[Fix] (Diligent): fix y inversion and front-face winding in viewport mapping

- Translate OpenGL bottom-left viewport/scissor rects to Diligent top-left origin
- Invert FrontCounterClockwise to compensate for the Y conversion
- Add DrawsTopHalfTriangleFromMobileGLState y-orientation test

.github/workflows/apk.yml

@@ -209,7 +209,7 @@ jobs:
       - name: Load trace cases
         id: trace-cases
         run: |
-          echo "android=$(python3 tools/trace_replay/trace_cases.py --ci --format github-apk)" >> "$GITHUB_OUTPUT"
+          echo "android=$(python3 tools/trace_replay/trace_cases.py --ci --format github-apk-matrix)" >> "$GITHUB_OUTPUT"
           echo "names=$(python3 tools/trace_replay/trace_cases.py --ci --format names)" >> "$GITHUB_OUTPUT"
 
   trace-fixtures:
@@ -337,13 +337,7 @@ jobs:
     strategy:
       fail-fast: false
       max-parallel: 4
-      matrix:
-        backend:
-          - name: DirectGLES
-            gpu: software
-          - name: DirectVulkan
-            gpu: lavapipe
-        case: ${{ fromJSON(needs.trace-cases.outputs.android) }}
+      matrix: ${{ fromJSON(needs.trace-cases.outputs.android) }}
     steps:
       - name: Set Swap Space
         uses: pierotofy/set-swap-space@v1.0

.github/workflows/test.yml

@@ -491,6 +491,7 @@ jobs:
       - benchmark
       - integration
     outputs:
+      matrix: ${{ steps.trace-cases.outputs.matrix }}
       names: ${{ steps.trace-cases.outputs.names }}
     steps:
       - name: Checkout repo
@@ -498,7 +499,9 @@ jobs:
 
       - name: Load trace cases
         id: trace-cases
-        run: echo "names=$(python3 tools/trace_replay/trace_cases.py --ci --format names)" >> "$GITHUB_OUTPUT"
+        run: |
+          echo "matrix=$(python3 tools/trace_replay/trace_cases.py --ci --format github-test-matrix)" >> "$GITHUB_OUTPUT"
+          echo "names=$(python3 tools/trace_replay/trace_cases.py --ci --format names)" >> "$GITHUB_OUTPUT"
 
   trace-fixtures:
     name: trace fixture (${{ matrix.case }})
@@ -577,11 +580,7 @@ jobs:
     strategy:
       fail-fast: false
       max-parallel: 4
-      matrix:
-        backend:
-          - DirectGLES
-          - DirectVulkan
-        case: ${{ fromJSON(needs.trace-cases.outputs.names) }}
+      matrix: ${{ fromJSON(needs.trace-cases.outputs.matrix) }}
 
     steps:
       - name: Set Swap Space

.gitignore

@@ -1,4 +1,4 @@
-################################################################################
+################################################################################
 # 此 .gitignore 文件已由 Microsoft(R) Visual Studio 自动创建。
 ################################################################################
 
@@ -16,6 +16,9 @@ MobileGLCodeManager
 MobileGL/MG_Test/build
 /build_*
 /cmake-build*
+/build-*/
+/local.properties
+/.jspace/
 .idea
 MobileGL/MG*/build*
 MobileGL/MG*/cmake-build*

CMakeLists.txt

@@ -199,7 +199,6 @@ set(SPIRV_REFLECT_ENABLE_ASSERTS OFF CACHE BOOL  "Enable asserts for debugging"
 set(SPIRV_REFLECT_ENABLE_ASAN    OFF CACHE BOOL "Use address sanitization" FORCE)
 set(SPIRV_REFLECT_INSTALL        OFF CACHE BOOL "Whether to install" FORCE)
 
-# add_subdirectory(3rdparty/DiligentCore)
 add_subdirectory(3rdparty/glslang)
 add_subdirectory(3rdparty/SPIRV-Cross)
 add_subdirectory(3rdparty/VulkanMemoryAllocator)
@@ -211,6 +210,23 @@ set(XXHASH_BUILD_XXHSUM OFF)
 option(BUILD_SHARED_LIBS OFF)
 add_subdirectory(3rdparty/xxHash/build/cmake xxhash_build EXCLUDE_FROM_ALL)
 
+# Diligent-based backend. Enabled by default on local builds; only the Vulkan
+# engine from DiligentCore is built. Added after the other 3rdparty projects so
+# DiligentCore reuses the glslang / SPIRV-Cross / SPIRV-Tools / xxHash targets
+# already defined by MobileGL instead of building its bundled copies.
+option(MOBILEGL_ENABLE_DILIGENT "Enable the Diligent/Vulkan backend" ON)
+if(MOBILEGL_ENABLE_DILIGENT)
+    set(DILIGENT_NO_DIRECT3D11 ON CACHE BOOL "Disable Direct3D11 backend" FORCE)
+    set(DILIGENT_NO_DIRECT3D12 ON CACHE BOOL "Disable Direct3D12 backend" FORCE)
+    set(DILIGENT_NO_OPENGL     ON CACHE BOOL "Disable OpenGL backend" FORCE)
+    set(DILIGENT_NO_METAL      ON CACHE BOOL "Disable Metal backend" FORCE)
+    set(DILIGENT_NO_WEBGPU     ON CACHE BOOL "Disable WebGPU backend" FORCE)
+    set(DILIGENT_NO_ARCHIVER   ON CACHE BOOL "Disable Archiver" FORCE)
+    set(DILIGENT_BUILD_TESTS   OFF CACHE BOOL "Build Diligent tests" FORCE)
+    set(DILIGENT_INSTALL_CORE  OFF CACHE BOOL "Install DiligentCore" FORCE)
+    add_subdirectory(3rdparty/DiligentCore)
+endif()
+
 set(TRACY_ENABLE ${MOBILEGL_ENABLE_TRACY} CACHE BOOL "Enable Tracy, this is an internal variable" FORCE)
 
 if (TRACY_ENABLE)
@@ -396,6 +412,14 @@ set(SOURCE_FILES
     MobileGL/MG_State/GLState/RenderbufferState/RenderbufferState.cpp
 )
 
+if(MOBILEGL_ENABLE_DILIGENT)
+    list(APPEND SOURCE_FILES
+        MobileGL/MG_Backend/Diligent/BackendObject_Diligent.cpp
+        MobileGL/MG_Backend/Diligent/DiligentVulkan.cpp
+        MobileGL/MG_Backend/Diligent/Renderer/DiligentRenderer.cpp
+    )
+endif()
+
 if (APPLE AND NOT MOBILEGL_IOS)
     list(APPEND SOURCE_FILES
         MobileGL/MG_Impl/CGLImpl/CGLImpl.cpp
@@ -436,6 +460,21 @@ set(MOBILEGL_LINK_LIBRARIES
         Threads::Threads
 )
 
+if(MOBILEGL_ENABLE_DILIGENT)
+    list(APPEND MOBILEGL_LINK_LIBRARIES
+        Diligent-GraphicsEngineVk-static
+        Diligent-GraphicsEngine
+        Diligent-GraphicsEngineNextGenBase
+        Diligent-GraphicsAccessories
+        Diligent-ShaderTools
+        Diligent-GraphicsTools
+        Diligent-Common
+        Diligent-Primitives
+        Diligent-TargetPlatform
+        Vulkan::Headers
+    )
+endif()
+
 set(MOBILEGL_COMPILE_DEF
         -DVMA_STATIC_VULKAN_FUNCTIONS=0
         -DVMA_DYNAMIC_VULKAN_FUNCTIONS=1
@@ -501,6 +540,7 @@ target_compile_definitions(${CMAKE_PROJECT_NAME}
         ${MOBILEGL_COMPILE_DEF}
         MOBILEGL_LOG_ACTIVE_LEVEL=${MOBILEGL_LOG_ACTIVE_LEVEL}
         $<$<BOOL:${MOBILEGL_TRACE_ANGLE_VARIANTS}>:MOBILEGL_TRACE_ANGLE_VARIANTS=1>
+        $<$<BOOL:${MOBILEGL_ENABLE_DILIGENT}>:MOBILEGL_ENABLE_DILIGENT=1>
 )
 
 if(UNIX AND NOT APPLE AND NOT ANDROID)
@@ -559,6 +599,7 @@ if(NOT ANDROID)
             PUBLIC
             ${MOBILEGL_COMPILE_DEF}
             MOBILEGL_LOG_ACTIVE_LEVEL=${MOBILEGL_LOG_ACTIVE_LEVEL}
+            $<$<BOOL:${MOBILEGL_ENABLE_DILIGENT}>:MOBILEGL_ENABLE_DILIGENT=1>
     )
 endif()
 

HANDOFF_DILIGENT.md

@@ -0,0 +1,278 @@
+# Handoff: Diligent/Vulkan GL3.2 Backend for MobileGL
+
+Date: 2026-08-18  
+Branch: `feat/diligent-vulkan-backend`  
+Repo: `~/MobileGL-dev`  
+Status: **Active work-in-progress. Do not mark complete yet.**
+
+---
+
+## 1. Goal
+
+Implement a complete OpenGL 3.2 front-end emulation on a new Diligent/Vulkan backend inside MobileGL, instead of the DirectVulkan / DirectGLES backends.
+
+Target state:
+- Fully wire MobileGL front-end `MG_State` (buffers, VAO, program, texture, sampler, framebuffer, render-state) into Diligent.
+- Implement all GL 3.2 core entry points through the Diligent backend.
+- Pass local non-Android GL3.2 tests on the Turnip Adreno 750 GPU.
+
+---
+
+## 2. Current Branch / Commits
+
+Latest 12 commits on `feat/diligent-vulkan-backend`:
+
+```
+2f5abf83 test(diligent): verify indexed DrawElements path from real frontend state
+c31b7381 feat(diligent): add basic texture binding and textured state-draw test
+8945c507 feat(diligent): clear depth in GL Clear when GL_DEPTH_BUFFER_BIT set
+558d3aea feat(diligent): add offscreen depth target and depth clear
+7e2f0bc8 feat(diligent): wire stencil and color-mask state into state PSO
+f99786f6 feat(diligent): wire viewport/scissor state into state draws
+be4cc3ce feat(diligent): wire blend/depth/cull render state into state PSO
+c855e6cf feat(diligent): verify state-driven draw with real MobileGL frontend state
+02e60bfa feat(diligent): add state-driven draw path (VAO/buffer/program to Diligent)
+a9515c92 feat(diligent): add dynamic vertex buffer upload path
+2f57582a feat(diligent): wire Clear/Draw/Present into GLFunctionsTable
+beb21123 feat(diligent): add real offscreen renderer with clear and triangle draw
+```
+
+Working tree is clean.
+
+---
+
+## 3. Key Files
+
+### Backend core
+
+- `MobileGL/MG_Backend/Diligent/BackendObject_Diligent.h/.cpp`
+  - `BackendObject_Diligent`
+  - Creates Diligent Vulkan device/context
+  - Owns `DiligentRenderer`
+  - Wires `GLFunctionsTable`:
+    - `Clear` (color + depth)
+    - `DrawArrays`
+    - `DrawElements`
+    - `Present`
+- `MobileGL/MG_Backend/Diligent/DiligentVulkan.h/.cpp`
+  - Backend identity helper / translation unit
+- `MobileGL/MG_Backend/Diligent/Renderer/DiligentRenderer.h/.cpp`
+  - Offscreen RGBA8 + D32F targets
+  - Clear / ClearDepth / DrawTriangle / DrawVertices
+  - `CreateTestTexture` (RGBA8 texture + SRV + sampler)
+  - `DrawFromState` (main front-end emulation draw path)
+  - `CreatePipelineFromState`:
+    - SPIR-V → Diligent shaders via SPIRV-Reflect
+    - VAO attributes → input layout
+    - primitive topology from GL mode
+    - blend / depth / cull / stencil / color-mask state
+  - `UploadVertexDataFromState`:
+    - packs enabled VAO attributes from `BufferObject` into interleaved vertex buffer
+    - supports `DrawArrays`, `DrawElements`, triangle-fan and line-loop expansion
+  - Static texture binding to `g_Texture` through PSO static variables + SRB
+
+### Integration changes
+
+- `CMakeLists.txt`
+  - New option `MOBILEGL_ENABLE_DILIGENT` (default ON for local)
+  - DiligentCore added **after** glslang/SPIRV-Cross/xxHash/Vulkan-Headers so it reuses existing CMake targets
+  - Diligent static libraries linked into `MobileGL` / `MobileGL_s`
+  - New Diligent backend sources added
+- `MobileGL/MG_Backend/BackendObject.h`
+  - New `BackendType::DiligentVulkan`
+- `MobileGL/MG_Backend/Init.cpp`
+  - New backend switch case
+- `MobileGL/ConfigLoader.cpp`
+  - `MOBILEGL_BACKEND_TYPE=DiligentVulkan` accepted
+- `MobileGL/MG_Test/CMakeLists.txt`
+  - New `MobileGL/MG_Test/Backend/Diligent` subdirectory
+- `MobileGL/MG_Test/Backend/Diligent/`
+  - `CMakeLists.txt`
+  - `SanityTest.cpp`
+
+### Local test files
+
+- `MobileGL/MG_Test/Backend/Diligent/SanityTest.cpp`
+  - `CreatesDiligentDeviceAndAdvertisesGL32`
+  - `ClearsAndDrawsTriangleOffscreen`
+  - `DrawsFromMobileGLState`
+  - `DrawsTexturedFromMobileGLState`
+  - `DrawsIndexedFromMobileGLState`
+  - `DrawsRealTexturedFromMobileGLState`
+  - `DrawsUniformFromMobileGLState`
+  - `DrawsToOffscreenFramebufferFromMobileGLState`
+  - `DrawsWithScissorFromMobileGLState`
+  - `DrawsWithBlendFromMobileGLState`
+  - `DrawsWithDepthTestFromMobileGLState`
+  - `DrawsNamedUniformBlockFromMobileGLState`
+  - `DrawsWithStencilTestFromMobileGLState`
+  - `DrawsToRenderbufferFramebufferFromMobileGLState`
+  - `DrawsToMultipleColorAttachmentsFromMobileGLState`
+  - `DrawsIndexedBaseVertexFromMobileGLState`
+
+---
+
+## 4. What Works Today
+
+Verified locally on Turnip Adreno 750:
+
+- Diligent device/context creation
+- EGL window-surface swapchain creation path through Diligent `ISwapChain` (offscreen tests still use the offscreen target)
+- GL 3.2 / GLSL 1.50 capability advertisement
+- Offscreen color + depth rendering
+- Clear color and depth
+- Real mobilegl front-end state-driven drawing:
+  - Program SPIR-V → Diligent shaders
+  - VAO attributes + bound GL buffer → interleaved vertex buffer
+  - `DrawArrays` path
+  - `DrawElements` path (index buffer)
+- Texture basics:
+  - Offscreen texture creation
+  - CPU → Diligent texture (`CreateTestTexture`)
+  - Static sampler2D binding to `g_Texture`
+  - Textured draw test passes
+- Render state:
+  - Blend enable/factors/equations
+  - Stencil clear + test enabled on a D24S8 default depth/stencil target
+  - Depth test enable/func/write mask
+  - Cull face enable/mode/front-face winding
+  - Stencil test enable/masks/ops/func/ref
+  - Color write mask
+  - Viewport
+  - Scissor rect
+- Texture/sampler full integration:
+  - `ITextureObject` → Diligent `ITexture` + SRV with automatic dirty upload
+  - `SamplerObject` / texture-object sampler → Diligent `ISampler`
+  - Real front-end `glTexImage2D` path (not only `CreateTestTexture`) verified
+- Global UBO upload:
+  - Front-end `glUniform*` shadow → Diligent uniform buffer bound as `MGL_GLOBAL_UBO`
+- User framebuffer mapping:
+  - Current draw/read FBO resolves texture attachments to Diligent RTV/DSV
+  - `ReadPixels` can read back from a user FBO color attachment
+- More GL entry points wired:
+  - `DrawRangeElements` / `DrawRangeElementsBaseVertex`
+  - `DrawElementsBaseVertex` with real baseVertex selection
+  - `MultiDrawArrays` / `MultiDrawElements` / `MultiDrawElementsBaseVertex`
+  - `DrawArraysInstanced` / `DrawElementsInstanced` family
+  - Indirect draw CPU fallback: `DrawArraysIndirect`, `DrawElementsIndirect`, `MultiDraw*Indirect`, `*IndirectCount`
+  - `ClearBufferfv` / `ClearBufferfi` / `ClearBufferiv` / `ClearBufferuiv` (incl. stencil clear)
+  - `BlitFramebuffer` / `BlitNamedFramebuffer` (same-size color copy between read/draw FBOs)
+  - `CopyTexImage2D` / `CopyTexSubImage2D` (whole-color copy fallback)
+  - `CopyImageSubData` (whole-texture copy between two texture objects)
+  - `GenerateMipmap` (Diligent GPU mip generation on state textures)
+  - `GetTexImage` / `GetTextureImage` (RGBA8 readback)
+  - Fence sync entries (`FenceSync` / `ClientWaitSync` / `WaitSync` / `DeleteSync` / `GetSyncStatus`) as CPU always-signaled fallback
+  - Timer query entries (`BeginTimeElapsedQuery` / `EndTimeElapsedQuery` / `QueryCounterTimestamp` / `GetQueryResult64` etc.) as CPU `steady_clock` fallback
+  - `ReadPixels` from default and user color attachments
+- Primitive expansion:
+  - `GL_TRIANGLE_FAN` expanded to triangle list
+  - `GL_LINE_LOOP` expanded to line strip
+- Local test result:
+
+```
+[  PASSED  ] 16 tests
+```
+
+---
+
+## 5. How to Build and Run Locally
+
+From repo root `~/MobileGL-dev`:
+
+```bash
+cmake -S . -B build-diligent -G Ninja \
+  -DCMAKE_BUILD_TYPE=Debug \
+  -DMOBILEGL_ENABLE_DILIGENT=ON \
+  -DMOBILEGL_BUILD_TEST=ON \
+  -DMOBILEGL_BUILD_BENCHMARK=OFF \
+  -DFETCHCONTENT_SOURCE_DIR_GOOGLETEST="$PWD/3rdparty/DiligentCore/ThirdParty/googletest"
+
+cmake --build build-diligent --target DiligentVulkanSanityTest -j 4
+
+./build-diligent/MobileGL/MG_Test/Backend/Diligent/DiligentVulkanSanityTest --gtest_color=no
+```
+
+Notes:
+- `MOBILEGL_BUILD_BENCHMARK=OFF` avoids network fetch of google/benchmark in this environment.
+- `FETCHCONTENT_SOURCE_DIR_GOOGLETEST` pins googletest to DiligentCore's bundled copy, avoiding flaky network clone.
+- Max 4 cores is intentional: use `-j 4`.
+
+---
+
+## 6. Environment Notes
+
+- Host: Linux `aarch64`, glibc 2.43 (Fedora container on Android/Droidspaces)
+- GPU: Turnip Adreno 750, Vulkan API 1.4.354
+- GPU nodes available:
+  - `/dev/dri/renderD128`
+  - `/dev/kgsl-3d0`
+- Android SDK/NDK: `~/android-sdk` (aarch64 glibc)
+  - NDK `27.3.13750724`
+  - CMake `3.22.1`
+- JDK/Gradle for APK builds:
+  - `~/android-build-tools/jdk17`
+  - `~/android-build-tools/gradle/gradle-8.10.2`
+
+---
+
+## 7. Known Limitations / Not Yet Implemented
+
+- User framebuffers now support texture color attachments, renderbuffer color readback, multiple simultaneous color targets, and depth/stencil texture or renderbuffer attachments.
+- Textures auto-sync `ITextureObject` → Diligent resources, including mip levels and sampler state; compressed textures and integer/3-channel formats that Diligent lacks are still skipped.
+- Global UBO (default-block `glUniform*`) and named application UBO blocks (through `glBindBufferBase`/`glUniformBlockBinding`) now upload and bind; SSBOs are still not fed from frontend buffer bindings.
+- Swapchain creation and resize are wired for native EGL window surfaces via `Diligent::ISwapChain`; `Present()` presents the active swap chain when present and otherwise flushes the offscreen target. Actual on-screen EGL presentation is still untested in this headless environment, and the X11 display/connection fields are not yet plumbed through `WindowHandle`. `SetSwapInterval` now forwards the requested sync interval to `ISwapChain::Present()`.
+- No transform feedback / GPU-accelerated queries / non-color readback; fence sync and timer queries use CPU fallbacks.
+- Draw range, multi-draw, instanced-draw wrappers, clear-buffer, blit, read-pixels, CopyTexImage*, CopyImageSubData, GenerateMipmap, GetTexImage/GetTextureImage and indirect draws are now wired; buffer subdata paths still remain.
+- A last-PSO cache now avoids recreating the pipeline when program/render-state/topology/VAO layout is unchanged; texture/UBO resources are still rebound dynamically per draw.
+- The `GLFunctionsTable` is only partially populated.
+
+---
+
+## 8. Recommended Next Steps
+
+1. **Framebuffer / Renderbuffer mapping**
+   - [x] Map `MG_State::GLState::FramebufferObject` attachments to Diligent `ITextureView` / `ITexture`.
+   - [x] Support default framebuffer as current offscreen target.
+   - [x] Support `glBindFramebuffer`, `glFramebufferTexture2D`, renderbuffer color/depth attachments and renderbuffer color readback.
+   - [x] Multiple simultaneous color attachments.
+
+2. **Texture / Sampler full integration**
+   - [x] Translate MobileGL `ITextureObject` to Diligent `ITexture` and cache by `GetLifetimeId()`.
+   - [x] Propagate texture unit bindings into the PSO SRB.
+   - [x] Translate `SamplerObject` state into Diligent `SamplerDesc`.
+
+3. **Uniform / UBO support**
+   - [x] Create Diligent buffer for `ProgramObject::GetUBOData()` / `GetUBOSize()`.
+   - [x] Bind the global UBO as a dynamic shader resource.
+   - [x] Handle per-program uniform block bindings / named UBO blocks.
+
+4. **PSO / resource caching**
+   - [~] Cache PSOs by program + VAO config + render state + topology (single last-PSO fast path).
+   - [~] Cache textures and samplers; buffers/SRBs can still be re-bound per draw.
+
+5. **More GL 3.2 entry points**
+   - [x] `DrawRangeElements`
+   - [x] `MultiDraw*`
+   - [x] `BlitFramebuffer` (same-size color copy)
+   - [x] `ReadPixels` from non-default framebuffer
+   - [x] `CopyTexImage*` / `CopyImageSubData` wired as whole-resource copies
+   - [x] `GetTexImage` / `GetTextureImage` (RGBA8)
+   - [x] Indirect draws (CPU fallback)
+
+6. **Expand local test suite**
+   - [x] Scissor test
+   - [x] Blend test
+   - [x] Texture filtering / sampler state test
+   - [x] framebuffer offscreen render-to-texture test
+   - [x] Depth test visual test
+   - [x] Stencil test
+
+---
+
+## 9. Handoff Notes for Next Agent
+
+- Do **not** reference `origin/Deprecated/Feat/Diligent`; that old implementation is intentionally ignored.
+- Work from this branch, keep tests green.
+- The command `./build-diligent/.../DiligentVulkanSanityTest` runs all 5 Diligent tests.
+- If a new test crashes during shader resource binding, remember Diligent texture SRVs need a sampler attached via `ITextureView::SetSampler()` before `InitializeStaticSRBResources()`.
+- When re-creating a PSO or buffer, call `Release()` (or assign `nullptr`) before the create call to avoid Diligent debug “Overwriting reference” assertions.

