[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

@@ -182,6 +182,7 @@ set(ENABLE_SPVREMAPPER         OFF CACHE BOOL "Enable SPVRemapper" FORCE)
 set(ENABLE_OPT                 ON  CACHE BOOL "Enable SPIRV-Tools opt usage in glslang" FORCE)
 set(BUILD_EXTERNAL             ON  CACHE BOOL "Build external deps in External/" FORCE)
 set(ENABLE_GLSLANG_INSTALL     OFF CACHE BOOL "Install glslang targets" FORCE)
+set(SPIRV_SKIP_EXECUTABLES     ON  CACHE BOOL "Skip building SPIRV-Tools executables" FORCE)
 
 set(SPIRV_CROSS_C_API          ON  CACHE BOOL "Enable C API" FORCE)
 set(SPIRV_CROSS_ENABLE_GLSL    ON  CACHE BOOL "Enable GLSL backend" FORCE)
@@ -198,7 +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)
@@ -210,6 +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)
@@ -395,6 +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
@@ -435,6 +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
@@ -500,6 +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)
@@ -558,6 +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

@@ -66,14 +66,12 @@ namespace MobileGL::MG_Config {
     //   - DISPLAY: X11 session variable, not MobileGL configuration.
     //   - MOBILEGL_LOG_FILE_PATH: log-file init runs before MG_ConfigLoader::Init
     //     (see MG_Util/Debug/Log.cpp).
-    //   - MOBILEGL_VALIDATE_SPIRV: test suites like SpirvPassTest exercise
-    //     ShaderCompiler without ever running MobileGL::Initialize(), and every
-    //     Initialize() re-runs MG_ConfigLoader::Init, which would clobber a
-    //     programmatic override stored here (see ShaderCompiler.cpp,
-    //     SpirvValidationEnabled).
     struct FeaturesTable {
         // MOBILEGL_DISABLE_TIMERQUERY: do not advertise or use GPU timer queries.
         Bool DisableTimerQuery = false;
+        // MOBILEGL_ENABLE_SPIRV_VALIDATION: validate generated and transformed SPIR-V.
+        // Disabled by default because validation is a diagnostics-only cost.
+        Bool EnableSpirvValidation = false;
         // MOBILEGL_USE_ANGLE: load ANGLE EGL/GLES libraries.
         Bool UseAngle = false;
 #if defined(MOBILEGL_TRACE_ANGLE_VARIANTS)
@@ -130,10 +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

@@ -162,6 +162,7 @@ namespace MobileGL::MG_ConfigLoader {
     inline void InitFeatures() {
         auto& features = MG_Config::Features;
         features.DisableTimerQuery = QueryEnvFlag("MOBILEGL_DISABLE_TIMERQUERY");
+        features.EnableSpirvValidation = QueryEnvFlag("MOBILEGL_ENABLE_SPIRV_VALIDATION");
         features.UseAngle = QueryEnvFlag("MOBILEGL_USE_ANGLE");
 #if defined(MOBILEGL_TRACE_ANGLE_VARIANTS)
         QueryEnvVariable("MOBILEGL_TRACE_ANGLE_VARIANT", features.TraceAngleVariant, "");
@@ -179,7 +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");
@@ -202,6 +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/DirectGLES/BackendObject_DirectGLES.cpp

@@ -712,9 +712,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     {
                         .TargetGLVersion = {4, 0, 0},   // GL target version
                         .TargetGLSLVersion = {4, 6, 0}, // Target Shading Language Version
-                        // Baseline advertisement (no timer queries / anisotropy yet); reconciled
-                        // once the ES capabilities exist, see UpdateAdvertisedCapabilityExtensions.
-                        .Extensions = BuildAdvertisedExtensions(false, false),
+                        // Baseline advertisement (no runtime capabilities yet); reconciled once
+                        // the ES capabilities exist, see UpdateAdvertisedCapabilityExtensions.
+                        .Extensions = BuildAdvertisedExtensions(false, false, false, false),
                         .IsCompatibilityProfile = false // Is Compatibility Profile
                     },
                 .StaticBackendCapability = {.AllowVSOnlyPrograms = false} // Backend Capability
@@ -734,9 +734,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
         // thread can only observe the extension string after the
         // advertisement for its context has settled; rebuilding the whole
         // list keeps the re-run after a context recreation idempotent.
-        void UpdateAdvertisedCapabilityExtensions(Bool anisotropicFilteringSupported) {
-            MutableRendererInfo().RendererGLInfo.Extensions =
-                BuildAdvertisedExtensions(AreTimerQueriesSupported(), anisotropicFilteringSupported);
+        void UpdateAdvertisedCapabilityExtensions(const MG_External::GLESCapabilities& capabilities) {
+            MutableRendererInfo().RendererGLInfo.Extensions = BuildAdvertisedExtensions(
+                AreTimerQueriesSupported(), capabilities.SupportsTextureFilterAnisotropy,
+                capabilities.SupportsDrawIndirect,
+                capabilities.SupportsDrawIndirect && capabilities.SupportsBaseInstance);
         }
     } // namespace
 
@@ -779,11 +781,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
             return false;
         }
         DirectGLES::SetGLESCapabilities(m_GLESCapabilities);
-        // Now that g_GLESCapabilities knows about GL_EXT_disjoint_timer_query and
-        // GL_EXT_texture_filter_anisotropic, reconcile the advertisement (see the comment on
-        // UpdateAdvertisedCapabilityExtensions for why it cannot happen when the extension
-        // list is first built).
-        UpdateAdvertisedCapabilityExtensions(m_GLESCapabilities.SupportsTextureFilterAnisotropy);
+        // Now that g_GLESCapabilities knows the host extensions, entry points, and ES version,
+        // reconcile every runtime-gated advertisement (see the comment on
+        // UpdateAdvertisedCapabilityExtensions for why this cannot happen when the list is first
+        // built).
+        UpdateAdvertisedCapabilityExtensions(m_GLESCapabilities);
         UpdateDynamicBackendParameters();
         PopulateFormatCapabilities(m_GLESFunctions, m_GLESCapabilities, MutableFormatCapabilities());
         PrintFormatCapabilities(GetFormatCapabilities());
@@ -924,11 +926,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
         return MutableRendererInfo();
     }
 
-    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported) {
+    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported,
+                                                  Bool drawIndirectSupported,
+                                                  Bool nonZeroIndirectBaseInstanceSupported) {
         Vector<GLExtension> extensions = {
             V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, E_GL_ARB_draw_buffers_blend,
             E_GL_ARB_compute_shader, E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
-            E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_EXT_framebuffer_object,
+            E_GL_ARB_clear_buffer_object, E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_EXT_framebuffer_object,
             E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage, E_GL_ARB_texture_storage,
             E_GL_ARB_texture_storage_multisample, E_GL_ARB_clear_texture, E_GL_ARB_direct_state_access,
             E_GL_ARB_multi_draw_indirect, E_GL_ARB_indirect_parameters, E_GL_ARB_shader_draw_parameters,
@@ -955,6 +959,19 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // extension explicitly permits. It is also the only thing that
             // exposes glProgramParameteri before GL 4.1.
             E_GL_ARB_get_program_binary};
+        // Minecraft 26.3 checks this prerequisite before it even considers
+        // GL_ARB_multi_draw_indirect. ES 3.1 supplies both single-draw entry points; the loader
+        // folds the version and pointer checks into SupportsDrawIndirect.
+        if (drawIndirectSupported) {
+            extensions.push_back(E_GL_ARB_draw_indirect);
+        }
+        // ARB_base_instance also defines the last word of an indirect command. Direct calls are
+        // emulated on every Espryt device, but without host GL_EXT_base_instance a native indirect
+        // draw cannot shift divisor attributes by a GPU-authored non-zero value, so do not promise
+        // that incomplete case.
+        if (drawIndirectSupported && nonZeroIndirectBaseInstanceSupported) {
+            extensions.push_back(E_GL_ARB_base_instance);
+        }
         // GL_KHR_parallel_shader_compile is MobileGL's own capability, not the host ES
         // driver's: the compiler threads are MobileGL's, and glCompileShader/glLinkProgram
         // are serviced entirely inside the frontend. Whether the device driver advertises

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.h

@@ -67,9 +67,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
     const RendererInfo& GetRendererIdentity();
 
     // The full OpenGL extension list Espryt advertises (glGetString(GL_EXTENSIONS))
-    // for a device whose timer queries / anisotropic filtering are (or are not) usable.
+    // for a device whose timer queries / anisotropic filtering / native indirect draws /
+    // non-zero indirect baseInstance semantics are (or are not) usable.
     // The MOBILEGL_DISABLE_TIMERQUERY escape hatch is applied inside.
-    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported);
+    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported,
+                                                  Bool drawIndirectSupported,
+                                                  Bool nonZeroIndirectBaseInstanceSupported);
 
     // Format: <OpenGL ES Renderer>, OpenGL ES <Major>.<Minor> — the exact string an
     // initialized backend returns from GetBackendAPIVersionString (and that ends up

MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp

@@ -3061,10 +3061,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
     }
 
     static Bool SupportsNativeIndirectDraws() {
-        const auto& version = g_GLESCapabilities.GLESVersion;
-        const Bool esVersionOk = version.Major > 3 || (version.Major == 3 && version.Minor >= 1);
-        return esVersionOk && g_GLESFuncs.glDrawElementsIndirect != nullptr &&
-               g_GLESFuncs.glDrawArraysIndirect != nullptr;
+        return g_GLESCapabilities.SupportsDrawIndirect;
     }
 
     // Runs an (indexed) indirect multi-draw. When a GL_DRAW_INDIRECT_BUFFER is bound the draws

MobileGL/MG_Backend/DirectGLES/Managers.cpp

@@ -729,8 +729,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
             void Ops_ReadbackFromGpu(BufferObject& bufferObject) {
                 auto* resource = ResourceOf(bufferObject);
                 if (!resource || resource->id == 0 || !resource->storageInitialized) return;
-                if (resource->persistentMapped) return; // shadow already IS the GPU storage
                 if (!CanTouchGLNow() || resource->contextGeneration != g_bufferContextGeneration) return;
+                if (resource->persistentMapped) {
+                    // Host writes to a persistent map must not race shader writes already queued
+                    // on this context. There is no backend copy to read back in this case.
+                    if (g_GLESFuncs.glFinish) g_GLESFuncs.glFinish();
+                    return;
+                }
                 if (!g_GLESFuncs.glMapBufferRange || !g_GLESFuncs.glUnmapBuffer) return;
                 const SizeT size = std::min<SizeT>(bufferObject.GetSize(), resource->storageSize);
                 if (size == 0) return;
@@ -4741,6 +4746,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 MGLOG_D("%s:", src.empty() ? "" : src.c_str());
             }
             auto& shaderSpirvs = stateProgramObject->GetGeneratedSpirv();
+            const Bool enableSpirvValidation = stateProgramObject->GetSpirvValidationEnabled();
 
             // Blocks a transform-feedback capture request names a member of ("StageData" of
             // "StageData.attrib[0]"). The Adreno ES driver accepts such a request, links, and
@@ -4794,7 +4800,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 Vector<unsigned int> loweredSpirv;
                 const Vector<unsigned int>* effectiveSpirv = &spirvCode;
                 if (glShaderType == GL_VERTEX_SHADER &&
-                    MG_Util::ShaderTranspiler::ShaderCompiler::LowerDrawParametersForEssl(spirvCode, loweredSpirv) &&
+                    MG_Util::ShaderTranspiler::ShaderCompiler::LowerDrawParametersForEssl(spirvCode, loweredSpirv, enableSpirvValidation) &&
                     !loweredSpirv.empty()) {
                     effectiveSpirv = &loweredSpirv;
                 }
@@ -4804,7 +4810,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 Vector<unsigned int> splitArrayInputSpirv;
                 if (glShaderType == GL_VERTEX_SHADER &&
                     MG_Util::ShaderTranspiler::ShaderCompiler::SplitArrayVertexInputsForEssl(
-                        *effectiveSpirv, splitArrayInputSpirv) &&
+                        *effectiveSpirv, splitArrayInputSpirv, enableSpirvValidation) &&
                     !splitArrayInputSpirv.empty() && splitArrayInputSpirv != *effectiveSpirv) {
                     // Only when the pass ACTUALLY split something. The optimizer hands back a
                     // re-serialised copy either way, and adopting that copy for every vertex
@@ -4826,7 +4832,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 if (!xfbCaptureBlockNames.empty() &&
                     MG_Util::ShaderTranspiler::ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(
                         *effectiveSpirv, xfbCaptureBlockNames, stageFlattenedXfbBlockNames,
-                        flattenedXfbSpirv) &&
+                        flattenedXfbSpirv, enableSpirvValidation) &&
                     !flattenedXfbSpirv.empty() && !stageFlattenedXfbBlockNames.empty()) {
                     effectiveSpirv = &flattenedXfbSpirv;
                     flattenedXfbBlockNames.insert(stageFlattenedXfbBlockNames.begin(),
@@ -4842,7 +4848,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 // declare the member highp; nothing else about emission changes.
                 Vector<unsigned int> uboPrecisionSpirv;
                 if (MG_Util::ShaderTranspiler::ShaderCompiler::StripUboMemberRelaxedPrecisionForEssl(
-                        *effectiveSpirv, uboPrecisionSpirv) &&
+                        *effectiveSpirv, uboPrecisionSpirv, enableSpirvValidation) &&
                     !uboPrecisionSpirv.empty()) {
                     effectiveSpirv = &uboPrecisionSpirv;
                 }
@@ -4857,7 +4863,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 Vector<unsigned int> noperspectiveSpirv;
                 if (!g_GLESCapabilities.SupportsNoperspectiveInterpolation &&
                     MG_Util::ShaderTranspiler::ShaderCompiler::EmulateNoPerspectiveForEssl(
-                        *effectiveSpirv, noperspectiveSpirv) &&
+                        *effectiveSpirv, noperspectiveSpirv, enableSpirvValidation) &&
                     !noperspectiveSpirv.empty()) {
                     effectiveSpirv = &noperspectiveSpirv;
                 }
@@ -4867,7 +4873,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 // divides the coordinate of every normalized-coordinate lookup by the texture
                 // size, which is the whole of the difference between the two.
                 Vector<unsigned int> rectLoweredSpirv;
-                if (MG_Util::ShaderTranspiler::ShaderCompiler::LowerRectImages(*effectiveSpirv, rectLoweredSpirv) &&
+                if (MG_Util::ShaderTranspiler::ShaderCompiler::LowerRectImages(*effectiveSpirv, rectLoweredSpirv, enableSpirvValidation) &&
                     !rectLoweredSpirv.empty()) {
                     effectiveSpirv = &rectLoweredSpirv;
                 }
@@ -4881,7 +4887,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 // coordinate to (u, 0, layer) - before SPIRV-Cross can apply its own.
                 Vector<unsigned int> arrayImageSpirv;
                 if (MG_Util::ShaderTranspiler::ShaderCompiler::Lower1DArrayImagesForEssl(*effectiveSpirv,
-                                                                                          arrayImageSpirv) &&
+                                                                                          arrayImageSpirv, enableSpirvValidation) &&
                     !arrayImageSpirv.empty()) {
                     effectiveSpirv = &arrayImageSpirv;
                 }
@@ -4900,7 +4906,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 if (!imageFormatBake.glFormatByUniformName.empty() &&
                     MG_Util::ShaderTranspiler::ShaderCompiler::DeclaresFormatlessStorageImage(*effectiveSpirv) &&
                     MG_Util::ShaderTranspiler::ShaderCompiler::BakeImageFormatsForEssl(
-                        *effectiveSpirv, imageFormatBake.glFormatByUniformName, imageFormatSpirv) &&
+                        *effectiveSpirv, imageFormatBake.glFormatByUniformName, imageFormatSpirv,
+                        enableSpirvValidation) &&
                     !imageFormatSpirv.empty()) {
                     effectiveSpirv = &imageFormatSpirv;
                 }
@@ -4916,7 +4923,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 Vector<unsigned int> outputIndexSpirv;
                 if (glShaderType == GL_FRAGMENT_SHADER &&
                     MG_Util::ShaderTranspiler::ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(
-                        *effectiveSpirv, outputIndexSpirv) &&
+                        *effectiveSpirv, outputIndexSpirv, enableSpirvValidation) &&
                     !outputIndexSpirv.empty()) {
                     effectiveSpirv = &outputIndexSpirv;
                 }

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());
@@ -497,20 +501,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             .ExtraVendor = Nullopt,
             .RendererGLInfo = {.TargetGLVersion = {4, 0, 0},
                                .TargetGLSLVersion = {4, 6, 0},
-                               // Baseline advertisement (no shader subgroup, no timer queries); a
-                               // live backend reconciles its copy in UpdateAdvertisedExtensions.
-                               .Extensions = BuildAdvertisedExtensions(false, false, false),
+                               // Baseline advertisement (no runtime-gated capabilities); a live
+                               // backend reconciles its copy in UpdateAdvertisedExtensions.
+                               .Extensions = BuildAdvertisedExtensions(false, false, false, false),
                                .IsCompatibilityProfile = false},
             .StaticBackendCapability = {.AllowVSOnlyPrograms = false}};
         return rendererInfo;
     }
 
     Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
-                                                  Bool anisotropicFilteringSupported) {
+                                                  Bool anisotropicFilteringSupported,
+                                                  Bool nonZeroIndirectBaseInstanceSupported) {
         Vector<GLExtension> extensions = {
             V_OpenGL30, V_OpenGL31, V_OpenGL32, V_OpenGL33, V_OpenGL40, E_GL_ARB_draw_buffers_blend,
             E_GL_ARB_compute_shader, E_GL_ARB_shader_storage_buffer_object, E_GL_ARB_shader_image_load_store,
-            E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_ARB_multi_draw_indirect,
+            E_GL_ARB_clear_buffer_object, E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object, E_GL_ARB_draw_indirect,
+            E_GL_ARB_multi_draw_indirect,
             E_GL_ARB_indirect_parameters, E_GL_EXT_framebuffer_object, E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage,
             E_GL_ARB_texture_storage, E_GL_ARB_texture_storage_multisample, E_GL_ARB_texture_multisample,
             E_GL_ARB_clear_texture, E_GL_ARB_direct_state_access, E_GL_ARB_shader_draw_parameters,
@@ -530,6 +536,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // extension explicitly permits. It is also the only thing that
             // exposes glProgramParameteri before GL 4.1.
             E_GL_ARB_get_program_binary};
+        // Vulkan's drawIndirectFirstInstance feature is optional. Direct base-instance calls work
+        // without it, but ARB_base_instance also promises non-zero firstInstance in GPU indirect
+        // commands; the renderer supplies true only when that word is legal and gl_InstanceID can
+        // be rebased to OpenGL's zero-based semantics.
+        if (nonZeroIndirectBaseInstanceSupported) {
+            extensions.push_back(E_GL_ARB_base_instance);
+        }
         if (shaderSubgroupSupported && !MG_Config::Features.DisableSubgroup) {
             extensions.push_back(E_GL_KHR_shader_subgroup);
         }
@@ -678,6 +691,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_vulkanCaps = capabilities;
         UpdateDynamicBackendParameters();
         UpdateAdvertisedExtensions();
+        if (MG_State::pGLContext) {
+            MG_State::pGLContext->InvalidateCompileEnv();
+        }
         MutableFormatCapabilities().Clear();
     }
 
@@ -690,7 +706,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // the whole list keeps re-runs idempotent.
         m_rendererInfo.RendererGLInfo.Extensions = BuildAdvertisedExtensions(
             m_vulkanCaps.SupportsShaderSubgroup, pVulkanRenderer && pVulkanRenderer->IsTimerQuerySupported(),
-            pVulkanRenderer && pVulkanRenderer->IsSamplerAnisotropySupported());
+            pVulkanRenderer && pVulkanRenderer->IsSamplerAnisotropySupported(),
+            pVulkanRenderer && pVulkanRenderer->IsNonZeroIndirectBaseInstanceSupported());
     }
 
     void BackendObject_DirectVulkan::UpdateDynamicBackendParameters() {

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.h

@@ -62,8 +62,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // POST screen shows.
 
