Updated to version 3.2.28 for release

quit: don't call signal() if we're using sigaction()
At best, this is a no-op. At worst, it might: - Clobber a signal handler someone registered after us - Overwrite the signal mask or flags - Cause unregistration to fail (sigaction() isn't guaranteed to return the exact pointer passed to signal()) (cherry picked from commit 6d99204a82)
2025-12-02 07:47:05 -08:00 · 2025-12-01 19:47:32 -06:00 · 2025-11-30 11:37:34 -08:00 · 2025-11-28 07:42:09 -08:00 · 2025-11-26 17:25:50 +03:00 · 2025-11-26 14:05:37 +01:00
228 changed files with 4584 additions and 2285 deletions
@@ -20,7 +20,6 @@ class AppleArch(Enum):
 class MsvcArch(Enum):
    X86 = "x86"
    X64 = "x64"
-    Arm32 = "arm"
    Arm64 = "arm64"


@@ -31,7 +30,9 @@ class JobOs(Enum):
    Ubuntu22_04 = "ubuntu-22.04"
    Ubuntu24_04 = "ubuntu-24.04"
    Ubuntu24_04_arm = "ubuntu-24.04-arm"
-    Macos13 = "macos-13"
+    Macos14 = "macos-14"  # macOS Sonoma (2023)
+    Macos15 = "macos-15"  # macOS Sequoia (2024)
+    Macos26 = "macos-26"  # macOS Tahoe (2025)


 class SdlPlatform(Enum):
@@ -54,6 +55,7 @@ class SdlPlatform(Enum):
    Riscos = "riscos"
    FreeBSD = "freebsd"
    NetBSD = "netbsd"
+    OpenBSD = "openbsd"


 class Msys2Platform(Enum):
@@ -108,7 +110,6 @@ JOB_SPECS = {
    "msvc-x86": JobSpec(name="Windows (MSVC, x86)",                         os=JobOs.WindowsLatest,     platform=SdlPlatform.Msvc,        artifact="SDL-VC-x86",             msvc_arch=MsvcArch.X86,   msvc_project="VisualC/SDL.sln", ),
    "msvc-clang-x64": JobSpec(name="Windows (MSVC, clang-cl x64)",          os=JobOs.WindowsLatest,     platform=SdlPlatform.Msvc,        artifact="SDL-clang-cl-x64",       msvc_arch=MsvcArch.X64,   clang_cl=True, ),
    "msvc-clang-x86": JobSpec(name="Windows (MSVC, clang-cl x86)",          os=JobOs.WindowsLatest,     platform=SdlPlatform.Msvc,        artifact="SDL-clang-cl-x86",       msvc_arch=MsvcArch.X86,   clang_cl=True, ),
-    "msvc-arm32": JobSpec(name="Windows (MSVC, ARM)",                       os=JobOs.WindowsLatest,     platform=SdlPlatform.Msvc,        artifact="SDL-VC-arm32",           msvc_arch=MsvcArch.Arm32, ),
    "msvc-arm64": JobSpec(name="Windows (MSVC, ARM64)",                     os=JobOs.WindowsLatest,     platform=SdlPlatform.Msvc,        artifact="SDL-VC-arm64",           msvc_arch=MsvcArch.Arm64, ),
    "msvc-gdk-x64": JobSpec(name="GDK (MSVC, x64)",                         os=JobOs.WindowsLatest,     platform=SdlPlatform.Msvc,        artifact="SDL-VC-GDK",             msvc_arch=MsvcArch.X64,   msvc_project="VisualC-GDK/SDL.sln", gdk=True, no_cmake=True, ),
    "ubuntu-22.04": JobSpec(name="Ubuntu 22.04",                            os=JobOs.Ubuntu22_04,       platform=SdlPlatform.Linux,       artifact="SDL-ubuntu22.04", ),
@@ -116,7 +117,7 @@ JOB_SPECS = {
    "steamrt-sniper": JobSpec(name="Steam Linux Runtime (Sniper)",          os=JobOs.UbuntuLatest,      platform=SdlPlatform.Linux,       artifact="SDL-slrsniper",          container="registry.gitlab.steamos.cloud/steamrt/sniper/sdk:beta", ),
    "ubuntu-intel-icx": JobSpec(name="Ubuntu 22.04 (Intel oneAPI)",         os=JobOs.Ubuntu22_04,       platform=SdlPlatform.Linux,       artifact="SDL-ubuntu22.04-oneapi", intel=IntelCompiler.Icx, ),
    "ubuntu-intel-icc": JobSpec(name="Ubuntu 22.04 (Intel Compiler)",       os=JobOs.Ubuntu22_04,       platform=SdlPlatform.Linux,       artifact="SDL-ubuntu22.04-icc",    intel=IntelCompiler.Icc, ),
-    "macos-framework-x64":  JobSpec(name="MacOS (Framework) (x64)",         os=JobOs.Macos13,           platform=SdlPlatform.MacOS,       artifact="SDL-macos-framework",    apple_framework=True,  apple_archs={AppleArch.Aarch64, AppleArch.X86_64, }, xcode=True, ),
+    "macos-framework-x64":  JobSpec(name="MacOS (Framework) (x64)",         os=JobOs.Macos14,           platform=SdlPlatform.MacOS,       artifact="SDL-macos-framework",    apple_framework=True,  apple_archs={AppleArch.Aarch64, AppleArch.X86_64, }, xcode=True, ),
    "macos-framework-arm64": JobSpec(name="MacOS (Framework) (arm64)",      os=JobOs.MacosLatest,       platform=SdlPlatform.MacOS,       artifact=None,                     apple_framework=True,  apple_archs={AppleArch.Aarch64, AppleArch.X86_64, }, ),
    "macos-gnu-arm64": JobSpec(name="MacOS (GNU prefix)",                   os=JobOs.MacosLatest,       platform=SdlPlatform.MacOS,       artifact="SDL-macos-arm64-gnu",    apple_framework=False, apple_archs={AppleArch.Aarch64, },  ),
    "ios": JobSpec(name="iOS (CMake & xcode)",                              os=JobOs.MacosLatest,       platform=SdlPlatform.Ios,         artifact="SDL-ios-arm64",          xcode=True, ),
@@ -137,6 +138,7 @@ JOB_SPECS = {
    "vita-pvr": JobSpec(name="Sony PlayStation Vita (GLES w/ PVR_PSP2)",    os=JobOs.UbuntuLatest,      platform=SdlPlatform.Vita,        artifact="SDL-vita-pvr",           container="vitasdk/vitasdk:latest", vita_gles=VitaGLES.Pvr, ),
    "riscos": JobSpec(name="RISC OS",                                       os=JobOs.UbuntuLatest,      platform=SdlPlatform.Riscos,      artifact="SDL-riscos",             container="riscosdotinfo/riscos-gccsdk-4.7:latest", ),
    "netbsd": JobSpec(name="NetBSD",                                        os=JobOs.UbuntuLatest,      platform=SdlPlatform.NetBSD,      artifact="SDL-netbsd-x64", ),
+    "openbsd": JobSpec(name="OpenBSD",                                      os=JobOs.UbuntuLatest,      platform=SdlPlatform.OpenBSD,     artifact="SDL-openbsd-x64", ),
    "freebsd": JobSpec(name="FreeBSD",                                      os=JobOs.UbuntuLatest,      platform=SdlPlatform.FreeBSD,     artifact="SDL-freebsd-x64", ),
 }

@@ -174,6 +176,7 @@ class JobDetails:
    brew_packages: list[str] = dataclasses.field(default_factory=list)
    cmake_toolchain_file: str = ""
    cmake_arguments: list[str] = dataclasses.field(default_factory=list)
+    cmake_generator: str = "Ninja"
    cmake_build_arguments: list[str] = dataclasses.field(default_factory=list)
    clang_tidy: bool = True
    cppflags: list[str] = dataclasses.field(default_factory=list)
@@ -222,6 +225,7 @@ class JobDetails:
    check_sources: bool = False
    setup_python: bool = False
    pypi_packages: list[str] = dataclasses.field(default_factory=list)
+    binutils_strings: str = "strings"

    def to_workflow(self, enable_artifacts: bool) -> dict[str, str|bool]:
        data = {
@@ -255,6 +259,7 @@ class JobDetails:
            "cflags": my_shlex_join(self.cppflags + self.cflags),
            "cxxflags": my_shlex_join(self.cppflags + self.cxxflags),
            "ldflags": my_shlex_join(self.ldflags),
+            "cmake-generator": self.cmake_generator,
            "cmake-toolchain-file": self.cmake_toolchain_file,
            "clang-tidy": self.clang_tidy,
            "cmake-arguments": my_shlex_join(self.cmake_arguments),
@@ -289,6 +294,7 @@ class JobDetails:
            "check-sources": self.check_sources,
            "setup-python": self.setup_python,
            "pypi-packages": my_shlex_join(self.pypi_packages),
+            "binutils-strings": self.binutils_strings,
        }
        return {k: v for k, v in data.items() if v != ""}

@@ -403,10 +409,6 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
                    job.msvc_vcvars_arch = "x64_x86"
                case MsvcArch.X64:
                    job.msvc_vcvars_arch = "x64"
-                case MsvcArch.Arm32:
-                    job.msvc_vcvars_arch = "x64_arm"
-                    job.msvc_vcvars_sdk = "10.0.22621.0"  # 10.0.26100.0 dropped ARM32 um and ucrt libraries
-                    job.run_tests = False
                case MsvcArch.Arm64:
                    job.msvc_vcvars_arch = "x64_arm64"
                    job.run_tests = False
@@ -514,9 +516,10 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
            job.ccache = True
            job.apt_packages = []
            job.brew_packages.extend((
-                "ccache",
                "ninja",
            ))
+            if job.ccache:
+                job.brew_packages.append("ccache")
            if job.clang_tidy:
                job.brew_packages.append("llvm")
            if spec.xcode:
@@ -550,6 +553,10 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
                    "testmultiaudio-apk",
                    "testsprite-apk",
                ]
+
+                # -fPIC is required after updating NDK from 21 to 28
+                job.cflags.append("-fPIC")
+                job.cxxflags.append("-fPIC")
        case SdlPlatform.Emscripten:
            job.clang_tidy = False  # clang-tidy does not understand -gsource-map
            job.shared = False
@@ -675,13 +682,16 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
            job.shared_lib = SharedLibType.SO_0
            job.static_lib = StaticLibType.A
        case SdlPlatform.N3ds:
-            job.ccache = True
+            job.cmake_generator = "Unix Makefiles"
+            job.cmake_build_arguments.append("-j$(nproc)")
+            job.ccache = False
            job.shared = False
-            job.apt_packages = ["ccache", "ninja-build", "binutils"]
+            job.apt_packages = []
            job.clang_tidy = False
            job.run_tests = False
            job.cc_from_cmake = True
            job.cmake_toolchain_file = "${DEVKITPRO}/cmake/3DS.cmake"
+            job.binutils_strings = "/opt/devkitpro/devkitARM/bin/arm-none-eabi-strings"
            job.static_lib = StaticLibType.A
        case SdlPlatform.Msys2:
            job.ccache = True
@@ -713,7 +723,7 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
            ))
            job.cmake_toolchain_file = "/home/riscos/env/toolchain-riscos.cmake"
            job.static_lib = StaticLibType.A
-        case SdlPlatform.FreeBSD | SdlPlatform.NetBSD:
+        case SdlPlatform.FreeBSD | SdlPlatform.NetBSD | SdlPlatform.OpenBSD:
            job.cpactions = True
            job.no_cmake = True
            job.run_tests = False
@@ -723,7 +733,7 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
            match spec.platform:
                case SdlPlatform.FreeBSD:
                    job.cpactions_os = "freebsd"
-                    job.cpactions_version = "14.2"
+                    job.cpactions_version = "14.3"
                    job.cpactions_arch = "x86-64"
                    job.cpactions_setup_cmd = "sudo pkg update"
                    job.cpactions_install_cmd = "sudo pkg install -y cmake ninja pkgconf libXcursor libXext libXinerama libXi libXfixes libXrandr libXScrnSaver libXxf86vm wayland wayland-protocols libxkbcommon mesa-libs libglvnd evdev-proto libinotify alsa-lib jackit pipewire pulseaudio sndio dbus zh-fcitx ibus libudev-devd"
@@ -737,6 +747,12 @@ def spec_to_job(spec: JobSpec, key: str, trackmem_symbol_names: bool) -> JobDeta
                    job.cpactions_arch = "x86-64"
                    job.cpactions_setup_cmd = "export PATH=\"/usr/pkg/sbin:/usr/pkg/bin:/sbin:$PATH\"; export PKG_CONFIG_PATH=\"/usr/pkg/lib/pkgconfig\";export PKG_PATH=\"https://cdn.netBSD.org/pub/pkgsrc/packages/NetBSD/$(uname -p)/$(uname -r|cut -f \"1 2\" -d.)/All/\";echo \"PKG_PATH=$PKG_PATH\";echo \"uname -a -> \"$(uname -a)\"\";sudo -E sysctl -w security.pax.aslr.enabled=0;sudo -E sysctl -w security.pax.aslr.global=0;sudo -E pkgin clean;sudo -E pkgin update"
                    job.cpactions_install_cmd = "sudo -E pkgin -y install cmake dbus pkgconf ninja-build pulseaudio libxkbcommon wayland wayland-protocols libinotify libusb1"
+                case SdlPlatform.OpenBSD:
+                    job.cpactions_os = "openbsd"
+                    job.cpactions_version = "7.7"
+                    job.cpactions_arch = "x86-64"
+                    job.cpactions_setup_cmd = "sudo pkg_add -u"
+                    job.cpactions_install_cmd = "sudo pkg_add cmake ninja pkgconf wayland wayland-protocols libxkbcommon libinotify pulseaudio dbus ibus"
        case _:
            raise ValueError(f"Unsupported platform={spec.platform}")

@@ -75,8 +75,8 @@ jobs:
        if: ${{ matrix.platform.android-ndk }}
        id: setup-ndk
        with:
-          local-cache: true
-          ndk-version: r21e
+          local-cache: false
+          ndk-version: r28c
      - name: 'Configure Android NDK variables'
        if: ${{ matrix.platform.android-ndk }}
        shell: sh
@@ -201,7 +201,7 @@ jobs:
        #shell: ${{ matrix.platform.shell }}
        run: |
          ${{ matrix.platform.source-cmd }}
-          ${{ matrix.platform.cmake-config-emulator }} cmake -S . -B build -GNinja \
+          ${{ matrix.platform.cmake-config-emulator }} cmake -S . -B build -G "${{ matrix.platform.cmake-generator }}" \
            -Wdeprecated -Wdev -Werror \
            ${{ matrix.platform.cmake-toolchain-file != '' && format('-DCMAKE_TOOLCHAIN_FILE={0}', matrix.platform.cmake-toolchain-file) || '' }} \
            -DSDL_WERROR=${{ matrix.platform.werror }} \
@@ -232,9 +232,9 @@ jobs:
        run: |
          echo "This should show us the SDL_REVISION"
          echo "Shared library:"
-          ${{ (matrix.platform.shared-lib && format('strings build/{0} | grep "Github Workflow"', matrix.platform.shared-lib)) || 'echo "<Shared library not supported by platform>"' }}
+          ${{ (matrix.platform.shared-lib && format('{0} build/{1} | grep "Github Workflow"', matrix.platform.binutils-strings, matrix.platform.shared-lib)) || 'echo "<Shared library not supported by platform>"' }}
          echo "Static library:"
-          ${{ (matrix.platform.static-lib && format('strings build/{0} | grep "Github Workflow"', matrix.platform.static-lib)) || 'echo "<Static library not supported by platform>"' }}
+          ${{ (matrix.platform.static-lib && format('{0} build/{1} | grep "Github Workflow"', matrix.platform.binutils-strings, matrix.platform.static-lib)) || 'echo "<Static library not supported by platform>"' }}
      - name: 'Run build-time tests (CMake)'
        id: tests
        if: ${{ !matrix.platform.no-cmake && matrix.platform.run-tests }}
@@ -321,7 +321,7 @@ jobs:
      - name: 'Build (cross-platform-actions, BSD)'
        id: cpactions
        if: ${{ matrix.platform.cpactions }}
-        uses: cross-platform-actions/action@v0.27.0
+        uses: cross-platform-actions/action@v0.29.0
        with:
          operating_system: '${{ matrix.platform.cpactions-os }}'
          architecture: '${{ matrix.platform.cpactions-arch }}'
@@ -406,6 +406,14 @@ jobs:
          build-scripts/test-versioning.sh
          python build-scripts/check_android_jni.py
          python build-scripts/check_stdlib_usage.py
+      - name: 'Verify alignment of Android test apks'
+        if: ${{ matrix.platform.android-ndk && !matrix.platform.no-cmake }}
+        run: |
+          find ./ -iname '*.apk' | xargs -L1 ./build-scripts/check_elf_alignment.sh
+      - name: 'Verify alignment of Android .so files'
+        if: ${{ matrix.platform.android-ndk && !matrix.platform.no-cmake }}
+        run: |
+          find ./ -iname '*.so' | xargs -L1 ./build-scripts/check_elf_alignment.sh
      - name: 'Upload binary package'
        uses: actions/upload-artifact@v4
        if: ${{ always() && matrix.platform.artifact != '' && (steps.package.outcome == 'success' || steps.cpactions.outcome == 'success') && (matrix.platform.enable-artifacts || steps.tests.outcome == 'failure') }}
@@ -428,4 +436,4 @@ jobs:
        with:
          if-no-files-found: error
          name: '${{ matrix.platform.artifact }}-apks'
-          path: build/test/*.apk
+          path: build/test/*.apk
@@ -204,10 +204,22 @@ jobs:
              -DCMAKE_PREFIX_PATH="${{ steps.mount.outputs.mount_point }}"  \
              -DCMAKE_SYSTEM_NAME=iOS                                       \
              -DCMAKE_OSX_ARCHITECTURES="arm64"                             \
-              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                             \
+              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                            \
              -Werror=dev                                                   \
              -B build_ios
          cmake --build build_ios --config Release --verbose
+
+          cmake -S "${{ steps.src.outputs.path }}/cmake/test"                                          \
+              -DTEST_FULL=FALSE                                                                        \
+              -DTEST_STATIC=FALSE                                                                      \
+              -DTEST_TEST=FALSE                                                                        \
+              -DCMAKE_PREFIX_PATH="${{ steps.mount.outputs.mount_point }}/SDL3.xcframework/ios-arm64"  \
+              -DCMAKE_SYSTEM_NAME=iOS                                                                  \
+              -DCMAKE_OSX_ARCHITECTURES="arm64"                                                        \
+              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                                                       \
+              -Werror=dev                                                                              \
+              -B build_ios2
+          cmake --build build_ios2 --config Release --verbose
      - name: 'CMake (configure + build) tvOS'
        run: |
          cmake -S "${{ steps.src.outputs.path }}/cmake/test"               \
@@ -217,10 +229,22 @@ jobs:
              -DCMAKE_PREFIX_PATH="${{ steps.mount.outputs.mount_point }}"  \
              -DCMAKE_SYSTEM_NAME=tvOS                                      \
              -DCMAKE_OSX_ARCHITECTURES="arm64"                             \
-              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                             \
+              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                            \
              -Werror=dev                                                   \
              -B build_tvos
          cmake --build build_tvos --config Release --verbose
+
+          cmake -S "${{ steps.src.outputs.path }}/cmake/test"                                          \
+              -DTEST_FULL=FALSE                                                                        \
+              -DTEST_STATIC=FALSE                                                                      \
+              -DTEST_TEST=FALSE                                                                        \
+              -DCMAKE_PREFIX_PATH="${{ steps.mount.outputs.mount_point }}/SDL3.xcframework/tvos-arm64" \
+              -DCMAKE_SYSTEM_NAME=tvOS                                                                 \
+              -DCMAKE_OSX_ARCHITECTURES="arm64"                                                        \
+              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                                                       \
+              -Werror=dev                                                                              \
+              -B build_tvos2
+          cmake --build build_tvos2 --config Release --verbose
      - name: 'CMake (configure + build) iOS simulator'
        run: |
          sysroot=$(xcodebuild -version -sdk iphonesimulator Path)
@@ -237,6 +261,19 @@ jobs:
              -Werror=dev                                                   \
              -B build_ios_simulator
          cmake --build build_ios_simulator --config Release --verbose
+
+          cmake -S "${{ steps.src.outputs.path }}/cmake/test"                                                          \
+              -DTEST_FULL=FALSE                                                                                        \
+              -DTEST_STATIC=FALSE                                                                                      \
+              -DTEST_TEST=FALSE                                                                                        \
+              -DCMAKE_PREFIX_PATH="${{ steps.mount.outputs.mount_point }}/SDL3.xcframework/ios-arm64_x86_64-simulator" \
+              -DCMAKE_SYSTEM_NAME=iOS                                                                                  \
+              -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"                                                                 \
+              -DCMAKE_OSX_SYSROOT="${sysroot}"                                                                         \
+              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                                                                       \
+              -Werror=dev                                                                                              \
+              -B build_ios_simulator2
+          cmake --build build_ios_simulator2 --config Release --verbose
      - name: 'CMake (configure + build) tvOS simulator'
        run: |
          sysroot=$(xcodebuild -version -sdk appletvsimulator Path)
@@ -254,9 +291,22 @@ jobs:
              -B build_tvos_simulator
          cmake --build build_tvos_simulator --config Release --verbose

+          cmake -S "${{ steps.src.outputs.path }}/cmake/test"                                                           \
+              -DTEST_FULL=FALSE                                                                                         \
+              -DTEST_STATIC=FALSE                                                                                       \
+              -DTEST_TEST=FALSE                                                                                         \
+              -DCMAKE_PREFIX_PATH="${{ steps.mount.outputs.mount_point }}/SDL3.xcframework/tvos-arm64_x86_64-simulator" \
+              -DCMAKE_SYSTEM_NAME=tvOS                                                                                  \
+              -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"                                                                  \
+              -DCMAKE_OSX_SYSROOT="${sysroot}"                                                                          \
+              -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0                                                                        \
+              -Werror=dev                                                                                               \
+              -B build_tvos_simulator2
+          cmake --build build_tvos_simulator2 --config Release --verbose
+
  msvc:
    needs: [src]
-    runs-on: windows-2019
+    runs-on: windows-2025
    outputs:
      VC-x86: ${{ steps.releaser.outputs.VC-x86 }}
      VC-x64: ${{ steps.releaser.outputs.VC-x64 }}
@@ -532,10 +582,11 @@ jobs:
        with:
          sparse-checkout: 'build-scripts/build-release.py'
      - name: 'Setup Android NDK'
+        id: setup-ndk
        uses: nttld/setup-ndk@v1
        with:
-          local-cache: true
-          ndk-version: r21e
+          local-cache: false
+          ndk-version: r28c
      - name: 'Setup Java JDK'
        uses: actions/setup-java@v4
        with:
@@ -561,6 +612,8 @@ jobs:
        run: |
          python build-scripts/build-release.py \
            --actions android \
+            --android-api 21 \
+            --android-ndk-home "${{ steps.setup-ndk.outputs.ndk-path }}" \
            --commit ${{ inputs.commit }} \
            --root "${{ steps.tar.outputs.path }}" \
            --github \
@@ -607,8 +660,14 @@ jobs:
          python "${{ needs.src.outputs.project }}-${{ needs.src.outputs.version }}.aar" -o /tmp/SDL3-android
          echo "prefix=/tmp/SDL3-android" >>$GITHUB_OUTPUT
          echo "sdl3-aar=/tmp/${{ needs.src.outputs.project }}-${{ needs.src.outputs.version }}.aar" >>$GITHUB_OUTPUT
+      - name: 'Verify alignment of libSDL3.so (arm64-v8a/x86_64)'
+        run: |
+          set -e
+          ${{ steps.src.outputs.path }}/build-scripts/check_elf_alignment.sh ${{ steps.sdk.outputs.prefix }}/lib/arm64-v8a/libSDL3.so
+          ${{ steps.src.outputs.path }}/build-scripts/check_elf_alignment.sh ${{ steps.sdk.outputs.prefix }}/lib/x86_64/libSDL3.so
      - name: 'CMake (configure + build) x86, x64, arm32, arm64'
        run: |
+          set -e
          android_abis="x86 x86_64 armeabi-v7a arm64-v8a"
          for android_abi in ${android_abis}; do
            echo "Configuring ${android_abi}..."
@@ -5,7 +5,7 @@ if(NOT DEFINED CMAKE_BUILD_TYPE)
 endif()

 # See docs/release_checklist.md
-project(SDL3 LANGUAGES C VERSION "3.2.12")
+project(SDL3 LANGUAGES C VERSION "3.2.28")

 if(CMAKE_SOURCE_DIR STREQUAL PROJECT_SOURCE_DIR)
  set(SDL3_MAINPROJECT ON)
@@ -79,6 +79,12 @@ include("${SDL3_SOURCE_DIR}/cmake/PreseedEmscriptenCache.cmake")

 SDL_DetectCompiler()
 SDL_DetectTargetCPUArchitectures(SDL_CPUS)
+if(APPLE AND CMAKE_OSX_ARCHITECTURES)
+  list(LENGTH CMAKE_OSX_ARCHITECTURES _num_arches)
+  if(_num_arches GREATER 1)
+    set(APPLE_MULTIARCH TRUE)
+  endif()
+endif()

 # Increment this if there is an incompatible change - but if that happens,
 # we should rename the library from SDL3 to SDL4, at which point this would
@@ -183,26 +189,6 @@ if(MSVC)
  set(SDL_RELOCATABLE_DEFAULT ON)
 endif()

-if(MSVC)
-  if(NOT SDL_LIBC)
-    # Make sure /RTC1 is disabled, otherwise it will use functions from the CRT
-    foreach(flag_var
-        CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
-        CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO)
-      string(REGEX REPLACE "/RTC(su|[1su])" "" ${flag_var} "${${flag_var}}")
-    endforeach(flag_var)
-  endif()
-
-  if(MSVC_CLANG)
-    # clang-cl treats /W4 as '-Wall -Wextra' -- we don't need -Wextra
-    foreach(flag_var
-        CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
-        CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO)
-      string(REGEX REPLACE "/W4" "/W3" ${flag_var} "${${flag_var}}")
-    endforeach(flag_var)
-  endif()
-endif()
-
 set(SDL_SHARED_DEFAULT ON)
 set(SDL_STATIC_DEFAULT ON)

@@ -370,7 +356,7 @@ dep_option(SDL_WASAPI              "Use the Windows WASAPI audio driver" ON "WIN
 dep_option(SDL_RENDER_D3D          "Enable the Direct3D 9 render driver" ON "SDL_RENDER;SDL_DIRECTX" OFF)
 dep_option(SDL_RENDER_D3D11        "Enable the Direct3D 11 render driver" ON "SDL_RENDER;SDL_DIRECTX" OFF)
 dep_option(SDL_RENDER_D3D12        "Enable the Direct3D 12 render driver" ON "SDL_RENDER;SDL_DIRECTX" OFF)
-dep_option(SDL_RENDER_METAL        "Enable the Metal render driver" ON "SDL_RENDER;${APPLE}" OFF)
+dep_option(SDL_RENDER_METAL        "Enable the Metal render driver" ON "SDL_RENDER;APPLE" OFF)
 dep_option(SDL_RENDER_GPU          "Enable the SDL_GPU render driver" ON "SDL_RENDER;SDL_GPU" OFF)
 dep_option(SDL_VIVANTE             "Use Vivante EGL video driver" ON "${UNIX_SYS};SDL_CPU_ARM32" OFF)
 dep_option(SDL_VULKAN              "Enable Vulkan support" ON "SDL_VIDEO;ANDROID OR APPLE OR LINUX OR FREEBSD OR WINDOWS" OFF)
@@ -421,6 +407,29 @@ if(SDL_PRESEED)
  SDL_Preseed_CMakeCache()
 endif()

+if(MSVC)
+  if(NOT SDL_LIBC)
+    # Make sure /RTC1 is disabled, otherwise it will use functions from the CRT
+    foreach(flag_var
+        CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
+        CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO
+        CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
+        CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
+      string(REGEX REPLACE "/RTC(su|[1su])" "" ${flag_var} "${${flag_var}}")
+    endforeach(flag_var)
+    set(CMAKE_MSVC_RUNTIME_CHECKS "")
+  endif()
+
+  if(MSVC_CLANG)
+    # clang-cl treats /W4 as '-Wall -Wextra' -- we don't need -Wextra
+    foreach(flag_var
+        CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
+        CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO)
+      string(REGEX REPLACE "/W4" "/W3" ${flag_var} "${${flag_var}}")
+    endforeach(flag_var)
+  endif()
+endif()
+
 if(SDL_SHARED)
  add_library(SDL3-shared SHARED)
  add_library(SDL3::SDL3-shared ALIAS SDL3-shared)
@@ -634,6 +643,11 @@ if(MSVC)
    # Mark SDL3.dll as compatible with Control-flow Enforcement Technology (CET)
    sdl_shared_link_options("-CETCOMPAT")
  endif()
+
+  # for VS >= 17.14 targeting ARM64: inline the Interlocked funcs
+  if(MSVC_VERSION GREATER 1943 AND SDL_CPU_ARM64 AND NOT SDL_LIBC)
+    sdl_compile_options(PRIVATE "/forceInterlockedFunctions-")
+  endif()
 endif()

 if(CMAKE_C_COMPILER_ID STREQUAL "MSVC")
@@ -650,7 +664,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -mmmx")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <mmintrin.h>
      void ints_add(int *dest, int *a, int *b, unsigned size) {
        for (; size >= 2; size -= 2, dest += 2, a += 2, b += 2) {
@@ -660,7 +674,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char *argv[]) {
        ints_add((int*)0, (int*)0, (int*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_MMX)
+      }]==] COMPILER_SUPPORTS_MMX)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_MMX)
      set(HAVE_MMX TRUE)
@@ -671,7 +685,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -msse")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <xmmintrin.h>
      void floats_add(float *dest, float *a, float *b, unsigned size) {
        for (; size >= 4; size -= 4, dest += 4, a += 4, b += 4) {
@@ -681,7 +695,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        floats_add((float*)0, (float*)0, (float*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_SSE)
+      }]==] COMPILER_SUPPORTS_SSE)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_SSE)
      set(HAVE_SSE TRUE)
@@ -692,7 +706,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -msse2")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <emmintrin.h>
      void doubles_add(double *dest, double *a, double *b, unsigned size) {
        for (; size >= 4; size -= 4, dest += 4, a += 4, b += 4) {
@@ -702,7 +716,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        doubles_add((double*)0, (double*)0, (double*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_SSE2)
+      }]==] COMPILER_SUPPORTS_SSE2)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_SSE2)
      set(HAVE_SSE2 TRUE)
@@ -713,7 +727,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -msse3")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <pmmintrin.h>
      void ints_add(int *dest, int *a, int *b, unsigned size) {
        for (; size >= 4; size -= 4, dest += 4, a += 4, b += 4) {
@@ -723,7 +737,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        ints_add((int*)0, (int*)0, (int*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_SSE3)
+      }]==] COMPILER_SUPPORTS_SSE3)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_SSE3)
      set(HAVE_SSE3 TRUE)
@@ -734,7 +748,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -msse4.1")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <smmintrin.h>
      void ints_mul(int *dest, int *a, int *b, unsigned size) {
        for (; size >= 4; size -= 4, dest += 4, a += 4, b += 4) {
@@ -744,7 +758,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        ints_mul((int*)0, (int*)0, (int*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_SSE4_1)
+      }]==] COMPILER_SUPPORTS_SSE4_1)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_SSE4_1)
      set(HAVE_SSE4_1 TRUE)
@@ -753,21 +767,16 @@ if(SDL_ASSEMBLY)
  if(SDL_SSE4_2)
    cmake_push_check_state()
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
-      string(APPEND CMAKE_REQUIRED_FLAGS " -msse4.2 -mcrc32")
+      string(APPEND CMAKE_REQUIRED_FLAGS " -msse4.2")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <nmmintrin.h>
-      unsigned calc_crc32c(const char *text, unsigned len) {
-          unsigned crc32c = ~0;
-          for (; len >= 4; len -= 4, text += 4) {
-            crc32c = (unsigned)_mm_crc32_u32(crc32c, *(unsigned*)text);
-          }
-          return crc32c;
-      }
+      __m128i bitmask;
+      char data[16];
      int main(int argc, char **argv) {
-        calc_crc32c(\"SDL_SSE4\",8);
+        bitmask = _mm_cmpgt_epi64(_mm_set1_epi64x(0), _mm_loadu_si128((void*)data));
        return 0;
-      }" COMPILER_SUPPORTS_SSE4_2)
+      }]==] COMPILER_SUPPORTS_SSE4_2)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_SSE4_2)
      set(HAVE_SSE4_2 TRUE)
@@ -778,7 +787,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -mavx")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <immintrin.h>
      void floats_add(float *dest, float *a, float *b, unsigned size) {
        for (; size >= 8; size -= 8, dest += 8, a += 8, b += 8) {
@@ -788,7 +797,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        floats_add((float*)0, (float*)0, (float*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_AVX)
+      }]==] COMPILER_SUPPORTS_AVX)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_AVX)
      set(HAVE_AVX TRUE)
@@ -799,7 +808,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -mavx2")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <immintrin.h>
      void ints_add(int *dest, int *a, int *b, unsigned size) {
        for (; size >= 8; size -= 8, dest += 8, a += 8, b += 8) {
@@ -809,7 +818,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        ints_add((int*)0, (int*)0, (int*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_AVX2)
+      }]==] COMPILER_SUPPORTS_AVX2)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_AVX2)
      set(HAVE_AVX2 TRUE)
@@ -820,7 +829,7 @@ if(SDL_ASSEMBLY)
    if(USE_GCC OR USE_CLANG OR USE_INTELCC)
      string(APPEND CMAKE_REQUIRED_FLAGS " -mavx512f")
    endif()
-    check_c_source_compiles("
+    check_x86_source_compiles([==[
      #include <immintrin.h>
      void floats_add(float *dest, float *a, float *b, unsigned size) {
        for (; size >= 16; size -= 16, dest += 16, a += 16, b += 16) {
@@ -830,7 +839,7 @@ if(SDL_ASSEMBLY)
      int main(int argc, char **argv) {
        floats_add((float*)0, (float*)0, (float*)0, 0);
        return 0;
-      }" COMPILER_SUPPORTS_AVX512F)
+      }]==] COMPILER_SUPPORTS_AVX512F)
    cmake_pop_check_state()
    if(COMPILER_SUPPORTS_AVX512F)
      set(HAVE_AVX512F TRUE)
@@ -838,18 +847,17 @@ if(SDL_ASSEMBLY)
  endif()

  if(SDL_ARMNEON)
-    check_c_source_compiles("
-        #include <arm_neon.h>
-        void floats_add(float *dest, float *a, float *b, unsigned size) {
-          for (; size >= 4; size -= 4, dest += 4, a += 4, b += 4) {
-            vst1q_f32(dest, vaddq_f32(vld1q_f32(a), vld1q_f32(b)));
-          }
+    check_arm_source_compiles([==[
+      #include <arm_neon.h>
+      void floats_add(float *dest, float *a, float *b, unsigned size) {
+        for (; size >= 4; size -= 4, dest += 4, a += 4, b += 4) {
+          vst1q_f32(dest, vaddq_f32(vld1q_f32(a), vld1q_f32(b)));
        }
-        int main(int argc, char *argv[]) {
-          floats_add((float*)0, (float*)0, (float*)0, 0);
-          return 0;
-         }" COMPILER_SUPPORTS_ARMNEON)
-
+      }
+      int main(int argc, char *argv[]) {
+        floats_add((float*)0, (float*)0, (float*)0, 0);
+        return 0;
+      }]==] COMPILER_SUPPORTS_ARMNEON)
    if(COMPILER_SUPPORTS_ARMNEON)
      set(HAVE_ARMNEON TRUE)
    endif()
@@ -890,7 +898,17 @@ if(SDL_ASSEMBLY)
      cmake_pop_check_state()

      if(COMPILER_SUPPORTS_LSX AND HAVE_LSXINTRIN_H)
-        set_property(SOURCE "${SDL3_SOURCE_DIR}/src/video/yuv2rgb/yuv_rgb_lsx.c" APPEND PROPERTY COMPILE_OPTIONS "-mlsx")
+        set_property(SOURCE
+            "${SDL3_SOURCE_DIR}/src/video/yuv2rgb/yuv_rgb_lsx.c"
+            "${SDL3_SOURCE_DIR}/src/video/SDL_blit_A.c"
+            "${SDL3_SOURCE_DIR}/src/video/SDL_fillrect.c"
+            APPEND PROPERTY COMPILE_OPTIONS "-mlsx")
+
+        set_property(SOURCE
+            "${SDL3_SOURCE_DIR}/src/video/yuv2rgb/yuv_rgb_lsx.c"
+            "${SDL3_SOURCE_DIR}/src/video/SDL_blit_A.c"
+            "${SDL3_SOURCE_DIR}/src/video/SDL_fillrect.c"
+            PROPERTY SKIP_PRECOMPILE_HEADERS 1)
        set(HAVE_LSX TRUE)
      endif()
    endif()
@@ -1052,8 +1070,10 @@ if(SDL_LIBC)
  cmake_push_check_state()
  if(MSVC)
    string(APPEND CMAKE_REQUIRED_FLAGS " -we4244 -WX")  # 'conversion' conversion from 'type1' to 'type2', possible loss of data
-  else()
+  elseif(HAVE_GCC_WFLOAT_CONVERSION)
    string(APPEND CMAKE_REQUIRED_FLAGS " -Wfloat-conversion -Werror")
+  else()
+    string(APPEND CMAKE_REQUIRED_FLAGS " -Wconversion -Werror")
  endif()
  foreach(math_fn isinf isnan)
    string(TOUPPER "${math_fn}" MATH_FN)
@@ -1091,6 +1111,7 @@ if(SDL_LIBC)
    check_symbol_exists(gethostname "unistd.h" HAVE_GETHOSTNAME)
    check_symbol_exists(getpagesize "unistd.h" HAVE_GETPAGESIZE)
    check_symbol_exists(sigaction "signal.h" HAVE_SIGACTION)
+    check_symbol_exists(sigtimedwait "signal.h" HAVE_SIGTIMEDWAIT)
    check_symbol_exists(setjmp "setjmp.h" HAVE_SETJMP)
    check_symbol_exists(nanosleep "time.h" HAVE_NANOSLEEP)
    check_symbol_exists(gmtime_r "time.h" HAVE_GMTIME_R)
@@ -1336,9 +1357,7 @@ if(ANDROID)
    set(HAVE_SDL_HAPTIC TRUE)
  endif()

-  if(SDL_HIDAPI)
-    CheckHIDAPI()
-  endif()
+  CheckHIDAPI()

  if(SDL_JOYSTICK)
    set(SDL_JOYSTICK_ANDROID 1)
@@ -1480,6 +1499,9 @@ if(ANDROID)
      endif()
    endif()
  endif()
+  if(TARGET SDL3-static)
+    target_link_options(SDL3-static INTERFACE "-Wl,-u,JNI_OnLoad")
+  endif()

 elseif(EMSCRIPTEN)
  # Hide noisy warnings that intend to aid mostly during initial stages of porting a new
@@ -1905,7 +1927,7 @@ elseif(WINDOWS)
    target_compile_options(SDL_uclibc PRIVATE $<$<COMPILE_LANGUAGE:C,CXX>:/GS-> $<$<COMPILE_LANGUAGE:C,CXX>:/Gs1048576>)
    if(SDL_CPU_X86)
      target_compile_options(SDL3-shared PRIVATE "/arch:SSE")
-      target_compile_options(SDL3-SDL_uclibc PRIVATE "/arch:SSE")
+      target_compile_options(SDL_uclibc PRIVATE "/arch:SSE")
    endif()
  endif()

