perf: terrain variation & parallax - #2220
Open
davo0411 wants to merge 72 commits into
Open
Conversation
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughRefactors terrain/landscape parallax and stochastic sampling: splits ExtendedMaterials into parallax core and terrain headers, reduces TerrainVariation sampling from 3→2 samples, removes stochastic mip noise from GetMipLevel, adds parallax/shadow tunables and terrain shadow helpers, consolidates per-layer blending, and removes the CPU ExtendShadows setting and its UI wiring. ChangesExtended Terrain & Parallax Refactor
Sequence Diagram(s)sequenceDiagram
participant PS as PixelShader (EMAT)
participant EM as ExtendedMaterials Parallax Core
participant ET as ExtendedMaterials Terrain
participant TV as TerrainVariation (stochastic offsets)
participant Tex as Terrain Textures
participant CPU as CPU Settings
PS->>EM: GetMipLevel(uv, parallaxTex)
EM-->>PS: mipLevel (deterministic)
PS->>TV: ComputeStochasticOffsets(sharedUV) [if TERRAIN_VARIATION]
TV-->>PS: sharedOffset
PS->>ET: InitializeTerrainMipLevels(coords)
ET-->>PS: mipLevels[6]
PS->>EM: GetParallaxCoords(..., mipLevels, viewDir, tbn, noise, params, sharedOffset)
EM->>Tex: sample displacement/color (stochastic or SampleLevel)
Tex-->>EM: height samples
EM->>EM: ray-march & secant refine -> pixelOffset, parallax UV
EM-->>PS: parallax UV, pixelOffset, hasPOM
PS->>ET: ComputeTerrainParallaxShadowBaseHeight(..., sharedOffset)
ET->>Tex: sample heights along light dir
Tex-->>ET: shadow samples
ET-->>PS: terrain parallax shadow multiplier
CPU->>PS: shared settings (parallax/shadow quality, EnableShadows)
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. Comment |
Automated formatting by clang-format, prettier, and other hooks. See https://pre-commit.ci for details.
|
No actionable suggestions for changed features. |
Automated formatting by clang-format, prettier, and other hooks. See https://pre-commit.ci for details.
Automated formatting by clang-format, prettier, and other hooks. See https://pre-commit.ci for details.
![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
🧹 Nitpick comments (3)
package/Shaders/Common/LightingLandscape.hlsli (1)
126-126: Align closing header-guard comment with macro nameThe closing comment references
__LANDSCAPE_LAYERS_HLSLI__, but the actual guard is__LIGHTING_LANDSCAPE_HLSLI__. Consider syncing for clarity.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@package/Shaders/Common/LightingLandscape.hlsli` at line 126, The closing header-guard comment is mismatched: update the comment after `#endif` to match the actual macro __LIGHTING_LANDSCAPE_HLSLI__ (replace __LANDSCAPE_LAYERS_HLSLI__ with __LIGHTING_LANDSCAPE_HLSLI__), ensuring the comment aligns with the guard macro used at the top of this file.features/Terrain Variation/Shaders/Features/TerrainVariation.ini (1)
1-2: Consider adding a linked issue keyword in the PR descriptionIf this PR implements tracked perf/feature work, add a reference like
Addresses #<id>for traceability.As per coding guidelines, feature PRs should include issue references using keywords such as “Addresses
#123” / “Implements#123”.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@features/Terrain` Variation/Shaders/Features/TerrainVariation.ini around lines 1 - 2, This PR is missing an issue reference for traceability—update the PR description to include a keywords-style reference like "Addresses #<issue-id>" or "Implements #<issue-id>" with the actual issue number; additionally, add a RelatedIssue field under the [Info] section in TerrainVariation.ini (e.g., add "RelatedIssue = `#123`" next to "Version = 1-1-0") so the feature (TerrainVariation.ini / [Info] / Version) explicitly records the linked issue.features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli (1)
23-27: PR metadata follow-up: add scope in title and link tracking issue.Suggested title:
perf(extended-materials): optimize terrain parallax path
If this maps to a tracked task, appendAddresses #<id>(orImplements #<id>) in the PR description.As per coding guidelines, "When reviewing PRs, please provide suggestions for: Conventional Commit Titles ... and Issue References ..."