     // Static identity of the Magma renderer (renderer/backend names, target GL/GLSL
-    // versions, ExtraVendor) with the baseline extension advertisement (no shader
-    // subgroup, no timer queries). A live backend copies this in its constructor and
+    // versions, ExtraVendor) with the baseline extension advertisement (no runtime-gated
+    // capabilities). A live backend copies this in its constructor and
     // reconciles the Extensions in UpdateAdvertisedExtensions once real capabilities
     // exist; callers that need the advertised list for a known capability set must
     // use BuildAdvertisedExtensions instead.
@@ -74,7 +74,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // MOBILEGL_DISABLE_TIMERQUERY escape hatches are applied inside, so callers pass
     // the detected device support (passing an already-gated value is harmless).
     Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
-                                                  Bool anisotropicFilteringSupported);
+                                                  Bool anisotropicFilteringSupported,
+                                                  Bool nonZeroIndirectBaseInstanceSupported);
 
     // Format: <GPU Name>, Vulkan <Vulkan Version>, Driver <Driver Version> — the exact
     // string an initialized backend returns from GetBackendAPIVersionString (and that

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp

@@ -2973,8 +2973,73 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             bindings.push_back(layoutBinding);
         }
 
+        // UPDATE_AFTER_BIND is strictly an optional per-layout acceleration. The GL
+        // descriptor model still resolves every sampler uniform element independently
+        // (including its texture-unit sampler-object override); selecting this path
+        // changes neither that resolution nor the set versioning in UniformManager.
+        // A conservative count keeps a layout on ordinary descriptors whenever any
+        // relevant update-after-bind limit is not large enough, rather than asking a
+        // driver to reject it during vkCreateDescriptorSetLayout.
+        Uint32 updateAfterBindSamplers = 0;
+        Uint32 updateAfterBindUniformBuffers = 0;
+        Uint32 updateAfterBindStorageBuffers = 0;
+        Uint32 updateAfterBindSampledImages = 0;
+        Uint32 updateAfterBindStorageImages = 0;
+        for (Uint32 binding = 0; binding < m_maxBindings; ++binding) {
+            const Uint32 count = entry.bindingDescriptorCounts[binding];
+            switch (entry.bindingKinds[binding]) {
+                case DescriptorBindingKind::UniformBufferDynamic:
+                    updateAfterBindUniformBuffers += count;
+                    break;
+                case DescriptorBindingKind::CombinedImageSampler:
+                    updateAfterBindSamplers += count;
+                    updateAfterBindSampledImages += count;
+                    break;
+                case DescriptorBindingKind::UniformTexelBuffer:
+                    updateAfterBindSampledImages += count;
+                    break;
+                case DescriptorBindingKind::StorageBuffer:
+                case DescriptorBindingKind::StorageTexelBuffer:
+                    updateAfterBindStorageBuffers += count;
+                    break;
+                case DescriptorBindingKind::StorageImage:
+                    updateAfterBindStorageImages += count;
+                    break;
+                case DescriptorBindingKind::None:
+                    break;
+            }
+        }
+        const Uint32 updateAfterBindResources = updateAfterBindUniformBuffers + updateAfterBindStorageBuffers +
+                                                updateAfterBindSampledImages + updateAfterBindStorageImages;
+        const auto& uab = m_updateAfterBindLimits;
+        entry.usesUpdateAfterBind =
+            uab.enabled && updateAfterBindSamplers <= uab.maxPerStageSamplers &&
+            updateAfterBindUniformBuffers <= uab.maxPerStageUniformBuffers &&
+            updateAfterBindStorageBuffers <= uab.maxPerStageStorageBuffers &&
+            updateAfterBindSampledImages <= uab.maxPerStageSampledImages &&
+            updateAfterBindStorageImages <= uab.maxPerStageStorageImages &&
+            updateAfterBindResources <= uab.maxPerStageResources &&
+            updateAfterBindSamplers <= uab.maxSetSamplers &&
+            updateAfterBindUniformBuffers <= uab.maxSetUniformBuffers &&
+            updateAfterBindUniformBuffers <= uab.maxSetUniformBuffersDynamic &&
+            updateAfterBindStorageBuffers <= uab.maxSetStorageBuffers &&
+            updateAfterBindStorageBuffers <= uab.maxSetStorageBuffersDynamic &&
+            updateAfterBindSampledImages <= uab.maxSetSampledImages &&
+            updateAfterBindStorageImages <= uab.maxSetStorageImages;
+
+        Vector<VkDescriptorBindingFlags> bindingFlags;
+        VkDescriptorSetLayoutBindingFlagsCreateInfo bindingFlagsInfo{};
+        if (entry.usesUpdateAfterBind) {
+            bindingFlags.assign(bindings.size(), VK_DESCRIPTOR_BINDING_UPDATE_AFTER_BIND_BIT);
+            bindingFlagsInfo.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_BINDING_FLAGS_CREATE_INFO;
+            bindingFlagsInfo.bindingCount = static_cast<Uint32>(bindingFlags.size());
+            bindingFlagsInfo.pBindingFlags = bindingFlags.data();
+        }
+
         VkDescriptorSetLayoutCreateInfo setLayoutInfo{};
         setLayoutInfo.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
+        setLayoutInfo.flags = entry.usesUpdateAfterBind ? VK_DESCRIPTOR_SET_LAYOUT_CREATE_UPDATE_AFTER_BIND_POOL_BIT : 0;
+        setLayoutInfo.pNext = entry.usesUpdateAfterBind ? &bindingFlagsInfo : nullptr;
         setLayoutInfo.bindingCount = static_cast<Uint32>(bindings.size());
         setLayoutInfo.pBindings = bindings.data();
         VK_VERIFY(vkCreateDescriptorSetLayout(m_device, &setLayoutInfo, nullptr, &entry.descriptorSetLayout),
@@ -3054,6 +3119,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         auto& shaders = program.GetAttachedShaders();
         auto& spirv = program.GetGeneratedSpirv();
         Vector<Vector<Uint>> moduleSpirvs(spirv.size());
+        const Bool enableSpirvValidation = program.GetSpirvValidationEnabled();
+        if (enableSpirvValidation) {
+            MG_Util::ShaderTranspiler::ShaderCompiler::PrepareSpirvValidation();
+        }
 
         const ShaderStage fixupStage = PickClipFixupStage(shaders);
 
@@ -3099,7 +3168,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // stored as - which addresses [0,1] where the application addressed texels.
             {
                 Vector<Uint> rectLoweredSpirv;
-                if (MG_Util::ShaderTranspiler::ShaderCompiler::LowerRectImages(moduleSpirvs[i], rectLoweredSpirv) &&
+                if (MG_Util::ShaderTranspiler::ShaderCompiler::LowerRectImages(moduleSpirvs[i], rectLoweredSpirv, enableSpirvValidation) &&
                     !rectLoweredSpirv.empty()) {
                     moduleSpirvs[i] = Move(rectLoweredSpirv);
                 }
@@ -3112,7 +3181,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             {
                 Vector<Uint> invariantSpirv;
                 if (MG_Util::ShaderTranspiler::ShaderCompiler::DecoratePositionInvariantForVulkan(
-                        moduleSpirvs[i], invariantSpirv)) {
+                        moduleSpirvs[i], invariantSpirv, enableSpirvValidation)) {
                     moduleSpirvs[i] = std::move(invariantSpirv);
                 } else {
                     // The pass round-trips through SPIRV-Tools IR, so an unparseable module
@@ -3136,7 +3205,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 m_shaderDrawParametersEnabled) {
                 Vector<Uint> rebasedSpirv;
                 if (MG_Util::ShaderTranspiler::ShaderCompiler::RebaseInstanceIndexForVulkan(moduleSpirvs[i],
-                                                                                            rebasedSpirv)) {
+                                                                                            rebasedSpirv, enableSpirvValidation)) {
                     moduleSpirvs[i] = std::move(rebasedSpirv);
                 } else {
                     MGLOG_E("ProgramFactory: failed to rebase gl_InstanceID for program %u; "
@@ -3154,7 +3223,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 (flags & CompileOptionBit::ZeroBaseVertex)) {
                 Vector<Uint> zeroedSpirv;
                 if (MG_Util::ShaderTranspiler::ShaderCompiler::ZeroBaseVertexForVulkan(moduleSpirvs[i],
-                                                                                       zeroedSpirv)) {
+                                                                                       zeroedSpirv, enableSpirvValidation)) {
                     moduleSpirvs[i] = std::move(zeroedSpirv);
                 } else {
                     // Failing open keeps the native builtin, which is the pre-fix behavior:
@@ -3177,7 +3246,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             if (shaders[i] && shaders[i]->GetShaderStage() == ShaderStage::Vertex) {
                 Vector<Uint> packedSpirv;
                 const Bool packOk = MG_Util::ShaderTranspiler::ShaderCompiler::PackDoubleVertexInputsForVulkan(
-                    moduleSpirvs[i], packedSpirv);
+                    moduleSpirvs[i], packedSpirv, enableSpirvValidation);
                 MOBILEGL_ASSERT(packOk,
                                 "ProgramFactory: 64-bit vertex input packing failed for program %u; the "
                                 "vertex-input format and the shader input type now disagree",
@@ -3201,7 +3270,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             if (m_unformattedFloatStorageImagesEnabled) {
                 Vector<Uint> unformattedSpirv;
                 if (MG_Util::ShaderTranspiler::ShaderCompiler::UseUnformattedFloatStorageImagesForVulkan(
-                        moduleSpirvs[i], unformattedSpirv)) {
+                        moduleSpirvs[i], unformattedSpirv, enableSpirvValidation)) {
                     moduleSpirvs[i] = std::move(unformattedSpirv);
                 } else {
                     MGLOG_E("ProgramFactory: failed to make float storage images unformatted for program %u",
@@ -3222,7 +3291,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 #else
             // Final module the driver receives; also checked in the INFO-level CI/test
             // lanes, where the DEBUG gate above is compiled out.
-            if (MG_Util::ShaderTranspiler::ShaderCompiler::SpirvValidationEnabled()) {
+            if (enableSpirvValidation) {
                 ValidateTransformedSpirv(moduleSpv, shaders[i]->GetShaderStage(), program.GetExternalIndex());
             }
 #endif
@@ -3299,14 +3368,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 const VkDescriptorSetLayout descriptorSetLayout = it->second.descriptorSetLayout;
                 MGLOG_D("ProgramFactory::OnFrameBoundary: evicting idle program entry hash=0x%llx",
                         static_cast<unsigned long long>(hash));
-                // erase runs ~VkProgramObject (modules/layouts destroyed); notify after
-                // so an observer never observes a half-destroyed entry through a lookup.
-                // Observers only need the handle values to purge their keyed caches.
-                ++m_cacheStructureEpoch; // erase moves/kills entries: memoised pointers die
-                it = m_cache.erase(it);
+                // The observer destroys dependent pipelines and frees descriptor sets while
+                // this entry still owns its layout. Vulkan requires every descriptor set to be
+                // freed before its VkDescriptorSetLayout is destroyed.
                 if (m_evictionObserver != nullptr) {
                     m_evictionObserver->OnProgramEvicted(hash, descriptorSetLayout);
                 }
+                ++m_cacheStructureEpoch; // erase moves/kills entries: memoised pointers die
+                it = m_cache.erase(it);
             } else {
                 ++it;
             }
@@ -3440,7 +3509,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 #if MOBILEGL_LOG_ACTIVE_LEVEL <= MOBILEGL_LOG_LEVEL_DEBUG
         ValidateTransformedSpirv(spirv, ShaderStage::TessControl, 0);
 #else
-        if (MG_Util::ShaderTranspiler::ShaderCompiler::SpirvValidationEnabled()) {
+        if (m_enableSpirvValidation) {
+            MG_Util::ShaderTranspiler::ShaderCompiler::PrepareSpirvValidation();
             ValidateTransformedSpirv(spirv, ShaderStage::TessControl, 0);
         }
 #endif

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.h

@@ -76,6 +76,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         using CompileOptionFlags = Flags<CompileOptionBit>;
         using HashType = Uint64;
 
+        struct UpdateAfterBindLimits {
+            Bool enabled = false;
+            Uint32 maxPerStageSamplers = 0;
+            Uint32 maxPerStageUniformBuffers = 0;
+            Uint32 maxPerStageStorageBuffers = 0;
+            Uint32 maxPerStageSampledImages = 0;
+            Uint32 maxPerStageStorageImages = 0;
+            Uint32 maxPerStageResources = 0;
+            Uint32 maxSetSamplers = 0;
+            Uint32 maxSetUniformBuffers = 0;
+            Uint32 maxSetUniformBuffersDynamic = 0;
+            Uint32 maxSetStorageBuffers = 0;
+            Uint32 maxSetStorageBuffersDynamic = 0;
+            Uint32 maxSetSampledImages = 0;
+            Uint32 maxSetStorageImages = 0;
+        };
+
         struct VkProgramObject {
             static constexpr Uint32 kMaxVertexInputLocations = 32;
 
@@ -88,6 +105,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
             // Layout data (previously in separate VkProgramLayout)
             VkDescriptorSetLayout descriptorSetLayout = VK_NULL_HANDLE;
+            // True only when this layout passed every descriptor-indexing feature and
+            // update-after-bind limit gate at reflection time. It controls both the
+            // layout/binding flags and the pool class used by UniformManager.
+            Bool usesUpdateAfterBind = false;
             VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
             Vector<DescriptorBindingKind> bindingKinds;
             // The bindings this program actually declares, ascending. bindingKinds is sized to the
@@ -196,6 +217,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 // a pipeline failure would be reported against the wrong SPIR-V.
                 stageSpirvDigests = std::move(other.stageSpirvDigests);
                 descriptorSetLayout = other.descriptorSetLayout;
+                usesUpdateAfterBind = other.usesUpdateAfterBind;
                 pipelineLayout = other.pipelineLayout;
                 bindingKinds = std::move(other.bindingKinds);
                 activeBindings = std::move(other.activeBindings);
@@ -224,11 +246,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 fragmentInputComponentCount = other.fragmentInputComponentCount;
                 fragmentReplacesDepth = other.fragmentReplacesDepth;
                 readsBaseVertexBuiltin = other.readsBaseVertexBuiltin;
+                writesViewportIndexBuiltin = other.writesViewportIndexBuiltin;
                 needsPassthroughTessControl = other.needsPassthroughTessControl;
                 passthroughTessControlEmulatable = other.passthroughTessControlEmulatable;
                 lastUsedFrame = other.lastUsedFrame;
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
+                other.usesUpdateAfterBind = false;
                 other.pipelineLayout = VK_NULL_HANDLE;
                 other.hasStorageImages = false;
                 other.declinedDescriptors = false;
@@ -240,6 +264,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.fragmentInputComponentCount = 0;
                 other.fragmentReplacesDepth = false;
                 other.readsBaseVertexBuiltin = false;
+                other.writesViewportIndexBuiltin = false;
                 other.needsPassthroughTessControl = false;
                 other.passthroughTessControlEmulatable = false;
                 other.lastUsedFrame = 0;
@@ -254,6 +279,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 modules = std::move(other.modules);
                 stageSpirvDigests = std::move(other.stageSpirvDigests); // travels with `modules` - see the move ctor
                 descriptorSetLayout = other.descriptorSetLayout;
+                usesUpdateAfterBind = other.usesUpdateAfterBind;
                 pipelineLayout = other.pipelineLayout;
                 bindingKinds = std::move(other.bindingKinds);
                 activeBindings = std::move(other.activeBindings);
@@ -282,11 +308,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 fragmentInputComponentCount = other.fragmentInputComponentCount;
                 fragmentReplacesDepth = other.fragmentReplacesDepth;
                 readsBaseVertexBuiltin = other.readsBaseVertexBuiltin;
+                writesViewportIndexBuiltin = other.writesViewportIndexBuiltin;
                 needsPassthroughTessControl = other.needsPassthroughTessControl;
                 passthroughTessControlEmulatable = other.passthroughTessControlEmulatable;
                 lastUsedFrame = other.lastUsedFrame;
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
+                other.usesUpdateAfterBind = false;
                 other.pipelineLayout = VK_NULL_HANDLE;
                 other.hasStorageImages = false;
                 other.declinedDescriptors = false;
@@ -298,6 +326,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.fragmentInputComponentCount = 0;
                 other.fragmentReplacesDepth = false;
                 other.readsBaseVertexBuiltin = false;
+                other.writesViewportIndexBuiltin = false;
                 other.needsPassthroughTessControl = false;
                 other.passthroughTessControlEmulatable = false;
                 other.lastUsedFrame = 0;
@@ -343,12 +372,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             virtual void OnProgramEvicted(HashType programHash, VkDescriptorSetLayout descriptorSetLayout) = 0;
         };
 
-        explicit ProgramFactory(VkDevice device, const VulkanRendererConfig& config, Uint32 maxBindings = 16,
-                                Bool shaderDrawParametersEnabled = false,
-                                Bool unformattedFloatStorageImagesEnabled = false)
+        explicit ProgramFactory(VkDevice device, const VulkanRendererConfig& config, Uint32 maxBindings,
+                                Bool shaderDrawParametersEnabled,
+                                Bool unformattedFloatStorageImagesEnabled,
+                                Bool enableSpirvValidation,
+                                UpdateAfterBindLimits updateAfterBindLimits)
             : m_device(device), m_maxBindings(maxBindings), m_config(config),
               m_shaderDrawParametersEnabled(shaderDrawParametersEnabled),
-              m_unformattedFloatStorageImagesEnabled(unformattedFloatStorageImagesEnabled) {
+              m_unformattedFloatStorageImagesEnabled(unformattedFloatStorageImagesEnabled),
+              m_enableSpirvValidation(enableSpirvValidation),
+              m_updateAfterBindLimits(updateAfterBindLimits) {
             VkProgramObject::s_device = device;
         }
         // Destroys the pass-through tessellation control modules. Runs while the device is
@@ -471,6 +504,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // True only when the logical device enabled both
         // shaderStorageImageReadWithoutFormat and shaderStorageImageWriteWithoutFormat.
         Bool m_unformattedFloatStorageImagesEnabled = false;
+        // Startup snapshot used only by internally synthesized shader modules, which do not
+        // originate from a ProgramLinkTask.
+        Bool m_enableSpirvValidation = false;
+        // Device feature and limit gate resolved before vkCreateDevice. Keeping it in
+        // the factory lets each reflected layout choose ordinary descriptors when its
+        // own counts would exceed the update-after-bind budget.
+        UpdateAfterBindLimits m_updateAfterBindLimits{};
         // See SetDefaultFramebufferHeight. 0 means "not known yet"; the FragCoordYFlip bit is
         // never set before the swapchain exists, so no variant can be compiled against it.
         Uint32 m_defaultFramebufferHeight = 0;

MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp

@@ -156,13 +156,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             frame.descriptorPools.clear();
 
             VkDescriptorPool initialPool = VK_NULL_HANDLE;
-            if (!CreateDescriptorPool(m_setsPerFrame, initialPool)) {
+            if (!CreateDescriptorPool(m_setsPerFrame, false, initialPool)) {
                 MGLOG_E_ONCE("UniformDescriptorBinder::Initialize failed: cannot create frame descriptor pool %u",
                         frameIndex);
                 Shutdown();
                 return false;
             }
-            frame.descriptorPools.push_back({initialPool, m_setsPerFrame, 0});
+            frame.descriptorPools.push_back({initialPool, m_setsPerFrame, 0, false});
             MGLOG_D("UniformDescriptorBinder: frame %u descriptor pool created (maxSets=%u)", frameIndex,
                     m_setsPerFrame);
         }
@@ -305,7 +305,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // texture/sampler resolution, completeness probe, sync, layout handling, sampler
         // and view lookups - would recompute the identical descriptor.
         if (trustUnchangedHint && descriptorMemoUsable && binding < m_samplerResolveMemo.size() &&
-            m_samplerResolveMemo[binding].infoValid) {
+            m_samplerResolveMemo[binding].infoValid &&
+            m_samplerResolveMemo[binding].infoProgramLifetimeId == program.GetLifetimeId()) {
             outImageInfo = m_samplerResolveMemo[binding].info;
             return true;
         }
@@ -504,6 +505,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         if (binding < m_samplerResolveMemo.size()) {
             if (descriptorMemoUsable) {
                 m_samplerResolveMemo[binding].info = outImageInfo;
+                m_samplerResolveMemo[binding].infoProgramLifetimeId = program.GetLifetimeId();
                 m_samplerResolveMemo[binding].infoValid = true;
             } else {
                 // An arrayed binding publishes nothing here, and clears what a previous program
@@ -1388,7 +1390,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return true;
     }
 
-    Bool UniformManager::CreateDescriptorPool(Uint32 maxSets, VkDescriptorPool& outPool) const {
+    Bool UniformManager::CreateDescriptorPool(Uint32 maxSets, Bool updateAfterBind, VkDescriptorPool& outPool) const {
         outPool = VK_NULL_HANDLE;
         if (m_device == VK_NULL_HANDLE || maxSets == 0 || m_maxBindings == 0) {
             return false;
@@ -1431,7 +1433,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // (OnDescriptorSetLayoutDestroyed) so program churn recycles pool capacity.
         // The cost is on set allocation only, which happens when a layout's per-frame
         // cache grows - never on the per-draw reuse path.
-        poolInfo.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT;
+        poolInfo.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT |
+                         (updateAfterBind ? VK_DESCRIPTOR_POOL_CREATE_UPDATE_AFTER_BIND_BIT : 0);
         poolInfo.maxSets = maxSets;
         poolInfo.poolSizeCount = static_cast<Uint32>(std::size(poolSizes));
         poolInfo.pPoolSizes = poolSizes;
@@ -1445,24 +1448,28 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return true;
     }
 
-    Bool UniformManager::GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex) {
+    Bool UniformManager::GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex, Bool updateAfterBind) {
         if (frame.descriptorPools.empty()) {
             return false;
         }
 
-        const auto& currentBucket = frame.descriptorPools[frame.activeDescriptorPoolIndex];
-        const Uint32 currentMaxSets = std::max<Uint32>(1, currentBucket.maxSets);
+        const auto matchingBucket = std::find_if(
+            frame.descriptorPools.begin(), frame.descriptorPools.end(),
+            [updateAfterBind](const DescriptorPoolBucket& candidate) { return candidate.updateAfterBind == updateAfterBind; });
+        const Uint32 currentMaxSets = matchingBucket != frame.descriptorPools.end()
+                                          ? std::max<Uint32>(1, matchingBucket->maxSets)
+                                          : m_setsPerFrame;
         const Uint32 grownMaxSets = currentMaxSets <= (std::numeric_limits<Uint32>::max() / 2) ? (currentMaxSets * 2)
                                                                                                  : currentMaxSets;
 
         VkDescriptorPool grownPool = VK_NULL_HANDLE;
-        if (!CreateDescriptorPool(grownMaxSets, grownPool)) {
+        if (!CreateDescriptorPool(grownMaxSets, updateAfterBind, grownPool)) {
             MGLOG_E_ONCE("UniformDescriptorBinder::GrowFrameDescriptorPool failed: cannot create grown pool (%u -> %u sets)",
                     currentMaxSets, grownMaxSets);
             return false;
         }
 
-        frame.descriptorPools.push_back({grownPool, grownMaxSets, 0});
+        frame.descriptorPools.push_back({grownPool, grownMaxSets, 0, updateAfterBind});
         frame.activeDescriptorPoolIndex = static_cast<Uint32>(frame.descriptorPools.size() - 1);
         MGLOG_D(
             "UniformDescriptorBinder: frame %u descriptor pool exhausted, grew pool (%u -> %u sets), poolCount=%zu",
@@ -1472,14 +1479,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
     VkResult UniformManager::AllocateDescriptorSetsFromActivePool(Uint32 frameIndex, const ProgramFactory::VkProgramObject& programObj, VkDescriptorSet& outDescriptorSet) {
         auto& frame = m_frames[frameIndex];
-        if (frame.activeDescriptorPoolIndex >= frame.descriptorPools.size()) {
-            frame.activeDescriptorPoolIndex = 0;
-        }
-        if (frame.descriptorPools[frame.activeDescriptorPoolIndex].allocatedSets >=
-            frame.descriptorPools[frame.activeDescriptorPoolIndex].maxSets) {
+        const Bool updateAfterBind = programObj.usesUpdateAfterBind;
+        if (frame.activeDescriptorPoolIndex >= frame.descriptorPools.size() ||
+            frame.descriptorPools[frame.activeDescriptorPoolIndex].updateAfterBind != updateAfterBind ||
+            frame.descriptorPools[frame.activeDescriptorPoolIndex].allocatedSets >=
+                frame.descriptorPools[frame.activeDescriptorPoolIndex].maxSets) {
             const auto availableBucket = std::find_if(
                 frame.descriptorPools.begin(), frame.descriptorPools.end(),
-                [](const DescriptorPoolBucket& candidate) { return candidate.allocatedSets < candidate.maxSets; });
+                [updateAfterBind](const DescriptorPoolBucket& candidate) {
+                    return candidate.updateAfterBind == updateAfterBind && candidate.allocatedSets < candidate.maxSets;
+                });
             if (availableBucket == frame.descriptorPools.end()) {
                 outDescriptorSet = VK_NULL_HANDLE;
                 return VK_ERROR_OUT_OF_POOL_MEMORY;
@@ -1515,7 +1524,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         } else {
             VkResult allocResult = AllocateDescriptorSetsFromActivePool(frameIndex, programObj, outDescriptorSet);
             if (allocResult == VK_ERROR_OUT_OF_POOL_MEMORY || allocResult == VK_ERROR_FRAGMENTED_POOL) {
-                if (!GrowFrameDescriptorPool(frame, frameIndex)) {
+                if (!GrowFrameDescriptorPool(frame, frameIndex, programObj.usesUpdateAfterBind)) {
                     MGLOG_E_ONCE("UniformDescriptorBinder::AcquireDescriptorSet failed: descriptor pool growth failed");
                     return allocResult;
                 }

MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.h

@@ -114,6 +114,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             VkDescriptorPool handle = VK_NULL_HANDLE;
             Uint32 maxSets = 0;
             Uint32 allocatedSets = 0;
+            Bool updateAfterBind = false;
         };
 
         // A cached descriptor set together with the pool it was allocated from, so a
@@ -223,8 +224,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         void BindDescriptorSetDeduped(VkCommandBuffer commandBuffer, VkPipelineBindPoint bindPoint,
                                       VkPipelineLayout pipelineLayout, VkDescriptorSet descriptorSet,
                                       const Vector<Uint32>& dynamicOffsets);
-        Bool CreateDescriptorPool(Uint32 maxSets, VkDescriptorPool& outPool) const;
-        Bool GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex);
+        Bool CreateDescriptorPool(Uint32 maxSets, Bool updateAfterBind, VkDescriptorPool& outPool) const;
+        Bool GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex, Bool updateAfterBind);
         VkResult AllocateDescriptorSetsFromActivePool(
             Uint32 frameIndex, const ProgramFactory::VkProgramObject& programObj, VkDescriptorSet& outDescriptorSet);
         VkResult AcquireDescriptorSet(Uint32 frameIndex,
@@ -341,8 +342,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // lifetime id, so a freed-and-reallocated sampler or texture at the same heap address
         // always gets a fresh id and misses (a raw pointer would false-hit that ABA) - so a
         // stale guess can only miss and fall through to the hash, never resolve wrong. Still
-        // reset each frame alongside the descriptor-set cache. Indexed by binding.
+        // reset each frame alongside the descriptor-set cache. Indexed by binding, but the
+        // whole-descriptor entry is additionally keyed by program lifetime: Vulkan binding
+        // numbers are layout-local and unrelated programs routinely reuse binding 0/1.
         struct SamplerResolveMemo {
+            Uint64 infoProgramLifetimeId = 0;
             Uint64 samplerLifetimeId = 0;
             Uint64 textureLifetimeId = 0;
             VkSampler sampler = VK_NULL_HANDLE;

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp

@@ -238,11 +238,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // nothing else across all of gl33.
     //
     // The mapping below is derived from - and at full extent exactly reproduces - the pixel
-    // mapping RemapDefaultFboReadbackToGLOrientation has always used:
+    // mapping VulkanRenderer::RemapDefaultFramebufferReadback uses:
     //     identity : image(x, H-1-y)      -> flip Y
     //     180      : image(W-1-x, y)      -> mirror X (the rotation already flips the rows)
-    // Quarter turns swap the axes; nothing in this renderer models that (the readback declines to
-    // remap them and the viewport path only rescales), so they are left exactly as they were.
+    // Quarter turns swap the axes and are handled by MapDefaultFramebufferReadbackRect rather than
+    // this same-axis helper.
     struct DefaultFramebufferRectMapping {
         Bool flipY = false;
         Bool mirrorX = false;
@@ -1480,7 +1480,8 @@ void main() {
             const char* label = nullptr;
         };
 
-        static Uint32 ComputeMaxProgramBindings(const VkPhysicalDeviceProperties& properties) {
+        static Uint32 ComputeMaxProgramBindings(const VkPhysicalDeviceProperties& properties,
+                                                const ProgramFactory::UpdateAfterBindLimits& updateAfterBindLimits) {
             const auto& limits = properties.limits;
             static constexpr Uint32 kMinProgramBindings = 16;
             static constexpr Uint32 kMaxProgramBindingsCap = 256;
@@ -1495,6 +1496,21 @@ void main() {
             maxBindings = std::min(maxBindings, maxCombinedImageSamplers);
             maxBindings = std::min(maxBindings, maxSampledImages + maxDynamicUniformBuffers);
 
+            if (updateAfterBindLimits.enabled) {
+                const Uint32 updateAfterBindSamplers = std::min(updateAfterBindLimits.maxPerStageSamplers,
+                                                                 updateAfterBindLimits.maxSetSamplers);
+                const Uint32 updateAfterBindSampledImages = std::min(updateAfterBindLimits.maxPerStageSampledImages,
+                                                                      updateAfterBindLimits.maxSetSampledImages);
+                const Uint32 updateAfterBindDynamicUniformBuffers =
+                    std::min(updateAfterBindLimits.maxPerStageUniformBuffers,
+                             updateAfterBindLimits.maxSetUniformBuffersDynamic);
+                Uint32 updateAfterBindBindings = updateAfterBindLimits.maxPerStageResources;
+                updateAfterBindBindings = std::min(updateAfterBindBindings, updateAfterBindSamplers);
+                updateAfterBindBindings =
+                    std::min(updateAfterBindBindings, updateAfterBindSampledImages + updateAfterBindDynamicUniformBuffers);
+                maxBindings = std::max(maxBindings, updateAfterBindBindings);
+            }
+
             maxBindings = std::max(kMinProgramBindings, maxBindings);
             maxBindings = std::min(kMaxProgramBindingsCap, maxBindings);
             return maxBindings;
@@ -2151,47 +2167,6 @@ void main() {
             return static_cast<Uint8>(value * 255.0f + 0.5f);
         }
 
-        // Re-order the copied BLOCK - not the whole image - from the default framebuffer's stored
-        // orientation into GL's. The caller has already aimed the copy at the right place with
-        // MapDefaultFramebufferRectAxis, so what arrives here is exactly the requested
-        // rectWidth x rectHeight rect, and all that is left is the order of rows (identity) or of
-        // columns (180) WITHIN it.
-        //
-        // This used to iterate the full swapchain extent and index both sides with that stride,
-        // which is why its caller could only use it on an exact full-extent read - and why every
-        // partial glReadPixels of the default framebuffer came back in Vulkan row order. Only
-        // identity/180 share the swapchain extent with the default framebuffer; 90/270 swap
-        // extents and are still declined.
-        static Bool RemapDefaultFboReadbackToGLOrientation(const Uint8* rawPixels,
-                                                            Uint32 rectWidth,
-                                                            Uint32 rectHeight,
-                                                            VkSurfaceTransformFlagBitsKHR preTransform,
-                                                            SizeT texelSize,
-                                                            Uint8* outPixels) {
-            if (IsQuarterTurnPreTransform(preTransform)) {
-                return false;
-            }
-            if (rectWidth == 0 || rectHeight == 0 || texelSize == 0) {
-                return false;
-            }
-            const DefaultFramebufferRectMapping mapping = GetDefaultFramebufferRectMapping(preTransform);
-            const SizeT rowBytes = static_cast<SizeT>(rectWidth) * texelSize;
-            for (Uint32 outY = 0; outY < rectHeight; ++outY) {
-                const Uint32 srcY = mapping.flipY ? (rectHeight - 1 - outY) : outY;
-                const Uint8* srcRow = rawPixels + static_cast<SizeT>(srcY) * rowBytes;
-                Uint8* dstRow = outPixels + static_cast<SizeT>(outY) * rowBytes;
-                if (!mapping.mirrorX) {
-                    Memcpy(dstRow, srcRow, rowBytes);
-                    continue;
-                }
-                for (Uint32 outX = 0; outX < rectWidth; ++outX) {
-                    Memcpy(dstRow + static_cast<SizeT>(outX) * texelSize,
-                           srcRow + static_cast<SizeT>(rectWidth - 1 - outX) * texelSize, texelSize);
-                }
-            }
-            return true;
-        }
-
         static SizeT AlignPixelRow(SizeT rowBytes, Int alignment) {
             const SizeT resolvedAlignment = static_cast<SizeT>(std::max(alignment, 1));
             return (rowBytes + resolvedAlignment - 1) & ~(resolvedAlignment - 1);
@@ -2738,6 +2713,95 @@ void main() {
         return formatInfo.texel_block_size;
     }
 
+    Bool VulkanRenderer::MapDefaultFramebufferReadbackRect(
+            GLint x, GLint y, GLsizei width, GLsizei height, VkExtent2D imageExtent,
+            VkSurfaceTransformFlagBitsKHR preTransform, VkOffset2D* imageOffset,
+            VkExtent2D* imageCopyExtent) {
+        if (width <= 0 || height <= 0 || imageOffset == nullptr || imageCopyExtent == nullptr) {
+            return false;
+        }
+
+        const Int imageWidth = static_cast<Int>(imageExtent.width);
+        const Int imageHeight = static_cast<Int>(imageExtent.height);
+        Int mappedX = x;
+        Int mappedY = y;
+        Uint32 mappedWidth = static_cast<Uint32>(width);
+        Uint32 mappedHeight = static_cast<Uint32>(height);
+
+        // InsertPositionFixup first flips GL Y and then applies the surface transform. In pixel
+        // coordinates that gives these half-open rectangle mappings into the stored image:
+        //   identity: (x, H-y-h), 90: (y, x), 180: (W-x-w, y), 270: (H-y-h, W-x-w).
+        // Quarter turns also transpose the copied block's extent.
+        switch (preTransform) {
+        case VK_SURFACE_TRANSFORM_ROTATE_90_BIT_KHR:
+            mappedX = y;
+            mappedY = x;
+            mappedWidth = static_cast<Uint32>(height);
+            mappedHeight = static_cast<Uint32>(width);
+            break;
+        case VK_SURFACE_TRANSFORM_ROTATE_180_BIT_KHR:
+            mappedX = imageWidth - x - width;
+            mappedY = y;
+            break;
+        case VK_SURFACE_TRANSFORM_ROTATE_270_BIT_KHR:
+            mappedX = imageWidth - y - height;
+            mappedY = imageHeight - x - width;
+            mappedWidth = static_cast<Uint32>(height);
+            mappedHeight = static_cast<Uint32>(width);
+            break;
+        default:
+            mappedY = imageHeight - y - height;
+            break;
+        }
+
+        if (mappedX < 0 || mappedY < 0 || mappedWidth > imageExtent.width ||
+            mappedHeight > imageExtent.height ||
+            static_cast<Uint64>(mappedX) + mappedWidth > imageExtent.width ||
+            static_cast<Uint64>(mappedY) + mappedHeight > imageExtent.height) {
+            return false;
+        }
+        *imageOffset = {mappedX, mappedY};
+        *imageCopyExtent = {mappedWidth, mappedHeight};
+        return true;
+    }
+
+    Bool VulkanRenderer::RemapDefaultFramebufferReadback(
+            const Uint8* rawPixels, Uint32 logicalWidth, Uint32 logicalHeight,
+            VkSurfaceTransformFlagBitsKHR preTransform, SizeT texelSize, Uint8* outPixels) {
+        if (rawPixels == nullptr || outPixels == nullptr || logicalWidth == 0 || logicalHeight == 0 ||
+            texelSize == 0) {
+            return false;
+        }
+
+        const Uint32 rawWidth = IsQuarterTurnPreTransform(preTransform) ? logicalHeight : logicalWidth;
+        for (Uint32 outY = 0; outY < logicalHeight; ++outY) {
+            for (Uint32 outX = 0; outX < logicalWidth; ++outX) {
+                Uint32 srcX = outX;
+                Uint32 srcY = outY;
+                switch (preTransform) {
+                case VK_SURFACE_TRANSFORM_ROTATE_90_BIT_KHR:
+                    srcX = outY;
+                    srcY = outX;
+                    break;
+                case VK_SURFACE_TRANSFORM_ROTATE_180_BIT_KHR:
+                    srcX = logicalWidth - 1 - outX;
+                    break;
+                case VK_SURFACE_TRANSFORM_ROTATE_270_BIT_KHR:
+                    srcX = logicalHeight - 1 - outY;
+                    srcY = logicalWidth - 1 - outX;
+                    break;
+                default:
+                    srcY = logicalHeight - 1 - outY;
+                    break;
+                }
+                Memcpy(outPixels + (static_cast<SizeT>(outY) * logicalWidth + outX) * texelSize,
+                       rawPixels + (static_cast<SizeT>(srcY) * rawWidth + srcX) * texelSize,
+                       texelSize);
+            }
+        }
+        return true;
+    }
+
     Bool VulkanRenderer::ConvertReadbackPixels(const Uint8* sourcePixels, VkFormat sourceFormat,
                                                GLsizei width, GLsizei height, GLenum destinationFormat,
                                                GLenum destinationType, SizeT destinationRowStride,
@@ -2962,7 +3026,7 @@ void main() {
         succeeded = m_renderPassManager->Initialize();
         MOBILEGL_ASSERT(succeeded, "VkRenderPassManager initialization failed.");
 
-        const Uint32 maxProgramBindings = ComputeMaxProgramBindings(m_physicalDevice.properties);
+        const Uint32 maxProgramBindings = ComputeMaxProgramBindings(m_physicalDevice.properties, m_updateAfterBindLimits);
         MGLOG_I("DirectVulkan: using %u program descriptor bindings", maxProgramBindings);
         if (IsPowerVRDevice(m_physicalDevice.properties)) {
             m_config.DisablePipelineCache = true;
@@ -2996,7 +3060,9 @@ void main() {
         }
         m_programFactory = MakeUnique<ProgramFactory>(m_device, m_config, maxProgramBindings,
                                                       m_shaderDrawParametersFeatureEnabled,
-                                                      m_unformattedFloatStorageImagesEnabled);
+                                                      m_unformattedFloatStorageImagesEnabled,
+                                                      MG_Config::Features.EnableSpirvValidation,
+                                                      m_updateAfterBindLimits);
         MOBILEGL_ASSERT(m_programFactory != nullptr, "ProgramFactory creation failed.");
         // The swapchain already exists at this point (Initialize creates it first), so seed the
         // height the factory could not be told about from CreateSwapchain.
@@ -9138,19 +9204,18 @@ void main() {
         // The GL rect, aimed at the default framebuffer's stored orientation. Using the GL y
         // verbatim copied rows [y, y+h) counted from the TOP of the image, i.e. the wrong band for
         // every read that was not full-height.
-        Int32 copyOffsetX = x;
-        Int32 copyOffsetY = y;
+        VkOffset2D copyOffset{x, y};
+        VkExtent2D copyExtent{static_cast<Uint32>(width), static_cast<Uint32>(height)};
         if (readIsDefaultFbo) {
             const VkExtent2D defaultFboExtent = m_swapchainObject.GetExtent();
-            const DefaultFramebufferRectMapping mapping =
-                GetDefaultFramebufferRectMapping(m_swapchainObject.GetPreTransform());
-            copyOffsetX = MapDefaultFramebufferRectAxis(x, width, static_cast<Int>(defaultFboExtent.width),
-                                                        mapping.mirrorX);
-            copyOffsetY = MapDefaultFramebufferRectAxis(y, height, static_cast<Int>(defaultFboExtent.height),
-                                                        mapping.flipY);
+            const Bool mapped = MapDefaultFramebufferReadbackRect(
+                x, y, width, height, defaultFboExtent, m_swapchainObject.GetPreTransform(), &copyOffset,
+                &copyExtent);
+            MOBILEGL_ASSERT(mapped, "ReadPixels: default framebuffer read rectangle is out of bounds");
+            if (!mapped) return;
         }
-        copyRegion.imageOffset = {copyOffsetX, copyOffsetY, static_cast<Int32>(srcBinding.depthOffset)};
-        copyRegion.imageExtent = {static_cast<Uint32>(width), static_cast<Uint32>(height), 1};
+        copyRegion.imageOffset = {copyOffset.x, copyOffset.y, static_cast<Int32>(srcBinding.depthOffset)};
+        copyRegion.imageExtent = {copyExtent.width, copyExtent.height, 1};
         vkCmdCopyImageToBuffer(frame.commandBuffer, srcBinding.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
                                readback.GetHandle(), 1, &copyRegion);
 
@@ -9192,16 +9257,14 @@ void main() {
             // already aimed with the same mapping. The gate is exactly what made every partial
             // read of the default framebuffer come back in Vulkan row order.
             Vector<Uint8> remapped(static_cast<SizeT>(width) * static_cast<SizeT>(height) * sourceTexelSize);
-            if (RemapDefaultFboReadbackToGLOrientation(mapped, static_cast<Uint32>(width),
-                                                       static_cast<Uint32>(height), preTransform, sourceTexelSize,
-                                                       remapped.data())) {
+            if (RemapDefaultFramebufferReadback(mapped, static_cast<Uint32>(width),
+                                                static_cast<Uint32>(height), preTransform, sourceTexelSize,
+                                                remapped.data())) {
                 PackReadbackToClientOrPbo(remapped.data(), srcFormat, width, height, 1, format, type, pixels,
                                           /*applyPackImageParams=*/false, /*applyReadColorClamp=*/true);
                 return;
             }
-            // Only a quarter-turn pre-transform reaches this, and nothing in this renderer models
-            // one. MGLOG_I because the INFO builds are the ones that run conformance.
-            MGLOG_D("DirectVulkan::ReadPixels: default-FBO remap declined (w=%d h=%d preTransform=%d); falling back "
+            MGLOG_D("DirectVulkan::ReadPixels: default-FBO remap failed (w=%d h=%d preTransform=%d); falling back "
                     "to raw readback",
                     width, height, static_cast<Int>(preTransform));
         }
@@ -9582,16 +9645,15 @@ void main() {
         // The swapchain's depth/stencil image is stored display-side-up like its colour twin, so
         // the GL rect has to be mapped into that space before the copy and the copied rows
         // re-oriented afterwards - the same two halves the colour ReadPixels path applies.
-        Int32 copyOffsetX = x;
-        Int32 copyOffsetY = y;
+        VkOffset2D copyOffset{x, y};
+        VkExtent2D copyExtent{static_cast<Uint32>(width), static_cast<Uint32>(height)};
         if (defaultFramebufferOrientation) {
             const VkExtent2D defaultFboExtent = m_swapchainObject.GetExtent();
-            const DefaultFramebufferRectMapping mapping =
-                GetDefaultFramebufferRectMapping(m_swapchainObject.GetPreTransform());
-            copyOffsetX = MapDefaultFramebufferRectAxis(x, width, static_cast<Int>(defaultFboExtent.width),
-                                                        mapping.mirrorX);
-            copyOffsetY = MapDefaultFramebufferRectAxis(y, height, static_cast<Int>(defaultFboExtent.height),
-                                                        mapping.flipY);
+            const Bool mapped = MapDefaultFramebufferReadbackRect(
+                x, y, width, height, defaultFboExtent, m_swapchainObject.GetPreTransform(), &copyOffset,
+                &copyExtent);
+            MOBILEGL_ASSERT(mapped, "ReadDepthStencilPixels: default framebuffer read rectangle is out of bounds");
+            if (!mapped) return;
         }
 
         VkBufferImageCopy regions[2]{};
@@ -9603,8 +9665,8 @@ void main() {
             region.imageSubresource.mipLevel = mipLevel;
             region.imageSubresource.baseArrayLayer = baseArrayLayer;
             region.imageSubresource.layerCount = 1;
-            region.imageOffset = {copyOffsetX, copyOffsetY, 0};
-            region.imageExtent = {static_cast<Uint32>(width), static_cast<Uint32>(height), 1};
+            region.imageOffset = {copyOffset.x, copyOffset.y, 0};
+            region.imageExtent = {copyExtent.width, copyExtent.height, 1};
         }
         if (wantStencil) {
             auto& region = regions[regionCount++];
@@ -9613,8 +9675,8 @@ void main() {
             region.imageSubresource.mipLevel = mipLevel;
             region.imageSubresource.baseArrayLayer = baseArrayLayer;
             region.imageSubresource.layerCount = 1;
-            region.imageOffset = {copyOffsetX, copyOffsetY, 0};
-            region.imageExtent = {static_cast<Uint32>(width), static_cast<Uint32>(height), 1};
+            region.imageOffset = {copyOffset.x, copyOffset.y, 0};
+            region.imageExtent = {copyExtent.width, copyExtent.height, 1};
         }
         vkCmdCopyImageToBuffer(frame.commandBuffer, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, readback.GetHandle(),
                                regionCount, regions);
@@ -9648,23 +9710,21 @@ void main() {
             Bool remapped = true;
             if (wantDepth && depthCopyBytes > 0) {
                 remappedDepth.resize(pixelCount * depthCopyBytes);
-                remapped = RemapDefaultFboReadbackToGLOrientation(depthSrc, static_cast<Uint32>(width),
-                                                                  static_cast<Uint32>(height), preTransform,
-                                                                  depthCopyBytes, remappedDepth.data());
+                remapped = RemapDefaultFramebufferReadback(depthSrc, static_cast<Uint32>(width),
+                                                           static_cast<Uint32>(height), preTransform,
+                                                           depthCopyBytes, remappedDepth.data());
             }
             if (remapped && wantStencil) {
                 remappedStencil.resize(pixelCount);
-                remapped = RemapDefaultFboReadbackToGLOrientation(stencilSrc, static_cast<Uint32>(width),
-                                                                  static_cast<Uint32>(height), preTransform, 1,
-                                                                  remappedStencil.data());
+                remapped = RemapDefaultFramebufferReadback(stencilSrc, static_cast<Uint32>(width),
+                                                           static_cast<Uint32>(height), preTransform, 1,
+                                                           remappedStencil.data());
             }
             if (remapped) {
                 if (!remappedDepth.empty()) depthSrc = remappedDepth.data();
                 if (!remappedStencil.empty()) stencilSrc = remappedStencil.data();
             } else {
-                // Only a quarter-turn pre-transform reaches this, and nothing in this renderer
-                // models one. MGLOG_I because the INFO builds are the ones that run conformance.
-                MGLOG_D("DirectVulkan::ReadDepthStencilPixels: default-FBO remap declined (w=%d h=%d "
+                MGLOG_D("DirectVulkan::ReadDepthStencilPixels: default-FBO remap failed (w=%d h=%d "
                         "preTransform=%d); falling back to raw readback",
                         width, height, static_cast<Int>(preTransform));
             }
@@ -11857,6 +11917,17 @@ void main() {
     }
 
     void VulkanRenderer::CreateInstance() {
+#if defined(VK_USE_PLATFORM_METAL_EXT)
+        // MoltenVK snapshots its configuration when the loader first discovers the ICD. Set
+        // this before instance-extension enumeration, while preserving an explicit user value.
+        if (std::getenv("MVK_CONFIG_USE_METAL_ARGUMENT_BUFFERS") == nullptr) {
+            if (::setenv("MVK_CONFIG_USE_METAL_ARGUMENT_BUFFERS", "1", 0) == 0) {
+                MGLOG_I("MoltenVK: enabling Metal argument buffers");
+            } else {
+                MGLOG_W("MoltenVK: could not enable Metal argument buffers before ICD discovery");
+            }
+        }
+#endif
         m_extensions = EnumerateInstanceExtensions();
         MGLOG_I("Got %d Vulkan instance extensions: ", m_extensions.size());
         for (auto& extension : m_extensions) {
@@ -11998,17 +12069,19 @@ void main() {
 
         auto debugMessengerCreateInfo = PopulateDebugMessengerCreateInfo();
         // Layers
+        const void* instanceCreatePNext = nullptr;
         if (m_validationLayersEnabled) {
             MGLOG_I("Enabling validation layer...");
             instanceInfo.enabledLayerCount = static_cast<uint32_t>(std::size(s_validationLayerNames));
             instanceInfo.ppEnabledLayerNames = s_validationLayerNames;
             // Chaining the messenger create-info is only legal with the extension on.
-            instanceInfo.pNext = debugUtilsAvailable ? &debugMessengerCreateInfo : nullptr;
+            instanceCreatePNext = debugUtilsAvailable ? &debugMessengerCreateInfo : nullptr;
         } else {
             instanceInfo.enabledLayerCount = 0;
-            instanceInfo.pNext = nullptr;
         }
 