@@ -2954,7 +2976,7 @@ if(WINDOWS)
  sdl_glob_sources("${SDL3_SOURCE_DIR}/src/process/windows/*.c")
  set(SDL_PROCESS_WINDOWS 1)
  set(HAVE_SDL_PROCESS TRUE)
-else()
+elseif(NOT ANDROID)
  check_c_source_compiles("
 #include <spawn.h>
 #include <unistd.h>
@@ -3144,14 +3166,15 @@ endforeach()
 if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/REVISION.txt")
  file(STRINGS "${CMAKE_CURRENT_SOURCE_DIR}/REVISION.txt" revisions)
  list(GET revisions 0 revisions_0)
-  string(STRIP "${revisions_0}" SDL_REVISION)
+  string(STRIP "${revisions_0}" revisions_0_stripped)
+  set(SDL_REVISION "SDL-${revisions_0_stripped}")
 else()
  set(SDL_REVISION "" CACHE STRING "Custom SDL revision (only used when REVISION.txt does not exist)")
 endif()
 if(NOT SDL_REVISION)
  # If SDL_REVISION is not overrided, use git to describe
  git_describe(SDL_REVISION_GIT)
-  set(SDL_REVISION "SDL3-${SDL3_VERSION}-${SDL_REVISION_GIT}")
+  set(SDL_REVISION "SDL-${SDL3_VERSION}-${SDL_REVISION_GIT}")
 endif()

 execute_process(COMMAND "${CMAKE_COMMAND}" -E make_directory "${SDL3_BINARY_DIR}/include-revision/SDL3")
@@ -781,7 +781,6 @@
    <ClCompile Include="..\..\src\render\opengl\SDL_shaders_gl.c" />
    <ClCompile Include="..\..\src\render\opengles2\SDL_render_gles2.c" />
    <ClCompile Include="..\..\src\render\opengles2\SDL_shaders_gles2.c" />
-    <ClCompile Include="..\..\src\render\SDL_d3dmath.c" />
    <ClCompile Include="..\..\src\render\SDL_render.c" />
    <ClCompile Include="..\..\src\render\SDL_render_unsupported.c" />
    <ClCompile Include="..\..\src\render\SDL_yuv_sw.c" />
@@ -108,7 +108,6 @@
    <ClCompile Include="..\..\src\render\opengl\SDL_shaders_gl.c" />
    <ClCompile Include="..\..\src\render\opengles2\SDL_render_gles2.c" />
    <ClCompile Include="..\..\src\render\opengles2\SDL_shaders_gles2.c" />
-    <ClCompile Include="..\..\src\render\SDL_d3dmath.c" />
    <ClCompile Include="..\..\src\render\SDL_render.c" />
    <ClCompile Include="..\..\src\render\SDL_render_unsupported.c" />
    <ClCompile Include="..\..\src\render\SDL_yuv_sw.c" />
@@ -627,7 +627,6 @@
    <ClCompile Include="..\..\src\render\opengl\SDL_shaders_gl.c" />
    <ClCompile Include="..\..\src\render\opengles2\SDL_render_gles2.c" />
    <ClCompile Include="..\..\src\render\opengles2\SDL_shaders_gles2.c" />
-    <ClCompile Include="..\..\src\render\SDL_d3dmath.c" />
    <ClCompile Include="..\..\src\render\SDL_render.c" />
    <ClCompile Include="..\..\src\render\SDL_render_unsupported.c" />
    <ClCompile Include="..\..\src\render\SDL_yuv_sw.c" />
@@ -1460,7 +1460,6 @@
    <ClCompile Include="..\..\src\sensor\windows\SDL_windowssensor.c">
      <Filter>sensor\windows</Filter>
    </ClCompile>
-    <ClCompile Include="..\..\src\render\SDL_d3dmath.c">
      <Filter>render</Filter>
    </ClCompile>
    <ClCompile Include="..\..\src\render\SDL_render.c">
@@ -19,10 +19,10 @@
 	<key>CFBundlePackageType</key>
 	<string>FMWK</string>
 	<key>CFBundleShortVersionString</key>
-	<string>3.2.12</string>
+	<string>3.2.28</string>
 	<key>CFBundleSignature</key>
 	<string>SDLX</string>
 	<key>CFBundleVersion</key>
-	<string>3.2.12</string>
+	<string>3.2.28</string>
 </dict>
 </plist>
@@ -292,7 +292,6 @@
 		A7D8BA2523E2514400DCD162 /* SDL_drawpoint.c in Sources */ = {isa = PBXBuildFile; fileRef = A7D8A8FC23E2514000DCD162 /* SDL_drawpoint.c */; };
 		A7D8BA2B23E2514400DCD162 /* SDL_blendfillrect.c in Sources */ = {isa = PBXBuildFile; fileRef = A7D8A8FD23E2514000DCD162 /* SDL_blendfillrect.c */; };
 		A7D8BA3123E2514400DCD162 /* SDL_rotate.h in Headers */ = {isa = PBXBuildFile; fileRef = A7D8A8FE23E2514000DCD162 /* SDL_rotate.h */; };
-		A7D8BA3723E2514400DCD162 /* SDL_d3dmath.c in Sources */ = {isa = PBXBuildFile; fileRef = A7D8A8FF23E2514000DCD162 /* SDL_d3dmath.c */; };
 		A7D8BA4923E2514400DCD162 /* SDL_render_gles2.c in Sources */ = {isa = PBXBuildFile; fileRef = A7D8A90423E2514000DCD162 /* SDL_render_gles2.c */; };
 		A7D8BA4F23E2514400DCD162 /* SDL_shaders_gles2.h in Headers */ = {isa = PBXBuildFile; fileRef = A7D8A90523E2514000DCD162 /* SDL_shaders_gles2.h */; };
 		A7D8BA5523E2514400DCD162 /* SDL_gles2funcs.h in Headers */ = {isa = PBXBuildFile; fileRef = A7D8A90623E2514000DCD162 /* SDL_gles2funcs.h */; };
@@ -869,7 +868,6 @@
 		A7D8A8FC23E2514000DCD162 /* SDL_drawpoint.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = SDL_drawpoint.c; sourceTree = "<group>"; };
 		A7D8A8FD23E2514000DCD162 /* SDL_blendfillrect.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = SDL_blendfillrect.c; sourceTree = "<group>"; };
 		A7D8A8FE23E2514000DCD162 /* SDL_rotate.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = SDL_rotate.h; sourceTree = "<group>"; };
-		A7D8A8FF23E2514000DCD162 /* SDL_d3dmath.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = SDL_d3dmath.c; sourceTree = "<group>"; };
 		A7D8A90423E2514000DCD162 /* SDL_render_gles2.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = SDL_render_gles2.c; sourceTree = "<group>"; };
 		A7D8A90523E2514000DCD162 /* SDL_shaders_gles2.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = SDL_shaders_gles2.h; sourceTree = "<group>"; };
 		A7D8A90623E2514000DCD162 /* SDL_gles2funcs.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = SDL_gles2funcs.h; sourceTree = "<group>"; };
@@ -2135,7 +2133,6 @@
 				A7D8A90C23E2514000DCD162 /* opengl */,
 				A7D8A90323E2514000DCD162 /* opengles2 */,
 				A7D8A8EF23E2514000DCD162 /* software */,
-				A7D8A8FF23E2514000DCD162 /* SDL_d3dmath.c */,
 				A7D8A8DC23E2514000DCD162 /* SDL_d3dmath.h */,
 				A7D8A8DB23E2514000DCD162 /* SDL_render.c */,
 				E4F7981D2AD8D86A00669F54 /* SDL_render_unsupported.c */,
@@ -2833,7 +2830,7 @@
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 			shellPath = /bin/sh;
-			shellScript = "# Build an xcframework with both device and simulator files for all platforms.\n# Adapted from an answer in\n# https://developer.apple.com/forums/thread/666335?answerId=685927022#685927022\n\nif [ \"$XCODE_VERSION_ACTUAL\" -lt 1100 ]\nthen\n    echo \"error: Building an xcframework requires Xcode 11 minimum.\"\n    exit 1\nfi\n\nFRAMEWORK_NAME=\"SDL3\"\nPROJECT_NAME=\"SDL\"\nSCHEME=\"SDL3\"\n\nMACOS_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-macosx.xcarchive\"\nIOS_SIMULATOR_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-iphonesimulator.xcarchive\"\nIOS_DEVICE_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-iphoneos.xcarchive\"\nTVOS_SIMULATOR_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-appletvsimulator.xcarchive\"\nTVOS_DEVICE_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-appletvos.xcarchive\"\n\nOUTPUT_DIR=\"./build/\"\n\n# macOS\nxcodebuild archive \\\n    ONLY_ACTIVE_ARCH=NO \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${MACOS_ARCHIVE_PATH} \\\n    -destination 'generic/platform=macOS,name=Any Mac' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n    \n# iOS simulator\nxcodebuild archive \\\n    ONLY_ACTIVE_ARCH=NO \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${IOS_SIMULATOR_ARCHIVE_PATH} \\\n    -destination 'generic/platform=iOS Simulator' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# iOS device\nxcodebuild archive \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${IOS_DEVICE_ARCHIVE_PATH} \\\n    -destination 'generic/platform=iOS' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# tvOS simulator\nxcodebuild archive \\\n    ONLY_ACTIVE_ARCH=NO \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${TVOS_SIMULATOR_ARCHIVE_PATH} \\\n    -destination 'generic/platform=tvOS Simulator' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# tvOS device\nxcodebuild archive \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${TVOS_DEVICE_ARCHIVE_PATH} \\\n    -destination 'generic/platform=tvOS' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# Clean-up any existing instance of this xcframework from the Products directory\nrm -rf \"${OUTPUT_DIR}${FRAMEWORK_NAME}.xcframework\"\n\n# Create final xcframework\nxcodebuild -create-xcframework \\\n    -framework \"${MACOS_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -framework \"${IOS_DEVICE_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -framework \"${IOS_SIMULATOR_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -framework \"${TVOS_DEVICE_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -framework \"${TVOS_SIMULATOR_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -output ${OUTPUT_DIR}/${FRAMEWORK_NAME}.xcframework\n\n# Ensure git doesn't pick up on our Products folder. \nrm -rf ${OUTPUT_DIR}/.gitignore\necho \"*\" >> ${OUTPUT_DIR}/.gitignore\n";
+			shellScript = "# Build an xcframework with both device and simulator files for all platforms.\n# Adapted from an answer in\n# https://developer.apple.com/forums/thread/666335?answerId=685927022#685927022\n\nif [ \"$XCODE_VERSION_ACTUAL\" -lt 1100 ]\nthen\n    echo \"error: Building an xcframework requires Xcode 11 minimum.\"\n    exit 1\nfi\n\nFRAMEWORK_NAME=\"SDL3\"\nPROJECT_NAME=\"SDL\"\nSCHEME=\"SDL3\"\n\nMACOS_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-macosx.xcarchive\"\nIOS_SIMULATOR_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-iphonesimulator.xcarchive\"\nIOS_DEVICE_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-iphoneos.xcarchive\"\nTVOS_SIMULATOR_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-appletvsimulator.xcarchive\"\nTVOS_DEVICE_ARCHIVE_PATH=\"${BUILD_DIR}/${CONFIGURATION}/${FRAMEWORK_NAME}-appletvos.xcarchive\"\n\nOUTPUT_DIR=\"./build/\"\n\n# macOS\nxcodebuild archive \\\n    ONLY_ACTIVE_ARCH=NO \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${MACOS_ARCHIVE_PATH} \\\n    -destination 'generic/platform=macOS,name=Any Mac' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n    \n# iOS simulator\nxcodebuild archive \\\n    ONLY_ACTIVE_ARCH=NO \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${IOS_SIMULATOR_ARCHIVE_PATH} \\\n    -destination 'generic/platform=iOS Simulator' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# iOS device\nxcodebuild archive \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${IOS_DEVICE_ARCHIVE_PATH} \\\n    -destination 'generic/platform=iOS' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# tvOS simulator\nxcodebuild archive \\\n    ONLY_ACTIVE_ARCH=NO \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${TVOS_SIMULATOR_ARCHIVE_PATH} \\\n    -destination 'generic/platform=tvOS Simulator' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# tvOS device\nxcodebuild archive \\\n    -scheme \"${SCHEME}\" \\\n    -project \"${PROJECT_NAME}.xcodeproj\" \\\n    -archivePath ${TVOS_DEVICE_ARCHIVE_PATH} \\\n    -destination 'generic/platform=tvOS' \\\n    BUILD_LIBRARY_FOR_DISTRIBUTION=YES \\\n    SKIP_INSTALL=NO || exit $?\n\n# Clean-up any existing instance of this xcframework from the Products directory\nrm -rf \"${OUTPUT_DIR}${FRAMEWORK_NAME}.xcframework\"\n\n# Create final xcframework\nxcodebuild -create-xcframework \\\n    -framework \"${MACOS_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -debug-symbols \"${MACOS_ARCHIVE_PATH}\"/dSYMs/$FRAMEWORK_NAME.framework.dSYM \\\n    -framework \"${IOS_DEVICE_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -debug-symbols \"${IOS_DEVICE_ARCHIVE_PATH}\"/dSYMs/$FRAMEWORK_NAME.framework.dSYM \\\n    -framework \"${IOS_SIMULATOR_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -debug-symbols \"${IOS_SIMULATOR_ARCHIVE_PATH}\"/dSYMs/$FRAMEWORK_NAME.framework.dSYM \\\n    -framework \"${TVOS_DEVICE_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -debug-symbols \"${TVOS_DEVICE_ARCHIVE_PATH}\"/dSYMs/$FRAMEWORK_NAME.framework.dSYM \\\n    -framework \"${TVOS_SIMULATOR_ARCHIVE_PATH}\"/Products/Library/Frameworks/${FRAMEWORK_NAME}.framework \\\n    -debug-symbols \"${TVOS_SIMULATOR_ARCHIVE_PATH}\"/dSYMs/$FRAMEWORK_NAME.framework.dSYM \\\n    -output ${OUTPUT_DIR}/${FRAMEWORK_NAME}.xcframework\n\n# Ensure git doesn't pick up on our Products folder. \nrm -rf ${OUTPUT_DIR}/.gitignore\necho \"*\" >> ${OUTPUT_DIR}/.gitignore\n";
 		};
 /* End PBXShellScriptBuildPhase section */

@@ -2903,7 +2900,6 @@
 				A7D8BBE723E2574800DCD162 /* SDL_uikitviewcontroller.m in Sources */,
 				A7D8ADF223E2514100DCD162 /* SDL_blit_A.c in Sources */,
 				A7D8BBDD23E2574800DCD162 /* SDL_uikitmodes.m in Sources */,
-				A7D8BA3723E2514400DCD162 /* SDL_d3dmath.c in Sources */,
 				F3A9AE9C2C8A13C100AAC390 /* SDL_pipeline_gpu.c in Sources */,
 				75E0915A241EA924004729E1 /* SDL_virtualjoystick.c in Sources */,
 				F338A11A2D1B37E4007CDFDF /* SDL_tray.c in Sources */,
@@ -3086,7 +3082,7 @@
 				CLANG_ENABLE_OBJC_ARC = YES;
 				DEPLOYMENT_POSTPROCESSING = YES;
 				DYLIB_COMPATIBILITY_VERSION = 201.0.0;
-				DYLIB_CURRENT_VERSION = 201.12.0;
+				DYLIB_CURRENT_VERSION = 201.28.0;
 				DYLIB_INSTALL_NAME_BASE = "@rpath";
 				ENABLE_STRICT_OBJC_MSGSEND = YES;
 				GCC_ALTIVEC_EXTENSIONS = YES;
@@ -3121,7 +3117,7 @@
 					"@loader_path/Frameworks",
 				);
 				MACOSX_DEPLOYMENT_TARGET = 10.13;
-				MARKETING_VERSION = 3.2.12;
+				MARKETING_VERSION = 3.2.28;
 				OTHER_LDFLAGS = "$(CONFIG_FRAMEWORK_LDFLAGS)";
 				PRODUCT_BUNDLE_IDENTIFIER = org.libsdl.SDL3;
 				PRODUCT_NAME = SDL3;
@@ -3138,6 +3134,8 @@
 			baseConfigurationReference = F3F7BE3B2CBD79D200C984AF /* config.xcconfig */;
 			buildSettings = {
 				CLANG_LINK_OBJC_RUNTIME = NO;
+				DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
+				GCC_GENERATE_DEBUGGING_SYMBOLS = YES;
 				OTHER_LDFLAGS = "-liconv";
 				SUPPORTS_MACCATALYST = YES;
 			};
@@ -3150,7 +3148,7 @@
 				CLANG_ENABLE_MODULES = YES;
 				CLANG_ENABLE_OBJC_ARC = YES;
 				DYLIB_COMPATIBILITY_VERSION = 201.0.0;
-				DYLIB_CURRENT_VERSION = 201.12.0;
+				DYLIB_CURRENT_VERSION = 201.28.0;
 				DYLIB_INSTALL_NAME_BASE = "@rpath";
 				ENABLE_STRICT_OBJC_MSGSEND = YES;
 				ENABLE_TESTABILITY = YES;
@@ -3182,7 +3180,7 @@
 					"@loader_path/Frameworks",
 				);
 				MACOSX_DEPLOYMENT_TARGET = 10.13;
-				MARKETING_VERSION = 3.2.12;
+				MARKETING_VERSION = 3.2.28;
 				ONLY_ACTIVE_ARCH = YES;
 				OTHER_LDFLAGS = "$(CONFIG_FRAMEWORK_LDFLAGS)";
 				PRODUCT_BUNDLE_IDENTIFIER = org.libsdl.SDL3;
@@ -3200,6 +3198,8 @@
 			baseConfigurationReference = F3F7BE3B2CBD79D200C984AF /* config.xcconfig */;
 			buildSettings = {
 				CLANG_LINK_OBJC_RUNTIME = NO;
+				DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
+				GCC_GENERATE_DEBUGGING_SYMBOLS = YES;
 				OTHER_LDFLAGS = "-liconv";
 				SUPPORTS_MACCATALYST = YES;
 			};
@@ -1,4 +1,4 @@
-Title SDL 3.2.12
+Title SDL 3.2.28
 Version 1
 Description SDL Library for macOS (http://www.libsdl.org)
 DefaultLocation /Library/Frameworks
@@ -1,8 +1,9 @@
 # SDL3 CMake configuration file:
-# This file is meant to be placed in Resources/CMake of a SDL3 framework
+# This file is meant to be placed in Resources/CMake of a SDL3 framework for macOS,
+# or in the CMake directory of a SDL3 framework for iOS / tvOS / visionOS.

 # INTERFACE_LINK_OPTIONS needs CMake 3.12
-cmake_minimum_required(VERSION 3.12)
+cmake_minimum_required(VERSION 3.12...4.0)

 include(FeatureSummary)
 set_package_properties(SDL3 PROPERTIES
@@ -31,16 +32,31 @@ endmacro()

 set(SDL3_FOUND TRUE)

-# Compute the installation prefix relative to this file.
-set(_sdl3_framework_path "${CMAKE_CURRENT_LIST_DIR}")                               # > /SDL3.framework/Resources/CMake/
-get_filename_component(_sdl3_framework_path "${_sdl3_framework_path}" REALPATH)     # > /SDL3.framework/Versions/Current/Resources/CMake
-get_filename_component(_sdl3_framework_path "${_sdl3_framework_path}" REALPATH)     # > /SDL3.framework/Versions/A/Resources/CMake/
-get_filename_component(_sdl3_framework_path "${_sdl3_framework_path}" PATH)         # > /SDL3.framework/Versions/A/Resources/
-get_filename_component(_sdl3_framework_path "${_sdl3_framework_path}" PATH)         # > /SDL3.framework/Versions/A/
-get_filename_component(_sdl3_framework_path "${_sdl3_framework_path}" PATH)         # > /SDL3.framework/Versions/
-get_filename_component(_sdl3_framework_path "${_sdl3_framework_path}" PATH)         # > /SDL3.framework/
-get_filename_component(_sdl3_framework_parent_path "${_sdl3_framework_path}" PATH)  # > /
+# Compute the installation prefix relative to this file:
+# search upwards for the .framework directory
+set(_current_path "${CMAKE_CURRENT_LIST_DIR}")
+get_filename_component(_current_path "${_current_path}" REALPATH)
+set(_sdl3_framework_path "")

+while(NOT _sdl3_framework_path)
+    if(IS_DIRECTORY "${_current_path}" AND "${_current_path}" MATCHES "/SDL3\\.framework$")
+        set(_sdl3_framework_path "${_current_path}")
+        break()
+    endif()
+    get_filename_component(_next_current_path "${_current_path}" DIRECTORY)
+    if("${_current_path}" STREQUAL "${_next_current_path}")
+        break()
+    endif()
+    set(_current_path "${_next_current_path}")
+endwhile()
+unset(_current_path)
+unset(_next_current_path)
+
+if(NOT _sdl3_framework_path)
+    message(FATAL_ERROR "Could not find SDL3.framework root from ${CMAKE_CURRENT_LIST_DIR}")
+endif()
+
+get_filename_component(_sdl3_framework_parent_path "${_sdl3_framework_path}" PATH)

 # All targets are created, even when some might not be requested though COMPONENTS.
 # This is done for compatibility with CMake generated SDL3-target.cmake files.
@@ -1,16 +1,26 @@
 # based on the files generated by CMake's write_basic_package_version_file

 # SDL CMake version configuration file:
-# This file is meant to be placed in Resources/CMake of a SDL3 framework
+# This file is meant to be placed in Resources/CMake of a SDL3 framework for macOS,
+# or in the CMake directory of a SDL3 framework for iOS / tvOS / visionOS.

-cmake_minimum_required(VERSION 3.12)
+cmake_minimum_required(VERSION 3.12...4.0)

-if(NOT EXISTS "${CMAKE_CURRENT_LIST_DIR}/../../Headers/SDL_version.h")
-    message(AUTHOR_WARNING "Could not find SDL_version.h. This script is meant to be placed in the Resources/CMake directory of SDL2.framework")
+# Find SDL_version.h
+set(_sdl_version_h_path "")
+if(EXISTS "${CMAKE_CURRENT_LIST_DIR}/../../Headers/SDL_version.h")
+    set(_sdl_version_h_path "${CMAKE_CURRENT_LIST_DIR}/../../Headers/SDL_version.h")
+elseif(EXISTS "${CMAKE_CURRENT_LIST_DIR}/../Headers/SDL_version.h")
+    set(_sdl_version_h_path "${CMAKE_CURRENT_LIST_DIR}/../Headers/SDL_version.h")
+endif()
+
+if(NOT _sdl_version_h_path)
+    message(AUTHOR_WARNING "Could not find SDL_version.h. This script is meant to be placed in the Resources/CMake directory or the CMake directory of SDL3.framework.")
+    set(PACKAGE_VERSION_UNSUITABLE TRUE)
    return()
 endif()

-file(READ "${CMAKE_CURRENT_LIST_DIR}/../../Headers/SDL_version.h" _sdl_version_h)
+file(READ "${_sdl_version_h_path}" _sdl_version_h)
 string(REGEX MATCH "#define[ \t]+SDL_MAJOR_VERSION[ \t]+([0-9]+)" _sdl_major_re "${_sdl_version_h}")
 set(_sdl_major "${CMAKE_MATCH_1}")
 string(REGEX MATCH "#define[ \t]+SDL_MINOR_VERSION[ \t]+([0-9]+)" _sdl_minor_re "${_sdl_version_h}")
@@ -21,9 +31,12 @@ if(_sdl_major_re AND _sdl_minor_re AND _sdl_micro_re)
    set(PACKAGE_VERSION "${_sdl_major}.${_sdl_minor}.${_sdl_micro}")
 else()
    message(AUTHOR_WARNING "Could not extract version from SDL_version.h.")
+    set(PACKAGE_VERSION_UNSUITABLE TRUE)
    return()
 endif()

+unset(_sdl_version_h)
+unset(_sdl_version_h_path)
 unset(_sdl_major_re)
 unset(_sdl_major)
 unset(_sdl_minor_re)
@@ -92,7 +92,7 @@ if(NOT TARGET SDL3::Headers)
    add_library(SDL3::Headers INTERFACE IMPORTED)
    set_target_properties(SDL3::Headers
        PROPERTIES
-            INTERFACE_COMPILE_OPTIONS "SHELL:-F \"${_sdl3_framework_parent_path}\""
+            INTERFACE_COMPILE_OPTIONS "-F${_sdl3_framework_parent_path}"
    )
 endif()
 set(SDL3_Headers_FOUND TRUE)
@@ -16,6 +16,7 @@ set(_sdl3_version_h "${_sdl3_framework}/Headers/SDL_version.h")

 if(NOT EXISTS "${_sdl3_version_h}")
    message(AUTHOR_WARNING "Cannot not find ${_sdl3_framework}. This script is meant to be placed in share/cmake/SDL3, next to SDL3.xcframework")
+    set(PACKAGE_VERSION_UNSUITABLE TRUE)
    return()
 endif()

@@ -36,6 +37,7 @@ if(_sdl_major_re AND _sdl_minor_re AND _sdl_micro_re)
    set(PACKAGE_VERSION "${_sdl_major}.${_sdl_minor}.${_sdl_micro}")
 else()
    message(AUTHOR_WARNING "Could not extract version from SDL_version.h.")
+    set(PACKAGE_VERSION_UNSUITABLE TRUE)
    return()
 endif()

@@ -7,4 +7,4 @@
 APP_ABI := armeabi-v7a arm64-v8a x86 x86_64

 # Min runtime API level
-APP_PLATFORM=android-16
+APP_PLATFORM=android-21
@@ -51,6 +51,8 @@
    boolean supportsRelativeMouse();
    int openFileDescriptor(java.lang.String, java.lang.String);
    boolean showFileDialog(java.lang.String[], boolean, boolean, int);
+    java.lang.String getPreferredLocales();
+    java.lang.String formatLocale(java.util.Locale);
 }

 -keep,includedescriptorclasses,allowoptimization class org.libsdl.app.HIDDeviceManager {
@@ -23,6 +23,7 @@ import android.net.Uri;
 import android.os.Build;
 import android.os.Bundle;
 import android.os.Handler;
+import android.os.LocaleList;
 import android.os.Message;
 import android.os.ParcelFileDescriptor;
 import android.util.DisplayMetrics;
@@ -60,7 +61,7 @@ public class SDLActivity extends Activity implements View.OnSystemUiVisibilityCh
    private static final String TAG = "SDL";
    private static final int SDL_MAJOR_VERSION = 3;
    private static final int SDL_MINOR_VERSION = 2;
-    private static final int SDL_MICRO_VERSION = 12;
+    private static final int SDL_MICRO_VERSION = 28;
 /*
    // Display InputType.SOURCE/CLASS of events and devices
    //
@@ -1127,6 +1128,11 @@ public class SDLActivity extends Activity implements View.OnSystemUiVisibilityCh
        int orientation_landscape = -1;
        int orientation_portrait = -1;

+        if (w <= 1 || h <= 1) {
+            // Invalid width/height, ignore this request
+            return;
+        }
+
        /* If set, hint "explicitly controls which UI orientations are allowed". */
        if (hint.contains("LandscapeRight") && hint.contains("LandscapeLeft")) {
            orientation_landscape = ActivityInfo.SCREEN_ORIENTATION_USER_LANDSCAPE;
@@ -2116,6 +2122,44 @@ public class SDLActivity extends Activity implements View.OnSystemUiVisibilityCh
        int requestCode;
        boolean multipleChoice;
    }
+    
+    /**
+     * This method is called by SDL using JNI.
+     */
+    public static String getPreferredLocales() {
+        String result = "";
+        if (Build.VERSION.SDK_INT >= 24 /* Android 7 (N) */) {
+            LocaleList locales = LocaleList.getAdjustedDefault();
+            for (int i = 0; i < locales.size(); i++) {
+                if (i != 0) result += ",";
+                result += formatLocale(locales.get(i));
+            }
+        } else if (mCurrentLocale != null) {
+            result = formatLocale(mCurrentLocale);
+        }
+        return result;
+    }
+
+    public static String formatLocale(Locale locale) {
+        String result = "";
+        String lang = "";
+        if (locale.getLanguage() == "in") {
+            // Indonesian is "id" according to ISO 639.2, but on Android is "in" because of Java backwards compatibility
+            lang = "id";
+        } else if (locale.getLanguage() == "") {
+            // Make sure language is never empty
+            lang = "und";
+        } else {
+            lang = locale.getLanguage();
+        }
+
+        if (locale.getCountry() == "") {
+            result = lang;
+        } else {
+            result = lang + "_" + locale.getCountry();
+        }
+        return result;
+    }
 }

 /**
@@ -2157,7 +2201,11 @@ class SDLClipboardHandler implements
    }

    public boolean clipboardHasText() {
-       return mClipMgr.hasPrimaryClip();
+        if (Build.VERSION.SDK_INT >= 28 /* Android 9 (P) */) {
+            return mClipMgr.hasPrimaryClip();
+        } else {
+            return mClipMgr.hasText();
+        }
    }

    public String clipboardGetText() {
@@ -2175,10 +2223,19 @@ class SDLClipboardHandler implements
    }

    public void clipboardSetText(String string) {
-       mClipMgr.removePrimaryClipChangedListener(this);
-       ClipData clip = ClipData.newPlainText(null, string);
-       mClipMgr.setPrimaryClip(clip);
-       mClipMgr.addPrimaryClipChangedListener(this);
+        mClipMgr.removePrimaryClipChangedListener(this);
+        if (string.isEmpty()) {
+            if (Build.VERSION.SDK_INT >= 28 /* Android 9 (P) */) {
+                mClipMgr.clearPrimaryClip();
+            } else {
+                ClipData clip = ClipData.newPlainText(null, "");
+                mClipMgr.setPrimaryClip(clip);
+            }
+        } else {
+            ClipData clip = ClipData.newPlainText(null, string);
+            mClipMgr.setPrimaryClip(clip);
+        }
+        mClipMgr.addPrimaryClipChangedListener(this);
    }

    @Override
@@ -15,6 +15,7 @@ import android.view.Display;
 import android.view.InputDevice;
 import android.view.KeyEvent;
 import android.view.MotionEvent;
+import android.view.PointerIcon;
 import android.view.Surface;
 import android.view.SurfaceHolder;
 import android.view.SurfaceView;
@@ -362,6 +363,16 @@ public class SDLSurface extends SurfaceView implements SurfaceHolder.Callback,
        }
    }

+    // Prevent android internal NullPointerException (https://github.com/libsdl-org/SDL/issues/13306)
+    @Override
+    public PointerIcon onResolvePointerIcon(MotionEvent event, int pointerIndex) {
+        try {
+            return super.onResolvePointerIcon(event, pointerIndex);
+        } catch (NullPointerException e) {
+            return null;
+        }
+    }
+
    // Captured pointer events for API 26.
    public boolean onCapturedPointerEvent(MotionEvent event)
    {
@@ -542,6 +542,7 @@ class AndroidApiVersion:
    def __repr__(self) -> str:
        return f"<{self.name} ({'.'.join(str(v) for v in self.ints)})>"

+ANDROID_ABI_EXTRA_LINK_OPTIONS = {}

 class Releaser:
    def __init__(self, release_info: dict, commit: str, revision: str, root: Path, dist_path: Path, section_printer: SectionPrinter, executer: Executer, cmake_generator: str, deps_path: Path, overwrite: bool, github: bool, fast: bool):
@@ -1013,6 +1014,7 @@ class Releaser:
        android_devel_file_tree = ArchiveFileTree()

        for android_abi in android_abis:
+            extra_link_options = ANDROID_ABI_EXTRA_LINK_OPTIONS.get(android_abi, "")
            with self.section_printer.group(f"Building for Android {android_api} {android_abi}"):
                build_dir = self.root / "build-android" / f"{android_abi}-build"
                install_dir = self.root / "install-android" / f"{android_abi}-install"
@@ -1023,8 +1025,12 @@ class Releaser:
                    "cmake",
                    "-S", str(self.root),
                    "-B", str(build_dir),
-                    f'''-DCMAKE_C_FLAGS="-ffile-prefix-map={self.root}=/src/{self.project}"''',
-                    f'''-DCMAKE_CXX_FLAGS="-ffile-prefix-map={self.root}=/src/{self.project}"''',
+                    # NDK 21e does not support -ffile-prefix-map
+                    # f'''-DCMAKE_C_FLAGS="-ffile-prefix-map={self.root}=/src/{self.project}"''',
+                    # f'''-DCMAKE_CXX_FLAGS="-ffile-prefix-map={self.root}=/src/{self.project}"''',
+                    f"-DANDROID_USE_LEGACY_TOOLCHAIN=0",
+                    f"-DCMAKE_EXE_LINKER_FLAGS={extra_link_options}",
+                    f"-DCMAKE_SHARED_LINKER_FLAGS={extra_link_options}",
                    f"-DCMAKE_TOOLCHAIN_FILE={cmake_toolchain_file}",
                    f"-DCMAKE_PREFIX_PATH={str(android_deps_path)}",
                    f"-DCMAKE_FIND_ROOT_PATH_MODE_PACKAGE=BOTH",
@@ -1121,8 +1127,7 @@ class Releaser:
        for dep, depinfo in self.release_info.get("dependencies", {}).items():
            startswith = depinfo["startswith"]
            dep_repo = depinfo["repo"]
-            # FIXME: dropped "--exclude-pre-releases"
-            dep_string_data = self.executer.check_output(["gh", "-R", dep_repo, "release", "list", "--exclude-drafts", "--json", "name,createdAt,tagName", "--jq", f'[.[]|select(.name|startswith("{startswith}"))]|max_by(.createdAt)']).strip()
+            dep_string_data = self.executer.check_output(["gh", "-R", dep_repo, "release", "list", "--exclude-drafts", "--exclude-pre-releases", "--json", "name,createdAt,tagName", "--jq", f'[.[]|select(.name|startswith("{startswith}"))]|max_by(.createdAt)']).strip()
            dep_data = json.loads(dep_string_data)
            dep_tag = dep_data["tagName"]
            dep_version = dep_data["name"]
@@ -1512,7 +1517,7 @@ def main(argv=None) -> int:
        if args.android_home is None or not Path(args.android_home).is_dir():
            parser.error("Invalid $ANDROID_HOME or --android-home: must be a directory containing the Android SDK")
        if args.android_ndk_home is None or not Path(args.android_ndk_home).is_dir():
-            parser.error("Invalid $ANDROID_NDK_HOME or --android_ndk_home: must be a directory containing the Android NDK")
+            parser.error("Invalid $ANDROID_NDK_HOME or --android-ndk-home: must be a directory containing the Android NDK")
        if args.android_api is None:
            with section_printer.group("Detect Android APIS"):
                args.android_api = releaser._detect_android_api(android_home=args.android_home)
@@ -1530,7 +1535,7 @@ def main(argv=None) -> int:
            parser.error("Invalid --android-api, and/or could not be detected")
        android_api_path = Path(args.android_home) / f"platforms/{args.android_api.name}"
        if not android_api_path.is_dir():
-            parser.error(f"Android API directory does not exist ({android_api_path})")
+            logger.warning(f"Android API directory does not exist ({android_api_path})")
        with section_printer.group("Android arguments"):
            print(f"android_home     = {args.android_home}")
            print(f"android_ndk_home = {args.android_ndk_home}")
@@ -0,0 +1,127 @@
+#!/bin/bash
+progname="${0##*/}"
+progname="${progname%.sh}"
+
+# usage: check_elf_alignment.sh [path to *.so files|path to *.apk]
+
+cleanup_trap() {
+  if [ -n "${tmp}" -a -d "${tmp}" ]; then
+    rm -rf ${tmp}
+  fi
+  exit $1
+}
+
+usage() {
+  echo "Host side script to check the ELF alignment of shared libraries."
+  echo "Shared libraries are reported ALIGNED when their ELF regions are"
+  echo "16 KB or 64 KB aligned. Otherwise they are reported as UNALIGNED."
+  echo
+  echo "Usage: ${progname} [input-path|input-APK|input-APEX]"
+}
+
+if [ ${#} -ne 1 ]; then
+  usage
+  exit
+fi
+
+case ${1} in
+  --help | -h | -\?)
+    usage
+    exit
+    ;;
+
+  *)
+    dir="${1}"
+    ;;
+esac
+
+if ! [ -f "${dir}" -o -d "${dir}" ]; then
+  echo "Invalid file: ${dir}" >&2
+  exit 1
+fi
+
+if [[ "${dir}" == *.apk ]]; then
+  trap 'cleanup_trap' EXIT
+
+  echo
+  echo "Recursively analyzing $dir"
+  echo
+
+  if { zipalign --help 2>&1 | grep -q "\-P <pagesize_kb>"; }; then
+    echo "=== APK zip-alignment ==="
+    zipalign -v -c -P 16 4 "${dir}" | egrep 'lib/arm64-v8a|lib/x86_64|Verification'
+    echo "========================="
+  else
+    echo "NOTICE: Zip alignment check requires build-tools version 35.0.0-rc3 or higher."
+    echo "  You can install the latest build-tools by running the below command"
+    echo "  and updating your \$PATH:"
+    echo
+    echo "    sdkmanager \"build-tools;35.0.0-rc3\""
+  fi
+
+  dir_filename=$(basename "${dir}")
+  tmp=$(mktemp -d -t "${dir_filename%.apk}_out_XXXXX")
+  unzip "${dir}" lib/* -d "${tmp}" >/dev/null 2>&1
+  dir="${tmp}"
+fi
+
+if [[ "${dir}" == *.apex ]]; then
+  trap 'cleanup_trap' EXIT
+
+  echo
+  echo "Recursively analyzing $dir"
+  echo
+
+  dir_filename=$(basename "${dir}")
+  tmp=$(mktemp -d -t "${dir_filename%.apex}_out_XXXXX")
+  deapexer extract "${dir}" "${tmp}" || { echo "Failed to deapex." && exit 1; }
+  dir="${tmp}"
+fi
+
+RED="\e[31m"
+GREEN="\e[32m"
+ENDCOLOR="\e[0m"
+
+unaligned_libs=()
+unaligned_critical_libs=()
+
+echo
+echo "=== ELF alignment ==="
+
+matches="$(find "${dir}" -type f)"
+IFS=$'\n'
+for match in $matches; do
+  # We could recursively call this script or rewrite it to though.
+  [[ "${match}" == *".apk" ]] && echo "WARNING: doesn't recursively inspect .apk file: ${match}"
+  [[ "${match}" == *".apex" ]] && echo "WARNING: doesn't recursively inspect .apex file: ${match}"
+
+  [[ $(file "${match}") == *"ELF"* ]] || continue
+
+  res="$(objdump -p "${match}" | grep LOAD | awk '{ print $NF }' | head -1)"
+  if [[ $res =~ 2\*\*(1[4-9]|[2-9][0-9]|[1-9][0-9]{2,}) ]]; then
+    echo -e "${match}: ${GREEN}ALIGNED${ENDCOLOR} ($res)"
+  else
+    unaligned_libs+=("${match}")
+    # Check if this is a critical architecture (arm64-v8a or x86_64)
+    if [[ "${match}" == *"arm64-v8a"* ]] || [[ "${match}" == *"x86_64"* ]]; then
+      unaligned_critical_libs+=("${match}")
+      echo -e "${match}: ${RED}UNALIGNED${ENDCOLOR} ($res)"
+    else
+      echo -e "${match}: UNALIGNED ($res)"
+    fi
+  fi
+done
+
+if [ ${#unaligned_libs[@]} -gt 0 ]; then
+  echo -e "Found ${#unaligned_libs[@]} unaligned libs (only arm64-v8a/x86_64 libs need to be aligned).${ENDCOLOR}"
+fi
+echo "====================="
+
+# Exit with appropriate code: 1 if critical unaligned libs found, 0 otherwise
+if [ ${#unaligned_critical_libs[@]} -gt 0 ]; then
+  echo -e "${RED}Found ${#unaligned_critical_libs[@]} critical unaligned libs.${ENDCOLOR}"
+  exit 1
+else
+  echo -e "${GREEN}ELF Verification Successful${ENDCOLOR}"
+  exit 0
+fi
@@ -183,7 +183,7 @@
    ],
    "api-minimum": 21,
    "api-target": 35,
-    "ndk-minimum": 21,
+    "ndk-minimum": 28,
    "aar-files": {
      "": [
        "android-project/app/proguard-rules.pro:proguard.txt",
@@ -1,21 +0,0 @@
-#!/bin/sh
-#
-# libtool assumes that the compiler can handle the -fPIC flag
-# This isn't always true (for example, nasm can't handle it)
-command=""
-while [ $# -gt 0 ]; do
-    case "$1" in
-        -?PIC)
-            # Ignore -fPIC and -DPIC options
-            ;;
-        -fno-common)
-            # Ignore -fPIC and -DPIC options
-            ;;
-        *)
-            command="$command $1"
-            ;;
-    esac
-    shift
-done
-echo $command
-exec $command
@@ -424,7 +424,11 @@ sub dewikify_chunk {
            $str .= "\n```$codelang\n$code\n```\n";
        }
    } elsif ($dewikify_mode eq 'manpage') {
-        $str =~ s/\./\\[char46]/gms;  # make sure these can't become control codes.
+        # make sure these can't become part of roff syntax.
+        $str =~ s/\./\\[char46]/gms;
+        $str =~ s/"/\\(dq/gms;
+        $str =~ s/'/\\(aq/gms;
+
        if ($wikitype eq 'mediawiki') {
            # Dump obvious wikilinks.
            if (defined $apiprefixregex) {
@@ -449,33 +453,52 @@ sub dewikify_chunk {
            # bullets
            $str =~ s/^\* /\n\\\(bu /gm;
        } elsif ($wikitype eq 'md') {
+            # bullets
+            $str =~ s/^\- /\n\\(bu /gm;
+            # merge paragraphs
+            $str =~ s/^[ \t]+//gm;
+            $str =~ s/([^\-\n])\n([^\-\n])/$1 $2/g;
+            $str =~ s/\n\n/\n.PP\n/g;
+
            # Dump obvious wikilinks.
            if (defined $apiprefixregex) {
-                $str =~ s/\[(\`?$apiprefixregex[a-zA-Z0-9_]+\`?)\]\($apiprefixregex[a-zA-Z0-9_]+\)/\n.BR $1\n/gms;
+                my $apr = $apiprefixregex;
+                if(!($apr =~ /\A\(.*\)\Z/s)) {
+                    # we're relying on the apiprefixregex having a capturing group.
+                    $apr = "(" . $apr . ")";
+                }
+                $str =~ s/(\S*?)\[\`?($apr[a-zA-Z0-9_]+)\`?\]\($apr[a-zA-Z0-9_]+\)(\S*)\s*/\n.BR "" "$1" "$2" "$5"\n/gm;
+                # handle cases like "[x](x), [y](y), [z](z)" being separated.
+                while($str =~ s/(\.BR[^\n]*)\n\n\.BR/$1\n.BR/gm) {}
            }

            # links
            $str =~ s/\[(.*?)]\((https?\:\/\/.*?)\)/\n.URL "$2" "$1"\n/g;

            # <code></code> is also popular.  :/
-            $str =~ s/\s*\`(.*?)\`\s*/\n.BR $1\n/gms;
+            $str =~ s/\s*(\S*?)\`([^\n]*?)\`(\S*)\s*/\n.BR "" "$1" "$2" "$3"\n/gms;

            # bold+italic (this looks bad, just make it bold).
-            $str =~ s/\s*\*\*\*(.*?)\*\*\*\s*/\n.B $1\n/gms;
+            $str =~ s/\s*(\S*?)\*\*\*([^\n]*?)\*\*\*(\S*)\s*/\n.BR "" "$1" "$2" "$3"\n/gms;

