[Test] (TraceReplay): register iterationRP in-world fixture (non-CI, fixture files pending LFS)

# Conflicts:

.github/scripts/fetch-trace-fixture-lfs.sh

@@ -10,11 +10,22 @@ case_name="$1"
 fixture_dir="${2:-tools/trace_replay/fixtures}"
 python_bin="${PYTHON:-python3}"
 mirror_base="${MOBILEGL_TRACE_FIXTURE_MIRROR_BASE:-https://repo.miawa.cn/mgl/tools/trace_replay/fixtures}"
+download_attempts="${MOBILEGL_TRACE_FIXTURE_DOWNLOAD_ATTEMPTS:-5}"
+retry_delay="${MOBILEGL_TRACE_FIXTURE_RETRY_DELAY:-2}"
 
 if ! command -v "${python_bin}" >/dev/null 2>&1 && command -v python >/dev/null 2>&1; then
   python_bin=python
 fi
 
+if ! [[ "${download_attempts}" =~ ^[1-9][0-9]*$ ]]; then
+  echo "MOBILEGL_TRACE_FIXTURE_DOWNLOAD_ATTEMPTS must be a positive integer: ${download_attempts}" >&2
+  exit 2
+fi
+if ! [[ "${retry_delay}" =~ ^[0-9]+$ ]]; then
+  echo "MOBILEGL_TRACE_FIXTURE_RETRY_DELAY must be a non-negative integer: ${retry_delay}" >&2
+  exit 2
+fi
+
 fixture_list="$("${python_bin}" tools/trace_replay/trace_cases.py \
   --format fixture-files \
   --case "${case_name}" \
@@ -34,6 +45,140 @@ if [ "${case_name}" = "OpenRA" ]; then
   exit 0
 fi
 
+get_lfs_metadata() {
+  local file="$1"
+  local pointer
+  local expected_oid
+  local expected_size
+
+  if ! pointer="$(git show "HEAD:${file}" 2>/dev/null)"; then
+    echo "failed to read tracked fixture metadata: ${file}" >&2
+    return 1
+  fi
+  if ! grep -q '^version https://git-lfs.github.com/spec/v1$' <<< "${pointer}"; then
+    echo "tracked fixture is not a Git LFS pointer: ${file}" >&2
+    return 1
+  fi
+
+  expected_oid="$(awk '$1 == "oid" && $2 ~ /^sha256:/ { sub(/^sha256:/, "", $2); print $2 }' <<< "${pointer}")"
+  expected_size="$(awk '$1 == "size" { print $2 }' <<< "${pointer}")"
+  if ! [[ "${expected_oid}" =~ ^[0-9a-f]{64}$ ]] || ! [[ "${expected_size}" =~ ^[0-9]+$ ]]; then
+    echo "invalid Git LFS pointer metadata: ${file}" >&2
+    return 1
+  fi
+
+  printf '%s %s\n' "${expected_oid}" "${expected_size}"
+}
+
+verify_fixture_file() {
+  local downloaded_file="$1"
+  local display_name="$2"
+  local expected_oid="$3"
+  local expected_size="$4"
+  local actual_oid
+  local actual_size
+
+  if [ ! -f "${downloaded_file}" ]; then
+    echo "fixture file is missing: ${display_name}" >&2
+    return 1
+  fi
+
+  actual_size="$(wc -c < "${downloaded_file}" | tr -d '[:space:]')"
+  if [ "${actual_size}" != "${expected_size}" ]; then
+    echo "fixture size mismatch for ${display_name}: expected ${expected_size}, got ${actual_size}" >&2
+    return 1
+  fi
+
+  actual_oid="$(sha256sum "${downloaded_file}" | awk '{ print $1 }')"
+  if [ "${actual_oid}" != "${expected_oid}" ]; then
+    echo "fixture SHA-256 mismatch for ${display_name}: expected ${expected_oid}, got ${actual_oid}" >&2
+    return 1
+  fi
+}
+
+fetch_file_from_mirror() {
+  local file="$1"
+  local url="$2"
+  local metadata
+  local expected_oid
+  local expected_size
+  local tmp_file="${file}.tmp"
+  local attempt
+  local partial_size
+  local curl_status
+
+  metadata="$(get_lfs_metadata "${file}")" || return 1
+  read -r expected_oid expected_size <<< "${metadata}"
+
+  if [ -f "${tmp_file}" ]; then
+    partial_size="$(wc -c < "${tmp_file}" | tr -d '[:space:]')"
+    if [ "${partial_size}" -gt "${expected_size}" ]; then
+      echo "Discarding oversized partial fixture ${tmp_file}: ${partial_size} > ${expected_size}" >&2
+      rm -f "${tmp_file}"
+    elif [ "${partial_size}" = "${expected_size}" ]; then
+      if verify_fixture_file "${tmp_file}" "${file}" "${expected_oid}" "${expected_size}"; then
+        mv "${tmp_file}" "${file}"
+        return 0
+      fi
+      rm -f "${tmp_file}"
+    fi
+  fi
+
+  for ((attempt = 1; attempt <= download_attempts; attempt++)); do
+    partial_size=0
+    if [ -f "${tmp_file}" ]; then
+      partial_size="$(wc -c < "${tmp_file}" | tr -d '[:space:]')"
+    fi
+
+    if [ "${partial_size}" -gt 0 ]; then
+      echo "Resuming mirror download for ${file} at byte ${partial_size} (attempt ${attempt}/${download_attempts})"
+    else
+      echo "Starting mirror download for ${file} (attempt ${attempt}/${download_attempts})"
+    fi
+
+    if curl -L --fail --show-error --continue-at - --output "${tmp_file}" "${url}"; then
+      if verify_fixture_file "${tmp_file}" "${file}" "${expected_oid}" "${expected_size}"; then
+        mv "${tmp_file}" "${file}"
+        return 0
+      fi
+      echo "Mirror download failed integrity verification; retrying from the beginning: ${file}" >&2
+      rm -f "${tmp_file}"
+    else
+      curl_status=$?
+      partial_size=0
+      if [ -f "${tmp_file}" ]; then
+        partial_size="$(wc -c < "${tmp_file}" | tr -d '[:space:]')"
+      fi
+
+      if [ "${partial_size}" = "${expected_size}" ]; then
+        if verify_fixture_file "${tmp_file}" "${file}" "${expected_oid}" "${expected_size}"; then
+          mv "${tmp_file}" "${file}"
+          return 0
+        fi
+        rm -f "${tmp_file}"
+        partial_size=0
+      elif [ "${partial_size}" -gt "${expected_size}" ]; then
+        echo "Discarding oversized partial fixture ${tmp_file}: ${partial_size} > ${expected_size}" >&2
+        rm -f "${tmp_file}"
+        partial_size=0
+      elif [ "${curl_status}" -eq 33 ]; then
+        echo "Mirror refused the resume request; retrying from the beginning: ${file}" >&2
+        rm -f "${tmp_file}"
+        partial_size=0
+      fi
+
+      echo "Mirror download attempt ${attempt}/${download_attempts} failed with curl exit ${curl_status}; retained ${partial_size} bytes for resume: ${file}" >&2
+    fi
+
+    if [ "${attempt}" -lt "${download_attempts}" ]; then
+      sleep "${retry_delay}"
+    fi
+  done
+
+  rm -f "${tmp_file}"
+  return 1
+}
+
 fetch_from_mirror() {
   mkdir -p "${fixture_dir}"
   for file in "${files[@]}"; do
@@ -42,11 +187,9 @@ fetch_from_mirror() {
     name="$(basename "${file}")"
     url="${mirror_base%/}/${name}"
     echo "Fetching trace fixture from mirror: ${url}"
-    if ! curl -L --fail --retry 3 --retry-delay 2 -o "${file}.tmp" "${url}"; then
-      rm -f "${file}.tmp"
+    if ! fetch_file_from_mirror "${file}" "${url}"; then
       return 1
     fi
-    mv "${file}.tmp" "${file}"
   done
 }
 
@@ -59,9 +202,7 @@ else
 fi
 
 for file in "${files[@]}"; do
-  test -s "${file}"
-  if head -n 1 "${file}" | grep -q "version https://git-lfs.github.com/spec/v1"; then
-    echo "failed to hydrate LFS fixture: ${file}" >&2
-    exit 1
-  fi
+  metadata="$(get_lfs_metadata "${file}")"
+  read -r expected_oid expected_size <<< "${metadata}"
+  verify_fixture_file "${file}" "${file}" "${expected_oid}" "${expected_size}"
 done

.github/workflows/apk.yml

@@ -227,6 +227,8 @@ jobs:
     env:
       AVD_NAME: mobilegl-ci
       ANDROID_AVD_HOME: ${{ github.workspace }}/.android/avd
+      ANDROID_HOME: ${{ github.workspace }}/.android/sdk
+      ANDROID_SDK_ROOT: ${{ github.workspace }}/.android/sdk
     steps:
       - name: Checkout repo
         uses: actions/checkout@v6
@@ -245,11 +247,11 @@ jobs:
         with:
           path: |
             ${{ env.ANDROID_AVD_HOME }}
-            /usr/local/lib/android/sdk/emulator
-            /usr/local/lib/android/sdk/platform-tools
-            /usr/local/lib/android/sdk/platforms/android-35
-            /usr/local/lib/android/sdk/system-images/android-35/google_apis/x86_64
-          key: ${{ runner.os }}-mobilegl-avd-api35-google_apis-x86_64-pixel_6-v1-${{ hashFiles('android-plugin/run-avd-ci.sh') }}
+            ${{ env.ANDROID_SDK_ROOT }}/emulator
+            ${{ env.ANDROID_SDK_ROOT }}/platform-tools
+            ${{ env.ANDROID_SDK_ROOT }}/platforms/android-35
+            ${{ env.ANDROID_SDK_ROOT }}/system-images/android-35/google_apis/x86_64
+          key: ${{ runner.os }}-mobilegl-avd-api35-google_apis-x86_64-pixel_6-v2-${{ hashFiles('android-plugin/run-avd-ci.sh') }}
 
       - name: Create AVD
         if: steps.android-avd-cache.outputs.cache-hit != 'true'
@@ -274,6 +276,8 @@ jobs:
     env:
       AVD_NAME: mobilegl-ci
       ANDROID_AVD_HOME: ${{ github.workspace }}/.android/avd
+      ANDROID_HOME: ${{ github.workspace }}/.android/sdk
+      ANDROID_SDK_ROOT: ${{ github.workspace }}/.android/sdk
     strategy:
       fail-fast: false
       max-parallel: 4
@@ -288,7 +292,7 @@ jobs:
       - name: Set Swap Space
         uses: pierotofy/set-swap-space@v1.0
         with:
-          swap-size-gb: 16
+          swap-size-gb: 8
 
       - name: Checkout repo
         uses: actions/checkout@v6
@@ -324,11 +328,11 @@ jobs:
         with:
           path: |
             ${{ env.ANDROID_AVD_HOME }}
-            /usr/local/lib/android/sdk/emulator
-            /usr/local/lib/android/sdk/platform-tools
-            /usr/local/lib/android/sdk/platforms/android-35
-            /usr/local/lib/android/sdk/system-images/android-35/google_apis/x86_64
-          key: ${{ runner.os }}-mobilegl-avd-api35-google_apis-x86_64-pixel_6-v1-${{ hashFiles('android-plugin/run-avd-ci.sh') }}
+            ${{ env.ANDROID_SDK_ROOT }}/emulator
+            ${{ env.ANDROID_SDK_ROOT }}/platform-tools
+            ${{ env.ANDROID_SDK_ROOT }}/platforms/android-35
+            ${{ env.ANDROID_SDK_ROOT }}/system-images/android-35/google_apis/x86_64
+          key: ${{ runner.os }}-mobilegl-avd-api35-google_apis-x86_64-pixel_6-v2-${{ hashFiles('android-plugin/run-avd-ci.sh') }}
 
       - name: Download retrace APK
         uses: actions/download-artifact@v8
@@ -378,27 +382,51 @@ jobs:
           if [ "${{ matrix.case.coherent_as_flush || false }}" = "true" ]; then
             extra_retrace_args+=(--coherent-as-flush)
           fi
-          timeout "$(( ${{ matrix.case.timeout_seconds }} + 300 ))" sh android-plugin/trace-replay-ci.sh \
-            --apk-file "${apk_file}" \
-            --package top.mobilegl.plugin.trace \
-            --backend "${{ matrix.backend.name }}" \
-            --result-root android-retrace-result \
-            --fixture-root android-retrace-fixture \
-            --case "${{ matrix.case.name }}" \
-            --trace-archive "${{ matrix.case.trace_archive }}" \
-            --trace-file "${{ matrix.case.trace_file }}" \
-            --golden "${{ matrix.case.golden }}" \
-            --alternate-golden "${{ matrix.case.alternate_golden || '' }}" \
-            --target-call "${{ matrix.case.target_call }}" \
-            --width "${{ matrix.case.width }}" \
-            --height "${{ matrix.case.height }}" \
-            --ssim-threshold "${{ matrix.case.ssim_threshold || '0.99' }}" \
-            --crop-x "${{ matrix.case.crop_x }}" \
-            --crop-y "${{ matrix.case.crop_y }}" \
-            --crop-width "${{ matrix.case.crop_width }}" \
-            --crop-height "${{ matrix.case.crop_height }}" \
-            --timeout-seconds "${{ matrix.case.timeout_seconds }}" \
-            "${extra_retrace_args[@]}"
+
+          run_retrace() {
+            timeout "$(( ${{ matrix.case.timeout_seconds }} + 300 ))" sh android-plugin/trace-replay-ci.sh \
+              --apk-file "${apk_file}" \
+              --package top.mobilegl.plugin.trace \
+              --backend "${{ matrix.backend.name }}" \
+              --result-root android-retrace-result \
+              --fixture-root android-retrace-fixture \
+              --case "${{ matrix.case.name }}" \
+              --trace-archive "${{ matrix.case.trace_archive }}" \
+              --trace-file "${{ matrix.case.trace_file }}" \
+              --golden "${{ matrix.case.golden }}" \
+              --alternate-golden "${{ matrix.case.alternate_golden || '' }}" \
+              --target-call "${{ matrix.case.target_call }}" \
+              --width "${{ matrix.case.width }}" \
+              --height "${{ matrix.case.height }}" \
+              --ssim-threshold "${{ matrix.case.ssim_threshold || '0.99' }}" \
+              --crop-x "${{ matrix.case.crop_x }}" \
+              --crop-y "${{ matrix.case.crop_y }}" \
+              --crop-width "${{ matrix.case.crop_width }}" \
+              --crop-height "${{ matrix.case.crop_height }}" \
+              --timeout-seconds "${{ matrix.case.timeout_seconds }}" \
+              "${extra_retrace_args[@]}"
+          }
+
+          retrace_status=0
+          run_retrace || retrace_status=$?
+          if [ "${retrace_status}" -eq 75 ]; then
+            echo "::warning::Android emulator infrastructure failed; restarting it and retrying this retrace once."
+            sh android-plugin/run-avd-ci.sh stop \
+              --avd-name "${AVD_NAME}" \
+              --emulator-log "${EMULATOR_LOG}" \
+              --pid-file "${EMULATOR_PID_FILE}"
+            adb kill-server || true
+            sleep 2
+            sh android-plugin/run-avd-ci.sh start \
+              --avd-name "${AVD_NAME}" \
+              --gpu "${{ matrix.backend.gpu }}" \
+              --emulator-log "${EMULATOR_LOG}" \
+              --pid-file "${EMULATOR_PID_FILE}" \
+              --boot-timeout 300
+            run_retrace
+          elif [ "${retrace_status}" -ne 0 ]; then
+            exit "${retrace_status}"
+          fi
 
       - name: Collect retrace summary inputs
         if: always()

.github/workflows/test.yml

@@ -107,6 +107,7 @@ jobs:
             --exclude='build.ninja' \
             --exclude='cmake_install.cmake' \
             -czf ci-artifacts/mobilegl-linux-runtime.tgz \
+            "${BUILD_DIR}/CTestTestfile.cmake" \
             "${BUILD_DIR}/MobileGL/MG_Test" \
             "${BUILD_DIR}/MobileGL/MG_Benchmark" \
             "${SHARED_LIBS[@]}"
@@ -156,12 +157,12 @@ jobs:
           PY
 
       - name: Test
-        working-directory: build-linux/MobileGL/MG_Test
+        working-directory: build-linux
         run: |
           if [ "${{ secrets.ACTIONS_STEP_DEBUG }}" = "true" ]; then
-            ctest -V
+            ctest -V -L unit --no-tests=error
           else
-            ctest --output-on-failure
+            ctest --output-on-failure -L unit --no-tests=error
           fi
 
   benchmark:
@@ -202,8 +203,8 @@ jobs:
           PY
 
       - name: Benchmark
-        working-directory: build-linux/MobileGL/MG_Benchmark
-        run: ctest -V -C Release
+        working-directory: build-linux
+        run: ctest -V -C Release -L benchmark --no-tests=error
 
   build-retrace:
     runs-on: ubuntu-latest

CMakeLists.txt

@@ -190,6 +190,7 @@ set(SOURCE_FILES
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/DecomposeWorkgroupVec3Pass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/LowerDrawParametersPass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/RebaseInstanceIndexPass.cpp
+    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/StripUboMemberRelaxedPrecisionPass.cpp
 
     MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.cpp
     MobileGL/MG_Util/BackendLoaders/Vulkan/Loader.cpp
@@ -475,6 +476,15 @@ if (NOT ANDROID AND NOT MOBILEGL_IOS)
 endif ()
 
 if (NOT ANDROID)
+    # Enable testing in the top-level scope so a CTestTestfile.cmake is emitted
+    # at the build-tree root. This lets `ctest` be invoked from the top-level
+    # build directory (IDE "run all tests", CI) and discover every test in the
+    # subdirectories below, instead of having to descend into each
+    # MG_Test/MG_Benchmark subdirectory. Tests are tagged with CTest labels
+    # (unit / benchmark / integration), so e.g. `ctest -L unit` selects just
+    # the unit suite.
+    enable_testing()
+
     if (MOBILEGL_BUILD_TEST)
         add_subdirectory(MobileGL/MG_Test)
     endif()
@@ -484,7 +494,6 @@ if (NOT ANDROID)
     endif()
 
     if (MOBILEGL_BUILD_TRACE_REPLAY)
-        enable_testing()
         add_subdirectory(tools/trace_replay)
     endif()
 endif()

MobileGL/Config.h

@@ -61,6 +61,12 @@ namespace MobileGL::MG_Config {
         // per-draw glBufferSubData path instead of the persistent-mapped ring allocator
         // (negative control / driver-bug escape hatch).
         Bool DisableUboRing = false;
+        // MOBILEGL_RELAXED_SEMANTICS: relax strict core-profile rules (e.g. VAO-0 draws,
+        // texture-name reuse after delete) even on contexts that explicitly requested a core
+        // profile. Without it, relaxed semantics still apply to every context that did not
+        // explicitly request a core profile via EGL_CONTEXT_OPENGL_PROFILE_MASK / a >=3.1
+        // version request.
+        Bool RelaxedSemantics = false;
     };
     extern FeaturesTable Features;
 } // namespace MobileGL::MG_Config

MobileGL/ConfigLoader.cpp

@@ -122,6 +122,7 @@ namespace MobileGL::MG_ConfigLoader {
         features.CoherentAsFlush = QueryEnvFlag("MOBILEGL_COHERENT_AS_FLUSH");
         features.TraceSkipAutodestroy = QueryEnvFlag("MOBILEGL_TRACE_SKIP_AUTODESTROY");
         features.DisableUboRing = QueryEnvFlag("MOBILEGL_DISABLE_UBO_RING");
+        features.RelaxedSemantics = QueryEnvFlag("MOBILEGL_RELAXED_SEMANTICS");
     }
 
     inline void InitBackendType() {

MobileGL/MG_Backend/BackendObject.h

@@ -232,6 +232,9 @@ namespace MobileGL {
 
         struct DynamicBackendParameters {
             SizeT UniformBufferOffsetAlignment = 256;
+            // GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT. 1.0 means the backend cannot filter anisotropically,
+            // which is also why the extension is not advertised in that case.
+            Float MaxTextureMaxAnisotropy = 1.0f;
             Float AliasedLineWidthRangeMin = 1.0f;
             Float AliasedLineWidthRangeMax = 1.0f;
             Float SmoothLineWidthRangeMin = 1.0f;

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp

@@ -605,9 +605,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     {
                         .TargetGLVersion = {3, 3, 0},   // Target OpenGL Version
                         .TargetGLSLVersion = {4, 6, 0}, // Target Shading Language Version
-                        // Baseline advertisement (no timer queries yet); reconciled once
-                        // the ES capabilities exist, see UpdateAdvertisedTimerQueryExtension.
-                        .Extensions = BuildAdvertisedExtensions(false),
+                        // Baseline advertisement (no timer queries / anisotropy yet); reconciled
+                        // once the ES capabilities exist, see UpdateAdvertisedCapabilityExtensions.
+                        .Extensions = BuildAdvertisedExtensions(false, false),
                         .IsCompatibilityProfile = false // Is Compatibility Profile
                     },
                 .StaticBackendCapability = {.AllowVSOnlyPrograms = false} // Backend Capability
@@ -627,8 +627,9 @@ 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 UpdateAdvertisedTimerQueryExtension() {
-            MutableRendererInfo().RendererGLInfo.Extensions = BuildAdvertisedExtensions(AreTimerQueriesSupported());
+        void UpdateAdvertisedCapabilityExtensions(Bool anisotropicFilteringSupported) {
+            MutableRendererInfo().RendererGLInfo.Extensions =
+                BuildAdvertisedExtensions(AreTimerQueriesSupported(), anisotropicFilteringSupported);
         }
     } // namespace
 
@@ -672,11 +673,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
             return false;
         }
         DirectGLES::SetGLESCapabilities(m_GLESCapabilities);
-        // Now that g_GLESCapabilities knows about GL_EXT_disjoint_timer_query,
-        // reconcile the E_GL_ARB_timer_query advertisement (see the comment on
-        // UpdateAdvertisedTimerQueryExtension for why it cannot happen when
-        // the extension list is first built).
-        UpdateAdvertisedTimerQueryExtension();
+        // 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);
         UpdateDynamicBackendParameters();
         PopulateFormatCapabilities(m_GLESFunctions, m_GLESCapabilities, MutableFormatCapabilities());
         PrintFormatCapabilities(GetFormatCapabilities());
@@ -818,7 +819,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
         return MutableRendererInfo();
     }
 
-    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported) {
+    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported) {
         Vector<GLExtension> extensions = {V_OpenGL30, V_OpenGL31, V_OpenGL32,
                                           V_OpenGL33, 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,
@@ -836,6 +837,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
         if (timerQueriesSupported && !MG_Config::Features.DisableTimerQuery) {
             extensions.push_back(E_GL_ARB_timer_query);
         }
+        // Only advertised when the host ES driver actually filters anisotropically: the sampler
+        // state is accepted regardless, but forwarding it would be a no-op without the extension,
+        // and an app that trusts the string (LWJGL builds GLCapabilities from it) would silently
+        // get plain trilinear.
+        if (anisotropicFilteringSupported) {
+            extensions.push_back(E_GL_EXT_texture_filter_anisotropic);
+            extensions.push_back(E_GL_ARB_texture_filter_anisotropic);
+        }
         return extensions;
     }
 
@@ -940,6 +949,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     void BackendObject_DirectGLES::UpdateDynamicBackendParameters() {
         m_dynamicParameters.UniformBufferOffsetAlignment = m_GLESCapabilities.UniformBufferOffsetAlignment;
+        m_dynamicParameters.MaxTextureMaxAnisotropy = m_GLESCapabilities.MaxTextureMaxAnisotropy;
         m_dynamicParameters.AliasedLineWidthRangeMin = m_GLESCapabilities.AliasedLineWidthRangeMin;
         m_dynamicParameters.AliasedLineWidthRangeMax = m_GLESCapabilities.AliasedLineWidthRangeMax;
         m_dynamicParameters.SmoothLineWidthRangeMin = m_GLESCapabilities.SmoothLineWidthRangeMin;

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.h

@@ -66,9 +66,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
     const RendererInfo& GetRendererIdentity();
 
     // The full OpenGL extension list Espryt advertises (glGetString(GL_EXTENSIONS))
-    // for a device whose timer queries are (or are not) usable. The
-    // MOBILEGL_DISABLE_TIMERQUERY escape hatch is applied inside.
-    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported);
+    // for a device whose timer queries / anisotropic filtering are (or are not) usable.
+    // The MOBILEGL_DISABLE_TIMERQUERY escape hatch is applied inside.
+    Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported);
 
     // 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.h

@@ -19,6 +19,10 @@
     operation Utils::CheckGLESError();
 
 namespace MobileGL::MG_Backend::DirectGLES {
+    // Re-establishes the frontend texture-unit bindings on the native ES context.
+    // Content uploads use scratch bindings, so draws and dispatches call this after
+    // texture synchronization.
+    void BindCurrentTextures();
     void ClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil);
     void ClearBufferfv(GLenum buffer, GLint drawbuffer, const GLfloat* value);
     void ClearBufferuiv(GLenum buffer, GLint drawbuffer, const GLuint* value);

MobileGL/MG_Backend/DirectGLES/Managers.cpp

@@ -1405,7 +1405,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 m_prevSkipPixels = s_skipPixels;
                 m_prevImageHeight = s_imageHeight;
                 m_prevSkipImages = s_skipImages;
-                Apply(4, 0, 0, 0, 0, 0);
+                // Shadow mip data is tightly packed (ProcessTexturePixelsDataUnpack emits
+                // width * bpp rows with no padding), so uploads must use UNPACK_ALIGNMENT = 1.
+                // Alignment 4 made the driver read e.g. 7-byte R8 rows at an 8-byte stride,
+                // shifting every row of a non-multiple-of-4 upload by one pixel.
+                Apply(1, 0, 0, 0, 0, 0);
             }
 
             ~ScopedDefaultUnpackState() {
@@ -1472,10 +1476,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 return 2;
             case TextureInternalFormat::RGB8Snorm:
             case TextureInternalFormat::RGB16:
+            case TextureInternalFormat::RGB10:  // stored as RGB16 (UNorm16 shadow)
+            case TextureInternalFormat::RGB12:  // stored as RGB16 (UNorm16 shadow)
             case TextureInternalFormat::RGB16Snorm:
                 return 3;
             case TextureInternalFormat::RGBA8Snorm:
             case TextureInternalFormat::RGBA16:
+            case TextureInternalFormat::RGBA12: // stored as RGBA16 (UNorm16 shadow)
             case TextureInternalFormat::RGBA16Snorm:
                 return 4;
             default:
@@ -1570,7 +1577,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
             MGLOG_D("Syncing texture mipmaps with backend ID %u to backend for state ID %u", m_backendTextureId,
                     stateTextureObject->GetExternalIndex());
 
-            GLenum target = MG_Util::ConvertTextureTargetToGLEnum(stateTextureObject->GetTarget());
+            GLenum target = ConvertTextureTargetToBackendGLEnum(stateTextureObject->GetTarget());
             auto targetInternal = stateTextureObject->GetTarget();
             MGLOG_D("    Texture target for syncing is %s",
                     MG_Util::ConvertTextureTargetToString(targetInternal).c_str());
@@ -1691,7 +1698,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                             auto levelTexelSize = textureMipmapObject->GetMipmapTexelSize(uploadTarget, level);
                             auto levelByteSize = textureMipmapObject->GetMipmapByteSize(uploadTarget, level);
                             bool levelDirty = textureMipmapObject->IsStorageDirty(uploadTarget, level);
-                            auto glUploadTarget = MG_Util::ConvertTextureUploadTargetToGLEnum(uploadTarget);
+                            auto glUploadTarget = ConvertTextureUploadTargetToBackendGLEnum(uploadTarget);
                             auto* pData = (levelDirty && levelByteSize != 0)
                                               ? textureMipmapObject->MapMipmapData(uploadTarget, level)
                                               : nullptr;
@@ -1702,19 +1709,22 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
                             DebugImpl::ErrorLopper::Clear();
                             g_GLESFuncs.glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
-                            switch (stateTextureObject->GetTarget()) {
+                            const IntVec3 uploadSize =
+                                GetBackendUploadSize(stateTextureObject->GetTarget(), levelTexelSize);
+                            switch (MapToBackendTextureTarget(stateTextureObject->GetTarget())) {
                             case TextureTarget::Texture2D:
                             case TextureTarget::TextureCubeMap:
                                 g_GLESFuncs.glTexImage2D(
                                     glUploadTarget, static_cast<GLint>(level), (GLint)glInternalFormat,
-                                    static_cast<GLsizei>(levelTexelSize.x()), static_cast<GLsizei>(levelTexelSize.y()),
+                                    static_cast<GLsizei>(uploadSize.x()), static_cast<GLsizei>(uploadSize.y()),
                                     0, glFormat, glType, uploadData);
                                 break;
                             case TextureTarget::Texture3D:
+                            case TextureTarget::Texture2DArray:
                                 g_GLESFuncs.glTexImage3D(
                                     glUploadTarget, static_cast<GLint>(level), (GLint)glInternalFormat,
-                                    static_cast<GLsizei>(levelTexelSize.x()), static_cast<GLsizei>(levelTexelSize.y()),
-                                    static_cast<GLsizei>(levelTexelSize.z()), 0, glFormat, glType, uploadData);
+                                    static_cast<GLsizei>(uploadSize.x()), static_cast<GLsizei>(uploadSize.y()),
+                                    static_cast<GLsizei>(uploadSize.z()), 0, glFormat, glType, uploadData);
                                 break;
                             default:
                                 MGLOG_E("Unhandled texture target %s",
@@ -1778,18 +1788,20 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     } else if (stateTextureObject->IsImmutable() || m_imageBindableStorageRequired) {
                         DebugImpl::ErrorLopper::Clear();
                         g_GLESFuncs.glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
-                        switch (targetInternal) {
+                        const IntVec3 storageSize = GetBackendUploadSize(targetInternal, baseSize);
+                        switch (MapToBackendTextureTarget(targetInternal)) {
                         case TextureTarget::Texture2D:
                         case TextureTarget::TextureCubeMap:
                             g_GLESFuncs.glTexStorage2D(target, static_cast<GLsizei>(mipmapCount), glInternalFormat,
-                                                       static_cast<GLsizei>(baseSize.x()),
-                                                       static_cast<GLsizei>(baseSize.y()));
+                                                       static_cast<GLsizei>(storageSize.x()),
+                                                       static_cast<GLsizei>(storageSize.y()));
                             break;
                         case TextureTarget::Texture3D:
+                        case TextureTarget::Texture2DArray:
                             g_GLESFuncs.glTexStorage3D(target, static_cast<GLsizei>(mipmapCount), glInternalFormat,
-                                                       static_cast<GLsizei>(baseSize.x()),
-                                                       static_cast<GLsizei>(baseSize.y()),
-                                                       static_cast<GLsizei>(baseSize.z()));
+                                                       static_cast<GLsizei>(storageSize.x()),
+                                                       static_cast<GLsizei>(storageSize.y()),
+                                                       static_cast<GLsizei>(storageSize.z()));
                             break;
                         default:
                             MGLOG_E("Unhandled immutable texture target %s",
@@ -1813,7 +1825,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                 if (levelDirty && levelByteSize != 0) {
                                     auto levelTexelSize =
                                         textureMipmapObject->GetMipmapTexelSize(uploadTarget, level);
-                                    auto glUploadTarget = MG_Util::ConvertTextureUploadTargetToGLEnum(uploadTarget);
+                                    auto glUploadTarget = ConvertTextureUploadTargetToBackendGLEnum(uploadTarget);
                                     auto* pData = textureMipmapObject->MapMipmapData(uploadTarget, level);
                                     Vector<Float> convertedUploadData;
                                     const void* uploadData = PrepareNormFloatFallbackUpload(
@@ -1822,20 +1834,23 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
                                     DebugImpl::ErrorLopper::Clear();
                                     g_GLESFuncs.glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
-                                    switch (targetInternal) {
+                                    const IntVec3 uploadSize =
+                                        GetBackendUploadSize(targetInternal, levelTexelSize);
+                                    switch (MapToBackendTextureTarget(targetInternal)) {
                                     case TextureTarget::Texture2D:
                                     case TextureTarget::TextureCubeMap:
                                         g_GLESFuncs.glTexSubImage2D(
                                             glUploadTarget, static_cast<GLint>(level), 0, 0,
-                                            static_cast<GLsizei>(levelTexelSize.x()),
-                                            static_cast<GLsizei>(levelTexelSize.y()), glFormat, glType, uploadData);
+                                            static_cast<GLsizei>(uploadSize.x()),
+                                            static_cast<GLsizei>(uploadSize.y()), glFormat, glType, uploadData);
                                         break;
                                     case TextureTarget::Texture3D:
+                                    case TextureTarget::Texture2DArray:
                                         g_GLESFuncs.glTexSubImage3D(
                                             glUploadTarget, static_cast<GLint>(level), 0, 0, 0,
-                                            static_cast<GLsizei>(levelTexelSize.x()),
-                                            static_cast<GLsizei>(levelTexelSize.y()),
-                                            static_cast<GLsizei>(levelTexelSize.z()), glFormat, glType, uploadData);
+                                            static_cast<GLsizei>(uploadSize.x()),
+                                            static_cast<GLsizei>(uploadSize.y()),
+                                            static_cast<GLsizei>(uploadSize.z()), glFormat, glType, uploadData);
                                         break;
                                     default:
                                         break;
@@ -1862,7 +1877,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                 auto levelTexelSize = textureMipmapObject->GetMipmapTexelSize(uploadTarget, level);
                                 auto levelByteSize = textureMipmapObject->GetMipmapByteSize(uploadTarget, level);
                                 bool levelDirty = textureMipmapObject->IsStorageDirty(uploadTarget, level);
-                                auto glUploadTarget = MG_Util::ConvertTextureUploadTargetToGLEnum(uploadTarget);
+                                auto glUploadTarget = ConvertTextureUploadTargetToBackendGLEnum(uploadTarget);
                                 auto* pData = (levelDirty && levelByteSize != 0)
                                                   ? textureMipmapObject->MapMipmapData(uploadTarget, level)
                                                   : nullptr;
@@ -1879,22 +1894,23 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                 DebugImpl::ErrorLopper::Clear();
                                 g_GLESFuncs.glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
                                 auto textureTarget = stateTextureObject->GetTarget();
-                                // TODO: handle more texture types
-                                switch (textureTarget) {
+                                const IntVec3 uploadSize = GetBackendUploadSize(textureTarget, levelTexelSize);
+                                switch (MapToBackendTextureTarget(textureTarget)) {
                                 case TextureTarget::Texture2D:
                                 case TextureTarget::TextureCubeMap: {
                                     g_GLESFuncs.glTexImage2D(
                                         glUploadTarget, static_cast<GLint>(level), (GLint)glInternalFormat,
-                                        static_cast<GLsizei>(levelTexelSize.x()),
-                                        static_cast<GLsizei>(levelTexelSize.y()), 0, glFormat, glType, uploadData);
+                                        static_cast<GLsizei>(uploadSize.x()),
+                                        static_cast<GLsizei>(uploadSize.y()), 0, glFormat, glType, uploadData);
                                     break;
                                 }
-                                case TextureTarget::Texture3D: {
+                                case TextureTarget::Texture3D:
+                                case TextureTarget::Texture2DArray: {
                                     g_GLESFuncs.glTexImage3D(
                                         glUploadTarget, static_cast<GLint>(level), (GLint)glInternalFormat,
-                                        static_cast<GLsizei>(levelTexelSize.x()),
-                                        static_cast<GLsizei>(levelTexelSize.y()),
-                                        static_cast<GLsizei>(levelTexelSize.z()), 0, glFormat, glType, uploadData);
+                                        static_cast<GLsizei>(uploadSize.x()),
+                                        static_cast<GLsizei>(uploadSize.y()),
+                                        static_cast<GLsizei>(uploadSize.z()), 0, glFormat, glType, uploadData);
                                     break;
                                 }
                                 default: {
@@ -1961,7 +1977,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                         textureMipmapObject->GetMipmapTexelSize(uploadTarget, level).x(),
                                         textureMipmapObject->GetMipmapTexelSize(uploadTarget, level).y(), byteSize);
 
-                            auto glUploadTarget = MG_Util::ConvertTextureUploadTargetToGLEnum(uploadTarget);
+                            auto glUploadTarget = ConvertTextureUploadTargetToBackendGLEnum(uploadTarget);
                             g_GLESFuncs.glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
                             DebugImpl::ErrorLopper::Loop(
                                 [file = __FILE__, line = __LINE__, func = __func__](GLenum err) {
@@ -1974,19 +1990,22 @@ namespace MobileGL::MG_Backend::DirectGLES {
                             const void* uploadData = PrepareNormFloatFallbackUpload(
                                 textureMipmapObject->GetFormat(), texelSize, mipData, byteSize, glType,
                                 convertedUploadData);
-                            switch (stateTextureObject->GetTarget()) {
+                            const IntVec3 uploadSize =
+                                GetBackendUploadSize(stateTextureObject->GetTarget(), texelSize);
+                            switch (MapToBackendTextureTarget(stateTextureObject->GetTarget())) {
                             case TextureTarget::Texture2D:
                             case TextureTarget::TextureCubeMap:
                                 g_GLESFuncs.glTexSubImage2D(glUploadTarget, static_cast<GLint>(level), 0, 0,
-                                                            static_cast<GLsizei>(texelSize.x()),
-                                                            static_cast<GLsizei>(texelSize.y()), glFormat, glType,
+                                                            static_cast<GLsizei>(uploadSize.x()),
+                                                            static_cast<GLsizei>(uploadSize.y()), glFormat, glType,
                                                             uploadData);
                                 break;
                             case TextureTarget::Texture3D:
+                            case TextureTarget::Texture2DArray:
                                 g_GLESFuncs.glTexSubImage3D(glUploadTarget, static_cast<GLint>(level), 0, 0, 0,
-                                                            static_cast<GLsizei>(texelSize.x()),
-                                                            static_cast<GLsizei>(texelSize.y()),
-                                                            static_cast<GLsizei>(texelSize.z()), glFormat, glType,
+                                                            static_cast<GLsizei>(uploadSize.x()),
+                                                            static_cast<GLsizei>(uploadSize.y()),
+                                                            static_cast<GLsizei>(uploadSize.z()), glFormat, glType,
                                                             uploadData);
                                 break;
                             default:
@@ -2079,7 +2098,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
             MGLOG_D("Syncing texture built-in sampler with backend ID %u to backend for state ID %u",
                     m_backendTextureId, stateTextureObject->GetExternalIndex());
 
-            GLenum target = MG_Util::ConvertTextureTargetToGLEnum(stateTextureObject->GetTarget());
+            GLenum target = ConvertTextureTargetToBackendGLEnum(stateTextureObject->GetTarget());
             auto targetInternal = stateTextureObject->GetTarget();
             MGLOG_D("    Texture target for syncing is %s",
                     MG_Util::ConvertTextureTargetToString(targetInternal).c_str());
@@ -2150,6 +2169,16 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 g_GLESFuncs.glTexParameterf(target, GL_TEXTURE_MAX_LOD, samplerParams.maxLod);
                 m_cacheSamplerParameters.maxLod = samplerParams.maxLod;
             }
+            if (m_cacheSamplerParameters.maxAnisotropy != samplerParams.maxAnisotropy) {
+                if (g_GLESCapabilities.SupportsTextureFilterAnisotropy) {
+                    g_GLESFuncs.glTexParameterf(target, GL_TEXTURE_MAX_ANISOTROPY_EXT,
+                                                samplerParams.maxAnisotropy);
+                }
+                // Unsupported GLES backends intentionally treat anisotropy as a
+                // frontend-only no-op; remember the observed value so the cache
+                // remains coherent without issuing an illegal enum every sync.
+                m_cacheSamplerParameters.maxAnisotropy = samplerParams.maxAnisotropy;
+            }
             DebugImpl::ErrorLopper::Loop([file = __FILE__, line = __LINE__, func = __func__](GLenum err) {
                 MGLOG_D("%s(%s:%d) ES error %s", func, file, line, MG_Util::ConvertGLEnumToString(err).c_str());
             });
@@ -2177,7 +2206,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
             MGLOG_D("Syncing texture params with backend ID %u to backend for state ID %u", m_backendTextureId,
                     stateTextureObject->GetExternalIndex());
 
-            GLenum target = MG_Util::ConvertTextureTargetToGLEnum(stateTextureObject->GetTarget());
+            GLenum target = ConvertTextureTargetToBackendGLEnum(stateTextureObject->GetTarget());
             auto targetInternal = stateTextureObject->GetTarget();
             MGLOG_D("    Texture target for syncing is %s",
                     MG_Util::ConvertTextureTargetToString(targetInternal).c_str());
@@ -2256,14 +2285,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
             g_activeTextureUnit = unit;
         }
 
-        void UnbindTexture(Uint unit, GLenum target) { // Active unit will be modified
-            if (unit != g_activeTextureUnit) {
-                ActivateTextureUnit(unit);
-            }
-
+        void UnbindTexture(Uint unit, GLenum target) { // Activates `unit` when an unbind is issued
             auto targetN = static_cast<SizeT>(MG_Util::ConvertGLEnumToTextureTarget(target));
             if (g_boundTexturesCache[unit][targetN] == nullptr) return;
 
+            ActivateTextureUnit(unit);
             g_GLESFuncs.glBindTexture(target, 0);
             g_boundTexturesCache[unit][targetN] = nullptr;
         }
@@ -2334,11 +2360,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     g_GLESFuncs.glFramebufferTexture(glFBOTarget, glBackendAttachment,
                                                      backendTextureObject->GetBackendTextureId(),
                                                      static_cast<GLint>(attachmentObject.GetTextureLevel()));
+                } else if (const auto uploadTarget = attachmentObject.GetTextureUploadTarget();
+                           uploadTarget == TextureUploadTarget::Texture3D ||
+                           uploadTarget == TextureUploadTarget::Texture2DArray ||
+                           uploadTarget == TextureUploadTarget::Texture2DMultisampleArray) {
+                    // Single slice/layer of a 3D or array texture: ES has no
+                    // glFramebufferTexture3D, layers attach via glFramebufferTextureLayer.
+                    g_GLESFuncs.glFramebufferTextureLayer(glFBOTarget, glBackendAttachment,
+                                                          backendTextureObject->GetBackendTextureId(),
+                                                          static_cast<GLint>(attachmentObject.GetTextureLevel()),
+                                                          static_cast<GLint>(attachmentObject.GetTextureLayer()));
                 } else {
-                    auto glTextureTarget =
-                        MG_Util::ConvertTextureUploadTargetToGLEnum(attachmentObject.GetTextureUploadTarget());
+                    auto glTextureTarget = TextureImpl::ConvertTextureUploadTargetToBackendGLEnum(
+                        attachmentObject.GetTextureUploadTarget());
                     if (glTextureTarget == GL_UNKNOWN_MGL) {
-                        glTextureTarget = MG_Util::ConvertTextureTargetToGLEnum(textureObject->GetTarget());
+                        glTextureTarget = TextureImpl::ConvertTextureTargetToBackendGLEnum(textureObject->GetTarget());
                     }
                     backendTextureObject->Bind(glTextureTarget);
                     g_GLESFuncs.glFramebufferTexture2D(glFBOTarget, glBackendAttachment, glTextureTarget,
@@ -2452,7 +2488,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 drawBufferClean = true;
             }
 
-            if (!drawBufferClean) {
+            // glDrawBuffers writes the state of the FBO bound to GL_DRAW_FRAMEBUFFER.
+            // When this object is only bound as the READ target the call would land on
+            // whatever framebuffer is draw-bound AND falsely stamp this object's memo,
+            // so the later draw-target sync skips as "clean" while the real state is
+            // stale (Minecraft 26.x OIT: the scratch clear-FBO kept draw buffers NONE
+            // from its blit-destination configuration, silently dropping every
+            // offscreen color clear).
+            if (!drawBufferClean && asTarget == FramebufferTarget::Draw) {
                 memcpy(m_frontendDrawBuffers, stateDrawBuffers.data(),
                        FramebufferObject::MAX_DRAW_BUFFERS * sizeof(FramebufferAttachmentType));
                 std::fill(m_backendDrawBuffers, m_backendDrawBuffers + FramebufferObject::MAX_DRAW_BUFFERS, GL_NONE);
@@ -2477,6 +2520,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     nEffectiveBuffers = i + 1;
                 }
                 g_GLESFuncs.glDrawBuffers(nEffectiveBuffers, m_backendDrawBuffers);
+                MGLOG_D("DBAPPLY beFbo=%u target=%d n=%d db0=0x%x feDb0=%d", m_backendFBOId, (int)asTarget,
+                        nEffectiveBuffers, m_backendDrawBuffers[0], (int)stateDrawBuffers[0]);
             }
 
             if (asTarget == FramebufferTarget::Draw) {
@@ -2499,9 +2544,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 PrgramImpl::g_unormFallbackClampOutputMask = unormClampOutputMask;
             }
 
-            // 2. Remap read buffer
+            // 2. Remap read buffer. glReadBuffer writes the READ-bound FBO's state, so
+            // only apply (and stamp the memo) when this object is bound as READ.
             auto frontendReadBuf = stateFBOObject->GetReadBuffer();
-            if (frontendReadBuf != m_frontendReadBuffer) {
+            if (frontendReadBuf != m_frontendReadBuffer && asTarget == FramebufferTarget::Read) {
                 m_frontendReadBuffer = frontendReadBuf;
 
                 GLenum glBackendReadBuffer = GetBackendAttachmentType(frontendReadBuf);
@@ -2608,6 +2654,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
         StateBackendObjectRegistry<MG_State::GLState::FramebufferObject, BackendFramebufferObject>
             g_backendFramebufferObjects;
         Array<Uint16, SizeT(FramebufferTarget::FramebufferTargetCount)> g_fboBindVersions = {0};
+        // Tracks the bound FBO's object version (bumped on any attachment/drawbuffer change)
+        // per target: re-attaching textures or changing draw buffers on an already-bound FBO
+        // must re-sync it even when the binding-slot version has not moved.
+        Array<Uint16, SizeT(FramebufferTarget::FramebufferTargetCount)> g_fboSyncedObjectVersions = {0};
+        Array<MG_State::GLState::FramebufferObject*, SizeT(FramebufferTarget::FramebufferTargetCount)>
+            g_fboSyncedObjects = {};
     } // namespace FramebufferImpl
 
     namespace PrgramImpl {
@@ -2718,6 +2770,20 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     effectiveSpirv = &loweredSpirv;
                 }
 
+                // ESSL stage-matches uniform blocks by member precision, but SPIRV-Cross prints
+                // a RelaxedPrecision member as explicit "mediump" in the vertex stage and as
+                // UNQUALIFIED (mediump-by-default) in the fragment stage; after
+                // ForceSupporterOutput swaps the fragment header to highp, that member reads
+                // back as highp and the ES driver refuses to link ("definitions of uniform
+                // block ... do not match"). Strip the hint from block structs so both stages
+                // declare the member highp; nothing else about emission changes.
+                Vector<unsigned int> uboPrecisionSpirv;
+                if (MG_Util::ShaderTranspiler::ShaderCompiler::StripUboMemberRelaxedPrecisionForEssl(
+                        *effectiveSpirv, uboPrecisionSpirv) &&
+                    !uboPrecisionSpirv.empty()) {
+                    effectiveSpirv = &uboPrecisionSpirv;
+                }
+
                 MG_Util::ShaderTranspiler::SpvcSession spvcSession(*effectiveSpirv,
                     MG_Util::ShaderTranspiler::SessionUsageBit::Transpile);
 
@@ -2882,6 +2948,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 }
                 m_uniformBlockBackendIndices[static_cast<SizeT>(i)] = static_cast<Int>(backendBlkIdx);
                 g_GLESFuncs.glUniformBlockBinding(m_backendProgramId, backendBlkIdx, lastUBOBinding);
+                MGLOG_D("CACHE prog=%u beProg=%u blk[%d]='%s' beIdx=%u -> bePoint=%u",
+                        stateProgramObject->GetExternalIndex(), m_backendProgramId, i, name.c_str(), backendBlkIdx,
+                        lastUBOBinding);
             }
 
             m_samplerUniformBindings.clear();
@@ -3017,6 +3086,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 g_GLESFuncs.glSamplerParameterf(m_backendSamplerId, GL_TEXTURE_MAX_LOD, samplerParams.maxLod);
                 m_cacheSamplerParameters.maxLod = samplerParams.maxLod;
             }
+            if (m_cacheSamplerParameters.maxAnisotropy != samplerParams.maxAnisotropy) {
+                if (g_GLESCapabilities.SupportsTextureFilterAnisotropy) {
+                    g_GLESFuncs.glSamplerParameterf(m_backendSamplerId, GL_TEXTURE_MAX_ANISOTROPY_EXT,
+                                                    samplerParams.maxAnisotropy);
+                }
+                m_cacheSamplerParameters.maxAnisotropy = samplerParams.maxAnisotropy;
+            }
 #undef SYNC_SAMPLER_PARAM_IF_CHANGED
             m_isInitialized = true;
         }

MobileGL/MG_Backend/DirectGLES/Managers.h

@@ -15,6 +15,7 @@
 #include "MG_State/GLState/TextureState/TextureEnum.h"
 #include <MG_State/GLState/TextureState/TextureObject.h>
 #include <MG_State/GLState/Core.h>
+#include <MG_Util/Converters/MGToGL/TextureEnumConverter.h>
 
 namespace MobileGL::MG_Backend::DirectGLES {
     String EmulateBaseInstanceInVertexShader(String source, GLenum shaderType);
@@ -254,10 +255,48 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     namespace TextureImpl {
         inline Bool IsSupportedTextureTarget(TextureTarget target) {
-            if (target == TextureTarget::Texture1D || target == TextureTarget::TextureRectangle ||
-                target == TextureTarget::Texture1DArray || target == TextureTarget::Texture2DArray)
-                return false;
-            return true;
+            // Rectangle textures need non-normalized sampling ES cannot express; everything else is
+            // either native or emulated (1D -> 2D with height 1, 1D array -> 2D array, see
+            // MapToBackendTextureTarget). SPIRV-Cross already emits the matching ESSL samplers and
+            // coordinate padding for 1D/1D-array shaders.
+            return target != TextureTarget::TextureRectangle;
+        }
+
+        // ES has no 1D targets: 1D textures are stored as 2D (height 1) and 1D arrays as 2D arrays
+        // (height 1, layers in depth). Must match SPIRV-Cross's ES 1D-as-2D shader emulation.
+        inline TextureTarget MapToBackendTextureTarget(TextureTarget target) {
+            switch (target) {
+            case TextureTarget::Texture1D:
+                return TextureTarget::Texture2D;
+            case TextureTarget::Texture1DArray:
+                return TextureTarget::Texture2DArray;
+            default:
+                return target;
+            }
+        }
+
+        inline GLenum ConvertTextureTargetToBackendGLEnum(TextureTarget target) {
+            return MG_Util::ConvertTextureTargetToGLEnum(MapToBackendTextureTarget(target));
+        }
+
+        inline GLenum ConvertTextureUploadTargetToBackendGLEnum(TextureUploadTarget uploadTarget) {
+            switch (uploadTarget) {
+            case TextureUploadTarget::Texture1D:
+                return GL_TEXTURE_2D;
+            case TextureUploadTarget::Texture1DArray:
+                return GL_TEXTURE_2D_ARRAY;
+            default:
+                return MG_Util::ConvertTextureUploadTargetToGLEnum(uploadTarget);
+            }
+        }
+
+        // 1D arrays store layers in the state-side height; the ES 2D-array image keeps height 1 and
+        // moves the layer count into depth.
+        inline IntVec3 GetBackendUploadSize(TextureTarget stateTarget, const IntVec3& texelSize) {
+            if (stateTarget == TextureTarget::Texture1DArray) {
+                return {texelSize.x(), 1, texelSize.y()};
+            }
+            return texelSize;
         }
 
         inline Bool IsMultisampleTextureTarget(TextureTarget target) {
@@ -369,6 +408,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
         extern StateBackendObjectRegistry<MG_State::GLState::FramebufferObject, BackendFramebufferObject>
             g_backendFramebufferObjects;
         extern Array<Uint16, SizeT(FramebufferTarget::FramebufferTargetCount)> g_fboBindVersions;
+        // Tracks the bound FBO's object version (bumped on any attachment/drawbuffer change)
+        // per target: re-attaching textures or changing draw buffers on an already-bound FBO
+        // must re-sync it even when the binding-slot version has not moved.
+        extern Array<Uint16, SizeT(FramebufferTarget::FramebufferTargetCount)> g_fboSyncedObjectVersions;
+        extern Array<MG_State::GLState::FramebufferObject*, SizeT(FramebufferTarget::FramebufferTargetCount)>
+            g_fboSyncedObjects;
     } // namespace FramebufferImpl
 
