[Feat] (DirectVulkan): relax fragment precision where the bound formats allow it

- WIP, parked: measures 80.9 -> 94.8 fps on Adreno 650 / MC 26.2 (same scene, device cooled to 38-40C), but is NOT validated. Desktop GLSL carries no precision qualifiers, so every fragment value reaches the driver as fp32 while Adreno runs fp16 at twice the rate. - RelaxTextureDerivedPrecisionPass taints the values a fragment shader derives from built-in inputs and decorates everything else RelaxedPrecision. The taint direction matters: whitelisting outward from texture reads captures nothing, because MC multiplies every texel by an interpolated colour and a UBO value and one un-relaxed operand vetoes the expression - measured at 80.4 fps, i.e. no gain, both with and without varyings seeded. Precision-critical sources are few (gl_FragCoord cannot even hold a 3044-pixel x exactly), so tainting them and relaxing the rest is what actually pays. - SPIR-V cannot see the bound formats - sampler2D yields vec4 whether the texture is RGBA8 or RGBA32F - so the decision is made per draw and passed in as a compile option, the same shape ExplicitLod0Sampling already uses. RelaxedFragmentPrecision is only requested when every sampled texture and every colour attachment is an 8-bit-or-less normalized format, where fp16's 11-bit mantissa already carries the value exactly. Shaderpack HDR gbuffers, float data textures and 16-bit normalized targets therefore keep full precision, as do shaders that write gl_FragDepth or gl_SampleMask. - LocalMultiStoreElim runs first: glslang emits function-local variables, and a load can never be relaxed, so without SSA promotion the analysis dies at the first temporary. - WHY THIS IS PARKED: the retrace correctness gate never ran green. Every DirectVulkan retrace on Adreno 650 dies with DEVICE_LOST in UploadDirtyMipLevels on unmodified dev (pre-existing, device-gated), and on Adreno 830 - where the gate does pass on dev - minecraft-1.21.4-in-world times out at 900s with this change, which still needs explaining. Do not merge until that is understood and vanilla plus non-Photon shaderpack cases pass. (photon-v1.3b is broken on Adreno independently of this work.) - The /sdcard/MG/exp_relaxed_precision_all and exp_no_relaxed_precision file toggles are development scaffolding for A/B measurement; they must go before this ships.
[Fix] (DirectVulkan): bound image mutability so Adreno keeps UBWC compression
2026-09-07 19:58:32 +09:00 · 2026-07-29 09:02:00 -04:00 · 2026-07-29 07:07:50 -04:00 · 2026-07-29 03:26:57 -04:00 · 2026-07-28 21:06:57 -04:00 · 2026-07-28 11:54:50 -04:00
216 changed files with 27309 additions and 1694 deletions
@@ -9,17 +9,45 @@ fi
 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}"
+# Fixture mirrors, tried in order before falling back to Git LFS. Override the
+# whole list with MOBILEGL_TRACE_FIXTURE_MIRROR_BASES (whitespace separated);
+# MOBILEGL_TRACE_FIXTURE_MIRROR_BASE still works and is tried first.
+default_mirror_bases=(
+  "https://git.hit.moe/swung0x48/MobileGL/media/branch/dev/tools/trace_replay/fixtures"
+  "https://repo.miawa.cn/mgl/tools/trace_replay/fixtures"
+)
+if [ -n "${MOBILEGL_TRACE_FIXTURE_MIRROR_BASES:-}" ]; then
+  read -r -a mirror_bases <<< "${MOBILEGL_TRACE_FIXTURE_MIRROR_BASES}"
+else
+  mirror_bases=("${default_mirror_bases[@]}")
+fi
+if [ -n "${MOBILEGL_TRACE_FIXTURE_MIRROR_BASE:-}" ]; then
+  mirror_bases=("${MOBILEGL_TRACE_FIXTURE_MIRROR_BASE}" "${mirror_bases[@]}")
+fi
+# Optional bearer token for mirrors that require authentication (private Gitea).
+mirror_token="${MOBILEGL_TRACE_FIXTURE_MIRROR_TOKEN:-}"
+download_attempts="${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}" \
  --fixture-root "${fixture_dir}")"
-mapfile -t files <<< "${fixture_list}"
+# Strip CR so the script also works when python emits CRLF (Git Bash on Windows).
+mapfile -t files < <(printf '%s\n' "${fixture_list}" | tr -d '\r')

 include="$(IFS=,; echo "${files[*]}")"
 if [ "${case_name}" = "OpenRA" ]; then
@@ -34,34 +62,178 @@ if [ "${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
+  local curl_auth
+
+  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
+
+    curl_auth=()
+    if [ -n "${mirror_token}" ]; then
+      curl_auth=(--header "Authorization: token ${mirror_token}")
+    fi
+    if curl -L --fail --show-error --continue-at - "${curl_auth[@]}" --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 "${files[@]}"; do
    local name
    local url
+    local base
+    local fetched=0
    name="$(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"
+    for base in "${mirror_bases[@]}"; do
+      url="${base%/}/${name}"
+      echo "Fetching trace fixture from mirror: ${url}"
+      if fetch_file_from_mirror "${file}" "${url}"; then
+        fetched=1
+        break
+      fi
+      echo "Mirror did not serve ${name}; trying the next mirror" >&2
+    done
+    if [ "${fetched}" -ne 1 ]; then
      return 1
    fi
-    mv "${file}.tmp" "${file}"
  done
 }

 if fetch_from_mirror; then
  echo "Fetched trace fixture files for ${case_name} from mirror: ${include}"
 else
-  echo "Mirror fetch failed for ${case_name}; falling back to Git LFS: ${include}"
+  echo "All mirrors failed for ${case_name}; falling back to Git LFS: ${include}"
  git lfs install --local
  git lfs pull --include="${include}" --exclude=""
 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
@@ -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()
@@ -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
@@ -454,6 +455,15 @@ jobs:
          if [ '${{ matrix.backend }}' = 'DirectVulkan' ]; then
            export MOBILEGL_MAGMA_R11G11B10F_FALLBACK=1
          fi
+          # The blended depth-write quirk auto-enables only on Qualcomm, which no CI
+          # runner has, so force it on for the OIT case it exists to fix. ForceOn
+          # bypasses only the vendor gate, so this exercises the real strip on
+          # lavapipe. The Android AVD lane deliberately leaves it off, keeping the
+          # unstripped path covered for the same trace.
+          if [ '${{ matrix.backend }}' = 'DirectVulkan' ] \
+             && [ '${{ matrix.case }}' = 'improved-transparency-minecraft-26.3' ]; then
+            export MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE=1
+          fi
          ctest -V --no-tests=error -R '^MobileGLTraceReplay\.${{ matrix.case }}\.${{ matrix.backend }}$'

      - name: Upload actual image
@@ -188,8 +188,12 @@ set(SOURCE_FILES
    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/EliminateFloatEqualsZeroPass.cpp
    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/RenameSamplerFunctionParameterPass.cpp
    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/DecomposeWorkgroupVec3Pass.cpp
+    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/DecoratePositionInvariantPass.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/ShaderTranspiler/SpirvPasses/StripNoPerspectivePass.cpp
+    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/EmulateNoPerspectivePass.cpp

    MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.cpp
    MobileGL/MG_Util/BackendLoaders/Vulkan/Loader.cpp
@@ -299,6 +303,13 @@ if (ANDROID)
    )
 endif()

+if (WIN32)
+    list(APPEND SOURCE_FILES
+        MobileGL/MG_Impl/WGLImpl/WGLImpl.cpp
+        MobileGL/MG_Impl/WGLImpl/Exporting/Definitions.cpp
+    )
+endif()
+
 set(MOBILEGL_LINK_LIBRARIES
        glslang::glslang
        spirv-cross-c
@@ -327,10 +338,18 @@ set(MOBILEGL_INCLUDE_DIR
        ${SPIRV-Headers_SOURCE_DIR}/include
 )

-add_library(${CMAKE_PROJECT_NAME} SHARED 
+add_library(${CMAKE_PROJECT_NAME} SHARED
    ${SOURCE_FILES}
 )

+if (WIN32)
+    # The wgl* entry points are exported via .def (see the comment in wgl.def);
+    # only the shared library links it.
+    target_sources(${CMAKE_PROJECT_NAME} PRIVATE
+        MobileGL/MG_Impl/WGLImpl/Exporting/wgl.def
+    )
+endif()
+
 if (CMAKE_BUILD_TYPE STREQUAL "Debug")
    set_target_properties(${CMAKE_PROJECT_NAME} PROPERTIES
        C_VISIBILITY_PRESET default
@@ -373,6 +392,18 @@ if(UNIX AND NOT APPLE AND NOT ANDROID)
    endforeach()
 endif()

+if(WIN32)
+    # Drop-in for the classic GL loader path: a copy named opengl32.dll placed
+    # next to a host executable is what LoadLibrary("opengl32.dll") and gdi32's
+    # pixel-format forwarding will resolve.
+    add_custom_command(TARGET ${CMAKE_PROJECT_NAME} POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different
+                "$<TARGET_FILE:${CMAKE_PROJECT_NAME}>"
+                "$<TARGET_FILE_DIR:${CMAKE_PROJECT_NAME}>/opengl32.dll"
+        COMMENT "Creating opengl32.dll drop-in copy"
+    )
+endif()
+
 if(NOT ANDROID)
    add_library(${CMAKE_PROJECT_NAME}_s STATIC 
        ${SOURCE_FILES}
@@ -424,8 +455,21 @@ if (ANDROID)
 endif()

 if (APPLE AND NOT MOBILEGL_IOS)
+    # MobileGL statically embeds glslang, SPIRV-Tools, and SPIRV-Cross. When
+    # this dylib is injected with DYLD_INSERT_LIBRARIES, exporting those C++
+    # symbols interposes incompatible copies embedded by host libraries such
+    # as shaderc. Keep only the public GL/EGL/CGL loader surface globally
+    # visible; GetProcAddress can still return pointers to hidden internals.
+    set(MOBILEGL_MACOS_EXPORTED_SYMBOLS
+        "${CMAKE_CURRENT_SOURCE_DIR}/MobileGL/MG_Impl/DyldInterpose/ExportedSymbols.txt")
+    target_link_options(${CMAKE_PROJECT_NAME} PRIVATE
+        "LINKER:-exported_symbols_list,${MOBILEGL_MACOS_EXPORTED_SYMBOLS}")
+    set_property(TARGET ${CMAKE_PROJECT_NAME} APPEND PROPERTY
+        LINK_DEPENDS "${MOBILEGL_MACOS_EXPORTED_SYMBOLS}")
+
    target_link_libraries(${CMAKE_PROJECT_NAME} PUBLIC
        "-framework Cocoa"
+        "-framework CoreVideo"
        "-framework QuartzCore"
        "-framework Foundation"
        "-framework OpenGL"
@@ -433,6 +477,7 @@ if (APPLE AND NOT MOBILEGL_IOS)
    if(TARGET ${CMAKE_PROJECT_NAME}_s)
        target_link_libraries(${CMAKE_PROJECT_NAME}_s PUBLIC
            "-framework Cocoa"
+            "-framework CoreVideo"
            "-framework QuartzCore"
            "-framework Foundation"
            "-framework OpenGL"
@@ -475,6 +520,15 @@ 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 +538,6 @@ if (NOT ANDROID)
    endif()

    if (MOBILEGL_BUILD_TRACE_REPLAY)
-        enable_testing()
        add_subdirectory(tools/trace_replay)
    endif()
 endif()
@@ -20,6 +20,15 @@ namespace MobileGL::MG_Config {

    extern BackendType ActiveBackendType;

+    // Tri-state override for device-specific quirks: Auto lets the detected device decide,
+    // ForceOn/ForceOff bypass the detection in either direction. ForceOn only bypasses the
+    // device gate - each quirk keeps its structural safety checks.
+    enum class QuirkOverride : Uint8 {
+        Auto = 0,
+        ForceOn,
+        ForceOff,
+    };
+
    // Feature toggles parsed once from environment variables in MG_ConfigLoader::Init()
    // (ConfigLoader.cpp), before the accepted-env map is destroyed. All Bool fields share
    // one truthy rule: the variable is set, non-empty, not "0", and not "false"
@@ -61,6 +70,27 @@ 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;
+        // MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN: overrides the shader-source quirk that
+        // rewrites the recognized workgroup prefix-scan template on Qualcomm devices with
+        // subgroups wider than 32 lanes (see ShaderSourceProcessor's quirk registry).
+        QuirkOverride SubgroupPrefixScanQuirk = QuirkOverride::Auto;
+        // MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE: overrides the DirectVulkan quirk that
+        // strips depth writes from accumulation-blended pipelines (MIN/MAX or additive
+        // ONE+ONE - the multi-pass depth-equality signature) on drivers without
+        // cross-pipeline vertex position invariance. Sorted-transparency "over" blends,
+        // gl_FragDepth writers, and fully color-masked attachments are exempt (see
+        // PipelineFactory::ShouldSuppressDepthWrite). Auto detects Qualcomm.
+        QuirkOverride MagmaDisableBlendedDepthWriteQuirk = QuirkOverride::Auto;
+        // MOBILEGL_DISABLE_ROBUST_BUFFER_ACCESS: leave the Vulkan robustBufferAccess device
+        // feature off. It is enabled by default to match GL's defined out-of-range fetch
+        // behavior; this escape hatch exists to measure or dodge its GPU cost on a device.
+        Bool DisableRobustBufferAccess = false;
    };
    extern FeaturesTable Features;
 } // namespace MobileGL::MG_Config
@@ -86,6 +86,17 @@ namespace MobileGL::MG_ConfigLoader {
        return it != acceptedEnvVariablesMap->end() && IsTruthyValue(it->second);
    }

+    // Quirk overrides are tri-state: an unset variable keeps device auto-detection, a truthy
+    // value forces the quirk on, anything else set ("0", "false", "") forces it off.
+    inline MG_Config::QuirkOverride QueryEnvQuirkOverride(const String& key) {
+        auto it = acceptedEnvVariablesMap->find(key);
+        if (it == acceptedEnvVariablesMap->end()) {
+            return MG_Config::QuirkOverride::Auto;
+        }
+        return IsTruthyValue(it->second) ? MG_Config::QuirkOverride::ForceOn
+                                         : MG_Config::QuirkOverride::ForceOff;
+    }
+
    inline Uint32 QueryEnvUint32(const String& key, Uint32 defaultValue, Uint32 minValue, Uint32 maxValue) {
        auto it = acceptedEnvVariablesMap->find(key);
        if (it == acceptedEnvVariablesMap->end()) {
@@ -122,6 +133,11 @@ 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");
+        features.SubgroupPrefixScanQuirk = QueryEnvQuirkOverride("MOBILEGL_QUIRK_SUBGROUP_PREFIX_SCAN");
+        features.MagmaDisableBlendedDepthWriteQuirk =
+            QueryEnvQuirkOverride("MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE");
+        features.DisableRobustBufferAccess = QueryEnvFlag("MOBILEGL_DISABLE_ROBUST_BUFFER_ACCESS");
    }

    inline void InitBackendType() {
@@ -34,6 +34,7 @@
 #define MOBILEGL_EGL_API MOBILEGL_API
 #define MOBILEGL_CGL_API MOBILEGL_API
 #define MOBILEGL_NSOPENGL_API MOBILEGL_API
+#define MOBILEGL_WGL_API MOBILEGL_API

 // ====================== MobileGL configurations ======================= //
 #ifndef MOBILEGL_LOG_ACTIVE_LEVEL
@@ -14,7 +14,13 @@ namespace MobileGL {
    } // namespace MG_Config

    namespace MG_Backend {
-        UniquePtr<BackendObject> pActiveBackendObject;
+        // Leak-at-exit storage: the UniquePtr itself lives on the heap and is
+        // never destroyed by the runtime, so process exit runs no backend
+        // destructors (static destruction order across TUs is undefined).
+        // Deterministic teardown happens inside the EGL lifecycle instead:
+        // the last eglTerminate calls MobileGL::Destroy(), which .reset()s
+        // these singletons while the process is still healthy.
+        UniquePtr<BackendObject>& pActiveBackendObject = *new UniquePtr<BackendObject>();
        GlobalBackendFunctionsTable gBackendFunctionsTable;
    } // namespace MG_Backend
 } // namespace MobileGL
@@ -9,14 +9,25 @@
 #include "Init.h"
 #include "Config.h"
 #include <MG_Backend/BackendObjects.h>
+#include <MG_Backend/DirectVulkan/DirectVulkan.h>
 #include <MG_State/GLState/Core.h>
 #include <MG_State/EGLState/Core.h>
 #include <MG_Impl/GLImpl/Texture/ProxyTexture.h>
 #include <MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h>
+#include <MG_Impl/GLImpl/Sync/GL_Sync.h>
+
+#include <atomic>
+#include <mutex>

 namespace MobileGL {
    namespace {
-        Bool g_isInitialized = false;
+        std::atomic<Bool> g_isInitialized = false;
+        thread_local Bool tl_initializing = false;
+
+        std::mutex& InitMutex() {
+            static std::mutex mutex;
+            return mutex;
+        }

        void DestroyImpl(Bool logLifecycle) {
            if (!g_isInitialized) {
@@ -27,6 +38,12 @@ namespace MobileGL {
                MGLOG_I("MobileGL closing...");
            }
            glslang::FinalizeProcess();
+            // GL syncs die with their contexts, and every context is gone by the
+            // time full teardown runs: drain the live-sync registry while the
+            // backend function table can still release the backend handles (and
+            // before a re-initialized library could pair them with the wrong
+            // backend's DeleteSync).
+            MG_Impl::GLImpl::DestroyAllSyncObjects();
            MG_Backend::pActiveBackendObject.reset();
            MG_State::pGLContext.reset();
            MG_State::pEGLContext.reset();
@@ -64,40 +81,37 @@ namespace MobileGL {
        MGLOG_I("MobileGL initialized");
    }

+    void EnsureInitialized() {
+        if (g_isInitialized.load(std::memory_order_acquire)) {
+            return;
+        }
+        // Re-entrant call while this thread is already inside Initialize()
+        // (e.g. an init step routing back through a public entry point).
+        if (tl_initializing) {
+            return;
+        }
+        const std::lock_guard<std::mutex> lock(InitMutex());
+        if (g_isInitialized.load(std::memory_order_acquire)) {
+            return;
+        }
+        tl_initializing = true;
+        Initialize();
+        tl_initializing = false;
+    }
+
    void Destroy() {
        DestroyImpl(true);
    }

-#if defined(__linux__) || defined(__APPLE__)
-    __attribute__((constructor)) static void AutoInit() {
-        Initialize();
-    }
-
-    __attribute__((destructor)) static void AutoDestroy() {
-        if (MG_Config::Features.TraceSkipAutodestroy) {
-            return;
-        }
-#if defined(__APPLE__)
-        // macOS injected dylibs can run destructors after logging/backend static state is already torn down.
-        return;
-#else
-        DestroyImpl(false);
-#endif
-    }
-#endif
-
-#ifdef _WIN32
-    BOOL WINAPI DllMain(HMODULE hModule, DWORD ul_reason_for_call, LPVOID lpReserved) {
-        switch (ul_reason_for_call) {
-        case DLL_PROCESS_ATTACH:
-            Initialize();
-            break;
-
-        case DLL_PROCESS_DETACH:
-            Destroy();
-            break;
-        }
-        return TRUE;
-    }
-#endif
+    // MobileGL's lifecycle is owned entirely by the host-API layers
+    // (EGL/WGL/CGL): initialization happens lazily on the first entry point
+    // via EnsureInitialized(), and full teardown happens deterministically
+    // when the last EGL display is terminated with nothing current (EGLImpl
+    // calls Destroy()). There is intentionally no backend-initializing static
+    // constructor, no static destructor, and no DllMain: the global singletons
+    // use leak-at-exit storage (see GlobalObjects.cpp), so a process that exits
+    // without eglTerminate simply leaks them to the OS instead of running
+    // backend destructors during static teardown. macOS has a lightweight
+    // dyld constructor that installs NSOpenGL dispatch hooks only; full backend
+    // initialization still enters here from the first hooked CGL context.
 } // namespace MobileGL
@@ -11,6 +11,13 @@

 namespace MobileGL {
    void Initialize();
+    // Thread-safe, idempotent, and re-entrant wrapper around Initialize().
+    // Host layers (EGL/WGL/CGL entry points) call this lazily on first use so
+    // full backend initialization never depends on ELF/DLL static constructors,
+    // and so a fresh init can follow a full Destroy() (e.g. after the last
+    // eglTerminate). The macOS dyld bootstrap installs only lightweight
+    // NSOpenGL method hooks.
+    void EnsureInitialized();
    void Destroy();

    namespace MG_Util::Debug {
@@ -230,8 +230,26 @@ namespace MobileGL {
            void (*SetSwapInterval)(Int interval);
        };

+        // Coarse GPU vendor identity for gating device-specific quirks. Detected from the
+        // Vulkan physical-device vendorID or the GLES GL_VENDOR/GL_RENDERER strings; stays
+        // Unknown when detection is inconclusive, in which case auto-gated quirks stay off.
+        enum class GpuVendorKind : Uint8 {
+            Unknown = 0,
+            Qualcomm,
+            Arm,
+            Nvidia,
+            Amd,
+            Intel,
+            ImgTec,
+            // Software rasterizers (llvmpipe/lavapipe, SwiftShader).
+            Software,
+        };
+
        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;
@@ -269,6 +287,9 @@ namespace MobileGL {
            Int MaxUniformBlockSize = 16384;
            Int MaxImageUnits = 8;
            Int MaxCombinedImageUniforms = 8;
+            Int MaxVertexImageUniforms = 0;
+            Int MaxGeometryImageUniforms = 0;
+            Int MaxFragmentImageUniforms = 8;
            Int MaxComputeImageUniforms = 8;
            Int MaxDrawBuffers = 8;
            Int MaxColorAttachments = 8;
@@ -285,13 +306,15 @@ namespace MobileGL {
            Uint32 SubgroupSupportedStages = 0;
            Uint32 SubgroupSupportedFeatures = 0;
            Bool SubgroupQuadOperationsInAllStages = false;
+            GpuVendorKind GpuVendor = GpuVendorKind::Unknown;
        };

        enum class WindowBackend {
            Android,
            X11,
            MetalLayer,
-            // TODO: Wayland, Windows, etc.
+            Win32, // Handle is an HWND
+            // TODO: Wayland, etc.
            WindowBackendCount,
            Unknown = -1
        };
@@ -13,6 +13,6 @@
 #include "DirectVulkan/BackendObject_DirectVulkan.h"

 namespace MobileGL::MG_Backend {
-    extern UniquePtr<BackendObject> pActiveBackendObject;
+    extern UniquePtr<BackendObject>& pActiveBackendObject;
    extern GlobalBackendFunctionsTable gBackendFunctionsTable;
 } // namespace MobileGL::MG_Backend
@@ -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());
@@ -700,9 +701,10 @@ namespace MobileGL::MG_Backend::DirectGLES {

        if ((handle.Backend != WindowBackend::Android &&
             handle.Backend != WindowBackend::X11 &&
-             handle.Backend != WindowBackend::MetalLayer) ||
+             handle.Backend != WindowBackend::MetalLayer &&
+             handle.Backend != WindowBackend::Win32) ||
            !handle.Handle) {
-            MGLOG_E("DirectGLES backend only supports Android, X11, and CAMetalLayer native windows");
+            MGLOG_E("DirectGLES backend only supports Android, X11, CAMetalLayer, and Win32 native windows");
            return false;
        }

@@ -818,14 +820,14 @@ 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,
                                          E_GL_ARB_program_interface_query, E_GL_ARB_framebuffer_object,
                                          E_GL_EXT_framebuffer_object, E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage,
                                          E_GL_ARB_texture_storage, E_GL_ARB_texture_storage_multisample,
-                                          E_GL_ARB_direct_state_access,
+                                          E_GL_ARB_clear_texture, E_GL_ARB_direct_state_access,
                                          E_GL_ARB_multi_draw_indirect, E_GL_ARB_indirect_parameters,
                                          E_GL_ARB_shader_draw_parameters, E_GL_ARB_gpu_shader5, E_GL_ARB_multi_bind,
                                          E_GL_ARB_shading_language_420pack, E_GL_ARB_vertex_attrib_binding,
@@ -836,6 +838,14 @@ 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;
    }

@@ -938,8 +948,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
        return m_dynamicParameters;
    }

+    void BackendObject_DirectGLES::ApplyGLESCapabilitiesForTesting(
+        const MG_External::GLESCapabilities& capabilities) {
+        m_GLESCapabilities = capabilities;
+        UpdateDynamicBackendParameters();
+    }
+
    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;
@@ -993,9 +1010,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
        m_dynamicParameters.MaxUniformBlockSize = m_GLESCapabilities.MaxUniformBlockSize;
        const Int maxSupportedTextureUnits =
            static_cast<Int>(MG_State::GLState::TextureState::MAX_TEXTURE_IMAGE_UNITS);
-        m_dynamicParameters.MaxImageUnits = std::min(m_GLESCapabilities.MaxImageUnits, maxSupportedTextureUnits);
-        m_dynamicParameters.MaxCombinedImageUniforms = m_GLESCapabilities.MaxCombinedImageUniforms;
-        m_dynamicParameters.MaxComputeImageUniforms = m_GLESCapabilities.MaxComputeImageUniforms;
+        m_dynamicParameters.MaxImageUnits =
+            std::max(std::min(m_GLESCapabilities.MaxImageUnits, maxSupportedTextureUnits), 0);
+        m_dynamicParameters.MaxCombinedImageUniforms = std::max(m_GLESCapabilities.MaxCombinedImageUniforms, 0);
+        const auto clampStageImageUniforms = [this](Int stageLimit) {
+            return std::min({std::max(stageLimit, 0), m_dynamicParameters.MaxImageUnits,
+                             m_dynamicParameters.MaxCombinedImageUniforms});
+        };
+        m_dynamicParameters.MaxVertexImageUniforms =
+            clampStageImageUniforms(m_GLESCapabilities.MaxVertexImageUniforms);
+        m_dynamicParameters.MaxGeometryImageUniforms =
+            clampStageImageUniforms(m_GLESCapabilities.MaxGeometryImageUniforms);
+        m_dynamicParameters.MaxFragmentImageUniforms =
+            clampStageImageUniforms(m_GLESCapabilities.MaxFragmentImageUniforms);
+        m_dynamicParameters.MaxComputeImageUniforms =
+            clampStageImageUniforms(m_GLESCapabilities.MaxComputeImageUniforms);
        m_dynamicParameters.MaxDrawBuffers = m_GLESCapabilities.MaxDrawBuffers;
        m_dynamicParameters.MaxColorAttachments = m_GLESCapabilities.MaxColorAttachments;
        m_dynamicParameters.MaxClipDistances = m_GLESCapabilities.MaxClipDistances;
@@ -1007,6 +1036,32 @@ namespace MobileGL::MG_Backend::DirectGLES {
        m_dynamicParameters.ViewportSubpixelBits = m_GLESCapabilities.ViewportSubpixelBits;
        m_dynamicParameters.SupportsWideLines =
            m_GLESCapabilities.AliasedLineWidthRangeMax > 1.0f || m_GLESCapabilities.SmoothLineWidthRangeMax > 1.0f;
+
+        const auto containsAny = [](const String& haystack, std::initializer_list<const char*> needles) {
+            return std::any_of(needles.begin(), needles.end(), [&](const char* needle) {
+                return haystack.find(needle) != String::npos;
+            });
+        };
+        const String vendorAndRenderer =
+            m_GLESCapabilities.GLESVendorString + " " + m_GLESCapabilities.GLESRendererString;
+        if (containsAny(vendorAndRenderer, {"llvmpipe", "SwiftShader", "softpipe"})) {
+            // Check software rasterizers first: ANGLE-on-llvmpipe reports both.
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Software;
+        } else if (containsAny(vendorAndRenderer, {"Qualcomm", "Adreno"})) {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Qualcomm;
+        } else if (containsAny(vendorAndRenderer, {"Mali", "ARM"})) {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Arm;
+        } else if (containsAny(vendorAndRenderer, {"NVIDIA"})) {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Nvidia;
+        } else if (containsAny(vendorAndRenderer, {"AMD", "Radeon"})) {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Amd;
+        } else if (containsAny(vendorAndRenderer, {"Intel"})) {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Intel;
+        } else if (containsAny(vendorAndRenderer, {"Imagination", "PowerVR"})) {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::ImgTec;
+        } else {
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Unknown;
+        }
    }

    const MG_External::GLESFunctionsTable& BackendObject_DirectGLES::GetGLESFunctions() const {
@@ -41,6 +41,7 @@ namespace MobileGL::MG_Backend::DirectGLES {

        const MG_External::GLESFunctionsTable& GetGLESFunctions() const;
        const MG_External::EGLFunctionsTable& GetEGLFunctions() const;
+        void ApplyGLESCapabilitiesForTesting(const MG_External::GLESCapabilities& capabilities);

    private:
        void UpdateDynamicBackendParameters();
@@ -66,9 +67,9 @@ 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
@@ -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);
@@ -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);
@@ -177,6 +178,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
        // glBindBuffer with a redundant-bind cache for GL_ARRAY_BUFFER.
        void BindBufferId(GLenum target, Uint id);
        void InvalidateArrayBufferBindingCache();
+        // Redundant-bind caches for the driver-level GL_PIXEL_PACK/UNPACK_BUFFER
+        // bindings. Every backend readback (glReadPixels / pack-PBO map) and pixel
+        // upload site routes its binding through these so the shadow always matches
+        // the driver; the resting state between operations is 0, which keeps any
+        // path that implicitly assumes "no PBO bound" correct. Scrubbed when a
+        // buffer id is deleted/pooled (GL resets a deleted buffer's bindings to 0,
+        // and a recycled name matching the shadow would false-skip the rebind) and
+        // invalidated on MakeCurrent (context may reset).
+        void BindPixelPackBufferId(Uint id);
+        void BindPixelUnpackBufferId(Uint id);
+        void InvalidatePixelBufferBindingCaches();
+        // A GL buffer id is being deleted by code outside BufferImpl (e.g. the VAO
+        // client-attribute staging buffers): scrub every buffer-binding shadow that
+        // could false-skip when the name is recycled.
+        void NoteBufferIdDeleted(Uint id);
        // Redundant-bind cache for INDEXED buffer bindings (glBindBufferBase/Range on
        // GL_UNIFORM_BUFFER / GL_SHADER_STORAGE_BUFFER): skips the GL call when the
        // (id, range) already at that index matches, like the array-buffer/texture/
@@ -254,10 +270,48 @@ 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) {
@@ -293,6 +347,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
        class BackendTextureObject {
        public:
            BackendTextureObject();
+            // Deletes the GL texture (frontend glDeleteTextures used to leak every
+            // backend id for the context lifetime) and scrubs the binding/scratch-FBO
+            // shadows so a recycled name or heap address cannot false-skip a rebind.
+            ~BackendTextureObject();
+            BackendTextureObject(const BackendTextureObject&) = delete;
+            BackendTextureObject& operator=(const BackendTextureObject&) = delete;
            void SyncMipmapsToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
            void SyncBuiltinSamplerToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
            void SyncTextureParamsToBackend(const SharedPtr<MG_State::GLState::ITextureObject>& stateTextureObject);
@@ -304,6 +364,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
            void RecreateBackendTexture();

            Uint m_backendTextureId = 0;
+            // ES context generation the id was created under; a dtor running after
+            // that context died must not delete a foreign (recycled) name.
+            Uint m_contextGeneration = 0;
            Bool m_isInitialized = false;
            Bool m_imageBindableStorageRequired = false;
            Bool m_backendStorageImmutable = false;
@@ -328,6 +391,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     MG_State::GLState::TextureState::MAX_TEXTURE_IMAGE_UNITS>
            g_boundTexturesCache;
        extern Uint g_activeTextureUnit;
+        // Bumped when the backend ES context is destroyed; texture ids stamped with
+        // an older generation belong to a dead context and must not be deleted.
+        extern Uint g_textureContextGeneration;
    } // namespace TextureImpl

    namespace FramebufferImpl {
@@ -336,6 +402,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
            BackendFramebufferObject();
            void SyncToBackend(const SharedPtr<MG_State::GLState::FramebufferObject>& stateFBOObject,
                               FramebufferTarget asTarget);
+            // Apply only this FBO's read buffer (glReadBuffer) to the backend. Split out so it can
+            // still run when SyncCurrentFBO skips the READ-target sync because the same GL FBO is
+            // bound as both draw and read (otherwise glReadBuffer changes would be silently dropped).
+            void SyncReadBufferToBackend(const SharedPtr<MG_State::GLState::FramebufferObject>& stateFBOObject);
            void InvalidateSyncedState();
            Uint GetBackendFramebufferId() const { return m_backendFBOId; }
            void Bind(FramebufferTarget target) const;
@@ -369,8 +439,105 @@ 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;
+
+        // Driver-level READ/DRAW framebuffer-binding shadow. Every backend
+        // glBindFramebuffer routes through BindFramebufferId so scoped helpers can
+        // save/restore the current binding without a glGetIntegerv round-trip (that
+        // query forces a driver pipeline sync) and so redundant rebinds no-op.
+        // Starts unknown; the first CurrentFramebufferBinding() query pins it from
+        // the driver once. Invalidated on MakeCurrent (context may reset).
+        // GL_FRAMEBUFFER binds both targets.
+        void BindFramebufferId(GLenum fbTarget, Uint id);
+        Uint CurrentFramebufferBinding(FramebufferTarget target);
+        void InvalidateFramebufferBindingCache();
    } // namespace FramebufferImpl

+    // Shared scratch framebuffers for the readback/copy/blit emulation paths, with a
+    // driver-side attachment shadow: repeated uses skip redundant detach/attach GL
+    // calls, and an attachment left by one use (e.g. a depth copy's DEPTH_STENCIL
+    // texture) is detached exactly when a later use of another aspect would
+    // otherwise inherit it (stale cross-aspect attachments made the shared temp FBO
+    // incomplete and silently degraded later readbacks).
+    namespace ScratchFBOImpl {
+        struct ScratchFramebuffer {
+            Uint id = 0;
+            // false => attachment state unknown; scrub every point on next use.
+            // A fresh FBO starts with nothing attached, so creation sets it true.
+            Bool attachmentsKnown = false;
+            Uint colorTex = 0;
+            GLenum colorTarget = 0;
+            GLint colorLevel = 0;
+            GLint colorLayer = -1; // >= 0 => attached via glFramebufferTextureLayer
+            Uint depthTex = 0;
+            GLenum depthTarget = 0;
+            GLint depthLevel = 0;
+            Bool depthHasStencil = false;
+            // Per-FBO read/draw buffer state (0 = unknown, set on first use).
+            GLenum readBuffer = 0;
+            GLenum drawBuffer = 0;
+        };
+        ScratchFramebuffer& TempFramebuffer();     // GetTexImage READ / CopyTex*Image2D depth DRAW
+        ScratchFramebuffer& BlitReadFramebuffer(); // texture-to-texture blit source
+        ScratchFramebuffer& BlitDrawFramebuffer(); // texture-to-texture blit destination
+        // Returns the GL id, generating it if needed (requires a current ES context).
+        Uint EnsureId(ScratchFramebuffer& fb);
+        // The fb must currently be bound at fbTarget (glReadBuffer/glDrawBuffers
+        // target the READ/DRAW binding respectively). Each Ensure* performs the
+        // minimal detach/attach set and keeps the shadow in sync; a failed attach
+        // records the point as detached so the completeness check fails instead of
+        // silently reading a stale attachment.
+        void EnsureColorAttachment2D(ScratchFramebuffer& fb, GLenum fbTarget, Uint tex, GLenum texTarget, GLint level);
+        void EnsureColorAttachmentLayer(ScratchFramebuffer& fb, GLenum fbTarget, Uint tex, GLint level, GLint layer);
+        void EnsureDepthAttachment2D(ScratchFramebuffer& fb, GLenum fbTarget, Uint tex, GLenum texTarget, GLint level,
+                                     Bool withStencil);
+        void EnsureNoColorAttachment(ScratchFramebuffer& fb, GLenum fbTarget);
+        void EnsureNoDepthAttachment(ScratchFramebuffer& fb, GLenum fbTarget);
+        void EnsureReadBuffer(ScratchFramebuffer& fb, GLenum readBuffer);
+        void EnsureDrawBuffer(ScratchFramebuffer& fb, GLenum drawBuffer);
+        // A 1x1 RGBA8-renderbuffer-complete FBO (GenerateMipmap needs a complete
+        // binding while respecifying texture storage). Attachment is set once at
+        // creation and never changes.
+        Uint EnsureCompleteTinyFramebufferId();
+        // A backend texture id is being deleted or respecified: a scratch FBO still
+        // referencing it would hold a dangling attachment (ES only auto-detaches
+        // from the *bound* framebuffer), and a recycled name could false-skip a
+        // re-attach; force a full scrub on next use.
+        void NoteTextureIdDeleted(Uint textureId);
+        // The ES context (and the scratch FBO ids with it) is going away.
+        void OnBackendContextDestroyed();
+    } // namespace ScratchFBOImpl
+
+    // Driver-level GL_PACK_* pixel-store shadow, the readback-side sibling of the
+    // upload path's ScopedDefaultUnpackState (Managers.cpp): the backend PACK state
+    // is written ONLY through ApplyPackState, so scoped helpers can save/restore it
+    // from the shadow instead of glGetIntegerv (which forces a driver pipeline
+    // sync), and redundant glPixelStorei calls no-op. The first Apply/Current call
+    // pins the driver to the shadow by writing all fields once. Invalidated on
+    // MakeCurrent (context may reset). PACK_IMAGE_HEIGHT/SKIP_IMAGES/SWAP_BYTES/
+    // LSB_FIRST have no ES equivalents; readbacks honor them on the CPU from the
+    // frontend context state instead.
+    namespace PixelStoreImpl {
+        struct PackState {
+            GLint Alignment = 4;
+            GLint RowLength = 0;
+            GLint SkipRows = 0;
+            GLint SkipPixels = 0;
+            Bool operator==(const PackState& o) const {
+                return Alignment == o.Alignment && RowLength == o.RowLength && SkipRows == o.SkipRows &&
+                       SkipPixels == o.SkipPixels;
+            }
+        };
+        void ApplyPackState(const PackState& desired);
+        PackState CurrentPackState();
+        void InvalidatePackStateCache();
+    } // namespace PixelStoreImpl
+
    // Image uniforms take their unit from the layout(binding=N) qualifier baked into
    // the transpiled ESSL; unlike samplers they must not (and in ES cannot) be
    // assigned through glUniform1i.
@@ -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 +452,407 @@ 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;
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        static SizeT AlignReadbackRow(SizeT rowBytes, Int alignment) {
+            const SizeT align = alignment > 0 ? static_cast<SizeT>(alignment) : 1;
+            return (rowBytes + align - 1) / align * align;
+        }
+
+    // Repacks wide RGBA(_INTEGER) rows into the client's (format, type) layout, honoring the
+        // client-side PACK parameters and the bound pixel-pack buffer. `wide` holds
+        // `sliceHeight * sliceCount` rows of `width` texels (slice-major, tightly stacked),
+        // 4 components x GetReadbackComponentSize(wideType) bytes each.
+        // applyPackImageParams: GL_PACK_IMAGE_HEIGHT / GL_PACK_SKIP_IMAGES apply only to GetTexImage
+        // of 3D/array images; ReadPixels and 2D GetTexImage ignore them (GL 3.3 sections 4.3.1, 6.1.4).
+        // Per the GL addressing rules, slice k row j lands at
+        // SKIP_IMAGES*imageStride + SKIP_ROWS*rowStride + SKIP_PIXELS*pixelBytes
+        //   + k*imageStride + j*rowStride, with imageStride = max(IMAGE_HEIGHT, sliceHeight)*rowStride.
+        Bool StoreWideRowsToClient(const Uint8* wide, GLenum wideType, GLsizei width, GLsizei sliceHeight,
+                                   GLsizei sliceCount, const ReadbackChannelMapping& mapping, GLenum type,
+                                   void* pixels, Bool applyPackImageParams) {
+        const SizeT dstPixelBytes = GetReadbackDstPixelSize(mapping, type);
+        if (dstPixelBytes == 0) {
+            return false;
+        }
+        PackedReadbackLayout packedLayout{};
+        const Bool isPackedType = GetPackedReadbackLayout(type, packedLayout);
+        const SizeT dstComponentSize = GetReadbackComponentSize(type);
+
+        const auto& pixelPackBufferObject =
+            MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::PixelPack).GetBoundObject();
+
+        // Destination layout is computed from the client-side PACK parameters; only the actual pixel
+        // rows are written so skip regions of the destination stay untouched.
+        const auto packParams = MG_State::pGLContext->GetPixelStoreParameters(false);
+        const SizeT rowPixels = static_cast<SizeT>(packParams.RowLength > 0 ? packParams.RowLength : width);
+        const SizeT dstRowStride = AlignReadbackRow(rowPixels * dstPixelBytes, packParams.Alignment);
+        const SizeT imageRows =
+            applyPackImageParams && packParams.ImageHeight > 0
+                ? static_cast<SizeT>(packParams.ImageHeight)
+                : static_cast<SizeT>(sliceHeight);
+        const SizeT dstImageStride = imageRows * dstRowStride;
+        const SizeT skipImages =
+            applyPackImageParams ? static_cast<SizeT>(std::max(packParams.SkipImages, 0)) : SizeT{0};
+        const SizeT dstSkipOffset = skipImages * dstImageStride +
+                                    static_cast<SizeT>(std::max(packParams.SkipRows, 0)) * dstRowStride +
+                                    static_cast<SizeT>(std::max(packParams.SkipPixels, 0)) * dstPixelBytes;
+        const SizeT dstRowBytes = static_cast<SizeT>(width) * dstPixelBytes;
+
+        const SizeT pboBaseOffset = reinterpret_cast<SizeT>(pixels); // with a PBO, `pixels` is an offset
+        if (pixelPackBufferObject) {
+            const SizeT requiredSize = pboBaseOffset + dstSkipOffset +
+                                    static_cast<SizeT>(sliceCount - 1) * dstImageStride +
+                                    static_cast<SizeT>(sliceHeight - 1) * dstRowStride + dstRowBytes;
+            if (requiredSize > pixelPackBufferObject->GetSize()) {
+                MGLOG_E("Readback conversion: pixel pack buffer is too small");
+                return true;
+            }
+        }
+
+        const SizeT srcComponentSize = GetReadbackComponentSize(wideType);
+        const SizeT srcPixelBytes = 4 * srcComponentSize;
+        Vector<Uint8> convertedRow(dstRowBytes);
+
+        for (GLsizei slice = 0; slice < sliceCount; ++slice) {
+            for (GLsizei row = 0; row < sliceHeight; ++row) {
+                const SizeT flatRow = static_cast<SizeT>(slice) * static_cast<SizeT>(sliceHeight) +
+                                   static_cast<SizeT>(row);
+                const Uint8* srcRow = wide + flatRow * static_cast<SizeT>(width) * srcPixelBytes;
+                ConvertWideReadbackRow(srcRow, convertedRow.data(), static_cast<SizeT>(width), wideType,
+                                                  mapping, type);
+
+                if (packParams.SwapBytes) {
+                    const SizeT groupSize = isPackedType ? packedLayout.byteSize : dstComponentSize;
+                    if (groupSize > 1) {
+                        for (SizeT offset = 0; offset + groupSize <= dstRowBytes; offset += groupSize) {
+                            std::reverse(convertedRow.data() + offset, convertedRow.data() + offset + groupSize);
+                        }
+                    }
+                }
+
+                const SizeT dstOffset = dstSkipOffset + static_cast<SizeT>(slice) * dstImageStride +
+                                     static_cast<SizeT>(row) * dstRowStride;
+                if (pixelPackBufferObject) {
+                    pixelPackBufferObject->WritebackFromBackend({convertedRow.data(), dstRowBytes},
+                                                             pboBaseOffset + dstOffset);
+                } else {
+                    Memcpy(static_cast<Uint8*>(pixels) + dstOffset, convertedRow.data(), dstRowBytes);
+                }
+            }
+        }
+            return true;
+        }
+    } // namespace ReadbackImpl
 } // namespace MobileGL::MG_Backend::DirectGLES
@@ -45,6 +45,59 @@ 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);
+
+        // Stores wide RGBA(_INTEGER) rows into the client pointer or the bound PACK pixel buffer,
+        // honoring the client-side PACK pixel-store parameters (row length, alignment, skips,
+        // swap-bytes, and - when applyPackImageParams - image height/skip images). Shared by the
+        // DirectGLES and DirectVulkan readback conversion paths.
+        Bool StoreWideRowsToClient(const Uint8* wide, GLenum wideType, GLsizei width, GLsizei sliceHeight,
+                                   GLsizei sliceCount, const ReadbackChannelMapping& mapping, GLenum type,
+                                   void* pixels, Bool applyPackImageParams);
+    } // namespace ReadbackImpl
+
    namespace PrgramImpl {
        String ProcessOutColorLocations(const String& glslCode);
        String ForceSupporterOutput(const String& glslCode);
@@ -140,6 +140,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            case TextureInternalFormat::RGB:
            case TextureInternalFormat::RGB8:
                return TextureInternalFormat::RGBA8;
+            // Legacy low-bit-depth formats with no (or rarely supported) native Vulkan
+            // encoding; a wider normalized fallback keeps at least the required precision.
+            case TextureInternalFormat::R3G3B2:
+            case TextureInternalFormat::RGB4:
+            case TextureInternalFormat::RGB5:
+            case TextureInternalFormat::RGBA2:
+            case TextureInternalFormat::RGBA4:
+            case TextureInternalFormat::RGB5A1:
+                return TextureInternalFormat::RGBA8;
+            case TextureInternalFormat::RGB10:
+                return TextureInternalFormat::RGB10A2;
+            case TextureInternalFormat::RGB12:
+            case TextureInternalFormat::RGBA12:
+                return TextureInternalFormat::RGBA16;
            case TextureInternalFormat::SRGB8:
                return TextureInternalFormat::SRGB8Alpha8;
            case TextureInternalFormat::RGB8Snorm:
@@ -397,8 +411,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        }
        if (!handle.Handle || (handle.Backend != WindowBackend::Android &&
                               handle.Backend != WindowBackend::X11 &&
-                               handle.Backend != WindowBackend::MetalLayer)) {
-            MGLOG_E("DirectVulkan backend only supports Android, X11, and CAMetalLayer native windows");
+                               handle.Backend != WindowBackend::MetalLayer &&
+                               handle.Backend != WindowBackend::Win32)) {
+            MGLOG_E("DirectVulkan backend only supports Android, X11, CAMetalLayer, and Win32 native windows");
            return false;
        }