            # bold
-            $str =~ s/\s*\*\*(.*?)\*\*\s*/\n.B $1\n/gms;
+            $str =~ s/\s*(\S*?)\*\*([^\n]*?)\*\*(\S*)\s*/\n.BR "" "$1" "$2" "$3"\n/gms;

            # italic
-            $str =~ s/\s*\*(.*?)\*\s*/\n.I $1\n/gms;
-
-            # bullets
-            $str =~ s/^\- /\n\\\(bu /gm;
+            $str =~ s/\s*(\S*?)\*([^\n]*?)\*(\S*)\s*/\n.IR "" "$1" "$2" "$3"\n/gms;
        }

+        # cleanup unnecessary quotes
+        $str =~ s/(\.[IB]R?)(.*?) ""\n/$1$2\n/gm;
+        $str =~ s/(\.[IB]R?) "" ""(.*?)\n/$1$2\n/gm;
+        $str =~ s/"(\S+)"/$1/gm;
+        # cleanup unnecessary whitespace
+        $str =~ s/ +\n/\n/gm;
+
        if (defined $code) {
            $code =~ s/\A\n+//gms;
            $code =~ s/\n+\Z//gms;
+            $code =~ s/\\/\\(rs/gms;
            if ($dewikify_manpage_code_indent) {
                $str .= "\n.IP\n"
            } else {
@@ -580,7 +603,7 @@ sub dewikify {
            $retval .= dewikify_chunk($wikitype, $1, $2, $3);
        }
    } elsif ($wikitype eq 'md') {
-        while ($str =~ s/\A(.*?)\n```(.*?)\n(.*?)\n```\n//ms) {
+        while ($str =~ s/\A(.*?)\n?```(.*?)\n(.*?)\n```\n//ms) {
            $retval .= dewikify_chunk($wikitype, $1, $2, $3);
        }
    }
@@ -2765,7 +2788,6 @@ __EOF__
        my $wikitype = $wikitypes{$sym};
        my $sectionsref = $wikisyms{$sym};
        my $remarks = $sectionsref->{'Remarks'};
-        my $params = $sectionsref->{'Function Parameters'};
        my $returns = $sectionsref->{'Return Value'};
        my $version = $sectionsref->{'Version'};
        my $threadsafety = $sectionsref->{'Thread Safety'};
@@ -2773,6 +2795,23 @@ __EOF__
        my $examples = $sectionsref->{'Code Examples'};
        my $deprecated = $sectionsref->{'Deprecated'};
        my $headerfile = $manpageheaderfiletext;
+
+        my $params = undef;
+
+        if ($symtype == -1) {  # category documentation block.
+            # nothing to be done here.
+        } elsif (($symtype == 1) || (($symtype == 5))) {  # we'll assume a typedef (5) with a \param is a function pointer typedef.
+            $params = $sectionsref->{'Function Parameters'};
+        } elsif ($symtype == 2) {
+            $params = $sectionsref->{'Macro Parameters'};
+        } elsif ($symtype == 3) {
+            $params = $sectionsref->{'Fields'};
+        } elsif ($symtype == 4) {
+            $params = $sectionsref->{'Values'};
+        } else {
+            die("Unexpected symtype $symtype");
+        }
+
        $headerfile =~ s/\%fname\%/$headersymslocation{$sym}/g;
        $headerfile .= "\n";

@@ -2839,18 +2878,22 @@ __EOF__
            $str .= dewikify($wikitype, $deprecated) . "\n";
        }

+        my $incfile = $mainincludefname;
        if (defined $headerfile) {
-            $str .= ".SH HEADER FILE\n";
-            $str .= dewikify($wikitype, $headerfile) . "\n";
+            if($headerfile =~ /Defined in (.*)/) {
+                $incfile = $1;
+            }
        }

        $str .= ".SH SYNOPSIS\n";
        $str .= ".nf\n";
-        $str .= ".B #include \\(dq$mainincludefname\\(dq\n";
+        $str .= ".B #include <$incfile>\n";
        $str .= ".PP\n";

        my @decllines = split /\n/, $decl;
        foreach (@decllines) {
+            $_ =~ s/\\/\\(rs/g;  # fix multiline macro defs
+            $_ =~ s/"/\\(dq/g;
            $str .= ".BI \"$_\n";
        }
        $str .= ".fi\n";
@@ -2938,8 +2981,11 @@ __EOF__
        }

        if (defined $returns) {
+            # Chop datatype in parentheses off the front.
+            if(!($returns =~ s/\A\([^\[]*\[[^\]]*\]\([^\)]*\)[^\)]*\) //ms)) {
+                $returns =~ s/\A\([^\)]*\) //ms;
+            }
            $returns = dewikify($wikitype, $returns);
-            $returns =~ s/\A\(.*?\)\s*//;  # Chop datatype in parentheses off the front.
            $str .= ".SH RETURN VALUE\n";
            $str .= "$returns\n";
        }
@@ -2975,6 +3021,8 @@ __EOF__
                s/\A\/*//;
                s/\A\.BR\s+//;  # dewikify added this, but we want to handle it.
                s/\A\.I\s+//;  # dewikify added this, but we want to handle it.
+                s/\A\.PP\s*//;  # dewikify added this, but we want to handle it.
+                s/\\\(bu//;  # dewikify added this, but we want to handle it.
                s/\A\s*[\:\*\-]\s*//;
                s/\A\s+//;
                s/\s+\Z//;
@@ -1,16 +1,21 @@
 # Helper for Find modules

 function(get_flags_from_pkg_config _library _pc_prefix _out_prefix)
-  if("${_library}" MATCHES "${CMAKE_STATIC_LIBRARY_SUFFIX}$")
-    set(_cflags ${_pc_prefix}_STATIC_CFLAGS_OTHER)
-    set(_link_libraries ${_pc_prefix}_STATIC_LIBRARIES)
-    set(_link_options ${_pc_prefix}_STATIC_LDFLAGS_OTHER)
-    set(_library_dirs ${_pc_prefix}_STATIC_LIBRARY_DIRS)
+  if(MINGW)
+    set(re_shared_suffix ".dll.a$")
  else()
+    set(re_shared_suffix "${CMAKE_SHARED_LIBRARY_SUFFIX}$")
+  endif()
+  if("${_library}" MATCHES "${re_shared_suffix}")
    set(_cflags ${_pc_prefix}_CFLAGS_OTHER)
    set(_link_libraries ${_pc_prefix}_LIBRARIES)
    set(_link_options ${_pc_prefix}_LDFLAGS_OTHER)
    set(_library_dirs ${_pc_prefix}_LIBRARY_DIRS)
+  else()
+    set(_cflags ${_pc_prefix}_STATIC_CFLAGS_OTHER)
+    set(_link_libraries ${_pc_prefix}_STATIC_LIBRARIES)
+    set(_link_options ${_pc_prefix}_STATIC_LDFLAGS_OTHER)
+    set(_library_dirs ${_pc_prefix}_STATIC_LIBRARY_DIRS)
  endif()

  # The *_LIBRARIES lists always start with the library itself
@@ -1077,6 +1077,14 @@ endmacro()

 # Check for HIDAPI support
 macro(CheckHIDAPI)
+  if(ANDROID)
+    enable_language(CXX)
+    sdl_sources("${SDL3_SOURCE_DIR}/src/hidapi/android/hid.cpp")
+  endif()
+  if(IOS OR TVOS)
+    sdl_sources("${SDL3_SOURCE_DIR}/src/hidapi/ios/hid.m")
+    set(SDL_FRAMEWORK_COREBLUETOOTH 1)
+  endif()
  if(SDL_HIDAPI)
    set(HAVE_HIDAPI ON)
    if(SDL_HIDAPI_LIBUSB)
@@ -1109,14 +1117,6 @@ macro(CheckHIDAPI)
    endif()

    if(HAVE_HIDAPI)
-      if(ANDROID)
-        enable_language(CXX)
-        sdl_sources("${SDL3_SOURCE_DIR}/src/hidapi/android/hid.cpp")
-      endif()
-      if(IOS OR TVOS)
-        sdl_sources("${SDL3_SOURCE_DIR}/src/hidapi/ios/hid.m")
-        set(SDL_FRAMEWORK_COREBLUETOOTH 1)
-      endif()
      set(HAVE_SDL_HIDAPI TRUE)

      if(SDL_JOYSTICK AND SDL_HIDAPI_JOYSTICK)
@@ -160,3 +160,63 @@ function(SDL_AddCommonCompilerFlags TARGET)
    endif()
  endif()
 endfunction()
+
+function(check_x86_source_compiles BODY VAR)
+  if(ARGN)
+    message(FATAL_ERROR "Unknown arguments: ${ARGN}")
+  endif()
+  if(APPLE_MULTIARCH AND (SDL_CPU_X86 OR SDL_CPU_X64))
+    set(test_conditional 1)
+  else()
+    set(test_conditional 0)
+  endif()
+  check_c_source_compiles("
+    #if ${test_conditional}
+    # if defined(__i386__) || defined(__x86_64__)
+    #  define test_enabled 1
+    # else
+    #  define test_enabled 0 /* feign success in Apple multi-arch configs */
+    # endif
+    #else                    /* test normally */
+    # define test_enabled 1
+    #endif
+    #if test_enabled
+    ${BODY}
+    #else
+    int main(int argc, char *argv[]) {
+      (void)argc;
+      (void)argv;
+      return 0;
+    }
+    #endif" ${VAR})
+endfunction()
+
+function(check_arm_source_compiles BODY VAR)
+  if(ARGN)
+    message(FATAL_ERROR "Unknown arguments: ${ARGN}")
+  endif()
+  if(APPLE_MULTIARCH AND (SDL_CPU_ARM32 OR SDL_CPU_ARM64))
+    set(test_conditional 1)
+  else()
+    set(test_conditional 0)
+  endif()
+  check_c_source_compiles("
+    #if ${test_conditional}
+    # if defined(__arm__) || defined(__aarch64__)
+    #  define test_enabled 1
+    # else
+    #  define test_enabled 0 /* feign success in Apple multi-arch configs */
+    # endif
+    #else                    /* test normally */
+    # define test_enabled 1
+    #endif
+    #if test_enabled
+    ${BODY}
+    #else
+    int main(int argc, char *argv[]) {
+      (void)argc;
+      (void)argv;
+      return 0;
+    }
+    #endif" ${VAR})
+endfunction()
@@ -4,15 +4,15 @@ function(SDL_DetectTargetCPUArchitectures DETECTED_ARCHS)

  if(APPLE AND CMAKE_OSX_ARCHITECTURES)
    foreach(known_arch IN LISTS known_archs)
-      set(SDL_CPU_${known_arch} "0")
+      set(SDL_CPU_${known_arch} "0" PARENT_SCOPE)
    endforeach()
    set(detected_archs)
    foreach(osx_arch IN LISTS CMAKE_OSX_ARCHITECTURES)
      if(osx_arch STREQUAL "x86_64")
-        set(SDL_CPU_X64 "1")
+        set(SDL_CPU_X64 "1" PARENT_SCOPE)
        list(APPEND detected_archs "X64")
      elseif(osx_arch STREQUAL "arm64")
-        set(SDL_CPU_ARM64 "1")
+        set(SDL_CPU_ARM64 "1" PARENT_SCOPE)
        list(APPEND detected_archs "ARM64")
      endif()
    endforeach()
@@ -131,7 +131,7 @@ function(sdl_compile_options)
    target_compile_options(SDL3-static ${visibility} ${escaped_opts})
  endif()
  if(NOT ARGS_NO_EXPORT AND (ARGS_PUBLIC OR ARGS_INTERFACE))
-    set_property(TARGET SDL3-collector APPEND PROPERTY INTERFACE_COMPILE_COMPILE_OPTIONS "${ARGS_UNPARSED_ARGUMENTS}")
+    set_property(TARGET SDL3-collector APPEND PROPERTY INTERFACE_COMPILE_OPTIONS "${ARGS_UNPARSED_ARGUMENTS}")
  endif()
 endfunction()