+        instanceInfo.pNext = instanceCreatePNext;
+
         VK_VERIFY(vkCreateInstance(&instanceInfo, nullptr, &m_instance), "vkCreateInstance failed");
 
         if (debugUtilsAvailable) {
@@ -12434,6 +12507,75 @@ void main() {
                 vkGetInstanceProcAddr(m_instance, "vkGetPhysicalDeviceFeatures2KHR"));
         }
 
+        m_updateAfterBindLimits = {};
+        VkPhysicalDeviceDescriptorIndexingFeatures descriptorIndexingFeatures{};
+        descriptorIndexingFeatures.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES;
+        VkPhysicalDeviceDescriptorIndexingProperties descriptorIndexingProperties{};
+        descriptorIndexingProperties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES;
+        const Bool descriptorIndexingCore = m_physicalDevice.properties.apiVersion >= VK_API_VERSION_1_2;
+        const Bool descriptorIndexingExtension =
+            IsExtensionSupported(availableExtensions, VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME);
+        auto getPhysicalDeviceProperties2 = reinterpret_cast<PFN_vkGetPhysicalDeviceProperties2>(
+            vkGetInstanceProcAddr(m_instance, "vkGetPhysicalDeviceProperties2"));
+        if (getPhysicalDeviceProperties2 == nullptr) {
+            getPhysicalDeviceProperties2 = reinterpret_cast<PFN_vkGetPhysicalDeviceProperties2>(
+                vkGetInstanceProcAddr(m_instance, "vkGetPhysicalDeviceProperties2KHR"));
+        }
+        if ((descriptorIndexingCore || descriptorIndexingExtension) && getPhysicalDeviceFeatures2 != nullptr &&
+            getPhysicalDeviceProperties2 != nullptr) {
+            VkPhysicalDeviceFeatures2 featureQuery{};
+            featureQuery.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
+            featureQuery.pNext = &descriptorIndexingFeatures;
+            getPhysicalDeviceFeatures2(m_physicalDevice.handle, &featureQuery);
+            VkPhysicalDeviceProperties2 propertyQuery{};
+            propertyQuery.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
+            propertyQuery.pNext = &descriptorIndexingProperties;
+            getPhysicalDeviceProperties2(m_physicalDevice.handle, &propertyQuery);
+
+            // This renderer emits every descriptor category listed below, including
+            // dynamic UBOs and combined image samplers. Do not enable a partial
+            // descriptor-indexing contract: it would make a later reflected program
+            // fail in the driver instead of choosing its ordinary descriptor layout.
+            const Bool allUpdateAfterBindFeatures =
+                descriptorIndexingFeatures.descriptorBindingUniformBufferUpdateAfterBind == VK_TRUE &&
+                descriptorIndexingFeatures.descriptorBindingSampledImageUpdateAfterBind == VK_TRUE &&
+                descriptorIndexingFeatures.descriptorBindingStorageImageUpdateAfterBind == VK_TRUE &&
+                descriptorIndexingFeatures.descriptorBindingStorageBufferUpdateAfterBind == VK_TRUE &&
+                descriptorIndexingFeatures.descriptorBindingUniformTexelBufferUpdateAfterBind == VK_TRUE &&
+                descriptorIndexingFeatures.descriptorBindingStorageTexelBufferUpdateAfterBind == VK_TRUE &&
+                (!deviceFeatures.robustBufferAccess || descriptorIndexingProperties.robustBufferAccessUpdateAfterBind);
+            if (allUpdateAfterBindFeatures) {
+                if (!descriptorIndexingCore && !IsExtensionAlreadyEnabled(
+                                                   enabledDeviceExtensions,
+                                                   VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME)) {
+                    enabledDeviceExtensions.push_back(VK_EXT_DESCRIPTOR_INDEXING_EXTENSION_NAME);
+                }
+                descriptorIndexingFeatures.pNext = const_cast<void*>(deviceCreateInfo.pNext);
+                deviceCreateInfo.pNext = &descriptorIndexingFeatures;
+                m_updateAfterBindLimits = {
+                    true,
+                    descriptorIndexingProperties.maxPerStageDescriptorUpdateAfterBindSamplers,
+                    descriptorIndexingProperties.maxPerStageDescriptorUpdateAfterBindUniformBuffers,
+                    descriptorIndexingProperties.maxPerStageDescriptorUpdateAfterBindStorageBuffers,
+                    descriptorIndexingProperties.maxPerStageDescriptorUpdateAfterBindSampledImages,
+                    descriptorIndexingProperties.maxPerStageDescriptorUpdateAfterBindStorageImages,
+                    descriptorIndexingProperties.maxPerStageUpdateAfterBindResources,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindSamplers,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindUniformBuffers,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindUniformBuffersDynamic,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindStorageBuffers,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindStorageBuffersDynamic,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindSampledImages,
+                    descriptorIndexingProperties.maxDescriptorSetUpdateAfterBindStorageImages};
+                MGLOG_I("Vulkan: update-after-bind descriptor layouts enabled");
+            } else {
+                MGLOG_I("Vulkan: descriptor indexing is present but lacks the complete update-after-bind feature set; "
+                        "using ordinary descriptor layouts");
+            }
+        } else {
+            MGLOG_I("Vulkan: descriptor indexing unavailable; using ordinary descriptor layouts");
+        }
+
         VkPhysicalDeviceIndexTypeUint8Features indexTypeUint8Features{};
         indexTypeUint8Features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES;
         if (indexTypeUint8ExtensionName != nullptr) {

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.h

@@ -229,6 +229,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                     GLint dstY, GLint width, GLint height, VkImageLayout srcRestoreLayout,
                                     VkImageLayout dstRestoreLayout, Bool stencilAspect);
         static SizeT GetReadbackTexelSize(VkFormat sourceFormat);
+        // Map a GL bottom-left-origin rectangle into the display-oriented swapchain image.
+        // Quarter-turn surface transforms swap the copy extent's axes.
+        static Bool MapDefaultFramebufferReadbackRect(GLint x, GLint y, GLsizei width, GLsizei height,
+                                                      VkExtent2D imageExtent,
+                                                      VkSurfaceTransformFlagBitsKHR preTransform,
+                                                      VkOffset2D* imageOffset, VkExtent2D* imageCopyExtent);
+        // Reorder a tightly packed block copied with MapDefaultFramebufferReadbackRect back into
+        // GL row order. The input block has swapped dimensions for 90/270 degree transforms.
+        static Bool RemapDefaultFramebufferReadback(const Uint8* rawPixels, Uint32 logicalWidth,
+                                                    Uint32 logicalHeight,
+                                                    VkSurfaceTransformFlagBitsKHR preTransform,
+                                                    SizeT texelSize, Uint8* outPixels);
         static Bool ConvertReadbackPixels(const Uint8* sourcePixels, VkFormat sourceFormat,
                                           GLsizei width, GLsizei height, GLenum destinationFormat,
                                           GLenum destinationType, SizeT destinationRowStride,
@@ -298,6 +310,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // The samplerAnisotropy device feature was granted, so GL_TEXTURE_MAX_ANISOTROPY_EXT is
         // honored rather than accepted-and-ignored.
         Bool IsSamplerAnisotropySupported() const { return m_samplerAnisotropyFeatureEnabled; }
+        // ARB_base_instance extends indirect command records with a non-zero firstInstance and
+        // requires gl_InstanceID to remain zero-based. Vulkan needs both features to honor that
+        // complete contract: one legalizes the command word, the other enables the shader rebase.
+        Bool IsNonZeroIndirectBaseInstanceSupported() const {
+            return m_drawIndirectFirstInstanceFeatureEnabled && m_shaderDrawParametersFeatureEnabled;
+        }
         // Ensures the frame command buffer is recording (same lazy pattern as
         // SetupDraw) and writes a bottom-of-pipe timestamp into the current
         // frame's pool. Null when unsupported or the pool is exhausted.
@@ -537,6 +555,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         Bool m_shaderDrawParametersExtensionEnabled = false;
         Bool m_shaderDrawParametersFeatureEnabled = false;
         Bool m_unformattedFloatStorageImagesEnabled = false;
+        // Set only after descriptor-indexing feature AND property queries prove that
+        // update-after-bind is legal for every descriptor category this renderer emits.
+        ProgramFactory::UpdateAfterBindLimits m_updateAfterBindLimits{};
         // fillModeNonSolid gates VK_POLYGON_MODE_LINE/_POINT (glPolygonMode); independentBlend gates
         // per-draw-buffer color write masks (glColorMaski). Both are cached at device creation and
         // drive a runtime fallback when the device lacks them.

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/GLImpl/Buffer/GL_Buffer.cpp

@@ -18,6 +18,7 @@
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
 #include <MG_Util/Converters/GLToMG/BufferEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/BufferEnumConverter.h>
+#include <MG_Util/Texture/PixelStoreProcessor.h>
 
 namespace MobileGL::MG_Impl::GLImpl {
     namespace {
@@ -31,6 +32,8 @@ namespace MobileGL::MG_Impl::GLImpl {
             NamedBufferData,
             NamedBufferSubData,
             CopyNamedBufferSubData,
+            ClearBufferData,
+            ClearBufferSubData,
             ClearNamedBufferData,
             ClearNamedBufferSubData,
             MapBufferRange,
@@ -65,6 +68,10 @@ namespace MobileGL::MG_Impl::GLImpl {
                 return "NamedBufferSubData";
             case BufferOp::CopyNamedBufferSubData:
                 return "CopyNamedBufferSubData";
+            case BufferOp::ClearBufferData:
+                return "ClearBufferData";
+            case BufferOp::ClearBufferSubData:
+                return "ClearBufferSubData";
             case BufferOp::ClearNamedBufferData:
                 return "ClearNamedBufferData";
             case BufferOp::ClearNamedBufferSubData:
@@ -143,16 +150,6 @@ namespace MobileGL::MG_Impl::GLImpl {
                 return 0;
             }
 
-            // The pattern is replicated verbatim, which is only the whole story while the client
-            // layout already matches the internal format - the case every entry point in practice
-            // uses, and the only one the conversion machinery here can express. Say so rather than
-            // quietly writing a differently-sized pattern.
-            const SizeT sourceSize = MG_Util::GetInputBytesPerPixel(inputFormat, pixelType);
-            if (sourceSize != elementSize) {
-                MGLOG_W_ONCE("%s: clear pattern is %zu bytes but internalformat 0x%X stores %zu; "
-                        "converting between them is not implemented",
-                        GetBufferOpName(op), sourceSize, internalformat, elementSize);
-            }
             return elementSize;
         }
 
@@ -194,27 +191,59 @@ namespace MobileGL::MG_Impl::GLImpl {
             return true;
         }
 
-        void ClearNamedBufferRange_State(GLuint buffer, GLenum internalformat, GLintptr offset, GLsizeiptr size,
-                                         GLenum format, GLenum type, const void* data, BufferOp op) {
+        Bool BuildClearPattern(GLenum internalformat, GLenum format, GLenum type, const void* data,
+                               SizeT patternSize, BufferOp op, Vector<Uint8>& pattern) {
+            const TextureInternalFormat internal = MG_Util::ConvertGLEnumToTextureInternalFormat(internalformat);
+            const TextureInputFormat inputFormat = MG_Util::ConvertGLEnumToTextureInputFormat(format);
+            const TexturePixelDataType inputType = MG_Util::ConvertGLEnumToTexturePixelDataType(type);
+
+            Vector<Uint8> zeroInput;
+            const void* inputPixel = data;
+            if (inputPixel == nullptr) {
+                const SizeT inputSize = MG_Util::GetInputBytesPerPixel(inputFormat, inputType);
+                if (inputSize == 0) {
+                    MG_State::pGLContext->RecordError(
+                        ErrorCode::InvalidValue,
+                        MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", GetBufferOpName(op),
+                                                     "format and type do not describe a source pixel."));
+                    return false;
+                }
+                zeroInput.resize(inputSize);
+                inputPixel = zeroInput.data();
+            }
+
+            if (!MG_Util::PixelStoreProcessor::ConvertOnePixelToInternal(
+                    internal, inputFormat, inputType, inputPixel, pattern)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>(
+                        "MG_Impl/GLImpl", GetBufferOpName(op),
+                        std::format("Cannot convert one ({}, {}) pixel into internalformat 0x{:X}.",
+                                    MG_Util::ConvertGLEnumToString(format), MG_Util::ConvertGLEnumToString(type),
+                                    internalformat)));
+                return false;
+            }
+
+            if (data == nullptr) {
+                // GL defines a null clear value as all zero bits in the destination store, while
+                // retaining the format/type validation above.
+                pattern.assign(patternSize, 0);
+            }
+            return true;
+        }
+
+        void ClearBufferRange_State(const SharedPtr<MG_State::GLState::BufferObject>& bufferObject,
+                                    GLenum internalformat, GLintptr offset, GLsizeiptr size,
+                                    GLenum format, GLenum type, const void* data, BufferOp op) {
             const SizeT patternSize = GetClearPatternSize(internalformat, format, type, op);
             if (patternSize == 0) return;
-
-            auto bufferObject = GetNamedBufferObject(buffer, op);
-            if (!bufferObject) return;
             if (!ValidateBufferClearRange(bufferObject, offset, size, patternSize, op)) return;
             if (size == 0) return;
 
-            Vector<Uint8> clearData(static_cast<SizeT>(size));
-            if (data) {
-                const auto* pattern = static_cast<const Uint8*>(data);
-                for (SizeT at = 0; at < clearData.size(); at += patternSize) {
-                    Memcpy(clearData.data() + at, pattern, patternSize);
-                }
-            } else {
-                Memset(clearData.data(), 0, clearData.size());
-            }
-
-            bufferObject->UploadSubData({clearData.data(), clearData.size()}, static_cast<SizeT>(offset));
+            Vector<Uint8> pattern;
+            if (!BuildClearPattern(internalformat, format, type, data, patternSize, op, pattern)) return;
+            bufferObject->FillSubData({pattern.data(), pattern.size()}, static_cast<SizeT>(offset),
+                                      static_cast<SizeT>(size));
         }
 
         auto& GetBufferBindingSlot(BufferTarget target) {
@@ -1197,17 +1226,34 @@ namespace MobileGL::MG_Impl::GLImpl {
                                         static_cast<SizeT>(writeOffset), static_cast<SizeT>(size));
     }
 
+    void ClearBufferData_State(GLenum target, GLenum internalformat, GLenum format, GLenum type, const void* data) {
+        auto bufferObject = GetBoundBufferObject(target, BufferOp::ClearBufferData);
+        if (!bufferObject) return;
+        ClearBufferRange_State(bufferObject, internalformat, 0, static_cast<GLsizeiptr>(bufferObject->GetSize()), format,
+                               type, data, BufferOp::ClearBufferData);
+    }
+
+    void ClearBufferSubData_State(GLenum target, GLenum internalformat, GLintptr offset, GLsizeiptr size,
+                                  GLenum format, GLenum type, const void* data) {
+        auto bufferObject = GetBoundBufferObject(target, BufferOp::ClearBufferSubData);
+        if (!bufferObject) return;
+        ClearBufferRange_State(bufferObject, internalformat, offset, size, format, type, data,
+                               BufferOp::ClearBufferSubData);
+    }
+
     void ClearNamedBufferData_State(GLuint buffer, GLenum internalformat, GLenum format, GLenum type, const void* data) {
         auto bufferObject = GetNamedBufferObject(buffer, BufferOp::ClearNamedBufferData);
         if (!bufferObject) return;
-        ClearNamedBufferRange_State(buffer, internalformat, 0, static_cast<GLsizeiptr>(bufferObject->GetSize()), format,
-                                    type, data, BufferOp::ClearNamedBufferData);
+        ClearBufferRange_State(bufferObject, internalformat, 0, static_cast<GLsizeiptr>(bufferObject->GetSize()), format,
+                               type, data, BufferOp::ClearNamedBufferData);
     }
 
     void ClearNamedBufferSubData_State(GLuint buffer, GLenum internalformat, GLintptr offset, GLsizeiptr size,
                                        GLenum format, GLenum type, const void* data) {
-        ClearNamedBufferRange_State(buffer, internalformat, offset, size, format, type, data,
-                                    BufferOp::ClearNamedBufferSubData);
+        auto bufferObject = GetNamedBufferObject(buffer, BufferOp::ClearNamedBufferSubData);
+        if (!bufferObject) return;
+        ClearBufferRange_State(bufferObject, internalformat, offset, size, format, type, data,
+                               BufferOp::ClearNamedBufferSubData);
     }
 
     void* MapNamedBuffer_State(GLuint buffer, GLenum access) {
@@ -1662,6 +1708,15 @@ namespace MobileGL::MG_Impl::GLImpl {
         CopyNamedBufferSubData_State(readBuffer, writeBuffer, readOffset, writeOffset, size);
     }
 
+    void ClearBufferData(GLenum target, GLenum internalformat, GLenum format, GLenum type, const void* data) {
+        ClearBufferData_State(target, internalformat, format, type, data);
+    }
+
+    void ClearBufferSubData(GLenum target, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format,
+                            GLenum type, const void* data) {
+        ClearBufferSubData_State(target, internalformat, offset, size, format, type, data);
+    }
+
     void ClearNamedBufferData(GLuint buffer, GLenum internalformat, GLenum format, GLenum type, const void* data) {
         ClearNamedBufferData_State(buffer, internalformat, format, type, data);
     }

MobileGL/MG_Impl/GLImpl/Buffer/GL_Buffer.h

@@ -27,6 +27,9 @@ namespace MobileGL::MG_Impl::GLImpl {
     void NamedBufferSubData(GLuint buffer, GLintptr offset, GLsizeiptr size, const void* data);
     void CopyNamedBufferSubData(GLuint readBuffer, GLuint writeBuffer, GLintptr readOffset, GLintptr writeOffset,
                                 GLsizeiptr size);
+    void ClearBufferData(GLenum target, GLenum internalformat, GLenum format, GLenum type, const void* data);
+    void ClearBufferSubData(GLenum target, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format,
+                            GLenum type, const void* data);
     void ClearNamedBufferData(GLuint buffer, GLenum internalformat, GLenum format, GLenum type, const void* data);
     void ClearNamedBufferSubData(GLuint buffer, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format,
                                  GLenum type, const void* data);

MobileGL/MG_Impl/GLImpl/Exporting/Definitions.cpp

@@ -985,8 +985,8 @@ DECLARE_GL_FUNCTION_HEAD(void, DrawElementsInstancedBaseVertexBaseInstance, GLen
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetActiveAtomicCounterBufferiv, GLuint program, GLuint bufferIndex, GLenum pname, GLint* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetActiveAtomicCounterBufferiv, program, bufferIndex, pname, params)
 DECLARE_GL_FUNCTION_HEAD(void, DrawTransformFeedbackInstanced, GLenum mode, GLuint id, GLsizei instancecount) DECLARE_GL_FUNCTION_END_NO_RETURN(void, DrawTransformFeedbackInstanced, mode, id, instancecount)
 DECLARE_GL_FUNCTION_HEAD(void, DrawTransformFeedbackStreamInstanced, GLenum mode, GLuint id, GLuint stream, GLsizei instancecount) DECLARE_GL_FUNCTION_END_NO_RETURN(void, DrawTransformFeedbackStreamInstanced, mode, id, stream, instancecount)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, ClearBufferData, GLenum target, GLenum internalformat, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ClearBufferData, target, internalformat, format, type, data)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, ClearBufferSubData, GLenum target, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ClearBufferSubData, target, internalformat, offset, size, format, type, data)
+DECLARE_GL_FUNCTION_HEAD(void, ClearBufferData, GLenum target, GLenum internalformat, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ClearBufferData, target, internalformat, format, type, data)
+DECLARE_GL_FUNCTION_HEAD(void, ClearBufferSubData, GLenum target, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ClearBufferSubData, target, internalformat, offset, size, format, type, data)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetInternalformati64v, GLenum target, GLenum internalformat, GLenum pname, GLsizei count, GLint64* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetInternalformati64v, target, internalformat, pname, count, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, InvalidateTexSubImage, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, InvalidateTexSubImage, texture, level, xoffset, yoffset, zoffset, width, height, depth)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, InvalidateTexImage, GLuint texture, GLint level) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, InvalidateTexImage, texture, level)

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/BufferState/BufferObject.cpp

@@ -250,6 +250,34 @@ namespace MobileGL::MG_State::GLState {
         NotifyContentWrite(atOffset, data.size);
     }
 
+    void BufferObject::FillSubData(DataPtr pattern, SizeT atOffset, SizeT size) {
+        MOBILEGL_ASSERT(pattern.data != nullptr && pattern.size > 0,
+                        "FillSubData requires a non-empty pattern.");
+        MOBILEGL_ASSERT(size % pattern.size == 0,
+                        "FillSubData size (%zu) must be a multiple of pattern size (%zu).", size, pattern.size);
+        MOBILEGL_ASSERT(atOffset <= m_size && size <= m_size - atOffset,
+                        "FillSubData out of bounds: atOffset (%zu) + size (%zu) > m_size (%zu)", atOffset, size,
+                        m_size);
+        MOBILEGL_ASSERT(!m_isMapped || (m_mappingAccess & BufferMappingAccessBit::Persistent),
+                        "Cannot fill data while buffer is non-persistently mapped.");
+        if (size == 0) return;
+
+        // A clear is ordered after all earlier GPU writes. Partial clears additionally need the
+        // retained shadow bytes; whole-store clears need the same synchronization before writing
+        // an adopted persistent mapping that the GPU may still be accessing.
+        SyncGpuWrites();
+
+        Uint8* dst = m_resource.Bytes() + atOffset;
+        if (pattern.size == 1) {
+            Memset(dst, *static_cast<const Uint8*>(pattern.data), size);
+        } else {
+            for (SizeT at = 0; at < size; at += pattern.size) {
+                Memcpy(dst + at, pattern.data, pattern.size);
+            }
+        }
+        NotifyContentWrite(atOffset, size);
+    }
+
     void BufferObject::DownloadSubData(void* dst, SizeT atOffset, SizeT size) const {
         MOBILEGL_ASSERT(atOffset + size <= m_size,
                         "DownloadSubData out of bounds: atOffset (%zu) + size (%zu) > m_size (%zu)", atOffset, size,

MobileGL/MG_State/GLState/BufferState/BufferObject.h

@@ -132,6 +132,9 @@ namespace MobileGL {
 
             void UploadData(DataPtr data, SizeT atOffset);
             void UploadSubData(DataPtr data, SizeT atOffset);
+            // Repeats one already-converted element through [atOffset, atOffset + size) and
+            // publishes the range as one content mutation.
+            void FillSubData(DataPtr pattern, SizeT atOffset, SizeT size);
             // Reads `size` bytes from the CPU shadow at `atOffset` into `dst` (glGetBufferSubData).
             // The shadow reflects CPU writes (BufferData/SubData/maps) and backend write-backs, but not
             // arbitrary GPU-side writes.