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@features/Extended` Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli around lines 23 - 27, Update the PR title to include scope and the suggested conventional-commit format — e.g. change it to "perf(extended-materials): optimize terrain parallax path" — and add an issue reference in the PR description like "Addresses #<id>" (or "Implements #<id>") if this change maps to a tracked task; note this PR touches constants such as ParallaxCheapDistance, ParallaxNearShadowQuality, ParallaxFarShadowQuality, and TerrainParallaxShadowMaxMipLevel so include the issue link and short summary referencing the terrain parallax/extended-materials change in the description.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@features/Extended`
Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli:
- Around line 127-139: Normalize guard: after summing weights[] into wsum, avoid
calling rcp(wsum) when wsum is zero/near-zero by computing invwsum =
rcp(max(wsum, 1e-6)) (or choose a small epsilon) and then multiply weights[k] by
invwsum; update both occurrences that use rcp(wsum) (the block that references
heightBlend, weights, wsum, invwsum and the similar block later in the file) so
degenerate/invalid blends no longer produce INF/NaN.
In `@package/Shaders/Common/LightingLandscape.hlsli`:
- Around line 91-94: The non-full-PBR branch double-applies weight by
multiplying landRMAOS with weight at assignment and again when accumulating into
blendedRMAOS; change the assignment of landRMAOS inside that branch to not
multiply by weight (e.g., set landRMAOS = float4(1 - glossiness.x, 0, 1, 0)) and
keep the accumulation blendedRMAOS += landRMAOS * weight so each layer’s RMAOS
is weighted exactly once (references: landRMAOS, blendedRMAOS, blendedRGB,
landColorRGB, glossiness.x, weight).
---
Nitpick comments:
In `@features/Extended`
Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli:
- Around line 23-27: Update the PR title to include scope and the suggested
conventional-commit format — e.g. change it to "perf(extended-materials):
optimize terrain parallax path" — and add an issue reference in the PR
description like "Addresses #<id>" (or "Implements #<id>") if this change maps
to a tracked task; note this PR touches constants such as ParallaxCheapDistance,
ParallaxNearShadowQuality, ParallaxFarShadowQuality, and
TerrainParallaxShadowMaxMipLevel so include the issue link and short summary
referencing the terrain parallax/extended-materials change in the description.
In `@features/Terrain` Variation/Shaders/Features/TerrainVariation.ini:
- Around line 1-2: This PR is missing an issue reference for traceability—update
the PR description to include a keywords-style reference like "Addresses
#<issue-id>" or "Implements #<issue-id>" with the actual issue number;
additionally, add a RelatedIssue field under the [Info] section in
TerrainVariation.ini (e.g., add "RelatedIssue = `#123`" next to "Version = 1-1-0")
so the feature (TerrainVariation.ini / [Info] / Version) explicitly records the
linked issue.
In `@package/Shaders/Common/LightingLandscape.hlsli`:
- Line 126: The closing header-guard comment is mismatched: update the comment
after `#endif` to match the actual macro __LIGHTING_LANDSCAPE_HLSLI__ (replace
__LANDSCAPE_LAYERS_HLSLI__ with __LIGHTING_LANDSCAPE_HLSLI__), ensuring the
comment aligns with the guard macro used at the top of this file.
🪄 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: 57ffdc06-34ee-4729-b25a-f3f07a561d70
📒 Files selected for processing (10)
features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlslifeatures/Terrain Variation/Shaders/Features/TerrainVariation.inifeatures/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlslipackage/Shaders/Common/LightingLandscape.hlslipackage/Shaders/Common/SharedData.hlslipackage/Shaders/Lighting.hlslsrc/Features/ExtendedMaterials.cppsrc/Features/ExtendedMaterials.hsrc/Features/TerrainVariation.cppsrc/Features/TerrainVariation.h
💤 Files with no reviewable changes (1)
- src/Features/ExtendedMaterials.cpp
Contributor
There was a problem hiding this comment.
Pull request overview
This PR restructures landscape/terrain shader paths to reduce branching and improve performance, while reworking Terrain Variation sampling and updating Extended Materials’ parallax/shadow logic (including secant-based refinement).
Changes:
- Refactors
Lighting.hlsllandscape layer blending into a shared include (LightingLandscape.hlsli) and streamlines Terrain Variation integration. - Removes
ExtendShadowssetting and adjusts Extended Materials shader constants/helpers for cheaper parallax + shadow evaluation. - Reworks Terrain Variation stochastic sampling to a consistent 2-sample approach and updates feature settings structs/padding accordingly.