@@ -49,7 +49,7 @@ NSApplicationDelegate implementation:
 ```objc
 - (NSApplicationTerminateReply)applicationShouldTerminate:(NSApplication *)sender
 {
-    if (SDL_GetEventState(SDL_EVENT_QUIT) == SDL_ENABLE) {
+    if (SDL_EventEnabled(SDL_EVENT_QUIT)) {
        SDL_Event event;
        SDL_zero(event);
        event.type = SDL_EVENT_QUIT;
@@ -61,7 +61,7 @@ NSApplicationDelegate implementation:

 - (BOOL)application:(NSApplication *)theApplication openFile:(NSString *)filename
 {
-    if (SDL_GetEventState(SDL_EVENT_DROP_FILE) == SDL_ENABLE) {
+    if (SDL_EventEnabled(SDL_EVENT_DROP_FILE)) {
        SDL_Event event;
        SDL_zero(event);
        event.type = SDL_EVENT_DROP_FILE;
@@ -11,7 +11,7 @@
 - [macOS](README-macos.md)
 - [NetBSD](README-bsd.md)
 - [Nintendo Switch](README-switch.md)
- [Nintendo 3DS](README-3ds.md)
+- [Nintendo 3DS](README-n3ds.md)
 - [OpenBSD](README-bsd.md)
 - [PlayStation 2](README-ps2.md)
 - [PlayStation 4](README-ps4.md)
@@ -20,7 +20,7 @@
 */

 /**
- * Main include header for the SDL library, version 3.2.12
+ * Main include header for the SDL library, version 3.2.28
 *
 * It is almost always best to include just this one header instead of
 * picking out individual headers included here. There are exceptions to
@@ -126,7 +126,7 @@ extern "C" {
 */
 #define SDL_TriggerBreakpoint() TriggerABreakpointInAPlatformSpecificManner

-#elif defined(_MSC_VER) && _MSC_VER >= 1310
+#elif defined(__MINGW32__) || (defined(_MSC_VER) && _MSC_VER >= 1310)
    /* Don't include intrin.h here because it contains C++ code */
    extern void __cdecl __debugbreak(void);
    #define SDL_TriggerBreakpoint() __debugbreak()
@@ -362,7 +362,7 @@ extern SDL_DECLSPEC SDL_AssertState SDLCALL SDL_ReportAssertion(SDL_AssertData *
 #define SDL_enabled_assert(condition) \
    do { \
        while ( !(condition) ) { \
-            static struct SDL_AssertData sdl_assert_data = { 0, 0, #condition, 0, 0, 0, 0 }; \
+            static struct SDL_AssertData sdl_assert_data = { false, 0, #condition, NULL, 0, NULL, NULL }; \
            const SDL_AssertState sdl_assert_state = SDL_ReportAssertion(&sdl_assert_data, SDL_FUNCTION, SDL_FILE, SDL_LINE); \
            if (sdl_assert_state == SDL_ASSERTION_RETRY) { \
                continue; /* go again. */ \
@@ -942,7 +942,10 @@ extern SDL_DECLSPEC void SDLCALL SDL_CloseAudioDevice(SDL_AudioDeviceID devid);
 * Binding a stream to a device will set its output format for playback
 * devices, and its input format for recording devices, so they match the
 * device's settings. The caller is welcome to change the other end of the
- * stream's format at any time with SDL_SetAudioStreamFormat().
+ * stream's format at any time with SDL_SetAudioStreamFormat(). If the other
+ * end of the stream's format has never been set (the audio stream was created
+ * with a NULL audio spec), this function will set it to match the device
+ * end's format.
 *
 * \param devid an audio device to bind a stream to.
 * \param streams an array of audio streams to bind.
@@ -1021,7 +1024,7 @@ extern SDL_DECLSPEC void SDLCALL SDL_UnbindAudioStream(SDL_AudioStream *stream);
 /**
 * Query an audio stream for its currently-bound device.
 *
- * This reports the audio device that an audio stream is currently bound to.
+ * This reports the logical audio device that an audio stream is currently bound to.
 *
 * If not bound, or invalid, this returns zero, which is not a valid device
 * ID.
@@ -119,7 +119,7 @@ typedef struct SDL_CameraSpec
    int width;                  /**< Frame width */
    int height;                 /**< Frame height */
    int framerate_numerator;     /**< Frame rate numerator ((num / denom) == FPS, (denom / num) == duration in seconds) */
-    int framerate_denominator;   /**< Frame rate demoninator ((num / denom) == FPS, (denom / num) == duration in seconds) */
+    int framerate_denominator;   /**< Frame rate denominator ((num / denom) == FPS, (denom / num) == duration in seconds) */
 } SDL_CameraSpec;

 /**
@@ -106,7 +106,7 @@ extern SDL_DECLSPEC bool SDLCALL SDL_SetClipboardText(const char *text);
 /**
 * Get UTF-8 text from the clipboard.
 *
- * This functions returns an empty string if there was not enough memory left
+ * This function returns an empty string if there is not enough memory left
 * for a copy of the clipboard's content.
 *
 * \returns the clipboard text on success or an empty string on failure; call
@@ -155,7 +155,7 @@ extern SDL_DECLSPEC bool SDLCALL SDL_SetPrimarySelectionText(const char *text);
 /**
 * Get UTF-8 text from the primary selection.
 *
- * This functions returns an empty string if there was not enough memory left
+ * This function returns an empty string if there is not enough memory left
 * for a copy of the primary selection's content.
 *
 * \returns the primary selection text on success or an empty string on
@@ -194,15 +194,14 @@ extern SDL_DECLSPEC bool SDLCALL SDL_HasPrimarySelectionText(void);
 * clipboard is cleared or new data is set. The clipboard is automatically
 * cleared in SDL_Quit().
 *
- * \param userdata a pointer to provided user data.
+ * \param userdata a pointer to the provided user data.
 * \param mime_type the requested mime-type.
 * \param size a pointer filled in with the length of the returned data.
 * \returns a pointer to the data for the provided mime-type. Returning NULL
- *          or setting length to 0 will cause no data to be sent to the
- *          "receiver". It is up to the receiver to handle this. Essentially
- *          returning no data is more or less undefined behavior and may cause
- *          breakage in receiving applications. The returned data will not be
- *          freed so it needs to be retained and dealt with internally.
+ *          or setting the length to 0 will cause zero length data to be sent
+ *          to the "receiver", which should be able to handle this. The
+ *          returned data will not be freed, so it needs to be retained and
+ *          dealt with internally.
 *
 * \since This function is available since SDL 3.2.0.
 *
@@ -211,10 +210,10 @@ extern SDL_DECLSPEC bool SDLCALL SDL_HasPrimarySelectionText(void);
 typedef const void *(SDLCALL *SDL_ClipboardDataCallback)(void *userdata, const char *mime_type, size_t *size);

 /**
- * Callback function that will be called when the clipboard is cleared, or new
+ * Callback function that will be called when the clipboard is cleared, or when new
 * data is set.
 *
- * \param userdata a pointer to provided user data.
+ * \param userdata a pointer to the provided user data.
 *
 * \since This function is available since SDL 3.2.0.
 *
@@ -231,7 +230,7 @@ typedef void (SDLCALL *SDL_ClipboardCleanupCallback)(void *userdata);
 * respond with the data for the requested mime-type.
 *
 * The size of text data does not include any terminator, and the text does
- * not need to be null terminated (e.g. you can directly copy a portion of a
+ * not need to be null-terminated (e.g., you can directly copy a portion of a
 * document).
 *
 * \param callback a function pointer to the function that provides the
@@ -239,7 +238,7 @@ typedef void (SDLCALL *SDL_ClipboardCleanupCallback)(void *userdata);
 * \param cleanup a function pointer to the function that cleans up the
 *                clipboard data.
 * \param userdata an opaque pointer that will be forwarded to the callbacks.
- * \param mime_types a list of mime-types that are being offered.
+ * \param mime_types a list of mime-types that are being offered. SDL copies the given list.
 * \param num_mime_types the number of mime-types in the mime_types list.
 * \returns true on success or false on failure; call SDL_GetError() for more
 *          information.
@@ -269,10 +268,10 @@ extern SDL_DECLSPEC bool SDLCALL SDL_SetClipboardData(SDL_ClipboardDataCallback
 extern SDL_DECLSPEC bool SDLCALL SDL_ClearClipboardData(void);

 /**
- * Get the data from clipboard for a given mime type.
+ * Get the data from the clipboard for a given mime type.
 *
 * The size of text data does not include the terminator, but the text is
- * guaranteed to be null terminated.
+ * guaranteed to be null-terminated.
 *
 * \param mime_type the mime type to read from the clipboard.
 * \param size a pointer filled in with the length of the returned data.
@@ -292,8 +291,8 @@ extern SDL_DECLSPEC void * SDLCALL SDL_GetClipboardData(const char *mime_type, s
 /**
 * Query whether there is data in the clipboard for the provided mime type.
 *
- * \param mime_type the mime type to check for data for.
- * \returns true if there exists data in clipboard for the provided mime type,
+ * \param mime_type the mime type to check for data.
+ * \returns true if data exists in the clipboard for the provided mime type,
 *          false if it does not.
 *
 * \threadsafety This function should only be called on the main thread.
@@ -310,7 +309,7 @@ extern SDL_DECLSPEC bool SDLCALL SDL_HasClipboardData(const char *mime_type);
 *
 * \param num_mime_types a pointer filled with the number of mime types, may
 *                       be NULL.
- * \returns a null terminated array of strings with mime types, or NULL on
+ * \returns a null-terminated array of strings with mime types, or NULL on
 *          failure; call SDL_GetError() for more information. This should be
 *          freed with SDL_free() when it is no longer needed.
 *
@@ -46,7 +46,7 @@
 #if defined(_MSC_VER) && (_MSC_VER >= 1400)
 /* As of Clang 11, '_m_prefetchw' is conflicting with the winnt.h's version,
   so we define the needed '_m_prefetch' here as a pseudo-header, until the issue is fixed. */
-#ifdef __clang__
+#if defined(__clang__) &&  !SDL_HAS_BUILTIN(_m_prefetch)
 #ifndef __PRFCHWINTRIN_H
 #define __PRFCHWINTRIN_H
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
@@ -128,7 +128,7 @@ _m_prefetch(void *__P)
 * \sa SDL_BIG_ENDIAN
 */
 #define SDL_BYTEORDER   SDL_LIL_ENDIAN___or_maybe___SDL_BIG_ENDIAN
-#elif defined(SDL_PLATFORM_LINUX)
+#elif defined(SDL_PLATFORM_LINUX) || defined(__GLIBC__)
 #include <endian.h>
 #define SDL_BYTEORDER  __BYTE_ORDER
 #elif defined(SDL_PLATFORM_SOLARIS)
@@ -486,7 +486,7 @@ SDL_FORCE_INLINE Uint32 SDL_Swap32(Uint32 x) { return x_but_byteswapped; }
 *
 * \since This function is available since SDL 3.2.0.
 */
-SDL_FORCE_INLINE Uint32 SDL_Swap64(Uint64 x) { return x_but_byteswapped; }
+SDL_FORCE_INLINE Uint64 SDL_Swap64(Uint64 x) { return x_but_byteswapped; }

 /**
 * Swap a 16-bit value from littleendian to native byte order.
@@ -1043,7 +1043,7 @@ typedef union SDL_Event
 } SDL_Event;

 /* Make sure we haven't broken binary compatibility */
-SDL_COMPILE_TIME_ASSERT(SDL_Event, sizeof(SDL_Event) == sizeof(((SDL_Event *)NULL)->padding));
+SDL_COMPILE_TIME_ASSERT(SDL_Event, sizeof(SDL_Event) == sizeof((SDL_static_cast(SDL_Event *, NULL))->padding));


 /* Function prototypes */
@@ -224,8 +224,8 @@
 * - `drawIndirectFirstInstance`
 *
 * **D3D12:** Supported on Windows 10 or newer, Xbox One (GDK), and Xbox
- * Series X|S (GDK). Requires a GPU that supports DirectX 12 Feature Level
- * 11_1.
+ * Series X|S (GDK). Requires a GPU that supports DirectX 12 Feature Level 11_0 and
+ * Resource Binding Tier 2 or above.
 *
 * **Metal:** Supported on macOS 10.14+ and iOS/tvOS 13.0+. Hardware
 * requirements vary by operating system:
@@ -1091,7 +1091,7 @@ typedef enum SDL_GPUCompareOp
    SDL_GPU_COMPAREOP_LESS_OR_EQUAL,     /**< The comparison evaluates reference <= test. */
    SDL_GPU_COMPAREOP_GREATER,           /**< The comparison evaluates reference > test. */
    SDL_GPU_COMPAREOP_NOT_EQUAL,         /**< The comparison evaluates reference != test. */
-    SDL_GPU_COMPAREOP_GREATER_OR_EQUAL,  /**< The comparison evalutes reference >= test. */
+    SDL_GPU_COMPAREOP_GREATER_OR_EQUAL,  /**< The comparison evaluates reference >= test. */
    SDL_GPU_COMPAREOP_ALWAYS             /**< The comparison always evaluates true. */
 } SDL_GPUCompareOp;

@@ -1549,7 +1549,7 @@ typedef struct SDL_GPUSamplerCreateInfo
 typedef struct SDL_GPUVertexBufferDescription
 {
    Uint32 slot;                        /**< The binding slot of the vertex buffer. */
-    Uint32 pitch;                       /**< The byte pitch between consecutive elements of the vertex buffer. */
+    Uint32 pitch;                       /**< The size of a single element + the offset between elements. */
    SDL_GPUVertexInputRate input_rate;  /**< Whether attribute addressing is a function of the vertex index or instance index. */
    Uint32 instance_step_rate;          /**< Reserved for future use. Must be set to 0. */
 } SDL_GPUVertexBufferDescription;
@@ -2648,7 +2648,7 @@ extern SDL_DECLSPEC void SDLCALL SDL_InsertGPUDebugLabel(
    const char *text);

 /**
- * Begins a debug group with an arbitary name.
+ * Begins a debug group with an arbitrary name.
 *
 * Used for denoting groups of calls when viewing the command buffer
 * callstream in a graphics debugging tool.
@@ -595,7 +595,7 @@ extern "C" {
 * A variable that limits what CPU features are available.
 *
 * By default, SDL marks all features the current CPU supports as available.
- * This hint allows to limit these to a subset.
+ * This hint allows the enabled features to be limited to a subset.
 *
 * When the hint is unset, or empty, SDL will enable all detected CPU
 * features.
@@ -2026,8 +2026,8 @@ extern "C" {
 *
 * The variable can be set to the following values:
 *
- * - "0": RAWINPUT drivers are not used.
- * - "1": RAWINPUT drivers are used. (default)
+ * - "0": RAWINPUT drivers are not used. (default)
+ * - "1": RAWINPUT drivers are used.
 *
 * This hint should be set before SDL is initialized.
 *
@@ -2126,8 +2126,8 @@ extern "C" {
 *
 * The variable can be set to the following values:
 *
- * - "0": WGI is not used.
- * - "1": WGI is used. (default)
+ * - "0": WGI is not used. (default)
+ * - "1": WGI is used.
 *
 * This hint should be set before SDL is initialized.
 *
@@ -79,7 +79,7 @@ typedef Uint32 SDL_InitFlags;

 #define SDL_INIT_AUDIO      0x00000010u /**< `SDL_INIT_AUDIO` implies `SDL_INIT_EVENTS` */
 #define SDL_INIT_VIDEO      0x00000020u /**< `SDL_INIT_VIDEO` implies `SDL_INIT_EVENTS`, should be initialized on the main thread */
-#define SDL_INIT_JOYSTICK   0x00000200u /**< `SDL_INIT_JOYSTICK` implies `SDL_INIT_EVENTS`, should be initialized on the same thread as SDL_INIT_VIDEO on Windows if you don't set SDL_HINT_JOYSTICK_THREAD */
+#define SDL_INIT_JOYSTICK   0x00000200u /**< `SDL_INIT_JOYSTICK` implies `SDL_INIT_EVENTS` */
 #define SDL_INIT_HAPTIC     0x00001000u
 #define SDL_INIT_GAMEPAD    0x00002000u /**< `SDL_INIT_GAMEPAD` implies `SDL_INIT_JOYSTICK` */
 #define SDL_INIT_EVENTS     0x00004000u
@@ -280,12 +280,14 @@ _m_prefetch(void *__P)
 * \sa SDL_TARGETING
 */
 #define SDL_HAS_TARGET_ATTRIBS
-
+#elif defined(__loongarch64) && defined(__GNUC__) && (__GNUC__ >= 15)
+/* LoongArch requires GCC 15+ for target attribute support */
+# define SDL_HAS_TARGET_ATTRIBS
 #elif defined(__clang__) && defined(__has_attribute)
 # if __has_attribute(target)
 # define SDL_HAS_TARGET_ATTRIBS
 # endif
-#elif defined(__GNUC__) && (__GNUC__ + (__GNUC_MINOR__ >= 9) > 4) /* gcc >= 4.9 */
+#elif defined(__GNUC__) && !defined(__loongarch64) && (__GNUC__ + (__GNUC_MINOR__ >= 9) > 4) /* gcc >= 4.9 */
 # define SDL_HAS_TARGET_ATTRIBS
 #elif defined(__ICC) && __ICC >= 1600
 # define SDL_HAS_TARGET_ATTRIBS
@@ -360,7 +360,7 @@ extern SDL_DECLSPEC void SDLCALL SDL_LockMutex(SDL_Mutex *mutex) SDL_ACQUIRE(mut
 * \sa SDL_LockMutex
 * \sa SDL_UnlockMutex
 */
-extern SDL_DECLSPEC bool SDLCALL SDL_TryLockMutex(SDL_Mutex *mutex) SDL_TRY_ACQUIRE(0, mutex);
+extern SDL_DECLSPEC bool SDLCALL SDL_TryLockMutex(SDL_Mutex *mutex) SDL_TRY_ACQUIRE(true, mutex);

 /**
 * Unlock the mutex.
@@ -559,7 +559,7 @@ extern SDL_DECLSPEC void SDLCALL SDL_LockRWLockForWriting(SDL_RWLock *rwlock) SD
 * \sa SDL_TryLockRWLockForWriting
 * \sa SDL_UnlockRWLock
 */
-extern SDL_DECLSPEC bool SDLCALL SDL_TryLockRWLockForReading(SDL_RWLock *rwlock) SDL_TRY_ACQUIRE_SHARED(0, rwlock);
+extern SDL_DECLSPEC bool SDLCALL SDL_TryLockRWLockForReading(SDL_RWLock *rwlock) SDL_TRY_ACQUIRE_SHARED(true, rwlock);

 /**
 * Try to lock a read/write lock _for writing_ without blocking.
@@ -589,7 +589,7 @@ extern SDL_DECLSPEC bool SDLCALL SDL_TryLockRWLockForReading(SDL_RWLock *rwlock)
 * \sa SDL_TryLockRWLockForReading
 * \sa SDL_UnlockRWLock
 */
-extern SDL_DECLSPEC bool SDLCALL SDL_TryLockRWLockForWriting(SDL_RWLock *rwlock) SDL_TRY_ACQUIRE(0, rwlock);
+extern SDL_DECLSPEC bool SDLCALL SDL_TryLockRWLockForWriting(SDL_RWLock *rwlock) SDL_TRY_ACQUIRE(true, rwlock);

 /**
 * Unlock the read/write lock.
@@ -517,7 +517,7 @@ typedef enum SDL_PackedLayout
 *   ABGR32, define a platform-independent encoding into bytes in the order
 *   specified. For example, in RGB24 data, each pixel is encoded in 3 bytes
 *   (red, green, blue) in that order, and in ABGR32 data, each pixel is
- *   encoded in 4 bytes alpha, blue, green, red) in that order. Use these
+ *   encoded in 4 bytes (alpha, blue, green, red) in that order. Use these
 *   names if the property of a format that is important to you is the order
 *   of the bytes in memory or on disk.
 * - Names with a bit count per component, such as ARGB8888 and XRGB1555, are
@@ -125,10 +125,10 @@ typedef struct SDL_FRect
 */
 SDL_FORCE_INLINE void SDL_RectToFRect(const SDL_Rect *rect, SDL_FRect *frect)
 {
-    frect->x = (float)rect->x;
-    frect->y = (float)rect->y;
-    frect->w = (float)rect->w;
-    frect->h = (float)rect->h;
+    frect->x = SDL_static_cast(float, rect->x);
+    frect->y = SDL_static_cast(float, rect->y);
+    frect->w = SDL_static_cast(float, rect->w);
+    frect->h = SDL_static_cast(float, rect->h);
 }

 /**
@@ -2612,7 +2612,7 @@ extern SDL_DECLSPEC bool SDLCALL SDL_RenderDebugText(SDL_Renderer *renderer, flo
 * Draw debug text to an SDL_Renderer.
 *
 * This function will render a printf()-style format string to a renderer.
- * Note that this is a convinence function for debugging, with severe
+ * Note that this is a convenience function for debugging, with severe
 * limitations, and is not intended to be used for production apps and games.
 *
 * For the full list of limitations and other useful information, see
@@ -208,7 +208,7 @@ typedef enum SDL_Scancode

    SDL_SCANCODE_NONUSBACKSLASH = 100, /**< This is the additional key that ISO
                                        *   keyboards have over ANSI ones,
-                                        *   located between left shift and Y.
+                                        *   located between left shift and Z.
                                        *   Produces GRAVE ACCENT and TILDE in a
                                        *   US or UK Mac layout, REVERSE SOLIDUS
                                        *   (backslash) and VERTICAL LINE in a
@@ -138,7 +138,8 @@ typedef enum SDL_SensorType
    SDL_SENSOR_ACCEL_L,         /**< Accelerometer for left Joy-Con controller and Wii nunchuk */
    SDL_SENSOR_GYRO_L,          /**< Gyroscope for left Joy-Con controller */
    SDL_SENSOR_ACCEL_R,         /**< Accelerometer for right Joy-Con controller */
-    SDL_SENSOR_GYRO_R           /**< Gyroscope for right Joy-Con controller */
+    SDL_SENSOR_GYRO_R,          /**< Gyroscope for right Joy-Con controller */
+    SDL_SENSOR_COUNT
 } SDL_SensorType;


@@ -1164,7 +1164,7 @@ typedef struct SDL_alignment_test
    void *b;
 } SDL_alignment_test;
 SDL_COMPILE_TIME_ASSERT(struct_alignment, sizeof(SDL_alignment_test) == (2 * sizeof(void *)));
-SDL_COMPILE_TIME_ASSERT(two_s_complement, (int)~(int)0 == (int)(-1));
+SDL_COMPILE_TIME_ASSERT(two_s_complement, SDL_static_cast(int, ~SDL_static_cast(int, 0)) == SDL_static_cast(int, -1));
 #endif /* DOXYGEN_SHOULD_IGNORE_THIS */
 /** \endcond */

@@ -3426,7 +3426,7 @@ extern SDL_DECLSPEC size_t SDLCALL SDL_utf8strnlen(const char *str, size_t bytes
 * Convert an integer into a string.
 *
 * This requires a radix to specified for string format. Specifying 10
- * produces a decimal number, 16 hexidecimal, etc. Must be in the range of 2
+ * produces a decimal number, 16 hexadecimal, etc. Must be in the range of 2
 * to 36.
 *
 * Note that this function will overflow a buffer if `str` is not large enough
@@ -3454,7 +3454,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_itoa(int value, char *str, int radix);
 * Convert an unsigned integer into a string.
 *
 * This requires a radix to specified for string format. Specifying 10
- * produces a decimal number, 16 hexidecimal, etc. Must be in the range of 2
+ * produces a decimal number, 16 hexadecimal, etc. Must be in the range of 2
 * to 36.
 *
 * Note that this function will overflow a buffer if `str` is not large enough
@@ -3482,7 +3482,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_uitoa(unsigned int value, char *str, int
 * Convert a long integer into a string.
 *
 * This requires a radix to specified for string format. Specifying 10
- * produces a decimal number, 16 hexidecimal, etc. Must be in the range of 2
+ * produces a decimal number, 16 hexadecimal, etc. Must be in the range of 2
 * to 36.
 *
 * Note that this function will overflow a buffer if `str` is not large enough
@@ -3510,7 +3510,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_ltoa(long value, char *str, int radix);
 * Convert an unsigned long integer into a string.
 *
 * This requires a radix to specified for string format. Specifying 10
- * produces a decimal number, 16 hexidecimal, etc. Must be in the range of 2
+ * produces a decimal number, 16 hexadecimal, etc. Must be in the range of 2
 * to 36.
 *
 * Note that this function will overflow a buffer if `str` is not large enough
@@ -3540,7 +3540,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_ultoa(unsigned long value, char *str, int
 * Convert a long long integer into a string.
 *
 * This requires a radix to specified for string format. Specifying 10
- * produces a decimal number, 16 hexidecimal, etc. Must be in the range of 2
+ * produces a decimal number, 16 hexadecimal, etc. Must be in the range of 2
 * to 36.
 *
 * Note that this function will overflow a buffer if `str` is not large enough
@@ -3568,7 +3568,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_lltoa(long long value, char *str, int rad
 * Convert an unsigned long long integer into a string.
 *
 * This requires a radix to specified for string format. Specifying 10
- * produces a decimal number, 16 hexidecimal, etc. Must be in the range of 2
+ * produces a decimal number, 16 hexadecimal, etc. Must be in the range of 2
 * to 36.
 *
 * Note that this function will overflow a buffer if `str` is not large enough
@@ -3923,7 +3923,7 @@ extern SDL_DECLSPEC int SDLCALL SDL_strcasecmp(const char *str1, const char *str
 extern SDL_DECLSPEC int SDLCALL SDL_strncasecmp(const char *str1, const char *str2, size_t maxlen);

 /**
- * Searches a string for the first occurence of any character contained in a
+ * Searches a string for the first occurrence of any character contained in a
 * breakset, and returns a pointer from the string to that character.
 *
 * \param str The null-terminated string to be searched. Must not be NULL, and
@@ -3931,7 +3931,7 @@ extern SDL_DECLSPEC int SDLCALL SDL_strncasecmp(const char *str1, const char *st
 * \param breakset A null-terminated string containing the list of characters
 *                 to look for. Must not be NULL, and must not overlap with
 *                 `str`.
- * \returns A pointer to the location, in str, of the first occurence of a
+ * \returns A pointer to the location, in str, of the first occurrence of a
 *          character present in the breakset, or NULL if none is found.
 *
 * \threadsafety It is safe to call this function from any thread.
@@ -4656,7 +4656,7 @@ extern SDL_DECLSPEC float SDLCALL SDL_atanf(float x);
 *
 * Domain: `-INF <= x <= INF`, `-INF <= y <= INF`
 *
- * Range: `-Pi/2 <= y <= Pi/2`
+ * Range: `-Pi <= y <= Pi`
 *
 * This function operates on double-precision floating point values, use
 * SDL_atan2f for single-precision floats.
@@ -4692,7 +4692,7 @@ extern SDL_DECLSPEC double SDLCALL SDL_atan2(double y, double x);
 *
 * Domain: `-INF <= x <= INF`, `-INF <= y <= INF`
 *
- * Range: `-Pi/2 <= y <= Pi/2`
+ * Range: `-Pi <= y <= Pi`
 *
 * This function operates on single-precision floating point values, use
 * SDL_atan2 for double-precision floats.
@@ -5821,7 +5821,7 @@ extern SDL_DECLSPEC int SDLCALL SDL_iconv_close(SDL_iconv_t cd);
 * This function converts text between encodings, reading from and writing to
 * a buffer.
 *
- * It returns the number of succesful conversions on success. On error,
+ * It returns the number of successful conversions on success. On error,
 * SDL_ICONV_E2BIG is returned when the output buffer is too small, or
 * SDL_ICONV_EILSEQ is returned when an invalid input sequence is encountered,
 * or SDL_ICONV_EINVAL is returned when an incomplete input sequence is
@@ -5921,7 +5921,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_iconv_string(const char *tocode,
 *
 * \since This macro is available since SDL 3.2.0.
 */
-#define SDL_iconv_utf8_ucs2(S)      (Uint16 *)SDL_iconv_string("UCS-2", "UTF-8", S, SDL_strlen(S)+1)
+#define SDL_iconv_utf8_ucs2(S)      SDL_reinterpret_cast(Uint16 *, SDL_iconv_string("UCS-2", "UTF-8", S, SDL_strlen(S)+1))

 /**
 * Convert a UTF-8 string to UCS-4.
@@ -5935,7 +5935,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_iconv_string(const char *tocode,
 *
 * \since This macro is available since SDL 3.2.0.
 */
-#define SDL_iconv_utf8_ucs4(S)      (Uint32 *)SDL_iconv_string("UCS-4", "UTF-8", S, SDL_strlen(S)+1)
+#define SDL_iconv_utf8_ucs4(S)      SDL_reinterpret_cast(Uint32 *, SDL_iconv_string("UCS-4", "UTF-8", S, SDL_strlen(S)+1))

 /**
 * Convert a wchar_t string to UTF-8.
@@ -5949,7 +5949,7 @@ extern SDL_DECLSPEC char * SDLCALL SDL_iconv_string(const char *tocode,
 *
 * \since This macro is available since SDL 3.2.0.
 */
-#define SDL_iconv_wchar_utf8(S)     SDL_iconv_string("UTF-8", "WCHAR_T", (char *)S, (SDL_wcslen(S)+1)*sizeof(wchar_t))
+#define SDL_iconv_wchar_utf8(S)     SDL_iconv_string("UTF-8", "WCHAR_T", SDL_reinterpret_cast(const char *, S), (SDL_wcslen(S)+1)*sizeof(wchar_t))


 /* force builds using Clang's static analysis tools to use literal C runtime
@@ -1279,10 +1279,11 @@ extern SDL_DECLSPEC bool SDLCALL SDL_BlitSurfaceUncheckedScaled(SDL_Surface *src
 *
 * \param src the SDL_Surface structure to be copied from.
 * \param srcrect the SDL_Rect structure representing the rectangle to be
- *                copied, may not be NULL.
+ *                copied, or NULL to copy the entire surface.
 * \param dst the SDL_Surface structure that is the blit target.
 * \param dstrect the SDL_Rect structure representing the target rectangle in
- *                the destination surface, may not be NULL.
+ *                the destination surface, or NULL to fill the entire
+ *                destination surface.
 * \param scaleMode the SDL_ScaleMode to be used.
 * \returns true on success or false on failure; call SDL_GetError() for more
 *          information.
@@ -62,7 +62,7 @@ extern "C" {
 *
 * \since This macro is available since SDL 3.2.0.
 */
-#define SDL_MICRO_VERSION   12
+#define SDL_MICRO_VERSION   28

 /**
 * This macro turns the version numbers into a numeric value.
@@ -1167,6 +1167,15 @@ extern SDL_DECLSPEC SDL_Window * SDLCALL SDL_CreateWindow(const char *title, int
 * Popup windows implicitly do not have a border/decorations and do not appear
 * on the taskbar/dock or in lists of windows such as alt-tab menus.
 *
+ * By default, popup window positions will automatically be constrained to keep
+ * the entire window within display bounds. This can be overridden with the
+ * `SDL_PROP_WINDOW_CREATE_CONSTRAIN_POPUP_BOOLEAN` property.
+ *
+ * By default, popup menus will automatically grab keyboard focus from the parent
+ * when shown. This behavior can be overridden by setting the `SDL_WINDOW_NOT_FOCUSABLE`
+ * flag, setting the `SDL_PROP_WINDOW_CREATE_FOCUSABLE_BOOLEAN` property to false, or
+ * toggling it after creation via the `SDL_SetWindowFocusable()` function.
+ *
 * If a parent window is hidden or destroyed, any child popup windows will be
 * recursively hidden or destroyed as well. Child popup windows not explicitly
 * hidden will be restored when the parent is shown.
@@ -1207,6 +1216,9 @@ extern SDL_DECLSPEC SDL_Window * SDLCALL SDL_CreatePopupWindow(SDL_Window *paren
 *   be always on top
 * - `SDL_PROP_WINDOW_CREATE_BORDERLESS_BOOLEAN`: true if the window has no
 *   window decoration
+ * - `SDL_PROP_WINDOW_CREATE_CONSTRAIN_POPUP_BOOLEAN`: true if the "tooltip" and
+ *   "menu" window types should be automatically constrained to be entirely within
+ *   display bounds (default), false if no constraints on the position are desired.
 * - `SDL_PROP_WINDOW_CREATE_EXTERNAL_GRAPHICS_CONTEXT_BOOLEAN`: true if the
 *   window will be used with an externally managed graphics context.
 * - `SDL_PROP_WINDOW_CREATE_FOCUSABLE_BOOLEAN`: true if the window should
@@ -1321,6 +1333,7 @@ extern SDL_DECLSPEC SDL_Window * SDLCALL SDL_CreateWindowWithProperties(SDL_Prop

 #define SDL_PROP_WINDOW_CREATE_ALWAYS_ON_TOP_BOOLEAN               "SDL.window.create.always_on_top"
 #define SDL_PROP_WINDOW_CREATE_BORDERLESS_BOOLEAN                  "SDL.window.create.borderless"
+#define SDL_PROP_WINDOW_CREATE_CONSTRAIN_POPUP_BOOLEAN             "SDL.window.create.constrain_popup"
 #define SDL_PROP_WINDOW_CREATE_FOCUSABLE_BOOLEAN                   "SDL.window.create.focusable"
 #define SDL_PROP_WINDOW_CREATE_EXTERNAL_GRAPHICS_CONTEXT_BOOLEAN   "SDL.window.create.external_graphics_context"
 #define SDL_PROP_WINDOW_CREATE_FLAGS_NUMBER                        "SDL.window.create.flags"
@@ -1460,7 +1473,7 @@ extern SDL_DECLSPEC SDL_Window * SDLCALL SDL_GetWindowParent(SDL_Window *window)
 * - `SDL_PROP_WINDOW_COCOA_WINDOW_POINTER`: the `(__unsafe_unretained)`
 *   NSWindow associated with the window
 * - `SDL_PROP_WINDOW_COCOA_METAL_VIEW_TAG_NUMBER`: the NSInteger tag
- *   assocated with metal views on the window
+ *   associated with metal views on the window
 *
 * On OpenVR:
 *
@@ -51,14 +51,14 @@
 extern "C" {
 #endif

-/* Avoid including vulkan.h, don't define VkInstance if it's already included */
-#ifdef VULKAN_H_
+/* Avoid including vulkan_core.h, don't define VkInstance if it's already included */
+#ifdef VULKAN_CORE_H_
 #define NO_SDL_VULKAN_TYPEDEFS
 #endif
 #ifndef NO_SDL_VULKAN_TYPEDEFS
 #define VK_DEFINE_HANDLE(object) typedef struct object##_T* object;

-#if defined(__LP64__) || defined(_WIN64) || defined(__x86_64__) || defined(_M_X64) || defined(__ia64) || defined (_M_IA64) || defined(__aarch64__) || defined(__powerpc64__)
+#if defined(__LP64__) || defined(_WIN64) || (defined(__x86_64__) && !defined(__ILP32__)) || defined(_M_X64) || defined(__ia64) || defined (_M_IA64) || defined(__aarch64__) || defined(__powerpc64__) || (defined(__riscv) && __riscv_xlen == 64)
 #define VK_DEFINE_NON_DISPATCHABLE_HANDLE(object) typedef struct object##_T *object;
 #else
 #define VK_DEFINE_NON_DISPATCHABLE_HANDLE(object) typedef uint64_t object;
@@ -174,6 +174,7 @@
 #cmakedefine HAVE_MEMFD_CREATE 1
 #cmakedefine HAVE_POSIX_FALLOCATE 1
 #cmakedefine HAVE_SIGACTION 1
+#cmakedefine HAVE_SIGTIMEDWAIT 1
 #cmakedefine HAVE_SA_SIGACTION 1
 #cmakedefine HAVE_ST_MTIM 1
 #cmakedefine HAVE_SETJMP 1
@@ -111,6 +111,9 @@ typedef unsigned int uintptr_t;
 # define SDL_DISABLE_AVX 1
 #endif

+#define HAVE_STDARG_H 1
+#define HAVE_STDDEF_H 1
+
 /* This can be disabled to avoid C runtime dependencies and manifest requirements */
 #ifndef HAVE_LIBC
 #define HAVE_LIBC 1
@@ -122,8 +125,6 @@ typedef unsigned int uintptr_t;
 #define HAVE_LIMITS_H 1
 #define HAVE_MATH_H 1
 #define HAVE_SIGNAL_H 1
-#define HAVE_STDARG_H 1
-#define HAVE_STDDEF_H 1
 #define HAVE_STDIO_H 1
 #define HAVE_STDLIB_H 1
 #define HAVE_STRING_H 1
@@ -153,7 +154,6 @@ typedef unsigned int uintptr_t;
 #define HAVE_STRCMP 1
 #define HAVE_STRNCMP 1
 #define HAVE_STRPBRK 1
-#define HAVE_VSSCANF 1
 #define HAVE_VSNPRINTF 1
 #define HAVE_ACOS 1
 #define HAVE_ASIN 1
@@ -211,10 +211,7 @@ typedef unsigned int uintptr_t;
 #if _MSC_VER >= 1400
 #define HAVE__FSEEKI64 1
 #endif
-#endif    /* _MSC_VER */
-#else
-#define HAVE_STDARG_H 1
-#define HAVE_STDDEF_H 1
+#endif /* _MSC_VER */
 #endif

 /* Enable various audio drivers */
@@ -65,7 +65,7 @@

 // Initialization/Cleanup routines
 #include "timer/SDL_timer_c.h"
-#ifdef SDL_VIDEO_DRIVER_WINDOWS
+#ifdef SDL_PLATFORM_WINDOWS
 extern bool SDL_HelperWindowCreate(void);
 extern void SDL_HelperWindowDestroy(void);
 #endif
@@ -317,7 +317,7 @@ bool SDL_InitSubSystem(SDL_InitFlags flags)
    SDL_DBus_Init();
 #endif

-#ifdef SDL_VIDEO_DRIVER_WINDOWS
+#ifdef SDL_PLATFORM_WINDOWS
    if (flags & (SDL_INIT_HAPTIC | SDL_INIT_JOYSTICK)) {
        if (!SDL_HelperWindowCreate()) {
            goto quit_and_error;
@@ -653,7 +653,7 @@ void SDL_Quit(void)
    SDL_bInMainQuit = true;

    // Quit all subsystems
-#ifdef SDL_VIDEO_DRIVER_WINDOWS
+#ifdef SDL_PLATFORM_WINDOWS
    SDL_HelperWindowDestroy();
 #endif
    SDL_QuitSubSystem(SDL_INIT_EVERYTHING);
@@ -265,6 +265,12 @@ extern "C" {
 #include "SDL_utils_c.h"
 #include "SDL_hashtable.h"

+#define PUSH_SDL_ERROR() \
+    { char *_error = SDL_strdup(SDL_GetError());
+
+#define POP_SDL_ERROR() \
+    SDL_SetError("%s", _error); SDL_free(_error); }
+
 // Do any initialization that needs to happen before threads are started
 extern void SDL_InitMainThread(void);

@@ -410,6 +410,7 @@ static SDL_LogicalAudioDevice *ObtainLogicalAudioDevice(SDL_AudioDeviceID devid,
        SDL_LockRWLockForReading(current_audio.device_hash_lock);
        SDL_FindInHashTable(current_audio.device_hash, (const void *) (uintptr_t) devid, (const void **) &logdev);
        if (logdev) {
+            SDL_assert(logdev->instance_id == devid);
            device = logdev->physical_device;
            SDL_assert(device != NULL);
            RefPhysicalAudioDevice(device);  // reference it, in case the logical device migrates to a new default.
@@ -459,6 +460,7 @@ static SDL_AudioDevice *ObtainPhysicalAudioDevice(SDL_AudioDeviceID devid)  // !
    } else {
        SDL_LockRWLockForReading(current_audio.device_hash_lock);
        SDL_FindInHashTable(current_audio.device_hash, (const void *) (uintptr_t) devid, (const void **) &device);
+        SDL_assert(device->instance_id == devid);
        SDL_UnlockRWLock(current_audio.device_hash_lock);

        if (!device) {
@@ -883,6 +885,7 @@ static bool SDLCALL FindLowestDeviceID(void *userdata, const SDL_HashTable *tabl
    if (isphysical && (devid_recording == data->recording) && (devid < data->highest)) {
        data->highest = devid;
        data->result = (SDL_AudioDevice *) value;
+        SDL_assert(data->result->instance_id == devid);
    }
    return true;  // keep iterating.
 }
@@ -1051,7 +1054,10 @@ static bool SDLCALL DestroyOnePhysicalAudioDevice(void *userdata, const SDL_Hash
    const SDL_AudioDeviceID devid = (SDL_AudioDeviceID) (uintptr_t) key;
    const bool isphysical = !!(devid & (1<<1));
    if (isphysical) {
-        DestroyPhysicalAudioDevice((SDL_AudioDevice *) value);
+        SDL_AudioDevice *dev = (SDL_AudioDevice *) value;
+
+        SDL_assert(dev->instance_id == devid);
+        DestroyPhysicalAudioDevice(dev);
    }
    return true;  // keep iterating.
 }
@@ -1147,7 +1153,10 @@ bool SDL_PlaybackAudioThreadIterate(SDL_AudioDevice *device)
            // We should have updated this elsewhere if the format changed!
            SDL_assert(SDL_AudioSpecsEqual(&stream->dst_spec, &device->spec, NULL, NULL));

+            SDL_assert(stream->src_spec.format != SDL_AUDIO_UNKNOWN);
+
            const int br = SDL_GetAtomicInt(&logdev->paused) ? 0 : SDL_GetAudioStreamDataAdjustGain(stream, device_buffer, buffer_size, logdev->gain);
+
            if (br < 0) {  // Probably OOM. Kill the audio device; the whole thing is likely dying soon anyhow.
                failed = true;
                SDL_memset(device_buffer, device->silence_value, buffer_size);  // just supply silence to the device before we die.
@@ -1189,6 +1198,8 @@ bool SDL_PlaybackAudioThreadIterate(SDL_AudioDevice *device)
                    // We should have updated this elsewhere if the format changed!
                    SDL_assert(SDL_AudioSpecsEqual(&stream->dst_spec, &outspec, NULL, NULL));

+                    SDL_assert(stream->src_spec.format != SDL_AUDIO_UNKNOWN);
+
                    /* this will hold a lock on `stream` while getting. We don't explicitly lock the streams
                       for iterating here because the binding linked list can only change while the device lock is held.
                       (we _do_ lock the stream during binding/unbinding to make sure that two threads can't try to bind
@@ -1243,7 +1254,11 @@ void SDL_PlaybackAudioThreadShutdown(SDL_AudioDevice *device)
    const int frames = device->buffer_size / SDL_AUDIO_FRAMESIZE(device->spec);
    // Wait for the audio to drain if device didn't die.
    if (!SDL_GetAtomicInt(&device->zombie)) {
-        SDL_Delay(((frames * 1000) / device->spec.freq) * 2);
+        int delay = ((frames * 1000) / device->spec.freq) * 2;
+        if (delay > 100) {
+            delay = 100;
+        }
+        SDL_Delay(delay);
    }
    current_audio.impl.ThreadDeinit(device);
    SDL_AudioThreadFinalize(device);
@@ -1324,6 +1339,7 @@ bool SDL_RecordingAudioThreadIterate(SDL_AudioDevice *device)
                    SDL_assert(stream->src_spec.format == ((logdev->postmix || (logdev->gain != 1.0f)) ? SDL_AUDIO_F32 : device->spec.format));
                    SDL_assert(stream->src_spec.channels == device->spec.channels);
                    SDL_assert(stream->src_spec.freq == device->spec.freq);
+                    SDL_assert(stream->dst_spec.format != SDL_AUDIO_UNKNOWN);

                    void *final_buf = output_buffer;

@@ -1385,6 +1401,7 @@ static int SDLCALL RecordingAudioThread(void *devicep)  // thread entry point
 typedef struct CountAudioDevicesData
 {
    int devs_seen;
+    int devs_skipped;
    const int num_devices;
    SDL_AudioDeviceID *result;
    const bool recording;
@@ -1400,7 +1417,13 @@ static bool SDLCALL CountAudioDevices(void *userdata, const SDL_HashTable *table
    const bool isphysical = !!(devid & (1<<1));
    if (isphysical && (devid_recording == data->recording)) {
        SDL_assert(data->devs_seen < data->num_devices);
-        data->result[data->devs_seen++] = devid;
+        SDL_AudioDevice *device = (SDL_AudioDevice *) value;  // this is normally risky, but we hold the device_hash_lock here.
+        const bool zombie = SDL_GetAtomicInt(&device->zombie) != 0;
+        if (zombie) {
+            data->devs_skipped++;
+        } else {
+            data->result[data->devs_seen++] = devid;
+        }
    }
    return true;  // keep iterating.
 }
@@ -1416,10 +1439,11 @@ static SDL_AudioDeviceID *GetAudioDevices(int *count, bool recording)
            num_devices = SDL_GetAtomicInt(recording ? &current_audio.recording_device_count : &current_audio.playback_device_count);
            result = (SDL_AudioDeviceID *) SDL_malloc((num_devices + 1) * sizeof (SDL_AudioDeviceID));
            if (result) {
-                CountAudioDevicesData data = { 0, num_devices, result, recording };
+                CountAudioDevicesData data = { 0, 0, num_devices, result, recording };
                SDL_IterateHashTable(current_audio.device_hash, CountAudioDevices, &data);
-                SDL_assert(data.devs_seen == num_devices);
-                result[data.devs_seen] = 0;  // null-terminated.
+                SDL_assert((data.devs_seen + data.devs_skipped) == num_devices);
+                num_devices = data.devs_seen;  // might be less if we skipped any.
+                result[num_devices] = 0;  // null-terminated.
            }
        }
        SDL_UnlockRWLock(current_audio.device_hash_lock);
@@ -1464,6 +1488,7 @@ static bool SDLCALL FindAudioDeviceByCallback(void *userdata, const SDL_HashTabl
        SDL_AudioDevice *device = (SDL_AudioDevice *) value;
        if (data->callback(device, data->userdata)) {  // found it?
            data->retval = device;
+            SDL_assert(data->retval->instance_id == devid);
            return false;  // stop iterating, we found it.
        }
    }
@@ -1502,8 +1527,10 @@ SDL_AudioDevice *SDL_FindPhysicalAudioDeviceByHandle(void *handle)

 const char *SDL_GetAudioDeviceName(SDL_AudioDeviceID devid)
 {
+    // bit #1 of devid is set for physical devices and unset for logical.
+    const bool islogical = !(devid & (1<<1));
    const char *result = NULL;
-    SDL_AudioDevice *device = NULL;
+    const void *vdev = NULL;

    if (!SDL_GetCurrentAudioDriver()) {
        SDL_SetError("Audio subsystem is not initialized");
@@ -1513,10 +1540,16 @@ const char *SDL_GetAudioDeviceName(SDL_AudioDeviceID devid)
        // remains valid (in case the device is unplugged at the wrong moment), we hold the
        // device_hash_lock while we copy the string.
        SDL_LockRWLockForReading(current_audio.device_hash_lock);
-        SDL_FindInHashTable(current_audio.device_hash, (const void *) (uintptr_t) devid, (const void **) &device);
-        if (!device) {
+        SDL_FindInHashTable(current_audio.device_hash, (const void *) (uintptr_t) devid, &vdev);
+        if (!vdev) {
            SDL_SetError("Invalid audio device instance ID");
+        } else if (islogical) {
+            const SDL_LogicalAudioDevice *logdev = (const SDL_LogicalAudioDevice *) vdev;
+            SDL_assert(logdev->instance_id == devid);
+            result = SDL_GetPersistentString(logdev->physical_device->name);
        } else {
+            const SDL_AudioDevice *device = (const SDL_AudioDevice *) vdev;
+            SDL_assert(device->instance_id == devid);
            result = SDL_GetPersistentString(device->name);
        }
        SDL_UnlockRWLock(current_audio.device_hash_lock);
@@ -1552,7 +1585,9 @@ int *SDL_GetAudioDeviceChannelMap(SDL_AudioDeviceID devid, int *count)
    SDL_AudioDevice *device = ObtainPhysicalAudioDeviceDefaultAllowed(devid);
    if (device) {
        channels = device->spec.channels;
-        result = SDL_ChannelMapDup(device->chmap, channels);
+        if (channels > 0 && device->chmap) {
+            result = SDL_ChannelMapDup(device->chmap, channels);
+        }
    }
    ReleaseAudioDevice(device);

@@ -1728,13 +1763,18 @@ static bool OpenPhysicalAudioDevice(SDL_AudioDevice *device, const SDL_AudioSpec