@@ -454,6 +469,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // treat them as signaled/available with zero results from here on.
        BumpRendererGeneration();
        pVulkanRenderer.reset();
+        // The reflection cache is file-scope, not renderer-owned; without this the
+        // deleted programs' reflection strings survive full context teardown.
+        ClearProgramResourceCaches();
        BackendObject::ReleaseEGLResources();
    }

@@ -463,6 +481,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // treat them as signaled/available with zero results from here on.
        BumpRendererGeneration();
        pVulkanRenderer.reset();
+        // The reflection cache is file-scope, not renderer-owned; without this the
+        // deleted programs' reflection strings survive full context teardown.
+        ClearProgramResourceCaches();
    }

    const RendererInfo& BackendObject_DirectVulkan::GetRendererInfo() const {
@@ -488,14 +509,15 @@ 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,
@@ -503,7 +525,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                          E_GL_ARB_multi_draw_indirect, E_GL_ARB_indirect_parameters,
                                          E_GL_EXT_framebuffer_object, E_GL_ARB_depth_texture, E_GL_ARB_buffer_storage,
                                          E_GL_ARB_texture_storage, E_GL_ARB_texture_storage_multisample,
-                                          E_GL_ARB_direct_state_access,
+                                          E_GL_ARB_clear_texture, E_GL_ARB_direct_state_access,
                                          E_GL_ARB_shader_draw_parameters, E_GL_ARB_gpu_shader_int64, E_GL_KHR_debug,
                                          E_GL_ARB_gpu_shader5, E_GL_ARB_multi_bind, E_GL_ARB_shading_language_420pack,
                                          E_GL_ARB_vertex_attrib_binding, E_GL_ARB_shader_image_size};
@@ -516,6 +538,13 @@ 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 +659,8 @@ 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 +710,11 @@ 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;
@@ -732,9 +767,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        m_dynamicParameters.MaxTextureBufferSize = m_vulkanCaps.MaxTextureBufferSize;
        m_dynamicParameters.MaxUniformBufferBindings = m_vulkanCaps.MaxUniformBufferBindings;
        m_dynamicParameters.MaxUniformBlockSize = m_vulkanCaps.MaxUniformBlockSize;
-        m_dynamicParameters.MaxImageUnits = std::min(m_vulkanCaps.MaxImageUnits, maxSupportedTextureUnits);
-        m_dynamicParameters.MaxCombinedImageUniforms = m_vulkanCaps.MaxCombinedImageUniforms;
-        m_dynamicParameters.MaxComputeImageUniforms = m_vulkanCaps.MaxComputeImageUniforms;
+        m_dynamicParameters.MaxImageUnits =
+            std::max(std::min(m_vulkanCaps.MaxImageUnits, maxSupportedTextureUnits), 0);
+        m_dynamicParameters.MaxCombinedImageUniforms = std::max(m_vulkanCaps.MaxCombinedImageUniforms, 0);
+        const Int maxPerStageImageUniforms =
+            std::min(m_dynamicParameters.MaxImageUnits, m_dynamicParameters.MaxCombinedImageUniforms);
+        // Vulkan uses one descriptor limit for every stage, but non-compute stores/atomics are
+        // optional device features. VulkanRenderer enables each feature whenever the physical
+        // device reports it, so these are the exact limits the logical device can compile and run.
+        m_dynamicParameters.MaxVertexImageUniforms =
+            m_vulkanCaps.SupportsVertexPipelineStoresAndAtomics ? maxPerStageImageUniforms : 0;
+        m_dynamicParameters.MaxGeometryImageUniforms =
+            m_vulkanCaps.SupportsVertexPipelineStoresAndAtomics && m_vulkanCaps.SupportsGeometryShader
+                ? maxPerStageImageUniforms
+                : 0;
+        m_dynamicParameters.MaxFragmentImageUniforms =
+            m_vulkanCaps.SupportsFragmentStoresAndAtomics ? maxPerStageImageUniforms : 0;
+        m_dynamicParameters.MaxComputeImageUniforms =
+            std::min(std::max(m_vulkanCaps.MaxComputeImageUniforms, 0), maxPerStageImageUniforms);
        const Int maxSupportedDrawBuffers =
            static_cast<Int>(MG_State::GLState::FramebufferObject::MAX_DRAW_BUFFERS);
        m_dynamicParameters.MaxDrawBuffers = std::min(m_vulkanCaps.MaxDrawBuffers, maxSupportedDrawBuffers);
@@ -765,5 +815,32 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                    m_vulkanCaps.MaxShaderStorageBlockSize,
                    m_dynamicParameters.MaxShaderStorageBlockSize);
        }
+        switch (m_vulkanCaps.VendorId) {
+        case 0x5143u: // VK_VENDOR_ID: Qualcomm
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Qualcomm;
+            break;
+        case 0x13B5u: // ARM
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Arm;
+            break;
+        case 0x10DEu: // NVIDIA
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Nvidia;
+            break;
+        case 0x1002u: // AMD
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Amd;
+            break;
+        case 0x8086u: // Intel
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Intel;
+            break;
+        case 0x1010u: // Imagination
+            m_dynamicParameters.GpuVendor = GpuVendorKind::ImgTec;
+            break;
+        case 0x10005u: // Mesa software (lavapipe)
+        case 0x1AE0u:  // Google (SwiftShader)
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Software;
+            break;
+        default:
+            m_dynamicParameters.GpuVendor = GpuVendorKind::Unknown;
+            break;
+        }
    }
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -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
@@ -20,7 +20,8 @@
 #include <spirv_reflect.h>

 namespace MobileGL::MG_Backend::DirectVulkan {
-    UniquePtr<VulkanRenderer> pVulkanRenderer = nullptr;
+    // Leak-at-exit storage; see GlobalObjects.cpp.
+    UniquePtr<VulkanRenderer>& pVulkanRenderer = *new UniquePtr<VulkanRenderer>();

    namespace {
        // Generation of the live VulkanRenderer instance, mirroring
@@ -60,6 +61,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        };

        struct ProgramResourceCache {
+            // Lifetime id of the program the cached reflection belongs to. GL names are
+            // recycled (IndexGenerator hands freed indices straight back), and a
+            // recreated program's backendStateVersion restarts at the same small values,
+            // so the version alone can collide; the never-reused lifetime id makes the
+            // slot's ownership unambiguous.
+            Uint64 programLifetimeId = 0;
            Uint32 backendStateVersion = 0;
            Vector<StorageBlockResource> storageBlocks;
            Vector<BufferVariableResource> bufferVariables;
@@ -81,6 +88,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            Uint32 baseInstance = 0;
        };

+        // Keyed by GL program name so the freed-name reuse in IndexGenerator bounds the
+        // map at the peak-simultaneous-program high-water mark; each slot's ownership is
+        // checked against the program's lifetime id before it is served (see
+        // GetProgramResourceCache). Cleared wholesale at EGL teardown via
+        // ClearProgramResourceCaches.
        UnorderedMap<GLuint, ProgramResourceCache> g_programResourceCaches;

        void ClearReadPixelsOutput(GLsizei width, GLsizei height, GLenum format, GLenum type, void* pixels) {
@@ -141,13 +153,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        ProgramResourceCache& GetProgramResourceCache(const MG_State::GLState::ProgramObject& program) {
            auto& cache = g_programResourceCaches[program.GetExternalIndex()];
+            const Uint64 programLifetimeId = program.GetLifetimeId();
            const Uint32 backendStateVersion = program.GetBackendStateVersion();
-            if (cache.backendStateVersion == backendStateVersion &&
+            // The lifetime id must match too: a new program that reuses a deleted
+            // program's name and happens to land on the same backendStateVersion (both
+            // count from zero) would otherwise be served the dead program's reflection.
+            if (cache.programLifetimeId == programLifetimeId &&
+                cache.backendStateVersion == backendStateVersion &&
                (!cache.storageBlocks.empty() || !cache.bufferVariables.empty())) {
                return cache;
            }

            cache = {};
+            cache.programLifetimeId = programLifetimeId;
            cache.backendStateVersion = backendStateVersion;

            Vector<SpvReflectShaderModule> modules;
@@ -365,6 +383,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        }
    } // namespace

+    void ClearProgramResourceCaches() {
+        // Called from EGL teardown while the backend's m_eglStateMutex is held; GL
+        // calls are serialized in this codebase (contexts migrate threads but never
+        // run concurrently), so no other thread can be inside the unsynchronized map.
+        // Live programs in another context self-heal: their entry rebuilds from the
+        // retained generated SPIR-V on the next resource query.
+        g_programResourceCaches.clear();
+    }
+
    GLuint GetShaderStorageBlockIndex(const MG_State::GLState::ProgramObject& program, const String& name) {
        auto& cache = GetProgramResourceCache(program);
        const auto it = std::find_if(cache.storageBlocks.begin(), cache.storageBlocks.end(),
@@ -12,7 +12,7 @@
 #include "Renderer/VulkanRenderer.h"

 namespace MobileGL::MG_Backend::DirectVulkan {
-    extern UniquePtr<VulkanRenderer> pVulkanRenderer;
+    extern UniquePtr<VulkanRenderer>& pVulkanRenderer;

    // Generation of the live VulkanRenderer instance, mirroring DirectGLES's
    // g_syncContextGeneration. BackendObject_DirectVulkan bumps it wherever
@@ -23,6 +23,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    Uint64 GetRendererGeneration();
    void BumpRendererGeneration();

+    // Drops every cached program-resource reflection entry (CPU-side strings/vectors
+    // only, no Vulkan handles). Called at EGL teardown next to the renderer reset;
+    // safe because GL calls are serialized in this codebase, and any still-live
+    // program rebuilds its entry from the retained generated SPIR-V on demand.
+    void ClearProgramResourceCaches();
+
    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);
@@ -150,12 +150,30 @@ namespace MobileGL::MG_Backend::DirectVulkan {

    Bool FrameContext::TransitionToPresent(VkImage image, VkImageLayout oldLayout, VkImageLayout presentLayout) {
        auto& frame = GetCurrent();
-        if (frame.hasCommandBufferRecorded || frame.isCommandRecording || oldLayout == presentLayout ||
-            oldLayout == VK_IMAGE_LAYOUT_SHARED_PRESENT_KHR) {
+        if (oldLayout == presentLayout || oldLayout == VK_IMAGE_LAYOUT_SHARED_PRESENT_KHR) {
            return false;
        }

-        auto& commandBuffer = BeginCommandRecording();
+        // The barrier belongs in the frame's own recording. Bailing out because
+        // something was already recorded (the previous behaviour) dropped the
+        // transition entirely for every frame that never ran a default-framebuffer
+        // render pass - the only other thing that carries the image to
+        // PRESENT_SRC_KHR, via that pass's finalLayout - so the swapchain image was
+        // handed to the WSI still in the layout it was acquired in.
+        // A closed-but-unsubmitted buffer can only come from a submit that already
+        // failed (SubmitPendingCommandBuffer leaves the flag set on error), and
+        // appending to it is illegal while reopening would reset the frame's own
+        // commands away. The device is gone on that path anyway - stay silent-safe
+        // rather than trade a lost device for a barrier into a closed buffer.
+        if (frame.hasCommandBufferRecorded) {
+            MGLOG_E("TransitionToPresent: command buffer already closed; skipping the present barrier");
+            return false;
+        }
+
+        // Reopening a recording here would vkResetCommandBuffer this frame's own
+        // commands away, so append to the open one and let the caller close it.
+        const Bool openedRecording = !frame.isCommandRecording;
+        VkCommandBuffer commandBuffer = openedRecording ? BeginCommandRecording() : frame.commandBuffer;

        VkImageMemoryBarrier presentBarrier{};
        presentBarrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
@@ -174,7 +192,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        vkCmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0,
                             nullptr, 0, nullptr, 1, &presentBarrier);

-        EndCommandRecording();
+        if (openedRecording) {
+            EndCommandRecording();
+        }
        return true;
    }

@@ -227,12 +247,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        result = vkAcquireNextImageKHR(device, swapchain, timeout, frame.imageAvailableSemaphore, acquireFence,
                                       &outImageIndex);
-        if (result != VK_SUCCESS) {
+        // VK_SUBOPTIMAL_KHR is a success code: an image *was* acquired and
+        // imageAvailableSemaphore *will* be signaled. Bailing out on it skipped both
+        // the consumed-flag reset (leaving a stale "already consumed", so the next
+        // submit never waited on the pending signal) and the fence reset (leaving
+        // the slot's fence signaled for the next submit to reuse). Only a genuine
+        // failure - VK_ERROR_OUT_OF_DATE_KHR and friends, where nothing is acquired
+        // and nothing is signaled - skips the bookkeeping.
+        if (result != VK_SUCCESS && result != VK_SUBOPTIMAL_KHR) {
            return result;
        }

        frame.imageAvailableSemaphoreConsumed = false;
-        return vkResetFences(device, 1, &frame.imageInFlightFence);
+        const VkResult resetResult = vkResetFences(device, 1, &frame.imageInFlightFence);
+        // Hand the acquire's own code back so the caller can schedule a rebuild.
+        return resetResult == VK_SUCCESS ? result : resetResult;
    }

    Uint32 FrameContext::GetCurrentFrameIndex() const {
@@ -264,7 +293,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        if (result != VK_SUCCESS) {
            return result;
        }
-        frame.retiredCommandBuffers.push_back(frame.commandBuffer);
+        // lastSubmitIndex was just written by the renderer for the submission
+        // that carried this command buffer.
+        frame.retiredCommandBuffers.push_back({frame.commandBuffer, frame.lastSubmitIndex});
        frame.commandBuffer = replacement;
        return VK_SUCCESS;
    }
@@ -274,12 +305,40 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            return;
        }
        if (m_device != VK_NULL_HANDLE && m_commandPool != VK_NULL_HANDLE) {
-            vkFreeCommandBuffers(m_device, m_commandPool, static_cast<Uint32>(frame.retiredCommandBuffers.size()),
-                                 frame.retiredCommandBuffers.data());
+            for (const auto& retired : frame.retiredCommandBuffers) {
+                vkFreeCommandBuffers(m_device, m_commandPool, 1, &retired.commandBuffer);
+            }
        }
        frame.retiredCommandBuffers.clear();
    }

+    void FrameContext::FreeRetiredCommandBuffersCompletedUpTo(Uint64 completedSubmitIndex) {
+        if (m_device == VK_NULL_HANDLE || m_commandPool == VK_NULL_HANDLE) {
+            return;
+        }
+        for (auto& frame : m_frames) {
+            // Retired buffers are appended in submit order, so the completed
+            // ones form a prefix.
+            SizeT completedCount = 0;
+            while (completedCount < frame.retiredCommandBuffers.size() &&
+                   frame.retiredCommandBuffers[completedCount].submitIndex <= completedSubmitIndex) {
+                vkFreeCommandBuffers(m_device, m_commandPool, 1,
+                                     &frame.retiredCommandBuffers[completedCount].commandBuffer);
+                ++completedCount;
+            }
+            if (completedCount > 0) {
+                frame.retiredCommandBuffers.erase(frame.retiredCommandBuffers.begin(),
+                                                  frame.retiredCommandBuffers.begin() + completedCount);
+            }
+        }
+    }
+
+    void FrameContext::FreeAllRetiredCommandBuffers() {
+        for (auto& frame : m_frames) {
+            FreeRetiredCommandBuffers(frame);
+        }
+    }
+
    void FrameContext::AssertValidFrameIndex(Uint32 frameIndex) const {
        MOBILEGL_ASSERT(frameIndex < m_frames.size(), "FrameContext index out of range");
    }
@@ -40,6 +40,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            VkPresentInfoKHR presentInfo{VK_STRUCTURE_TYPE_PRESENT_INFO_KHR};
        };

+        // A command buffer submitted mid-frame (FlushPendingCommands), tagged
+        // with the submit-tracker index it was submitted under so it can be
+        // freed as soon as that submission is observed complete - without
+        // waiting for the slot's fence to be waited again (present-less flush
+        // loops never wait it).
+        struct RetiredCommandBuffer {
+            VkCommandBuffer commandBuffer = VK_NULL_HANDLE;
+            Uint64 submitIndex = 0;
+        };
+
        struct FrameData {
            VkCommandBuffer commandBuffer = VK_NULL_HANDLE;
            VkSemaphore imageAvailableSemaphore = VK_NULL_HANDLE;
@@ -47,10 +57,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            Bool isCommandRecording = false;
            Bool hasCommandBufferRecorded = false;
            Bool imageAvailableSemaphoreConsumed = false;
-            // Command buffers submitted mid-frame (FlushPendingCommands) whose
-            // execution is only known complete once this slot's fence has been
-            // waited again; freed at that point.
-            Vector<VkCommandBuffer> retiredCommandBuffers;
+            // Command buffers submitted mid-frame (FlushPendingCommands),
+            // appended in submit order; freed once their submission is known
+            // complete (fence wait or completion poll).
+            Vector<RetiredCommandBuffer> retiredCommandBuffers;
            // Submit-tracker index of this slot's most recent queue submission
            // (written by the renderer at submit time).
            Uint64 lastSubmitIndex = 0;
@@ -79,9 +89,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // Parks the current (already ended and submitted) command buffer on the
        // slot's retired list and installs a freshly allocated one, so recording
        // can restart while the submitted buffer is still executing. Retired
-        // buffers are freed after the slot's fence is next waited.
+        // buffers are freed after the slot's fence is next waited, or as soon
+        // as their submission is observed complete.
        VkResult RetireCurrentCommandBuffer();

+        // Frees every retired command buffer whose tagged submission index is
+        // known complete. Driven by the renderer's submit tracker on completion
+        // events (fence waits and non-blocking polls), so present-less flush
+        // loops reclaim their buffers without any extra wait.
+        void FreeRetiredCommandBuffersCompletedUpTo(Uint64 completedSubmitIndex);
+        // Frees every slot's retired command buffers. Only valid when the
+        // caller has proven every queue submission complete.
+        void FreeAllRetiredCommandBuffers();
+
        Uint32 GetCurrentFrameIndex() const;
        Uint32 GetFrameCount() const;

@@ -8,6 +8,8 @@

 #include "PipelineFactory.h"

+#include <algorithm>
+
 namespace MobileGL::MG_Backend::DirectVulkan {
    static const char* PrimitiveTopologyToString(VkPrimitiveTopology topology) {
        switch (topology) {
@@ -108,6 +110,81 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                  "vkCreatePipelineCache");
    }

+    // Must be called once, before any pipeline is created: the flag is not part of the
+    // pipeline hash, so flipping it mid-life would serve cached pipelines built under the
+    // old value.
+    void PipelineFactory::SetSuppressBlendedDepthWrite(Bool enabled) {
+        s_suppressBlendedDepthWrite = enabled;
+    }
+
+    Bool PipelineFactory::ShouldSuppressBlendedDepthWriteForDevice(MG_Config::QuirkOverride quirkOverride,
+                                                                   Uint32 vendorId) {
+        static constexpr Uint32 kVendorIdQualcomm = 0x5143;
+        switch (quirkOverride) {
+        case MG_Config::QuirkOverride::ForceOn:
+            return true;
+        case MG_Config::QuirkOverride::ForceOff:
+            return false;
+        case MG_Config::QuirkOverride::Auto:
+        default:
+            return vendorId == kVendorIdQualcomm;
+        }
+    }
+
+    namespace {
+        // MIN/MAX extremum blending: the signature of a depth-bounds accumulation pass
+        // (MC 26.3 OIT writes vec4(-linD, linD, deviceZ, 0) under GL_MAX while writing
+        // depth for its equality chain). MIN/MAX ignore blend factors per the Vulkan spec.
+        //
+        // Deliberately the ONLY shape stripped. A quirk should touch as little unrelated
+        // content as possible, and a trace sweep of every fixture showed the wider
+        // alternatives all cost more than they fix:
+        //   - additive ONE+ONE with a depth write matched zero draws of the 26.3 chain
+        //     (its transmittance/accumulate passes disable depth writes themselves) - the
+        //     only real content it caught was harmless additive glow effects (Create);
+        //   - sorted-transparency "over" blends (SRC_ALPHA-style) are order-dependent,
+        //     drawn once per surface, and rely on their depth writes for occlusion;
+        //   - separate-alpha accumulation over an over-blending color channel has no
+        //     known pairing with a depth-equality chain (color channel only, see tests).
+        // If a future workload pairs another blend shape with an equality chain, widen
+        // this with that evidence in hand rather than pre-emptively.
+        Bool IsAccumulationBlend(const VkPipelineColorBlendAttachmentState& attachment) {
+            return attachment.colorBlendOp == VK_BLEND_OP_MIN ||
+                   attachment.colorBlendOp == VK_BLEND_OP_MAX;
+        }
+    } // namespace
+
+    Bool PipelineFactory::ShouldSuppressDepthWrite(const PipelineCreatePayload& payload) {
+        if (!payload.depthWriteEnable) {
+            return false;
+        }
+        // A shader that assigns gl_FragDepth supplies depth itself rather than taking the
+        // pipeline's interpolated Z, so a driver that varies the vertex position math
+        // between pipelines cannot desynchronize it. (A gl_FragDepth = gl_FragCoord.z
+        // passthrough is the exception that stays exposed; no known content pairs one with
+        // an equality chain, and 26.3's composite is a genuine computed-depth writer.)
+        if (payload.fragmentReplacesDepth) {
+            return false;
+        }
+        for (Uint32 i = 0; i < payload.colorAttachmentCount; ++i) {
+            const VkPipelineColorBlendAttachmentState& attachment = payload.colorBlendAttachments[i];
+            if (attachment.blendEnable != VK_TRUE) {
+                continue;
+            }
+            // All color writes masked: blending is moot (depth-prepass pattern that left
+            // GL_BLEND enabled); stripping the depth write would delete the whole prepass.
+            if (attachment.colorWriteMask == 0) {
+                continue;
+            }
+            // Any attachment qualifies, not just attachment 0: the 26.3 transmittance pass
+            // accumulates into a 2-target MRT and must stay stripped.
+            if (IsAccumulationBlend(attachment)) {
+                return true;
+            }
+        }
+        return false;
+    }
+
    PipelineFactory::~PipelineFactory() {
        DestroyAll();
        if (m_pipelineCache != VK_NULL_HANDLE) {
@@ -152,6 +229,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            XXH64_update(m_hashState, &payload.backStencilDepthFailOp, sizeof(payload.backStencilDepthFailOp)));
        XXHASH_VERIFY(
            XXH64_update(m_hashState, &payload.backStencilCompareOp, sizeof(payload.backStencilCompareOp)));
+        XXHASH_VERIFY(
+            XXH64_update(m_hashState, &payload.fragmentReplacesDepth, sizeof(payload.fragmentReplacesDepth)));
        if (payload.colorAttachmentCount > 0) {
            XXHASH_VERIFY(XXH64_update(
                m_hashState,
@@ -165,23 +244,108 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const HashType hash = ComputeHash(payload);
        auto it = m_cache.find(hash);
        if (it != m_cache.end()) {
-            return it->second;
+            it->second.lastUsedFrame = m_frameCounter;
+            return it->second.pipeline;
        }

        VkPipeline pipeline = CreatePipeline(payload);
-        m_cache.emplace(hash, pipeline);
+        m_cache.emplace(hash, PipelineCacheEntry{pipeline, payload.programHash, payload.renderPass,
+                                                 m_frameCounter});
        return pipeline;
    }

    void PipelineFactory::DestroyAll() {
        for (auto& pair : m_cache) {
-            if (pair.second != VK_NULL_HANDLE) {
-                vkDestroyPipeline(m_device, pair.second, nullptr);
+            if (pair.second.pipeline != VK_NULL_HANDLE) {
+                vkDestroyPipeline(m_device, pair.second.pipeline, nullptr);
            }
        }
        m_cache.clear();
    }

+    Uint32 PipelineFactory::OnFrameBoundary() {
+        ++m_frameCounter;
+
+        // Sweep cadence and retire age mirror VkRenderPassManager::OnPresent: an entry
+        // idle for more than kRetireAgeFrames frame boundaries cannot be referenced by
+        // any in-flight command buffer (frames-in-flight <= MOBILEGL_MAGMA_FRAMESINFLIGHT),
+        // so immediate vkDestroyPipeline is safe. The caller must drop its "last
+        // pipeline" memo when this returns non-zero: the memo can return a cached
+        // handle without touching this cache, so an evicted pipeline may still be
+        // memoized (present-less flush loops never reset the memo per frame).
+        constexpr Uint64 kSweepInterval = 256;
+        constexpr Uint64 kRetireAgeFrames = 1024;
+        if ((m_frameCounter % kSweepInterval) != 0) {
+            return 0;
+        }
+
+        Uint32 evicted = 0;
+        for (auto it = m_cache.begin(); it != m_cache.end();) {
+            if (m_frameCounter - it->second.lastUsedFrame > kRetireAgeFrames) {
+                if (it->second.pipeline != VK_NULL_HANDLE) {
+                    vkDestroyPipeline(m_device, it->second.pipeline, nullptr);
+                }
+                it = m_cache.erase(it);
+                ++evicted;
+            } else {
+                ++it;
+            }
+        }
+        if (evicted > 0) {
+            MGLOG_D("PipelineFactory::OnFrameBoundary: evicted %u idle pipelines (%zu remain)", evicted,
+                    m_cache.size());
+        }
+        return evicted;
+    }
+
+    Uint32 PipelineFactory::EvictByRenderPasses(const Vector<VkRenderPass>& renderPasses) {
+        if (renderPasses.empty() || m_cache.empty()) {
+            return 0;
+        }
+        // Sorted-batch membership test keeps a mass eviction (shader-pack switch,
+        // dimension exit) at one O(cache * log batch) scan instead of one full scan
+        // per dying pass.
+        Vector<VkRenderPass> sortedPasses = renderPasses;
+        std::sort(sortedPasses.begin(), sortedPasses.end());
+        Uint32 evicted = 0;
+        for (auto it = m_cache.begin(); it != m_cache.end();) {
+            if (std::binary_search(sortedPasses.begin(), sortedPasses.end(), it->second.renderPass)) {
+                if (it->second.pipeline != VK_NULL_HANDLE) {
+                    vkDestroyPipeline(m_device, it->second.pipeline, nullptr);
+                }
+                it = m_cache.erase(it);
+                ++evicted;
+            } else {
+                ++it;
+            }
+        }
+        if (evicted > 0) {
+            MGLOG_D("PipelineFactory::EvictByRenderPasses: evicted %u pipelines for %zu destroyed render passes",
+                    evicted, sortedPasses.size());
+        }
+        return evicted;
+    }
+
+    Uint32 PipelineFactory::EvictByProgramHash(HashType programHash) {
+        Uint32 evicted = 0;
+        for (auto it = m_cache.begin(); it != m_cache.end();) {
+            if (it->second.programHash == programHash) {
+                if (it->second.pipeline != VK_NULL_HANDLE) {
+                    vkDestroyPipeline(m_device, it->second.pipeline, nullptr);
+                }
+                it = m_cache.erase(it);
+                ++evicted;
+            } else {
+                ++it;
+            }
+        }
+        if (evicted > 0) {
+            MGLOG_D("PipelineFactory::EvictByProgramHash: evicted %u pipelines for program hash 0x%llx",
+                    evicted, static_cast<unsigned long long>(programHash));
+        }
+        return evicted;
+    }
+
    VkPipeline PipelineFactory::CreatePipeline(const PipelineCreatePayload& payload) const {
        MOBILEGL_ASSERT(payload.stages != nullptr && !payload.stages->empty(), "PipelineFactory: stages are empty");
        MOBILEGL_ASSERT(payload.vertexInputState != nullptr, "PipelineFactory: vertexInputState is null");
@@ -258,6 +422,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        for (Uint32 i = 0; i < payload.colorAttachmentCount; ++i) {
            colorAttachments[i] = payload.colorBlendAttachments[i];
        }
+        // Suppress depth writes on accumulation-blended pipelines when the active driver
+        // cannot keep vertex positions invariant across the pipelines of a multi-pass
+        // depth-equality chain (see SetSuppressBlendedDepthWrite). The decision is narrowed
+        // in ShouldSuppressDepthWrite: sorted-transparency "over" blends (vanilla MC water),
+        // gl_FragDepth writers, and masked-out attachments keep their depth writes.
+        // This bakes the decision into the pipeline, which only works because depth write is
+        // static state here - adding VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE to kDynamicStates
+        // would let the record-time value override it and silently disable the quirk.
+        if (s_suppressBlendedDepthWrite && ShouldSuppressDepthWrite(payload)) {
+            depthStencil.depthWriteEnable = VK_FALSE;
+        }
        VkPipelineColorBlendStateCreateInfo blend{VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO};
        blend.logicOpEnable = payload.logicOpEnable ? VK_TRUE : VK_FALSE;
        blend.logicOp = payload.logicOp;
@@ -49,6 +49,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            VkStencilOp backStencilPassOp = VK_STENCIL_OP_KEEP;
            VkStencilOp backStencilDepthFailOp = VK_STENCIL_OP_KEEP;
            VkCompareOp backStencilCompareOp = VK_COMPARE_OP_ALWAYS;
+            // The fragment module writes gl_FragDepth (SPIR-V DepthReplacing); exempts the
+            // pipeline from the blended depth-write quirk (see ShouldSuppressDepthWrite).
+            Bool fragmentReplacesDepth = false;
            Array<VkPipelineColorBlendAttachmentState, kMaxColorAttachments> colorBlendAttachments{};
            const Vector<VkPipelineShaderStageCreateInfo>* stages = nullptr;
            const VkPipelineVertexInputStateCreateInfo* vertexInputState = nullptr;
@@ -62,13 +65,69 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        VkPipeline GetOrCreatePipeline(const PipelineCreatePayload& payload);
        void DestroyAll();

+        // Frame boundary hook: ages the pipeline cache and destroys long-unused entries
+        // (their command buffers retired many frames ago), mirroring
+        // VkRenderPassManager::OnPresent's sweep. Returns the number of pipelines
+        // destroyed so the caller can drop any memoized VkPipeline handle.
+        Uint32 OnFrameBoundary();
+        // Destroys every cached pipeline hashed on one of `renderPasses`. Only safe
+        // when the caller guarantees GPU idleness for them - the render-pass manager
+        // calls this (via the renderer) for passes its own >1024-boundary-idle sweep
+        // just evicted, and a pipeline hashed on those handles is only ever bound by
+        // draws that also hit the render-pass entries. Also closes the handle-recycling
+        // hazard: a recycled VkRenderPass value must never serve a stale pipeline.
+        // Batched: one cache scan regardless of how many passes died in the sweep.
+        // Returns the number destroyed (callers invalidate memos when non-zero).
+        Uint32 EvictByRenderPasses(const Vector<VkRenderPass>& renderPasses);
+        // Destroys every cached pipeline built from the program with content hash
+        // `programHash`. Called from the ProgramFactory eviction path, which proves the
+        // same >1024-boundary idleness (the program's pipelines are only bound by draws
+        // that stamp its factory entry). Returns the number destroyed.
+        Uint32 EvictByProgramHash(HashType programHash);
+
+        // Driver quirk: suppress depth writes on accumulation-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. Only MIN/MAX extremum blends are stripped - the signature of such a
+        // chain's depth-bounds pass (MC 26.3 OIT), and per a fixture-wide trace sweep the
+        // only depth-writing shape the chain actually uses - so every other blend
+        // (sorted-transparency "over" like vanilla MC water, additive glows, ...) keeps
+        // its depth writes. Set at renderer initialization based on the active driver.
+        static void SetSuppressBlendedDepthWrite(Bool enabled);
+        static Bool IsSuppressBlendedDepthWriteEnabled() { return s_suppressBlendedDepthWrite; }
+        // Device gate for the quirk: ForceOn/ForceOff bypass detection, Auto enables it on
+        // the known-affected vendor (Qualcomm).
+        static Bool ShouldSuppressBlendedDepthWriteForDevice(MG_Config::QuirkOverride quirkOverride,
+                                                             Uint32 vendorId);
+        // Pure per-pipeline strip decision (exempts gl_FragDepth writers, masked-out and
+        // non-accumulation blends); combined with the device flag in CreatePipeline. Static
+        // and payload-only so tests can pin the contract without a VkDevice.
+        static Bool ShouldSuppressDepthWrite(const PipelineCreatePayload& payload);
+
    private:
+        struct PipelineCacheEntry {
+            VkPipeline pipeline = VK_NULL_HANDLE;
+            // The hashed inputs the eviction paths key on: programHash ties the entry to
+            // its ProgramFactory entry, renderPass records the exact handle the hash
+            // folded in (the hash is one-way, so targeted eviction needs them verbatim).
+            HashType programHash = 0;
+            VkRenderPass renderPass = VK_NULL_HANDLE;
+            // Frame-boundary counter value of the last GetOrCreatePipeline hit; drives
+            // cache eviction (see OnFrameBoundary).
+            Uint64 lastUsedFrame = 0;
+        };
+
        VkPipeline CreatePipeline(const PipelineCreatePayload& payload) const;

        VkDevice m_device = VK_NULL_HANDLE;
        const VulkanRendererConfig& m_config;
        VkPipelineCache m_pipelineCache = VK_NULL_HANDLE;
-        UnorderedMap<HashType, VkPipeline> m_cache;
+        UnorderedMap<HashType, PipelineCacheEntry> m_cache;
+        // Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
+        Uint64 m_frameCounter = 0;
        static inline XXH64_state_t* m_hashState = XXH64_createState();
+        static inline Bool s_suppressBlendedDepthWrite = false;
    };
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -14,8 +14,16 @@
 #include "MG_State/GLState/TextureState/TextureEnum.h"

 #include <Includes.h>
+#include <spirv_reflect.h>

 namespace MobileGL::MG_Backend::DirectVulkan {
+    enum class SamplerNumericDomain : Uint8 {
+        Unknown = 0,
+        Float,
+        SignedInteger,
+        UnsignedInteger,
+    };
+
    class ProgramFactory {
    public:
        enum class DescriptorBindingKind : Uint8 {
@@ -34,6 +42,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            SurfaceRotate90 = 1 << 2,
            SurfaceRotate180 = 1 << 3,
            SurfaceRotate270 = 1 << 4,
+            // Rewrites the fragment stage's implicit-LOD image samples to explicit LOD 0.
+            // Only ever set for a draw whose every sampler binding is clamped to a single mip
+            // level, which makes the two forms produce identical texels (the implicit lambda is
+            // clamped into [minLod, maxLod] = [0, 0] regardless of derivatives or bias).
+            ExplicitLod0Sampling = 1 << 5,
+            // Fragment arithmetic may run at relaxed (fp16) precision. Only requested for draws
+            // where every sampled texture and every colour attachment is an 8-bit-or-less
+            // normalized format, so nothing the shader reads or writes carries more precision
+            // than fp16 already represents exactly.
+            RelaxedFragmentPrecision = 1 << 6,
        };
        using CompileOptionFlags = Flags<CompileOptionBit>;
        using HashType = Uint64;
@@ -51,11 +69,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            Vector<DescriptorBindingKind> bindingKinds;
            Vector<Uint32> dynamicBindings;
            Vector<Int> uniformBlockIndexByBinding;
+            // Descriptor count per binding (1 except for UBO instance arrays, which occupy one
+            // binding with descriptorCount = N).
+            Vector<Uint16> bindingDescriptorCounts;
+            // Per-element GL uniform block indices for arrayed UBO bindings (count > 1);
+            // element 0 of a non-arrayed binding stays in uniformBlockIndexByBinding.
+            UnorderedMap<Uint32, Vector<Int>> arrayedUniformBlockIndicesByBinding;
            Vector<String> samplerNameByBinding;
            Vector<Int> samplerUniformLocationByBinding;
            Vector<TextureTarget> samplerTextureTargetByBinding;
+            Vector<SamplerNumericDomain> samplerNumericDomainByBinding;
+            Vector<VkFormat> storageImageFormatByBinding;
+            Vector<Bool> storageImageUsesBindingFormatByBinding;
            Vector<String> storageBlockNameByBinding;
            Vector<Int> storageBlockIndexByBinding;
+            // Set once during ReflectLayout so the per-draw path can skip the whole
+            // storage-image preparation for the overwhelming majority of programs.
+            Bool hasStorageImages = false;
            Int globalUboBinding = -1;
            Uint32 activeVertexInputLocationMask = 0;
            Array<GLenum, kMaxVertexInputLocations> vertexInputTypes{};
@@ -64,6 +94,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            ShaderStage rasterizationProducerStage = ShaderStage::Unknown;
            Uint32 producerOutputComponentCount = 0;
            Uint32 fragmentInputComponentCount = 0;
+            // The fragment module declares the DepthReplacing execution mode (writes
+            // gl_FragDepth); shader-computed depth is immune to the cross-pipeline
+            // position-invariance quirk (see PipelineFactory::ShouldSuppressDepthWrite).
+            Bool fragmentReplacesDepth = false;
+            // Frame-boundary counter value of the last GetOrCreateProgram hit; drives
+            // cache eviction (see OnFrameBoundary).
+            Uint64 lastUsedFrame = 0;

            static inline VkDevice s_device = VK_NULL_HANDLE;

@@ -79,11 +116,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                bindingKinds = std::move(other.bindingKinds);
                dynamicBindings = std::move(other.dynamicBindings);
                uniformBlockIndexByBinding = std::move(other.uniformBlockIndexByBinding);
+                bindingDescriptorCounts = std::move(other.bindingDescriptorCounts);
+                arrayedUniformBlockIndicesByBinding = std::move(other.arrayedUniformBlockIndicesByBinding);
                samplerNameByBinding = std::move(other.samplerNameByBinding);
                samplerUniformLocationByBinding = std::move(other.samplerUniformLocationByBinding);
                samplerTextureTargetByBinding = std::move(other.samplerTextureTargetByBinding);
+                samplerNumericDomainByBinding = std::move(other.samplerNumericDomainByBinding);
+                storageImageFormatByBinding = std::move(other.storageImageFormatByBinding);
+                storageImageUsesBindingFormatByBinding =
+                    std::move(other.storageImageUsesBindingFormatByBinding);
                storageBlockNameByBinding = std::move(other.storageBlockNameByBinding);
                storageBlockIndexByBinding = std::move(other.storageBlockIndexByBinding);
+                hasStorageImages = other.hasStorageImages;
                globalUboBinding = other.globalUboBinding;
                activeVertexInputLocationMask = other.activeVertexInputLocationMask;
                vertexInputTypes = other.vertexInputTypes;
@@ -92,15 +136,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                rasterizationProducerStage = other.rasterizationProducerStage;
                producerOutputComponentCount = other.producerOutputComponentCount;
                fragmentInputComponentCount = other.fragmentInputComponentCount;
+                fragmentReplacesDepth = other.fragmentReplacesDepth;
+                lastUsedFrame = other.lastUsedFrame;
                other.hash = 0;
                other.descriptorSetLayout = VK_NULL_HANDLE;
                other.pipelineLayout = VK_NULL_HANDLE;
+                other.hasStorageImages = false;
                other.globalUboBinding = -1;
                other.activeVertexInputLocationMask = 0;
                other.activeFragmentOutputLocationMask = 0;
                other.rasterizationProducerStage = ShaderStage::Unknown;
                other.producerOutputComponentCount = 0;
                other.fragmentInputComponentCount = 0;
+                other.fragmentReplacesDepth = false;
+                other.lastUsedFrame = 0;
            }
            VkProgramObject& operator=(VkProgramObject&& other) noexcept {
                if (this == &other) {
@@ -115,11 +164,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                bindingKinds = std::move(other.bindingKinds);
                dynamicBindings = std::move(other.dynamicBindings);
                uniformBlockIndexByBinding = std::move(other.uniformBlockIndexByBinding);
+                bindingDescriptorCounts = std::move(other.bindingDescriptorCounts);
+                arrayedUniformBlockIndicesByBinding = std::move(other.arrayedUniformBlockIndicesByBinding);
                samplerNameByBinding = std::move(other.samplerNameByBinding);
                samplerUniformLocationByBinding = std::move(other.samplerUniformLocationByBinding);
                samplerTextureTargetByBinding = std::move(other.samplerTextureTargetByBinding);
+                samplerNumericDomainByBinding = std::move(other.samplerNumericDomainByBinding);
+                storageImageFormatByBinding = std::move(other.storageImageFormatByBinding);
+                storageImageUsesBindingFormatByBinding =
+                    std::move(other.storageImageUsesBindingFormatByBinding);
                storageBlockNameByBinding = std::move(other.storageBlockNameByBinding);
                storageBlockIndexByBinding = std::move(other.storageBlockIndexByBinding);
+                hasStorageImages = other.hasStorageImages;
                globalUboBinding = other.globalUboBinding;
                activeVertexInputLocationMask = other.activeVertexInputLocationMask;
                vertexInputTypes = other.vertexInputTypes;
@@ -128,15 +184,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                rasterizationProducerStage = other.rasterizationProducerStage;
                producerOutputComponentCount = other.producerOutputComponentCount;
                fragmentInputComponentCount = other.fragmentInputComponentCount;
+                fragmentReplacesDepth = other.fragmentReplacesDepth;
+                lastUsedFrame = other.lastUsedFrame;
                other.hash = 0;
                other.descriptorSetLayout = VK_NULL_HANDLE;
                other.pipelineLayout = VK_NULL_HANDLE;
+                other.hasStorageImages = false;
                other.globalUboBinding = -1;
                other.activeVertexInputLocationMask = 0;
                other.activeFragmentOutputLocationMask = 0;
                other.rasterizationProducerStage = ShaderStage::Unknown;
                other.producerOutputComponentCount = 0;
                other.fragmentInputComponentCount = 0;
+                other.fragmentReplacesDepth = false;
+                other.lastUsedFrame = 0;
                return *this;
            }

