[PR #1763] [MERGED] graphics: Improve handling of color buffer and storage image swizzles #2286

New issue

Closed

opened 2026-02-27 21:15:54 +03:00 by kerem · 0 comments

kerem commented

2026-02-27 21:15:54 +03:00

Owner

📋 Pull Request Information

Original PR: https://github.com/shadps4-emu/shadPS4/pull/1763
Author: @raphaelthegreat
Created: 12/13/2024
Status: ✅ Merged
Merged: 12/13/2024
Merged by: @raphaelthegreat

Base: main ← Head: better-swizzle

📝 Commits (4)

08ed66f liverpool_to_vk: Remove wrong component swap formats
eba2549 shader_recompiler: Handle storage and buffer format swizzles
3963ad9 shader_recompiler: Skip unsupported depth export
7d7352c image_view: Remove image format swizzle

📊 Changes

6 files changed (+66 additions, -49 deletions)

View changed files

📝 src/shader_recompiler/frontend/translate/export.cpp (+5 -0)
📝 src/shader_recompiler/ir/passes/resource_tracking_pass.cpp (+42 -0)
📝 src/shader_recompiler/specialization.h (+10 -1)
📝 src/video_core/amdgpu/resource.h (+9 -0)
📝 src/video_core/renderer_vulkan/liverpool_to_vk.cpp (+0 -9)
📝 src/video_core/texture_cache/image_view.cpp (+0 -39)

📄 Description

Removed some color buffer component swaps that were wrong since eR8G8B8A8Unorm and eA8B8G8R8UnormPack32 are technically the same format for little-endian GPUs, which both AMD and NVIDIA are. This lets the shader specialization mechanism handle it instead.
Handle swizzles for storage image writes by swizzling incoming texel in the shader, as well as format buffer swizzles since BUFFER_LOAD_FORMAT and BUFFER_STORE_FORMAT use dst_sel from the sharp. ~~For storage images specifically we only perform manual swizzling if the format isn't 8_8_8_8 since those can be swizzled with image view and is already mostly handled~~
Ignore mrtz stencil exports since those are unsupported for now

_{🔄 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/1763 **Author:** [@raphaelthegreat](https://github.com/raphaelthegreat) **Created:** 12/13/2024 **Status:** ✅ Merged **Merged:** 12/13/2024 **Merged by:** [@raphaelthegreat](https://github.com/raphaelthegreat) **Base:** `main` ← **Head:** `better-swizzle` --- ### 📝 Commits (4) - [`08ed66f`](https://github.com/shadps4-emu/shadPS4/commit/08ed66f9001997a3c139d359c8ae43304501a0d2) liverpool_to_vk: Remove wrong component swap formats - [`eba2549`](https://github.com/shadps4-emu/shadPS4/commit/eba2549d3a694aafd73e7793d3ed895da252fc75) shader_recompiler: Handle storage and buffer format swizzles - [`3963ad9`](https://github.com/shadps4-emu/shadPS4/commit/3963ad93b169a2d038b944788be2ab5add1e1597) shader_recompiler: Skip unsupported depth export - [`7d7352c`](https://github.com/shadps4-emu/shadPS4/commit/7d7352c1f20fd79e79ae9db415833a30cdd2d313) image_view: Remove image format swizzle ### 📊 Changes **6 files changed** (+66 additions, -49 deletions) <details> <summary>View changed files</summary> 📝 `src/shader_recompiler/frontend/translate/export.cpp` (+5 -0) 📝 `src/shader_recompiler/ir/passes/resource_tracking_pass.cpp` (+42 -0) 📝 `src/shader_recompiler/specialization.h` (+10 -1) 📝 `src/video_core/amdgpu/resource.h` (+9 -0) 📝 `src/video_core/renderer_vulkan/liverpool_to_vk.cpp` (+0 -9) 📝 `src/video_core/texture_cache/image_view.cpp` (+0 -39) </details> ### 📄 Description * Removed some color buffer component swaps that were wrong since eR8G8B8A8Unorm and eA8B8G8R8UnormPack32 are technically the same format for little-endian GPUs, which both AMD and NVIDIA are. This lets the shader specialization mechanism handle it instead. * Handle swizzles for storage image writes by swizzling incoming texel in the shader, as well as format buffer swizzles since BUFFER_LOAD_FORMAT and BUFFER_STORE_FORMAT use dst_sel from the sharp. ~For storage images specifically we only perform manual swizzling if the format isn't 8_8_8_8 since those can be swizzled with image view and is already mostly handled~ * Ignore mrtz stencil exports since those are unsupported for now --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>