Reviewed changes
Copilot reviewed 10 out of 10 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| src/Features/TerrainVariation.h | Adjusts shader-define gating and aligns Settings layout to 16 bytes. |
| src/Features/TerrainVariation.cpp | Updates UI toggles to write uint32_t settings and removes vertex-desc “dirty” forcing. |
| src/Features/ExtendedMaterials.h | Removes ExtendShadows and fixes Settings padding/alignment. |
| src/Features/ExtendedMaterials.cpp | Drops JSON/UI handling for ExtendShadows. |
| package/Shaders/Lighting.hlsl | Major landscape/parallax/TV integration changes; adds landscape blend helpers and terrain shadow caching. |
| package/Shaders/Common/SharedData.hlsli | Removes ExtendShadows from shader-visible CPM settings and adjusts padding. |
| package/Shaders/Common/LightingLandscape.hlsli | New shared landscape blending macro include to reduce duplicated per-layer code. |
| features/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlsli | Reworks TV stochastic sampling (2-sample blend, distance/minification fallbacks, LOD jitter helpers). |
| features/Terrain Variation/Shaders/Features/TerrainVariation.ini | Bumps Terrain Variation feature version. |
| features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli | Adds new parallax/shadow helpers, secant refinement, and terrain shadow evaluation helpers. |
Collaborator
Author
|
@coderabbitai full review please be mean like doodlum and no niceties. Point out any magic numbers and sloppily written code. Suggest improvements where you feel appropriate. Suggest ideas to further reduce compile times |
|
✅ A pre-release build is available for this PR: |
doodlum
reviewed
Apr 30, 2026
doodlum
marked this pull request as draft
April 30, 2026 23:33
davo0411
marked this pull request as ready for review
May 4, 2026 04:31
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (7)
features/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlsli (2)
27-33: 💤 Low valueConsider documenting or removing
offset3fromStochasticOffsets.With the transition to a 2-sample model,
offset3is always set to0(lines 101, 118). The field is retained for struct compatibility, but this could confuse future maintainers who might expect it to be used.Consider adding a comment explaining that
offset3is unused (kept for ABI stability), or remove it if no external consumers depend on the struct layout.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@features/Terrain` Variation/Shaders/TerrainVariation/TerrainVariation.hlsli around lines 27 - 33, The struct StochasticOffsets contains an unused field offset3 (always set to zero under the new 2-sample model); update the declaration of StochasticOffsets to either (a) add a clear comment above offset3 explaining it is intentionally unused and kept only for ABI/struct compatibility, referencing the struct name StochasticOffsets and the field offset3, or (b) remove offset3 entirely if you confirm no external consumers rely on the existing layout (search for StochasticOffsets usages before removing). Ensure the chosen change is applied where StochasticOffsets is defined so future maintainers understand why offset3 is present or has been removed.
111-113: 💤 Low valueMagic numbers in hash computation.
Lines 111 and 113 use unexplained constants (
255437.0,127.0) for the LOD hash computation. While these are typical in hash functions, brief comments would help future maintainers understand the rationale (e.g., "large prime for uniform distribution" or "offset for second independent sample").🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@features/Terrain` Variation/Shaders/TerrainVariation/TerrainVariation.hlsli around lines 111 - 113, The hash computation uses unexplained magic constants (255437.0 and 127.0) in the cellID and second sample; update the code around the cellID calculation and the hashLOD calls to either replace these literals with named constants (e.g., LARGE_PRIME_FOR_HASH and SECOND_SAMPLE_OFFSET) or add brief comments above the lines mentioning why the values were chosen (e.g., "large prime for uniform distribution" and "offset for independent second sample"), referencing the variables cellID, landscapeUV and the hashLOD(...) calls so maintainers can locate and understand the rationale.features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli (2)
735-737: 💤 Low valueIncorrect closing comment.