     // Image uniforms take their unit from the layout(binding=N) qualifier baked into

MobileGL/MG_Backend/DirectGLES/Utils.cpp

@@ -18,6 +18,10 @@
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/TextureEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/FramebufferEnumConverter.h>
+#include <MG_Util/Math/HalfFloat.h>
+#include <MG_Util/Math/SmallFloat.h>
+
+#include <cmath>
 
 namespace MobileGL::MG_Backend::DirectGLES {
     namespace {
@@ -448,4 +452,317 @@ namespace MobileGL::MG_Backend::DirectGLES {
             }
         }
     } // namespace Utils
+
+    // ---- Client-format readback conversion helpers -------------------------------------------------
+    // ReadPixels/GetTexImage read a guaranteed wide RGBA(_INTEGER) layout from the ES driver and repack
+    // it on the CPU into the client's (format, type) layout. Everything here is pure byte shuffling so
+    // unit tests can assert the exact packed words; field positions follow GL 3.3 table 3.6 and mirror
+    // the GL CTS packed_pixels oracle (glcPackedPixelsTests.cpp pack_UNSIGNED_* helpers).
+    namespace ReadbackImpl {
+        using MG_Util::DecodeHalfBitsToFloat;
+        using MG_Util::EncodeFloatToHalfBits;
+
+        Bool GetReadbackChannelMapping(GLenum format, ReadbackChannelMapping& outMapping) {
+            switch (format) {
+            case GL_RED:          outMapping = {{0, 0, 0, 0}, 1, false}; return true;
+            case GL_RED_INTEGER:  outMapping = {{0, 0, 0, 0}, 1, true};  return true;
+            // Desktop-GL single-channel client formats (GL CTS packed_pixels rgba8_format_green/blue):
+            // the destination holds one component sourced from the named channel of the wide RGBA read.
+            // GL_ALPHA is mapped here from the raw enum because the state layer folds it into Red for the
+            // legacy alpha-texture upload hack.
+            case GL_GREEN:         outMapping = {{1, 0, 0, 0}, 1, false}; return true;
+            case GL_GREEN_INTEGER: outMapping = {{1, 0, 0, 0}, 1, true};  return true;
+            case GL_BLUE:          outMapping = {{2, 0, 0, 0}, 1, false}; return true;
+            case GL_BLUE_INTEGER:  outMapping = {{2, 0, 0, 0}, 1, true};  return true;
+            case GL_ALPHA:         outMapping = {{3, 0, 0, 0}, 1, false}; return true;
+            case GL_ALPHA_INTEGER: outMapping = {{3, 0, 0, 0}, 1, true};  return true;
+            case GL_RG:           outMapping = {{0, 1, 0, 0}, 2, false}; return true;
+            case GL_RG_INTEGER:   outMapping = {{0, 1, 0, 0}, 2, true};  return true;
+            case GL_RGB:          outMapping = {{0, 1, 2, 0}, 3, false}; return true;
+            case GL_RGB_INTEGER:  outMapping = {{0, 1, 2, 0}, 3, true};  return true;
+            case GL_BGR:          outMapping = {{2, 1, 0, 0}, 3, false}; return true;
+            case GL_BGR_INTEGER:  outMapping = {{2, 1, 0, 0}, 3, true};  return true;
+            case GL_RGBA:         outMapping = {{0, 1, 2, 3}, 4, false}; return true;
+            case GL_RGBA_INTEGER: outMapping = {{0, 1, 2, 3}, 4, true};  return true;
+            case GL_BGRA:         outMapping = {{2, 1, 0, 3}, 4, false}; return true;
+            case GL_BGRA_INTEGER: outMapping = {{2, 1, 0, 3}, 4, true};  return true;
+            default:
+                return false;
+            }
+        }
+
+        Bool GetPackedReadbackLayout(GLenum type, PackedReadbackLayout& out) {
+            switch (type) {
+            // Non-REV types pack the first format component starting at the most significant bit,
+            // *_REV types starting at the least significant bit (GL CTS pack_UNSIGNED_SHORT_5_6_5:
+            // R bits 15-11; pack_UNSIGNED_SHORT_1_5_5_5_REV: R bits 4-0, A bit 15).
+            case GL_UNSIGNED_BYTE_3_3_2:          out = {3, {3, 3, 2, 0},    {5, 2, 0, 0},    1, false}; return true;
+            case GL_UNSIGNED_BYTE_2_3_3_REV:      out = {3, {3, 3, 2, 0},    {0, 3, 6, 0},    1, false}; return true;
+            case GL_UNSIGNED_SHORT_5_6_5:         out = {3, {5, 6, 5, 0},    {11, 5, 0, 0},   2, false}; return true;
+            case GL_UNSIGNED_SHORT_5_6_5_REV:     out = {3, {5, 6, 5, 0},    {0, 5, 11, 0},   2, false}; return true;
+            case GL_UNSIGNED_SHORT_4_4_4_4:       out = {4, {4, 4, 4, 4},    {12, 8, 4, 0},   2, false}; return true;
+            case GL_UNSIGNED_SHORT_4_4_4_4_REV:   out = {4, {4, 4, 4, 4},    {0, 4, 8, 12},   2, false}; return true;
+            case GL_UNSIGNED_SHORT_5_5_5_1:       out = {4, {5, 5, 5, 1},    {11, 6, 1, 0},   2, false}; return true;
+            case GL_UNSIGNED_SHORT_1_5_5_5_REV:   out = {4, {5, 5, 5, 1},    {0, 5, 10, 15},  2, false}; return true;
+            case GL_UNSIGNED_INT_8_8_8_8:         out = {4, {8, 8, 8, 8},    {24, 16, 8, 0},  4, false}; return true;
+            case GL_UNSIGNED_INT_8_8_8_8_REV:     out = {4, {8, 8, 8, 8},    {0, 8, 16, 24},  4, false}; return true;
+            case GL_UNSIGNED_INT_10_10_10_2:      out = {4, {10, 10, 10, 2}, {22, 12, 2, 0},  4, false}; return true;
+            case GL_UNSIGNED_INT_2_10_10_10_REV:  out = {4, {10, 10, 10, 2}, {0, 10, 20, 30}, 4, false}; return true;
+            // Packed-float RGB types: fields hold unsigned small floats; 5_9_9_9_REV's shared 5-bit
+            // exponent (bits 31-27) is emitted by EncodeSharedExponentRGB9E5, not a component field.
+            case GL_UNSIGNED_INT_10F_11F_11F_REV: out = {3, {11, 11, 10, 0}, {0, 11, 22, 0},  4, true};  return true;
+            case GL_UNSIGNED_INT_5_9_9_9_REV:     out = {3, {9, 9, 9, 0},    {0, 9, 18, 0},   4, true};  return true;
+            default:
+                return false;
+            }
+        }
+
+        SizeT GetReadbackComponentSize(GLenum type) {
+            PackedReadbackLayout packedLayout{};
+            if (GetPackedReadbackLayout(type, packedLayout)) {
+                return packedLayout.byteSize;
+            }
+            switch (type) {
+            case GL_UNSIGNED_BYTE:
+            case GL_BYTE:
+                return 1;
+            case GL_UNSIGNED_SHORT:
+            case GL_SHORT:
+            case GL_HALF_FLOAT:
+                return 2;
+            case GL_UNSIGNED_INT:
+            case GL_INT:
+            case GL_FLOAT:
+                return 4;
+            default:
+                return 0;
+            }
+        }
+
+        SizeT GetReadbackDstPixelSize(const ReadbackChannelMapping& mapping, GLenum type) {
+            PackedReadbackLayout packedLayout{};
+            if (GetPackedReadbackLayout(type, packedLayout)) {
+                if (packedLayout.fieldCount != mapping.channelCount) {
+                    return 0; // 3-field packed types pair with 3-component formats only, 4 with 4
+                }
+                if (mapping.isInteger && packedLayout.isFloatPacked) {
+                    return 0; // packed-float RGB types never pair with integer formats
+                }
+                return packedLayout.byteSize;
+            }
+            if (mapping.isInteger && (type == GL_FLOAT || type == GL_HALF_FLOAT)) {
+                return 0;
+            }
+            const SizeT componentSize = GetReadbackComponentSize(type);
+            return componentSize == 0 ? 0 : static_cast<SizeT>(mapping.channelCount) * componentSize;
+        }
+
+        namespace {
+            void WritePackedReadbackWord(Uint8* dst, Uint32 word, SizeT byteSize) {
+                switch (byteSize) {
+                case 1: {
+                    const auto out = static_cast<Uint8>(word);
+                    Memcpy(dst, &out, sizeof(out));
+                    break;
+                }
+                case 2: {
+                    const auto out = static_cast<Uint16>(word);
+                    Memcpy(dst, &out, sizeof(out));
+                    break;
+                }
+                default:
+                    Memcpy(dst, &word, sizeof(word));
+                    break;
+                }
+            }
+        } // namespace
+
+        // Shared encoders live in MG_Util/Math/SmallFloat.h so the upload conversion
+        // (PixelStoreProcessor) uses byte-identical packing; kept exported here for unit tests.
+        Uint32 EncodeFloatToUnsignedF11(Float value) { return MG_Util::EncodeFloatToUnsignedF11(value); }
+        Uint32 EncodeFloatToUnsignedF10(Float value) { return MG_Util::EncodeFloatToUnsignedF10(value); }
+        Uint32 EncodeSharedExponentRGB9E5(const Float rgb[3]) { return MG_Util::EncodeSharedExponentRGB9E5(rgb); }
+
+        void ConvertWideReadbackRow(const Uint8* src, Uint8* dst, SizeT width, GLenum wideType,
+                                    const ReadbackChannelMapping& mapping, GLenum type) {
+            PackedReadbackLayout packedLayout{};
+            const Bool isPacked = GetPackedReadbackLayout(type, packedLayout);
+            const SizeT dstComponentSize = GetReadbackComponentSize(type);
+            const SizeT dstPixelBytes = GetReadbackDstPixelSize(mapping, type);
+            const SizeT srcPixelBytes = 4 * GetReadbackComponentSize(wideType);
+
+            for (SizeT col = 0; col < width; ++col) {
+                const Uint8* srcPixel = src + col * srcPixelBytes;
+                Uint8* dstPixel = dst + col * dstPixelBytes;
+                if (mapping.isInteger) {
+                    Int64 srcValues[4];
+                    for (Int c = 0; c < 4; ++c) {
+                        srcValues[c] = wideType == GL_INT
+                                           ? static_cast<Int64>(reinterpret_cast<const Int32*>(srcPixel)[c])
+                                           : static_cast<Int64>(reinterpret_cast<const Uint32*>(srcPixel)[c]);
+                    }
+                    if (isPacked) {
+                        // Integer sources clamp each component to the unsigned range of its field
+                        // (GL 3.3 section 4.3.1 final conversion).
+                        Uint32 word = 0;
+                        for (Int ch = 0; ch < packedLayout.fieldCount; ++ch) {
+                            const Int64 fieldMax = (Int64{1} << packedLayout.width[ch]) - 1;
+                            const auto v = static_cast<Uint32>(
+                                std::clamp<Int64>(srcValues[mapping.sourceChannel[ch]], 0, fieldMax));
+                            word |= v << packedLayout.shift[ch];
+                        }
+                        WritePackedReadbackWord(dstPixel, word, packedLayout.byteSize);
+                    } else {
+                        for (Int ch = 0; ch < mapping.channelCount; ++ch) {
+                            const Int64 v = srcValues[mapping.sourceChannel[ch]];
+                            Uint8* dstComponent = dstPixel + static_cast<SizeT>(ch) * dstComponentSize;
+                            switch (type) {
+                            case GL_UNSIGNED_BYTE:
+                                *dstComponent = static_cast<Uint8>(std::clamp<Int64>(v, 0, 255));
+                                break;
+                            case GL_BYTE: {
+                                const auto out = static_cast<Int8>(std::clamp<Int64>(v, -128, 127));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_UNSIGNED_SHORT: {
+                                const auto out = static_cast<Uint16>(std::clamp<Int64>(v, 0, 65535));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_SHORT: {
+                                const auto out = static_cast<Int16>(std::clamp<Int64>(v, -32768, 32767));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_UNSIGNED_INT: {
+                                const auto out = static_cast<Uint32>(std::clamp<Int64>(v, 0, 4294967295LL));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_INT: {
+                                const auto out =
+                                    static_cast<Int32>(std::clamp<Int64>(v, -2147483648LL, 2147483647LL));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            default:
+                                break;
+                            }
+                        }
+                    }
+                } else {
+                    Float srcValues[4];
+                    switch (wideType) {
+                    case GL_UNSIGNED_BYTE:
+                        for (Int c = 0; c < 4; ++c) {
+                            srcValues[c] = static_cast<Float>(srcPixel[c]) / 255.0f;
+                        }
+                        break;
+                    case GL_BYTE:
+                        for (Int c = 0; c < 4; ++c) {
+                            srcValues[c] = std::max(
+                                static_cast<Float>(reinterpret_cast<const Int8*>(srcPixel)[c]) / 127.0f, -1.0f);
+                        }
+                        break;
+                    case GL_UNSIGNED_SHORT:
+                        for (Int c = 0; c < 4; ++c) {
+                            srcValues[c] =
+                                static_cast<Float>(reinterpret_cast<const Uint16*>(srcPixel)[c]) / 65535.0f;
+                        }
+                        break;
+                    case GL_SHORT:
+                        for (Int c = 0; c < 4; ++c) {
+                            srcValues[c] = std::max(
+                                static_cast<Float>(reinterpret_cast<const Int16*>(srcPixel)[c]) / 32767.0f, -1.0f);
+                        }
+                        break;
+                    case GL_HALF_FLOAT:
+                        for (Int c = 0; c < 4; ++c) {
+                            srcValues[c] = DecodeHalfBitsToFloat(reinterpret_cast<const Uint16*>(srcPixel)[c]);
+                        }
+                        break;
+                    default: // GL_FLOAT
+                        for (Int c = 0; c < 4; ++c) {
+                            srcValues[c] = reinterpret_cast<const Float*>(srcPixel)[c];
+                        }
+                        break;
+                    }
+                    if (isPacked) {
+                        Uint32 word = 0;
+                        if (packedLayout.isFloatPacked) {
+                            const Float fields[3] = {srcValues[mapping.sourceChannel[0]],
+                                                     srcValues[mapping.sourceChannel[1]],
+                                                     srcValues[mapping.sourceChannel[2]]};
+                            word = type == GL_UNSIGNED_INT_5_9_9_9_REV
+                                       ? EncodeSharedExponentRGB9E5(fields)
+                                       : (EncodeFloatToUnsignedF11(fields[0]) << packedLayout.shift[0]) |
+                                             (EncodeFloatToUnsignedF11(fields[1]) << packedLayout.shift[1]) |
+                                             (EncodeFloatToUnsignedF10(fields[2]) << packedLayout.shift[2]);
+                        } else {
+                            // Normalized encode: round(clamp(v, 0, 1) * (2^bits - 1)) into each field.
+                            for (Int ch = 0; ch < packedLayout.fieldCount; ++ch) {
+                                const auto fieldMax = static_cast<Float>((1u << packedLayout.width[ch]) - 1u);
+                                const auto v = static_cast<Uint32>(std::llround(
+                                    std::clamp(srcValues[mapping.sourceChannel[ch]], 0.0f, 1.0f) * fieldMax));
+                                word |= v << packedLayout.shift[ch];
+                            }
+                        }
+                        WritePackedReadbackWord(dstPixel, word, packedLayout.byteSize);
+                    } else {
+                        for (Int ch = 0; ch < mapping.channelCount; ++ch) {
+                            const Float v = srcValues[mapping.sourceChannel[ch]];
+                            Uint8* dstComponent = dstPixel + static_cast<SizeT>(ch) * dstComponentSize;
+                            switch (type) {
+                            case GL_UNSIGNED_BYTE:
+                                *dstComponent =
+                                    static_cast<Uint8>(std::llround(std::clamp(v, 0.0f, 1.0f) * 255.0));
+                                break;
+                            case GL_BYTE: {
+                                const auto out =
+                                    static_cast<Int8>(std::llround(std::clamp(v, -1.0f, 1.0f) * 127.0));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_UNSIGNED_SHORT: {
+                                const auto out =
+                                    static_cast<Uint16>(std::llround(std::clamp(v, 0.0f, 1.0f) * 65535.0));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_SHORT: {
+                                const auto out =
+                                    static_cast<Int16>(std::llround(std::clamp(v, -1.0f, 1.0f) * 32767.0));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_UNSIGNED_INT: {
+                                const auto out = static_cast<Uint32>(
+                                    std::llround(static_cast<Double>(std::clamp(v, 0.0f, 1.0f)) * 4294967295.0));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_INT: {
+                                const auto out = static_cast<Int32>(
+                                    std::llround(static_cast<Double>(std::clamp(v, -1.0f, 1.0f)) * 2147483647.0));
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            case GL_FLOAT:
+                                Memcpy(dstComponent, &v, sizeof(v));
+                                break;
+                            case GL_HALF_FLOAT: {
+                                const Uint16 out = EncodeFloatToHalfBits(v);
+                                Memcpy(dstComponent, &out, sizeof(out));
+                                break;
+                            }
+                            default:
+                                break;
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    } // namespace ReadbackImpl
 } // namespace MobileGL::MG_Backend::DirectGLES

MobileGL/MG_Backend/DirectGLES/Utils.h

@@ -45,6 +45,51 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     namespace FramebufferImpl {} // namespace FramebufferImpl
 
+    // Pure CPU helpers of the client-format readback conversion (ReadPixels/GetTexImage repack a wide
+    // RGBA(_INTEGER) read into the caller's (format, type) layout). Kept context-free so unit tests can
+    // exercise the exact packing the GL CTS packed_pixels oracle compares against.
+    namespace ReadbackImpl {
+        struct ReadbackChannelMapping {
+            Int sourceChannel[4]; // RGBA source channel feeding each destination component
+            Int channelCount;     // destination component count
+            Bool isInteger;
+        };
+        Bool GetReadbackChannelMapping(GLenum format, ReadbackChannelMapping& outMapping);
+
+        // Byte size of one destination component of `type`; packed types report the packed word size.
+        // 0 = type not supported by the conversion path.
+        SizeT GetReadbackComponentSize(GLenum type);
+
+        // Bit-field layout of a GL packed pixel type. width/shift are indexed in the client format's
+        // component order (matching ReadbackChannelMapping); shift is the LSB position of the field in
+        // the packed word: non-REV types pack the first component from the MSB, *_REV types from the
+        // LSB (GL 3.3 table 3.6; field positions mirror the GL CTS glcPackedPixelsTests pack_* oracle).
+        struct PackedReadbackLayout {
+            Int fieldCount;     // format components stored in the packed word
+            Int width[4];       // bit width of each component's field
+            Int shift[4];       // LSB bit position of each component's field
+            SizeT byteSize;     // packed word size in bytes (1, 2 or 4)
+            Bool isFloatPacked; // 10F_11F_11F_REV / 5_9_9_9_REV: fields hold unsigned small floats
+        };
+        Bool GetPackedReadbackLayout(GLenum type, PackedReadbackLayout& out);
+
+        // Unsigned small-float encoders (EXT_packed_float / EXT_texture_shared_exponent semantics).
+        Uint32 EncodeFloatToUnsignedF11(Float value);
+        Uint32 EncodeFloatToUnsignedF10(Float value);
+        Uint32 EncodeSharedExponentRGB9E5(const Float rgb[3]);
+
+        // Destination bytes per pixel for a (format mapping, type) readback pair; 0 when the pair is
+        // not convertible (unknown type, packed field count != format component count, floating-point
+        // or packed-float type with an integer format).
+        SizeT GetReadbackDstPixelSize(const ReadbackChannelMapping& mapping, GLenum type);
+
+        // Repacks one row of wide RGBA(_INTEGER) texels (4 components of wideType each) into the
+        // client's (format, type) layout. src holds width * 4 * GetReadbackComponentSize(wideType)
+        // bytes, dst receives width * GetReadbackDstPixelSize(mapping, type) bytes.
+        void ConvertWideReadbackRow(const Uint8* src, Uint8* dst, SizeT width, GLenum wideType,
+                                    const ReadbackChannelMapping& mapping, GLenum type);
+    } // namespace ReadbackImpl
+
     namespace PrgramImpl {
         String ProcessOutColorLocations(const String& glslCode);
         String ForceSupporterOutput(const String& glslCode);

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.cpp

@@ -488,14 +488,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     .TargetGLSLVersion = {4, 6, 0},
                     // Baseline advertisement (no shader subgroup, no timer queries); a
                     // live backend reconciles its copy in UpdateAdvertisedExtensions.
-                    .Extensions = BuildAdvertisedExtensions(false, false),
+                    .Extensions = BuildAdvertisedExtensions(false, false, false),
                     .IsCompatibilityProfile = false
                 },
             .StaticBackendCapability = {.AllowVSOnlyPrograms = false}};
         return rendererInfo;
     }
 
-    Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported) {
+    Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
+                                                  Bool anisotropicFilteringSupported) {
         Vector<GLExtension> extensions = {V_OpenGL30, V_OpenGL31, V_OpenGL32,
                                           V_OpenGL33, 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,
@@ -516,6 +517,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         if (timerQueriesSupported && !MG_Config::Features.DisableTimerQuery) {
             extensions.push_back(E_GL_ARB_timer_query);
         }
+        // Only advertised when the samplerAnisotropy device feature was granted: without it the
+        // sampler state is accepted but never applied, and an app trusting the string (LWJGL builds
+        // GLCapabilities from it) would think it enabled anisotropic filtering.
+        if (anisotropicFilteringSupported) {
+            extensions.push_back(E_GL_EXT_texture_filter_anisotropic);
+            extensions.push_back(E_GL_ARB_texture_filter_anisotropic);
+        }
         return extensions;
     }
 
@@ -630,7 +638,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // run without a renderer; no timer query is advertised then. Rebuilding
         // the whole list keeps re-runs idempotent.
         m_rendererInfo.RendererGLInfo.Extensions = BuildAdvertisedExtensions(
-            m_vulkanCaps.SupportsShaderSubgroup, pVulkanRenderer && pVulkanRenderer->IsTimerQuerySupported());
+            m_vulkanCaps.SupportsShaderSubgroup, pVulkanRenderer && pVulkanRenderer->IsTimerQuerySupported(),
+            pVulkanRenderer && pVulkanRenderer->IsSamplerAnisotropySupported());
     }
 
     void BackendObject_DirectVulkan::UpdateDynamicBackendParameters() {
@@ -680,6 +689,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_dynamicParameters.UniformBufferOffsetAlignment = m_vulkanCaps.UniformBufferOffsetAlignment;
         m_dynamicParameters.AliasedLineWidthRangeMin = m_vulkanCaps.AliasedLineWidthRangeMin;
         m_dynamicParameters.AliasedLineWidthRangeMax = m_vulkanCaps.AliasedLineWidthRangeMax;
+        // Without the samplerAnisotropy feature the limit is unusable, so report 1.0 (no anisotropy)
+        // rather than a maximum the sampler manager will never apply.
+        m_dynamicParameters.MaxTextureMaxAnisotropy =
+            (pVulkanRenderer && pVulkanRenderer->IsSamplerAnisotropySupported()) ? m_vulkanCaps.MaxSamplerAnisotropy
+                                                                                : 1.0f;
         m_dynamicParameters.SmoothLineWidthRangeMin = m_vulkanCaps.SmoothLineWidthRangeMin;
         m_dynamicParameters.SmoothLineWidthRangeMax = m_vulkanCaps.SmoothLineWidthRangeMax;
         m_dynamicParameters.SmoothLineWidthGranularity = m_vulkanCaps.SmoothLineWidthGranularity;

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.h

@@ -73,7 +73,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // a device with the given raw capabilities. The MOBILEGL_DISABLE_SUBGROUP and
     // 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);
+    Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
+                                                  Bool anisotropicFilteringSupported);
 
     // 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/PipelineFactory.cpp

@@ -8,6 +8,7 @@
 
 #include "PipelineFactory.h"
 
+
 namespace MobileGL::MG_Backend::DirectVulkan {
     static const char* PrimitiveTopologyToString(VkPrimitiveTopology topology) {
         switch (topology) {
@@ -108,6 +109,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                   "vkCreatePipelineCache");
     }
 
+    void PipelineFactory::SetSuppressBlendedDepthWrite(Bool enabled) {
+        s_suppressBlendedDepthWrite = enabled;
+    }
+
     PipelineFactory::~PipelineFactory() {
         DestroyAll();
         if (m_pipelineCache != VK_NULL_HANDLE) {
@@ -258,6 +263,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         for (Uint32 i = 0; i < payload.colorAttachmentCount; ++i) {
             colorAttachments[i] = payload.colorBlendAttachments[i];
         }
+        // Suppress depth writes on blended pipelines when the active driver cannot keep
+        // vertex positions invariant across the pipelines of a multi-pass depth-equality
+        // chain (see SetSuppressBlendedDepthWrite). Blended draws that write depth are rare
+        // and the equality-dependent prepass pattern is exactly the case that breaks.
+        if (s_suppressBlendedDepthWrite && depthStencil.depthWriteEnable == VK_TRUE) {
+            for (Uint32 i = 0; i < payload.colorAttachmentCount; ++i) {
+                if (colorAttachments[i].blendEnable == VK_TRUE) {
+                    depthStencil.depthWriteEnable = VK_FALSE;
+                    break;
+                }
+            }
+        }
         VkPipelineColorBlendStateCreateInfo blend{VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO};
         blend.logicOpEnable = payload.logicOpEnable ? VK_TRUE : VK_FALSE;
         blend.logicOp = payload.logicOp;

MobileGL/MG_Backend/DirectVulkan/Renderer/PipelineFactory.h

@@ -62,6 +62,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         VkPipeline GetOrCreatePipeline(const PipelineCreatePayload& payload);
         void DestroyAll();
 
+        // Driver quirk: suppress depth writes on blended pipelines. Multi-pass depth-equality
+        // rendering (a blended prepass writes depth that later passes re-test with an
+        // equality-inclusive compare on the re-rasterized geometry) requires cross-pipeline
+        // position invariance that some mobile compilers do not provide, even with the
+        // SPIR-V Invariant decoration; whole primitives then drop out of the later passes.
+        // Set at renderer initialization based on the active driver.
+        static void SetSuppressBlendedDepthWrite(Bool enabled);
+
     private:
         VkPipeline CreatePipeline(const PipelineCreatePayload& payload) const;
 
@@ -70,5 +78,6 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         VkPipelineCache m_pipelineCache = VK_NULL_HANDLE;
         UnorderedMap<HashType, VkPipeline> m_cache;
         static inline XXH64_state_t* m_hashState = XXH64_createState();
+        static inline Bool s_suppressBlendedDepthWrite = false;
     };
 } // namespace MobileGL::MG_Backend::DirectVulkan

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp

@@ -1697,6 +1697,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 moduleSpirvs[i] = spv;
             }
 
+            // GL apps depend on cross-program position invariance for multi-pass equality
+            // depth tests (MC 26.3's OIT re-draws the cloud geometry with GEQUAL against the
+            // depth its own first pass wrote); decorate Position outputs Invariant so
+            // per-pipeline compilers cannot vary the position math between passes.
+            {
+                Vector<Uint> invariantSpirv;
+                if (MG_Util::ShaderTranspiler::ShaderCompiler::DecoratePositionInvariantForVulkan(
+                        moduleSpirvs[i], invariantSpirv)) {
+                    moduleSpirvs[i] = std::move(invariantSpirv);
+                } else {
+                    MGLOG_W("ProgramFactory: position-invariant decoration failed for program %u; "
+                            "keeping the original module",
+                            program.GetExternalIndex());
+                }
+            }
+
             // glslang's relaxed-Vulkan mode aliases GL's zero-based gl_InstanceID to Vulkan's
             // gl_InstanceIndex, which wrongly includes the draw's baseInstance. Rebase vertex-stage
             // loads to (InstanceIndex - BaseInstance) so shaders observe GL semantics. Reflection

MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp

@@ -360,6 +360,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         auto& textureUnit = MG_State::pGLContext->GetTextureUnitObject(unit);
         const TextureTarget preferredTarget = programObj.samplerTextureTargetByBinding[binding];
         outTexture = textureUnit.GetBindingSlot(preferredTarget).GetBoundObject();
+        // The slot always holds at least the target's default texture (name 0). While that
+        // default has no image it is unsampleable; report it as "unbound" so callers keep
+        // taking their fallback paths instead of trying to sync a storage-less texture.
+        if (MG_State::GLState::IsUndefinedDefaultTexture(outTexture.get())) {
+            outTexture.reset();
+        }
 
         return true;
     }
@@ -380,7 +386,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         const TextureTarget preferredTarget = programObj.samplerTextureTargetByBinding[binding];
         // GetBoundObject() returns the SharedPtr by const ref; .get() reads the pointer without
         // touching the refcount (no atomic inc/dec per binding per draw).
-        return textureUnit.GetBindingSlot(preferredTarget).GetBoundObject().get();
+        MG_State::GLState::ITextureObject* texture =
+            textureUnit.GetBindingSlot(preferredTarget).GetBoundObject().get();
+        // The slot always holds at least the target's default texture (name 0). While that
+        // default has no image it is unsampleable; report it as "unbound" so the caller
+        // substitutes its fallback texture exactly like it did for the old null slot.
+        if (MG_State::GLState::IsUndefinedDefaultTexture(texture)) {
+            return nullptr;
+        }
+        return texture;
     }
 
     Bool UniformManager::ResolveTexelBufferDescriptor(const MG_State::GLState::ProgramObject& program,
@@ -599,7 +613,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
             MG_State::GLState::ITextureObject* texture = ResolveSamplerTextureRaw(program, programObj, binding);
             if (!texture) {
-                continue;
+                // ResolveSamplerDescriptor will substitute the fallback texture for this binding;
+                // include it in the sampled set so the pre-render-pass sync/transition pass covers
+                // its first use instead of leaving that work to happen inside an active pass.
+                const TextureTarget preferredTarget = programObj.samplerTextureTargetByBinding[binding];
+                if (preferredTarget != TextureTarget::Texture2D &&
+                    preferredTarget != TextureTarget::TextureRectangle) {
+                    continue;
+                }
+                texture = GetFallbackTexture(preferredTarget).get();
             }
 
             auto found = std::find(outTextures.begin(), outTextures.end(), texture);

MobileGL/MG_Backend/DirectVulkan/Renderer/VkBufferManager.cpp

@@ -197,16 +197,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return static_cast<VkBufferResource*>(bufferObject.GetBackendResource().get());
     }
 
-    SharedPtr<VkBufferResource> VkBufferManager::GetOrCreateResource(
+    VkBufferResource* VkBufferManager::GetOrCreateResource(
         const SharedPtr<MG_State::GLState::BufferObject>& bufferObject) {
-        auto existing = std::static_pointer_cast<VkBufferResource>(bufferObject->GetBackendResource());
+        // Return by raw pointer: the resource is owned for its whole lifetime by the BufferObject's
+        // backend-resource SharedPtr (already set, or set below), so callers that only dereference
+        // it avoid a static_pointer_cast + SharedPtr refcount inc/dec on every per-draw buffer bind.
+        const auto& existing = bufferObject->GetBackendResource();
         if (existing) {
-            return existing;
+            return static_cast<VkBufferResource*>(existing.get());
         }
         auto resource = MakeShared<VkBufferResource>();
+        VkBufferResource* raw = resource.get();
         bufferObject->SetBackendResource(resource);
         TrackLiveResource(resource);
-        return resource;
+        return raw;
     }
 
     void VkBufferManager::TrackLiveResource(const SharedPtr<VkBufferResource>& resource) {

MobileGL/MG_Backend/DirectVulkan/Renderer/VkBufferManager.h

@@ -124,7 +124,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     private:
         Bool InitializeTransientArenas();
         static VkBufferUsageFlags GetVkBufferUsage(BufferKind kind);
-        SharedPtr<VkBufferResource> GetOrCreateResource(const SharedPtr<MG_State::GLState::BufferObject>& bufferObject);
+        VkBufferResource* GetOrCreateResource(const SharedPtr<MG_State::GLState::BufferObject>& bufferObject);
         static VkBufferResource* ResourceOf(MG_State::GLState::BufferObject& bufferObject);
         Bool CreateResidentStorage(VkBufferResource& resource, VkDeviceSize size, VkBufferUsageFlags usage,
                                    VkMemoryPropertyFlags requiredFlags = 0);

MobileGL/MG_Backend/DirectVulkan/Renderer/VkRenderPassManager.cpp

@@ -602,6 +602,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             m_rpFastRenderPassHash == activeRenderPass->hash && !hasPendingClearOnFramebuffer()) {
             auto activeIt = m_renderPasses.find(activeRenderPass->hash);
             if (activeIt != m_renderPasses.end()) {
+                activeIt->second.lastUsedFrame = m_frameCounter;
                 return activeIt->second;
             }
         }
@@ -623,12 +624,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             m_rpFastTexEpoch = m_textureManager.GetTextureImageEpoch();
             m_rpFastRbEpoch = m_renderbufferImageEpoch;
             m_rpFastRenderPassHash = activeRenderPass->hash;
+            activeIt->second.lastUsedFrame = m_frameCounter;
             return activeIt->second;
         }
         auto hash = ComputeHash(fbo, swapchainImageIndex, true);
         auto it = m_renderPasses.find(hash);
-        if (it != m_renderPasses.end())
+        if (it != m_renderPasses.end()) {
+            it->second.lastUsedFrame = m_frameCounter;
             return it->second;
+        }
 
         Bool isDefaultFbo = fbo.IsDefaultFramebuffer();
         // Color attachment
@@ -718,6 +722,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     if (hasClear) {
                         pendingClearAttachments.emplace_back(PendingClearAttachmentInfo {
                             .attachmentIndex = attachmentIndex,
+                            .colorAttachmentSlot = i,
                             .key = VkClearManager::MakePendingClearKey(att)
                         });
                     }
@@ -974,6 +979,52 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         subpassDesc.preserveAttachmentCount = 0;
         subpassDesc.pPreserveAttachments = nullptr;
 
+        // External subpass dependencies. Without them there is NO execution/memory
+        // dependency between consecutive render passes (or a pass and a transfer)
+        // touching the same attachments when the image layout does not change — the
+        // layout-transition helper no-ops on identical layouts and no other barrier
+        // exists. Tile-based GPUs then race tile loads against the previous pass's
+        // stores (multi-pass FBO chains like MC 26.3's OIT flicker on Adreno).
+        // Conservative both-ways dependencies: prior writes (attachment output,
+        // depth/stencil, transfer, shader) are made visible to this pass's loads,
+        // and this pass's attachment writes to subsequent sampling/transfer/loads.
+        VkSubpassDependency subpassDependencies[2];
+        subpassDependencies[0] = {};
+        subpassDependencies[0].srcSubpass = VK_SUBPASS_EXTERNAL;
+        subpassDependencies[0].dstSubpass = 0;
+        subpassDependencies[0].srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
+                                              VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                                              VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT;
+        subpassDependencies[0].srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                                               VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
+                                               VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
+        subpassDependencies[0].dstStageMask = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                              VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                                              VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                                              VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
+        subpassDependencies[0].dstAccessMask =
+            VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+            VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
+            VK_ACCESS_SHADER_READ_BIT;
+        subpassDependencies[0].dependencyFlags = 0;
+        subpassDependencies[1] = {};
+        subpassDependencies[1].srcSubpass = 0;
+        subpassDependencies[1].dstSubpass = VK_SUBPASS_EXTERNAL;
+        subpassDependencies[1].srcStageMask =
+            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
+        subpassDependencies[1].srcAccessMask =
+            VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+        subpassDependencies[1].dstStageMask = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                                              VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
+                                              VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                                              VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
+                                              VK_PIPELINE_STAGE_TRANSFER_BIT;
+        subpassDependencies[1].dstAccessMask =
+            VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+            VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
+            VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
+        subpassDependencies[1].dependencyFlags = 0;
+
         // Render Pass
         VkRenderPassCreateInfo renderPassCreateInfo;
         renderPassCreateInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
@@ -983,8 +1034,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         renderPassCreateInfo.pAttachments = attachmentDescriptions.data();
         renderPassCreateInfo.subpassCount = 1;
         renderPassCreateInfo.pSubpasses = &subpassDesc;
-        renderPassCreateInfo.dependencyCount = 0;
-        renderPassCreateInfo.pDependencies = nullptr;
+        renderPassCreateInfo.dependencyCount = 2;
+        renderPassCreateInfo.pDependencies = subpassDependencies;
 
         VkRenderPass renderPass = VK_NULL_HANDLE;
         VK_VERIFY(vkCreateRenderPass(m_device, &renderPassCreateInfo, nullptr, &renderPass));
@@ -1015,7 +1066,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             hasDepthStencilAttachment,
             renderPassSampleCount,
             extent,
-            static_cast<Int>(framebufferLayers) };
+            framebufferLayers };
         MGLOG_D("VkRenderPassManager::GetOrCreateRenderPass: hash=0x%llx compatibilityHash=0x%llx attachmentCount=%u colorAttachmentCount=%u samples=%d extent=%dx%d",
                 static_cast<unsigned long long>(hash),
                 static_cast<unsigned long long>(compatibilityHash),
@@ -1025,9 +1076,36 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 extent.x(),
                 extent.y());
         auto [insertedIt, _] = m_renderPasses.emplace(hash, Move(renderPassEntry));
+        insertedIt->second.lastUsedFrame = m_frameCounter;
         return insertedIt->second;
     }
 
+    void VkRenderPassManager::OnPresent() {
+        ++m_frameCounter;
+
+        // Sweep occasionally; evict entries whose last use is far past every
+        // in-flight frame so their VkRenderPass/VkFramebuffer can be destroyed
+        // safely (RenderPassEntry's destructor releases the handles).
+        constexpr Uint64 kSweepInterval = 256;
+        constexpr Uint64 kRetireAgeFrames = 1024;
+        if ((m_frameCounter % kSweepInterval) != 0) {
+            return;
+        }
+
+        const Uint64 activeHash = s_hasActiveRenderPass ? s_activeRenderPass.hash : 0;
+        for (auto it = m_renderPasses.begin(); it != m_renderPasses.end();) {
+            const Bool isActive = s_hasActiveRenderPass && it->first == activeHash;
+            if (!isActive && m_frameCounter - it->second.lastUsedFrame > kRetireAgeFrames) {
+                if (m_rpFastValid && m_rpFastRenderPassHash == it->first) {
+                    m_rpFastValid = false;
+                }
+                it = m_renderPasses.erase(it);
+            } else {
+                ++it;
+            }
+        }
+    }
+
     Bool VkRenderPassManager::BeginRenderPass(VkCommandBuffer commandBuffer, RenderPassEntry& renderPassEntry) {
         // TODO: Transition all the attachments into proper layout before starting the render pass
         VkRenderPassBeginInfo renderPassBeginInfo;

MobileGL/MG_Backend/DirectVulkan/Renderer/VkRenderPassManager.h

@@ -27,7 +27,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     };
 
     struct PendingClearAttachmentInfo {
+        // Index into the render pass attachment descriptions (VkRenderPassBeginInfo::pClearValues space).
         Uint32 attachmentIndex = 0;
+        // Index into the subpass pColorAttachments (VkClearAttachment::colorAttachment space) — the GL
+        // draw-buffer slot. Differs from attachmentIndex when earlier slots are GL_NONE/incomplete.
+        // Only meaningful for color clears.
+        Uint32 colorAttachmentSlot = 0;
         PendingClearKey key{};
         MG_State::GLState::RenderbufferObject* renderbuffer = nullptr;
         Bool hasInlinePayload = false;
@@ -68,7 +73,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         Bool hasDepthStencilAttachment = false;
         VkSampleCountFlagBits sampleCount = VK_SAMPLE_COUNT_1_BIT;
         IntVec2 extent = {0, 0};
-        Uint32 subpass = 0;
+        // VkFramebufferCreateInfo::layers of the entry's framebuffer (>1 for layered GL attachments).
+        Uint32 layers = 1;
+        // Frame counter value of the last GetOrCreateRenderPass hit; drives cache eviction.
+        Uint64 lastUsedFrame = 0;
 
         RenderPassEntry() = default;
         RenderPassEntry(const RenderPassEntry&) = delete;
@@ -84,7 +92,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             std::swap(hasDepthStencilAttachment, that.hasDepthStencilAttachment);
             std::swap(sampleCount, that.sampleCount);
             std::swap(extent, that.extent);
-            std::swap(subpass, that.subpass);
+            std::swap(layers, that.layers);
+            std::swap(lastUsedFrame, that.lastUsedFrame);
         }
         RenderPassEntry(
             Uint64 hash,
@@ -97,7 +106,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             Uint32 colorAttachmentCount,
             Bool hasDepthStencilAttachment,
             VkSampleCountFlagBits sampleCount,
-            IntVec2 extent, int subpass):
+            IntVec2 extent, Uint32 layers):
             hash(hash),
             renderPass(renderpass),
             framebuffer(framebuffer),
@@ -109,7 +118,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             hasDepthStencilAttachment(hasDepthStencilAttachment),
             sampleCount(sampleCount),
             extent(extent),
-            subpass(subpass)
+            layers(layers)
         {}
 
         ~RenderPassEntry() {
@@ -166,6 +175,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         void QueueRenderbufferClear(const ClearAttachmentPayload& clearPayload,
                                     const MG_State::GLState::FramebufferAttachmentObject& attachment);
         void PopPendingRenderbufferClear(MG_State::GLState::RenderbufferObject* renderbuffer);
+        // Frame boundary hook: ages the render-pass cache and evicts long-unused
+        // entries (their command buffers retired many frames ago).
+        void OnPresent();
         static Bool BeginRenderPass(VkCommandBuffer commandBuffer, RenderPassEntry& renderPassEntry);
         static Bool EndRenderPass(VkCommandBuffer commandBuffer);
         static ActiveRenderPassInfo* GetActiveRenderPass();
@@ -178,6 +190,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         VkTextureManager& m_textureManager;
         SwapchainObject& m_swapchainObject;
         UnorderedMap<Uint64, RenderPassEntry> m_renderPasses;
+        // Monotonic frame counter (bumped in OnPresent) for render-pass cache aging.
+        Uint64 m_frameCounter = 0;
 
         // Bumped whenever a renderbuffer VkImage is (re)created; together with the texture
         // manager's image epoch this invalidates the render-pass fast path on any attachment

MobileGL/MG_Backend/DirectVulkan/Renderer/VkSamplerManager.cpp

@@ -58,11 +58,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
         m_device = initInfo.device;
         m_config = initInfo.config;
+        m_samplerAnisotropySupported = initInfo.samplerAnisotropySupported;
+        m_maxSamplerAnisotropy = std::max(initInfo.maxSamplerAnisotropy, 1.0f);
         MOBILEGL_ASSERT(m_device != VK_NULL_HANDLE && m_config != nullptr,
                         "VkSamplerManager::Initialize failed: invalid initialization info");
         return true;
     }
 
+    Float VkSamplerManager::ResolveEffectiveMaxAnisotropy(const MG_State::GLState::SamplerObject& sampler) const {
+        if (!m_samplerAnisotropySupported) return 1.0f;
+        // VUID-VkSamplerCreateInfo-anisotropyEnable-01071/01072: anisotropy requires both filters to
+        // be LINEAR and the value to sit within [1, limits.maxSamplerAnisotropy].
+        if (sampler.GetMinFilter() != SamplerFilterMode::Linear ||
+            sampler.GetMagFilter() != SamplerFilterMode::Linear) {
+            return 1.0f;
+        }
+        return std::clamp(sampler.GetMaxAnisotropy(), 1.0f, m_maxSamplerAnisotropy);
+    }
+
     void VkSamplerManager::Shutdown() {
         for (auto& [_, sampler] : m_samplers) {
             if (m_device != VK_NULL_HANDLE && sampler.handle != VK_NULL_HANDLE) {
@@ -99,6 +112,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         XXHASH_VERIFY(XXH64_update(m_hashState, &maxLod, sizeof(maxLod)));
         const auto lodBias = sampler.GetLodBias();
         XXHASH_VERIFY(XXH64_update(m_hashState, &lodBias, sizeof(lodBias)));
+        // The RESOLVED value, not the GL request: samplers that only differ in an anisotropy Vulkan
+        // will not apply (NEAREST filtering, or requests past the device limit) must still share one
+        // VkSampler, while two samplers that really do differ must not collide onto the first one's.
+        const auto maxAnisotropy = ResolveEffectiveMaxAnisotropy(sampler);
+        XXHASH_VERIFY(XXH64_update(m_hashState, &maxAnisotropy, sizeof(maxAnisotropy)));
         const auto compareMode = sampler.GetCompareMode();
         XXHASH_VERIFY(XXH64_update(m_hashState, &compareMode, sizeof(compareMode)));
         const auto compareFunc = ResolveCompareFunc(sampler, texture);
@@ -125,8 +143,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         samplerInfo.addressModeV = ToVkAddressMode(sampler.GetWrapT());
         samplerInfo.addressModeW = ToVkAddressMode(sampler.GetWrapR());
         samplerInfo.mipLodBias = sampler.GetLodBias();
-        samplerInfo.anisotropyEnable = VK_FALSE;
-        samplerInfo.maxAnisotropy = 1.0f;
+        // Must use the same resolver as BuildSamplerKey - a divergence would either collide two
+        // different samplers or silently create duplicates.
+        const Float maxAnisotropy = ResolveEffectiveMaxAnisotropy(sampler);
+        samplerInfo.anisotropyEnable = maxAnisotropy > 1.0f ? VK_TRUE : VK_FALSE;
+        samplerInfo.maxAnisotropy = maxAnisotropy;
         samplerInfo.compareEnable = sampler.GetCompareMode() == SamplerCompareMode::CompareToTexture ? VK_TRUE : VK_FALSE;
         samplerInfo.compareOp = ToVkCompareOp(ResolveCompareFunc(sampler, texture));
         samplerInfo.maxLod = ResolveEffectiveMaxLod(sampler);

MobileGL/MG_Backend/DirectVulkan/Renderer/VkSamplerManager.h

@@ -24,6 +24,10 @@ public:
     struct InitInfo {
         VkDevice device = VK_NULL_HANDLE;
         const VulkanRendererConfig* config = nullptr;
+        // The samplerAnisotropy device feature was requested and granted at vkCreateDevice.
+        Bool samplerAnisotropySupported = false;
+        // VkPhysicalDeviceLimits::maxSamplerAnisotropy.
+        Float maxSamplerAnisotropy = 1.0f;
     };
 
     Bool Initialize(const InitInfo& initInfo);
@@ -49,9 +53,16 @@ private:
                                                  const MG_State::GLState::ITextureObject& texture);
     static VkBorderColor ResolveVkBorderColor(const MG_State::GLState::SamplerObject& sampler,
                                               const MG_State::GLState::ITextureObject& texture);
+    // The anisotropy Vulkan will actually apply: 1.0 (i.e. disabled) unless the feature is on and
+    // the sampler filters linearly both ways, otherwise the GL request clamped to the device limit.
+    // GL happily carries GL_TEXTURE_MAX_ANISOTROPY on a NEAREST sampler (Blaze3D's blocks do exactly
+    // that) while Vulkan forbids anisotropyEnable there, so the GL value must never be forwarded raw.
+    Float ResolveEffectiveMaxAnisotropy(const MG_State::GLState::SamplerObject& sampler) const;
 
     VkDevice m_device = VK_NULL_HANDLE;
     const VulkanRendererConfig* m_config = nullptr;
+    Bool m_samplerAnisotropySupported = false;
+    Float m_maxSamplerAnisotropy = 1.0f;
     UnorderedMap<Uint64, SamplerCacheEntry> m_samplers;
     static inline XXH64_state_t* m_hashState = XXH64_createState();
 };

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp

@@ -21,9 +21,13 @@
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
 #include "MG_Util/Metrics/TextureMetrics.h"
 #include <Config.h>
+#include <algorithm>
 #include <cstdlib>
 #include <cstring>
 #include <vulkan/vulkan_core.h>
+#ifdef __ANDROID__
+#include <sys/system_properties.h>
+#endif
 
 #if defined(__APPLE__)
 #include <CoreGraphics/CoreGraphics.h>
@@ -946,6 +950,7 @@ void main() {
 }
 )";
 
+
         static Uint32 ComputeFullMipLevelCount(const IntVec3& baseTexelSize) {
             Int maxDimension = std::max<Int>(
                 baseTexelSize.x(),
@@ -1883,13 +1888,36 @@ void main() {
 
         m_pipelineFactory = MakeUnique<PipelineFactory>(m_device, m_config);
         MOBILEGL_ASSERT(m_pipelineFactory != nullptr, "PipelineFactory creation failed.");
+        {
+            // Qualcomm's pipeline compiler does not keep vertex positions invariant across
+            // the pipelines of a multi-pass depth-equality chain (even with the SPIR-V
+            // Invariant decoration), so a blended depth-writing prepass makes later
+            // equality-compare passes drop whole primitives (MC 26.3 improved-transparency
+            // clouds flicker black). Suppress blended depth writes there; the env variable
+            // forces the quirk on ("0") or off ("1") on any driver.
+            static constexpr Uint32 kVendorIdQualcomm = 0x5143;
+            Bool suppressBlendedDepthWrite = m_physicalDevice.properties.vendorID == kVendorIdQualcomm;
+            if (const char* env = getenv("MOBILEGL_MAGMA_BLENDED_DEPTH_WRITE")) {
+                if (env[0] == '0') {
+                    suppressBlendedDepthWrite = true;
+                } else if (env[0] == '1') {
+                    suppressBlendedDepthWrite = false;
+                }
+            }
+            if (suppressBlendedDepthWrite) {
+                MGLOG_I("DirectVulkan: suppressing depth writes on blended pipelines "
+                        "(driver lacks cross-pipeline position invariance)");
+            }
+            PipelineFactory::SetSuppressBlendedDepthWrite(suppressBlendedDepthWrite);
+        }
         m_programFactory = MakeUnique<ProgramFactory>(m_device, m_config, maxProgramBindings,
                                                       m_shaderDrawParametersFeatureEnabled);
         MOBILEGL_ASSERT(m_programFactory != nullptr, "ProgramFactory creation failed.");
 
         m_samplerManager = MakeUnique<VkSamplerManager>();
         MOBILEGL_ASSERT(m_samplerManager != nullptr, "VkSamplerManager creation failed.");
-        succeeded = m_samplerManager->Initialize({m_device, &m_config});
+        succeeded = m_samplerManager->Initialize({m_device, &m_config, m_samplerAnisotropyFeatureEnabled,
+                                                  m_physicalDevice.properties.limits.maxSamplerAnisotropy});
         MOBILEGL_ASSERT(succeeded, "VkSamplerManager initialization failed.");
         succeeded = InitializeBlitResources();
         MOBILEGL_ASSERT(succeeded, "Blit pipeline resource initialization failed.");
@@ -2150,8 +2178,8 @@ void main() {
             if (!supported) {
                 // SetupDraw's pre-flight should have rejected this already; never upload a null payload.
                 MGLOG_E("UploadAndBindVertexStreams skipped: unsupported current generic vertex attribute type: "
-                        "program=%u location=%u type=0x%x",
-                        program.GetExternalIndex(), location, glType);
+                        "programHash=%llu location=%u type=0x%x",
+                        static_cast<unsigned long long>(programObj.hash), location, glType);
                 return false;
             }
 
@@ -3181,17 +3209,30 @@ void main() {
                     colorAttachmentFormat = textureResource->format;
                 }
 