    SDL_copyp(&spec, inspec ? inspec : &device->default_spec);
    PrepareAudioFormat(device->recording, &spec);

-    /* We allow the device format to change if it's better than the current settings (by various definitions of "better"). This prevents
-       something low quality, like an old game using S8/8000Hz audio, from ruining a music thing playing at CD quality that tries to open later.
-       (or some VoIP library that opens for mono output ruining your surround-sound game because it got there first).
+    /* We impose a simple minimum on device formats. This prevents something low quality, like an old game using S8/8000Hz audio,
+       from ruining a music thing playing at CD quality that tries to open later, or some VoIP library that opens for mono output
+       ruining your surround-sound game because it got there first.
       These are just requests! The backend may change any of these values during OpenDevice method! */
-    device->spec.format = (SDL_AUDIO_BITSIZE(device->default_spec.format) >= SDL_AUDIO_BITSIZE(spec.format)) ? device->default_spec.format : spec.format;
-    device->spec.freq = SDL_max(device->default_spec.freq, spec.freq);
-    device->spec.channels = SDL_max(device->default_spec.channels, spec.channels);
+
+    const SDL_AudioFormat minimum_format = device->recording ? DEFAULT_AUDIO_RECORDING_FORMAT : DEFAULT_AUDIO_PLAYBACK_FORMAT;
+    const int minimum_channels = device->recording ? DEFAULT_AUDIO_RECORDING_CHANNELS : DEFAULT_AUDIO_PLAYBACK_CHANNELS;
+    const int minimum_freq = device->recording ? DEFAULT_AUDIO_RECORDING_FREQUENCY : DEFAULT_AUDIO_PLAYBACK_FREQUENCY;
+
+    device->spec.format = (SDL_AUDIO_BITSIZE(minimum_format) >= SDL_AUDIO_BITSIZE(spec.format)) ? minimum_format : spec.format;
+    device->spec.channels = SDL_max(minimum_channels, spec.channels);
+    device->spec.freq = SDL_max(minimum_freq, spec.freq);
    device->sample_frames = SDL_GetDefaultSampleFramesFromFreq(device->spec.freq);
    SDL_UpdatedAudioDeviceFormat(device);  // start this off sane.

@@ -1945,10 +1985,6 @@ bool SDL_BindAudioStreams(SDL_AudioDeviceID devid, SDL_AudioStream * const *stre
    } else if (logdev->simplified) {
        result = SDL_SetError("Cannot change stream bindings on device opened with SDL_OpenAudioDeviceStream");
    } else {
-
-        // !!! FIXME: We'll set the device's side's format below, but maybe we should refuse to bind a stream if the app's side doesn't have a format set yet.
-        // !!! FIXME: Actually, why do we allow there to be an invalid format, again?
-
        // make sure start of list is sane.
        SDL_assert(!logdev->bound_streams || (logdev->bound_streams->prev_binding == NULL));

@@ -1983,9 +2019,17 @@ bool SDL_BindAudioStreams(SDL_AudioDeviceID devid, SDL_AudioStream * const *stre

    if (result) {
        // Now that everything is verified, chain everything together.
+        const bool recording = device->recording;
        for (int i = 0; i < num_streams; i++) {
            SDL_AudioStream *stream = streams[i];
            if (stream) {  // shouldn't be NULL, but just in case...
+                // if the stream never had its non-device-end format set, just set it to the device end's format.
+                if (recording && (stream->dst_spec.format == SDL_AUDIO_UNKNOWN)) {
+                    SDL_copyp(&stream->dst_spec, &device->spec);
+                } else if (!recording && (stream->src_spec.format == SDL_AUDIO_UNKNOWN)) {
+                    SDL_copyp(&stream->src_spec, &device->spec);
+                }
+
                stream->bound_device = logdev;
                stream->prev_binding = NULL;
                stream->next_binding = logdev->bound_streams;
@@ -537,8 +537,10 @@ static void SDL_TARGETING("ssse3") SDL_Convert_Swap32_SSSE3(Uint32* dst, const U
 // behavior. However, the compiler support for this pragma is bad.
 #if defined(__clang__)
 #if __clang_major__ >= 12
+#if defined(__aarch64__)
 #pragma STDC FENV_ACCESS ON
 #endif
+#endif
 #elif defined(_MSC_VER)
 #pragma fenv_access (on)
 #elif defined(__GNUC__)
@@ -814,8 +816,10 @@ static void SDL_Convert_Swap32_NEON(Uint32* dst, const Uint32* src, int num_samp

 #if defined(__clang__)
 #if __clang_major__ >= 12
+#if defined(__aarch64__)
 #pragma STDC FENV_ACCESS DEFAULT
 #endif
+#endif
 #elif defined(_MSC_VER)
 #pragma fenv_access (off)
 #elif defined(__GNUC__)
@@ -1775,6 +1775,7 @@ static bool WaveLoad(SDL_IOStream *src, WaveFile *file, SDL_AudioSpec *spec, Uin
    int result;
    Uint32 chunkcount = 0;
    Uint32 chunkcountlimit = 10000;
+    const Sint64 flen = SDL_GetIOSize(src);   // this might be -1 if the IOStream can't determine the total size.
    const char *hint;
    Sint64 RIFFstart, RIFFend, lastchunkpos;
    bool RIFFlengthknown = false;
@@ -1883,6 +1884,14 @@ static bool WaveLoad(SDL_IOStream *src, WaveFile *file, SDL_AudioSpec *spec, Uin
                fmtchunk = *chunk;
            }
        } else if (chunk->fourcc == DATA) {
+            /* If the data chunk is bigger than the file, it might be corrupt
+               or the file is truncated. Try to recover by clamping the file
+               size. This also means a malicious file can't allocate 4 gigabytes
+               for the chunks without actually supplying a 4 gigabyte file. */
+            if ((flen > 0) && ((chunk->position + chunk->length) > flen)) {
+                chunk->length = (Uint32) (flen - chunk->position);
+            }
+
            /* Only use the first data chunk. Handling the wavl list madness
             * may require a different approach.
             */
@@ -2114,8 +2123,8 @@ bool SDL_LoadWAV_IO(SDL_IOStream *src, bool closeio, SDL_AudioSpec *spec, Uint8
    result = WaveLoad(src, &file, spec, audio_buf, audio_len);
    if (!result) {
        SDL_free(*audio_buf);
-        audio_buf = NULL;
-        audio_len = 0;
+        *audio_buf = NULL;
+        *audio_len = 0;
    }

    // Cleanup
@@ -308,6 +308,12 @@ static bool BuildAAudioStream(SDL_AudioDevice *device)
    ctx.AAudioStreamBuilder_setFormat(builder, format);
    ctx.AAudioStreamBuilder_setSampleRate(builder, device->spec.freq);
    ctx.AAudioStreamBuilder_setChannelCount(builder, device->spec.channels);
+	
+    // If no specific buffer size has been requested, the device will pick the optimal 
+	if(SDL_GetHint(SDL_HINT_AUDIO_DEVICE_SAMPLE_FRAMES)) {
+	    ctx.AAudioStreamBuilder_setBufferCapacityInFrames(builder, 2 * device->sample_frames); // AAudio requires that the buffer capacity is at least
+	    ctx.AAudioStreamBuilder_setFramesPerDataCallback(builder, device->sample_frames);      // twice the size of the data callback buffer size
+	}

    const aaudio_direction_t direction = (recording ? AAUDIO_DIRECTION_INPUT : AAUDIO_DIRECTION_OUTPUT);
    ctx.AAudioStreamBuilder_setDirection(builder, direction);
@@ -31,7 +31,7 @@ SDL_PROC_UNUSED(void, AAudioStreamBuilder_setSamplesPerFrame, (AAudioStreamBuild
 SDL_PROC(void, AAudioStreamBuilder_setFormat, (AAudioStreamBuilder * builder, aaudio_format_t format))
 SDL_PROC_UNUSED(void, AAudioStreamBuilder_setSharingMode, (AAudioStreamBuilder * builder, aaudio_sharing_mode_t sharingMode))
 SDL_PROC(void, AAudioStreamBuilder_setDirection, (AAudioStreamBuilder * builder, aaudio_direction_t direction))
-SDL_PROC_UNUSED(void, AAudioStreamBuilder_setBufferCapacityInFrames, (AAudioStreamBuilder * builder, int32_t numFrames))
+SDL_PROC(void, AAudioStreamBuilder_setBufferCapacityInFrames, (AAudioStreamBuilder * builder, int32_t numFrames))
 SDL_PROC(void, AAudioStreamBuilder_setPerformanceMode, (AAudioStreamBuilder * builder, aaudio_performance_mode_t mode))
 SDL_PROC_UNUSED(void, AAudioStreamBuilder_setUsage, (AAudioStreamBuilder * builder, aaudio_usage_t usage))                                         // API 28
 SDL_PROC_UNUSED(void, AAudioStreamBuilder_setContentType, (AAudioStreamBuilder * builder, aaudio_content_type_t contentType))                      // API 28
@@ -461,8 +461,6 @@ static void ALSA_CloseDevice(SDL_AudioDevice *device)
 {
    if (device->hidden) {
        if (device->hidden->pcm) {
-            // Wait for the submitted audio to drain. ALSA_snd_pcm_drop() can hang, so don't use that.
-            SDL_Delay(((device->sample_frames * 1000) / device->spec.freq) * 2);
            ALSA_snd_pcm_close(device->hidden->pcm);
        }
        SDL_free(device->hidden->mixbuf);
@@ -1156,7 +1154,7 @@ static bool ALSA_OpenDevice(SDL_AudioDevice *device)
    #if SDL_ALSA_DEBUG
    snd_pcm_uframes_t bufsize;
    ALSA_snd_pcm_hw_params_get_buffer_size(cfg_ctx.hwparams, &bufsize);
-    SDL_LogError(SDL_LOG_CATEGORY_AUDIO,
+    SDL_LogDebug(SDL_LOG_CATEGORY_AUDIO,
                     "ALSA: period size = %ld, periods = %u, buffer size = %lu",
                     cfg_ctx.persize, cfg_ctx.periods, bufsize);
    #endif
@@ -420,7 +420,8 @@ static bool UpdateAudioSession(SDL_AudioDevice *device, bool open, bool allow_pl

        hint = SDL_GetHint(SDL_HINT_AUDIO_CATEGORY);
        if (hint) {
-            if (SDL_strcasecmp(hint, "AVAudioSessionCategoryAmbient") == 0) {
+            if (SDL_strcasecmp(hint, "AVAudioSessionCategoryAmbient") == 0 ||
+                SDL_strcasecmp(hint, "ambient") == 0) {
                category = AVAudioSessionCategoryAmbient;
            } else if (SDL_strcasecmp(hint, "AVAudioSessionCategorySoloAmbient") == 0) {
                category = AVAudioSessionCategorySoloAmbient;
@@ -206,7 +206,7 @@ static void DSOUND_DetectDevices(SDL_AudioDevice **default_playback, SDL_AudioDe
 {
 #ifdef HAVE_MMDEVICEAPI_H
    if (SupportsIMMDevice) {
-        SDL_IMMDevice_EnumerateEndpoints(default_playback, default_recording);
+        SDL_IMMDevice_EnumerateEndpoints(default_playback, default_recording, SDL_AUDIO_UNKNOWN);
    } else
 #endif
    {
@@ -189,7 +189,7 @@ static bool EMSCRIPTENAUDIO_OpenDevice(SDL_AudioDevice *device)
    }

    // limit to native freq
-    device->spec.freq = EM_ASM_INT({ return Module['SDL3'].audioContext.sampleRate; });
+    device->spec.freq = MAIN_THREAD_EM_ASM_INT({ return Module['SDL3'].audioContext.sampleRate; });
    device->sample_frames = SDL_GetDefaultSampleFramesFromFreq(device->spec.freq) * 2;  // double the buffer size, some browsers need more, and we'll just have to live with the latency.

    SDL_UpdatedAudioDeviceFormat(device);
@@ -426,28 +426,33 @@ static void OPENSLES_DestroyPCMPlayer(SDL_AudioDevice *device)

 static bool OPENSLES_CreatePCMPlayer(SDL_AudioDevice *device)
 {
-    /* If we want to add floating point audio support (requires API level 21)
-       it can be done as described here:
-        https://developer.android.com/ndk/guides/audio/opensl/android-extensions.html#floating-point
-    */
+    /* according to https://developer.android.com/ndk/guides/audio/opensl/opensl-for-android,
+       Android's OpenSL ES only supports Uint8 and _littleendian_ Sint16.
+       (and float32, with an extension we use, below.) */
    if (SDL_GetAndroidSDKVersion() >= 21) {
        const SDL_AudioFormat *closefmts = SDL_ClosestAudioFormats(device->spec.format);
        SDL_AudioFormat test_format;
        while ((test_format = *(closefmts++)) != 0) {
-            if (SDL_AUDIO_ISSIGNED(test_format)) {
+            switch (test_format) {
+            case SDL_AUDIO_U8:
+            case SDL_AUDIO_S16LE:
+            case SDL_AUDIO_F32:
                break;
+            default:
+                continue;
            }
+            break;
        }

        if (!test_format) {
            // Didn't find a compatible format :
-            LOGI("No compatible audio format, using signed 16-bit audio");
-            test_format = SDL_AUDIO_S16;
+            LOGI("No compatible audio format, using signed 16-bit LE audio");
+            test_format = SDL_AUDIO_S16LE;
        }
        device->spec.format = test_format;
    } else {
        // Just go with signed 16-bit audio as it's the most compatible
-        device->spec.format = SDL_AUDIO_S16;
+        device->spec.format = SDL_AUDIO_S16LE;
    }

    // Update the fragment size as size in bytes
@@ -114,7 +114,7 @@ static bool PSPAUDIO_PlayDevice(SDL_AudioDevice *device, const Uint8 *buffer, in
    } else {
        rc = sceAudioOutputPannedBlocking(device->hidden->channel, PSP_AUDIO_VOLUME_MAX, PSP_AUDIO_VOLUME_MAX, (void *) buffer);
    }
-    return (rc == 0);
+    return (rc >= 0);
 }

 static bool PSPAUDIO_WaitDevice(SDL_AudioDevice *device)
@@ -269,19 +269,9 @@ static const char *getAppName(void)
    return SDL_GetAppMetadataProperty(SDL_PROP_APP_METADATA_NAME_STRING);
 }

-static void ThreadedMainloopSignal(void)
-{
-    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);  // alert waiting threads to unblock.
-
-    // we need to kill any SDL_SetError state; we didn't create this thread
-    //  so its SDL TLS slot will leak otherwise, so we do this every time
-    //  we're (presumably) losing control of the thread.
-    SDL_CleanupTLS();
-}
-
 static void OperationStateChangeCallback(pa_operation *o, void *userdata)
 {
-    ThreadedMainloopSignal();  // just signal any waiting code, it can look up the details.
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);  // just signal any waiting code, it can look up the details.
 }

 /* This function assume you are holding `mainloop`'s lock. The operation is unref'd in here, assuming
@@ -323,7 +313,7 @@ static void DisconnectFromPulseServer(void)

 static void PulseContextStateChangeCallback(pa_context *context, void *userdata)
 {
-    ThreadedMainloopSignal();  // just signal any waiting code, it can look up the details.
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);  // just signal any waiting code, it can look up the details.
 }

 static bool ConnectToPulseServer(void)
@@ -410,7 +400,7 @@ static void WriteCallback(pa_stream *p, size_t nbytes, void *userdata)
    struct SDL_PrivateAudioData *h = (struct SDL_PrivateAudioData *)userdata;
    //SDL_Log("PULSEAUDIO WRITE CALLBACK! nbytes=%u", (unsigned int) nbytes);
    h->bytes_requested += nbytes;
-    ThreadedMainloopSignal();
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);
 }

 // This function waits until it is possible to write a full sound buffer
@@ -481,7 +471,7 @@ static Uint8 *PULSEAUDIO_GetDeviceBuf(SDL_AudioDevice *device, int *buffer_size)
 static void ReadCallback(pa_stream *p, size_t nbytes, void *userdata)
 {
    //SDL_Log("PULSEAUDIO READ CALLBACK! nbytes=%u", (unsigned int) nbytes);
-    ThreadedMainloopSignal();  // the recording code queries what it needs, we just need to signal to end any wait
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);  // the recording code queries what it needs, we just need to signal to end any wait
 }

 static bool PULSEAUDIO_WaitRecordingDevice(SDL_AudioDevice *device)
@@ -602,7 +592,72 @@ static void PULSEAUDIO_CloseDevice(SDL_AudioDevice *device)

 static void PulseStreamStateChangeCallback(pa_stream *stream, void *userdata)
 {
-    ThreadedMainloopSignal();  // just signal any waiting code, it can look up the details.
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);  // just signal any waiting code, it can look up the details.
+}
+
+// Channel maps that match the order in SDL_Audio.h
+static const pa_channel_position_t Pulse_map_1[] = { PA_CHANNEL_POSITION_MONO };
+static const pa_channel_position_t Pulse_map_2[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT };
+
+static const pa_channel_position_t Pulse_map_3[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT,
+                                                     PA_CHANNEL_POSITION_LFE };
+
+static const pa_channel_position_t Pulse_map_4[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT,
+                                                     PA_CHANNEL_POSITION_REAR_LEFT, PA_CHANNEL_POSITION_REAR_RIGHT };
+
+static const pa_channel_position_t Pulse_map_5[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT,
+                                                     PA_CHANNEL_POSITION_LFE,
+                                                     PA_CHANNEL_POSITION_REAR_LEFT, PA_CHANNEL_POSITION_REAR_RIGHT };
+
+static const pa_channel_position_t Pulse_map_6[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT,
+                                                     PA_CHANNEL_POSITION_FRONT_CENTER, PA_CHANNEL_POSITION_LFE,
+                                                     PA_CHANNEL_POSITION_REAR_LEFT, PA_CHANNEL_POSITION_REAR_RIGHT };
+
+static const pa_channel_position_t Pulse_map_7[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT,
+                                                     PA_CHANNEL_POSITION_FRONT_CENTER, PA_CHANNEL_POSITION_LFE,
+                                                     PA_CHANNEL_POSITION_REAR_CENTER,
+                                                     PA_CHANNEL_POSITION_SIDE_LEFT, PA_CHANNEL_POSITION_SIDE_RIGHT };
+
+static const pa_channel_position_t Pulse_map_8[] = { PA_CHANNEL_POSITION_FRONT_LEFT, PA_CHANNEL_POSITION_FRONT_RIGHT,
+                                                     PA_CHANNEL_POSITION_FRONT_CENTER, PA_CHANNEL_POSITION_LFE,
+                                                     PA_CHANNEL_POSITION_REAR_LEFT, PA_CHANNEL_POSITION_REAR_RIGHT,
+                                                     PA_CHANNEL_POSITION_SIDE_LEFT, PA_CHANNEL_POSITION_SIDE_RIGHT };
+
+#define COPY_CHANNEL_MAP(c) SDL_memcpy(pacmap->map, Pulse_map_##c, sizeof(Pulse_map_##c))
+
+static void PulseCreateChannelMap(pa_channel_map *pacmap, uint8_t channels)
+{
+    SDL_assert(channels <= PA_CHANNELS_MAX);
+
+    pacmap->channels = channels;
+
+    switch (channels) {
+    case 1:
+        COPY_CHANNEL_MAP(1);
+        break;
+    case 2:
+        COPY_CHANNEL_MAP(2);
+        break;
+    case 3:
+        COPY_CHANNEL_MAP(3);
+        break;
+    case 4:
+        COPY_CHANNEL_MAP(4);
+        break;
+    case 5:
+        COPY_CHANNEL_MAP(5);
+        break;
+    case 6:
+        COPY_CHANNEL_MAP(6);
+        break;
+    case 7:
+        COPY_CHANNEL_MAP(7);
+        break;
+    case 8:
+        COPY_CHANNEL_MAP(8);
+        break;
+    }
+
 }

 static bool PULSEAUDIO_OpenDevice(SDL_AudioDevice *device)
@@ -682,7 +737,8 @@ static bool PULSEAUDIO_OpenDevice(SDL_AudioDevice *device)
    paspec.rate = device->spec.freq;

    // Reduced prebuffering compared to the defaults.
-    paattr.fragsize = device->buffer_size;   // despite the name, this is only used for recording devices, according to PulseAudio docs!
+
+    paattr.fragsize = device->buffer_size * 2;   // despite the name, this is only used for recording devices, according to PulseAudio docs!  (times 2 because we want _more_ than our buffer size sent from the server at a time, which helps some drivers).
    paattr.tlength = device->buffer_size;
    paattr.prebuf = -1;
    paattr.maxlength = -1;
@@ -692,9 +748,8 @@ static bool PULSEAUDIO_OpenDevice(SDL_AudioDevice *device)
    PULSEAUDIO_pa_threaded_mainloop_lock(pulseaudio_threaded_mainloop);

    const char *name = SDL_GetHint(SDL_HINT_AUDIO_DEVICE_STREAM_NAME);
-    // The SDL ALSA output hints us that we use Windows' channel mapping
-    // https://bugzilla.libsdl.org/show_bug.cgi?id=110
-    PULSEAUDIO_pa_channel_map_init_auto(&pacmap, device->spec.channels, PA_CHANNEL_MAP_WAVEEX);
+
+    PulseCreateChannelMap(&pacmap, device->spec.channels);

    h->stream = PULSEAUDIO_pa_stream_new(
        pulseaudio_context,
@@ -803,7 +858,7 @@ static void SinkInfoCallback(pa_context *c, const pa_sink_info *i, int is_last,
    if (i) {
        AddPulseAudioDevice(false, i->description, i->name, i->index, &i->sample_spec);
    }
-    ThreadedMainloopSignal();
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);
 }

 // This is called when PulseAudio adds a recording ("source") device.
@@ -813,7 +868,7 @@ static void SourceInfoCallback(pa_context *c, const pa_source_info *i, int is_la
    if (i && (include_monitors || (i->monitor_of_sink == PA_INVALID_INDEX))) {
        AddPulseAudioDevice(true, i->description, i->name, i->index, &i->sample_spec);
    }
-    ThreadedMainloopSignal();
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);
 }

 static void ServerInfoCallback(pa_context *c, const pa_server_info *i, void *data)
@@ -838,7 +893,7 @@ static void ServerInfoCallback(pa_context *c, const pa_server_info *i, void *dat
        }
    }

-    ThreadedMainloopSignal();
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);
 }

 static bool FindAudioDeviceByIndex(SDL_AudioDevice *device, void *userdata)
@@ -882,7 +937,7 @@ static void HotplugCallback(pa_context *c, pa_subscription_event_type_t t, uint3
            SDL_AudioDeviceDisconnected(SDL_FindPhysicalAudioDeviceByCallback(FindAudioDeviceByIndex, (void *)(uintptr_t)idx));
        }
    }
-    ThreadedMainloopSignal();
+    PULSEAUDIO_pa_threaded_mainloop_signal(pulseaudio_threaded_mainloop, 0);
 }

 static bool CheckDefaultDevice(const bool changed, char *device_path)
@@ -130,7 +130,8 @@ static bool VITAAUD_OpenDevice(SDL_AudioDevice *device)

 static bool VITAAUD_PlayDevice(SDL_AudioDevice *device, const Uint8 *buffer, int buffer_size)
 {
-    return (sceAudioOutOutput(device->hidden->port, buffer) == 0);
+    // sceAudioOutOutput returns amount of samples queued or < 0 on error
+    return (sceAudioOutOutput(device->hidden->port, buffer) >= 0);
 }

 // This function waits until it is possible to write a full sound buffer
@@ -337,7 +337,7 @@ typedef struct
 static bool mgmtthrtask_DetectDevices(void *userdata)
 {
    mgmtthrtask_DetectDevicesData *data = (mgmtthrtask_DetectDevicesData *)userdata;
-    SDL_IMMDevice_EnumerateEndpoints(data->default_playback, data->default_recording);
+    SDL_IMMDevice_EnumerateEndpoints(data->default_playback, data->default_recording, SDL_AUDIO_F32);
    return true;
 }

@@ -814,7 +814,7 @@ static void ANDROIDCAMERA_Deinitialize(void)
 static bool ANDROIDCAMERA_Init(SDL_CameraDriverImpl *impl)
 {
    // !!! FIXME: slide this off into a subroutine
-    // system libraries are in android-24 and later; we currently target android-16 and later, so check if they exist at runtime.
+    // system libraries are in android-24 and later; we currently target android-21 and later, so check if they exist at runtime.
    void *libcamera2 = dlopen("libcamera2ndk.so", RTLD_NOW | RTLD_LOCAL);
    if (!libcamera2) {
        SDL_Log("CAMERA: libcamera2ndk.so can't be loaded: %s", dlerror());
@@ -61,7 +61,7 @@ static SDL_CameraFrameResult EMSCRIPTENCAMERA_AcquireFrame(SDL_Camera *device, S

        SDL3.camera.ctx2d.drawImage(SDL3.camera.video, 0, 0, w, h);
        const imgrgba = SDL3.camera.ctx2d.getImageData(0, 0, w, h).data;
-        Module.HEAPU8.set(imgrgba, rgba);
+        HEAPU8.set(imgrgba, rgba);

        return 1;
    }, device->actual_spec.width, device->actual_spec.height, rgba);
@@ -570,7 +570,7 @@ static bool V4L2_OpenDevice(SDL_Camera *device, const SDL_CameraSpec *spec)
        setfps.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
        if (xioctl(fd, VIDIOC_G_PARM, &setfps) == 0) {
            if ( (setfps.parm.capture.timeperframe.denominator != spec->framerate_numerator) ||
-                 (setfps.parm.capture.timeperframe.numerator = spec->framerate_denominator) ) {
+                 (setfps.parm.capture.timeperframe.numerator != spec->framerate_denominator) ) {
                setfps.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;
                setfps.parm.capture.timeperframe.numerator = spec->framerate_denominator;
                setfps.parm.capture.timeperframe.denominator = spec->framerate_numerator;
@@ -207,7 +207,6 @@ Sint32 JNI_OnLoad(void *vm, void *reserved)
 {
    (void)vm;
    (void)reserved;
-    SDL_Unsupported();
-    return -1; // JNI_ERR
+    return 0x00010004; // JNI_VERSION_1_4
 }
 #endif
@@ -371,6 +371,7 @@ static jmethodID midShowTextInput;
 static jmethodID midSupportsRelativeMouse;
 static jmethodID midOpenFileDescriptor;
 static jmethodID midShowFileDialog;
+static jmethodID midGetPreferredLocales;

 // audio manager
 static jclass mAudioManagerClass;
@@ -660,6 +661,7 @@ JNIEXPORT void JNICALL SDL_JAVA_INTERFACE(nativeSetupJNI)(JNIEnv *env, jclass cl
    midSupportsRelativeMouse = (*env)->GetStaticMethodID(env, mActivityClass, "supportsRelativeMouse", "()Z");
    midOpenFileDescriptor = (*env)->GetStaticMethodID(env, mActivityClass, "openFileDescriptor", "(Ljava/lang/String;Ljava/lang/String;)I");
    midShowFileDialog = (*env)->GetStaticMethodID(env, mActivityClass, "showFileDialog", "([Ljava/lang/String;ZZI)Z");
+    midGetPreferredLocales = (*env)->GetStaticMethodID(env, mActivityClass, "getPreferredLocales", "()Ljava/lang/String;");

    if (!midClipboardGetText ||
        !midClipboardHasText ||
@@ -691,7 +693,8 @@ JNIEXPORT void JNICALL SDL_JAVA_INTERFACE(nativeSetupJNI)(JNIEnv *env, jclass cl
        !midShowTextInput ||
        !midSupportsRelativeMouse ||
        !midOpenFileDescriptor ||
-        !midShowFileDialog) {
+        !midShowFileDialog ||
+        !midGetPreferredLocales) {
        __android_log_print(ANDROID_LOG_WARN, "SDL", "Missing some Java callbacks, do you have the latest version of SDLActivity.java?");
    }

@@ -1077,7 +1080,8 @@ JNIEXPORT void JNICALL SDL_JAVA_INTERFACE(nativeAddTouch)(
 {
    const char *utfname = (*env)->GetStringUTFChars(env, name, NULL);

-    SDL_AddTouch((SDL_TouchID)touchId, SDL_TOUCH_DEVICE_DIRECT, utfname);
+    SDL_AddTouch(Android_ConvertJavaTouchID(touchId),
+            SDL_TOUCH_DEVICE_DIRECT, utfname);

    (*env)->ReleaseStringUTFChars(env, name, utfname);
 }
@@ -1725,18 +1729,24 @@ static bool Android_JNI_ExceptionOccurred(bool silent)
            exceptionName = (jstring)(*env)->CallObjectMethod(env, exceptionClass, mid);
            exceptionNameUTF8 = (*env)->GetStringUTFChars(env, exceptionName, 0);

+            (*env)->DeleteLocalRef(env, classClass);
+
            mid = (*env)->GetMethodID(env, exceptionClass, "getMessage", "()Ljava/lang/String;");
            exceptionMessage = (jstring)(*env)->CallObjectMethod(env, exception, mid);

+            (*env)->DeleteLocalRef(env, exceptionClass);
+
            if (exceptionMessage != NULL) {
                const char *exceptionMessageUTF8 = (*env)->GetStringUTFChars(env, exceptionMessage, 0);
                SDL_SetError("%s: %s", exceptionNameUTF8, exceptionMessageUTF8);
                (*env)->ReleaseStringUTFChars(env, exceptionMessage, exceptionMessageUTF8);
+                (*env)->DeleteLocalRef(env, exceptionMessage);
            } else {
                SDL_SetError("%s", exceptionNameUTF8);
            }

            (*env)->ReleaseStringUTFChars(env, exceptionName, exceptionNameUTF8);
+            (*env)->DeleteLocalRef(env, exceptionName);
        }

        return true;
@@ -2160,7 +2170,7 @@ bool Android_JNI_ShowMessageBox(const SDL_MessageBoxData *messageboxdata, int *b
    mid = (*env)->GetMethodID(env, clazz,
                              "messageboxShowMessageBox", "(ILjava/lang/String;Ljava/lang/String;[I[I[Ljava/lang/String;[I)I");
    *buttonID = (*env)->CallIntMethod(env, context, mid,
-                                      messageboxdata->flags,
+                                      (jint)messageboxdata->flags,
                                      title,
                                      message,
                                      button_flags,
@@ -2414,7 +2424,7 @@ const char *SDL_GetAndroidCachePath(void)
        // fileObj = context.getExternalFilesDir();
        mid = (*env)->GetMethodID(env, (*env)->GetObjectClass(env, context),
                                  "getCacheDir", "()Ljava/io/File;");
-        fileObject = (*env)->CallObjectMethod(env, context, mid, NULL);
+        fileObject = (*env)->CallObjectMethod(env, context, mid);
        if (!fileObject) {
            SDL_SetError("Couldn't get cache directory");
            LocalReferenceHolder_Cleanup(&refs);
@@ -2585,65 +2595,22 @@ bool Android_JNI_ShowToast(const char *message, int duration, int gravity, int x

 bool Android_JNI_GetLocale(char *buf, size_t buflen)
 {
-    AConfiguration *cfg;
-
-    SDL_assert(buflen > 6);
-
-    // Need to re-create the asset manager if locale has changed (SDL_EVENT_LOCALE_CHANGED)
-    Internal_Android_Destroy_AssetManager();
-
-    if (!asset_manager) {
-        Internal_Android_Create_AssetManager();
-    }
-
-    if (!asset_manager) {
-        return false;
-    }
-
-    cfg = AConfiguration_new();
-    if (!cfg) {
-        return false;
-    }
-
-    {
-        char language[2] = {};
-        char country[2] = {};
-        size_t id = 0;
-
-        AConfiguration_fromAssetManager(cfg, asset_manager);
-        AConfiguration_getLanguage(cfg, language);
-        AConfiguration_getCountry(cfg, country);
-
-        // Indonesian is "id" according to ISO 639.2, but on Android is "in" because of Java backwards compatibility
-        if (language[0] == 'i' && language[1] == 'n') {
-            language[1] = 'd';
-        }
-
-        // copy language (not null terminated)
-        if (language[0]) {
-            buf[id++] = language[0];
-            if (language[1]) {
-                buf[id++] = language[1];
+    bool result = false;
+    if (buf && buflen > 0) {
+        *buf = '\0';
+        JNIEnv *env = Android_JNI_GetEnv();
+        jstring string = (jstring)(*env)->CallStaticObjectMethod(env, mActivityClass, midGetPreferredLocales);
+        if (string) {
+            const char *utf8string = (*env)->GetStringUTFChars(env, string, NULL);
+            if (utf8string) {
+                result = true;
+                SDL_strlcpy(buf, utf8string, buflen);
+                (*env)->ReleaseStringUTFChars(env, string, utf8string);
            }
+            (*env)->DeleteLocalRef(env, string);
        }
-
-        buf[id++] = '_';
-
-        // copy country (not null terminated)
-        if (country[0]) {
-            buf[id++] = country[0];
-            if (country[1]) {
-                buf[id++] = country[1];
-            }
-        }
-
-        buf[id++] = '\0';
-        SDL_assert(id <= buflen);
    }
-
-    AConfiguration_delete(cfg);
-
-    return true;
+    return result;
 }

 bool Android_JNI_OpenURL(const char *url)
@@ -2807,6 +2774,7 @@ bool Android_JNI_OpenFileDialog(
    if (filters) {
        jclass stringClass = (*env)->FindClass(env, "java/lang/String");
        filtersArray = (*env)->NewObjectArray(env, nfilters, stringClass, NULL);
+        (*env)->DeleteLocalRef(env, stringClass);

        // Convert to string
        for (int i = 0; i < nfilters; i++) {
@@ -105,7 +105,7 @@ bool GDK_RegisterChangeNotifications(void)
        SDL_LogDebug(SDL_LOG_CATEGORY_APPLICATION, "[GDK] in RegisterAppConstrainedChangeNotification handler");
        SDL_VideoDevice *_this = SDL_GetVideoDevice();
        if (_this) {
-            if (constrained) {
+            if (constrained && !((_this->windows) && _this->windows->text_input_active)) {
                SDL_SetKeyboardFocus(NULL);
            } else {
                SDL_SetKeyboardFocus(_this->windows);
@@ -25,6 +25,7 @@
 #include "SDL_fcitx.h"
 #include "../../video/SDL_sysvideo.h"
 #include "../../events/SDL_keyboard_c.h"
+#include "../../core/unix/SDL_appid.h"
 #include "SDL_dbus.h"

 #ifdef SDL_VIDEO_DRIVER_X11
@@ -53,32 +54,14 @@ typedef struct FcitxClient

 static FcitxClient fcitx_client;

-static char *GetAppName(void)
+static const char *GetAppName(void)
 {
-#if defined(SDL_PLATFORM_LINUX) || defined(SDL_PLATFORM_FREEBSD)
-    char *spot;
-    char procfile[1024];
-    char linkfile[1024];
-    int linksize;
-
-#ifdef SDL_PLATFORM_LINUX
-    (void)SDL_snprintf(procfile, sizeof(procfile), "/proc/%d/exe", getpid());
-#elif defined(SDL_PLATFORM_FREEBSD)
-    (void)SDL_snprintf(procfile, sizeof(procfile), "/proc/%d/file", getpid());
-#endif
-    linksize = readlink(procfile, linkfile, sizeof(linkfile) - 1);
-    if (linksize > 0) {
-        linkfile[linksize] = '\0';
-        spot = SDL_strrchr(linkfile, '/');
-        if (spot) {
-            return SDL_strdup(spot + 1);
-        } else {
-            return SDL_strdup(linkfile);
-        }
+    const char *exe_name = SDL_GetExeName();
+    if (exe_name) {
+        return exe_name;
    }
-#endif // SDL_PLATFORM_LINUX || SDL_PLATFORM_FREEBSD

-    return SDL_strdup("SDL_App");
+    return "SDL_App";
 }

 static size_t Fcitx_GetPreeditString(SDL_DBusContext *dbus,
@@ -281,7 +264,7 @@ static bool FcitxCreateInputContext(SDL_DBusContext *dbus, const char *appname,

 static bool FcitxClientCreateIC(FcitxClient *client)
 {
-    char *appname = GetAppName();
+    const char *appname = GetAppName();
    char *ic_path = NULL;
    SDL_DBusContext *dbus = client->dbus;

@@ -290,8 +273,6 @@ static bool FcitxClientCreateIC(FcitxClient *client)
        ic_path = NULL; // just in case.
    }

-    SDL_free(appname);
-
    if (ic_path) {
        SDL_free(client->ic_path);
        client->ic_path = SDL_strdup(ic_path);
@@ -120,7 +120,7 @@ void SDL_IMMDevice_FreeDeviceHandle(SDL_AudioDevice *device)
    }
 }

-static SDL_AudioDevice *SDL_IMMDevice_Add(const bool recording, const char *devname, WAVEFORMATEXTENSIBLE *fmt, LPCWSTR devid, GUID *dsoundguid)
+static SDL_AudioDevice *SDL_IMMDevice_Add(const bool recording, const char *devname, WAVEFORMATEXTENSIBLE *fmt, LPCWSTR devid, GUID *dsoundguid, SDL_AudioFormat force_format)
 {
    /* You can have multiple endpoints on a device that are mutually exclusive ("Speakers" vs "Line Out" or whatever).
       In a perfect world, things that are unplugged won't be in this collection. The only gotcha is probably for
@@ -162,7 +162,7 @@ static SDL_AudioDevice *SDL_IMMDevice_Add(const bool recording, const char *devn
        SDL_zero(spec);
        spec.channels = (Uint8)fmt->Format.nChannels;
        spec.freq = fmt->Format.nSamplesPerSec;
-        spec.format = SDL_WaveFormatExToSDLFormat((WAVEFORMATEX *)fmt);
+        spec.format = (force_format != SDL_AUDIO_UNKNOWN) ? force_format : SDL_WaveFormatExToSDLFormat((WAVEFORMATEX *)fmt);

        device = SDL_AddAudioDevice(recording, devname, &spec, handle);
        if (!device) {
@@ -183,6 +183,7 @@ typedef struct SDLMMNotificationClient
 {
    const IMMNotificationClientVtbl *lpVtbl;
    SDL_AtomicInt refcount;
+    SDL_AudioFormat force_format;
 } SDLMMNotificationClient;

 static HRESULT STDMETHODCALLTYPE SDLMMNotificationClient_QueryInterface(IMMNotificationClient *client, REFIID iid, void **ppv)
@@ -241,6 +242,7 @@ static HRESULT STDMETHODCALLTYPE SDLMMNotificationClient_OnDeviceRemoved(IMMNoti

 static HRESULT STDMETHODCALLTYPE SDLMMNotificationClient_OnDeviceStateChanged(IMMNotificationClient *iclient, LPCWSTR pwstrDeviceId, DWORD dwNewState)
 {
+    SDLMMNotificationClient *client = (SDLMMNotificationClient *)iclient;
    IMMDevice *device = NULL;

    if (SUCCEEDED(IMMDeviceEnumerator_GetDevice(enumerator, pwstrDeviceId, &device))) {
@@ -255,7 +257,7 @@ static HRESULT STDMETHODCALLTYPE SDLMMNotificationClient_OnDeviceStateChanged(IM
                    GUID dsoundguid;
                    GetMMDeviceInfo(device, &utf8dev, &fmt, &dsoundguid);
                    if (utf8dev) {
-                        SDL_IMMDevice_Add(recording, utf8dev, &fmt, pwstrDeviceId, &dsoundguid);
+                        SDL_IMMDevice_Add(recording, utf8dev, &fmt, pwstrDeviceId, &dsoundguid, client->force_format);
                        SDL_free(utf8dev);
                    }
                } else {
@@ -286,7 +288,7 @@ static const IMMNotificationClientVtbl notification_client_vtbl = {
    SDLMMNotificationClient_OnPropertyValueChanged
 };

-static SDLMMNotificationClient notification_client = { &notification_client_vtbl, { 1 } };
+static SDLMMNotificationClient notification_client = { &notification_client_vtbl, { 1 }, SDL_AUDIO_UNKNOWN };

 bool SDL_IMMDevice_Init(const SDL_IMMDevice_callbacks *callbacks)
 {
@@ -363,7 +365,7 @@ bool SDL_IMMDevice_Get(SDL_AudioDevice *device, IMMDevice **immdevice, bool reco
    return true;
 }

-static void EnumerateEndpointsForFlow(const bool recording, SDL_AudioDevice **default_device)
+static void EnumerateEndpointsForFlow(const bool recording, SDL_AudioDevice **default_device, SDL_AudioFormat force_format)
 {
    /* Note that WASAPI separates "adapter devices" from "audio endpoint devices"
       ...one adapter device ("SoundBlaster Pro") might have multiple endpoint devices ("Speakers", "Line-Out"). */
@@ -405,7 +407,7 @@ static void EnumerateEndpointsForFlow(const bool recording, SDL_AudioDevice **de
                SDL_zero(dsoundguid);
                GetMMDeviceInfo(immdevice, &devname, &fmt, &dsoundguid);
                if (devname) {
-                    SDL_AudioDevice *sdldevice = SDL_IMMDevice_Add(recording, devname, &fmt, devid, &dsoundguid);
+                    SDL_AudioDevice *sdldevice = SDL_IMMDevice_Add(recording, devname, &fmt, devid, &dsoundguid, force_format);
                    if (default_device && default_devid && SDL_wcscmp(default_devid, devid) == 0) {
                        *default_device = sdldevice;
                    }
@@ -422,10 +424,12 @@ static void EnumerateEndpointsForFlow(const bool recording, SDL_AudioDevice **de
    IMMDeviceCollection_Release(collection);
 }

-void SDL_IMMDevice_EnumerateEndpoints(SDL_AudioDevice **default_playback, SDL_AudioDevice **default_recording)
+void SDL_IMMDevice_EnumerateEndpoints(SDL_AudioDevice **default_playback, SDL_AudioDevice **default_recording, SDL_AudioFormat force_format)
 {
-    EnumerateEndpointsForFlow(false, default_playback);
-    EnumerateEndpointsForFlow(true, default_recording);
+    EnumerateEndpointsForFlow(false, default_playback, force_format);
+    EnumerateEndpointsForFlow(true, default_recording, force_format);
+
+    notification_client.force_format = force_format;

    // if this fails, we just won't get hotplug events. Carry on anyhow.
    IMMDeviceEnumerator_RegisterEndpointNotificationCallback(enumerator, (IMMNotificationClient *)&notification_client);
@@ -37,7 +37,7 @@ typedef struct SDL_IMMDevice_callbacks
 bool SDL_IMMDevice_Init(const SDL_IMMDevice_callbacks *callbacks);
 void SDL_IMMDevice_Quit(void);
 bool SDL_IMMDevice_Get(struct SDL_AudioDevice *device, IMMDevice **immdevice, bool recording);
-void SDL_IMMDevice_EnumerateEndpoints(struct SDL_AudioDevice **default_playback, struct SDL_AudioDevice **default_recording);
+void SDL_IMMDevice_EnumerateEndpoints(struct SDL_AudioDevice **default_playback, struct SDL_AudioDevice **default_recording, SDL_AudioFormat force_format);
 LPGUID SDL_IMMDevice_GetDirectSoundGUID(struct SDL_AudioDevice *device);
 LPCWSTR SDL_IMMDevice_GetDevID(struct SDL_AudioDevice *device);
 void SDL_IMMDevice_FreeDeviceHandle(struct SDL_AudioDevice *device);
@@ -53,6 +53,78 @@ typedef enum RO_INIT_TYPE
 #define WC_ERR_INVALID_CHARS 0x00000080
 #endif

+// Fake window to help with DirectInput events.
+HWND SDL_HelperWindow = NULL;
+static const TCHAR *SDL_HelperWindowClassName = TEXT("SDLHelperWindowInputCatcher");
+static const TCHAR *SDL_HelperWindowName = TEXT("SDLHelperWindowInputMsgWindow");
+static ATOM SDL_HelperWindowClass = 0;
+
+/*
+ * Creates a HelperWindow used for DirectInput.
+ */
+bool SDL_HelperWindowCreate(void)
+{
+    HINSTANCE hInstance = GetModuleHandle(NULL);
+    WNDCLASS wce;
+
+    // Make sure window isn't created twice.
+    if (SDL_HelperWindow != NULL) {
+        return true;
+    }
+
+    // Create the class.
+    SDL_zero(wce);
+    wce.lpfnWndProc = DefWindowProc;
+    wce.lpszClassName = SDL_HelperWindowClassName;
+    wce.hInstance = hInstance;
+
+    // Register the class.
+    SDL_HelperWindowClass = RegisterClass(&wce);
+    if (SDL_HelperWindowClass == 0 && GetLastError() != ERROR_CLASS_ALREADY_EXISTS) {
+        return WIN_SetError("Unable to create Helper Window Class");
+    }
+
+    // Create the window.
+    SDL_HelperWindow = CreateWindowEx(0, SDL_HelperWindowClassName,
+                                      SDL_HelperWindowName,
+                                      WS_OVERLAPPED, CW_USEDEFAULT,
+                                      CW_USEDEFAULT, CW_USEDEFAULT,
+                                      CW_USEDEFAULT, HWND_MESSAGE, NULL,
+                                      hInstance, NULL);
+    if (!SDL_HelperWindow) {
+        UnregisterClass(SDL_HelperWindowClassName, hInstance);
+        return WIN_SetError("Unable to create Helper Window");
+    }
+
+    return true;
+}
+
+/*
+ * Destroys the HelperWindow previously created with SDL_HelperWindowCreate.
+ */
+void SDL_HelperWindowDestroy(void)
+{
+    HINSTANCE hInstance = GetModuleHandle(NULL);
+
+    // Destroy the window.
+    if (SDL_HelperWindow != NULL) {
+        if (DestroyWindow(SDL_HelperWindow) == 0) {
+            WIN_SetError("Unable to destroy Helper Window");
+            return;
+        }
+        SDL_HelperWindow = NULL;
+    }
+
+    // Unregister the class.
+    if (SDL_HelperWindowClass != 0) {
+        if ((UnregisterClass(SDL_HelperWindowClassName, hInstance)) == 0) {
+            WIN_SetError("Unable to destroy Helper Window Class");
+            return;
+        }
+        SDL_HelperWindowClass = 0;
+    }
+}
+
 // Sets an error message based on an HRESULT
 bool WIN_SetErrorFromHRESULT(const char *prefix, HRESULT hr)
 {
@@ -198,6 +270,24 @@ static BOOL IsWindowsVersionOrGreater(WORD wMajorVersion, WORD wMinorVersion, WO
        return result;
 #endif

+BOOL WIN_IsWine(void)
+{
+    static bool checked;
+    static bool is_wine;
+
+    if (!checked) {
+        HMODULE ntdll = LoadLibrary(TEXT("ntdll.dll"));
+        if (ntdll) {
+            if (GetProcAddress(ntdll, "wine_get_version") != NULL) {
+                is_wine = true;
+            }
+            FreeLibrary(ntdll);
+        }
+        checked = true;
+    }
+    return is_wine;
+}
+
 // this is the oldest thing we run on (and we may lose support for this in SDL3 at any time!),
 //  so there's no "OrGreater" as that would always be TRUE. The other functions are here to
 //  ask "can we support a specific feature?" but this function is here to ask "do we need to do
@@ -133,6 +133,9 @@ extern void WIN_CoUninitialize(void);
 extern HRESULT WIN_RoInitialize(void);
 extern void WIN_RoUninitialize(void);

+// Returns true if we're running on Wine
+extern BOOL WIN_IsWine(void);
+
 // Returns true if we're running on Windows XP (any service pack). DOES NOT CHECK XP "OR GREATER"!
 extern BOOL WIN_IsWindowsXP(void);

@@ -9,8 +9,8 @@ LANGUAGE LANG_ENGLISH, SUBLANG_ENGLISH_US
 //

 VS_VERSION_INFO VERSIONINFO
- FILEVERSION 3,2,12,0
- PRODUCTVERSION 3,2,12,0
