swung0x48/MobileGL
1
0
Fork 0
mirror of https://github.com/MobileGL-Dev/MobileGL synced 2026-09-13 06:38:31 +09:00
Code Issues Packages Projects Releases Wiki Activity

[Refactor] (ShaderTranspiler, MG_Backend, MG_Util): gate the subgroup prefix-scan rewrite behind a generic device-quirk registry with GPU vendor detection and MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN override, warn on template mismatch, and block ARB/NV subgroup spellings

Browse Source
This commit is contained in:
swung0x48
2026-07-20 02:36:26 -04:00
parent 293f64b3c2
commit bf7b5755cc
10 changed files with 219 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
MobileGL/MG_Util/ShaderTranspiler/ShaderSourceProcessor.h
+4 -2
View File
@@ -27,8 +27,10 @@ namespace MobileGL {
// wider than the capture's 32 lanes. For the narrowly recognized, uniform-control-
// flow template, replace the subgroup-local scan with a shared-memory, strict
// left-fold over virtual 32-lane segments. Returns true only when the complete safe
// template was recognized and rewritten. DirectVulkan calls this through
// PreprocessShaderSource; the explicit entry point exists for deterministic tests.
// template was recognized and rewritten. PreprocessShaderSource reaches this through
// its device-quirk registry: by default only on detected Qualcomm Vulkan devices,
// overridable either way with MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN=1/0. The explicit
// entry point exists for deterministic tests.
Bool RewriteLinearSubgroupPrefixScanForVulkan(ShaderStage stage, Uint32 nativeSubgroupSize, String& source);
// Rewrites a "#version 330 core" directive that PreprocessShaderSource normalized down