-#if MOBILEGL_LOG_ACTIVE_LEVEL <= MOBILEGL_LOG_LEVEL_DEBUG
-                VkFormatProperties formatProperties{};
-                vkGetPhysicalDeviceFormatProperties(m_physicalDevice.handle, colorAttachmentFormat, &formatProperties);
-                MOBILEGL_ASSERT(
-                    (formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT) != 0,
-                    "GetOrCreatePipeline: blend is enabled on color attachment %u for format=%d textureId=%d, but the format lacks VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT (program=%u)",
-                    i,
-                    static_cast<Int>(colorAttachmentFormat),
-                    textureExternalIndex,
-                    program.GetExternalIndex());
-#endif
+                // Blending on an attachment whose format lacks
+                // VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT is invalid pipeline state
+                // (blend support is optional for e.g. 32-bit float formats on some GPUs);
+                // force-disable it instead of baking undefined behavior into the pipeline.
+                static UnorderedMap<Int, Bool> formatBlendSupport;
+                auto blendSupportIt = formatBlendSupport.find(static_cast<Int>(colorAttachmentFormat));
+                if (blendSupportIt == formatBlendSupport.end()) {
+                    VkFormatProperties formatProperties{};
+                    vkGetPhysicalDeviceFormatProperties(m_physicalDevice.handle, colorAttachmentFormat,
+                                                        &formatProperties);
+                    const Bool blendable =
+                        (formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT) != 0;
+                    blendSupportIt =
+                        formatBlendSupport.emplace(static_cast<Int>(colorAttachmentFormat), blendable).first;
+                    if (!blendable) {
+                        MGLOG_E("GetOrCreatePipeline: format=%d lacks VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT; "
+                                "disabling blending on attachments with this format (first hit: attachment %u textureId=%d program=%u)",
+                                static_cast<Int>(colorAttachmentFormat), i, textureExternalIndex,
+                                program.GetExternalIndex());
+                    }
+                }
+                if (!blendSupportIt->second) {
+                    effectiveBlendEnabled = false;
+                }
             }
             // Dual-source blending (GL_SRC1_* factors from glBlendFunc paired with
             // glBindFragDataLocationIndexed) requires the dualSrcBlend device feature. It is detected at
@@ -3584,8 +3625,73 @@ void main() {
                              1, &memoryBarrier, 0, nullptr, 0, nullptr);
     }
 
+    VulkanRenderer::ScissoredClearPrep VulkanRenderer::PrepareScissoredClear(
+            const MG_State::GLState::FramebufferObject& framebuffer, VkClearRect& outClearRect) {
+        auto& frame = m_frameContext.GetCurrent();
+        if (!frame.isCommandRecording) {
+            m_frameContext.BeginCommandRecording();
+        }
+
+        auto* activeRenderPass = VkRenderPassManager::GetActiveRenderPass();
+        auto* renderPassEntry = &m_renderPassManager->GetOrCreateRenderPass(framebuffer, m_imageIndexAcquired);
+        if (renderPassEntry->attachmentCount == 0 ||
+            renderPassEntry->extent.x() <= 0 || renderPassEntry->extent.y() <= 0) {
+            return ScissoredClearPrep::NoOp;
+        }
+
+        VkClearRect clearRect{};
+        clearRect.rect = framebuffer.IsDefaultFramebuffer()
+            ? MakeDefaultFramebufferScissorRect(MG_State::pGLContext->GetScissorBox(),
+                                                renderPassEntry->extent,
+                                                m_swapchainObject.GetPreTransform())
+            : MakeClampedScissorRect(MG_State::pGLContext->GetScissorBox(), renderPassEntry->extent);
+        clearRect.baseArrayLayer = 0;
+        // GL 3.3 §4.4.7: clearing a layered framebuffer clears every layer.
+        clearRect.layerCount = renderPassEntry->layers;
+        if (clearRect.rect.extent.width == 0 || clearRect.rect.extent.height == 0) {
+            return ScissoredClearPrep::NoOp;
+        }
+        // A scissor that covers the whole target is a whole-surface clear; the deferred loadOp
+        // path is equivalent and cheaper (no render pass churn, loadOp=CLEAR on tilers).
+        if (clearRect.rect.offset.x == 0 && clearRect.rect.offset.y == 0 &&
+            clearRect.rect.extent.width == static_cast<Uint32>(renderPassEntry->extent.x()) &&
+            clearRect.rect.extent.height == static_cast<Uint32>(renderPassEntry->extent.y())) {
+            return ScissoredClearPrep::NotNeeded;
+        }
+
+        if (activeRenderPass && !activeRenderPass->CompatibleWith(*renderPassEntry)) {
+            VkRenderPassManager::EndRenderPass(frame.commandBuffer);
+            activeRenderPass = nullptr;
+            // Re-resolve: ending the pass updates tracked attachment layouts, which feed the
+            // entry's load ops and initial layouts.
+            renderPassEntry = &m_renderPassManager->GetOrCreateRenderPass(framebuffer, m_imageIndexAcquired);
+        }
+        // A still-active pass is necessarily compatible here: the block above ended any
+        // incompatible one and nothing since can change the active pass.
+        if (activeRenderPass) {
+            // Materialize any older whole-attachment clear before applying this
+            // ordered, scissored clear.
+            ClearAttachmentsOnActiveRenderPass(frame.commandBuffer, *renderPassEntry);
+        } else {
+            const Bool began = VkRenderPassManager::BeginRenderPass(frame.commandBuffer, *renderPassEntry);
+            MOBILEGL_ASSERT(began, "%s: BeginRenderPass failed", __func__);
+            if (!began) {
+                return ScissoredClearPrep::NoOp;
+            }
+        }
+        outClearRect = clearRect;
+        return ScissoredClearPrep::Ready;
+    }
+
     void VulkanRenderer::Clear(GLbitfield mask) {
         m_clearManager->CollectGarbage();
+        if ((mask & (GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT)) == 0) {
+            return;
+        }
+        // GL 3.3 §3.1: when RASTERIZER_DISCARD is enabled, Clear and ClearBuffer* are ignored.
+        if (MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::RasterizerDiscard)) {
+            return;
+        }
         auto* fbo = MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Draw).GetBoundObject().get();
         MOBILEGL_ASSERT(fbo, "VulkanRenderer::Clear: draw framebuffer not found (fbo == nullptr)");
         if (IsUnsupportedFramebufferForDirectVulkan(*fbo)) {
@@ -3598,70 +3704,358 @@ void main() {
             .depth = MG_State::pGLContext->GetClearDepth(),
             .stencil = MG_State::pGLContext->GetClearStencil()
         };
-        m_clearManager->QueueClear(mask, payload, *fbo);
-        m_renderPassManager->QueueRenderbufferClear(mask, payload, *fbo);
+
+        // A render-pass loadOp clear always covers the complete attachment, while
+        // OpenGL glClear is clipped by GL_SCISSOR_TEST. Blaze3D relies on this for
+        // GuiItemAtlas: animated items clear only their atlas slot before being
+        // redrawn. Queueing that clear as a loadOp erases every cached static item.
+        if (MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::ScissorTest)) {
+            VkClearRect clearRect{};
+            switch (PrepareScissoredClear(*fbo, clearRect)) {
+            case ScissoredClearPrep::NoOp:
+                return;
+            case ScissoredClearPrep::NotNeeded:
+                break;  // full-coverage scissor: the deferred whole-surface path below is equivalent
+            case ScissoredClearPrep::Ready: {
+                VkClearAttachment clearAttachments[MG_State::GLState::FramebufferObject::MAX_DRAW_BUFFERS + 1];
+                Uint32 clearAttachmentCount = 0;
+
+                if ((mask & GL_COLOR_BUFFER_BIT) != 0) {
+                    const auto& drawBuffers = fbo->GetDrawBuffers();
+                    for (Uint32 drawBufferIndex = 0; drawBufferIndex < drawBuffers.size(); ++drawBufferIndex) {
+                        const auto attachmentType = drawBuffers[drawBufferIndex];
+                        if (attachmentType == FramebufferAttachmentType::None) {
+                            continue;
+                        }
+                        const auto& attachment = fbo->GetAttachment(attachmentType);
+                        if (!attachment.IsComplete()) {
+                            continue;
+                        }
+
+                        const BoolVec4 colorMask = MG_State::pGLContext->GetColorMaskIndexed(drawBufferIndex);
+                        if (!colorMask.r() && !colorMask.g() && !colorMask.b() && !colorMask.a()) {
+                            continue;
+                        }
+                        if (!colorMask.r() || !colorMask.g() || !colorMask.b() || !colorMask.a()) {
+                            MGLOG_W("DirectVulkan: scissored glClear with a partial color mask is not supported");
+                            continue;
+                        }
+
+                        MG_State::GLState::ITextureObject* colorTexture = nullptr;
+                        if (attachment.IsTexture()) {
+                            colorTexture = attachment.GetTexture().get();
+                        }
+                        VkClearAttachment clearAttachment{};
+                        clearAttachment.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+                        clearAttachment.colorAttachment = drawBufferIndex;
+                        clearAttachment.clearValue.color = {
+                            payload.color.x(), payload.color.y(), payload.color.z(),
+                            ResolveColorClearAlpha(colorTexture, payload.color.w())
+                        };
+                        clearAttachments[clearAttachmentCount++] = clearAttachment;
+                    }
+                }
+
+                VkImageAspectFlags depthStencilAspects = 0;
+                if ((mask & GL_DEPTH_BUFFER_BIT) != 0 && MG_State::pGLContext->GetDepthMask()) {
+                    const auto& depthAttachment = fbo->GetAttachment(FramebufferAttachmentType::Depth);
+                    if (depthAttachment.IsComplete()) {
+                        depthStencilAspects |= VK_IMAGE_ASPECT_DEPTH_BIT;
+                    }
+                }
+                if ((mask & GL_STENCIL_BUFFER_BIT) != 0) {
+                    const auto& stencilAttachment = fbo->GetAttachment(FramebufferAttachmentType::Stencil);
+                    if (stencilAttachment.IsComplete()) {
+                        // GL 3.3 §4.2.3: the clear is masked by the front stencil write mask.
+                        // vkCmdClearAttachments writes every bit, so only a full (8-bit stencil) or
+                        // zero mask can be expressed; treat a partial mask like a partial color mask.
+                        const Uint32 stencilWriteMask =
+                            MG_State::pGLContext->GetStencilState(StencilFace::Front).WriteMask;
+                        if ((stencilWriteMask & 0xFFu) == 0xFFu) {
+                            depthStencilAspects |= VK_IMAGE_ASPECT_STENCIL_BIT;
+                        } else if (stencilWriteMask != 0) {
+                            MGLOG_W("DirectVulkan: scissored glClear with a partial stencil write mask is not supported");
+                        }
+                    }
+                }
+                if (depthStencilAspects != 0) {
+                    VkClearAttachment clearAttachment{};
+                    clearAttachment.aspectMask = depthStencilAspects;
+                    clearAttachment.clearValue.depthStencil = {payload.depth, payload.stencil};
+                    clearAttachments[clearAttachmentCount++] = clearAttachment;
+                }
+
+                if (clearAttachmentCount != 0) {
+                    vkCmdClearAttachments(m_frameContext.GetCurrent().commandBuffer,
+                                          clearAttachmentCount, clearAttachments,
+                                          1, &clearRect);
+                }
+                return;
+            }
+            }
+        }
+
+        // GL 3.3 §4.2.3: glClear honors the write masks. Mirror the scissored path's
+        // gating for the deferred path: drop fully-masked planes, warn on partial
+        // masks vkCmdClear*/loadOp clears cannot express.
+        GLbitfield deferredMask = mask;
+        if ((deferredMask & GL_DEPTH_BUFFER_BIT) != 0 && !MG_State::pGLContext->GetDepthMask()) {
+            deferredMask &= ~static_cast<GLbitfield>(GL_DEPTH_BUFFER_BIT);
+        }
+        if ((deferredMask & GL_STENCIL_BUFFER_BIT) != 0) {
+            const Uint32 stencilWriteMask = MG_State::pGLContext->GetStencilState(StencilFace::Front).WriteMask;
+            if ((stencilWriteMask & 0xFFu) != 0xFFu) {
+                if (stencilWriteMask != 0) {
+                    MGLOG_W("DirectVulkan: deferred glClear with a partial stencil write mask is not supported");
+                }
+                deferredMask &= ~static_cast<GLbitfield>(GL_STENCIL_BUFFER_BIT);
+            }
+        }
+        if ((deferredMask & GL_COLOR_BUFFER_BIT) != 0) {
+            const auto& drawBuffers = fbo->GetDrawBuffers();
+            Bool anyFullMask = false;
+            Bool anyRestrictedMask = false;
+            for (Uint32 drawBufferIndex = 0; drawBufferIndex < drawBuffers.size(); ++drawBufferIndex) {
+                if (drawBuffers[drawBufferIndex] == FramebufferAttachmentType::None) {
+                    continue;
+                }
+                const BoolVec4 colorMask = MG_State::pGLContext->GetColorMaskIndexed(drawBufferIndex);
+                const Bool full = colorMask.r() && colorMask.g() && colorMask.b() && colorMask.a();
+                if (full) {
+                    anyFullMask = true;
+                } else {
+                    anyRestrictedMask = true;
+                    if (colorMask.r() || colorMask.g() || colorMask.b() || colorMask.a()) {
+                        MGLOG_W("DirectVulkan: deferred glClear with a partial color mask is not supported");
+                    }
+                }
+            }
+            if (!anyFullMask) {
+                deferredMask &= ~static_cast<GLbitfield>(GL_COLOR_BUFFER_BIT);
+            } else if (anyRestrictedMask) {
+                // Mixed per-buffer masks: queue only the fully-writable texture targets
+                // individually and drop the framebuffer-level color clear.
+                for (Uint32 drawBufferIndex = 0; drawBufferIndex < drawBuffers.size(); ++drawBufferIndex) {
+                    const auto attachmentType = drawBuffers[drawBufferIndex];
+                    if (attachmentType == FramebufferAttachmentType::None) {
+                        continue;
+                    }
+                    const BoolVec4 colorMask = MG_State::pGLContext->GetColorMaskIndexed(drawBufferIndex);
+                    if (!(colorMask.r() && colorMask.g() && colorMask.b() && colorMask.a())) {
+                        continue;
+                    }
+                    const auto& attachment = fbo->GetAttachment(attachmentType);
+                    if (attachment.IsRenderbuffer()) {
+                        m_renderPassManager->QueueRenderbufferClear(
+                            {.mask = GL_COLOR_BUFFER_BIT, .color = payload.color}, attachment);
+                    } else if (attachment.IsTexture()) {
+                        m_clearManager->QueueClear({.mask = GL_COLOR_BUFFER_BIT, .color = payload.color},
+                                                   attachment);
+                    }
+                }
+                deferredMask &= ~static_cast<GLbitfield>(GL_COLOR_BUFFER_BIT);
+            }
+        }
+        if (deferredMask == 0) {
+            return;
+        }
+
+        m_clearManager->QueueClear(deferredMask, payload, *fbo);
+        m_renderPassManager->QueueRenderbufferClear(deferredMask, payload, *fbo);
     }
 
     void VulkanRenderer::QueueClearBufferPayloadForFramebuffer(
             const MG_State::GLState::FramebufferObject& framebuffer, GLenum buffer, GLint drawbuffer,
             const ClearAttachmentPayload& clearPayload) {
         m_clearManager->CollectGarbage();
+        // GL 3.3 §3.1: when RASTERIZER_DISCARD is enabled, Clear and ClearBuffer* are ignored.
+        if (MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::RasterizerDiscard)) {
+            return;
+        }
         if (IsUnsupportedFramebufferForDirectVulkan(framebuffer)) {
             RecordUnsupportedFramebufferError(__func__);
             return;
         }
 
-        auto queueAttachmentClear = [&](FramebufferAttachmentType attachmentType) {
-            if (attachmentType == FramebufferAttachmentType::None) {
-                return;
-            }
-            const auto& attachment = framebuffer.GetAttachment(attachmentType);
-            if (attachment.IsRenderbuffer()) {
-                m_renderPassManager->QueueRenderbufferClear(clearPayload, attachment);
-                return;
-            }
-            if (!attachment.IsTexture()) {
-                return;
-            }
-            m_clearManager->QueueClear(clearPayload, attachment);
-        };
-
+        // Validate (buffer, drawbuffer) up front so GL errors fire regardless of which clear
+        // path is taken below.
         switch (buffer) {
-            case GL_COLOR: {
+            case GL_COLOR:
                 if (drawbuffer < 0 ||
                     drawbuffer >= static_cast<GLint>(MG_State::GLState::FramebufferObject::MAX_DRAW_BUFFERS)) {
                     RecordClearBufferError(__func__, ErrorCode::InvalidValue, "color drawbuffer index is out of range");
                     return;
                 }
-                queueAttachmentClear(framebuffer.GetDrawBuffers()[drawbuffer]);
-                return;
-            }
+                break;
             case GL_DEPTH:
                 if (drawbuffer != 0) {
                     RecordClearBufferError(__func__, ErrorCode::InvalidValue, "depth clear requires drawbuffer 0");
                     return;
                 }
-                queueAttachmentClear(FramebufferAttachmentType::Depth);
-                return;
+                break;
             case GL_STENCIL:
                 if (drawbuffer != 0) {
                     RecordClearBufferError(__func__, ErrorCode::InvalidValue, "stencil clear requires drawbuffer 0");
                     return;
                 }
-                queueAttachmentClear(FramebufferAttachmentType::Stencil);
-                return;
+                break;
             case GL_DEPTH_STENCIL:
                 if (drawbuffer != 0) {
                     RecordClearBufferError(__func__, ErrorCode::InvalidValue, "depth/stencil clear requires drawbuffer 0");
                     return;
                 }
-                queueAttachmentClear(FramebufferAttachmentType::Depth);
-                queueAttachmentClear(FramebufferAttachmentType::Stencil);
-                return;
+                break;
             default:
                 RecordClearBufferError(__func__, ErrorCode::InvalidEnum, "unsupported clear buffer target");
                 return;
         }
+
+        // GL 3.3 §4.2.3: ClearBuffer* is clipped by GL_SCISSOR_TEST exactly like Clear.
+        if (MG_State::pGLContext->IsCapabilityEnabled(CapabilityInput::ScissorTest)) {
+            VkClearRect clearRect{};
+            switch (PrepareScissoredClear(framebuffer, clearRect)) {
+            case ScissoredClearPrep::NoOp:
+                return;
+            case ScissoredClearPrep::NotNeeded:
+                break;  // full-coverage scissor: the deferred whole-surface path below is equivalent
+            case ScissoredClearPrep::Ready:
+                RecordScissoredClearBuffer(framebuffer, buffer, drawbuffer, clearPayload, clearRect);
+                return;
+            }
+        }
+
+        auto queueAttachmentClear = [&](FramebufferAttachmentType attachmentType,
+                                        const ClearAttachmentPayload& payload) {
+            if (attachmentType == FramebufferAttachmentType::None || payload.mask == 0) {
+                return;
+            }
+            const auto& attachment = framebuffer.GetAttachment(attachmentType);
+            if (attachment.IsRenderbuffer()) {
+                m_renderPassManager->QueueRenderbufferClear(payload, attachment);
+                return;
+            }
+            if (!attachment.IsTexture()) {
+                return;
+            }
+            m_clearManager->QueueClear(payload, attachment);
+        };
+
+        // GL 3.3 §4.2.3: ClearBuffer* honors the write masks like Clear. Deferred
+        // clears cannot express partial masks; warn and skip those.
+        const auto depthClearAllowed = [&]() -> Bool { return MG_State::pGLContext->GetDepthMask(); };
+        const auto stencilClearAllowed = [&]() -> Bool {
+            const Uint32 stencilWriteMask = MG_State::pGLContext->GetStencilState(StencilFace::Front).WriteMask;
+            if ((stencilWriteMask & 0xFFu) == 0xFFu) {
+                return true;
+            }
+            if (stencilWriteMask != 0) {
+                MGLOG_W("DirectVulkan: deferred glClearBuffer with a partial stencil write mask is not supported");
+            }
+            return false;
+        };
+
+        switch (buffer) {
+            case GL_COLOR: {
+                const BoolVec4 colorMask = MG_State::pGLContext->GetColorMaskIndexed(static_cast<Uint32>(drawbuffer));
+                if (!colorMask.r() && !colorMask.g() && !colorMask.b() && !colorMask.a()) {
+                    return;
+                }
+                if (!(colorMask.r() && colorMask.g() && colorMask.b() && colorMask.a())) {
+                    MGLOG_W("DirectVulkan: deferred glClearBuffer with a partial color mask is not supported");
+                    return;
+                }
+                queueAttachmentClear(framebuffer.GetDrawBuffers()[drawbuffer], clearPayload);
+                return;
+            }
+            case GL_DEPTH:
+                if (depthClearAllowed()) {
+                    queueAttachmentClear(FramebufferAttachmentType::Depth, clearPayload);
+                }
+                return;
+            case GL_STENCIL:
+                if (stencilClearAllowed()) {
+                    queueAttachmentClear(FramebufferAttachmentType::Stencil, clearPayload);
+                }
+                return;
+            case GL_DEPTH_STENCIL: {
+                ClearAttachmentPayload allowedPayload = clearPayload;
+                if (!depthClearAllowed()) {
+                    allowedPayload.mask &= ~static_cast<GLbitfield>(GL_DEPTH_BUFFER_BIT);
+                }
+                if (!stencilClearAllowed()) {
+                    allowedPayload.mask &= ~static_cast<GLbitfield>(GL_STENCIL_BUFFER_BIT);
+                }
+                if ((allowedPayload.mask & GL_DEPTH_BUFFER_BIT) != 0) {
+                    queueAttachmentClear(FramebufferAttachmentType::Depth, allowedPayload);
+                }
+                if ((allowedPayload.mask & GL_STENCIL_BUFFER_BIT) != 0) {
+                    queueAttachmentClear(FramebufferAttachmentType::Stencil, allowedPayload);
+                }
+                return;
+            }
+            default:
+                return;
+        }
+    }
+
+    void VulkanRenderer::RecordScissoredClearBuffer(const MG_State::GLState::FramebufferObject& framebuffer,
+                                                    GLenum buffer, GLint drawbuffer,
+                                                    const ClearAttachmentPayload& clearPayload,
+                                                    const VkClearRect& clearRect) {
+        VkClearAttachment clearAttachment{};
+
+        if (buffer == GL_COLOR) {
+            const auto attachmentType = framebuffer.GetDrawBuffers()[drawbuffer];
+            if (attachmentType == FramebufferAttachmentType::None) {
+                return;
+            }
+            const auto& attachment = framebuffer.GetAttachment(attachmentType);
+            if (!attachment.IsComplete()) {
+                return;
+            }
+            const BoolVec4 colorMask = MG_State::pGLContext->GetColorMaskIndexed(static_cast<Uint>(drawbuffer));
+            if (!colorMask.r() && !colorMask.g() && !colorMask.b() && !colorMask.a()) {
+                return;
+            }
+            if (!colorMask.r() || !colorMask.g() || !colorMask.b() || !colorMask.a()) {
+                MGLOG_W("DirectVulkan: scissored glClearBuffer with a partial color mask is not supported");
+                return;
+            }
+            MG_State::GLState::ITextureObject* colorTexture = nullptr;
+            if (attachment.IsTexture()) {
+                colorTexture = attachment.GetTexture().get();
+            }
+            clearAttachment.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+            clearAttachment.colorAttachment = static_cast<Uint32>(drawbuffer);
+            clearAttachment.clearValue.color = {
+                clearPayload.color.x(), clearPayload.color.y(), clearPayload.color.z(),
+                ResolveColorClearAlpha(colorTexture, clearPayload.color.w())
+            };
+        } else {
+            VkImageAspectFlags aspects = 0;
+            if ((clearPayload.mask & GL_DEPTH_BUFFER_BIT) != 0 && MG_State::pGLContext->GetDepthMask() &&
+                framebuffer.GetAttachment(FramebufferAttachmentType::Depth).IsComplete()) {
+                aspects |= VK_IMAGE_ASPECT_DEPTH_BIT;
+            }
+            if ((clearPayload.mask & GL_STENCIL_BUFFER_BIT) != 0 &&
+                framebuffer.GetAttachment(FramebufferAttachmentType::Stencil).IsComplete()) {
+                // GL 3.3 §4.2.3: the clear is masked by the front stencil write mask (see Clear).
+                const Uint32 stencilWriteMask =
+                    MG_State::pGLContext->GetStencilState(StencilFace::Front).WriteMask;
+                if ((stencilWriteMask & 0xFFu) == 0xFFu) {
+                    aspects |= VK_IMAGE_ASPECT_STENCIL_BIT;
+                } else if (stencilWriteMask != 0) {
+                    MGLOG_W("DirectVulkan: scissored glClearBuffer with a partial stencil write mask is not supported");
+                }
+            }
+            if (aspects == 0) {
+                return;
+            }
+            clearAttachment.aspectMask = aspects;
+            clearAttachment.clearValue.depthStencil = {clearPayload.depth, clearPayload.stencil};
+        }
+
+        vkCmdClearAttachments(m_frameContext.GetCurrent().commandBuffer, 1, &clearAttachment, 1, &clearRect);
     }
 
     void VulkanRenderer::QueueClearBufferPayload(GLenum buffer, GLint drawbuffer,
@@ -3676,7 +4070,8 @@ void main() {
     void VulkanRenderer::ClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil) {
         ClearAttachmentPayload payload{};
         payload.mask = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
-        payload.depth = depth;
+        // Vulkan clear values require depth in [0,1] (VUID-VkClearDepthStencilValue-depth-00022).
+        payload.depth = std::clamp(depth, 0.0f, 1.0f);
         payload.stencil = static_cast<Uint32>(stencil);
         QueueClearBufferPayload(buffer, drawbuffer, payload);
     }
@@ -3693,7 +4088,7 @@ void main() {
                 break;
             case GL_DEPTH:
                 payload.mask = GL_DEPTH_BUFFER_BIT;
-                payload.depth = value[0];
+                payload.depth = std::clamp(value[0], 0.0f, 1.0f);
                 break;
             default:
                 break;
@@ -3715,7 +4110,7 @@ void main() {
                 break;
             case GL_DEPTH:
                 payload.mask = GL_DEPTH_BUFFER_BIT;
-                payload.depth = value[0];
+                payload.depth = std::clamp(value[0], 0.0f, 1.0f);
                 break;
             default:
                 break;
@@ -3731,7 +4126,8 @@ void main() {
         }
         ClearAttachmentPayload payload{};
         payload.mask = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
-        payload.depth = depth;
+        // Vulkan clear values require depth in [0,1] (VUID-VkClearDepthStencilValue-depth-00022).
+        payload.depth = std::clamp(depth, 0.0f, 1.0f);
         payload.stencil = static_cast<Uint32>(stencil);
         QueueClearBufferPayloadForFramebuffer(*framebuffer, buffer, drawbuffer, payload);
     }
@@ -4101,6 +4497,21 @@ void main() {
                                 sourceTexture->GetExternalIndex());
             }
 
+            if (!drawIsDefaultFbo) {
+                // A clear queued for the destination predates this blit in API order;
+                // execute it now, or its deferred materialization would later stomp the
+                // copied contents (MC 26.3 OIT clears cloud_depth, then blits the main
+                // depth into it - the stale loadOp=CLEAR erased the copy).
+                const auto destAttachmentType = ResolveFramebufferCopyAttachmentType(*drawFbo, false, dstBinding.aspectMask);
+                const auto& destAttachment = drawFbo->GetAttachment(destAttachmentType);
+                auto destTexture = destAttachment.GetTexture();
+                MOBILEGL_ASSERT(destTexture != nullptr, "BlitFramebuffer: depth destination texture attachment is null");
+                const Bool dstClearReady = MaterializePendingClearForTexture(frame.commandBuffer, *destTexture);
+                MOBILEGL_ASSERT(dstClearReady,
+                                "BlitFramebuffer: failed to materialize pending clear for depth destination textureId=%d",
+                                destTexture->GetExternalIndex());
+            }
+
             const VkImageLayout srcOriginalLayout = readIsDefaultFbo
                 ? m_swapchainObject.GetDepthStencilImageLayout(m_imageIndexAcquired)
                 : *srcBinding.trackedLayout;
@@ -4237,6 +4648,19 @@ void main() {
                             sourceTexture->GetExternalIndex());
         }
 
+        if (!drawIsDefaultFbo) {
+            // A clear queued for the destination predates this blit in API order; execute
+            // it now, or its deferred materialization would later stomp the blitted color.
+            const auto& destAttachment = drawFbo->GetAttachment(drawFbo->GetDrawBuffers()[0]);
+            auto destTexture = destAttachment.GetTexture();
+            if (destTexture != nullptr) {
+                const Bool dstClearReady = MaterializePendingClearForTexture(frame.commandBuffer, *destTexture);
+                MOBILEGL_ASSERT(dstClearReady,
+                                "BlitFramebuffer: failed to materialize pending clear for destination textureId=%d",
+                                destTexture->GetExternalIndex());
+            }
+        }
+
         VkImageLayout srcLayout = readIsDefaultFbo
             ? m_swapchainObject.GetImageLayout(m_imageIndexAcquired)
             : *srcBinding.trackedLayout;
@@ -4431,6 +4855,15 @@ void main() {
                             sourceTexture->GetExternalIndex());
         }
 
+        {
+            // A clear queued for the destination predates this copy in API order;
+            // execute it now so the deferred materialization cannot stomp the copy.
+            const Bool dstClearReady = MaterializePendingClearForTexture(frame.commandBuffer, *destinationTexture);
+            MOBILEGL_ASSERT(dstClearReady,
+                            "CopyTexSubImage2D: failed to materialize pending clear for destination textureId=%d",
+                            destinationTexture->GetExternalIndex());
+        }
+
         const Bool srcUsesSwapchainDepth = readIsDefaultFbo && (srcBinding.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) == 0;
         const VkImageLayout srcOriginalLayout = readIsDefaultFbo
             ? (srcUsesSwapchainDepth
@@ -5861,6 +6294,7 @@ void main() {
     void VulkanRenderer::Present() {
         MOBILEGL_ASSERT(m_imageIndexAcquired < m_swapchainObject.GetImageCount(),
                         "Present, acquired image index out of range");
+        m_renderPassManager->OnPresent();
         auto& frame = m_frameContext.GetCurrent();
         auto* activeRenderPass = VkRenderPassManager::GetActiveRenderPass();
         if (activeRenderPass)
@@ -6257,6 +6691,10 @@ void main() {
         deviceFeatures.multiDrawIndirect = supportedDeviceFeatures.multiDrawIndirect;
         m_multiDrawIndirectFeatureEnabled = deviceFeatures.multiDrawIndirect == VK_TRUE;
         m_logicOpFeatureEnabled = deviceFeatures.logicOp == VK_TRUE;
+        // Backs GL_TEXTURE_MAX_ANISOTROPY_EXT; optional in Vulkan, so the sampler manager falls back
+        // to isotropic filtering (and the extension goes unadvertised) when the device lacks it.
+        deviceFeatures.samplerAnisotropy = supportedDeviceFeatures.samplerAnisotropy;
+        m_samplerAnisotropyFeatureEnabled = deviceFeatures.samplerAnisotropy == VK_TRUE;
 
         VkDeviceCreateInfo deviceCreateInfo{};
         deviceCreateInfo.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
@@ -6752,7 +7190,8 @@ void main() {
                 static_cast<Uint32>(activeRenderPass->extent.y())
         };
         clearRect.baseArrayLayer = 0;
-        clearRect.layerCount = 1;
+        // Compatible entries share the framebuffer layer count; layered attachments clear every layer.
+        clearRect.layerCount = compatibleRenderPassEntry.layers;
 
         for (const auto& pending : compatibleRenderPassEntry.pendingClearAttachments) {
             if (!pending.hasInlinePayload && pending.key.texture == nullptr) {
@@ -6773,7 +7212,9 @@ void main() {
             clearAttachment.clearValue.depthStencil = {1.0f, 0};
             if ((clearPayload.mask & GL_COLOR_BUFFER_BIT) != 0) {
                 clearAttachment.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
-                clearAttachment.colorAttachment = pending.attachmentIndex;
+                // VkClearAttachment::colorAttachment indexes the subpass pColorAttachments (draw-buffer
+                // slot space, with UNUSED holes), not the compacted attachment descriptions.
+                clearAttachment.colorAttachment = pending.colorAttachmentSlot;
                 clearAttachment.clearValue.color = {
                         clearPayload.color.x(),
                         clearPayload.color.y(),

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.h

@@ -131,6 +131,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         void ClearAttachmentsOnActiveRenderPass(VkCommandBuffer commandBuffer,
                                                 const RenderPassEntry& compatibleRenderPassEntry);
 
+        enum class ScissoredClearPrep {
+            NotNeeded,  // scissor covers the whole target — take the deferred whole-surface path instead
+            NoOp,       // nothing to clear (degenerate target or empty scissor rect)
+            Ready,      // a render pass is active; record vkCmdClearAttachments with the returned rect
+        };
+        ScissoredClearPrep PrepareScissoredClear(const MG_State::GLState::FramebufferObject& framebuffer,
+                                                 VkClearRect& outClearRect);
+
         void Clear(GLbitfield mask);
         void ClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil);
         void ClearBufferfv(GLenum buffer, GLint drawbuffer, const GLfloat* value);
@@ -219,6 +227,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // frontend. Timestamp support (queue timestampValidBits > 0 and a
         // non-zero timestampPeriod) is cached at device creation.
         Bool IsTimerQuerySupported() const;
+        // 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; }
         // 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.
@@ -275,6 +286,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         void QueueClearBufferPayloadForFramebuffer(const MG_State::GLState::FramebufferObject& framebuffer,
                                                   GLenum buffer, GLint drawbuffer,
                                                   const ClearAttachmentPayload& clearPayload);
+        void RecordScissoredClearBuffer(const MG_State::GLState::FramebufferObject& framebuffer,
+                                        GLenum buffer, GLint drawbuffer,
+                                        const ClearAttachmentPayload& clearPayload,
+                                        const VkClearRect& clearRect);
 
         // ---- Submission fence tracking (GL sync objects) ----
         // One record per vkQueueSubmit still in flight, in ascending submit
@@ -347,6 +362,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         Bool m_indexTypeUint8ExtensionEnabled = false;
         Bool m_logicOpFeatureEnabled = false;
         Bool m_multiDrawIndirectFeatureEnabled = false;
+        Bool m_samplerAnisotropyFeatureEnabled = false;
         Bool m_shaderDrawParametersExtensionEnabled = false;
         Bool m_shaderDrawParametersFeatureEnabled = false;
         // fillModeNonSolid gates VK_POLYGON_MODE_LINE/_POINT (glPolygonMode); independentBlend gates

MobileGL/MG_Benchmark/Buffer/CMakeLists.txt

@@ -17,3 +17,4 @@ target_link_libraries(
 )
 
 add_test(NAME BufferBench COMMAND BufferBench --benchmark_counters_tabular=true)
+set_tests_properties(BufferBench PROPERTIES LABELS benchmark)

MobileGL/MG_Benchmark/CMakeLists.txt

@@ -38,6 +38,7 @@ target_link_libraries(
 )
 
 add_test(NAME SanityBench COMMAND SanityBench --benchmark_counters_tabular=true)
+set_tests_properties(SanityBench PROPERTIES LABELS benchmark)
 
 add_subdirectory(Program)
 add_subdirectory(Buffer)

MobileGL/MG_Benchmark/Program/CMakeLists.txt

@@ -17,3 +17,4 @@ target_link_libraries(
 )
 
 add_test(NAME ProgramBench COMMAND ProgramBench --benchmark_counters_tabular=true)
+set_tests_properties(ProgramBench PROPERTIES LABELS benchmark)

MobileGL/MG_Impl/GLImpl/Buffer/GL_Buffer.cpp

@@ -441,7 +441,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         for (SizeT i = 0; i < static_cast<SizeT>(n); ++i) {
             Uint bufferName = buffers[i];
             if (bufferName == 0) continue;
-            if (!BufferImpl::ValidateBufferName(bufferName, true)) continue;
+            // GL 3.3 core 2.9: names that do not correspond to an existing buffer are silently
+            // ignored here, so probe with the non-recording query - the shared validator would
+            // record INVALID_OPERATION, which is only correct on the bind path.
+            if (!MG_State::pGLContext->ValidateBufferName(bufferName)) continue;
             MG_State::pGLContext->MarkBufferObjectForDeletion(bufferName);
         }
     }

MobileGL/MG_Impl/GLImpl/Drawing/GL_Drawing.cpp

@@ -67,7 +67,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         const auto& vao = MG_State::pGLContext->GetBoundVertexArray();
-        if (MG_State::pEGLContext->IsCurrentContextOpenGLCoreProfile() && vao && vao->GetExternalIndex() == 0) {
+        if (vao && vao->GetExternalIndex() == 0 && !MG_State::IsRelaxedSemanticsActive()) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,

MobileGL/MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.cpp

@@ -60,6 +60,87 @@ namespace MobileGL::MG_Impl::GLImpl {
             return false;
         }
 
+        // Whether the backend can actually attach this color format to a framebuffer. Preferred
+        // source of truth is the backend's probed format-capability cache (real glCheckFramebufferStatus
+        // probes, so extensions like EXT_render_snorm are respected). Formats a probe-less backend
+        // cannot answer for fall back to a conservative static list of formats no ES driver renders to:
+        // shared-exponent, SNORM, three-channel norm16/float32/sRGB and three-channel integer formats.
+        // Desktop GL treats those as texture-only too (not in the GL 3.3 required-renderable list), so
+        // reporting GL_FRAMEBUFFER_UNSUPPORTED for them is legal.
+        Bool IsColorInternalFormatRenderable(TextureInternalFormat format) {
+            const SizeT formatIndex = static_cast<SizeT>(format);
+            if (MG_Backend::pActiveBackendObject && formatIndex < MG_Backend::kFormatCapabilityFormatCount) {
+                const auto& cache = MG_Backend::pActiveBackendObject->GetFormatCapabilities();
+                const SizeT sentinelFormat = static_cast<SizeT>(TextureInternalFormat::RGBA8);
+                Bool cachePopulated = false;
+                for (SizeT targetIndex = 0; targetIndex < MG_Backend::kFormatCapabilityTargetCount && !cachePopulated;
+                     ++targetIndex) {
+                    cachePopulated = MG_Backend::HasFormatCapability(cache.FullCaps[targetIndex][sentinelFormat],
+                                                                     MG_Backend::FormatCapability::Creatable);
+                }
+                if (cachePopulated) {
+                    for (SizeT targetIndex = 0; targetIndex < MG_Backend::kFormatCapabilityTargetCount;
+                         ++targetIndex) {
+                        if (MG_Backend::HasFormatCapability(cache.FullCaps[targetIndex][formatIndex],
+                                                            MG_Backend::FormatCapability::FramebufferRenderable) ||
+                            MG_Backend::HasFormatCapability(cache.CaveatCaps[targetIndex][formatIndex],
+                                                            MG_Backend::FormatCapability::FramebufferRenderable)) {
+                            return true;
+                        }
+                    }
+                    return false;
+                }
+            }
+            switch (format) {
+            case TextureInternalFormat::RGB9E5:
+            case TextureInternalFormat::R8Snorm:
+            case TextureInternalFormat::RG8Snorm:
+            case TextureInternalFormat::RGB8Snorm:
+            case TextureInternalFormat::RGBA8Snorm:
+            case TextureInternalFormat::R16Snorm:
+            case TextureInternalFormat::RG16Snorm:
+            case TextureInternalFormat::RGB16Snorm:
+            case TextureInternalFormat::RGBA16Snorm:
+            case TextureInternalFormat::RGB16:
+            case TextureInternalFormat::RGB10: // stored as RGB16
+            case TextureInternalFormat::RGB12: // stored as RGB16
+            case TextureInternalFormat::RGB16F:
+            case TextureInternalFormat::RGB32F:
+            case TextureInternalFormat::RGB8I:
+            case TextureInternalFormat::RGB8UI:
+            case TextureInternalFormat::RGB16I:
+            case TextureInternalFormat::RGB16UI:
+            case TextureInternalFormat::RGB32I:
+            case TextureInternalFormat::RGB32UI:
+            case TextureInternalFormat::SRGB8:
+                return false;
+            default:
+                return true;
+            }
+        }
+
+        Bool HasNonRenderableColorAttachment(const MG_State::GLState::FramebufferObject& framebufferObject) {
+            const auto& attachments = framebufferObject.GetAllAttachmentObjects();
+            for (SizeT i = 0; i < attachments.size(); ++i) {
+                const auto type = static_cast<FramebufferAttachmentType>(i);
+                if (type < FramebufferAttachmentType::Color0 || type > FramebufferAttachmentType::Color31) {
+                    continue;
+                }
+                const auto& attachment = attachments[i];
+                if (!attachment.IsValid()) continue;
+                TextureInternalFormat format = TextureInternalFormat::Unknown;
+                if (attachment.IsTexture() && attachment.GetTexture()) {
+                    format = attachment.GetTexture()->GetFormat();
+                } else if (attachment.IsRenderbuffer() && attachment.GetRenderbuffer()) {
+                    format = attachment.GetRenderbuffer()->GetInternalFormat();
+                }
+                if (format != TextureInternalFormat::Unknown && !IsColorInternalFormatRenderable(format)) {
+                    return true;
+                }
+            }
+            return false;
+        }
+
         void RecordUnsupportedFramebufferTextureAttachmentError(const char* functionName, const char* detail) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
@@ -556,6 +637,62 @@ namespace MobileGL::MG_Impl::GLImpl {
         return framebufferObject;
     }
 
+    // Attaches a single layer/slice of a 3D or array texture. The attachment model stores the layer
+    // index; the DirectGLES backend attaches it with glFramebufferTextureLayer.
+    static void AttachFramebufferTextureLayer(const char* functionName, GLenum target, GLenum attachment,
+                                              GLuint texture, GLint level, GLint layer,
+                                              TextureUploadTarget textureUploadTarget) {
+        if (target == GL_FRAMEBUFFER) {
+            target = GL_DRAW_FRAMEBUFFER;
+        }
+        if (attachment == GL_DEPTH_STENCIL_ATTACHMENT) {
+            AttachFramebufferTextureLayer(functionName, target, GL_DEPTH_ATTACHMENT, texture, level, layer,
+                                          textureUploadTarget);
+            AttachFramebufferTextureLayer(functionName, target, GL_STENCIL_ATTACHMENT, texture, level, layer,
+                                          textureUploadTarget);
+            return;
+        }
+
+        const FramebufferAttachmentType attachmentType = MG_Util::ConvertGLEnumToFramebufferAttachmentType(attachment);
+        const FramebufferTarget framebufferTarget = MG_Util::ConvertGLEnumToFramebufferTarget(target);
+        if (!FramebufferImpl::ValidateFramebufferAttachmentType(attachmentType)) return;
+        if (!FramebufferImpl::ValidateFramebufferTarget(framebufferTarget)) return;
+        if (!TextureImpl::ValidateTextureName(texture, true)) return;
+
+        auto& bindingSlot = MG_State::pGLContext->GetFramebufferBindingSlot(framebufferTarget);
+        auto& framebufferObject = bindingSlot.GetBoundObject();
+        if (!framebufferObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
+                                             "Framebuffer target is bound to no framebuffer object."));
+            return;
+        }
+
+        if (texture == 0) {
+            framebufferObject->Detach(attachmentType);
+            return;
+        }
+
+        auto& textureObject = MG_State::pGLContext->GetTextureObject(texture);
+        if (!textureObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName,
+                                             std::format("Texture object {} is not valid.", texture)));
+            return;
+        }
+        if (layer < 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", functionName, "Layer must be non-negative."));
+            return;
+        }
+
+        framebufferObject->AttachTexture(attachmentType, textureObject, textureUploadTarget, level, layer,
+                                         /*layered=*/false);
+    }
+
     void FramebufferTextureLayer_State(GLenum target, GLenum attachment, GLuint texture, GLint level, GLint layer) {
         if (texture == 0) {
             const TextureUploadTarget detachTarget = TextureUploadTarget::Texture2D;
@@ -563,10 +700,31 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
-        static_cast<void>(layer);
-        RecordUnsupportedFramebufferTextureAttachmentError(
-            __func__,
-            "Layered framebuffer texture attachments are not represented by the current framebuffer attachment model.");
+        auto& textureObject = MG_State::pGLContext->GetTextureObject(texture);
+        if (!textureObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             std::format("Texture object {} is not valid.", texture)));
+            return;
+        }
+        TextureUploadTarget textureUploadTarget = TextureUploadTarget::Unknown;
+        switch (textureObject->GetTarget()) {
+        case TextureTarget::Texture3D:
+            textureUploadTarget = TextureUploadTarget::Texture3D;
+            break;
+        case TextureTarget::Texture2DArray:
+            textureUploadTarget = TextureUploadTarget::Texture2DArray;
+            break;
+        case TextureTarget::Texture2DMultisampleArray:
+            textureUploadTarget = TextureUploadTarget::Texture2DMultisampleArray;
+            break;
+        default:
+            RecordUnsupportedFramebufferTextureAttachmentError(
+                __func__, "FramebufferTextureLayer requires a 3D, 2D array or 2D multisample array texture.");
+            return;
+        }
+        AttachFramebufferTextureLayer(__func__, target, attachment, texture, level, layer, textureUploadTarget);
     }
 
     void FramebufferTexture3D_State(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level,
@@ -578,10 +736,15 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
-        static_cast<void>(zoffset);
-        RecordUnsupportedFramebufferTextureAttachmentError(
-            __func__,
-            "3D framebuffer texture slice attachments are not represented by the current framebuffer attachment model.");
+        if (textarget != GL_TEXTURE_3D) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "FramebufferTexture3D requires GL_TEXTURE_3D."));
+            return;
+        }
+        AttachFramebufferTextureLayer(__func__, target, attachment, texture, level, zoffset,
+                                      TextureUploadTarget::Texture3D);
     }
 
     void FramebufferTexture2D_State(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level) {
@@ -1014,7 +1177,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             auto& bindingSlot = MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Draw);
             auto& fbo = bindingSlot.GetBoundObject();
             const bool isDefaultFBO = (fbo == FramebufferImpl::pDefaultFramebufferInfo->defaultFBO);
-            static GLenum bufs[] = {buf};
+            const GLenum bufs[] = {buf};
             DrawBuffersForFramebuffer_State(fbo, isDefaultFBO, 1, bufs, true);
         }
     }
@@ -1201,7 +1364,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         for (SizeT i = 0; i < static_cast<SizeT>(n); ++i) {
             Uint bufferName = renderbuffers[i];
             if (bufferName == 0) continue;
-            if (!FramebufferImpl::ValidateRenderbufferName(bufferName)) continue;
+            // GL 3.3 core 4.4.2: unknown names are silently ignored on delete; the shared bind-path
+            // validator would record INVALID_OPERATION instead.
+            if (!MG_State::pGLContext->ValidateRenderbufferName(bufferName)) continue;
             MG_State::pGLContext->MarkRenderbufferObjectForDeletion(bufferName);
         }
     }
@@ -1224,7 +1389,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         for (SizeT i = 0; i < static_cast<SizeT>(n); ++i) {
             Uint bufferName = framebuffers[i];
             if (bufferName == 0) continue;
-            if (!FramebufferImpl::ValidateFramebufferName(bufferName)) continue;
+            // GL 3.3 core 4.4.1: unknown names are silently ignored on delete; the shared bind-path
+            // validator would record INVALID_OPERATION instead.
+            if (!MG_State::pGLContext->ValidateFramebufferName(bufferName)) continue;
             MG_State::pGLContext->MarkFramebufferObjectForDeletion(bufferName);
         }
     }
@@ -1252,6 +1419,9 @@ namespace MobileGL::MG_Impl::GLImpl {
                 GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT :
                 GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT;
         }
+        if (HasNonRenderableColorAttachment(*framebufferObject)) {
+            return GL_FRAMEBUFFER_UNSUPPORTED;
+        }
         if (IsActiveBackendDirectVulkan() &&
             IsUnsupportedFramebufferForDirectVulkan(*framebufferObject)) {
             return GL_FRAMEBUFFER_UNSUPPORTED;
@@ -1277,6 +1447,9 @@ namespace MobileGL::MG_Impl::GLImpl {
                 GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT :
                 GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT;
         }
+        if (HasNonRenderableColorAttachment(*framebufferObject)) {
+            return GL_FRAMEBUFFER_UNSUPPORTED;
+        }
         if (IsActiveBackendDirectVulkan() &&
             IsUnsupportedFramebufferForDirectVulkan(*framebufferObject)) {
             return GL_FRAMEBUFFER_UNSUPPORTED;
@@ -1626,8 +1799,8 @@ namespace MobileGL::MG_Impl::GLImpl {
             return false;
         }
 
-        // Check framebuffer completeness
-        if (!framebufferObject->CheckCompleteness()) {
+        // Check framebuffer completeness (including formats the ES pipeline cannot attach)
+        if (!framebufferObject->CheckCompleteness() || HasNonRenderableColorAttachment(*framebufferObject)) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidFramebufferOperation,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ReadPixels_State", "Framebuffer is incomplete"));
@@ -1679,6 +1852,40 @@ namespace MobileGL::MG_Impl::GLImpl {
                                                  "No color buffer for color format"));
                 return false;
             }
+
+            // GL 3.3 section 4.3.1: GL_INVALID_OPERATION if format is an integer format and the read
+            // buffer is not an integer format, or vice versa (GL CTS packed_pixels expects the error
+            // for every *_INTEGER readback from a normalized attachment).
+            const auto& readAttachment = framebufferObject->GetAttachment(readBuffer);
+            TextureInternalFormat attachmentFormat = TextureInternalFormat::Unknown;
+            if (readAttachment.IsTexture() && readAttachment.GetTexture()) {
+                attachmentFormat = readAttachment.GetTexture()->GetFormat();
+            } else if (readAttachment.IsRenderbuffer() && readAttachment.GetRenderbuffer()) {
+                attachmentFormat = readAttachment.GetRenderbuffer()->GetInternalFormat();
+            }
+            if (attachmentFormat != TextureInternalFormat::Unknown &&
+                TextureImpl::IsIntegerColorInputFormat(textureInputFormat) !=
+                    TextureImpl::IsIntegerColorInternalFormat(attachmentFormat)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidOperation,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ReadPixels_State",
+                                                 "Integer-ness of format does not match the read buffer"));
+                return false;
+            }
+        }
+
+        // Packed-type/format pairing (GL CTS packed_pixels: e.g. GL_RED with GL_UNSIGNED_SHORT_5_6_5 must
+        // raise an error instead of reaching the backend). Shared with the TexImage/GetTexImage validators;
+        // runs after the depth-stencil branch above so DEPTH_STENCIL with a wrong type keeps GL_INVALID_ENUM.
+        if (!TextureImpl::ValidateClientFormatTypePairing(textureInputFormat, texturePixelDataType)) {
+            return false;
+        }
+
+        // Packed-type/format pairing (GL CTS packed_pixels: e.g. GL_RED with GL_UNSIGNED_SHORT_5_6_5 must
+        // raise an error instead of reaching the backend). Shared with the TexImage/GetTexImage validators;
+        // runs after the depth-stencil branch above so DEPTH_STENCIL with a wrong type keeps GL_INVALID_ENUM.
+        if (!TextureImpl::ValidateClientFormatTypePairing(textureInputFormat, texturePixelDataType)) {
+            return false;
         }
 
         // Check PBO state

MobileGL/MG_Impl/GLImpl/Framebuffer/Validators.cpp

@@ -76,7 +76,13 @@ namespace MobileGL::MG_Impl::GLImpl::FramebufferImpl {
     }
 
     Bool ValidateRenderbufferName(Uint index, Bool allowZero) {
-        if (index == 0 && !allowZero) {
+        if (index == 0) {
+            // Zero is never a GenRenderbuffers name, so it must not reach the name-table lookup
+            // below: where it is allowed (glBindRenderbuffer / FramebufferRenderbuffer detach) it
+            // means "unbind", and looking it up would record a bogus INVALID_OPERATION - GL CTS's
+            // per-case state reset calls glBindRenderbuffer(GL_RENDERBUFFER, 0) after every case.
+            if (allowZero) return true;
+
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidValue,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl/FramebufferImpl", "ValidateRenderbufferName",