+ FILEVERSION 3,2,28,0
+ PRODUCTVERSION 3,2,28,0
 FILEFLAGSMASK 0x3fL
 FILEFLAGS 0x0L
 FILEOS 0x40004L
@@ -23,12 +23,12 @@ BEGIN
        BEGIN
            VALUE "CompanyName", "\0"
            VALUE "FileDescription", "SDL\0"
-            VALUE "FileVersion", "3, 2, 12, 0\0"
+            VALUE "FileVersion", "3, 2, 28, 0\0"
            VALUE "InternalName", "SDL\0"
            VALUE "LegalCopyright", "Copyright (C) 2025 Sam Lantinga\0"
            VALUE "OriginalFilename", "SDL3.dll\0"
            VALUE "ProductName", "Simple DirectMedia Layer\0"
-            VALUE "ProductVersion", "3, 2, 12, 0\0"
+            VALUE "ProductVersion", "3, 2, 28, 0\0"
        END
    END
    BLOCK "VarFileInfo"
@@ -115,7 +115,11 @@
 #define CPU_CFG2_LSX  (1 << 6)
 #define CPU_CFG2_LASX (1 << 7)

-#if defined(SDL_ALTIVEC_BLITTERS) && defined(HAVE_SETJMP) && !defined(SDL_PLATFORM_MACOS) && !defined(SDL_PLATFORM_OPENBSD) && !defined(SDL_PLATFORM_FREEBSD)
+#if !defined(SDL_CPUINFO_DISABLED) && \
+    !((defined(SDL_PLATFORM_MACOS) && (defined(__ppc__) || defined(__ppc64__))) || (defined(SDL_PLATFORM_OPENBSD) && defined(__powerpc__))) && \
+    !(defined(SDL_PLATFORM_FREEBSD) && defined(__powerpc__)) && \
+    !(defined(SDL_PLATFORM_LINUX) && defined(__powerpc__) && defined(HAVE_GETAUXVAL)) && \
+    defined(SDL_ALTIVEC_BLITTERS) && defined(HAVE_SETJMP)
 /* This is the brute force way of detecting instruction sets...
   the idea is borrowed from the libmpeg2 library - thanks!
 */
@@ -344,6 +348,8 @@ static int CPU_haveAltiVec(void)
    elf_aux_info(AT_HWCAP, &cpufeatures, sizeof(cpufeatures));
    altivec = cpufeatures & PPC_FEATURE_HAS_ALTIVEC;
    return altivec;
+#elif defined(SDL_PLATFORM_LINUX) && defined(__powerpc__) && defined(HAVE_GETAUXVAL)
+    altivec = getauxval(AT_HWCAP) & PPC_FEATURE_HAS_ALTIVEC;
 #elif defined(SDL_ALTIVEC_BLITTERS) && defined(HAVE_SETJMP)
    void (*handler)(int sig);
    handler = signal(SIGILL, illegal_instruction);
@@ -27,6 +27,8 @@
 #import <Cocoa/Cocoa.h>
 #import <UniformTypeIdentifiers/UTType.h>

+extern void Cocoa_SetWindowHasModalDialog(SDL_Window *window, bool has_modal);
+
 static void AddFileExtensionType(NSMutableArray *types, const char *pattern_ptr)
 {
    if (!*pattern_ptr) {
@@ -163,6 +165,9 @@ void SDL_SYS_ShowFileDialogWithProperties(SDL_FileDialogType type, SDL_DialogFil

    if (window) {
        w = (__bridge NSWindow *)SDL_GetPointerProperty(SDL_GetWindowProperties(window), SDL_PROP_WINDOW_COCOA_WINDOW_POINTER, NULL);
+        if (w) {
+            Cocoa_SetWindowHasModalDialog(window, true);
+        }
    }

    if (w) {
@@ -186,6 +191,7 @@ void SDL_SYS_ShowFileDialogWithProperties(SDL_FileDialogType type, SDL_DialogFil
                callback(userdata, files, -1);
            }

+            Cocoa_SetWindowHasModalDialog(window, false);
            ReactivateAfterDialog();
        }];
    } else {
@@ -206,6 +212,7 @@ void SDL_SYS_ShowFileDialogWithProperties(SDL_FileDialogType type, SDL_DialogFil
            const char *files[1] = { NULL };
            callback(userdata, files, -1);
        }
+
        ReactivateAfterDialog();
    }
 }
@@ -261,7 +261,7 @@ void windows_ShowFileDialog(void *ptr)

            chosen_files_list[nfiles] = NULL;

-            if (WideCharToMultiByte(CP_UTF8, 0, file_ptr, -1, chosen_folder, MAX_PATH, NULL, NULL) >= MAX_PATH) {
+            if (WideCharToMultiByte(CP_UTF8, 0, file_ptr, -1, chosen_folder, MAX_PATH, NULL, NULL) == 0) {
                SDL_SetError("Path too long or invalid character in path");
                SDL_free(chosen_files_list);
                callback(userdata, NULL, -1);
@@ -273,7 +273,7 @@ void windows_ShowFileDialog(void *ptr)
            SDL_strlcpy(chosen_file, chosen_folder, MAX_PATH);
            chosen_file[chosen_folder_size] = '\\';

-            file_ptr += SDL_strlen(chosen_folder) + 1;
+            file_ptr += SDL_wcslen(file_ptr) + 1;

            while (*file_ptr) {
                nfiles++;
@@ -295,7 +295,7 @@ void windows_ShowFileDialog(void *ptr)

                int diff = ((int) chosen_folder_size) + 1;

-                if (WideCharToMultiByte(CP_UTF8, 0, file_ptr, -1, chosen_file + diff, MAX_PATH - diff, NULL, NULL) >= MAX_PATH - diff) {
+                if (WideCharToMultiByte(CP_UTF8, 0, file_ptr, -1, chosen_file + diff, MAX_PATH - diff, NULL, NULL) == 0) {
                    SDL_SetError("Path too long or invalid character in path");

                    for (size_t i = 0; i < nfiles - 1; i++) {
@@ -308,7 +308,7 @@ void windows_ShowFileDialog(void *ptr)
                    return;
                }

-                file_ptr += SDL_strlen(chosen_file) + 1 - diff;
+                file_ptr += SDL_wcslen(file_ptr) + 1;

                chosen_files_list[nfiles - 1] = SDL_strdup(chosen_file);

@@ -1077,16 +1077,11 @@ static void SDL_SendWakeupEvent(void)
        return;
    }

-    SDL_LockMutex(_this->wakeup_lock);
-    {
-        if (_this->wakeup_window) {
-            _this->SendWakeupEvent(_this, _this->wakeup_window);
-
-            // No more wakeup events needed until we enter a new wait
-            _this->wakeup_window = NULL;
-        }
+    // We only want to do this once while waiting for an event, so set it to NULL atomically here
+    SDL_Window *wakeup_window = (SDL_Window *)SDL_SetAtomicPointer(&_this->wakeup_window, NULL);
+    if (wakeup_window) {
+        _this->SendWakeupEvent(_this, wakeup_window);
    }
-    SDL_UnlockMutex(_this->wakeup_lock);
 #endif
 }

@@ -1113,7 +1108,8 @@ static int SDL_PeepEventsInternal(SDL_Event *events, int numevents, SDL_EventAct
        if (action == SDL_ADDEVENT) {
            if (!events) {
                SDL_UnlockMutex(SDL_EventQ.lock);
-                return SDL_InvalidParamError("events");
+                SDL_InvalidParamError("events");
+                return -1;
            }
            for (i = 0; i < numevents; ++i) {
                used += SDL_AddEvent(&events[i]);
@@ -1524,18 +1520,7 @@ static int SDL_WaitEventTimeout_Device(SDL_VideoDevice *_this, SDL_Window *wakeu
        */
        SDL_PumpEventsInternal(true);

-        SDL_LockMutex(_this->wakeup_lock);
-        {
-            status = SDL_PeepEvents(event, 1, SDL_GETEVENT, SDL_EVENT_FIRST, SDL_EVENT_LAST);
-            // If status == 0 we are going to block so wakeup will be needed.
-            if (status == 0) {
-                _this->wakeup_window = wakeup_window;
-            } else {
-                _this->wakeup_window = NULL;
-            }
-        }
-        SDL_UnlockMutex(_this->wakeup_lock);
-
+        status = SDL_PeepEvents(event, 1, SDL_GETEVENT, SDL_EVENT_FIRST, SDL_EVENT_LAST);
        if (status < 0) {
            // Got an error: return
            break;
@@ -1548,8 +1533,6 @@ static int SDL_WaitEventTimeout_Device(SDL_VideoDevice *_this, SDL_Window *wakeu
        if (timeoutNS > 0) {
            Sint64 elapsed = SDL_GetTicksNS() - start;
            if (elapsed >= timeoutNS) {
-                // Set wakeup_window to NULL without holding the lock.
-                _this->wakeup_window = NULL;
                return 0;
            }
            loop_timeoutNS = (timeoutNS - elapsed);
@@ -1562,9 +1545,9 @@ static int SDL_WaitEventTimeout_Device(SDL_VideoDevice *_this, SDL_Window *wakeu
                loop_timeoutNS = poll_intervalNS;
            }
        }
+        SDL_SetAtomicPointer(&_this->wakeup_window, wakeup_window);
        status = _this->WaitEventTimeout(_this, loop_timeoutNS);
-        // Set wakeup_window to NULL without holding the lock.
-        _this->wakeup_window = NULL;
+        SDL_SetAtomicPointer(&_this->wakeup_window, NULL);
        if (status == 0 && poll_intervalNS != SDL_MAX_SINT64 && loop_timeoutNS == poll_intervalNS) {
            // We may have woken up to poll. Try again
            continue;
@@ -1664,6 +1647,8 @@ bool SDL_WaitEventTimeoutNS(SDL_Event *event, Sint64 timeoutNS)

 #ifdef SDL_PLATFORM_ANDROID
    for (;;) {
+        SDL_PumpEventsInternal(true);
+
        if (SDL_PeepEvents(event, 1, SDL_GETEVENT, SDL_EVENT_FIRST, SDL_EVENT_LAST) > 0) {
            return true;
        }
@@ -333,6 +333,18 @@ bool SDL_SetKeyboardFocus(SDL_Window *window)
        }
    }

+    // See if the current window has lost focus
+    if (keyboard->focus && keyboard->focus != window) {
+        SDL_SendWindowEvent(keyboard->focus, SDL_EVENT_WINDOW_FOCUS_LOST, 0, 0);
+
+        // Ensures IME compositions are committed
+        if (SDL_TextInputActive(keyboard->focus)) {
+            if (video && video->StopTextInput) {
+                video->StopTextInput(video, keyboard->focus);
+            }
+        }
+    }
+
    if (keyboard->focus && !window) {
        // We won't get anymore keyboard messages, so reset keyboard state
        SDL_ResetKeyboard();
@@ -351,18 +363,6 @@ bool SDL_SetKeyboardFocus(SDL_Window *window)
        }
    }

-    // See if the current window has lost focus
-    if (keyboard->focus && keyboard->focus != window) {
-        SDL_SendWindowEvent(keyboard->focus, SDL_EVENT_WINDOW_FOCUS_LOST, 0, 0);
-
-        // Ensures IME compositions are committed
-        if (SDL_TextInputActive(keyboard->focus)) {
-            if (video && video->StopTextInput) {
-                video->StopTextInput(video, keyboard->focus);
-            }
-        }
-    }
-
    keyboard->focus = window;

    if (keyboard->focus) {
@@ -138,21 +138,25 @@ static void SDLCALL SDL_TouchMouseEventsChanged(void *userdata, const char *name
 #ifdef SDL_PLATFORM_VITA
 static void SDLCALL SDL_VitaTouchMouseDeviceChanged(void *userdata, const char *name, const char *oldValue, const char *hint)
 {
+    Uint8 vita_touch_mouse_device = 1;
+
    SDL_Mouse *mouse = (SDL_Mouse *)userdata;
    if (hint) {
        switch (*hint) {
-        default:
        case '0':
-            mouse->vita_touch_mouse_device = 1;
+            vita_touch_mouse_device = 1;
            break;
        case '1':
-            mouse->vita_touch_mouse_device = 2;
+            vita_touch_mouse_device = 2;
            break;
        case '2':
-            mouse->vita_touch_mouse_device = 3;
+            vita_touch_mouse_device = 3;
+            break;
+        default:
            break;
        }
    }
+    mouse->vita_touch_mouse_device = vita_touch_mouse_device;
 }
 #endif

@@ -46,8 +46,10 @@ static bool send_foregrounding_pending = false;

 static void SDL_HandleSIG(int sig)
 {
-    // Reset the signal handler
+#ifndef HAVE_SIGACTION
+    // Reset the signal handler if it was installed with signal()
    (void)signal(sig, SDL_HandleSIG);
+#endif

    // Send a quit event next time the event loop pumps.
    // We can't send it in signal handler; SDL_malloc() might be interrupted!
@@ -77,9 +77,6 @@ bool SDL_SendWindowEvent(SDL_Window *window, SDL_EventType windowevent, int data
    }
    SDL_assert(SDL_ObjectValid(window, SDL_OBJECT_TYPE_WINDOW));

-    if (window->is_destroying && windowevent != SDL_EVENT_WINDOW_DESTROYED) {
-        return false;
-    }
    switch (windowevent) {
    case SDL_EVENT_WINDOW_SHOWN:
        if (!(window->flags & SDL_WINDOW_HIDDEN)) {
@@ -212,6 +209,10 @@ bool SDL_SendWindowEvent(SDL_Window *window, SDL_EventType windowevent, int data
        break;
    }

+    if (window->is_destroying && windowevent != SDL_EVENT_WINDOW_DESTROYED) {
+        return false;
+    }
+
    // Post the event, if desired
    SDL_Event event;
    event.type = windowevent;
@@ -371,10 +371,10 @@ static const SDL_Scancode xfree86_scancode_table2[] = {
    /* 188, 0x0bc */   SDL_SCANCODE_F18,                // XF86Launch9
    /* 189, 0x0bd */   SDL_SCANCODE_F19,                // NoSymbol
    /* 190, 0x0be */   SDL_SCANCODE_F20,                // XF86AudioMicMute
-    /* 191, 0x0bf */   SDL_SCANCODE_UNKNOWN,            // XF86TouchpadToggle
-    /* 192, 0x0c0 */   SDL_SCANCODE_UNKNOWN,            // XF86TouchpadOn
-    /* 193, 0x0c1 */   SDL_SCANCODE_UNKNOWN,            // XF86TouchpadOff
-    /* 194, 0x0c2 */   SDL_SCANCODE_UNKNOWN,            // NoSymbol
+    /* 191, 0x0bf */   SDL_SCANCODE_F21,                // XF86TouchpadToggle
+    /* 192, 0x0c0 */   SDL_SCANCODE_F22,                // XF86TouchpadOn
+    /* 193, 0x0c1 */   SDL_SCANCODE_F23,                // XF86TouchpadOff
+    /* 194, 0x0c2 */   SDL_SCANCODE_F24,                // NoSymbol
    /* 195, 0x0c3 */   SDL_SCANCODE_MODE,               // Mode_switch
    /* 196, 0x0c4 */   SDL_SCANCODE_UNKNOWN,            // NoSymbol
    /* 197, 0x0c5 */   SDL_SCANCODE_UNKNOWN,            // NoSymbol
@@ -78,7 +78,7 @@ static char *search_path_for_binary(const char *bin)
    char *envr;
    size_t alloc_size;
    char *exe = NULL;
-    char *start = envr;
+    char *start;
    char *ptr;

    if (!envr_real) {
@@ -86,7 +86,7 @@ static char *search_path_for_binary(const char *bin)
        return NULL;
    }

-    envr = SDL_strdup(envr_real);
+    start = envr = SDL_strdup(envr_real);
    if (!envr) {
        return NULL;
    }
@@ -61,6 +61,10 @@
        return;                                              \
    }

+#if 0
+// The below validation is too aggressive, since there are advanced situations
+// where this is legal. This is being temporarily disabled for further review.
+// See: https://github.com/libsdl-org/SDL/issues/13871
 #define CHECK_SAMPLER_TEXTURES                                                                                                          \
    RenderPass *rp = (RenderPass *)render_pass;                                                                                         \
    for (Uint32 color_target_index = 0; color_target_index < rp->num_color_targets; color_target_index += 1) {                          \
@@ -92,11 +96,15 @@
            SDL_assert_release(!"Texture cannot be simultaneously bound as a depth stencil target and a storage texture!"); \
        }                                                                                                                   \
    }
+#else
+#define CHECK_SAMPLER_TEXTURES
+#define CHECK_STORAGE_TEXTURES
+#endif

-#define CHECK_GRAPHICS_PIPELINE_BOUND                                                       \
-    if (!((CommandBufferCommonHeader *)RENDERPASS_COMMAND_BUFFER)->graphics_pipeline_bound) { \
-        SDL_assert_release(!"Graphics pipeline not bound!");                                \
-        return;                                                                             \
+#define CHECK_GRAPHICS_PIPELINE_BOUND                                                   \
+    if (!((RenderPass *)render_pass)->graphics_pipeline) { \
+        SDL_assert_release(!"Graphics pipeline not bound!");                            \
+        return;                                                                         \
    }

 #define CHECK_COMPUTEPASS                                     \
@@ -106,7 +114,7 @@
    }

 #define CHECK_COMPUTE_PIPELINE_BOUND                                                        \
-    if (!((CommandBufferCommonHeader *)COMPUTEPASS_COMMAND_BUFFER)->compute_pipeline_bound) { \
+    if (!((ComputePass *)compute_pass)->compute_pipeline) { \
        SDL_assert_release(!"Compute pipeline not bound!");                                 \
        return;                                                                             \
    }
@@ -174,18 +182,132 @@
 #define RENDERPASS_DEVICE \
    ((CommandBufferCommonHeader *)RENDERPASS_COMMAND_BUFFER)->device

+#define RENDERPASS_BOUND_PIPELINE \
+    ((RenderPass *)render_pass)->graphics_pipeline
+
 #define COMPUTEPASS_COMMAND_BUFFER \
    ((Pass *)compute_pass)->command_buffer

 #define COMPUTEPASS_DEVICE \
    ((CommandBufferCommonHeader *)COMPUTEPASS_COMMAND_BUFFER)->device

+#define COMPUTEPASS_BOUND_PIPELINE \
+    ((ComputePass *)compute_pass)->compute_pipeline
+
 #define COPYPASS_COMMAND_BUFFER \
    ((Pass *)copy_pass)->command_buffer

 #define COPYPASS_DEVICE \
    ((CommandBufferCommonHeader *)COPYPASS_COMMAND_BUFFER)->device

+static bool TextureFormatIsComputeWritable[] = {
+    false, // INVALID
+    false, // A8_UNORM
+    true,  // R8_UNORM
+    true,  // R8G8_UNORM
+    true,  // R8G8B8A8_UNORM
+    true,  // R16_UNORM
+    true,  // R16G16_UNORM
+    true,  // R16G16B16A16_UNORM
+    true,  // R10G10B10A2_UNORM
+    false, // B5G6R5_UNORM
+    false, // B5G5R5A1_UNORM
+    false, // B4G4R4A4_UNORM
+    false, // B8G8R8A8_UNORM
+    false, // BC1_UNORM
+    false, // BC2_UNORM
+    false, // BC3_UNORM
+    false, // BC4_UNORM
+    false, // BC5_UNORM
+    false, // BC7_UNORM
+    false, // BC6H_FLOAT
+    false, // BC6H_UFLOAT
+    true,  // R8_SNORM
+    true,  // R8G8_SNORM
+    true,  // R8G8B8A8_SNORM
+    true,  // R16_SNORM
+    true,  // R16G16_SNORM
+    true,  // R16G16B16A16_SNORM
+    true,  // R16_FLOAT
+    true,  // R16G16_FLOAT
+    true,  // R16G16B16A16_FLOAT
+    true,  // R32_FLOAT
+    true,  // R32G32_FLOAT
+    true,  // R32G32B32A32_FLOAT
+    true,  // R11G11B10_UFLOAT
+    true,  // R8_UINT
+    true,  // R8G8_UINT
+    true,  // R8G8B8A8_UINT
+    true,  // R16_UINT
+    true,  // R16G16_UINT
+    true,  // R16G16B16A16_UINT
+    true,  // R32_UINT
+    true,  // R32G32_UINT
+    true,  // R32G32B32A32_UINT
+    true,  // R8_INT
+    true,  // R8G8_INT
+    true,  // R8G8B8A8_INT
+    true,  // R16_INT
+    true,  // R16G16_INT
+    true,  // R16G16B16A16_INT
+    true,  // R32_INT
+    true,  // R32G32_INT
+    true,  // R32G32B32A32_INT
+    false, // R8G8B8A8_UNORM_SRGB
+    false, // B8G8R8A8_UNORM_SRGB
+    false, // BC1_UNORM_SRGB
+    false, // BC3_UNORM_SRGB
+    false, // BC3_UNORM_SRGB
+    false, // BC7_UNORM_SRGB
+    false, // D16_UNORM
+    false, // D24_UNORM
+    false, // D32_FLOAT
+    false, // D24_UNORM_S8_UINT
+    false, // D32_FLOAT_S8_UINT
+    false, // ASTC_4x4_UNORM
+    false, // ASTC_5x4_UNORM
+    false, // ASTC_5x5_UNORM
+    false, // ASTC_6x5_UNORM
+    false, // ASTC_6x6_UNORM
+    false, // ASTC_8x5_UNORM
+    false, // ASTC_8x6_UNORM
+    false, // ASTC_8x8_UNORM
+    false, // ASTC_10x5_UNORM
+    false, // ASTC_10x6_UNORM
+    false, // ASTC_10x8_UNORM
+    false, // ASTC_10x10_UNORM
+    false, // ASTC_12x10_UNORM
+    false, // ASTC_12x12_UNORM
+    false, // ASTC_4x4_UNORM_SRGB
+    false, // ASTC_5x4_UNORM_SRGB
+    false, // ASTC_5x5_UNORM_SRGB
+    false, // ASTC_6x5_UNORM_SRGB
+    false, // ASTC_6x6_UNORM_SRGB
+    false, // ASTC_8x5_UNORM_SRGB
+    false, // ASTC_8x6_UNORM_SRGB
+    false, // ASTC_8x8_UNORM_SRGB
+    false, // ASTC_10x5_UNORM_SRGB
+    false, // ASTC_10x6_UNORM_SRGB
+    false, // ASTC_10x8_UNORM_SRGB
+    false, // ASTC_10x10_UNORM_SRGB
+    false, // ASTC_12x10_UNORM_SRGB
+    false, // ASTC_12x12_UNORM_SRGB
+    false, // ASTC_4x4_FLOAT
+    false, // ASTC_5x4_FLOAT
+    false, // ASTC_5x5_FLOAT
+    false, // ASTC_6x5_FLOAT
+    false, // ASTC_6x6_FLOAT
+    false, // ASTC_8x5_FLOAT
+    false, // ASTC_8x6_FLOAT
+    false, // ASTC_8x8_FLOAT
+    false, // ASTC_10x5_FLOAT
+    false, // ASTC_10x6_FLOAT
+    false, // ASTC_10x8_FLOAT
+    false, // ASTC_10x10_FLOAT
+    false, // ASTC_12x10_FLOAT
+    false  // ASTC_12x12_FLOAT
+};
+
 // Drivers

 #ifndef SDL_GPU_DISABLED
@@ -403,6 +525,73 @@ void SDL_GPU_BlitCommon(
    SDL_EndGPURenderPass(render_pass);
 }

+static void SDL_GPU_CheckGraphicsBindings(SDL_GPURenderPass *render_pass)
+{
+    RenderPass *rp = (RenderPass *)render_pass;
+    GraphicsPipelineCommonHeader *pipeline = (GraphicsPipelineCommonHeader *)RENDERPASS_BOUND_PIPELINE;
+    for (Uint32 i = 0; i < pipeline->num_vertex_samplers; i += 1) {
+        if (!rp->vertex_sampler_bound[i]) {
+            SDL_assert_release(!"Missing vertex sampler binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->num_vertex_storage_textures; i += 1) {
+        if (!rp->vertex_storage_texture_bound[i]) {
+            SDL_assert_release(!"Missing vertex storage texture binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->num_vertex_storage_buffers; i += 1) {
+        if (!rp->vertex_storage_buffer_bound[i]) {
+            SDL_assert_release(!"Missing vertex storage buffer binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->num_fragment_samplers; i += 1) {
+        if (!rp->fragment_sampler_bound[i]) {
+            SDL_assert_release(!"Missing fragment sampler binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->num_fragment_storage_textures; i += 1) {
+        if (!rp->fragment_storage_texture_bound[i]) {
+            SDL_assert_release(!"Missing fragment storage texture binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->num_fragment_storage_buffers; i += 1) {
+        if (!rp->fragment_storage_buffer_bound[i]) {
+            SDL_assert_release(!"Missing fragment storage buffer binding!");
+        }
+    }
+}
+
+static void SDL_GPU_CheckComputeBindings(SDL_GPUComputePass *compute_pass)
+{
+    ComputePass *cp = (ComputePass *)compute_pass;
+    ComputePipelineCommonHeader *pipeline = (ComputePipelineCommonHeader *)COMPUTEPASS_BOUND_PIPELINE;
+    for (Uint32 i = 0; i < pipeline->numSamplers; i += 1) {
+        if (!cp->sampler_bound[i]) {
+            SDL_assert_release(!"Missing compute sampler binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->numReadonlyStorageTextures; i += 1) {
+        if (!cp->read_only_storage_texture_bound[i]) {
+            SDL_assert_release(!"Missing compute readonly storage texture binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->numReadonlyStorageBuffers; i += 1) {
+        if (!cp->read_only_storage_buffer_bound[i]) {
+            SDL_assert_release(!"Missing compute readonly storage buffer binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->numReadWriteStorageTextures; i += 1) {
+        if (!cp->read_write_storage_texture_bound[i]) {
+            SDL_assert_release(!"Missing compute read-write storage texture binding!");
+        }
+    }
+    for (Uint32 i = 0; i < pipeline->numReadWriteStorageBuffers; i += 1) {
+        if (!cp->read_write_storage_buffer_bound[i]) {
+            SDL_assert_release(!"Missing compute read-write storage buffer bbinding!");
+        }
+    }
+}
+
 // Driver Functions

 #ifndef SDL_GPU_DISABLED
@@ -564,7 +753,6 @@ SDL_GPUDevice *SDL_CreateGPUDeviceWithProperties(SDL_PropertiesID props)
        result = selectedBackend->CreateDevice(debug_mode, preferLowPower, props);
        if (result != NULL) {
            result->backend = selectedBackend->name;
-            result->shader_formats = selectedBackend->shader_formats;
            result->debug_mode = debug_mode;
        }
    }
@@ -753,6 +941,13 @@ bool SDL_GPUTextureSupportsFormat(
        CHECK_TEXTUREFORMAT_ENUM_INVALID(format, false)
    }

+    if ((usage & SDL_GPU_TEXTUREUSAGE_COMPUTE_STORAGE_WRITE) ||
+        (usage & SDL_GPU_TEXTUREUSAGE_COMPUTE_STORAGE_SIMULTANEOUS_READ_WRITE)) {
+        if (!TextureFormatIsComputeWritable[format]) {
+            return false;
+        }
+    }
+
    return device->SupportsTextureFormat(
        device->driverData,
        format,
@@ -1125,6 +1320,12 @@ SDL_GPUBuffer *SDL_CreateGPUBuffer(
        return NULL;
    }

+    if (device->debug_mode) {
+        if (createinfo->size < 4) {
+            SDL_assert_release(!"Cannot create a buffer with size less than 4 bytes!");
+        }
+    }
+
    const char *debugName = SDL_GetStringProperty(createinfo->props, SDL_PROP_GPU_BUFFER_CREATE_NAME_STRING, NULL);

    return device->CreateBuffer(
@@ -1375,15 +1576,30 @@ SDL_GPUCommandBuffer *SDL_AcquireGPUCommandBuffer(
    commandBufferHeader = (CommandBufferCommonHeader *)command_buffer;
    commandBufferHeader->device = device;
    commandBufferHeader->render_pass.command_buffer = command_buffer;
-    commandBufferHeader->render_pass.in_progress = false;
-    commandBufferHeader->graphics_pipeline_bound = false;
    commandBufferHeader->compute_pass.command_buffer = command_buffer;
-    commandBufferHeader->compute_pass.in_progress = false;
-    commandBufferHeader->compute_pipeline_bound = false;
    commandBufferHeader->copy_pass.command_buffer = command_buffer;
-    commandBufferHeader->copy_pass.in_progress = false;
-    commandBufferHeader->swapchain_texture_acquired = false;
-    commandBufferHeader->submitted = false;
+
+    if (device->debug_mode) {
+        commandBufferHeader->render_pass.in_progress = false;
+        commandBufferHeader->render_pass.graphics_pipeline = NULL;
+        commandBufferHeader->compute_pass.in_progress = false;
+        commandBufferHeader->compute_pass.compute_pipeline = NULL;
+        commandBufferHeader->copy_pass.in_progress = false;
+        commandBufferHeader->swapchain_texture_acquired = false;
+        commandBufferHeader->submitted = false;
+        commandBufferHeader->ignore_render_pass_texture_validation = false;
+        SDL_zeroa(commandBufferHeader->render_pass.vertex_sampler_bound);
+        SDL_zeroa(commandBufferHeader->render_pass.vertex_storage_texture_bound);
+        SDL_zeroa(commandBufferHeader->render_pass.vertex_storage_buffer_bound);
+        SDL_zeroa(commandBufferHeader->render_pass.fragment_sampler_bound);
+        SDL_zeroa(commandBufferHeader->render_pass.fragment_storage_texture_bound);
+        SDL_zeroa(commandBufferHeader->render_pass.fragment_storage_buffer_bound);
+        SDL_zeroa(commandBufferHeader->compute_pass.sampler_bound);
+        SDL_zeroa(commandBufferHeader->compute_pass.read_only_storage_texture_bound);
+        SDL_zeroa(commandBufferHeader->compute_pass.read_only_storage_buffer_bound);
+        SDL_zeroa(commandBufferHeader->compute_pass.read_write_storage_texture_bound);
+        SDL_zeroa(commandBufferHeader->compute_pass.read_write_storage_buffer_bound);
+    }

    return command_buffer;
 }
@@ -1501,30 +1717,47 @@ SDL_GPURenderPass *SDL_BeginGPURenderPass(

            if (color_target_infos[i].cycle && color_target_infos[i].load_op == SDL_GPU_LOADOP_LOAD) {
                SDL_assert_release(!"Cannot cycle color target when load op is LOAD!");
+                return NULL;
            }

            if (color_target_infos[i].store_op == SDL_GPU_STOREOP_RESOLVE || color_target_infos[i].store_op == SDL_GPU_STOREOP_RESOLVE_AND_STORE) {
                if (color_target_infos[i].resolve_texture == NULL) {
                    SDL_assert_release(!"Store op is RESOLVE or RESOLVE_AND_STORE but resolve_texture is NULL!");
+                    return NULL;
                } else {
                    TextureCommonHeader *resolveTextureHeader = (TextureCommonHeader *)color_target_infos[i].resolve_texture;
                    if (textureHeader->info.sample_count == SDL_GPU_SAMPLECOUNT_1) {
                        SDL_assert_release(!"Store op is RESOLVE or RESOLVE_AND_STORE but texture is not multisample!");
+                        return NULL;
                    }
                    if (resolveTextureHeader->info.sample_count != SDL_GPU_SAMPLECOUNT_1) {
                        SDL_assert_release(!"Resolve texture must have a sample count of 1!");
+                        return NULL;
                    }
                    if (resolveTextureHeader->info.format != textureHeader->info.format) {
                        SDL_assert_release(!"Resolve texture must have the same format as its corresponding color target!");
+                        return NULL;
                    }
                    if (resolveTextureHeader->info.type == SDL_GPU_TEXTURETYPE_3D) {
                        SDL_assert_release(!"Resolve texture must not be of TEXTURETYPE_3D!");
+                        return NULL;
                    }
                    if (!(resolveTextureHeader->info.usage & SDL_GPU_TEXTUREUSAGE_COLOR_TARGET)) {
                        SDL_assert_release(!"Resolve texture usage must include COLOR_TARGET!");
+                        return NULL;
                    }
                }
            }
+
+            if (color_target_infos[i].layer_or_depth_plane >= textureHeader->info.layer_count_or_depth) {
+                SDL_assert_release(!"Color target layer index must be less than the texture's layer count!");
+                return NULL;
+            }
+
+            if (color_target_infos[i].mip_level >= textureHeader->info.num_levels) {
+                SDL_assert_release(!"Color target mip level must be less than the texture's level count!");
+                return NULL;
+            }
        }

        if (depth_stencil_target_info != NULL) {
@@ -1532,10 +1765,12 @@ SDL_GPURenderPass *SDL_BeginGPURenderPass(
            TextureCommonHeader *textureHeader = (TextureCommonHeader *)depth_stencil_target_info->texture;
            if (!(textureHeader->info.usage & SDL_GPU_TEXTUREUSAGE_DEPTH_STENCIL_TARGET)) {
                SDL_assert_release(!"Depth target must have been created with the DEPTH_STENCIL_TARGET usage flag!");
+                return NULL;
            }

            if (depth_stencil_target_info->cycle && (depth_stencil_target_info->load_op == SDL_GPU_LOADOP_LOAD || depth_stencil_target_info->stencil_load_op == SDL_GPU_LOADOP_LOAD)) {
                SDL_assert_release(!"Cannot cycle depth target when load op or stencil load op is LOAD!");
+                return NULL;
            }

            if (depth_stencil_target_info->store_op == SDL_GPU_STOREOP_RESOLVE ||
@@ -1543,6 +1778,7 @@ SDL_GPURenderPass *SDL_BeginGPURenderPass(
                depth_stencil_target_info->store_op == SDL_GPU_STOREOP_RESOLVE_AND_STORE ||
                depth_stencil_target_info->stencil_store_op == SDL_GPU_STOREOP_RESOLVE_AND_STORE) {
                SDL_assert_release(!"RESOLVE store ops are not supported for depth-stencil targets!");
+                return NULL;
            }
        }
    }
@@ -1554,14 +1790,20 @@ SDL_GPURenderPass *SDL_BeginGPURenderPass(
        depth_stencil_target_info);

    commandBufferHeader = (CommandBufferCommonHeader *)command_buffer;
-    commandBufferHeader->render_pass.in_progress = true;
-    for (Uint32 i = 0; i < num_color_targets; i += 1) {
-        commandBufferHeader->render_pass.color_targets[i] = color_target_infos[i].texture;
-    }
-    commandBufferHeader->render_pass.num_color_targets = num_color_targets;
-    if (depth_stencil_target_info != NULL) {
-        commandBufferHeader->render_pass.depth_stencil_target = depth_stencil_target_info->texture;
+
+    if (COMMAND_BUFFER_DEVICE->debug_mode) {
+        commandBufferHeader->render_pass.in_progress = true;
+        for (Uint32 i = 0; i < num_color_targets; i += 1) {
+            commandBufferHeader->render_pass.color_targets[i] = color_target_infos[i].texture;
+        }
+        commandBufferHeader->render_pass.num_color_targets = num_color_targets;
+        if (depth_stencil_target_info != NULL) {
+            commandBufferHeader->render_pass.depth_stencil_target = depth_stencil_target_info->texture;
+        } else {
+            commandBufferHeader->render_pass.depth_stencil_target = NULL;
+        }
    }
+
    return (SDL_GPURenderPass *)&(commandBufferHeader->render_pass);
 }

@@ -1569,8 +1811,6 @@ void SDL_BindGPUGraphicsPipeline(
    SDL_GPURenderPass *render_pass,
    SDL_GPUGraphicsPipeline *graphics_pipeline)
 {
-    CommandBufferCommonHeader *commandBufferHeader;
-
    if (render_pass == NULL) {
        SDL_InvalidParamError("render_pass");
        return;
@@ -1584,8 +1824,10 @@ void SDL_BindGPUGraphicsPipeline(
        RENDERPASS_COMMAND_BUFFER,
        graphics_pipeline);

-    commandBufferHeader = (CommandBufferCommonHeader *)RENDERPASS_COMMAND_BUFFER;
-    commandBufferHeader->graphics_pipeline_bound = true;
+
+    if (RENDERPASS_DEVICE->debug_mode) {
+        RENDERPASS_BOUND_PIPELINE = graphics_pipeline;
+    }
 }

 void SDL_SetGPUViewport(
@@ -1735,7 +1977,15 @@ void SDL_BindGPUVertexSamplers(

    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
-        CHECK_SAMPLER_TEXTURES
+
+        if (!((CommandBufferCommonHeader*)RENDERPASS_COMMAND_BUFFER)->ignore_render_pass_texture_validation)
+        {
+            CHECK_SAMPLER_TEXTURES
+        }
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((RenderPass *)render_pass)->vertex_sampler_bound[first_slot + i] = true;
+        }
    }

    RENDERPASS_DEVICE->BindVertexSamplers(
@@ -1763,6 +2013,10 @@ void SDL_BindGPUVertexStorageTextures(
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
        CHECK_STORAGE_TEXTURES
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((RenderPass *)render_pass)->vertex_storage_texture_bound[first_slot + i] = true;
+        }
    }

    RENDERPASS_DEVICE->BindVertexStorageTextures(
@@ -1789,6 +2043,10 @@ void SDL_BindGPUVertexStorageBuffers(

    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((RenderPass *)render_pass)->vertex_storage_buffer_bound[first_slot + i] = true;
+        }
    }

    RENDERPASS_DEVICE->BindVertexStorageBuffers(
@@ -1815,7 +2073,14 @@ void SDL_BindGPUFragmentSamplers(

    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
-        CHECK_SAMPLER_TEXTURES
+
+        if (!((CommandBufferCommonHeader*)RENDERPASS_COMMAND_BUFFER)->ignore_render_pass_texture_validation) {
+            CHECK_SAMPLER_TEXTURES
+        }
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((RenderPass *)render_pass)->fragment_sampler_bound[first_slot + i] = true;
+        }
    }

    RENDERPASS_DEVICE->BindFragmentSamplers(
@@ -1843,6 +2108,10 @@ void SDL_BindGPUFragmentStorageTextures(
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
        CHECK_STORAGE_TEXTURES
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((RenderPass *)render_pass)->fragment_storage_texture_bound[first_slot + i] = true;
+        }
    }

    RENDERPASS_DEVICE->BindFragmentStorageTextures(
@@ -1869,6 +2138,10 @@ void SDL_BindGPUFragmentStorageBuffers(

    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((RenderPass *)render_pass)->fragment_storage_buffer_bound[first_slot + i] = true;
+        }
    }

    RENDERPASS_DEVICE->BindFragmentStorageBuffers(
@@ -1894,6 +2167,7 @@ void SDL_DrawGPUIndexedPrimitives(
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
        CHECK_GRAPHICS_PIPELINE_BOUND
+        SDL_GPU_CheckGraphicsBindings(render_pass);
    }

    RENDERPASS_DEVICE->DrawIndexedPrimitives(
@@ -1920,6 +2194,7 @@ void SDL_DrawGPUPrimitives(
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
        CHECK_GRAPHICS_PIPELINE_BOUND
+        SDL_GPU_CheckGraphicsBindings(render_pass);
    }

    RENDERPASS_DEVICE->DrawPrimitives(
@@ -1948,6 +2223,7 @@ void SDL_DrawGPUPrimitivesIndirect(
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
        CHECK_GRAPHICS_PIPELINE_BOUND
+        SDL_GPU_CheckGraphicsBindings(render_pass);
    }

    RENDERPASS_DEVICE->DrawPrimitivesIndirect(
@@ -1975,6 +2251,7 @@ void SDL_DrawGPUIndexedPrimitivesIndirect(
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
        CHECK_GRAPHICS_PIPELINE_BOUND
+        SDL_GPU_CheckGraphicsBindings(render_pass);
    }

    RENDERPASS_DEVICE->DrawIndexedPrimitivesIndirect(
@@ -1987,13 +2264,14 @@ void SDL_DrawGPUIndexedPrimitivesIndirect(
 void SDL_EndGPURenderPass(
    SDL_GPURenderPass *render_pass)
 {
-    CommandBufferCommonHeader *commandBufferCommonHeader;
-
    if (render_pass == NULL) {
        SDL_InvalidParamError("render_pass");
        return;
    }

+    CommandBufferCommonHeader *commandBufferCommonHeader;
+    commandBufferCommonHeader = (CommandBufferCommonHeader *)RENDERPASS_COMMAND_BUFFER;
+
    if (RENDERPASS_DEVICE->debug_mode) {
        CHECK_RENDERPASS
    }
@@ -2001,15 +2279,22 @@ void SDL_EndGPURenderPass(
    RENDERPASS_DEVICE->EndRenderPass(
        RENDERPASS_COMMAND_BUFFER);

-    commandBufferCommonHeader = (CommandBufferCommonHeader *)RENDERPASS_COMMAND_BUFFER;
-    commandBufferCommonHeader->render_pass.in_progress = false;
-    for (Uint32 i = 0; i < MAX_COLOR_TARGET_BINDINGS; i += 1)
-    {
-        commandBufferCommonHeader->render_pass.color_targets[i] = NULL;
+    if (RENDERPASS_DEVICE->debug_mode) {
+        commandBufferCommonHeader->render_pass.in_progress = false;
+        for (Uint32 i = 0; i < MAX_COLOR_TARGET_BINDINGS; i += 1)
+        {
+            commandBufferCommonHeader->render_pass.color_targets[i] = NULL;
+        }
+        commandBufferCommonHeader->render_pass.num_color_targets = 0;
+        commandBufferCommonHeader->render_pass.depth_stencil_target = NULL;
+        commandBufferCommonHeader->render_pass.graphics_pipeline = NULL;
+        SDL_zeroa(commandBufferCommonHeader->render_pass.vertex_sampler_bound);
+        SDL_zeroa(commandBufferCommonHeader->render_pass.vertex_storage_texture_bound);
+        SDL_zeroa(commandBufferCommonHeader->render_pass.vertex_storage_buffer_bound);
+        SDL_zeroa(commandBufferCommonHeader->render_pass.fragment_sampler_bound);
+        SDL_zeroa(commandBufferCommonHeader->render_pass.fragment_storage_texture_bound);
+        SDL_zeroa(commandBufferCommonHeader->render_pass.fragment_storage_buffer_bound);
    }
-    commandBufferCommonHeader->render_pass.num_color_targets = 0;
-    commandBufferCommonHeader->render_pass.depth_stencil_target = NULL;
-    commandBufferCommonHeader->graphics_pipeline_bound = false;
 }

 // Compute Pass
@@ -2053,6 +2338,16 @@ SDL_GPUComputePass *SDL_BeginGPUComputePass(
                SDL_assert_release(!"Texture must be created with COMPUTE_STORAGE_WRITE or COMPUTE_STORAGE_SIMULTANEOUS_READ_WRITE flag");
                return NULL;
            }
+
+            if (storage_texture_bindings[i].layer >= header->info.layer_count_or_depth) {
+                SDL_assert_release(!"Storage texture layer index must be less than the texture's layer count!");
+                return NULL;
+            }
+
+            if (storage_texture_bindings[i].mip_level >= header->info.num_levels) {
+                SDL_assert_release(!"Storage texture mip level must be less than the texture's level count!");
+                return NULL;
+            }
        }

        // TODO: validate buffer usage?
@@ -2066,7 +2361,19 @@ SDL_GPUComputePass *SDL_BeginGPUComputePass(
        num_storage_buffer_bindings);

    commandBufferHeader = (CommandBufferCommonHeader *)command_buffer;
-    commandBufferHeader->compute_pass.in_progress = true;
+
+    if (COMMAND_BUFFER_DEVICE->debug_mode) {
+        commandBufferHeader->compute_pass.in_progress = true;
+
+        for (Uint32 i = 0; i < num_storage_texture_bindings; i += 1) {
+            commandBufferHeader->compute_pass.read_write_storage_texture_bound[i] = true;
+        }
+
+        for (Uint32 i = 0; i < num_storage_buffer_bindings; i += 1) {
+            commandBufferHeader->compute_pass.read_write_storage_buffer_bound[i] = true;
+        }
+    }
+
    return (SDL_GPUComputePass *)&(commandBufferHeader->compute_pass);
 }

@@ -2074,8 +2381,6 @@ void SDL_BindGPUComputePipeline(
    SDL_GPUComputePass *compute_pass,
    SDL_GPUComputePipeline *compute_pipeline)
 {
-    CommandBufferCommonHeader *commandBufferHeader;
-
    if (compute_pass == NULL) {
        SDL_InvalidParamError("compute_pass");
        return;
@@ -2093,8 +2398,10 @@ void SDL_BindGPUComputePipeline(
        COMPUTEPASS_COMMAND_BUFFER,
        compute_pipeline);

-    commandBufferHeader = (CommandBufferCommonHeader *)COMPUTEPASS_COMMAND_BUFFER;
-    commandBufferHeader->compute_pipeline_bound = true;
+
+    if (COMPUTEPASS_DEVICE->debug_mode) {
+        COMPUTEPASS_BOUND_PIPELINE = compute_pipeline;
+    }
 }

 void SDL_BindGPUComputeSamplers(
@@ -2114,6 +2421,10 @@ void SDL_BindGPUComputeSamplers(

    if (COMPUTEPASS_DEVICE->debug_mode) {
        CHECK_COMPUTEPASS
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((ComputePass *)compute_pass)->sampler_bound[first_slot + i] = true;
+        }
    }

    COMPUTEPASS_DEVICE->BindComputeSamplers(
@@ -2140,6 +2451,10 @@ void SDL_BindGPUComputeStorageTextures(

    if (COMPUTEPASS_DEVICE->debug_mode) {
        CHECK_COMPUTEPASS
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((ComputePass *)compute_pass)->read_only_storage_texture_bound[first_slot + i] = true;
+        }
    }

    COMPUTEPASS_DEVICE->BindComputeStorageTextures(
@@ -2166,6 +2481,10 @@ void SDL_BindGPUComputeStorageBuffers(

    if (COMPUTEPASS_DEVICE->debug_mode) {
        CHECK_COMPUTEPASS
+
+        for (Uint32 i = 0; i < num_bindings; i += 1) {
+            ((ComputePass *)compute_pass)->read_only_storage_buffer_bound[first_slot + i] = true;
+        }
    }

    COMPUTEPASS_DEVICE->BindComputeStorageBuffers(
@@ -2189,6 +2508,7 @@ void SDL_DispatchGPUCompute(
    if (COMPUTEPASS_DEVICE->debug_mode) {
        CHECK_COMPUTEPASS
        CHECK_COMPUTE_PIPELINE_BOUND
+        SDL_GPU_CheckComputeBindings(compute_pass);
    }

    COMPUTEPASS_DEVICE->DispatchCompute(
@@ -2211,6 +2531,7 @@ void SDL_DispatchGPUComputeIndirect(
    if (COMPUTEPASS_DEVICE->debug_mode) {
        CHECK_COMPUTEPASS
        CHECK_COMPUTE_PIPELINE_BOUND
+        SDL_GPU_CheckComputeBindings(compute_pass);
    }

    COMPUTEPASS_DEVICE->DispatchComputeIndirect(
@@ -2236,9 +2557,16 @@ void SDL_EndGPUComputePass(
    COMPUTEPASS_DEVICE->EndComputePass(
        COMPUTEPASS_COMMAND_BUFFER);

-    commandBufferCommonHeader = (CommandBufferCommonHeader *)COMPUTEPASS_COMMAND_BUFFER;
-    commandBufferCommonHeader->compute_pass.in_progress = false;
-    commandBufferCommonHeader->compute_pipeline_bound = false;
+    if (COMPUTEPASS_DEVICE->debug_mode) {
+        commandBufferCommonHeader = (CommandBufferCommonHeader *)COMPUTEPASS_COMMAND_BUFFER;
+        commandBufferCommonHeader->compute_pass.in_progress = false;
+        commandBufferCommonHeader->compute_pass.compute_pipeline = NULL;
+        SDL_zeroa(commandBufferCommonHeader->compute_pass.sampler_bound);
+        SDL_zeroa(commandBufferCommonHeader->compute_pass.read_only_storage_texture_bound);
+        SDL_zeroa(commandBufferCommonHeader->compute_pass.read_only_storage_buffer_bound);
+        SDL_zeroa(commandBufferCommonHeader->compute_pass.read_write_storage_texture_bound);
+        SDL_zeroa(commandBufferCommonHeader->compute_pass.read_write_storage_buffer_bound);
+    }
 }

 // TransferBuffer Data
@@ -2296,7 +2624,11 @@ SDL_GPUCopyPass *SDL_BeginGPUCopyPass(
        command_buffer);

    commandBufferHeader = (CommandBufferCommonHeader *)command_buffer;
-    commandBufferHeader->copy_pass.in_progress = true;
+
+    if (COMMAND_BUFFER_DEVICE->debug_mode) {
+        commandBufferHeader->copy_pass.in_progress = true;
+    }
+
    return (SDL_GPUCopyPass *)&(commandBufferHeader->copy_pass);
 }

@@ -2547,7 +2879,9 @@ void SDL_EndGPUCopyPass(
    COPYPASS_DEVICE->EndCopyPass(
        COPYPASS_COMMAND_BUFFER);