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_State/GLState/ProgramState/ProgramLinkTask.h

@@ -60,6 +60,8 @@ namespace MobileGL::MG_State::GLState {
             Uint externalIndex = 0; // logs only
             Vector<LinkShaderInput> shaders; // already stage-sorted
             SharedPtr<const MG_Util::ShaderTranspiler::CompileEnv> env;
+            // Startup configuration copied with the task, never read from worker code.
+            Bool enableSpirvValidation = false;
             // The four "takes effect at the next link" request maps. Snapshotted rather than
             // referenced, which is precisely what makes glBindAttribLocation and friends
             // legal to call over a pending link without cancelling it: the pending link keeps

MobileGL/MG_State/GLState/ProgramState/ProgramObject.cpp

@@ -494,6 +494,7 @@ namespace MobileGL::MG_State::GLState {
         auto task = MakeShared<ProgramLinkTask>();
         task->in.externalIndex = m_externalIndex;
         task->in.env = MG_Util::ShaderTranspiler::GetCurrentCompileEnv();
+        task->in.enableSpirvValidation = MG_Config::Features.EnableSpirvValidation;
         task->in.explicitAttribLocations = m_explicitAttribLocations;
         task->in.explicitFragDataLocation = m_explicitFragDataLocation;
         task->in.explicitFragDataIndex = m_explicitFragDataIndex;

MobileGL/MG_State/GLState/ProgramState/ProgramObject.h

@@ -819,6 +819,9 @@ namespace MobileGL::MG_State::GLState {
         // backend asks this exactly where it used to ask GetLinkStatus(), i.e. right before
         // it builds or draws with the program.
         Bool GetSpirvStatus() const { return Spirv().spirvStatus; }
+        // Copied from the link task that generated this program's SPIR-V. Backends use it for
+        // their final transforms, which must honor the same diagnostic setting as phase B.
+        Bool GetSpirvValidationEnabled() const { return Spirv().enableSpirvValidation; }
 
         // The linked glslang reflection itself, for the ONE consumer that needs resource
         // lists no typed getter above exposes: the GL program-interface query layer
@@ -985,6 +988,7 @@ namespace MobileGL::MG_State::GLState {
         // cannot be lifted out of glslang's reflection instead.
         struct SpirvArtifacts {
             Vector<Vector<unsigned>> generatedSpirv;
+            Bool enableSpirvValidation = false;
             // Byte offset of each uniform location inside globalUboScratch, or
             // kInvalidUniformOffset. Sized maxUniformLocation + 1 by the routing pass.
             Vector<Uint> uniformOffsets;

MobileGL/MG_State/GLState/ProgramState/ProgramSpirvTask.cpp

@@ -102,7 +102,11 @@ namespace MobileGL::MG_State::GLState {
         }
 
         MGLOG_D("ProgramObject %u: Starting SPIR-V generation", externalIndex);
-        GenerateSpirv(handoff, externalIndex);
+        const Bool deferOutputValidationForDirectVulkan =
+            m_phaseA->in.env != nullptr && m_phaseA->in.env->backend == BackendType::DirectVulkan;
+        const Bool enableSpirvValidation = m_phaseA->in.enableSpirvValidation;
+        artifacts.enableSpirvValidation = enableSpirvValidation;
+        GenerateSpirv(handoff, externalIndex, deferOutputValidationForDirectVulkan, enableSpirvValidation);
         // GlslangToSpv was the only consumer of the parsed ASTs; everything after this point
         // works on the SPIR-V and on the TProgram's own self-contained reflection pool. Drop
         // them here rather than at the end of the body, which is ~87% of this node's runtime
@@ -137,7 +141,9 @@ namespace MobileGL::MG_State::GLState {
                 artifacts.generatedSpirv.size());
     }
 
-    void ProgramSpirvTask::GenerateSpirv(const ProgramLinkTask::SpirvHandoff& handoff, const Uint externalIndex) {
+    void ProgramSpirvTask::GenerateSpirv(const ProgramLinkTask::SpirvHandoff& handoff, const Uint externalIndex,
+                                         const Bool deferOutputValidationForDirectVulkan,
+                                         const Bool enableSpirvValidation) {
         /* As we passed first stage compilation/linking,
          * we'll assume all the operations here should
          * pass. We may be able to employ some optimizations
@@ -169,7 +175,8 @@ namespace MobileGL::MG_State::GLState {
         Bool allOptimized = true;
         {
             for (auto& spv : artifacts.generatedSpirv) {
-                auto success = ShaderCompiler::SanitizeAndOptimizeBinary(spv, spv);
+                auto success = ShaderCompiler::SanitizeAndOptimizeBinary(
+                    spv, spv, !deferOutputValidationForDirectVulkan, enableSpirvValidation);
                 if (!success) {
                     // The one genuine phase-B failure mode: one of the seven optimizer passes
                     // reported failure, so `spv` is whatever the run left behind. A fordebug

MobileGL/MG_State/GLState/ProgramState/ProgramSpirvTask.h

@@ -65,7 +65,8 @@ namespace MobileGL::MG_State::GLState {
     private:
         void RunBody() override;
 
-        void GenerateSpirv(const ProgramLinkTask::SpirvHandoff& handoff, Uint externalIndex);
+        void GenerateSpirv(const ProgramLinkTask::SpirvHandoff& handoff, Uint externalIndex,
+                           Bool deferOutputValidationForDirectVulkan, Bool enableSpirvValidation);
         void BuildGlobalUboRouting(const ProgramLinkTask::SpirvHandoff& handoff, Uint externalIndex);
 
         // Worker-side MGLOG replacement, replayed by the join on the GL thread. Same reason as

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/Backend/DirectVulkan/SanityTest.cpp

@@ -8,9 +8,28 @@
 
 #include <gtest/gtest.h>
 #include <iostream>
+#include <utility>
 #include <vector>
 
-#include <vulkan/vulkan.h>
+#include <MG_Backend/DirectVulkan/Renderer/ProgramFactory.h>
+
+TEST(DirectVulkanSanity, ProgramMovePreservesViewportIndexUsage) {
+    using VkProgramObject = MobileGL::MG_Backend::DirectVulkan::ProgramFactory::VkProgramObject;
+
+    VkProgramObject moveConstructedSource;
+    moveConstructedSource.writesViewportIndexBuiltin = true;
+    VkProgramObject moveConstructed(std::move(moveConstructedSource));
+    EXPECT_TRUE(moveConstructed.writesViewportIndexBuiltin);
+    EXPECT_FALSE(moveConstructedSource.writesViewportIndexBuiltin);
+
+    VkProgramObject moveAssignedSource;
+    moveAssignedSource.writesViewportIndexBuiltin = true;
+    VkProgramObject moveAssigned;
+    moveAssigned = std::move(moveAssignedSource);
+    EXPECT_TRUE(moveAssigned.writesViewportIndexBuiltin);
+    EXPECT_FALSE(moveAssignedSource.writesViewportIndexBuiltin);
+}
+
 TEST(DirectVulkanSanity, ExtensionEnumeration) {
     uint32_t extensionCount = 0;
     vkEnumerateInstanceExtensionProperties(nullptr, &extensionCount, nullptr);

MobileGL/MG_Test/BackendLoader/BackendLoaderTest.cpp

@@ -725,22 +725,57 @@ TEST(TextureAnisotropyCapabilities, ExtensionIsAdvertisedOnlyWhenTheHostDriverSu
         return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
     };
 
-    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false);
+    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false);
     EXPECT_FALSE(contains(without, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_FALSE(contains(without, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 
-    const auto with = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, true);
+    const auto with = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, true, false, false);
     EXPECT_TRUE(contains(with, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_TRUE(contains(with, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 
     // Same rule on the Vulkan backend, where the gate is the samplerAnisotropy device feature.
-    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false);
+    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false);
     EXPECT_FALSE(contains(vkWithout, MobileGL::E_GL_EXT_texture_filter_anisotropic));
-    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true);
+    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true, false);
     EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 }
 
+// Minecraft 26.3 checks ARB_draw_indirect before it considers the already-advertised
+// ARB_multi_draw_indirect, then separately requires ARB_base_instance before enabling its terrain
+// indirect path. Pin both strings and, just as importantly, the non-zero firstInstance gate.
+TEST(IndirectDrawAdvertisement, MatchesEachBackendsUsableCommandSemantics) {
+    const auto contains = [](const MobileGL::Vector<MobileGL::GLExtension>& extensions,
+                             MobileGL::GLExtension wanted) {
+        return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
+    };
+
+    const auto esWithoutIndirect =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false);
+    EXPECT_FALSE(contains(esWithoutIndirect, MobileGL::E_GL_ARB_draw_indirect));
+    EXPECT_FALSE(contains(esWithoutIndirect, MobileGL::E_GL_ARB_base_instance));
+
+    const auto esWithoutBaseInstance =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, false);
+    EXPECT_TRUE(contains(esWithoutBaseInstance, MobileGL::E_GL_ARB_draw_indirect));
+    EXPECT_FALSE(contains(esWithoutBaseInstance, MobileGL::E_GL_ARB_base_instance));
+
+    const auto esWithBoth =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, true);
+    EXPECT_TRUE(contains(esWithBoth, MobileGL::E_GL_ARB_draw_indirect));
+    EXPECT_TRUE(contains(esWithBoth, MobileGL::E_GL_ARB_base_instance));
+
+    const auto vkWithoutBaseInstance =
+        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false);
+    EXPECT_TRUE(contains(vkWithoutBaseInstance, MobileGL::E_GL_ARB_draw_indirect));
+    EXPECT_FALSE(contains(vkWithoutBaseInstance, MobileGL::E_GL_ARB_base_instance));
+
+    const auto vkWithBoth =
+        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, true);
+    EXPECT_TRUE(contains(vkWithBoth, MobileGL::E_GL_ARB_draw_indirect));
+    EXPECT_TRUE(contains(vkWithBoth, MobileGL::E_GL_ARB_base_instance));
+}
+
 TEST(TextureAnisotropyCapabilities, MaxAnisotropyIsQueriedOnlyWhenTheExtensionIsPresent) {
     ResetFakeDriver();
     g_fake.maxVertexSsboBlocks = 0;
@@ -836,3 +871,63 @@ TEST(MultiDrawCapabilities, ExtensionWithoutResolvedPointerIsNotSupport) {
     EXPECT_FALSE(caps.SupportsMultiDrawIndirect);
     EXPECT_FALSE(caps.SupportsMultiDrawElementsBaseVertex);
 }
+
+TEST(DrawIndirectCapabilities, RequiresEs31AndBothCoreEntryPoints) {
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    auto funcs = MakeFakeGLESFunctions();
+    funcs.glDrawElementsIndirect = [](GLenum, GLenum, const void*) {};
+
+    MobileGL::MG_External::GLESCapabilities supportedCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(supportedCaps, funcs));
+    EXPECT_TRUE(supportedCaps.SupportsDrawIndirect);
+
+    // The same pointers on an ES 3.0 context are not core entry points and cannot back the
+    // desktop extension contract.
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    g_fake.glesMinorVersion = 0;
+    MobileGL::MG_External::GLESCapabilities es30Caps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(es30Caps, funcs));
+    EXPECT_FALSE(es30Caps.SupportsDrawIndirect);
+
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    const auto missingElements = MakeFakeGLESFunctions();
+    MobileGL::MG_External::GLESCapabilities missingEntryPointCaps;
+    ASSERT_TRUE(
+        MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(missingEntryPointCaps, missingElements));
+    EXPECT_FALSE(missingEntryPointCaps.SupportsDrawIndirect);
+}
+
+TEST(BaseInstanceCapabilities, RequiresTheExtensionAndAllThreeEntryPoints) {
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    auto funcs = MakeFakeGLESFunctions();
+    funcs.glDrawArraysInstancedBaseInstanceEXT = [](GLenum, GLint, GLsizei, GLsizei, GLuint) {};
+    funcs.glDrawElementsInstancedBaseInstanceEXT =
+        [](GLenum, GLsizei, GLenum, const void*, GLsizei, GLuint) {};
+    funcs.glDrawElementsInstancedBaseVertexBaseInstanceEXT =
+        [](GLenum, GLsizei, GLenum, const void*, GLsizei, GLint, GLuint) {};
+
+    // Resolved stubs alone must never make the capability true.
+    MobileGL::MG_External::GLESCapabilities pointersOnlyCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(pointersOnlyCaps, funcs));
+    EXPECT_FALSE(pointersOnlyCaps.SupportsBaseInstance);
+
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    g_fake.extensions.emplace_back("GL_EXT_base_instance");
+    MobileGL::MG_External::GLESCapabilities supportedCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(supportedCaps, funcs));
+    EXPECT_TRUE(supportedCaps.SupportsBaseInstance);
+
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    g_fake.extensions.emplace_back("GL_EXT_base_instance");
+    funcs.glDrawElementsInstancedBaseInstanceEXT = nullptr;
+    MobileGL::MG_External::GLESCapabilities missingEntryPointCaps;
+    ASSERT_TRUE(
+        MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(missingEntryPointCaps, funcs));
+    EXPECT_FALSE(missingEntryPointCaps.SupportsBaseInstance);
+}

MobileGL/MG_Test/Buffer/BufferTest.cpp

@@ -16,6 +16,7 @@
 #include <MG_State/GLState/Core.h>
 
 #include <MG_Impl/GLImpl/Buffer/GL_Buffer.h>
+#include <MG_Impl/GetProcAddress.h>
 #include <MG_Impl/GLImpl/Getter/GL_Getter.h>
 
 using namespace MobileGL;
@@ -599,6 +600,117 @@ TEST_F(BufferTest, ClearNamedBufferSubDataRepeatsPattern) {
     EXPECT_EQ(actual, (Vector<Uint32>{0, pattern, pattern, pattern, 0}));
     EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
+TEST_F(BufferTest, ClearBufferSubDataInitializesIrisStaticSsboRange) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_SHADER_STORAGE_BUFFER, buffer);
+
+    Vector<Uint8> initial(32, 0x7F);
+    MobileGL::MG_Impl::GLImpl::BufferData(
+        GL_SHADER_STORAGE_BUFFER, initial.size(), initial.data(), GL_STATIC_DRAW);
+    const GLbyte zero = 0;
+    const auto clear = reinterpret_cast<PFNGLCLEARBUFFERSUBDATAPROC>(
+        MobileGL::MG_Impl::GetProcAddress("glClearBufferSubData"));
+    ASSERT_NE(clear, nullptr);
+    clear(GL_SHADER_STORAGE_BUFFER, GL_R8, 4, 24, GL_RED, GL_BYTE, &zero);
+
+    Vector<Uint8> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    EXPECT_EQ(actual, (Vector<Uint8>{0x7F, 0x7F, 0x7F, 0x7F,
+                                    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                    0x7F, 0x7F, 0x7F, 0x7F}));
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
+TEST_F(BufferTest, ClearBufferSubDataInitializesCompleteIrisStaticSsbo) {
+    constexpr SizeT irisStaticSsboSize = 5'000'192;
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_SHADER_STORAGE_BUFFER, buffer);
+
+    Vector<Uint8> initial(irisStaticSsboSize, 0x7F);
+    MobileGL::MG_Impl::GLImpl::BufferData(
+        GL_SHADER_STORAGE_BUFFER, initial.size(), initial.data(), GL_STATIC_DRAW);
+    const GLbyte zero = 0;
+    MobileGL::MG_Impl::GLImpl::ClearBufferSubData(
+        GL_SHADER_STORAGE_BUFFER, GL_R8, 0, irisStaticSsboSize, GL_RED, GL_BYTE, &zero);
+
+    Vector<Uint8> actual(irisStaticSsboSize);
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    EXPECT_EQ(actual, Vector<Uint8>(irisStaticSsboSize, 0));
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
+TEST_F(BufferTest, ClearBufferDataConvertsOneClientPixelBeforeRepeatingIt) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, buffer);
+
+    Vector<Uint32> initial(4, 0u);
+    MobileGL::MG_Impl::GLImpl::BufferData(GL_ARRAY_BUFFER, initial.size() * sizeof(Uint32), initial.data(),
+                                           GL_STATIC_DRAW);
+    const Uint8 value = 0xAB;
+    MobileGL::MG_Impl::GLImpl::ClearBufferData(
+        GL_ARRAY_BUFFER, GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE, &value);
+
+    Vector<Uint32> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size() * sizeof(Uint32));
+    EXPECT_EQ(actual, Vector<Uint32>(initial.size(), value));
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, 0);
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
+TEST_F(BufferTest, ClearBufferSubDataRejectsUnboundTarget) {
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
+    const GLbyte zero = 0;
+    MobileGL::MG_Impl::GLImpl::ClearBufferSubData(
+        GL_SHADER_STORAGE_BUFFER, GL_R8, 0, 1, GL_RED, GL_BYTE, &zero);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+}
+
+TEST_F(BufferTest, ClearBufferDataRejectsInvalidPixelFormatTypePairs) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, buffer);
+
+    const Vector<Uint8> initial{0x7F, 0x7F};
+    MobileGL::MG_Impl::GLImpl::BufferData(GL_ARRAY_BUFFER, initial.size(), initial.data(), GL_STATIC_DRAW);
+    const Uint16 packed = 0;
+    MobileGL::MG_Impl::GLImpl::ClearBufferData(
+        GL_ARRAY_BUFFER, GL_R16, GL_RED, GL_UNSIGNED_SHORT_5_6_5, &packed);
+    ExpectSingleGlError(GL_INVALID_VALUE);
+    MobileGL::MG_Impl::GLImpl::ClearBufferData(
+        GL_ARRAY_BUFFER, GL_R16, GL_RED, GL_UNSIGNED_SHORT_5_6_5, nullptr);
+    ExpectSingleGlError(GL_INVALID_VALUE);
+
+    Vector<Uint8> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    EXPECT_EQ(actual, initial);
+
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, 0);
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
 
 
 // GL 4.6 core 6.5: glBufferSubData fails only when the written range OVERLAPS the mapped range.