MobileGL/MG_Impl/GLImpl/Getter/GL_Getter.cpp

@@ -529,6 +529,13 @@ namespace MobileGL::MG_Impl::GLImpl {
             params[1] = dynamicParameters.AliasedLineWidthRangeMax;
             return;
         }
+        case GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT: {
+            // EXT_texture_filter_anisotropic queries this as a float; the integer path below widens
+            // from here, so this case is the authoritative one.
+            const auto& dynamicParameters = MG_Backend::pActiveBackendObject->GetDynamicParameters();
+            params[0] = dynamicParameters.MaxTextureMaxAnisotropy;
+            return;
+        }
         case GL_ALIASED_POINT_SIZE_RANGE:
         case GL_POINT_SIZE_RANGE: {
             const auto& dynamicParameters = MG_Backend::pActiveBackendObject->GetDynamicParameters();
@@ -1727,7 +1734,11 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = 1024 * 1024; // TODO
             return;
         case GL_CONTEXT_PROFILE_MASK:
-            *params = GL_CONTEXT_CORE_PROFILE_BIT;
+            // Reports the requested context profile (EGL defaults 3.x contexts to core);
+            // MOBILEGL_RELAXED_SEMANTICS loosens behavior without changing the identity.
+            *params = MG_State::pEGLContext && MG_State::pEGLContext->IsCurrentContextOpenGLCompatibilityProfile()
+                          ? GL_CONTEXT_COMPATIBILITY_PROFILE_BIT
+                          : GL_CONTEXT_CORE_PROFILE_BIT;
             return;
         default:
             break;
@@ -1894,7 +1905,13 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = dynamicParameters.MaxTextureSize;
             break;
         case GL_MAX_UNIFORM_BUFFER_BINDINGS:
-            *params = std::max(dynamicParameters.MaxUniformBufferBindings, kFrontendMinUniformBufferBindings);
+            // Never advertise more bindings than the state layer's indexed-binding array can track
+            // (BufferState::BufferBindingPointCount): glBindBufferBase rejects indices past that
+            // capacity, and the GL CTS per-case state reset calls glBindBufferBase on every
+            // advertised index and expects no error. The floor equals the GL 3.3 core minimum
+            // (36), so the clamp never under-advertises.
+            *params = std::clamp(dynamicParameters.MaxUniformBufferBindings, kFrontendMinUniformBufferBindings,
+                                 static_cast<GLint>(MG_State::GLState::BufferBindingPointCount));
             break;
         case GL_MAX_UNIFORM_BLOCK_SIZE:
             *params = dynamicParameters.MaxUniformBlockSize;
@@ -1955,6 +1972,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_MAX_SAMPLES:
             *params = std::max(dynamicParameters.MaxSamples, kFrontendMaxSamples);
             break;
+        case GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT:
+            // Float state (see GetFloatv); rounded to nearest for the integer query per GL 3.3 6.1.2.
+            *params = static_cast<GLint>(std::lround(dynamicParameters.MaxTextureMaxAnisotropy));
+            break;
         default:
             MGLOG_E("glGetIntegerv: Invalid enum %s (0x%X)", MG_Util::ConvertGLEnumToString(pname).c_str(), pname);
             MG_State::pGLContext->RecordError(ErrorCode::InvalidEnum,

MobileGL/MG_Impl/GLImpl/Program/GL_Program.cpp

@@ -339,6 +339,9 @@ namespace MobileGL::MG_Impl::GLImpl {
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "Shader is not attached to program."));
             return;
         }
+        // A shader flagged with glDeleteShader lives on while attached; this detach may
+        // have been its last GL-visible attachment.
+        MG_State::pGLContext->ReleaseShaderNameIfOrphaned(shader);
     }
 
     void GetActiveAttrib_State(GLuint program, GLuint index, GLsizei bufSize, GLsizei* length, GLint* size,
@@ -735,6 +738,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             auto size = programObject->GetUniformSizesInBytes(location);
             char* pUBO = (char*)programObject->MapUBO();
             auto* ttype = programObject->GetUniformTType(location);
+            if (pUBO == nullptr || offset == MG_State::GLState::ProgramObject::kInvalidUniformOffset ||
+                offset + size > programObject->GetUBOSize()) {
+                MGLOG_E("%s: uniform at program %u location %d has no backing storage; returning nothing", __func__,
+                        program, location);
+                return;
+            }
 
             if (!ttype->isMatrix() || ttype->getMatrixCols() != 3)
                 Memcpy(params, pUBO + offset, size);
@@ -784,6 +793,12 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto size = programObject->GetUniformSizesInBytes(location);
         char* pUBO = static_cast<char*>(programObject->MapUBO());
         auto* ttype = programObject->GetUniformTType(location);
+        if (pUBO == nullptr || offset == MG_State::GLState::ProgramObject::kInvalidUniformOffset ||
+            offset + size > programObject->GetUBOSize()) {
+            MGLOG_E("%s: uniform at program %u location %d has no backing storage; returning nothing", __func__,
+                    program, location);
+            return;
+        }
 
         if constexpr (std::is_same_v<T, GLfloat>) {
             if (ttype->isMatrix() && ttype->getMatrixCols() == 3) {
@@ -900,14 +915,31 @@ namespace MobileGL::MG_Impl::GLImpl {
         if (!programObject.IsUniformOpaqueAtLocation(location)) {
             MGLOG_D("%s: program = %d, location = %d, maxLocation = %d", __func__, programObject.GetExternalIndex(),
                     location, programObject.GetMaxUniformLocation());
-            auto size = programObject.GetUniformSizesInBytes(location);
-            auto offset = programObject.GetUniformOffset(location);
-            MOBILEGL_ASSERT(size >= ItemCount * sizeof(T),
-                            "Uniform size mismatch, expected at least %zu bytes, got %zu bytes.", ItemCount * sizeof(T),
-                            size);
+            const SizeT size = programObject.GetUniformSizesInBytes(location);
+            const Uint offset = programObject.GetUniformOffset(location);
+            char* pUBO = static_cast<char*>(programObject.MapUBO());
+            const SizeT uboSize = programObject.GetUBOSize();
+            SizeT writeSize = ItemCount * sizeof(T);
+            if (size < writeSize) {
+                // Metadata bug: degrade to a clamped copy instead of killing the process.
+                MGLOG_E("%s: uniform size mismatch at program %u location %u: expected at least %zu bytes, got %zu "
+                        "bytes; clamping",
+                        __func__, programObject.GetExternalIndex(), location, ItemCount * sizeof(T), size);
+                writeSize = size;
+            }
+            if (pUBO == nullptr || offset == MG_State::GLState::ProgramObject::kInvalidUniformOffset ||
+                offset + byteOffsetInsideUniform + writeSize > uboSize) {
+                // Should not happen: linking gives every settable uniform backing
+                // storage. Log and drop the write instead of faulting.
+                MGLOG_E("%s: uniform at program %u location %u has no backing storage (ubo=%p offset=%u size=%zu "
+                        "uboSize=%zu); dropping write",
+                        __func__, programObject.GetExternalIndex(), location, static_cast<void*>(pUBO), offset,
+                        writeSize, uboSize);
+                return;
+            }
             MGLOG_D("%s: program = %d, location = %d, byteOffset = %d", __func__, programObject.GetExternalIndex(),
                     location, offset + byteOffsetInsideUniform);
-            Memcpy((char*)programObject.MapUBO() + offset + byteOffsetInsideUniform, value, ItemCount * sizeof(T));
+            Memcpy(pUBO + offset + byteOffsetInsideUniform, value, writeSize);
             programObject.MarkUBOContentDirty();
         } else {
             auto* ttype = programObject.GetUniformTType(location);
@@ -940,6 +972,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         for (GLint offset = 0; offset < count; offset++) {
+            if (offset > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + offset)) {
+                // GL 3.3 §2.11.4: values for elements beyond the end of the uniform
+                // array are ignored. Never step onto a neighboring uniform's location.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + offset)) {
                 RecordInvalidUniformLocationError(__func__, location + offset, "the current program object");
                 return;
@@ -964,6 +1001,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         for (GLint offset = 0; offset < count; offset++) {
+            if (offset > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + offset)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + offset)) {
                 RecordInvalidUniformLocationError(__func__, location + offset,
                                                   "program " + std::to_string(program));
@@ -1092,6 +1133,10 @@ namespace MobileGL::MG_Impl::GLImpl {
 
         // For matrix uniforms, we handle each matrix individually
         for (GLint i = 0; i < count; i++) {
+            if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + i)) {
                 RecordInvalidUniformLocationError(__func__, location + i, "the current program object");
                 return;
@@ -1124,6 +1169,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         // For matrix uniforms, we handle each matrix individually
         // Handle padding in mat3 correctly!!
         for (GLint i = 0; i < count; i++) {
+            if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + i)) {
                 RecordInvalidUniformLocationError(__func__, location + i, "the current program object");
                 return;
@@ -1159,6 +1208,10 @@ namespace MobileGL::MG_Impl::GLImpl {
 
         // For matrix uniforms, we handle each matrix individually
         for (GLint i = 0; i < count; i++) {
+            if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + i)) {
                 RecordInvalidUniformLocationError(__func__, location + i, "the current program object");
                 return;
@@ -1219,6 +1272,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         for (GLint i = 0; i < count; i++) {
+            if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + i)) {
                 RecordInvalidUniformLocationError(__func__, location + i, "program " + std::to_string(program));
                 return;
@@ -1249,6 +1306,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         for (GLint i = 0; i < count; i++) {
+            if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + i)) {
                 RecordInvalidUniformLocationError(__func__, location + i, "program " + std::to_string(program));
                 return;
@@ -1283,6 +1344,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         for (GLint i = 0; i < count; i++) {
+            if (i > 0 && !programObject->UniformLocationsAliasSameUniform(location, location + i)) {
+                // Values for elements beyond the end of the uniform array are ignored.
+                break;
+            }
             if (!programObject->IsValidUniformLocation(location + i)) {
                 RecordInvalidUniformLocationError(__func__, location + i, "program " + std::to_string(program));
                 return;
@@ -1341,6 +1406,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         const auto& index = programObject->GetUniformBlockIndex(uniformBlockName);
+        MGLOG_D("GBI prog=%u name='%s' -> %d", program, uniformBlockName ? uniformBlockName : "(null)", (Int)index);
         return index;
     }
 
@@ -1364,6 +1430,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                         std::to_string(program) + "."));
             return;
         }
+        MGLOG_D("UBB prog=%u idx=%u binding=%u", program, uniformBlockIndex, uniformBlockBinding);
         programObject->SetUniformBlockBinding(uniformBlockIndex, uniformBlockBinding);
     }
 
@@ -1435,9 +1502,13 @@ namespace MobileGL::MG_Impl::GLImpl {
             MGLOG_D("%s: GL_UNIFORM_BLOCK_REFERENCED_BY_COMPUTE_SHADER = %d", __func__, *params);
             break;
         case GL_UNIFORM_BLOCK_ACTIVE_UNIFORM_INDICES: {
+            // Member entries of an arrayed block are recorded against the first instance;
+            // every instance of the array reports that shared member set (matches
+            // GL_UNIFORM_BLOCK_ACTIVE_UNIFORMS, which scans with the same owner index).
+            const Int ownerIndex = static_cast<Int>(programObject->GetUniformBlockMemberOwnerIndex(uniformBlockIndex));
             GLint uniformIndexCount = 0;
             for (Uint uniformIndex = 0; uniformIndex < programObject->GetUniformCount(); ++uniformIndex) {
-                if (programObject->GetActiveUniformBlockIndex(uniformIndex) != static_cast<Int>(uniformBlockIndex)) {
+                if (programObject->GetActiveUniformBlockIndex(uniformIndex) != ownerIndex) {
                     continue;
                 }
                 params[uniformIndexCount++] = static_cast<GLint>(uniformIndex);

MobileGL/MG_Impl/GLImpl/RenderState/GL_RenderState.cpp

@@ -649,7 +649,9 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void ClearDepth_State(GLclampd depth) {
-        MG_State::pGLContext->SetClearDepth(static_cast<Float>(depth));
+        // GL 3.3 §4.2.3: the clear depth is clamped to [0,1] at specification time (Vulkan clear
+        // values additionally require it: VUID-VkClearDepthStencilValue-depth-00022).
+        MG_State::pGLContext->SetClearDepth(ClampUnitFloat(static_cast<Float>(depth)));
     }
 
     void ClearColor_State(GLfloat red, GLfloat green, GLfloat blue, GLfloat alpha) {

MobileGL/MG_Impl/GLImpl/Sampler/GL_Sampler.cpp

@@ -14,7 +14,13 @@
 
 namespace MobileGL::MG_Impl::GLImpl {
     namespace {
-        Bool ValidateSamplerParameterValue(GLenum pname, const void* param, Bool isFloat, Bool isInteger) {
+        Float ReadSamplerScalar(const void* param, Bool isFloat, Bool isUnsignedInteger) {
+            if (isFloat) return *(const GLfloat*)param;
+            if (isUnsignedInteger) return static_cast<Float>(*(const GLuint*)param);
+            return static_cast<Float>(*(const GLint*)param);
+        }
+
+        Bool ValidateSamplerParameterValue(GLenum pname, const void* param, Bool isFloat, Bool isUnsignedInteger) {
             if (param == nullptr) return false;
 
             switch (pname) {
@@ -22,6 +28,13 @@ namespace MobileGL::MG_Impl::GLImpl {
             case GL_TEXTURE_MAX_LOD:
             case GL_TEXTURE_LOD_BIAS:
                 return true;
+            case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+                if (ReadSamplerScalar(param, isFloat, isUnsignedInteger) >= 1.0f) return true;
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "SetSamplerParam_State",
+                                                 "GL_TEXTURE_MAX_ANISOTROPY_EXT must be at least 1.0."));
+                return false;
             default:
                 break;
             }
@@ -29,14 +42,15 @@ namespace MobileGL::MG_Impl::GLImpl {
             if (isFloat) {
                 return SamplerImpl::ValidateSamplerFloatParam(pname, *(const GLfloat*)param);
             }
-            if (isInteger) {
+            if (isUnsignedInteger) {
                 return SamplerImpl::ValidateSamplerIntParam(pname, static_cast<GLint>(*(const GLuint*)param));
             }
             return SamplerImpl::ValidateSamplerIntParam(pname, *(const GLint*)param);
         }
     } // namespace
 
-    void SetSamplerParam_State(GLuint sampler, GLenum pname, const void* param, bool isFloat, bool isInteger) {
+    void SetSamplerParam_State(GLuint sampler, GLenum pname, const void* param, bool isFloat,
+                               bool isUnsignedInteger) {
         if (param == nullptr) return;
         if (!SamplerImpl::ValidateSamplerName(sampler)) return;
 
@@ -47,7 +61,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
         auto& samplerObj = MG_State::pGLContext->GetSamplerObject(sampler);
         if (!SamplerImpl::ValidateSamplerObject(sampler)) return;
-        if (!ValidateSamplerParameterValue(pname, param, isFloat, isInteger)) return;
+        if (!ValidateSamplerParameterValue(pname, param, isFloat, isUnsignedInteger)) return;
 
         using namespace MG_Util;
         switch (pname) {
@@ -76,6 +90,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_LOD_BIAS:
             samplerObj->SetLodBias(*(const GLfloat*)param);
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            samplerObj->SetMaxAnisotropy(ReadSamplerScalar(param, isFloat, isUnsignedInteger));
+            break;
         case GL_TEXTURE_COMPARE_MODE:
             samplerObj->SetCompareMode(MG_Util::ConvertGLEnumToSamplerCompareMode(*(const GLint*)param));
             break;
@@ -89,7 +106,8 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
     }
 
-    void GetSamplerParam_State(GLuint sampler, GLenum pname, void* params, bool isFloat, bool isInteger) {
+    void GetSamplerParam_State(GLuint sampler, GLenum pname, void* params, bool isFloat,
+                               bool isUnsignedInteger) {
         if (params == nullptr) return;
         if (!SamplerImpl::ValidateSamplerName(sampler)) return;
 
@@ -129,6 +147,15 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_LOD_BIAS:
             *(GLfloat*)params = samplerObj->GetLodBias();
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            if (isFloat) {
+                *(GLfloat*)params = samplerObj->GetMaxAnisotropy();
+            } else if (isUnsignedInteger) {
+                *(GLuint*)params = static_cast<GLuint>(samplerObj->GetMaxAnisotropy());
+            } else {
+                *(GLint*)params = static_cast<GLint>(samplerObj->GetMaxAnisotropy());
+            }
+            break;
         case GL_TEXTURE_COMPARE_MODE:
             *(GLuint*)params = MG_Util::ConvertSamplerCompareModeToGLEnum(samplerObj->GetCompareMode());
             break;
@@ -206,7 +233,16 @@ namespace MobileGL::MG_Impl::GLImpl {
         if (sampler == 0) {
             textureUnit.SetSamplerObject(nullptr);
         } else {
-            if (!SamplerImpl::ValidateSamplerName(sampler)) return;
+            // GL 3.3 core 3.8.2: BindSampler on a name GenSamplers never returned - or one already
+            // deleted - is INVALID_OPERATION. SamplerParameter* raises INVALID_VALUE for the same
+            // name, which is why this cannot go through the shared SamplerImpl validator.
+            if (!MG_State::pGLContext->ValidateSamplerName(sampler)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidOperation,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "BindSampler_State",
+                                                 std::format("Invalid sampler name {}", sampler)));
+                return;
+            }
             Bool doesSamplerObjectCreated = MG_State::pGLContext->ValidateSamplerObject(sampler);
             if (!doesSamplerObjectCreated) {
                 MG_State::pGLContext->CreateSamplerObject(sampler);
@@ -240,7 +276,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void SamplerParameterIiv(GLuint sampler, GLenum pname, const GLint* param) {
-        SetSamplerParam_State(sampler, pname, param, false, true);
+        SetSamplerParam_State(sampler, pname, param, false, false);
     }
 
     void SamplerParameteriv(GLuint sampler, GLenum pname, const GLint* param) {
@@ -268,7 +304,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void GetSamplerParameterIiv(GLuint sampler, GLenum pname, GLint* params) {
-        GetSamplerParam_State(sampler, pname, params, false, true);
+        GetSamplerParam_State(sampler, pname, params, false, false);
     }
 
     void GetSamplerParameterfv(GLuint sampler, GLenum pname, GLfloat* params) {

MobileGL/MG_Impl/GLImpl/Sampler/Validators.cpp

@@ -99,6 +99,16 @@ namespace MobileGL::MG_Impl::GLImpl::SamplerImpl {
         case GL_TEXTURE_LOD_BIAS:
             return true;
 
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            if (!(param >= 1.0f)) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateSamplerFloatParam",
+                                                 "GL_TEXTURE_MAX_ANISOTROPY_EXT must be at least 1.0."));
+                return false;
+            }
+            return true;
+
         case GL_TEXTURE_BORDER_COLOR:
             if (param < 0.0f || param > 1.0f) {
                 MG_State::pGLContext->RecordError(
@@ -125,6 +135,16 @@ namespace MobileGL::MG_Impl::GLImpl::SamplerImpl {
             }
             return true;
 
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            if (param < 1) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateSamplerIntParam",
+                                                 "GL_TEXTURE_MAX_ANISOTROPY_EXT must be at least 1."));
+                return false;
+            }
+            return true;
+
         default:
             return ValidateSamplerParam(pname, static_cast<GLenum>(param));
         }

MobileGL/MG_Impl/GLImpl/Texture/GL_Texture.cpp

@@ -74,6 +74,16 @@ namespace MobileGL::MG_Impl::GLImpl {
             return true;
         }
 
+        Bool ValidateMaxAnisotropy(Float maxAnisotropy, const char* caller) {
+            if (maxAnisotropy >= 1.0f) return true;
+
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
+                                             "GL_TEXTURE_MAX_ANISOTROPY_EXT must be at least 1.0."));
+            return false;
+        }
+
         template <typename Fn>
         void WithTemporarilyBoundNamedTexture(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
                                              Fn&& fn) {
@@ -284,10 +294,16 @@ namespace MobileGL::MG_Impl::GLImpl {
                 MG_State::GLState::TextureState::MAX_TEXTURE_IMAGE_UNITS));
         }
 
-        Uint ComputeFullMipmapLevelCount(const IntVec3& baseTexelSize) {
+        // Array targets store their layer count in z; layers never participate in mip
+        // reduction (GL 3.3 §3.8.14), only true 3D textures halve their depth per level.
+        Bool DepthParticipatesInMipmapping(TextureTarget target) {
+            return target == TextureTarget::Texture3D;
+        }
+
+        Uint ComputeFullMipmapLevelCount(const IntVec3& baseTexelSize, Bool depthMips) {
             Int maxDimension = std::max<Int>(
                 baseTexelSize.x(),
-                std::max<Int>(baseTexelSize.y(), std::max<Int>(baseTexelSize.z(), 1)));
+                std::max<Int>(baseTexelSize.y(), depthMips ? std::max<Int>(baseTexelSize.z(), 1) : 1));
             Uint mipLevelCount = 1;
             while (maxDimension > 1) {
                 maxDimension = std::max<Int>(maxDimension / 2, 1);
@@ -296,11 +312,12 @@ namespace MobileGL::MG_Impl::GLImpl {
             return mipLevelCount;
         }
 
-        IntVec3 ComputeMipmapTexelSize(const IntVec3& baseTexelSize, Uint relativeLevel) {
+        IntVec3 ComputeMipmapTexelSize(const IntVec3& baseTexelSize, Uint relativeLevel, Bool depthMips) {
             return {
                 std::max<Int>(baseTexelSize.x() >> static_cast<Int>(relativeLevel), 1),
                 std::max<Int>(baseTexelSize.y() >> static_cast<Int>(relativeLevel), 1),
-                std::max<Int>(baseTexelSize.z() >> static_cast<Int>(relativeLevel), 1),
+                depthMips ? std::max<Int>(baseTexelSize.z() >> static_cast<Int>(relativeLevel), 1)
+                          : std::max<Int>(baseTexelSize.z(), 1),
             };
         }
 
@@ -323,9 +340,10 @@ namespace MobileGL::MG_Impl::GLImpl {
             }
 
             const SizeT bytesPerTexel = baseByteSize / baseTexelCount;
-            const Uint requiredLevelCount = ComputeFullMipmapLevelCount(baseTexelSize);
+            const Bool depthMips = DepthParticipatesInMipmapping(texture.GetTarget());
+            const Uint requiredLevelCount = ComputeFullMipmapLevelCount(baseTexelSize, depthMips);
             for (Uint level = 1; level < requiredLevelCount; ++level) {
-                const IntVec3 levelTexelSize = ComputeMipmapTexelSize(baseTexelSize, level);
+                const IntVec3 levelTexelSize = ComputeMipmapTexelSize(baseTexelSize, level, depthMips);
                 const SizeT levelByteSize = bytesPerTexel * static_cast<SizeT>(levelTexelSize.x()) *
                                             static_cast<SizeT>(levelTexelSize.y()) *
                                             static_cast<SizeT>(levelTexelSize.z());
@@ -404,13 +422,10 @@ namespace MobileGL::MG_Impl::GLImpl {
                                                  "2D multisample textures must use depth 1."));
                 return false;
             }
-            if (textureTarget == TextureTarget::Texture2DMultisampleArray && depth == 0) {
-                MG_State::pGLContext->RecordError(
-                    ErrorCode::InvalidValue,
-                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
-                                                 "2D multisample array textures must have at least one layer."));
-                return false;
-            }
+            // Zero layers is NOT an error for multisample arrays: depth == 0 (like width/height
+            // == 0) deallocates the image - GL 4.5 8.8 only raises INVALID_VALUE for negative
+            // dimensions, and GL CTS's per-case state reset (gluStateReset) clears the default
+            // GL_TEXTURE_2D_MULTISAMPLE_ARRAY texture with glTexImage3DMultisample(..., 0, 0, 0).
 
             const Int maxSamples = GetMaxSupportedTextureSamples(textureInternalFormat);
             if (samples > maxSamples) {
@@ -458,6 +473,9 @@ namespace MobileGL::MG_Impl::GLImpl {
     Bool ValidateTextureParameterForTarget(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
                                            GLenum pname, GLint param, const char* caller) {
         const auto target = textureObject->GetTarget();
+        if (pname == GL_TEXTURE_MAX_ANISOTROPY_EXT && !ValidateMaxAnisotropy(param, caller)) {
+            return false;
+        }
         if ((pname == GL_TEXTURE_BASE_LEVEL || pname == GL_TEXTURE_MAX_LEVEL) && param < 0) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidValue,
@@ -487,7 +505,8 @@ namespace MobileGL::MG_Impl::GLImpl {
             (pname == GL_TEXTURE_WRAP_S || pname == GL_TEXTURE_WRAP_T || pname == GL_TEXTURE_WRAP_R ||
              pname == GL_TEXTURE_MIN_FILTER || pname == GL_TEXTURE_MAG_FILTER || pname == GL_TEXTURE_MIN_LOD ||
              pname == GL_TEXTURE_MAX_LOD || pname == GL_TEXTURE_LOD_BIAS || pname == GL_TEXTURE_COMPARE_MODE ||
-             pname == GL_TEXTURE_COMPARE_FUNC || pname == GL_TEXTURE_BORDER_COLOR)) {
+             pname == GL_TEXTURE_COMPARE_FUNC || pname == GL_TEXTURE_BORDER_COLOR ||
+             pname == GL_TEXTURE_MAX_ANISOTROPY_EXT)) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
@@ -557,6 +576,14 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_SWIZZLE_A: {
             auto swizzleParam = MG_Util::ConvertGLEnumPnameToTextureSwizzleParam(pname);
             auto swizzleValue = MG_Util::ConvertGLEnumToTextureSwizzleParam(param);
+            if (swizzleValue == TextureSwizzleParam::Unknown) {
+                // GL CTS texture_swizzle.api_errors: single-value TexParameter* with a value outside
+                // [RED, GREEN, BLUE, ALPHA, ZERO, ONE] must raise GL_INVALID_ENUM.
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidEnum,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller, "Invalid texture swizzle value."));
+                return;
+            }
             textureObject->SetSwizzleParam(swizzleParam, swizzleValue);
             break;
         }
@@ -578,6 +605,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_LOD_BIAS:
             textureObject->GetSamplerObject()->SetLodBias((GLfloat)param);
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            textureObject->GetSamplerObject()->SetMaxAnisotropy(static_cast<GLfloat>(param));
+            break;
         case GL_GENERATE_MIPMAP:
             g_autoGenerateMipmapByTextureId[textureObject->GetExternalIndex()] = (param != GL_FALSE);
             break;
@@ -594,7 +624,10 @@ namespace MobileGL::MG_Impl::GLImpl {
     void TextureParameterObjectf_State(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject, GLenum pname,
                                        GLfloat param, const char* caller) {
         if (!textureObject) return;
-        if (!ValidateTextureParameterForTarget(textureObject, pname, static_cast<GLint>(param), caller)) return;
+        if (pname == GL_TEXTURE_MAX_ANISOTROPY_EXT && !ValidateMaxAnisotropy(param, caller)) return;
+        const GLint validationParam =
+            pname == GL_TEXTURE_MAX_ANISOTROPY_EXT ? 1 : static_cast<GLint>(param);
+        if (!ValidateTextureParameterForTarget(textureObject, pname, validationParam, caller)) return;
 
         switch (pname) {
         case GL_TEXTURE_MAG_FILTER:
@@ -640,6 +673,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_LOD_BIAS:
             textureObject->GetSamplerObject()->SetLodBias(param);
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            textureObject->GetSamplerObject()->SetMaxAnisotropy(param);
+            break;
         case GL_GENERATE_MIPMAP:
             g_autoGenerateMipmapByTextureId[textureObject->GetExternalIndex()] = (param != 0.0f);
             break;
@@ -709,6 +745,9 @@ namespace MobileGL::MG_Impl::GLImpl {
             *params = (GLint)MG_Util::ConvertSamplerCompareFuncToGLEnum(
                 textureObject->GetSamplerObject()->GetSamplerCompareFunc());
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            *params = static_cast<GLint>(textureObject->GetSamplerObject()->GetMaxAnisotropy());
+            break;
         default:
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidEnum,
@@ -734,6 +773,23 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
     }
 
+    // Texture-parameter lookups must not raise GL_INVALID_OPERATION when the default texture
+    // (name 0) is bound: glTexParameter* on default textures is legal GL (the GL CTS state reset
+    // sets swizzles/levels on texture 0 for every unit x target and expects glGetError() to stay
+    // clean). Name 0 resolves to the target's real default texture object, so parameters set on
+    // it are stored and queryable like on any texture.
+    const SharedPtr<MG_State::GLState::ITextureObject>& GetTextureObjectByTargetForParameter(
+        TextureUploadTarget textureUploadTarget, TextureTarget textureTarget) {
+        if (TextureImpl::IsProxyTextureTarget(textureUploadTarget)) {
+            return TextureImpl::pProxyTextureManager->GetProxyTextureObject(textureUploadTarget);
+        }
+        if (textureTarget == TextureTarget::Unknown) {
+            return nullTextureObject;
+        }
+        auto& activeUnit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
+        return activeUnit.GetBindingSlot(textureTarget).GetBoundObject();
+    }
+
     void GenerateMipmap_Backend(GLenum target) {
         MG_Backend::gBackendFunctionsTable.GL.GenerateMipmap(target);
     }
@@ -779,6 +835,12 @@ namespace MobileGL::MG_Impl::GLImpl {
         MOBILEGL_ASSERT(nullptr != static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get()),
                         "Texture object here should always be an object with mipmap");
         auto textureMipmapObject = static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get());
+        if (static_cast<Uint>(level) >= textureMipmapObject->GetMipmapLevelCount()) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "Texture level is out of range."));
+            return;
+        }
 
         const void* originalPixels = pixels;
         const auto& pixelUnpackBufferObject =
@@ -811,6 +873,14 @@ namespace MobileGL::MG_Impl::GLImpl {
         const SizeT destRowSize = static_cast<SizeT>(texelSize.x()) * internalBpp;
         const SizeT destSliceSize = static_cast<SizeT>(texelSize.y()) * destRowSize;
 
+        if (xoffset + width > static_cast<GLsizei>(texelSize.x()) ||
+            yoffset + height > static_cast<GLsizei>(texelSize.y()) ||
+            zoffset + depth > static_cast<GLsizei>(texelSize.z())) {
+            MGLOG_E("TexSubImage3D_State: Specified region exceeds texture level dimensions");
+            free(processedPixels);
+            return;
+        }
+
         const auto* srcData = static_cast<const Uint8*>(processedPixels);
         Uint8* destData = static_cast<Uint8*>(textureMipmapObject->MapMipmapData(textureUploadTarget, level));
         if (destData) {
@@ -865,11 +935,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& activeUnit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
         auto& bindingSlot = activeUnit.GetBindingSlot(textureTarget);
         auto& textureObject = bindingSlot.GetBoundObject();
+        if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         TextureInternalFormat textureInternalFormat = textureObject->GetFormat();
         MGLOG_D("%s: working on texture %d", __func__, textureObject->GetExternalIndex());
 
         // ===================== Error Checking ==============================
-        if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (!TextureImpl::ValidateTextureSubImageOffsets(textureObject, xoffset, width, yoffset, height)) return;
         if (!TextureImpl::ValidateTextureInternalFormatCompatibleWithInput(textureInputFormat, textureInternalFormat,
                                                                            texturePixelDataType))
@@ -1027,7 +1097,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
 
         // ======================= Processing ================================
-        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+        auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
         if (!textureObject) return;
 
         switch (pname) {
@@ -1060,6 +1130,12 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_SWIZZLE_A: {
             auto swizzleParam = MG_Util::ConvertGLEnumPnameToTextureSwizzleParam(pname);
             auto swizzleValue = MG_Util::ConvertGLEnumToTextureSwizzleParam((GLenum)param);
+            if (swizzleValue == TextureSwizzleParam::Unknown) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidEnum,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "Invalid texture swizzle value."));
+                return;
+            }
             textureObject->SetSwizzleParam(swizzleParam, swizzleValue);
             break;
         }
@@ -1083,6 +1159,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_LOD_BIAS:
             textureObject->GetSamplerObject()->SetLodBias(param);
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            if (!ValidateMaxAnisotropy(param, __func__)) return;
+            textureObject->GetSamplerObject()->SetMaxAnisotropy(param);
+            break;
         case GL_GENERATE_MIPMAP:
             g_autoGenerateMipmapByTextureId[textureObject->GetExternalIndex()] = (param != 0.0f);
             break;
@@ -1106,7 +1186,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
 
         // ======================= Processing ================================
-        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+        auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
         if (!textureObject) return;
 
         TextureParameterObject_State(textureObject, pname, param, __func__);
@@ -1122,7 +1202,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
 
             // ======================= Processing ================================
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             SetTextureBorderColorFromFloats(textureObject, params);
             break;
@@ -1130,7 +1210,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_SWIZZLE_RGBA: {
             TextureUploadTarget textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             GLint signedParams[4] = {static_cast<GLint>(params[0]), static_cast<GLint>(params[1]),
                                      static_cast<GLint>(params[2]), static_cast<GLint>(params[3])};
@@ -1153,7 +1233,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
 
             // ======================= Processing ================================
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             SetTextureBorderColorFromInts(textureObject, params);
             break;
@@ -1161,7 +1241,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_SWIZZLE_RGBA: {
             TextureUploadTarget textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             if (!SetTextureSwizzleParamsFromInts(textureObject, params, __func__)) {
                 return;
@@ -1179,7 +1259,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_BORDER_COLOR: {
             TextureUploadTarget textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             SetTextureBorderColorFromIntegerInts(textureObject, params);
             break;
@@ -1187,7 +1267,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_SWIZZLE_RGBA: {
             TextureUploadTarget textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             GLint signedParams[4] = {static_cast<GLint>(params[0]), static_cast<GLint>(params[1]),
                                      static_cast<GLint>(params[2]), static_cast<GLint>(params[3])};
@@ -1207,7 +1287,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         case GL_TEXTURE_BORDER_COLOR: {
             TextureUploadTarget textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
             SetTextureBorderColorFromUnsignedInts(textureObject, params);
             break;
@@ -1218,7 +1298,7 @@ namespace MobileGL::MG_Impl::GLImpl {
             TextureTarget textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
 
             // ======================= Processing ================================
-            auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+            auto& textureObject = GetTextureObjectByTargetForParameter(textureUploadTarget, textureTarget);
             if (!textureObject) return;
 
             Vec4<TextureSwizzleParam> swizzleParams;
@@ -1269,6 +1349,8 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& textureObject =
             isProxy ? TextureImpl::pProxyTextureManager->CreateOrReplaceProxyTextureObject(textureUploadTarget)
                     : bindingSlot.GetBoundObject();
+        // Name 0 resolves to the target's default texture object - a real texture this call
+        // (re)specifies like any other; the slot is never empty anymore.
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (textureObject->GetStorageType() != TextureStorageType::Mipmap) {
             MG_State::pGLContext->RecordError(
@@ -1310,6 +1392,8 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& textureObject =
             isProxy ? TextureImpl::pProxyTextureManager->CreateOrReplaceProxyTextureObject(textureUploadTarget)
                     : bindingSlot.GetBoundObject();
+        // Name 0 resolves to the target's default texture object - a real texture this call
+        // (re)specifies like any other; the slot is never empty anymore.
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (textureObject->GetStorageType() != TextureStorageType::Mipmap) {
             MG_State::pGLContext->RecordError(
@@ -1378,6 +1462,8 @@ namespace MobileGL::MG_Impl::GLImpl {
         // target and data is not evenly divisible into the number of bytes needed to store in memory a datum
         // indicated by type.
         // ======================= Processing ================================
+        textureInternalFormat =
+            MG_Util::ConvertInternalFormatToSized(textureInternalFormat, textureInputFormat, texturePixelDataType);
         auto& activeUnit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
         auto& bindingSlot = activeUnit.GetBindingSlot(textureTarget);
         Bool isProxy = TextureImpl::IsProxyTextureTarget(textureUploadTarget);
@@ -1386,6 +1472,8 @@ namespace MobileGL::MG_Impl::GLImpl {
                     : bindingSlot.GetBoundObject();
 
         // ===================== Error Checking ==============================
+        // Name 0 resolves to the target's default texture object - a real texture this call
+        // (re)specifies like any other; the slot is never empty anymore.
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (!ValidateTextureMutable(textureObject, __func__)) return;
 
@@ -1419,7 +1507,11 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto textureMipmapObject = static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get());
 
         // Allocate in TextureObject
-        textureMipmapObject->AllocateStorage(textureUploadTarget, level, {{width, height, depth}, internalBytes});
+        if (isProxy) {
+            MGLOG_D("%s: isProxy = true, not allocating", __func__);
+        } else {
+            textureMipmapObject->AllocateStorage(textureUploadTarget, level, {{width, height, depth}, internalBytes});
+        }
 
         if (!originalPixels) {
             MGLOG_D("%s: No input pixel and no PBO bound, no pixel transfer", __func__);
@@ -1446,6 +1538,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         textureMipmapObject->MarkStorageDirty(textureUploadTarget, level, true);
 
         free(processedPixels);
+        MaybeAutoGenerateMipmap(target, textureObject, isProxy, level);
     }
 
     void TexImage2D_State(GLenum target, GLint level, GLint internalformat, GLsizei width, GLsizei height, GLint border,
@@ -1500,6 +1593,8 @@ namespace MobileGL::MG_Impl::GLImpl {
                     : bindingSlot.GetBoundObject();
 
         // ===================== Error Checking ==============================
+        // Name 0 resolves to the target's default texture object - a real texture this call
+        // (re)specifies like any other; the slot is never empty anymore.
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (!ValidateTextureMutable(textureObject, __func__)) return;
 
@@ -1604,6 +1699,8 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& textureObject =
             isProxy ? TextureImpl::pProxyTextureManager->CreateOrReplaceProxyTextureObject(textureUploadTarget)
                     : bindingSlot.GetBoundObject();
+        // Name 0 resolves to the target's default texture object - a real texture this call
+        // (re)specifies like any other; the slot is never empty anymore.
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (!ValidateTextureMutable(textureObject, __func__)) return;
 
@@ -1659,8 +1756,12 @@ namespace MobileGL::MG_Impl::GLImpl {
         if (!TextureImpl::ValidateTextureUploadTarget(textureUploadTarget)) return;
         if (!TextureImpl::ValidateTextureInternalFormat(textureInternalFormat)) return;
         // TODO: make sure `internalformat` is in one of supported format for TexBuffer
+        // GL 3.3 core 3.8.5: buffer zero detaches any buffer from the buffer texture - only a
+        // nonzero name that is not an existing buffer object is an error. This is reachable on
+        // the default buffer texture (bound whenever texture 0 is bound to GL_TEXTURE_BUFFER),
+        // which the GL CTS state reset detaches with glTexBuffer(..., 0) after every case.
         auto& bufferObject = MG_State::pGLContext->GetBufferObject(buffer);
-        if (!bufferObject) {
+        if (buffer != 0 && !bufferObject) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
                 MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
@@ -1674,6 +1775,8 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& textureObject = bindingSlot.GetBoundObject();
 
         // ===================== Error Checking ==============================
+        // Name 0 is the default buffer texture - a real object the (de)attach operates on, not a
+        // silent no-op; the slot is never empty now that every unit/target holds its default.
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (textureObject->GetStorageType() != TextureStorageType::Buffer) {
             MG_State::pGLContext->RecordError(
@@ -1694,7 +1797,9 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     GLboolean IsTexture_State(GLuint texture) {
         // ======================= Processing ================================
-        if (!TextureImpl::ValidateTextureName(texture, true)) return GL_FALSE;
+        // GL 3.3 core 6.1.4: IsTexture generates no error - an unknown, deleted or merely reserved
+        // name is just GL_FALSE. Probing with the recording validator (as every other Is* entry
+        // point already avoids doing) would leave a spurious INVALID_VALUE behind.
         return MG_State::pGLContext->ValidateTextureObject(texture) ? GL_TRUE : GL_FALSE;
     }
 
@@ -1883,6 +1988,11 @@ namespace MobileGL::MG_Impl::GLImpl {
                     textureObject->GetSamplerObject()->GetSamplerCompareFunc());
             }
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            if (params) {
+                *params = static_cast<GLint>(textureObject->GetSamplerObject()->GetMaxAnisotropy());
+            }
+            break;
         case GL_IMAGE_FORMAT_COMPATIBILITY_TYPE:
             if (params) {
                 *params = GL_IMAGE_FORMAT_COMPATIBILITY_BY_SIZE;
@@ -2029,6 +2139,11 @@ namespace MobileGL::MG_Impl::GLImpl {
                     textureObject->GetSamplerObject()->GetSamplerCompareFunc());
             }
             break;
+        case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+            if (params) {
+                *params = textureObject->GetSamplerObject()->GetMaxAnisotropy();
+            }
+            break;
         default:
             MG_State::pGLContext->RecordError(ErrorCode::InvalidEnum,
                                               MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "GetTexParameterfv_State",
@@ -2300,7 +2415,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         for (SizeT i = 0; i < static_cast<SizeT>(n); ++i) {
             Uint textureName = textures[i];
             if (textureName == 0) continue;
-            if (!TextureImpl::ValidateTextureName(textureName, true)) continue;
+            if (!MG_State::pGLContext->ValidateTextureName(textureName)) continue;
             MG_State::pGLContext->MarkTextureObjectForDeletion(textureName);
         }
     }
@@ -2507,11 +2622,13 @@ namespace MobileGL::MG_Impl::GLImpl {
         // ===================== Error Checking ==============================
         if (!TextureImpl::ValidateTextureTarget(textureTarget)) return;
 
-        // Name 0 unbinds the current target from the active texture unit.
+        // GL 3.3 core 3.8: name 0 is the target's default texture object - a real texture that
+        // glTexImage*/glTexParameter*/glGetTex* must operate on - not "nothing bound". Binding it
+        // restores the unit/target slot to its initial state.
         if (texture == 0) {
             auto& currentUnit = MG_State::pGLContext->GetTextureUnitObject(activeUnit);
             auto& bindingSlot = currentUnit.GetBindingSlot(textureTarget);
-            bindingSlot.Bind(nullptr);
+            bindingSlot.Bind(MG_State::pGLContext->GetDefaultTextureObject(textureTarget));
             MG_State::pGLContext->NoteTextureUnitTouched(activeUnit);
             return;
         }
@@ -2523,6 +2640,9 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
+        // GL 3.3 core 3.8.1: a name that GenTextures never returned - or that has since been deleted -
+        // is not a legal bind target in the core profile (no application-generated names), and the error
+        // is INVALID_OPERATION, not INVALID_VALUE.
         if (!MG_State::pGLContext->ValidateTextureName(texture)) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
@@ -2530,8 +2650,6 @@ namespace MobileGL::MG_Impl::GLImpl {
             return;
         }
 
-        if (!TextureImpl::ValidateTextureName(texture, true)) return;
-
         // ======================= Processing ================================
         Bool doesTextureExist = MG_State::pGLContext->ValidateTextureObject(texture);
         if (!doesTextureExist) {
@@ -2551,14 +2669,25 @@ namespace MobileGL::MG_Impl::GLImpl {
 
     void ActiveTexture_State(GLenum texture) {
         // ===================== Error Checking ==============================
-        if (texture < GL_TEXTURE0 || texture > GL_TEXTURE31) {
+        // GL 3.3 core 3.8: the valid range is [GL_TEXTURE0, GL_TEXTURE0 +
+        // GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS) - NOT a fixed 0..31 range. GL CTS's per-case
+        // state reset iterates every advertised combined unit, so rejecting units the
+        // implementation itself reports would leave a sticky GL_INVALID_ENUM behind and abort
+        // whole test batches. The backend already clamps its advertised value to the state
+        // layer's MAX_TEXTURE_IMAGE_UNITS capacity.
+        Int maxCombinedUnits = static_cast<Int>(MG_State::GLState::TextureState::MAX_TEXTURE_IMAGE_UNITS);
+        if (MG_Backend::pActiveBackendObject) {
+            maxCombinedUnits = std::min(
+                maxCombinedUnits, MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxCombinedTextureImageUnits);
+        }
+        if (texture < GL_TEXTURE0 || static_cast<Int>(texture - GL_TEXTURE0) >= maxCombinedUnits) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidEnum,
                 MakeUnique<GenericErrorInfo>(
                     "MG_Impl/GLImpl", "ActiveTexture_State",
-                    std::format("Texture must be one of GL_TEXTUREi, where i is in the range 0 to 31, but got "
+                    std::format("Texture must be one of GL_TEXTUREi, where i is in the range 0 to {}, but got "
                                 "invalid enum: 0x{:X}, which may stand for unit {}.",
-                                texture, texture - GL_TEXTURE0)));
+                                maxCombinedUnits - 1, texture, texture - GL_TEXTURE0)));
             return;
         }
 
@@ -2918,10 +3047,13 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto* textureMipmapObject = static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get());
 
         textureObject->SetInternalFormat(textureInternalFormat);
+        // Array targets keep their layer count constant across levels; only true 3D
+        // textures halve depth per level (GL 3.3 §3.9 glTexStorage3D).
+        const Bool depthMips = DepthParticipatesInMipmapping(textureObject->GetTarget());
         for (GLsizei level = 0; level < levels; ++level) {
             const GLsizei levelWidth = std::max<GLsizei>(1, width >> level);
             const GLsizei levelHeight = std::max<GLsizei>(1, height >> level);
-            const GLsizei levelDepth = std::max<GLsizei>(1, depth >> level);
+            const GLsizei levelDepth = depthMips ? std::max<GLsizei>(1, depth >> level) : depth;
             const SizeT byteSize = ComputeTextureStorageByteSize(textureInternalFormat, levelWidth, levelHeight,
                                                                  levelDepth);
             textureMipmapObject->AllocateStorage(textureUploadTarget, level,
@@ -2957,6 +3089,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& bindingSlot = activeUnit.GetBindingSlot(textureTarget);
         auto& textureObject = bindingSlot.GetBoundObject();
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
+        if (!TextureImpl::ValidateTextureNotDefault(textureObject, __func__)) return;
 
         TextureStorage1D(textureObject->GetExternalIndex(), levels, internalformat, width);
     }
@@ -2971,6 +3104,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& bindingSlot = activeUnit.GetBindingSlot(textureTarget);
         auto& textureObject = bindingSlot.GetBoundObject();
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
+        if (!TextureImpl::ValidateTextureNotDefault(textureObject, __func__)) return;
 
         TextureStorage2D(textureObject->GetExternalIndex(), levels, internalformat, width, height);
     }
@@ -2986,6 +3120,7 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& bindingSlot = activeUnit.GetBindingSlot(textureTarget);
         auto& textureObject = bindingSlot.GetBoundObject();
         if (!TextureImpl::ValidateTextureObject(textureObject)) return;
+        if (!TextureImpl::ValidateTextureNotDefault(textureObject, __func__)) return;
 
         TextureStorage3D(textureObject->GetExternalIndex(), levels, internalformat, width, height, depth);
     }
@@ -3141,8 +3276,10 @@ namespace MobileGL::MG_Impl::GLImpl {
         auto& textureUnit = MG_State::pGLContext->GetTextureUnitObject(static_cast<Int>(unit));
         MG_State::pGLContext->NoteTextureUnitTouched(static_cast<Int>(unit));
         if (texture == 0) {
+            // GL 4.5 8.1: texture zero unbinds every target of the unit, i.e. rebinds each
+            // target's default texture object (the unit's initial state).
             for (auto& slot : textureUnit.GetAllBindingSlots()) {
-                slot.Bind(nullptr);
+                slot.Bind(MG_State::pGLContext->GetDefaultTextureObject(slot.GetTarget()));
             }
             return;
         }

MobileGL/MG_Impl/GLImpl/Texture/Validators.cpp

@@ -175,13 +175,15 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
         return true;
     }
 
-    static Bool IsIntegerColorInputFormat(TextureInputFormat format) {
+    Bool IsIntegerColorInputFormat(TextureInputFormat format) {
         return format == TextureInputFormat::RInteger || format == TextureInputFormat::RGInteger ||
                format == TextureInputFormat::RGBInteger || format == TextureInputFormat::BGRInteger ||
-               format == TextureInputFormat::RGBAInteger || format == TextureInputFormat::BGRAInteger;
+               format == TextureInputFormat::RGBAInteger || format == TextureInputFormat::BGRAInteger ||
+               format == TextureInputFormat::GreenInteger || format == TextureInputFormat::BlueInteger ||
+               format == TextureInputFormat::AlphaInteger;
     }
 
-    static Bool IsIntegerColorInternalFormat(TextureInternalFormat internalFormat) {
+    Bool IsIntegerColorInternalFormat(TextureInternalFormat internalFormat) {
         switch (internalFormat) {
             case TextureInternalFormat::R8I:
             case TextureInternalFormat::R8UI:
@@ -235,17 +237,15 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
                format == TextureInputFormat::StencilIndex;
     }
 
-    // Mirrors the desktop-GL validity matrix used by GL CTS packed_pixels (glcPackedPixelsTests
-    // isFormatValid, INPUT_TEXIMAGE): packed-type/format pairing, depth-vs-color mismatch, and
-    // integer-ness matching all raise GL_INVALID_OPERATION instead of reaching the upload path.
-    Bool ValidateTextureInternalFormatCompatibleWithInput(TextureInputFormat format,
-                                                          TextureInternalFormat internalFormat,
-                                                          TexturePixelDataType type) {
+    // Client-memory format<->type pairing rules shared by pixel uploads (TexImage*) and readbacks
+    // (ReadPixels, GetTexImage). Mirrors the desktop-GL validity matrix used by GL CTS packed_pixels
+    // (glcPackedPixelsTests isFormatValid): packed types constrain the formats they may pair with, and
+    // integer formats reject floating-point types; violations raise GL_INVALID_OPERATION.
+    Bool ValidateClientFormatTypePairing(TextureInputFormat format, TexturePixelDataType type) {
         const auto recordInvalidOperation = [](const char* message) {
             MG_State::pGLContext->RecordError(
                 ErrorCode::InvalidOperation,
-                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateTextureInternalFormatCompatibleWithInput",
-                                             message));
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateClientFormatTypePairing", message));
             return false;
         };
 
@@ -288,6 +288,27 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
             return recordInvalidOperation("Integer format cannot be used with a floating-point type");
         }
 
+        return true;
+    }
+
+    // Mirrors the desktop-GL validity matrix used by GL CTS packed_pixels (glcPackedPixelsTests
+    // isFormatValid, INPUT_TEXIMAGE): packed-type/format pairing, depth-vs-color mismatch, and
+    // integer-ness matching all raise GL_INVALID_OPERATION instead of reaching the upload path.
+    Bool ValidateTextureInternalFormatCompatibleWithInput(TextureInputFormat format,
+                                                          TextureInternalFormat internalFormat,
+                                                          TexturePixelDataType type) {
+        const auto recordInvalidOperation = [](const char* message) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", "ValidateTextureInternalFormatCompatibleWithInput",
+                                             message));
+            return false;
+        };
+
+        if (!ValidateClientFormatTypePairing(format, type)) {
+            return false;
+        }
+
         // TexImage in core 3.3 has no stencil-only upload path (that arrived with GL 4.4).
         if (format == TextureInputFormat::StencilIndex) {
             return recordInvalidOperation("STENCIL_INDEX is not a valid texture upload format");
@@ -339,6 +360,19 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
         return true;
     }
 
+    Bool ValidateTextureNotDefault(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
+                                   const char* caller) {
+        if (textureObject && textureObject->GetExternalIndex() == 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
+                                             "This operation is not allowed on the default texture (zero is "
+                                             "bound to the target)."));
+            return false;
+        }
+        return true;
+    }
+
     Bool ValidateTextureTargetUniformity(SharedPtr<MG_State::GLState::ITextureObject> textureObject,
                                          TextureTarget target) {
         if (!textureObject) return true; // should be created later