MobileGL/Config.h

@@ -128,10 +128,6 @@ namespace MobileGL::MG_Config {
         // explicitly request a core profile via EGL_CONTEXT_OPENGL_PROFILE_MASK / a >=3.1
         // version request.
         Bool RelaxedSemantics = false;
-        // MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN: overrides the shader-source quirk that
-        // rewrites the recognized workgroup prefix-scan template on Qualcomm devices with
-        // subgroups wider than 32 lanes (see ShaderSourceProcessor's quirk registry).
-        QuirkOverride SubgroupPrefixScanQuirk = QuirkOverride::Auto;
         // MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE: overrides the DirectVulkan quirk that
         // strips depth writes from accumulation-blended pipelines (MIN/MAX or additive
         // ONE+ONE - the multi-pass depth-equality signature) on drivers without

MobileGL/ConfigLoader.cpp

@@ -180,7 +180,6 @@ namespace MobileGL::MG_ConfigLoader {
         features.EsprytForceDepthStencilReadbackEmulation =
             QueryEnvFlag("MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION");
         features.RelaxedSemantics = QueryEnvFlag("MOBILEGL_RELAXED_SEMANTICS");
-        features.SubgroupPrefixScanQuirk = QueryEnvQuirkOverride("MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN");
         features.MagmaDisableBlendedDepthWriteQuirk =
             QueryEnvQuirkOverride("MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE");
         features.DisableRobustBufferAccess = QueryEnvFlag("MOBILEGL_DISABLE_ROBUST_BUFFER_ACCESS");
@@ -203,6 +202,7 @@ namespace MobileGL::MG_ConfigLoader {
     }
         ENTRY(DirectGLES)
         ENTRY(DirectVulkan)
+        ENTRY(DiligentVulkan)
         ENTRY(Unknown)
         MG_Config::ActiveBackendType = BackendType::Unknown;
 #undef ENTRY

MobileGL/MG_Backend/BackendObject.h

@@ -19,6 +19,7 @@ namespace MobileGL {
     enum class BackendType {
         DirectGLES,
         DirectVulkan,
+        DiligentVulkan,
         BackendTypeCount,
         Unknown = -1
     };