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/ParallelShaderCompileTest.cpp

@@ -516,17 +516,17 @@ TEST_F(ParallelShaderCompileTest, MaxShaderCompilerThreadsIgnoresTheCurrentBudge
 TEST_F(ParallelShaderCompileTest, BothBackendsAdvertiseTheExtensionIffAsyncIsEnabled) {
     {
         const AsyncModeScope async(true);
-        EXPECT_TRUE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false),
+        EXPECT_TRUE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false),
                                E_GL_KHR_parallel_shader_compile));
-        EXPECT_TRUE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false),
+        EXPECT_TRUE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false),
                                E_GL_KHR_parallel_shader_compile));
     }
     {
         const AsyncModeScope async(false);
-        EXPECT_FALSE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false),
+        EXPECT_FALSE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false),
                                 E_GL_KHR_parallel_shader_compile))
             << "MOBILEGL_ASYNC_SHADER_COMPILE=0 must withdraw the extension, not only the threading";
-        EXPECT_FALSE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false),
+        EXPECT_FALSE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false),
                                 E_GL_KHR_parallel_shader_compile))
             << "MOBILEGL_ASYNC_SHADER_COMPILE=0 must withdraw the extension, not only the threading";
     }

MobileGL/MG_Test/Program/ProgramTest.cpp

@@ -2692,11 +2692,13 @@ out vec4 fragColor;
 float fma
     (float a, float b, float c) { return a * b + c; }
 float sinh(float x, float y) { return x * y; }
+float length_squared(vec3 value) { return dot(value, value); }
 float round(float x) { return floor(x + 0.5); }
 float min3(float a, float b, float c) { return min(min(a, b), c); }
 
 void main() {
-    fragColor = vec4(fma(0.1, 0.2, 0.3), sinh(0.4, 2.0), round(1.25), min3(0.1, 0.2, 0.3));
+    fragColor = vec4(fma(0.1, 0.2, 0.3), sinh(0.4, 2.0), round(1.25),
+                     min3(0.1, 0.2, 0.3) + length_squared(vec3(0.1, 0.2, 0.3)));
 }
 )";
     GLuint vs = CompileShaderChecked(GL_VERTEX_SHADER, vsSource);
@@ -2707,6 +2709,7 @@ void main() {
         if (essl.find("fragColor") == String::npos) continue; // fragment module only
         EXPECT_NE(essl.find("mg_fma("), String::npos) << essl;
         EXPECT_NE(essl.find("mg_sinh("), String::npos) << essl;
+        EXPECT_NE(essl.find("mg_length_squared("), String::npos) << essl;
         EXPECT_NE(essl.find("mg_round("), String::npos) << essl;
         EXPECT_NE(essl.find("mg_min3("), String::npos) << essl;
         EXPECT_EQ(essl.find("float fma("), String::npos) << essl;

MobileGL/MG_Test/Program/ProgramUtilTest.cpp

@@ -52,20 +52,24 @@ TEST_F(ProgramUtilTest, RenameSamplerFunctionParameterInSpirvPass) {
                OpEntryPoint Fragment %main "main" %outColor
                OpExecutionMode %main OriginUpperLeft
                OpName %globalSampler "sampler"
+               OpName %globalNew "new"
                OpName %paramSampler "sampler"
+               OpName %paramNew "new"
                OpName %main "main"
                OpDecorate %outColor Location 0
        %void = OpTypeVoid
       %float = OpTypeFloat 32
     %v4float = OpTypeVector %float 4
      %mainFn = OpTypeFunction %void
-    %paramFn = OpTypeFunction %void %float
+    %paramFn = OpTypeFunction %void %float %float
   %outV4Ptr = OpTypePointer Output %v4float
  %privatePtr = OpTypePointer Private %float
    %outColor = OpVariable %outV4Ptr Output
 %globalSampler = OpVariable %privatePtr Private
+    %globalNew = OpVariable %privatePtr Private
      %helper = OpFunction %void None %paramFn
 %paramSampler = OpFunctionParameter %float
+    %paramNew = OpFunctionParameter %float
  %helperBody = OpLabel
                OpReturn
                OpFunctionEnd
@@ -91,6 +95,7 @@ TEST_F(ProgramUtilTest, RenameSamplerFunctionParameterInSpirvPass) {
     ASSERT_TRUE(tools.Disassemble(outputBinary, &outputText));
 
     EXPECT_NE(outputText.find("\"MGL_COMPAT_sampler\""), String::npos);
+    EXPECT_NE(outputText.find("\"MGL_COMPAT_new\""), String::npos);
 
     SizeT exactSamplerNameCount = 0;
     SizeT searchOffset = 0;
@@ -99,6 +104,14 @@ TEST_F(ProgramUtilTest, RenameSamplerFunctionParameterInSpirvPass) {
         searchOffset += std::strlen("\"sampler\"");
     }
     EXPECT_EQ(exactSamplerNameCount, 1u);
+
+    SizeT exactNewNameCount = 0;
+    searchOffset = 0;
+    while ((searchOffset = outputText.find("\"new\"", searchOffset)) != String::npos) {
+        ++exactNewNameCount;
+        searchOffset += std::strlen("\"new\"");
+    }
+    EXPECT_EQ(exactNewNameCount, 1u);
 }
 
 TEST_F(ProgramUtilTest, UnformattedFloatStorageImagesKeepIntegerAtomicImagesTyped) {
@@ -2060,153 +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.
@@ -2435,9 +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:
@@ -2836,16 +2699,6 @@ vec4 helperTint() { return vec4(1.0); }
         return binaryResult->front();
     }
 
-    struct SpirvValidationScope {
-        bool previous;
-        explicit SpirvValidationScope(bool enabled)
-            : previous(MG_Util::ShaderTranspiler::ShaderCompiler::SpirvValidationEnabled()) {
-            MG_Util::ShaderTranspiler::ShaderCompiler::SetSpirvValidationEnabled(enabled);
-        }
-        ~SpirvValidationScope() {
-            MG_Util::ShaderTranspiler::ShaderCompiler::SetSpirvValidationEnabled(previous);
-        }
-    };
 } // namespace
 
 TEST_F(ProgramUtilTest, DeadPrivateChainVertexInputIsEliminatedFromOptimizedBinary) {
@@ -2867,7 +2720,7 @@ TEST_F(ProgramUtilTest, DeadPrivateChainVertexInputIsEliminatedFromOptimizedBina
         << "entry-point-with-calls shape it exists for";
 
     Vector<Uint32> optimized;
-    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized));
+    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized, true, true));
 
     const SpirvVariableCensus after = TakeVariableCensus(optimized);
     EXPECT_EQ(after.inputCount, 1u)
@@ -2905,7 +2758,7 @@ void main() {
     ASSERT_GE(before.outputCount, 3u);
 
     Vector<Uint32> optimized;
-    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized));
+    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized, true, true));
     EXPECT_EQ(TakeVariableCensus(optimized).outputCount, before.outputCount)
         << "a declared-but-unwritten output was deleted; a fragment stage reading it now "
         << "fails to link (ES) or breaks the Vulkan stage interface";
@@ -2964,17 +2817,15 @@ void main() {
         // succeeds - fail-open call sites downstream must not see a different world),
         // and the failure latch is the signal. This is the catch that took a device
         // bisect to find when the validator was off everywhere.
-        SpirvValidationScope validationOn(true);
         const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
-        EXPECT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized));
+        EXPECT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized, true, true));
         EXPECT_GT(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
             << "an invalid optimized module must bump the validation-failure latch";
     }
     {
         // The shipping configuration: same result, no validation, latch untouched.
-        SpirvValidationScope validationOff(false);
         const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
-        EXPECT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized));
+        EXPECT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized, true, false));
         EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore);
     }
 }
@@ -3044,10 +2895,9 @@ void main() {
     ASSERT_FALSE(raw.empty());
     ASSERT_GE(CountRectImageTypes(raw), 1u) << "glslang no longer emits Dim::Rect for sampler2DRect";
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
     Vector<Uint32> optimized;
-    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized));
+    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized, true, true));
     EXPECT_EQ(CountRectImageTypes(optimized), 0u);
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
         << "a rectangle module must leave the chain valid, not latched as a failure";
@@ -3072,10 +2922,9 @@ void main() {
     ASSERT_TRUE(AnyLocationOnUniformStorage(raw))
         << "glslang no longer keeps the explicit uniform location; the strip pass may be obsolete";
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
     Vector<Uint32> optimized;
-    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized));
+    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, optimized, true, true));
     EXPECT_FALSE(AnyLocationOnUniformStorage(optimized));
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
         << "the stripped module must validate clean";
@@ -3172,11 +3021,10 @@ void main() {
         << "the fixture must reproduce the defect before the fix is asked to remove it:\n"
         << DisassembleSpirv(raw);
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> legalized;
-    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized));
+    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized, true));
     ASSERT_FALSE(legalized.empty());
 
     const String disassembly = DisassembleSpirv(legalized);
@@ -3213,11 +3061,10 @@ void main() {
     ASSERT_TRUE(LegalizeFragmentOutputIndexPass::BinaryHasDynamicOutputIndexing(raw))
         << DisassembleSpirv(raw);
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> legalized;
-    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized));
+    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized, true));
     ASSERT_FALSE(legalized.empty());
 
     const String disassembly = DisassembleSpirv(legalized);
@@ -3257,11 +3104,10 @@ void main() {
     ASSERT_FALSE(raw.empty());
     ASSERT_TRUE(LegalizeFragmentOutputIndexPass::BinaryHasDynamicOutputIndexing(raw));
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> legalized;
-    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized));
+    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized, true));
     ASSERT_FALSE(legalized.empty());
 
     const String disassembly = DisassembleSpirv(legalized);
@@ -3300,7 +3146,7 @@ void main() {
     ASSERT_FALSE(LegalizeFragmentOutputIndexPass::BinaryHasDynamicOutputIndexing(raw));
 
     Vector<Uint32> legalized;
-    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized));
+    ASSERT_TRUE(ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(raw, legalized, true));
     EXPECT_EQ(legalized, raw) << "the module must not be rewritten - not even re-serialized - when "
                                  "nothing indexes a fragment output dynamically";
 }
@@ -3550,11 +3396,10 @@ TEST_F(ProgramUtilTest, Lower1DArrayImagesRewritesTheTypeAndWidensTheCoordinate)
     ASSERT_EQ(Count1DArrayStorageImageTypes(spirv), 1u)
         << "the shared chain must leave the 1D-array image for this pass to handle";
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> lowered;
-    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered));
+    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered, true));
     ASSERT_FALSE(lowered.empty());
 
     EXPECT_EQ(Count1DArrayStorageImageTypes(lowered), 0u)
@@ -3602,11 +3447,10 @@ void main() { ssb.sum = imageLoad(i0, ivec2(2, 3)).r + imageLoad(i1, ivec3(1, 1,
     ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(raw, spirv));
     ASSERT_EQ(Count1DArrayStorageImageTypes(spirv), 1u);
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> lowered;
-    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered));
+    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered, true));
     ASSERT_FALSE(lowered.empty());
 
     EXPECT_EQ(Count1DArrayStorageImageTypes(lowered), 0u) << DisassembleSpirv(lowered);
@@ -3636,7 +3480,7 @@ void main() { ssb.sum = imageLoad(i0, 2).r; }
     ASSERT_FALSE(spirv.empty());
 
     Vector<Uint32> lowered;
-    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered));
+    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered, true));
     EXPECT_EQ(lowered, spirv) << "a non-arrayed 1D storage image must pass through byte for byte";
 
     const String essl = DecompileToEssl(lowered);
@@ -3660,7 +3504,7 @@ void main() { fragColor = texture(uTex, vUv); }
     ASSERT_FALSE(spirv.empty());
 
     Vector<Uint32> lowered;
-    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered));
+    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered, true));
     EXPECT_EQ(lowered, spirv) << "a sampled 1D-array image must pass through byte for byte";
 }
 
@@ -3684,7 +3528,7 @@ void main() { ssb.sum = uint(imageSize(i0).x) + imageLoad(i0, ivec2(0, 0)).r; }
         << "the fixture must contain the shape the pass declines";
 
     Vector<Uint32> lowered;
-    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered));
+    ASSERT_TRUE(ShaderCompiler::Lower1DArrayImagesForEssl(spirv, lowered, true));
     EXPECT_EQ(lowered, spirv) << "a declined module must be handed back untouched, not partly rewritten";
     EXPECT_EQ(Count1DArrayStorageImageTypes(lowered), 1u)
         << "declining means the 1D-array type is still there for the driver to reject";
@@ -3735,11 +3579,10 @@ void main() { imageStore(uni_image, ivec2(gl_GlobalInvocationID.xy), uvec4(15u,
     // Precondition: SPIRV-Cross prints no format for it, which is the ESSL the driver refuses.
     EXPECT_EQ(DecompileToEssl(spirv).find("r32ui"), String::npos);
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> baked;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked, true));
     ASSERT_FALSE(baked.empty());
     EXPECT_FALSE(ShaderCompiler::DeclaresFormatlessStorageImage(baked)) << DisassembleSpirv(baked);
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
@@ -3800,7 +3643,7 @@ void main() { imageStore(uni_image, ivec2(0), uvec4(1u)); }
 
     Vector<Uint32> baked;
     // Even asked to, with a format of the right component class.
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked, true));
     EXPECT_EQ(baked, spirv) << "a module with nothing format-less must pass through byte for byte";
     EXPECT_NE(DecompileToEssl(baked).find("rgba32ui"), String::npos);
 }
@@ -3822,11 +3665,10 @@ void main() { writeIt(uni_image); }
     ASSERT_FALSE(spirv.empty());
     ASSERT_TRUE(ShaderCompiler::DeclaresFormatlessStorageImage(spirv));
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> baked;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked, true));
     EXPECT_EQ(baked, spirv) << "a shape the retype cannot follow must leave the module untouched, "
                                "not partly rewritten:\n"
                             << DisassembleSpirv(baked);
@@ -3848,18 +3690,17 @@ void main() { imageStore(uni_image, ivec2(0), vec4(1.0)); }
                                                    GL_COMPUTE_SHADER);
     ASSERT_FALSE(spirv.empty());
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> baked;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32ui}}, baked, true));
     EXPECT_EQ(baked, spirv) << "a declined module must be handed back untouched, not partly rewritten";
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore);
 
     // ...and the same image with a float bind format is baked, so the decline above is about the
     // class and not about the pass refusing float images.
     Vector<Uint32> bakedFloat;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32f}}, bakedFloat));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_image", kGlR32f}}, bakedFloat, true));
     EXPECT_NE(DecompileToEssl(bakedFloat).find("r32f"), String::npos) << DisassembleSpirv(bakedFloat);
 }
 
@@ -3883,12 +3724,11 @@ void main() {
     ASSERT_EQ(CountSpirvOpcode(DisassembleSpirv(spirv), "OpTypeImage"), 1u)
         << "the fixture must have the two images sharing one type:\n" << DisassembleSpirv(spirv);
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> baked;
     ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(
-        spirv, {{"imgA", kGlR32ui}, {"imgB", kGlRgba32ui}}, baked));
+        spirv, {{"imgA", kGlR32ui}, {"imgB", kGlRgba32ui}}, baked, true));
     ASSERT_FALSE(baked.empty());
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
         << "splitting the shared type must not leave a dangling or duplicate declaration:\n"
@@ -3921,11 +3761,10 @@ void main() {
     ASSERT_EQ(CountSpirvOpcode(DisassembleSpirv(spirv), "OpTypeImage"), 2u)
         << "the fixture needs one Unknown-format and one r32ui image type:\n" << DisassembleSpirv(spirv);
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> baked;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"formatless", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"formatless", kGlR32ui}}, baked, true));
     ASSERT_FALSE(baked.empty());
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
         << "the baked image collided with the module's own r32ui image and left a duplicate type:\n"
@@ -3950,11 +3789,10 @@ void main() {
     ASSERT_FALSE(spirv.empty());
     ASSERT_TRUE(ShaderCompiler::DeclaresFormatlessStorageImage(spirv));
 
-    SpirvValidationScope validationOn(true);
     const Uint64 failuresBefore = ShaderCompiler::SpirvValidationFailureCount();
 
     Vector<Uint32> baked;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"imgs", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"imgs", kGlR32ui}}, baked, true));
     ASSERT_FALSE(baked.empty());
     EXPECT_FALSE(ShaderCompiler::DeclaresFormatlessStorageImage(baked)) << DisassembleSpirv(baked);
     EXPECT_EQ(ShaderCompiler::SpirvValidationFailureCount(), failuresBefore)
@@ -3980,7 +3818,7 @@ void main() { fragColor = texture(uni_sampler, vUv); }
         << "a sampled image must not read as a format-less STORAGE image:\n" << DisassembleSpirv(spirv);
 
     Vector<Uint32> baked;
-    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_sampler", kGlR32ui}}, baked));
+    ASSERT_TRUE(ShaderCompiler::BakeImageFormatsForEssl(spirv, {{"uni_sampler", kGlR32ui}}, baked, true));
     EXPECT_EQ(baked, spirv) << "a sampled image must pass through byte for byte";
 }
 

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;
 
@@ -936,6 +968,45 @@ TEST(DirectVulkanSanity, ReadbackUsesTheSourceFormatTexelSize) {
     EXPECT_EQ(VulkanRenderer::GetReadbackTexelSize(VK_FORMAT_R32G32B32A32_SFLOAT), 16u);
 }
 
+TEST(DirectVulkanSanity, DefaultFramebufferQuarterTurnReadbackMapsRectAndPixels) {
+    using MobileGL::MG_Backend::DirectVulkan::VulkanRenderer;
+    using MobileGL::Uint8;
+
+    VkOffset2D offset{};
+    VkExtent2D copyExtent{};
+    ASSERT_TRUE(VulkanRenderer::MapDefaultFramebufferReadbackRect(
+        1, 0, 2, 1, VkExtent2D{2, 3}, VK_SURFACE_TRANSFORM_ROTATE_90_BIT_KHR,
+        &offset, &copyExtent));
+    EXPECT_EQ(offset.x, 0);
+    EXPECT_EQ(offset.y, 1);
+    EXPECT_EQ(copyExtent.width, 1u);
+    EXPECT_EQ(copyExtent.height, 2u);
+
+    ASSERT_TRUE(VulkanRenderer::MapDefaultFramebufferReadbackRect(
+        1, 0, 2, 1, VkExtent2D{2, 3}, VK_SURFACE_TRANSFORM_ROTATE_270_BIT_KHR,
+        &offset, &copyExtent));
+    EXPECT_EQ(offset.x, 1);
+    EXPECT_EQ(offset.y, 0);
+    EXPECT_EQ(copyExtent.width, 1u);
+    EXPECT_EQ(copyExtent.height, 2u);
+
+    // Logical GL rows, bottom to top, are abc / def. The display-oriented swapchain blocks are
+    // transposed in opposite directions for 90 and 270 degrees.
+    const Uint8 raw90[] = {'a', 'd', 'b', 'e', 'c', 'f'};
+    const Uint8 raw270[] = {'f', 'c', 'e', 'b', 'd', 'a'};
+    const Uint8 expected[] = {'a', 'b', 'c', 'd', 'e', 'f'};
+    Uint8 result[sizeof(expected)]{};
+
+    ASSERT_TRUE(VulkanRenderer::RemapDefaultFramebufferReadback(
+        raw90, 3, 2, VK_SURFACE_TRANSFORM_ROTATE_90_BIT_KHR, 1, result));
+    EXPECT_TRUE(std::equal(std::begin(expected), std::end(expected), std::begin(result)));
+
+    std::fill(std::begin(result), std::end(result), 0);
+    ASSERT_TRUE(VulkanRenderer::RemapDefaultFramebufferReadback(
+        raw270, 3, 2, VK_SURFACE_TRANSFORM_ROTATE_270_BIT_KHR, 1, result));
+    EXPECT_TRUE(std::equal(std::begin(expected), std::end(expected), std::begin(result)));
+}
+
 TEST(DirectVulkanSanity, ReadbackConvertsRgba8AndRgba16fPixels) {
     using MobileGL::MG_Backend::DirectVulkan::VulkanRenderer;
     using MobileGL::MG_Util::EncodeFloatToHalfBits;

MobileGL/MG_Test/ShaderTranspiler/DemoteFloat64Test.cpp

@@ -154,7 +154,6 @@ class DemoteFloat64Test : public ::testing::Test {
 protected:
     void SetUp() override {
         MobileGL::Initialize();
-        ShaderCompiler::SetSpirvValidationEnabled(true);
         m_validationFailuresAtStart = ShaderCompiler::SpirvValidationFailureCount();
     }
 
@@ -175,7 +174,7 @@ TEST_F(DemoteFloat64Test, DemotesEveryWidthAndDropsTheCapability) {
     ASSERT_TRUE(DeclaresFloat64Capability(input));
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
 
     EXPECT_EQ(CountFloatTypesOfWidth(output, 64), 0u) << Disassemble(output);
     // And exactly one 32-bit float type survives: the merge has to happen, or spirv-val rejects
@@ -210,7 +209,7 @@ void main() {
         << Disassemble(input);
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
 
     // std140 for the demoted members: float at 4, vec2 at 8, vec3 at 16 (aligned like a vec4),
     // vec4 at 32, mat4 at 48 with a 16-byte column stride, the array at 112 with the std140
@@ -241,7 +240,7 @@ void main() {
     EXPECT_EQ(CollectOffsetsOf(input, "Ssbo"), (Vector<Uint32>{0, 32, 64})) << Disassemble(input);
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
 
     // std430, so the array packs at its element size rather than being rounded to 16: float at 0,
     // vec4 at 16, float[4] at 32 with a 4-byte stride. A storage block must NOT come out std140,
@@ -273,7 +272,7 @@ void main() {
     ASSERT_FALSE(before.empty());
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
 
     // Only the block that actually narrowed is re-laid-out. Touching the other one would be
     // churn at best, and a disagreement with glslang's own layout at worst.
@@ -287,7 +286,7 @@ TEST_F(DemoteFloat64Test, FoldsTheConversionsThatBecameIdentities) {
     ASSERT_GT(CountFConverts(input), 0u) << "the fixture no longer converts between the two widths";
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
 
     // SPIR-V requires the two component widths of an OpFConvert to differ, so every one of them
     // has to be gone: both sides are 32 bits now.
@@ -307,7 +306,7 @@ void main() {
     ASSERT_FALSE(input.empty());
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
 
     // A 64-bit literal is two words wide and a 32-bit one is a single word, so a constant left
     // unconverted is not merely imprecise - it is an unparseable instruction. Disassembling both
@@ -326,7 +325,7 @@ void main() { gl_Position = inPos; }
     ASSERT_FALSE(input.empty());
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
     // The pass reports SuccessWithoutChange here, and SPIRV-Tools asserts (in assert-enabled
     // builds) that such a run round-trips byte-identically.
     EXPECT_EQ(output, input);
@@ -351,7 +350,7 @@ void main() {
     ASSERT_EQ(CountFloatTypesOfWidth(input, 64), 1u) << Disassemble(input);
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(input, output, true));
     EXPECT_EQ(output, input) << Disassemble(output);
     EXPECT_TRUE(ShaderCompiler::ModuleDeclaresFloat64(output));
 }
@@ -362,7 +361,7 @@ TEST_F(DemoteFloat64Test, ModuleDeclaresFloat64AnswersBothWays) {
     EXPECT_TRUE(ShaderCompiler::ModuleDeclaresFloat64(wide));
 
     Vector<Uint32> demoted;
-    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(wide, demoted));
+    ASSERT_TRUE(ShaderCompiler::DemoteFloat64ToFloat32(wide, demoted, true));
     EXPECT_FALSE(ShaderCompiler::ModuleDeclaresFloat64(demoted));
 
     EXPECT_FALSE(ShaderCompiler::ModuleDeclaresFloat64({}));
@@ -375,7 +374,7 @@ TEST_F(DemoteFloat64Test, TheSharedChainDemotesToo) {
     ASSERT_FALSE(input.empty());
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(input, output));
+    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(input, output, true, true));
     EXPECT_FALSE(ShaderCompiler::ModuleDeclaresFloat64(output)) << Disassemble(output);
 }
 
@@ -459,7 +458,7 @@ TEST_P(DemoteFloat64EsslTest, TheDemotedModuleCanBeEmittedAsEssl) {
     ASSERT_FALSE(input.empty());
 
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(input, output));
+    ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(input, output, true, true));
 
     SpvcSession session(output, SessionUsageBit::Transpile);
     spvc_compiler_options options;
@@ -482,7 +481,7 @@ TEST_P(DemoteFloat64EsslTest, TheDemotedModuleCanBeEmittedAsEssl) {
 TEST_F(DemoteFloat64Test, RejectsGarbageInput) {
     const Vector<Uint32> notSpirv{0xdeadbeefu, 0u, 0u, 0u, 0u};
     Vector<Uint32> output;
-    EXPECT_FALSE(ShaderCompiler::DemoteFloat64ToFloat32(notSpirv, output));
+    EXPECT_FALSE(ShaderCompiler::DemoteFloat64ToFloat32(notSpirv, output, true));
 }
 