MobileGL/MG_Impl/GLImpl/Texture/Validators.h

@@ -23,11 +23,19 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
     Bool ValidateTextureSizeRange(Int width, Int height, Int depth);
     Bool ValidateTextureInternalFormat(TextureInternalFormat format);
     Bool ValidateTextureBorderNumber(Int border);
+    Bool IsIntegerColorInputFormat(TextureInputFormat format);
+    Bool IsIntegerColorInternalFormat(TextureInternalFormat internalFormat);
+    Bool ValidateClientFormatTypePairing(TextureInputFormat format, TexturePixelDataType type);
     Bool ValidateTextureInternalFormatCompatibleWithInput(TextureInputFormat format,
                                                           TextureInternalFormat internalFormat,
                                                           TexturePixelDataType type);
     Bool ValidateTextureLevelWithUploadTarget(TextureUploadTarget target, Int level);
     Bool ValidateTextureObject(SharedPtr<MG_State::GLState::ITextureObject> textureObject);
+    // Rejects the per-target default texture objects (name 0) with GL_INVALID_OPERATION for entry
+    // points that require a GenTextures-created texture, e.g. TexStorage* ("An INVALID_OPERATION
+    // error is generated if zero is bound to target", ARB_texture_storage).
+    Bool ValidateTextureNotDefault(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
+                                   const char* caller);
     Bool ValidateTextureTargetUniformity(SharedPtr<MG_State::GLState::ITextureObject> textureObject,
                                          TextureTarget target);
     Bool ValidateTextureSubImageOffsets(SharedPtr<MG_State::GLState::ITextureObject> textureObject, Int xoffset,

MobileGL/MG_Impl/GLImpl/VertexArray/GL_VertexArray.cpp

@@ -78,15 +78,12 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
 
         static bool ValidateCurrentVertexAttribIndex(GLuint index, const char* funcName) {
-            if (!VertexArrayImpl::ValidateVertexAttributeIndex(index)) return false;
-            if (index == 0) {
-                MG_State::pGLContext->RecordError(
-                    ErrorCode::InvalidOperation,
-                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", funcName,
-                                                 "Generic vertex attribute 0 current value cannot be modified."));
-                return false;
-            }
-            return true;
+            // GL 3.3 core 2.7: VertexAttrib* sets the current value of ANY generic attribute,
+            // including index 0 - only an out-of-range index is an error (INVALID_VALUE).
+            // "Attribute 0 is immutable" was legacy immediate-mode lore; rejecting it broke GL
+            // CTS's per-case state reset, which writes vertexAttrib4f(0, 0,0,0,1) after every case.
+            static_cast<void>(funcName);
+            return VertexArrayImpl::ValidateVertexAttributeIndex(index);
         }
 
         static bool TryGetVertexAttribute(GLuint index, const MG_State::GLState::VertexAttribute** outAttr) {
@@ -288,7 +285,9 @@ namespace MobileGL::MG_Impl::GLImpl {
             GLuint vao = arrays[i];
             if (vao == 0) continue;
 
-            if (!VertexArrayImpl::ValidateVertexArrayName(vao)) continue;
+            // GL 3.3 core 2.10: unknown names are silently ignored on delete; the shared bind-path
+            // validator would record INVALID_OPERATION instead.
+            if (!MG_State::pGLContext->ValidateVertexArrayName(vao)) continue;
 
             if (MG_State::pGLContext->GetBoundVertexArray() &&
                 MG_State::pGLContext->GetBoundVertexArray() == MG_State::pGLContext->GetVertexArrayObject(vao)) {

MobileGL/MG_State/EGLState/Core.cpp

@@ -764,6 +764,20 @@ namespace MobileGL {
                        ctx->MajorVersion > 3 || (ctx->MajorVersion == 3 && ctx->MinorVersion >= 1);
             }
 
+            Bool EGLContext::IsCurrentContextOpenGLCompatibilityProfile() const {
+                const std::lock_guard<std::recursive_mutex> lock(m_mutex);
+                auto currentIt = m_threadCurrents.find(CurrentThreadKey());
+                if (currentIt == m_threadCurrents.end()) {
+                    return false;
+                }
+                const auto* ctx = TryGetContext(currentIt->second.Context);
+                // Affirmative check: true only when the host explicitly requested the
+                // compatibility bit. Attrib-less contexts report false and thus read as core
+                // for GL_CONTEXT_PROFILE_MASK (EGL defaults 3.x contexts to the core profile).
+                return ctx && ctx->ClientAPI == EGL_OPENGL_API &&
+                       (ctx->OpenGLProfileMask & EGL_CONTEXT_OPENGL_COMPATIBILITY_PROFILE_BIT);
+            }
+
             EGLint EGLContext::GetCurrentContextFlags() const {
                 const std::lock_guard<std::recursive_mutex> lock(m_mutex);
                 auto currentIt = m_threadCurrents.find(CurrentThreadKey());

MobileGL/MG_State/EGLState/Core.h

@@ -59,6 +59,7 @@ namespace MobileGL {
                 Bool ValidateContext(EGLContextHandle context) const;
                 Bool ValidateContextOnDisplay(EGLDisplayHandle display, EGLContextHandle context) const;
                 Bool IsCurrentContextOpenGLCoreProfile() const;
+                Bool IsCurrentContextOpenGLCompatibilityProfile() const;
                 EGLint GetCurrentContextFlags() const;
 
                 // Surface

MobileGL/MG_State/GLState/Core.cpp

@@ -9,6 +9,7 @@
 #include "Core.h"
 #include "MG_State/GLState/RenderbufferState/RenderbufferObject.h"
 #include "MG_State/EGLState/Core.h"
+#include <Config.h>
 
 namespace MobileGL::MG_State {
     void Init() {
@@ -17,6 +18,11 @@ namespace MobileGL::MG_State {
         pEGLContext = MakeUnique<EGLState::EGLContext>();
     }
 
+    Bool IsRelaxedSemanticsActive() {
+        return MG_Config::Features.RelaxedSemantics ||
+               !(pEGLContext && pEGLContext->IsCurrentContextOpenGLCoreProfile());
+    }
+
     namespace GLState {
         // Error
         void GLContext::RecordError(ErrorCode code, UniquePtr<ErrorInfo> info) {
@@ -226,12 +232,16 @@ namespace MobileGL::MG_State {
             return m_textureState.GetTextureObject(index);
         }
 
+        const SharedPtr<ITextureObject>& GLContext::GetDefaultTextureObject(TextureTarget target) const {
+            return m_textureState.GetDefaultTextureObject(target);
+        }
+
         const SharedPtr<ITextureObject>& GLContext::CreateTextureObject(Uint index, TextureTarget target) {
             return m_textureState.CreateTextureObject(index, target);
         }
 
         void GLContext::MarkTextureObjectForDeletion(Uint index) {
-            m_textureState.MarkTextureObjectForDeletion(index);
+            m_textureState.MarkTextureObjectForDeletion(index, IsRelaxedSemanticsActive());
         }
 
         TextureUnit& GLContext::GetTextureUnitObject(Int unit) {
@@ -279,6 +289,10 @@ namespace MobileGL::MG_State {
             return m_programState.MarkShaderObjectForDeletion(index);
         }
 
+        void GLContext::ReleaseShaderNameIfOrphaned(const Uint index) {
+            return m_programState.ReleaseShaderNameIfOrphaned(index);
+        }
+
         Bool GLContext::ValidateProgramName(const Uint index) const {
             return m_programState.ValidateProgramObject(index);
         }

MobileGL/MG_State/GLState/Core.h

@@ -95,6 +95,8 @@ namespace MobileGL {
                 // Texture
                 void GenTextureNames(Uint number, Vector<Uint>& textures);
                 const SharedPtr<ITextureObject>& GetTextureObject(Uint index);
+                // Per-target default texture object (name 0); see TextureState::GetDefaultTextureObject.
+                const SharedPtr<ITextureObject>& GetDefaultTextureObject(TextureTarget target) const;
                 const SharedPtr<ITextureObject>& CreateTextureObject(Uint index, TextureTarget target);
                 void MarkTextureObjectForDeletion(Uint index);
                 TextureUnit& GetTextureUnitObject(Int unit);
@@ -106,6 +108,7 @@ namespace MobileGL {
                 // texture is bound at a unit; lets a backend skip re-resolving an unchanged
                 // per-draw sampled-texture set.
                 Uint64 GetTextureBindGeneration() const { return m_textureState.GetTextureBindGeneration(); }
+                void BumpTextureBindGeneration() { m_textureState.BumpTextureBindGeneration(); }
                 Bool ValidateTextureName(Uint index) const;
                 Bool ValidateTextureObject(Uint index) const;
                 Int GetActiveTextureUnit() const;
@@ -116,6 +119,9 @@ namespace MobileGL {
                 Uint CreateShader(ShaderStage stage);
                 void MarkProgramForDeletion(Uint index);
                 void MarkShaderForDeletion(Uint index);
+                // Frees a deletion-flagged shader's name once it lost its last GL-visible
+                // attachment (call after glDetachShader).
+                void ReleaseShaderNameIfOrphaned(Uint index);
                 Bool ValidateProgramName(Uint index) const;
                 Bool ValidateShaderName(Uint index) const;
                 const SharedPtr<ProgramObject>& GetProgramObject(Uint index);
@@ -247,5 +253,12 @@ namespace MobileGL {
         } // namespace GLState
 
         extern UniquePtr<GLState::GLContext> pGLContext;
+
+        // True when relaxed GL semantics apply. Strict core rules are enforced only when the
+        // current EGL context explicitly requested a core profile (core bit in
+        // EGL_CONTEXT_OPENGL_PROFILE_MASK, or a >=3.1 version request without the compatibility
+        // bit) and MOBILEGL_RELAXED_SEMANTICS is off; no current context, legacy version
+        // requests, and the compatibility bit all relax.
+        Bool IsRelaxedSemanticsActive();
     } // namespace MG_State
 } // namespace MobileGL

MobileGL/MG_State/GLState/ErrorState/Error.cpp

@@ -7,6 +7,7 @@
 // End of Source File Header
 
 #include "Error.h"
+#include <algorithm>
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/ErrorCodeConverter.h>
 
@@ -19,7 +20,15 @@ namespace MobileGL::MG_State::GLState {
             MGLOG_E("Recording OpenGL error (%s):\n%s",
                     MG_Util::ConvertGLEnumToString(MG_Util::ConvertErrorCodeToGLEnum(code)).c_str(),
                     info->toString().c_str());
-            m_errors.push_back(MakeUnique<Error>(code, Move(info)));
+            // GL error semantics are sticky flags, not a queue (GL 3.3 core §2.5): with multiple
+            // error flags, each is set only while currently unset — repeated errors of the same
+            // code are discarded until glGetError reads the flag. Unbounded accumulation leaked
+            // stale errors into later, unrelated glGetError checks (GL CTS deinit noise).
+            const Bool alreadyPending = std::any_of(m_errors.begin(), m_errors.end(),
+                                                    [code](const auto& e) { return e->code == code; });
+            if (!alreadyPending) {
+                m_errors.push_back(MakeUnique<Error>(code, Move(info)));
+            }
         }
     }
 

MobileGL/MG_State/GLState/ProgramState/ProgramObject.cpp

@@ -8,6 +8,7 @@
 
 #include "ProgramObject.h"
 #include <atomic>
+#include <cstring>
 #include <MG_Backend/BackendObjects.h>
 #include <MG_State/GLState/VertexArrayState/VertexArrayObject.h>
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
@@ -87,6 +88,19 @@ namespace {
         }
     }
 
+    // How many consecutive uniform locations a uniform occupies. Array uniforms (opaque
+    // or not) span one location per element so glUniform*v(count > 1) and
+    // glGetUniformLocation("arr[k]") can address elements individually; everything else
+    // spans a single location. TObjectReflection.size only carries the element count for
+    // non-block arrays, so prefer the TType, which is authoritative for both.
+    static MobileGL::Int GetUniformLocationSpan(const glslang::TObjectReflection& uniform) {
+        const glslang::TType* type = uniform.getType();
+        if (type != nullptr && type->isSizedArray()) {
+            return std::max(1, type->getOuterArraySize());
+        }
+        return std::max(1, uniform.size);
+    }
+
     static bool ComputeShaderDeclaresLocalSize(const MobileGL::String& source) {
         bool inLineComment = false;
         bool inBlockComment = false;
@@ -346,7 +360,18 @@ namespace MobileGL::MG_State::GLState {
         }
 
         MGLOG_D("ProgramObject %u: DoReflection - building reflection", m_externalIndex);
-        if (!m_program->buildReflection()) {
+        // GL-style reflection naming (GL CTS uniform_block relies on all four):
+        //  - BasicArraySuffix: an array uniform is reported as "arr[0]" per the GL spec.
+        //  - StrictArraySuffix: named-block struct arrays expand per element ("s[0].a",
+        //    "s[1].a", ...) following ARB_program_interface_query rules. Default-block
+        //    (loose) uniforms already expand per element without this option.
+        //  - AllBlockVariables: every member of an active named block is active even when
+        //    no shader statement reads it (ES 3.0/GL 3.3 named-block semantics).
+        //  - SharedStd140UBO: a DECLARED uniform block is active even when no member is
+        //    ever read (reflected from the linker objects). PreprocessShaderSource coerces
+        //    every block to std140, so this covers all of them.
+        if (!m_program->buildReflection(EShReflectionStrictArraySuffix | EShReflectionBasicArraySuffix |
+                                        EShReflectionAllBlockVariables | EShReflectionSharedStd140UBO)) {
             m_linkStatus = false;
             m_infoLog = "Build reflection failed.";
             MGLOG_E("ProgramObject %u: DoReflection - buildReflection() returned false", m_externalIndex);
@@ -361,8 +386,7 @@ namespace MobileGL::MG_State::GLState {
         for (int i = 0; i < m_activeUniformCount; i++) {
             auto& uniform = m_program->getUniform(i);
             auto location = uniform.layoutLocation();
-            const Int locationSpan =
-                (uniform.getType() && uniform.getType()->isOpaque()) ? std::max(1, uniform.size) : 1;
+            const Int locationSpan = GetUniformLocationSpan(uniform);
             requiredUniformLocations += locationSpan;
             if (location != glslang::TQualifier::layoutLocationEnd) {
                 m_maxUniformLocation = std::max(m_maxUniformLocation, location + locationSpan - 1);
@@ -401,8 +425,7 @@ namespace MobileGL::MG_State::GLState {
                         m_externalIndex, uniform.name.c_str());
                 continue; // will allocate unallocated uniforms later
             }
-            const Int locationSpan =
-                (uniform.getType() && uniform.getType()->isOpaque()) ? std::max(1, uniform.size) : 1;
+            const Int locationSpan = GetUniformLocationSpan(uniform);
             for (Int element = 0; element < locationSpan; ++element) {
                 m_uniformIndexInTProgram[location + element] = i;
             }
@@ -419,8 +442,8 @@ namespace MobileGL::MG_State::GLState {
         });
         for (auto index : unallocatedUniformIndex) {
             auto& uniform = m_program->getUniform(index);
-            const Int locationSpan =
-                (uniform.getType() && uniform.getType()->isOpaque()) ? std::max(1, uniform.size) : 1;
+            const Int locationSpan = GetUniformLocationSpan(uniform);
+            Bool placed = false;
             for (; locNeedle <= m_maxUniformLocation; locNeedle++) {
                 bool hasRoom = locNeedle + locationSpan - 1 <= m_maxUniformLocation;
                 for (Int element = 0; hasRoom && element < locationSpan; ++element) {
@@ -437,8 +460,25 @@ namespace MobileGL::MG_State::GLState {
                         "(index %d)",
                         m_externalIndex, uniform.name.c_str(), locNeedle, locNeedle + locationSpan - 1, index);
                 locNeedle += locationSpan;
+                placed = true;
                 break;
             }
+            if (!placed) {
+                // Explicit-location uniforms can fragment the space so no contiguous
+                // span is left; grow the table instead of leaving the uniform without
+                // a location (which would make it unsettable via glUniform*).
+                const SizeT base = m_uniformIndexInTProgram.size();
+                m_uniformIndexInTProgram.resize(base + locationSpan, glslang::TQualifier::layoutLocationEnd);
+                m_uniformSamplerOrImageUnitIndex.resize(base + locationSpan, -1);
+                m_maxUniformLocation = static_cast<Uint>(base + locationSpan - 1);
+                for (Int element = 0; element < locationSpan; ++element) {
+                    m_uniformIndexInTProgram[base + element] = index;
+                }
+                m_uniformLocations[uniform.name] = static_cast<Uint>(base);
+                MGLOG_D("ProgramObject %u: Reflection - grew location table to place uniform '%s' at %zu..%zu",
+                        m_externalIndex, uniform.name.c_str(), base, base + locationSpan - 1);
+                locNeedle = base + locationSpan;
+            }
         }
 
         for (int i = 0; i < m_activeUniformCount; i++) {
@@ -454,10 +494,17 @@ namespace MobileGL::MG_State::GLState {
                 continue;
             }
 
-            const auto explicitBinding = m_explicitOpaqueUniformBindings.find(uniform.name);
+            // Reflection names an array "texs[0]" while the layout(binding = N) map from the IO
+            // resolver is keyed by the declared name ("texs"); look up both spellings.
+            auto explicitBinding = m_explicitOpaqueUniformBindings.find(uniform.name);
+            if (explicitBinding == m_explicitOpaqueUniformBindings.end() && uniform.name.length() > 3 &&
+                uniform.name.compare(uniform.name.length() - 3, 3, "[0]") == 0) {
+                explicitBinding =
+                    m_explicitOpaqueUniformBindings.find(uniform.name.substr(0, uniform.name.length() - 3));
+            }
             const int initialUnit =
                 explicitBinding != m_explicitOpaqueUniformBindings.end() ? static_cast<int>(explicitBinding->second) : 0;
-            const Int locationSpan = std::max(1, uniform.size);
+            const Int locationSpan = GetUniformLocationSpan(uniform);
             for (Int element = 0; element < locationSpan &&
                                   location + element < m_uniformSamplerOrImageUnitIndex.size(); ++element) {
                 m_uniformSamplerOrImageUnitIndex[location + element] =
@@ -622,8 +669,11 @@ namespace MobileGL::MG_State::GLState {
         m_uniformSizesInBytes.clear();
         m_uniformOffsets.clear();
         m_globalUboScratch.clear();
-        m_uniformOffsets.resize(m_maxUniformLocation + 1);
-        m_uniformSizesInBytes.resize(m_maxUniformLocation + 1);
+        // kInvalidUniformOffset marks locations that end up without global-UBO backing
+        // (e.g. the optimizer eliminated every use of the uniform); the fallback pass
+        // below gives those locations tail storage so glUniform* always has a target.
+        m_uniformOffsets.resize(m_maxUniformLocation + 1, kInvalidUniformOffset);
+        m_uniformSizesInBytes.resize(m_maxUniformLocation + 1, 0);
         for (SizeT i = 0; i < m_generatedSpirv.size(); i++) {
             auto& spv = m_generatedSpirv[i];
 
@@ -653,31 +703,95 @@ namespace MobileGL::MG_State::GLState {
                     m_globalUboScratch.resize(size);
                 }
                 for (const auto& [name, offset] : meta.plainUniformOffsetsInUBO) {
-                    if (m_uniformLocations.find(name) != m_uniformLocations.end()) {
-                        m_uniformOffsets[m_uniformLocations[name]] = offset;
-                        MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' offset=%u assigned to location %u",
-                                m_externalIndex, name.c_str(), offset, m_uniformLocations[name]);
-            } else {
-                MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' offset=%u but not found in "
-                        "m_uniformLocations",
-                        m_externalIndex, name.c_str(), offset);
-            }
-                }
-                for (const auto& [name, size] : meta.plainUniformMemberSizesInBytes) {
-                    if (m_uniformLocations.find(name) != m_uniformLocations.end()) {
-                        m_uniformSizesInBytes[m_uniformLocations[name]] = size;
-                        MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' size=%u assigned to location %u",
-                                m_externalIndex, name.c_str(), size, m_uniformLocations[name]);
-            } else {
-                MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' size=%u but not found in "
-                        "m_uniformLocations",
-                        m_externalIndex, name.c_str(), size);
-            }
+                    // SPIRV-Reflect leaf names never carry a "[0]" suffix; frontend
+                    // reflection keys arrays as "arr[0]" (GL naming), so retry with the
+                    // suffix before declaring the uniform unbacked.
+                    auto locationIt = m_uniformLocations.find(name);
+                    if (locationIt == m_uniformLocations.end()) {
+                        locationIt = m_uniformLocations.find(name + "[0]");
+                    }
+                    if (locationIt == m_uniformLocations.end()) {
+                        MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' offset=%u but not found in "
+                                "m_uniformLocations",
+                                m_externalIndex, name.c_str(), offset);
+                        continue;
+                    }
+                    const Uint baseLocation = locationIt->second;
+                    if (!IsValidUniformLocation(static_cast<Int>(baseLocation))) {
+                        continue;
+                    }
+
+                    const Int uniformIndex = m_uniformIndexInTProgram[baseLocation];
+                    const GLint arraySize = GetActiveUniformArraySize(uniformIndex);
+                    SizeT memberSize = 0;
+                    const auto sizeIt = meta.plainUniformMemberSizesInBytes.find(name);
+                    if (sizeIt != meta.plainUniformMemberSizesInBytes.end()) {
+                        memberSize = sizeIt->second;
+                    }
+                    Uint arrayStride = 0;
+                    const auto strideIt = meta.plainUniformArrayStridesInUBO.find(name);
+                    if (strideIt != meta.plainUniformArrayStridesInUBO.end()) {
+                        arrayStride = strideIt->second;
+                    }
+
+                    // Array uniforms span one location per element (see DoReflection);
+                    // give each element its real byte offset inside the UBO.
+                    const GLint elementCount = (arraySize > 1 && arrayStride == 0) ? 1 : std::max(arraySize, 1);
+                    for (GLint element = 0; element < elementCount; ++element) {
+                        const Uint location = baseLocation + static_cast<Uint>(element);
+                        if (location > m_maxUniformLocation || m_uniformIndexInTProgram[location] != uniformIndex) {
+                            break;
+                        }
+                        m_uniformOffsets[location] = offset + static_cast<Uint>(element) * arrayStride;
+                        const SizeT consumed = static_cast<SizeT>(element) * arrayStride;
+                        m_uniformSizesInBytes[location] = memberSize > consumed ? memberSize - consumed : 0;
+                    }
+                    MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' offset=%u stride=%u size=%zu assigned "
+                            "to locations %u..%u",
+                            m_externalIndex, name.c_str(), offset, arrayStride, memberSize, baseLocation,
+                            baseLocation + static_cast<Uint>(elementCount) - 1);
                 }
                 MGLOG_D("ProgramObject %u: GenerateBinary - finished parsing module %zu metadata",
                         m_externalIndex, i);
             }
         }
+
+        // Fallback pass: a linked program's active non-opaque uniforms must accept
+        // glUniform*/glGetUniform* even when the optimized SPIR-V no longer contains
+        // them (AggressiveDCE can remove a dead loop together with the only loads of a
+        // uniform -- or the entire global UBO, leaving the scratch unallocated). Hand
+        // such locations CPU-side storage at the (16-byte aligned) tail of the shadow
+        // buffer; backends bind at least the SPIR-V-declared UBO range, and the GPU
+        // never reads these bytes, so this only keeps the GL-visible state coherent.
+        for (Uint location = 0; location <= m_maxUniformLocation; ++location) {
+            if (m_uniformOffsets[location] != kInvalidUniformOffset) continue;
+            if (!IsValidUniformLocation(static_cast<Int>(location))) continue;
+            const auto& uniform = m_program->getUniform(m_uniformIndexInTProgram[location]);
+            const glslang::TType* type = uniform.getType();
+            if (type != nullptr && type->isOpaque()) continue;
+            if (uniform.index >= 0 && uniform.index < m_program->getNumUniformBlocks() &&
+                std::strstr(m_program->getUniformBlock(uniform.index).name.c_str(),
+                            MG_Util::ShaderTranspiler::GLOBAL_UBO_NAME) == nullptr) {
+                // Member of a named uniform block: not settable through glUniform*, so it
+                // needs no global-UBO shadow storage.
+                continue;
+            }
+
+            // std140-style slot: the matrix upload paths write column vectors at
+            // 16-byte strides, so a matrix slot must cover cols * 16 bytes.
+            SizeT slotSize = MG_Util::GetGLTypeSize(uniform.glDefineType);
+            if (type != nullptr && type->isMatrix()) {
+                slotSize = static_cast<SizeT>(type->getMatrixCols()) * 16u;
+            }
+            slotSize = (slotSize + 15u) & ~static_cast<SizeT>(15u);
+            const SizeT slotOffset = (m_globalUboScratch.size() + 15u) & ~static_cast<SizeT>(15u);
+            m_globalUboScratch.resize(slotOffset + slotSize, 0);
+            m_uniformOffsets[location] = static_cast<Uint>(slotOffset);
+            m_uniformSizesInBytes[location] = slotSize;
+            MGLOG_D("ProgramObject %u: GenerateBinary - uniform '%s' location %u has no UBO backing in the "
+                    "generated SPIR-V (optimized out?); allocated %zu fallback bytes at scratch offset %zu",
+                    m_externalIndex, uniform.name.c_str(), location, slotSize, slotOffset);
+        }
     }
 
     void ProgramObject::WaitUntilGenerationCompleted() const {

MobileGL/MG_State/GLState/ProgramState/ProgramObject.h

@@ -18,6 +18,13 @@ namespace MobileGL::MG_State::GLState {
     public:
         ProgramObject(Uint externalIndex) : m_externalIndex(externalIndex), m_lifetimeId(AllocateLifetimeId()) {}
         bool ShaderIsAttached(const SharedPtr<ShaderObject>& shader);
+        // GL-visible attachment: in the attach list and not pending detach (glDetachShader
+        // defers the actual removal to the next link).
+        Bool ShaderIsAttachedGLVisible(const SharedPtr<ShaderObject>& shader) const {
+            const auto matches = [&shader](const SharedPtr<ShaderObject>& s) { return s.get() == shader.get(); };
+            if (std::none_of(m_shaders.begin(), m_shaders.end(), matches)) return false;
+            return std::none_of(m_detachedShaders.begin(), m_detachedShaders.end(), matches);
+        }
         bool AttachShader(const SharedPtr<ShaderObject>& shader);
         SizeT DetachShader(const SharedPtr<ShaderObject>& shader);
         SizeT RemoveShader(const SharedPtr<ShaderObject>& shader);
@@ -43,8 +50,51 @@ namespace MobileGL::MG_State::GLState {
         Uint GetMaxUniformLocation() const { return m_maxUniformLocation; }
         Int GetUniformLocation(const String& name) const {
             const auto it = m_uniformLocations.find(name);
-            if (it == m_uniformLocations.end()) return -1;
-            return (Int)it->second;
+            if (it != m_uniformLocations.end()) return (Int)it->second;
+
+            // Reflection stores GL-style names: an array uniform is keyed "arr[0]" (its base
+            // location). A bare "arr" query resolves to that entry; an "arr[k]" query resolves
+            // to base + k because DoReflection reserves one location per array element.
+            if (name.empty()) return -1;
+            if (name.back() != ']') {
+                const auto suffixedIt = m_uniformLocations.find(name + "[0]");
+                if (suffixedIt != m_uniformLocations.end()) return (Int)suffixedIt->second;
+                return -1;
+            }
+            if (name.length() < 4) return -1;
+            const SizeT bracket = name.rfind('[');
+            // Require at least one digit between the brackets.
+            if (bracket == String::npos || bracket + 1 >= name.length() - 1) return -1;
+            Uint element = 0;
+            for (SizeT i = bracket + 1; i < name.length() - 1; ++i) {
+                if (name[i] < '0' || name[i] > '9') return -1;
+                element = element * 10 + static_cast<Uint>(name[i] - '0');
+                if (element > 0x0FFFFFFFu) return -1;
+            }
+            auto baseIt = m_uniformLocations.find(name.substr(0, bracket) + "[0]");
+            if (baseIt == m_uniformLocations.end()) {
+                // Legacy key without the "[0]" suffix (defensive; reflection normally
+                // stores the suffixed form for arrays).
+                baseIt = m_uniformLocations.find(name.substr(0, bracket));
+                if (baseIt == m_uniformLocations.end()) return -1;
+            }
+            const Int base = (Int)baseIt->second;
+            if (!IsValidUniformLocation(base)) return -1;
+            const Int index = m_uniformIndexInTProgram[base];
+            // "[k]" only addresses arrays ("scalar[0]" is not a uniform name), and only
+            // in-range elements.
+            const glslang::TType* type = m_program->getUniform(index).getType();
+            if (type == nullptr || !type->isArray()) return -1;
+            if (static_cast<GLint>(element) >= GetActiveUniformArraySize(index)) return -1;
+            const Int location = base + (Int)element;
+            if (!UniformLocationsAliasSameUniform(base, location)) return -1;
+            return location;
+        }
+
+        // True when both locations are element slots of the same uniform variable.
+        Bool UniformLocationsAliasSameUniform(Int a, Int b) const {
+            if (!IsValidUniformLocation(a) || !IsValidUniformLocation(b)) return false;
+            return m_uniformIndexInTProgram[a] == m_uniformIndexInTProgram[b];
         }
 
         Int GetActiveUniformIndex(const String& name) const {
@@ -54,6 +104,19 @@ namespace MobileGL::MG_State::GLState {
                 return uniformIndex;
             }
 
+            // Reflection stores an array uniform under "arr[0]"; accept the bare "arr"
+            // spelling too. The reverse ("arr[0]" against a bare "arr" entry) is kept for
+            // robustness against non-suffixed reflection entries.
+            if (!name.empty() && name.back() != ']') {
+                const String suffixedName = name + "[0]";
+                const Int suffixedIndex = m_program->getUniformIndex(suffixedName.c_str());
+                if (suffixedIndex >= 0 && suffixedIndex < m_activeUniformCount &&
+                    m_program->getUniform(suffixedIndex).name == suffixedName) {
+                    return suffixedIndex;
+                }
+                return -1;
+            }
+
             if (name.length() <= 3 || name.compare(name.length() - 3, 3, "[0]") != 0) return -1;
             const String baseName = name.substr(0, name.length() - 3);
             const Int baseIndex = m_program->getUniformIndex(baseName.c_str());
@@ -104,11 +167,24 @@ namespace MobileGL::MG_State::GLState {
         }
 
         // GL_UNIFORM_ARRAY_STRIDE: byte stride of an array member in a named block; 0 for a non-array
-        // block member; -1 for a default-block uniform. glslang yields arrayStride==0 for the
-        // default-block case, so gate on block membership to return the spec-mandated -1.
+        // block member; -1 for a default-block uniform (glslang yields arrayStride==0 there, so gate
+        // on block membership for the spec-mandated -1). The stride itself is derived from the type
+        // instead of glslang's reflected arrayStride: for an array nested inside a struct member,
+        // glslang computes that field against the enclosing STRUCT's (unset) packing and reports a
+        // tight std430-like stride (ivec2 a[7] -> 8), even though its own member offsets and the
+        // generated SPIR-V lay the array out with std140 16-byte-rounded strides. MobileGL's UBO
+        // layout is always std140, where every array element stride rounds up to a vec4.
         GLint GetActiveUniformArrayStride(Uint index) const {
             const auto& uniform = m_program->getUniform(static_cast<Int>(index));
-            return (uniform.index < 0) ? -1 : uniform.arrayStride;
+            if (uniform.index < 0) return -1;
+            const glslang::TType* type = uniform.getType();
+            if (type == nullptr || !type->isArray()) return 0;
+            if (type->isMatrix()) {
+                const bool rowMajor = GetActiveUniformIsRowMajor(index) != 0;
+                const int vectors = rowMajor ? type->getMatrixRows() : type->getMatrixCols();
+                return GetActiveUniformMatrixStride(index) * vectors;
+            }
+            return 16; // scalars and vectors: std140 rounds the element stride up to a vec4
         }
 
         // GL_UNIFORM_IS_ROW_MAJOR: 1 only for a row-major matrix in a named block, else 0. The
@@ -170,6 +246,9 @@ namespace MobileGL::MG_State::GLState {
             auto& uniform = m_program->getUniform(static_cast<Int>(index));
             return uniform.name;
         }
+        // Sentinel for a uniform location without global-UBO backing storage (should not
+        // survive linking: GenerateBinary falls back to tail-allocated scratch storage).
+        static constexpr Uint kInvalidUniformOffset = ~0u;
         Uint GetUniformOffset(Uint location) const { return m_uniformOffsets[location]; }
         Uint GetUniformSizesInBytes(Uint location) const { return MG_Util::GetGLTypeSize(GetUniformType(location)); }
 
@@ -292,6 +371,11 @@ namespace MobileGL::MG_State::GLState {
         Uint GetUniformBlockIndex(const char* name) const {
             auto it = m_uniformBlockIndexByName.find(name);
             if (it != m_uniformBlockIndexByName.end()) return it->second;
+            // Instances of an arrayed block are reflected as "Block[0]".."Block[N-1]";
+            // a bare "Block" query resolves to the first instance per GL semantics.
+            const String suffixedName = String(name) + "[0]";
+            it = m_uniformBlockIndexByName.find(suffixedName);
+            if (it != m_uniformBlockIndexByName.end()) return it->second;
             return 0xFFFFFFFFu; // GL_INVALID_INDEX
         }
         Bool IsActiveUniformBlock(Uint index) const {
@@ -300,7 +384,11 @@ namespace MobileGL::MG_State::GLState {
         }
         Uint GetUBOSizeAt(Uint index) const {
             if (!IsActiveUniformBlock(index)) return 0;
-            return m_program->getUniformBlock((Int)index).size;
+            // glslang reports the unpadded end offset of the last member, but a std140 block
+            // (like a std140 struct) occupies a vec4-rounded size, and that is what the
+            // backend compiles: ES drivers reject draws whose bound UBO range is smaller
+            // than the block (a block ending in ivec3 reported 12 while the driver needs 16).
+            return (m_program->getUniformBlock((Int)index).size + 15u) & ~15u;
         }
 
         const String& GetUniformBlockName(Uint index) const {
@@ -308,8 +396,30 @@ namespace MobileGL::MG_State::GLState {
             return ubo.name;
         }
 
+        // Uniform entries that belong to an arrayed uniform block are reflected once, against
+        // the first instance ("Block[0]"); per GL semantics every other instance shares that
+        // member set. Maps any instance's block index to the index owning the member entries.
+        Uint GetUniformBlockMemberOwnerIndex(Uint index) const {
+            const String& name = GetUniformBlockName(index);
+            if (name.empty() || name.back() != ']') return index;
+            const SizeT bracket = name.rfind('[');
+            if (bracket == String::npos) return index;
+            const auto it = m_uniformBlockIndexByName.find(name.substr(0, bracket) + "[0]");
+            if (it != m_uniformBlockIndexByName.end()) return it->second;
+            return index;
+        }
+
+        // GL_UNIFORM_BLOCK_ACTIVE_UNIFORMS: derived from the same active-uniform scan that
+        // fills GL_UNIFORM_BLOCK_ACTIVE_UNIFORM_INDICES, so the two queries always agree
+        // (glslang's numMembers counts declared members, which diverges from the reflected
+        // entry list for struct arrays and arrayed block instances).
         Int GetUniformBlockActiveUniformCount(Uint index) const {
-            return m_program->getUniformBlock((Int)index).numMembers;
+            const Int ownerIndex = static_cast<Int>(GetUniformBlockMemberOwnerIndex(index));
+            Int count = 0;
+            for (Uint uniformIndex = 0; uniformIndex < m_activeUniformCount; ++uniformIndex) {
+                if (GetActiveUniformBlockIndex(uniformIndex) == ownerIndex) ++count;
+            }
+            return count;
         }
 
         Bool IsUniformBlockReferencedByStage(Uint index, EShLanguage stage) const {

MobileGL/MG_State/GLState/ProgramState/ProgramState.cpp

@@ -29,9 +29,18 @@ namespace MobileGL::MG_State::GLState {
         if (!CheckIndexAvail(program, m_programObjects)) return; // FIXME: add error reporting here
         auto& programObject = m_programObjects[program];
         if (programObject != nullptr) {
+            // Snapshot the attachments: deleting the program is a detach point for shaders
+            // that were flagged with glDeleteShader while still attached.
+            const Vector<SharedPtr<ShaderObject>> attachedShaders = programObject->GetAttachedShaders();
             programObject->MarkAsDeleted();
             programObject.reset();
             m_programIndexGenerator.Delete(program);
+            for (const auto& shader : attachedShaders) {
+                const Uint shaderName = shader->GetExternalIndex();
+                if (CheckIndexAvail(shaderName, m_shaderObjects) && m_shaderObjects[shaderName] == shader) {
+                    ReleaseShaderNameIfOrphaned(shaderName);
+                }
+            }
         }
     }
 
@@ -66,12 +75,35 @@ namespace MobileGL::MG_State::GLState {
         if (!CheckIndexAvail(shader, m_shaderObjects)) return;
         auto& shaderObject = m_shaderObjects[shader];
         if (shaderObject != nullptr) {
-            m_shaderObjects[shader]->MarkAsDeleted();
-            m_shaderObjects[shader].reset();
-            m_shaderIndexGenerator.Delete(shader);
+            // glDeleteShader on an attached shader only FLAGS it; the name stays valid (and
+            // glShaderSource/glCompileShader keep working on it) until the shader is detached
+            // from every program. The GL CTS compiles shaders through exactly this
+            // create-attach-delete-source-compile sequence (uniform_block.common.name_matching).
+            shaderObject->MarkAsDeleted();
+            ReleaseShaderNameIfOrphaned(shader);
         }
     }
 
+    Bool ProgramState::ShaderHasGLVisibleAttachment(const SharedPtr<ShaderObject>& shaderObject) const {
+        for (const auto& programObject : m_programObjects) {
+            if (programObject != nullptr && programObject->ShaderIsAttachedGLVisible(shaderObject)) {
+                return true;
+            }
+        }
+        // A program deleted while current vacates its table slot but stays alive as the
+        // current program; its attachments still count.
+        return m_currentProgram != nullptr && m_currentProgram->ShaderIsAttachedGLVisible(shaderObject);
+    }
+
+    void ProgramState::ReleaseShaderNameIfOrphaned(Uint shader) {
+        if (!CheckIndexAvail(shader, m_shaderObjects)) return;
+        auto& shaderObject = m_shaderObjects[shader];
+        if (shaderObject == nullptr || !shaderObject->GetDeleteStatus()) return;
+        if (ShaderHasGLVisibleAttachment(shaderObject)) return;
+        shaderObject.reset();
+        m_shaderIndexGenerator.Delete(shader);
+    }
+
     Bool ProgramState::ValidateShaderObject(Uint shader) const {
         return CheckIndexAvail(shader, m_shaderObjects) && m_shaderObjects[shader] != nullptr;
     }

MobileGL/MG_State/GLState/ProgramState/ProgramState.h

@@ -26,11 +26,16 @@ namespace MobileGL::MG_State::GLState {
         Uint CreateShader(ShaderStage stage);
         const SharedPtr<ShaderObject>& GetShaderObject(Uint shader);
         void MarkShaderObjectForDeletion(Uint shader);
+        // Frees a deletion-flagged shader's name once no program holds a GL-visible
+        // attachment to it (the deferred half of glDeleteShader-while-attached).
+        void ReleaseShaderNameIfOrphaned(Uint shader);
         Bool ValidateShaderObject(Uint shader) const;
 
         const SharedPtr<ProgramObject>& GetCurrentProgram() const { return m_currentProgram; }
 
     private:
+        Bool ShaderHasGLVisibleAttachment(const SharedPtr<ShaderObject>& shaderObject) const;
+
         template <typename T>
         static Bool CheckIndexAvail(const SizeT idx, const Vector<T>& vec) {
             return idx < vec.size();

MobileGL/MG_State/GLState/SamplerState/SamplerObject.cpp

@@ -78,6 +78,13 @@ namespace MobileGL {
                 ++m_version;
             }
 
+            void SamplerObject::SetMaxAnisotropy(Float maxAnisotropy) {
+                if (maxAnisotropy == m_samplerParameters.maxAnisotropy) return;
+
+                m_samplerParameters.maxAnisotropy = maxAnisotropy;
+                ++m_version;
+            }
+
             void SamplerObject::SetSamplerCompareFunc(SamplerCompareFunc func) {
                 if (func == m_samplerParameters.compareFunc) return;
 
@@ -128,6 +135,10 @@ namespace MobileGL {
                 return m_samplerParameters.lodBias;
             }
 
+            Float SamplerObject::GetMaxAnisotropy() const {
+                return m_samplerParameters.maxAnisotropy;
+            }
+
             SamplerCompareMode SamplerObject::GetCompareMode() const {
                 return m_samplerParameters.compareMode;
             }

MobileGL/MG_State/GLState/SamplerState/SamplerObject.h

@@ -65,6 +65,7 @@ namespace MobileGL {
         Float minLod = -1000.0f;
         Float maxLod = 1000.0f;
         Float lodBias = 0.0f;
+        Float maxAnisotropy = 1.0f;
         SamplerCompareFunc compareFunc = SamplerCompareFunc::Always;
         SamplerCompareMode compareMode = SamplerCompareMode::None;
     };
@@ -83,6 +84,7 @@ namespace MobileGL {
                 void SetMipmapMode(SamplerMipmapMode mode);
                 void SetLodRange(Float minLod, Float maxLod);
                 void SetLodBias(Float bias);
+                void SetMaxAnisotropy(Float maxAnisotropy);
                 void SetSamplerCompareFunc(SamplerCompareFunc func);
                 void SetCompareMode(SamplerCompareMode mode);
 
@@ -95,6 +97,7 @@ namespace MobileGL {
                 Float GetMinLod() const;
                 Float GetMaxLod() const;
                 Float GetLodBias() const;
+                Float GetMaxAnisotropy() const;
                 SamplerCompareMode GetCompareMode() const;
                 SamplerCompareFunc GetSamplerCompareFunc() const;
                 Uint GetExternalIndex() const;

MobileGL/MG_State/GLState/TextureState/TextureEnum.h

@@ -76,6 +76,14 @@ namespace MobileGL {
         BGRInteger,
         RGBAInteger,
         BGRAInteger,
+        // Desktop-GL single-channel client formats (table 3.3): the data holds one component that
+        // feeds the G, B or A channel; the remaining channels default to 0 (color) / 1 (alpha).
+        Green,
+        Blue,
+        Alpha,
+        GreenInteger,
+        BlueInteger,
+        AlphaInteger,
         StencilIndex,
         DepthComponent,
         DepthStencil,

MobileGL/MG_State/GLState/TextureState/TextureObject.cpp

@@ -7,6 +7,7 @@
 // End of Source File Header
 
 #include "TextureObject.h"
+#include "MG_State/GLState/Core.h"
 #include "MG_Util/Types.h"
 #include <MG_Util/Metrics/TextureMetrics.h>
 
@@ -52,6 +53,20 @@ namespace MobileGL {
             void TextureObjectBase::SetInternalFormat(TextureInternalFormat format) {
                 if (format == m_internalFormat) return;
 
+                // A default texture (name 0) changes IsUndefinedDefaultTexture on the
+                // Unknown<->defined transition, which changes per-draw sampled-set membership
+                // without any bind happening; bump the bind generation so cached sampled sets
+                // re-resolve instead of replaying the stale membership. The identity check
+                // excludes the other externalIndex-0 objects (proxy textures, default-FBO
+                // attachments) whose definedness never feeds sampled-set membership, so e.g.
+                // proxy probes cannot churn the cache.
+                if (m_externalIndex == 0 && pGLContext &&
+                    (m_internalFormat == TextureInternalFormat::Unknown) !=
+                        (format == TextureInternalFormat::Unknown) &&
+                    pGLContext->GetDefaultTextureObject(GetTarget()).get() == this) {
+                    pGLContext->BumpTextureBindGeneration();
+                }
+
                 m_internalFormat = format;
                 ++m_textureParamsVersion;
             }

MobileGL/MG_State/GLState/TextureState/TextureObject.h

@@ -154,6 +154,17 @@ namespace MobileGL::MG_State::GLState {
                    : nullptr;
     }
 
+    // The per-target default texture objects (name 0) sit permanently in every texture unit's
+    // binding slots, so "nothing useful bound" is no longer a null slot. While a default texture
+    // has never been given an image (its internal format is still Unknown) it can contribute
+    // nothing to sampling; backends treat such a binding exactly like the old empty slot and
+    // skip per-draw sync/bind work for it. Once an application defines an image on a default
+    // texture it loses this shortcut and is synced like any other texture.
+    inline Bool IsUndefinedDefaultTexture(const ITextureObject* texture) {
+        return texture != nullptr && texture->GetExternalIndex() == 0 &&
+               texture->GetFormat() == TextureInternalFormat::Unknown;
+    }
+
     class TextureObjectWithOneMipmap : public TextureObjectMipmap {
     public:
         TextureObjectWithOneMipmap(TextureTarget target, Uint externalIndex)

MobileGL/MG_State/GLState/TextureState/TextureState.cpp

@@ -18,9 +18,51 @@
 #include "TextureObjectStubs.h"
 
 namespace MobileGL::MG_State::GLState {
+    static SharedPtr<ITextureObject> MakeTextureObjectForTarget(Uint index, TextureTarget target) {
+        switch (target) {
+        case TextureTarget::Texture1D:
+            return MakeShared<TextureObject1D>(index);
+        case TextureTarget::TextureCubeMap:
+            return MakeShared<TextureObject2DCube>(index);
+        case TextureTarget::Texture2D:
+            return MakeShared<TextureObject2D>(index);
+        case TextureTarget::Texture3D:
+            return MakeShared<TextureObject3D>(index);
+        case TextureTarget::TextureBuffer:
+            return MakeShared<TextureObjectBuffer>(index);
+
+            // These texture types are still stubbed:
+        case TextureTarget::TextureRectangle:
+            return MakeShared<TextureObjectRectangle>(index);
+        case TextureTarget::Texture2DMultisample:
+            return MakeShared<TextureObject2DMultisample>(index);
+        case TextureTarget::Texture1DArray:
+            return MakeShared<TextureObject1DArray>(index);
+        case TextureTarget::Texture2DArray:
+            return MakeShared<TextureObject2DArray>(index);
+        case TextureTarget::TextureCubeMapArray:
+            return MakeShared<TextureObjectCubeMapArray>(index);
+        case TextureTarget::Texture2DMultisampleArray:
+            return MakeShared<TextureObject2DMultisampleArray>(index);
+        default:
+            MOBILEGL_ASSERT(false, "Unimplemented texture type when creating texture object!: %d", (int)target);
+            return nullptr;
+        }
+    }
+
     TextureState::TextureState() : m_indexGenerator(1024, 1) {
+        // GL 3.3 core 3.8: each target owns one default texture object (name 0) per context,
+        // shared across all texture units, and it is the initial binding of every unit/target
+        // slot. It is created outside m_textureObjects so name-based paths (glIsTexture,
+        // GenTextures/DeleteTextures, by-name DSA lookups) never see it.
+        for (int i = 0; i < (int)TextureTarget::TextureTargetCount; ++i) {
+            m_defaultTextureObjects[i] = MakeTextureObjectForTarget(0, static_cast<TextureTarget>(i));
+        }
         for (int i = 0; i < MAX_TEXTURE_IMAGE_UNITS; ++i) {
             m_textureUnits[i] = TextureUnit();
+            for (auto& bindingSlot : m_textureUnits[i].GetAllBindingSlots()) {
+                bindingSlot.Bind(m_defaultTextureObjects[(int)bindingSlot.GetTarget()]);
+            }
         }
     }
 
@@ -33,6 +75,12 @@ namespace MobileGL::MG_State::GLState {
         return nullTextureObject;
     }
 
+    const SharedPtr<ITextureObject>& TextureState::GetDefaultTextureObject(TextureTarget target) const {
+        MOBILEGL_ASSERT(target > TextureTarget::Unknown && target < TextureTarget::TextureTargetCount,
+                        "GetDefaultTextureObject: invalid texture target %d", (int)target);
+        return m_defaultTextureObjects[(int)target];
+    }
+
     void TextureState::GenerateNames(Uint number, Vector<Uint>& textures) {
         textures.resize(number);
         m_indexGenerator.Generate(number, textures.data());
@@ -40,52 +88,15 @@ namespace MobileGL::MG_State::GLState {
 
     const SharedPtr<ITextureObject>& TextureState::CreateTextureObject(Uint index, TextureTarget target) {
         auto& textureObject = m_textureObjects[index];
-        switch (target) {
-        case TextureTarget::Texture1D:
-            textureObject = MakeShared<TextureObject1D>(index);
-            break;
-        case TextureTarget::TextureCubeMap:
-            textureObject = MakeShared<TextureObject2DCube>(index);
-            break;
-        case TextureTarget::Texture2D:
-            textureObject = MakeShared<TextureObject2D>(index);
-            break;
-        case TextureTarget::Texture3D:
-            textureObject = MakeShared<TextureObject3D>(index);
-            break;
-        case TextureTarget::TextureBuffer:
-            textureObject = MakeShared<TextureObjectBuffer>(index);
-            break;
-
-            // These texture types are still stubbed:
-        case TextureTarget::TextureRectangle:
-            textureObject = MakeShared<TextureObjectRectangle>(index);
-            break;
-        case TextureTarget::Texture2DMultisample:
-            textureObject = MakeShared<TextureObject2DMultisample>(index);
-            break;
-        case TextureTarget::Texture1DArray:
-            textureObject = MakeShared<TextureObject1DArray>(index);
-            break;
-        case TextureTarget::Texture2DArray:
-            textureObject = MakeShared<TextureObject2DArray>(index);
-            break;
-        case TextureTarget::TextureCubeMapArray:
-            textureObject = MakeShared<TextureObjectCubeMapArray>(index);
-            break;
-        case TextureTarget::Texture2DMultisampleArray:
-            textureObject = MakeShared<TextureObject2DMultisampleArray>(index);
-            break;
-        default:
-            MOBILEGL_ASSERT(false, "Unimplemented texture type when creating texture object!: %d", (int)target);
+        textureObject = MakeTextureObjectForTarget(index, target);
+        if (!textureObject) {
             static SharedPtr<ITextureObject> nullTextureObject = nullptr;
             return nullTextureObject;
         }
-
         return textureObject;
     }
 
-    void TextureState::MarkTextureObjectForDeletion(Uint index) {
+    void TextureState::MarkTextureObjectForDeletion(Uint index, Bool keepUnboundReservation) {
         if (m_indexGenerator.IsValid(index)) {
             auto it = m_textureObjects.find(index);
             if (it != m_textureObjects.end()) {
@@ -94,7 +105,9 @@ namespace MobileGL::MG_State::GLState {
                     auto& bindingSlots = m_textureUnits[unit].GetAllBindingSlots();
                     for (auto& bindingSlot : bindingSlots) {
                         if (bindingSlot.GetBoundObject() == it->second) {
-                            bindingSlot.Bind(nullptr);
+                            // GL 3.3 core 3.8.1: deleting a bound texture rebinds zero, i.e. the
+                            // target's default texture object, on every unit it was bound to.
+                            bindingSlot.Bind(m_defaultTextureObjects[(int)bindingSlot.GetTarget()]);
                         }
                     }
                 }
@@ -109,8 +122,16 @@ namespace MobileGL::MG_State::GLState {
                 // re-resolved instead of dangling.
                 BumpTextureBindGeneration();
                 m_textureObjects.erase(index);
+                m_indexGenerator.Delete(index);
+            } else if (!keepUnboundReservation) {
+                // GL 3.3 core 3.8.1 makes a deleted name unused again even when GenTextures only
+                // reserved it and no bind ever instantiated an object (so a later bind of it must
+                // fail), and the reservation has to return to the free list.
+                m_indexGenerator.Delete(index);
             }
-            m_indexGenerator.Delete(index);
+            // Relaxed semantics: legacy apps may delete a generated name before its first
+            // bind, then bind and populate that same name. Keep such a reservation alive there;
+            // a real texture object reaching deletion still releases its index above.
         }
     }
 