@@ -166,10 +227,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            }
        };

+        // Notified when the OnFrameBoundary sweep destroys an aged-out cache entry,
+        // carrying the entry's content hash and the VkDescriptorSetLayout it owned.
+        // Dependent caches (compute pipelines, PipelineFactory entries, UniformManager's
+        // per-layout descriptor sets) must purge in the same step: after vkDestroy the
+        // layout handle value may be recycled for an unrelated layout, and the program
+        // hash may be re-inserted by a later rebuild of the same content.
+        class IEvictionObserver {
+        public:
+            virtual ~IEvictionObserver() = default;
+            virtual void OnProgramEvicted(HashType programHash, VkDescriptorSetLayout descriptorSetLayout) = 0;
+        };
+
        explicit ProgramFactory(VkDevice device, const VulkanRendererConfig& config, Uint32 maxBindings = 16,
-                                Bool shaderDrawParametersEnabled = false)
+                                Bool shaderDrawParametersEnabled = false,
+                                Bool unformattedFloatStorageImagesEnabled = false)
            : m_device(device), m_maxBindings(maxBindings), m_config(config),
-              m_shaderDrawParametersEnabled(shaderDrawParametersEnabled) {
+              m_shaderDrawParametersEnabled(shaderDrawParametersEnabled),
+              m_unformattedFloatStorageImagesEnabled(unformattedFloatStorageImagesEnabled) {
            VkProgramObject::s_device = device;
        }
        ~ProgramFactory() = default;
@@ -179,7 +254,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const VkProgramObject& GetOrCreateProgram(
            const MG_State::GLState::ProgramObject& program, CompileOptionFlags flags);

+        // Observer may be null (no notifications). Not owned.
+        void SetEvictionObserver(IEvictionObserver* observer) { m_evictionObserver = observer; }
+        // Frame boundary hook: ages the program cache and evicts long-unused entries
+        // (their command buffers retired many frames ago), mirroring
+        // VkRenderPassManager::OnPresent's sweep.
+        void OnFrameBoundary();
+
        static VkShaderStageFlagBits ToVkStage(ShaderStage stage);
+        static VkFormat ConvertSpirvImageFormatToVkFormat(SpvImageFormat format);
+        static SamplerNumericDomain UniformTypeToSamplerNumericDomain(GLenum glType);
+        // True when any entry point declares the DepthReplacing execution mode, i.e. the
+        // shader assigns gl_FragDepth. Exposed so the blended depth-write quirk's exemption
+        // can be pinned by tests. A false negative loses the exemption, so such a shader is
+        // stripped conservatively and forfeits its depth write.
+        static Bool ReflectedFragmentReplacesDepth(const SpvReflectShaderModule& reflectModule);
+        // True when an entry point reads the InstanceIndex builtin. Only gates a diagnostic:
+        // without shaderDrawParameters such a shader cannot have gl_InstanceID rebased.
+        static Bool ReflectedReadsInstanceIndexBuiltin(const SpvReflectShaderModule& reflectModule);

    private:
        struct ProgramLookupCache {
@@ -206,7 +298,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // True when the device enabled shaderDrawParameters; gates the InstanceIndex rebase pass
        // (which needs the DrawParameters capability / gl_BaseInstance builtin).
        Bool m_shaderDrawParametersEnabled = false;
+        // True only when the logical device enabled both
+        // shaderStorageImageReadWithoutFormat and shaderStorageImageWriteWithoutFormat.
+        Bool m_unformattedFloatStorageImagesEnabled = false;
        mutable ProgramLookupCache m_lastLookup;
+        // Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
+        Uint64 m_frameCounter = 0;
+        IEvictionObserver* m_evictionObserver = nullptr;
        static inline XXH64_state_t* m_hashState = XXH64_createState();
    };
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -247,6 +247,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        m_surfaceFormat = {createInfo.imageFormat, createInfo.imageColorSpace};
        m_extent = createInfo.imageExtent;
+        // The surface-space extent this swapchain was built from, i.e. before the
+        // quarter-turn swap above. Out-of-date checks must compare in THIS space: comparing a
+        // freshly queried currentExtent against the swapped m_extent flips axes every rotation
+        // and makes the comparison alternate forever.
+        m_surfaceExtent = defaultFramebufferExtent;
        m_preTransform = createInfo.preTransform;

        VK_VERIFY(vkCreateSwapchainKHR(device, &createInfo, nullptr, &m_swapchain));
@@ -35,6 +35,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        VkSwapchainKHR GetHandle() const { return m_swapchain; }
        const VkSurfaceFormatKHR& GetSurfaceFormat() const { return m_surfaceFormat; }
        VkExtent2D GetExtent() const { return m_extent; }
+        // Surface-space extent (before the pre-rotation quarter-turn swap) this swapchain was
+        // created from - the value to compare a freshly queried currentExtent against.
+        VkExtent2D GetSurfaceExtent() const { return m_surfaceExtent; }
        VkSurfaceTransformFlagBitsKHR GetPreTransform() const { return m_preTransform; }
        const Vector<VkImage>& GetImages() const { return m_images; }
        const Vector<VkImageView>& GetImageViews() const { return m_imageViews; }
@@ -63,6 +66,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        VkSwapchainKHR m_swapchain = VK_NULL_HANDLE;
        VkSurfaceFormatKHR m_surfaceFormat{};
        VkExtent2D m_extent{};
+        VkExtent2D m_surfaceExtent{};
        VkSurfaceTransformFlagBitsKHR m_preTransform = VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR;
        Vector<VkImage> m_images;
        Vector<VkImageView> m_imageViews;
@@ -13,8 +13,10 @@
 #include "MG_State/GLState/ProgramState/ProgramObject.h"
 #include "MG_State/GLState/TextureState/TextureObject2D.h"
 #include "MG_State/GLState/TextureState/TextureObjectBuffer.h"
+#include "MG_Util/Converters/GLToMG/TextureEnumConverter.h"
 #include "MG_Util/Converters/MGToStr/FramebufferEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
+#include <vulkan/utility/vk_format_utils.h>
 #include "MG_Util/Metrics/TextureMetrics.h"
 #include <Config.h>
 #include <cstdio>
@@ -78,6 +80,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return uniformUnit >= 0 ? uniformUnit : 0;
    }

+    VkFormat UniformManager::ResolveStorageImageViewFormat(VkFormat reflectedFormat, GLenum bindingFormat,
+                                                           VkFormat resourceFormat, Bool useBindingFormat) {
+        if (useBindingFormat) {
+            const TextureInternalFormat bindingInternalFormat =
+                MG_Util::ConvertGLEnumToTextureInternalFormat(bindingFormat);
+            return MG_Util::ConvertTextureInternalFormatToVkEnum(bindingInternalFormat);
+        }
+        return reflectedFormat != VK_FORMAT_UNDEFINED ? reflectedFormat : resourceFormat;
+    }
+
    Bool UniformManager::Initialize(VkDevice device, VkBufferManager* bufferManager,
                                             ProgramFactory* programFactory,
                                             VkDeviceSize minUniformBufferOffsetAlignment, Uint32 frameCount,
@@ -200,6 +212,40 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        }
    }

+    void UniformManager::OnDescriptorSetLayoutDestroyed(VkDescriptorSetLayout descriptorSetLayout) {
+        SizeT purgedSets = 0;
+        for (auto& frame : m_frames) {
+            const auto it = frame.descriptorSetCacheByLayout.find(descriptorSetLayout);
+            if (it == frame.descriptorSetCacheByLayout.end()) {
+                continue;
+            }
+            // Free the sets back to their pools and credit the bucket accounting, so
+            // program churn recycles pool capacity instead of abandoning the slots.
+            // GPU-safe: the layout only dies after >1024 idle frame boundaries, so no
+            // in-flight command buffer references these sets.
+            for (const auto& cached : it->second.sets) {
+                if (cached.set == VK_NULL_HANDLE) {
+                    continue;
+                }
+                vkFreeDescriptorSets(m_device, cached.pool, 1, &cached.set);
+                const auto bucket = std::find_if(
+                    frame.descriptorPools.begin(), frame.descriptorPools.end(),
+                    [&cached](const DescriptorPoolBucket& candidate) { return candidate.handle == cached.pool; });
+                if (bucket != frame.descriptorPools.end() && bucket->allocatedSets > 0) {
+                    --bucket->allocatedSets;
+                }
+            }
+            purgedSets += it->second.sets.size();
+            frame.descriptorSetCacheByLayout.erase(it);
+        }
+        if (purgedSets > 0) {
+            // The per-draw reuse memo folds the layout handle into its signature; drop
+            // it so a recycled handle value cannot revive a purged set mid-frame.
+            m_hasLastDescriptor = false;
+            MGLOG_D("UniformDescriptorBinder: freed %zu descriptor sets for destroyed layout", purgedSets);
+        }
+    }
+
    Bool UniformManager::ResolveSamplerDescriptor(VkCommandBuffer commandBuffer,
                                                            const MG_State::GLState::ProgramObject& program,
                                                            const ProgramFactory::VkProgramObject& programObj,
@@ -276,6 +322,55 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                            "ResolveSamplerDescriptor: invalid sampled image layout=%d for textureId=%d, binding=%u",
                            static_cast<Int>(resource->layout), texture->GetExternalIndex(), binding);
        }
+
+        MOBILEGL_ASSERT(binding < programObj.samplerNumericDomainByBinding.size(),
+                        "ResolveSamplerDescriptor: sampler numeric-domain binding %u out of range", binding);
+        const SamplerNumericDomain numericDomain = programObj.samplerNumericDomainByBinding[binding];
+        // Vulkan forbids linear filtering and anisotropy for integer sampled-image formats.
+        // Some desktop GL shader packs deliberately bit-read a mutable float texture through a
+        // usampler and still leave the texture's ordinary linear parameters in place; texelFetch
+        // ignores filtering, so a nearest VkSampler preserves the operation while keeping the
+        // descriptor valid.
+        const Bool forceNearestFiltering = numericDomain == SamplerNumericDomain::SignedInteger ||
+                                           numericDomain == SamplerNumericDomain::UnsignedInteger;
+        SamplerResolveMemo* viewFormatMemo =
+            binding < m_samplerResolveMemo.size() ? &m_samplerResolveMemo[binding] : nullptr;
+        VkFormat sampledViewFormat;
+        if (viewFormatMemo != nullptr && viewFormatMemo->viewFormatValid &&
+            viewFormatMemo->viewFormatSource == resource->format &&
+            viewFormatMemo->viewFormatDomain == numericDomain) {
+            sampledViewFormat = viewFormatMemo->viewFormat;
+        } else {
+            sampledViewFormat =
+                VkTextureManager::ResolveSampledImageViewFormat(resource->format, numericDomain);
+            if (viewFormatMemo != nullptr) {
+                viewFormatMemo->viewFormatSource = resource->format;
+                viewFormatMemo->viewFormatDomain = numericDomain;
+                viewFormatMemo->viewFormat = sampledViewFormat;
+                viewFormatMemo->viewFormatValid = true;
+            }
+        }
+        if (sampledViewFormat == VK_FORMAT_UNDEFINED) {
+            MGLOG_E("ResolveSamplerDescriptor: no compatible sampled view for binding=%u ('%s') "
+                    "textureId=%d imageFormat=%d numericDomain=%d",
+                    binding, programObj.samplerNameByBinding[binding].c_str(), texture->GetExternalIndex(),
+                    static_cast<Int>(resource->format), static_cast<Int>(numericDomain));
+            return false;
+        }
+        // No reinterpretation requested: bind the depth-or-color aspect view the sync above
+        // already produced instead of re-entering GetOrCreateSampledImageView's sync path.
+        const VkImageView sampledImageView =
+            sampledViewFormat == resource->format
+                ? resource->sampledView
+                : m_textureManager->GetOrCreateSampledImageView(*texture, sampledViewFormat);
+        if (sampledImageView == VK_NULL_HANDLE) {
+            MGLOG_E("ResolveSamplerDescriptor: failed to resolve sampled view for binding=%u ('%s') "
+                    "textureId=%d imageFormat=%d viewFormat=%d numericDomain=%d",
+                    binding, programObj.samplerNameByBinding[binding].c_str(), texture->GetExternalIndex(),
+                    static_cast<Int>(resource->format), static_cast<Int>(sampledViewFormat),
+                    static_cast<Int>(numericDomain));
+            return false;
+        }
        // Skip GetOrCreateSampler's per-draw key hash + map lookup when this binding's
        // sampler object and texture (both by lifetime id + version) are unchanged from the
        // last draw that resolved it: the resulting sampler key, and therefore the VkSampler
@@ -290,24 +385,32 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            const Uint16 samplerVersion = samplerToUse->GetVersion();
            const Uint64 textureLifetimeId = texture->GetLifetimeId();
            const Uint16 textureParamsVersion = texture->GetTextureParamsVersion();
+            // The sampler's LOD clamp depends on how many levels the sampled view exposes, and that
+            // follows uploads as well as GL parameters - so it belongs in the memo key too.
+            const Uint32 viewLevelCount = resource->sampledLevelCount;
            if (memo.valid && memo.samplerLifetimeId == samplerLifetimeId && memo.samplerVersion == samplerVersion &&
-                memo.textureLifetimeId == textureLifetimeId && memo.textureParamsVersion == textureParamsVersion) {
+                memo.textureLifetimeId == textureLifetimeId && memo.textureParamsVersion == textureParamsVersion &&
+                memo.forceNearestFiltering == forceNearestFiltering && memo.viewLevelCount == viewLevelCount) {
                resolvedSampler = memo.sampler;
            } else {
-                resolvedSampler = m_samplerManager->GetOrCreateSampler(*samplerToUse, *texture);
+                resolvedSampler = m_samplerManager->GetOrCreateSampler(*samplerToUse, *texture,
+                                                                       forceNearestFiltering, viewLevelCount);
                memo.samplerLifetimeId = samplerLifetimeId;
                memo.samplerVersion = samplerVersion;
                memo.textureLifetimeId = textureLifetimeId;
                memo.textureParamsVersion = textureParamsVersion;
+                memo.forceNearestFiltering = forceNearestFiltering;
+                memo.viewLevelCount = viewLevelCount;
                memo.sampler = resolvedSampler;
                memo.valid = true;
            }
        } else {
-            resolvedSampler = m_samplerManager->GetOrCreateSampler(*samplerToUse, *texture);
+            resolvedSampler = m_samplerManager->GetOrCreateSampler(*samplerToUse, *texture, forceNearestFiltering,
+                                                                   resource->sampledLevelCount);
        }
        outImageInfo = {
            .sampler = resolvedSampler,
-            .imageView = resource->sampledView != VK_NULL_HANDLE ? resource->sampledView : resource->fullView,
+            .imageView = sampledImageView,
            .imageLayout = resource->layout,
        };
        return outImageInfo.sampler != VK_NULL_HANDLE;
@@ -344,6 +447,106 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return outImageInfo.sampler != VK_NULL_HANDLE;
    }

+    namespace {
+        // fp16 carries an 11-bit mantissa, so an 8-bit normalized channel round-trips exactly.
+        // Anything wider - 16-bit normalized, half float, full float, and every packed HDR
+        // encoding - holds precision or range that relaxing the arithmetic would throw away.
+        Bool IsLowPrecisionNormalizedFormat(VkFormat format) {
+            if (format == VK_FORMAT_UNDEFINED) return false;
+            if (!vkuFormatIsUNORM(format) && !vkuFormatIsSNORM(format) && !vkuFormatIsSRGB(format)) {
+                return false;
+            }
+            const struct VKU_FORMAT_INFO info = vkuGetFormatInfo(format);
+            for (Uint32 i = 0; i < info.component_count; ++i) {
+                if (info.components[i].size > 8) return false;
+            }
+            return info.component_count > 0;
+        }
+    } // namespace
+
+    Bool UniformManager::DrawTargetIsLowPrecision(const MG_State::GLState::FramebufferObject* drawFramebuffer) {
+        // Default framebuffer: the swapchain is an 8-bit normalized surface.
+        if (drawFramebuffer == nullptr) return true;
+
+        Bool sawColour = false;
+        for (Int i = static_cast<Int>(FramebufferAttachmentType::Color0);
+             i < static_cast<Int>(FramebufferAttachmentType::FramebufferAttachmentTypeCount);
+             ++i) {
+            const auto& attachment =
+                drawFramebuffer->GetAttachment(static_cast<FramebufferAttachmentType>(i));
+            VkFormat format = VK_FORMAT_UNDEFINED;
+            if (const auto& texture = attachment.GetTexture()) {
+                format = MG_Util::ConvertTextureInternalFormatToVkEnum(texture->GetFormat());
+            } else if (const auto& renderbuffer = attachment.GetRenderbuffer()) {
+                format = MG_Util::ConvertTextureInternalFormatToVkEnum(
+                    renderbuffer->GetInternalFormat());
+            } else {
+                continue;
+            }
+            if (!IsLowPrecisionNormalizedFormat(format)) return false;
+            sawColour = true;
+        }
+        return sawColour;
+    }
+
+    Bool UniformManager::ProgramSamplesOnlyLowPrecisionTextures(
+        const MG_State::GLState::ProgramObject& program, const ProgramFactory::VkProgramObject& programObj) {
+        for (Uint32 binding = 0; binding < programObj.bindingKinds.size(); ++binding) {
+            if (programObj.bindingKinds[binding] != ProgramFactory::DescriptorBindingKind::CombinedImageSampler) {
+                continue;
+            }
+            const auto* texture = ResolveSamplerTextureRaw(program, programObj, binding);
+            // An unresolvable binding is unknown territory, not licence to relax.
+            if (texture == nullptr) return false;
+            const VkFormat format =
+                MG_Util::ConvertTextureInternalFormatToVkEnum(texture->GetFormat());
+            if (!IsLowPrecisionNormalizedFormat(format)) return false;
+        }
+        return true;
+    }
+
+    Bool UniformManager::ProgramSamplesOnlySingleLevelTextures(
+        const MG_State::GLState::ProgramObject& program, const ProgramFactory::VkProgramObject& programObj) {
+        Bool sawSampler = false;
+        for (Uint32 binding = 0; binding < programObj.bindingKinds.size(); ++binding) {
+            if (programObj.bindingKinds[binding] != ProgramFactory::DescriptorBindingKind::CombinedImageSampler) {
+                continue;
+            }
+            const auto* texture = ResolveSamplerTextureRaw(program, programObj, binding);
+            if (texture == nullptr) return false;
+            const auto& levelRange = texture->GetLevelRange();
+            if (levelRange.x() != levelRange.y()) return false;
+
+            // An explicit-LOD sample is a single filtered tap, so it also gives up anisotropic
+            // filtering - which a single-level view can still have. Resolve the sampler exactly
+            // the way ResolveSamplerDescriptor does and bail if anisotropy would apply.
+            const Int location = programObj.samplerUniformLocationByBinding[binding];
+            const Int unit = ResolveSamplerUnitIndex(program, location, binding);
+            const auto& samplerOverride = MG_State::pGLContext->GetTextureUnitObject(unit).GetSamplerObject();
+            const auto* effectiveSampler =
+                samplerOverride ? samplerOverride.get() : texture->GetSamplerObject().get();
+            if (effectiveSampler == nullptr) return false;
+            if (effectiveSampler->GetMaxAnisotropy() > 1.0f &&
+                effectiveSampler->GetMinFilter() == SamplerFilterMode::Linear &&
+                effectiveSampler->GetMagFilter() == SamplerFilterMode::Linear) {
+                return false;
+            }
+
+            // An explicit LOD 0 makes lambda exactly 0, which is the magnification side of the
+            // min/mag decision. That only matches the implicit form when lambda could not have been
+            // positive anyway (the LOD clamp already pins it at or below 0), or when the two
+            // filters are the same and the choice cannot be observed.
+            const Float effectiveMaxLod = effectiveSampler->GetMipmapMode() == SamplerMipmapMode::None
+                                              ? 0.0f
+                                              : effectiveSampler->GetMaxLod();
+            if (effectiveMaxLod > 0.0f && effectiveSampler->GetMinFilter() != effectiveSampler->GetMagFilter()) {
+                return false;
+            }
+            sawSampler = true;
+        }
+        return sawSampler;
+    }
+
    Bool UniformManager::ResolveSamplerTexture(const MG_State::GLState::ProgramObject& program,
                                                         const ProgramFactory::VkProgramObject& programObj, Uint32 binding,
                                                         SharedPtr<MG_State::GLState::ITextureObject>& outTexture) {
@@ -360,6 +563,12 @@ 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 +589,15 @@ 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,
@@ -558,9 +775,32 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        }

        const Uint32 mipLevel = static_cast<Uint32>(std::max<GLint>(0, imageBinding.Level));
-        VkImageView view = m_textureManager->GetOrCreateViewAtMipLevel(*imageBinding.Texture, mipLevel);
+        MOBILEGL_ASSERT(binding < programObj.storageImageFormatByBinding.size(),
+                        "ResolveStorageImageDescriptor: storage image format binding %u out of range", binding);
+        MOBILEGL_ASSERT(binding < programObj.storageImageUsesBindingFormatByBinding.size(),
+                        "ResolveStorageImageDescriptor: storage image format policy binding %u out of range",
+                        binding);
+        const VkFormat reflectedFormat = programObj.storageImageFormatByBinding[binding];
+        const Bool useBindingFormat = programObj.storageImageUsesBindingFormatByBinding[binding];
+        const VkFormat viewFormat = ResolveStorageImageViewFormat(
+            reflectedFormat, imageBinding.Format, resource->format, useBindingFormat);
+        if (viewFormat == VK_FORMAT_UNDEFINED) {
+            MGLOG_E("ResolveStorageImageDescriptor: unsupported glBindImageTexture format=0x%x "
+                    "for binding=%u imageUnit=%d textureId=%d bindingPolicy=%s",
+                    imageBinding.Format, binding, imageUnit, imageBinding.Texture->GetExternalIndex(),
+                    useBindingFormat ? "true" : "false");
+            return false;
+        }
+        const VkImageView view = m_textureManager->GetOrCreateStorageImageView(
+            *imageBinding.Texture, mipLevel, viewFormat, imageBinding.Layered != GL_FALSE, imageBinding.Layer);
        if (view == VK_NULL_HANDLE) {
-            view = resource->fullView;
+            MGLOG_E("ResolveStorageImageDescriptor: failed to resolve storage view textureId=%d mip=%u "
+                    "bindingFormat=0x%x imageFormat=%d reflectedFormat=%d selectedFormat=%d bindingPolicy=%s",
+                    imageBinding.Texture->GetExternalIndex(), mipLevel, imageBinding.Format,
+                    static_cast<Int>(resource->format), static_cast<Int>(reflectedFormat),
+                    static_cast<Int>(viewFormat),
+                    useBindingFormat ? "true" : "false");
+            return false;
        }
        outImageInfo.sampler = VK_NULL_HANDLE;
        outImageInfo.imageView = view;
@@ -599,7 +839,15 @@ 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);
@@ -610,9 +858,53 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return true;
    }

+    Bool UniformManager::CollectStorageImageTextures(
+        const MG_State::GLState::ProgramObject& program,
+        const ProgramFactory::VkProgramObject& programObj,
+        Vector<MG_State::GLState::ITextureObject*>& outTextures) const {
+        outTextures.clear();
+        MOBILEGL_ASSERT(MG_State::pGLContext != nullptr,
+                        "CollectStorageImageTextures: GL context is null");
+
+        const Uint32 bindingCount =
+            std::min<Uint32>(m_maxBindings, static_cast<Uint32>(programObj.bindingKinds.size()));
+        for (Uint32 binding = 0; binding < bindingCount; ++binding) {
+            if (programObj.bindingKinds[binding] != ProgramFactory::DescriptorBindingKind::StorageImage) {
+                continue;
+            }
+            if (binding >= programObj.samplerUniformLocationByBinding.size()) {
+                MGLOG_E("CollectStorageImageTextures: binding %u has no uniform-location mapping", binding);
+                return false;
+            }
+
+            const Int location = programObj.samplerUniformLocationByBinding[binding];
+            if (location < 0) {
+                MGLOG_E("CollectStorageImageTextures: binding %u has no image uniform location", binding);
+                return false;
+            }
+            const Int imageUnit = program.GetUniformSamplerOrImageUnitIndex(static_cast<Uint>(location));
+            if (imageUnit < 0 || imageUnit >= MG_State::GLState::TextureState::MAX_TEXTURE_IMAGE_UNITS) {
+                MGLOG_E("CollectStorageImageTextures: image unit %d is invalid for binding %u",
+                        imageUnit, binding);
+                return false;
+            }
+
+            auto* texture = MG_State::pGLContext->GetImageTextureBinding(imageUnit).Texture.get();
+            if (texture == nullptr) {
+                MGLOG_E("CollectStorageImageTextures: image unit %d is unbound for binding %u",
+                        imageUnit, binding);
+                return false;
+            }
+            if (std::find(outTextures.begin(), outTextures.end(), texture) == outTextures.end()) {
+                outTextures.push_back(texture);
+            }
+        }
+        return true;
+    }
+
    Bool UniformManager::ResolveUniformBufferPayload(const MG_State::GLState::ProgramObject& program,
                                                     const ProgramFactory::VkProgramObject& programObj, Uint32 binding,
-                                                     UboBindResult& out) const {
+                                                     Uint32 arrayElement, UboBindResult& out) const {
        const void* outData = nullptr;
        VkDeviceSize outSize = 0;

@@ -638,7 +930,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        MOBILEGL_ASSERT(binding < programObj.uniformBlockIndexByBinding.size(),
                        "ResolveUniformBufferPayload: UBO mapping binding %u out of range", binding);
-        const Int blockIndex = programObj.uniformBlockIndexByBinding[binding];
+        Int blockIndex = programObj.uniformBlockIndexByBinding[binding];
+        if (arrayElement > 0) {
+            const auto arrayIt = programObj.arrayedUniformBlockIndicesByBinding.find(binding);
+            const Bool elementValid = arrayIt != programObj.arrayedUniformBlockIndicesByBinding.end() &&
+                                      arrayElement < arrayIt->second.size();
+            MOBILEGL_ASSERT(elementValid,
+                            "ResolveUniformBufferPayload: UBO binding %u has no array element %u", binding,
+                            arrayElement);
+            if (!elementValid) {
+                return false;
+            }
+            blockIndex = arrayIt->second[arrayElement];
+        }
        MOBILEGL_ASSERT(blockIndex >= 0,
                        "ResolveUniformBufferPayload: no uniform block mapped to descriptor binding %u", binding);

@@ -746,6 +1050,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        VkDescriptorPoolCreateInfo poolInfo{};
        poolInfo.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO;
+        // FREE_DESCRIPTOR_SET_BIT lets a destroyed layout's cached sets be freed back
+        // (OnDescriptorSetLayoutDestroyed) so program churn recycles pool capacity.
+        // The cost is on set allocation only, which happens when a layout's per-frame
+        // cache grows - never on the per-draw reuse path.
+        poolInfo.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT;
        poolInfo.maxSets = maxSets;
        poolInfo.poolSizeCount = static_cast<Uint32>(std::size(poolSizes));
        poolInfo.pPoolSizes = poolSizes;
@@ -825,7 +1134,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        auto& frame = m_frames[frameIndex];
        auto& cache = frame.descriptorSetCacheByLayout[programObj.descriptorSetLayout];
        if (cache.cursor < cache.sets.size()) {
-            outDescriptorSet = cache.sets[cache.cursor++];
+            outDescriptorSet = cache.sets[cache.cursor++].set;
        } else {
            VkResult allocResult = AllocateDescriptorSetsFromActivePool(frameIndex, programObj, outDescriptorSet);
            if (allocResult == VK_ERROR_OUT_OF_POOL_MEMORY || allocResult == VK_ERROR_FRAGMENTED_POOL) {
@@ -839,7 +1148,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                return allocResult;
            }

-            cache.sets.push_back(outDescriptorSet);
+            // The successful allocation came from the bucket the alloc helper left
+            // active; record it so a layout-destroyed purge can free the set back.
+            cache.sets.push_back({outDescriptorSet, frame.descriptorPools[frame.activeDescriptorPoolIndex].handle});
            ++cache.cursor;
            MGLOG_D("UniformDescriptorBinder: cached descriptor set count for frame=%u grew to %zu", frameIndex,
                    cache.sets.size());
@@ -885,11 +1196,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        imageInfos.clear();
        texelBufferViews.clear();
        dynamicOffsets.clear();
+        // Arrayed UBO bindings contribute extra buffer infos and dynamic offsets; reserve for
+        // the worst case so the pBufferInfo pointers taken below never dangle on reallocation.
+        Uint32 uboArrayExtra = 0;
+        for (const auto& arrayEntry : programObj.arrayedUniformBlockIndicesByBinding) {
+            uboArrayExtra += static_cast<Uint32>(arrayEntry.second.size()) - 1u;
+        }
        writes.reserve(m_maxBindings);
-        bufferInfos.reserve(m_maxBindings);
+        bufferInfos.reserve(m_maxBindings + uboArrayExtra);
        imageInfos.reserve(m_maxBindings);
        texelBufferViews.reserve(m_maxBindings);
-        dynamicOffsets.reserve(programObj.dynamicBindings.size());
+        dynamicOffsets.reserve(programObj.dynamicBindings.size() + uboArrayExtra);

        const Uint32 bindingCount =
            std::min<Uint32>(m_maxBindings, static_cast<Uint32>(programObj.bindingKinds.size()));
@@ -907,40 +1224,51 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            write.descriptorCount = 1;

            if (kind == ProgramFactory::DescriptorBindingKind::UniformBufferDynamic) {
-                UboBindResult ubo{};
-                const Bool hasPayload = ResolveUniformBufferPayload(program, programObj, binding, ubo);
-                MOBILEGL_ASSERT(hasPayload && ubo.payload != nullptr && ubo.payloadSize > 0,
-                                "UniformDescriptorBinder::BindProgramUniformBuffers failed: missing UBO payload on binding %u",
-                                binding);
+                const Uint32 descriptorCount =
+                    binding < programObj.bindingDescriptorCounts.size()
+                        ? std::max<Uint32>(1, programObj.bindingDescriptorCounts[binding])
+                        : 1u;
+                const SizeT firstBufferInfoIndex = bufferInfos.size();
+                for (Uint32 element = 0; element < descriptorCount; ++element) {
+                    UboBindResult ubo{};
+                    const Bool hasPayload =
+                        ResolveUniformBufferPayload(program, programObj, binding, element, ubo);
+                    MOBILEGL_ASSERT(hasPayload && ubo.payload != nullptr && ubo.payloadSize > 0,
+                                    "UniformDescriptorBinder::BindProgramUniformBuffers failed: missing UBO payload on binding %u element %u",
+                                    binding, element);

-                VkDescriptorBufferInfo bufferInfo{};
-                // Keep offset 0 (sub-range selected via the dynamic offset) so the hashed bufferInfo
-                // is stable across draws and the descriptor-set reuse cache keeps hitting.
-                bufferInfo.offset = 0;
-                Uint32 dynOffset;
-                if (ubo.directBindable) {
-                    // Zero-copy: bind the app's resident VkBuffer directly, no per-draw memcpy.
-                    bufferInfo.buffer = ubo.buffer;
-                    bufferInfo.range = ubo.range;
-                    dynOffset = static_cast<Uint32>(ubo.dynamicOffset);
-                } else {
-                    BufferSlice slice{};
-                    if (!m_bufferManager->UploadTransient(BufferKind::Uniform, frameIndex, ubo.payload,
-                                                          ubo.payloadSize, m_minDynamicOffsetAlignment, slice)) {
-                        MOBILEGL_ASSERT(false, "UniformDescriptorBinder::BindProgramUniformBuffers failed: UBO upload failed on binding %u",
-                                binding);
-                        return false;
+                    VkDescriptorBufferInfo bufferInfo{};
+                    // Keep offset 0 (sub-range selected via the dynamic offset) so the hashed bufferInfo
+                    // is stable across draws and the descriptor-set reuse cache keeps hitting.
+                    bufferInfo.offset = 0;
+                    Uint32 dynOffset;
+                    if (ubo.directBindable) {
+                        // Zero-copy: bind the app's resident VkBuffer directly, no per-draw memcpy.
+                        bufferInfo.buffer = ubo.buffer;
+                        bufferInfo.range = ubo.range;
+                        dynOffset = static_cast<Uint32>(ubo.dynamicOffset);
+                    } else {
+                        BufferSlice slice{};
+                        if (!m_bufferManager->UploadTransient(BufferKind::Uniform, frameIndex, ubo.payload,
+                                                              ubo.payloadSize, m_minDynamicOffsetAlignment, slice)) {
+                            MOBILEGL_ASSERT(false, "UniformDescriptorBinder::BindProgramUniformBuffers failed: UBO upload failed on binding %u element %u",
+                                    binding, element);
+                            return false;
+                        }
+                        bufferInfo.buffer = slice.buffer;
+                        bufferInfo.range = ubo.payloadSize;
+                        dynOffset = static_cast<Uint32>(slice.offset);
                    }
-                    bufferInfo.buffer = slice.buffer;
-                    bufferInfo.range = ubo.payloadSize;
-                    dynOffset = static_cast<Uint32>(slice.offset);
+                    bufferInfos.push_back(bufferInfo);
+                    // Dynamic offsets are consumed in binding order, then array element order,
+                    // matching Vulkan's dynamic-offset consumption rules.
+                    dynamicOffsets.push_back(dynOffset);
                }
-                bufferInfos.push_back(bufferInfo);

                write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC;
-                write.pBufferInfo = &bufferInfos.back();
+                write.descriptorCount = descriptorCount;
+                write.pBufferInfo = &bufferInfos[firstBufferInfoIndex];
                writes.push_back(write);
-                dynamicOffsets.push_back(dynOffset);
            } else if (kind == ProgramFactory::DescriptorBindingKind::UniformTexelBuffer) {
                VkBufferView bufferView = VK_NULL_HANDLE;
                if (!ResolveTexelBufferDescriptor(program, programObj, binding, frameIndex, bufferView) ||
@@ -39,9 +39,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        void Shutdown();

        void BeginFrame(Uint32 frameIndex);
+        // A ProgramFactory eviction just destroyed this layout: purge every frame
+        // slot's cached descriptor sets for it, so a recycled handle value can never
+        // stale-hit sets written for the dead layout's bindings. The sets are
+        // vkFreeDescriptorSets'd back to their pools (created with
+        // FREE_DESCRIPTOR_SET_BIT) and the pool accounting is credited, so program
+        // churn recycles pool capacity instead of abandoning it. GPU-safe: the layout
+        // only dies after >1024 idle frame boundaries, so no in-flight command buffer
+        // references its sets. This is the only eviction path for the per-layout
+        // caches - a live layout's entry must never be purged (its sets would be
+        // unreachable pool slots), so there is deliberately no age-based sweep here.
+        void OnDescriptorSetLayoutDestroyed(VkDescriptorSetLayout descriptorSetLayout);
        Bool CollectSampledTextures(const MG_State::GLState::ProgramObject& program,
                                    const ProgramFactory::VkProgramObject& programObj,
                                    Vector<MG_State::GLState::ITextureObject*>& outTextures);
+        Bool CollectStorageImageTextures(const MG_State::GLState::ProgramObject& program,
+                                         const ProgramFactory::VkProgramObject& programObj,
+                                         Vector<MG_State::GLState::ITextureObject*>& outTextures) const;
        Bool BindProgramUniformBuffers(VkCommandBuffer commandBuffer,
                                       const MG_State::GLState::ProgramObject& program,
                                       const ProgramFactory::VkProgramObject& programObj,
@@ -49,6 +63,31 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                       VkPipelineBindPoint bindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
                                       const SamplerBindingOverride* samplerBindingOverride = nullptr);

+        // Pure format-policy helper kept public for host regression tests. Formatted storage
+        // images use their shader qualifier; transformed float images use glBindImageTexture's
+        // format and never silently fall back to the backing image format.
+        static VkFormat ResolveStorageImageViewFormat(VkFormat reflectedFormat, GLenum bindingFormat,
+                                                      VkFormat resourceFormat, Bool useBindingFormat);
+
+        // True when the program reads at least one sampler and every one of them is bound to a
+        // texture whose GL level range is a single level. Such a sampler resolves to
+        // minLod = maxLod = 0 (see VkSamplerManager::GetOrCreateSampler), so an implicit-LOD sample
+        // and an explicit LOD 0 sample must read the same texel - which is what makes the
+        // ExplicitLod0Sampling SPIR-V rewrite safe to request. Deliberately conservative: it reads
+        // only GL state, so a texture that ends up single-level for another reason (one uploaded
+        // level under a wide level range) merely misses the rewrite.
+        // True when every texture this program samples is an 8-bit-or-less normalized format, so
+        // relaxing the fragment stage to fp16 cannot lose a bit the texel ever carried. Says
+        // nothing about the render target - the caller must check that too.
+        static Bool ProgramSamplesOnlyLowPrecisionTextures(const MG_State::GLState::ProgramObject& program,
+                                                          const ProgramFactory::VkProgramObject& programObj);
+        // True when every colour attachment the draw writes is an 8-bit-or-less normalized
+        // format (nullptr = default framebuffer, which is). Blending happens at attachment
+        // precision, so a wider target must keep the fragment stage at full precision.
+        static Bool DrawTargetIsLowPrecision(const MG_State::GLState::FramebufferObject* drawFramebuffer);
+        static Bool ProgramSamplesOnlySingleLevelTextures(const MG_State::GLState::ProgramObject& program,
+                                                          const ProgramFactory::VkProgramObject& programObj);
+
    private:
        struct DescriptorPoolBucket {
            VkDescriptorPool handle = VK_NULL_HANDLE;
@@ -56,8 +95,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            Uint32 allocatedSets = 0;
        };

+        // A cached descriptor set together with the pool it was allocated from, so a
+        // layout-destroyed purge can vkFreeDescriptorSets it back and credit the
+        // owning bucket's accounting.
+        struct CachedDescriptorSet {
+            VkDescriptorSet set = VK_NULL_HANDLE;
+            VkDescriptorPool pool = VK_NULL_HANDLE;
+        };
+
        struct DescriptorSetCacheEntry {
-            Vector<VkDescriptorSet> sets;
+            Vector<CachedDescriptorSet> sets;
            Uint32 cursor = 0;
        };

@@ -107,7 +154,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        };
        Bool ResolveUniformBufferPayload(const MG_State::GLState::ProgramObject& program,
                                         const ProgramFactory::VkProgramObject& programObj, Uint32 binding,
-                                         UboBindResult& out) const;
+                                         Uint32 arrayElement, UboBindResult& out) const;
        Bool CreateDescriptorPool(Uint32 maxSets, VkDescriptorPool& outPool) const;
        Bool GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex);
        VkResult AllocateDescriptorSetsFromActivePool(
@@ -163,11 +210,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            Uint64 samplerLifetimeId = 0;
            Uint64 textureLifetimeId = 0;
            VkSampler sampler = VK_NULL_HANDLE;
+            Uint32 viewLevelCount = 0;
            Uint16 samplerVersion = 0;
            Uint16 textureParamsVersion = 0;
+            Bool forceNearestFiltering = false;
            Bool valid = false;
+            // ResolveSampledImageViewFormat is pure in (image format, numeric domain), but a
+            // domain mismatch walks a ~184-entry format table. Memo the resolution per binding
+            // so a reinterpreted sampler pays that scan once, not once per draw.
+            VkFormat viewFormatSource = VK_FORMAT_UNDEFINED;
+            SamplerNumericDomain viewFormatDomain = SamplerNumericDomain::Unknown;
+            VkFormat viewFormat = VK_FORMAT_UNDEFINED;
+            Bool viewFormatValid = false;
        };
        mutable Vector<SamplerResolveMemo> m_samplerResolveMemo;
    };
 } // namespace MobileGL::MG_Backend::DirectVulkan
-
@@ -32,6 +32,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            XXHASH_VERIFY(XXH64_update(m_hashState, &attr.IsBgra, sizeof(attr.IsBgra)));
            XXHASH_VERIFY(XXH64_update(m_hashState, &attr.Divisor, sizeof(attr.Divisor)));

+            // The buffer's heap address is an identity component of the key: a freed
+            // buffer's reused address can alias an old cache entry, but only under a
+            // byte-identical attribute layout - and the entry payload is a pure function
+            // of the hashed inputs, with the draw path re-resolving bindingBufferKeys
+            // against the live VAO attribute pointers, so an aliased hit returns exactly
+            // what a rebuild would. Address drift only grows the map; the OnFrameBoundary
+            // aging sweep bounds that.
            const SizeT bufferKey = reinterpret_cast<SizeT>(attr.Buffer.get());
            XXHASH_VERIFY(XXH64_update(m_hashState, &bufferKey, sizeof(bufferKey)));
        }
@@ -58,6 +65,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const MG_State::GLState::VertexArrayObject& vao, HashType hash) {
        auto it = m_cache.find(hash);
        if (it != m_cache.end()) {
+            it->second.lastUsedFrameBoundary = m_frameBoundaryCounter;
            return it->second;
        }

@@ -66,6 +74,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        Vector<SizeT> bindingBaseOffsets;
        Vector<Uint32> bindingAttributeLocations;
        Vector<Bool> bindingUsesClientMemory;
+        Vector<VertexStreamConversion> bindingConversions;
        Uint32 unsupportedAttribMask = 0;

        for (Uint32 location = 0; location < MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS; ++location) {
@@ -74,8 +83,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                continue;
            }

-            const auto vkFormat = ToVkVertexFormat(attr.Type, attr.Size, attr.Normalized, attr.IsInteger, attr.IsBgra);
-            if (vkFormat == VK_FORMAT_UNDEFINED) {
+            const VkFormat sourceVkFormat =
+                ToVkVertexFormat(attr.Type, attr.Size, attr.Normalized, attr.IsInteger, attr.IsBgra);
+            if (sourceVkFormat == VK_FORMAT_UNDEFINED) {
                MGLOG_E("Unsupported vertex attribute layout (location=%u, type=%s, size=%d): the array is "
                        "enabled but cannot be mapped to a VkFormat",
                        location, MG_Util::ConvertDataTypeToString(attr.Type).c_str(), attr.Size);
@@ -83,6 +93,33 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                continue;
            }

+            VkFormat vkFormat = sourceVkFormat;
+            VertexStreamConversion conversion = VertexStreamConversion::None;
+            if (!SupportsVertexBufferFormat(vkFormat)) {
+                if (IsScaledIntegerVertexFormat(vkFormat)) {
+                    const VkFormat fallbackFormat = ToFloat32VertexFormat(attr.Size);
+                    if (fallbackFormat != VK_FORMAT_UNDEFINED && SupportsVertexBufferFormat(fallbackFormat)) {
+                        vkFormat = fallbackFormat;
+                        conversion = VertexStreamConversion::ScaledIntegerToFloat32;
+                        MGLOG_W("Vertex attribute location=%u format=%d lacks "
+                                "VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT; using float32 stream format=%d "
+                                "(type=%s size=%d normalized=%s integer=%s)",
+                                location, static_cast<Int>(sourceVkFormat), static_cast<Int>(vkFormat),
+                                MG_Util::ConvertDataTypeToString(attr.Type).c_str(), attr.Size,
+                                attr.Normalized ? "true" : "false", attr.IsInteger ? "true" : "false");
+                    }
+                }
+
+                if (conversion == VertexStreamConversion::None) {
+                    MGLOG_E("Unsupported Vulkan vertex format (location=%u, format=%d, type=%s, size=%d): "
+                            "VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT is unavailable and no semantic fallback exists",
+                            location, static_cast<Int>(sourceVkFormat),
+                            MG_Util::ConvertDataTypeToString(attr.Type).c_str(), attr.Size);
+                    unsupportedAttribMask |= (1u << location);
+                    continue;
+                }
+            }
+
            const SizeT attribByteSize = GetAttributeByteSize(attr.Type, attr.Size, attr.IsBgra);
            if (attribByteSize == 0) {
                MGLOG_E("Vertex attribute with unknown component size (location=%u, type=%s): the array is "
@@ -92,8 +129,33 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                continue;
            }