MobileGL/MG_Backend/Diligent/BackendObject_Diligent.cpp

@@ -0,0 +1,906 @@
+// MobileGL - MobileGL/MG_Backend/Diligent/BackendObject_Diligent.cpp
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+
+#include "BackendObject_Diligent.h"
+#include "DiligentVulkan.h"
+#include "Renderer/DiligentRenderer.h"
+#include <MG_Backend/BackendObject.h>
+#include <MG_Backend/BackendObjects.h>
+#include <MG_State/GLState/Core.h>
+
+#include <EngineFactoryVk.h>
+#include <RenderDevice.h>
+#include <DeviceContext.h>
+
+#include <exception>
+#include <chrono>
+
+namespace MobileGL::MG_Backend::DiligentBackend {
+    namespace {
+        const RendererInfo BuildInitialRendererInfo() {
+            RendererInfo info;
+            info.RendererName = "MobileGL (Diligent/Vulkan)";
+            info.BackendName = "Diligent Vulkan";
+            info.RendererGLInfo.TargetGLVersion = {3, 2, 0};
+            info.RendererGLInfo.TargetGLSLVersion = {1, 50, 0};
+            info.RendererGLInfo.IsCompatibilityProfile = false;
+            return info;
+        }
+
+        DiligentRenderer* GetActiveRenderer() {
+            auto* backend = dynamic_cast<BackendObject_Diligent*>(pActiveBackendObject.get());
+            return backend != nullptr ? backend->GetRenderer() : nullptr;
+        }
+
+        struct DrawArraysIndirectCommand {
+            Uint32 Count = 0;
+            Uint32 InstanceCount = 0;
+            Uint32 First = 0;
+            Uint32 BaseInstance = 0;
+        };
+
+        struct DrawElementsIndirectCommand {
+            Uint32 Count = 0;
+            Uint32 InstanceCount = 0;
+            Uint32 FirstIndex = 0;
+            Int32 BaseVertex = 0;
+            Uint32 BaseInstance = 0;
+        };
+
+        struct CpuTimerQuery {
+            std::chrono::steady_clock::time_point Start;
+            Uint64 TimestampNs = 0;
+            Bool Available = false;
+        };
+
+        const Uint8* ResolveIndirectCommandBytes(const void* indirect, SizeT requiredBytes, const char* label) {
+            auto drawBuffer = MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::DrawIndirect).GetBoundObject();
+            if (drawBuffer) {
+                drawBuffer->SyncPersistentMappedRange();
+                const SizeT commandOffset = reinterpret_cast<SizeT>(indirect);
+                if (drawBuffer->MappedData() == nullptr || commandOffset + requiredBytes > drawBuffer->GetSize()) {
+                    MGLOG_E_ONCE("%s skipped: invalid GL_DRAW_INDIRECT_BUFFER binding or range", label);
+                    return nullptr;
+                }
+                return drawBuffer->MappedData() + commandOffset;
+            }
+            if (indirect == nullptr) {
+                MGLOG_E_ONCE("%s skipped: indirect pointer is null", label);
+                return nullptr;
+            }
+            return reinterpret_cast<const Uint8*>(indirect);
+        }
+
+        void Clear(GLbitfield mask) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr) {
+                return;
+            }
+            if ((mask & GL_COLOR_BUFFER_BIT) != 0) {
+                const auto& color = MG_State::pGLContext->GetClearColor();
+                renderer->Clear(color.x(), color.y(), color.z(), color.w());
+            }
+            if ((mask & GL_DEPTH_BUFFER_BIT) != 0) {
+                renderer->ClearDepth(MG_State::pGLContext->GetClearDepth());
+            }
+            if ((mask & GL_STENCIL_BUFFER_BIT) != 0) {
+                renderer->ClearStencil(MG_State::pGLContext->GetClearStencil());
+            }
+        }
+
+        void DrawArrays(GLenum mode, GLint first, GLsizei count) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->DrawFromState(mode, first, count, 0, nullptr);
+            }
+        }
+
+        void DrawElements(GLenum mode, GLsizei count, GLenum type, const void* indices) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->DrawFromState(mode, 0, count, type, indices);
+            }
+        }
+
+        void DrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type,
+                               const void* indices) {
+            // The CPU-side UploadVertexDataFromState path already honors the selected index
+            // range. start/end only restrict which indices may be referenced; they do not
+            // change the vertex buffer layout for this backend.
+            (void)start;
+            (void)end;
+            DrawElements(mode, count, type, indices);
+        }
+
+        void DrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type,
+                                         const void* indices, GLint basevertex) {
+            (void)start;
+            (void)end;
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->DrawFromState(mode, 0, count, type, indices, basevertex);
+            }
+        }
+
+        void MultiDrawArrays(GLenum mode, const GLint* first, const GLsizei* count, GLsizei drawcount) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr) {
+                return;
+            }
+            for (GLsizei i = 0; i < drawcount; ++i) {
+                if (count[i] > 0) {
+                    renderer->DrawFromState(mode, first[i], count[i], 0, nullptr);
+                }
+            }
+        }
+
+        void MultiDrawElements(GLenum mode, const GLsizei* count, GLenum type, const GLvoid* const* indices,
+                               GLsizei drawcount) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr) {
+                return;
+            }
+            for (GLsizei i = 0; i < drawcount; ++i) {
+                if (count[i] > 0) {
+                    renderer->DrawFromState(mode, 0, count[i], type, indices[i]);
+                }
+            }
+        }
+
+        void DrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type, const void* indices,
+                                    GLint basevertex) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->DrawFromState(mode, 0, count, type, indices, basevertex);
+            }
+        }
+
+        void MultiDrawElementsBaseVertex(GLenum mode, const GLsizei* count, GLenum type,
+                                         const GLvoid* const* indices, GLsizei drawcount,
+                                         const GLint* basevertex) {
+            for (GLsizei i = 0; i < drawcount; ++i) {
+                if (count[i] > 0) {
+                    DrawElementsBaseVertex(mode, count[i], type, indices[i],
+                                           basevertex != nullptr ? basevertex[i] : 0);
+                }
+            }
+        }
+
+        void DrawArraysIndirect(GLenum mode, const void* indirect) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr) {
+                return;
+            }
+            const auto* bytes = ResolveIndirectCommandBytes(indirect, sizeof(DrawArraysIndirectCommand),
+                                                            "DrawArraysIndirect");
+            if (bytes == nullptr) {
+                return;
+            }
+            DrawArraysIndirectCommand cmd{};
+            std::memcpy(&cmd, bytes, sizeof(cmd));
+            if (cmd.Count == 0 || cmd.InstanceCount == 0) {
+                return;
+            }
+            for (Uint32 i = 0; i < cmd.InstanceCount; ++i) {
+                renderer->DrawFromState(mode, static_cast<GLint>(cmd.First), static_cast<GLsizei>(cmd.Count),
+                                        0, nullptr);
+            }
+        }
+
+        void DrawElementsIndirect(GLenum mode, GLenum type, const void* indirect) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr) {
+                return;
+            }
+            const SizeT indexSize = MG_Util::GetGLTypeSize(type);
+            if (indexSize == 0) {
+                return;
+            }
+            const auto* bytes = ResolveIndirectCommandBytes(indirect, sizeof(DrawElementsIndirectCommand),
+                                                            "DrawElementsIndirect");
+            if (bytes == nullptr) {
+                return;
+            }
+            DrawElementsIndirectCommand cmd{};
+            std::memcpy(&cmd, bytes, sizeof(cmd));
+            if (cmd.Count == 0 || cmd.InstanceCount == 0) {
+                return;
+            }
+            const void* indices = reinterpret_cast<const void*>(static_cast<SizeT>(cmd.FirstIndex) * indexSize);
+            for (Uint32 i = 0; i < cmd.InstanceCount; ++i) {
+                renderer->DrawFromState(mode, 0, static_cast<GLsizei>(cmd.Count), type, indices,
+                                        static_cast<GLint>(cmd.BaseVertex));
+            }
+        }
+
+        void MultiDrawArraysIndirect(GLenum mode, const void* indirect, GLsizei drawcount, GLsizei stride) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr || drawcount <= 0) {
+                return;
+            }
+            const GLsizei realStride = stride == 0 ? static_cast<GLsizei>(sizeof(DrawArraysIndirectCommand)) : stride;
+            for (GLsizei i = 0; i < drawcount; ++i) {
+                const auto* bytes = ResolveIndirectCommandBytes(
+                    static_cast<const Uint8*>(indirect) + static_cast<SizeT>(i) * static_cast<SizeT>(realStride),
+                    sizeof(DrawArraysIndirectCommand), "MultiDrawArraysIndirect");
+                if (bytes == nullptr) {
+                    continue;
+                }
+                DrawArraysIndirectCommand cmd{};
+                std::memcpy(&cmd, bytes, sizeof(cmd));
+                if (cmd.Count == 0 || cmd.InstanceCount == 0) {
+                    continue;
+                }
+                for (Uint32 instance = 0; instance < cmd.InstanceCount; ++instance) {
+                    renderer->DrawFromState(mode, static_cast<GLint>(cmd.First),
+                                            static_cast<GLsizei>(cmd.Count), 0, nullptr);
+                }
+            }
+        }
+
+        void MultiDrawElementsIndirect(GLenum mode, GLenum type, const void* indirect, GLsizei drawcount,
+                                       GLsizei stride) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr || drawcount <= 0) {
+                return;
+            }
+            const SizeT indexSize = MG_Util::GetGLTypeSize(type);
+            if (indexSize == 0) {
+                return;
+            }
+            const GLsizei realStride = stride == 0 ? static_cast<GLsizei>(sizeof(DrawElementsIndirectCommand)) : stride;
+            for (GLsizei i = 0; i < drawcount; ++i) {
+                const auto* bytes = ResolveIndirectCommandBytes(
+                    static_cast<const Uint8*>(indirect) + static_cast<SizeT>(i) * static_cast<SizeT>(realStride),
+                    sizeof(DrawElementsIndirectCommand), "MultiDrawElementsIndirect");
+                if (bytes == nullptr) {
+                    continue;
+                }
+                DrawElementsIndirectCommand cmd{};
+                std::memcpy(&cmd, bytes, sizeof(cmd));
+                if (cmd.Count == 0 || cmd.InstanceCount == 0) {
+                    continue;
+                }
+                const void* indices = reinterpret_cast<const void*>(static_cast<SizeT>(cmd.FirstIndex) * indexSize);
+                for (Uint32 instance = 0; instance < cmd.InstanceCount; ++instance) {
+                    renderer->DrawFromState(mode, 0, static_cast<GLsizei>(cmd.Count), type, indices,
+                                            static_cast<GLint>(cmd.BaseVertex));
+                }
+            }
+        }
+
+        void MultiDrawArraysIndirectCount(GLenum mode, const void* indirect, GLintptr drawcount,
+                                          GLsizei maxdrawcount, GLsizei stride) {
+            if (MG_State::pGLContext == nullptr) {
+                return;
+            }
+            auto paramBuffer = MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::Parameter).GetBoundObject();
+            if (!paramBuffer) {
+                return;
+            }
+            paramBuffer->SyncPersistentMappedRange();
+            const Uint8* paramData = paramBuffer->MappedData();
+            if (paramData == nullptr) {
+                return;
+            }
+            Uint32 actualDrawCount = 0;
+            std::memcpy(&actualDrawCount, paramData + static_cast<SizeT>(drawcount), sizeof(actualDrawCount));
+            actualDrawCount = std::min<Uint32>(actualDrawCount, static_cast<Uint32>(maxdrawcount));
+            MultiDrawArraysIndirect(mode, indirect, static_cast<GLsizei>(actualDrawCount), stride);
+        }
+
+        void MultiDrawElementsIndirectCount(GLenum mode, GLenum type, const void* indirect,
+                                            GLintptr drawcount, GLsizei maxdrawcount, GLsizei stride) {
+            if (MG_State::pGLContext == nullptr) {
+                return;
+            }
+            auto paramBuffer = MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::Parameter).GetBoundObject();
+            if (!paramBuffer) {
+                return;
+            }
+            paramBuffer->SyncPersistentMappedRange();
+            const Uint8* paramData = paramBuffer->MappedData();
+            if (paramData == nullptr) {
+                return;
+            }
+            Uint32 actualDrawCount = 0;
+            std::memcpy(&actualDrawCount, paramData + static_cast<SizeT>(drawcount), sizeof(actualDrawCount));
+            actualDrawCount = std::min<Uint32>(actualDrawCount, static_cast<Uint32>(maxdrawcount));
+            MultiDrawElementsIndirect(mode, type, indirect, static_cast<GLsizei>(actualDrawCount), stride);
+        }
+
+        void DrawArraysInstanced(GLenum mode, GLint first, GLsizei count, GLsizei instancecount) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || instancecount <= 0) {
+                return;
+            }
+            for (GLsizei i = 0; i < instancecount; ++i) {
+                renderer->DrawFromState(mode, first, count, 0, nullptr);
+            }
+        }
+
+        void DrawArraysInstancedBaseInstance(GLenum mode, GLint first, GLsizei count, GLsizei instancecount,
+                                             GLuint baseinstance) {
+            (void)baseinstance;
+            DrawArraysInstanced(mode, first, count, instancecount);
+        }
+
+        void DrawElementsInstanced(GLenum mode, GLsizei count, GLenum type, const void* indices,
+                                   GLsizei instancecount) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || instancecount <= 0) {
+                return;
+            }
+            for (GLsizei i = 0; i < instancecount; ++i) {
+                renderer->DrawFromState(mode, 0, count, type, indices);
+            }
+        }
+
+        void DrawElementsInstancedBaseVertex(GLenum mode, GLsizei count, GLenum type, const void* indices,
+                                             GLsizei instancecount, GLint basevertex) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || instancecount <= 0) {
+                return;
+            }
+            for (GLsizei i = 0; i < instancecount; ++i) {
+                renderer->DrawFromState(mode, 0, count, type, indices, basevertex);
+            }
+        }
+
+        void DrawElementsInstancedBaseInstance(GLenum mode, GLsizei count, GLenum type, const void* indices,
+                                               GLsizei instancecount, GLuint baseinstance) {
+            (void)baseinstance;
+            DrawElementsInstanced(mode, count, type, indices, instancecount);
+        }
+
+        void DrawElementsInstancedBaseVertexBaseInstance(GLenum mode, GLsizei count, GLenum type,
+                                                         const void* indices, GLsizei instancecount,
+                                                         GLint basevertex, GLuint baseinstance) {
+            (void)baseinstance;
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || instancecount <= 0) {
+                return;
+            }
+            for (GLsizei i = 0; i < instancecount; ++i) {
+                renderer->DrawFromState(mode, 0, count, type, indices, basevertex);
+            }
+        }
+
+        void ClearBufferfv(GLenum buffer, GLint drawbuffer, const GLfloat* value) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || value == nullptr) {
+                return;
+            }
+            if (buffer == GL_COLOR && drawbuffer == 0) {
+                renderer->Clear(value[0], value[1], value[2], value[3]);
+            } else if (buffer == GL_DEPTH && drawbuffer == 0) {
+                renderer->ClearDepth(value[0]);
+            }
+        }
+
+        void ClearBufferiv(GLenum buffer, GLint drawbuffer, const GLint* value) {
+            if (value == nullptr) {
+                return;
+            }
+            if (buffer == GL_STENCIL) {
+                auto* renderer = GetActiveRenderer();
+                if (renderer != nullptr) {
+                    renderer->ClearStencil(static_cast<Uint32>(value[0]));
+                }
+                return;
+            }
+            Float color[4] = {
+                static_cast<Float>(value[0]) / 255.0f,
+                static_cast<Float>(value[1]) / 255.0f,
+                static_cast<Float>(value[2]) / 255.0f,
+                static_cast<Float>(value[3]) / 255.0f,
+            };
+            ClearBufferfv(buffer, drawbuffer, color);
+        }
+
+        void ClearBufferuiv(GLenum buffer, GLint drawbuffer, const GLuint* value) {
+            if (value == nullptr) {
+                return;
+            }
+            if (buffer == GL_STENCIL) {
+                auto* renderer = GetActiveRenderer();
+                if (renderer != nullptr) {
+                    renderer->ClearStencil(value[0]);
+                }
+                return;
+            }
+            Float color[4] = {
+                static_cast<Float>(value[0]) / 255.0f,
+                static_cast<Float>(value[1]) / 255.0f,
+                static_cast<Float>(value[2]) / 255.0f,
+                static_cast<Float>(value[3]) / 255.0f,
+            };
+            ClearBufferfv(buffer, drawbuffer, color);
+        }
+
+        void ClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || buffer != GL_DEPTH_STENCIL) {
+                return;
+            }
+            (void)drawbuffer;
+            renderer->ClearDepth(depth);
+            renderer->ClearStencil(static_cast<Uint32>(stencil));
+        }
+
+        void ReadPixels(GLint x, GLint y, GLsizei width, GLsizei height, GLenum format, GLenum type, void* pixels) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || pixels == nullptr) {
+                return;
+            }
+            // The Diligent backend's offscreen targets are RGBA8; the frontend currently
+            // uses this entry for the common GL_RGBA/GL_UNSIGNED_BYTE readback path.
+            if (format != GL_RGBA || type != GL_UNSIGNED_BYTE) {
+                return;
+            }
+            renderer->ReadPixels(static_cast<Uint32>(x), static_cast<Uint32>(y),
+                                 static_cast<Uint32>(width), static_cast<Uint32>(height), pixels);
+        }
+
+        void BlitFramebuffer(GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
+                             GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+                             GLbitfield mask, GLenum filter) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->BlitFramebuffer(srcX0, srcY0, srcX1, srcY1, dstX0, dstY0, dstX1, dstY1,
+                                          mask, filter);
+            }
+        }
+
+        void BlitNamedFramebuffer(const SharedPtr<MG_State::GLState::FramebufferObject>& readFramebuffer,
+                                  const SharedPtr<MG_State::GLState::FramebufferObject>& drawFramebuffer,
+                                  GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
+                                  GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+                                  GLbitfield mask, GLenum filter) {
+            (void)srcX0;
+            (void)srcY0;
+            (void)srcX1;
+            (void)srcY1;
+            (void)dstX0;
+            (void)dstY0;
+            (void)dstX1;
+            (void)dstY1;
+            (void)filter;
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->BlitNamedFramebuffer(readFramebuffer, drawFramebuffer, mask);
+            }
+        }
+
+        void CopyTexImage2D(GLenum target, GLint level, GLenum internalformat, GLint x, GLint y,
+                            GLsizei width, GLsizei height, GLint border) {
+            (void)level;
+            (void)internalformat;
+            (void)x;
+            (void)y;
+            (void)width;
+            (void)height;
+            (void)border;
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr || target != GL_TEXTURE_2D) {
+                return;
+            }
+            auto& unit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
+            auto texture = unit.GetBindingSlot(TextureTarget::Texture2D).GetBoundObject();
+            if (texture) {
+                renderer->CopyReadFramebufferToTexture(*texture);
+            }
+        }
+
+        void CopyTexSubImage2D(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y,
+                               GLsizei width, GLsizei height) {
+            (void)level;
+            (void)xoffset;
+            (void)yoffset;
+            (void)x;
+            (void)y;
+            (void)width;
+            (void)height;
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr || target != GL_TEXTURE_2D) {
+                return;
+            }
+            auto& unit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
+            auto texture = unit.GetBindingSlot(TextureTarget::Texture2D).GetBoundObject();
+            if (texture) {
+                renderer->CopyReadFramebufferToTexture(*texture);
+            }
+        }
+
+        void GetTexImage(GLenum target, GLint level, GLenum format, GLenum type, GLvoid* pixels) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr || target != GL_TEXTURE_2D ||
+                format != GL_RGBA || type != GL_UNSIGNED_BYTE || pixels == nullptr) {
+                return;
+            }
+            auto& unit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
+            auto texture = unit.GetBindingSlot(TextureTarget::Texture2D).GetBoundObject();
+            if (texture) {
+                renderer->ReadTextureImage(*texture, static_cast<Uint32>(level), pixels);
+            }
+        }
+
+        void GetTextureImage(const SharedPtr<MG_State::GLState::ITextureObject>& texture,
+                             TextureUploadTarget uploadTarget, GLint level, GLenum format, GLenum type,
+                             GLsizei bufSize, GLvoid* pixels) {
+            (void)uploadTarget;
+            (void)bufSize;
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || !texture || format != GL_RGBA || type != GL_UNSIGNED_BYTE ||
+                pixels == nullptr) {
+                return;
+            }
+            renderer->ReadTextureImage(*texture, static_cast<Uint32>(level), pixels);
+        }
+
+        void CopyImageSubData(const SharedPtr<MG_State::GLState::ITextureObject>& srcTexture,
+                              GLenum srcTarget, GLint srcLevel, GLint srcX, GLint srcY, GLint srcZ,
+                              const SharedPtr<MG_State::GLState::ITextureObject>& dstTexture,
+                              GLenum dstTarget, GLint dstLevel, GLint dstX, GLint dstY, GLint dstZ,
+                              GLsizei srcWidth, GLsizei srcHeight, GLsizei srcDepth) {
+            (void)srcTarget;
+            (void)srcLevel;
+            (void)srcX;
+            (void)srcY;
+            (void)srcZ;
+            (void)dstTarget;
+            (void)dstLevel;
+            (void)dstX;
+            (void)dstY;
+            (void)dstZ;
+            (void)srcWidth;
+            (void)srcHeight;
+            (void)srcDepth;
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr && srcTexture && dstTexture) {
+                renderer->CopyTextureSubData(*srcTexture, *dstTexture);
+            }
+        }
+
+        void GenerateMipmap(GLenum target) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer == nullptr || MG_State::pGLContext == nullptr || target != GL_TEXTURE_2D) {
+                return;
+            }
+            auto& unit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
+            auto texture = unit.GetBindingSlot(TextureTarget::Texture2D).GetBoundObject();
+            if (texture) {
+                renderer->GenerateMipmap(*texture);
+            }
+        }
+
+        Bool IsTimerQuerySupported() {
+            return true;
+        }
+
+        BackendQueryHandle BeginTimeElapsedQuery() {
+            auto* query = new CpuTimerQuery;
+            query->Start = std::chrono::steady_clock::now();
+            query->Available = false;
+            return query;
+        }
+
+        void EndTimeElapsedQuery(BackendQueryHandle query) {
+            if (query == nullptr) {
+                return;
+            }
+            auto* cpuQuery = static_cast<CpuTimerQuery*>(query);
+            const auto now = std::chrono::steady_clock::now();
+            cpuQuery->TimestampNs = static_cast<Uint64>(
+                std::chrono::duration_cast<std::chrono::nanoseconds>(now - cpuQuery->Start).count());
+            cpuQuery->Available = true;
+        }
+
+        BackendQueryHandle QueryCounterTimestamp() {
+            auto* query = new CpuTimerQuery;
+            query->TimestampNs = static_cast<Uint64>(
+                std::chrono::duration_cast<std::chrono::nanoseconds>(
+                    std::chrono::steady_clock::now().time_since_epoch()).count());
+            query->Available = true;
+            return query;
+        }
+
+        Bool IsQueryResultAvailable(BackendQueryHandle query) {
+            return query != nullptr && static_cast<CpuTimerQuery*>(query)->Available;
+        }
+
+        Bool GetQueryResult64(BackendQueryHandle query, Bool wait, Uint64* outNanoseconds) {
+            if (query == nullptr || outNanoseconds == nullptr) {
+                return false;
+            }
+            auto* cpuQuery = static_cast<CpuTimerQuery*>(query);
+            if (!cpuQuery->Available && !wait) {
+                return false;
+            }
+            *outNanoseconds = cpuQuery->TimestampNs;
+            return true;
+        }
+
+        void DeleteBackendQuery(BackendQueryHandle query) {
+            delete static_cast<CpuTimerQuery*>(query);
+        }
+
+        BackendSyncHandle FenceSync() {
+            // CPU fallback fence: always signaled is a valid implementation for a
+            // backend without native sync primitives. The handle still round-trips
+            // through ClientWaitSync/DeleteSync so frontend state stays balanced.
+            return new int(0);
+        }
+
+        GLenum ClientWaitSync(BackendSyncHandle sync, GLbitfield flags, GLuint64 timeout) {
+            (void)sync;
+            (void)flags;
+            (void)timeout;
+            return GL_ALREADY_SIGNALED;
+        }
+
+        void WaitSync(BackendSyncHandle sync, GLbitfield flags, GLuint64 timeout) {
+            (void)sync;
+            (void)flags;
+            (void)timeout;
+        }
+
+        void DeleteSync(BackendSyncHandle sync) {
+            delete static_cast<int*>(sync);
+        }
+
+        Bool GetSyncStatus(BackendSyncHandle sync) {
+            (void)sync;
+            return true;
+        }
+
+        void SetSwapInterval(Int interval) {
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->SetSwapInterval(interval > 0 ? static_cast<Uint32>(interval) : 0);
+            }
+        }
+
+        void Present() {
+            auto* renderer = GetActiveRenderer();
+            if (renderer != nullptr) {
+                renderer->Present();
+            }
+        }
+    } // namespace
+
+    BackendObject_Diligent::BackendObject_Diligent()
+        : m_rendererInfo(BuildInitialRendererInfo()) {}
+
+    BackendObject_Diligent::~BackendObject_Diligent() {
+        m_pRenderer.reset();
+        m_pContext.Release();
+        m_pDevice.Release();
+        m_pFactoryVk = nullptr;
+    }
+
+    Bool BackendObject_Diligent::CreateDiligentDevice() {
+        if (m_pDevice && m_pContext) {
+            return true;
+        }
+
+        try {
+            if (m_pFactoryVk == nullptr) {
+                m_pFactoryVk = ::Diligent::GetEngineFactoryVk();
+                if (m_pFactoryVk == nullptr) {
+                    MGLOG_E("Diligent: failed to load Vulkan engine factory");
+                    return false;
+                }
+                m_pFactoryVk->SetBreakOnError(false);
+            }
+
+            ::Diligent::Uint32 numAdapters = 0;
+            m_pFactoryVk->EnumerateAdapters(::Diligent::Version{}, numAdapters, nullptr);
+            if (numAdapters == 0) {
+                MGLOG_W("Diligent: no Vulkan adapters available; skipping device creation");
+                return false;
+            }
+
+            ::Diligent::EngineVkCreateInfo engineCI;
+            ::Diligent::ImmediateContextCreateInfo ctxCI;
+            ctxCI.Name = "MobileGL Diligent Main Context";
+            ctxCI.QueueId = 0;
+            ctxCI.Priority = ::Diligent::QUEUE_PRIORITY_MEDIUM;
+            engineCI.NumImmediateContexts = 1;
+            engineCI.pImmediateContextInfo = &ctxCI;
+
+            ::Diligent::IRenderDevice* pDevice = nullptr;
+            ::Diligent::IDeviceContext* pContext = nullptr;
+            m_pFactoryVk->CreateDeviceAndContextsVk(engineCI, &pDevice, &pContext);
+            if (pDevice == nullptr || pContext == nullptr) {
+                MGLOG_E("Diligent: failed to create Vulkan device/context");
+                return false;
+            }
+
+            m_pDevice.Attach(pDevice);
+            m_pContext.Attach(pContext);
+            MGLOG_I("Diligent: Vulkan device created");
+            return true;
+        } catch (const std::exception& e) {
+            MGLOG_W("Diligent: Vulkan device creation failed: %s", e.what());
+            return false;
+        } catch (...) {
+            MGLOG_W("Diligent: Vulkan device creation failed");
+            return false;
+        }
+    }
+
+    void BackendObject_Diligent::Initialize() {
+        if (m_initialized) {
+            return;
+        }
+        if (!CreateDiligentDevice()) {
+            MGLOG_W("Diligent: backend initialization failed");
+            return;
+        }
+
+        m_pRenderer = std::make_unique<DiligentRenderer>(m_pDevice, m_pContext);
+        if (!m_pRenderer->Initialize(256, 256)) {
+            MGLOG_W("Diligent: renderer initialization failed");
+            m_pRenderer.reset();
+            return;
+        }
+
+        m_functions.GL.Clear = Clear;
+        m_functions.GL.DrawArrays = DrawArrays;
+        m_functions.GL.DrawElements = DrawElements;
+        m_functions.GL.DrawElementsBaseVertex = DrawElementsBaseVertex;
+        m_functions.GL.DrawRangeElements = DrawRangeElements;
+        m_functions.GL.DrawRangeElementsBaseVertex = DrawRangeElementsBaseVertex;
+        m_functions.GL.MultiDrawArrays = MultiDrawArrays;
+        m_functions.GL.MultiDrawElements = MultiDrawElements;
+        m_functions.GL.MultiDrawElementsBaseVertex = MultiDrawElementsBaseVertex;
+        m_functions.GL.DrawArraysInstanced = DrawArraysInstanced;
+        m_functions.GL.DrawArraysInstancedBaseInstance = DrawArraysInstancedBaseInstance;
+        m_functions.GL.DrawElementsInstanced = DrawElementsInstanced;
+        m_functions.GL.DrawElementsInstancedBaseVertex = DrawElementsInstancedBaseVertex;
+        m_functions.GL.DrawElementsInstancedBaseInstance = DrawElementsInstancedBaseInstance;
+        m_functions.GL.DrawElementsInstancedBaseVertexBaseInstance = DrawElementsInstancedBaseVertexBaseInstance;
+        m_functions.GL.DrawArraysIndirect = DrawArraysIndirect;
+        m_functions.GL.DrawElementsIndirect = DrawElementsIndirect;
+        m_functions.GL.MultiDrawArraysIndirect = MultiDrawArraysIndirect;
+        m_functions.GL.MultiDrawElementsIndirect = MultiDrawElementsIndirect;
+        m_functions.GL.MultiDrawArraysIndirectCount = MultiDrawArraysIndirectCount;
+        m_functions.GL.MultiDrawElementsIndirectCount = MultiDrawElementsIndirectCount;
+        m_functions.GL.ClearBufferfv = ClearBufferfv;
+        m_functions.GL.ClearBufferfi = ClearBufferfi;
+        m_functions.GL.ClearBufferiv = ClearBufferiv;
+        m_functions.GL.ClearBufferuiv = ClearBufferuiv;
+        m_functions.GL.BlitFramebuffer = BlitFramebuffer;
+        m_functions.GL.BlitNamedFramebuffer = BlitNamedFramebuffer;
+        m_functions.GL.CopyTexImage2D = CopyTexImage2D;
+        m_functions.GL.CopyTexSubImage2D = CopyTexSubImage2D;
+        m_functions.GL.CopyImageSubData = CopyImageSubData;
+        m_functions.GL.GenerateMipmap = GenerateMipmap;
+        m_functions.GL.GetTexImage = GetTexImage;
+        m_functions.GL.GetTextureImage = GetTextureImage;
+        m_functions.GL.ReadPixels = ReadPixels;
+        m_functions.GL.FenceSync = FenceSync;
+        m_functions.GL.ClientWaitSync = ClientWaitSync;
+        m_functions.GL.WaitSync = WaitSync;
+        m_functions.GL.DeleteSync = DeleteSync;
+        m_functions.GL.GetSyncStatus = GetSyncStatus;
+        m_functions.GL.IsTimerQuerySupported = IsTimerQuerySupported;
+        m_functions.GL.BeginTimeElapsedQuery = BeginTimeElapsedQuery;
+        m_functions.GL.EndTimeElapsedQuery = EndTimeElapsedQuery;
+        m_functions.GL.QueryCounterTimestamp = QueryCounterTimestamp;
+        m_functions.GL.IsQueryResultAvailable = IsQueryResultAvailable;
+        m_functions.GL.GetQueryResult64 = GetQueryResult64;
+        m_functions.GL.DeleteBackendQuery = DeleteBackendQuery;
+        m_functions.Present = Present;
+        m_functions.SetSwapInterval = SetSwapInterval;
+
+        m_initialized = true;
+    }
+
+    DiligentRenderer* BackendObject_Diligent::GetRenderer() {
+        return m_pRenderer.get();
+    }
+
+    Bool BackendObject_Diligent::InitCapabilities() {
+        // Skeleton: no format probing yet. The backend advertises GL 3.2 core
+        // capability, and the capability tables will be filled as resource
+        // creation paths are ported.
+        m_backendCapabilitiesInitialized = true;
+        return true;
+    }
+
+    Bool BackendObject_Diligent::InitWindowSurface() {
+        if (!m_windowHandle.Handle) {
+            MGLOG_E("BackendObject_Diligent::InitWindowSurface failed: native window handle is null");
+            return false;
+        }
+        if (m_pRenderer == nullptr || m_pFactoryVk == nullptr) {
+            MGLOG_E("BackendObject_Diligent::InitWindowSurface failed: renderer/factory is not ready");
+            return false;
+        }
+        return m_pRenderer->CreateSwapChain(m_pFactoryVk, m_windowHandle,
+                                            m_windowHandle.Width, m_windowHandle.Height);
+    }
+
+    Bool BackendObject_Diligent::InitPbufferSurface(EGLint width, EGLint height) {
+        // The Diligent backend keeps its offscreen target for pbuffer EGL surfaces.
+        // A future enhancement can resize/recreate the offscreen target to match the
+        // pbuffer dimensions.
+        (void)width;
+        (void)height;
+        return m_pRenderer != nullptr;
+    }
+
+    void BackendObject_Diligent::ReleaseEGLResources() {
+        if (m_pRenderer != nullptr) {
+            m_pRenderer->ReleaseSwapChain();
+        }
+        BackendObject::ReleaseEGLResources();
+    }
+
+    void BackendObject_Diligent::OnEGLSurfaceReleased(EGLSurface surface) {
+        (void)surface;
+        if (m_pRenderer != nullptr) {
+            m_pRenderer->ReleaseSwapChain();
+        }
+    }
+
+    Bool BackendObject_Diligent::CreateEGLWindowSurface(EGLSurface surface, const WindowHandle& handle) {
+        const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
+        if (!m_initialized) {
+            MGLOG_E("BackendObject_Diligent::CreateEGLWindowSurface failed: backend not initialized");
+            return false;
+        }
+        if (!handle.Handle || (handle.Backend != WindowBackend::Android && handle.Backend != WindowBackend::X11 &&
+                               handle.Backend != WindowBackend::MetalLayer && handle.Backend != WindowBackend::Win32)) {
+            MGLOG_E("BackendObject_Diligent::CreateEGLWindowSurface failed: unsupported native window backend");
+            return false;
+        }
+        return RegisterEGLWindowSurface(surface, handle);
+    }
+
+    Bool BackendObject_Diligent::CreateEGLPbufferSurface(EGLSurface surface, EGLint width, EGLint height) {
+        const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
+        if (!m_initialized) {
+            MGLOG_E("BackendObject_Diligent::CreateEGLPbufferSurface failed: backend not initialized");
+            return false;
+        }
+        return RegisterEGLPbufferSurface(surface, width, height);
+    }
+
+    Bool BackendObject_Diligent::ResizeEGLWindowSurface(EGLSurface surface, Uint32 width, Uint32 height) {
+        const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
+        if (!BackendObject::ResizeEGLWindowSurface(surface, width, height)) {
+            return false;
+        }
+        if (m_eglSurface == surface && m_pRenderer != nullptr) {
+            return m_pRenderer->ResizeSwapChain(width, height);
+        }
+        return true;
+    }
+
+    const RendererInfo& BackendObject_Diligent::GetRendererInfo() const {
+        return m_rendererInfo;
+    }
+
+    String BackendObject_Diligent::GetBackendAPIVersionString() const {
+        return "Diligent Vulkan 0.1 (GL 3.2 skeleton)";
+    }
+
+    const GlobalBackendFunctionsTable& BackendObject_Diligent::GetBackendFunctions() const {
+        return m_functions;
+    }
+
+    const DynamicBackendParameters& BackendObject_Diligent::GetDynamicParameters() const {
+        return m_dynamicParameters;
+    }
+
+    BackendType BackendObject_Diligent::GetBackendType() const {
+        return BackendType::DiligentVulkan;
+    }
+} // namespace MobileGL::MG_Backend::DiligentBackend