Line 736's comment says
// defined(LANDSCAPE) && defined(TERRAIN_VARIATION)but this#endifcloses the#if defined(LANDSCAPE)block that started at line 346 (and implicitly includes line 584). TheTERRAIN_VARIATIONpart is not required for this guard.♻️ Suggested fix
# undef TERRAIN_HEIGHT_AT -#endif // defined(LANDSCAPE) && defined(TERRAIN_VARIATION) +#endif // defined(LANDSCAPE)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@features/Extended` Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli around lines 735 - 737, The closing preprocessor comment on the `#endif` after the `#undef` TERRAIN_HEIGHT_AT is wrong: it references "defined(LANDSCAPE) && defined(TERRAIN_VARIATION)" but the block being closed is the `#if` defined(LANDSCAPE) block (and not dependent on TERRAIN_VARIATION). Update the comment on that `#endif` to accurately reflect the opening guard (e.g., // defined(LANDSCAPE)) so the `#endif` matches the `#if` that began the LANDSCAPE section; refer to the surrounding symbols (`#if` defined(LANDSCAPE), TERRAIN_VARIATION, `#undef` TERRAIN_HEIGHT_AT) to locate the correct closing directive.
619-628: 💤 Low valueInconsistent tap-count thresholds between functions.
TerrainDirectionalShadowTapCountuses thresholds[0.7, 0.25, 0.0]whileParallaxShadowTapCountuses[0.75, 0.5, 0.25]. The inconsistency may be intentional (directional shadows use fewer taps), but the differing threshold values (0.7 vs 0.75, 0.25 vs 0.5) could benefit from a brief comment explaining the rationale.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@features/Extended` Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli around lines 619 - 628, TerrainDirectionalShadowTapCount currently uses thresholds (0.7, 0.25, 0.0) that differ from ParallaxShadowTapCount's (0.75, 0.5, 0.25); either align the thresholds if this was accidental or add a concise comment above the TerrainDirectionalShadowTapCount definition explaining the intentional rationale (e.g., fewer taps for directional shadows, performance/quality tradeoffs and why 0.7/0.25 were chosen vs 0.75/0.5/0.25), referencing the function names TerrainDirectionalShadowTapCount and ParallaxShadowTapCount so future readers know the difference is deliberate.package/Shaders/Lighting.hlsl (3)
2257-2268: 💤 Low valueVariable
lightDirectionTSshadows outer declaration.Line 2261 redeclares
lightDirectionTSinside the conditional block, but it's already declared and computed at line 2252 with the same expression. This variable shadowing is unnecessary and could confuse maintainers.♻️ Suggested fix
if (LANDSCAPE_PARALLAX_ENABLED) { if (ExtendedMaterials::TerrainHasSignificantBlend(input.LandBlendWeights1, input.LandBlendWeights2.xy) && ExtendedMaterials::TerrainHasAnyDisplacement()) { float terrainHeightScale = ExtendedMaterials::TerrainMaxWeightedHeightScale(input, displacementParams); if (terrainHeightScale > 0.01) { - float3 lightDirectionTS = normalize(mul(refractedLightDirection, tbn).xyz); # if defined(TERRAIN_VARIATION) parallaxShadow = ExtendedMaterials::GetParallaxSoftShadowMultiplierTerrain(input, uv, terrainShadowMipLevels, lightDirectionTS, sh0, terrainDirectionalShadowQuality, screenNoise, displacementParams, sharedOffset); # else🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@package/Shaders/Lighting.hlsl` around lines 2257 - 2268, The variable lightDirectionTS is being redeclared inside the LANDSCAPE_PARALLAX_ENABLED branch although it was already computed earlier (using normalize(mul(refractedLightDirection, tbn).xyz);), causing shadowing; remove the local re-declaration and reuse the previously declared lightDirectionTS before calling ExtendedMaterials::GetParallaxSoftShadowMultiplierTerrain (and keep the existing arguments: input, uv, terrainShadowMipLevels, lightDirectionTS, sh0, terrainDirectionalShadowQuality, screenNoise, displacementParams, sharedOffset) so behavior is unchanged.