-            const Uint32 stride =
+            const Uint32 sourceStride =
                attr.Stride > 0 ? static_cast<Uint32>(attr.Stride) : static_cast<Uint32>(attribByteSize);
+            const Bool packedAttribute = attr.Type == DataType::Int2101010Rev ||
+                                         attr.Type == DataType::Uint2101010Rev;
+            const SizeT requiredAlignment = packedAttribute ? attribByteSize : GetComponentSize(attr.Type);
+            // For a client-memory array attr.Offset holds the raw client pointer, and the
+            // draw path re-uploads the data to a 16-aligned transient slice with attribute
+            // offset 0, so only the stride can violate Vulkan's fetch alignment there.
+            const Bool clientMemoryAttribute = attr.Buffer == nullptr;
+            if (conversion == VertexStreamConversion::None && requiredAlignment > 1 &&
+                ((sourceStride % requiredAlignment) != 0 ||
+                 (!clientMemoryAttribute && (attr.Offset % requiredAlignment) != 0))) {
+                // GL accepts arbitrary byte strides and offsets. Core Vulkan vertex fetches do not
+                // unless VK_EXT_legacy_vertex_attributes is available, so deinterleave this one
+                // attribute into a tightly packed transient stream without changing its format.
+                conversion = VertexStreamConversion::Repack;
+                MGLOG_W("Vertex attribute location=%u uses Vulkan-incompatible alignment "
+                        "(offset=%zu stride=%u required=%zu); using a tightly packed stream",
+                        location, attr.Offset, sourceStride, requiredAlignment);
+            }
+
+            Uint32 stride = sourceStride;
+            if (conversion == VertexStreamConversion::Repack) {
+                stride = static_cast<Uint32>(attribByteSize);
+            } else if (conversion == VertexStreamConversion::ScaledIntegerToFloat32) {
+                stride = static_cast<Uint32>(attr.Size * static_cast<Int>(sizeof(Float)));
+            }
            const VkVertexInputRate inputRate =
                (attr.Divisor == 0) ? VK_VERTEX_INPUT_RATE_VERTEX : VK_VERTEX_INPUT_RATE_INSTANCE;

@@ -103,6 +165,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            bindingBaseOffsets.push_back(attr.Buffer ? attr.Offset : 0);
            bindingAttributeLocations.push_back(location);
            bindingUsesClientMemory.push_back(attr.Buffer == nullptr);
+            bindingConversions.push_back(conversion);
            builder.AddBinding(binding, stride, inputRate);
            builder.AddAttribute(location, binding, vkFormat, 0);
        }
@@ -111,12 +174,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        auto& entry = m_cache[hash];
        entry.hash = hash;
+        entry.lastUsedFrameBoundary = m_frameBoundaryCounter;
        entry.bindings = builder.GetBindings();
        entry.attributes = builder.GetAttributes();
        entry.bindingBufferKeys = std::move(bindingBufferKeys);
        entry.bindingBaseOffsets = std::move(bindingBaseOffsets);
        entry.bindingAttributeLocations = std::move(bindingAttributeLocations);
        entry.bindingUsesClientMemory = std::move(bindingUsesClientMemory);
+        entry.bindingConversions = std::move(bindingConversions);
        entry.unsupportedAttribMask = unsupportedAttribMask;
        entry.state = state;
        entry.state.pVertexBindingDescriptions = entry.bindings.empty() ? nullptr : entry.bindings.data();
@@ -124,6 +189,30 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return entry;
    }

+    void VertexInputStateFactory::OnFrameBoundary() {
+        ++m_frameBoundaryCounter;
+
+        // Sweep occasionally; evict entries whose last hit is far in the past.
+        // Erasure happens only here, never mid-frame: the draw path holds a
+        // reference into the current entry across its setup, and unordered_map
+        // erase would invalidate it. Entries are CPU-side only, so no GPU-idle
+        // proof is needed; an evicted entry that is used again is simply rebuilt
+        // from the VAO state (same hash, same content).
+        constexpr Uint64 kSweepInterval = 256;
+        constexpr Uint64 kRetireAgeBoundaries = 1024;
+        if ((m_frameBoundaryCounter % kSweepInterval) != 0) {
+            return;
+        }
+
+        for (auto it = m_cache.begin(); it != m_cache.end();) {
+            if (m_frameBoundaryCounter - it->second.lastUsedFrameBoundary > kRetireAgeBoundaries) {
+                it = m_cache.erase(it);
+            } else {
+                ++it;
+            }
+        }
+    }
+
    VkFormat VertexInputStateFactory::ToVkVertexFormat(DataType type, Int size, Bool normalized, Bool isInteger,
                                                       Bool isBgra) {
        if (isBgra) {
@@ -282,4 +371,47 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const SizeT componentSize = GetComponentSize(type);
        return componentSize == 0 ? 0 : componentSize * static_cast<SizeT>(size);
    }
+
+    Bool VertexInputStateFactory::IsScaledIntegerVertexFormat(VkFormat format) {
+        switch (format) {
+        case VK_FORMAT_R8_USCALED:
+        case VK_FORMAT_R8_SSCALED:
+        case VK_FORMAT_R8G8_USCALED:
+        case VK_FORMAT_R8G8_SSCALED:
+        case VK_FORMAT_R8G8B8_USCALED:
+        case VK_FORMAT_R8G8B8_SSCALED:
+        case VK_FORMAT_R8G8B8A8_USCALED:
+        case VK_FORMAT_R8G8B8A8_SSCALED:
+        case VK_FORMAT_R16_USCALED:
+        case VK_FORMAT_R16_SSCALED:
+        case VK_FORMAT_R16G16_USCALED:
+        case VK_FORMAT_R16G16_SSCALED:
+        case VK_FORMAT_R16G16B16_USCALED:
+        case VK_FORMAT_R16G16B16_SSCALED:
+        case VK_FORMAT_R16G16B16A16_USCALED:
+        case VK_FORMAT_R16G16B16A16_SSCALED:
+            return true;
+        default:
+            return false;
+        }
+    }
+
+    VkFormat VertexInputStateFactory::ToFloat32VertexFormat(Int componentCount) {
+        switch (componentCount) {
+        case 1: return VK_FORMAT_R32_SFLOAT;
+        case 2: return VK_FORMAT_R32G32_SFLOAT;
+        case 3: return VK_FORMAT_R32G32B32_SFLOAT;
+        case 4: return VK_FORMAT_R32G32B32A32_SFLOAT;
+        default: return VK_FORMAT_UNDEFINED;
+        }
+    }
+
+    Bool VertexInputStateFactory::SupportsVertexBufferFormat(VkFormat format) const {
+        if (m_physicalDevice == VK_NULL_HANDLE || format == VK_FORMAT_UNDEFINED) {
+            return false;
+        }
+        VkFormatProperties properties{};
+        vkGetPhysicalDeviceFormatProperties(m_physicalDevice, format, &properties);
+        return (properties.bufferFeatures & VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT) != 0;
+    }
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -19,14 +19,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    public:
        using HashType = Uint64;

+        enum class VertexStreamConversion : Uint8 {
+            None = 0,
+            Repack,
+            ScaledIntegerToFloat32,
+        };
+
        struct BackendVertexInputState {
            HashType hash = 0;
+            // Frame boundary of the last cache hit; entries idle past the
+            // OnFrameBoundary retirement age are evicted (CPU heap only).
+            Uint64 lastUsedFrameBoundary = 0;
            Vector<VkVertexInputBindingDescription> bindings;
            Vector<VkVertexInputAttributeDescription> attributes;
            Vector<SizeT> bindingBufferKeys;
            Vector<SizeT> bindingBaseOffsets;
            Vector<Uint32> bindingAttributeLocations;
            Vector<Bool> bindingUsesClientMemory;
+            Vector<VertexStreamConversion> bindingConversions;
            // Locations whose array is ENABLED but whose GL format has no VkFormat mapping. They are
            // absent from `attributes`, so without this mask the draw path cannot tell them apart from
            // a genuinely disabled array and would silently feed the shader the current attribute value.
@@ -36,8 +46,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            };
        };

-        explicit VertexInputStateFactory(const VulkanRendererConfig& config):
-            m_config(config) {}
+        VertexInputStateFactory(const VulkanRendererConfig& config, VkPhysicalDevice physicalDevice):
+            m_config(config), m_physicalDevice(physicalDevice) {}
        ~VertexInputStateFactory() = default;
        VertexInputStateFactory(const VertexInputStateFactory&) = delete;

@@ -48,6 +58,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const BackendVertexInputState& GetOrCreateVertexInputState(
            const MG_State::GLState::VertexArrayObject& vao, HashType hash);
        const BackendVertexInputState& GetOrCreateVertexInputState(const MG_State::GLState::VertexArrayObject& vao);
+        // Frame boundary hook: ages the cache and evicts entries not hit for many
+        // frames. The key mixes buffer heap addresses, so buffer/VAO churn keeps
+        // minting fresh keys; without eviction the map grows for the whole session.
+        // Entries hold no Vulkan handles (pipeline creation copies the descriptions)
+        // and the draw path's entry reference never spans a frame boundary, so
+        // eviction here needs no GPU-idle proof. Self-gated: one counter bump and
+        // compare except on sweep boundaries.
+        void OnFrameBoundary();
        static SizeT GetComponentSize(DataType type);
        // Tightly-packed byte size of one vertex element for this attribute: componentSize * size for
        // normal types, and 4 (one packed word) for the 2_10_10_10 types and GL_BGRA. Returns 0 for
@@ -56,9 +74,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {

    private:
        static VkFormat ToVkVertexFormat(DataType type, Int size, Bool normalized, Bool isInteger, Bool isBgra = false);
+        static Bool IsScaledIntegerVertexFormat(VkFormat format);
+        static VkFormat ToFloat32VertexFormat(Int componentCount);
+        Bool SupportsVertexBufferFormat(VkFormat format) const;

        const VulkanRendererConfig& m_config;
+        VkPhysicalDevice m_physicalDevice = VK_NULL_HANDLE;
        UnorderedMap<HashType, BackendVertexInputState> m_cache;
+        // Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
+        Uint64 m_frameBoundaryCounter = 0;
        static inline XXH64_state_t* m_hashState = XXH64_createState();
    };
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -141,6 +141,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        m_transientUploadArena.BeginFrame(frameIndex);
    }

+    void VkBufferManager::CollectAllDeferredReleases() {
+        for (Uint32 frameIndex = 0; frameIndex < m_deferredBufferReleases.size(); ++frameIndex) {
+            CollectDeferredReleases(frameIndex);
+        }
+        for (Uint32 frameIndex = 0; frameIndex < m_transientUploadArena.GetFrameCount(); ++frameIndex) {
+            m_transientUploadArena.CollectDeferredReleases(frameIndex);
+        }
+    }
+
    void VkBufferManager::NotifyDeviceIdle() {
        // Everything submitted so far has completed. Work recorded for the
        // current frame has not been submitted yet, so the current serial
@@ -197,16 +206,20 @@ 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) {
@@ -77,6 +77,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // Recreate all per-frame transient arenas
        Bool RecreateTransientArenas(Uint32 frameCount);
        void BeginFrame(Uint32 frameIndex);
+        // Drains every frame slot's deferred buffer/resource releases (and the
+        // transient arena's parked superseded blocks). Only valid when the
+        // caller has proven every queue submission complete; used by the
+        // present-less frame-boundary drain.
+        void CollectAllDeferredReleases();
        // All previously submitted GPU work has completed (vkDeviceWaitIdle).
        void NotifyDeviceIdle();
        // A frame slot's submission fence has been waited: every serial up to
@@ -124,7 +129,7 @@ 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);
@@ -157,6 +157,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return true;
    }

+    Bool VkBufferObject::Invalidate(VkDeviceSize size, VkDeviceSize offset) {
+        MOBILEGL_ASSERT(IsValid(), "VkBufferObject::Invalidate called on invalid buffer");
+        MOBILEGL_ASSERT(IsMapped(), "VkBufferObject::Invalidate requires mapped memory");
+        MOBILEGL_ASSERT(offset <= m_size, "VkBufferObject::Invalidate offset out of range");
+
+        const VkDeviceSize resolvedSize = size == VK_WHOLE_SIZE ? m_size - offset : size;
+        MOBILEGL_ASSERT(offset + resolvedSize <= m_size, "VkBufferObject::Invalidate range out of bounds");
+        if (resolvedSize == 0) {
+            return true;
+        }
+
+        const VkResult result = vmaInvalidateAllocation(m_allocator, m_allocation, offset, resolvedSize);
+        if (result != VK_SUCCESS) {
+            MGLOG_E("VkBufferObject::Invalidate failed: vmaInvalidateAllocation returned %d", result);
+            return false;
+        }
+        return true;
+    }
+
    BufferSlice VkBufferObject::GetSlice(VkDeviceSize offset, VkDeviceSize size) const {
        MOBILEGL_ASSERT(offset <= m_size, "VkBufferObject::GetSlice offset out of range");
        const VkDeviceSize resolvedSize = (size == VK_WHOLE_SIZE) ? (m_size - offset) : size;
@@ -44,6 +44,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        void* Map();
        void Unmap();
        Bool Upload(const void* data, VkDeviceSize size, VkDeviceSize offset = 0);
+        Bool Invalidate(VkDeviceSize size = VK_WHOLE_SIZE, VkDeviceSize offset = 0);

        VkBuffer GetHandle() const { return m_buffer; }
        VkDeviceSize GetSize() const { return m_size; }
@@ -180,6 +180,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        sampleCount = VK_SAMPLE_COUNT_1_BIT;
        internalFormat = TextureInternalFormat::Unknown;
        samples = 0;
+        deadSinceFrame = kNeverObservedDead;
    }

    VkRenderPassManager::VkRenderPassManager(VkDevice device,
@@ -206,6 +207,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            resource.Destroy(m_device, m_allocator);
        }
        m_renderbufferResources.clear();
+        CollectDeferredRenderbufferReleases(/*destroyAll=*/true); // caller guarantees device idle
        m_pendingRenderbufferClears.clear();
        RenderPassEntry::s_textureResourcesScratch.clear();
        s_activeRenderPass = {};
@@ -213,22 +215,75 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        m_rpFastValid = false;
    }

-    void VkRenderPassManager::CollectRenderbufferGarbage() {
-        Vector<MG_State::GLState::RenderbufferObject*> deadRenderbuffers;
-        deadRenderbuffers.reserve(m_renderbufferResources.size());
-        for (auto& [renderbuffer, resource] : m_renderbufferResources) {
-            const auto liveRenderbuffer = resource.renderbuffer.lock();
-            if (!liveRenderbuffer || liveRenderbuffer.get() != renderbuffer) {
-                deadRenderbuffers.emplace_back(renderbuffer);
-            }
+    Uint64 VkRenderPassManager::RetireAgeFrames() const {
+        // MaxFramesInFlight + 2 covers the frame ring plus one boundary for the
+        // recording-to-submit gap and one because OnPresent runs ahead of Present's
+        // fence wait; the floor of 8 keeps a margin over the default ring of 3 while
+        // still releasing multi-MB attachment memory promptly (the render-pass cache's
+        // 1024-frame retirement would pin it for no additional safety).
+        return std::max<Uint64>(8, static_cast<Uint64>(m_config.MaxFramesInFlight) + 2);
+    }
+
+    void VkRenderPassManager::DeferRenderbufferBackingRelease(RenderbufferResource& resource) {
+        // The superseded backing may still be referenced by in-flight command buffers
+        // (glRenderbufferStorage can respecify a renderbuffer drawn this very frame),
+        // so it is parked and destroyed only after RetireAgeFrames() boundaries.
+        if (resource.image == VK_NULL_HANDLE && resource.view == VK_NULL_HANDLE) {
+            return;
        }
-        for (auto* renderbuffer : deadRenderbuffers) {
-            auto resourceIt = m_renderbufferResources.find(renderbuffer);
-            if (resourceIt != m_renderbufferResources.end()) {
-                resourceIt->second.Destroy(m_device, m_allocator);
-                m_renderbufferResources.erase(resourceIt);
+        m_deferredRenderbufferReleases.push_back({resource.image, resource.allocation, resource.view, m_frameCounter});
+        resource.image = VK_NULL_HANDLE;
+        resource.allocation = nullptr;
+        resource.view = VK_NULL_HANDLE;
+    }
+
+    void VkRenderPassManager::CollectDeferredRenderbufferReleases(Bool destroyAll) {
+        if (m_deferredRenderbufferReleases.empty()) {
+            return;
+        }
+        const Uint64 retireAgeFrames = RetireAgeFrames();
+        std::erase_if(m_deferredRenderbufferReleases, [&](DeferredRenderbufferRelease& release) {
+            if (!destroyAll && m_frameCounter - release.deferredAtFrame < retireAgeFrames) {
+                return false;
            }
-            m_pendingRenderbufferClears.erase(renderbuffer);
+            if (release.view != VK_NULL_HANDLE) {
+                vkDestroyImageView(m_device, release.view, nullptr);
+            }
+            if (release.image != VK_NULL_HANDLE) {
+                vmaDestroyImage(m_allocator, release.image, release.allocation);
+            }
+            return true;
+        });
+    }
+
+    void VkRenderPassManager::CollectRenderbufferGarbage() {
+        // Two-phase reclamation: a dead renderbuffer's VkImage may still be referenced by
+        // command buffers submitted up to frames-in-flight frames ago (it was legally
+        // attached and drawn right up to its deletion), so the first observation of an
+        // expired weak reference only stamps the current frame counter; Destroy runs once
+        // enough frame boundaries have passed that the stamping frame's submission fence
+        // has provably been waited (see RetireAgeFrames).
+        const Uint64 retireAgeFrames = RetireAgeFrames();
+        for (auto it = m_renderbufferResources.begin(); it != m_renderbufferResources.end();) {
+            auto& resource = it->second;
+            const auto liveRenderbuffer = resource.renderbuffer.lock();
+            if (liveRenderbuffer && liveRenderbuffer.get() == it->first) {
+                resource.deadSinceFrame = RenderbufferResource::kNeverObservedDead;
+                ++it;
+                continue;
+            }
+            if (resource.deadSinceFrame == RenderbufferResource::kNeverObservedDead) {
+                resource.deadSinceFrame = m_frameCounter;
+                ++it;
+                continue;
+            }
+            if (m_frameCounter - resource.deadSinceFrame < retireAgeFrames) {
+                ++it;
+                continue;
+            }
+            m_pendingRenderbufferClears.erase(it->first);
+            resource.Destroy(m_device, m_allocator);
+            it = m_renderbufferResources.erase(it);
        }
    }

@@ -251,11 +306,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const auto internalFormat = renderbuffer->GetInternalFormat();
        const VkFormat format = MG_Util::ConvertTextureInternalFormatToVkEnum(internalFormat);
        const VkImageAspectFlags aspect = ResolveImageAspectMaskForFormat(format);
-        if ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) != 0) {
-            MGLOG_E("GetOrCreateRenderbufferResource: color renderbuffer %u is not supported by DirectVulkan render passes yet",
-                    renderbuffer->GetExternalIndex());
-            return nullptr;
-        }
+        // Renderbuffers are never sampled (GL has no way to bind one to a sampler), so the
+        // usage set is attachment + transfer: transfer covers readback (vkCmdCopyImageToBuffer),
+        // BlitFramebuffer, CopyTexImage sources, and out-of-render-pass clear materialization.
+        const VkImageUsageFlags imageUsage =
+            ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) != 0 ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT
+                                                       : VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) |
+            VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;

        auto& resource = m_renderbufferResources[renderbuffer.get()];
        const Bool needsCreate =
@@ -268,9 +325,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            resource.samples != renderbuffer->GetSamples();
        if (!needsCreate) {
            resource.renderbuffer = renderbuffer;
+            // A new renderbuffer at a recycled address may adopt a compatible entry that
+            // was already stamped dead; it is alive again, so cancel the aging.
+            resource.deadSinceFrame = RenderbufferResource::kNeverObservedDead;
            return &resource;
        }

+        // Respecify: park the old backing for aged destruction instead of destroying
+        // inline - it may still be referenced by in-flight command buffers.
+        DeferRenderbufferBackingRelease(resource);
        resource.Destroy(m_device, m_allocator);
        resource.renderbuffer = renderbuffer;

@@ -285,7 +348,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        imageInfo.format = format;
        imageInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
        imageInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
-        imageInfo.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
+        imageInfo.usage = imageUsage;
        imageInfo.samples = sampleCount;
        imageInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;

@@ -386,6 +449,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    void VkRenderPassManager::QueueRenderbufferClear(
        GLbitfield mask, const ClearFramebufferPayload& clearPayload,
        const MG_State::GLState::FramebufferObject& drawFbo) {
+        if ((mask & GL_COLOR_BUFFER_BIT) != 0) {
+            // Color renderbuffer draw buffers take the framebuffer-level clear too; texture
+            // attachments are skipped by the per-attachment overload's IsRenderbuffer guard.
+            for (const auto attachmentType : drawFbo.GetDrawBuffers()) {
+                if (attachmentType == FramebufferAttachmentType::None) {
+                    continue;
+                }
+                QueueRenderbufferClear(
+                    ClearAttachmentPayload{.mask = GL_COLOR_BUFFER_BIT, .color = clearPayload.color},
+                    drawFbo.GetAttachment(attachmentType));
+            }
+        }
        if ((mask & GL_DEPTH_BUFFER_BIT) != 0) {
            QueueRenderbufferClear(
                ClearAttachmentPayload{.mask = GL_DEPTH_BUFFER_BIT, .depth = clearPayload.depth},
@@ -602,6 +677,7 @@ 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 +699,15 @@ 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
@@ -678,6 +757,83 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // assuming default FBO has the right param
        for (Uint32 i = 0; i < colorAttachmentSlotCount; ++i) {
            auto drawbuf = drawbufs[i];
+
+            // Renderbuffer color attachments mirror the texture path below, with the
+            // resource (image/view/format/layout) coming from the render-pass manager's
+            // renderbuffer store instead of the texture manager.
+            if (drawbuf != FramebufferAttachmentType::None && !isDefaultFbo) {
+                const auto& rbAtt = fbo.GetAttachment(drawbuf);
+                if (rbAtt.IsRenderbuffer() && rbAtt.IsComplete()) {
+                    const auto& renderbuffer = rbAtt.GetRenderbuffer();
+                    auto* rbResource = GetOrCreateRenderbufferResource(renderbuffer);
+                    if (rbResource == nullptr || (rbResource->aspect & VK_IMAGE_ASPECT_COLOR_BIT) == 0) {
+                        MGLOG_E("GetOrCreateRenderPass: draw buffer slot %u on FBO %u has an unsupported color "
+                                "renderbuffer %u; using VK_ATTACHMENT_UNUSED",
+                                i, fbo.GetExternalIndex(), renderbuffer->GetExternalIndex());
+                        continue;
+                    }
+
+                    const Uint32 rbAttachmentIndex = static_cast<Uint32>(attachmentDescriptions.size());
+                    attachmentDescriptions.emplace_back();
+                    VkAttachmentDescription& rbDesc = attachmentDescriptions.back();
+
+                    ClearAttachmentPayload rbClearPayload{};
+                    Bool rbHasClear = GetPendingRenderbufferClear(renderbuffer.get(), rbClearPayload) &&
+                                      (rbClearPayload.mask & GL_COLOR_BUFFER_BIT) != 0;
+                    if (rbHasClear &&
+                        MG_Util::GetBaseInternalFormatComponentCount(renderbuffer->GetInternalFormat()) == 3) {
+                        // RGB renderbuffers are backed by an RGBA image; the missing alpha reads as 1.
+                        rbClearPayload.color =
+                            FloatVec4(rbClearPayload.color.x(), rbClearPayload.color.y(),
+                                      rbClearPayload.color.z(), 1.0f);
+                    }
+
+                    const VkImageLayout trackedRbLayout = rbResource->layout;
+                    rbDesc.flags = 0;
+                    rbDesc.format = rbResource->format;
+                    rbDesc.samples = rbResource->sampleCount;
+                    rbDesc.loadOp = rbHasClear ? VK_ATTACHMENT_LOAD_OP_CLEAR :
+                                    (trackedRbLayout == VK_IMAGE_LAYOUT_UNDEFINED ? VK_ATTACHMENT_LOAD_OP_DONT_CARE
+                                                                                  : VK_ATTACHMENT_LOAD_OP_LOAD);
+                    rbDesc.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
+                    rbDesc.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
+                    rbDesc.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
+                    rbDesc.initialLayout = (rbHasClear || trackedRbLayout == VK_IMAGE_LAYOUT_UNDEFINED) ?
+                        VK_IMAGE_LAYOUT_UNDEFINED : trackedRbLayout;
+                    rbDesc.finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
+                    adoptRenderPassSampleCount(rbResource->sampleCount, "color",
+                                               static_cast<Int>(renderbuffer->GetExternalIndex()));
+
+                    if (rbHasClear) {
+                        pendingClearAttachments.emplace_back(PendingClearAttachmentInfo {
+                            .attachmentIndex = rbAttachmentIndex,
+                            .colorAttachmentSlot = i,
+                            .renderbuffer = renderbuffer.get(),
+                            .hasInlinePayload = true,
+                            .inlinePayload = rbClearPayload,
+                        });
+                    }
+
+                    if (width == 0)
+                        width = static_cast<Int>(rbResource->extent.width);
+                    if (height == 0)
+                        height = static_cast<Int>(rbResource->extent.height);
+
+                    trackedAttachmentLayouts.emplace_back(TrackedAttachmentLayoutInfo {
+                        .target = TrackedAttachmentTarget::Renderbuffer,
+                        .renderbuffer = renderbuffer,
+                        .finalLayout = rbDesc.finalLayout,
+                    });
+                    textureResources.emplace_back(nullptr);
+                    attachmentViews.emplace_back(rbResource->view);
+                    MOBILEGL_ASSERT(attachmentViews.back() != VK_NULL_HANDLE,
+                                    "GetOrCreateRenderPass: renderbuffer view missing at color attachment %d", i);
+
+                    colorAttachmentRefs[i].attachment = rbAttachmentIndex;
+                    continue;
+                }
+            }
+
            auto* texture = ResolveCompleteColorAttachmentTexture(fbo, drawbuf, i);
            if (texture == nullptr)
                continue;
@@ -696,6 +852,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                case TextureTarget::Texture2D:
                case TextureTarget::Texture2DArray:
                case TextureTarget::Texture2DMultisample:
+                case TextureTarget::Texture2DMultisampleArray:
+                case TextureTarget::Texture3D:
+                case TextureTarget::TextureCubeMap:
+                case TextureTarget::TextureCubeMapArray:
                case TextureTarget::TextureRectangle: {
                    desc.flags = 0;
                    desc.format = isDefaultFbo ?
@@ -718,6 +878,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                    if (hasClear) {
                        pendingClearAttachments.emplace_back(PendingClearAttachmentInfo {
                            .attachmentIndex = attachmentIndex,
+                            .colorAttachmentSlot = i,
                            .key = VkClearManager::MakePendingClearKey(att)
                        });
                    }
@@ -974,6 +1135,52 @@ 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 +1190,8 @@ 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 +1222,7 @@ 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 +1232,54 @@ 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;
+
+        // Runs every frame boundary, ahead of the render-pass sweep gate below: the walk
+        // is O(#renderbuffer resources) — single digits in practice — and per-frame
+        // invocation keeps dead-resource reclaim latency at the aging bound instead of
+        // coupling it to renderbuffer *use* (the GetOrCreateRenderbufferResource call
+        // site never runs again once an app stops using renderbuffers).
+        CollectRenderbufferGarbage();
+        CollectDeferredRenderbufferReleases(/*destroyAll=*/false);
+
+        // 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;
+        }
+
+        // Collect the dying handles and notify once after the loop: pipelines hashed
+        // on them share the entries' >kRetireAgeFrames idleness (they are only bound
+        // by draws that hit those entries), so the observer may destroy them
+        // immediately - and a single batched notification costs one pipeline-cache
+        // scan instead of one per evicted pass.
+        Vector<VkRenderPass> destroyedRenderPasses;
+        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;
+                }
+                destroyedRenderPasses.push_back(it->second.renderPass);
+                it = m_renderPasses.erase(it);
+            } else {
+                ++it;
+            }
+        }
+        if (!destroyedRenderPasses.empty() && m_evictionObserver != nullptr) {
+            m_evictionObserver->OnRenderPassesDestroyed(destroyedRenderPasses);
+        }
+    }
+
    Bool VkRenderPassManager::BeginRenderPass(VkCommandBuffer commandBuffer, RenderPassEntry& renderPassEntry) {
        // TODO: Transition all the attachments into proper layout before starting the render pass
        VkRenderPassBeginInfo renderPassBeginInfo;
@@ -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() {
@@ -148,11 +157,31 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    class VkRenderPassManager {
    public:
        using HashType = Uint64;
+
+        // Notified once per OnPresent sweep with every aged-out entry's VkRenderPass
+        // value: pipelines are hashed on the raw handle, and once destroyed the value
+        // may be recycled for an incompatible pass, so dependent caches must purge
+        // everything keyed on them before any new pass can be created (the sweep and
+        // the notification run back-to-back with no creation in between; observers
+        // compare the values, never dereference them). Batched so a mass-idle cohort
+        // (shader-pack switch, dimension exit) costs the observer one pipeline-cache
+        // scan, not one per dying pass. The wholesale paths
+        // (Shutdown/RecreateSwapchain) do not notify - their callers already drop
+        // every pipeline outright.
+        class IEvictionObserver {
+        public:
+            virtual ~IEvictionObserver() = default;
+            virtual void OnRenderPassesDestroyed(const Vector<VkRenderPass>& renderPasses) = 0;
+        };
+
        VkRenderPassManager(VkDevice device,
            VkPhysicalDevice physicalDevice, VmaAllocator allocator, const VulkanRendererConfig& config,
            VkClearManager& clearManager, VkTextureManager& textureManager, SwapchainObject& swapchainObject);
        ~VkRenderPassManager();

+        // Observer may be null (no notifications). Not owned.
+        void SetEvictionObserver(IEvictionObserver* observer) { m_evictionObserver = observer; }
+
        Bool Initialize();
        void Shutdown();

@@ -166,6 +195,9 @@ 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 +210,9 @@ 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;
+        IEvictionObserver* m_evictionObserver = nullptr;

        // 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
@@ -197,7 +232,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        Uint64 m_rpFastRbEpoch = 0;
        Uint64 m_rpFastRenderPassHash = 0;

+    public:
        struct RenderbufferResource {
+            // deadSinceFrame sentinel: the owning weak reference has not been observed
+            // expired. Dead resources age past every in-flight frame before Destroy
+            // (see CollectRenderbufferGarbage); the GPU may still reference the image
+            // for frames-in-flight frames after the GL object dies.
+            static constexpr Uint64 kNeverObservedDead = UINT64_MAX;
+
            WeakPtr<MG_State::GLState::RenderbufferObject> renderbuffer;
            VkImage image = VK_NULL_HANDLE;
            VmaAllocation allocation = nullptr;
@@ -209,25 +251,47 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            VkSampleCountFlagBits sampleCount = VK_SAMPLE_COUNT_1_BIT;
            TextureInternalFormat internalFormat = TextureInternalFormat::Unknown;
            Int samples = 0;
+            // m_frameCounter value at which the weak reference was first seen expired.
+            Uint64 deadSinceFrame = kNeverObservedDead;

            void Destroy(VkDevice device, VmaAllocator allocator);
        };

+        // Public so the renderer's blit/copy/readback bindings can source renderbuffer
+        // attachments the same way texture attachments go through the texture manager.
+        RenderbufferResource* GetOrCreateRenderbufferResource(
+            const SharedPtr<MG_State::GLState::RenderbufferObject>& renderbuffer);
+        Bool GetPendingRenderbufferClear(MG_State::GLState::RenderbufferObject* renderbuffer,
+                                         ClearAttachmentPayload& outPayload) const;
+
+    private:
        struct PendingRenderbufferClear {
            WeakPtr<MG_State::GLState::RenderbufferObject> renderbuffer;
            ClearAttachmentPayload payload{};
        };

+        // A superseded renderbuffer backing (glRenderbufferStorage respecify) parked
+        // until enough frame boundaries have passed that no in-flight command buffer
+        // can still reference it; destroyed in OnPresent (see RetireAgeFrames).
+        struct DeferredRenderbufferRelease {
+            VkImage image = VK_NULL_HANDLE;
+            VmaAllocation allocation = nullptr;
+            VkImageView view = VK_NULL_HANDLE;
+            Uint64 deferredAtFrame = 0;
+        };
+
        UnorderedMap<MG_State::GLState::RenderbufferObject*, RenderbufferResource> m_renderbufferResources;
        UnorderedMap<MG_State::GLState::RenderbufferObject*, PendingRenderbufferClear> m_pendingRenderbufferClears;
+        Vector<DeferredRenderbufferRelease> m_deferredRenderbufferReleases;

-        RenderbufferResource* GetOrCreateRenderbufferResource(
-            const SharedPtr<MG_State::GLState::RenderbufferObject>& renderbuffer);
-        Bool GetPendingRenderbufferClear(MG_State::GLState::RenderbufferObject* renderbuffer,
-                                         ClearAttachmentPayload& outPayload) const;
        Bool HasPendingRenderbufferClear(
            const MG_State::GLState::FramebufferAttachmentObject& attachment) const;
        void CollectRenderbufferGarbage();
+        // Frame-boundary margin after which a resource last referenced by a retired
+        // GL object (or superseded backing) is provably past every in-flight frame.
+        Uint64 RetireAgeFrames() const;
+        void DeferRenderbufferBackingRelease(RenderbufferResource& resource);
+        void CollectDeferredRenderbufferReleases(Bool destroyAll);

        static inline XXH64_state_t* m_hashState = XXH64_createState();
        static inline ActiveRenderPassInfo s_activeRenderPass{};
@@ -51,6 +51,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        Float ResolveEffectiveMinLod(const MG_State::GLState::SamplerObject& sampler, Float effectiveMaxLod) {
            return std::min(sampler.GetMinLod(), effectiveMaxLod);
        }
+
+        // A single-level view can only ever deliver the base level, but the LOD clamp must not be
+        // collapsed to exactly 0: both GL and Vulkan pick magFilter over minFilter from the
+        // *clamped* lambda, so maxLod = 0 would make every fragment magnify and quietly retire the
+        // min filter. 0.25 is the value VkSamplerCreateInfo's own note prescribes for emulating
+        // GL's non-mipmapped minification - large enough for lambda to stay positive, small enough
+        // that a NEAREST mip mode still rounds down to level 0. Clamped rather than assigned, so a
+        // texture whose GL_TEXTURE_MAX_LOD really is 0 keeps magnifying as GL says it must.
+        Float ResolveSingleLevelMaxLod(const MG_State::GLState::SamplerObject& sampler, Bool singleLevelView) {
+            const Float maxLod = ResolveEffectiveMaxLod(sampler);
+            return singleLevelView ? std::min(maxLod, 0.25f) : maxLod;
+        }
    } // namespace

    Bool VkSamplerManager::Initialize(const InitInfo& initInfo) {
@@ -58,11 +70,26 @@ 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,
+                                                           Bool forceNearestFiltering) const {
+        if (!m_samplerAnisotropySupported) return 1.0f;
+        if (forceNearestFiltering) 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) {
@@ -74,13 +101,44 @@ namespace MobileGL::MG_Backend::DirectVulkan {

        m_device = VK_NULL_HANDLE;
        m_config = nullptr;
+        m_frameBoundaryCounter = 0;
+    }
+
+    void VkSamplerManager::OnFrameBoundary() {
+        ++m_frameBoundaryCounter;
+
+        // Sweep occasionally; destroy samplers whose last use is far past every
+        // in-flight frame. Destroy and erase must stay atomic, or Shutdown would
+        // double-free the handle; an evicted key that recurs simply re-creates
+        // its sampler on the next miss.
+        constexpr Uint64 kSweepInterval = 256;
+        constexpr Uint64 kRetireAgeBoundaries = 1024;
+        if ((m_frameBoundaryCounter % kSweepInterval) != 0) {
+            return;
+        }
+
+        for (auto it = m_samplers.begin(); it != m_samplers.end();) {
+            auto& entry = it->second;
+            if (m_frameBoundaryCounter - entry.lastUsedFrameBoundary > kRetireAgeBoundaries) {
+                if (m_device != VK_NULL_HANDLE && entry.handle != VK_NULL_HANDLE) {
+                    vkDestroySampler(m_device, entry.handle, nullptr);
+                }
+                it = m_samplers.erase(it);
+            } else {
+                ++it;
+            }
+        }
    }

    Uint64 VkSamplerManager::BuildSamplerKey(const MG_State::GLState::SamplerObject& sampler,
-                                             const MG_State::GLState::ITextureObject& texture) const {
+                                             const MG_State::GLState::ITextureObject& texture,
+                                             Bool forceNearestFiltering, Bool singleLevelView) const {
        MOBILEGL_ASSERT(m_config != nullptr, "VkSamplerManager::BuildSamplerKey: m_config is null");
        XXHASH_VERIFY(XXH64_reset(m_hashState, m_config->CacheVersion));

+        XXHASH_VERIFY(XXH64_update(m_hashState, &forceNearestFiltering, sizeof(forceNearestFiltering)));
+        XXHASH_VERIFY(XXH64_update(m_hashState, &singleLevelView, sizeof(singleLevelView)));
+
        const auto minFilter = sampler.GetMinFilter();
        XXHASH_VERIFY(XXH64_update(m_hashState, &minFilter, sizeof(minFilter)));
        const auto magFilter = sampler.GetMagFilter();
@@ -93,12 +151,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        XXHASH_VERIFY(XXH64_update(m_hashState, &wrapT, sizeof(wrapT)));
        const auto wrapR = sampler.GetWrapR();
        XXHASH_VERIFY(XXH64_update(m_hashState, &wrapR, sizeof(wrapR)));
-        const auto maxLod = ResolveEffectiveMaxLod(sampler);
+        const auto maxLod = ResolveSingleLevelMaxLod(sampler, singleLevelView);
        const auto minLod = ResolveEffectiveMinLod(sampler, maxLod);
        XXHASH_VERIFY(XXH64_update(m_hashState, &minLod, sizeof(minLod)));
        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, forceNearestFiltering);
+        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);
@@ -109,27 +172,44 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    }

    VkSampler VkSamplerManager::GetOrCreateSampler(const MG_State::GLState::SamplerObject& sampler,
-                                                   const MG_State::GLState::ITextureObject& texture) {
-        const Uint64 key = BuildSamplerKey(sampler, texture);
+                                                   const MG_State::GLState::ITextureObject& texture,
+                                                   Bool forceNearestFiltering, Uint32 viewLevelCount) {
+        // A view that exposes a single mip level has no second level to blend with, so GL's
+        // *_MIPMAP_* minification filters degenerate to plain filtering on the base level -
+        // sampling is unchanged by pinning the Vulkan sampler to NEAREST mip mode at LOD 0.
+        // It is not cosmetic: MobileGL backs such a view with a fully allocated mip chain whose
+        // tail is never written, and a LINEAR mip mode lets the texture unit issue the level+1
+        // fetch anyway. On Adreno that fetch lands in uninitialized UBWC pages (or past the
+        // allocation for a genuinely single-level image) and faults the GPU - the same failure
+        // the default-framebuffer blit shader had to work around with an explicit-LOD sample.
+        const Bool singleLevelView = viewLevelCount == 1;
+        const Uint64 key = BuildSamplerKey(sampler, texture, forceNearestFiltering, singleLevelView);
        auto it = m_samplers.find(key);
        if (it != m_samplers.end()) {
+            it->second.lastUsedFrameBoundary = m_frameBoundaryCounter;
            return it->second.handle;
        }

        VkSamplerCreateInfo samplerInfo{};
        samplerInfo.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
-        samplerInfo.magFilter = ToVkFilter(sampler.GetMagFilter());
-        samplerInfo.minFilter = ToVkFilter(sampler.GetMinFilter());
-        samplerInfo.mipmapMode = ToVkMipmapMode(sampler.GetMipmapMode());
+        samplerInfo.magFilter = forceNearestFiltering ? VK_FILTER_NEAREST : ToVkFilter(sampler.GetMagFilter());
+        samplerInfo.minFilter = forceNearestFiltering ? VK_FILTER_NEAREST : ToVkFilter(sampler.GetMinFilter());
+        samplerInfo.mipmapMode = (forceNearestFiltering || singleLevelView)
+                                     ? VK_SAMPLER_MIPMAP_MODE_NEAREST
+                                     : ToVkMipmapMode(sampler.GetMipmapMode());
        samplerInfo.addressModeU = ToVkAddressMode(sampler.GetWrapS());
        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, forceNearestFiltering);
+        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);
+        // Must match BuildSamplerKey's resolution exactly.
+        samplerInfo.maxLod = ResolveSingleLevelMaxLod(sampler, singleLevelView);
        samplerInfo.minLod = ResolveEffectiveMinLod(sampler, samplerInfo.maxLod);
        samplerInfo.borderColor = ResolveVkBorderColor(sampler, texture);
        samplerInfo.unnormalizedCoordinates = VK_FALSE;
@@ -141,6 +221,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        entry.handle = vkSampler;
        entry.externalIndex = sampler.GetExternalIndex();
        entry.version = sampler.GetVersion();
+        entry.lastUsedFrameBoundary = m_frameBoundaryCounter;
        m_samplers[key] = entry;
        return vkSampler;
    }
@@ -24,23 +24,47 @@ 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);
    void Shutdown();

+    // viewLevelCount is the mip-level count of the image view this sampler will be paired
+    // with; 0 means "unknown, do not narrow". See GetOrCreateSampler for why it matters.
    VkSampler GetOrCreateSampler(const MG_State::GLState::SamplerObject& sampler,
-                                 const MG_State::GLState::ITextureObject& texture);
+                                 const MG_State::GLState::ITextureObject& texture,
+                                 Bool forceNearestFiltering = false,
+                                 Uint32 viewLevelCount = 0);
+    // Frame boundary hook: ages the sampler cache and destroys samplers not used
+    // for many frames. The key hashes continuous float state (lodBias, LOD clamps,
+    // anisotropy), so an app animating those would otherwise mint an unbounded
+    // stream of never-destroyed VkSamplers and eventually exhaust the device's
+    // maxSamplerAllocationCount. A sampler idle for over a thousand frame
+    // boundaries cannot be referenced by any in-flight command buffer (frames in
+    // flight are single digits), and every descriptor set the GPU consumes is
+    // written that same frame with live handles (the per-binding resolve memo and
+    // descriptor-set reuse are both frame-reset), so destruction here needs no
+    // fence wait. Self-gated: one counter bump and compare except on sweep
+    // boundaries.
+    void OnFrameBoundary();

 private:
    struct SamplerCacheEntry {
        VkSampler handle = VK_NULL_HANDLE;
        Uint externalIndex = 0;
        Uint16 version = 0;
+        // Frame boundary of the last cache hit; entries idle past the
+        // OnFrameBoundary retirement age have their VkSampler destroyed.
+        Uint64 lastUsedFrameBoundary = 0;
    };

    Uint64 BuildSamplerKey(const MG_State::GLState::SamplerObject& sampler,
-                           const MG_State::GLState::ITextureObject& texture) const;
+                           const MG_State::GLState::ITextureObject& texture,
+                           Bool forceNearestFiltering, Bool singleLevelView) const;
    static VkFilter ToVkFilter(SamplerFilterMode mode);
    static VkSamplerMipmapMode ToVkMipmapMode(SamplerMipmapMode mode);
    static VkSamplerAddressMode ToVkAddressMode(SamplerWrapMode mode);
@@ -49,10 +73,20 @@ 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,
+                                        Bool forceNearestFiltering) 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;
+    // Monotonic frame-boundary counter (bumped in OnFrameBoundary) for cache aging.
+    Uint64 m_frameBoundaryCounter = 0;
    static inline XXH64_state_t* m_hashState = XXH64_createState();
 };
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -8,6 +8,8 @@

 #include "VkTextureManager.h"