-    ((CommandBufferCommonHeader *)COPYPASS_COMMAND_BUFFER)->copy_pass.in_progress = false;
+    if (COPYPASS_DEVICE->debug_mode) {
+        ((CommandBufferCommonHeader *)COPYPASS_COMMAND_BUFFER)->copy_pass.in_progress = false;
+    }
 }

 void SDL_GenerateMipmapsForGPUTexture(
@@ -2577,11 +2911,19 @@ void SDL_GenerateMipmapsForGPUTexture(
            SDL_assert_release(!"GenerateMipmaps texture must be created with SAMPLER and COLOR_TARGET usage flags!");
            return;
        }
+
+        CommandBufferCommonHeader *commandBufferHeader = (CommandBufferCommonHeader *)command_buffer;
+        commandBufferHeader->ignore_render_pass_texture_validation = true;
    }

    COMMAND_BUFFER_DEVICE->GenerateMipmaps(
        command_buffer,
        texture);
+
+    if (COMMAND_BUFFER_DEVICE->debug_mode) {
+        CommandBufferCommonHeader *commandBufferHeader = (CommandBufferCommonHeader *)command_buffer;
+        commandBufferHeader->ignore_render_pass_texture_validation = false;
+    }
 }

 void SDL_BlitGPUTexture(
@@ -35,7 +35,7 @@
 #define UNIFORM_BUFFER_SIZE            32768
 #define MAX_VERTEX_BUFFERS             16
 #define MAX_VERTEX_ATTRIBUTES          16
-#define MAX_COLOR_TARGET_BINDINGS      4
+#define MAX_COLOR_TARGET_BINDINGS      8
 #define MAX_PRESENT_COUNT              16
 #define MAX_FRAMES_IN_FLIGHT           3

@@ -47,6 +47,20 @@ typedef struct Pass
    bool in_progress;
 } Pass;

+typedef struct ComputePass
+{
+    SDL_GPUCommandBuffer *command_buffer;
+    bool in_progress;
+
+    SDL_GPUComputePipeline *compute_pipeline;
+
+    bool sampler_bound[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    bool read_only_storage_texture_bound[MAX_STORAGE_TEXTURES_PER_STAGE];
+    bool read_only_storage_buffer_bound[MAX_STORAGE_BUFFERS_PER_STAGE];
+    bool read_write_storage_texture_bound[MAX_COMPUTE_WRITE_TEXTURES];
+    bool read_write_storage_buffer_bound[MAX_COMPUTE_WRITE_BUFFERS];
+} ComputePass;
+
 typedef struct RenderPass
 {
    SDL_GPUCommandBuffer *command_buffer;
@@ -54,18 +68,30 @@ typedef struct RenderPass
    SDL_GPUTexture *color_targets[MAX_COLOR_TARGET_BINDINGS];
    Uint32 num_color_targets;
    SDL_GPUTexture *depth_stencil_target;
+
+    SDL_GPUGraphicsPipeline *graphics_pipeline;
+
+    bool vertex_sampler_bound[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    bool vertex_storage_texture_bound[MAX_STORAGE_TEXTURES_PER_STAGE];
+    bool vertex_storage_buffer_bound[MAX_STORAGE_BUFFERS_PER_STAGE];
+
+    bool fragment_sampler_bound[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    bool fragment_storage_texture_bound[MAX_STORAGE_TEXTURES_PER_STAGE];
+    bool fragment_storage_buffer_bound[MAX_STORAGE_BUFFERS_PER_STAGE];
 } RenderPass;

 typedef struct CommandBufferCommonHeader
 {
    SDL_GPUDevice *device;
+
    RenderPass render_pass;
-    bool graphics_pipeline_bound;
-    Pass compute_pass;
-    bool compute_pipeline_bound;
+    ComputePass compute_pass;
+
    Pass copy_pass;
    bool swapchain_texture_acquired;
    bool submitted;
+    // used to avoid tripping assert on GenerateMipmaps
+    bool ignore_render_pass_texture_validation;
 } CommandBufferCommonHeader;

 typedef struct TextureCommonHeader
@@ -73,6 +99,29 @@ typedef struct TextureCommonHeader
    SDL_GPUTextureCreateInfo info;
 } TextureCommonHeader;

+typedef struct GraphicsPipelineCommonHeader
+{
+    Uint32 num_vertex_samplers;
+    Uint32 num_vertex_storage_textures;
+    Uint32 num_vertex_storage_buffers;
+    Uint32 num_vertex_uniform_buffers;
+
+    Uint32 num_fragment_samplers;
+    Uint32 num_fragment_storage_textures;
+    Uint32 num_fragment_storage_buffers;
+    Uint32 num_fragment_uniform_buffers;
+} GraphicsPipelineCommonHeader;
+
+typedef struct ComputePipelineCommonHeader
+{
+    Uint32 numSamplers;
+    Uint32 numReadonlyStorageTextures;
+    Uint32 numReadonlyStorageBuffers;
+    Uint32 numReadWriteStorageTextures;
+    Uint32 numReadWriteStorageBuffers;
+    Uint32 numUniformBuffers;
+} ComputePipelineCommonHeader;
+
 typedef struct BlitFragmentUniforms
 {
    // texcoord space
@@ -160,6 +209,7 @@ static inline Sint32 Texture_GetBlockWidth(
    case SDL_GPU_TEXTUREFORMAT_BC6H_RGB_FLOAT:
    case SDL_GPU_TEXTUREFORMAT_BC6H_RGB_UFLOAT:
    case SDL_GPU_TEXTUREFORMAT_BC1_RGBA_UNORM_SRGB:
+    case SDL_GPU_TEXTUREFORMAT_BC2_RGBA_UNORM_SRGB:
    case SDL_GPU_TEXTUREFORMAT_BC3_RGBA_UNORM_SRGB:
    case SDL_GPU_TEXTUREFORMAT_BC7_RGBA_UNORM_SRGB:
    case SDL_GPU_TEXTUREFORMAT_ASTC_4x4_UNORM:
@@ -277,6 +327,7 @@ static inline Sint32 Texture_GetBlockHeight(
    case SDL_GPU_TEXTUREFORMAT_BC6H_RGB_FLOAT:
    case SDL_GPU_TEXTUREFORMAT_BC6H_RGB_UFLOAT:
    case SDL_GPU_TEXTUREFORMAT_BC1_RGBA_UNORM_SRGB:
+    case SDL_GPU_TEXTUREFORMAT_BC2_RGBA_UNORM_SRGB:
    case SDL_GPU_TEXTUREFORMAT_BC3_RGBA_UNORM_SRGB:
    case SDL_GPU_TEXTUREFORMAT_BC7_RGBA_UNORM_SRGB:
    case SDL_GPU_TEXTUREFORMAT_ASTC_5x4_UNORM:
@@ -109,8 +109,8 @@
 #define DXGI_GET_DEBUG_INTERFACE_FUNC       "DXGIGetDebugInterface"
 #define D3D12_GET_DEBUG_INTERFACE_FUNC      "D3D12GetDebugInterface"
 #define WINDOW_PROPERTY_DATA                "SDL_GPUD3D12WindowPropertyData"
-#define D3D_FEATURE_LEVEL_CHOICE            D3D_FEATURE_LEVEL_11_1
-#define D3D_FEATURE_LEVEL_CHOICE_STR        "11_1"
+#define D3D_FEATURE_LEVEL_CHOICE            D3D_FEATURE_LEVEL_11_0
+#define D3D_FEATURE_LEVEL_CHOICE_STR        "11_0"
 #define MAX_ROOT_SIGNATURE_PARAMETERS         64
 #define D3D12_FENCE_UNSIGNALED_VALUE          0
 #define D3D12_FENCE_SIGNAL_VALUE              1
@@ -876,7 +876,7 @@ struct D3D12Renderer
    SDL_SharedObject *d3d12_dll;
    ID3D12Device *device;
    PFN_D3D12_SERIALIZE_ROOT_SIGNATURE D3D12SerializeRootSignature_func;
-    const char *semantic;
+    char *semantic;
    SDL_iconv_t iconv;

    ID3D12CommandQueue *commandQueue;
@@ -991,6 +991,10 @@ struct D3D12CommandBuffer
    // Set at acquire time
    D3D12DescriptorHeap *gpuDescriptorHeaps[D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER + 1];

+    D3D12DescriptorHeap **usedDescriptorHeaps;
+    Uint32 usedDescriptorHeapCount;
+    Uint32 usedDescriptorHeapCapacity;
+
    D3D12UniformBuffer **usedUniformBuffers;
    Uint32 usedUniformBufferCount;
    Uint32 usedUniformBufferCapacity;
@@ -1015,26 +1019,38 @@ struct D3D12CommandBuffer
    Uint32 vertexBufferOffsets[MAX_VERTEX_BUFFERS];
    Uint32 vertexBufferCount;

-    D3D12Texture *vertexSamplerTextures[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    D3D12Sampler *vertexSamplers[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    D3D12Texture *vertexStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
-    D3D12Buffer *vertexStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE vertexSamplerTextureDescriptorHandles[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE vertexSamplerDescriptorHandles[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE vertexStorageTextureDescriptorHandles[MAX_STORAGE_TEXTURES_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE vertexStorageBufferDescriptorHandles[MAX_STORAGE_BUFFERS_PER_STAGE];
+
    D3D12UniformBuffer *vertexUniformBuffers[MAX_UNIFORM_BUFFERS_PER_STAGE];

-    D3D12Texture *fragmentSamplerTextures[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    D3D12Sampler *fragmentSamplers[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    D3D12Texture *fragmentStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
-    D3D12Buffer *fragmentStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE fragmentSamplerTextureDescriptorHandles[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE fragmentSamplerDescriptorHandles[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE fragmentStorageTextureDescriptorHandles[MAX_STORAGE_TEXTURES_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE fragmentStorageBufferDescriptorHandles[MAX_STORAGE_BUFFERS_PER_STAGE];
+
    D3D12UniformBuffer *fragmentUniformBuffers[MAX_UNIFORM_BUFFERS_PER_STAGE];

-    D3D12Texture *computeSamplerTextures[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    D3D12Sampler *computeSamplers[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE computeSamplerTextureDescriptorHandles[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE computeSamplerDescriptorHandles[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE computeReadOnlyStorageTextureDescriptorHandles[MAX_STORAGE_TEXTURES_PER_STAGE];
+    D3D12_CPU_DESCRIPTOR_HANDLE computeReadOnlyStorageBufferDescriptorHandles[MAX_STORAGE_BUFFERS_PER_STAGE];
+
+    // Track these separately because barriers can happen mid compute pass
    D3D12Texture *computeReadOnlyStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
    D3D12Buffer *computeReadOnlyStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
+
+    D3D12_CPU_DESCRIPTOR_HANDLE computeReadWriteStorageTextureDescriptorHandles[MAX_COMPUTE_WRITE_TEXTURES];
+    D3D12_CPU_DESCRIPTOR_HANDLE computeReadWriteStorageBufferDescriptorHandles[MAX_COMPUTE_WRITE_BUFFERS];
+
+    // Track these separately because they are bound when the compute pass begins
    D3D12TextureSubresource *computeReadWriteStorageTextureSubresources[MAX_COMPUTE_WRITE_TEXTURES];
    Uint32 computeReadWriteStorageTextureSubresourceCount;
    D3D12Buffer *computeReadWriteStorageBuffers[MAX_COMPUTE_WRITE_BUFFERS];
    Uint32 computeReadWriteStorageBufferCount;
+
    D3D12UniformBuffer *computeUniformBuffers[MAX_UNIFORM_BUFFERS_PER_STAGE];

    // Resource tracking
@@ -1098,22 +1114,14 @@ typedef struct D3D12GraphicsRootSignature

 struct D3D12GraphicsPipeline
 {
+    GraphicsPipelineCommonHeader header;
+
    ID3D12PipelineState *pipelineState;
    D3D12GraphicsRootSignature *rootSignature;
    SDL_GPUPrimitiveType primitiveType;

    Uint32 vertexStrides[MAX_VERTEX_BUFFERS];

-    Uint32 vertexSamplerCount;
-    Uint32 vertexUniformBufferCount;
-    Uint32 vertexStorageBufferCount;
-    Uint32 vertexStorageTextureCount;
-
-    Uint32 fragmentSamplerCount;
-    Uint32 fragmentUniformBufferCount;
-    Uint32 fragmentStorageBufferCount;
-    Uint32 fragmentStorageTextureCount;
-
    SDL_AtomicInt referenceCount;
 };

@@ -1132,16 +1140,11 @@ typedef struct D3D12ComputeRootSignature

 struct D3D12ComputePipeline
 {
+    ComputePipelineCommonHeader header;
+
    ID3D12PipelineState *pipelineState;
    D3D12ComputeRootSignature *rootSignature;

-    Uint32 numSamplers;
-    Uint32 numReadOnlyStorageTextures;
-    Uint32 numReadOnlyStorageBuffers;
-    Uint32 numReadWriteStorageTextures;
-    Uint32 numReadWriteStorageBuffers;
-    Uint32 numUniformBuffers;
-
    SDL_AtomicInt referenceCount;
 };

@@ -1195,7 +1198,6 @@ struct D3D12UniformBuffer
    D3D12Buffer *buffer;
    Uint32 writeOffset;
    Uint32 drawOffset;
-    Uint32 currentBlockSize;
 };

 // Forward function declarations
@@ -1442,6 +1444,8 @@ static void D3D12_INTERNAL_ReleaseTextureContainer(
            container->textures[i]);
    }

+    SDL_DestroyProperties(container->header.info.props);
+
    // Containers are just client handles, so we can destroy immediately
    if (container->debugName) {
        SDL_free(container->debugName);
@@ -1588,6 +1592,7 @@ static void D3D12_INTERNAL_DestroyCommandBuffer(D3D12CommandBuffer *commandBuffe
    SDL_free(commandBuffer->usedSamplers);
    SDL_free(commandBuffer->usedGraphicsPipelines);
    SDL_free(commandBuffer->usedComputePipelines);
+    SDL_free(commandBuffer->usedDescriptorHeaps);
    SDL_free(commandBuffer->usedUniformBuffers);
    SDL_free(commandBuffer->textureDownloads);
    SDL_free(commandBuffer);
@@ -1734,6 +1739,7 @@ static void D3D12_INTERNAL_DestroyRenderer(D3D12Renderer *renderer)
    SDL_DestroyMutex(renderer->windowLock);
    SDL_DestroyMutex(renderer->fenceLock);
    SDL_DestroyMutex(renderer->disposeLock);
+    SDL_free(renderer->semantic);
    SDL_free(renderer);
 }

@@ -2284,6 +2290,28 @@ static bool D3D12_INTERNAL_ExpandStagingDescriptorPool(
    return true;
 }

+static void D3D12_INTERNAL_TrackGPUDescriptorHeap(
+    D3D12CommandBuffer *commandBuffer,
+    D3D12DescriptorHeap *descriptorHeap)
+{
+    Uint32 i;
+    for (i = 0; i < commandBuffer->usedDescriptorHeapCount; i += 1) {
+        if (commandBuffer->usedDescriptorHeaps[i] == descriptorHeap) {
+            return;
+        }
+    }
+
+    if (commandBuffer->usedDescriptorHeapCount == commandBuffer->usedDescriptorHeapCapacity) {
+        commandBuffer->usedDescriptorHeapCapacity += 1;
+        commandBuffer->usedDescriptorHeaps = (D3D12DescriptorHeap **)SDL_realloc(
+            commandBuffer->usedDescriptorHeaps,
+            commandBuffer->usedDescriptorHeapCapacity * sizeof(D3D12DescriptorHeap *));
+    }
+
+    commandBuffer->usedDescriptorHeaps[commandBuffer->usedDescriptorHeapCount] = descriptorHeap;
+    commandBuffer->usedDescriptorHeapCount += 1;
+}
+
 static D3D12DescriptorHeap *D3D12_INTERNAL_AcquireGPUDescriptorHeapFromPool(
    D3D12CommandBuffer *commandBuffer,
    D3D12_DESCRIPTOR_HEAP_TYPE descriptorHeapType)
@@ -2305,6 +2333,7 @@ static D3D12DescriptorHeap *D3D12_INTERNAL_AcquireGPUDescriptorHeapFromPool(
    }
    SDL_UnlockMutex(pool->lock);

+    D3D12_INTERNAL_TrackGPUDescriptorHeap(commandBuffer, result);
    return result;
 }

@@ -2886,12 +2915,12 @@ static SDL_GPUComputePipeline *D3D12_CreateComputePipeline(

    computePipeline->pipelineState = pipelineState;
    computePipeline->rootSignature = rootSignature;
-    computePipeline->numSamplers = createinfo->num_samplers;
-    computePipeline->numReadOnlyStorageTextures = createinfo->num_readonly_storage_textures;
-    computePipeline->numReadOnlyStorageBuffers = createinfo->num_readonly_storage_buffers;
-    computePipeline->numReadWriteStorageTextures = createinfo->num_readwrite_storage_textures;
-    computePipeline->numReadWriteStorageBuffers = createinfo->num_readwrite_storage_buffers;
-    computePipeline->numUniformBuffers = createinfo->num_uniform_buffers;
+    computePipeline->header.numSamplers = createinfo->num_samplers;
+    computePipeline->header.numReadonlyStorageTextures = createinfo->num_readonly_storage_textures;
+    computePipeline->header.numReadonlyStorageBuffers = createinfo->num_readonly_storage_buffers;
+    computePipeline->header.numReadWriteStorageTextures = createinfo->num_readwrite_storage_textures;
+    computePipeline->header.numReadWriteStorageBuffers = createinfo->num_readwrite_storage_buffers;
+    computePipeline->header.numUniformBuffers = createinfo->num_uniform_buffers;
    SDL_SetAtomicInt(&computePipeline->referenceCount, 0);

    if (renderer->debug_mode && SDL_HasProperty(createinfo->props, SDL_PROP_GPU_COMPUTEPIPELINE_CREATE_NAME_STRING)) {
@@ -3172,15 +3201,15 @@ static SDL_GPUGraphicsPipeline *D3D12_CreateGraphicsPipeline(

    pipeline->primitiveType = createinfo->primitive_type;

-    pipeline->vertexSamplerCount = vertShader->num_samplers;
-    pipeline->vertexStorageTextureCount = vertShader->numStorageTextures;
-    pipeline->vertexStorageBufferCount = vertShader->numStorageBuffers;
-    pipeline->vertexUniformBufferCount = vertShader->numUniformBuffers;
+    pipeline->header.num_vertex_samplers = vertShader->num_samplers;
+    pipeline->header.num_vertex_storage_textures = vertShader->numStorageTextures;
+    pipeline->header.num_vertex_storage_buffers = vertShader->numStorageBuffers;
+    pipeline->header.num_vertex_uniform_buffers = vertShader->numUniformBuffers;

-    pipeline->fragmentSamplerCount = fragShader->num_samplers;
-    pipeline->fragmentStorageTextureCount = fragShader->numStorageTextures;
-    pipeline->fragmentStorageBufferCount = fragShader->numStorageBuffers;
-    pipeline->fragmentUniformBufferCount = fragShader->numUniformBuffers;
+    pipeline->header.num_fragment_samplers = fragShader->num_samplers;
+    pipeline->header.num_fragment_storage_textures = fragShader->numStorageTextures;
+    pipeline->header.num_fragment_storage_buffers = fragShader->numStorageBuffers;
+    pipeline->header.num_fragment_uniform_buffers = fragShader->numUniformBuffers;

    SDL_SetAtomicInt(&pipeline->referenceCount, 0);

@@ -3350,7 +3379,7 @@ static D3D12Texture *D3D12_INTERNAL_CreateTexture(

    if (createinfo->type != SDL_GPU_TEXTURETYPE_3D) {
        desc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D;
-        desc.Alignment = isSwapchainTexture ? 0 : D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT;
+        desc.Alignment = isSwapchainTexture ? 0 : isMultisample ? D3D12_DEFAULT_MSAA_RESOURCE_PLACEMENT_ALIGNMENT : D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT;
        desc.Width = createinfo->width;
        desc.Height = createinfo->height;
        desc.DepthOrArraySize = (UINT16)createinfo->layer_count_or_depth;
@@ -4497,7 +4526,6 @@ static D3D12UniformBuffer *D3D12_INTERNAL_AcquireUniformBufferFromPool(

    SDL_UnlockMutex(renderer->acquireUniformBufferLock);

-    uniformBuffer->currentBlockSize = 0;
    uniformBuffer->drawOffset = 0;
    uniformBuffer->writeOffset = 0;

@@ -4536,6 +4564,7 @@ static void D3D12_INTERNAL_PushUniformData(
    Uint32 length)
 {
    D3D12UniformBuffer *uniformBuffer;
+    Uint32 blockSize;

    if (shaderStage == SDL_GPU_SHADERSTAGE_VERTEX) {
        if (commandBuffer->vertexUniformBuffers[slotIndex] == NULL) {
@@ -4560,13 +4589,13 @@ static void D3D12_INTERNAL_PushUniformData(
        return;
    }

-    uniformBuffer->currentBlockSize =
+    blockSize =
        D3D12_INTERNAL_Align(
            length,
            256);

    // If there is no more room, acquire a new uniform buffer
-    if (uniformBuffer->writeOffset + uniformBuffer->currentBlockSize >= UNIFORM_BUFFER_SIZE) {
+    if (uniformBuffer->writeOffset + blockSize >= UNIFORM_BUFFER_SIZE) {
        ID3D12Resource_Unmap(
            uniformBuffer->buffer->handle,
            0,
@@ -4596,7 +4625,7 @@ static void D3D12_INTERNAL_PushUniformData(
        data,
        length);

-    uniformBuffer->writeOffset += uniformBuffer->currentBlockSize;
+    uniformBuffer->writeOffset += blockSize;

    if (shaderStage == SDL_GPU_SHADERSTAGE_VERTEX) {
        commandBuffer->needVertexUniformBufferBind[slotIndex] = true;
@@ -4637,14 +4666,14 @@ static void D3D12_BindGraphicsPipeline(
        d3d12CommandBuffer->needFragmentUniformBufferBind[i] = true;
    }

-    for (i = 0; i < pipeline->vertexUniformBufferCount; i += 1) {
+    for (i = 0; i < pipeline->header.num_vertex_uniform_buffers; i += 1) {
        if (d3d12CommandBuffer->vertexUniformBuffers[i] == NULL) {
            d3d12CommandBuffer->vertexUniformBuffers[i] = D3D12_INTERNAL_AcquireUniformBufferFromPool(
                d3d12CommandBuffer);
        }
    }

-    for (i = 0; i < pipeline->fragmentUniformBufferCount; i += 1) {
+    for (i = 0; i < pipeline->header.num_fragment_uniform_buffers; i += 1) {
        if (d3d12CommandBuffer->fragmentUniformBuffers[i] == NULL) {
            d3d12CommandBuffer->fragmentUniformBuffers[i] = D3D12_INTERNAL_AcquireUniformBufferFromPool(
                d3d12CommandBuffer);
@@ -4711,21 +4740,21 @@ static void D3D12_BindVertexSamplers(
        D3D12TextureContainer *container = (D3D12TextureContainer *)textureSamplerBindings[i].texture;
        D3D12Sampler *sampler = (D3D12Sampler *)textureSamplerBindings[i].sampler;

-        if (d3d12CommandBuffer->vertexSamplers[firstSlot + i] != sampler) {
+        if (d3d12CommandBuffer->vertexSamplerDescriptorHandles[firstSlot + i].ptr != sampler->handle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackSampler(
                d3d12CommandBuffer,
                sampler);

-            d3d12CommandBuffer->vertexSamplers[firstSlot + i] = sampler;
+            d3d12CommandBuffer->vertexSamplerDescriptorHandles[firstSlot + i] = sampler->handle.cpuHandle;
            d3d12CommandBuffer->needVertexSamplerBind = true;
        }

-        if (d3d12CommandBuffer->vertexSamplerTextures[firstSlot + i] != container->activeTexture) {
+        if (d3d12CommandBuffer->vertexSamplerTextureDescriptorHandles[firstSlot + i].ptr != container->activeTexture->srvHandle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackTexture(
                d3d12CommandBuffer,
                container->activeTexture);

-            d3d12CommandBuffer->vertexSamplerTextures[firstSlot + i] = container->activeTexture;
+            d3d12CommandBuffer->vertexSamplerTextureDescriptorHandles[firstSlot + i] = container->activeTexture->srvHandle.cpuHandle;
            d3d12CommandBuffer->needVertexSamplerBind = true;
        }
    }
@@ -4743,10 +4772,10 @@ static void D3D12_BindVertexStorageTextures(
        D3D12TextureContainer *container = (D3D12TextureContainer *)storageTextures[i];
        D3D12Texture *texture = container->activeTexture;

-        if (d3d12CommandBuffer->vertexStorageTextures[firstSlot + i] != texture) {
+        if (d3d12CommandBuffer->vertexStorageTextureDescriptorHandles[firstSlot + i].ptr != texture->srvHandle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackTexture(d3d12CommandBuffer, texture);

-            d3d12CommandBuffer->vertexStorageTextures[firstSlot + i] = texture;
+            d3d12CommandBuffer->vertexStorageTextureDescriptorHandles[firstSlot + i] = texture->srvHandle.cpuHandle;
            d3d12CommandBuffer->needVertexStorageTextureBind = true;
        }
    }
@@ -4762,12 +4791,12 @@ static void D3D12_BindVertexStorageBuffers(

    for (Uint32 i = 0; i < numBindings; i += 1) {
        D3D12BufferContainer *container = (D3D12BufferContainer *)storageBuffers[i];
-        if (d3d12CommandBuffer->vertexStorageBuffers[firstSlot + i] != container->activeBuffer) {
+        if (d3d12CommandBuffer->vertexStorageBufferDescriptorHandles[firstSlot + i].ptr != container->activeBuffer->srvDescriptor.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackBuffer(
                d3d12CommandBuffer,
                container->activeBuffer);

-            d3d12CommandBuffer->vertexStorageBuffers[firstSlot + i] = container->activeBuffer;
+            d3d12CommandBuffer->vertexStorageBufferDescriptorHandles[firstSlot + i] = container->activeBuffer->srvDescriptor.cpuHandle;
            d3d12CommandBuffer->needVertexStorageBufferBind = true;
        }
    }
@@ -4785,21 +4814,21 @@ static void D3D12_BindFragmentSamplers(
        D3D12TextureContainer *container = (D3D12TextureContainer *)textureSamplerBindings[i].texture;
        D3D12Sampler *sampler = (D3D12Sampler *)textureSamplerBindings[i].sampler;

-        if (d3d12CommandBuffer->fragmentSamplers[firstSlot + i] != sampler) {
+        if (d3d12CommandBuffer->fragmentSamplerDescriptorHandles[firstSlot + i].ptr != sampler->handle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackSampler(
                d3d12CommandBuffer,
                sampler);

-            d3d12CommandBuffer->fragmentSamplers[firstSlot + i] = sampler;
+            d3d12CommandBuffer->fragmentSamplerDescriptorHandles[firstSlot + i] = sampler->handle.cpuHandle;
            d3d12CommandBuffer->needFragmentSamplerBind = true;
        }

-        if (d3d12CommandBuffer->fragmentSamplerTextures[firstSlot + i] != container->activeTexture) {
+        if (d3d12CommandBuffer->fragmentSamplerTextureDescriptorHandles[firstSlot + i].ptr != container->activeTexture->srvHandle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackTexture(
                d3d12CommandBuffer,
                container->activeTexture);

-            d3d12CommandBuffer->fragmentSamplerTextures[firstSlot + i] = container->activeTexture;
+            d3d12CommandBuffer->fragmentSamplerTextureDescriptorHandles[firstSlot + i] = container->activeTexture->srvHandle.cpuHandle;
            d3d12CommandBuffer->needFragmentSamplerBind = true;
        }
    }
@@ -4817,10 +4846,10 @@ static void D3D12_BindFragmentStorageTextures(
        D3D12TextureContainer *container = (D3D12TextureContainer *)storageTextures[i];
        D3D12Texture *texture = container->activeTexture;

-        if (d3d12CommandBuffer->fragmentStorageTextures[firstSlot + i] != texture) {
+        if (d3d12CommandBuffer->fragmentStorageTextureDescriptorHandles[firstSlot + i].ptr != texture->srvHandle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackTexture(d3d12CommandBuffer, texture);

-            d3d12CommandBuffer->fragmentStorageTextures[firstSlot + i] = texture;
+            d3d12CommandBuffer->fragmentStorageTextureDescriptorHandles[firstSlot + i] = texture->srvHandle.cpuHandle;
            d3d12CommandBuffer->needFragmentStorageTextureBind = true;
        }
    }
@@ -4837,12 +4866,12 @@ static void D3D12_BindFragmentStorageBuffers(
    for (Uint32 i = 0; i < numBindings; i += 1) {
        D3D12BufferContainer *container = (D3D12BufferContainer *)storageBuffers[i];

-        if (d3d12CommandBuffer->fragmentStorageBuffers[firstSlot + i] != container->activeBuffer) {
+        if (d3d12CommandBuffer->fragmentStorageBufferDescriptorHandles[firstSlot + i].ptr != container->activeBuffer->srvDescriptor.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackBuffer(
                d3d12CommandBuffer,
                container->activeBuffer);

-            d3d12CommandBuffer->fragmentStorageBuffers[firstSlot + i] = container->activeBuffer;
+            d3d12CommandBuffer->fragmentStorageBufferDescriptorHandles[firstSlot + i] = container->activeBuffer->srvDescriptor.cpuHandle;
            d3d12CommandBuffer->needFragmentStorageBufferBind = true;
        }
    }
@@ -4923,15 +4952,19 @@ static void D3D12_INTERNAL_WriteGPUDescriptors(
    gpuBaseDescriptor->ptr = heap->descriptorHeapGPUStart.ptr + (heap->currentDescriptorIndex * heap->descriptorSize);

    for (Uint32 i = 0; i < resourceHandleCount; i += 1) {
-        ID3D12Device_CopyDescriptorsSimple(
-            commandBuffer->renderer->device,
-            1,
-            gpuHeapCpuHandle,
-            resourceDescriptorHandles[i],
-            heapType);
+        // This will crash the driver if it gets a null handle! Cool!
+        if (resourceDescriptorHandles[i].ptr != 0)
+        {
+            ID3D12Device_CopyDescriptorsSimple(
+                commandBuffer->renderer->device,
+                1,
+                gpuHeapCpuHandle,
+                resourceDescriptorHandles[i],
+                heapType);

-        heap->currentDescriptorIndex += 1;
-        gpuHeapCpuHandle.ptr += heap->descriptorSize;
+            heap->currentDescriptorIndex += 1;
+            gpuHeapCpuHandle.ptr += heap->descriptorSize;
+        }
    }
 }

@@ -4961,19 +4994,21 @@ static void D3D12_INTERNAL_BindGraphicsResources(
            0,
            commandBuffer->vertexBufferCount,
            vertexBufferViews);
+
+        commandBuffer->needVertexBufferBind = false;
    }

    if (commandBuffer->needVertexSamplerBind) {
-        if (graphicsPipeline->vertexSamplerCount > 0) {
-            for (Uint32 i = 0; i < graphicsPipeline->vertexSamplerCount; i += 1) {
-                cpuHandles[i] = commandBuffer->vertexSamplers[i]->handle.cpuHandle;
+        if (graphicsPipeline->header.num_vertex_samplers > 0) {
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_vertex_samplers; i += 1) {
+                cpuHandles[i] = commandBuffer->vertexSamplerDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER,
                cpuHandles,
-                graphicsPipeline->vertexSamplerCount,
+                graphicsPipeline->header.num_vertex_samplers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -4981,15 +5016,15 @@ static void D3D12_INTERNAL_BindGraphicsResources(
                graphicsPipeline->rootSignature->vertexSamplerRootIndex,
                gpuDescriptorHandle);

-            for (Uint32 i = 0; i < graphicsPipeline->vertexSamplerCount; i += 1) {
-                cpuHandles[i] = commandBuffer->vertexSamplerTextures[i]->srvHandle.cpuHandle;
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_vertex_samplers; i += 1) {
+                cpuHandles[i] = commandBuffer->vertexSamplerTextureDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                graphicsPipeline->vertexSamplerCount,
+                graphicsPipeline->header.num_vertex_samplers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5001,16 +5036,16 @@ static void D3D12_INTERNAL_BindGraphicsResources(
    }

    if (commandBuffer->needVertexStorageTextureBind) {
-        if (graphicsPipeline->vertexStorageTextureCount > 0) {
-            for (Uint32 i = 0; i < graphicsPipeline->vertexStorageTextureCount; i += 1) {
-                cpuHandles[i] = commandBuffer->vertexStorageTextures[i]->srvHandle.cpuHandle;
+        if (graphicsPipeline->header.num_vertex_storage_textures > 0) {
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_vertex_storage_textures; i += 1) {
+                cpuHandles[i] = commandBuffer->vertexStorageTextureDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                graphicsPipeline->vertexStorageTextureCount,
+                graphicsPipeline->header.num_vertex_storage_textures,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5022,16 +5057,16 @@ static void D3D12_INTERNAL_BindGraphicsResources(
    }

    if (commandBuffer->needVertexStorageBufferBind) {
-        if (graphicsPipeline->vertexStorageBufferCount > 0) {
-            for (Uint32 i = 0; i < graphicsPipeline->vertexStorageBufferCount; i += 1) {
-                cpuHandles[i] = commandBuffer->vertexStorageBuffers[i]->srvDescriptor.cpuHandle;
+        if (graphicsPipeline->header.num_vertex_storage_buffers > 0) {
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_vertex_storage_buffers; i += 1) {
+                cpuHandles[i] = commandBuffer->vertexStorageBufferDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                graphicsPipeline->vertexStorageBufferCount,
+                graphicsPipeline->header.num_vertex_storage_buffers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5044,7 +5079,7 @@ static void D3D12_INTERNAL_BindGraphicsResources(

    for (Uint32 i = 0; i < MAX_UNIFORM_BUFFERS_PER_STAGE; i += 1) {
        if (commandBuffer->needVertexUniformBufferBind[i]) {
-            if (graphicsPipeline->vertexUniformBufferCount > i) {
+            if (graphicsPipeline->header.num_vertex_uniform_buffers > i) {
                ID3D12GraphicsCommandList_SetGraphicsRootConstantBufferView(
                    commandBuffer->graphicsCommandList,
                    graphicsPipeline->rootSignature->vertexUniformBufferRootIndex[i],
@@ -5055,16 +5090,16 @@ static void D3D12_INTERNAL_BindGraphicsResources(
    }

    if (commandBuffer->needFragmentSamplerBind) {
-        if (graphicsPipeline->fragmentSamplerCount > 0) {
-            for (Uint32 i = 0; i < graphicsPipeline->fragmentSamplerCount; i += 1) {
-                cpuHandles[i] = commandBuffer->fragmentSamplers[i]->handle.cpuHandle;
+        if (graphicsPipeline->header.num_fragment_samplers > 0) {
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_fragment_samplers; i += 1) {
+                cpuHandles[i] = commandBuffer->fragmentSamplerDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER,
                cpuHandles,
-                graphicsPipeline->fragmentSamplerCount,
+                graphicsPipeline->header.num_fragment_samplers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5072,15 +5107,15 @@ static void D3D12_INTERNAL_BindGraphicsResources(
                graphicsPipeline->rootSignature->fragmentSamplerRootIndex,
                gpuDescriptorHandle);

-            for (Uint32 i = 0; i < graphicsPipeline->fragmentSamplerCount; i += 1) {
-                cpuHandles[i] = commandBuffer->fragmentSamplerTextures[i]->srvHandle.cpuHandle;
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_fragment_samplers; i += 1) {
+                cpuHandles[i] = commandBuffer->fragmentSamplerTextureDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                graphicsPipeline->fragmentSamplerCount,
+                graphicsPipeline->header.num_fragment_samplers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5092,16 +5127,16 @@ static void D3D12_INTERNAL_BindGraphicsResources(
    }

    if (commandBuffer->needFragmentStorageTextureBind) {
-        if (graphicsPipeline->fragmentStorageTextureCount > 0) {
-            for (Uint32 i = 0; i < graphicsPipeline->fragmentStorageTextureCount; i += 1) {
-                cpuHandles[i] = commandBuffer->fragmentStorageTextures[i]->srvHandle.cpuHandle;
+        if (graphicsPipeline->header.num_fragment_storage_textures > 0) {
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_fragment_storage_textures; i += 1) {
+                cpuHandles[i] = commandBuffer->fragmentStorageTextureDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                graphicsPipeline->fragmentStorageTextureCount,
+                graphicsPipeline->header.num_fragment_storage_textures,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5113,16 +5148,16 @@ static void D3D12_INTERNAL_BindGraphicsResources(
    }

    if (commandBuffer->needFragmentStorageBufferBind) {
-        if (graphicsPipeline->fragmentStorageBufferCount > 0) {
-            for (Uint32 i = 0; i < graphicsPipeline->fragmentStorageBufferCount; i += 1) {
-                cpuHandles[i] = commandBuffer->fragmentStorageBuffers[i]->srvDescriptor.cpuHandle;
+        if (graphicsPipeline->header.num_fragment_storage_buffers > 0) {
+            for (Uint32 i = 0; i < graphicsPipeline->header.num_fragment_storage_buffers; i += 1) {
+                cpuHandles[i] = commandBuffer->fragmentStorageBufferDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                graphicsPipeline->fragmentStorageBufferCount,
+                graphicsPipeline->header.num_fragment_storage_buffers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetGraphicsRootDescriptorTable(
@@ -5135,7 +5170,7 @@ static void D3D12_INTERNAL_BindGraphicsResources(

    for (Uint32 i = 0; i < MAX_UNIFORM_BUFFERS_PER_STAGE; i += 1) {
        if (commandBuffer->needFragmentUniformBufferBind[i]) {
-            if (graphicsPipeline->fragmentUniformBufferCount > i) {
+            if (graphicsPipeline->header.num_fragment_uniform_buffers > i) {
                ID3D12GraphicsCommandList_SetGraphicsRootConstantBufferView(
                    commandBuffer->graphicsCommandList,
                    graphicsPipeline->rootSignature->fragmentUniformBufferRootIndex[i],
@@ -5300,15 +5335,15 @@ static void D3D12_EndRenderPass(
    SDL_zeroa(d3d12CommandBuffer->vertexBufferOffsets);
    d3d12CommandBuffer->vertexBufferCount = 0;

-    SDL_zeroa(d3d12CommandBuffer->vertexSamplerTextures);
-    SDL_zeroa(d3d12CommandBuffer->vertexSamplers);
-    SDL_zeroa(d3d12CommandBuffer->vertexStorageTextures);
-    SDL_zeroa(d3d12CommandBuffer->vertexStorageBuffers);
+    SDL_zeroa(d3d12CommandBuffer->vertexSamplerTextureDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->vertexSamplerDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->vertexStorageTextureDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->vertexStorageBufferDescriptorHandles);

-    SDL_zeroa(d3d12CommandBuffer->fragmentSamplerTextures);
-    SDL_zeroa(d3d12CommandBuffer->fragmentSamplers);
-    SDL_zeroa(d3d12CommandBuffer->fragmentStorageTextures);
-    SDL_zeroa(d3d12CommandBuffer->fragmentStorageBuffers);
+    SDL_zeroa(d3d12CommandBuffer->fragmentSamplerTextureDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->fragmentSamplerDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->fragmentStorageTextureDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->fragmentStorageBufferDescriptorHandles);
 }

 // Compute Pass
@@ -5342,6 +5377,7 @@ static void D3D12_BeginComputePass(
                D3D12_RESOURCE_STATE_UNORDERED_ACCESS);

            d3d12CommandBuffer->computeReadWriteStorageTextureSubresources[i] = subresource;
+            d3d12CommandBuffer->computeReadWriteStorageTextureDescriptorHandles[i] = subresource->uavHandle.cpuHandle;

            D3D12_INTERNAL_TrackTexture(
                d3d12CommandBuffer,
@@ -5360,6 +5396,7 @@ static void D3D12_BeginComputePass(
                D3D12_RESOURCE_STATE_UNORDERED_ACCESS);

            d3d12CommandBuffer->computeReadWriteStorageBuffers[i] = buffer;
+            d3d12CommandBuffer->computeReadWriteStorageBufferDescriptorHandles[i] = buffer->uavDescriptor.cpuHandle;

            D3D12_INTERNAL_TrackBuffer(
                d3d12CommandBuffer,
@@ -5401,7 +5438,7 @@ static void D3D12_BindComputePipeline(
        d3d12CommandBuffer->needComputeUniformBufferBind[i] = true;
    }

-    for (Uint32 i = 0; i < pipeline->numUniformBuffers; i += 1) {
+    for (Uint32 i = 0; i < pipeline->header.numUniformBuffers; i += 1) {
        if (d3d12CommandBuffer->computeUniformBuffers[i] == NULL) {
            d3d12CommandBuffer->computeUniformBuffers[i] = D3D12_INTERNAL_AcquireUniformBufferFromPool(
                d3d12CommandBuffer);
@@ -5411,9 +5448,9 @@ static void D3D12_BindComputePipeline(
    D3D12_INTERNAL_TrackComputePipeline(d3d12CommandBuffer, pipeline);

    // Bind write-only resources after setting root signature
-    if (pipeline->numReadWriteStorageTextures > 0) {
-        for (Uint32 i = 0; i < pipeline->numReadWriteStorageTextures; i += 1) {
-            cpuHandles[i] = d3d12CommandBuffer->computeReadWriteStorageTextureSubresources[i]->uavHandle.cpuHandle;
+    if (pipeline->header.numReadWriteStorageTextures > 0) {
+        for (Uint32 i = 0; i < pipeline->header.numReadWriteStorageTextures; i += 1) {
+            cpuHandles[i] = d3d12CommandBuffer->computeReadWriteStorageTextureDescriptorHandles[i];
        }

        D3D12_INTERNAL_WriteGPUDescriptors(
@@ -5429,9 +5466,9 @@ static void D3D12_BindComputePipeline(
            gpuDescriptorHandle);
    }

-    if (pipeline->numReadWriteStorageBuffers > 0) {
-        for (Uint32 i = 0; i < pipeline->numReadWriteStorageBuffers; i += 1) {
-            cpuHandles[i] = d3d12CommandBuffer->computeReadWriteStorageBuffers[i]->uavDescriptor.cpuHandle;
+    if (pipeline->header.numReadWriteStorageBuffers > 0) {
+        for (Uint32 i = 0; i < pipeline->header.numReadWriteStorageBuffers; i += 1) {
+            cpuHandles[i] = d3d12CommandBuffer->computeReadWriteStorageBufferDescriptorHandles[i];
        }

        D3D12_INTERNAL_WriteGPUDescriptors(
@@ -5460,21 +5497,21 @@ static void D3D12_BindComputeSamplers(
        D3D12TextureContainer *container = (D3D12TextureContainer *)textureSamplerBindings[i].texture;
        D3D12Sampler *sampler = (D3D12Sampler *)textureSamplerBindings[i].sampler;

-        if (d3d12CommandBuffer->computeSamplers[firstSlot + i] != sampler) {
+        if (d3d12CommandBuffer->computeSamplerDescriptorHandles[firstSlot + i].ptr != sampler->handle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackSampler(
                d3d12CommandBuffer,
                (D3D12Sampler *)textureSamplerBindings[i].sampler);

-            d3d12CommandBuffer->computeSamplers[firstSlot + i] = (D3D12Sampler *)textureSamplerBindings[i].sampler;
+            d3d12CommandBuffer->computeSamplerDescriptorHandles[firstSlot + i] = sampler->handle.cpuHandle;
            d3d12CommandBuffer->needComputeSamplerBind = true;
        }

-        if (d3d12CommandBuffer->computeSamplerTextures[firstSlot + i] != container->activeTexture) {
+        if (d3d12CommandBuffer->computeSamplerTextureDescriptorHandles[firstSlot + i].ptr != container->activeTexture->srvHandle.cpuHandle.ptr) {
            D3D12_INTERNAL_TrackTexture(
                d3d12CommandBuffer,
                container->activeTexture);

-            d3d12CommandBuffer->computeSamplerTextures[firstSlot + i] = container->activeTexture;
+            d3d12CommandBuffer->computeSamplerTextureDescriptorHandles[firstSlot + i] = container->activeTexture->srvHandle.cpuHandle;
            d3d12CommandBuffer->needComputeSamplerBind = true;
        }
    }
@@ -5511,6 +5548,7 @@ static void D3D12_BindComputeStorageTextures(
                container->activeTexture);

            d3d12CommandBuffer->computeReadOnlyStorageTextures[firstSlot + i] = container->activeTexture;
+            d3d12CommandBuffer->computeReadOnlyStorageTextureDescriptorHandles[firstSlot + i] = container->activeTexture->srvHandle.cpuHandle;
            d3d12CommandBuffer->needComputeReadOnlyStorageTextureBind = true;
        }
    }
@@ -5548,6 +5586,7 @@ static void D3D12_BindComputeStorageBuffers(
                buffer);

            d3d12CommandBuffer->computeReadOnlyStorageBuffers[firstSlot + i] = buffer;
+            d3d12CommandBuffer->computeReadOnlyStorageBufferDescriptorHandles[firstSlot + i] = buffer->srvDescriptor.cpuHandle;
            d3d12CommandBuffer->needComputeReadOnlyStorageBufferBind = true;
        }
    }
@@ -5583,16 +5622,16 @@ static void D3D12_INTERNAL_BindComputeResources(
    D3D12_GPU_DESCRIPTOR_HANDLE gpuDescriptorHandle;

    if (commandBuffer->needComputeSamplerBind) {
-        if (computePipeline->numSamplers > 0) {
-            for (Uint32 i = 0; i < computePipeline->numSamplers; i += 1) {
-                cpuHandles[i] = commandBuffer->computeSamplers[i]->handle.cpuHandle;
+        if (computePipeline->header.numSamplers > 0) {
+            for (Uint32 i = 0; i < computePipeline->header.numSamplers; i += 1) {
+                cpuHandles[i] = commandBuffer->computeSamplerDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER,
                cpuHandles,
-                computePipeline->numSamplers,
+                computePipeline->header.numSamplers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetComputeRootDescriptorTable(
@@ -5600,15 +5639,15 @@ static void D3D12_INTERNAL_BindComputeResources(
                computePipeline->rootSignature->samplerRootIndex,
                gpuDescriptorHandle);

-            for (Uint32 i = 0; i < computePipeline->numSamplers; i += 1) {
-                cpuHandles[i] = commandBuffer->computeSamplerTextures[i]->srvHandle.cpuHandle;
+            for (Uint32 i = 0; i < computePipeline->header.numSamplers; i += 1) {
+                cpuHandles[i] = commandBuffer->computeSamplerTextureDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                computePipeline->numSamplers,
+                computePipeline->header.numSamplers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetComputeRootDescriptorTable(
@@ -5620,16 +5659,16 @@ static void D3D12_INTERNAL_BindComputeResources(
    }

    if (commandBuffer->needComputeReadOnlyStorageTextureBind) {
-        if (computePipeline->numReadOnlyStorageTextures > 0) {
-            for (Uint32 i = 0; i < computePipeline->numReadOnlyStorageTextures; i += 1) {