1025-1029: 💤 Low valueRedundant assignment of
terrainDirectionalShadowQuality.On line 1026,
terrainDirectionalShadowQualityis initialized toparallaxShadowQuality, and immediately on lines 1027-1029, it's conditionally reassigned to the same value whenLANDSCAPEis defined. The second assignment is a no-op.♻️ Suggested simplification
float parallaxShadowQuality = ExtendedMaterials::ParallaxNearShadowQuality; - float terrainDirectionalShadowQuality = parallaxShadowQuality; -# if defined(LANDSCAPE) - terrainDirectionalShadowQuality = ExtendedMaterials::ParallaxNearShadowQuality; -# endif + float terrainDirectionalShadowQuality = ExtendedMaterials::ParallaxNearShadowQuality;🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@package/Shaders/Lighting.hlsl` around lines 1025 - 1029, The variable terrainDirectionalShadowQuality is initialized to parallaxShadowQuality and then conditionally reassigned to the same ExtendedMaterials::ParallaxNearShadowQuality inside the LANDSCAPE block, which is redundant; remove the conditional reassignment (the lines inside the `#if` defined(LANDSCAPE) / `#endif` that set terrainDirectionalShadowQuality = ExtendedMaterials::ParallaxNearShadowQuality) so terrainDirectionalShadowQuality remains initialized from parallaxShadowQuality without a no-op repeat, or alternatively move the initial assignment into the conditional and drop the unconditional one—update references to parallaxShadowQuality, terrainDirectionalShadowQuality, and the LANDSCAPE conditional accordingly.
1031-1036: 💤 Low valueConsider documenting implicit macro dependencies.
The
COMPUTE_TERRAIN_SHADOW_BASEandEVAL_TERRAIN_DIR_SHADOWmacros capture multiple variables implicitly (input,uv,terrainShadowMipLevels,terrainDirectionalShadowQuality,screenNoise,displacementParams,sharedOffset). While this reduces duplication, it makes the code harder to reason about.A brief comment listing the expected variables in scope would help maintainability.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@package/Shaders/Lighting.hlsl` around lines 1031 - 1036, Add a short comment above the COMPUTE_TERRAIN_SHADOW_BASE and EVAL_TERRAIN_DIR_SHADOW macro definitions that lists the implicit variables those macros expect in scope (e.g., input, uv, terrainShadowMipLevels, terrainDirectionalShadowQuality, screenNoise, displacementParams, sharedOffset) and mention which ExtendedMaterials functions they map to (ExtendedMaterials::ComputeTerrainParallaxShadowBaseHeight and ExtendedMaterials::EvaluateTerrainDirectionalParallaxShadowMultiplier) so future readers can quickly see required context before the macros are used.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@features/Extended`
Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli:
- Around line 735-737: The closing preprocessor comment on the `#endif` after the
`#undef` TERRAIN_HEIGHT_AT is wrong: it references "defined(LANDSCAPE) &&
defined(TERRAIN_VARIATION)" but the block being closed is the `#if`
defined(LANDSCAPE) block (and not dependent on TERRAIN_VARIATION). Update the
comment on that `#endif` to accurately reflect the opening guard (e.g., //
defined(LANDSCAPE)) so the `#endif` matches the `#if` that began the LANDSCAPE
section; refer to the surrounding symbols (`#if` defined(LANDSCAPE),
TERRAIN_VARIATION, `#undef` TERRAIN_HEIGHT_AT) to locate the correct closing
directive.
- Around line 619-628: TerrainDirectionalShadowTapCount currently uses
thresholds (0.7, 0.25, 0.0) that differ from ParallaxShadowTapCount's (0.75,
0.5, 0.25); either align the thresholds if this was accidental or add a concise
comment above the TerrainDirectionalShadowTapCount definition explaining the
intentional rationale (e.g., fewer taps for directional shadows,
performance/quality tradeoffs and why 0.7/0.25 were chosen vs 0.75/0.5/0.25),
referencing the function names TerrainDirectionalShadowTapCount and
ParallaxShadowTapCount so future readers know the difference is deliberate.
In `@features/Terrain` Variation/Shaders/TerrainVariation/TerrainVariation.hlsli:
- Around line 27-33: The struct StochasticOffsets contains an unused field
offset3 (always set to zero under the new 2-sample model); update the
declaration of StochasticOffsets to either (a) add a clear comment above offset3
explaining it is intentionally unused and kept only for ABI/struct
compatibility, referencing the struct name StochasticOffsets and the field
offset3, or (b) remove offset3 entirely if you confirm no external consumers
rely on the existing layout (search for StochasticOffsets usages before
removing). Ensure the chosen change is applied where StochasticOffsets is
defined so future maintainers understand why offset3 is present or has been
removed.