MobileGL/MG_Backend/Diligent/BackendObject_Diligent.h

@@ -0,0 +1,72 @@
+// MobileGL - MobileGL/MG_Backend/Diligent/BackendObject_Diligent.h
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+
+#pragma once
+
+#include <Includes.h>
+#include "../BackendObject.h"
+
+// X11 (pulled in by Includes.h through Vulkan-Headers) defines True/False as
+// macros, which collide with Diligent's Bool constants in BasicTypes.h.
+#if defined(True)
+#undef True
+#endif
+#if defined(False)
+#undef False
+#endif
+
+#include <RefCntAutoPtr.hpp>
+
+namespace Diligent {
+    struct IEngineFactoryVk;
+    struct IRenderDevice;
+    struct IDeviceContext;
+}
+
+namespace MobileGL::MG_Backend::DiligentBackend {
+    class DiligentRenderer;
+
+    // New Diligent/Vulkan backend, implemented from scratch on top of
+    // DiligentCore. The backend object owns the Diligent device/context and
+    // currently advertises OpenGL 3.2 core capability; the GL function table
+    // is intentionally empty until drawing/resource paths are ported.
+    class BackendObject_Diligent : public BackendObject {
+    public:
+        BackendObject_Diligent();
+        ~BackendObject_Diligent() override;
+
+        void Initialize() override;
+        Bool InitCapabilities() override;
+        Bool InitWindowSurface() override;
+        Bool InitPbufferSurface(EGLint width, EGLint height) override;
+        Bool CreateEGLWindowSurface(EGLSurface surface, const WindowHandle& handle) override;
+        Bool CreateEGLPbufferSurface(EGLSurface surface, EGLint width, EGLint height) override;
+        Bool ResizeEGLWindowSurface(EGLSurface surface, Uint32 width, Uint32 height) override;
+        void OnEGLSurfaceReleased(EGLSurface surface) override;
+
+        const RendererInfo& GetRendererInfo() const override;
+        String GetBackendAPIVersionString() const override;
+        const GlobalBackendFunctionsTable& GetBackendFunctions() const override;
+        const DynamicBackendParameters& GetDynamicParameters() const override;
+        BackendType GetBackendType() const override;
+        void ReleaseEGLResources() override;
+
+        DiligentRenderer* GetRenderer();
+
+    private:
+        Bool CreateDiligentDevice();
+
+        RendererInfo m_rendererInfo;
+        DynamicBackendParameters m_dynamicParameters;
+        GlobalBackendFunctionsTable m_functions{};
+        ::Diligent::IEngineFactoryVk* m_pFactoryVk = nullptr;
+        ::Diligent::RefCntAutoPtr<::Diligent::IRenderDevice> m_pDevice;
+        ::Diligent::RefCntAutoPtr<::Diligent::IDeviceContext> m_pContext;
+        std::unique_ptr<DiligentRenderer> m_pRenderer;
+        Bool m_initialized = false;
+    };
+} // namespace MobileGL::MG_Backend::DiligentBackend

MobileGL/MG_Backend/Diligent/DiligentVulkan.cpp

@@ -0,0 +1,8 @@
+// MobileGL - MobileGL/MG_Backend/Diligent/DiligentVulkan.cpp
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+
+#include "DiligentVulkan.h"

