[PR #2266] [MERGED] shader_recompiler: Fix cube sampling coordinates. #2608

New issue

Closed

opened 2026-02-27 22:00:34 +03:00 by kerem · 0 comments

kerem commented

2026-02-27 22:00:34 +03:00

Owner

📋 Pull Request Information

Original PR: https://github.com/shadps4-emu/shadPS4/pull/2266
Author: @squidbus
Created: 1/29/2025
Status: ✅ Merged
Merged: 1/30/2025
Merged by: @squidbus

Base: main ← Head: fix-cube

📝 Commits (1)

cd20260 shader_recompiler: Fix cube sampling coordinates.

📊 Changes

2 files changed (+21 additions, -6 deletions)

View changed files

📝 src/shader_recompiler/ir/passes/resource_tracking_pass.cpp (+19 -6)
📝 src/shader_recompiler/specialization.h (+2 -0)

📄 Description

AMD adds 1.5 to cube coordinates to put them in the range [1.0, 2.0], see here for more details: https://gpuopen.com/learn/fetching-from-cubes-and-octahedrons/

We need the coordinates in the range [0.0, 1.0] for sampling, so subtract 1.f from each coordinate. As a result we also need shader specialization for whether an image is a cube, to emit the right coordinate fix-up. This likely won't vary too much, as when cube calculations are used the image should basically always be a cube.

Fixes incorrect character lighting in CUSA06093.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/shadps4-emu/shadPS4/pull/2266 **Author:** [@squidbus](https://github.com/squidbus) **Created:** 1/29/2025 **Status:** ✅ Merged **Merged:** 1/30/2025 **Merged by:** [@squidbus](https://github.com/squidbus) **Base:** `main` ← **Head:** `fix-cube` --- ### 📝 Commits (1) - [`cd20260`](https://github.com/shadps4-emu/shadPS4/commit/cd20260bf63502b3b9acd73ae634deedce41ed15) shader_recompiler: Fix cube sampling coordinates. ### 📊 Changes **2 files changed** (+21 additions, -6 deletions) <details> <summary>View changed files</summary> 📝 `src/shader_recompiler/ir/passes/resource_tracking_pass.cpp` (+19 -6) 📝 `src/shader_recompiler/specialization.h` (+2 -0) </details> ### 📄 Description AMD adds `1.5` to cube coordinates to put them in the range `[1.0, 2.0]`, see here for more details: https://gpuopen.com/learn/fetching-from-cubes-and-octahedrons/ We need the coordinates in the range `[0.0, 1.0]` for sampling, so subtract `1.f` from each coordinate. As a result we also need shader specialization for whether an image is a cube, to emit the right coordinate fix-up. This likely won't vary too much, as when cube calculations are used the image should basically always be a cube. Fixes incorrect character lighting in CUSA06093. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>