MobileGL/MG_State/GLState/TextureState/TextureState.h

@@ -49,12 +49,18 @@ namespace MobileGL::MG_State::GLState {
         void GenerateNames(Uint number, Vector<Uint>& textures);
         const SharedPtr<ITextureObject>& CreateTextureObject(Uint index, TextureTarget target);
         const SharedPtr<ITextureObject>& GetTextureObject(Uint index);
+        // The context's default texture object (name 0) for `target`. GL 3.3 core 3.8: texture
+        // zero names a real, per-target texture object shared by every texture unit; binding 0
+        // binds it, and image/parameter calls on it must work like on any texture. It is not a
+        // GenTextures name: it lives outside m_textureObjects (so glIsTexture(0) stays GL_FALSE
+        // and by-name lookups keep failing for 0) and can never be deleted.
+        const SharedPtr<ITextureObject>& GetDefaultTextureObject(TextureTarget target) const;
         TextureUnit& GetUnitObject(Int unit);
         ImageTextureBinding& GetImageTextureBinding(Int unit);
         const ImageTextureBinding& GetImageTextureBinding(Int unit) const;
         Int GetActiveTextureUnit() const;
         void SetActiveTextureUnit(Int unit);
-        void MarkTextureObjectForDeletion(Uint index);
+        void MarkTextureObjectForDeletion(Uint index, Bool keepUnboundReservation);
         Bool ValidateName(Uint index) const;
         Bool ValidateTextureObject(Uint index) const;
 
@@ -83,5 +89,8 @@ namespace MobileGL::MG_State::GLState {
         Array<ImageTextureBinding, MAX_TEXTURE_IMAGE_UNITS> m_imageTextureBindings;
         IndexGenerator<Uint> m_indexGenerator;
         UnorderedMap<GLuint, SharedPtr<ITextureObject>> m_textureObjects;
+        // One default texture object (external name 0) per target, created with the context and
+        // immortal for its lifetime; the initial binding of every unit/target slot.
+        Array<SharedPtr<ITextureObject>, (int)TextureTarget::TextureTargetCount> m_defaultTextureObjects;
     };
 } // namespace MobileGL::MG_State::GLState

MobileGL/MG_Test/Backend/DirectVulkan/CMakeLists.txt

@@ -40,10 +40,16 @@ if (APPLE)
     target_link_libraries(DirectVulkanSanityTest PRIVATE objc)
     target_link_libraries(DirectVulkanSanityTest PRIVATE ${QUARTZCORE_FRAMEWORK})
 endif()
+if (MSVC)
+    # The GLES headers declare gl* as dllimport on Windows, so objects like GetProcAddress.cpp
+    # reference __imp_gl*. Those resolve against the in-library GL entry-point definitions only
+    # if their objects are part of the link; pull the whole static library like the DLL does.
+    target_link_options(DirectVulkanSanityTest PRIVATE /WHOLEARCHIVE:MobileGL_s)
+endif()
 target_compile_definitions(DirectVulkanSanityTest PRIVATE -DNOMINMAX)
 
 include(GoogleTest)
-gtest_discover_tests(DirectVulkanSanityTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(DirectVulkanSanityTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS integration)
 
 add_executable(
         DirectVulkanTestExec

MobileGL/MG_Test/BackendLoader/BackendLoaderTest.cpp

@@ -9,8 +9,13 @@
 #include <gtest/gtest.h>
 #include <cstring>
 #include <map>
+#include <string>
 #include <vector>
 
+#include <algorithm>
+
+#include <MG_Backend/DirectGLES/BackendObject_DirectGLES.h>
+#include <MG_Backend/DirectVulkan/BackendObject_DirectVulkan.h>
 #include <MG_Util/BackendLoaders/OpenGL/Loader.h>
 
 // ProbeIndirectInstanceIdIncludesBaseInstance is driven against a fake GLES driver:
@@ -28,6 +33,12 @@ namespace {
         GLenum errorRaisedByDraw = GL_NO_ERROR;
 
         GLenum pendingError = GL_NO_ERROR;
+        std::vector<std::string> extensions;
+
+        // GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT the fake reports, and whether it was ever asked:
+        // querying it on a driver without the extension would raise GL_INVALID_ENUM.
+        GLfloat maxTextureMaxAnisotropy = 16.0f;
+        bool maxTextureMaxAnisotropyQueried = false;
 
         GLuint nextBufferId = 1;
         GLuint nextShaderId = 1;
@@ -86,7 +97,7 @@ namespace {
                 *data = 1;
                 break;
             case GL_NUM_EXTENSIONS:
-                *data = 0;
+                *data = static_cast<GLint>(g_fake.extensions.size());
                 break;
             default:
                 // Leave the caller's defaults for every other capability query.
@@ -114,11 +125,16 @@ namespace {
                 return reinterpret_cast<const GLubyte*>("");
             }
         };
-        // GL_NUM_EXTENSIONS reports 0 above, so this is never reached; it exists so the
-        // table stays complete if the extension loop ever runs.
-        funcs.glGetStringi = [](GLenum, GLuint) -> const GLubyte* { return nullptr; };
+        funcs.glGetStringi = [](GLenum name, GLuint index) -> const GLubyte* {
+            if (name != GL_EXTENSIONS || index >= g_fake.extensions.size()) return nullptr;
+            return reinterpret_cast<const GLubyte*>(g_fake.extensions[index].c_str());
+        };
         funcs.glGetFloatv = [](GLenum pname, GLfloat* data) {
             switch (pname) {
+            case GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT:
+                g_fake.maxTextureMaxAnisotropyQueried = true;
+                data[0] = g_fake.maxTextureMaxAnisotropy;
+                break;
             // Two-component range queries.
             case GL_ALIASED_LINE_WIDTH_RANGE:
             case GL_SMOOTH_LINE_WIDTH_RANGE:
@@ -400,3 +416,65 @@ TEST(IndirectInstanceIdProbe, FillInCapabilitiesWiresProbeResult) {
     EXPECT_FALSE(conformingCaps.IndirectDrawInstanceIdIncludesBaseInstance);
     ExpectProbeReleasedAllObjects();
 }
+
+// The extension string is what apps gate on (LWJGL builds GLCapabilities from it), so advertising
+// it on a driver that cannot filter anisotropically would leave them silently on trilinear.
+TEST(TextureAnisotropyCapabilities, ExtensionIsAdvertisedOnlyWhenTheHostDriverSupportsIt) {
+    const auto contains = [](const MobileGL::Vector<MobileGL::GLExtension>& extensions,
+                             MobileGL::GLExtension wanted) {
+        return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
+    };
+
+    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(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);
+    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);
+    EXPECT_FALSE(contains(vkWithout, MobileGL::E_GL_EXT_texture_filter_anisotropic));
+    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true);
+    EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_EXT_texture_filter_anisotropic));
+    EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_ARB_texture_filter_anisotropic));
+}
+
+TEST(TextureAnisotropyCapabilities, MaxAnisotropyIsQueriedOnlyWhenTheExtensionIsPresent) {
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    const auto funcs = MakeFakeGLESFunctions();
+
+    MobileGL::MG_External::GLESCapabilities absentCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(absentCaps, funcs));
+    // Never probed (it would be GL_INVALID_ENUM), and reported as "no anisotropy".
+    EXPECT_FALSE(g_fake.maxTextureMaxAnisotropyQueried);
+    EXPECT_FLOAT_EQ(absentCaps.MaxTextureMaxAnisotropy, 1.0f);
+
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    g_fake.maxTextureMaxAnisotropy = 16.0f;
+    g_fake.extensions.emplace_back("GL_EXT_texture_filter_anisotropic");
+    MobileGL::MG_External::GLESCapabilities presentCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(presentCaps, funcs));
+    EXPECT_TRUE(g_fake.maxTextureMaxAnisotropyQueried);
+    EXPECT_FLOAT_EQ(presentCaps.MaxTextureMaxAnisotropy, 16.0f);
+}
+
+TEST(TextureAnisotropyCapabilities, ExtensionPresenceIsDetectedExactly) {
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    const auto funcs = MakeFakeGLESFunctions();
+
+    MobileGL::MG_External::GLESCapabilities absentCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(absentCaps, funcs));
+    EXPECT_FALSE(absentCaps.SupportsTextureFilterAnisotropy);
+
+    ResetFakeDriver();
+    g_fake.maxVertexSsboBlocks = 0;
+    g_fake.extensions.emplace_back("GL_EXT_texture_filter_anisotropic");
+    MobileGL::MG_External::GLESCapabilities presentCaps;
+    ASSERT_TRUE(MobileGL::MG_Util::BackendLoader::FillInGLESCapabilities(presentCaps, funcs));
+    EXPECT_TRUE(presentCaps.SupportsTextureFilterAnisotropy);
+}

MobileGL/MG_Test/BackendLoader/CMakeLists.txt

@@ -17,4 +17,4 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(BackendLoaderTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(BackendLoaderTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/Buffer/BufferTest.cpp

@@ -8,6 +8,8 @@
 
 #include <gtest/gtest.h>
 
+#include <limits>
+
 #include "Includes.h"
 #include "Init.h"
 #include <Config.h>
@@ -20,9 +22,31 @@ using namespace MobileGL;
 
 class BufferTest : public ::testing::Test {
 protected:
-    void SetUp() override { MobileGL::Initialize(); }
+    // GL error flags are sticky per error code and the context outlives an individual test in this
+    // binary, so drain whatever an earlier test left pending - otherwise an error-code assertion
+    // here reads someone else's error. Bounded: one flag per code, so this cannot hang the suite.
+    static void DrainPendingGlErrors() {
+        for (Int drained = 0; drained < 16 && MG_Impl::GLImpl::GetError() != GL_NO_ERROR; ++drained) {
+        }
+    }
 
-    void TearDown() override {}
+    // The call under test must raise exactly the expected error and nothing more: a second pending
+    // error means one entry point queued several, which GetError() would hand out at an unrelated
+    // call site later on.
+    static void ExpectSingleGlError(GLenum expected) {
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), expected);
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "the call recorded more than one error";
+    }
+
+    void SetUp() override {
+        MobileGL::Initialize();
+        DrainPendingGlErrors();
+    }
+
+    void TearDown() override {
+        // Attribute a leaked error to the test that caused it instead of to whoever runs next.
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "test left an unconsumed GL error behind";
+    }
 };
 
 TEST_F(BufferTest, Binding) {
@@ -105,6 +129,53 @@ TEST_F(BufferTest, GenerateManyNames_NoPrematureCreation) {
     }
 }
 
+// GL 3.3 core 2.9 name lifecycle. The same three rules are asserted per object family (see the
+// texture/vertex-array/framebuffer/renderbuffer suites): a deleted or never-generated name is
+// INVALID_OPERATION to bind, deleting one is silent, and a generated-but-never-bound reservation
+// is still released so the name gets recycled.
+TEST_F(BufferTest, DeleteOfUnknownOrAlreadyDeletedBufferNameIsSilent) {
+    GLuint buffer = 0;
+    MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    ASSERT_NE(buffer, 0u);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // Double delete, name 0 and a never-generated name must all be ignored without an error.
+    MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const GLuint unknownNames[] = {0u, std::numeric_limits<GLuint>::max()};
+    MG_Impl::GLImpl::DeleteBuffers(2, unknownNames);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(BufferTest, DeleteGeneratedButUnboundBufferNameReleasesReservationAndBindFails) {
+    GLuint buffer = 0;
+    MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    ASSERT_NE(buffer, 0u);
+    ASSERT_TRUE(MG_State::pGLContext->ValidateBufferName(buffer));
+
+    MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+    EXPECT_FALSE(MG_State::pGLContext->ValidateBufferName(buffer));
+
+    MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, buffer);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    GLuint recycled = 0;
+    MG_Impl::GLImpl::GenBuffers(1, &recycled);
+    EXPECT_EQ(recycled, buffer);
+}
+
+TEST_F(BufferTest, BindNeverGeneratedBufferNameIsInvalidOperation) {
+    // Not a small literal: other tests in this binary share the context and generate names in
+    // bulk, so a low number may well be a legitimately reserved name here.
+    MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, std::numeric_limits<GLuint>::max());
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+}
+
 TEST_F(BufferTest, AcquireMemory) {
     auto& slot = MobileGL::MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::Uniform);
     Vector<Uint> bufferNames;

MobileGL/MG_Test/Buffer/CMakeLists.txt

@@ -24,4 +24,4 @@ if (MSVC)
 endif()
 
 include(GoogleTest)
-gtest_discover_tests(BufferTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(BufferTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/CMakeLists.txt

@@ -62,7 +62,7 @@ set(LINK_LIBRARIES
 )
 
 include(GoogleTest)
-gtest_discover_tests(SanityTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(SanityTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)
 
 add_subdirectory(BackendLoader)
 add_subdirectory(Buffer)

MobileGL/MG_Test/EGLState/CMakeLists.txt

@@ -17,4 +17,4 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(EGLStateTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(EGLStateTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/EGLState/EGLStateTest.cpp

@@ -102,3 +102,33 @@ TEST(EGLStateMakeCurrent, SameThreadRepeatedAttachReleaseDoesNotLeaveStaleOwner)
     EXPECT_TRUE(fixture->State.MakeCurrent(EGL_NO_DISPLAY, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT));
     EXPECT_EQ(fixture->State.ConsumeError(), EGL_SUCCESS);
 }
+
+// The compatibility-profile accessor is affirmative-only (it backs GL_CONTEXT_PROFILE_MASK
+// reporting): attrib-less contexts (profile mask 0) and released threads both answer "not
+// compat" and therefore read as core-profile contexts.
+TEST(EGLStateProfile, CompatibilityProfileRequiresExplicitCompatBit) {
+    auto fixture = CreateFixture();
+
+    EXPECT_FALSE(fixture->State.IsCurrentContextOpenGLCompatibilityProfile());
+
+    EXPECT_TRUE(fixture->State.MakeCurrent(fixture->Display, fixture->Surface, fixture->Surface, fixture->Context));
+    EXPECT_FALSE(fixture->State.IsCurrentContextOpenGLCompatibilityProfile());
+
+    const EGLint compatAttribs[] = {EGL_CONTEXT_MAJOR_VERSION,
+                                    3,
+                                    EGL_CONTEXT_MINOR_VERSION,
+                                    3,
+                                    EGL_CONTEXT_OPENGL_PROFILE_MASK,
+                                    EGL_CONTEXT_OPENGL_COMPATIBILITY_PROFILE_BIT,
+                                    EGL_NONE};
+    const auto compatContext =
+        fixture->State.CreateContext(fixture->Display, fixture->Config, EGL_NO_CONTEXT, compatAttribs);
+    ASSERT_NE(compatContext, EGL_NO_CONTEXT);
+    EXPECT_TRUE(fixture->State.MakeCurrent(fixture->Display, fixture->Surface, fixture->Surface, compatContext));
+    EXPECT_TRUE(fixture->State.IsCurrentContextOpenGLCompatibilityProfile());
+    EXPECT_FALSE(fixture->State.IsCurrentContextOpenGLCoreProfile());
+
+    EXPECT_TRUE(fixture->State.MakeCurrent(EGL_NO_DISPLAY, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT));
+    EXPECT_FALSE(fixture->State.IsCurrentContextOpenGLCompatibilityProfile());
+    EXPECT_EQ(fixture->State.ConsumeError(), EGL_SUCCESS);
+}

MobileGL/MG_Test/Framebuffer/CMakeLists.txt

@@ -17,4 +17,4 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(FramebufferTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(FramebufferTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/Framebuffer/FramebufferTest.cpp

@@ -8,9 +8,12 @@
 
 #include <gtest/gtest.h>
 
+#include <limits>
+
 #include "Includes.h"
 #include "Init.h"
 #include <MG_Backend/BackendObjects.h>
+#include <MG_Backend/DirectGLES/Utils.h>
 #include <MG_Impl/GLImpl/Buffer/GL_Buffer.h>
 #include <MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h>
 #include <MG_Impl/GLImpl/Getter/GL_Getter.h>
@@ -32,6 +35,8 @@ namespace {
     Int g_clearNamedFramebufferfvCallCount = 0;
     Int g_clearNamedFramebufferfiCallCount = 0;
     Int g_readPixelsCallCount = 0;
+    GLenum g_lastReadPixelsFormat = GL_NONE;
+    GLenum g_lastReadPixelsType = GL_NONE;
 
     void RecordBlitNamedFramebuffer(const SharedPtr<MG_State::GLState::FramebufferObject>& readFramebuffer,
                                     const SharedPtr<MG_State::GLState::FramebufferObject>& drawFramebuffer,
@@ -66,15 +71,39 @@ namespace {
         ++g_clearNamedFramebufferfiCallCount;
     }
 
-    void RecordReadPixels(GLint, GLint, GLsizei, GLsizei, GLenum, GLenum, void*) {
+    void RecordReadPixels(GLint, GLint, GLsizei, GLsizei, GLenum format, GLenum type, void*) {
         ++g_readPixelsCallCount;
+        g_lastReadPixelsFormat = format;
+        g_lastReadPixelsType = type;
     }
 } // namespace
 
 class FramebufferTest : public ::testing::Test {
 protected:
+    // GL error flags are sticky per error code and the context outlives an individual test in this
+    // binary, so drain whatever an earlier test left pending - otherwise an error-code assertion
+    // here reads someone else's error. Bounded: one flag per code, so this cannot hang the suite.
+    static void DrainPendingGlErrors() {
+        for (Int drained = 0; drained < 16 && MG_Impl::GLImpl::GetError() != GL_NO_ERROR; ++drained) {
+        }
+    }
+
+    // The call under test must raise exactly the expected error and nothing more: a second pending
+    // error means one entry point queued several, which GetError() would hand out at an unrelated
+    // call site later on.
+    static void ExpectSingleGlError(GLenum expected) {
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), expected);
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "the call recorded more than one error";
+    }
+
+    void TearDown() override {
+        // Attribute a leaked error to the test that caused it instead of to whoever runs next.
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "test left an unconsumed GL error behind";
+    }
+
     void SetUp() override {
         MobileGL::Initialize();
+        DrainPendingGlErrors();
         const auto defaultFramebuffer = MG_State::pGLContext->GetFramebufferObject(0);
         ASSERT_NE(defaultFramebuffer, nullptr);
         MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Draw).Bind(defaultFramebuffer);
@@ -97,6 +126,8 @@ protected:
         g_clearNamedFramebufferfvCallCount = 0;
         g_clearNamedFramebufferfiCallCount = 0;
         g_readPixelsCallCount = 0;
+        g_lastReadPixelsFormat = GL_NONE;
+        g_lastReadPixelsType = GL_NONE;
         MG_Backend::gBackendFunctionsTable.GL.BlitNamedFramebuffer = nullptr;
         MG_Backend::gBackendFunctionsTable.GL.ClearNamedFramebufferfv = nullptr;
         MG_Backend::gBackendFunctionsTable.GL.ClearNamedFramebufferfi = nullptr;
@@ -115,6 +146,83 @@ TEST_F(FramebufferTest, CreateFramebuffersCreatesObjectsImmediately) {
     EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
 
+// GL 3.3 core 4.4.1/4.4.2 name lifecycle - mirrors the rules asserted for the other object
+// families: deleting an unknown name is silent, a released reservation is recycled, and binding
+// a dead name is INVALID_OPERATION.
+TEST_F(FramebufferTest, DeleteOfUnknownOrAlreadyDeletedFramebufferNameIsSilent) {
+    GLuint framebuffer = 0;
+    MG_Impl::GLImpl::GenFramebuffers(1, &framebuffer);
+    ASSERT_NE(framebuffer, 0u);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteFramebuffers(1, &framebuffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteFramebuffers(1, &framebuffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // Not a small literal: other tests in this binary share the context and generate names in
+    // bulk, so a low number may well be a legitimately reserved name here.
+    const GLuint unknownNames[] = {0u, std::numeric_limits<GLuint>::max()};
+    MG_Impl::GLImpl::DeleteFramebuffers(2, unknownNames);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(FramebufferTest, DeleteGeneratedButUnboundFramebufferNameReleasesReservationAndBindFails) {
+    GLuint framebuffer = 0;
+    MG_Impl::GLImpl::GenFramebuffers(1, &framebuffer);
+    ASSERT_NE(framebuffer, 0u);
+    ASSERT_TRUE(MG_State::pGLContext->ValidateFramebufferName(framebuffer));
+
+    MG_Impl::GLImpl::DeleteFramebuffers(1, &framebuffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+    EXPECT_FALSE(MG_State::pGLContext->ValidateFramebufferName(framebuffer));
+
+    MG_Impl::GLImpl::BindFramebuffer(GL_FRAMEBUFFER, framebuffer);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    GLuint recycled = 0;
+    MG_Impl::GLImpl::GenFramebuffers(1, &recycled);
+    EXPECT_EQ(recycled, framebuffer);
+}
+
+TEST_F(FramebufferTest, DeleteOfUnknownOrAlreadyDeletedRenderbufferNameIsSilent) {
+    GLuint renderbuffer = 0;
+    MG_Impl::GLImpl::GenRenderbuffers(1, &renderbuffer);
+    ASSERT_NE(renderbuffer, 0u);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteRenderbuffers(1, &renderbuffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteRenderbuffers(1, &renderbuffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // Not a small literal: other tests in this binary share the context and generate names in
+    // bulk, so a low number may well be a legitimately reserved name here.
+    const GLuint unknownNames[] = {0u, std::numeric_limits<GLuint>::max()};
+    MG_Impl::GLImpl::DeleteRenderbuffers(2, unknownNames);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(FramebufferTest, DeleteGeneratedButUnboundRenderbufferNameReleasesReservationAndBindFails) {
+    GLuint renderbuffer = 0;
+    MG_Impl::GLImpl::GenRenderbuffers(1, &renderbuffer);
+    ASSERT_NE(renderbuffer, 0u);
+    ASSERT_TRUE(MG_State::pGLContext->ValidateRenderbufferName(renderbuffer));
+
+    MG_Impl::GLImpl::DeleteRenderbuffers(1, &renderbuffer);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+    EXPECT_FALSE(MG_State::pGLContext->ValidateRenderbufferName(renderbuffer));
+
+    MG_Impl::GLImpl::BindRenderbuffer(GL_RENDERBUFFER, renderbuffer);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    GLuint recycled = 0;
+    MG_Impl::GLImpl::GenRenderbuffers(1, &recycled);
+    EXPECT_EQ(recycled, renderbuffer);
+}
+
 TEST_F(FramebufferTest, DefaultFramebufferIdentityTracksFramebufferNameZero) {
     const auto defaultFramebuffer = MG_State::pGLContext->GetFramebufferObject(0);
     ASSERT_NE(defaultFramebuffer, nullptr);
@@ -221,6 +329,77 @@ TEST_F(FramebufferTest, ReadPixelsAllowsPersistentMappedPixelPackBuffer) {
     EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
 
+TEST_F(FramebufferTest, ReadPixelsRejectsMismatchedPackedTypeFormatPairs) {
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 1, GL_RGBA8, 4, 4);
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, texture, 0);
+    MG_Impl::GLImpl::BindFramebuffer(GL_READ_FRAMEBUFFER, framebuffer);
+
+    MG_Backend::gBackendFunctionsTable.GL.ReadPixels = RecordReadPixels;
+    Uint8 pixelStorage[4 * 4 * 4] = {};
+
+    // Packed RGB type with a non-RGB format must never reach the backend (GL CTS packed_pixels
+    // reads GL_RED with GL_UNSIGNED_SHORT_5_6_5 and expects an error).
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RED, GL_UNSIGNED_SHORT_5_6_5, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+
+    // Packed RGBA type with a non-RGBA/BGRA format is rejected as well.
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+
+    // Packed depth-stencil type requires the DEPTH_STENCIL format.
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGBA, GL_UNSIGNED_INT_24_8, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+
+    // A plain RGBA/UNSIGNED_BYTE readback keeps working.
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGBA, GL_UNSIGNED_BYTE, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(FramebufferTest, ReadPixelsForwardsSingleChannelDesktopClientFormats) {
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 1, GL_RGBA8, 4, 4);
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, texture, 0);
+    MG_Impl::GLImpl::BindFramebuffer(GL_READ_FRAMEBUFFER, framebuffer);
+
+    MG_Backend::gBackendFunctionsTable.GL.ReadPixels = RecordReadPixels;
+    Uint8 pixelStorage[4 * 4 * 4] = {};
+
+    // Desktop GL treats GL_GREEN/GL_BLUE/GL_ALPHA as valid ReadPixels client formats (GL CTS
+    // packed_pixels rgba8_format_green failed with GL_INVALID_ENUM before). The state layer must
+    // validate them and forward the raw enum to the backend, which extracts the source channel
+    // from a wide RGBA read.
+    const GLenum singleChannelFormats[] = {GL_GREEN, GL_BLUE, GL_ALPHA};
+    Int expectedCallCount = 0;
+    for (const GLenum format : singleChannelFormats) {
+        MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, format, GL_UNSIGNED_BYTE, pixelStorage);
+        EXPECT_EQ(g_readPixelsCallCount, ++expectedCallCount);
+        EXPECT_EQ(g_lastReadPixelsFormat, format);
+        EXPECT_EQ(g_lastReadPixelsType, static_cast<GLenum>(GL_UNSIGNED_BYTE));
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+    }
+
+    // Packed-type pairing still applies: packed RGB/RGBA types never pair with single-channel formats.
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_GREEN, GL_UNSIGNED_SHORT_5_6_5, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, expectedCallCount);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+
+    // Integer client formats reject floating-point types.
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_GREEN_INTEGER, GL_FLOAT, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, expectedCallCount);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+}
+
 TEST_F(FramebufferTest, NamedRenderbufferStorageAndFramebufferAttachDoNotChangeBindings) {
     GLuint framebuffer = 0;
     GLuint renderbuffer = 0;
@@ -419,3 +598,263 @@ TEST_F(FramebufferTest, BlitNamedFramebufferAllowsDefaultFramebufferZero) {
     EXPECT_EQ(MG_State::pGLContext->GetFramebufferBindingSlot(FramebufferTarget::Read).GetBoundObject(), defaultRead);
     EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
+
+// ---- Packed-type readback encoding ------------------------------------------------------------------
+// Oracle-independent guard for the DirectGLES client-format readback conversion: feeds known wide RGBA
+// rows through ReadbackImpl::ConvertWideReadbackRow and asserts the exact packed words. Field positions
+// were hand-computed from GL 3.3 table 3.6 and match the GL CTS packed_pixels comparison functions
+// (glcPackedPixelsTests.cpp pack_UNSIGNED_*): non-REV types pack the first format component from the
+// most significant bit, *_REV types from the least significant bit.
+
+namespace {
+    namespace ReadbackImpl = MG_Backend::DirectGLES::ReadbackImpl;
+
+    // Converts a row of wide pixels (4 components of wideType each) into `format`/`type` words.
+    template <typename WordT, typename SrcT>
+    Vector<WordT> ConvertWideRowToPackedWords(const Vector<SrcT>& wide, GLenum wideType, GLenum format,
+                                              GLenum type) {
+        ReadbackImpl::ReadbackChannelMapping mapping{};
+        EXPECT_TRUE(ReadbackImpl::GetReadbackChannelMapping(format, mapping));
+        EXPECT_EQ(ReadbackImpl::GetReadbackDstPixelSize(mapping, type), sizeof(WordT));
+        const SizeT width = wide.size() / 4;
+        Vector<WordT> out(width, static_cast<WordT>(0));
+        ReadbackImpl::ConvertWideReadbackRow(reinterpret_cast<const Uint8*>(wide.data()),
+                                             reinterpret_cast<Uint8*>(out.data()), width, wideType, mapping,
+                                             type);
+        return out;
+    }
+
+    // Normalized encodes read the wide row as RGBA8 (values are v / 255).
+    template <typename WordT>
+    Vector<WordT> ConvertRGBA8Row(const Vector<Uint8>& rgba, GLenum format, GLenum type) {
+        return ConvertWideRowToPackedWords<WordT>(rgba, GL_UNSIGNED_BYTE, format, type);
+    }
+
+    // Wide RGBA8 pattern shared by the normalized-encode tests. Expected fields below are
+    // round(v / 255 * (2^bits - 1)), computed by hand per pixel.
+    //                            R     G     B     A
+    const Vector<Uint8> kRGBA8Row{255,  0,    128,  64,     // P0
+                                  10,   250,  33,   200,    // P1
+                                  85,   170,  255,  0};     // P2
+} // namespace
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedShort565) {
+    // P0: R=31 G=0 B=round(128*31/255)=16      -> 31<<11 | 0<<5 | 16      = 0xF810
+    // P1: R=round(10*31/255)=1 G=round(250*63/255)=62 B=round(33*31/255)=4 -> 1<<11|62<<5|4 = 0x0FC4
+    // P2: R=round(85*31/255)=10 G=round(170*63/255)=42 B=31               -> 10<<11|42<<5|31 = 0x555F
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_RGB, GL_UNSIGNED_SHORT_5_6_5);
+    EXPECT_EQ(words[0], 0xF810u);
+    EXPECT_EQ(words[1], 0x0FC4u);
+    EXPECT_EQ(words[2], 0x555Fu);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedShort565Rev) {
+    // REV packs R from the LSB: P2 -> 10 | 42<<5 | 31<<11 = 0xFD4A
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV);
+    EXPECT_EQ(words[2], 0xFD4Au);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedShort4444) {
+    // P0: R=15 G=0 B=round(128*15/255)=8 A=round(64*15/255)=4 -> 0xF084
+    // P2: R=round(85*15/255)=5 G=round(170*15/255)=10 B=15 A=0 -> 0x5AF0
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4);
+    EXPECT_EQ(words[0], 0xF084u);
+    EXPECT_EQ(words[2], 0x5AF0u);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedShort4444Rev) {
+    // P0 fields R=15 G=0 B=8 A=4 packed from the LSB -> 15 | 0<<4 | 8<<8 | 4<<12 = 0x480F
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV);
+    EXPECT_EQ(words[0], 0x480Fu);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedShort5551) {
+    // P0: R=31 G=0 B=16 A=round(64/255)=0  -> 31<<11 | 16<<1        = 0xF820
+    // P1: R=1 G=round(250*31/255)=30 B=4 A=round(200/255)=1 -> 1<<11|30<<6|4<<1|1 = 0x0F89
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1);
+    EXPECT_EQ(words[0], 0xF820u);
+    EXPECT_EQ(words[1], 0x0F89u);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedShort1555Rev) {
+    // P1 fields R=1 G=30 B=4 A=1 packed from the LSB -> 1 | 30<<5 | 4<<10 | 1<<15 = 0x93C1
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV);
+    EXPECT_EQ(words[1], 0x93C1u);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedInt2101010Rev) {
+    // P0: R=1023 G=0 B=round(128*1023/255)=514 A=round(64*3/255)=1 -> 1023|514<<20|1<<30 = 0x602003FF
+    // P2: R=round(85*1023/255)=341 G=round(170*1023/255)=682 B=1023 A=0 -> 0x3FFAA955
+    const auto words = ConvertRGBA8Row<Uint32>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV);
+    EXPECT_EQ(words[0], 0x602003FFu);
+    EXPECT_EQ(words[2], 0x3FFAA955u);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedInt1010102) {
+    // P2 fields R=341 G=682 B=1023 A=0 packed from the MSB -> 341<<22 | 682<<12 | 1023<<2 = 0x556AAFFC
+    const auto words = ConvertRGBA8Row<Uint32>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_INT_10_10_10_2);
+    EXPECT_EQ(words[2], 0x556AAFFCu);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedByte332) {
+    // P0: R=7 G=0 B=round(128*3/255)=2                        -> 7<<5 | 2   = 0xE2
+    // P1: R=round(10*7/255)=0 G=round(250*7/255)=7 B=round(33*3/255)=0 -> 7<<2 = 0x1C
+    const auto words = ConvertRGBA8Row<Uint8>(kRGBA8Row, GL_RGB, GL_UNSIGNED_BYTE_3_3_2);
+    EXPECT_EQ(words[0], 0xE2u);
+    EXPECT_EQ(words[1], 0x1Cu);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesUnsignedByte233Rev) {
+    // P0 fields R=7 G=0 B=2 packed from the LSB -> 7 | 0<<3 | 2<<6 = 0x87
+    const auto words = ConvertRGBA8Row<Uint8>(kRGBA8Row, GL_RGB, GL_UNSIGNED_BYTE_2_3_3_REV);
+    EXPECT_EQ(words[0], 0x87u);
+}
+
+TEST(PackedReadbackEncodeTest, Encodes8888KeepsLegacyByteOrder) {
+    // Regression for the previously supported types: P0 = (255, 0, 128, 64).
+    const auto msbFirst = ConvertRGBA8Row<Uint32>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8);
+    EXPECT_EQ(msbFirst[0], 0xFF008040u);
+    const auto lsbFirst = ConvertRGBA8Row<Uint32>(kRGBA8Row, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV);
+    EXPECT_EQ(lsbFirst[0], 0x408000FFu);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesBGRAWithChannelMapping) {
+    // BGRA's first format component is Blue: P0 fields B=8 G=0 R=15 A=4 -> 8<<12 | 15<<4 | 4 = 0x80F4
+    const auto words = ConvertRGBA8Row<Uint16>(kRGBA8Row, GL_BGRA, GL_UNSIGNED_SHORT_4_4_4_4);
+    EXPECT_EQ(words[0], 0x80F4u);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesIntegerRGBA2101010RevWithFieldClamp) {
+    // Integer sources clamp to each field's unsigned range (10/10/10/2 bits).
+    const Vector<Uint32> wide{1023u, 1024u, 5u, 4u};
+    const auto words =
+        ConvertWideRowToPackedWords<Uint32>(wide, GL_UNSIGNED_INT, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV);
+    EXPECT_EQ(words[0], 0xC05FFFFFu); // 1023 | 1023<<10 | 5<<20 | 3<<30
+}
+
+TEST(PackedReadbackEncodeTest, EncodesIntegerNegativeValuesClampToZero) {
+    const Vector<Int32> wide{-5, 2, 100000, 1};
+    const auto words =
+        ConvertWideRowToPackedWords<Uint32>(wide, GL_INT, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV);
+    EXPECT_EQ(words[0], 0x7FF00800u); // 0 | 2<<10 | 1023<<20 | 1<<30
+}
+
+TEST(PackedReadbackEncodeTest, EncodesIntegerRGB565) {
+    const Vector<Uint32> wide{31u, 64u, 2u, 0u};
+    const auto words =
+        ConvertWideRowToPackedWords<Uint16>(wide, GL_UNSIGNED_INT, GL_RGB_INTEGER, GL_UNSIGNED_SHORT_5_6_5);
+    EXPECT_EQ(words[0], 0xFFE2u); // 31<<11 | 63<<5 | 2 (G clamps 64 -> 63)
+}
+
+TEST(PackedReadbackEncodeTest, EncodesPackedFloat10F11F11FRev) {
+    // F11(1.0)=0x3C0 F11(0.5)=0x380 F10(0.25)=0x1A0 -> 0x3C0 | 0x380<<11 | 0x1A0<<22 = 0x681C03C0.
+    // Second pixel: values above 65024 clamp to the max finite F11 (0x7BF), negatives go to zero.
+    const Vector<Float> wide{1.0f, 0.5f, 0.25f, 1.0f, 100000.0f, -1.0f, 0.25f, 1.0f};
+    const auto words = ConvertWideRowToPackedWords<Uint32>(wide, GL_FLOAT, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV);
+    EXPECT_EQ(words[0], 0x681C03C0u);
+    EXPECT_EQ(words[1], 0x680007BFu);
+}
+
+TEST(PackedReadbackEncodeTest, EncodesSharedExponent5999Rev) {
+    // (1.0, 0.5, 0.25): shared exponent 16, fields 256/128/64 -> 256 | 128<<9 | 64<<18 | 16<<27
+    const Vector<Float> wide{1.0f, 0.5f, 0.25f, 1.0f};
+    const auto words = ConvertWideRowToPackedWords<Uint32>(wide, GL_FLOAT, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV);
+    EXPECT_EQ(words[0], 0x81010100u);
+}
+
+TEST(PackedReadbackEncodeTest, RejectsMismatchedPackedFieldCounts) {
+    ReadbackImpl::ReadbackChannelMapping rgba{};
+    ASSERT_TRUE(ReadbackImpl::GetReadbackChannelMapping(GL_RGBA, rgba));
+    ReadbackImpl::ReadbackChannelMapping rgbInteger{};
+    ASSERT_TRUE(ReadbackImpl::GetReadbackChannelMapping(GL_RGB_INTEGER, rgbInteger));
+
+    // 3-field packed types never pair with 4-component formats and vice versa.
+    EXPECT_EQ(ReadbackImpl::GetReadbackDstPixelSize(rgba, GL_UNSIGNED_SHORT_5_6_5), 0u);
+    EXPECT_EQ(ReadbackImpl::GetReadbackDstPixelSize(rgbInteger, GL_UNSIGNED_SHORT_4_4_4_4), 0u);
+    // Packed-float RGB types never pair with integer formats.
+    EXPECT_EQ(ReadbackImpl::GetReadbackDstPixelSize(rgbInteger, GL_UNSIGNED_INT_5_9_9_9_REV), 0u);
+    EXPECT_EQ(ReadbackImpl::GetReadbackDstPixelSize(rgbInteger, GL_UNSIGNED_INT_10F_11F_11F_REV), 0u);
+}
+
+// ---- GL CTS packed_pixels readback root-cause regressions --------------------------------------
+
+TEST_F(FramebufferTest, ReadPixelsRejectsIntegerFormatMismatchWithReadBuffer) {
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 1, GL_RGBA8UI, 4, 4);
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, texture, 0);
+    MG_Impl::GLImpl::BindFramebuffer(GL_READ_FRAMEBUFFER, framebuffer);
+
+    MG_Backend::gBackendFunctionsTable.GL.ReadPixels = RecordReadPixels;
+    Uint8 pixelStorage[4 * 4 * 4] = {};
+
+    // GL 3.3 section 4.3.1: normalized format on an integer read buffer -> GL_INVALID_OPERATION
+    // (GL CTS packed_pixels expects the error for every mismatched combination).
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGBA, GL_UNSIGNED_BYTE, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+
+    // The matching integer readback stays valid.
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGBA_INTEGER, GL_UNSIGNED_INT, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // And the inverse mismatch: integer format on a normalized attachment.
+    GLuint normalizedTexture = 0;
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &normalizedTexture);
+    MG_Impl::GLImpl::TextureStorage2D(normalizedTexture, 1, GL_RGBA8, 4, 4);
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, normalizedTexture, 0);
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGBA_INTEGER, GL_UNSIGNED_INT, pixelStorage);
+    EXPECT_EQ(g_readPixelsCallCount, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION);
+}
+
+TEST_F(FramebufferTest, BindRenderbufferZeroUnbindsWithoutError) {
+    // The GL CTS state reset calls glBindRenderbuffer(GL_RENDERBUFFER, 0) and expects no error;
+    // name 0 used to be reported as an invalid renderbuffer name.
+    MG_Impl::GLImpl::BindRenderbuffer(GL_RENDERBUFFER, 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(FramebufferTest, FramebufferTexture3DAttachesSliceWithLayerTracking) {
+    // glFramebufferTexture3D with zoffset used to be rejected outright, leaving a sticky
+    // GL_INVALID_OPERATION behind (GL CTS packed_pixels varied_rectangle runs on GL_TEXTURE_3D).
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_3D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage3D(texture, 1, GL_RGBA8, 4, 4, 2);
+
+    MG_Impl::GLImpl::BindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
+    MG_Impl::GLImpl::FramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_3D, texture, 0, 1);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const auto framebufferObject = MG_State::pGLContext->GetFramebufferObject(framebuffer);
+    ASSERT_NE(framebufferObject, nullptr);
+    const auto& attachment = framebufferObject->GetAttachment(FramebufferAttachmentType::Color0);
+    ASSERT_TRUE(attachment.IsTexture());
+    EXPECT_EQ(attachment.GetTextureLayer(), 1);
+    EXPECT_FALSE(attachment.IsLayered());
+}
+
+TEST_F(FramebufferTest, NonRenderableColorFormatsReportUnsupportedFramebuffer) {
+    // Without a probing backend the conservative list applies: RGB9_E5 is texture-only, so
+    // attaching it must not report GL_FRAMEBUFFER_COMPLETE (GL CTS packed_pixels rgb9_e5 expects
+    // read errors instead of silent unwritten readbacks).
+    GLuint framebuffer = 0;
+    GLuint texture = 0;
+    MG_Impl::GLImpl::CreateFramebuffers(1, &framebuffer);
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_2D, 1, &texture);
+    MG_Impl::GLImpl::TextureStorage2D(texture, 1, GL_RGB9_E5, 4, 4);
+    MG_Impl::GLImpl::NamedFramebufferTexture(framebuffer, GL_COLOR_ATTACHMENT0, texture, 0);
+    MG_Impl::GLImpl::BindFramebuffer(GL_READ_FRAMEBUFFER, framebuffer);
+
+    EXPECT_EQ(MG_Impl::GLImpl::CheckFramebufferStatus(GL_READ_FRAMEBUFFER),
+              static_cast<GLenum>(GL_FRAMEBUFFER_UNSUPPORTED));
+
+    Uint8 pixelStorage[4 * 4 * 4] = {};
+    MG_Impl::GLImpl::ReadPixels(0, 0, 4, 4, GL_RGBA, GL_UNSIGNED_BYTE, pixelStorage);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_FRAMEBUFFER_OPERATION);
+}

MobileGL/MG_Test/Program/CMakeLists.txt

@@ -39,5 +39,5 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(ProgramUtilTest DISCOVERY_TIMEOUT 30)
-gtest_discover_tests(ProgramTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(ProgramUtilTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)
+gtest_discover_tests(ProgramTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/Program/ProgramTest.cpp

