || t.water) && count < REF_SAMPLE_COUNT. Water then mixes it in as an automatic probe, as before.

Every entry point taps the same probes in the same order as before, at both probe levels. This replaces three tap sites with one.

Viewer impact: the viewer lists 32 probes a point, so its water met this only where 32 probes hold a pixel. It would show as dark 8×8 blocks in the water's irradiance on AMD. I haven't looked for this in the viewer.

Testing

On the RX 6800:

• reflectionprobe_gl_test passes, three runs in random order.
• All 9 alrender_gl_tests pass.
• reflectionprobe_gl_bench with the old and new shader alternated in the same binary: every fused time is unchanged within noise. Level-3 water is 2.605 ms both ways, and the rest are within ±0.4%.
• Not run: NVIDIA, and Mesa (CI will run llvmpipe).

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Review was skipped due to path filters

⛔ Files ignored due to path filters: 1

• indra/newview/app_settings/shaders/class3/deferred/reflectionProbeF.glsl is excluded by !**/*.glsl

CodeRabbit blocks several paths by default. You can override this behavior by explicitly including those paths in the path filters. For example, including **/dist/** will override the default block on the dist directory, by removing the pattern from both the lists.

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: e805fe12-d36e-449e-a583-12df806b6bdd

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.}