+#include "ProgramFactory.h"
+
 #include "MG_State/GLState/Core.h"
 #include "MG_Util/Converters/MGToStr/TextureEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
@@ -17,6 +19,7 @@
 #include <cstdlib>
 #include <cstring>
 #include <memory>
+#include <vulkan/utility/vk_format_utils.h>

 namespace MobileGL::MG_Backend::DirectVulkan {
    // Compute shaders may legally sample framebuffer-attached textures (the GL feedback-loop rule
@@ -63,6 +66,59 @@ namespace MobileGL::MG_Backend::DirectVulkan {
               target == TextureUploadTarget::ProxyTexture2DMultisampleArray;
    }

+    static Bool IsMutableStorageImageFormat(VkFormat format) {
+        if (!vkuFormatIsColor(format) || vkuFormatIsCompressed(format)) {
+            return false;
+        }
+
+        // These are the uncompressed color compatibility classes covered by the core GLSL/SPIR-V
+        // storage-image formats. OpenGL mutable texture storage uses image-format compatibility by
+        // size, so a shader may legally reinterpret (for example) RGBA16_UNORM storage as rgba16f. Vulkan
+        // requires the image to be mutable and the view formats to share this exact compatibility
+        // class for the equivalent operation.
+        switch (vkuFormatCompatibilityClass(format)) {
+        case VKU_FORMAT_COMPATIBILITY_CLASS_8BIT:
+        case VKU_FORMAT_COMPATIBILITY_CLASS_16BIT:
+        case VKU_FORMAT_COMPATIBILITY_CLASS_32BIT:
+        case VKU_FORMAT_COMPATIBILITY_CLASS_64BIT:
+        case VKU_FORMAT_COMPATIBILITY_CLASS_128BIT:
+            return true;
+        default:
+            return false;
+        }
+    }
+
+    static Bool HasMatchingColorComponentLayout(VkFormat lhs, VkFormat rhs) {
+        const VKU_FORMAT_INFO lhsInfo = vkuGetFormatInfo(lhs);
+        const VKU_FORMAT_INFO rhsInfo = vkuGetFormatInfo(rhs);
+        if (lhsInfo.component_count == 0 || lhsInfo.component_count != rhsInfo.component_count ||
+            lhsInfo.texel_block_size != rhsInfo.texel_block_size ||
+            lhsInfo.texels_per_block != 1 || rhsInfo.texels_per_block != 1) {
+            return false;
+        }
+        for (Uint32 component = 0; component < lhsInfo.component_count; ++component) {
+            if (lhsInfo.components[component].type != rhsInfo.components[component].type ||
+                lhsInfo.components[component].size != rhsInfo.components[component].size) {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    static Bool FormatMatchesSamplerNumericDomain(VkFormat format, SamplerNumericDomain numericDomain) {
+        switch (numericDomain) {
+        case SamplerNumericDomain::Float:
+            return vkuFormatIsSampledFloat(format);
+        case SamplerNumericDomain::SignedInteger:
+            return vkuFormatIsSINT(format);
+        case SamplerNumericDomain::UnsignedInteger:
+            return vkuFormatIsUINT(format);
+        case SamplerNumericDomain::Unknown:
+            return true;
+        }
+        return false;
+    }
+
    static Bool TryResolveSampleCountFlagBits(Int requestedSamples, VkSampleCountFlagBits& outSampleCount) {
        switch (requestedSamples) {
        case 1:
@@ -319,7 +375,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        switch (format) {
        case TextureInternalFormat::RGB:
        case TextureInternalFormat::RGB8:
+        // Legacy low-bit RGB formats share the UNorm8 canonical shadow layout (see
+        // TextureFormatProcessor), so they upload exactly like RGB8 with an alpha expand.
+        case TextureInternalFormat::R3G3B2:
+        case TextureInternalFormat::RGB4:
+        case TextureInternalFormat::RGB5:
            return {VK_FORMAT_R8G8B8A8_UNORM, true, 1, {0xFF, 0x00, 0x00, 0x00}};
+        // Low-bit RGBA formats: UNorm8x4 canonical shadow, no expansion needed.
+        case TextureInternalFormat::RGBA2:
+        case TextureInternalFormat::RGBA4:
+        case TextureInternalFormat::RGB5A1:
+            return {VK_FORMAT_R8G8B8A8_UNORM, false, 0, {0, 0, 0, 0}};
+        // 10/12-bit RGB(A): UNorm16 canonical shadow.
+        case TextureInternalFormat::RGB10:
+        case TextureInternalFormat::RGB12:
+            return {VK_FORMAT_R16G16B16A16_UNORM, true, 2, {0xFF, 0xFF, 0x00, 0x00}};
+        case TextureInternalFormat::RGBA12:
+            return {VK_FORMAT_R16G16B16A16_UNORM, false, 0, {0, 0, 0, 0}};
        case TextureInternalFormat::SRGB8:
            return {VK_FORMAT_R8G8B8A8_SRGB, true, 1, {0xFF, 0x00, 0x00, 0x00}};
        case TextureInternalFormat::RGB8Snorm:
@@ -515,6 +587,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        m_allocator = initInfo.allocator;
        m_commandPool = initInfo.commandPool;
        m_graphicsQueue = initInfo.graphicsQueue;
+        m_imageFormatListSupported = initInfo.imageFormatListSupported;
        m_currentFrameIndex = 0;
        m_deferredReleases.clear();
        m_deferredReleases.resize(initInfo.frameCount);
@@ -537,6 +610,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        DestroyDeferredReleases();
        m_textureResources.clear();
        m_aliveObjects.clear();
+        m_storageImageTextures.clear();

        m_device = VK_NULL_HANDLE;
        m_physicalDevice = VK_NULL_HANDLE;
@@ -555,6 +629,26 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                        frameIndex, m_deferredViewReleases.size());
        m_currentFrameIndex = frameIndex;
        CollectDeferredReleases(frameIndex);
+
+        // Frame-boundary GC: every 64 frame boundaries (~1 s at 60 fps) bounds the reclaim
+        // latency for dead textures regardless of draw traffic — workloads that churn
+        // textures through clears/readbacks alone never reach the draw-gated
+        // CollectGarbage. Must run after CollectDeferredReleases above: the prune defers
+        // its releases into this frame's slot, which was just drained, so they are
+        // destroyed only after the slot's fence has been waited again one full frame-ring
+        // cycle from now (never while an in-flight frame may still reference them).
+        constexpr Uint32 kGcFrameInterval = 64;
+        ++m_gcFrameCounter;
+        if (m_gcFrameCounter % kGcFrameInterval == 0) {
+            PruneDeadTextures();
+        }
+    }
+
+    void VkTextureManager::CollectAllDeferredReleases() {
+        const SizeT frameCount = std::min(m_deferredReleases.size(), m_deferredViewReleases.size());
+        for (SizeT frameIndex = 0; frameIndex < frameCount; ++frameIndex) {
+            CollectDeferredReleases(static_cast<Uint32>(frameIndex));
+        }
    }

    void VkTextureManager::EraseTrackedTexture(const TextureIdentity& identity) {
@@ -564,6 +658,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            m_textureResources.erase(resourceIt);
        }
        m_aliveObjects.erase(identity);
+        m_storageImageTextures.erase(identity);
    }

    void VkTextureManager::PruneStaleTextureAliases(MG_State::GLState::ITextureObject* texture) {
@@ -630,9 +725,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // construction introduces a new identity. Doing this unconditionally made every
        // sampled-texture sync scan the entire alive-texture map per draw.
        if (aliveIt == m_aliveObjects.end()) {
+            WeakPtr<MG_State::GLState::ITextureObject> aliveTexture;
            const auto& liveTexture = MG_State::pGLContext->GetTextureObject(texture.GetExternalIndex());
            if (liveTexture && liveTexture.get() == &texture) {
-                m_aliveObjects[identity] = WeakPtr<MG_State::GLState::ITextureObject>(liveTexture);
+                aliveTexture = liveTexture;
+            } else {
+                // The name lookup legally fails while the object is alive: the name was
+                // deleted with the texture still attached to an FBO (the attachment's
+                // SharedPtr keeps it alive), or the name was reused by a new texture, or
+                // this is a default texture object (name 0 lives outside the name map).
+                // Register through the object's own control block so the resource created
+                // below still participates in weak-expiry GC instead of becoming an
+                // orphan no reclamation path can reach until Shutdown.
+                aliveTexture = texture.weak_from_this();
+            }
+            if (!aliveTexture.expired()) {
+                m_aliveObjects[identity] = Move(aliveTexture);
                PruneStaleTextureAliases(&texture);
            }
        }
@@ -767,6 +875,180 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return perMipSampledView;
    }

+    VkImageView VkTextureManager::GetOrCreateSampledImageView(MG_State::GLState::ITextureObject& texture,
+                                                               VkFormat format) {
+        TextureResource* resource = SyncTextureAndGetDescriptor(texture);
+        if (resource == nullptr || resource->image == VK_NULL_HANDLE ||
+            resource->sampledView == VK_NULL_HANDLE) {
+            return VK_NULL_HANDLE;
+        }
+
+        if (format == VK_FORMAT_UNDEFINED || format == resource->format) {
+            return resource->sampledView;
+        }
+        if (!AreSampledImageViewFormatsCompatible(resource->format, format)) {
+            MGLOG_E("%s: incompatible sampled image view format=%d for textureId=%d imageFormat=%d",
+                    __func__, static_cast<Int>(format), texture.GetExternalIndex(),
+                    static_cast<Int>(resource->format));
+            return VK_NULL_HANDLE;
+        }
+        if ((resource->imageCreateFlags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) == 0) {
+            MGLOG_E("%s: textureId=%d needs mutable image format=%d for sampled view format=%d",
+                    __func__, texture.GetExternalIndex(), static_cast<Int>(resource->format),
+                    static_cast<Int>(format));
+            return VK_NULL_HANDLE;
+        }
+
+        const TextureResource::SampledImageViewKey key{
+            .baseMipLevel = resource->sampledBaseMipLevel,
+            .levelCount = resource->sampledLevelCount,
+            .viewType = resource->viewType,
+            .format = format,
+        };
+        const auto existing = resource->alternateSampledViews.find(key);
+        if (existing != resource->alternateSampledViews.end()) {
+            return existing->second;
+        }
+
+        VkFormatProperties formatProperties{};
+        vkGetPhysicalDeviceFormatProperties(m_physicalDevice, format, &formatProperties);
+        if ((formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT) == 0) {
+            MGLOG_E("%s: sampled image view format=%d lacks VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT "
+                    "for textureId=%d (available=0x%x)",
+                    __func__, static_cast<Int>(format), texture.GetExternalIndex(),
+                    static_cast<Uint32>(formatProperties.optimalTilingFeatures));
+            return VK_NULL_HANDLE;
+        }
+
+        const TextureFormatInfo formatInfo = ResolveTextureFormatInfo(texture.GetFormat());
+        const VkComponentMapping sampledComponents = ResolveSampledViewComponents(texture, formatInfo);
+        const VkImageView view = CreateImageView(
+            resource->image, format, VK_IMAGE_ASPECT_COLOR_BIT, resource->viewType,
+            resource->sampledBaseMipLevel, resource->sampledLevelCount, 0, resource->arrayLayers,
+            &sampledComponents, VK_IMAGE_USAGE_SAMPLED_BIT);
+        if (view == VK_NULL_HANDLE) {
+            MGLOG_E("%s: failed to create sampled image view textureId=%d imageFormat=%d viewFormat=%d",
+                    __func__, texture.GetExternalIndex(), static_cast<Int>(resource->format),
+                    static_cast<Int>(format));
+            return VK_NULL_HANDLE;
+        }
+
+        resource->alternateSampledViews.emplace(key, view);
+        MGLOG_D("%s: created sampled image view textureId=%d imageFormat=%d viewFormat=%d mip=[%u,%u)",
+                __func__, texture.GetExternalIndex(), static_cast<Int>(resource->format),
+                static_cast<Int>(format), resource->sampledBaseMipLevel,
+                resource->sampledBaseMipLevel + resource->sampledLevelCount);
+        return view;
+    }
+
+    VkImageView VkTextureManager::GetOrCreateStorageImageView(MG_State::GLState::ITextureObject& texture,
+                                                               Uint32 mipLevel, VkFormat format,
+                                                               Bool layered, Int32 layer) {
+        TextureResource* resource = SyncTextureAndGetDescriptor(texture);
+        if (resource == nullptr || resource->image == VK_NULL_HANDLE || mipLevel >= resource->mipLevels ||
+            resource->sampleCount != VK_SAMPLE_COUNT_1_BIT ||
+            (resource->aspect & VK_IMAGE_ASPECT_COLOR_BIT) == 0) {
+            return VK_NULL_HANDLE;
+        }
+
+        if (format == VK_FORMAT_UNDEFINED) {
+            format = resource->format;
+        }
+        if (!AreStorageImageViewFormatsCompatible(resource->format, format)) {
+            MGLOG_E("%s: incompatible storage image view format=%d for textureId=%d imageFormat=%d",
+                    __func__, static_cast<Int>(format), texture.GetExternalIndex(),
+                    static_cast<Int>(resource->format));
+            return VK_NULL_HANDLE;
+        }
+        if (format != resource->format &&
+            (resource->imageCreateFlags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) == 0) {
+            MGLOG_E("%s: textureId=%d needs mutable image format=%d for storage view format=%d",
+                    __func__, texture.GetExternalIndex(), static_cast<Int>(resource->format),
+                    static_cast<Int>(format));
+            return VK_NULL_HANDLE;
+        }
+
+        Uint32 baseArrayLayer = 0;
+        Uint32 layerCount = resource->arrayLayers;
+        VkImageViewType viewType = resource->viewType;
+        if (!layered) {
+            switch (resource->viewType) {
+            case VK_IMAGE_VIEW_TYPE_1D_ARRAY:
+                viewType = VK_IMAGE_VIEW_TYPE_1D;
+                break;
+            case VK_IMAGE_VIEW_TYPE_2D_ARRAY:
+            case VK_IMAGE_VIEW_TYPE_CUBE:
+            case VK_IMAGE_VIEW_TYPE_CUBE_ARRAY:
+                viewType = VK_IMAGE_VIEW_TYPE_2D;
+                break;
+            case VK_IMAGE_VIEW_TYPE_3D:
+                MGLOG_E("%s: non-layered 3D storage views are unsupported for textureId=%d",
+                        __func__, texture.GetExternalIndex());
+                return VK_NULL_HANDLE;
+            default:
+                break;
+            }
+
+            if (viewType != resource->viewType) {
+                if (layer < 0 || static_cast<Uint32>(layer) >= resource->arrayLayers) {
+                    MGLOG_E("%s: storage image layer=%d is out of range for textureId=%d arrayLayers=%u",
+                            __func__, layer, texture.GetExternalIndex(), resource->arrayLayers);
+                    return VK_NULL_HANDLE;
+                }
+                baseArrayLayer = static_cast<Uint32>(layer);
+                layerCount = 1;
+            }
+        }
+
+        const Bool isFullResourceView = baseArrayLayer == 0 && layerCount == resource->arrayLayers &&
+                                        viewType == resource->viewType;
+        if (format == resource->format && isFullResourceView) {
+            return GetOrCreateViewAtMipLevel(texture, mipLevel);
+        }
+
+        const TextureResource::StorageImageViewKey key{
+            .mipLevel = mipLevel,
+            .baseArrayLayer = baseArrayLayer,
+            .layerCount = layerCount,
+            .viewType = viewType,
+            .format = format,
+        };
+        auto it = resource->storageImageViews.find(key);
+        if (it != resource->storageImageViews.end()) {
+            return it->second;
+        }
+
+        VkFormatFeatureFlags requiredFormatFeatures = VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT;
+        if (format != resource->format &&
+            (format == VK_FORMAT_R32_UINT || format == VK_FORMAT_R32_SINT)) {
+            requiredFormatFeatures |= VK_FORMAT_FEATURE_STORAGE_IMAGE_ATOMIC_BIT;
+        }
+        VkFormatProperties formatProperties{};
+        vkGetPhysicalDeviceFormatProperties(m_physicalDevice, format, &formatProperties);
+        if ((formatProperties.optimalTilingFeatures & requiredFormatFeatures) != requiredFormatFeatures) {
+            MGLOG_E("%s: storage image view format=%d lacks required features=0x%x for textureId=%d "
+                    "(available=0x%x)",
+                    __func__, static_cast<Int>(format), static_cast<Uint32>(requiredFormatFeatures),
+                    texture.GetExternalIndex(), static_cast<Uint32>(formatProperties.optimalTilingFeatures));
+            return VK_NULL_HANDLE;
+        }
+
+        const VkImageView view = CreateImageView(resource->image, format, VK_IMAGE_ASPECT_COLOR_BIT, viewType,
+                                                 mipLevel, 1, baseArrayLayer, layerCount, nullptr,
+                                                 VK_IMAGE_USAGE_STORAGE_BIT);
+        if (view == VK_NULL_HANDLE) {
+            MGLOG_E("%s: failed to create storage image view for textureId=%d mip=%u imageFormat=%d viewFormat=%d",
+                    __func__, texture.GetExternalIndex(), mipLevel, static_cast<Int>(resource->format),
+                    static_cast<Int>(format));
+            return VK_NULL_HANDLE;
+        }
+        resource->storageImageViews.emplace(key, view);
+        MGLOG_D("%s: created storage image view textureId=%d mip=%u imageFormat=%d viewFormat=%d",
+                __func__, texture.GetExternalIndex(), mipLevel, static_cast<Int>(resource->format),
+                static_cast<Int>(format));
+        return view;
+    }
+
    void VkTextureManager::UpdateTrackedImageLayout(MG_State::GLState::ITextureObject* texture, VkImageLayout newLayout) {
        MOBILEGL_ASSERT(texture != nullptr, "UpdateTrackedImageLayout: texture is null");
        auto it = m_textureResources.find(MakeTextureIdentity(texture));
@@ -910,6 +1192,47 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        return ok;
    }

+    void VkTextureManager::MarkStorageImageTexture(MG_State::GLState::ITextureObject& texture) {
+        m_storageImageTextures.insert(MakeTextureIdentity(&texture));
+    }
+
+    Bool VkTextureManager::NeedsStorageUsageUpgrade(MG_State::GLState::ITextureObject& texture) const {
+        const TextureIdentity identity = MakeTextureIdentity(&texture);
+        if (m_storageImageTextures.find(identity) == m_storageImageTextures.end()) {
+            return false;
+        }
+        const auto it = m_textureResources.find(identity);
+        // No image yet: the first sync creates it with STORAGE straight away, so there is nothing
+        // to preserve and nothing to order against.
+        return it != m_textureResources.end() && it->second.image != VK_NULL_HANDLE &&
+               !it->second.storageUsageResolved;
+    }
+
+    Bool VkTextureManager::NeedsStorageImagePreparation(MG_State::GLState::ITextureObject& texture) const {
+        const TextureIdentity identity = MakeTextureIdentity(&texture);
+        const auto it = m_textureResources.find(identity);
+        if (it == m_textureResources.end()) {
+            return true;
+        }
+        const TextureResource& resource = it->second;
+        if (resource.image == VK_NULL_HANDLE || resource.layout != VK_IMAGE_LAYOUT_GENERAL) {
+            return true;
+        }
+        // The image predates this texture's first image-unit binding, so it was created without
+        // STORAGE usage and has to be recreated - which is illegal inside a render pass.
+        if (!resource.storageUsageResolved &&
+            m_storageImageTextures.find(identity) != m_storageImageTextures.end()) {
+            return true;
+        }
+        // Mirror SyncTexture's cross-draw skip condition: any version drift means the sync
+        // path may upload or rebuild, both of which need the render pass ended first.
+        const auto* mipTexture = MG_State::GLState::AsMipmapTexture(&texture);
+        const Uint32 mipLevelCount = mipTexture != nullptr ? mipTexture->GetMipmapLevelCount() : 0u;
+        return resource.syncedContentVersion != texture.GetContentVersion() ||
+               resource.syncedTextureParamsVersion != texture.GetTextureParamsVersion() ||
+               resource.syncedMipLevelCount != mipLevelCount;
+    }
+
    Bool VkTextureManager::TransitionImageLayout(VkCommandBuffer commandBuffer, VkImage image,
                                                 VkImageLayout& trackedLayout, VkImageLayout newLayout,
                                                 VkPipelineStageFlags srcStageMask, VkPipelineStageFlags dstStageMask,
@@ -948,10 +1271,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    }

    SizeT VkTextureManager::CollectGarbage() {
+        // Draw-gated stagger (1 in 256 calls): keeps the per-draw cost at one counter
+        // bump. The guaranteed reclaim path is the frame-boundary prune in BeginFrame;
+        // this remains as a cheap assist so draw-heavy workloads reclaim sooner.
        m_gcCounter++;
        if (m_gcCounter != 0) {
            return 0;
        }
+        return PruneDeadTextures();
+    }
+
+    SizeT VkTextureManager::PruneDeadTextures() {
+        // Erasing entries would dangle the raw TextureResource pointers memoized for the
+        // current draw; every call path (BeginFrame, and CollectGarbage at the top of a
+        // freshly opened draw-sync scope) runs before any memo entry is recorded.
+        MOBILEGL_ASSERT(m_drawSyncedThisDraw.empty(),
+                        "PruneDeadTextures: draw-sync memo holds raw resource pointers an erase would dangle");

        Vector<MG_State::GLState::ITextureObject*> expiredTextures;
        expiredTextures.reserve(m_aliveObjects.size());
@@ -963,7 +1298,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        for (auto* texture : expiredTextures) {
            PruneStaleTextureAliases(texture);
        }
-        return expiredTextures.size();
+        SizeT prunedCount = expiredTextures.size();
+
+        // Orphan sweep: after the pass above, m_aliveObjects holds only live entries.
+        // Registration in SyncTextureAndGetDescriptor cannot fail for a SharedPtr-owned
+        // texture (weak_from_this fallback), so a resource whose identity has no alive
+        // entry has no trackable owner: its GL-side object is gone, or was never
+        // shared-owned, in which case recreation on a later sync is the safe fallback.
+        // Destruction goes through the per-frame deferred queues, never immediate.
+        Vector<TextureIdentity> orphanIdentities;
+        for (auto it = m_textureResources.begin(); it != m_textureResources.end(); ++it) {
+            if (m_aliveObjects.find(it->first) == m_aliveObjects.end()) {
+                orphanIdentities.emplace_back(it->first);
+            }
+        }
+        for (const auto& identity : orphanIdentities) {
+            EraseTrackedTexture(identity);
+        }
+        prunedCount += orphanIdentities.size();
+        return prunedCount;
    }

    Bool VkTextureManager::SyncTexture(MG_State::GLState::ITextureObject &texture,
@@ -977,7 +1330,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        const auto* syncingMipTexture = MG_State::GLState::AsMipmapTexture(&texture);
        const Uint32 syncingMipLevelCount =
            syncingMipTexture != nullptr ? syncingMipTexture->GetMipmapLevelCount() : 0u;
-        if (outResource.image != VK_NULL_HANDLE &&
+        // A pending storage-usage upgrade also has to bust the skip: nothing about the texture's
+        // content or params changed, but the image itself must be recreated with STORAGE usage
+        // before it can back an image-unit descriptor.
+        const Bool storageUpgradePending =
+            !outResource.storageUsageResolved &&
+            m_storageImageTextures.find(MakeTextureIdentity(&texture)) != m_storageImageTextures.end();
+        if (outResource.image != VK_NULL_HANDLE && !storageUpgradePending &&
            outResource.syncedContentVersion == syncingContentVersion &&
            outResource.syncedTextureParamsVersion == texture.GetTextureParamsVersion() &&
            outResource.syncedMipLevelCount == syncingMipLevelCount) {
@@ -1085,6 +1444,47 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            return false;
        }

+        const VkImageAspectFlags aspect = GetAspectMaskForFormat(format);
+        VkFormatProperties formatProperties{};
+        vkGetPhysicalDeviceFormatProperties(m_physicalDevice, format, &formatProperties);
+        // Only textures that have actually been bound to a GL image unit get STORAGE usage (and
+        // the MUTABLE_FORMAT it drags in for format-reinterpreting image views). Requesting it
+        // for every storage-capable colour texture costs real bandwidth: Adreno cannot keep UBWC
+        // compression on an image that may be written through a storage descriptor, so the whole
+        // render target - MC's included - runs uncompressed. MarkStorageImageTexture upgrades a
+        // texture before its first image-unit draw, and the usage below feeds the compatibility
+        // check so the upgrade recreates the image.
+        const Bool markedAsStorageImage =
+            m_storageImageTextures.find(MakeTextureIdentity(
+                const_cast<MG_State::GLState::ITextureObject*>(&texture))) != m_storageImageTextures.end();
+        // Storage-image CAPABILITY (does the format allow it at all) is deliberately separate from
+        // whether this texture actually needs the usage. MUTABLE_FORMAT keys off capability, as
+        // before: format-reinterpreting views are not a storage-only concern - the SAMPLED path
+        // needs them too (GetOrCreateSampledImageView bails out without it, see ~line 892), so
+        // tying MUTABLE_FORMAT to the image-unit mark would break sampled format reinterpretation
+        // for every texture that never becomes a storage image.
+        const Bool storageImageCapable =
+            !isMultisampleTexture &&
+            (aspect & VK_IMAGE_ASPECT_COLOR_BIT) != 0 &&
+            (formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT) != 0;
+        const Bool supportsStorageImage = storageImageCapable && markedAsStorageImage;
+        VkImageCreateFlags imageCreateFlags = shapeInfo.imageFlags;
+        if (storageImageCapable && IsMutableStorageImageFormat(format) &&
+            m_mutableFormatUnsupported.find(format) == m_mutableFormatUnsupported.end()) {
+            imageCreateFlags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+        }
+
+        VkImageUsageFlags desiredUsage =
+            VK_IMAGE_USAGE_SAMPLED_BIT |
+            (supportsStorageImage ? VK_IMAGE_USAGE_STORAGE_BIT : 0) |
+            ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : 0) |
+            (((aspect & VK_IMAGE_ASPECT_DEPTH_BIT) || (aspect & VK_IMAGE_ASPECT_STENCIL_BIT)) ?
+                 VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT :
+                 0);
+        if (!isMultisampleTexture) {
+            desiredUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+        }
+
        const Bool compatible = resource.image != VK_NULL_HANDLE && resource.format == format &&
                                resource.extent.width == static_cast<Uint32>(texelSize.x()) &&
                                resource.extent.height == static_cast<Uint32>(texelSize.y()) &&
@@ -1092,6 +1492,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                resource.arrayLayers == shapeInfo.arrayLayers &&
                                resource.viewType == shapeInfo.viewType &&
                                resource.sampleCount == resolvedSampleCount &&
+                                resource.imageCreateFlags == imageCreateFlags &&
+                                resource.usageFlags == desiredUsage &&
                                resource.mipLevels == backingMipLevels;
        if (compatible) {
            if (resource.perMipViews.size() != backingMipLevels) {
@@ -1100,6 +1502,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            if (resource.perMipSampledViews.size() != backingMipLevels) {
                resource.perMipSampledViews.resize(backingMipLevels, VK_NULL_HANDLE);
            }
+            // Keeping the image is itself the answer to the mark: either it already carries
+            // STORAGE, or this format can never carry it. Either way there is nothing left to
+            // recreate, so stop reporting the texture as needing preparation.
+            resource.storageUsageResolved = markedAsStorageImage;
            return true;
        }

@@ -1112,8 +1518,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            resource.arrayLayers == shapeInfo.arrayLayers &&
            resource.viewType == shapeInfo.viewType &&
            resource.sampleCount == resolvedSampleCount &&
+            resource.imageCreateFlags == imageCreateFlags &&
            resolvedSampleCount == VK_SAMPLE_COUNT_1_BIT &&
-            resource.mipLevels < backingMipLevels &&
+            // '<=' rather than '<': a storage-usage upgrade recreates the image with an
+            // unchanged mip count, and its contents (a render target's pixels live only on the
+            // GPU) still have to survive. The vkCmdCopyImage below copies min(mipLevels).
+            resource.mipLevels <= backingMipLevels &&
            resource.layout != VK_IMAGE_LAYOUT_UNDEFINED;

        std::unique_ptr<TextureResource> preservedResource;
@@ -1123,45 +1533,78 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            DeferResourceRelease(Move(resource));
        }

-        auto aspect = GetAspectMaskForFormat(format);
-
        VkImageCreateInfo imageInfo{};
        imageInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
-    imageInfo.flags = shapeInfo.imageFlags;
-    imageInfo.imageType = shapeInfo.imageType;
+        imageInfo.flags = imageCreateFlags;
+        imageInfo.imageType = shapeInfo.imageType;
        imageInfo.extent.width = static_cast<Uint32>(texelSize.x());
        imageInfo.extent.height = static_cast<Uint32>(texelSize.y());
-    imageInfo.extent.depth = shapeInfo.depth;
+        imageInfo.extent.depth = shapeInfo.depth;
        imageInfo.mipLevels = backingMipLevels;
-    imageInfo.arrayLayers = shapeInfo.arrayLayers;
+        imageInfo.arrayLayers = shapeInfo.arrayLayers;
        imageInfo.format = format;
        imageInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
        imageInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
-        VkFormatProperties formatProperties{};
-        vkGetPhysicalDeviceFormatProperties(m_physicalDevice, format, &formatProperties);
-        const Bool supportsStorageImage =
-            !isMultisampleTexture &&
-            (aspect & VK_IMAGE_ASPECT_COLOR_BIT) != 0 &&
-            (formatProperties.optimalTilingFeatures & VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT) != 0;
-        imageInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT |
-                          (supportsStorageImage ? VK_IMAGE_USAGE_STORAGE_BIT : 0) |
-                          ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : 0) |
-                          (((aspect & VK_IMAGE_ASPECT_DEPTH_BIT) || (aspect & VK_IMAGE_ASPECT_STENCIL_BIT)) ?
-                               VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT :
-                               0);
-        if (!isMultisampleTexture) {
-            imageInfo.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
-        }
+        imageInfo.usage = desiredUsage;
        imageInfo.samples = resolvedSampleCount;
-        if (isMultisampleTexture) {
+
+        // Bound the mutability. A blindly-mutable image has to be laid out so that ANY format in
+        // its compatibility class can be viewed, which costs bandwidth compression on tilers;
+        // naming the exact set instead lets the driver keep it. Only safe when that set really is
+        // exhaustive, so it is restricted to textures that are not image-unit bound: sampled views
+        // can only ever ask for ResolveSampledImageViewFormat's output, whereas glBindImageTexture
+        // may name any compatible format, which nothing here can enumerate ahead of time.
+        Vector<VkFormat> viewFormats;
+        VkImageFormatListCreateInfo formatListInfo{};
+        if (m_imageFormatListSupported && !supportsStorageImage &&
+            (imageInfo.flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) != 0) {
+            viewFormats.push_back(format);
+            for (const SamplerNumericDomain domain : {SamplerNumericDomain::Float,
+                                                      SamplerNumericDomain::SignedInteger,
+                                                      SamplerNumericDomain::UnsignedInteger}) {
+                const VkFormat viewFormat = ResolveSampledImageViewFormat(format, domain);
+                if (viewFormat == VK_FORMAT_UNDEFINED) {
+                    continue;
+                }
+                if (std::find(viewFormats.begin(), viewFormats.end(), viewFormat) == viewFormats.end()) {
+                    viewFormats.push_back(viewFormat);
+                }
+            }
+            formatListInfo.sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO;
+            formatListInfo.viewFormatCount = static_cast<Uint32>(viewFormats.size());
+            formatListInfo.pViewFormats = viewFormats.data();
+            imageInfo.pNext = &formatListInfo;
+        }
+
+        if (isMultisampleTexture || (imageInfo.flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) != 0) {
            VkImageFormatProperties imageFormatProperties{};
-            const VkResult imageFormatResult = vkGetPhysicalDeviceImageFormatProperties(
+            VkResult imageFormatResult = vkGetPhysicalDeviceImageFormatProperties(
                m_physicalDevice, format, imageInfo.imageType, imageInfo.tiling, imageInfo.usage,
                imageInfo.flags, &imageFormatProperties);
+            if (imageFormatResult != VK_SUCCESS && !isMultisampleTexture &&
+                (imageInfo.flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) != 0) {
+                // Losing reinterpreted views only degrades the formatless-image feature for
+                // this texture; failing creation would lose the texture entirely, so retry
+                // as a plain immutable-format image.
+                MGLOG_W("%s: mutable image format=%d is unsupported for textureId=%d; creating "
+                        "without VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT (format reinterpretation "
+                        "will be unavailable for it)",
+                        __func__, static_cast<Int>(format), texture.GetExternalIndex());
+                // Remember the verdict so later syncs of same-format textures neither retry
+                // the probe nor flag-mismatch against this image and recreate it.
+                m_mutableFormatUnsupported.insert(format);
+                imageInfo.flags &= ~VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+                imageCreateFlags = imageInfo.flags;
+                imageFormatResult = vkGetPhysicalDeviceImageFormatProperties(
+                    m_physicalDevice, format, imageInfo.imageType, imageInfo.tiling, imageInfo.usage,
+                    imageInfo.flags, &imageFormatProperties);
+            }
            if (imageFormatResult != VK_SUCCESS ||
-                (imageFormatProperties.sampleCounts & resolvedSampleCount) == 0) {
-                MGLOG_D("%s: sampleCount=%d is unsupported for textureId=%d target=%s format=%d usage=0x%x",
-                        __func__, texture.GetSamples(), texture.GetExternalIndex(),
+                (isMultisampleTexture && (imageFormatProperties.sampleCounts & resolvedSampleCount) == 0)) {
+                MGLOG_D("%s: image flags=0x%x sampleCount=%d are unsupported for textureId=%d target=%s "
+                        "format=%d usage=0x%x",
+                        __func__, static_cast<Uint32>(imageInfo.flags), texture.GetSamples(),
+                        texture.GetExternalIndex(),
                        MG_Util::ConvertTextureUploadTargetToString(uploadTarget).c_str(),
                        static_cast<Int>(format), static_cast<Uint32>(imageInfo.usage));
                return false;
@@ -1188,6 +1631,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        resource.aspect = aspect;
        resource.viewType = shapeInfo.viewType;
        resource.sampleCount = resolvedSampleCount;
+        resource.imageCreateFlags = imageCreateFlags;
+        resource.usageFlags = imageInfo.usage;
+        resource.storageUsageResolved = markedAsStorageImage;
        resource.syncedTextureParamsVersion = 0;

        if (preservedResource) {
@@ -1203,7 +1649,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
    void VkTextureManager::DeferResourceRelease(TextureResource&& resource) {
        if (resource.image == VK_NULL_HANDLE && resource.fullView == VK_NULL_HANDLE &&
            resource.sampledView == VK_NULL_HANDLE &&
-            resource.perMipViews.empty() && resource.perMipSampledViews.empty()) {
+            resource.perMipViews.empty() && resource.perMipSampledViews.empty() &&
+            resource.attachmentViews.empty() && resource.alternateSampledViews.empty() &&
+            resource.storageImageViews.empty()) {
            return;
        }

@@ -1299,6 +1747,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                sampledView = VK_NULL_HANDLE;
            }
        }
+        for (const auto& [_, sampledView] : resource.alternateSampledViews) {
+            DeferViewRelease(sampledView);
+        }
+        resource.alternateSampledViews.clear();

        const TextureFormatInfo formatInfo = ResolveTextureFormatInfo(texture.GetFormat());
        const VkComponentMapping sampledComponents = ResolveSampledViewComponents(texture, formatInfo);
@@ -1324,7 +1776,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                                  VkImageViewType viewType, Uint32 baseMipLevel, Uint32 levelCount,
                                                  Uint32 baseArrayLayer,
                                                  Uint32 layerCount,
-                                                  const VkComponentMapping* components) const {
+                                                  const VkComponentMapping* components,
+                                                  VkImageUsageFlags viewUsage) const {
        VkImageViewCreateInfo viewInfo{};
        viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
        viewInfo.image = image;
@@ -1340,6 +1793,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        viewInfo.subresourceRange.baseArrayLayer = baseArrayLayer;
        viewInfo.subresourceRange.layerCount = layerCount;

+        VkImageViewUsageCreateInfo usageInfo{};
+        if (viewUsage != 0) {
+            usageInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO;
+            usageInfo.usage = viewUsage;
+            viewInfo.pNext = &usageInfo;
+        }
+
        VkImageView view = VK_NULL_HANDLE;
        VK_VERIFY(vkCreateImageView(m_device, &viewInfo, nullptr, &view), "vkCreateImageView(texture)");
        return view;
@@ -1665,4 +2125,64 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        }
        return imageAspect;
    }
+
+    VkFormat VkTextureManager::ResolveSampledImageViewFormat(VkFormat imageFormat,
+                                                              SamplerNumericDomain numericDomain) {
+        // Depth/stencil images always sample through the existing depth-aspect sampledView.
+        // Combined formats (D24S8, D32FS8) are multi-numeric, so vkuFormatIsSampledFloat is
+        // false for them by design, yet their depth aspect reads as float in every GL depth
+        // texture mode; Vulkan also forbids reinterpreting them through color-class views.
+        // Integer domains keep the same view (pre-reinterpretation behavior for stencil-index
+        // style access) rather than failing the draw.
+        if (vkuFormatIsDepthOrStencil(imageFormat)) {
+            return imageFormat;
+        }
+        if (imageFormat == VK_FORMAT_UNDEFINED || numericDomain == SamplerNumericDomain::Unknown ||
+            FormatMatchesSamplerNumericDomain(imageFormat, numericDomain)) {
+            return imageFormat;
+        }
+        if (!IsMutableStorageImageFormat(imageFormat)) {
+            return VK_FORMAT_UNDEFINED;
+        }
+
+        // Preserve component ordering and bit widths. This selects R32_UINT for an R32_SFLOAT
+        // texture sampled by a usampler rather than an arbitrary member (such as
+        // R8G8B8A8_UINT) of Vulkan's broad 32-bit compatibility class.
+        for (Int candidateValue = static_cast<Int>(VK_FORMAT_R4G4_UNORM_PACK8);
+             candidateValue <= static_cast<Int>(VK_FORMAT_ASTC_12x12_SRGB_BLOCK);
+             ++candidateValue) {
+            const VkFormat candidate = static_cast<VkFormat>(candidateValue);
+            if (!IsMutableStorageImageFormat(candidate) ||
+                !FormatMatchesSamplerNumericDomain(candidate, numericDomain) ||
+                !HasMatchingColorComponentLayout(imageFormat, candidate) ||
+                !AreSampledImageViewFormatsCompatible(imageFormat, candidate)) {
+                continue;
+            }
+
+            // If an integer backing is intentionally bit-read through a float sampler, require
+            // a true floating-point view. Normalized/scaled views satisfy OpTypeFloat but apply
+            // an unrelated numeric conversion to those bits.
+            if (numericDomain == SamplerNumericDomain::Float && !vkuFormatIsSFLOAT(candidate)) {
+                continue;
+            }
+            return candidate;
+        }
+        return VK_FORMAT_UNDEFINED;
+    }
+
+    Bool VkTextureManager::AreSampledImageViewFormatsCompatible(VkFormat imageFormat, VkFormat viewFormat) {
+        if (imageFormat == viewFormat) {
+            return true;
+        }
+        return IsMutableStorageImageFormat(imageFormat) && IsMutableStorageImageFormat(viewFormat) &&
+               vkuFormatCompatibilityClass(imageFormat) == vkuFormatCompatibilityClass(viewFormat);
+    }
+
+    Bool VkTextureManager::AreStorageImageViewFormatsCompatible(VkFormat imageFormat, VkFormat viewFormat) {
+        if (imageFormat == viewFormat) {
+            return true;
+        }
+        return IsMutableStorageImageFormat(imageFormat) && IsMutableStorageImageFormat(viewFormat) &&
+               vkuFormatCompatibilityClass(imageFormat) == vkuFormatCompatibilityClass(viewFormat);
+    }
 } // namespace MobileGL::MG_Backend::DirectVulkan
@@ -13,12 +13,15 @@
 #include <MG_State/GLState/TextureState/TextureObject.h>
 #include <vk_mem_alloc.h>
 #include <unordered_map>
+#include <unordered_set>

 namespace MobileGL::MG_State::GLState {
 class ITextureObject;
 }

 namespace MobileGL::MG_Backend::DirectVulkan {
+enum class SamplerNumericDomain : Uint8;
+
 class VkTextureManager {
 public:
    // Monotonic epoch bumped whenever a texture VkImage is (re)created. The render-pass
@@ -50,6 +53,9 @@ public:
        VkCommandPool commandPool = VK_NULL_HANDLE;
        VkQueue graphicsQueue = VK_NULL_HANDLE;
        Uint32 frameCount = 0;
+        // VK_KHR_image_format_list is enabled: MUTABLE_FORMAT images can name the exact set of
+        // formats they will be viewed as, which is what lets a tiler keep them compressed.
+        Bool imageFormatListSupported = false;
    };

    struct TextureResource {
@@ -78,6 +84,61 @@ public:
            }
        };