- Around line 111-113: The hash computation uses unexplained magic constants
(255437.0 and 127.0) in the cellID and second sample; update the code around the
cellID calculation and the hashLOD calls to either replace these literals with
named constants (e.g., LARGE_PRIME_FOR_HASH and SECOND_SAMPLE_OFFSET) or add
brief comments above the lines mentioning why the values were chosen (e.g.,
"large prime for uniform distribution" and "offset for independent second
sample"), referencing the variables cellID, landscapeUV and the hashLOD(...)
calls so maintainers can locate and understand the rationale.
In `@package/Shaders/Lighting.hlsl`:
- Around line 2257-2268: The variable lightDirectionTS is being redeclared
inside the LANDSCAPE_PARALLAX_ENABLED branch although it was already computed
earlier (using normalize(mul(refractedLightDirection, tbn).xyz);), causing
shadowing; remove the local re-declaration and reuse the previously declared
lightDirectionTS before calling
ExtendedMaterials::GetParallaxSoftShadowMultiplierTerrain (and keep the existing
arguments: input, uv, terrainShadowMipLevels, lightDirectionTS, sh0,
terrainDirectionalShadowQuality, screenNoise, displacementParams, sharedOffset)
so behavior is unchanged.
- Around line 1025-1029: The variable terrainDirectionalShadowQuality is
initialized to parallaxShadowQuality and then conditionally reassigned to the
same ExtendedMaterials::ParallaxNearShadowQuality inside the LANDSCAPE block,
which is redundant; remove the conditional reassignment (the lines inside the
`#if` defined(LANDSCAPE) / `#endif` that set terrainDirectionalShadowQuality =
ExtendedMaterials::ParallaxNearShadowQuality) so terrainDirectionalShadowQuality
remains initialized from parallaxShadowQuality without a no-op repeat, or
alternatively move the initial assignment into the conditional and drop the
unconditional one—update references to parallaxShadowQuality,
terrainDirectionalShadowQuality, and the LANDSCAPE conditional accordingly.
- Around line 1031-1036: Add a short comment above the
COMPUTE_TERRAIN_SHADOW_BASE and EVAL_TERRAIN_DIR_SHADOW macro definitions that
lists the implicit variables those macros expect in scope (e.g., input, uv,
terrainShadowMipLevels, terrainDirectionalShadowQuality, screenNoise,
displacementParams, sharedOffset) and mention which ExtendedMaterials functions
they map to (ExtendedMaterials::ComputeTerrainParallaxShadowBaseHeight and
ExtendedMaterials::EvaluateTerrainDirectionalParallaxShadowMultiplier) so future
readers can quickly see required context before the macros are used.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro Plus
Run ID: e416c1af-4478-4a0b-a78b-1dd58c13a03b
📒 Files selected for processing (6)
features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlslifeatures/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlslipackage/Shaders/Common/SharedData.hlslipackage/Shaders/Lighting.hlslsrc/Features/TerrainVariation.cppsrc/Features/TerrainVariation.h
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 20, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 20, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 20, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 20, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 20, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 20, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 21, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 21, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 21, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 21, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 21, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 21, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
Collaborator
|
@coderabbitai review this source code, will it improve performance? |
Contributor
|
Tip For best results, initiate chat on the files or code changes. 🧩 Analysis chain🏁 Script executed: #!/bin/bash
set -euo pipefail
echo '=== Changed files and statistics ==='
git diff --stat HEAD^ HEAD
git diff --name-status HEAD^ HEAD
echo
echo '=== Outline: core parallax implementation ==='
ast-grep outline 'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsParallaxCore.hlsli' --items all --view expanded
echo
echo '=== Outline: terrain implementation ==='
ast-grep outline 'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsTerrain.hlsli' --items all --view expanded
echo
echo '=== Outline: lighting integration ==='
ast-grep outline package/Shaders/Lighting.hlsl --items all --view expanded
echo
echo '=== Targeted performance-relevant diff ==='
