[PR #1532] [CLOSED] shader_recompiler: shaer patch to manualy normalize on unsupported Vulkan UNORM/SNORM formats #2156

New issue

Closed

opened 2026-02-27 21:15:24 +03:00 by kerem · 0 comments

kerem commented

2026-02-27 21:15:24 +03:00

Owner

📋 Pull Request Information

Original PR: https://github.com/shadps4-emu/shadPS4/pull/1532
Author: @LNDF
Created: 11/14/2024
Status: ❌ Closed

Base: main ← Head: normalization

📝 Commits (8)

ed5da02 Copy IR instruction
e498875 Initial prep
a4010b3 Add surface formats
78ad50b Patch read and write
008d2ac Surface format fixup
fe3b41c Shader specialization
0fd4c3b Handle atomic instructions
a4dba67 Specialize normalization component number

📊 Changes

6 files changed (+136 additions, -6 deletions)

View changed files

📝 src/shader_recompiler/ir/ir_emitter.cpp (+5 -0)
📝 src/shader_recompiler/ir/ir_emitter.h (+2 -0)
📝 src/shader_recompiler/ir/passes/resource_tracking_pass.cpp (+65 -0)
📝 src/shader_recompiler/specialization.h (+21 -6)
📝 src/video_core/amdgpu/resource.h (+16 -0)
📝 src/video_core/renderer_vulkan/liverpool_to_vk.cpp (+27 -0)

📄 Description

Vulkan doesn't have support for SNORM/UNORM number formats for 32, 32_32, 32_32_32 and 32_32_32_32 data formats. Insteed use SINT/UINT and normalize/unormalize on shader when reading/writing.

Maybe we don't need to specialize by number of components, I can revert that.

_{🔄 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/1532 **Author:** [@LNDF](https://github.com/LNDF) **Created:** 11/14/2024 **Status:** ❌ Closed **Base:** `main` ← **Head:** `normalization` --- ### 📝 Commits (8) - [`ed5da02`](https://github.com/shadps4-emu/shadPS4/commit/ed5da0263cf213d8a1a893ad71858221f857a506) Copy IR instruction - [`e498875`](https://github.com/shadps4-emu/shadPS4/commit/e49887535f545b6b236f254d19f1aff164dc8e7e) Initial prep - [`a4010b3`](https://github.com/shadps4-emu/shadPS4/commit/a4010b33125fa01fad270529691269dbd761284f) Add surface formats - [`78ad50b`](https://github.com/shadps4-emu/shadPS4/commit/78ad50bf56c56001383ed30afcc4e39d22358ca3) Patch read and write - [`008d2ac`](https://github.com/shadps4-emu/shadPS4/commit/008d2ac1db0225379020ee4935feac1fa2f801b7) Surface format fixup - [`fe3b41c`](https://github.com/shadps4-emu/shadPS4/commit/fe3b41c64bfd50f15cf7f1ede6b6a2fe4d2b20f1) Shader specialization - [`0fd4c3b`](https://github.com/shadps4-emu/shadPS4/commit/0fd4c3b307bb97ff8bb583effb63bd4909f9b421) Handle atomic instructions - [`a4dba67`](https://github.com/shadps4-emu/shadPS4/commit/a4dba67e0c43f84c4650a38d16af04cec568802f) Specialize normalization component number ### 📊 Changes **6 files changed** (+136 additions, -6 deletions) <details> <summary>View changed files</summary> 📝 `src/shader_recompiler/ir/ir_emitter.cpp` (+5 -0) 📝 `src/shader_recompiler/ir/ir_emitter.h` (+2 -0) 📝 `src/shader_recompiler/ir/passes/resource_tracking_pass.cpp` (+65 -0) 📝 `src/shader_recompiler/specialization.h` (+21 -6) 📝 `src/video_core/amdgpu/resource.h` (+16 -0) 📝 `src/video_core/renderer_vulkan/liverpool_to_vk.cpp` (+27 -0) </details> ### 📄 Description Vulkan doesn't have support for SNORM/UNORM number formats for 32, 32_32, 32_32_32 and 32_32_32_32 data formats. Insteed use SINT/UINT and normalize/unormalize on shader when reading/writing. Maybe we don't need to specialize by number of components, I can revert that. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>