@@ -1845,3 +1845,440 @@ TEST_F(ProgramTest, GetActiveUniformsivErrors) {
     EXPECT_EQ(GetError(), GL_NO_ERROR);
     EXPECT_EQ(params[0], -999);
 }
+
+namespace {
+    GLuint LinkVsFsProgram(const char* vsSource, const char* fsSource) {
+        char infoLog[4096] = "";
+        GLuint vs = CreateShader(GL_VERTEX_SHADER);
+        ShaderSource(vs, 1, &vsSource, nullptr);
+        CompileShader(vs);
+        GLint vsStatus = GL_FALSE;
+        GetShaderiv(vs, GL_COMPILE_STATUS, &vsStatus);
+        GetShaderInfoLog(vs, sizeof(infoLog), nullptr, infoLog);
+        EXPECT_EQ(vsStatus, GL_TRUE) << infoLog;
+
+        GLuint fs = CreateShader(GL_FRAGMENT_SHADER);
+        ShaderSource(fs, 1, &fsSource, nullptr);
+        CompileShader(fs);
+        GLint fsStatus = GL_FALSE;
+        GetShaderiv(fs, GL_COMPILE_STATUS, &fsStatus);
+        GetShaderInfoLog(fs, sizeof(infoLog), nullptr, infoLog);
+        EXPECT_EQ(fsStatus, GL_TRUE) << infoLog;
+
+        GLuint program = CreateProgram();
+        AttachShader(program, vs);
+        AttachShader(program, fs);
+        LinkProgram(program);
+        GLint linkStatus = GL_FALSE;
+        GetProgramiv(program, GL_LINK_STATUS, &linkStatus);
+        GetProgramInfoLog(program, sizeof(infoLog), nullptr, infoLog);
+        EXPECT_EQ(linkStatus, GL_TRUE) << infoLog;
+        return program;
+    }
+
+    const char* kPassthroughCoordsVs = R"(#version 330
+in vec4 a_position;
+in vec4 a_coords;
+out vec4 coords_in;
+void main() {
+    gl_Position = a_position;
+    coords_in = a_coords;
+})";
+} // namespace
+
+// Repro for KHR-GL33.shaders.loops.do_while_dynamic_iterations.empty_body_* (and the
+// only_continue / unconditional_break variants): the loop is dead code, so the SPIR-V
+// optimizer eliminates it together with the only loads of `one` / `ui_one` -- and with
+// them the entire global UBO. The uniforms stay active in link reflection, so
+// glUniform1i on them must still have backing storage instead of memcpy-ing to null.
+TEST_F(ProgramTest, DoWhileDeadLoopUniformsKeepBackingStorage) {
+    const char* loopBodies[] = {"", "continue;", "break;"};
+    for (const char* body : loopBodies) {
+        const String fsSource = String(R"(#version 330
+uniform int ui_one;
+uniform mediump int one;
+in vec4 coords_in;
+out vec4 o_color;
+void main() {
+    vec4 res = coords_in;
+    mediump int i = 0;
+    do {)") + body + R"(} while (i++ < one*ui_one);
+    o_color = res;
+})";
+        GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource.c_str());
+
+        const GLint locOne = GetUniformLocation(program, "one");
+        const GLint locUiOne = GetUniformLocation(program, "ui_one");
+        ASSERT_GE(locOne, 0) << "body: '" << body << "'";
+        ASSERT_GE(locUiOne, 0) << "body: '" << body << "'";
+
+        UseProgram(program);
+        Uniform1i(locOne, 1);   // crashed with a null MapUBO() before the fallback storage
+        Uniform1i(locUiOne, 2);
+        EXPECT_EQ(GetError(), GL_NO_ERROR) << "body: '" << body << "'";
+
+        GLint readback = -1;
+        GetUniformiv(program, locOne, &readback);
+        EXPECT_EQ(readback, 1) << "body: '" << body << "'";
+        readback = -1;
+        GetUniformiv(program, locUiOne, &readback);
+        EXPECT_EQ(readback, 2) << "body: '" << body << "'";
+        EXPECT_EQ(GetError(), GL_NO_ERROR) << "body: '" << body << "'";
+    }
+}
+
+// Repro for KHR-GL33.shaders.struct.uniform.*nested_struct_array_*: leaf uniforms of
+// nested struct arrays need (a) one location per array element and (b) real byte
+// offsets inside the global UBO. Before the fix every leaf had a single location and
+// offset 0, so glUniform2fv(loc, 2, ...) tripped the size assert on the neighboring
+// float uniform (and corrupted it in release builds).
+TEST_F(ProgramTest, NestedStructArrayUniformElementWrites) {
+    // Struct shape from CTS glcShaderStructTests nested_struct_array (uniform case).
+    const char* fsSource = R"(#version 330
+struct T {
+    mediump float   a;
+    mediump vec2    b[2];
+};
+struct S {
+    mediump float   a;
+    T               b[3];
+    int             c;
+};
+uniform S s[2];
+in vec4 coords_in;
+out vec4 o_color;
+void main() {
+    mediump float r = (s[0].b[1].b[0].x + s[1].b[2].b[1].y) * s[0].b[0].a;
+    mediump float g = s[1].b[0].b[0].y * s[0].b[2].a * s[1].b[2].a;
+    mediump float b = (s[0].b[2].b[1].y + s[0].b[1].b[0].y + s[1].a) * s[0].b[1].a;
+    mediump float a = float(s[0].c) + s[1].b[2].a - s[1].b[1].a;
+    o_color = vec4(r, g, b, a);
+})";
+    GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource);
+    UseProgram(program);
+
+    const GLint locVecArray = GetUniformLocation(program, "s[0].b[1].b");
+    ASSERT_GE(locVecArray, 0);
+    // Element locations are consecutive and reachable via the "[k]" suffix.
+    EXPECT_EQ(GetUniformLocation(program, "s[0].b[1].b[0]"), locVecArray);
+    EXPECT_EQ(GetUniformLocation(program, "s[0].b[1].b[1]"), locVecArray + 1);
+    EXPECT_EQ(GetUniformLocation(program, "s[0].b[1].b[2]"), -1);
+
+    // Distinct scalar leaves must land at distinct UBO offsets (they all aliased
+    // offset 0 before the fix).
+    const char* scalarLeaves[] = {"s[0].b[0].a", "s[0].b[1].a", "s[0].b[2].a", "s[1].a", "s[1].b[1].a",
+                                  "s[1].b[2].a"};
+    const GLfloat scalarValues[] = {0.5f, 0.25f, 0.125f, 7.0f, 3.0f, 4.0f};
+    for (SizeT i = 0; i < std::size(scalarLeaves); ++i) {
+        const GLint loc = GetUniformLocation(program, scalarLeaves[i]);
+        ASSERT_GE(loc, 0) << scalarLeaves[i];
+        Uniform1f(loc, scalarValues[i]);
+    }
+
+    // CTS-style whole-array write: glUniform2fv with count = 2 on a vec2[2] leaf.
+    // Before the fix this asserted/corrupted the next uniform ("s[0].b[2].a").
+    const GLfloat vecData[4] = {1.0f, 2.0f, 3.0f, 4.0f};
+    Uniform2fv(locVecArray, 2, vecData);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+
+    GLfloat vecReadback[2] = {};
+    GetUniformfv(program, locVecArray, vecReadback);
+    EXPECT_EQ(vecReadback[0], 1.0f);
+    EXPECT_EQ(vecReadback[1], 2.0f);
+    GetUniformfv(program, locVecArray + 1, vecReadback);
+    EXPECT_EQ(vecReadback[0], 3.0f);
+    EXPECT_EQ(vecReadback[1], 4.0f);
+
+    // All scalar leaves survived the array write intact.
+    for (SizeT i = 0; i < std::size(scalarLeaves); ++i) {
+        GLfloat readback = -1.0f;
+        GetUniformfv(program, GetUniformLocation(program, scalarLeaves[i]), &readback);
+        EXPECT_EQ(readback, scalarValues[i]) << scalarLeaves[i];
+    }
+
+    // std140: vec2 array elements inside the struct are 16 bytes apart, and the
+    // per-element offsets differ.
+    auto programObject = MG_State::pGLContext->GetProgramObject(program);
+    ASSERT_NE(programObject, nullptr);
+    const Uint offsetElement0 = programObject->GetUniformOffset(static_cast<Uint>(locVecArray));
+    const Uint offsetElement1 = programObject->GetUniformOffset(static_cast<Uint>(locVecArray + 1));
+    EXPECT_EQ(offsetElement1, offsetElement0 + 16u);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}
+
+// Plain top-level uniform arrays share the same per-element location machinery.
+TEST_F(ProgramTest, PlainArrayUniformElementLocationsAndWrites) {
+    const char* fsSource = R"(#version 330
+uniform float arr[4];
+uniform float guard;
+in vec4 coords_in;
+out vec4 o_color;
+void main() {
+    o_color = vec4(arr[0] + arr[1], arr[2] + arr[3], guard, 1.0);
+})";
+    GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource);
+    UseProgram(program);
+
+    const GLint locArr = GetUniformLocation(program, "arr");
+    ASSERT_GE(locArr, 0);
+    EXPECT_EQ(GetUniformLocation(program, "arr[0]"), locArr);
+    EXPECT_EQ(GetUniformLocation(program, "arr[2]"), locArr + 2);
+    EXPECT_EQ(GetUniformLocation(program, "arr[4]"), -1);
+
+    const GLint locGuard = GetUniformLocation(program, "guard");
+    ASSERT_GE(locGuard, 0);
+    EXPECT_EQ(GetUniformLocation(program, "guard[0]"), -1); // not an array
+
+    Uniform1f(locGuard, 9.0f);
+
+    const GLfloat values[4] = {1.0f, 2.0f, 3.0f, 4.0f};
+    Uniform1fv(locArr, 4, values);
+    for (int i = 0; i < 4; ++i) {
+        GLfloat readback = -1.0f;
+        GetUniformfv(program, locArr + i, &readback);
+        EXPECT_EQ(readback, values[i]) << "arr[" << i << "]";
+    }
+
+    // Overlong writes stop at the end of the array (GL 3.3 §2.11.4) instead of
+    // spilling into the next uniform.
+    const GLfloat tail[3] = {30.0f, 40.0f, 50.0f};
+    Uniform1fv(GetUniformLocation(program, "arr[2]"), 3, tail);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+    GLfloat readback = -1.0f;
+    GetUniformfv(program, locArr + 2, &readback);
+    EXPECT_EQ(readback, 30.0f);
+    GetUniformfv(program, locArr + 3, &readback);
+    EXPECT_EQ(readback, 40.0f);
+    GetUniformfv(program, locGuard, &readback);
+    EXPECT_EQ(readback, 9.0f); // untouched by the overlong write
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}
+
+// ---------------------------------------------------------------------------
+// GL CTS KHR-GL33.shaders.uniform_block regression pack. MobileGL's SPIR-V
+// pipeline lays every uniform block out as std140; the frontend implements the
+// GL-visible consequences of that choice: packed/shared qualifiers compile (as
+// std140), reflection uses GL naming ("arr[0]", per-element struct arrays),
+// unused block members stay active, block sizes are vec4-padded, and array
+// strides are std140 even for arrays nested inside struct members.
+// ---------------------------------------------------------------------------
+
+TEST_F(ProgramTest, UniformBlockPackedAndSharedLayoutsCompileAsStd140) {
+    const char* fsSource = R"(#version 330
+layout(packed) uniform PackedBlock {
+    vec4 pv;
+};
+layout(shared, row_major) uniform SharedBlock {
+    float sf;
+    mat4 sm;
+};
+out vec4 o_color;
+void main() {
+    o_color = pv + vec4(sf) + vec4(sm[0][0]);
+})";
+    GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource);
+
+    // The blocks land on the implementation's chosen layout: std140 offsets.
+    const GLuint pv = UniformIndexByName(program, "pv");
+    const GLuint sf = UniformIndexByName(program, "sf");
+    const GLuint sm = UniformIndexByName(program, "sm");
+    ASSERT_NE(pv, GL_INVALID_INDEX);
+    ASSERT_NE(sf, GL_INVALID_INDEX);
+    ASSERT_NE(sm, GL_INVALID_INDEX);
+    EXPECT_EQ(QueryUniformiv(program, pv, GL_UNIFORM_OFFSET), 0);
+    EXPECT_EQ(QueryUniformiv(program, sf, GL_UNIFORM_OFFSET), 0);
+    EXPECT_EQ(QueryUniformiv(program, sm, GL_UNIFORM_OFFSET), 16);
+    // The remaining qualifiers in the rewritten layout() list survive.
+    EXPECT_EQ(QueryUniformiv(program, sm, GL_UNIFORM_IS_ROW_MAJOR), 1);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}
+
+TEST_F(ProgramTest, UniformBlockReflectsUnusedMembersWithGLNamesAndPaddedSize) {
+    const char* fsSource = R"(#version 330
+layout(std140) uniform Blk {
+    float used;
+    vec4 unusedArr[3];
+    ivec3 tail;
+};
+out vec4 o_color;
+void main() {
+    o_color = vec4(used);
+})";
+    GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource);
+
+    const GLuint blockIndex = GetUniformBlockIndex(program, "Blk");
+    ASSERT_NE(blockIndex, GL_INVALID_INDEX);
+
+    // All three members are active (unusedArr and tail are never read), the array is
+    // reported under its GL name "unusedArr[0]", and both spellings resolve.
+    const GLuint used = UniformIndexByName(program, "used");
+    const GLuint unusedSuffixed = UniformIndexByName(program, "unusedArr[0]");
+    const GLuint unusedBare = UniformIndexByName(program, "unusedArr");
+    const GLuint tail = UniformIndexByName(program, "tail");
+    ASSERT_NE(used, GL_INVALID_INDEX);
+    ASSERT_NE(unusedSuffixed, GL_INVALID_INDEX);
+    ASSERT_NE(tail, GL_INVALID_INDEX);
+    EXPECT_EQ(unusedSuffixed, unusedBare);
+
+    char nameBuf[64] = "";
+    GLsizei nameLen = 0;
+    GLint arraySize = 0;
+    GLenum type = 0;
+    GetActiveUniform(program, unusedSuffixed, sizeof(nameBuf), &nameLen, &arraySize, &type, nameBuf);
+    EXPECT_STREQ(nameBuf, "unusedArr[0]");
+    EXPECT_EQ(arraySize, 3);
+    EXPECT_EQ(type, static_cast<GLenum>(GL_FLOAT_VEC4));
+
+    // std140 layout of the unused members.
+    EXPECT_EQ(QueryUniformiv(program, unusedSuffixed, GL_UNIFORM_OFFSET), 16);
+    EXPECT_EQ(QueryUniformiv(program, unusedSuffixed, GL_UNIFORM_ARRAY_STRIDE), 16);
+    EXPECT_EQ(QueryUniformiv(program, tail, GL_UNIFORM_OFFSET), 64);
+
+    // GL_UNIFORM_BLOCK_ACTIVE_UNIFORMS agrees with the INDICES list and counts all members.
+    GLint activeInBlock = 0;
+    GetActiveUniformBlockiv(program, blockIndex, GL_UNIFORM_BLOCK_ACTIVE_UNIFORMS, &activeInBlock);
+    ASSERT_EQ(activeInBlock, 3);
+    GLint indices[3] = {-1, -1, -1};
+    GetActiveUniformBlockiv(program, blockIndex, GL_UNIFORM_BLOCK_ACTIVE_UNIFORM_INDICES, indices);
+    for (GLint index : indices) {
+        EXPECT_TRUE(index == static_cast<GLint>(used) || index == static_cast<GLint>(unusedSuffixed) ||
+                    index == static_cast<GLint>(tail));
+    }
+
+    // The block ends with an ivec3 at offset 64 (unpadded end 76); the backend compiles
+    // the std140 block at its vec4-padded size, and the reported size must cover it or
+    // buffers sized from this query are too small to draw with.
+    GLint dataSize = 0;
+    GetActiveUniformBlockiv(program, blockIndex, GL_UNIFORM_BLOCK_DATA_SIZE, &dataSize);
+    EXPECT_EQ(dataSize, 80);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}
+
+TEST_F(ProgramTest, UniformBlockStructArrayExpandsPerElementWithStd140Strides) {
+    const char* fsSource = R"(#version 330
+struct S {
+    ivec2 v[2];
+    float f;
+};
+layout(std140) uniform Blk2 {
+    S s[2];
+} inst;
+out vec4 o_color;
+void main() {
+    o_color = vec4(inst.s[0].f);
+})";
+    GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource);
+
+    // ARB_program_interface_query naming: one entry per struct array element, prefixed
+    // with the BLOCK name (not the instance name), basic arrays suffixed with "[0]".
+    const GLuint v0 = UniformIndexByName(program, "Blk2.s[0].v[0]");
+    const GLuint f0 = UniformIndexByName(program, "Blk2.s[0].f");
+    const GLuint v1 = UniformIndexByName(program, "Blk2.s[1].v[0]");
+    const GLuint f1 = UniformIndexByName(program, "Blk2.s[1].f");
+    ASSERT_NE(v0, GL_INVALID_INDEX);
+    ASSERT_NE(f0, GL_INVALID_INDEX);
+    ASSERT_NE(v1, GL_INVALID_INDEX);
+    ASSERT_NE(f1, GL_INVALID_INDEX);
+
+    // std140: ivec2 v[2] rounds each element up to a vec4 (stride 16, NOT the tight 8
+    // glslang reflects for arrays nested inside a struct member); struct size rounds to
+    // 48, giving s[1] members a 48-byte bias.
+    EXPECT_EQ(QueryUniformiv(program, v0, GL_UNIFORM_OFFSET), 0);
+    EXPECT_EQ(QueryUniformiv(program, v0, GL_UNIFORM_ARRAY_STRIDE), 16);
+    EXPECT_EQ(QueryUniformiv(program, v0, GL_UNIFORM_SIZE), 2);
+    EXPECT_EQ(QueryUniformiv(program, f0, GL_UNIFORM_OFFSET), 32);
+    EXPECT_EQ(QueryUniformiv(program, v1, GL_UNIFORM_OFFSET), 48);
+    EXPECT_EQ(QueryUniformiv(program, f1, GL_UNIFORM_OFFSET), 80);
+
+    GLint dataSize = 0;
+    const GLuint blockIndex = GetUniformBlockIndex(program, "Blk2");
+    ASSERT_NE(blockIndex, GL_INVALID_INDEX);
+    GetActiveUniformBlockiv(program, blockIndex, GL_UNIFORM_BLOCK_DATA_SIZE, &dataSize);
+    EXPECT_EQ(dataSize, 96);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}
+
+TEST_F(ProgramTest, UniformBlockInstanceArrayReportsPerInstanceBlocks) {
+    const char* fsSource = R"(#version 330
+layout(std140) uniform ArrBlk {
+    vec4 av;
+} insts[2];
+out vec4 o_color;
+void main() {
+    o_color = insts[0].av + insts[1].av;
+})";
+    GLuint program = LinkVsFsProgram(kPassthroughCoordsVs, fsSource);
+
+    const GLuint inst0 = GetUniformBlockIndex(program, "ArrBlk[0]");
+    const GLuint inst1 = GetUniformBlockIndex(program, "ArrBlk[1]");
+    ASSERT_NE(inst0, GL_INVALID_INDEX);
+    ASSERT_NE(inst1, GL_INVALID_INDEX);
+    EXPECT_NE(inst0, inst1);
+    // A bare block name resolves to the first instance.
+    EXPECT_EQ(GetUniformBlockIndex(program, "ArrBlk"), inst0);
+
+    // Every instance of the array shares the single reflected member set.
+    GLint count0 = 0;
+    GLint count1 = 0;
+    GetActiveUniformBlockiv(program, inst0, GL_UNIFORM_BLOCK_ACTIVE_UNIFORMS, &count0);
+    GetActiveUniformBlockiv(program, inst1, GL_UNIFORM_BLOCK_ACTIVE_UNIFORMS, &count1);
+    EXPECT_EQ(count0, 1);
+    EXPECT_EQ(count1, 1);
+    GLint index0 = -1;
+    GLint index1 = -1;
+    GetActiveUniformBlockiv(program, inst0, GL_UNIFORM_BLOCK_ACTIVE_UNIFORM_INDICES, &index0);
+    GetActiveUniformBlockiv(program, inst1, GL_UNIFORM_BLOCK_ACTIVE_UNIFORM_INDICES, &index1);
+    EXPECT_EQ(index0, index1);
+    EXPECT_EQ(static_cast<GLuint>(index0), UniformIndexByName(program, "ArrBlk.av"));
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}
+
+TEST_F(ProgramTest, DeleteShaderWhileAttachedKeepsNameUsableUntilDetach) {
+    // GL CTS compiles through exactly this sequence (create, attach, DELETE, source,
+    // compile): glDeleteShader on an attached shader only flags it, and the name must
+    // keep working until the last detach.
+    const char* vsSource = R"(#version 330
+void main() { gl_Position = vec4(0.0); }
+)";
+    const char* fsSource = R"(#version 330
+out vec4 o_color;
+void main() { o_color = vec4(1.0); }
+)";
+
+    GLuint program = CreateProgram();
+    GLuint vs = CreateShader(GL_VERTEX_SHADER);
+    AttachShader(program, vs);
+    DeleteShader(vs);
+    EXPECT_EQ(IsShader(vs), GL_TRUE); // still alive: attached
+
+    ShaderSource(vs, 1, &vsSource, nullptr);
+    CompileShader(vs);
+    GLint status = GL_FALSE;
+    GetShaderiv(vs, GL_COMPILE_STATUS, &status);
+    EXPECT_EQ(status, GL_TRUE);
+    status = GL_FALSE;
+    GetShaderiv(vs, GL_DELETE_STATUS, &status);
+    EXPECT_EQ(status, GL_TRUE);
+
+    GLuint fs = CreateShader(GL_FRAGMENT_SHADER);
+    AttachShader(program, fs);
+    DeleteShader(fs);
+    ShaderSource(fs, 1, &fsSource, nullptr);
+    CompileShader(fs);
+
+    LinkProgram(program);
+    GLint linkStatus = GL_FALSE;
+    char infoLog[1024] = "";
+    GetProgramiv(program, GL_LINK_STATUS, &linkStatus);
+    GetProgramInfoLog(program, sizeof(infoLog), nullptr, infoLog);
+    EXPECT_EQ(linkStatus, GL_TRUE) << infoLog;
+
+    // The last GL-visible detach releases the flagged shader's name.
+    DetachShader(program, vs);
+    EXPECT_EQ(IsShader(vs), GL_FALSE);
+
+    // Deleting the program releases the other flagged shader.
+    DeleteProgram(program);
+    EXPECT_EQ(IsShader(fs), GL_FALSE);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+}

MobileGL/MG_Test/Program/ProgramUtilTest.cpp

@@ -107,7 +107,7 @@ void main() {
 
     PreprocessShaderSource(ShaderStage::Vertex, source);
 
-    EXPECT_EQ(source.find("#version 460 core\n"), 0);
+    EXPECT_EQ(source.find("#version 330 core\n"), 0);
     EXPECT_NE(source.find("in vec3 position;"), String::npos);
     EXPECT_NE(source.find("out vec2 uv;"), String::npos);
     EXPECT_EQ(source.find("attribute"), String::npos);
@@ -136,7 +136,7 @@ void main() {
 
     PreprocessShaderSource(ShaderStage::Fragment, source);
 
-    EXPECT_EQ(source.find("#version 460 core\n"), 0);
+    EXPECT_EQ(source.find("#version 330 core\n"), 0);
     EXPECT_NE(source.find("out vec4 mg_FragColor;\n"), String::npos);
     EXPECT_NE(source.find("in vec2 uv;"), String::npos);
     EXPECT_NE(source.find("texture(texture0, uv)"), String::npos);
@@ -153,6 +153,243 @@ void main() {
     }
 }
 
+TEST_F(ProgramUtilTest, PreprocessMinecraft112BlurShaderKeepsLegacySampleIdentifier) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    // assets/minecraft/shaders/program/blur.fsh from the unmodified Minecraft 1.12 client jar.
+    String source = R"(#version 120
+
+uniform sampler2D DiffuseSampler;
+
+varying vec2 texCoord;
+varying vec2 oneTexel;
+
+uniform vec2 InSize;
+
+uniform vec2 BlurDir;
+uniform float Radius;
+
+void main() {
+    vec4 blurred = vec4(0.0);
+    float totalStrength = 0.0;
+    float totalAlpha = 0.0;
+    float totalSamples = 0.0;
+    for(float r = -Radius; r <= Radius; r += 1.0) {
+        vec4 sample = texture2D(DiffuseSampler, texCoord + oneTexel * r * BlurDir);
+
+		// Accumulate average alpha
+        totalAlpha = totalAlpha + sample.a;
+        totalSamples = totalSamples + 1.0;
+
+		// Accumulate smoothed blur
+        float strength = 1.0 - abs(r / Radius);
+        totalStrength = totalStrength + strength;
+        blurred = blurred + sample;
+    }
+    gl_FragColor = vec4(blurred.rgb / (Radius * 2.0 + 1.0), totalAlpha);
+}
+)";
+
+    PreprocessShaderSource(ShaderStage::Fragment, source);
+
+    EXPECT_EQ(source.find("#version 330 core\n"), 0);
+    EXPECT_NE(source.find("vec4 sample = texture(DiffuseSampler"), String::npos);
+    EXPECT_NE(source.find("totalAlpha = totalAlpha + sample.a;"), String::npos);
+    EXPECT_NE(source.find("float totalSamples = 0.0;"), String::npos);
+    EXPECT_NE(source.find("totalSamples = totalSamples + 1.0;"), String::npos);
+    EXPECT_NE(source.find("blurred = blurred + sample;"), String::npos);
+
+    ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+    auto res = ShaderCompiler::CompileShader(attrib);
+    if (!res) {
+        FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+    }
+}
+
+TEST_F(ProgramUtilTest, PreprocessLegacySampleInterfaceIdentifiersKeepNames) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    String vertexSource = R"(#version 150
+attribute vec3 sample;
+
+void main() {
+    gl_Position = vec4(sample, 1.0);
+}
+)";
+    PreprocessShaderSource(ShaderStage::Vertex, vertexSource);
+
+    EXPECT_EQ(vertexSource.find("#version 330 core\n"), 0);
+    EXPECT_NE(vertexSource.find("in vec3 sample;"), String::npos);
+
+    ShaderAttrib vertexAttrib{.shaderType = GL_VERTEX_SHADER, .sourceStr = vertexSource};
+    auto vertexResult = ShaderCompiler::CompileShader(vertexAttrib);
+    if (!vertexResult) {
+        FAIL() << "errc: " << vertexResult.error().errc << "\nlog: " << vertexResult.error().log
+               << "\nsource:\n" << vertexSource;
+    }
+
+    String fragmentSource = R"(#version 150
+uniform sampler2D sample;
+varying vec2 texCoord;
+
+void main() {
+    gl_FragColor = texture2D(sample, texCoord);
+}
+)";
+    PreprocessShaderSource(ShaderStage::Fragment, fragmentSource);
+
+    EXPECT_EQ(fragmentSource.find("#version 330 core\n"), 0);
+    EXPECT_NE(fragmentSource.find("uniform sampler2D sample;"), String::npos);
+    EXPECT_NE(fragmentSource.find("texture(sample, texCoord)"), String::npos);
+
+    ShaderAttrib fragmentAttrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = fragmentSource};
+    auto fragmentResult = ShaderCompiler::CompileShader(fragmentAttrib);
+    if (!fragmentResult) {
+        FAIL() << "errc: " << fragmentResult.error().errc << "\nlog: " << fragmentResult.error().log
+               << "\nsource:\n" << fragmentSource;
+    }
+}
+
+TEST_F(ProgramUtilTest, PreprocessEsslVersionsRemainVulkanCompatible) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    const auto verifyVersion = [](const char* inputVersion, const char* expectedVersion) {
+        SCOPED_TRACE(inputVersion);
+        String source = inputVersion;
+        source += R"(
+precision mediump float;
+out vec4 fragColor;
+
+void main() {
+    fragColor = vec4(1.0);
+}
+)";
+        PreprocessShaderSource(ShaderStage::Fragment, source);
+
+        EXPECT_EQ(source.find(expectedVersion), 0);
+
+        ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+        auto res = ShaderCompiler::CompileShader(attrib);
+        if (!res) {
+            FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+        }
+    };
+
+    // Preserve the pre-existing desktop-core route: the current resource table cannot parse ESSL built-ins.
+    verifyVersion("#version 300 es", "#version 460 core\n");
+    verifyVersion("#version 310 es", "#version 460 core\n");
+}
+
+TEST_F(ProgramUtilTest, PreprocessModernDesktopVersionsRecognizesUtf8Bom) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    const auto verifyVersion = [](const char* inputVersion) {
+        SCOPED_TRACE(inputVersion);
+        String source = "\xef\xbb\xbf";
+        source += inputVersion;
+        source += R"(
+out vec4 fragColor;
+
+void main() {
+    fragColor = vec4(1.0);
+}
+)";
+        PreprocessShaderSource(ShaderStage::Fragment, source);
+
+        EXPECT_EQ(source.find("#version 460 core\n"), 0);
+        EXPECT_EQ(source.find("\xef\xbb\xbf"), String::npos);
+
+        ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+        auto res = ShaderCompiler::CompileShader(attrib);
+        if (!res) {
+            FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+        }
+    };
+
+    verifyVersion("#version 400 core");
+    verifyVersion("#version 460 core");
+}
+
+TEST_F(ProgramUtilTest, PreprocessUsesRealSpacedVersionDirectiveForInjectedOutput) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    String source = R"(// #version 460 core
+/* "#version 400 core" */
+#line 7 "#version 460 core"
+# version 120
+varying vec2 uv;
+
+void main() {
+    gl_FragColor = vec4(uv, 0.0, 1.0);
+}
+)";
+    PreprocessShaderSource(ShaderStage::Fragment, source);
+
+    const SizeT versionPos = source.find("#version 330 core\n");
+    const SizeT outputPos = source.find("out vec4 mg_FragColor;\n");
+    EXPECT_NE(versionPos, String::npos);
+    EXPECT_EQ(outputPos, versionPos + std::strlen("#version 330 core\n"));
+    EXPECT_NE(source.find("// #version 460 core"), String::npos);
+    EXPECT_EQ(source.find("#line"), String::npos);
+
+    ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+    auto res = ShaderCompiler::CompileShader(attrib);
+    if (!res) {
+        FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+    }
+}
+
+TEST_F(ProgramUtilTest, PreprocessModernSampleQualifierStaysAtVersion460) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    String source = R"(#version 400 core
+sample in vec4 interpolatedColor;
+out vec4 fragColor;
+
+void main() {
+    fragColor = interpolatedColor;
+}
+)";
+    PreprocessShaderSource(ShaderStage::Fragment, source);
+
+    EXPECT_EQ(source.find("#version 460 core\n"), 0);
+    EXPECT_NE(source.find("sample in vec4 interpolatedColor;"), String::npos);
+
+    ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+    auto res = ShaderCompiler::CompileShader(attrib);
+    if (!res) {
+        FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+    }
+}
+
+TEST_F(ProgramUtilTest, PreprocessGpuShader5SampleQualifierUsesVersion460) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    for (const char* extension : {"GL_ARB_gpu_shader5", "GL_NV_gpu_shader5"}) {
+        SCOPED_TRACE(extension);
+        String source = "#version 150\n#extension ";
+        source += extension;
+        source += R"( : enable
+sample in vec4 interpolatedColor;
+out vec4 fragColor;
+
+void main() {
+    fragColor = interpolatedColor;
+}
+)";
+        PreprocessShaderSource(ShaderStage::Fragment, source);
+
+        EXPECT_EQ(source.find("#version 460 core\n"), 0);
+        EXPECT_NE(source.find("sample in vec4 interpolatedColor;"), String::npos);
+
+        ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+        auto res = ShaderCompiler::CompileShader(attrib);
+        if (!res) {
+            FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+        }
+    }
+}
+
 TEST_F(ProgramUtilTest, PreprocessLegacyFragmentShaderModernizesFragData) {
     using namespace MG_Util::ShaderTranspiler;
 
@@ -164,7 +401,7 @@ void main() {
 
     PreprocessShaderSource(ShaderStage::Fragment, source);
 
-    EXPECT_EQ(source.find("#version 460 core\n"), 0);
+    EXPECT_EQ(source.find("#version 330 core\n"), 0);
     EXPECT_NE(source.find("layout(location = 0) out vec4 mg_FragData[8];\n"), String::npos);
     EXPECT_NE(source.find("mg_FragData[0] = vec4(1.0);"), String::npos);
     EXPECT_NE(source.find("mg_FragData[1].a = 0.5;"), String::npos);
@@ -182,7 +419,7 @@ TEST_F(ProgramUtilTest, PreprocessKeepsDefaultPrecisionStatements) {
 
     // Mirrors the GL CTS helper shaders (e.g. glcPixelStorageModesTests): the old qualifier strip
     // turned "precision highp float;" into invalid "precision  float;". Precision qualifiers are
-    // legal (and ignored) in the forced 460 core profile, so they now pass through untouched.
+    // legal (and ignored) in the normalized desktop core profile, so they now pass through untouched.
     String source = R"(#version 330
 precision highp float;
 precision mediump int;
@@ -212,7 +449,7 @@ TEST_F(ProgramUtilTest, PreprocessKeepsPrecisionInLegacyShaderForGlslang) {
     using namespace MG_Util::ShaderTranspiler;
 
     // Legacy ES-style shader: precision statements and qualifier macros are left for glslang
-    // (its preprocessor expands the #define; the 460 core parse ignores the qualifiers).
+    // (its preprocessor expands the #define; the normalized 330 core parse ignores the qualifiers).
     String source = R"(#define HIGHP_OR_DEFAULT highp
 precision HIGHP_OR_DEFAULT float;
 precision mediump int;
@@ -319,6 +556,81 @@ TEST_F(ProgramUtilTest, CompileSimpleVertexShader) {
     }
 }
 
+// Legacy desktop sources are normalized to "#version 330 core", which is stricter than the 460 they
+// used to be forced to. A shader declaring 330 while using 420-era syntax without the matching
+// #extension line is accepted by real drivers, so CompileShader retries it at 460 instead of failing.
+TEST_F(ProgramUtilTest, CompileShaderRetriesAt460WhenLegacyVersionRejects420Syntax) {
+    using namespace MG_Util::ShaderTranspiler;
+    String source = R"(#version 330
+layout(binding = 0) uniform sampler2D InSampler;
+in vec2 texCoord;
+out vec4 fragColor;
+void main() {
+    fragColor = texture(InSampler, texCoord);
+})";
+    PreprocessShaderSource(ShaderStage::Fragment, source);
+    // The normal path still emits 330 - the retry must not become the default.
+    ASSERT_EQ(source.find("#version 330 core"), 0u);
+
+    ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+    auto res = ShaderCompiler::CompileShader(attrib);
+    if (!res) {
+        FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log;
+    }
+
+    // Same source compiled for the OpenGL environment must take the retry too.
+    ShaderAttrib glAttrib{
+        .shaderType = GL_FRAGMENT_SHADER, .sourceStr = source, .flags = ShaderCompileBits::CompileForOpenGL};
+    auto glRes = ShaderCompiler::CompileShader(glAttrib);
+    if (!glRes) {
+        FAIL() << "errc: " << glRes.error().errc << "\nlog: " << glRes.error().log;
+    }
+}
+
+TEST_F(ProgramUtilTest, CompileShaderStillFailsWithOriginalDiagnosticsWhenRetryCannotHelp) {
+    using namespace MG_Util::ShaderTranspiler;
+    String source = R"(#version 330
+in vec2 texCoord;
+out vec4 fragColor;
+void main() {
+    fragColor = thisFunctionDoesNotExist(texCoord);
+})";
+    PreprocessShaderSource(ShaderStage::Fragment, source);
+
+    ShaderAttrib attrib{.shaderType = GL_FRAGMENT_SHADER, .sourceStr = source};
+    auto res = ShaderCompiler::CompileShader(attrib);
+    ASSERT_FALSE(res);
+    EXPECT_EQ(res.error().errc, -2);
+    EXPECT_NE(res.error().log.find("thisFunctionDoesNotExist"), String::npos) << res.error().log;
+}
+
+TEST_F(ProgramUtilTest, RetargetLegacyVersionDirectiveOnlyTouchesNormalizedDesktopCore) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    String normalized = "#version 330 core\nvoid main() {}\n";
+    EXPECT_TRUE(RetargetLegacyVersionDirectiveTo460(normalized));
+    EXPECT_EQ(normalized.find("#version 460 core"), 0u);
+
+    // Already modern: nothing to retarget.
+    String modern = "#version 460 core\nvoid main() {}\n";
+    EXPECT_FALSE(RetargetLegacyVersionDirectiveTo460(modern));
+    EXPECT_EQ(modern.find("#version 460 core"), 0u);
+
+    // ES and compatibility sources keep what they declared.
+    String es = "#version 300 es\nvoid main() {}\n";
+    EXPECT_FALSE(RetargetLegacyVersionDirectiveTo460(es));
+    EXPECT_EQ(es.find("#version 300 es"), 0u);
+
+    String compat = "#version 330 compatibility\nvoid main() {}\n";
+    EXPECT_FALSE(RetargetLegacyVersionDirectiveTo460(compat));
+    EXPECT_EQ(compat.find("#version 330 compatibility"), 0u);
+
+    // A commented-out directive is not the real one.
+    String commented = "// #version 330 core\nvoid main() {}\n";
+    EXPECT_FALSE(RetargetLegacyVersionDirectiveTo460(commented));
+    EXPECT_EQ(commented.find("#version 460"), String::npos);
+}
+
 const char* fs = R"(#version 150
 
 uniform sampler2D InSampler;
@@ -1017,3 +1329,57 @@ void main() {
     ASSERT_TRUE(ShaderCompiler::SanitizeAndOptimizeBinary(binRes->at(0), optimized));
 }
 
+
+TEST_F(ProgramUtilTest, PreprocessCoercesBlockPackingQualifiersToStd140) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    // glslang rejects `packed`/`shared` outright when generating SPIR-V, and MobileGL's
+    // UBO layout is always std140 anyway; the preprocessor rewrites the qualifiers so the
+    // validation compile, reflection, and generated SPIR-V all agree on std140 (GL CTS
+    // KHR-GL33.shaders.uniform_block.*.packed/shared).
+    String source = R"(#version 330
+layout(packed) uniform PackedBlock { vec4 pv; };
+layout(shared, row_major) uniform SharedBlock { mat4 sm; };
+layout ( shared ) uniform SpacedBlock { float sx; };
+layout(std140) uniform KeptBlock { float kx; };
+// A non-layout use of the identifier stays untouched (compute storage qualifier).
+void main() {
+    gl_Position = pv + vec4(sm[0][0]) + vec4(sx) + vec4(kx);
+})";
+
+    PreprocessShaderSource(ShaderStage::Vertex, source);
+
+    EXPECT_EQ(source.find("packed"), String::npos);
+    EXPECT_EQ(source.find("layout(shared"), String::npos);
+    EXPECT_NE(source.find("layout(std140) uniform PackedBlock"), String::npos);
+    EXPECT_NE(source.find("layout(std140, row_major) uniform SharedBlock"), String::npos);
+    EXPECT_NE(source.find("layout ( std140 ) uniform SpacedBlock"), String::npos);
+    EXPECT_NE(source.find("layout(std140) uniform KeptBlock"), String::npos);
+
+    ShaderAttrib attrib{.shaderType = GL_VERTEX_SHADER,
+                        .sourceStr = source,
+                        .flags = ShaderCompileBits::CompileForOpenGL};
+    auto res = ShaderCompiler::CompileShader(attrib);
+    if (!res) {
+        FAIL() << "errc: " << res.error().errc << "\nlog: " << res.error().log << "\nsource:\n" << source;
+    }
+}
+
+TEST_F(ProgramUtilTest, PreprocessLeavesComputeSharedStorageQualifierAlone) {
+    using namespace MG_Util::ShaderTranspiler;
+
+    // `shared` is only a packing qualifier inside layout(...); the compute-shader storage
+    // qualifier of the same spelling must survive.
+    String source = R"(#version 430
+layout(local_size_x = 8) in;
+shared float sharedScratch[8];
+layout(shared) uniform Blk { float bx; };
+void main() {
+    sharedScratch[gl_LocalInvocationIndex] = bx;
+})";
+
+    PreprocessShaderSource(ShaderStage::Compute, source);
+
+    EXPECT_NE(source.find("shared float sharedScratch[8];"), String::npos);
+    EXPECT_NE(source.find("layout(std140) uniform Blk"), String::npos);
+}

MobileGL/MG_Test/Query/CMakeLists.txt

@@ -17,4 +17,4 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(QueryTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(QueryTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/SanityTest.cpp

@@ -19,6 +19,7 @@
 #include <MG_Backend/BackendObjects.h>
 #include <MG_Impl/GLImpl/Getter/GL_Getter.h>
 #include <MG_Impl/GLImpl/RenderState/GL_RenderState.h>
+#include <MG_Impl/GLImpl/Texture/GL_Texture.h>
 #include <MG_Impl/GLImpl/VertexArray/Validators.h>
 #include <MG_State/GLState/Core.h>
 #include <MG_State/GLState/FramebufferState/FramebufferObject.h>
@@ -103,6 +104,79 @@ namespace {
     private:
         MobileGL::MG_External::GLESCapabilities m_snapshot;
     };
+
+    struct TextureBindCall {
+        GLenum target;
+        GLuint texture;
+    };
+
+    MobileGL::Vector<TextureBindCall>* g_textureBindCalls = nullptr;
+    GLuint g_nextBackendTextureId = 73;
+
+    void RecordTextureBind(GLenum target, GLuint texture) {
+        if (g_textureBindCalls) {
+            g_textureBindCalls->push_back({target, texture});
+        }
+    }
+
+    void GenerateBackendTextures(GLsizei count, GLuint* textures) {
+        for (GLsizei i = 0; i < count; ++i) {
+            textures[i] = g_nextBackendTextureId++;
+        }
+    }
+
+    void DeleteBackendTextures(GLsizei, const GLuint*) {}
+
+    GLenum NoBackendError() {
+        return GL_NO_ERROR;
+    }
+
+    // Isolates the DirectGLES globals touched by the binding-cache regression test. The test
+    // installs only the native ES entry points needed to construct/bind a backend texture and
+    // restores the process-wide state even when a gtest assertion unwinds the test body.
+    struct ScopedDirectGLESTextureBindings {
+        ScopedDirectGLESTextureBindings():
+            previousContext(MobileGL::Move(MobileGL::MG_State::pGLContext)),
+            previousFunctions(MobileGL::MG_Backend::DirectGLES::g_GLESFuncs),
+            previousActiveUnit(MobileGL::MG_Backend::DirectGLES::TextureImpl::g_activeTextureUnit),
+            previousCache(MobileGL::MG_Backend::DirectGLES::TextureImpl::g_boundTexturesCache),
+            previousRegistry(MobileGL::MG_Backend::DirectGLES::TextureImpl::g_backendTextureObjects) {
+            MobileGL::MG_State::pGLContext = MobileGL::MakeUnique<MobileGL::MG_State::GLState::GLContext>();
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_activeTextureUnit = 0;
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_boundTexturesCache = {};
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_backendTextureObjects = {};
+
+            MobileGL::MG_External::GLESFunctionsTable functions{};
+            functions.glBindTexture = RecordTextureBind;
+            functions.glDeleteTextures = DeleteBackendTextures;
+            functions.glGenTextures = GenerateBackendTextures;
+            functions.glGetError = NoBackendError;
+            MobileGL::MG_Backend::DirectGLES::SetGLESFuncsTable(functions);
+            g_textureBindCalls = &bindCalls;
+        }
+
+        ~ScopedDirectGLESTextureBindings() {
+            g_textureBindCalls = nullptr;
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_boundTexturesCache = {};
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_backendTextureObjects = previousRegistry;
+            MobileGL::MG_Backend::DirectGLES::SetGLESFuncsTable(previousFunctions);
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_activeTextureUnit = previousActiveUnit;
+            MobileGL::MG_Backend::DirectGLES::TextureImpl::g_boundTexturesCache = previousCache;
+            MobileGL::MG_State::pGLContext = MobileGL::Move(previousContext);
+        }
+
+        ScopedDirectGLESTextureBindings(const ScopedDirectGLESTextureBindings&) = delete;
+        ScopedDirectGLESTextureBindings& operator=(const ScopedDirectGLESTextureBindings&) = delete;
+
+        MobileGL::Vector<TextureBindCall> bindCalls;
+
+    private:
+        MobileGL::UniquePtr<MobileGL::MG_State::GLState::GLContext> previousContext;
+        MobileGL::MG_External::GLESFunctionsTable previousFunctions;
+        MobileGL::Uint previousActiveUnit;
+        decltype(MobileGL::MG_Backend::DirectGLES::TextureImpl::g_boundTexturesCache) previousCache;
+        decltype(MobileGL::MG_Backend::DirectGLES::TextureImpl::g_backendTextureObjects) previousRegistry;
+    };
 } // namespace
 
 TEST(Sanity, BasicAssertions) {
@@ -148,6 +222,52 @@ TEST(DirectGLESSanity, AdvertisesVoxyRequiredRenderingExtensionsWithoutRaisingGL
               extensions.end());
 }
 
+TEST(DirectGLESSanity, BindingZeroClearsPreviousNativeTextureBinding) {
+    using namespace MobileGL;
+    namespace DirectGLES = MG_Backend::DirectGLES;
+
+    ScopedDirectGLESTextureBindings state;
+
+    GLuint frontendTexture = 0;
+    MG_Impl::GLImpl::GenTextures(1, &frontendTexture);
+    ASSERT_NE(frontendTexture, 0u);
+    MG_Impl::GLImpl::BindTexture(GL_TEXTURE_2D, frontendTexture);
+    const auto& frontendTextureObject = MG_State::pGLContext->GetTextureUnitObject(0)
+                                            .GetBindingSlot(TextureTarget::Texture2D)
+                                            .GetBoundObject();
+    ASSERT_NE(frontendTextureObject, nullptr);
+    ASSERT_EQ(frontendTextureObject->GetExternalIndex(), frontendTexture);
+
+    auto& backendTexture = DirectGLES::TextureImpl::g_backendTextureObjects.GetOrCreate(frontendTextureObject);
+    backendTexture = MakeShared<DirectGLES::TextureImpl::BackendTextureObject>();
+    const GLuint backendTextureId = backendTexture->GetBackendTextureId();
+
+    DirectGLES::BindCurrentTextures();
+    ASSERT_EQ(state.bindCalls.size(), 1u);
+    EXPECT_EQ(state.bindCalls[0].target, GL_TEXTURE_2D);
+    EXPECT_EQ(state.bindCalls[0].texture, backendTextureId);
+
+    // The default 1D slot maps to the same native ES target as 2D. It must not clear and force a
+    // redundant rebind while the real 2D frontend object remains current.
+    DirectGLES::BindCurrentTextures();
+    EXPECT_EQ(state.bindCalls.size(), 1u);
+
+    MG_Impl::GLImpl::BindTexture(GL_TEXTURE_2D, 0);
+    ASSERT_TRUE(MG_State::GLState::IsUndefinedDefaultTexture(
+        MG_State::pGLContext->GetTextureUnitObject(0)
+            .GetBindingSlot(TextureTarget::Texture2D)
+            .GetBoundObject()
+            .get()));
+
+    DirectGLES::BindCurrentTextures();
+
+    ASSERT_EQ(state.bindCalls.size(), 2u);
+    EXPECT_EQ(state.bindCalls[1].target, GL_TEXTURE_2D);
+    EXPECT_EQ(state.bindCalls[1].texture, 0u);
+    EXPECT_EQ(DirectGLES::TextureImpl::g_boundTexturesCache[0][static_cast<SizeT>(TextureTarget::Texture2D)],
+              nullptr);
+}
+
 TEST(DirectGLESSanity, ProvidesNamedFramebufferBlitForDirectStateAccess) {
     MobileGL::MG_Backend::DirectGLES::BackendObject_DirectGLES backend;
     const auto& funcs = backend.GetBackendFunctions().GL;

MobileGL/MG_Test/Texture/CMakeLists.txt

@@ -17,7 +17,7 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(TextureTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(TextureTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)
 
 add_executable(
         VkClearManagerTest
@@ -35,4 +35,4 @@ target_link_libraries(
         ${LINK_LIBRARIES}
 )
 
-gtest_discover_tests(VkClearManagerTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(VkClearManagerTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/VertexArray/CMakeLists.txt

@@ -17,4 +17,4 @@ target_link_libraries(
 )
 
 include(GoogleTest)
-gtest_discover_tests(VertexArrayTest DISCOVERY_TIMEOUT 30)
+gtest_discover_tests(VertexArrayTest DISCOVERY_TIMEOUT 30 PROPERTIES LABELS unit)

MobileGL/MG_Test/VertexArray/VertexArrayTest.cpp

@@ -8,6 +8,8 @@
 
 #include <gtest/gtest.h>
 
+#include <limits>
+
 #include "Includes.h"
 #include "Init.h"
 
@@ -35,9 +37,31 @@ protected:
 
         return vbo;
     }
-    void SetUp() override { MobileGL::Initialize(); }
+    // GL error flags are sticky per error code and the context outlives an individual test in this
+    // binary, so drain whatever an earlier test left pending - otherwise an error-code assertion
+    // here reads someone else's error. Bounded: one flag per code, so this cannot hang the suite.
+    static void DrainPendingGlErrors() {
+        for (Int drained = 0; drained < 16 && MG_Impl::GLImpl::GetError() != GL_NO_ERROR; ++drained) {
+        }
+    }
 
-    void TearDown() override {}
+    // The call under test must raise exactly the expected error and nothing more: a second pending
+    // error means one entry point queued several, which GetError() would hand out at an unrelated
+    // call site later on.
+    static void ExpectSingleGlError(GLenum expected) {
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), expected);
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "the call recorded more than one error";
+    }
+
+    void SetUp() override {
+        MobileGL::Initialize();
+        DrainPendingGlErrors();
+    }
+
+    void TearDown() override {
+        // Attribute a leaked error to the test that caused it instead of to whoever runs next.
+        EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR) << "test left an unconsumed GL error behind";
+    }
 };
 
 TEST_F(VertexArrayTest, GenerateAndBindVAO) {
@@ -57,6 +81,46 @@ TEST_F(VertexArrayTest, GenerateAndBindVAO) {
     // Do not detect if it supports default VAO
 }
 
+// GL 3.3 core 2.10 name lifecycle - the same three rules the other object families assert:
+// deleting an unknown name is silent, a released reservation is recycled, and binding a dead
+// name is INVALID_OPERATION.
+TEST_F(VertexArrayTest, DeleteOfUnknownOrAlreadyDeletedVertexArrayNameIsSilent) {
+    GLuint vao = 0;
+    MG_Impl::GLImpl::GenVertexArrays(1, &vao);
+    ASSERT_NE(vao, 0u);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteVertexArrays(1, &vao);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    MG_Impl::GLImpl::DeleteVertexArrays(1, &vao);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // Not a small literal: other tests in this binary share the context and generate names in
+    // bulk, so a low number may well be a legitimately reserved name here.
+    const GLuint unknownNames[] = {0u, std::numeric_limits<GLuint>::max()};
+    MG_Impl::GLImpl::DeleteVertexArrays(2, unknownNames);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(VertexArrayTest, DeleteGeneratedButUnboundVertexArrayNameReleasesReservationAndBindFails) {
+    GLuint vao = 0;
+    MG_Impl::GLImpl::GenVertexArrays(1, &vao);
+    ASSERT_NE(vao, 0u);
+    ASSERT_TRUE(MG_State::pGLContext->ValidateVertexArrayName(vao));
+
+    MG_Impl::GLImpl::DeleteVertexArrays(1, &vao);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+    EXPECT_FALSE(MG_State::pGLContext->ValidateVertexArrayName(vao));
+
+    MG_Impl::GLImpl::BindVertexArray(vao);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    GLuint recycled = 0;
+    MG_Impl::GLImpl::GenVertexArrays(1, &recycled);
+    EXPECT_EQ(recycled, vao);
+}
+
 TEST_F(VertexArrayTest, VertexAttributeSetup) {
     Vector<Uint> vaoNames;
     MobileGL::MG_State::pGLContext->GenVertexArrayNames(1, vaoNames);
@@ -264,7 +328,9 @@ TEST_F(VertexArrayTest, VertexBindingIndexIsBoundedByTheAdvertisedAttribLimit) {
 
     const GLuint outOfRange = MG_Impl::GLImpl::VertexArrayImpl::GetMaxVertexAttribs();
     MG_Impl::GLImpl::VertexAttribBinding(0, outOfRange);
-    EXPECT_TRUE(MG_State::pGLContext->HasGLError());
+    // Asserting the exact code (rather than just "some error") also consumes it, so the next test
+    // does not inherit it - GL error flags are sticky and this context is shared.
+    ExpectSingleGlError(GL_INVALID_VALUE);
 }
 
 // The default attribute -> binding-point mapping is the identity. It used to be a 16-element literal
@@ -1107,9 +1173,17 @@ TEST_F(GeneralVertexArrayTest, CurrentAttrib_PackedValidation) {
     GetVertexAttribfv(1, GL_CURRENT_VERTEX_ATTRIB, out);
     EXPECT_FLOAT_EQ(out[0], 42.0f); // unchanged by the failed call
 
-    // Attribute 0 is rejected by MobileGL policy (GL_INVALID_OPERATION).
+    // GL 3.3 core 2.7: attribute 0's current value is writable like any other generic
+    // attribute - no error. (An earlier MobileGL policy rejected index 0 with
+    // GL_INVALID_OPERATION, which broke GL CTS's per-case state reset: gluStateReset writes
+    // vertexAttrib4f(0, 0,0,0,1) for every attribute after every case.)
     VertexAttribP4ui(0, GL_UNSIGNED_INT_2_10_10_10_REV, GL_FALSE, 1u);
-    EXPECT_EQ(GetError(), GL_INVALID_OPERATION);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+    GetVertexAttribfv(0, GL_CURRENT_VERTEX_ATTRIB, out);
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
+    EXPECT_FLOAT_EQ(out[0], 1.0f); // x field of the packed word
+    VertexAttrib4f(0, 0.0f, 0.0f, 0.0f, 1.0f); // restore the initial current value
+    EXPECT_EQ(GetError(), GL_NO_ERROR);
 
     // Out-of-range index -> GL_INVALID_VALUE.
     VertexAttribP4ui(VertexArrayImpl::GetMaxVertexAttribs(), GL_UNSIGNED_INT_2_10_10_10_REV, GL_FALSE, 1u);

MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.cpp

@@ -799,6 +799,9 @@ namespace MobileGL::MG_Util::BackendLoader {
                 if (std::strcmp(extension, "GL_EXT_texture_norm16") == 0) {
                     caps.SupportsNorm16Texture = true;
                 }
+                if (std::strcmp(extension, "GL_EXT_texture_filter_anisotropic") == 0) {
+                    caps.SupportsTextureFilterAnisotropy = true;
+                }
                 if (std::strcmp(extension, "GL_EXT_base_instance") == 0) {
                     caps.SupportsBaseInstance = true;
                 }
@@ -908,6 +911,13 @@ namespace MobileGL::MG_Util::BackendLoader {
         glesFuncs.glGetIntegerv(GL_MAX_VIEWPORTS, &maxViewports);
         glesFuncs.glGetIntegerv(GL_MAX_VIEWPORT_DIMS, maxViewportDims);
         glesFuncs.glGetIntegerv(GL_VIEWPORT_SUBPIXEL_BITS, &viewportSubpixelBits);
+        // Only legal to query once the extension has been seen in the loop above, hence not batched
+        // with the unconditional probes: on a driver without it this raises GL_INVALID_ENUM.
+        if (caps.SupportsTextureFilterAnisotropy) {
+            GLfloat maxTextureMaxAnisotropy = 1.0f;
+            glesFuncs.glGetFloatv(GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &maxTextureMaxAnisotropy);
+            caps.MaxTextureMaxAnisotropy = std::max(maxTextureMaxAnisotropy, 1.0f);
+        }
         caps.AliasedLineWidthRangeMin = aliasedLineWidthRange[0];
         caps.AliasedLineWidthRangeMax = aliasedLineWidthRange[1];
         caps.SmoothLineWidthRangeMin = smoothLineWidthRange[0];

MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.h

@@ -1031,6 +1031,12 @@ namespace MobileGL {
             String GLESShadingLanguageVersionString;
             Bool SupportsPersistentMapping = false;
             Bool SupportsNorm16Texture = false;
+            // GL_EXT_texture_filter_anisotropic is present, so sampler/texture
+            // anisotropy may be forwarded without raising GL_INVALID_ENUM in GLES.
+            Bool SupportsTextureFilterAnisotropy = false;
+            // GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT of the host driver; only queried when the
+            // extension above is present, and left at 1.0 (no anisotropy) otherwise.
+            Float MaxTextureMaxAnisotropy = 1.0f;
             Bool SupportsBaseInstance = false;
             // GL_EXT_disjoint_timer_query is present in the extension string.
             Bool SupportsDisjointTimerQuery = false;

MobileGL/MG_Util/BackendLoaders/Vulkan/Loader.cpp

@@ -124,6 +124,7 @@ namespace MobileGL::MG_Util::BackendLoader {
         caps.UniformBufferOffsetAlignment = static_cast<int>(p.limits.minUniformBufferOffsetAlignment);
         caps.AliasedLineWidthRangeMin = p.limits.lineWidthRange[0];
         caps.AliasedLineWidthRangeMax = p.limits.lineWidthRange[1];
+        caps.MaxSamplerAnisotropy = p.limits.maxSamplerAnisotropy;
         caps.SmoothLineWidthRangeMin = p.limits.lineWidthRange[0];
         caps.SmoothLineWidthRangeMax = p.limits.lineWidthRange[1];
         caps.SmoothLineWidthGranularity = p.limits.lineWidthGranularity;
@@ -208,6 +209,7 @@ namespace MobileGL::MG_Util::BackendLoader {
         caps.UniformBufferOffsetAlignment = static_cast<int>(properties.limits.minUniformBufferOffsetAlignment);
         caps.AliasedLineWidthRangeMin = properties.limits.lineWidthRange[0];
         caps.AliasedLineWidthRangeMax = properties.limits.lineWidthRange[1];
+        caps.MaxSamplerAnisotropy = properties.limits.maxSamplerAnisotropy;
         caps.SmoothLineWidthRangeMin = properties.limits.lineWidthRange[0];
         caps.SmoothLineWidthRangeMax = properties.limits.lineWidthRange[1];
         caps.SmoothLineWidthGranularity = properties.limits.lineWidthGranularity;

MobileGL/MG_Util/BackendLoaders/Vulkan/Loader.h

@@ -18,6 +18,9 @@ namespace MobileGL {
             Int UniformBufferOffsetAlignment = 256;
             Float AliasedLineWidthRangeMin = 1.0f;
             Float AliasedLineWidthRangeMax = 1.0f;
+            // VkPhysicalDeviceLimits::maxSamplerAnisotropy. Whether it can be used at all depends on
+            // the samplerAnisotropy feature, which the renderer decides at device creation.
+            Float MaxSamplerAnisotropy = 1.0f;
             Float SmoothLineWidthRangeMin = 1.0f;
             Float SmoothLineWidthRangeMax = 1.0f;
             Float SmoothLineWidthGranularity = 1.0f;

MobileGL/MG_Util/Converters/GLToMG/TextureEnumConverter.cpp

@@ -58,9 +58,17 @@ namespace MobileGL {
 
         TextureInputFormat ConvertGLEnumToTextureInputFormat(GLenum format) {
             switch (format) {
+            // Legacy carve-out: GL_ALPHA stays mapped to Red so alpha-texture uploads keep landing in
+            // the R channel of the R8-backed storage (TexImage*_State pairs this with a 0,0,0,R
+            // swizzle). Readback of GL_ALPHA is corrected at the backend, which maps the raw enum to
+            // source channel 3 (DirectGLES GetReadbackChannelMapping).
             case GL_ALPHA:
             case GL_RED:
                 return TextureInputFormat::Red;
+            case GL_GREEN:
+                return TextureInputFormat::Green;
+            case GL_BLUE:
+                return TextureInputFormat::Blue;
             case GL_RG:
                 return TextureInputFormat::RG;
             case GL_RGB:
@@ -73,6 +81,12 @@ namespace MobileGL {
                 return TextureInputFormat::BGRA;
             case GL_RED_INTEGER:
                 return TextureInputFormat::RInteger;
+            case GL_GREEN_INTEGER:
+                return TextureInputFormat::GreenInteger;
+            case GL_BLUE_INTEGER:
+                return TextureInputFormat::BlueInteger;
+            case GL_ALPHA_INTEGER:
+                return TextureInputFormat::AlphaInteger;
             case GL_RG_INTEGER:
                 return TextureInputFormat::RGInteger;
             case GL_RGB_INTEGER:
@@ -117,6 +131,9 @@ namespace MobileGL {
             case GL_RGB4:
                 return TextureInternalFormat::RGB4;
             case GL_RGB5:
+            // GL_RGB565 (GL 4.1 / ARB_ES2_compatibility, used directly by the GL CTS) is the
+            // ES-facing rendition of the legacy RGB5 resolution.
+            case GL_RGB565:
                 return TextureInternalFormat::RGB5;
             case GL_RGB8:
                 return TextureInternalFormat::RGB8;
@@ -294,6 +311,8 @@ namespace MobileGL {
                 return TexturePixelDataType::UnsignedInt8888;
             case GL_UNSIGNED_INT_8_8_8_8_REV:
                 return TexturePixelDataType::UnsignedInt8888Rev;
+            case GL_UNSIGNED_INT_10_10_10_2:
+                return TexturePixelDataType::UnsignedInt1010102;
             case GL_UNSIGNED_INT_10F_11F_11F_REV:
                 return TexturePixelDataType::UnsignedInt101111Rev;
             case GL_UNSIGNED_INT_2_10_10_10_REV:

MobileGL/MG_Util/Converters/GLToStr/GLEnumConverter.cpp

@@ -535,6 +535,7 @@ namespace MobileGL {
                 CASE(GL_RED_INTEGER)
                 CASE(GL_GREEN_INTEGER)
                 CASE(GL_BLUE_INTEGER)
+                CASE(GL_ALPHA_INTEGER)
                 CASE(GL_RGB_INTEGER)
                 CASE(GL_RGBA_INTEGER)
                 CASE(GL_BGR_INTEGER)