git diff --unified=80 HEAD^ HEAD -- \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsParallaxCore.hlsli' \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsTerrain.hlsli' \
'features/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlsli' \
package/Shaders/Lighting.hlsl \
package/Shaders/Common/LightingLandscape.hlsli \
package/Shaders/Common/SharedData.hlsli \
src/Features/ExtendedMaterials.cpp \
src/Features/ExtendedMaterials.h \
src/Features/TerrainVariation.cpp \
src/Features/TerrainVariation.hLength of output: 3172 🏁 Script executed: #!/bin/bash
set -euo pipefail
repo='community-shaders/skyrim-community-shaders'
pr=2220
echo '=== PR head/base metadata ==='
gh pr view "$pr" --repo "$repo" --json number,title,baseRefName,headRefName,baseRefOid,headRefOid,commits \
--jq '{number,title,base:.baseRefName,head:.headRefName,baseOid:.baseRefOid,headOid:.headRefOid,commits:[.commits[]|{oid:.oid,messageHeadline:.messageHeadline}]}'
echo
echo '=== PR changed files ==='
gh api "repos/$repo/pulls/$pr/files?per_page=100" --paginate --jq '.[] | [.status, .filename, (.additions|tostring), (.deletions|tostring)] | `@tsv`'
echo
echo '=== Local checkout identity and available refs ==='
git rev-parse HEAD
git branch -a --no-color | sed -n '1,100p'
git remote -v
echo
echo '=== Locate expected shader files in this checkout (if present under a different path) ==='
fd -HI 'ExtendedMaterialsParallaxCore\.hlsli|ExtendedMaterialsTerrain\.hlsli|Lighting\.hlsl|TerrainVariation\.hlsli' . | sed -n '1,120p'Length of output: 10463 🏁 Script executed: #!/bin/bash
set -euo pipefail
base='04482fcbc1746bd1f6875723180dd5e03e7e05d7'
head='5c591c25c6ab413447b924245569852f2aff428e'
echo '=== Aggregate diff: parallax core ==='
git diff --unified=35 "$base...$head" -- \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsParallaxCore.hlsli'
echo
echo '=== Aggregate diff: terrain + terrain variation ==='
git diff --unified=30 "$base...$head" -- \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsTerrain.hlsli' \
'features/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlsli'
echo
echo '=== Aggregate diff: lighting call sites and layer blending ==='
git diff --unified=35 "$base...$head" -- \
package/Shaders/Lighting.hlsl \
package/Shaders/Common/LightingLandscape.hlsli
echo
echo '=== Current implementation: key symbols and call sites ==='
rg -n -C 12 \
'GetParallaxCoords\s*\(|GetParallaxSoftShadowMultiplier\s*\(|GetTerrainHeight\s*\(|GetTerrainHeightQuadRayMarch\s*\(|ComputeTerrainParallaxShadowBaseHeight\s*\(|EvaluateTerrain(Directional)?ParallaxShadowMultiplier\s*\(|TerrainMaxWeightedHeightScale\s*\(|TerrainHasSignificantBlend\s*\(' \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsParallaxCore.hlsli' \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsTerrain.hlsli' \
package/Shaders/Lighting.hlsl
echo
echo '=== Loop context around lighting and expensive terrain/parallax calls ==='
rg -n -C 20 \
'for\s*\(|while\s*\(|GetTerrainHeight\s*\(|EvaluateTerrain(Directional)?ParallaxShadowMultiplier\s*\(|GetParallaxSoftShadowMultiplier\s*\(' \
package/Shaders/Lighting.hlslLength of output: 198125 🏁 Script executed: #!/bin/bash
set -euo pipefail
base='04482fcbc1746bd1f6875723180dd5e03e7e05d7'
echo '=== Base: original parallax and terrain code ==='
git show "$base:features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli" \
| rg -n -C 20 \
'GetParallaxCoords\s*\(|GetParallaxSoftShadowMultiplier\s*\(|GetTerrainHeight\s*\(|GetTerrainHeightQuadRayMarch\s*\(|StochasticEffectParallax\s*\(|StochasticEffect\s*\(|GetMipLevel\s*\('
echo
echo '=== Current: include placement, mip helpers, and tuning values ==='
rg -n -C 12 \
'`#include`|GetMipLevel\s*\(|GetMipLevelFromDims\s*\(|ComputeParallaxMarchMip\s*\(|ParallaxCheapDistance|ParallaxNearShadowQuality|ParallaxFarShadowQuality|TerrainParallaxShadowMaxMipLevel|ParallaxShadowTapCount\s*\(' \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterials.hlsli' \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsParallaxCore.hlsli' \
'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsTerrain.hlsli'
echo
echo '=== Line-numbered hot paths for review findings ==='
nl -ba 'features/Extended Materials/Shaders/ExtendedMaterials/ExtendedMaterialsTerrain.hlsli' \
| sed -n '315,395p'
nl -ba package/Shaders/Lighting.hlsl | sed -n '1178,1240p'
nl -ba package/Shaders/Lighting.hlsl | sed -n '2318,2350p'
nl -ba 'features/Terrain Variation/Shaders/TerrainVariation/TerrainVariation.hlsli' | sed -n '115,205p'
echo
echo '=== Settings layouts and writer-side synchronization ==='
rg -n -C 10 \
'struct.*Settings|TerrainVariationSettings|enableLODTerrainTilingFix|enableTilingFix|ExtendShadows|UpdateShaderSettings' \