+        struct StorageImageViewKey {
+            Uint32 mipLevel = 0;
+            Uint32 baseArrayLayer = 0;
+            Uint32 layerCount = 1;
+            VkImageViewType viewType = VK_IMAGE_VIEW_TYPE_2D;
+            VkFormat format = VK_FORMAT_UNDEFINED;
+
+            Bool operator==(const StorageImageViewKey& other) const {
+                return mipLevel == other.mipLevel &&
+                       baseArrayLayer == other.baseArrayLayer &&
+                       layerCount == other.layerCount &&
+                       viewType == other.viewType &&
+                       format == other.format;
+            }
+        };
+
+        struct SampledImageViewKey {
+            Uint32 baseMipLevel = 0;
+            Uint32 levelCount = 1;
+            VkImageViewType viewType = VK_IMAGE_VIEW_TYPE_2D;
+            VkFormat format = VK_FORMAT_UNDEFINED;
+
+            Bool operator==(const SampledImageViewKey& other) const {
+                return baseMipLevel == other.baseMipLevel &&
+                       levelCount == other.levelCount &&
+                       viewType == other.viewType &&
+                       format == other.format;
+            }
+        };
+
+        struct SampledImageViewKeyHash {
+            SizeT operator()(const SampledImageViewKey& key) const {
+                SizeT hash = std::hash<Uint32>{}(key.baseMipLevel);
+                hash ^= std::hash<Uint32>{}(key.levelCount) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.viewType)) +
+                        0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.format)) +
+                        0x9e3779b9u + (hash << 6) + (hash >> 2);
+                return hash;
+            }
+        };
+
+        struct StorageImageViewKeyHash {
+            SizeT operator()(const StorageImageViewKey& key) const {
+                SizeT hash = std::hash<Uint32>{}(key.mipLevel);
+                hash ^= std::hash<Uint32>{}(key.baseArrayLayer) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(key.layerCount) + 0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.viewType)) +
+                        0x9e3779b9u + (hash << 6) + (hash >> 2);
+                hash ^= std::hash<Uint32>{}(static_cast<Uint32>(key.format)) +
+                        0x9e3779b9u + (hash << 6) + (hash >> 2);
+                return hash;
+            }
+        };
+
        VkImage image = VK_NULL_HANDLE;
        VmaAllocation allocation = nullptr;
        VkImageView fullView = VK_NULL_HANDLE;
@@ -85,6 +146,8 @@ public:
        Vector<VkImageView> perMipViews;
        Vector<VkImageView> perMipSampledViews;
        UnorderedMap<AttachmentViewKey, VkImageView, AttachmentViewKeyHash> attachmentViews;
+        UnorderedMap<SampledImageViewKey, VkImageView, SampledImageViewKeyHash> alternateSampledViews;
+        UnorderedMap<StorageImageViewKey, VkImageView, StorageImageViewKeyHash> storageImageViews;
        VkImageLayout layout = VK_IMAGE_LAYOUT_UNDEFINED;
        VkExtent2D extent = {0, 0};
        Uint32 depth = 1;
@@ -96,6 +159,18 @@ public:
        VkImageAspectFlags aspect = VK_IMAGE_ASPECT_NONE;
        VkImageViewType viewType = VK_IMAGE_VIEW_TYPE_2D;
        VkSampleCountFlagBits sampleCount = VK_SAMPLE_COUNT_1_BIT;
+        VkImageCreateFlags imageCreateFlags = 0;
+        // Usage the live image was created with. STORAGE is only requested for textures that
+        // have actually been bound to a GL image unit, because on Adreno a storage-capable
+        // image loses UBWC bandwidth compression; a later image binding upgrades the usage
+        // and recreates the image, so the resolved usage has to be part of the compatibility
+        // check that decides whether the existing image can be kept.
+        VkImageUsageFlags usageFlags = 0;
+        // True once this image was (re)resolved while the texture was already marked as an
+        // image-unit texture. Distinguishes "not upgraded yet" from "cannot be upgraded"
+        // (a format whose optimalTilingFeatures lack STORAGE_IMAGE never gains the bit), so
+        // NeedsStorageImagePreparation cannot ask for a recreate that will never happen.
+        Bool storageUsageResolved = false;
        Uint16 syncedTextureParamsVersion = 0;
        // Snapshot of ITextureObject::GetContentVersion() at the last successful sync;
        // lets SyncTexture skip the whole re-check/re-upload when content is unchanged.
@@ -115,6 +190,8 @@ public:
            std::swap(this->perMipViews, that.perMipViews);
            std::swap(this->perMipSampledViews, that.perMipSampledViews);
            std::swap(this->attachmentViews, that.attachmentViews);
+            std::swap(this->alternateSampledViews, that.alternateSampledViews);
+            std::swap(this->storageImageViews, that.storageImageViews);
            std::swap(this->layout, that.layout);
            std::swap(this->extent, that.extent);
            std::swap(this->depth, that.depth);
@@ -126,6 +203,9 @@ public:
            std::swap(this->aspect, that.aspect);
            std::swap(this->viewType, that.viewType);
            std::swap(this->sampleCount, that.sampleCount);
+            std::swap(this->imageCreateFlags, that.imageCreateFlags);
+            std::swap(this->usageFlags, that.usageFlags);
+            std::swap(this->storageUsageResolved, that.storageUsageResolved);
            std::swap(this->syncedTextureParamsVersion, that.syncedTextureParamsVersion);
            std::swap(this->syncedContentVersion, that.syncedContentVersion);
            std::swap(this->syncedMipLevelCount, that.syncedMipLevelCount);
@@ -153,6 +233,16 @@ public:
                    vkDestroyImageView(s_device, attachmentView, nullptr);
                }
            }
+            for (const auto& [_, sampledView] : alternateSampledViews) {
+                if (sampledView != VK_NULL_HANDLE) {
+                    vkDestroyImageView(s_device, sampledView, nullptr);
+                }
+            }
+            for (const auto& [_, storageImageView] : storageImageViews) {
+                if (storageImageView != VK_NULL_HANDLE) {
+                    vkDestroyImageView(s_device, storageImageView, nullptr);
+                }
+            }
            if (image != VK_NULL_HANDLE && allocation != nullptr) {
                vmaDestroyImage(s_allocator, image, allocation);
            }
@@ -161,6 +251,8 @@ public:
            perMipViews.clear();
            perMipSampledViews.clear();
            attachmentViews.clear();
+            alternateSampledViews.clear();
+            storageImageViews.clear();
            image = VK_NULL_HANDLE;
            allocation = nullptr;
            layout = VK_IMAGE_LAYOUT_UNDEFINED;
@@ -174,6 +266,9 @@ public:
            aspect = VK_IMAGE_ASPECT_NONE;
            viewType = VK_IMAGE_VIEW_TYPE_2D;
            sampleCount = VK_SAMPLE_COUNT_1_BIT;
+            imageCreateFlags = 0;
+            usageFlags = 0;
+            storageUsageResolved = false;
            syncedTextureParamsVersion = 0;
            syncedContentVersion = 0;
            syncedMipLevelCount = 0;
@@ -190,6 +285,10 @@ public:
    Bool Initialize(const InitInfo& initInfo);
    void Shutdown();
    void BeginFrame(Uint32 frameIndex);
+    // Drains every frame slot's deferred image/view releases. Only valid when
+    // the caller has proven every queue submission complete; used by the
+    // present-less frame-boundary drain.
+    void CollectAllDeferredReleases();

    TextureResource* SyncTextureAndGetDescriptor(
        MG_State::GLState::ITextureObject& texture);
@@ -198,6 +297,9 @@ public:
                                                    Uint32 baseArrayLayer, Uint32 layerCount,
                                                    VkImageViewType viewType);
    VkImageView GetOrCreateSampledViewAtMipLevel(MG_State::GLState::ITextureObject& texture, Uint32 mipLevel);
+    VkImageView GetOrCreateSampledImageView(MG_State::GLState::ITextureObject& texture, VkFormat format);
+    VkImageView GetOrCreateStorageImageView(MG_State::GLState::ITextureObject& texture, Uint32 mipLevel,
+                                            VkFormat format, Bool layered, Int32 layer);
    void UpdateTrackedImageLayout(MG_State::GLState::ITextureObject* texture, VkImageLayout newLayout);
    void UpdateTrackedImageLayoutAfterAttachmentWrite(VkCommandBuffer commandBuffer,
                                                      MG_State::GLState::ITextureObject* texture,
@@ -205,8 +307,27 @@ public:
                                                      VkImageLayout newLayout);
    Bool TransitionTextureForSampling(VkCommandBuffer commandBuffer, MG_State::GLState::ITextureObject& texture);
    Bool TransitionTextureForStorageImage(VkCommandBuffer commandBuffer, MG_State::GLState::ITextureObject& texture);
+    // Records that this texture is bound to a GL image unit, so its image must carry
+    // VK_IMAGE_USAGE_STORAGE_BIT. Must be called before NeedsStorageImagePreparation, and
+    // therefore before the render pass is committed: an image that has to be upgraded is
+    // recreated, which is illegal inside a render pass. Sticky for the texture's lifetime -
+    // GL lets an image binding come and go, and re-creating the image every time it does
+    // would cost far more than the compression it wins back.
+    void MarkStorageImageTexture(MG_State::GLState::ITextureObject& texture);
+    // True when this texture is marked but its live image predates the mark, i.e. the next sync
+    // will recreate it with STORAGE usage and copy the old contents forward. Callers use this to
+    // submit their pending recording first, so that copy cannot read pre-flush content.
+    Bool NeedsStorageUsageUpgrade(MG_State::GLState::ITextureObject& texture) const;
+    // Non-mutating probe for the per-draw storage-image fast path: true when preparing this
+    // texture as a storage image may need work that is illegal inside a render pass (resource
+    // creation, dirty-content upload, or a layout transition to GENERAL). Unknown state reports
+    // true - a false positive merely ends the render pass, a false negative would skip a barrier.
+    Bool NeedsStorageImagePreparation(MG_State::GLState::ITextureObject& texture) const;

    static VkImageAspectFlags ResolveSampledImageViewAspectMask(VkImageAspectFlags imageAspect);
+    static VkFormat ResolveSampledImageViewFormat(VkFormat imageFormat, SamplerNumericDomain numericDomain);
+    static Bool AreSampledImageViewFormatsCompatible(VkFormat imageFormat, VkFormat viewFormat);
+    static Bool AreStorageImageViewFormatsCompatible(VkFormat imageFormat, VkFormat viewFormat);

    static Bool TransitionImageLayout(VkCommandBuffer commandBuffer, VkImage image, VkImageLayout& trackedLayout,
                               VkImageLayout newLayout, VkPipelineStageFlags srcStageMask,
@@ -255,7 +376,8 @@ private:
                                VkImageViewType viewType, Uint32 baseMipLevel, Uint32 levelCount,
                                Uint32 baseArrayLayer,
                                Uint32 layerCount,
-                                const VkComponentMapping* components = nullptr) const;
+                                const VkComponentMapping* components = nullptr,
+                                VkImageUsageFlags viewUsage = 0) const;
    Bool UploadDirtyMipLevels(MG_State::GLState::TextureObjectMipmap &mipmapTexture,
                      TextureUploadTarget uploadTarget,
                      TextureResource &outResource);
@@ -275,15 +397,20 @@ private:
    static TextureIdentity MakeTextureIdentity(MG_State::GLState::ITextureObject* texture);
    void EraseTrackedTexture(const TextureIdentity& identity);
    void PruneStaleTextureAliases(MG_State::GLState::ITextureObject* texture);
+    SizeT PruneDeadTextures();

    VkDevice m_device = VK_NULL_HANDLE;
    VkPhysicalDevice m_physicalDevice = VK_NULL_HANDLE;
    VmaAllocator m_allocator = nullptr;
    VkCommandPool m_commandPool = VK_NULL_HANDLE;
    VkQueue m_graphicsQueue = VK_NULL_HANDLE;
+    Bool m_imageFormatListSupported = false;
    Uint32 m_currentFrameIndex = 0;

    Uint8 m_gcCounter = 0;
+    // Frame-boundary GC gate: counts BeginFrame calls, not draws, so texture churn
+    // through non-draw paths (FBO clears, readbacks) still reaches the prune.
+    Uint32 m_gcFrameCounter = 0;
    // Active only between BeginDrawSyncScope/EndDrawSyncScope; identities of
    // textures already fully synced in the current draw (small N -> flat scan).
    Bool m_drawSyncScopeActive = false;
@@ -296,8 +423,13 @@ private:
        TextureResource* resource = nullptr;
    };
    Vector<DrawSyncedTexture> m_drawSyncedThisDraw;
+    // Formats whose mutable-image probe failed on this device; their images are created
+    // without MUTABLE_FORMAT_BIT so repeat syncs neither re-probe nor flag-mismatch.
+    std::unordered_set<VkFormat> m_mutableFormatUnsupported;
    std::unordered_map<TextureIdentity, WeakPtr<MG_State::GLState::ITextureObject>, TextureIdentityHash> m_aliveObjects;
    std::unordered_map<TextureIdentity, TextureResource, TextureIdentityHash> m_textureResources;
+    // Textures that have been bound to a GL image unit (see MarkStorageImageTexture).
+    std::unordered_set<TextureIdentity, TextureIdentityHash> m_storageImageTextures;
    Vector<Vector<TextureResource>> m_deferredReleases;
    Vector<Vector<VkImageView>> m_deferredViewReleases;
 };
@@ -51,6 +51,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        Uint32 instanceCount = 1;
        Uint32 firstVertex = 0;
        Uint32 firstInstance = 0;
+        // Indexed-draw metadata for bounding vertex-stream conversion. baseVertex is the
+        // draw's base-vertex offset; indexRangeIsExactView is true only when the draw
+        // fetches exactly the indices its IndexBufferView describes (direct DrawElements;
+        // multi/indirect forms leave it false because the CPU cannot bound their ranges).
+        Int32 baseVertex = 0;
+        Bool indexRangeIsExactView = false;
    };

    struct DrawIndexedCmdParam {
@@ -108,7 +114,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        }
    };

-    class VulkanRenderer : public IBufferCopyCommandProvider, public FrameContext::IRecordingObserver {
+    class VulkanRenderer : public IBufferCopyCommandProvider,
+                           public FrameContext::IRecordingObserver,
+                           public VkRenderPassManager::IEvictionObserver,
+                           public ProgramFactory::IEvictionObserver {
    public:
        VulkanRenderer(NativeWindowType window, const VulkanRendererConfig& cfg = {});
        ~VulkanRenderer();
@@ -125,12 +134,34 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // recording, before any render pass.
        void OnFrameCommandRecordingBegan(VkCommandBuffer commandBuffer) override;

+        // VkRenderPassManager::IEvictionObserver: the render-pass aging sweep just
+        // destroyed these VkRenderPasses; evict every graphics pipeline hashed on a
+        // dying handle (they share its >1024-boundary idleness, so immediate
+        // destruction is safe) and drop the last-pipeline memo if any went.
+        void OnRenderPassesDestroyed(const Vector<VkRenderPass>& renderPasses) override;
+
+        // ProgramFactory::IEvictionObserver: an aged-out program entry was
+        // destroyed; evict its compute pipeline and graphics pipelines (same
+        // idleness guarantee - they are only bound through draws/dispatches that
+        // stamp the program entry) and purge the descriptor-set cache entries
+        // keyed by its now-recyclable VkDescriptorSetLayout handle.
+        void OnProgramEvicted(ProgramFactory::HashType programHash,
+                              VkDescriptorSetLayout descriptorSetLayout) override;
+
        Bool SetupDraw(FrameContext::FrameData& frame, GLenum mode, Flags<DrawSetupAspect> aspects,
                       const DrawCmdParam& drawParams,
                       const IndexBufferView* pIndexBufferView = nullptr);
        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);
@@ -157,6 +188,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                              GLsizei srcWidth, GLsizei srcHeight, GLsizei srcDepth);
        void GenerateMipmap(GLenum target);
        void ReadPixels(GLint x, GLint y, GLsizei width, GLsizei height, GLenum format, GLenum type, void* pixels);
+        static SizeT GetReadbackTexelSize(VkFormat sourceFormat);
+        static Bool ConvertReadbackPixels(const Uint8* sourcePixels, VkFormat sourceFormat,
+                                          GLsizei width, GLsizei height, GLenum destinationFormat,
+                                          GLenum destinationType, SizeT destinationRowStride,
+                                          Uint8* destinationPixels);
        void GetTexImage(GLenum target, GLint level, GLenum format, GLenum type, GLvoid* pixels);
        void GetTextureImage(const SharedPtr<MG_State::GLState::ITextureObject>& texture,
                             TextureUploadTarget uploadTarget, GLint level, GLenum format, GLenum type,
@@ -219,6 +255,9 @@ 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.
@@ -235,7 +274,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        Uint64 GetTimerQueryTimestampNs(const VkTimerQueryManager::TimestampRecord& record) const;

        void RequestSwapchainResize(Uint32 width, Uint32 height);
-        void RecreateSwapchain();
+        // Re-query the surface and report whether the live swapchain no longer matches it
+        // (size or orientation). This - not a VK_SUBOPTIMAL_KHR result - is what decides a
+        // rebuild, so a surface the driver merely considers suboptimal cannot thrash.
+        Bool SwapchainIsOutOfDate();
+        // Returns false when the surface is zero-area (minimized/hidden window):
+        // no new swapchain is installed and presentation must stay suspended.
+        Bool RecreateSwapchain();

    private:
        struct BlitUniformData {
@@ -275,6 +320,10 @@ 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
@@ -317,11 +366,26 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        VkFence AcquirePooledSubmitFence();
        void DestroySubmitFencePool();
        Bool HasPendingRecordedWork() const;
+        // Frame-boundary housekeeping for paths that never reach Present's
+        // tail (present-less readback loops, suspended presentation, blocking
+        // sync waits): runs the same per-frame drains Present performs, but
+        // only when every queue submission has been observed complete AND no
+        // recorded-but-unsubmitted commands exist - i.e. when CPU-GPU overlap
+        // is provably already zero. Never blocks (non-blocking fence poll
+        // only), so the presenting path's frames-in-flight pipelining is
+        // untouched. Returns true when the drain ran.
+        Bool TryDrainFrameTransients();

        Vector<SubmitRecord> m_inFlightSubmits;
        Vector<VkFence> m_freeSubmitFences;
        Uint64 m_submitCounter = 0;
        Uint64 m_completedSubmitCounter = 0;
+        // Drains since the last Present, gating the drain's frame-boundary-equivalent
+        // work (arena rewind + cache aging): a presenting app's mid-frame
+        // readbacks/waits must neither churn the transient caches nor accelerate the
+        // aging clocks, while present-less loops still cross a boundary every few
+        // iterations. Reset in Present.
+        Uint32 m_drainsSinceLastPresent = 0;

        NativeWindowType m_window = 0;
        void* m_platformDisplay = nullptr;
@@ -329,12 +393,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        void* m_platformCloseDisplay = nullptr;
        VulkanRendererConfig m_config;
        Bool m_swapchainResizeRequested = false;
+        // Presentation is suspended while the window is zero-area (minimized): the
+        // swapchain is unusable/out of date, so Present drops frames instead of
+        // submitting on a signaled fence / presenting never-acquired images.
+        Bool m_presentSuspended = false;

        // Vulkan objects
        Bool m_validationLayersEnabled = false;
        Vector<VkExtensionProperties> m_extensions;
        VkInstance m_instance = VK_NULL_HANDLE;
        VkDebugUtilsMessengerEXT m_debugMessenger = VK_NULL_HANDLE;
+        // Fallback reporting channel for drivers that ship the validation layers but
+        // only expose the older VK_EXT_debug_report (Adreno 650 / Vulkan 1.1.128).
+        VkDebugReportCallbackEXT m_debugReportCallback = VK_NULL_HANDLE;
        PhysicalDevice m_physicalDevice;
        VkDevice m_device = VK_NULL_HANDLE;
        VmaAllocator m_allocator = nullptr;
@@ -347,8 +418,10 @@ 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;
+        Bool m_unformattedFloatStorageImagesEnabled = false;
        // fillModeNonSolid gates VK_POLYGON_MODE_LINE/_POINT (glPolygonMode); independentBlend gates
        // per-draw-buffer color write masks (glColorMaski). Both are cached at device creation and
        // drive a runtime fallback when the device lacks them.
@@ -421,13 +494,70 @@ namespace MobileGL::MG_Backend::DirectVulkan {
        // Per-draw scratch buffers (clear keeps capacity) — these paths run for every
        // draw call and must not allocate.
        Vector<MG_State::GLState::ITextureObject*> m_sampledTexturesScratch;
+        Vector<MG_State::GLState::ITextureObject*> m_storageImageTexturesScratch;
        Vector<VkBuffer> m_vertexBuffersScratch;
        Vector<VkDeviceSize> m_vertexOffsetsScratch;
        Vector<VkVertexInputAttributeDescription> m_patchedAttributesScratch;
+        Vector<Float> m_vertexConversionScratch;
+        Vector<Uint8> m_vertexRepackScratch;
+
+        struct ConvertedVertexStreamKey {
+            const MG_State::GLState::BufferObject* buffer = nullptr;
+            Uint64 changeSerial = 0;
+            SizeT baseOffset = 0;
+            Uint32 sourceStride = 0;
+            DataType type = DataType::Float32;
+            Int size = 0;
+            Bool normalized = false;
+            Bool isInteger = false;
+            VertexInputStateFactory::VertexStreamConversion conversion =
+                VertexInputStateFactory::VertexStreamConversion::None;
+
+            Bool operator==(const ConvertedVertexStreamKey& other) const {
+                return buffer == other.buffer && changeSerial == other.changeSerial &&
+                       baseOffset == other.baseOffset && sourceStride == other.sourceStride &&
+                       type == other.type && size == other.size && normalized == other.normalized &&
+                       isInteger == other.isInteger && conversion == other.conversion;
+            }
+        };
+
+        struct ConvertedVertexStreamKeyHash {
+            SizeT operator()(const ConvertedVertexStreamKey& key) const {
+                SizeT hash = std::hash<const void*>{}(key.buffer);
+                auto combine = [&hash](SizeT value) {
+                    hash ^= value + static_cast<SizeT>(0x9e3779b97f4a7c15ull) + (hash << 6) + (hash >> 2);
+                };
+                combine(std::hash<Uint64>{}(key.changeSerial));
+                combine(std::hash<SizeT>{}(key.baseOffset));
+                combine(std::hash<Uint32>{}(key.sourceStride));
+                combine(std::hash<Uint32>{}(static_cast<Uint32>(key.type)));
+                combine(std::hash<Int>{}(key.size));
+                combine(std::hash<Bool>{}(key.normalized));
+                combine(std::hash<Bool>{}(key.isInteger));
+                combine(std::hash<Uint32>{}(static_cast<Uint32>(key.conversion)));
+                return hash;
+            }
+        };
+
+        struct ConvertedVertexStream {
+            BufferSlice slice;
+            // Number of source elements the cached slice covers. A draw needing a prefix of
+            // this range reuses the slice (converted streams are tightly packed); a draw
+            // needing more reconverts and replaces the entry, so per (buffer, layout) a
+            // frame converts at most the largest range any draw asked for.
+            SizeT elementCount = 0;
+            // Pins the source buffer for the frame so its heap address cannot be reused by
+            // a new BufferObject while this pointer-keyed entry is alive.
+            SharedPtr<const MG_State::GLState::BufferObject> sourcePin;
+        };
+        UnorderedMap<ConvertedVertexStreamKey, ConvertedVertexStream, ConvertedVertexStreamKeyHash>
+            m_convertedVertexStreams;

        void CreateInstance();
        VkResult SetupDebugMessenger();
        VkResult DestroyDebugMessenger();
+        VkResult SetupDebugReportCallback();
+        void DestroyDebugReportCallback();
        VkDebugUtilsMessengerCreateInfoEXT PopulateDebugMessengerCreateInfo();
        void CreateSurface();
        void PickPhysicalDevice();
@@ -446,10 +576,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
            const RenderPassEntry& renderPassEntry);
        VkPipeline GetOrCreateComputePipeline(const ProgramFactory::VkProgramObject& programObj);
        void DestroyComputePipelines();
+        // Takes the frame rather than a command buffer: a first-time storage-usage upgrade has to
+        // flush the pending recording (see the body), which retires the current command buffer.
+        Bool PrepareStorageImageTextures(
+            FrameContext::FrameData& frame,
+            const MG_State::GLState::ProgramObject& program,
+            const ProgramFactory::VkProgramObject& programObj);

        Bool UploadAndBindVertexBuffers(VkCommandBuffer commandBuffer, const MG_State::GLState::VertexArrayObject& vao,
                                        const ProgramFactory::VkProgramObject& programObj,
-                                        const DrawCmdParam& drawParams);
+                                        const DrawCmdParam& drawParams,
+                                        const IndexBufferView* pIndexBufferView);
        Bool UploadAndBindIndexBuffer(FrameContext::FrameData& frame,
                                     const MG_State::GLState::VertexArrayObject& vao,
                                      const IndexBufferView* pIndexBufferView = nullptr);
@@ -467,6 +604,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                                   GLenum filter);
        Bool MaterializePendingClearForTexture(VkCommandBuffer commandBuffer,
                                               MG_State::GLState::ITextureObject& texture);
+        Bool MaterializePendingClearForRenderbuffer(
+            VkCommandBuffer commandBuffer,
+            const SharedPtr<MG_State::GLState::RenderbufferObject>& renderbuffer);
        VkPipeline GetOrCreateBlitPipeline(const RenderPassEntry& renderPassEntry);
        Bool GenerateDepthMipmapWithShader(FrameContext::FrameData& frame,
                                           MG_State::GLState::ITextureObject& texture,
@@ -501,6 +641,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                                 const PhysicalDevice& compareWithDevice,
                                                 PhysicalDevice& outBetterDevice);
        static constexpr const char* s_validationLayerNames[] = {"VK_LAYER_KHRONOS_validation"};
+        // VK_KHR_image_format_list: lets MUTABLE_FORMAT images declare their exact view-format
+        // set so the driver can keep bandwidth compression (see CreateLogicalDeviceAndQueues).
+        Bool m_imageFormatListExtensionEnabled = false;
+
        static constexpr const char* s_deviceExtensionNames[] = {VK_KHR_SWAPCHAIN_EXTENSION_NAME};
        static Bool CheckValidationLayerSupport();

@@ -16,4 +16,5 @@ target_link_libraries(
        ${LINK_LIBRARIES}
 )

-add_test(NAME BufferBench COMMAND BufferBench --benchmark_counters_tabular=true)
+add_test(NAME BufferBench COMMAND BufferBench --benchmark_counters_tabular=true)
+set_tests_properties(BufferBench PROPERTIES LABELS benchmark)
@@ -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)
@@ -16,4 +16,5 @@ target_link_libraries(
        ${LINK_LIBRARIES}
 )

-add_test(NAME ProgramBench COMMAND ProgramBench --benchmark_counters_tabular=true)
+add_test(NAME ProgramBench COMMAND ProgramBench --benchmark_counters_tabular=true)
+set_tests_properties(ProgramBench PROPERTIES LABELS benchmark)
@@ -24,6 +24,7 @@ namespace MobileGL::MG_Impl::CGLImpl {
            GLint Samples = 0;
            GLint Profile = kCGLOGLPVersion_3_2_Core;
            GLint RendererId = 0x4d474c;
+            GLint DisplayMask = 0;
        };

        struct ContextObject {
@@ -134,6 +135,9 @@ namespace MobileGL::MG_Impl::CGLImpl {
            case kCGLPFARendererID:
                pixelFormat.RendererId = value;
                break;
+            case kCGLPFADisplayMask:
+                pixelFormat.DisplayMask = value;
+                break;
            default:
                break;
            }
@@ -343,6 +347,9 @@ namespace MobileGL::MG_Impl::CGLImpl {
        case kCGLPFARendererID:
            *value = pixelFormat->RendererId;
            return kCGLNoError;
+        case kCGLPFADisplayMask:
+            *value = pixelFormat->DisplayMask;
+            return kCGLNoError;
        case kCGLPFAOpenGLProfile:
            *value = pixelFormat->Profile;
            return kCGLNoError;
@@ -481,6 +488,32 @@ namespace MobileGL::MG_Impl::CGLImpl {
        return it == currentContexts.end() ? nullptr : it->second;
    }

+    CGLError SetVirtualScreen(CGLContextObj ctx, GLint screen) {
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        auto* object = TryGetContext(ctx);
+        if (!object) {
+            return kCGLBadContext;
+        }
+        if (screen != 0) {
+            return kCGLBadValue;
+        }
+        object->VirtualScreen = screen;
+        return kCGLNoError;
+    }
+
+    CGLError GetVirtualScreen(CGLContextObj ctx, GLint* screen) {
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        auto* object = TryGetContext(ctx);
+        if (!object) {
+            return kCGLBadContext;
+        }
+        if (!screen) {
+            return kCGLBadAddress;
+        }
+        *screen = object->VirtualScreen;
+        return kCGLNoError;
+    }
+
    CGLError SetParameter(CGLContextObj ctx, CGLContextParameter pname, const GLint* params) {
        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
        auto* object = TryGetContext(ctx);
@@ -32,6 +32,8 @@ namespace MobileGL::MG_Impl::CGLImpl {

    CGLError SetCurrentContext(CGLContextObj ctx);
    CGLContextObj GetCurrentContext();
+    CGLError SetVirtualScreen(CGLContextObj ctx, GLint screen);
+    CGLError GetVirtualScreen(CGLContextObj ctx, GLint* screen);
    CGLError SetParameter(CGLContextObj ctx, CGLContextParameter pname, const GLint* params);
    CGLError GetParameter(CGLContextObj ctx, CGLContextParameter pname, GLint* params);
    CGLError UpdateContext(CGLContextObj ctx);
@@ -71,6 +71,14 @@ MOBILEGL_CGL_API CGLContextObj CGLGetCurrentContext(void) {
    return MobileGL::MG_Impl::CGLImpl::GetCurrentContext();
 }

+MOBILEGL_CGL_API CGLError CGLSetVirtualScreen(CGLContextObj ctx, GLint screen) {
+    return MobileGL::MG_Impl::CGLImpl::SetVirtualScreen(ctx, screen);
+}
+
+MOBILEGL_CGL_API CGLError CGLGetVirtualScreen(CGLContextObj ctx, GLint* screen) {
+    return MobileGL::MG_Impl::CGLImpl::GetVirtualScreen(ctx, screen);
+}
+
 MOBILEGL_CGL_API CGLError CGLSetParameter(CGLContextObj ctx, CGLContextParameter pname, const GLint* params) {
    return MobileGL::MG_Impl::CGLImpl::SetParameter(ctx, pname, params);
 }
@@ -10,8 +10,12 @@

 #if defined(__APPLE__)

+#include "MG_Impl/CGLImpl/CGLImpl.h"
 #include "MG_Impl/GetProcAddress.h"

+#include <CoreGraphics/CoreGraphics.h>
+#include <CoreVideo/CVDisplayLink.h>
+#include <cstdint>
 #include <dlfcn.h>

 namespace {
@@ -47,10 +51,52 @@ namespace {
        return dlsym(handle, symbol);
    }

+    CGDirectDisplayID DisplayForMask(GLint displayMask) {
+        constexpr std::uint32_t MaxDisplays = sizeof(CGOpenGLDisplayMask) * 8;
+        CGDirectDisplayID displays[MaxDisplays] = {};
+        std::uint32_t displayCount = 0;
+        if (displayMask != 0 &&
+            CGGetActiveDisplayList(MaxDisplays, displays, &displayCount) == kCGErrorSuccess) {
+            const auto mask = static_cast<CGOpenGLDisplayMask>(displayMask);
+            for (std::uint32_t i = 0; i < displayCount; ++i) {
+                if ((CGDisplayIDToOpenGLDisplayMask(displays[i]) & mask) != 0) {
+                    return displays[i];
+                }
+            }
+        }
+        return CGMainDisplayID();
+    }
+
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+    CVReturn MobileGLCVDisplayLinkSetCurrentCGDisplayFromOpenGLContext(
+        CVDisplayLinkRef displayLink,
+        CGLContextObj context,
+        CGLPixelFormatObj pixelFormat) {
+        GLint virtualScreen = 0;
+        if (MobileGL::MG_Impl::CGLImpl::GetVirtualScreen(context, &virtualScreen) == kCGLNoError) {
+            GLint displayMask = 0;
+            if (!displayLink ||
+                MobileGL::MG_Impl::CGLImpl::DescribePixelFormat(
+                    pixelFormat, virtualScreen, kCGLPFADisplayMask, &displayMask) != kCGLNoError) {
+                return kCVReturnInvalidArgument;
+            }
+            return CVDisplayLinkSetCurrentCGDisplay(displayLink, DisplayForMask(displayMask));
+        }
+
+        using OriginalFunction = CVReturn (*)(CVDisplayLinkRef, CGLContextObj, CGLPixelFormatObj);
+        static const auto original = reinterpret_cast<OriginalFunction>(
+            dlsym(RTLD_NEXT, "CVDisplayLinkSetCurrentCGDisplayFromOpenGLContext"));
+        return original ? original(displayLink, context, pixelFormat) : kCVReturnError;
+    }
+
    __attribute__((used)) static const DyldInterposeEntry kMobileGLDyldInterpose[]
        __attribute__((section("__DATA,__interpose"))) = {
            {reinterpret_cast<const void*>(MobileGLDlsym), reinterpret_cast<const void*>(dlsym)},
+            {reinterpret_cast<const void*>(MobileGLCVDisplayLinkSetCurrentCGDisplayFromOpenGLContext),
+             reinterpret_cast<const void*>(CVDisplayLinkSetCurrentCGDisplayFromOpenGLContext)},
        };
+#pragma clang diagnostic pop
 } // namespace

 #endif
@@ -0,0 +1,10 @@
+# Public CGL entry points.
+_CGL*
+
+# Public EGL entry points.
+_egl*
+
+# Public OpenGL and GLX entry points. OpenGL function names always use an
+# uppercase letter or digit after the "gl" prefix; excluding lowercase here
+# deliberately prevents glslang_* from matching this pattern.
+_gl[A-Z0-9]*
@@ -8,6 +8,7 @@

 #include "EGLImpl.h"
 #include "../GetProcAddress.h"
+#include <Init.h>
 #include <MG_Backend/BackendObjects.h>
 #include <MG_State/EGLState/Core.h>
 #include <mutex>
@@ -25,6 +26,17 @@ namespace MobileGL::MG_Impl::EGLImpl {
            return MG_State::pEGLContext.get();
        }

+        // Entry points that can legitimately be an application's FIRST EGL
+        // call (display/proc-address/string queries) lazily bring MobileGL
+        // up here, so the library needs no static constructor and can
+        // re-initialize after the last eglTerminate tore everything down.
+        // Teardown-ish entry points keep using GetState() and fail benignly
+        // when MobileGL is not initialized.
+        EGLStateContext* GetStateEnsureInitialized() {
+            MobileGL::EnsureInitialized();
+            return GetState();
+        }
+
        MG_Backend::BackendObject* GetBackendObject(EGLStateContext* state) {
            auto* backendObject = MG_Backend::pActiveBackendObject.get();
            if (!backendObject && state) {
@@ -49,6 +61,8 @@ namespace MobileGL::MG_Impl::EGLImpl {
            return MG_Backend::WindowBackend::Android;
 #elif defined(__APPLE__)
            return MG_Backend::WindowBackend::MetalLayer;
+#elif defined(_WIN32)
+            return MG_Backend::WindowBackend::Win32;
 #elif defined(__linux__)
            return MG_Backend::WindowBackend::X11;
 #else
@@ -187,7 +201,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
    }

    EGLBoolean Initialize(EGLDisplay dpy, EGLint* major, EGLint* minor) {
-        auto* state = GetState();
+        auto* state = GetStateEnsureInitialized();
        if (!state) {
            return EGL_FALSE;
        }
@@ -208,7 +222,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
    }

    EGLDisplay GetDisplay(NativeDisplayType display) {
-        auto* state = GetState();
+        auto* state = GetStateEnsureInitialized();
        if (!state) {
            return EGL_NO_DISPLAY;
        }
@@ -313,6 +327,14 @@ namespace MobileGL::MG_Impl::EGLImpl {
        if (auto* backendObject = MG_Backend::pActiveBackendObject.get()) {
            backendObject->ReleaseEGLResources();
        }
+        // The last initialized display is gone and nothing is current on any
+        // thread: tear the whole library down deterministically inside the
+        // EGL lifecycle (backend, GL/EGL state, glslang). A later EGL call
+        // re-initializes lazily via GetStateEnsureInitialized(); process exit
+        // then has nothing left to destroy.
+        if (!state->HasAnyInitializedDisplay() && !state->HasAnyCurrentContext()) {
+            MobileGL::Destroy();
+        }
        return EGL_TRUE;
    }

@@ -345,7 +367,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
    }

    EGLBoolean BindAPI(EGLenum api) {
-        auto* state = GetState();
+        auto* state = GetStateEnsureInitialized();
        if (!state) {
            return EGL_FALSE;
        }
@@ -378,7 +400,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
    }

    char const* QueryString(EGLDisplay display, EGLint name) {
-        auto* state = GetState();
+        auto* state = GetStateEnsureInitialized();
        if (!state) {
            return nullptr;
        }
@@ -641,7 +663,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
    EGLDisplay GetPlatformDisplay(EGLenum platform, void* native_display, const EGLAttrib* attrib_list) {
        (void)attrib_list;

-        auto* state = GetState();
+        auto* state = GetStateEnsureInitialized();
        if (!state) {
            return EGL_NO_DISPLAY;
        }
@@ -737,6 +759,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
        if (!name) {
            return nullptr;
        }
+        MobileGL::EnsureInitialized();

        MGLOG_D("eglGetProcAddress(%s)", name);
        void* proc = MG_Impl::GetProcAddress(name);
@@ -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);
        }
    }
@@ -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,
@@ -295,7 +295,13 @@ DECLARE_GL_FUNCTION_HEAD(void, VertexAttribDivisor, GLuint index, GLuint divisor
 DECLARE_GL_FUNCTION_STUB_HEAD(void, BindTransformFeedback, GLenum target, GLuint id) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, BindTransformFeedback, target, id)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, DeleteTransformFeedbacks, GLsizei n, const GLuint* ids) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DeleteTransformFeedbacks, n, ids)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GenTransformFeedbacks, GLsizei n, GLuint* ids) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GenTransformFeedbacks, n, ids)
-DECLARE_GL_FUNCTION_STUB_HEAD(GLboolean, IsTransformFeedback, GLuint id) DECLARE_GL_FUNCTION_STUB_END(GLboolean, IsTransformFeedback, id)
+// Transform feedback objects are not implemented, so no name is ever a live object. The shared
+// stub returns (type)1, telling a probing caller that every id it invents already exists; GL_FALSE
+// is both truthful and what the spec requires for a name that was never generated.
+MOBILEGL_GL_API GLboolean glIsTransformFeedback(GLuint id) {
+    MGLOG_W("Stub function: %s(...)", __FUNCTION__);
+    return GL_FALSE;
+}
 DECLARE_GL_FUNCTION_STUB_HEAD(void, PauseTransformFeedback) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PauseTransformFeedback)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ResumeTransformFeedback) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ResumeTransformFeedback)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetProgramBinary, GLuint program, GLsizei bufSize, GLsizei* length, GLenum* binaryFormat, void* binary) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetProgramBinary, program, bufSize, length, binaryFormat, binary)
@@ -418,7 +424,7 @@ DECLARE_GL_FUNCTION_HEAD(void, DrawRangeElementsBaseVertex, GLenum mode, GLuint
 DECLARE_GL_FUNCTION_HEAD(void, DrawElementsInstancedBaseVertex, GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount, GLint basevertex) DECLARE_GL_FUNCTION_END_NO_RETURN(void, DrawElementsInstancedBaseVertex, mode, count, type, indices, instancecount, basevertex)
 DECLARE_GL_FUNCTION_HEAD(void, FramebufferTexture, GLenum target, GLenum attachment, GLuint texture, GLint level) DECLARE_GL_FUNCTION_END_NO_RETURN(void, FramebufferTexture, target, attachment, texture, level)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, PrimitiveBoundingBox, GLfloat minX, GLfloat minY, GLfloat minZ, GLfloat minW, GLfloat maxX, GLfloat maxY, GLfloat maxZ, GLfloat maxW) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PrimitiveBoundingBox, minX, minY, minZ, minW, maxX, maxY, maxZ, maxW)
-DECLARE_GL_FUNCTION_STUB_HEAD(GLenum, GetGraphicsResetStatus) DECLARE_GL_FUNCTION_STUB_END(GLenum, GetGraphicsResetStatus)
+DECLARE_GL_FUNCTION_HEAD(GLenum, GetGraphicsResetStatus) DECLARE_GL_FUNCTION_END(GLenum, GetGraphicsResetStatus)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ReadnPixels, GLint x, GLint y, GLsizei width, GLsizei height, GLenum format, GLenum type, GLsizei bufSize, void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ReadnPixels, x, y, width, height, format, type, bufSize, data)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnUniformfv, GLuint program, GLint location, GLsizei bufSize, GLfloat* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnUniformfv, program, location, bufSize, params)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnUniformiv, GLuint program, GLint location, GLsizei bufSize, GLint* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetnUniformiv, program, location, bufSize, params)
@@ -998,8 +1004,8 @@ DECLARE_GL_FUNCTION_HEAD(void, ShaderStorageBlockBinding, GLuint program, GLuint
 DECLARE_GL_FUNCTION_STUB_HEAD(void, TextureView, GLuint texture, GLenum target, GLuint origtexture, GLenum internalformat, GLuint minlevel, GLuint numlevels, GLuint minlayer, GLuint numlayers) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, TextureView, texture, target, origtexture, internalformat, minlevel, numlevels, minlayer, numlayers)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, VertexAttribLFormat, GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, VertexAttribLFormat, attribindex, size, type, relativeoffset)
 DECLARE_GL_FUNCTION_HEAD(void, BufferStorage, GLenum target, GLsizeiptr size, const void* data, GLbitfield flags) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BufferStorage, target, size, data, flags)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, ClearTexImage, GLuint texture, GLint level, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ClearTexImage, texture, level, format, type, data)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, ClearTexSubImage, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ClearTexSubImage, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, type, data)
+DECLARE_GL_FUNCTION_HEAD(void, ClearTexImage, GLuint texture, GLint level, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ClearTexImage, texture, level, format, type, data)
+DECLARE_GL_FUNCTION_HEAD(void, ClearTexSubImage, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, ClearTexSubImage, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, type, data)
 DECLARE_GL_FUNCTION_HEAD(void, BindBuffersBase, GLenum target, GLuint first, GLsizei count, const GLuint* buffers) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BindBuffersBase, target, first, count, buffers)
 DECLARE_GL_FUNCTION_HEAD(void, BindBuffersRange, GLenum target, GLuint first, GLsizei count, const GLuint* buffers, const GLintptr* offsets, const GLsizeiptr* sizes) DECLARE_GL_FUNCTION_END_NO_RETURN(void, BindBuffersRange, target, first, count, buffers, offsets, sizes)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, BindTextures, GLuint first, GLsizei count, const GLuint* textures) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, BindTextures, first, count, textures)
@@ -1063,7 +1069,7 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage1D, GLuint texture,
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, imageSize, data)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, data)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CopyTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CopyTextureSubImage1D, texture, level, xoffset, x, y, width)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CopyTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CopyTextureSubImage2D, texture, level, xoffset, yoffset, x, y, width, height)
+DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage2D, texture, level, xoffset, yoffset, x, y, width, height)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CopyTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CopyTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, x, y, width, height)
 DECLARE_GL_FUNCTION_HEAD(void, TextureParameterf, GLuint texture, GLenum pname, GLfloat param) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureParameterf, texture, pname, param)
 DECLARE_GL_FUNCTION_HEAD(void, TextureParameterfv, GLuint texture, GLenum pname, const GLfloat* param) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureParameterfv, texture, pname, param)
@@ -2583,7 +2589,10 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, TransformFeedbackStreamAttribsNV, GLsizei co
 DECLARE_GL_FUNCTION_STUB_HEAD(void, BindTransformFeedbackNV, GLenum target, GLuint id) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, BindTransformFeedbackNV, target, id)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, DeleteTransformFeedbacksNV, GLsizei n, const GLuint* ids) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DeleteTransformFeedbacksNV, n, ids)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GenTransformFeedbacksNV, GLsizei n, GLuint* ids) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GenTransformFeedbacksNV, n, ids)
-DECLARE_GL_FUNCTION_STUB_HEAD(GLboolean, IsTransformFeedbackNV, GLuint id) DECLARE_GL_FUNCTION_STUB_END(GLboolean, IsTransformFeedbackNV, id)
+MOBILEGL_GL_API GLboolean glIsTransformFeedbackNV(GLuint id) {
+    MGLOG_W("Stub function: %s(...)", __FUNCTION__);
+    return GL_FALSE;
+}
 DECLARE_GL_FUNCTION_STUB_HEAD(void, PauseTransformFeedbackNV, void) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, PauseTransformFeedbackNV, )
 DECLARE_GL_FUNCTION_STUB_HEAD(void, ResumeTransformFeedbackNV, void) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, ResumeTransformFeedbackNV, )
 DECLARE_GL_FUNCTION_STUB_HEAD(void, DrawTransformFeedbackNV, GLenum mode, GLuint id) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DrawTransformFeedbackNV, mode, id)
@@ -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
@@ -1937,6 +2144,7 @@ namespace MobileGL::MG_Impl::GLImpl {
    }

    namespace FramebufferImpl {
-        UniquePtr<DefaultFramebufferInfo> pDefaultFramebufferInfo;
+        // Leak-at-exit storage; see GlobalObjects.cpp.
+        UniquePtr<DefaultFramebufferInfo>& pDefaultFramebufferInfo = *new UniquePtr<DefaultFramebufferInfo>();
    } // namespace FramebufferImpl
 } // namespace MobileGL::MG_Impl::GLImpl
@@ -78,6 +78,6 @@ namespace MobileGL::MG_Impl::GLImpl {
            SharedPtr<MG_State::GLState::ITextureObject> stencilAttachment;
        };