MobileGL/MG_Util/Converters/MGToGL/TextureEnumConverter.cpp

@@ -67,6 +67,18 @@ namespace MobileGL {
                 return GL_RGBA_INTEGER;
             case TextureInputFormat::BGRAInteger:
                 return GL_BGRA_INTEGER;
+            case TextureInputFormat::Green:
+                return GL_GREEN;
+            case TextureInputFormat::Blue:
+                return GL_BLUE;
+            case TextureInputFormat::Alpha:
+                return GL_ALPHA;
+            case TextureInputFormat::GreenInteger:
+                return GL_GREEN_INTEGER;
+            case TextureInputFormat::BlueInteger:
+                return GL_BLUE_INTEGER;
+            case TextureInputFormat::AlphaInteger:
+                return GL_ALPHA_INTEGER;
             case TextureInputFormat::StencilIndex:
                 return GL_STENCIL_INDEX;
             case TextureInputFormat::DepthComponent:
@@ -101,7 +113,10 @@ namespace MobileGL {
             case TextureInternalFormat::RGB4:
                 return GL_RGB4;
             case TextureInternalFormat::RGB5:
-                return GL_RGB5;
+                // Emit the ES-compatible GL_RGB565 rendition: desktop GL_RGB5 is not a legal
+                // sized internalformat on OpenGL ES backends, GL_RGB565 is (and GL 4.1+
+                // accepts it too via ARB_ES2_compatibility).
+                return GL_RGB565;
             case TextureInternalFormat::RGB8:
                 return GL_RGB8;
             case TextureInternalFormat::RGB8Snorm:

MobileGL/MG_Util/Converters/MGToMG/TextureEnumConverter.cpp

@@ -129,6 +129,8 @@ namespace MobileGL {
             case TextureInternalFormat::RGBA: {
                 switch (type) {
                 case TexturePixelDataType::UnsignedByte:
+                case TexturePixelDataType::UnsignedInt8888:
+                case TexturePixelDataType::UnsignedInt8888Rev:
                     return TextureInternalFormat::RGBA8;
                 case TexturePixelDataType::UnsignedShort:
                     return TextureInternalFormat::RGBA16;

MobileGL/MG_Util/Converters/MGToStr/TextureEnumConverter.cpp

@@ -67,6 +67,18 @@ namespace MobileGL {
                 return "RGBAInteger";
             case TextureInputFormat::BGRAInteger:
                 return "BGRAInteger";
+            case TextureInputFormat::Green:
+                return "Green";
+            case TextureInputFormat::Blue:
+                return "Blue";
+            case TextureInputFormat::Alpha:
+                return "Alpha";
+            case TextureInputFormat::GreenInteger:
+                return "GreenInteger";
+            case TextureInputFormat::BlueInteger:
+                return "BlueInteger";
+            case TextureInputFormat::AlphaInteger:
+                return "AlphaInteger";
             case TextureInputFormat::StencilIndex:
                 return "StencilIndex";
             case TextureInputFormat::DepthComponent:

MobileGL/MG_Util/Converters/MGToVk/TextureEnumConverter.cpp

@@ -65,6 +65,16 @@ namespace MobileGL {
                 return VK_FORMAT_R8G8B8A8_UINT;
             case TextureInputFormat::BGRAInteger:
                 return VK_FORMAT_B8G8R8A8_UINT;
+            // Single-channel desktop client formats: the client memory holds one component per pixel,
+            // matching the R8 layouts (the channel it feeds is a pixel-transfer concern, not a layout one).
+            case TextureInputFormat::Green:
+            case TextureInputFormat::Blue:
+            case TextureInputFormat::Alpha:
+                return VK_FORMAT_R8_UNORM;
+            case TextureInputFormat::GreenInteger:
+            case TextureInputFormat::BlueInteger:
+            case TextureInputFormat::AlphaInteger:
+                return VK_FORMAT_R8_UINT;
             case TextureInputFormat::StencilIndex:
                 return VK_FORMAT_S8_UINT;
             case TextureInputFormat::DepthComponent:

MobileGL/MG_Util/Math/HalfFloat.h

@@ -0,0 +1,69 @@
+// MobileGL - MobileGL/MG_Util/Math/HalfFloat.h
+// Copyright (c) 2025-2026 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
+// End of Source File Header
+
+#pragma once
+#include <Includes.h>
+
+namespace MobileGL::MG_Util {
+    inline Float DecodeHalfBitsToFloat(Uint16 half) {
+        const Uint32 sign = static_cast<Uint32>(half & 0x8000u) << 16;
+        const Uint32 exponent = (half >> 10) & 0x1Fu;
+        const Uint32 mantissa = half & 0x3FFu;
+        Uint32 bits;
+        if (exponent == 0) {
+            if (mantissa == 0) {
+                bits = sign; // signed zero
+            } else {
+                // Subnormal half: renormalize into a float exponent.
+                Uint32 e = 127 - 15 + 1;
+                Uint32 m = mantissa;
+                while ((m & 0x400u) == 0) {
+                    m <<= 1;
+                    --e;
+                }
+                bits = sign | (e << 23) | ((m & 0x3FFu) << 13);
+            }
+        } else if (exponent == 31) {
+            bits = sign | 0x7F800000u | (mantissa << 13); // Inf / NaN
+        } else {
+            bits = sign | ((exponent + 112) << 23) | (mantissa << 13);
+        }
+        return std::bit_cast<Float>(bits);
+    }
+
+    inline Uint16 EncodeFloatToHalfBits(Float value) {
+        const Uint32 bits = std::bit_cast<Uint32>(value);
+        const auto sign = static_cast<Uint16>((bits >> 16) & 0x8000u);
+        const Uint32 exponent = (bits >> 23) & 0xFFu;
+        const Uint32 mantissa = bits & 0x7FFFFFu;
+        if (exponent == 0xFF) { // Inf / NaN
+            return static_cast<Uint16>(sign | 0x7C00u | (mantissa != 0 ? 0x200u : 0u));
+        }
+        const Int32 halfExponent = static_cast<Int32>(exponent) - 127 + 15;
+        if (halfExponent >= 31) {
+            return static_cast<Uint16>(sign | 0x7C00u); // overflow -> Inf
+        }
+        if (halfExponent <= 0) {
+            if (halfExponent < -10) {
+                return sign; // underflow -> signed zero
+            }
+            const Uint32 m = mantissa | 0x800000u;
+            const Uint32 shift = static_cast<Uint32>(14 - halfExponent);
+            Uint32 half = m >> shift;
+            if ((m >> (shift - 1)) & 1u) {
+                ++half; // round to nearest
+            }
+            return static_cast<Uint16>(sign | half);
+        }
+        Uint32 half = (static_cast<Uint32>(halfExponent) << 10) | (mantissa >> 13);
+        if (mantissa & 0x1000u) {
+            ++half; // round to nearest; a carry into the exponent is the correct result
+        }
+        return static_cast<Uint16>(sign | half);
+    }
+} // namespace MobileGL::MG_Util

MobileGL/MG_Util/Math/SmallFloat.h

@@ -0,0 +1,116 @@
+// MobileGL - MobileGL/MG_Util/Math/SmallFloat.h
+// Copyright (c) 2025-2026 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
+// End of Source File Header
+
+#pragma once
+#include <Includes.h>
+
+#include <bit>
+#include <cmath>
+#include <limits>
+
+namespace MobileGL::MG_Util {
+    // Encodes an unsigned small float with a 5-bit exponent (bias 15) and mantissaBits mantissa
+    // bits, per the EXT_packed_float conversion rules: negatives (including -Inf) go to zero,
+    // +Inf stays +Inf, NaN stays NaN, and finite values above the largest representable value
+    // clamp to it. The mantissa is truncated (rounding mode is implementation-defined).
+    inline Uint32 EncodeFloatToUnsignedSmallFloat(Float value, Int mantissaBits) {
+        const Uint32 bits = std::bit_cast<Uint32>(value);
+        const Bool negative = (bits & 0x80000000u) != 0;
+        const Uint32 exponent = (bits >> 23) & 0xFFu;
+        const Uint32 mantissa = bits & 0x7FFFFFu;
+        const Uint32 exponentMask = 0x1Fu << mantissaBits;
+        if (exponent == 0xFFu) {
+            if (mantissa != 0) {
+                return exponentMask | 1u; // NaN keeps NaN
+            }
+            return negative ? 0u : exponentMask; // -Inf -> 0, +Inf -> +Inf
+        }
+        if (negative) {
+            return 0u;
+        }
+        const Int32 smallExponent = static_cast<Int32>(exponent) - 127 + 15;
+        if (smallExponent >= 31) { // above the largest finite value -> clamp to it
+            return ((31u - 1u) << mantissaBits) | ((1u << mantissaBits) - 1u);
+        }
+        if (smallExponent <= 0) { // subnormal range: renormalize, flushing tiny values to zero
+            const Uint32 fullMantissa = mantissa | 0x800000u;
+            const Int32 shift = (23 - mantissaBits) + 1 - smallExponent;
+            return shift > 23 ? 0u : fullMantissa >> shift;
+        }
+        return (static_cast<Uint32>(smallExponent) << mantissaBits) |
+               (mantissa >> (23u - static_cast<Uint32>(mantissaBits)));
+    }
+
+    inline Uint32 EncodeFloatToUnsignedF11(Float value) { return EncodeFloatToUnsignedSmallFloat(value, 6); }
+    inline Uint32 EncodeFloatToUnsignedF10(Float value) { return EncodeFloatToUnsignedSmallFloat(value, 5); }
+
+    // Decodes an unsigned small float (5-bit exponent, bias 15, mantissaBits mantissa bits).
+    inline Float DecodeUnsignedSmallFloatToFloat(Uint32 field, Int mantissaBits) {
+        const Uint32 exponent = (field >> mantissaBits) & 0x1Fu;
+        const Uint32 mantissa = field & ((1u << mantissaBits) - 1u);
+        const Float mantissaScale = 1.0f / static_cast<Float>(1u << mantissaBits);
+        if (exponent == 0) {
+            return std::exp2(-14.0f) * static_cast<Float>(mantissa) * mantissaScale;
+        }
+        if (exponent == 31) {
+            return mantissa == 0 ? std::numeric_limits<Float>::infinity()
+                                 : std::numeric_limits<Float>::quiet_NaN();
+        }
+        return std::exp2(static_cast<Float>(exponent) - 15.0f) *
+               (1.0f + static_cast<Float>(mantissa) * mantissaScale);
+    }
+
+    inline Float DecodeUnsignedF11ToFloat(Uint32 field) { return DecodeUnsignedSmallFloatToFloat(field, 6); }
+    inline Float DecodeUnsignedF10ToFloat(Uint32 field) { return DecodeUnsignedSmallFloatToFloat(field, 5); }
+
+    // RGB9E5 shared-exponent encode, following the EXT_texture_shared_exponent spec algorithm
+    // (N = 9 mantissa bits, B = 15 exponent bias, Emax = 31).
+    inline Uint32 EncodeSharedExponentRGB9E5(const Float rgb[3]) {
+        constexpr Int kMantissaBits = 9;
+        constexpr Int kExponentBias = 15;
+        constexpr Float kSharedExpMax = 511.0f / 512.0f * 65536.0f; // (2^N-1)/2^N * 2^(Emax-B)
+
+        Float clamped[3];
+        for (Int i = 0; i < 3; ++i) {
+            const Float v = rgb[i];
+            clamped[i] = (std::isnan(v) || v < 0.0f) ? 0.0f : std::min(v, kSharedExpMax);
+        }
+        const Float maxComponent = std::max(clamped[0], std::max(clamped[1], clamped[2]));
+
+        Int sharedExponent = 0; // all-zero input keeps the all-zero word
+        if (maxComponent > 0.0f) {
+            sharedExponent = std::max(-kExponentBias - 1, static_cast<Int>(std::floor(std::log2(maxComponent)))) +
+                             1 + kExponentBias;
+            const Float maxScaled = std::floor(
+                maxComponent / std::exp2(static_cast<Float>(sharedExponent - kExponentBias - kMantissaBits)) +
+                0.5f);
+            if (maxScaled >= 512.0f) { // rounded up to 2^N: bump the shared exponent instead
+                ++sharedExponent;
+            }
+        }
+
+        const Float scale = std::exp2(static_cast<Float>(sharedExponent - kExponentBias - kMantissaBits));
+        Uint32 word = static_cast<Uint32>(sharedExponent) << 27;
+        for (Int i = 0; i < 3; ++i) {
+            const auto field = static_cast<Uint32>(std::floor(clamped[i] / scale + 0.5f));
+            word |= std::min(field, 511u) << (i * kMantissaBits);
+        }
+        return word;
+    }
+
+    // RGB9E5 shared-exponent decode.
+    inline void DecodeSharedExponentRGB9E5(Uint32 word, Float outRgb[3]) {
+        constexpr Int kMantissaBits = 9;
+        constexpr Int kExponentBias = 15;
+        const Int exponent = static_cast<Int>(word >> 27) - kExponentBias - kMantissaBits;
+        const Float scale = std::exp2(static_cast<Float>(exponent));
+        for (Int i = 0; i < 3; ++i) {
+            outRgb[i] = static_cast<Float>((word >> (i * kMantissaBits)) & 0x1FFu) * scale;
+        }
+    }
+} // namespace MobileGL::MG_Util

MobileGL/MG_Util/Metrics/TextureMetrics.cpp

@@ -15,10 +15,10 @@ namespace MobileGL {
         SizeT GetSizedInternalFormatSizeInBytes(TextureInternalFormat internal) {
             switch (internal) {
             case TextureInternalFormat::R8:
+            case TextureInternalFormat::Red: // UNorm8 shadow layout
             case TextureInternalFormat::R8Snorm:
             case TextureInternalFormat::R8I:
             case TextureInternalFormat::R8UI:
-            case TextureInternalFormat::R3G3B2:
                 return 1;
 
             case TextureInternalFormat::R16:
@@ -27,14 +27,17 @@ namespace MobileGL {
             case TextureInternalFormat::R16UI:
             case TextureInternalFormat::R16F:
             case TextureInternalFormat::RG8:
+            case TextureInternalFormat::RG: // UNorm8x2 shadow layout
             case TextureInternalFormat::RG8Snorm:
             case TextureInternalFormat::RG8I:
             case TextureInternalFormat::RG8UI:
             case TextureInternalFormat::DepthComponent16:
                 return 2;
 
+            case TextureInternalFormat::R3G3B2: // UNorm8x3 shadow layout
             case TextureInternalFormat::RGB4:
             case TextureInternalFormat::RGB5:
+            case TextureInternalFormat::RGB: // UNorm8x3 shadow layout
             case TextureInternalFormat::RGB8:
             case TextureInternalFormat::RGB8Snorm:
             case TextureInternalFormat::SRGB8:
@@ -43,11 +46,10 @@ namespace MobileGL {
             case TextureInternalFormat::DepthComponent24:
                 return 3;
 
-            case TextureInternalFormat::RGB10:
-            case TextureInternalFormat::RGB12:
             case TextureInternalFormat::RGBA2:
             case TextureInternalFormat::RGBA4:
             case TextureInternalFormat::RGB5A1:
+            case TextureInternalFormat::RGBA: // UNorm8x4 shadow layout
             case TextureInternalFormat::RGBA8:
             case TextureInternalFormat::RGBA8Snorm:
             case TextureInternalFormat::RGBA8I:
@@ -72,6 +74,8 @@ namespace MobileGL {
                 return 4;
 
             case TextureInternalFormat::RGB16:
+            case TextureInternalFormat::RGB10: // UNorm16x3 shadow layout
+            case TextureInternalFormat::RGB12: // UNorm16x3 shadow layout
             case TextureInternalFormat::RGB16Snorm:
             case TextureInternalFormat::RGB16F:
             case TextureInternalFormat::RGB16I:
@@ -113,6 +117,12 @@ namespace MobileGL {
             switch (format) {
             case TextureInputFormat::Red:
             case TextureInputFormat::RInteger:
+            case TextureInputFormat::Green:
+            case TextureInputFormat::GreenInteger:
+            case TextureInputFormat::Blue:
+            case TextureInputFormat::BlueInteger:
+            case TextureInputFormat::Alpha:
+            case TextureInputFormat::AlphaInteger:
                 return 1;
             case TextureInputFormat::RG:
             case TextureInputFormat::RGInteger:

MobileGL/MG_Util/SelfTest/DriverPost.cpp

@@ -548,8 +548,8 @@ namespace MobileGL::MG_Util::SelfTest {
         if (summary.capsValid) {
             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));
+            advertisedExtensions = JoinAdvertisedExtensions(MG_Backend::DirectGLES::BuildAdvertisedExtensions(
+                summary.caps.SupportsDisjointTimerQuery, summary.caps.SupportsTextureFilterAnisotropy));
         }
         AppendMobileGLReportedRows(builder, MG_Backend::DirectGLES::GetRendererIdentity(), backendApiVersionString,
                                    advertisedExtensions);
@@ -841,6 +841,7 @@ namespace MobileGL::MG_Util::SelfTest {
             String driverVersionString; // raw hex, vendor-encoded (see RunVulkanDriverPost)
             Bool shaderSubgroupUsable = false;
             Bool timerQueriesSupported = false;
+            Bool samplerAnisotropySupported = false;
         };
     } // namespace
 
@@ -1107,6 +1108,7 @@ namespace MobileGL::MG_Util::SelfTest {
 
         VkPhysicalDeviceFeatures features{};
         vkGetPhysicalDeviceFeaturesFn(physicalDevice, &features);
+        summary.samplerAnisotropySupported = features.samplerAnisotropy == VK_TRUE;
         if (features.multiDrawIndirect == VK_TRUE) {
             builder.Pass("multiDrawIndirect", "indirect multi-draw batches run as single native commands");
         } else {
@@ -1279,7 +1281,7 @@ 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.shaderSubgroupUsable, summary.timerQueriesSupported, summary.samplerAnisotropySupported));
         }
         AppendMobileGLReportedRows(builder, MG_Backend::DirectVulkan::GetRendererIdentity(), backendApiVersionString,
                                    advertisedExtensions);

MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.cpp

@@ -14,11 +14,14 @@
 #include "SpirvPasses/DecomposeWorkgroupVec3Pass.h"
 #include "SpirvPasses/LowerDrawParametersPass.h"
 #include "SpirvPasses/RebaseInstanceIndexPass.h"
+#include "SpirvPasses/StripUboMemberRelaxedPrecisionPass.h"
 #include "spirv-tools/libspirv.h"
 #include "spirv-tools/optimizer.hpp"
 
+#include "ShaderSourceProcessor.h"
 #include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
 #include <MG_Util/Converters/GLToGlslang/ProgramEnumConverter.h>
+#include <cstdlib>
 
 namespace MobileGL {
     namespace MG_Util {
@@ -132,27 +135,23 @@ namespace MobileGL {
                 return Resources;
             }
 
-            Result<SharedPtr<glslang::TShader>> ShaderCompiler::CompileShader(const ShaderAttrib& attrib) {
-                auto shaderType = attrib.shaderType;
-                auto& sourceStr = attrib.sourceStr;
-
-                auto lang = MG_Util::ConvertGLEnumToEShLanguage(shaderType);
-                if (lang == EShLanguage::EShLangCount) {
-                    ResultInfo r;
-                    r.log += "Error: [Preprocess] Unsupported shader type: " + ConvertGLEnumToString(shaderType);
-                    r.errc = -1;
-                    return std::unexpected(r);
-                }
-
+            // One parse attempt. A glslang::TShader cannot be re-parsed, so a retry has to build a
+            // fresh one with byte-identical setup - hence a single factored body rather than two
+            // copies that could drift apart.
+            static Result<SharedPtr<glslang::TShader>> ParseShaderSource(EShLanguage lang, GLenum shaderType,
+                                                                         const String& source,
+                                                                         Flags<ShaderCompileBits> flags) {
                 SharedPtr<glslang::TShader> res;
                 auto& tshader = res;
                 tshader = MakeShared<glslang::TShader>(lang);
-                const char* src[] = {sourceStr.data()};
+                // setStrings gets no length array, so it relies on NUL termination: source must be an
+                // owning buffer that outlives parse(), never a StringView's substring.
+                const char* src[] = {source.c_str()};
                 tshader->setStrings(src, 1);
                 tshader->setNanMinMaxClamp(true);
                 tshader->setInvertY(true);
                 tshader->setPreamble("#undef VULKAN\n");
-                if (attrib.flags & ShaderCompileBits::CompileForOpenGL) {
+                if (flags & ShaderCompileBits::CompileForOpenGL) {
                     tshader->setEnvInput(glslang::EShSourceGlsl, lang, glslang::EShClientVulkan, 450);
                     tshader->setEnvClient(glslang::EShClientOpenGL, glslang::EShTargetOpenGL_450);
                     tshader->setEnvTarget(glslang::EShTargetSpv, glslang::EShTargetSpv_1_3);
@@ -181,6 +180,39 @@ namespace MobileGL {
                 return res;
             }
 
+            Result<SharedPtr<glslang::TShader>> ShaderCompiler::CompileShader(const ShaderAttrib& attrib) {
+                auto shaderType = attrib.shaderType;
+
+                auto lang = MG_Util::ConvertGLEnumToEShLanguage(shaderType);
+                if (lang == EShLanguage::EShLangCount) {
+                    ResultInfo r;
+                    r.log += "Error: [Preprocess] Unsupported shader type: " + ConvertGLEnumToString(shaderType);
+                    r.errc = -1;
+                    return std::unexpected(r);
+                }
+
+                const String source(attrib.sourceStr);
+                auto result = ParseShaderSource(lang, shaderType, source, attrib.flags);
+                if (result) return result;
+
+                // Legacy desktop sources are normalized to "#version 330 core", which parses under
+                // stricter rules than the 460 they used to be forced to: a shader declaring 330 while
+                // using e.g. layout(binding=...) without the matching #extension line compiles on real
+                // drivers but is rejected here. Retry once at 460 before reporting failure; a genuinely
+                // broken shader fails both attempts and keeps its original diagnostics.
+                String retrySource = source;
+                if (!MG_Util::ShaderTranspiler::RetargetLegacyVersionDirectiveTo460(retrySource)) {
+                    return result;
+                }
+
+                auto retryResult = ParseShaderSource(lang, shaderType, retrySource, attrib.flags);
+                if (!retryResult) return result;
+
+                MGLOG_D("CompileShader: %s only parsed after retargeting its legacy #version to 460",
+                        ConvertGLEnumToString(shaderType).c_str());
+                return retryResult;
+            }
+
             Result<SharedPtr<glslang::TProgram>> ShaderCompiler::LinkProgram(const ProgramAttrib& attrib) {
                 SharedPtr<glslang::TProgram> program = MakeShared<glslang::TProgram>();
                 for (auto& s : attrib.shaders) {
@@ -266,6 +298,19 @@ namespace MobileGL {
                 return optimizer.Run(inputBinary.data(), inputBinary.size(), &outputBinary, options);
             }
 
+            bool ShaderCompiler::StripUboMemberRelaxedPrecisionForEssl(const Vector<Uint32>& inputBinary,
+                                                                       Vector<uint32_t>& outputBinary) {
+                using namespace spvtools;
+                OptimizerOptions options;
+                options.set_run_validator(false);
+
+                Optimizer optimizer(SPV_ENV_VULKAN_1_1);
+                optimizer.RegisterPass(
+                    StripUboMemberRelaxedPrecisionPass::CreateStripUboMemberRelaxedPrecisionPass());
+
+                return optimizer.Run(inputBinary.data(), inputBinary.size(), &outputBinary, options);
+            }
+
             bool ShaderCompiler::RebaseInstanceIndexForVulkan(const Vector<Uint32>& inputBinary,
                                                               Vector<uint32_t>& outputBinary) {
                 using namespace spvtools;
@@ -278,6 +323,75 @@ namespace MobileGL {
                 return optimizer.Run(inputBinary.data(), inputBinary.size(), &outputBinary, options);
             }
 
+            bool ShaderCompiler::DecoratePositionInvariantForVulkan(const Vector<Uint32>& inputBinary,
+                                                                    Vector<uint32_t>& outputBinary) {
+                static constexpr Uint32 kHeaderWords = 5;
+                static constexpr Uint32 kOpDecorate = 71;
+                static constexpr Uint32 kOpMemberDecorate = 72;
+                static constexpr Uint32 kDecorationInvariant = 18;
+                static constexpr Uint32 kDecorationBuiltIn = 11;
+                static constexpr Uint32 kBuiltInPosition = 0;
+                if (inputBinary.size() < kHeaderWords) {
+                    return false;
+                }
+
+                // First pass: find targets that already carry Invariant so we never duplicate.
+                struct MemberKey {
+                    Uint32 id;
+                    Uint32 member;
+                    bool operator==(const MemberKey& o) const { return id == o.id && member == o.member; }
+                };
+                Vector<Uint32> invariantIds;
+                Vector<MemberKey> invariantMembers;
+                for (SizeT i = kHeaderWords; i < inputBinary.size();) {
+                    const Uint32 word0 = inputBinary[i];
+                    const Uint32 opcode = word0 & 0xFFFFu;
+                    const Uint32 length = word0 >> 16;
+                    if (length == 0 || i + length > inputBinary.size()) {
+                        return false;
+                    }
+                    if (opcode == kOpDecorate && length >= 3 && inputBinary[i + 2] == kDecorationInvariant) {
+                        invariantIds.push_back(inputBinary[i + 1]);
+                    } else if (opcode == kOpMemberDecorate && length >= 4 &&
+                               inputBinary[i + 3] == kDecorationInvariant) {
+                        invariantMembers.push_back({inputBinary[i + 1], inputBinary[i + 2]});
+                    }
+                    i += length;
+                }
+
+                outputBinary.clear();
+                outputBinary.reserve(inputBinary.size() + 8);
+                outputBinary.insert(outputBinary.end(), inputBinary.begin(), inputBinary.begin() + kHeaderWords);
+                for (SizeT i = kHeaderWords; i < inputBinary.size();) {
+                    const Uint32 word0 = inputBinary[i];
+                    const Uint32 opcode = word0 & 0xFFFFu;
+                    const Uint32 length = word0 >> 16;
+                    outputBinary.insert(outputBinary.end(), inputBinary.begin() + i,
+                                        inputBinary.begin() + i + length);
+                    if (opcode == kOpDecorate && length == 4 &&
+                        inputBinary[i + 2] == kDecorationBuiltIn && inputBinary[i + 3] == kBuiltInPosition) {
+                        const Uint32 target = inputBinary[i + 1];
+                        if (std::find(invariantIds.begin(), invariantIds.end(), target) == invariantIds.end()) {
+                            outputBinary.push_back((3u << 16) | kOpDecorate);
+                            outputBinary.push_back(target);
+                            outputBinary.push_back(kDecorationInvariant);
+                        }
+                    } else if (opcode == kOpMemberDecorate && length == 5 &&
+                               inputBinary[i + 3] == kDecorationBuiltIn && inputBinary[i + 4] == kBuiltInPosition) {
+                        const MemberKey key{inputBinary[i + 1], inputBinary[i + 2]};
+                        if (std::find(invariantMembers.begin(), invariantMembers.end(), key) ==
+                            invariantMembers.end()) {
+                            outputBinary.push_back((4u << 16) | kOpMemberDecorate);
+                            outputBinary.push_back(key.id);
+                            outputBinary.push_back(key.member);
+                            outputBinary.push_back(kDecorationInvariant);
+                        }
+                    }
+                    i += length;
+                }
+                return true;
+            }
+
             Result<String> ShaderCompiler::DecompileShader(SpvcSession& session) {
                 spvc_compiler_options options;
                 session.CreateOptions(&options);

MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.h

@@ -27,12 +27,25 @@ namespace MobileGL {
                 // Only for backends without native draw-parameter support (DirectGLES).
                 static bool LowerDrawParametersForEssl(const Vector<Uint32>& inputBinary,
                                                        Vector<uint32_t>& outputBinary);
+                // Drops RelaxedPrecision member decorations from uniform-block structs so
+                // SPIRV-Cross prints the same (highp) member precision in every stage; ES
+                // 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);
                 // 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,
                 // which wrongly includes baseInstance).
                 static bool RebaseInstanceIndexForVulkan(const Vector<Uint32>& inputBinary,
                                                          Vector<uint32_t>& outputBinary);
+                // 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);
                 static Result<String> DecompileShader(SpvcSession& session);
             };
         } // namespace ShaderTranspiler

MobileGL/MG_Util/ShaderTranspiler/ShaderSourceProcessor.cpp

@@ -19,6 +19,220 @@ namespace {
         return (ch >= '0' && ch <= '9') || (ch >= 'A' && ch <= 'Z') || (ch >= 'a' && ch <= 'z') || ch == '_';
     }
 
+    bool IsIdentifierStart(char ch) {
+        return (ch >= 'A' && ch <= 'Z') || (ch >= 'a' && ch <= 'z') || ch == '_';
+    }
+
+    MobileGL::String MaskCommentsAndQuotedText(const MobileGL::String& source) {
+        enum class Region { Code, SingleLineComment, MultiLineComment, QuotedText };
+
+        MobileGL::String masked = source;
+        Region region = Region::Code;
+        char quote = '\0';
+        bool escaped = false;
+
+        for (SizeT pos = 0; pos < source.size(); pos++) {
+            const char ch = source[pos];
+            const char next = pos + 1 < source.size() ? source[pos + 1] : '\0';
+
+            if (region == Region::Code) {
+                if (ch == '/' && next == '/') {
+                    masked[pos] = ' ';
+                    masked[pos + 1] = ' ';
+                    pos++;
+                    region = Region::SingleLineComment;
+                } else if (ch == '/' && next == '*') {
+                    masked[pos] = ' ';
+                    masked[pos + 1] = ' ';
+                    pos++;
+                    region = Region::MultiLineComment;
+                } else if (ch == '"' || ch == '\'') {
+                    masked[pos] = ' ';
+                    quote = ch;
+                    escaped = false;
+                    region = Region::QuotedText;
+                }
+                continue;
+            }
+
+            if (region == Region::SingleLineComment) {
+                if (ch == '\n' || ch == '\r') {
+                    region = Region::Code;
+                } else {
+                    masked[pos] = ' ';
+                }
+                continue;
+            }
+
+            if (region == Region::MultiLineComment) {
+                if (ch == '*' && next == '/') {
+                    masked[pos] = ' ';
+                    masked[pos + 1] = ' ';
+                    pos++;
+                    region = Region::Code;
+                } else if (ch != '\n' && ch != '\r') {
+                    masked[pos] = ' ';
+                }
+                continue;
+            }
+
+            if (ch != '\n' && ch != '\r') {
+                masked[pos] = ' ';
+            }
+            if (escaped) {
+                escaped = false;
+            } else if (ch == '\\') {
+                escaped = true;
+            } else if (ch == quote) {
+                region = Region::Code;
+            }
+        }
+
+        return masked;
+    }
+
+    void SkipDirectiveWhitespace(const MobileGL::String& source, SizeT& pos, SizeT lineEnd) {
+        while (pos < lineEnd && std::isspace(static_cast<unsigned char>(source[pos]))) {
+            pos++;
+        }
+    }
+
+    MobileGL::String ReadDirectiveIdentifier(const MobileGL::String& source, SizeT& pos, SizeT lineEnd) {
+        if (pos >= lineEnd || !IsIdentifierStart(source[pos])) {
+            return {};
+        }
+
+        const SizeT start = pos++;
+        while (pos < lineEnd && IsIdentifierChar(source[pos])) {
+            pos++;
+        }
+        return source.substr(start, pos - start);
+    }
+
+    bool HasUtf8Bom(const MobileGL::String& source) {
+        return source.size() >= 3 && static_cast<unsigned char>(source[0]) == 0xef &&
+               static_cast<unsigned char>(source[1]) == 0xbb && static_cast<unsigned char>(source[2]) == 0xbf;
+    }
+
+    struct ShaderLanguageInfo {
+        unsigned version = 110;
+        MobileGL::ShaderProfile profile = MobileGL::ShaderProfile::Core;
+        SizeT versionDirectiveStart = MobileGL::String::npos;
+        SizeT versionDirectiveEnd = MobileGL::String::npos;
+        bool hasUtf8Bom = false;
+        bool enablesGpuShader5 = false;
+
+        bool HasVersionDirective() const { return versionDirectiveStart != MobileGL::String::npos; }
+    };
+
+    ShaderLanguageInfo InspectShaderLanguage(const MobileGL::String& source) {
+        const MobileGL::String code = MaskCommentsAndQuotedText(source);
+        ShaderLanguageInfo info;
+        info.hasUtf8Bom = HasUtf8Bom(source);
+
+        SizeT lineStart = 0;
+        while (lineStart < code.size()) {
+            SizeT lineEnd = code.find('\n', lineStart);
+            const bool hasLineBreak = lineEnd != MobileGL::String::npos;
+            if (!hasLineBreak) {
+                lineEnd = code.size();
+            }
+
+            SizeT probe = lineStart;
+            if (lineStart == 0 && info.hasUtf8Bom) {
+                probe = 3;
+            }
+            SkipDirectiveWhitespace(code, probe, lineEnd);
+            if (probe < lineEnd && code[probe] == '#') {
+                const SizeT directiveStart = probe;
+                probe++;
+                SkipDirectiveWhitespace(code, probe, lineEnd);
+                const MobileGL::String directive = ReadDirectiveIdentifier(code, probe, lineEnd);
+
+                if (directive == "version" && !info.HasVersionDirective()) {
+                    SkipDirectiveWhitespace(code, probe, lineEnd);
+                    unsigned version = 0;
+                    bool hasVersionDigits = false;
+                    while (probe < lineEnd && code[probe] >= '0' && code[probe] <= '9') {
+                        hasVersionDigits = true;
+                        version = version * 10 + static_cast<unsigned>(code[probe] - '0');
+                        probe++;
+                    }
+                    if (hasVersionDigits) {
+                        info.version = version;
+                        info.versionDirectiveStart = directiveStart;
+                        info.versionDirectiveEnd = lineEnd + (hasLineBreak ? 1 : 0);
+                        SkipDirectiveWhitespace(code, probe, lineEnd);
+                        const MobileGL::String profile = ReadDirectiveIdentifier(code, probe, lineEnd);
+                        if (profile == "es" || profile == "ES") {
+                            info.profile = MobileGL::ShaderProfile::ES;
+                        } else if (profile == "compatibility") {
+                            info.profile = MobileGL::ShaderProfile::Compatibility;
+                        } else {
+                            info.profile = MobileGL::ShaderProfile::Core;
+                        }
+                    }
+                } else if (directive == "extension") {
+                    SkipDirectiveWhitespace(code, probe, lineEnd);
+                    const MobileGL::String extension = ReadDirectiveIdentifier(code, probe, lineEnd);
+                    SkipDirectiveWhitespace(code, probe, lineEnd);
+                    if (probe < lineEnd && code[probe] == ':') {
+                        probe++;
+                        SkipDirectiveWhitespace(code, probe, lineEnd);
+                        const MobileGL::String behavior = ReadDirectiveIdentifier(code, probe, lineEnd);
+                        const bool isGpuShader5 = extension == "GL_ARB_gpu_shader5" ||
+                                                  extension == "GL_NV_gpu_shader5";
+                        const bool enablesExtension = behavior == "enable" || behavior == "require" ||
+                                                      behavior == "warn";
+                        // Gate the whole source if it ever opts into either extension. This is deliberately
+                        // conservative around conditional directives and keeps legal sample qualifiers intact.
+                        info.enablesGpuShader5 = info.enablesGpuShader5 || (isGpuShader5 && enablesExtension);
+                    }
+                }
+            }
+
+            lineStart = lineEnd + (hasLineBreak ? 1 : 0);
+        }
+
+        return info;
+    }
+
+    MobileGL::String GetNormalizedVersionDirective(const ShaderLanguageInfo& info) {
+        if (info.profile == MobileGL::ShaderProfile::ES) {
+            // Preserve the pre-existing behavior for standard lowercase "es" directives. MobileGL's Vulkan
+            // glslang resource table cannot parse its ESSL built-ins today, even at ESSL 310, whereas the same
+            // source is accepted through the normalized desktop core path.
+            return "#version 460 core\n";
+        }
+
+        // Keep compatibility-profile handling on its pre-existing 460 path. Vulkan glslang does not accept that
+        // profile today, and this legacy-sample fix must not broaden or otherwise alter that separate limitation.
+        if (info.profile == MobileGL::ShaderProfile::Compatibility) {
+            return "#version 460 compatibility\n";
+        }
+
+        const bool useLegacyDesktopVersion =
+            info.version < 400 && !info.enablesGpuShader5;
+        return useLegacyDesktopVersion ? "#version 330 core\n" : "#version 460 core\n";
+    }
+
+    void NormalizeVersionDirective(MobileGL::String& source, const ShaderLanguageInfo& info) {
+        const MobileGL::String replacement = GetNormalizedVersionDirective(info);
+        if (info.HasVersionDirective()) {
+            source.replace(info.versionDirectiveStart, info.versionDirectiveEnd - info.versionDirectiveStart,
+                           replacement);
+            if (info.hasUtf8Bom) {
+                source.erase(0, 3);
+            }
+            return;
+        }
+
+        if (info.hasUtf8Bom) {
+            source.erase(0, 3);
+        }
+        source.insert(0, replacement);
+    }
+
     bool HasSingleLineFunctionDefinition(const MobileGL::String& source, const MobileGL::String& functionName) {
         SizeT lineStart = 0;
         while (lineStart < source.size()) {
@@ -153,12 +367,8 @@ namespace {
     }
 
     SizeT FindAfterVersionDirective(const MobileGL::String& source) {
-        const SizeT versionPos = source.find("#version");
-        if (versionPos == MobileGL::String::npos) {
-            return 0;
-        }
-        const SizeT lineEnd = source.find('\n', versionPos);
-        return lineEnd == MobileGL::String::npos ? source.size() : lineEnd + 1;
+        const ShaderLanguageInfo info = InspectShaderLanguage(source);
+        return info.HasVersionDirective() ? info.versionDirectiveEnd : 0;
     }
 
     bool IsExtensionAdvertised(MobileGL::GLExtension extension) {
@@ -261,9 +471,68 @@ namespace {
         ReplaceIdentifier(source, "GL_ARB_gpu_shader_int64", "MG_DISABLED_GL_ARB_gpu_shader_int64");
     }
 
+    // Rewrite the `packed` / `shared` block-packing qualifiers inside layout(...) declarations to
+    // `std140`. Desktop GL leaves the memory layout of such blocks to the implementation and the
+    // app must query member offsets; MobileGL's SPIR-V pipeline always lays uniform blocks out as
+    // std140 (glslang under a SPIR-V target rejects `packed`/`shared` outright and SPIRV-Cross has
+    // no other packing for UBOs), so std140 IS this implementation's chosen layout. Rewriting at
+    // the source level keeps the validation compile, the reflection the app queries, and the
+    // generated SPIR-V all agreeing on that choice. Both replacement tokens are 6 characters, so
+    // the rewrite is done in place.
+    void CoerceUniformBlockPackingToStd140(MobileGL::String& source) {
+        constexpr const char* layoutToken = "layout";
+        constexpr SizeT layoutLen = 6;
+
+        SizeT pos = 0;
+        while ((pos = source.find(layoutToken, pos)) != MobileGL::String::npos) {
+            const bool hasLeftBoundary = pos == 0 || !IsIdentifierChar(source[pos - 1]);
+            SizeT probe = pos + layoutLen;
+            const bool hasRightBoundary = probe >= source.size() || !IsIdentifierChar(source[probe]);
+            if (!hasLeftBoundary || !hasRightBoundary) {
+                pos = probe;
+                continue;
+            }
+
+            while (probe < source.size() && std::isspace(static_cast<unsigned char>(source[probe]))) {
+                probe++;
+            }
+            if (probe >= source.size() || source[probe] != '(') {
+                pos = probe;
+                continue;
+            }
+
+            // Scan the qualifier list; layout qualifier values may contain parenthesized
+            // constant expressions, so track nesting until the matching ')'.
+            SizeT cursor = probe + 1;
+            int depth = 1;
+            while (cursor < source.size() && depth > 0) {
+                const char ch = source[cursor];
+                if (ch == '(') {
+                    depth++;
+                } else if (ch == ')') {
+                    depth--;
+                } else if (IsIdentifierChar(ch) && (cursor == 0 || !IsIdentifierChar(source[cursor - 1]))) {
+                    SizeT identifierEnd = cursor;
+                    while (identifierEnd < source.size() && IsIdentifierChar(source[identifierEnd])) {
+                        identifierEnd++;
+                    }
+                    const SizeT identifierLen = identifierEnd - cursor;
+                    if (identifierLen == 6 && (source.compare(cursor, 6, "packed") == 0 ||
+                                               source.compare(cursor, 6, "shared") == 0)) {
+                        source.replace(cursor, 6, "std140");
+                    }
+                    cursor = identifierEnd;
+                    continue;
+                }
+                cursor++;
+            }
+            pos = cursor;
+        }
+    }
+
     void ModernizeLegacyGLSL(MobileGL::ShaderStage stage, MobileGL::String& source) {
         // Precision qualifiers (highp/mediump/lowp and default-precision statements) are legal and
-        // ignored in the forced "#version 460 core" profile, so glslang handles them natively.
+        // ignored in the normalized desktop core profiles, so glslang handles them natively.
 
         ReplaceIdentifier(source, "texture2D", "texture");
         ReplaceIdentifier(source, "texture2DProj", "textureProj");
@@ -308,6 +577,11 @@ namespace MobileGL {
     namespace MG_Util {
         namespace ShaderTranspiler {
             void PreprocessShaderSource(ShaderStage stage, String& source) {
+                // Normalize while the inspector's source span still refers to the untouched input. Later passes
+                // remove comments and directives, so any subsequent insertion re-inspects the current source.
+                const ShaderLanguageInfo originalLanguage = InspectShaderLanguage(source);
+                NormalizeVersionDirective(source, originalLanguage);
+
                 // remove multi-line comment
                 size_t commentStartPos = source.find("/*");
                 while (commentStartPos != String::npos) {
@@ -345,44 +619,8 @@ namespace MobileGL {
                     noperspectivePos = source.find(str_np);
                 }
 
-                // force #version
-                ShaderProfile profile = ShaderProfile::Core;
-                SizeT versionPos = source.find("#version");
-                SizeT lineEnd = source.find('\n', versionPos);
-
-                if (versionPos != String::npos) {
-                    String versionLine = source.substr(versionPos, lineEnd - versionPos);
-
-                    if (versionLine.find("ES") != String::npos)
-                        profile = ShaderProfile::ES;
-                    else if (versionLine.find("compatibility") != String::npos)
-                        profile = ShaderProfile::Compatibility;
-                    else
-                        profile = ShaderProfile::Core;
-                } else {
-                    profile = ShaderProfile::Core;
-                    source.insert(0, "#version 460 core\n");
-                    versionPos = 0;
-                    lineEnd = source.find('\n', versionPos);
-                }
-
-                SizeT firstLineEnd = lineEnd;
-
-                if (profile != ShaderProfile::ES) {
-                    constexpr const char* versionDirectiveCore = "#version 460 core\n";
-                    constexpr const char* versionDirectiveCompat = "#version 460 compatibility\n";
-
-                    const char* replacement =
-                        (profile == ShaderProfile::Compatibility) ? versionDirectiveCompat : versionDirectiveCore;
-
-                    if (firstLineEnd != String::npos) {
-                        source.replace(versionPos, firstLineEnd - versionPos + 1, replacement);
-                    } else {
-                        source = replacement;
-                    }
-                }
-
                 FilterUnsupportedGpuShaderInt64(source);
+                CoerceUniformBlockPackingToStd140(source);
 
                 // Some shader packs define helpers with built-in GLSL names such as round(), tanh(), or fma().
                 // These may pass OpenGL-style validation but fail when recompiled for Vulkan/SPIR-V generation.
@@ -395,6 +633,21 @@ namespace MobileGL {
                 InjectDepthRangeBuiltinShim(stage, source);
             }
 
+            Bool RetargetLegacyVersionDirectiveTo460(String& source) {
+                // Re-inspect rather than searching for the literal directive: it is not necessarily at
+                // offset 0 (a BOM or comments may precede it) and a commented-out "#version" elsewhere
+                // must not be mistaken for the real one.
+                const ShaderLanguageInfo info = InspectShaderLanguage(source);
+                if (!info.HasVersionDirective()) return false;
+                // Only the set NormalizeVersionDirective downgraded: desktop core below 400. ES and
+                // compatibility shaders keep whatever they declared.
+                if (info.profile != ShaderProfile::Core || info.version >= 400) return false;
+
+                source.replace(info.versionDirectiveStart, info.versionDirectiveEnd - info.versionDirectiveStart,
+                               "#version 460 core\n");
+                return true;
+            }
+
         } // namespace ShaderTranspiler
     } // namespace MG_Util
 } // namespace MobileGL

MobileGL/MG_Util/ShaderTranspiler/ShaderSourceProcessor.h

@@ -20,6 +20,14 @@ namespace MobileGL {
     namespace MG_Util {
         namespace ShaderTranspiler {
             void PreprocessShaderSource(ShaderStage stage, 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
+            // declaration. Exists so a shader that only parses under the laxer 460 rules - e.g. it
+            // uses 420-era syntax without the matching #extension line, which real drivers tend to
+            // accept - can be retried instead of failing to compile.
+            Bool RetargetLegacyVersionDirectiveTo460(String& source);
         } // namespace ShaderTranspiler
     } // namespace MG_Util
 } // namespace MobileGL

MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/StripUboMemberRelaxedPrecisionPass.cpp

@@ -0,0 +1,119 @@
+// MobileGL - MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/StripUboMemberRelaxedPrecisionPass.cpp
+// Copyright (c) 2025-2026 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
+// End of Source File Header
+
+#include "StripUboMemberRelaxedPrecisionPass.h"
+
+#include "spirv.hpp"
+#include "source/opt/def_use_manager.h"
+#include "source/opt/instruction.h"
+#include "source/opt/ir_context.h"
+#include "source/opt/module.h"
+
+#include <unordered_set>
+#include <vector>
+
+namespace MobileGL {
+    namespace MG_Util {
+        namespace ShaderTranspiler {
+            namespace {
+                using spvtools::opt::Instruction;
+                using spvtools::opt::IRContext;
+
+                // Marks `typeId` and every struct type reachable through its members
+                // (following arrays) for decoration stripping.
+                void CollectStructTypes(IRContext* context, uint32_t typeId,
+                                        std::unordered_set<uint32_t>& structTypeIds) {
+                    Instruction* typeInst = context->get_def_use_mgr()->GetDef(typeId);
+                    if (typeInst == nullptr) return;
+
+                    switch (typeInst->opcode()) {
+                    case spv::Op::OpTypeStruct: {
+                        if (!structTypeIds.insert(typeId).second) return; // already visited
+                        for (uint32_t member = 0; member < typeInst->NumInOperands(); ++member) {
+                            CollectStructTypes(context, typeInst->GetSingleWordInOperand(member), structTypeIds);
+                        }
+                        break;
+                    }
+                    case spv::Op::OpTypeArray:
+                    case spv::Op::OpTypeRuntimeArray:
+                        CollectStructTypes(context, typeInst->GetSingleWordInOperand(0), structTypeIds);
+                        break;
+                    default:
+                        break;
+                    }
+                }
+            } // namespace
+
+            spvtools::opt::Pass::Status StripUboMemberRelaxedPrecisionPass::Process() {
+                auto* irContext = context();
+                auto* defUseMgr = irContext->get_def_use_mgr();
+
+                // Uniform blocks: StorageClass Uniform variables whose pointee struct carries
+                // the Block decoration (BufferBlock/StorageBuffer SSBOs are left alone - they
+                // are not stage-matched by member precision in this pipeline's ESSL output).
+                std::unordered_set<uint32_t> blockStructIds;
+                for (Instruction& annotation : irContext->module()->annotations()) {
+                    if (annotation.opcode() != spv::Op::OpDecorate) continue;
+                    if (static_cast<spv::Decoration>(annotation.GetSingleWordInOperand(1)) != spv::Decoration::Block) {
+                        continue;
+                    }
+                    blockStructIds.insert(annotation.GetSingleWordInOperand(0));
+                }
+                if (blockStructIds.empty()) return Status::SuccessWithoutChange;
+
+                std::unordered_set<uint32_t> structTypeIds;
+                for (Instruction& variable : irContext->module()->types_values()) {
+                    if (variable.opcode() != spv::Op::OpVariable) continue;
+                    if (static_cast<spv::StorageClass>(variable.GetSingleWordInOperand(0)) !=
+                        spv::StorageClass::Uniform) {
+                        continue;
+                    }
+
+                    Instruction* pointerType = defUseMgr->GetDef(variable.type_id());
+                    if (pointerType == nullptr || pointerType->opcode() != spv::Op::OpTypePointer) continue;
+                    uint32_t pointeeId = pointerType->GetSingleWordInOperand(1);
+
+                    // Instance-arrayed blocks: unwrap the array around the block struct.
+                    Instruction* pointee = defUseMgr->GetDef(pointeeId);
+                    while (pointee != nullptr && (pointee->opcode() == spv::Op::OpTypeArray ||
+                                                  pointee->opcode() == spv::Op::OpTypeRuntimeArray)) {
+                        pointeeId = pointee->GetSingleWordInOperand(0);
+                        pointee = defUseMgr->GetDef(pointeeId);
+                    }
+                    if (pointee == nullptr || pointee->opcode() != spv::Op::OpTypeStruct) continue;
+                    if (blockStructIds.find(pointeeId) == blockStructIds.end()) continue;
+
+                    CollectStructTypes(irContext, pointeeId, structTypeIds);
+                }
+                if (structTypeIds.empty()) return Status::SuccessWithoutChange;
+
+                std::vector<Instruction*> decorationsToRemove;
+                for (Instruction& annotation : irContext->module()->annotations()) {
+                    if (annotation.opcode() != spv::Op::OpMemberDecorate) continue;
+                    if (static_cast<spv::Decoration>(annotation.GetSingleWordInOperand(2)) !=
+                        spv::Decoration::RelaxedPrecision) {
+                        continue;
+                    }
+                    if (structTypeIds.find(annotation.GetSingleWordInOperand(0)) == structTypeIds.end()) continue;
+                    decorationsToRemove.push_back(&annotation);
+                }
+                if (decorationsToRemove.empty()) return Status::SuccessWithoutChange;
+
+                for (Instruction* decoration : decorationsToRemove) {
+                    irContext->KillInst(decoration);
+                }
+                return Status::SuccessWithChange;
+            }
+
+            spvtools::Optimizer::PassToken
+            StripUboMemberRelaxedPrecisionPass::CreateStripUboMemberRelaxedPrecisionPass() {
+                return spvtools::Optimizer::PassToken(MakeUnique<StripUboMemberRelaxedPrecisionPass>());
+            }
+        } // namespace ShaderTranspiler
+    } // namespace MG_Util
+} // namespace MobileGL

MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/StripUboMemberRelaxedPrecisionPass.h

@@ -0,0 +1,44 @@
+// MobileGL - MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/StripUboMemberRelaxedPrecisionPass.h
+// Copyright (c) 2025-2026 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
+// End of Source File Header
+
+#pragma once
+#include "source/opt/pass.h"
+#include "spirv-tools/optimizer.hpp"
+
+#include <Includes.h>
+
+namespace MobileGL {
+    namespace MG_Util {
+        namespace ShaderTranspiler {
+            // Removes RelaxedPrecision member decorations from every struct type reachable
+            // from a uniform-block variable (the block struct itself and any structs nested
+            // in it through members or arrays).
+            //
+            // Rationale: ESSL requires matched uniform blocks to declare members with
+            // identical precision in every stage, but SPIRV-Cross prints a member's
+            // qualifier relative to the stage's DEFAULT precision (highp in the vertex
+            // stage, mediump in the fragment stage). A RelaxedPrecision member therefore
+            // comes out as an explicit "mediump" in the vertex shader but UNQUALIFIED in
+            // the fragment shader - and once ForceSupporterOutput swaps the fragment
+            // header to "precision highp float;", that unqualified member reads back as
+            // highp and the ES driver refuses to link ("definitions of uniform block ...
+            // do not match", GL CTS KHR-GL33.shaders.uniform_block struct sub-groups).
+            // Dropping the hint promotes the member to highp in BOTH stages, which is
+            // always conformant and matches the std140 data layout either way. Only meant
+            // for the DirectGLES transpile path - block member precision is a per-member
+            // hint with no layout effect, and no other emission behavior changes.
+            class StripUboMemberRelaxedPrecisionPass : public spvtools::opt::Pass {
+            public:
+                const char* name() const override { return "strip-ubo-member-relaxed-precision"; }
+                Status Process() override;
+
+                static spvtools::Optimizer::PassToken CreateStripUboMemberRelaxedPrecisionPass();
+            };
+        } // namespace ShaderTranspiler
+    } // namespace MG_Util
+} // namespace MobileGL