MobileGL/MG_Backend/Diligent/DiligentVulkan.h

@@ -0,0 +1,17 @@
+// MobileGL - MobileGL/MG_Backend/Diligent/DiligentVulkan.h
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+
+#pragma once
+
+#include <Includes.h>
+
+namespace MobileGL::MG_Backend::DiligentBackend {
+    // Backend identity string used by the backend object and local smoke tests.
+    inline String GetDiligentVulkanBackendName() {
+        return "DiligentVulkan";
+    }
+} // namespace MobileGL::MG_Backend::DiligentBackend

MobileGL/MG_Backend/Diligent/Renderer/DiligentRenderer.h

@@ -0,0 +1,153 @@
+// MobileGL - MobileGL/MG_Backend/Diligent/Renderer/DiligentRenderer.h
+// Copyright (c) 2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+
+#pragma once
+
+#include <Includes.h>
+
+// X11 (pulled in by Includes.h through Vulkan-Headers) defines True/False as
+// macros, which collide with Diligent's Bool constants in BasicTypes.h.
+#if defined(True)
+#undef True
+#endif
+#if defined(False)
+#undef False
+#endif
+
+#include <RefCntAutoPtr.hpp>
+
+namespace MobileGL::MG_Backend {
+    struct WindowHandle;
+}
+
+namespace Diligent {
+    struct IRenderDevice;
+    struct IDeviceContext;
+    struct ITexture;
+    struct ITextureView;
+    struct IPipelineState;
+    struct IBuffer;
+    struct ISampler;
+    struct IShaderResourceBinding;
+    struct ISwapChain;
+    struct IEngineFactoryVk;
+}
+
+namespace MobileGL::MG_State::GLState {
+    class ITextureObject;
+    class SamplerObject;
+    class ProgramObject;
+    class RenderbufferObject;
+    class FramebufferObject;
+}
+
+namespace MobileGL::MG_Backend::DiligentBackend {
+    // Minimal real Diligent renderer used to prove the GL 3.2 basic path:
+    // clear an offscreen color target, draw a hardcoded triangle, and read
+    // pixels back. This is the first concrete rendering layer on top of the
+    // Diligent device; it will be expanded into the full MobileGL backend.
+    class DiligentRenderer {
+    public:
+        DiligentRenderer(::Diligent::IRenderDevice* device, ::Diligent::IDeviceContext* context);
+        ~DiligentRenderer();
+
+        Bool Initialize(Uint32 width, Uint32 height);
+        void Clear(Float r, Float g, Float b, Float a);
+        void ClearDepth(Float depth);
+        void ClearStencil(Uint32 stencil);
+        void DrawTriangle();
+        void DrawVertices(const Float* vertices, Uint32 vertexCount);
+        // Creates a real Diligent swap chain for a native EGL window surface.
+        Bool CreateSwapChain(::Diligent::IEngineFactoryVk* factory, const WindowHandle& handle,
+                             Uint32 width, Uint32 height);
+        Bool ResizeSwapChain(Uint32 width, Uint32 height);
+        void SetSwapInterval(Uint32 interval);
+        // Creates a simple 2D RGBA8 texture from CPU data and makes it available
+        // to state PSOs under the shader variable name "g_Texture".
+        Bool CreateTestTexture(const void* data, Uint32 width, Uint32 height);
+        // Draws using the live MG_State GL context: current program, VAO and
+        // bound buffers. This is the front-end emulation entry point.
+        void DrawFromState(GLenum mode, GLint first, GLsizei count, GLenum type, const void* indices,
+                           GLint baseVertex = 0);
+        void ReadPixels(Uint32 x, Uint32 y, Uint32 width, Uint32 height, void* pixels);
+        void BlitFramebuffer(GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
+                             GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+                             GLbitfield mask, GLenum filter);
+        void BlitNamedFramebuffer(const SharedPtr<MG_State::GLState::FramebufferObject>& readFbo,
+                                  const SharedPtr<MG_State::GLState::FramebufferObject>& drawFbo,
+                                  GLbitfield mask);
+        void CopyReadFramebufferToTexture(MG_State::GLState::ITextureObject& dst);
+        void CopyTextureSubData(MG_State::GLState::ITextureObject& src, MG_State::GLState::ITextureObject& dst);
+        void GenerateMipmap(MG_State::GLState::ITextureObject& texture);
+        Bool ReadTextureImage(MG_State::GLState::ITextureObject& texture, Uint32 level, void* pixels);
+        void ReleaseSwapChain();
+        void Present();
+
+        ::Diligent::IRenderDevice* GetDevice() const { return m_pDevice; }
+        ::Diligent::IDeviceContext* GetContext() const { return m_pContext; }
+
+    private:
+        struct TextureResource {
+            ::Diligent::RefCntAutoPtr<::Diligent::ITexture> Texture;
+            ::Diligent::RefCntAutoPtr<::Diligent::ITextureView> SRV;
+            ::Diligent::RefCntAutoPtr<::Diligent::ITextureView> RTV;
+            ::Diligent::RefCntAutoPtr<::Diligent::ITextureView> DSV;
+            Uint64 ContentVersion = 0;
+            Uint16 ParamsVersion = 0;
+            Bool IsDepth = false;
+        };
+
+        struct SamplerResource {
+            ::Diligent::RefCntAutoPtr<::Diligent::ISampler> Sampler;
+            Uint16 Version = 0;
+        };
+
+        Bool CreateOffscreenTargets();
+        Bool CreatePipeline();
+        Bool CreateVertexBuffer();
+        Bool CreatePipelineFromState(GLenum mode);
+        Bool UploadVertexDataFromState(GLenum mode, GLint first, GLsizei count, GLenum type, const void* indices,
+                                        GLint baseVertex = 0);
+        ::Diligent::ITextureView* SyncTexture(MG_State::GLState::ITextureObject& texture);
+        ::Diligent::ITextureView* SyncTextureForAttachment(MG_State::GLState::ITextureObject& texture, Bool depth);
+        ::Diligent::ITextureView* SyncRenderbuffer(MG_State::GLState::RenderbufferObject& renderbuffer);
+        ::Diligent::ISampler* SyncSampler(const MG_State::GLState::SamplerObject& sampler);
+        Bool BindShaderResourcesFromState(const MG_State::GLState::ProgramObject& program);
+        Bool UploadUBOFromState(const MG_State::GLState::ProgramObject& program);
+        Bool ResolveCurrentRenderTargets(Vector<::Diligent::ITextureView*>& rtvs,
+                                         ::Diligent::ITextureView*& dsv);
+
+        ::Diligent::IRenderDevice* m_pDevice = nullptr;
+        ::Diligent::IDeviceContext* m_pContext = nullptr;
+        ::Diligent::RefCntAutoPtr<::Diligent::ITexture> m_pColorTarget;
+        ::Diligent::RefCntAutoPtr<::Diligent::ITextureView> m_pColorRTV;
+        ::Diligent::RefCntAutoPtr<::Diligent::ITexture> m_pDepthTarget;
+        ::Diligent::RefCntAutoPtr<::Diligent::ITextureView> m_pDepthDSV;
+        ::Diligent::RefCntAutoPtr<::Diligent::ISwapChain> m_pSwapChain;
+        ::Diligent::RefCntAutoPtr<::Diligent::ITexture> m_pTestTexture;
+        ::Diligent::RefCntAutoPtr<::Diligent::ITextureView> m_pTestSRV;
+        ::Diligent::RefCntAutoPtr<::Diligent::ISampler> m_pTestSampler;
+        ::Diligent::RefCntAutoPtr<::Diligent::IShaderResourceBinding> m_pStateSRB;
+        ::Diligent::RefCntAutoPtr<::Diligent::IPipelineState> m_pPSO;
+        ::Diligent::RefCntAutoPtr<::Diligent::IBuffer> m_pVertexBuffer;
+        ::Diligent::RefCntAutoPtr<::Diligent::IBuffer> m_pUBO;
+        Uint32 m_uboSize = 0;
+        Uint32 m_uboContentVersion = 0;
+        Uint64 m_uboProgramLifetimeId = 0;
+        UnorderedMap<Uint64, TextureResource> m_textureCache;
+        UnorderedMap<Uint64, SamplerResource> m_samplerCache;
+        UnorderedMap<Uint32, TextureResource> m_renderbufferCache;
+        UnorderedMap<Uint64, ::Diligent::RefCntAutoPtr<::Diligent::IBuffer>> m_namedUboCache;
+        Uint32 m_width = 256;
+        Uint32 m_height = 256;
+        Uint32 m_swapInterval = 0;
+        Uint32 m_lastDrawVertexCount = 0;
+        Uint64 m_lastPSOKey = 0;
+        Bool m_hasCachedPSO = false;
+        Bool m_initialized = false;
+    };
+} // namespace MobileGL::MG_Backend::DiligentBackend

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.cpp

@@ -10,6 +10,7 @@
 #include "MG_Backend/BackendObject.h"
 #include "DirectVulkan.h"
 #include "MG_State/GLState/FramebufferState/FramebufferObject.h"
+#include "MG_State/GLState/Core.h"
 #include "MG_State/GLState/TextureState/TextureState.h"
 #include "MG_Util/Classifiers/TextureEnumClassifier.h"
 #include "MG_Util/Converters/MGToGL/TextureEnumConverter.h"
@@ -383,6 +384,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         }
         UpdateDynamicBackendParameters();
         UpdateAdvertisedExtensions();
+        if (MG_State::pGLContext) {
+            MG_State::pGLContext->InvalidateCompileEnv();
+        }
         PopulateFormatCapabilities(physicalDevice.handle, vkGetPhysicalDeviceFormatProperties, m_vulkanCaps,
                                    MutableFormatCapabilities());
         PrintFormatCapabilities(GetFormatCapabilities());
@@ -687,6 +691,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_vulkanCaps = capabilities;
         UpdateDynamicBackendParameters();
         UpdateAdvertisedExtensions();
+        if (MG_State::pGLContext) {
+            MG_State::pGLContext->InvalidateCompileEnv();
+        }
         MutableFormatCapabilities().Clear();
     }
 

MobileGL/MG_Backend/Init.cpp

@@ -10,6 +10,9 @@
 #include <Config.h>
 #include <MG_Util/BackendLoaders/OpenGL/Loader.h>
 #include <MG_Util/Converters/MGToStr/GLExtensionConverter.h>
+#if defined(MOBILEGL_ENABLE_DILIGENT)
+#include <MG_Backend/Diligent/BackendObject_Diligent.h>
+#endif
 
 namespace MobileGL::MG_Backend {
     void LogBackendInfo() {
@@ -55,6 +58,11 @@ namespace MobileGL::MG_Backend {
         case BackendType::DirectVulkan:
             pActiveBackendObject = MakeUnique<DirectVulkan::BackendObject_DirectVulkan>();
             break;
+#if defined(MOBILEGL_ENABLE_DILIGENT)
+        case BackendType::DiligentVulkan:
+            pActiveBackendObject = MakeUnique<DiligentBackend::BackendObject_Diligent>();
+            break;
+#endif
         case BackendType::Unknown:
         default:
             MGLOG_W("Unknown backend type, defaulting to unknown backend");

MobileGL/MG_Impl/GLXImpl/GLXImpl.cpp

@@ -175,14 +175,14 @@ namespace MobileGL::MG_Impl::GLXImpl {
 
         struct ContextObject {
             Display* XDisplay = nullptr;
-            EGLDisplay Display = EGL_NO_DISPLAY;
+            EGLDisplay Dpy = EGL_NO_DISPLAY;
             EGLConfig Config = nullptr;
             EGLContext Context = EGL_NO_CONTEXT;
             const FBConfigInfo* FBConfig = nullptr;
         };
 
         struct DrawableSurface {
-            EGLDisplay Display = EGL_NO_DISPLAY;
+            EGLDisplay Dpy = EGL_NO_DISPLAY;
             EGLSurface Surface = EGL_NO_SURFACE;
             Uint32 Width = 0;
             Uint32 Height = 0;
@@ -294,7 +294,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
             if (width == surface.Width && height == surface.Height) {
                 return;
             }
-            if (EGLImpl::ResizePlatformWindowSurface(surface.Display, surface.Surface,
+            if (EGLImpl::ResizePlatformWindowSurface(surface.Dpy, surface.Surface,
                                                      static_cast<EGLint>(width),
                                                      static_cast<EGLint>(height))) {
                 surface.Width = width;
@@ -324,7 +324,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
                 EGL_NONE,
             };
             EGLSurface surface = EGLImpl::CreatePlatformWindowSurface(
-                context.Display, context.Config, reinterpret_cast<void*>(drawable), attribs);
+                context.Dpy, context.Config, reinterpret_cast<void*>(drawable), attribs);
             if (surface == EGL_NO_SURFACE) {
                 MGLOG_E_ONCE("glx: failed to create window surface for drawable 0x%lx (%ux%u)", drawable,
                         width, height);
@@ -332,7 +332,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
             }
 
             DrawableSurface record;
-            record.Display = context.Display;
+            record.Dpy = context.Dpy;
             record.Surface = surface;
             record.Width = width;
             record.Height = height;
@@ -388,7 +388,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
 
             ContextObject object;
             object.XDisplay = dpy;
-            object.Display = display;
+            object.Dpy = display;
             object.Config = config;
             object.Context = eglContext;
             object.FBConfig = fbconfig;
@@ -895,7 +895,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
             return;
         }
         if (object->Context != EGL_NO_CONTEXT) {
-            EGLImpl::DestroyContext(object->Display, object->Context);
+            EGLImpl::DestroyContext(object->Dpy, object->Context);
         }
         Contexts().erase(context);
     }
@@ -929,7 +929,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
             return 0;
         }
 
-        if (!EGLImpl::MakeCurrent(object->Display, surface->Surface, surface->Surface,
+        if (!EGLImpl::MakeCurrent(object->Dpy, surface->Surface, surface->Surface,
                                   object->Context)) {
             MGLOG_E_ONCE("glx: eglMakeCurrent failed (drawable=0x%lx, ctx=%p)", drawable, context);
             return 0;
@@ -962,7 +962,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
             return;
         }
         SyncSurfaceSize(dpy, drawable, it->second);
-        EGLImpl::SwapBuffers(it->second.Display, it->second.Surface);
+        EGLImpl::SwapBuffers(it->second.Dpy, it->second.Surface);
     }
 
     GLXDrawableHandle CreateWindow(Display*, GLXFBConfigHandle config, GLXDrawableHandle window,
@@ -988,7 +988,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
         if (it == surfaces.end()) {
             return;
         }
-        EGLImpl::DestroySurface(it->second.Display, it->second.Surface);
+        EGLImpl::DestroySurface(it->second.Dpy, it->second.Surface);
         surfaces.erase(it);
     }
 

MobileGL/MG_State/GLState/Core.cpp

@@ -39,6 +39,11 @@ namespace MobileGL::MG_State {
             return m_compileEnv;
         }
 
+        void GLContext::InvalidateCompileEnv() {
+            m_compileEnv.reset();
+            m_compileEnvBackend = nullptr;
+        }
+
         // Error
         void GLContext::RecordError(ErrorCode code, UniquePtr<ErrorInfo> info) {
             // Invariant I1, mechanically enforced: the GL error state is GL-thread-owned.