-        extern UniquePtr<DefaultFramebufferInfo> pDefaultFramebufferInfo;
+        extern UniquePtr<DefaultFramebufferInfo>& pDefaultFramebufferInfo;
    } // namespace FramebufferImpl
 } // namespace MobileGL::MG_Impl::GLImpl
@@ -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",
@@ -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();
@@ -1168,10 +1175,9 @@ namespace MobileGL::MG_Impl::GLImpl {
            *params = kFrontendMaxFragmentInputComponents;
            return;
        case GL_MAX_FRAGMENT_IMAGE_UNIFORMS:
-            // TODO: Track per-stage image uniform limits separately instead of reusing the compute/backend stage cap.
            *params = MG_Backend::pActiveBackendObject
-                          ? MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxComputeImageUniforms
-                          : MG_Backend::DynamicBackendParameters{}.MaxComputeImageUniforms;
+                          ? MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxFragmentImageUniforms
+                          : MG_Backend::DynamicBackendParameters{}.MaxFragmentImageUniforms;
            return;
        case GL_MAX_FRAGMENT_UNIFORM_COMPONENTS:
            *params = kFrontendMaxFragmentUniformComponents;
@@ -1201,7 +1207,9 @@ namespace MobileGL::MG_Impl::GLImpl {
            *params = kFrontendMaxGeometryTextureImageUnits;
            return;
        case GL_MAX_GEOMETRY_IMAGE_UNIFORMS:
-            *params = 0;
+            *params = MG_Backend::pActiveBackendObject
+                          ? MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxGeometryImageUniforms
+                          : MG_Backend::DynamicBackendParameters{}.MaxGeometryImageUniforms;
            return;
        case GL_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
            *params = kFrontendMaxGeometryTotalOutputComponents;
@@ -1270,7 +1278,9 @@ namespace MobileGL::MG_Impl::GLImpl {
            *params = kFrontendMaxVertexAtomicCounters;
            return;
        case GL_MAX_VERTEX_IMAGE_UNIFORMS:
-            *params = 0;
+            *params = MG_Backend::pActiveBackendObject
+                          ? MG_Backend::pActiveBackendObject->GetDynamicParameters().MaxVertexImageUniforms
+                          : MG_Backend::DynamicBackendParameters{}.MaxVertexImageUniforms;
            return;
        case GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS:
            *params = 16; // TODO
@@ -1727,7 +1737,11 @@ 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 +1908,13 @@ 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 +1975,10 @@ 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,
@@ -1972,4 +1996,12 @@ namespace MobileGL::MG_Impl::GLImpl {
        }
        return MG_Util::ConvertErrorCodeToGLEnum(error->get()->code);
    }
+
+    GLenum GetGraphicsResetStatus() {
+        // MobileGL does not implement robustness reset notification, so report GL_NO_ERROR
+        // ("no reset detected"). Returning the generic stub's (GLenum)1 makes dEQP read a lost
+        // device after every case (gl3cTestPackages.cpp:121) and, under the default
+        // --deqp-terminate-on-device-lost=enable, tear the whole CTS run down.
+        return GL_NO_ERROR;
+    }
 } // namespace MobileGL::MG_Impl::GLImpl
@@ -21,4 +21,5 @@ namespace MobileGL::MG_Impl::GLImpl {
    void GetIntegeri_v(GLenum target, GLuint index, GLint* data);
    void GetInteger64i_v(GLenum target, GLuint index, GLint64* data);
    GLenum GetError();
+    GLenum GetGraphicsResetStatus();
 } // namespace MobileGL::MG_Impl::GLImpl
@@ -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);
@@ -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) {
@@ -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) {
@@ -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));
        }
@@ -133,4 +133,33 @@ namespace MobileGL::MG_Impl::GLImpl {
            values[0] = value;
        }
    }
+
+    void DestroyAllSyncObjects() {
+        // Detach the registry under the lock, release outside it. Entries the app
+        // already deleted were erased by DeleteSync, so nothing here double-frees;
+        // a DeleteSync racing this sweep finds an empty registry and returns. A
+        // thread still blocked inside ClientWaitSync/GetSynciv during teardown
+        // holds a raw SyncObject* these deletes invalidate - the same undefined
+        // race an app-driven DeleteSync already has.
+        UnorderedMap<GLsync, SyncObject*> orphans;
+        {
+            const std::lock_guard<std::mutex> lock(g_syncObjectsMutex);
+            orphans.swap(g_liveSyncObjects);
+        }
+        if (orphans.empty()) {
+            return;
+        }
+        // Both backends' DeleteSync only free the heap wrapper once their GL
+        // context/renderer is gone (generation/current-thread guards), so this is
+        // safe after the backend has released its EGL resources - but not after
+        // the function table itself is cleared.
+        const auto backendDeleteSync = MG_Backend::gBackendFunctionsTable.GL.DeleteSync;
+        for (const auto& [_, syncObject] : orphans) {
+            if (backendDeleteSync && syncObject->backendHandle) {
+                backendDeleteSync(syncObject->backendHandle);
+            }
+            delete syncObject;
+        }
+        MGLOG_D("DestroyAllSyncObjects: reclaimed %zu sync object(s) the app left undeleted", orphans.size());
+    }
 } // namespace MobileGL::MG_Impl::GLImpl
@@ -16,4 +16,12 @@ namespace MobileGL::MG_Impl::GLImpl {
    void WaitSync(GLsync sync, GLbitfield flags, GLuint64 timeout);
    void DeleteSync(GLsync sync);
    void GetSynciv(GLsync sync, GLenum pname, GLsizei bufSize, GLsizei* length, GLint* values);
+    // Destroys every still-registered sync object exactly as DeleteSync would.
+    // GL requires syncs to die with their context; called only from full library
+    // teardown (DestroyImpl), where no context survives on any thread, so the
+    // process-global registry can be drained wholesale. Must run while the
+    // backend function table is still populated: each backend handle has to be
+    // released by the backend that created it, never by a later re-initialized
+    // one.
+    void DestroyAllSyncObjects();
 } // namespace MobileGL::MG_Impl::GLImpl
@@ -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),
            };
        }

@@ -323,15 +340,20 @@ 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());
                texture.AllocateStorage(uploadTarget, level, {levelTexelSize, levelByteSize});
                texture.MarkStorageDirty(uploadTarget, level, false);
            }
+            // glGenerateMipmap defines exactly levels 0..requiredLevelCount-1. AllocateStorage only
+            // grows, so a previously longer chain (a bigger base image before respecification) would
+            // otherwise keep a tail of stale levels here and read as incomplete.
+            texture.TruncateMipmapLevels(uploadTarget, requiredLevelCount);
            // Mip generation grows/regenerates the level set on the GPU without marking any CPU
            // level dirty (MarkStorageDirty(...,false) above). Bump the content version so the
            // backend re-syncs: a cached sampled VkImageView built for the pre-generate level
@@ -404,18 +426,17 @@ 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) {
+                // GL specifies INVALID_OPERATION - not INVALID_VALUE - when the sample count
+                // exceeds what the format supports, and the native Adreno driver agrees.
                MG_State::pGLContext->RecordError(
-                    ErrorCode::InvalidValue,
+                    ErrorCode::InvalidOperation,
                    MakeUnique<GenericErrorInfo>(
                        "MG_Impl/GLImpl", caller,
                        std::format("Sample count {} exceeds the supported maximum {} for this texture format.",
@@ -439,8 +460,56 @@ namespace MobileGL::MG_Impl::GLImpl {
            textureObject->SetSamples(samples);
            textureObject->SetFixedSampleLocations(fixedsamplelocations == GL_TRUE);
            textureMipmapObject->AllocateStorage(textureUploadTarget, 0, {{width, height, depth}, 0});
+            // Multisample textures are single-level by definition, so a name that previously held a
+            // mip chain must not keep its tail now that AllocateStorage only grows.
+            textureMipmapObject->TruncateMipmapLevels(textureUploadTarget, 1);
            textureMipmapObject->MarkStorageDirty(textureUploadTarget, 0, false);
        }
+
+        // Redefining level 0 of a texture that already had a base image drops the rest of the chain,
+        // which is exactly what AllocateLevel used to do implicitly for every level. Keeping that
+        // behaviour for level 0 - and only for level 0 - is what makes the grow-only change safe:
+        // any level-0 respecification leaves the chain in precisely the state it would have had
+        // before, while an upload to level N no longer destroys the levels beneath it.
+        //
+        // Why it has to be *every* level-0 respecification and not just a size change: Minecraft's
+        // Mipmap Levels setting rebuilds the block atlas at the SAME dimensions with a different
+        // level count. A size-only test would leave the old tail in place, and because Mojang
+        // terminates its chains with a 0x0 level the result is the zero-then-nonzero pattern that
+        // IsComplete() rejects (TextureObject.cpp) - whereupon DirectGLES skips syncing the texture
+        // entirely (Managers.cpp) and the atlas samples black.
+        //
+        // The "already has a base image" test is what lets the fix work at all: a level that was
+        // never written reads back as {0,0,0}, so building a chain top-down - upload level N first,
+        // then level 0 - must not discard the levels just uploaded. That ordering is what
+        // KHR-GL33.texture_repeat_mode does.
+        // Scoped to the respecified upload target only, which is what AllocateLevel already did.
+        // Cube maps keep six independent chains while reporting a single level count (face +X), so
+        // respecifying a face other than +X can leave the count longer than that face - but that
+        // asymmetry predates this change and widening the truncation to all six faces would destroy
+        // mip data for faces the application never touched. Left alone deliberately.
+        void DiscardMipmapChainOnBaseRespecification(MG_State::GLState::TextureObjectMipmap* texture,
+                                                     TextureUploadTarget uploadTarget, Uint level) {
+            if (level != 0) return;
+
+            const IntVec3 existingBaseSize = texture->GetMipmapTexelSize(uploadTarget, 0);
+            const Bool hasExistingBaseImage =
+                existingBaseSize.x() > 0 && existingBaseSize.y() > 0 && existingBaseSize.z() > 0;
+            if (!hasExistingBaseImage) return;
+
+            texture->TruncateMipmapLevels(uploadTarget, 1);
+        }
+
+        // Compressed texture upload is not implemented yet. GL_NUM_COMPRESSED_TEXTURE_FORMATS
+        // reports 0, so every compressed internalformat is by definition unsupported and
+        // GL_INVALID_ENUM is the specified error - unlike THROW_UNIMPL_EXCEPTION, which unwinds
+        // a C++ exception through the C GL ABI and takes the process down.
+        void RecordUnsupportedCompressedFormat(const char* caller) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller,
+                                             "Compressed texture formats are not supported."));
+        }
    } // namespace

    const SharedPtr<MG_State::GLState::ITextureObject>& GetTextureObjectByName(GLuint texture, const char* caller) {
@@ -455,9 +524,229 @@ namespace MobileGL::MG_Impl::GLImpl {
        return textureObject;
    }

+    namespace {
+        void RecordClearTextureError(const char* caller, ErrorCode code, const String& message) {
+            MG_State::pGLContext->RecordError(
+                code, MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", caller, message));
+        }
+
+        SharedPtr<MG_State::GLState::TextureObjectMipmap> GetClearTextureObject(GLuint texture, GLint level,
+                                                                                const char* caller) {
+            if (texture == 0) {
+                RecordClearTextureError(caller, ErrorCode::InvalidOperation,
+                                        "Clear texture operations require a non-zero texture name.");
+                return nullptr;
+            }
+
+            auto textureObject = GetTextureObjectByName(texture, caller);
+            if (!textureObject) return nullptr;
+            if (textureObject->GetStorageType() != TextureStorageType::Mipmap) {
+                RecordClearTextureError(caller, ErrorCode::InvalidOperation,
+                                        "Buffer textures cannot be cleared with glClearTexImage.");
+                return nullptr;
+            }
+
+            auto mipmapTexture = std::static_pointer_cast<MG_State::GLState::TextureObjectMipmap>(textureObject);
+            if (level < 0) {
+                RecordClearTextureError(caller, ErrorCode::InvalidValue,
+                                        std::format("Texture level {} is negative.", level));
+                return nullptr;
+            }
+            // ARB_clear_texture: clearing an image that was never defined by TexImage*/
+            // TexStorage* is INVALID_OPERATION, not INVALID_VALUE.
+            if (static_cast<Uint>(level) >= mipmapTexture->GetMipmapLevelCount()) {
+                RecordClearTextureError(caller, ErrorCode::InvalidOperation,
+                                        std::format("Texture level {} is not defined.", level));
+                return nullptr;
+            }
+            return mipmapTexture;
+        }
+
+        Bool BuildClearPixel(const SharedPtr<MG_State::GLState::TextureObjectMipmap>& textureObject,
+                             GLenum format, GLenum type, const void* data, Vector<Uint8>& clearPixel) {
+            const TextureInputFormat inputFormat = MG_Util::ConvertGLEnumToTextureInputFormat(format);
+            const TexturePixelDataType inputType = MG_Util::ConvertGLEnumToTexturePixelDataType(type);
+            if (!TextureImpl::ValidateTextureInputFormat(inputFormat) ||
+                !TextureImpl::ValidateTexturePixelDataType(inputType) ||
+                !TextureImpl::ValidateTextureInternalFormatCompatibleWithInput(
+                    inputFormat, textureObject->GetFormat(), inputType)) {
+                return false;
+            }
+
+            clearPixel.clear();
+            if (data == nullptr) {
+                // ARB_clear_texture defines a null clear value as all zeroes. Keeping the
+                // pattern empty lets the region writer use a fast memset path.
+                return true;
+            }
+
+            PixelStoreParameters clearPixelStore{};
+            clearPixelStore.Alignment = 1;
+            SizeT clearPixelSize = 0;
+            void* converted = MG_Util::PixelStoreProcessor::ProcessTexturePixelsDataUnpack(
+                data, clearPixelStore, textureObject->GetFormat(), inputFormat, inputType,
+                {1, 1, 1}, false, clearPixelSize);
+            if (!converted || clearPixelSize == 0) {
+                if (converted) free(converted);
+                return false;
+            }
+
+            clearPixel.resize(clearPixelSize);
+            Memcpy(clearPixel.data(), converted, clearPixelSize);
+            free(converted);
+            return true;
+        }
+
+        // Writes the clear into the CPU shadow and marks the whole level dirty, exactly like
+        // TexSubImage*_State does. Shared limitation of the level-granular shadow sync: the
+        // shadow does not reflect GPU-side writes (FBO rendering, imageStore), so a PARTIAL
+        // clear of a GPU-written level re-uploads stale shadow bytes outside the region on
+        // the next sync. Full-level clears (glClearTexImage, or a sub-clear covering the
+        // level) rewrite the entire shadow and are always correct.
+        Bool ClearMipmapRegion(const SharedPtr<MG_State::GLState::TextureObjectMipmap>& textureObject,
+                               TextureUploadTarget uploadTarget, GLint level,
+                               GLint xoffset, GLint yoffset, GLint zoffset,
+                               GLsizei width, GLsizei height, GLsizei depth,
+                               const Vector<Uint8>& clearPixel, const char* caller) {
+            const IntVec3 texelSize = textureObject->GetMipmapTexelSize(uploadTarget, static_cast<Uint>(level));
+            if (texelSize.x() <= 0 || texelSize.y() <= 0 || texelSize.z() <= 0) {
+                RecordClearTextureError(caller, ErrorCode::InvalidOperation,
+                                        "The requested texture level has no storage.");
+                return false;
+            }
+            if (xoffset < 0 || yoffset < 0 || zoffset < 0 ||
+                width < 0 || height < 0 || depth < 0 ||
+                width > texelSize.x() - xoffset ||
+                height > texelSize.y() - yoffset ||
+                depth > texelSize.z() - zoffset) {
+                RecordClearTextureError(caller, ErrorCode::InvalidValue,
+                                        "The clear region lies outside the requested texture level.");
+                return false;
+            }
+            if (width == 0 || height == 0 || depth == 0) return true;
+
+            const SizeT texelCount = static_cast<SizeT>(texelSize.x()) *
+                                     static_cast<SizeT>(texelSize.y()) *
+                                     static_cast<SizeT>(texelSize.z());
+            const SizeT byteSize = textureObject->GetMipmapByteSize(uploadTarget, static_cast<Uint>(level));
+            if (byteSize == 0 || byteSize % texelCount != 0) {
+                RecordClearTextureError(caller, ErrorCode::InvalidOperation,
+                                        "The requested texture storage cannot be cleared.");
+                return false;
+            }
+
+            const SizeT bytesPerTexel = byteSize / texelCount;
+            if (!clearPixel.empty() && clearPixel.size() != bytesPerTexel) {
+                RecordClearTextureError(
+                    caller, ErrorCode::InvalidOperation,
+                    std::format("Converted clear value is {} bytes, but the texture stores {} bytes per texel.",
+                                clearPixel.size(), bytesPerTexel));
+                return false;
+            }
+
+            auto* destination = static_cast<Uint8*>(
+                textureObject->MapMipmapData(uploadTarget, static_cast<Uint>(level)));
+            if (!destination) {
+                RecordClearTextureError(caller, ErrorCode::InvalidOperation,
+                                        "The requested texture level could not be mapped.");
+                return false;
+            }
+
+            const SizeT fullRowBytes = static_cast<SizeT>(texelSize.x()) * bytesPerTexel;
+            const SizeT fullSliceBytes = static_cast<SizeT>(texelSize.y()) * fullRowBytes;
+            const SizeT clearRowBytes = static_cast<SizeT>(width) * bytesPerTexel;
+            Uint8* firstClearRow = nullptr;
+
+            for (GLsizei z = 0; z < depth; ++z) {
+                for (GLsizei y = 0; y < height; ++y) {
+                    Uint8* row = destination +
+                                 static_cast<SizeT>(zoffset + z) * fullSliceBytes +
+                                 static_cast<SizeT>(yoffset + y) * fullRowBytes +
+                                 static_cast<SizeT>(xoffset) * bytesPerTexel;
+                    if (firstClearRow) {
+                        Memcpy(row, firstClearRow, clearRowBytes);
+                        continue;
+                    }
+
+                    firstClearRow = row;
+                    if (clearPixel.empty()) {
+                        Memset(row, 0, clearRowBytes);
+                        continue;
+                    }
+
+                    Memcpy(row, clearPixel.data(), bytesPerTexel);
+                    SizeT filled = bytesPerTexel;
+                    while (filled < clearRowBytes) {
+                        const SizeT copySize = std::min(filled, clearRowBytes - filled);
+                        Memcpy(row + filled, row, copySize);
+                        filled += copySize;
+                    }
+                }
+            }
+
+            textureObject->MarkStorageDirty(uploadTarget, static_cast<Uint>(level), true);
+            return true;
+        }
+    } // namespace
+
+    void ClearTexImage(GLuint texture, GLint level, GLenum format, GLenum type, const void* data) {
+        auto textureObject = GetClearTextureObject(texture, level, __func__);
+        if (!textureObject) return;
+
+        Vector<Uint8> clearPixel;
+        if (!BuildClearPixel(textureObject, format, type, data, clearPixel)) return;
+
+        for (TextureUploadTarget uploadTarget : textureObject->GetUploadTargets()) {
+            const IntVec3 size = textureObject->GetMipmapTexelSize(uploadTarget, static_cast<Uint>(level));
+            if (!ClearMipmapRegion(textureObject, uploadTarget, level, 0, 0, 0,
+                                   size.x(), size.y(), size.z(), clearPixel, __func__)) {
+                return;
+            }
+        }
+    }
+
+    void ClearTexSubImage(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset,
+                          GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLenum type,
+                          const void* data) {
+        auto textureObject = GetClearTextureObject(texture, level, __func__);
+        if (!textureObject) return;
+
+        Vector<Uint8> clearPixel;
+        if (!BuildClearPixel(textureObject, format, type, data, clearPixel)) return;
+
+        const auto& uploadTargets = textureObject->GetUploadTargets();
+        if (textureObject->GetTarget() == TextureTarget::TextureCubeMap) {
+            if (zoffset < 0 || depth < 0 ||
+                static_cast<SizeT>(zoffset) > uploadTargets.size() ||
+                static_cast<SizeT>(depth) > uploadTargets.size() - static_cast<SizeT>(zoffset)) {
+                RecordClearTextureError(__func__, ErrorCode::InvalidValue,
+                                        "The cube-map clear region selects invalid faces.");
+                return;
+            }
+            for (GLsizei face = 0; face < depth; ++face) {
+                if (!ClearMipmapRegion(textureObject, uploadTargets[static_cast<SizeT>(zoffset + face)], level,
+                                       xoffset, yoffset, 0, width, height, 1, clearPixel, __func__)) {
+                    return;
+                }
+            }
+            return;
+        }
+
+        if (uploadTargets.empty()) {
+            RecordClearTextureError(__func__, ErrorCode::InvalidOperation,
+                                    "The requested texture has no upload target.");
+            return;
+        }
+        ClearMipmapRegion(textureObject, uploadTargets.front(), level, xoffset, yoffset, zoffset,
+                          width, height, depth, clearPixel, __func__);
+    }
+
    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 +776,8 @@ 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 +847,14 @@ 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 +876,9 @@ 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 +895,10 @@ 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 +944,9 @@ 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 +1016,9 @@ 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 +1044,23 @@ 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 +1106,12 @@ 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 +1144,14 @@ 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 +1206,11 @@ 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 +1368,7 @@ 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 +1401,12 @@ 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 +1430,10 @@ 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 +1457,7 @@ 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 +1473,7 @@ 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 +1481,7 @@ 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 +1504,7 @@ 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 +1512,7 @@ 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 +1530,7 @@ 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 +1538,7 @@ 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 +1558,7 @@ 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 +1569,7 @@ 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 +1620,8 @@ 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 +1663,8 @@ 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(
@@ -1369,6 +1724,21 @@ namespace MobileGL::MG_Impl::GLImpl {
            return;
        }

+        // RGTC is a 2D-only compression scheme, so a 3D target rejects it. This has to be tested on
+        // the raw enum: the RGTC formats resolve to plain R8/RG8/SNORM storage on the way in (see
+        // GLToMG's TextureEnumConverter), so once the internal format is converted there is nothing
+        // left to distinguish them from an ordinary one- or two-channel upload.
+        if ((textureUploadTarget == TextureUploadTarget::Texture3D ||
+             textureUploadTarget == TextureUploadTarget::ProxyTexture3D) &&
+            (internalformat == GL_COMPRESSED_RED_RGTC1 || internalformat == GL_COMPRESSED_SIGNED_RED_RGTC1 ||
+             internalformat == GL_COMPRESSED_RG_RGTC2 || internalformat == GL_COMPRESSED_SIGNED_RG_RGTC2)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "RGTC compressed formats are invalid for 3D texture targets"));
+            return;
+        }
+
        // TODO: GL_INVALID_OPERATION is generated if a non-zero buffer object name is bound to the
        // GL_PIXEL_UNPACK_BUFFER target and the buffer object's data store is currently mapped.
        // GL_INVALID_OPERATION is generated if a non-zero buffer object name is bound to the GL_PIXEL_UNPACK_BUFFER
@@ -1378,6 +1748,8 @@ 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 +1758,8 @@ 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 +1793,12 @@ 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 {
+            DiscardMipmapChainOnBaseRespecification(textureMipmapObject, textureUploadTarget, level);
+            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 +1825,7 @@ 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 +1880,8 @@ 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;

@@ -1542,6 +1924,7 @@ namespace MobileGL::MG_Impl::GLImpl {
            MGLOG_D("%s: isProxy = true, not allocating", __func__);
        } else {
            MGLOG_D("%s: Allocating %d bytes at mip %d", __func__, internalBytes, level);
+            DiscardMipmapChainOnBaseRespecification(textureMipmapObject, textureUploadTarget, level);
            textureMipmapObject->AllocateStorage(textureUploadTarget, level,
                                                 {{width, height, 1}, internalBytes});
        }
@@ -1604,6 +1987,8 @@ 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;

@@ -1628,6 +2013,7 @@ namespace MobileGL::MG_Impl::GLImpl {
                        "Texture object here should always be an object with mipmap");
        auto textureMipmapObject = static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get());
        if (!isProxy) {
+            DiscardMipmapChainOnBaseRespecification(textureMipmapObject, textureUploadTarget, level);
            textureMipmapObject->AllocateStorage(textureUploadTarget, level, {{width, 1, 1}, internalBytes});
        }

@@ -1659,8 +2045,12 @@ 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 +2064,8 @@ 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 +2086,9 @@ 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 +2277,11 @@ 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 +2428,11 @@ 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",
@@ -2262,7 +2666,13 @@ namespace MobileGL::MG_Impl::GLImpl {
    }

    void GetCompressedTexImage_State(GLenum target, GLint level, void* img) {
-        // TODO: implement
+        // TODO: implement compressed readback. Reporting success while writing nothing hands
+        // the caller stale memory with GL_NO_ERROR; no texture can be compressed yet, and GL
+        // specifies GL_INVALID_OPERATION when the bound level is not compressed.
+        MG_State::pGLContext->RecordError(
+            ErrorCode::InvalidOperation,
+            MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                         "Texture level is not stored in a compressed format."));
    }

    void GenTextures_State(GLsizei n, GLuint* textures) {
@@ -2300,7 +2710,7 @@ 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);
        }
    }
@@ -2449,20 +2859,20 @@ namespace MobileGL::MG_Impl::GLImpl {
    void CompressedTexSubImage3D_State(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset,
                                       GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize,
                                       const void* data) {
-        // TODO: implement
-        THROW_UNIMPL_EXCEPTION;
+        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        RecordUnsupportedCompressedFormat(__func__);
    }

    void CompressedTexSubImage2D_State(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLsizei width,
                                       GLsizei height, GLenum format, GLsizei imageSize, const void* data) {
-        // TODO: implement
-        THROW_UNIMPL_EXCEPTION;
+        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        RecordUnsupportedCompressedFormat(__func__);
    }

    void CompressedTexSubImage1D_State(GLenum target, GLint level, GLint xoffset, GLsizei width, GLenum format,
                                       GLsizei imageSize, const void* data) {
-        // TODO: implement
-        THROW_UNIMPL_EXCEPTION;
+        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        RecordUnsupportedCompressedFormat(__func__);
    }

    void CompressedTexImage3D_State(GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height,
@@ -2472,8 +2882,8 @@ namespace MobileGL::MG_Impl::GLImpl {
        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
        if (!ValidateTextureMutable(textureObject, __func__)) return;

-        // TODO: implement
-        THROW_UNIMPL_EXCEPTION;
+        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        RecordUnsupportedCompressedFormat(__func__);
    }

    void CompressedTexImage2D_State(GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height,
@@ -2483,8 +2893,11 @@ namespace MobileGL::MG_Impl::GLImpl {
        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
        if (!ValidateTextureMutable(textureObject, __func__)) return;

-        // TODO: implement
-        THROW_UNIMPL_EXCEPTION;
+        // TODO: implement compressed upload. Until then report the spec error for an
+        // unsupported compressed format rather than throwing - a C++ exception unwinding
+        // through the C GL ABI is a hard crash for the caller, while GL_INVALID_ENUM is
+        // exactly what GL_NUM_COMPRESSED_TEXTURE_FORMATS == 0 promises.
+        RecordUnsupportedCompressedFormat(__func__);
    }

    void CompressedTexImage1D_State(GLenum target, GLint level, GLenum internalformat, GLsizei width, GLint border,
@@ -2494,8 +2907,8 @@ namespace MobileGL::MG_Impl::GLImpl {
        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
        if (!ValidateTextureMutable(textureObject, __func__)) return;

-        // TODO: implement
-        THROW_UNIMPL_EXCEPTION;
+        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        RecordUnsupportedCompressedFormat(__func__);
    }

    void BindTexture_State(GLenum target, GLuint texture) {
@@ -2507,11 +2920,13 @@ 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 +2938,9 @@ 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 +2948,6 @@ namespace MobileGL::MG_Impl::GLImpl {
            return;
        }

-        if (!TextureImpl::ValidateTextureName(texture, true)) return;
-
        // ======================= Processing ================================
        Bool doesTextureExist = MG_State::pGLContext->ValidateTextureObject(texture);
        if (!doesTextureExist) {
@@ -2551,14 +2967,25 @@ 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;
        }

@@ -2827,6 +3254,9 @@ namespace MobileGL::MG_Impl::GLImpl {
            textureMipmapObject->AllocateStorage(textureUploadTarget, level, {{levelWidth, 1, 1}, byteSize});
            textureMipmapObject->MarkStorageDirty(textureUploadTarget, level, false);
        }
+        // Immutable storage defines exactly `levels` levels; AllocateStorage only grows, so a
+        // longer pre-existing chain has to be dropped explicitly.
+        textureMipmapObject->TruncateMipmapLevels(textureUploadTarget, static_cast<Uint>(levels));
        textureObject->SetImmutableLevels(static_cast<Uint>(levels));
    }

@@ -2879,6 +3309,8 @@ namespace MobileGL::MG_Impl::GLImpl {
            textureMipmapObject->AllocateStorage(textureUploadTarget, level, {{levelWidth, levelHeight, 1}, byteSize});
            textureMipmapObject->MarkStorageDirty(textureUploadTarget, level, false);
        }
+        // See TextureStorage1D.
+        textureMipmapObject->TruncateMipmapLevels(textureUploadTarget, static_cast<Uint>(levels));
        textureObject->SetImmutableLevels(static_cast<Uint>(levels));
    }

@@ -2918,16 +3350,21 @@ 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,
                                                 {{levelWidth, levelHeight, levelDepth}, byteSize});
            textureMipmapObject->MarkStorageDirty(textureUploadTarget, level, false);
        }
+        // See TextureStorage1D.
+        textureMipmapObject->TruncateMipmapLevels(textureUploadTarget, static_cast<Uint>(levels));
        textureObject->SetImmutableLevels(static_cast<Uint>(levels));
    }

@@ -2957,6 +3394,7 @@ 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 +3409,7 @@ 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 +3425,7 @@ 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 +3581,10 @@ 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;
        }
@@ -3681,6 +4123,27 @@ namespace MobileGL::MG_Impl::GLImpl {
        CopyTexSubImage2D_Backend(target, level, xoffset, yoffset, x, y, width, height);
    }

+    void CopyTextureSubImage2D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y,
+                               GLsizei width, GLsizei height) {
+        auto textureObject = GetTextureObjectByName(texture, __func__);
+        if (!textureObject) return;
+        // GL 4.6 sec. 8.8: the 2D form only accepts these effective targets; cube maps must
+        // go through CopyTextureSubImage3D with the face as a layer.
+        const auto target = textureObject->GetTarget();
+        if (target != TextureTarget::Texture2D && target != TextureTarget::Texture1DArray &&
+            target != TextureTarget::TextureRectangle) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "CopyTextureSubImage2D requires a 2D, 1D-array, or "
+                                             "rectangle texture."));
+            return;
+        }
+        WithTemporarilyBoundNamedTexture(textureObject, [&](GLenum glTarget) {
+            CopyTexSubImage2D_Backend(glTarget, level, xoffset, yoffset, x, y, width, height);
+        });
+    }
+
    void CopyTexSubImage1D(GLenum target, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width) {
        CopyTexSubImage1D_State(target, level, xoffset, x, y, width);
    }
@@ -11,6 +11,9 @@

 namespace MobileGL::MG_Impl::GLImpl {
    /* @INSERTION_POINT:FUNCTION_DECLARATION@ */
+    void ClearTexImage(GLuint texture, GLint level, GLenum format, GLenum type, const void* data);
+    void ClearTexSubImage(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width,
+                          GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* data);
    void BindImageTexture(GLuint unit, GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum access,
                          GLenum format);
    void GenerateMipmap(GLenum target);
@@ -95,6 +98,8 @@ namespace MobileGL::MG_Impl::GLImpl {
                           GLsizei width, GLsizei height);
    void CopyTexSubImage2D(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width,
                           GLsizei height);
+    void CopyTextureSubImage2D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y,
+                               GLsizei width, GLsizei height);
    void CopyTexSubImage1D(GLenum target, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width);
    void CopyTexImage2D(GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width,
                        GLsizei height, GLint border);
@@ -14,7 +14,8 @@
 #include <MG_State/GLState/TextureState/TextureObjectStubs.h>

 namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
-    UniquePtr<ProxyTextureManager> pProxyTextureManager;
+    // Leak-at-exit storage; see GlobalObjects.cpp.
+    UniquePtr<ProxyTextureManager>& pProxyTextureManager = *new UniquePtr<ProxyTextureManager>();

    Bool IsProxyTextureTarget(TextureUploadTarget target) {
        switch (target) {
@@ -23,5 +23,5 @@ namespace MobileGL::MG_Impl::GLImpl::TextureImpl {
        UnorderedMap<TextureUploadTarget, SharedPtr<MG_State::GLState::ITextureObject>> m_proxyTexturesMap;
    };

-    extern UniquePtr<ProxyTextureManager> pProxyTextureManager;
+    extern UniquePtr<ProxyTextureManager>& pProxyTextureManager;
 } // namespace MobileGL::MG_Impl::GLImpl::TextureImpl
@@ -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
@@ -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,
@@ -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)) {
@@ -85,6 +85,8 @@ namespace MobileGL::MG_Impl {
        GETPROC(CGLGetPixelFormat, name);
        GETPROC(CGLSetCurrentContext, name);
        GETPROC(CGLGetCurrentContext, name);
+        GETPROC(CGLSetVirtualScreen, name);
+        GETPROC(CGLGetVirtualScreen, name);
        GETPROC(CGLSetParameter, name);
        GETPROC(CGLGetParameter, name);
        GETPROC(CGLUpdateContext, name);
@@ -29,10 +29,19 @@ namespace MobileGL::MG_Impl::NSOpenGLImpl {
        char kContextViewKey;
        char kContextLayerKey;

-        std::once_flag g_installOnce;
        IMP g_pixelFormatDealloc = nullptr;
        IMP g_contextDealloc = nullptr;

+        std::mutex& HookInstallMutex() {
+            static auto* mutex = new std::mutex();
+            return *mutex;
+        }
+
+        Bool& HooksInstalled() {
+            static auto* installed = new Bool(false);
+            return *installed;
+        }
+
        template <typename Fn>
        Fn ObjcMsgSend() {
            return reinterpret_cast<Fn>(objc_msgSend);
@@ -431,12 +440,12 @@ namespace MobileGL::MG_Impl::NSOpenGLImpl {
            method_setImplementation(method, replacement);
        }

-        void InstallHooksOnce() {
+        Bool InstallHooksOnce() {
            Class pixelFormatClass = objc_getClass("NSOpenGLPixelFormat");
            Class contextClass = objc_getClass("NSOpenGLContext");
            if (!pixelFormatClass || !contextClass) {
                MGLOG_W("NSOpenGLImpl: NSOpenGL classes are not loaded; hooks not installed");
-                return;
+                return false;
            }

            ReplaceInstanceMethod(pixelFormatClass, "initWithAttributes:",
@@ -471,11 +480,34 @@ namespace MobileGL::MG_Impl::NSOpenGLImpl {
            ReplaceInstanceMethod(contextClass, "dealloc", reinterpret_cast<IMP>(ContextDealloc), &g_contextDealloc);

            MGLOG_I("NSOpenGLImpl hooks installed");
+            return true;
        }
    } // namespace