 // EliminateFloatEqualsZeroPass turns a comparison against 0.0 into an epsilon test, a
@@ -502,7 +501,7 @@ namespace {
         EXPECT_FALSE(input.empty());
         if (input.empty()) return false;
         Vector<Uint32> output;
-        EXPECT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(input, output));
+        EXPECT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(input, output, true, true));
         return Disassemble(output).find("FAbs") != String::npos;
     }
 

MobileGL/MG_Test/ShaderTranspiler/FlattenXfbInterfaceBlocksTest.cpp

@@ -98,7 +98,6 @@ class FlattenXfbInterfaceBlocksTest : public ::testing::Test {
 protected:
     void SetUp() override {
         MobileGL::Initialize();
-        ShaderCompiler::SetSpirvValidationEnabled(true);
         m_validationFailuresAtStart = ShaderCompiler::SpirvValidationFailureCount();
     }
 
@@ -116,7 +115,7 @@ TEST_F(FlattenXfbInterfaceBlocksTest, FlattensACapturedBlockIntoOneVariablePerMe
 
     std::set<String> flattened;
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"StageData"}, flattened, output));
+    ASSERT_TRUE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"StageData"}, flattened, output, true));
     ASSERT_FALSE(output.empty());
     EXPECT_EQ(flattened, (std::set<String>{"StageData"}));
 
@@ -145,7 +144,7 @@ TEST_F(FlattenXfbInterfaceBlocksTest, TheEmittedDeclarationIsAPlainArrayNotABloc
 
     std::set<String> flattened;
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"StageData"}, flattened, output));
+    ASSERT_TRUE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"StageData"}, flattened, output, true));
 
     const String after = Transpile(output);
     EXPECT_NE(after.find("StageData_attrib[16]"), String::npos) << after;
@@ -162,7 +161,7 @@ TEST_F(FlattenXfbInterfaceBlocksTest, GivesEachMemberItsOwnConsecutiveLocations)
 
     std::set<String> flattened;
     Vector<Uint32> output;
-    ASSERT_TRUE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"StageData"}, flattened, output));
+    ASSERT_TRUE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"StageData"}, flattened, output, true));
     ASSERT_FALSE(output.empty());
 
     const String dis = Disassemble(output);
@@ -184,7 +183,7 @@ TEST_F(FlattenXfbInterfaceBlocksTest, LeavesABlockNoCaptureNamesAlone) {
     std::set<String> flattened;
     Vector<Uint32> output;
     ASSERT_TRUE(
-        ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"SomeOtherBlock"}, flattened, output));
+        ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {"SomeOtherBlock"}, flattened, output, true));
     EXPECT_TRUE(flattened.empty());
 
     const String after = Transpile(output);
@@ -200,7 +199,7 @@ TEST_F(FlattenXfbInterfaceBlocksTest, DeclinesAnEmptyRequestWithoutRewriting) {
 
     std::set<String> flattened;
     Vector<Uint32> output;
-    EXPECT_FALSE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {}, flattened, output));
+    EXPECT_FALSE(ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(input, {}, flattened, output, true));
     EXPECT_TRUE(flattened.empty());
     EXPECT_TRUE(output.empty());
 }

MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.cpp

@@ -955,6 +955,8 @@ namespace MobileGL::MG_Util::BackendLoader {
                                  (caps.GLESVersion.Major == 3 && caps.GLESVersion.Minor >= 2);
         const Bool esAtLeast31 = caps.GLESVersion.Major > 3 ||
                                  (caps.GLESVersion.Major == 3 && caps.GLESVersion.Minor >= 1);
+        caps.SupportsDrawIndirect = esAtLeast31 && glesFuncs.glDrawArraysIndirect != nullptr &&
+                                    glesFuncs.glDrawElementsIndirect != nullptr;
         caps.SupportsDrawElementsBaseVertex = (esAtLeast32 || hasDrawElementsBaseVertexExtension) &&
                                               glesFuncs.glDrawElementsBaseVertex != nullptr;
         caps.SupportsComputeShader = esAtLeast31 && glesFuncs.glDispatchCompute != nullptr &&
@@ -976,6 +978,7 @@ namespace MobileGL::MG_Util::BackendLoader {
         MGLOG_I("    indexed glColorMaski: %s", caps.SupportsIndexedColorMask ? "yes" : "no");
         MGLOG_I("    dual-source blend (EXT_blend_func_extended): %s",
                 caps.SupportsDualSourceBlend ? "yes" : "no");
+        MGLOG_I("    draw indirect (ES 3.1 core): %s", caps.SupportsDrawIndirect ? "yes" : "no");
         MGLOG_I("    multi-draw indirect (EXT_multi_draw_indirect): %s",
                 caps.SupportsMultiDrawIndirect ? "yes" : "no");
         MGLOG_I("    multi-draw base vertex (EXT/OES_draw_elements_base_vertex + EXT_multi_draw_arrays): %s",

MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.h

@@ -1149,6 +1149,10 @@ namespace MobileGL {
             // GLES 3.2 core or GL_OES_shader_multisample_interpolation exposes
             // interpolateAtOffset and the three fragment-offset limit queries.
             Bool SupportsShaderMultisampleInterpolation = false;
+            // ES 3.1+ exposes glDrawArraysIndirect / glDrawElementsIndirect in core. Keep the
+            // version and both entry-point checks together so extension advertisement and the
+            // DirectGLES dispatch path cannot disagree on whether native indirect draws exist.
+            Bool SupportsDrawIndirect = false;
             // GL_EXT_multi_draw_indirect is present AND glMultiDrawArraysIndirectEXT /
             // glMultiDrawElementsIndirectEXT both resolved. Multi-draw is not core in any ES
             // version, and eglGetProcAddress may return a live-looking stub on drivers without

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/SelfTest/DriverPost.cpp

@@ -1168,7 +1168,9 @@ namespace MobileGL::MG_Util::SelfTest {
             backendApiVersionString = MG_Backend::DirectGLES::FormatBackendAPIVersionString(
                 summary.caps.GLESRendererString, summary.caps.GLESVersion.Major, summary.caps.GLESVersion.Minor);
             advertisedExtensions = JoinAdvertisedExtensions(MG_Backend::DirectGLES::BuildAdvertisedExtensions(
-                summary.caps.SupportsDisjointTimerQuery, summary.caps.SupportsTextureFilterAnisotropy));
+                summary.caps.SupportsDisjointTimerQuery, summary.caps.SupportsTextureFilterAnisotropy,
+                summary.caps.SupportsDrawIndirect,
+                summary.caps.SupportsDrawIndirect && summary.caps.SupportsBaseInstance));
         }
         AppendMobileGLReportedRows(builder, MG_Backend::DirectGLES::GetRendererIdentity(), backendApiVersionString,
                                    advertisedExtensions);
@@ -1461,6 +1463,8 @@ namespace MobileGL::MG_Util::SelfTest {
             Bool shaderSubgroupUsable = false;
             Bool timerQueriesSupported = false;
             Bool samplerAnisotropySupported = false;
+            Bool drawIndirectFirstInstanceSupported = false;
+            Bool shaderDrawParametersSupported = false;
         };
     } // namespace
 
@@ -1744,6 +1748,7 @@ namespace MobileGL::MG_Util::SelfTest {
         VkPhysicalDeviceFeatures features{};
         vkGetPhysicalDeviceFeaturesFn(physicalDevice, &features);
         summary.samplerAnisotropySupported = features.samplerAnisotropy == VK_TRUE;
+        summary.drawIndirectFirstInstanceSupported = features.drawIndirectFirstInstance == VK_TRUE;
         if (features.multiDrawIndirect == VK_TRUE) {
             builder.Pass("multiDrawIndirect", "indirect multi-draw batches run as single native commands");
         } else {
@@ -1910,6 +1915,7 @@ namespace MobileGL::MG_Util::SelfTest {
             builder.Warn("shaderDrawParameters",
                          "unavailable; shaders using gl_DrawID/gl_BaseInstance will not work");
         }
+        summary.shaderDrawParametersSupported = shaderDrawParameters;
 
         Bool provokingVertexLast = false;
         Bool transformFeedbackPreservesProvokingVertex = false;
@@ -2108,7 +2114,8 @@ namespace MobileGL::MG_Util::SelfTest {
             backendApiVersionString = MG_Backend::DirectVulkan::FormatBackendAPIVersionString(
                 summary.deviceName, summary.apiVersionString, summary.driverVersionString);
             advertisedExtensions = JoinAdvertisedExtensions(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(
-                summary.shaderSubgroupUsable, summary.timerQueriesSupported, summary.samplerAnisotropySupported));
+                summary.shaderSubgroupUsable, summary.timerQueriesSupported, summary.samplerAnisotropySupported,
+                summary.drawIndirectFirstInstanceSupported && summary.shaderDrawParametersSupported));
         }
         AppendMobileGLReportedRows(builder, MG_Backend::DirectVulkan::GetRendererIdentity(), backendApiVersionString,
                                    advertisedExtensions);

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/EsslBuiltinFunctionNames.h

@@ -61,7 +61,7 @@ namespace MobileGL {
                 "imageAtomicXor", "imageLoad", "imageSize", "imageStore", "imulExtended",
                 "intBitsToFloat", "interpolateAtCentroid", "interpolateAtOffset",
                 "interpolateAtSample", "inverse", "inversesqrt", "isinf", "isnan",
-                "ldexp", "length", "lessThan", "lessThanEqual", "log", "log2",
+                "ldexp", "length", "length_squared", "lessThan", "lessThanEqual", "log", "log2",
                 "matrixCompMult", "max", "max3", "memoryBarrier",
                 "memoryBarrierAtomicCounter", "memoryBarrierBuffer", "memoryBarrierImage",
                 "memoryBarrierShared", "mid3", "min", "min3", "mix", "mod", "modf",

MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.cpp

@@ -369,12 +369,6 @@ namespace MobileGL {
                 return allSpirv;
             }
 
-            // -1 unresolved, 0 off, 1 on. Resolved once from MOBILEGL_VALIDATE_SPIRV on first
-            // use. A live getenv rather than an MG_Config::Features field, for the same reason
-            // Config.h already exempts MOBILEGL_LOG_FILE_PATH: suites like SpirvPassTest never
-            // run MobileGL::Initialize(), and every Initialize() re-runs MG_ConfigLoader::Init,
-            // which would clobber a programmatic override stored in the feature table.
-            static std::atomic<int> g_validateSpirv{-1};
             // Total validation failures observed this process. This latch - not the wrappers'
             // return values - is the test-lane signal: validation must never change what a
             // wrapper returns, or the validating lanes would render differently from the
@@ -383,28 +377,6 @@ namespace MobileGL {
             static std::atomic<Uint64> g_spirvValidationFailures{0};
 
             namespace {
-                // Test lanes (desktop/CI/WSL) validate by default; device builds do not -
-                // validation costs real time per module, and on device the driver is the
-                // final validator anyway. MOBILEGL_VALIDATE_SPIRV overrides in either
-                // direction, using the ConfigLoader truthy rule.
-                constexpr bool kValidateSpirvDefault =
-#if defined(__ANDROID__)
-                    false;
-#else
-                    true;
-#endif
-
-                bool IsTruthySpirvEnvValue(const char* value) {
-                    if (value == nullptr || value[0] == '\0') {
-                        return false;
-                    }
-                    String lowered(value);
-                    for (auto& c : lowered) {
-                        c = static_cast<char>(std::tolower(static_cast<unsigned char>(c)));
-                    }
-                    return lowered != "0" && lowered != "false";
-                }
-
                 // spirv-tools' validator lazily constructs function-local static tables on
                 // its first run, which on this codebase happens on a ShaderCompilePool
                 // worker. Function-local statics are destroyed in reverse construction
@@ -445,11 +417,6 @@ namespace MobileGL {
                             tools.Validate(warmup);
                         }
                         std::atexit(+[] {
-                            // Flip validation off first: a validator table this warmup does
-                            // not know about (a future spirv-tools bump) would still be
-                            // destroyed before this handler, and workers must stop entering
-                            // Validate before the drain waits for them.
-                            g_validateSpirv.store(0, std::memory_order_release);
                             Async::ShaderCompilePool::StopAndDrainProcessPoolAtExit();
                         });
                     });
@@ -478,10 +445,10 @@ namespace MobileGL {
                 // Validation is decoupled from control flow on purpose: a failure logs and
                 // bumps the latch, and the caller proceeds exactly as the shipping (non-
                 // validating) configuration would. Tests assert on the latch delta.
-                void ValidateOrLatch(const char* site, const Vector<Uint32>& binary) {
-                    if (!ShaderCompiler::SpirvValidationEnabled()) {
-                        return;
-                    }
+                void ValidateOrLatch(const char* site, const Vector<Uint32>& binary,
+                                     const bool enableSpirvValidation) {
+                    if (!enableSpirvValidation) return;
+                    PinValidatorTablesForProcessExit();
                     spvtools::SpirvTools tools(SPV_ENV_VULKAN_1_1);
                     tools.SetMessageConsumer(MakeSpirvMessageConsumer(site));
                     if (!tools.Validate(binary)) {
@@ -502,39 +469,23 @@ namespace MobileGL {
                 // spirv-tools drops pass diagnostics on the floor.
                 bool RunOptimizerChecked(const char* site, spvtools::Optimizer& optimizer,
                                          const Vector<Uint32>& inputBinary,
-                                         Vector<uint32_t>& outputBinary) {
+                                         Vector<uint32_t>& outputBinary, const bool validateOutput,
+                                         const bool enableSpirvValidation) {
                     spvtools::OptimizerOptions options;
                     options.set_run_validator(false);
                     optimizer.SetMessageConsumer(MakeSpirvMessageConsumer(site));
                     if (!optimizer.Run(inputBinary.data(), inputBinary.size(), &outputBinary, options)) {
                         return false;
                     }
-                    ValidateOrLatch(site, outputBinary);
+                    if (validateOutput) {
+                        ValidateOrLatch(site, outputBinary, enableSpirvValidation);
+                    }
                     return true;
                 }
             } // namespace
 
-            bool ShaderCompiler::SpirvValidationEnabled() {
-                int state = g_validateSpirv.load(std::memory_order_acquire);
-                if (state < 0) {
-                    const char* env = std::getenv("MOBILEGL_VALIDATE_SPIRV");
-                    const bool resolved = env != nullptr ? IsTruthySpirvEnvValue(env) : kValidateSpirvDefault;
-                    int expected = -1;
-                    g_validateSpirv.compare_exchange_strong(expected, resolved ? 1 : 0,
-                                                            std::memory_order_acq_rel);
-                    state = g_validateSpirv.load(std::memory_order_acquire);
-                    if (state == 1) {
-                        PinValidatorTablesForProcessExit();
-                    }
-                }
-                return state == 1;
-            }
-
-            void ShaderCompiler::SetSpirvValidationEnabled(bool enabled) {
-                g_validateSpirv.store(enabled ? 1 : 0, std::memory_order_release);
-                if (enabled) {
-                    PinValidatorTablesForProcessExit();
-                }
+            void ShaderCompiler::PrepareSpirvValidation() {
+                PinValidatorTablesForProcessExit();
             }
 
             Uint64 ShaderCompiler::NoteSpirvValidationFailure() {
@@ -604,16 +555,19 @@ namespace MobileGL {
             }
 
             bool ShaderCompiler::DemoteFloat64ToFloat32(const Vector<Uint32>& inputBinary,
-                                                        Vector<uint32_t>& outputBinary) {
+                                                        Vector<uint32_t>& outputBinary,
+                                                        const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(DemoteFloat64Pass::CreateDemoteFloat64Pass());
 
-                return RunOptimizerChecked("DemoteFloat64ToFloat32", optimizer, inputBinary, outputBinary);
+                return RunOptimizerChecked("DemoteFloat64ToFloat32", optimizer, inputBinary, outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::SanitizeAndOptimizeBinary(const Vector<Uint32>& inputBinary,
-                                                           Vector<uint32_t>& outputBinary) {
+                                                           Vector<uint32_t>& outputBinary,
+                                                           const bool validateOutput,
+                                                           const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
 
@@ -663,38 +617,41 @@ namespace MobileGL {
                 optimizer.RegisterPass(DemoteFloat64Pass::CreateDemoteFloat64Pass());
 
                 return RunOptimizerChecked("SanitizeAndOptimizeBinary", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, validateOutput, enableSpirvValidation);
             }
 
             bool ShaderCompiler::LowerDrawParametersForEssl(const Vector<Uint32>& inputBinary,
-                                                            Vector<uint32_t>& outputBinary) {
+                                                            Vector<uint32_t>& outputBinary,
+                                                            const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(LowerDrawParametersPass::CreateLowerDrawParametersPass());
 
                 return RunOptimizerChecked("LowerDrawParametersForEssl", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::SplitArrayVertexInputsForEssl(const Vector<Uint32>& inputBinary,
-                                                               Vector<uint32_t>& outputBinary) {
+                                                               Vector<uint32_t>& outputBinary,
+                                                               const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(SplitArrayVertexInputsPass::CreateSplitArrayVertexInputsPass());
 
                 return RunOptimizerChecked("SplitArrayVertexInputsForEssl", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::BakeImageFormatsForEssl(const Vector<Uint32>& inputBinary,
                                                          const UnorderedMap<String, Uint>& glFormatByName,
-                                                         Vector<uint32_t>& outputBinary) {
+                                                         Vector<uint32_t>& outputBinary,
+                                                         const bool enableSpirvValidation) {
                 using namespace spvtools;
                 if (glFormatByName.empty()) return false;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(BakeImageFormatsPass::CreateBakeImageFormatsPass(glFormatByName));
 
-                return RunOptimizerChecked("BakeImageFormatsForEssl", optimizer, inputBinary, outputBinary);
+                return RunOptimizerChecked("BakeImageFormatsForEssl", optimizer, inputBinary, outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::DeclaresFormatlessStorageImage(const Vector<Uint32>& binary) {
@@ -718,7 +675,8 @@ namespace MobileGL {
             bool ShaderCompiler::FlattenXfbInterfaceBlocksForEssl(const Vector<Uint32>& inputBinary,
                                                                   const std::set<String>& blockNames,
                                                                   std::set<String>& flattenedBlockNames,
-                                                                  Vector<uint32_t>& outputBinary) {
+                                                                  Vector<uint32_t>& outputBinary,
+                                                                  const bool enableSpirvValidation) {
                 using namespace spvtools;
                 if (blockNames.empty()) return false;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
@@ -726,7 +684,7 @@ namespace MobileGL {
                     blockNames, &flattenedBlockNames));
 
                 return RunOptimizerChecked("FlattenXfbInterfaceBlocksForEssl", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::RewriteXfbCaptureNameForFlattenedBlock(
@@ -736,48 +694,53 @@ namespace MobileGL {
             }
 
             bool ShaderCompiler::PackDoubleVertexInputsForVulkan(const Vector<Uint32>& inputBinary,
-                                                                 Vector<uint32_t>& outputBinary) {
+                                                                 Vector<uint32_t>& outputBinary,
+                                                                 const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(PackDoubleVertexInputsPass::CreatePackDoubleVertexInputsPass());
 
                 return RunOptimizerChecked("PackDoubleVertexInputsForVulkan", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::StripUboMemberRelaxedPrecisionForEssl(const Vector<Uint32>& inputBinary,
-                                                                       Vector<uint32_t>& outputBinary) {
+                                                                       Vector<uint32_t>& outputBinary,
+                                                                       const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(
                     StripUboMemberRelaxedPrecisionPass::CreateStripUboMemberRelaxedPrecisionPass());
 
                 return RunOptimizerChecked("StripUboMemberRelaxedPrecisionForEssl", optimizer,
-                                           inputBinary, outputBinary);
+                                           inputBinary, outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::StripNoPerspectiveForEssl(const Vector<Uint32>& inputBinary,
-                                                           Vector<uint32_t>& outputBinary) {
+                                                           Vector<uint32_t>& outputBinary,
+                                                           const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(StripNoPerspectivePass::CreateStripNoPerspectivePass());
 
                 return RunOptimizerChecked("StripNoPerspectiveForEssl", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::EmulateNoPerspectiveForEssl(const Vector<Uint32>& inputBinary,
-                                                             Vector<uint32_t>& outputBinary) {
+                                                             Vector<uint32_t>& outputBinary,
+                                                             const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(EmulateNoPerspectivePass::CreateEmulateNoPerspectivePass());
 
                 return RunOptimizerChecked("EmulateNoPerspectiveForEssl", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::LegalizeFragmentOutputIndexingForEssl(const Vector<Uint32>& inputBinary,
-                                                                       Vector<uint32_t>& outputBinary) {
+                                                                       Vector<uint32_t>& outputBinary,
+                                                                       const bool enableSpirvValidation) {
                 using namespace spvtools;
 
                 // Detection gates everything: a module with no dynamically indexed fragment
@@ -810,7 +773,7 @@ namespace MobileGL {
 
                 Vector<uint32_t> folded;
                 if (!RunOptimizerChecked("LegalizeFragmentOutputIndexingForEssl.fold", folder, inputBinary,
-                                         folded) ||
+                                         folded, true, enableSpirvValidation) ||
                     folded.empty()) {
                     // Fail open onto the fallback rather than onto the illegal module.
                     folded = inputBinary;
@@ -829,7 +792,7 @@ namespace MobileGL {
                 lowerer.RegisterPass(CreateAggressiveDCEPass(false));
 
                 if (!RunOptimizerChecked("LegalizeFragmentOutputIndexingForEssl.lower", lowerer, folded,
-                                         outputBinary) ||
+                                         outputBinary, true, enableSpirvValidation) ||
                     outputBinary.empty()) {
                     outputBinary = folded;
                     return true;
@@ -846,16 +809,17 @@ namespace MobileGL {
             }
 
             bool ShaderCompiler::LowerRectImages(const Vector<Uint32>& inputBinary,
-                                                 Vector<uint32_t>& outputBinary) {
+                                                 Vector<uint32_t>& outputBinary,
+                                                 const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(NormalizeRectCoordinatesPass::CreateNormalizeRectCoordinatesPass());
 
-                return RunOptimizerChecked("LowerRectImages", optimizer, inputBinary, outputBinary);
+                return RunOptimizerChecked("LowerRectImages", optimizer, inputBinary, outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::Lower1DArrayImagesForEssl(const Vector<Uint32>& inputBinary,
-                                                            Vector<uint32_t>& outputBinary) {
+                                                            Vector<uint32_t>& outputBinary, const bool enableSpirvValidation) {
                 using namespace spvtools;
 