MobileGL/MG_State/GLState/Core.h

@@ -413,9 +413,12 @@ namespace MobileGL {
                 // cannot be captured in MG_State::Init() - that runs BEFORE MG_Backend::Init(),
                 // so there is no backend to query yet. Re-captured whenever the active backend
                 // object changes, which also rolls the fingerprint and therefore invalidates
-                // every P0b preprocess memo keyed against the old one.
+                // every P0b preprocess memo keyed against the old one. A backend whose dynamic
+                // capabilities become available without changing object identity must call
+                // InvalidateCompileEnv() after publishing them.
                 // GL thread only.
                 const SharedPtr<const MG_Util::ShaderTranspiler::CompileEnv>& GetCompileEnv();
+                void InvalidateCompileEnv();
 
             private:
                 // State Components

MobileGL/MG_Test/Backend/Diligent/CMakeLists.txt

@@ -0,0 +1,22 @@
+cmake_minimum_required(VERSION 3.14)
+
+message(STATUS "Generating build files for MobileGL Diligent Backend Test...")
+
+add_executable(
+        DiligentVulkanSanityTest
+        SanityTest.cpp
+)
+
+target_include_directories(DiligentVulkanSanityTest PRIVATE
+        ${MGL_ROOT}/include
+        ${MGL_ROOT}/MobileGL
+)
+
+target_link_libraries(
+        DiligentVulkanSanityTest PRIVATE
+        GTest::gtest_main
+        ${LINK_LIBRARIES}
+)
+
+include(GoogleTest)
+gtest_discover_tests(DiligentVulkanSanityTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS integration)

MobileGL/MG_Test/CMakeLists.txt

@@ -84,3 +84,6 @@ add_subdirectory(Backend/DirectGLES)
 if (ENABLE_INTEGRATION_TESTS)
     add_subdirectory(Backend/DirectVulkan)
 endif()
+if (MOBILEGL_ENABLE_DILIGENT)
+    add_subdirectory(Backend/Diligent)
+endif()

MobileGL/MG_Test/Program/ProgramUtilTest.cpp

@@ -2073,153 +2073,6 @@ void main() {
     EXPECT_NE(source.find("layout(std140) uniform Blk"), String::npos);
 }
 
-namespace {
-    String MakeLinearSubgroupPrefixScanShader() {
-        return R"(#version 460 core
-#extension GL_KHR_shader_subgroup_arithmetic : enable
-layout(local_size_x = 1024) in;
-shared float prefixSumCache[64];
-
-layout(std430, binding = 0) writeonly buffer OutputBuffer {
-    float outputValues[];
-};
-
-void main() {
-    float importance = 1.0f;
-    float prefixSum = subgroupInclusiveAdd(importance);
-    if (gl_SubgroupInvocationID == gl_SubgroupSize - 1u) prefixSumCache[gl_SubgroupID] = prefixSum;
-    barrier();
-    uint loopLength = uint(findMSB(gl_NumSubgroups));
-    loopLength += uint(gl_NumSubgroups - (1u << (loopLength - 1u)) > 0u);
-    for (uint i = 0; i < loopLength; i++) {
-        if ((gl_SubgroupID & (1u << i)) > 0u) {
-            prefixSum += prefixSumCache[(gl_SubgroupID >> i << i) - 1u];
-            if (gl_SubgroupInvocationID == gl_SubgroupSize - 1u) prefixSumCache[gl_SubgroupID] = prefixSum;
-        }
-        barrier();
-    }
-    if (gl_LocalInvocationID.x == uint(1024 - 1)) prefixSumCache[0] = prefixSum;
-    barrier();
-    float sum = prefixSumCache[0];
-    float warp = (prefixSum - importance) / sum - float(gl_LocalInvocationID.x + 1u) / float(1024);
-    outputValues[gl_GlobalInvocationID.x] = warp;
-}
-)";
-    }
-} // namespace
-
-TEST_F(ProgramUtilTest, RewriteLinearSubgroupPrefixScanUsesSharedMemoryAndProducesValidSpirv) {
-    using namespace MG_Util::ShaderTranspiler;
-
-    String source = MakeLinearSubgroupPrefixScanShader();
-    ASSERT_TRUE(RewriteLinearSubgroupPrefixScanForVulkan(ShaderStage::Compute, 64, source));
-
-    EXPECT_NE(source.find("shared float prefixSumCache[1024]"), String::npos) << source;
-    EXPECT_NE(source.find("mglVirtualSubgroupInvocation"), String::npos) << source;
-    EXPECT_NE(source.find("for (uint mglPrefixLane"), String::npos) << source;
-    EXPECT_EQ(source.find("subgroupInclusiveAdd"), String::npos) << source;
-    EXPECT_EQ(source.find("gl_Subgroup"), String::npos) << source;
-
-    const String onceRewritten = source;
-    EXPECT_FALSE(RewriteLinearSubgroupPrefixScanForVulkan(ShaderStage::Compute, 64, source));
-    EXPECT_EQ(source, onceRewritten);
-
-    ShaderAttrib shaderAttrib{.shaderType = GL_COMPUTE_SHADER, .sourceStr = source};
-    auto shaderResult = ShaderCompiler::CompileShader(shaderAttrib);
-    ASSERT_TRUE(shaderResult) << shaderResult.error().log << "\nsource:\n" << source;
-
-    ProgramAttrib programAttrib{.shaders = {shaderResult.value()}};
-    auto programResult = ShaderCompiler::LinkProgram(programAttrib);
-    ASSERT_TRUE(programResult) << programResult.error().log;
-
-    ProgramBinaryAttrib binaryAttrib{.shaderTypes = {GL_COMPUTE_SHADER}, .program = *programResult.value()};
-    auto binaryResult = ShaderCompiler::GetSpirvBinaryFromProgram(binaryAttrib);
-    ASSERT_TRUE(binaryResult) << binaryResult.error().log;
-    ASSERT_EQ(binaryResult->size(), 1u);
-
-    String validationDiagnostics;
-    spvtools::SpirvTools tools(SPV_ENV_VULKAN_1_1);
-    tools.SetMessageConsumer([&](spv_message_level_t, const char*, const spv_position_t&, const char* message) {
-        validationDiagnostics += message;
-        validationDiagnostics += '\n';
-    });
-    EXPECT_TRUE(tools.Validate(binaryResult->front())) << validationDiagnostics;
-
-    String spirvText;
-    ASSERT_TRUE(tools.Disassemble(binaryResult->front(), &spirvText));
-    EXPECT_EQ(spirvText.find("OpGroupNonUniform"), String::npos) << spirvText;
-}
-
-TEST_F(ProgramUtilTest, RewriteLinearSubgroupPrefixScanRejectsOtherStagesAndSubgroupWidths) {
-    using namespace MG_Util::ShaderTranspiler;
-
-    const String original = MakeLinearSubgroupPrefixScanShader();
-    for (const auto& [stage, subgroupSize] :
-         {std::pair{ShaderStage::Compute, Uint32{32}}, std::pair{ShaderStage::Fragment, Uint32{64}},
-          std::pair{ShaderStage::Compute, Uint32{96}}}) {
-        String source = original;
-        EXPECT_FALSE(RewriteLinearSubgroupPrefixScanForVulkan(stage, subgroupSize, source));
-        EXPECT_EQ(source, original);
-    }
-}
-
-TEST_F(ProgramUtilTest, RewriteLinearSubgroupPrefixScanRejectsPartialOrUnsafeTemplateMatches) {
-    using namespace MG_Util::ShaderTranspiler;
-
-    const auto expectUnchanged = [](String source) {
-        const String original = source;
-        EXPECT_FALSE(RewriteLinearSubgroupPrefixScanForVulkan(ShaderStage::Compute, 64, source));
-        EXPECT_EQ(source, original);
-    };
-
-    String wrongLocalSize = MakeLinearSubgroupPrefixScanShader();
-    wrongLocalSize.replace(wrongLocalSize.find("local_size_x = 1024"), std::strlen("local_size_x = 1024"),
-                           "local_size_x = 512");
-    expectUnchanged(std::move(wrongLocalSize));
-
-    String cacheHasAnotherUse = MakeLinearSubgroupPrefixScanShader();
-    cacheHasAnotherUse.insert(cacheHasAnotherUse.find("float importance"), "prefixSumCache[0] = 0.0f;\n    ");
-    expectUnchanged(std::move(cacheHasAnotherUse));
-
-    String extraSubgroupBuiltin = MakeLinearSubgroupPrefixScanShader();
-    extraSubgroupBuiltin.insert(extraSubgroupBuiltin.find("float importance"),
-                                "uvec4 extraMask = gl_SubgroupEqMask;\n    ");
-    expectUnchanged(std::move(extraSubgroupBuiltin));
-
-    String alteredBarrier = MakeLinearSubgroupPrefixScanShader();
-    alteredBarrier.replace(alteredBarrier.find("barrier();"), std::strlen("barrier();"), "memoryBarrierShared();");
-    expectUnchanged(std::move(alteredBarrier));
-
-    String nestedScan = MakeLinearSubgroupPrefixScanShader();
-    nestedScan.insert(nestedScan.find("float prefixSum ="), "if (importance > 0.0f) {\n    ");
-    const SizeT consumerEnd = nestedScan.find(';', nestedScan.find("float warp ="));
-    ASSERT_NE(consumerEnd, String::npos);
-    nestedScan.insert(consumerEnd + 1, "\n    }");
-    expectUnchanged(std::move(nestedScan));
-
-    // ARB/NV spellings of lane-width-sensitive builtins must block the rewrite exactly
-    // like their KHR counterparts.
-    String arbSubgroupBuiltin = MakeLinearSubgroupPrefixScanShader();
-    arbSubgroupBuiltin.insert(arbSubgroupBuiltin.find("float importance"),
-                              "uint arbLane = gl_SubGroupInvocationARB;\n    ");
-    expectUnchanged(std::move(arbSubgroupBuiltin));
-
-    String arbBallotCall = MakeLinearSubgroupPrefixScanShader();
-    arbBallotCall.insert(arbBallotCall.find("float importance"),
-                         "uint64_t arbMask = ballotARB(true);\n    ");
-    expectUnchanged(std::move(arbBallotCall));
-
-    String nvWarpBuiltin = MakeLinearSubgroupPrefixScanShader();
-    nvWarpBuiltin.insert(nvWarpBuiltin.find("float importance"),
-                         "uint warpSize = gl_WarpSizeNV;\n    ");
-    expectUnchanged(std::move(nvWarpBuiltin));
-
-    String nvShuffleCall = MakeLinearSubgroupPrefixScanShader();
-    nvShuffleCall.insert(nvShuffleCall.find("float importance"),
-                         "float other = shuffleNV(1.0f, 0u, 32u);\n    ");
-    expectUnchanged(std::move(nvShuffleCall));
-}
-
 // The LEXICAL half must fire at the source level (before the parse) for the
 // preempt-list names - the end-to-end ESSL tests cannot tell which half did the
 // rename, and for these names the parse would fail without the source rewrite.
@@ -2448,9 +2301,6 @@ TEST_F(ProgramUtilTest, CompileEnvFingerprintTracksEveryInput) {
     otherExtensions.advertisedExtensions.push_back(MobileGL::E_GL_ARB_gpu_shader_int64);
     EXPECT_NE(ComputeCompileEnvFingerprint(otherExtensions), baseline);
 
-    CompileEnv otherQuirk = base;
-    otherQuirk.subgroupPrefixScanQuirk = MobileGL::MG_Config::QuirkOverride::ForceOn;
-    EXPECT_NE(ComputeCompileEnvFingerprint(otherQuirk), baseline);
 }
 
 // The no-backend fallback must stay exactly what the pipeline used to do inline:

MobileGL/MG_Test/SanityTest.cpp

@@ -31,6 +31,7 @@
 #include <MG_Backend/DirectVulkan/Renderer/VulkanRenderer.h>
 #include <MG_Util/Math/HalfFloat.h>
 #include <MG_Util/ShaderTranspiler/ShaderCompiler.h>
+#include <MG_Util/ShaderTranspiler/CompileEnv.h>
 #include <MG_Util/ShaderTranspiler/ShaderSourceProcessor.h>
 #include <MG_Util/Debug/Log.h>
 #include <MG_Util/Types.h>
@@ -710,6 +711,37 @@ TEST(DirectVulkanSanity, AdvertisesSubgroupOnlyWhenVulkanReportsUsableSupport) {
     EXPECT_TRUE(backend.GetDynamicParameters().SubgroupQuadOperationsInAllStages);
 }
 
+TEST(DirectVulkanSanity, CapabilityRefreshInvalidatesTheCachedCompileEnvironment) {
+    using namespace MobileGL;
+
+    auto previousContext = Move(MG_State::pGLContext);
+    auto previousBackend = Move(MG_Backend::pActiveBackendObject);
+    MG_State::pGLContext = MakeUnique<MG_State::GLState::GLContext>();
+
+    auto backend = MakeUnique<MG_Backend::DirectVulkan::BackendObject_DirectVulkan>();
+    auto* backendPtr = backend.get();
+    MG_Backend::pActiveBackendObject = Move(backend);
+
+    const auto before = MG_State::pGLContext->GetCompileEnv();
+    EXPECT_EQ(before->params.SubgroupSize, 0u);
+
+    MG_External::VulkanCapabilities caps;
+    caps.SupportsShaderSubgroup = true;
+    caps.SubgroupSize = 8;
+    caps.SubgroupSupportedStages = VK_SHADER_STAGE_COMPUTE_BIT;
+    caps.SubgroupSupportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT | VK_SUBGROUP_FEATURE_ARITHMETIC_BIT;
+    backendPtr->ApplyVulkanCapabilitiesForTesting(caps);
+
+    const auto after = MG_State::pGLContext->GetCompileEnv();
+    EXPECT_NE(after.get(), before.get());
+    EXPECT_NE(after->fingerprint, before->fingerprint);
+    EXPECT_EQ(after->backend, BackendType::DirectVulkan);
+    EXPECT_EQ(after->params.SubgroupSize, 8u);
+
+    MG_Backend::pActiveBackendObject = Move(previousBackend);
+    MG_State::pGLContext = Move(previousContext);
+}
+
 TEST(DirectVulkanSanity, KeepsOptionalGpuShaderInt64BranchForVoxyQuadDecode) {
     using namespace MobileGL;
 

MobileGL/MG_Util/Debug/Log.cpp

@@ -33,13 +33,13 @@ namespace MobileGL {
 
         std::string GetThreadName() {
             char buffer[64] = {0};
-#if defined(_WIN32) && !defined(__MINGW32__)
+#if defined(_WIN32)
             PWSTR desc = nullptr;
             if (SUCCEEDED(GetThreadDescription(GetCurrentThread(), &desc))) {
                 WideCharToMultiByte(CP_UTF8, 0, desc, -1, buffer, sizeof(buffer), nullptr, nullptr);
                 LocalFree(desc);
             }
-#elif defined(__ANDROID__) || defined(__linux__) || defined(__APPLE__) || defined(__MINGW32__)
+#elif defined(__ANDROID__) || defined(__linux__) || defined(__APPLE__)
             pthread_getname_np(pthread_self(), buffer, sizeof(buffer));
 #endif
             return buffer[0] ? buffer : "UnknownThread";

MobileGL/MG_Util/ShaderTranspiler/CompileEnv.cpp

@@ -38,7 +38,6 @@ namespace MobileGL::MG_Util::ShaderTranspiler {
             HashBytes(state, env.advertisedExtensions.data(),
                       env.advertisedExtensions.size() * sizeof(GLExtension));
         }
-        HashValue(state, env.subgroupPrefixScanQuirk);
         return state;
     }
 
@@ -73,8 +72,6 @@ namespace MobileGL::MG_Util::ShaderTranspiler {
                                      kFrontendMaxComputeWorkGroupInvocations)
                           : kFrontendMaxComputeWorkGroupInvocations;
 
-        env->subgroupPrefixScanQuirk = MG_Config::Features.SubgroupPrefixScanQuirk;
-
         env->fingerprint = ComputeCompileEnvFingerprint(*env);
         return env;
     }
@@ -84,7 +81,6 @@ namespace MobileGL::MG_Util::ShaderTranspiler {
         // computed, and this must not run before MG_Config is loaded.
         static const SharedPtr<const CompileEnv> kDefault = [] {
             auto env = MakeShared<CompileEnv>();
-            env->subgroupPrefixScanQuirk = MG_Config::Features.SubgroupPrefixScanQuirk;
             env->fingerprint = ComputeCompileEnvFingerprint(*env);
             return SharedPtr<const CompileEnv>(Move(env));
         }();

MobileGL/MG_Util/ShaderTranspiler/CompileEnv.h

@@ -12,9 +12,9 @@
 #include <MG_Backend/BackendObject.h>
 
 namespace MobileGL::MG_Util::ShaderTranspiler {
-    // Everything the shader compile/link pipeline reads from OUTSIDE its own (stage, source)
-    // inputs: backend identity, backend limits, the advertised extension list, and the one
-    // config quirk the source rewriter branches on.
+    // everything outside (stage, source) this reads - advertised extensions and backend limits -
+    // so the transformation is a pure function of its three arguments and can run on a worker
+    // thread.
     //
     // Why it exists (P1): every one of those reads is a reach-back into
     // MG_Backend::pActiveBackendObject / gBackendFunctionsTable, and one of them
@@ -46,9 +46,6 @@ namespace MobileGL::MG_Util::ShaderTranspiler {
         MG_Backend::DynamicBackendParameters params{};   // by value, never by reference
         Vector<GLExtension> advertisedExtensions;
 