package/Shaders/Common/SharedData.hlsli \
src/Features/TerrainVariation.h \
src/Features/TerrainVariation.cpp \
src/Features/ExtendedMaterials.h \
src/Features/ExtendedMaterials.cppLength of output: 49788 The PR description image 1 is a hand-drawn secant-interpolation diagram. It derives The PR description image 2 is a CAPFRAMEX chart comparing
Changes that should improve GPU performance
What prevents calling this a universal win
Remaining hot-path issue
The first and third are pixel-invariant after the terrain setup. Compute them once before the point-light loop, for example as Moreover, return (w1.x + w1.y + w1.z + w1.w + w2.x + w2.y) > 0.01;The weights were normalized earlier, so their sum is normally Also, VerdictYes, this PR should improve runtime performance in the intended Terrain Variation + terrain-parallax workloads. The two strongest reasons are the 2-tap/1-tap stochastic sampling and the single-tap cached terrain shadowing. The CAPFRAMEX result is consistent with that. The remaining point-light-loop work should be hoisted before claiming the implementation is fully optimized, and the performance claim needs repeated GPU profiling to verify that reduced texture work was not offset by register pressure or divergence. |
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 22, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 22, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 22, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 22, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 22, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 22, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 23, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 23, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 23, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 23, 2026
Squashed sync of community-shaders#2220: a full rework of Extended Materials parallax (secant-stepping march) and Terrain Variation (deterministic 2-sample stochastic blend), plus a landscape-layer-blend refactor split out of Lighting.hlsl into LightingLandscape.hlsli / ExtendedMaterialsParallaxCore.hlsli / ExtendedMaterialsTerrain.hlsli. 1. Refactored lighting/landscape layers to facilitate small compiler wins for terrain variation (TV) and PBR, reducing branching inside lighting.hlsl. 2. Adopted secant stepping instead of linear for parallax occlusion mapping, per graphics discord discussion. 3. Full terrain variation rework: now always takes 2 samples. 4. General comment/code cleanup and magic-number fixes. Fork-specific adaptations (VR mip bias, distance-based parallax quality falloff, dead-code cleanup, standards fixes) are layered on top as separate commits, not part of this squashed import. Co-Authored-By: davo0411 <davidkehoe0411@outlook.com>
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 23, 2026
Upstream PR 2220 made ExtendedMaterials::GetMipLevel deterministic (dropped the screen-noise dither), which silently swallowed our second VR-only mip bump alongside it. VR's texel-density halving at the top of the function survived the merge; this restores the additive mipLevel++ that ran after log2() as well, matching our prior VR-conservative mip bias. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
alandtse
added a commit
to alandtse/open-shaders
that referenced
this pull request
Jul 23, 2026
Upstream PR 2220's secant-stepping rewrite of GetParallaxCoords dropped the distance-parameterized near/far quality falloff that had landed on our dev independently after the PR's base (upstream removed it entirely as part of the rework, calling it "distance shenanigans" in their own history). The new mip-based step scaling serves a related purpose for near/mid distances, but had no hard cutoff for far geometry, silently regressing the cost savings on distant terrain/meshes. Restores the `distance` parameter, a far-distance early-out (skip the march entirely past the cutoff), and a smooth near-band blend of the final result and pixel offset so the transition doesn't pop. The far-distance early-out also needs an exception: pre-merge dev's GetParallaxCoords always ran the march for TRUE_PBR materials flagged PBR::Flags::InterlayerParallax regardless of distance (it has no distance-cheap approximation), so the restored cutoff carries that bypass in the non-LANDSCAPE TRUE_PBR path. Part of syncing community-shaders#2220 ("perf: terrain variation & parallax", davo0411).
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
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.
Summary by CodeRabbit
New Features
Refactor
Chores