    void InstallHooks() {
-        std::call_once(g_installOnce, InstallHooksOnce);
+        const std::lock_guard<std::mutex> lock(HookInstallMutex());
+        if (!HooksInstalled()) {
+            // Do not permanently consume the install attempt when the OpenGL
+            // framework has not registered its Objective-C classes yet. The
+            // dyld bootstrap normally runs after framework dependencies, but
+            // an explicitly loaded/static-linked MobileGL can arrive earlier.
+            HooksInstalled() = InstallHooksOnce();
+        }
    }
 } // namespace MobileGL::MG_Impl::NSOpenGLImpl
+
+namespace {
+    // SDL's Cocoa backend creates NSOpenGLPixelFormat/NSOpenGLContext before
+    // its first dlsym("glGetString") or other MobileGL host-API call. Install
+    // only the lightweight Objective-C dispatch hooks while the injected dylib
+    // is loading so those first Cocoa objects are routed through CGLImpl. The
+    // hooked context constructor reaches EGLImpl::GetDisplay(), which performs
+    // the full, thread-safe MobileGL initialization outside this bootstrap.
+    //
+    // There is intentionally no matching destructor: backend teardown remains
+    // owned by the EGL lifecycle and process-exit globals remain leak-at-exit.
+    __attribute__((constructor)) void BootstrapNSOpenGLHooks() {
+        MobileGL::MG_Impl::NSOpenGLImpl::InstallHooks();
+    }
+} // namespace
 #endif
@@ -0,0 +1,156 @@
+// MobileGL - MobileGL/MG_Impl/WGLImpl/Exporting/Definitions.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
+
+// wingdi.h declares most wgl* entry points as WINGDIAPI (__declspec(dllimport)),
+// which would reject our definitions. _GDI32_ is the SDK's "I am the module that
+// implements these" switch: it turns WINGDIAPI into a plain declaration. It must
+// be defined before the first windows.h inclusion in this translation unit.
+#if defined(_WIN32) && !defined(_GDI32_)
+#define _GDI32_ 1
+#endif
+
+#include <Includes.h>
+
+#if defined(_WIN32)
+#include "../WGLImpl.h"
+
+namespace WGL = MobileGL::MG_Impl::WGLImpl;
+
+// ---- Pixel-format entry points (gdi32 forwards ChoosePixelFormat/SetPixelFormat/
+// ---- DescribePixelFormat/GetPixelFormat/SwapBuffers into these exports) ----
+
+extern "C" int WINAPI wglChoosePixelFormat(HDC hdc, CONST PIXELFORMATDESCRIPTOR* ppfd) {
+    return WGL::ChoosePixelFormat(hdc, ppfd);
+}
+
+extern "C" int WINAPI wglDescribePixelFormat(HDC hdc, int iPixelFormat, UINT nBytes,
+                                                   LPPIXELFORMATDESCRIPTOR ppfd) {
+    return WGL::DescribePixelFormat(hdc, iPixelFormat, nBytes, ppfd);
+}
+
+extern "C" int WINAPI wglGetPixelFormat(HDC hdc) {
+    return WGL::GetPixelFormat(hdc);
+}
+
+extern "C" BOOL WINAPI wglSetPixelFormat(HDC hdc, int iPixelFormat, CONST PIXELFORMATDESCRIPTOR* ppfd) {
+    return WGL::SetPixelFormat(hdc, iPixelFormat, ppfd);
+}
+
+extern "C" BOOL WINAPI wglSwapBuffers(HDC hdc) {
+    return WGL::SwapBuffers(hdc);
+}
+
+// ---- Context management ----
+
+extern "C" HGLRC WINAPI wglCreateContext(HDC hdc) {
+    return WGL::CreateContext(hdc);
+}
+
+extern "C" HGLRC WINAPI wglCreateLayerContext(HDC hdc, int iLayerPlane) {
+    return iLayerPlane == 0 ? WGL::CreateContext(hdc) : nullptr;
+}
+
+extern "C" BOOL WINAPI wglCopyContext(HGLRC, HGLRC, UINT) {
+    MGLOG_W("wglCopyContext is not supported");
+    SetLastError(ERROR_NOT_SUPPORTED);
+    return FALSE;
+}
+
+extern "C" BOOL WINAPI wglDeleteContext(HGLRC hglrc) {
+    return WGL::DeleteContext(hglrc);
+}
+
+extern "C" HGLRC WINAPI wglGetCurrentContext(VOID) {
+    return WGL::GetCurrentContext();
+}
+
+extern "C" HDC WINAPI wglGetCurrentDC(VOID) {
+    return WGL::GetCurrentDC();
+}
+
+extern "C" BOOL WINAPI wglMakeCurrent(HDC hdc, HGLRC hglrc) {
+    return WGL::MakeCurrent(hdc, hglrc);
+}
+
+extern "C" BOOL WINAPI wglShareLists(HGLRC hglrcShare, HGLRC hglrcDest) {
+    return WGL::ShareLists(hglrcShare, hglrcDest);
+}
+
+// ---- Proc address ----
+
+extern "C" PROC WINAPI wglGetProcAddress(LPCSTR lpszProc) {
+    return WGL::GetProcAddress(lpszProc);
+}
+
+extern "C" PROC WINAPI wglGetDefaultProcAddress(LPCSTR lpszProc) {
+    return WGL::GetProcAddress(lpszProc);
+}
+
+// ---- Layer planes and palettes (unsupported; overlay planes do not exist here) ----
+
+extern "C" BOOL WINAPI wglDescribeLayerPlane(HDC, int, int, UINT, LPLAYERPLANEDESCRIPTOR) {
+    return FALSE;
+}
+
+extern "C" int WINAPI wglSetLayerPaletteEntries(HDC, int, int, int, CONST COLORREF*) {
+    return 0;
+}
+
+extern "C" int WINAPI wglGetLayerPaletteEntries(HDC, int, int, int, COLORREF*) {
+    return 0;
+}
+
+extern "C" BOOL WINAPI wglRealizeLayerPalette(HDC, int, BOOL) {
+    return FALSE;
+}
+
+extern "C" BOOL WINAPI wglSwapLayerBuffers(HDC hdc, UINT fuPlanes) {
+    if (fuPlanes & WGL_SWAP_MAIN_PLANE) {
+        return WGL::SwapBuffers(hdc);
+    }
+    return FALSE;
+}
+
+extern "C" DWORD WINAPI wglSwapMultipleBuffers(UINT n, CONST WGLSWAP* ps) {
+    if (!ps) {
+        return 0;
+    }
+    DWORD swapped = 0;
+    for (UINT i = 0; i < n; ++i) {
+        if (WGL::SwapBuffers(ps[i].hdc)) {
+            ++swapped;
+        }
+    }
+    return swapped;
+}
+
+// ---- Font rendering (legacy immediate-mode feature; not supported) ----
+
+extern "C" BOOL WINAPI wglUseFontBitmapsA(HDC, DWORD, DWORD, DWORD) {
+    MGLOG_W("wglUseFontBitmapsA is not supported");
+    return FALSE;
+}
+
+extern "C" BOOL WINAPI wglUseFontBitmapsW(HDC, DWORD, DWORD, DWORD) {
+    MGLOG_W("wglUseFontBitmapsW is not supported");
+    return FALSE;
+}
+
+extern "C" BOOL WINAPI wglUseFontOutlinesA(HDC, DWORD, DWORD, DWORD, FLOAT, FLOAT, int,
+                                                 LPGLYPHMETRICSFLOAT) {
+    MGLOG_W("wglUseFontOutlinesA is not supported");
+    return FALSE;
+}
+
+extern "C" BOOL WINAPI wglUseFontOutlinesW(HDC, DWORD, DWORD, DWORD, FLOAT, FLOAT, int,
+                                                 LPGLYPHMETRICSFLOAT) {
+    MGLOG_W("wglUseFontOutlinesW is not supported");
+    return FALSE;
+}
+
+#endif // _WIN32
@@ -0,0 +1,30 @@
+; MobileGL WGL exports. The wgl* entry points are defined without
+; __declspec(dllexport) because wingdi.h pre-declares them (with _GDI32_ they
+; become plain declarations, and MSVC rejects adding dllexport afterwards),
+; so this .def file is what actually exports them from the DLL.
+EXPORTS
+    wglChoosePixelFormat
+    wglCopyContext
+    wglCreateContext
+    wglCreateLayerContext
+    wglDeleteContext
+    wglDescribeLayerPlane
+    wglDescribePixelFormat
+    wglGetCurrentContext
+    wglGetCurrentDC
+    wglGetDefaultProcAddress
+    wglGetLayerPaletteEntries
+    wglGetPixelFormat
+    wglGetProcAddress
+    wglMakeCurrent
+    wglRealizeLayerPalette
+    wglSetLayerPaletteEntries
+    wglSetPixelFormat
+    wglShareLists
+    wglSwapBuffers
+    wglSwapLayerBuffers
+    wglSwapMultipleBuffers
+    wglUseFontBitmapsA
+    wglUseFontBitmapsW
+    wglUseFontOutlinesA
+    wglUseFontOutlinesW
@@ -0,0 +1,732 @@
+// MobileGL - MobileGL/MG_Impl/WGLImpl/WGLImpl.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 "WGLImpl.h"
+
+#if defined(_WIN32)
+#include "../EGLImpl/EGLImpl.h"
+#include "../GetProcAddress.h"
+#include <Init.h>
+
+namespace MobileGL::MG_Impl::WGLImpl {
+    namespace {
+        // WGL_ARB_pixel_format
+        constexpr int WGL_NUMBER_PIXEL_FORMATS_ARB = 0x2000;
+        constexpr int WGL_DRAW_TO_WINDOW_ARB = 0x2001;
+        constexpr int WGL_DRAW_TO_BITMAP_ARB = 0x2002;
+        constexpr int WGL_ACCELERATION_ARB = 0x2003;
+        constexpr int WGL_NEED_PALETTE_ARB = 0x2004;
+        constexpr int WGL_NEED_SYSTEM_PALETTE_ARB = 0x2005;
+        constexpr int WGL_SWAP_LAYER_BUFFERS_ARB = 0x2006;
+        constexpr int WGL_SWAP_METHOD_ARB = 0x2007;
+        constexpr int WGL_NUMBER_OVERLAYS_ARB = 0x2008;
+        constexpr int WGL_NUMBER_UNDERLAYS_ARB = 0x2009;
+        constexpr int WGL_TRANSPARENT_ARB = 0x200A;
+        constexpr int WGL_SHARE_DEPTH_ARB = 0x200C;
+        constexpr int WGL_SHARE_STENCIL_ARB = 0x200D;
+        constexpr int WGL_SHARE_ACCUM_ARB = 0x200E;
+        constexpr int WGL_SUPPORT_GDI_ARB = 0x200F;
+        constexpr int WGL_SUPPORT_OPENGL_ARB = 0x2010;
+        constexpr int WGL_DOUBLE_BUFFER_ARB = 0x2011;
+        constexpr int WGL_STEREO_ARB = 0x2012;
+        constexpr int WGL_PIXEL_TYPE_ARB = 0x2013;
+        constexpr int WGL_COLOR_BITS_ARB = 0x2014;
+        constexpr int WGL_RED_BITS_ARB = 0x2015;
+        constexpr int WGL_RED_SHIFT_ARB = 0x2016;
+        constexpr int WGL_GREEN_BITS_ARB = 0x2017;
+        constexpr int WGL_GREEN_SHIFT_ARB = 0x2018;
+        constexpr int WGL_BLUE_BITS_ARB = 0x2019;
+        constexpr int WGL_BLUE_SHIFT_ARB = 0x201A;
+        constexpr int WGL_ALPHA_BITS_ARB = 0x201B;
+        constexpr int WGL_ALPHA_SHIFT_ARB = 0x201C;
+        constexpr int WGL_ACCUM_BITS_ARB = 0x201D;
+        constexpr int WGL_ACCUM_RED_BITS_ARB = 0x201E;
+        constexpr int WGL_ACCUM_GREEN_BITS_ARB = 0x201F;
+        constexpr int WGL_ACCUM_BLUE_BITS_ARB = 0x2020;
+        constexpr int WGL_ACCUM_ALPHA_BITS_ARB = 0x2021;
+        constexpr int WGL_DEPTH_BITS_ARB = 0x2022;
+        constexpr int WGL_STENCIL_BITS_ARB = 0x2023;
+        constexpr int WGL_AUX_BUFFERS_ARB = 0x2024;
+        constexpr int WGL_NO_ACCELERATION_ARB = 0x2025;
+        constexpr int WGL_FULL_ACCELERATION_ARB = 0x2027;
+        constexpr int WGL_SWAP_EXCHANGE_ARB = 0x2028;
+        constexpr int WGL_TYPE_RGBA_ARB = 0x202B;
+        // WGL_ARB_multisample
+        constexpr int WGL_SAMPLE_BUFFERS_ARB = 0x2041;
+        constexpr int WGL_SAMPLES_ARB = 0x2042;
+        // WGL_ARB_create_context / _profile / _no_error
+        constexpr int WGL_CONTEXT_MAJOR_VERSION_ARB = 0x2091;
+        constexpr int WGL_CONTEXT_MINOR_VERSION_ARB = 0x2092;
+        constexpr int WGL_CONTEXT_LAYER_PLANE_ARB = 0x2093;
+        constexpr int WGL_CONTEXT_FLAGS_ARB = 0x2094;
+        constexpr int WGL_CONTEXT_PROFILE_MASK_ARB = 0x9126;
+        constexpr int WGL_CONTEXT_DEBUG_BIT_ARB = 0x0001;
+        constexpr int WGL_CONTEXT_FORWARD_COMPATIBLE_BIT_ARB = 0x0002;
+        constexpr int WGL_CONTEXT_CORE_PROFILE_BIT_ARB = 0x00000001;
+        constexpr int WGL_CONTEXT_COMPATIBILITY_PROFILE_BIT_ARB = 0x00000002;
+        constexpr int WGL_CONTEXT_OPENGL_NO_ERROR_ARB = 0x31B3;
+        constexpr DWORD ERROR_INVALID_VERSION_ARB = 0x2095;
+        constexpr DWORD ERROR_INVALID_PROFILE_ARB = 0x2096;
+
+        struct PixelFormatInfo {
+            GLint AlphaBits;
+            GLint DepthBits;
+            GLint StencilBits;
+        };
+
+        // Mirrors the two EGLState configs (RGBA8 + depth24, stencil 8 / stencil 0).
+        constexpr PixelFormatInfo kPixelFormats[] = {
+            {8, 24, 8},
+            {8, 24, 0},
+        };
+        constexpr int kPixelFormatCount = static_cast<int>(std::size(kPixelFormats));
+
+        struct ContextObject {
+            EGLDisplay Display = EGL_NO_DISPLAY;
+            EGLConfig Config = nullptr;
+            EGLContext Context = EGL_NO_CONTEXT;
+        };
+
+        struct WindowSurface {
+            EGLDisplay Display = EGL_NO_DISPLAY;
+            EGLSurface Surface = EGL_NO_SURFACE;
+            Uint32 Width = 0;
+            Uint32 Height = 0;
+        };
+
+        std::recursive_mutex& RegistryMutex() {
+            static auto* mutex = new std::recursive_mutex();
+            return *mutex;
+        }
+
+        UnorderedMap<HGLRC, ContextObject>& Contexts() {
+            static auto* contexts = new UnorderedMap<HGLRC, ContextObject>();
+            return *contexts;
+        }
+
+        UnorderedMap<HWND, WindowSurface>& WindowSurfaces() {
+            static auto* surfaces = new UnorderedMap<HWND, WindowSurface>();
+            return *surfaces;
+        }
+
+        UnorderedMap<HWND, int>& WindowPixelFormats() {
+            static auto* formats = new UnorderedMap<HWND, int>();
+            return *formats;
+        }
+
+        Uint64& NextContextHandle() {
+            static auto* handle = new Uint64(0x10000);
+            return *handle;
+        }
+
+        struct ThreadCurrent {
+            HDC DC = nullptr;
+            HGLRC Context = nullptr;
+        };
+        thread_local ThreadCurrent t_current;
+
+        Int& SwapIntervalShadow() {
+            static auto* interval = new Int(1);
+            return *interval;
+        }
+
+        void EnsureInitialized() {
+            // Initialize() loads backend libraries and glslang, which must not run
+            // under the loader lock; first WGL call is the earliest safe moment.
+            // MobileGL::EnsureInitialized (not a local once_flag) so a fresh init
+            // can follow a full teardown from the last eglTerminate.
+            MobileGL::EnsureInitialized();
+        }
+
+        EGLDisplay EnsureDisplay() {
+            EnsureInitialized();
+            EGLDisplay display = EGLImpl::GetDisplay(EGL_DEFAULT_DISPLAY);
+            if (display == EGL_NO_DISPLAY) {
+                return EGL_NO_DISPLAY;
+            }
+            if (!EGLImpl::Initialize(display, nullptr, nullptr)) {
+                return EGL_NO_DISPLAY;
+            }
+            return display;
+        }
+
+        HGLRC EncodeContext(Uint64 handle) {
+            return reinterpret_cast<HGLRC>(static_cast<SizeT>(handle));
+        }
+
+        ContextObject* TryGetContext(HGLRC hglrc) {
+            auto& contexts = Contexts();
+            auto it = contexts.find(hglrc);
+            return it == contexts.end() ? nullptr : &it->second;
+        }
+
+        const PixelFormatInfo& PixelFormatForWindow(HWND hwnd) {
+            auto& formats = WindowPixelFormats();
+            auto it = formats.find(hwnd);
+            int index = it == formats.end() ? 1 : it->second;
+            if (index < 1 || index > kPixelFormatCount) {
+                index = 1;
+            }
+            return kPixelFormats[index - 1];
+        }
+
+        Bool QueryClientSize(HWND hwnd, Uint32& width, Uint32& height) {
+            RECT rect{};
+            if (!GetClientRect(hwnd, &rect)) {
+                return false;
+            }
+            width = static_cast<Uint32>(std::max<LONG>(rect.right - rect.left, 1));
+            height = static_cast<Uint32>(std::max<LONG>(rect.bottom - rect.top, 1));
+            return true;
+        }
+
+        // The backends never query the HWND client size themselves; the WGL layer
+        // owns size discovery and pushes changes through the internal resize hook
+        // (same contract as the macOS CGL layer).
+        void SyncSurfaceSize(HWND hwnd, WindowSurface& surface) {
+            Uint32 width = 0;
+            Uint32 height = 0;
+            if (!QueryClientSize(hwnd, width, height)) {
+                return;
+            }
+            if (width == surface.Width && height == surface.Height) {
+                return;
+            }
+            if (EGLImpl::ResizePlatformWindowSurface(surface.Display, surface.Surface,
+                                                     static_cast<EGLint>(width), static_cast<EGLint>(height))) {
+                surface.Width = width;
+                surface.Height = height;
+            }
+        }
+
+        WindowSurface* EnsureWindowSurface(HWND hwnd, const ContextObject& context) {
+            auto& surfaces = WindowSurfaces();
+            auto it = surfaces.find(hwnd);
+            if (it != surfaces.end()) {
+                SyncSurfaceSize(hwnd, it->second);
+                return &it->second;
+            }
+
+            Uint32 width = 0;
+            Uint32 height = 0;
+            if (!QueryClientSize(hwnd, width, height)) {
+                MGLOG_E("wgl: GetClientRect failed for HWND %p", hwnd);
+                return nullptr;
+            }
+
+            const EGLAttrib attribs[] = {
+                EGL_WIDTH, static_cast<EGLAttrib>(width),
+                EGL_HEIGHT, static_cast<EGLAttrib>(height),
+                EGL_NONE,
+            };
+            EGLSurface surface =
+                EGLImpl::CreatePlatformWindowSurface(context.Display, context.Config, hwnd, attribs);
+            if (surface == EGL_NO_SURFACE) {
+                MGLOG_E("wgl: failed to create window surface for HWND %p (%ux%u)", hwnd, width, height);
+                return nullptr;
+            }
+
+            WindowSurface record;
+            record.Display = context.Display;
+            record.Surface = surface;
+            record.Width = width;
+            record.Height = height;
+            auto [inserted, _] = surfaces.emplace(hwnd, record);
+            return &inserted->second;
+        }
+
+        HGLRC CreateContextFromEGLAttribs(HDC hdc, HGLRC share, const EGLint* contextAttribs) {
+            const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+            EGLDisplay display = EnsureDisplay();
+            if (display == EGL_NO_DISPLAY) {
+                MGLOG_E("wgl: no EGL display");
+                return nullptr;
+            }
+            EGLImpl::BindAPI(EGL_OPENGL_API);
+
+            EGLContext shareContext = EGL_NO_CONTEXT;
+            if (share) {
+                auto* shareObject = TryGetContext(share);
+                if (!shareObject) {
+                    SetLastError(ERROR_INVALID_HANDLE);
+                    return nullptr;
+                }
+                shareContext = shareObject->Context;
+            }
+
+            HWND hwnd = WindowFromDC(hdc);
+            const PixelFormatInfo& pixelFormat = PixelFormatForWindow(hwnd);
+            const EGLint configAttribs[] = {
+                EGL_RED_SIZE, 8,
+                EGL_GREEN_SIZE, 8,
+                EGL_BLUE_SIZE, 8,
+                EGL_ALPHA_SIZE, pixelFormat.AlphaBits,
+                EGL_DEPTH_SIZE, pixelFormat.DepthBits,
+                EGL_STENCIL_SIZE, pixelFormat.StencilBits,
+                EGL_SURFACE_TYPE, EGL_WINDOW_BIT | EGL_PBUFFER_BIT,
+                EGL_RENDERABLE_TYPE, EGL_OPENGL_BIT,
+                EGL_NONE,
+            };
+            EGLConfig config = nullptr;
+            EGLint configCount = 0;
+            if (!EGLImpl::ChooseConfig(display, configAttribs, &config, 1, &configCount) || configCount <= 0) {
+                MGLOG_E("wgl: eglChooseConfig failed");
+                return nullptr;
+            }
+
+            EGLContext eglContext = EGLImpl::CreateContext(display, config, shareContext, contextAttribs);
+            if (eglContext == EGL_NO_CONTEXT) {
+                MGLOG_E("wgl: eglCreateContext failed");
+                return nullptr;
+            }
+
+            ContextObject object;
+            object.Display = display;
+            object.Config = config;
+            object.Context = eglContext;
+            const auto handle = EncodeContext(NextContextHandle()++);
+            Contexts()[handle] = object;
+            MGLOG_I("wgl: created context %p (EGL context %p)", handle, eglContext);
+            return handle;
+        }
+
+        // ---- WGL extension entry points (resolved via wglGetProcAddress only) ----
+
+        const char* WINAPI Ext_GetExtensionsStringARB(HDC) {
+            return "WGL_ARB_create_context WGL_ARB_create_context_no_error WGL_ARB_create_context_profile "
+                   "WGL_ARB_extensions_string WGL_ARB_pixel_format WGL_EXT_extensions_string WGL_EXT_swap_control";
+        }
+
+        const char* WINAPI Ext_GetExtensionsStringEXT() {
+            return Ext_GetExtensionsStringARB(nullptr);
+        }
+
+        HGLRC WINAPI Ext_CreateContextAttribsARB(HDC hdc, HGLRC hShareContext, const int* attribList) {
+            EnsureInitialized();
+            int major = 1;
+            int minor = 0;
+            int profileMask = 0;
+            int flags = 0;
+            if (attribList) {
+                for (SizeT i = 0; attribList[i] != 0; i += 2) {
+                    const int attrib = attribList[i];
+                    const int value = attribList[i + 1];
+                    switch (attrib) {
+                    case WGL_CONTEXT_MAJOR_VERSION_ARB:
+                        major = value;
+                        break;
+                    case WGL_CONTEXT_MINOR_VERSION_ARB:
+                        minor = value;
+                        break;
+                    case WGL_CONTEXT_PROFILE_MASK_ARB:
+                        profileMask = value;
+                        break;
+                    case WGL_CONTEXT_FLAGS_ARB:
+                        flags = value;
+                        break;
+                    case WGL_CONTEXT_LAYER_PLANE_ARB:
+                        if (value != 0) {
+                            SetLastError(ERROR_INVALID_PARAMETER);
+                            return nullptr;
+                        }
+                        break;
+                    case WGL_CONTEXT_OPENGL_NO_ERROR_ARB:
+                        // Accepted and ignored: MobileGL always validates.
+                        break;
+                    default:
+                        MGLOG_D("wglCreateContextAttribsARB: ignoring attrib 0x%04x = 0x%x", attrib, value);
+                        break;
+                    }
+                }
+            }
+
+            if (major < 1 || (profileMask & ~(WGL_CONTEXT_CORE_PROFILE_BIT_ARB |
+                                              WGL_CONTEXT_COMPATIBILITY_PROFILE_BIT_ARB))) {
+                SetLastError(profileMask ? ERROR_INVALID_PROFILE_ARB : ERROR_INVALID_VERSION_ARB);
+                return nullptr;
+            }
+
+            Vector<EGLint> attribs = {
+                EGL_CONTEXT_MAJOR_VERSION, major,
+                EGL_CONTEXT_MINOR_VERSION, minor,
+            };
+            const Bool wantsCompat = (profileMask & WGL_CONTEXT_COMPATIBILITY_PROFILE_BIT_ARB) != 0;
+            if (major > 3 || (major == 3 && minor >= 2) || profileMask != 0) {
+                attribs.push_back(EGL_CONTEXT_OPENGL_PROFILE_MASK);
+                attribs.push_back(wantsCompat ? EGL_CONTEXT_OPENGL_COMPATIBILITY_PROFILE_BIT
+                                              : EGL_CONTEXT_OPENGL_CORE_PROFILE_BIT);
+            }
+            if (flags & WGL_CONTEXT_FORWARD_COMPATIBLE_BIT_ARB) {
+                attribs.push_back(EGL_CONTEXT_OPENGL_FORWARD_COMPATIBLE);
+                attribs.push_back(EGL_TRUE);
+            }
+            if (flags & WGL_CONTEXT_DEBUG_BIT_ARB) {
+                attribs.push_back(EGL_CONTEXT_OPENGL_DEBUG);
+                attribs.push_back(EGL_TRUE);
+            }
+            attribs.push_back(EGL_NONE);
+
+            return CreateContextFromEGLAttribs(hdc, hShareContext, attribs.data());
+        }
+
+        BOOL WINAPI Ext_SwapIntervalEXT(int interval) {
+            const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+            EGLDisplay display = EnsureDisplay();
+            if (display == EGL_NO_DISPLAY) {
+                return FALSE;
+            }
+            if (interval < 0) {
+                // Adaptive vsync is not supported; clamp to regular vsync.
+                interval = 1;
+            }
+            EGLImpl::SwapInterval(display, interval);
+            SwapIntervalShadow() = interval;
+            return TRUE;
+        }
+
+        int WINAPI Ext_GetSwapIntervalEXT() {
+            const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+            return SwapIntervalShadow();
+        }
+
+        int PixelFormatAttribValue(int format, int attrib) {
+            const PixelFormatInfo& info = kPixelFormats[format - 1];
+            switch (attrib) {
+            case WGL_NUMBER_PIXEL_FORMATS_ARB:
+                return kPixelFormatCount;
+            case WGL_SUPPORT_OPENGL_ARB:
+            case WGL_DRAW_TO_WINDOW_ARB:
+            case WGL_DOUBLE_BUFFER_ARB:
+                return 1;
+            case WGL_ACCELERATION_ARB:
+                return WGL_FULL_ACCELERATION_ARB;
+            case WGL_PIXEL_TYPE_ARB:
+                return WGL_TYPE_RGBA_ARB;
+            case WGL_COLOR_BITS_ARB:
+                return 32;
+            case WGL_RED_BITS_ARB:
+            case WGL_GREEN_BITS_ARB:
+            case WGL_BLUE_BITS_ARB:
+                return 8;
+            case WGL_RED_SHIFT_ARB:
+                return 16;
+            case WGL_GREEN_SHIFT_ARB:
+                return 8;
+            case WGL_BLUE_SHIFT_ARB:
+                return 0;
+            case WGL_ALPHA_BITS_ARB:
+                return info.AlphaBits;
+            case WGL_ALPHA_SHIFT_ARB:
+                return 24;
+            case WGL_DEPTH_BITS_ARB:
+                return info.DepthBits;
+            case WGL_STENCIL_BITS_ARB:
+                return info.StencilBits;
+            case WGL_SWAP_METHOD_ARB:
+                return WGL_SWAP_EXCHANGE_ARB;
+            case WGL_DRAW_TO_BITMAP_ARB:
+            case WGL_NEED_PALETTE_ARB:
+            case WGL_NEED_SYSTEM_PALETTE_ARB:
+            case WGL_SWAP_LAYER_BUFFERS_ARB:
+            case WGL_NUMBER_OVERLAYS_ARB:
+            case WGL_NUMBER_UNDERLAYS_ARB:
+            case WGL_TRANSPARENT_ARB:
+            case WGL_SHARE_DEPTH_ARB:
+            case WGL_SHARE_STENCIL_ARB:
+            case WGL_SHARE_ACCUM_ARB:
+            case WGL_SUPPORT_GDI_ARB:
+            case WGL_STEREO_ARB:
+            case WGL_ACCUM_BITS_ARB:
+            case WGL_ACCUM_RED_BITS_ARB:
+            case WGL_ACCUM_GREEN_BITS_ARB:
+            case WGL_ACCUM_BLUE_BITS_ARB:
+            case WGL_ACCUM_ALPHA_BITS_ARB:
+            case WGL_AUX_BUFFERS_ARB:
+            case WGL_SAMPLE_BUFFERS_ARB:
+            case WGL_SAMPLES_ARB:
+            default:
+                return 0;
+            }
+        }
+
+        BOOL WINAPI Ext_GetPixelFormatAttribivARB(HDC, int iPixelFormat, int iLayerPlane, UINT nAttributes,
+                                                  const int* piAttributes, int* piValues) {
+            if (iLayerPlane != 0 || !piAttributes || !piValues) {
+                return FALSE;
+            }
+            // Format 0 is only valid for WGL_NUMBER_PIXEL_FORMATS_ARB queries.
+            if (iPixelFormat < 0 || iPixelFormat > kPixelFormatCount) {
+                return FALSE;
+            }
+            const int format = iPixelFormat == 0 ? 1 : iPixelFormat;
+            for (UINT i = 0; i < nAttributes; ++i) {
+                piValues[i] = PixelFormatAttribValue(format, piAttributes[i]);
+            }
+            return TRUE;
+        }
+
+        BOOL WINAPI Ext_GetPixelFormatAttribfvARB(HDC hdc, int iPixelFormat, int iLayerPlane, UINT nAttributes,
+                                                  const int* piAttributes, FLOAT* pfValues) {
+            if (!pfValues) {
+                return FALSE;
+            }
+            Vector<int> values(nAttributes);
+            if (!Ext_GetPixelFormatAttribivARB(hdc, iPixelFormat, iLayerPlane, nAttributes, piAttributes,
+                                               values.data())) {
+                return FALSE;
+            }
+            for (UINT i = 0; i < nAttributes; ++i) {
+                pfValues[i] = static_cast<FLOAT>(values[i]);
+            }
+            return TRUE;
+        }
+
+        BOOL WINAPI Ext_ChoosePixelFormatARB(HDC, const int* piAttribIList, const FLOAT*, UINT nMaxFormats,
+                                             int* piFormats, UINT* nNumFormats) {
+            if (!piFormats || !nNumFormats) {
+                return FALSE;
+            }
+            int wantedStencil = 0;
+            if (piAttribIList) {
+                for (SizeT i = 0; piAttribIList[i] != 0; i += 2) {
+                    if (piAttribIList[i] == WGL_STENCIL_BITS_ARB) {
+                        wantedStencil = piAttribIList[i + 1];
+                    }
+                }
+            }
+            UINT count = 0;
+            const int preferred = wantedStencil > 0 ? 1 : 2;
+            const int fallback = wantedStencil > 0 ? 2 : 1;
+            if (count < nMaxFormats) {
+                piFormats[count++] = preferred;
+            }
+            if (count < nMaxFormats) {
+                piFormats[count++] = fallback;
+            }
+            *nNumFormats = count;
+            return TRUE;
+        }
+
+        struct WGLExtensionProc {
+            const char* Name;
+            PROC Proc;
+        };
+
+        const WGLExtensionProc kWGLExtensionProcs[] = {
+            {"wglGetExtensionsStringARB", reinterpret_cast<PROC>(Ext_GetExtensionsStringARB)},
+            {"wglGetExtensionsStringEXT", reinterpret_cast<PROC>(Ext_GetExtensionsStringEXT)},
+            {"wglCreateContextAttribsARB", reinterpret_cast<PROC>(Ext_CreateContextAttribsARB)},
+            {"wglSwapIntervalEXT", reinterpret_cast<PROC>(Ext_SwapIntervalEXT)},
+            {"wglGetSwapIntervalEXT", reinterpret_cast<PROC>(Ext_GetSwapIntervalEXT)},
+            {"wglGetPixelFormatAttribivARB", reinterpret_cast<PROC>(Ext_GetPixelFormatAttribivARB)},
+            {"wglGetPixelFormatAttribfvARB", reinterpret_cast<PROC>(Ext_GetPixelFormatAttribfvARB)},
+            {"wglChoosePixelFormatARB", reinterpret_cast<PROC>(Ext_ChoosePixelFormatARB)},
+        };
+    } // namespace
+
+    int ChoosePixelFormat(HDC hdc, const PIXELFORMATDESCRIPTOR* pfd) {
+        EnsureInitialized();
+        MGLOG_D("wglChoosePixelFormat(hdc=%p)", hdc);
+        // Format 1 (RGBA8 + depth24/stencil8) satisfies every request; a format
+        // exceeding the asked-for capabilities is a legal ChoosePixelFormat answer.
+        (void)pfd;
+        return 1;
+    }
+
+    int DescribePixelFormat(HDC hdc, int format, UINT size, PIXELFORMATDESCRIPTOR* pfd) {
+        EnsureInitialized();
+        MGLOG_D("wglDescribePixelFormat(hdc=%p, format=%d)", hdc, format);
+        if (!pfd) {
+            return kPixelFormatCount;
+        }
+        if (size < sizeof(PIXELFORMATDESCRIPTOR) || format < 1 || format > kPixelFormatCount) {
+            return 0;
+        }
+
+        const PixelFormatInfo& info = kPixelFormats[format - 1];
+        std::memset(pfd, 0, sizeof(PIXELFORMATDESCRIPTOR));
+        pfd->nSize = sizeof(PIXELFORMATDESCRIPTOR);
+        pfd->nVersion = 1;
+        pfd->dwFlags = PFD_DRAW_TO_WINDOW | PFD_SUPPORT_OPENGL | PFD_DOUBLEBUFFER | PFD_SWAP_EXCHANGE
+#if defined(PFD_SUPPORT_COMPOSITION)
+                       | PFD_SUPPORT_COMPOSITION
+#endif
+            ;
+        pfd->iPixelType = PFD_TYPE_RGBA;
+        pfd->cColorBits = 32;
+        pfd->cRedBits = 8;
+        pfd->cRedShift = 16;
+        pfd->cGreenBits = 8;
+        pfd->cGreenShift = 8;
+        pfd->cBlueBits = 8;
+        pfd->cBlueShift = 0;
+        pfd->cAlphaBits = static_cast<BYTE>(info.AlphaBits);
+        pfd->cAlphaShift = 24;
+        pfd->cDepthBits = static_cast<BYTE>(info.DepthBits);
+        pfd->cStencilBits = static_cast<BYTE>(info.StencilBits);
+        pfd->iLayerType = PFD_MAIN_PLANE;
+        return kPixelFormatCount;
+    }
+
+    int GetPixelFormat(HDC hdc) {
+        EnsureInitialized();
+        HWND hwnd = WindowFromDC(hdc);
+        if (!hwnd) {
+            return 0;
+        }
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        auto& formats = WindowPixelFormats();
+        auto it = formats.find(hwnd);
+        return it == formats.end() ? 0 : it->second;
+    }
+
+    BOOL SetPixelFormat(HDC hdc, int format, const PIXELFORMATDESCRIPTOR*) {
+        EnsureInitialized();
+        MGLOG_D("wglSetPixelFormat(hdc=%p, format=%d)", hdc, format);
+        if (format < 1 || format > kPixelFormatCount) {
+            SetLastError(ERROR_INVALID_PARAMETER);
+            return FALSE;
+        }
+        HWND hwnd = WindowFromDC(hdc);
+        if (!hwnd) {
+            SetLastError(ERROR_INVALID_HANDLE);
+            return FALSE;
+        }
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        WindowPixelFormats()[hwnd] = format;
+        return TRUE;
+    }
+
+    BOOL SwapBuffers(HDC hdc) {
+        HWND hwnd = WindowFromDC(hdc);
+        if (!hwnd) {
+            SetLastError(ERROR_INVALID_HANDLE);
+            return FALSE;
+        }
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        auto& surfaces = WindowSurfaces();
+        auto it = surfaces.find(hwnd);
+        if (it == surfaces.end()) {
+            MGLOG_W("wglSwapBuffers: no surface for HWND %p", hwnd);
+            return FALSE;
+        }
+        SyncSurfaceSize(hwnd, it->second);
+        return EGLImpl::SwapBuffers(it->second.Display, it->second.Surface) == EGL_TRUE ? TRUE : FALSE;
+    }
+
+    HGLRC CreateContext(HDC hdc) {
+        EnsureInitialized();
+        MGLOG_I("wglCreateContext(hdc=%p)", hdc);
+        // A legacy WGL context is a compatibility-profile context; MobileGL keys
+        // its relaxed-semantics mode off the explicit compatibility bit.
+        const EGLint attribs[] = {
+            EGL_CONTEXT_MAJOR_VERSION, 3,
+            EGL_CONTEXT_MINOR_VERSION, 3,
+            EGL_CONTEXT_OPENGL_PROFILE_MASK, EGL_CONTEXT_OPENGL_COMPATIBILITY_PROFILE_BIT,
+            EGL_NONE,
+        };
+        return CreateContextFromEGLAttribs(hdc, nullptr, attribs);
+    }
+
+    BOOL DeleteContext(HGLRC hglrc) {
+        EnsureInitialized();
+        MGLOG_I("wglDeleteContext(%p)", hglrc);
+        if (t_current.Context == hglrc) {
+            MakeCurrent(nullptr, nullptr);
+        }
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        auto* object = TryGetContext(hglrc);
+        if (!object) {
+            SetLastError(ERROR_INVALID_HANDLE);
+            return FALSE;
+        }
+        if (object->Context != EGL_NO_CONTEXT) {
+            EGLImpl::DestroyContext(object->Display, object->Context);
+        }
+        Contexts().erase(hglrc);
+        return TRUE;
+    }
+
+    BOOL MakeCurrent(HDC hdc, HGLRC hglrc) {
+        EnsureInitialized();
+        MGLOG_D("wglMakeCurrent(hdc=%p, hglrc=%p)", hdc, hglrc);
+        if (!hglrc) {
+            if (!t_current.Context) {
+                t_current = {};
+                return TRUE;
+            }
+            const EGLBoolean released =
+                EGLImpl::MakeCurrent(EGL_NO_DISPLAY, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT);
+            t_current = {};
+            return released == EGL_TRUE ? TRUE : FALSE;
+        }
+
+        HWND hwnd = WindowFromDC(hdc);
+        if (!hwnd) {
+            SetLastError(ERROR_INVALID_HANDLE);
+            return FALSE;
+        }
+
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        auto* object = TryGetContext(hglrc);
+        if (!object) {
+            SetLastError(ERROR_INVALID_HANDLE);
+            return FALSE;
+        }
+
+        WindowSurface* surface = EnsureWindowSurface(hwnd, *object);
+        if (!surface) {
+            return FALSE;
+        }
+
+        if (!EGLImpl::MakeCurrent(object->Display, surface->Surface, surface->Surface, object->Context)) {
+            MGLOG_E("wglMakeCurrent: eglMakeCurrent failed (hdc=%p, hglrc=%p)", hdc, hglrc);
+            return FALSE;
+        }
+        t_current = {hdc, hglrc};
+        return TRUE;
+    }
+
+    HGLRC GetCurrentContext() {
+        return t_current.Context;
+    }
+
+    HDC GetCurrentDC() {
+        return t_current.DC;
+    }
+
+    BOOL ShareLists(HGLRC hglrcShare, HGLRC hglrcDest) {
+        // All MobileGL contexts alias one global GL object namespace, so every
+        // pair of contexts already shares.
+        const std::lock_guard<std::recursive_mutex> lock(RegistryMutex());
+        if (!TryGetContext(hglrcShare) || !TryGetContext(hglrcDest)) {
+            SetLastError(ERROR_INVALID_HANDLE);
+            return FALSE;
+        }
+        return TRUE;
+    }
+
+    PROC GetProcAddress(const char* name) {
+        EnsureInitialized();
+        if (!name) {
+            return nullptr;
+        }
+        if (name[0] == 'w' && name[1] == 'g' && name[2] == 'l') {
+            for (const auto& entry : kWGLExtensionProcs) {
+                if (std::strcmp(entry.Name, name) == 0) {
+                    return entry.Proc;
+                }
+            }
+            MGLOG_D("wglGetProcAddress: unknown wgl entry point %s", name);
+            return nullptr;
+        }
+        return reinterpret_cast<PROC>(MG_Impl::GetProcAddress(name));
+    }
+} // namespace MobileGL::MG_Impl::WGLImpl
+
+#endif // _WIN32
@@ -0,0 +1,34 @@
+// MobileGL - MobileGL/MG_Impl/WGLImpl/WGLImpl.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>
+
+#if defined(_WIN32)
+
+namespace MobileGL::MG_Impl::WGLImpl {
+    // Classic opengl32.dll surface. gdi32's ChoosePixelFormat/SetPixelFormat/
+    // DescribePixelFormat/GetPixelFormat/SwapBuffers forward into the loaded
+    // opengl32.dll's wgl* exports, so these back both call paths.
+    int ChoosePixelFormat(HDC hdc, const PIXELFORMATDESCRIPTOR* pfd);
+    int DescribePixelFormat(HDC hdc, int format, UINT size, PIXELFORMATDESCRIPTOR* pfd);
+    int GetPixelFormat(HDC hdc);
+    BOOL SetPixelFormat(HDC hdc, int format, const PIXELFORMATDESCRIPTOR* pfd);
+    BOOL SwapBuffers(HDC hdc);
+
+    HGLRC CreateContext(HDC hdc);
+    BOOL DeleteContext(HGLRC hglrc);
+    BOOL MakeCurrent(HDC hdc, HGLRC hglrc);
+    HGLRC GetCurrentContext();
+    HDC GetCurrentDC();
+    BOOL ShareLists(HGLRC hglrcShare, HGLRC hglrcDest);
+
+    PROC GetProcAddress(const char* name);
+} // namespace MobileGL::MG_Impl::WGLImpl
+
+#endif // _WIN32
@@ -368,6 +368,26 @@ namespace MobileGL {
                return true;
            }

+            Bool EGLContext::HasAnyInitializedDisplay() const {
+                const std::lock_guard<std::recursive_mutex> lock(m_mutex);
+                for (const auto& [handle, displayObject] : m_displays) {
+                    if (displayObject.Initialized) {
+                        return true;
+                    }
+                }
+                return false;
+            }
+
+            Bool EGLContext::HasAnyCurrentContext() const {
+                const std::lock_guard<std::recursive_mutex> lock(m_mutex);
+                for (const auto& [threadId, current] : m_threadCurrents) {
+                    if (current.Context != nullptr) {
+                        return true;
+                    }
+                }
+                return false;
+            }
+
            Bool EGLContext::ChooseConfig(EGLDisplayHandle display, const EGLint* attribList, EGLConfigHandle* configs,
                                          EGLint configSize, EGLint* numConfig) {
                const std::lock_guard<std::recursive_mutex> lock(m_mutex);
@@ -764,6 +784,20 @@ 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());
@@ -1401,6 +1435,7 @@ namespace MobileGL {
            }
        } // namespace EGLState

-        UniquePtr<EGLState::EGLContext> pEGLContext;
+        // Leak-at-exit storage; see GlobalObjects.cpp.
+        UniquePtr<EGLState::EGLContext>& pEGLContext = *new UniquePtr<EGLState::EGLContext>();
    } // namespace MG_State
 } // namespace MobileGL
@@ -39,6 +39,10 @@ namespace MobileGL {
                Bool IsDisplayInitialized(EGLDisplayHandle display) const;
                Bool InitializeDisplay(EGLDisplayHandle display, EGLint* major, EGLint* minor);
                Bool TerminateDisplay(EGLDisplayHandle display);
+                // Whole-library idle checks used by EGLImpl::Terminate to decide
+                // when the last eglTerminate may tear MobileGL down entirely.
+                Bool HasAnyInitializedDisplay() const;
+                Bool HasAnyCurrentContext() const;

                // Config
                Bool ChooseConfig(EGLDisplayHandle display, const EGLint* attribList, EGLConfigHandle* configs,
@@ -59,6 +63,7 @@ namespace MobileGL {
                Bool ValidateContext(EGLContextHandle context) const;
                Bool ValidateContextOnDisplay(EGLDisplayHandle display, EGLContextHandle context) const;
                Bool IsCurrentContextOpenGLCoreProfile() const;
+                Bool IsCurrentContextOpenGLCompatibilityProfile() const;
                EGLint GetCurrentContextFlags() const;

                // Surface
@@ -261,6 +266,6 @@ namespace MobileGL {
            };
        } // namespace EGLState

-        extern UniquePtr<EGLState::EGLContext> pEGLContext;
+        extern UniquePtr<EGLState::EGLContext>& pEGLContext;
    } // namespace MG_State
 } // namespace MobileGL
@@ -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);
        }
@@ -707,5 +721,6 @@ namespace MobileGL::MG_State {
        }
    } // namespace GLState

-    UniquePtr<GLState::GLContext> pGLContext;
+    // Leak-at-exit storage; see GlobalObjects.cpp.
+    UniquePtr<GLState::GLContext>& pGLContext = *new UniquePtr<GLState::GLContext>();
 } // namespace MobileGL::MG_State
@@ -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);
@@ -246,6 +252,13 @@ namespace MobileGL {
            };
        } // namespace GLState

-        extern UniquePtr<GLState::GLContext> pGLContext;
+        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
@@ -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)));
+            }
        }
    }

@@ -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 {
@@ -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)); }

@@ -255,16 +334,32 @@ namespace MobileGL::MG_State::GLState {
        // draw. The memo is keyed by (backendStateVersion, flags); ResetLinkArtifacts and
        // the binding setters below invalidate it by bumping m_backendStateVersion.
        Bool GetBackendHashMemo(Uint flags, Uint64& outHash) const {
-            if (m_backendHashMemoVersion != m_backendStateVersion || m_backendHashMemoFlags != flags) {
-                return false;
+            if (m_backendHashMemoVersion != m_backendStateVersion) return false;
+            for (const auto& slot : m_backendHashMemoSlots) {
+                if (slot.valid && slot.flags == flags) {
+                    outHash = slot.hash;
+                    return true;
+                }
            }
-            outHash = m_backendHashMemo;
-            return true;
+            return false;
        }
        void SetBackendHashMemo(Uint flags, Uint64 hash) const {
-            m_backendHashMemo = hash;
-            m_backendHashMemoVersion = m_backendStateVersion;
-            m_backendHashMemoFlags = flags;
+            if (m_backendHashMemoVersion != m_backendStateVersion) {
+                for (auto& slot : m_backendHashMemoSlots) slot.valid = false;
+                m_backendHashMemoVersion = m_backendStateVersion;
+                m_backendHashMemoNextSlot = 0;
+            }
+            for (auto& slot : m_backendHashMemoSlots) {
+                if (slot.valid && slot.flags == flags) {
+                    slot.hash = hash;
+                    return;
+                }
+            }
+            auto& slot = m_backendHashMemoSlots[m_backendHashMemoNextSlot];
+            slot.flags = flags;
+            slot.hash = hash;
+            slot.valid = true;
+            m_backendHashMemoNextSlot = (m_backendHashMemoNextSlot + 1) % kBackendHashMemoSlotCount;
        }

        void SetUniformSamplerOrImageUnitIndex(Uint location, Int unit) {
@@ -292,6 +387,11 @@ 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 +400,11 @@ 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 +412,30 @@ 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 {
@@ -417,10 +543,19 @@ namespace MobileGL::MG_State::GLState {
        Uint32 m_backendStateVersion = 0;

        // Backend-owned content-hash memo (see GetBackendHashMemo): valid only while
-        // m_backendStateVersion and the compile flags match the recorded values.
-        mutable Uint64 m_backendHashMemo = 0;
+        // m_backendStateVersion matches. Several slots, not one: a backend may resolve the same
+        // program under more than one compile-flag set within a frame (surface rotation, and the
+        // explicit-LOD sampling variant), and a single slot would then miss on every lookup and
+        // re-hash the program's whole SPIR-V once per draw.
+        static constexpr SizeT kBackendHashMemoSlotCount = 4;
+        struct BackendHashMemoSlot {
+            Uint64 hash = 0;
+            Uint flags = 0;
+            Bool valid = false;
+        };
+        mutable Array<BackendHashMemoSlot, kBackendHashMemoSlotCount> m_backendHashMemoSlots{};
+        mutable SizeT m_backendHashMemoNextSlot = 0;
        mutable Uint32 m_backendHashMemoVersion = ~0u;
-        mutable Uint m_backendHashMemoFlags = 0;
        Uint32 m_uboContentVersion = 0;
        Uint32 m_linkVersion = 0;
    };
@@ -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;
    }
@@ -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();
@@ -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;
            }
@@ -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;
@@ -16,17 +16,32 @@ namespace MobileGL {
            }

            void MipmapStorage::AllocateLevel(Uint level, MipmapInput input) {
-                m_data.reserve(std::bit_ceil(level + 1));
-                m_data.resize(level + 1);
-                m_texelSizes.reserve(std::bit_ceil(level + 1));
-                m_texelSizes.resize(level + 1);
-                m_texelSizes[level] = input.texelSize;
-                m_isDirty.resize(level + 1, false);
+                // Grow only. GL respecifies exactly the level it is handed, so allocating level 0
+                // must not disturb the levels above it - but resize() shrinks as readily as it
+                // grows, so this used to truncate the whole chain to a single level. Callers that
+                // genuinely redefine the complete level set say so with TruncateToLevelCount.
+                const SizeT requiredLevelCount = static_cast<SizeT>(level) + 1;
+                if (m_data.size() < requiredLevelCount) {
+                    m_data.reserve(std::bit_ceil(requiredLevelCount));
+                    m_data.resize(requiredLevelCount);
+                    m_texelSizes.reserve(std::bit_ceil(requiredLevelCount));
+                    m_texelSizes.resize(requiredLevelCount);
+                    m_isDirty.resize(requiredLevelCount, false);
+                }

+                m_texelSizes[level] = input.texelSize;
                auto& data = m_data[level];
                data.resize(input.byteSize, 0);
            }

+            void MipmapStorage::TruncateToLevelCount(SizeT levelCount) {
+                if (levelCount >= m_data.size()) return;
+
+                m_data.resize(levelCount);
+                m_texelSizes.resize(levelCount);
+                m_isDirty.resize(levelCount);
+            }
+
            void MipmapStorage::UpdateSubData(Uint level, DataPtr input) {
                auto& targetData = m_data;
                MOBILEGL_ASSERT(level < targetData.size(), "UpdateSubData: level out of range");
@@ -55,6 +70,7 @@ namespace MobileGL {
            }

            SizeT MipmapStorage::GetByteSize(Uint level) const {
+                if (level >= m_data.size()) return 0;
                return m_data[level].size();
            }

@@ -19,6 +19,10 @@ namespace MobileGL {
            public:
                SizeT GetLevelCount() const;
                void AllocateLevel(Uint level, MipmapInput input);
+                // Discard every level at or above levelCount. AllocateLevel never shrinks, so this
+                // is the only way a chain gets shorter - use it where the caller defines the whole
+                // level set (glTexStorage*, mip regeneration, atlas respecification).
+                void TruncateToLevelCount(SizeT levelCount);
                void UpdateSubData(Uint level, DataPtr input);
                void* MapData(Uint level);
                IntVec3 GetTexelSize(Uint level) const;
@@ -29,6 +29,14 @@ namespace MobileGL {
                    m_storage[targetIndex].AllocateLevel(level, input);
                }

+                // Per-target, like AllocateLevel: cube-map faces are respecified independently, so
+                // truncating one face must not disturb the others.
+                void TruncateToLevelCount(Uint targetIndex, SizeT levelCount) {
+                    MOBILEGL_ASSERT(targetIndex < TargetCount, "TruncateToLevelCount: target invalid");
+
+                    m_storage[targetIndex].TruncateToLevelCount(levelCount);
+                }
+
                void UpdateSubData(Uint targetIndex, Uint level, DataPtr input) {
                    MOBILEGL_ASSERT(targetIndex < TargetCount, "UpdateSubData: target invalid");
                    m_storage[targetIndex].UpdateSubData(level, input);
@@ -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,
@@ -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;
            }
@@ -256,6 +271,10 @@ namespace MobileGL {
                m_textureStorage.AllocateLevel(GetIndexOfTextureUploadTarget(uploadTarget), mipmapLevel, input);
            }

+            void TextureObjectWithOneMipmap::TruncateMipmapLevels(TextureUploadTarget uploadTarget, Uint levelCount) {
+                m_textureStorage.TruncateToLevelCount(GetIndexOfTextureUploadTarget(uploadTarget), levelCount);
+            }
+
            void TextureObjectWithOneMipmap::UpdateMipmapSubData(TextureUploadTarget uploadTarget, Uint mipmapLevel,
                                                                 DataPtr input) {
                m_textureStorage.UpdateSubData(GetIndexOfTextureUploadTarget(uploadTarget), mipmapLevel, input);
--- a/Show More
+++ b/Show More