-        // --- config the source rewriter branches on ---
-        MG_Config::QuirkOverride subgroupPrefixScanQuirk = MG_Config::QuirkOverride::Auto;
-
         Uint64 fingerprint = 0;   // set by CaptureCompileEnv()
 
         Bool HasBackend() const { return backend != BackendType::Unknown; }

MobileGL/MG_Util/ShaderTranspiler/ShaderSourceProcessor.cpp

@@ -23,6 +23,7 @@
 namespace {
     using MobileGL::SizeT;
     using MobileGL::String;
+    using MobileGL::Uint32;
     using MobileGL::Vector;
 
     bool IsIdentifierChar(char ch) {
@@ -181,331 +182,6 @@ namespace {
         return std::all_of(token.text.begin() + 1, token.text.end(), IsIdentifierChar);
     }
 
-    class TokenCursor {
-    public:
-        TokenCursor(const Vector<CodeToken>& tokens, SizeT position) : m_tokens(tokens), m_position(position) {}
-
-        bool Consume(const char* expected) {
-            if (m_position >= m_tokens.size() || m_tokens[m_position].text != expected) {
-                return false;
-            }
-            ++m_position;
-            return true;
-        }
-
-        bool ConsumeAnyIdentifier(String& identifier) {
-            if (m_position >= m_tokens.size() || !IsIdentifierToken(m_tokens[m_position])) {
-                return false;
-            }
-            identifier = m_tokens[m_position++].text;
-            return true;
-        }
-
-        bool ConsumeAnyIdentifier() {
-            if (m_position >= m_tokens.size() || !IsIdentifierToken(m_tokens[m_position])) {
-                return false;
-            }
-            ++m_position;
-            return true;
-        }
-
-        bool ConsumeIdentifier(const String& expected) {
-            if (m_position >= m_tokens.size() || !IsIdentifierToken(m_tokens[m_position]) ||
-                m_tokens[m_position].text != expected) {
-                return false;
-            }
-            ++m_position;
-            return true;
-        }
-
-        SizeT Position() const { return m_position; }
-
-    private:
-        const Vector<CodeToken>& m_tokens;
-        SizeT m_position;
-    };
-
-    SizeT CountToken(const Vector<CodeToken>& tokens, const String& tokenText) {
-        return static_cast<SizeT>(std::count_if(tokens.begin(), tokens.end(),
-                                                [&](const CodeToken& token) { return token.text == tokenText; }));
-    }
-
-    bool HasIdentifierWithPrefixOutsideAllowed(const Vector<CodeToken>& tokens, const String& prefix,
-                                               std::initializer_list<const char*> allowedIdentifiers) {
-        return std::any_of(tokens.begin(), tokens.end(), [&](const CodeToken& token) {
-            if (!IsIdentifierToken(token) || !token.text.starts_with(prefix)) {
-                return false;
-            }
-            return std::none_of(allowedIdentifiers.begin(), allowedIdentifiers.end(),
-                                [&](const char* allowed) { return token.text == allowed; });
-        });
-    }
-
-    bool MatchTokenSequence(const Vector<CodeToken>& tokens, SizeT position,
-                            std::initializer_list<const char*> expected) {
-        if (position + expected.size() > tokens.size()) {
-            return false;
-        }
-        for (const char* token : expected) {
-            if (tokens[position++].text != token) {
-                return false;
-            }
-        }
-        return true;
-    }
-
-    struct LinearPrefixScanMatch {
-        SizeT sharedArraySizeBegin = 0;
-        SizeT sharedArraySizeEnd = 0;
-        SizeT scanBegin = 0;
-        SizeT scanEnd = 0;
-        String cache;
-        String importance;
-        String prefixSum;
-        String loopLength;
-        String loopIndex;
-        String sum;
-    };
-
-    bool ParseLinearPrefixScanTemplate(const Vector<CodeToken>& tokens, LinearPrefixScanMatch& match) {
-        // The workaround deliberately recognizes one complete algorithm, not merely the
-        // subgroupInclusiveAdd token. Changing scratch storage is only safe when that storage is
-        // private to this scan and the workgroup has exactly 1024 X invocations.
-        SizeT localSizeDeclarationCount = 0;
-        for (SizeT i = 0; i < tokens.size(); ++i) {
-            if (MatchTokenSequence(tokens, i, {"layout", "(", "local_size_x", "=", "1024", ")", "in", ";"})) {
-                ++localSizeDeclarationCount;
-            }
-        }
-        if (localSizeDeclarationCount != 1) {
-            return false;
-        }
-
-        SizeT sharedDeclarationIndex = String::npos;
-        SizeT sharedDeclarationCount = 0;
-        String cacheName;
-        for (SizeT i = 0; i + 6 < tokens.size(); ++i) {
-            if (tokens[i].text != "shared" || tokens[i + 1].text != "float" || !IsIdentifierToken(tokens[i + 2]) ||
-                tokens[i + 3].text != "[" || tokens[i + 4].text != "64" || tokens[i + 5].text != "]" ||
-                tokens[i + 6].text != ";") {
-                continue;
-            }
-            ++sharedDeclarationCount;
-            sharedDeclarationIndex = i;
-            cacheName = tokens[i + 2].text;
-        }
-        if (sharedDeclarationCount != 1) {
-            return false;
-        }
-
-        SizeT scanTokenIndex = String::npos;
-        SizeT scanCount = 0;
-        for (SizeT i = 0; i + 7 < tokens.size(); ++i) {
-            if (tokens[i].text == "float" && IsIdentifierToken(tokens[i + 1]) && tokens[i + 2].text == "=" &&
-                tokens[i + 3].text == "subgroupInclusiveAdd" && tokens[i + 4].text == "(" &&
-                IsIdentifierToken(tokens[i + 5]) && tokens[i + 6].text == ")" && tokens[i + 7].text == ";") {
-                ++scanCount;
-                scanTokenIndex = i;
-            }
-        }
-        if (scanCount != 1 || sharedDeclarationIndex >= scanTokenIndex) {
-            return false;
-        }
-
-        TokenCursor cursor(tokens, scanTokenIndex);
-        String prefixSum;
-        String importance;
-        String loopLength;
-        String loopIndex;
-        String sum;
-        if (!cursor.Consume("float") || !cursor.ConsumeAnyIdentifier(prefixSum) || !cursor.Consume("=") ||
-            !cursor.Consume("subgroupInclusiveAdd") || !cursor.Consume("(") ||
-            !cursor.ConsumeAnyIdentifier(importance) || !cursor.Consume(")") || !cursor.Consume(";") ||
-            !cursor.Consume("if") || !cursor.Consume("(") || !cursor.Consume("gl_SubgroupInvocationID") ||
-            !cursor.Consume("==") || !cursor.Consume("gl_SubgroupSize") || !cursor.Consume("-") ||
-            !cursor.Consume("1u") || !cursor.Consume(")") || !cursor.ConsumeIdentifier(cacheName) ||
-            !cursor.Consume("[") || !cursor.Consume("gl_SubgroupID") || !cursor.Consume("]") || !cursor.Consume("=") ||
-            !cursor.ConsumeIdentifier(prefixSum) || !cursor.Consume(";") || !cursor.Consume("barrier") ||
-            !cursor.Consume("(") || !cursor.Consume(")") || !cursor.Consume(";") || !cursor.Consume("uint") ||
-            !cursor.ConsumeAnyIdentifier(loopLength) || !cursor.Consume("=") || !cursor.Consume("uint") ||
-            !cursor.Consume("(") || !cursor.Consume("findMSB") || !cursor.Consume("(") ||
-            !cursor.Consume("gl_NumSubgroups") || !cursor.Consume(")") || !cursor.Consume(")") ||
-            !cursor.Consume(";") || !cursor.ConsumeIdentifier(loopLength) || !cursor.Consume("+=") ||
-            !cursor.Consume("uint") || !cursor.Consume("(") || !cursor.Consume("gl_NumSubgroups") ||
-            !cursor.Consume("-") || !cursor.Consume("(") || !cursor.Consume("1u") || !cursor.Consume("<<") ||
-            !cursor.Consume("(") || !cursor.ConsumeIdentifier(loopLength) || !cursor.Consume("-") ||
-            !cursor.Consume("1u") || !cursor.Consume(")") || !cursor.Consume(")") || !cursor.Consume(">") ||
-            !cursor.Consume("0u") || !cursor.Consume(")") || !cursor.Consume(";") || !cursor.Consume("for") ||
-            !cursor.Consume("(") || !cursor.Consume("uint") || !cursor.ConsumeAnyIdentifier(loopIndex) ||
-            !cursor.Consume("=") || !cursor.Consume("0") || !cursor.Consume(";") ||
-            !cursor.ConsumeIdentifier(loopIndex) || !cursor.Consume("<") || !cursor.ConsumeIdentifier(loopLength) ||
-            !cursor.Consume(";") || !cursor.ConsumeIdentifier(loopIndex) || !cursor.Consume("++") ||
-            !cursor.Consume(")") || !cursor.Consume("{") || !cursor.Consume("if") || !cursor.Consume("(") ||
-            !cursor.Consume("(") || !cursor.Consume("gl_SubgroupID") || !cursor.Consume("&") || !cursor.Consume("(") ||
-            !cursor.Consume("1u") || !cursor.Consume("<<") || !cursor.ConsumeIdentifier(loopIndex) ||
-            !cursor.Consume(")") || !cursor.Consume(")") || !cursor.Consume(">") || !cursor.Consume("0u") ||
-            !cursor.Consume(")") || !cursor.Consume("{") || !cursor.ConsumeIdentifier(prefixSum) ||
-            !cursor.Consume("+=") || !cursor.ConsumeIdentifier(cacheName) || !cursor.Consume("[") ||
-            !cursor.Consume("(") || !cursor.Consume("gl_SubgroupID") || !cursor.Consume(">>") ||
-            !cursor.ConsumeIdentifier(loopIndex) || !cursor.Consume("<<") || !cursor.ConsumeIdentifier(loopIndex) ||
-            !cursor.Consume(")") || !cursor.Consume("-") || !cursor.Consume("1u") || !cursor.Consume("]") ||
-            !cursor.Consume(";") || !cursor.Consume("if") || !cursor.Consume("(") ||
-            !cursor.Consume("gl_SubgroupInvocationID") || !cursor.Consume("==") || !cursor.Consume("gl_SubgroupSize") ||
-            !cursor.Consume("-") || !cursor.Consume("1u") || !cursor.Consume(")") ||
-            !cursor.ConsumeIdentifier(cacheName) || !cursor.Consume("[") || !cursor.Consume("gl_SubgroupID") ||
-            !cursor.Consume("]") || !cursor.Consume("=") || !cursor.ConsumeIdentifier(prefixSum) ||
-            !cursor.Consume(";") || !cursor.Consume("}") || !cursor.Consume("barrier") || !cursor.Consume("(") ||
-            !cursor.Consume(")") || !cursor.Consume(";") || !cursor.Consume("}") || !cursor.Consume("if") ||
-            !cursor.Consume("(") || !cursor.Consume("gl_LocalInvocationID") || !cursor.Consume(".") ||
-            !cursor.Consume("x") || !cursor.Consume("==") || !cursor.Consume("uint") || !cursor.Consume("(") ||
-            !cursor.Consume("1024") || !cursor.Consume("-") || !cursor.Consume("1") || !cursor.Consume(")") ||
-            !cursor.Consume(")") || !cursor.ConsumeIdentifier(cacheName) || !cursor.Consume("[") ||
-            !cursor.Consume("0") || !cursor.Consume("]") || !cursor.Consume("=") ||
-            !cursor.ConsumeIdentifier(prefixSum) || !cursor.Consume(";") || !cursor.Consume("barrier") ||
-            !cursor.Consume("(") || !cursor.Consume(")") || !cursor.Consume(";") || !cursor.Consume("float") ||
-            !cursor.ConsumeAnyIdentifier(sum) || !cursor.Consume("=") || !cursor.ConsumeIdentifier(cacheName) ||
-            !cursor.Consume("[") || !cursor.Consume("0") || !cursor.Consume("]") || !cursor.Consume(";")) {
-            return false;
-        }
-        const SizeT scanEndToken = cursor.Position() - 1;
-
-        // Require the scan's immediate consumer as well. This makes the match specific to a
-        // linear distribution warp, and avoids changing unrelated prefix scans which may rely on
-        // the implementation's native subgroup partitioning.
-        if (!cursor.Consume("float") || !cursor.ConsumeAnyIdentifier() || !cursor.Consume("=") ||
-            !cursor.Consume("(") || !cursor.ConsumeIdentifier(prefixSum) || !cursor.Consume("-") ||
-            !cursor.ConsumeIdentifier(importance) || !cursor.Consume(")") || !cursor.Consume("/") ||
-            !cursor.ConsumeIdentifier(sum) || !cursor.Consume("-") || !cursor.Consume("float") ||
-            !cursor.Consume("(") || !cursor.Consume("gl_LocalInvocationID") || !cursor.Consume(".") ||
-            !cursor.Consume("x") || !cursor.Consume("+") || !cursor.Consume("1u") || !cursor.Consume(")") ||
-            !cursor.Consume("/") || !cursor.Consume("float") || !cursor.Consume("(") || !cursor.Consume("1024") ||
-            !cursor.Consume(")") || !cursor.Consume(";")) {
-            return false;
-        }
-
-        // No other use may share the scratch array, and no additional subgroup operation or
-        // builtin may silently retain native-64 semantics after this module becomes virtual-32.
-        if (CountToken(tokens, cacheName) != 6 || CountToken(tokens, "subgroupInclusiveAdd") != 1 ||
-            CountToken(tokens, "gl_SubgroupInvocationID") != 2 || CountToken(tokens, "gl_SubgroupSize") != 2 ||
-            CountToken(tokens, "gl_SubgroupID") != 4 || CountToken(tokens, "gl_NumSubgroups") != 2 ||
-            CountToken(tokens, "gl_LocalInvocationID") != 2 || CountToken(tokens, "barrier") != 3 ||
-            CountToken(tokens, "findMSB") != 1 ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "subgroup", {"subgroupInclusiveAdd"}) ||
-            HasIdentifierWithPrefixOutsideAllowed(
-                tokens, "gl_Subgroup",
-                {"gl_SubgroupInvocationID", "gl_SubgroupSize", "gl_SubgroupID", "gl_NumSubgroups"}) ||
-            // ARB/NV spellings of lane-width-sensitive builtins and functions
-            // (gl_SubGroupSizeARB, ballotARB, gl_WarpSizeNV, shuffleNV, ...) must block the
-            // rewrite just like their KHR counterparts: they would silently keep native-width
-            // semantics in a module rewritten to the virtual 32-lane model.
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "gl_SubGroup", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "gl_Warp", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "gl_Thread", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "gl_SMID", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "ballot", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "shuffle", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "readInvocation", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "readFirstInvocation", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "anyInvocation", {}) ||
-            HasIdentifierWithPrefixOutsideAllowed(tokens, "allInvocations", {})) {
-            return false;
-        }
-
-        // The scan must be at the top level of the sole main() body. Its existing barriers already
-        // require uniform control flow; this check prevents us from introducing extra barriers in
-        // a nested branch or loop.
-        SizeT mainOpenBrace = String::npos;
-        SizeT mainCloseBrace = String::npos;
-        SizeT mainCount = 0;
-        for (SizeT i = 0; i + 4 < tokens.size(); ++i) {
-            if (!MatchTokenSequence(tokens, i, {"void", "main", "(", ")", "{"})) {
-                continue;
-            }
-            ++mainCount;
-            mainOpenBrace = i + 4;
-            int depth = 1;
-            for (SizeT j = mainOpenBrace + 1; j < tokens.size(); ++j) {
-                if (tokens[j].text == "{")
-                    ++depth;
-                else if (tokens[j].text == "}" && --depth == 0) {
-                    mainCloseBrace = j;
-                    break;
-                }
-            }
-        }
-        if (mainCount != 1 || mainCloseBrace == String::npos || scanTokenIndex <= mainOpenBrace ||
-            scanEndToken >= mainCloseBrace) {
-            return false;
-        }
-        int depthAtScan = 1;
-        for (SizeT i = mainOpenBrace + 1; i < scanTokenIndex; ++i) {
-            if (tokens[i].text == "{")
-                ++depthAtScan;
-            else if (tokens[i].text == "}")
-                --depthAtScan;
-        }
-        if (depthAtScan != 1) {
-            return false;
-        }
-
-        constexpr const char* injectedNames[] = {"mglPrefixScanLane",  "mglVirtualSubgroupInvocation",
-                                                 "mglVirtualSubgroup", "mglVirtualSubgroupBase",
-                                                 "mglPrefixLane",      "mglVirtualSubgroupCount"};
-        for (const char* injectedName : injectedNames) {
-            if (CountToken(tokens, injectedName) != 0) {
-                return false;
-            }
-        }
-
-        match.sharedArraySizeBegin = tokens[sharedDeclarationIndex + 4].begin;
-        match.sharedArraySizeEnd = tokens[sharedDeclarationIndex + 4].end;
-        match.scanBegin = tokens[scanTokenIndex].begin;
-        match.scanEnd = tokens[scanEndToken].end;
-        match.cache = std::move(cacheName);
-        match.importance = std::move(importance);
-        match.prefixSum = std::move(prefixSum);
-        match.loopLength = std::move(loopLength);
-        match.loopIndex = std::move(loopIndex);
-        match.sum = std::move(sum);
-        return true;
-    }
-
-    String BuildLinearPrefixScanReplacement(const LinearPrefixScanMatch& match) {
-        String replacement;
-        replacement.reserve(1800);
-        replacement += "uint mglPrefixScanLane = gl_LocalInvocationID.x;\n";
-        replacement += "uint mglVirtualSubgroupInvocation = mglPrefixScanLane & 31u;\n";
-        replacement += "uint mglVirtualSubgroup = mglPrefixScanLane >> 5u;\n";
-        replacement += "const uint mglVirtualSubgroupCount = 32u;\n";
-        replacement += match.cache + "[mglPrefixScanLane] = " + match.importance + ";\n";
-        replacement += "barrier();\n";
-        replacement += "float " + match.prefixSum + " = 0.0f;\n";
-        replacement += "uint mglVirtualSubgroupBase = mglVirtualSubgroup << 5u;\n";
-        replacement += "for (uint mglPrefixLane = mglVirtualSubgroupBase; "
-                       "mglPrefixLane <= mglPrefixScanLane; ++mglPrefixLane) {\n";
-        replacement += match.prefixSum + " += " + match.cache + "[mglPrefixLane];\n";
-        replacement += "}\n";
-        replacement += "barrier();\n";
-        replacement += "if (mglVirtualSubgroupInvocation == 31u) " + match.cache +
-                       "[mglVirtualSubgroup] = " + match.prefixSum + ";\n";
-        replacement += "barrier();\n";
-        replacement += "uint " + match.loopLength + " = uint(findMSB(mglVirtualSubgroupCount));\n";
-        replacement +=
-            match.loopLength + " += uint(mglVirtualSubgroupCount - (1u << (" + match.loopLength + " - 1u)) > 0u);\n";
-        replacement += "for (uint " + match.loopIndex + " = 0u; " + match.loopIndex + " < " + match.loopLength +
-                       "; ++" + match.loopIndex + ") {\n";
-        replacement += "if ((mglVirtualSubgroup & (1u << " + match.loopIndex + ")) > 0u) {\n";
-        replacement += match.prefixSum + " += " + match.cache + "[(mglVirtualSubgroup >> " + match.loopIndex + " << " +
-                       match.loopIndex + ") - 1u];\n";
-        replacement += "if (mglVirtualSubgroupInvocation == 31u) " + match.cache +
-                       "[mglVirtualSubgroup] = " + match.prefixSum + ";\n";
-        replacement += "}\nbarrier();\n}\n";
-        replacement += "if (mglPrefixScanLane == 1023u) " + match.cache + "[0] = " + match.prefixSum + ";\n";
-        replacement += "barrier();\n";
-        replacement += "float " + match.sum + " = " + match.cache + "[0];";
-        return replacement;
-    }
-
     void SkipDirectiveWhitespace(const MobileGL::String& source, SizeT& pos, SizeT lineEnd) {
         while (pos < lineEnd && std::isspace(static_cast<unsigned char>(source[pos]))) {
             pos++;
@@ -1253,117 +929,6 @@ namespace {
 namespace MobileGL {
     namespace MG_Util {
         namespace ShaderTranspiler {
-            Bool RewriteLinearSubgroupPrefixScanForVulkan(ShaderStage stage, Uint32 nativeSubgroupSize,
-                                                          String& source) {
-                constexpr Uint32 capturedSubgroupSize = 32;
-                if (stage != ShaderStage::Compute || nativeSubgroupSize <= capturedSubgroupSize ||
-                    nativeSubgroupSize % capturedSubgroupSize != 0) {
-                    return false;
-                }
-
-                // Vulkan subgroup widths are powers of two. Keep the workaround restricted to
-                // wider widths which are a power-of-two multiple of the captured 32-lane model.
-                const Uint32 subgroupScale = nativeSubgroupSize / capturedSubgroupSize;
-                if ((subgroupScale & (subgroupScale - 1u)) != 0u) {
-                    return false;
-                }
-
-                const Vector<CodeToken> tokens = TokenizeCode(source);
-                LinearPrefixScanMatch match;
-                if (!ParseLinearPrefixScanTemplate(tokens, match)) {
-                    // Diagnosability: when the trigger op is present but the template no longer
-                    // matches (e.g. the pack shipped a new shader revision), the affected device
-                    // silently falls back to the driver's miscompiled path. Make that visible.
-                    if (CountToken(tokens, "subgroupInclusiveAdd") > 0) {
-                        MGLOG_W_ONCE("%s: subgroupInclusiveAdd present but the linear prefix-scan template "
-                                "did not match; the wide-subgroup rewrite was NOT applied",
-                                __func__);
-                    }
-                    return false;
-                }
-
-                const String replacement = BuildLinearPrefixScanReplacement(match);
-                source.replace(match.scanBegin, match.scanEnd - match.scanBegin, replacement);
-                // The declaration occurs before the replaced scan, so its original offsets remain
-                // valid after the first replacement.
-                source.replace(match.sharedArraySizeBegin, match.sharedArraySizeEnd - match.sharedArraySizeBegin,
-                               "1024");
-                return true;
-            }
-
-            namespace {
-                struct ShaderSourceQuirkContext {
-                    ShaderStage stage = ShaderStage::Unknown;
-                    BackendType backend = BackendType::Unknown;
-                    MG_Backend::GpuVendorKind vendor = MG_Backend::GpuVendorKind::Unknown;
-                    Uint32 subgroupSize = 0;
-                };
-
-                // Device-quirk registry. Every entry is a narrowly scoped source rewrite that
-                // works around a specific driver defect. A quirk runs when its env override
-                // forces it on, or when the override is Auto and DeviceApplies matches the
-                // detected device. ForceOn bypasses only the device gate - each Apply keeps
-                // its own structural safety checks. Add new per-device workarounds here
-                // instead of open-coding them in PreprocessShaderSource.
-                struct ShaderSourceQuirk {
-                    const char* name;
-                    // Reads the override out of the captured env, never out of the live
-                    // MG_Config table: a worker must see the same config the GL thread saw.
-                    MG_Config::QuirkOverride (*GetOverride)(const CompileEnv&);
-                    Bool (*DeviceApplies)(const ShaderSourceQuirkContext&);
-                    Bool (*Apply)(const ShaderSourceQuirkContext&, String&);
-                };
-
-                constexpr ShaderSourceQuirk kShaderSourceQuirks[] = {
-                    {
-                        // MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN
-                        "subgroup-prefix-scan-rewrite",
-                        [](const CompileEnv& env) { return env.subgroupPrefixScanQuirk; },
-                        [](const ShaderSourceQuirkContext& ctx) {
-                            // Qualcomm's Vulkan driver miscompiles the recognized float
-                            // InclusiveScan pattern for native subgroups wider than the
-                            // captured 32 lanes; other vendors compile it correctly and
-                            // should keep their native scan.
-                            return ctx.backend == BackendType::DirectVulkan &&
-                                   ctx.vendor == MG_Backend::GpuVendorKind::Qualcomm;
-                        },
-                        [](const ShaderSourceQuirkContext& ctx, String& source) {
-                            return RewriteLinearSubgroupPrefixScanForVulkan(ctx.stage, ctx.subgroupSize,
-                                                                            source);
-                        },
-                    },
-                };
-
-                void ApplyShaderSourceQuirks(const CompileEnv& env, ShaderStage stage, String& source) {
-                    // No backend at capture time means no device to match a quirk against,
-                    // and (as before) no quirk can fire - not even a forced one, because
-                    // every Apply reads device parameters that do not exist yet.
-                    if (!env.HasBackend()) {
-                        return;
-                    }
-                    const ShaderSourceQuirkContext quirkContext{
-                        stage,
-                        env.backend,
-                        env.params.GpuVendor,
-                        env.params.SubgroupSize,
-                    };
-                    for (const ShaderSourceQuirk& quirk : kShaderSourceQuirks) {
-                        const MG_Config::QuirkOverride quirkOverride = quirk.GetOverride(env);
-                        if (quirkOverride == MG_Config::QuirkOverride::ForceOff) {
-                            continue;
-                        }
-                        if (quirkOverride == MG_Config::QuirkOverride::Auto &&
-                            !quirk.DeviceApplies(quirkContext)) {
-                            continue;
-                        }
-                        if (quirk.Apply(quirkContext, source)) {
-                            MGLOG_D("ApplyShaderSourceQuirks: applied '%s'%s", quirk.name,
-                                    quirkOverride == MG_Config::QuirkOverride::ForceOn ? " (forced on)" : "");
-                        }
-                    }
-                }
-            } // namespace
-
             void PreprocessShaderSource(ShaderStage stage, String& source) {
                 PreprocessShaderSource(stage, source, *GetCurrentCompileEnv());
             }
@@ -1403,7 +968,6 @@ namespace MobileGL {
                 ModernizeLegacyGLSL(stage, source, afterVersion);
                 InjectDepthRangeBuiltinShim(stage, source, afterVersion);
 