                 // Declined rather than half-translated: after the rewrite the image is a 2D
@@ -897,40 +861,41 @@ namespace MobileGL {
                 // second Shader. Deduplicating afterwards collapses all three at once.
                 optimizer.RegisterPass(CreateRemoveDuplicatesPass());
 
-                return RunOptimizerChecked("Lower1DArrayImagesForEssl", optimizer, inputBinary, outputBinary);
+                return RunOptimizerChecked("Lower1DArrayImagesForEssl", optimizer, inputBinary, outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::RebaseInstanceIndexForVulkan(const Vector<Uint32>& inputBinary,
-                                                              Vector<uint32_t>& outputBinary) {
+                                                              Vector<uint32_t>& outputBinary, const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(RebaseInstanceIndexPass::CreateRebaseInstanceIndexPass());
 
                 return RunOptimizerChecked("RebaseInstanceIndexForVulkan", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::ZeroBaseVertexForVulkan(const Vector<Uint32>& inputBinary,
-                                                         Vector<uint32_t>& outputBinary) {
+                                                         Vector<uint32_t>& outputBinary, const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(ZeroBaseVertexPass::CreateZeroBaseVertexPass());
 
-                return RunOptimizerChecked("ZeroBaseVertexForVulkan", optimizer, inputBinary, outputBinary);
+                return RunOptimizerChecked("ZeroBaseVertexForVulkan", optimizer, inputBinary, outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::DecoratePositionInvariantForVulkan(const Vector<Uint32>& inputBinary,
-                                                                    Vector<uint32_t>& outputBinary) {
+                                                                    Vector<uint32_t>& outputBinary, const bool enableSpirvValidation) {
                 using namespace spvtools;
                 Optimizer optimizer(SPV_ENV_VULKAN_1_1);
                 optimizer.RegisterPass(DecoratePositionInvariantPass::CreateDecoratePositionInvariantPass());
 
                 return RunOptimizerChecked("DecoratePositionInvariantForVulkan", optimizer, inputBinary,
-                                           outputBinary);
+                                           outputBinary, true, enableSpirvValidation);
             }
 
             bool ShaderCompiler::UseUnformattedFloatStorageImagesForVulkan(
-                const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary) {
+                const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary,
+                const bool enableSpirvValidation) {
                 constexpr SizeT kSpirvHeaderWordCount = 5;
                 outputBinary.clear();
                 if (inputBinary.size() < kSpirvHeaderWordCount || inputBinary[0] != spv::MagicNumber) {
@@ -1058,7 +1023,8 @@ namespace MobileGL {
                                     addedCapabilities.begin(), addedCapabilities.end());
                 // Hand-rolled word walk, so no Optimizer wrapper ever sees this rewrite;
                 // check the modified module explicitly in validating lanes.
-                ValidateOrLatch("UseUnformattedFloatStorageImagesForVulkan", outputBinary);
+                ValidateOrLatch("UseUnformattedFloatStorageImagesForVulkan", outputBinary,
+                                enableSpirvValidation);
                 return true;
             }
 

MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.h

@@ -23,19 +23,23 @@ namespace MobileGL {
                 static Result<SharedPtr<glslang::TProgram>> LinkProgram(const ProgramAttrib& attrib);
                 static Result<Vector<Vector<unsigned>>> GetSpirvBinaryFromProgram(const ProgramBinaryAttrib& attrib);
                 static bool SanitizeAndOptimizeBinary(const Vector<Uint32>& inputBinary,
-                                                      Vector<uint32_t>& outputBinary);
+                                                      Vector<uint32_t>& outputBinary,
+                                                      bool validateOutput = true,
+                                                      bool enableSpirvValidation = false);
                 // Demotes DrawIndex/BaseInstance/BaseVertex builtins to plain Private globals
                 // (mg_DrawID/mg_BaseInstance/mg_BaseVertex) so SPIRV-Cross can emit ESSL.
                 // Only for backends without native draw-parameter support (DirectGLES).
                 static bool LowerDrawParametersForEssl(const Vector<Uint32>& inputBinary,
-                                                       Vector<uint32_t>& outputBinary);
+                                                       Vector<uint32_t>& outputBinary,
+                                                       bool enableSpirvValidation = false);
                 // Replaces an ARRAY vertex input with one input per element at consecutive
                 // locations, seeding a Private copy of the array so indexed reads still work.
                 // GLSL ES has no array vertex inputs and SPIRV-Cross refuses the whole module
                 // rather than emulating them, so without this the stage never reaches the
                 // driver. Only for the DirectGLES transpile path.
                 static bool SplitArrayVertexInputsForEssl(const Vector<Uint32>& inputBinary,
-                                                          Vector<uint32_t>& outputBinary);
+                                                          Vector<uint32_t>& outputBinary,
+                                                          bool enableSpirvValidation = false);
                 // Replaces the named interface BLOCKS with one variable per member, named
                 // "<Block>_<member>", shadowing the block itself so the body is untouched. The
                 // Adreno ES driver silently captures NOTHING for a transform-feedback varying
@@ -46,7 +50,8 @@ namespace MobileGL {
                 static bool FlattenXfbInterfaceBlocksForEssl(const Vector<Uint32>& inputBinary,
                                                              const std::set<String>& blockNames,
                                                              std::set<String>& flattenedBlockNames,
-                                                             Vector<uint32_t>& outputBinary);
+                                                             Vector<uint32_t>& outputBinary,
+                                                             bool enableSpirvValidation = false);
                 // The capture request "StageData.attrib[0]" as the pass above renamed it,
                 // "StageData_attrib[0]", or false when it does not name a member of a block
                 // that was flattened.
@@ -58,17 +63,20 @@ namespace MobileGL {
                 // drivers reject cross-stage uniform blocks whose member precisions differ.
                 // Only for the DirectGLES transpile path.
                 static bool StripUboMemberRelaxedPrecisionForEssl(const Vector<Uint32>& inputBinary,
-                                                                  Vector<uint32_t>& outputBinary);
+                                                                  Vector<uint32_t>& outputBinary,
+                                                             bool enableSpirvValidation = false);
                 // Removes NoPerspective decorations so SPIRV-Cross emits plain (smooth) ESSL varyings.
                 // DirectGLES fallback only, for devices lacking GL_NV_shader_noperspective_interpolation
                 // (SPIRV-Cross would otherwise require that extension and the driver would reject it).
                 static bool StripNoPerspectiveForEssl(const Vector<Uint32>& inputBinary,
-                                                      Vector<uint32_t>& outputBinary);
+                                                      Vector<uint32_t>& outputBinary,
+                                                      bool enableSpirvValidation = false);
                 // Emulates noperspective (screen-linear) interpolation via gl_Position.w / gl_FragCoord.w
                 // so no NV extension is needed; strips what it cannot emulate. DirectGLES fallback for
                 // devices lacking GL_NV_shader_noperspective_interpolation. See EmulateNoPerspectivePass.
                 static bool EmulateNoPerspectiveForEssl(const Vector<Uint32>& inputBinary,
-                                                        Vector<uint32_t>& outputBinary);
+                                                        Vector<uint32_t>& outputBinary,
+                                                      bool enableSpirvValidation = false);
                 // Makes every index into a fragment-output array a constant integral
                 // expression, which is what GLSL ES requires and SPIR-V does not. Runs the
                 // stock folding chain first (loop unrolling folds the loop-derived indices
@@ -79,7 +87,8 @@ namespace MobileGL {
                 // dynamically, which is every shader but a handful.
                 // See LegalizeFragmentOutputIndexPass.
                 static bool LegalizeFragmentOutputIndexingForEssl(const Vector<Uint32>& inputBinary,
-                                                                  Vector<uint32_t>& outputBinary);
+                                                                  Vector<uint32_t>& outputBinary,
+                                                             bool enableSpirvValidation = false);
                 // Rebases loads of the InstanceIndex builtin to (InstanceIndex - BaseInstance) so
                 // shaders see GL's zero-based gl_InstanceID. Vertex shaders only; DirectVulkan
                 // backend only (glslang's relaxed mode aliases gl_InstanceID to gl_InstanceIndex,
@@ -88,7 +97,8 @@ namespace MobileGL {
                 // divides the coordinate of each normalized-coordinate lookup by the texture
                 // size and rewrites the image type to 2D. See NormalizeRectCoordinatesPass for
                 // what it declines and why.
-                static bool LowerRectImages(const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary);
+                static bool LowerRectImages(const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary,
+                                            bool enableSpirvValidation = false);
                 // GL_TEXTURE_1D_ARRAY storage images rewritten to the 2D-array shape the texture
                 // is actually stored in on ES, with the layer moved from the coordinate's second
                 // component to its third. DirectGLES transpile path only - Vulkan binds a real
@@ -96,7 +106,8 @@ namespace MobileGL {
                 // through untouched when the module declares no such image, which is every shader
                 // but a handful. See Lower1DArrayImagesPass for what it declines and why.
                 static bool Lower1DArrayImagesForEssl(const Vector<Uint32>& inputBinary,
-                                                      Vector<uint32_t>& outputBinary);
+                                                      Vector<uint32_t>& outputBinary,
+                                                      bool enableSpirvValidation = false);
                 // Gives each format-less storage image the format bound to its image unit, so
                 // the emitted ESSL can carry the format layout qualifier GLSL ES requires of
                 // every image and desktop GLSL lets a writeonly declaration omit. `glFormatByName`
@@ -105,7 +116,8 @@ namespace MobileGL {
                 // natively. See BakeImageFormatsPass for what it declines and why.
                 static bool BakeImageFormatsForEssl(const Vector<Uint32>& inputBinary,
                                                     const UnorderedMap<String, Uint>& glFormatByName,
-                                                    Vector<uint32_t>& outputBinary);
+                                                    Vector<uint32_t>& outputBinary,
+                                                    bool enableSpirvValidation = false);
                 // Whether the module declares a storage image with no format qualifier at all,
                 // i.e. whether BakeImageFormatsForEssl could change anything. One module parse,
                 // so the ~every shader that declares none pays no optimizer run.
@@ -124,27 +136,31 @@ namespace MobileGL {
                 // emitted text instead.
                 static bool SpirvCrossCanPrintEsslImageFormat(Uint glInternalFormat);
                 static bool RebaseInstanceIndexForVulkan(const Vector<Uint32>& inputBinary,
-                                                         Vector<uint32_t>& outputBinary);
+                                                         Vector<uint32_t>& outputBinary,
+                                                      bool enableSpirvValidation = false);
                 // Builds the non-indexed-draw variant of a vertex shader: every gl_BaseVertex
                 // read becomes zero, which is what GL defines for a command carrying no
                 // baseVertex parameter while Vulkan's builtin would report firstVertex.
                 // See ZeroBaseVertexPass.
                 static bool ZeroBaseVertexForVulkan(const Vector<Uint32>& inputBinary,
-                                                    Vector<uint32_t>& outputBinary);
+                                                    Vector<uint32_t>& outputBinary,
+                                                    bool enableSpirvValidation = false);
                 // Re-declares 64-bit float vertex inputs as their 32-bit unsigned word pair
                 // (double -> uvec2, dvec2 -> uvec4) and bitcasts them back to double at entry, so no
                 // VK_FORMAT_R64*_SFLOAT is needed - lavapipe advertises none of them for vertex
                 // buffers. Vertex stage, DirectVulkan only; pairs with the Float64 case in
                 // VertexInputStateFactory::ToVkVertexFormat.
                 static bool PackDoubleVertexInputsForVulkan(const Vector<Uint32>& inputBinary,
-                                                            Vector<uint32_t>& outputBinary);
+                                                            Vector<uint32_t>& outputBinary,
+                                                      bool enableSpirvValidation = false);
                 // Adds the Invariant decoration to every Position builtin output. GL apps
                 // routinely rely on cross-program position invariance for multi-pass
                 // equality depth tests (e.g. GEQUAL re-draws of the same geometry), and
                 // mobile drivers that optimize per-pipeline break that without the
                 // decoration. DirectVulkan only.
                 static bool DecoratePositionInvariantForVulkan(const Vector<Uint32>& inputBinary,
-                                                               Vector<uint32_t>& outputBinary);
+                                                               Vector<uint32_t>& outputBinary,
+                                                             bool enableSpirvValidation = false);
                 // Replaces the declared format of float storage images with Unknown and adds the
                 // matching SPIR-V capabilities. DirectVulkan uses this only when both Vulkan
                 // shaderStorageImage*WithoutFormat features are enabled, allowing the
@@ -152,13 +168,15 @@ namespace MobileGL {
                 // storage images deliberately keep their declared format for GL-compatible bit
                 // reinterpretation paths (for example, R32F storage accessed as r32ui).
                 static bool UseUnformattedFloatStorageImagesForVulkan(
-                    const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary);
+                    const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary,
+                    bool enableSpirvValidation = false);
                 // Rewrites every 64-bit float in the module to a 32-bit one, preserving every
                 // block offset and stride exactly (see DemoteFloat64Pass). Already part of
                 // SanitizeAndOptimizeBinary, which is where production reaches it; exposed
                 // separately so a test can drive the demotion on its own.
                 static bool DemoteFloat64ToFloat32(const Vector<Uint32>& inputBinary,
-                                                   Vector<uint32_t>& outputBinary);
+                                                   Vector<uint32_t>& outputBinary,
+                                                   bool enableSpirvValidation = false);
                 static Result<String> DecompileShader(SpvcSession& session);
 
                 // Parses one trivial shader in each configuration the production path can
@@ -185,18 +203,16 @@ namespace MobileGL {
                 // no way left to warm it.
                 static void ResetPrewarmLatch();
 
-                // Test-environment SPIR-V validation. When enabled, every Optimizer wrapper
-                // in this file validates its OUTPUT binary - the bytes a driver can actually
-                // receive - and a failure logs the VUID (via MGLOG_I; see the consumer for
-                // why not MGLOG_E) and bumps the failure latch below WITHOUT changing the
-                // wrapper's return value: control flow must stay identical between the
-                // validating and shipping configurations, or fail-open call sites would make
-                // the two render differently. Resolved lazily from MOBILEGL_VALIDATE_SPIRV;
-                // defaults on for desktop/CI/WSL builds and off for device (__ANDROID__)
-                // builds. The setter wins over the environment and is safe to call from test
-                // fixtures at any time.
-                static bool SpirvValidationEnabled();
-                static void SetSpirvValidationEnabled(bool enabled);
+                // Validation is an explicit immutable option of each compiler operation. The
+                // program-link task snapshots MOBILEGL_ENABLE_SPIRV_VALIDATION before it can run
+                // on a worker; standalone callers pass true directly. A failure logs the VUID and
+                // bumps the latch below WITHOUT changing a wrapper's return value, so validating
+                // and shipping configurations preserve identical rendering control flow.
+
+                // Makes validator table lifetime safe before an external final-module validator
+                // runs. This has no configuration state; callers invoke it only for an enabled
+                // task-local validation option.
+                static void PrepareSpirvValidation();
 
                 // The test-lane enforcement signal: total validation failures observed this
                 // process. Tests snapshot it, run the operation under scrutiny, and assert

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

MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/RenameSamplerFunctionParameterPass.cpp

@@ -19,23 +19,27 @@ namespace MobileGL {
     namespace MG_Util {
         namespace ShaderTranspiler {
             namespace {
-                constexpr const char* kConflictingName = "sampler";
-                constexpr const char* kCompatName = "MGL_COMPAT_sampler";
+                const char* GetCompatName(StringView name) {
+                    if (name == "sampler") return "MGL_COMPAT_sampler";
+                    if (name == "new") return "MGL_COMPAT_new";
+                    return nullptr;
+                }
 
-                Bool IsNamedSamplerFunctionParameter(spvtools::opt::IRContext* context,
-                                                     spvtools::opt::Instruction& nameInst) {
+                const char* GetConflictingFunctionParameterCompatName(spvtools::opt::IRContext* context,
+                                                                       spvtools::opt::Instruction& nameInst) {
                     if (nameInst.opcode() != spv::Op::OpName || nameInst.NumInOperands() < 2) {
-                        return false;
+                        return nullptr;
                     }
 
-                    if (nameInst.GetInOperand(1).AsString() != kConflictingName) {
-                        return false;
+                    const char* compatName = GetCompatName(nameInst.GetInOperand(1).AsString());
+                    if (compatName == nullptr) {
+                        return nullptr;
                     }
 
                     auto* defUseMgr = context->get_def_use_mgr();
                     const Uint32 targetId = nameInst.GetSingleWordInOperand(0);
                     const auto* target = defUseMgr->GetDef(targetId);
-                    return target != nullptr && target->opcode() == spv::Op::OpFunctionParameter;
+                    return target != nullptr && target->opcode() == spv::Op::OpFunctionParameter ? compatName : nullptr;
                 }
             } // namespace
 
@@ -44,12 +48,13 @@ namespace MobileGL {
                 auto* irContext = context();
 
                 for (auto& debugInst : irContext->debugs2()) {
-                    if (!IsNamedSamplerFunctionParameter(irContext, debugInst)) {
+                    const char* compatName = GetConflictingFunctionParameterCompatName(irContext, debugInst);
+                    if (compatName == nullptr) {
                         continue;
                     }
 
                     debugInst.SetInOperand(
-                        1, spvtools::utils::MakeVector<spvtools::opt::Operand::OperandData>(kCompatName));
+                        1, spvtools::utils::MakeVector<spvtools::opt::Operand::OperandData>(compatName));
                     modified = true;
                 }
 

MobileGL/MG_Util/Texture/PixelStoreProcessor.cpp

@@ -423,6 +423,26 @@ namespace MobileGL::MG_Util::PixelStoreProcessor {
             InternalPackedLayout internalPacked;
         };
 
+        Bool IsValidUnpackPixelPair(TextureInputFormat format, TexturePixelDataType type) {
+            UnpackChannelMapping mapping{};
+            if (!GetUnpackChannelMapping(format, mapping)) return false;
+
+            PackedTypeLayout packed{};
+            if (GetPackedTypeLayout(type, packed)) {
+                return packed.fieldCount == mapping.channelCount;
+            }
+
+            switch (type) {
+            case TexturePixelDataType::UnsignedInt5999Rev:
+            case TexturePixelDataType::UnsignedInt101111Rev:
+                return !mapping.isInteger && mapping.channelCount == 3;
+            default: {
+                ShadowComponent component{};
+                return GetDirectShadowComponentForType(type, mapping.isInteger, component);
+            }
+            }
+        }
+
         // Returns true when the (format, type) -> internal-format upload needs a per-texel conversion;
         // returns false both for layouts that already match the shadow bytes (memcpy fast path) and for
         // combinations the converter does not support (legacy copy behavior).
@@ -964,6 +984,32 @@ namespace MobileGL::MG_Util::PixelStoreProcessor {
         return outputPixels;
     }
 
+    Bool ConvertOnePixelToInternal(TextureInternalFormat targetInternalFormat,
+                                   TextureInputFormat textureInputFormat,
+                                   TexturePixelDataType inputDataType,
+                                   const void* inputPixel,
+                                   Vector<Uint8>& outputPixel) {
+        outputPixel.clear();
+        if (inputPixel == nullptr || !IsValidUnpackPixelPair(textureInputFormat, inputDataType)) return false;
+
+        PixelStoreParameters params{};
+        params.Alignment = 1;
+        SizeT convertedSize = 0;
+        void* converted = ProcessTexturePixelsDataUnpack(
+            inputPixel, params, targetInternalFormat, textureInputFormat, inputDataType, {1, 1, 1}, false,
+            convertedSize);
+        const SizeT expectedSize = MG_Util::GetSizedInternalFormatSizeInBytes(targetInternalFormat);
+        if (converted == nullptr || convertedSize != expectedSize || expectedSize == 0) {
+            if (converted != nullptr) free(converted);
+            return false;
+        }
+
+        outputPixel.resize(convertedSize);
+        Memcpy(outputPixel.data(), converted, convertedSize);
+        free(converted);
+        return true;
+    }
+
     void* ProcessTexturePixelsDataPack(const void* inputPixels, const PixelStoreParameters& params,
                                        TextureInternalFormat srcInternalFormat, TexturePixelDataType srcDataType,
                                        TextureInputFormat dstInputFormat, TexturePixelDataType dstDataType,

MobileGL/MG_Util/Texture/PixelStoreProcessor.h

@@ -21,6 +21,12 @@ namespace MobileGL::MG_Util::PixelStoreProcessor {
                                        TextureInternalFormat srcInternalFormat, TexturePixelDataType srcDataType,
                                        TextureInputFormat dstInputFormat, TexturePixelDataType dstDataType,
                                        IntVec3 dimension, Bool isBitmap, SizeT& outSize);
+    Bool ConvertOnePixelToInternal(TextureInternalFormat targetInternalFormat,
+                                   TextureInputFormat textureInputFormat,
+                                   TexturePixelDataType inputDataType,
+                                   const void* inputPixel,
+                                   Vector<Uint8>& outputPixel);
+
     void ProcessColorSwizzle(void* data, SizeT pixelCount, const Vector<TextureSwizzleParam>& swizzle);
 
     // True when a packed internal format's 32-bit storage word IS the client (format, type) word,

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/README.md

@@ -11,6 +11,8 @@ The bundled fixtures cover:
   ![Minecraft 1.21.4 startup golden](fixtures/minecraft-1.21.4-startup.0000092195.png)
 - minecraft-1.21.4-main-menu: captured from Minecraft 1.21.4's main menu.
   ![Minecraft 1.21.4 main menu golden](fixtures/minecraft-1.21.4-main-menu.0000481787.png)
+- minecraft-1.21.11-main-menu: captured from Minecraft 1.21.11's main menu on a Pixel 8 Pro through FCL MobileGL.
+  ![Minecraft 1.21.11 main menu golden](fixtures/minecraft-1.21.11-main-menu.0000205347.png)
 - minecraft-1.17-main-menu-854: captured from Minecraft 1.17's 854x480 main menu through FCL MobileGL capture.
   ![Minecraft 1.17 854x480 main menu golden](fixtures/minecraft-1.17-main-menu-854.0000117757.png)
 - minecraft-1.21.4-in-world: captured from Minecraft 1.21.4 after entering a singleplayer world.

tools/trace_replay/trace_cases.json

@@ -40,6 +40,13 @@
       "target_call": 481787,
       "timeout_seconds": 180
     },
+    {
+      "name": "minecraft-1.21.11-main-menu",
+      "trace_archive": "minecraft-1.21.11-main-menu.tgz",
+      "golden": "minecraft-1.21.11-main-menu.0000205347.png",
+      "target_call": 205347,
+      "timeout_seconds": 180
+    },
     {
       "name": "minecraft-1.17-main-menu-854",
       "trace_archive": "minecraft-1.17-main-menu-854.tgz",
@@ -270,11 +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)