-                ApplyShaderSourceQuirks(env, stage, source);
             }
 
             Bool RetargetLegacyVersionDirectiveTo460(String& source) {

MobileGL/MG_Util/ShaderTranspiler/ShaderSourceProcessor.h

@@ -30,18 +30,6 @@ namespace MobileGL {
             // tests and diagnostics that drive the preprocessor standalone.
             void PreprocessShaderSource(ShaderStage stage, String& source);
 
-            // Some desktop-captured compute shaders build a workgroup-wide linear prefix scan
-            // from subgroupInclusiveAdd plus a shared array of subgroup totals. Qualcomm's
-            // Vulkan driver miscompiles that exact float InclusiveScan path for native subgroups
-            // 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. 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
             // from a legacy desktop version back up to "#version 460 core". Returns false (leaving
             // the source untouched) for anything else: ES, compatibility, or an already-modern

android-plugin/app/build.gradle.kts

@@ -86,7 +86,7 @@ val pluginRendererConfig = buildJsonValue {
             selectable(
                 key = "MOBILEGL_BACKEND_TYPE",
                 title = RendererConfig.MetaString("mobilegl_backend_type_title"),
-                items = RendererConfig.EnvItems("DirectGLES", listOf("DirectVulkan")),
+                items = RendererConfig.EnvItems("DirectGLES", listOf("DirectVulkan", "DiligentVulkan")),
             )
             toggleable("MOBILEGL_DISABLE_TIMERQUERY", "1", false, RendererConfig.MetaString("mobilegl_disable_timerquery_title"))
             toggleable("MOBILEGL_DISABLE_SUBGROUP", "1", false, RendererConfig.MetaString("mobilegl_disable_subgroup_title"))

include/KHR/khrplatform.h

@@ -221,10 +221,13 @@ typedef signed   char          khronos_int8_t;
 typedef unsigned char          khronos_uint8_t;
 typedef signed   short int     khronos_int16_t;
 typedef unsigned short int     khronos_uint16_t;
-typedef signed   long  int     khronos_intptr_t;
-typedef unsigned long  int     khronos_uintptr_t;
-typedef signed   long  int     khronos_ssize_t;
-typedef unsigned long  int     khronos_usize_t;
+/* `long` is 32-bit on LLP64 Windows, including 64-bit MinGW.  Use the
+ * standard pointer-sized integer types so these remain pointer-width there. */
+#include <stdint.h>
+typedef intptr_t                   khronos_intptr_t;
+typedef uintptr_t                  khronos_uintptr_t;
+typedef intptr_t                   khronos_ssize_t;
+typedef uintptr_t                  khronos_usize_t;
 
 #if KHRONOS_SUPPORT_FLOAT
 /*

tools/trace_replay/trace_cases.json

@@ -277,11 +277,14 @@
     },
     {
       "name": "minecraft-1.21.4-fabric-iris-iterationrp-in-world",
-      "ci": false,
+      "ci_backends": [
+        "DirectVulkan"
+      ],
       "trace_archive": "minecraft-1.21.4-fabric-iris-iterationrp-in-world.tgz",
       "golden": "minecraft-1.21.4-fabric-iris-iterationrp-in-world.0000202020.png",
       "target_call": 202020,
-      "timeout_seconds": 1800
+      "timeout_seconds": 1800,
+      "ssim_threshold": 0.98
     },
     {
       "name": "minecraft-1.21.4-fabric-iris-bsl-esc-menu-854",

tools/trace_replay/trace_cases.py

@@ -6,6 +6,7 @@ from pathlib import Path
 
 
 TRACE_CASES_JSON = Path(__file__).with_name("trace_cases.json")
+CI_BACKENDS = ("DirectGLES", "DirectVulkan")
 
 
 def load_trace_case_manifest(path=TRACE_CASES_JSON):
@@ -71,6 +72,46 @@ def ci_trace_cases(cases):
     return [case for case in cases if case.get("ci", True)]
 
 
+def ci_backends(case):
+    backends = case.get("ci_backends")
+    if backends is None:
+        return CI_BACKENDS
+    if not isinstance(backends, list) or not backends:
+        raise ValueError(f"ci_backends must be a non-empty list for {case['name']}")
+    unknown = [backend for backend in backends if backend not in CI_BACKENDS]
+    if unknown:
+        raise ValueError(
+            f"unknown ci_backends for {case['name']}: {', '.join(unknown)}"
+        )
+    if len(set(backends)) != len(backends):
+        raise ValueError(f"ci_backends contains duplicates for {case['name']}")
+    return backends
+
+
+def github_test_matrix(cases):
+    return {
+        "include": [
+            {"backend": backend, "case": case["name"]}
+            for case in cases
+            for backend in ci_backends(case)
+        ]
+    }
+
+
+def github_apk_matrix(cases):
+    backends = {
+        "DirectGLES": {"name": "DirectGLES", "gpu": "software"},
+        "DirectVulkan": {"name": "DirectVulkan", "gpu": "lavapipe"},
+    }
+    return {
+        "include": [
+            {"backend": backends[backend], "case": github_apk_case(case)}
+            for case in cases
+            for backend in ci_backends(case)
+        ]
+    }
+
+
 def cmake_quote(value):
     return '"' + str(value).replace("\\", "/").replace('"', '\\"') + '"'
 
@@ -114,7 +155,14 @@ def parse_args():
     parser.add_argument("--fixture-root", default="tools/trace_replay/fixtures")
     parser.add_argument(
         "--format",
-        choices=("names", "github-apk", "fixture-files", "cmake"),
+        choices=(
+            "names",
+            "github-test-matrix",
+            "github-apk",
+            "github-apk-matrix",
+            "fixture-files",
+            "cmake",
+        ),
         default="names",
     )
     return parser.parse_args()
@@ -127,8 +175,12 @@ def main():
         cases = ci_trace_cases(cases)
     if args.format == "names":
         print(json.dumps([case["name"] for case in cases], separators=(",", ":")))
+    elif args.format == "github-test-matrix":
+        print(json.dumps(github_test_matrix(cases), separators=(",", ":")))
     elif args.format == "github-apk":
         print(json.dumps([github_apk_case(case) for case in cases], separators=(",", ":")))
+    elif args.format == "github-apk-matrix":
+        print(json.dumps(github_apk_matrix(cases), separators=(",", ":")))
     elif args.format == "fixture-files":
         if not args.case_name:
             print("--case is required for --format fixture-files", file=sys.stderr)