-                cpuHandles[i] = commandBuffer->computeReadOnlyStorageTextures[i]->srvHandle.cpuHandle;
+        if (computePipeline->header.numReadonlyStorageTextures > 0) {
+            for (Uint32 i = 0; i < computePipeline->header.numReadonlyStorageTextures; i += 1) {
+                cpuHandles[i] = commandBuffer->computeReadOnlyStorageTextureDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                computePipeline->numReadOnlyStorageTextures,
+                computePipeline->header.numReadonlyStorageTextures,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetComputeRootDescriptorTable(
@@ -5641,16 +5680,16 @@ static void D3D12_INTERNAL_BindComputeResources(
    }

    if (commandBuffer->needComputeReadOnlyStorageBufferBind) {
-        if (computePipeline->numReadOnlyStorageBuffers > 0) {
-            for (Uint32 i = 0; i < computePipeline->numReadOnlyStorageBuffers; i += 1) {
-                cpuHandles[i] = commandBuffer->computeReadOnlyStorageBuffers[i]->srvDescriptor.cpuHandle;
+        if (computePipeline->header.numReadonlyStorageBuffers > 0) {
+            for (Uint32 i = 0; i < computePipeline->header.numReadonlyStorageBuffers; i += 1) {
+                cpuHandles[i] = commandBuffer->computeReadOnlyStorageBufferDescriptorHandles[i];
            }

            D3D12_INTERNAL_WriteGPUDescriptors(
                commandBuffer,
                D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV,
                cpuHandles,
-                computePipeline->numReadOnlyStorageBuffers,
+                computePipeline->header.numReadonlyStorageBuffers,
                &gpuDescriptorHandle);

            ID3D12GraphicsCommandList_SetComputeRootDescriptorTable(
@@ -5663,7 +5702,7 @@ static void D3D12_INTERNAL_BindComputeResources(

    for (Uint32 i = 0; i < MAX_UNIFORM_BUFFERS_PER_STAGE; i += 1) {
        if (commandBuffer->needComputeUniformBufferBind[i]) {
-            if (computePipeline->numUniformBuffers > i) {
+            if (computePipeline->header.numUniformBuffers > i) {
                ID3D12GraphicsCommandList_SetComputeRootConstantBufferView(
                    commandBuffer->graphicsCommandList,
                    computePipeline->rootSignature->uniformBufferRootIndex[i],
@@ -5762,8 +5801,11 @@ static void D3D12_EndComputePass(
        }
    }

-    SDL_zeroa(d3d12CommandBuffer->computeSamplerTextures);
-    SDL_zeroa(d3d12CommandBuffer->computeSamplers);
+    SDL_zeroa(d3d12CommandBuffer->computeSamplerTextureDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->computeSamplerDescriptorHandles);
+
+    SDL_zeroa(d3d12CommandBuffer->computeReadWriteStorageTextureDescriptorHandles);
+    SDL_zeroa(d3d12CommandBuffer->computeReadWriteStorageBufferDescriptorHandles);

    d3d12CommandBuffer->currentComputePipeline = NULL;
 }
@@ -5911,7 +5953,7 @@ static void D3D12_UploadToTexture(
                SDL_memcpy(
                    temporaryBuffer->mapPointer + (sliceIndex * rowsPerSlice) + (rowIndex * alignedRowPitch),
                    transferBufferContainer->activeBuffer->mapPointer + source->offset + (sliceIndex * bytesPerSlice) + (rowIndex * rowPitch),
-                    alignedRowPitch);
+                    rowPitch);
            }
            Uint32 offset = source->offset + (sliceIndex * bytesPerSlice) + ((rowsPerSlice - 1) * rowPitch);
            SDL_memcpy(
@@ -7267,6 +7309,11 @@ static bool D3D12_INTERNAL_AllocateCommandBuffer(
    commandBuffer->usedComputePipelines = (D3D12ComputePipeline **)SDL_calloc(
        commandBuffer->usedComputePipelineCapacity, sizeof(D3D12ComputePipeline *));

+    commandBuffer->usedDescriptorHeapCapacity = 4;
+    commandBuffer->usedDescriptorHeapCount = 0;
+    commandBuffer->usedDescriptorHeaps = (D3D12DescriptorHeap **)SDL_calloc(
+        commandBuffer->usedDescriptorHeapCapacity, sizeof(D3D12DescriptorHeap *));
+
    commandBuffer->usedUniformBufferCapacity = 4;
    commandBuffer->usedUniformBufferCount = 0;
    commandBuffer->usedUniformBuffers = (D3D12UniformBuffer **)SDL_calloc(
@@ -7352,20 +7399,22 @@ static SDL_GPUCommandBuffer *D3D12_AcquireCommandBuffer(
    SDL_zeroa(commandBuffer->vertexBufferOffsets);
    commandBuffer->vertexBufferCount = 0;

-    SDL_zeroa(commandBuffer->vertexSamplerTextures);
-    SDL_zeroa(commandBuffer->vertexSamplers);
-    SDL_zeroa(commandBuffer->vertexStorageTextures);
-    SDL_zeroa(commandBuffer->vertexStorageBuffers);
+    SDL_zeroa(commandBuffer->vertexSamplerTextureDescriptorHandles);
+    SDL_zeroa(commandBuffer->vertexSamplerDescriptorHandles);
+    SDL_zeroa(commandBuffer->vertexStorageTextureDescriptorHandles);
+    SDL_zeroa(commandBuffer->vertexStorageBufferDescriptorHandles);
    SDL_zeroa(commandBuffer->vertexUniformBuffers);

-    SDL_zeroa(commandBuffer->fragmentSamplerTextures);
-    SDL_zeroa(commandBuffer->fragmentSamplers);
-    SDL_zeroa(commandBuffer->fragmentStorageTextures);
-    SDL_zeroa(commandBuffer->fragmentStorageBuffers);
+    SDL_zeroa(commandBuffer->fragmentSamplerTextureDescriptorHandles);
+    SDL_zeroa(commandBuffer->fragmentSamplerDescriptorHandles);
+    SDL_zeroa(commandBuffer->fragmentStorageTextureDescriptorHandles);
+    SDL_zeroa(commandBuffer->fragmentStorageBufferDescriptorHandles);
    SDL_zeroa(commandBuffer->fragmentUniformBuffers);

-    SDL_zeroa(commandBuffer->computeSamplerTextures);
-    SDL_zeroa(commandBuffer->computeSamplers);
+    SDL_zeroa(commandBuffer->computeSamplerTextureDescriptorHandles);
+    SDL_zeroa(commandBuffer->computeSamplerDescriptorHandles);
+    SDL_zeroa(commandBuffer->computeReadOnlyStorageTextureDescriptorHandles);
+    SDL_zeroa(commandBuffer->computeReadOnlyStorageBufferDescriptorHandles);
    SDL_zeroa(commandBuffer->computeReadOnlyStorageTextures);
    SDL_zeroa(commandBuffer->computeReadOnlyStorageBuffers);
    SDL_zeroa(commandBuffer->computeReadWriteStorageTextureSubresources);
@@ -7686,13 +7735,13 @@ static bool D3D12_INTERNAL_CleanCommandBuffer(
        NULL);
    CHECK_D3D12_ERROR_AND_RETURN("Could not reset command list", false);

-    // Return descriptor heaps to pool
-    D3D12_INTERNAL_ReturnGPUDescriptorHeapToPool(
-        renderer,
-        commandBuffer->gpuDescriptorHeaps[D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV]);
-    D3D12_INTERNAL_ReturnGPUDescriptorHeapToPool(
-        renderer,
-        commandBuffer->gpuDescriptorHeaps[D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER]);
+    // Return descriptor heaps to pool, pools own their own locks
+    for (i = 0; i < commandBuffer->usedDescriptorHeapCount; i += 1) {
+        D3D12_INTERNAL_ReturnGPUDescriptorHeapToPool(
+            renderer,
+            commandBuffer->usedDescriptorHeaps[i]);
+    }
+    commandBuffer->usedDescriptorHeapCount = 0;

    commandBuffer->gpuDescriptorHeaps[D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV] = NULL;
    commandBuffer->gpuDescriptorHeaps[D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER] = NULL;
@@ -8338,6 +8387,7 @@ static bool D3D12_PrepareDriver(SDL_VideoDevice *_this)
    IDXGIFactory4 *factory4;
    IDXGIFactory6 *factory6;
    IDXGIAdapter1 *adapter;
+    bool supports_64UAVs = false;

    // Can we load D3D12?

@@ -8426,12 +8476,39 @@ static bool D3D12_PrepareDriver(SDL_VideoDevice *_this)
        return false;
    }

+    SDL_COMPILE_TIME_ASSERT(featurelevel, D3D_FEATURE_LEVEL_CHOICE < D3D_FEATURE_LEVEL_11_1);
+
+    // Check feature level 11_1 first, guarantees 64+ UAVs unlike 11_0 Tier1
    res = D3D12CreateDeviceFunc(
        (IUnknown *)adapter,
-        D3D_FEATURE_LEVEL_CHOICE,
+        D3D_FEATURE_LEVEL_11_1,
        D3D_GUID(D3D_IID_ID3D12Device),
        (void **)&device);

+    if (SUCCEEDED(res)) {
+        supports_64UAVs = true;
+    } else {
+        res = D3D12CreateDeviceFunc(
+            (IUnknown *)adapter,
+            D3D_FEATURE_LEVEL_CHOICE,
+            D3D_GUID(D3D_IID_ID3D12Device),
+            (void **)&device);
+
+        if (SUCCEEDED(res)) {
+            D3D12_FEATURE_DATA_D3D12_OPTIONS featureOptions;
+            SDL_zero(featureOptions);
+
+            res = ID3D12Device_CheckFeatureSupport(
+                device,
+                D3D12_FEATURE_D3D12_OPTIONS,
+                &featureOptions,
+                sizeof(featureOptions));
+            if (SUCCEEDED(res) && featureOptions.ResourceBindingTier >= D3D12_RESOURCE_BINDING_TIER_2) {
+                supports_64UAVs = true;
+            }
+        }
+    }
+
    if (SUCCEEDED(res)) {
        ID3D12Device_Release(device);
    }
@@ -8441,6 +8518,11 @@ static bool D3D12_PrepareDriver(SDL_VideoDevice *_this)
    SDL_UnloadObject(d3d12Dll);
    SDL_UnloadObject(dxgiDll);

+    if (!supports_64UAVs) {
+        SDL_LogWarn(SDL_LOG_CATEGORY_GPU, "D3D12: Tier 2 Resource binding is not supported");
+        return false;
+    }
+
    if (FAILED(res)) {
        SDL_LogWarn(SDL_LOG_CATEGORY_GPU, "D3D12: Could not create D3D12Device with feature level " D3D_FEATURE_LEVEL_CHOICE_STR);
        return false;
@@ -8819,6 +8901,12 @@ static SDL_GPUDevice *D3D12_CreateDevice(bool debugMode, bool preferLowPower, SD
    // Initialize the D3D12 debug layer, if applicable
    if (debugMode) {
        bool hasD3d12Debug = D3D12_INTERNAL_TryInitializeD3D12Debug(renderer);
+#if (defined(SDL_PLATFORM_XBOXONE) || defined(SDL_PLATFORM_XBOXSERIES))
+        if (hasD3d12Debug) {
+            SDL_LogInfo(
+                SDL_LOG_CATEGORY_GPU,
+                "Validation layers enabled, expect debug level performance!");
+#else
        if (hasDxgiDebug && hasD3d12Debug) {
            SDL_LogInfo(
                SDL_LOG_CATEGORY_GPU,
@@ -8827,6 +8915,7 @@ static SDL_GPUDevice *D3D12_CreateDevice(bool debugMode, bool preferLowPower, SD
            SDL_LogWarn(
                SDL_LOG_CATEGORY_GPU,
                "Validation layers partially enabled, some warnings may not be available");
+#endif
        } else {
            SDL_LogWarn(
                SDL_LOG_CATEGORY_GPU,
@@ -9124,7 +9213,7 @@ static SDL_GPUDevice *D3D12_CreateDevice(bool debugMode, bool preferLowPower, SD
    renderer->debug_mode = debugMode;
    renderer->allowedFramesInFlight = 2;

-    renderer->semantic = SDL_GetStringProperty(props, SDL_PROP_GPU_DEVICE_CREATE_D3D12_SEMANTIC_NAME_STRING, "TEXCOORD");
+    renderer->semantic = SDL_strdup(SDL_GetStringProperty(props, SDL_PROP_GPU_DEVICE_CREATE_D3D12_SEMANTIC_NAME_STRING, "TEXCOORD"));

    // Blit resources
    D3D12_INTERNAL_InitBlitResources(renderer);
@@ -9159,8 +9248,23 @@ static SDL_GPUDevice *D3D12_CreateDevice(bool debugMode, bool preferLowPower, SD
        return NULL;
    }

+    SDL_GPUShaderFormat shaderFormats = SDL_GPU_SHADERFORMAT_DXBC;
+
+    D3D12_FEATURE_DATA_SHADER_MODEL shaderModel;
+    shaderModel.HighestShaderModel = D3D_SHADER_MODEL_6_0;
+
+    res = ID3D12Device_CheckFeatureSupport(
+        renderer->device,
+        D3D12_FEATURE_SHADER_MODEL,
+        &shaderModel,
+        sizeof(shaderModel));
+    if (SUCCEEDED(res) && shaderModel.HighestShaderModel >= D3D_SHADER_MODEL_6_0) {
+        shaderFormats |= SDL_GPU_SHADERFORMAT_DXIL;
+    }
+
    ASSIGN_DRIVER(D3D12)
    result->driverData = (SDL_GPURenderer *)renderer;
+    result->shader_formats = shaderFormats;
    result->debug_mode = debugMode;
    renderer->sdlGPUDevice = result;

@@ -476,33 +476,21 @@ typedef struct MetalShader

 typedef struct MetalGraphicsPipeline
 {
+    GraphicsPipelineCommonHeader header;
+
    id<MTLRenderPipelineState> handle;

    SDL_GPURasterizerState rasterizerState;
    SDL_GPUPrimitiveType primitiveType;

    id<MTLDepthStencilState> depth_stencil_state;
-
-    Uint32 vertexSamplerCount;
-    Uint32 vertexUniformBufferCount;
-    Uint32 vertexStorageBufferCount;
-    Uint32 vertexStorageTextureCount;
-
-    Uint32 fragmentSamplerCount;
-    Uint32 fragmentUniformBufferCount;
-    Uint32 fragmentStorageBufferCount;
-    Uint32 fragmentStorageTextureCount;
 } MetalGraphicsPipeline;

 typedef struct MetalComputePipeline
 {
+    ComputePipelineCommonHeader header;
+
    id<MTLComputePipelineState> handle;
-    Uint32 numSamplers;
-    Uint32 numReadonlyStorageTextures;
-    Uint32 numReadWriteStorageTextures;
-    Uint32 numReadonlyStorageBuffers;
-    Uint32 numReadWriteStorageBuffers;
-    Uint32 numUniformBuffers;
    Uint32 threadcountX;
    Uint32 threadcountY;
    Uint32 threadcountZ;
@@ -900,6 +888,7 @@ static void METAL_INTERNAL_DestroyTextureContainer(
        container->textures[i]->handle = nil;
        SDL_free(container->textures[i]);
    }
+    SDL_DestroyProperties(container->header.info.props);
    if (container->debugName != NULL) {
        SDL_free(container->debugName);
    }
@@ -1059,12 +1048,12 @@ static SDL_GPUComputePipeline *METAL_CreateComputePipeline(

        pipeline = SDL_calloc(1, sizeof(MetalComputePipeline));
        pipeline->handle = handle;
-        pipeline->numSamplers = createinfo->num_samplers;
-        pipeline->numReadonlyStorageTextures = createinfo->num_readonly_storage_textures;
-        pipeline->numReadWriteStorageTextures = createinfo->num_readwrite_storage_textures;
-        pipeline->numReadonlyStorageBuffers = createinfo->num_readonly_storage_buffers;
-        pipeline->numReadWriteStorageBuffers = createinfo->num_readwrite_storage_buffers;
-        pipeline->numUniformBuffers = createinfo->num_uniform_buffers;
+        pipeline->header.numSamplers = createinfo->num_samplers;
+        pipeline->header.numReadonlyStorageTextures = createinfo->num_readonly_storage_textures;
+        pipeline->header.numReadWriteStorageTextures = createinfo->num_readwrite_storage_textures;
+        pipeline->header.numReadonlyStorageBuffers = createinfo->num_readonly_storage_buffers;
+        pipeline->header.numReadWriteStorageBuffers = createinfo->num_readwrite_storage_buffers;
+        pipeline->header.numUniformBuffers = createinfo->num_uniform_buffers;
        pipeline->threadcountX = createinfo->threadcount_x;
        pipeline->threadcountY = createinfo->threadcount_y;
        pipeline->threadcountZ = createinfo->threadcount_z;
@@ -1207,14 +1196,14 @@ static SDL_GPUGraphicsPipeline *METAL_CreateGraphicsPipeline(
        result->depth_stencil_state = depthStencilState;
        result->rasterizerState = createinfo->rasterizer_state;
        result->primitiveType = createinfo->primitive_type;
-        result->vertexSamplerCount = vertexShader->numSamplers;
-        result->vertexUniformBufferCount = vertexShader->numUniformBuffers;
-        result->vertexStorageBufferCount = vertexShader->numStorageBuffers;
-        result->vertexStorageTextureCount = vertexShader->numStorageTextures;
-        result->fragmentSamplerCount = fragmentShader->numSamplers;
-        result->fragmentUniformBufferCount = fragmentShader->numUniformBuffers;
-        result->fragmentStorageBufferCount = fragmentShader->numStorageBuffers;
-        result->fragmentStorageTextureCount = fragmentShader->numStorageTextures;
+        result->header.num_vertex_samplers = vertexShader->numSamplers;
+        result->header.num_vertex_uniform_buffers = vertexShader->numUniformBuffers;
+        result->header.num_vertex_storage_buffers = vertexShader->numStorageBuffers;
+        result->header.num_vertex_storage_textures = vertexShader->numStorageTextures;
+        result->header.num_fragment_samplers = fragmentShader->numSamplers;
+        result->header.num_fragment_uniform_buffers = fragmentShader->numUniformBuffers;
+        result->header.num_fragment_storage_buffers = fragmentShader->numStorageBuffers;
+        result->header.num_fragment_storage_textures = fragmentShader->numStorageTextures;
        return (SDL_GPUGraphicsPipeline *)result;
    }
 }
@@ -1789,7 +1778,8 @@ static void METAL_UploadToTexture(
                 copyFromBuffer:bufferContainer->activeBuffer->handle
                   sourceOffset:source->offset
              sourceBytesPerRow:BytesPerRow(destination->w, textureContainer->header.info.format)
-            sourceBytesPerImage:SDL_CalculateGPUTextureFormatSize(textureContainer->header.info.format, destination->w, destination->h, destination->d)
+            // sourceBytesPerImage expects the stride between 2D images (slices) of a 3D texture, not the size of the entire region
+            sourceBytesPerImage:SDL_CalculateGPUTextureFormatSize(textureContainer->header.info.format, destination->w, destination->h, 1)
                     sourceSize:MTLSizeMake(destination->w, destination->h, destination->d)
                      toTexture:metalTexture->handle
               destinationSlice:destination->layer
@@ -2412,14 +2402,14 @@ static void METAL_BindGraphicsPipeline(
            metalCommandBuffer->needFragmentUniformBufferBind[i] = true;
        }

-        for (i = 0; i < pipeline->vertexUniformBufferCount; i += 1) {
+        for (i = 0; i < pipeline->header.num_vertex_uniform_buffers; i += 1) {
            if (metalCommandBuffer->vertexUniformBuffers[i] == NULL) {
                metalCommandBuffer->vertexUniformBuffers[i] = METAL_INTERNAL_AcquireUniformBufferFromPool(
                    metalCommandBuffer);
            }
        }

-        for (i = 0; i < pipeline->fragmentUniformBufferCount; i += 1) {
+        for (i = 0; i < pipeline->header.num_fragment_uniform_buffers; i += 1) {
            if (metalCommandBuffer->fragmentUniformBuffers[i] == NULL) {
                metalCommandBuffer->fragmentUniformBuffers[i] = METAL_INTERNAL_AcquireUniformBufferFromPool(
                    metalCommandBuffer);
@@ -2650,11 +2640,11 @@ static void METAL_INTERNAL_BindGraphicsResources(
    // Vertex Samplers+Textures

    if (commandBuffer->needVertexSamplerBind) {
-        if (graphicsPipeline->vertexSamplerCount > 0) {
+        if (graphicsPipeline->header.num_vertex_samplers > 0) {
            [commandBuffer->renderEncoder setVertexSamplerStates:commandBuffer->vertexSamplers
-                                                       withRange:NSMakeRange(0, graphicsPipeline->vertexSamplerCount)];
+                                                       withRange:NSMakeRange(0, graphicsPipeline->header.num_vertex_samplers)];
            [commandBuffer->renderEncoder setVertexTextures:commandBuffer->vertexTextures
-                                                  withRange:NSMakeRange(0, graphicsPipeline->vertexSamplerCount)];
+                                                  withRange:NSMakeRange(0, graphicsPipeline->header.num_vertex_samplers)];
        }
        commandBuffer->needVertexSamplerBind = false;
    }
@@ -2662,10 +2652,10 @@ static void METAL_INTERNAL_BindGraphicsResources(
    // Vertex Storage Textures

    if (commandBuffer->needVertexStorageTextureBind) {
-        if (graphicsPipeline->vertexStorageTextureCount > 0) {
+        if (graphicsPipeline->header.num_vertex_storage_textures > 0) {
            [commandBuffer->renderEncoder setVertexTextures:commandBuffer->vertexStorageTextures
-                                                  withRange:NSMakeRange(graphicsPipeline->vertexSamplerCount,
-                                                                        graphicsPipeline->vertexStorageTextureCount)];
+                                                  withRange:NSMakeRange(graphicsPipeline->header.num_vertex_samplers,
+                                                                        graphicsPipeline->header.num_vertex_storage_textures)];
        }
        commandBuffer->needVertexStorageTextureBind = false;
    }
@@ -2673,20 +2663,20 @@ static void METAL_INTERNAL_BindGraphicsResources(
    // Vertex Storage Buffers

    if (commandBuffer->needVertexStorageBufferBind) {
-        if (graphicsPipeline->vertexStorageBufferCount > 0) {
+        if (graphicsPipeline->header.num_vertex_storage_buffers > 0) {
            [commandBuffer->renderEncoder setVertexBuffers:commandBuffer->vertexStorageBuffers
                                                   offsets:offsets
-                                                 withRange:NSMakeRange(graphicsPipeline->vertexUniformBufferCount,
-                                                                       graphicsPipeline->vertexStorageBufferCount)];
+                                                 withRange:NSMakeRange(graphicsPipeline->header.num_vertex_uniform_buffers,
+                                                                       graphicsPipeline->header.num_vertex_storage_buffers)];
        }
        commandBuffer->needVertexStorageBufferBind = false;
    }

    // Vertex Uniform Buffers

-    for (Uint32 i = 0; i < graphicsPipeline->vertexUniformBufferCount; i += 1) {
+    for (Uint32 i = 0; i < graphicsPipeline->header.num_vertex_uniform_buffers; i += 1) {
        if (commandBuffer->needVertexUniformBufferBind[i]) {
-            if (graphicsPipeline->vertexUniformBufferCount > i) {
+            if (graphicsPipeline->header.num_vertex_uniform_buffers > i) {
                [commandBuffer->renderEncoder
                    setVertexBuffer:commandBuffer->vertexUniformBuffers[i]->handle
                             offset:commandBuffer->vertexUniformBuffers[i]->drawOffset
@@ -2699,11 +2689,11 @@ static void METAL_INTERNAL_BindGraphicsResources(
    // Fragment Samplers+Textures

    if (commandBuffer->needFragmentSamplerBind) {
-        if (graphicsPipeline->fragmentSamplerCount > 0) {
+        if (graphicsPipeline->header.num_fragment_samplers > 0) {
            [commandBuffer->renderEncoder setFragmentSamplerStates:commandBuffer->fragmentSamplers
-                                                         withRange:NSMakeRange(0, graphicsPipeline->fragmentSamplerCount)];
+                                                         withRange:NSMakeRange(0, graphicsPipeline->header.num_fragment_samplers)];
            [commandBuffer->renderEncoder setFragmentTextures:commandBuffer->fragmentTextures
-                                                    withRange:NSMakeRange(0, graphicsPipeline->fragmentSamplerCount)];
+                                                    withRange:NSMakeRange(0, graphicsPipeline->header.num_fragment_samplers)];
        }
        commandBuffer->needFragmentSamplerBind = false;
    }
@@ -2711,10 +2701,10 @@ static void METAL_INTERNAL_BindGraphicsResources(
    // Fragment Storage Textures

    if (commandBuffer->needFragmentStorageTextureBind) {
-        if (graphicsPipeline->fragmentStorageTextureCount > 0) {
+        if (graphicsPipeline->header.num_fragment_storage_textures > 0) {
            [commandBuffer->renderEncoder setFragmentTextures:commandBuffer->fragmentStorageTextures
-                                                    withRange:NSMakeRange(graphicsPipeline->fragmentSamplerCount,
-                                                                          graphicsPipeline->fragmentStorageTextureCount)];
+                                                    withRange:NSMakeRange(graphicsPipeline->header.num_fragment_samplers,
+                                                                          graphicsPipeline->header.num_fragment_storage_textures)];
        }
        commandBuffer->needFragmentStorageTextureBind = false;
    }
@@ -2722,20 +2712,20 @@ static void METAL_INTERNAL_BindGraphicsResources(
    // Fragment Storage Buffers

    if (commandBuffer->needFragmentStorageBufferBind) {
-        if (graphicsPipeline->fragmentStorageBufferCount > 0) {
+        if (graphicsPipeline->header.num_fragment_storage_buffers > 0) {
            [commandBuffer->renderEncoder setFragmentBuffers:commandBuffer->fragmentStorageBuffers
                                                     offsets:offsets
-                                                   withRange:NSMakeRange(graphicsPipeline->fragmentUniformBufferCount,
-                                                                         graphicsPipeline->fragmentStorageBufferCount)];
+                                                   withRange:NSMakeRange(graphicsPipeline->header.num_fragment_uniform_buffers,
+                                                                         graphicsPipeline->header.num_fragment_storage_buffers)];
        }
        commandBuffer->needFragmentStorageBufferBind = false;
    }

    // Fragment Uniform Buffers

-    for (Uint32 i = 0; i < graphicsPipeline->fragmentUniformBufferCount; i += 1) {
+    for (Uint32 i = 0; i < graphicsPipeline->header.num_fragment_uniform_buffers; i += 1) {
        if (commandBuffer->needFragmentUniformBufferBind[i]) {
-            if (graphicsPipeline->fragmentUniformBufferCount > i) {
+            if (graphicsPipeline->header.num_fragment_uniform_buffers > i) {
                [commandBuffer->renderEncoder
                    setFragmentBuffer:commandBuffer->fragmentUniformBuffers[i]->handle
                            offset:commandBuffer->fragmentUniformBuffers[i]->drawOffset
@@ -2754,38 +2744,38 @@ static void METAL_INTERNAL_BindComputeResources(
    NSUInteger offsets[MAX_STORAGE_BUFFERS_PER_STAGE] = { 0 };

    if (commandBuffer->needComputeSamplerBind) {
-        if (computePipeline->numSamplers > 0) {
+        if (computePipeline->header.numSamplers > 0) {
            [commandBuffer->computeEncoder setTextures:commandBuffer->computeSamplerTextures
-                                             withRange:NSMakeRange(0, computePipeline->numSamplers)];
+                                             withRange:NSMakeRange(0, computePipeline->header.numSamplers)];
            [commandBuffer->computeEncoder setSamplerStates:commandBuffer->computeSamplers
-                                                  withRange:NSMakeRange(0, computePipeline->numSamplers)];
+                                                  withRange:NSMakeRange(0, computePipeline->header.numSamplers)];
        }
        commandBuffer->needComputeSamplerBind = false;
    }

    if (commandBuffer->needComputeReadOnlyStorageTextureBind) {
-        if (computePipeline->numReadonlyStorageTextures > 0) {
+        if (computePipeline->header.numReadonlyStorageTextures > 0) {
            [commandBuffer->computeEncoder setTextures:commandBuffer->computeReadOnlyTextures
                                             withRange:NSMakeRange(
-                                                           computePipeline->numSamplers,
-                                                           computePipeline->numReadonlyStorageTextures)];
+                                                           computePipeline->header.numSamplers,
+                                                           computePipeline->header.numReadonlyStorageTextures)];
        }
        commandBuffer->needComputeReadOnlyStorageTextureBind = false;
    }

    if (commandBuffer->needComputeReadOnlyStorageBufferBind) {
-        if (computePipeline->numReadonlyStorageBuffers > 0) {
+        if (computePipeline->header.numReadonlyStorageBuffers > 0) {
            [commandBuffer->computeEncoder setBuffers:commandBuffer->computeReadOnlyBuffers
                                              offsets:offsets
-                                            withRange:NSMakeRange(computePipeline->numUniformBuffers,
-                                                                  computePipeline->numReadonlyStorageBuffers)];
+                                            withRange:NSMakeRange(computePipeline->header.numUniformBuffers,
+                                                                  computePipeline->header.numReadonlyStorageBuffers)];
        }
        commandBuffer->needComputeReadOnlyStorageBufferBind = false;
    }

    for (Uint32 i = 0; i < MAX_UNIFORM_BUFFERS_PER_STAGE; i += 1) {
        if (commandBuffer->needComputeUniformBufferBind[i]) {
-            if (computePipeline->numUniformBuffers > i) {
+            if (computePipeline->header.numUniformBuffers > i) {
                [commandBuffer->computeEncoder
                    setBuffer:commandBuffer->computeUniformBuffers[i]->handle
                    offset:commandBuffer->computeUniformBuffers[i]->drawOffset
@@ -3133,7 +3123,7 @@ static void METAL_BindComputePipeline(
            metalCommandBuffer->needComputeUniformBufferBind[i] = true;
        }

-        for (Uint32 i = 0; i < pipeline->numUniformBuffers; i += 1) {
+        for (Uint32 i = 0; i < pipeline->header.numUniformBuffers; i += 1) {
            if (metalCommandBuffer->computeUniformBuffers[i] == NULL) {
                metalCommandBuffer->computeUniformBuffers[i] = METAL_INTERNAL_AcquireUniformBufferFromPool(
                    metalCommandBuffer);
@@ -3141,22 +3131,22 @@ static void METAL_BindComputePipeline(
        }

        // Bind write-only resources
-        if (pipeline->numReadWriteStorageTextures > 0) {
+        if (pipeline->header.numReadWriteStorageTextures > 0) {
            [metalCommandBuffer->computeEncoder setTextures:metalCommandBuffer->computeReadWriteTextures
                                                  withRange:NSMakeRange(
-                                                        pipeline->numSamplers +
-                                                            pipeline->numReadonlyStorageTextures,
-                                                        pipeline->numReadWriteStorageTextures)];
+                                                        pipeline->header.numSamplers +
+                                                            pipeline->header.numReadonlyStorageTextures,
+                                                        pipeline->header.numReadWriteStorageTextures)];
        }

        NSUInteger offsets[MAX_COMPUTE_WRITE_BUFFERS] = { 0 };
-        if (pipeline->numReadWriteStorageBuffers > 0) {
+        if (pipeline->header.numReadWriteStorageBuffers > 0) {
            [metalCommandBuffer->computeEncoder setBuffers:metalCommandBuffer->computeReadWriteBuffers
                                                   offsets:offsets
                                                 withRange:NSMakeRange(
-                                                        pipeline->numUniformBuffers +
-                                                            pipeline->numReadonlyStorageBuffers,
-                                                        pipeline->numReadWriteStorageBuffers)];
+                                                        pipeline->header.numUniformBuffers +
+                                                            pipeline->header.numReadonlyStorageBuffers,
+                                                        pipeline->header.numReadWriteStorageBuffers)];
        }
    }
 }
@@ -4570,6 +4560,7 @@ static SDL_GPUDevice *METAL_CreateDevice(bool debugMode, bool preferLowPower, SD
        SDL_GPUDevice *result = SDL_calloc(1, sizeof(SDL_GPUDevice));
        ASSIGN_DRIVER(METAL)
        result->driverData = (SDL_GPURenderer *)renderer;
+        result->shader_formats = SDL_GPU_SHADERFORMAT_MSL | SDL_GPU_SHADERFORMAT_METALLIB;
        renderer->sdlGPUDevice = result;

        return result;
@@ -732,7 +732,7 @@ typedef struct WindowData

    // Synchronization primitives
    VkSemaphore imageAvailableSemaphore[MAX_FRAMES_IN_FLIGHT];
-    VkSemaphore renderFinishedSemaphore[MAX_FRAMES_IN_FLIGHT];
+    VkSemaphore *renderFinishedSemaphore;
    SDL_GPUFence *inFlightFences[MAX_FRAMES_IN_FLIGHT];

    Uint32 frameCounter;
@@ -822,13 +822,13 @@ typedef struct DescriptorSetLayout
 typedef struct GraphicsPipelineResourceLayoutHashTableKey
 {
    Uint32 vertexSamplerCount;
-    Uint32 vertexStorageBufferCount;
    Uint32 vertexStorageTextureCount;
+    Uint32 vertexStorageBufferCount;
    Uint32 vertexUniformBufferCount;

    Uint32 fragmentSamplerCount;
-    Uint32 fragmentStorageBufferCount;
    Uint32 fragmentStorageTextureCount;
+    Uint32 fragmentStorageBufferCount;
    Uint32 fragmentUniformBufferCount;
 } GraphicsPipelineResourceLayoutHashTableKey;

@@ -846,18 +846,20 @@ typedef struct VulkanGraphicsPipelineResourceLayout
    DescriptorSetLayout *descriptorSetLayouts[4];

    Uint32 vertexSamplerCount;
-    Uint32 vertexStorageBufferCount;
    Uint32 vertexStorageTextureCount;
+    Uint32 vertexStorageBufferCount;
    Uint32 vertexUniformBufferCount;

    Uint32 fragmentSamplerCount;
-    Uint32 fragmentStorageBufferCount;
    Uint32 fragmentStorageTextureCount;
+    Uint32 fragmentStorageBufferCount;
    Uint32 fragmentUniformBufferCount;
 } VulkanGraphicsPipelineResourceLayout;

 typedef struct VulkanGraphicsPipeline
 {
+    GraphicsPipelineCommonHeader header;
+
    VkPipeline pipeline;
    SDL_GPUPrimitiveType primitiveType;

@@ -901,6 +903,8 @@ typedef struct VulkanComputePipelineResourceLayout

 typedef struct VulkanComputePipeline
 {
+    ComputePipelineCommonHeader header;
+
    VkShaderModule shaderModule;
    VkPipeline pipeline;
    VulkanComputePipelineResourceLayout *resourceLayout;
@@ -1038,25 +1042,33 @@ typedef struct VulkanCommandBuffer
    Uint32 vertexBufferCount;
    bool needVertexBufferBind;

-    VulkanTexture *vertexSamplerTextures[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    VulkanSampler *vertexSamplers[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    VulkanTexture *vertexStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
-    VulkanBuffer *vertexStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
+    VkImageView vertexSamplerTextureViewBindings[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    VkSampler vertexSamplerBindings[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    VkImageView vertexStorageTextureViewBindings[MAX_STORAGE_TEXTURES_PER_STAGE];
+    VkBuffer vertexStorageBufferBindings[MAX_STORAGE_BUFFERS_PER_STAGE];

-    VulkanTexture *fragmentSamplerTextures[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    VulkanSampler *fragmentSamplers[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    VulkanTexture *fragmentStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
-    VulkanBuffer *fragmentStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
+    VkImageView fragmentSamplerTextureViewBindings[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    VkSampler fragmentSamplerBindings[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    VkImageView fragmentStorageTextureViewBindings[MAX_STORAGE_TEXTURES_PER_STAGE];
+    VkBuffer fragmentStorageBufferBindings[MAX_STORAGE_BUFFERS_PER_STAGE];

+    VkImageView computeSamplerTextureViewBindings[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    VkSampler computeSamplerBindings[MAX_TEXTURE_SAMPLERS_PER_STAGE];
+    VkImageView readOnlyComputeStorageTextureViewBindings[MAX_STORAGE_TEXTURES_PER_STAGE];
+    VkBuffer readOnlyComputeStorageBufferBindings[MAX_STORAGE_BUFFERS_PER_STAGE];
+
+    // Track these separately because barriers can happen mid compute pass
+    VulkanTexture *readOnlyComputeStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
+    VulkanBuffer *readOnlyComputeStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
+
+    VkImageView readWriteComputeStorageTextureViewBindings[MAX_COMPUTE_WRITE_TEXTURES];
+    VkBuffer readWriteComputeStorageBufferBindings[MAX_COMPUTE_WRITE_BUFFERS];
+
+    // Track these separately because they are barriered when the compute pass begins
    VulkanTextureSubresource *readWriteComputeStorageTextureSubresources[MAX_COMPUTE_WRITE_TEXTURES];
    Uint32 readWriteComputeStorageTextureSubresourceCount;
    VulkanBuffer *readWriteComputeStorageBuffers[MAX_COMPUTE_WRITE_BUFFERS];

-    VulkanTexture *computeSamplerTextures[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    VulkanSampler *computeSamplers[MAX_TEXTURE_SAMPLERS_PER_STAGE];
-    VulkanTexture *readOnlyComputeStorageTextures[MAX_STORAGE_TEXTURES_PER_STAGE];
-    VulkanBuffer *readOnlyComputeStorageBuffers[MAX_STORAGE_BUFFERS_PER_STAGE];
-
    // Uniform buffers

    VulkanUniformBuffer *vertexUniformBuffers[MAX_UNIFORM_BUFFERS_PER_STAGE];
@@ -1096,6 +1108,7 @@ typedef struct VulkanCommandBuffer
    VulkanFenceHandle *inFlightFence;
    bool autoReleaseFence;

+    bool swapchainRequested;
    bool isDefrag; // Whether this CB was created for defragging
 } VulkanCommandBuffer;

@@ -1130,6 +1143,7 @@ struct VulkanRenderer
    VulkanExtensions supports;
    bool supportsDebugUtils;
    bool supportsColorspace;
+    bool supportsPhysicalDeviceProperties2;
    bool supportsFillModeNonSolid;
    bool supportsMultiDrawIndirect;

@@ -1265,13 +1279,20 @@ static inline const char *VkErrorMessages(VkResult code)
 #undef ERR_TO_STR
 }

-#define SET_ERROR_AND_RETURN(fmt, msg, ret)               \
+#define SET_ERROR(fmt, msg)                               \
    do {                                                  \
        if (renderer->debugMode) {                        \
            SDL_LogError(SDL_LOG_CATEGORY_GPU, fmt, msg); \
        }                                                 \
        SDL_SetError((fmt), (msg));                       \
-        return ret;                                       \
+    } while (0)
+
+#define SET_STRING_ERROR(msg) SET_ERROR("%s", msg)
+
+#define SET_ERROR_AND_RETURN(fmt, msg, ret) \
+    do {                                    \
+        SET_ERROR(fmt, msg);                \
+        return ret;                         \
    } while (0)

 #define SET_STRING_ERROR_AND_RETURN(msg, ret) SET_ERROR_AND_RETURN("%s", msg, ret)
@@ -3217,7 +3238,6 @@ static void VULKAN_INTERNAL_DestroySwapchain(
        SDL_free(windowData->textureContainers[i].activeTexture->subresources);
        SDL_free(windowData->textureContainers[i].activeTexture);
    }
-    windowData->imageCount = 0;

    SDL_free(windowData->textureContainers);
    windowData->textureContainers = NULL;
@@ -3246,7 +3266,8 @@ static void VULKAN_INTERNAL_DestroySwapchain(
                NULL);
            windowData->imageAvailableSemaphore[i] = VK_NULL_HANDLE;
        }
-
+    }
+    for (i = 0; i < windowData->imageCount; i += 1) {
        if (windowData->renderFinishedSemaphore[i]) {
            renderer->vkDestroySemaphore(
                renderer->logicalDevice,
@@ -3255,6 +3276,10 @@ static void VULKAN_INTERNAL_DestroySwapchain(
            windowData->renderFinishedSemaphore[i] = VK_NULL_HANDLE;
        }
    }
+    SDL_free(windowData->renderFinishedSemaphore);
+    windowData->renderFinishedSemaphore = NULL;
+
+    windowData->imageCount = 0;
 }

 static void VULKAN_INTERNAL_DestroyGraphicsPipelineResourceLayout(
@@ -4812,6 +4837,12 @@ static Uint32 VULKAN_INTERNAL_CreateSwapchain(
            CHECK_VULKAN_ERROR_AND_RETURN(vulkanResult, vkCreateSemaphore, false);
        }

+        windowData->inFlightFences[i] = NULL;
+    }
+
+    windowData->renderFinishedSemaphore = SDL_malloc(
+        sizeof(VkSemaphore) * windowData->imageCount);
+    for (i = 0; i < windowData->imageCount; i += 1) {
        vulkanResult = renderer->vkCreateSemaphore(
            renderer->logicalDevice,
            &semaphoreCreateInfo,
@@ -4831,8 +4862,6 @@ static Uint32 VULKAN_INTERNAL_CreateSwapchain(
            windowData->swapchain = VK_NULL_HANDLE;
            CHECK_VULKAN_ERROR_AND_RETURN(vulkanResult, vkCreateSemaphore, false);
        }
-
-        windowData->inFlightFences[i] = NULL;
    }

    windowData->needsSwapchainRecreate = false;
@@ -5118,8 +5147,8 @@ static void VULKAN_INTERNAL_BindGraphicsDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pBufferInfo = NULL;

-            imageInfos[imageInfoCount].sampler = commandBuffer->vertexSamplers[i]->sampler;
-            imageInfos[imageInfoCount].imageView = commandBuffer->vertexSamplerTextures[i]->fullView;
+            imageInfos[imageInfoCount].sampler = commandBuffer->vertexSamplerBindings[i];
+            imageInfos[imageInfoCount].imageView = commandBuffer->vertexSamplerTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -5142,7 +5171,7 @@ static void VULKAN_INTERNAL_BindGraphicsDescriptorSets(
            currentWriteDescriptorSet->pBufferInfo = NULL;

            imageInfos[imageInfoCount].sampler = VK_NULL_HANDLE;
-            imageInfos[imageInfoCount].imageView = commandBuffer->vertexStorageTextures[i]->fullView;
+            imageInfos[imageInfoCount].imageView = commandBuffer->vertexStorageTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_GENERAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -5164,7 +5193,7 @@ static void VULKAN_INTERNAL_BindGraphicsDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pImageInfo = NULL;

-            bufferInfos[bufferInfoCount].buffer = commandBuffer->vertexStorageBuffers[i]->buffer;
+            bufferInfos[bufferInfoCount].buffer = commandBuffer->vertexStorageBufferBindings[i];
            bufferInfos[bufferInfoCount].offset = 0;
            bufferInfos[bufferInfoCount].range = VK_WHOLE_SIZE;

@@ -5237,8 +5266,8 @@ static void VULKAN_INTERNAL_BindGraphicsDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pBufferInfo = NULL;

-            imageInfos[imageInfoCount].sampler = commandBuffer->fragmentSamplers[i]->sampler;
-            imageInfos[imageInfoCount].imageView = commandBuffer->fragmentSamplerTextures[i]->fullView;
+            imageInfos[imageInfoCount].sampler = commandBuffer->fragmentSamplerBindings[i];
+            imageInfos[imageInfoCount].imageView = commandBuffer->fragmentSamplerTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -5261,7 +5290,7 @@ static void VULKAN_INTERNAL_BindGraphicsDescriptorSets(
            currentWriteDescriptorSet->pBufferInfo = NULL;

            imageInfos[imageInfoCount].sampler = VK_NULL_HANDLE;
-            imageInfos[imageInfoCount].imageView = commandBuffer->fragmentStorageTextures[i]->fullView;
+            imageInfos[imageInfoCount].imageView = commandBuffer->fragmentStorageTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_GENERAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -5283,7 +5312,7 @@ static void VULKAN_INTERNAL_BindGraphicsDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pImageInfo = NULL;

-            bufferInfos[bufferInfoCount].buffer = commandBuffer->fragmentStorageBuffers[i]->buffer;
+            bufferInfos[bufferInfoCount].buffer = commandBuffer->fragmentStorageBufferBindings[i];
            bufferInfos[bufferInfoCount].offset = 0;
            bufferInfos[bufferInfoCount].range = VK_WHOLE_SIZE;

@@ -5864,7 +5893,10 @@ static VulkanTexture *VULKAN_INTERNAL_CreateTexture(
            VULKAN_TEXTURE_USAGE_MODE_UNINITIALIZED,
            texture);
        VULKAN_INTERNAL_TrackTexture(barrierCommandBuffer, texture);
-        VULKAN_Submit((SDL_GPUCommandBuffer *)barrierCommandBuffer);
+        if (!VULKAN_Submit((SDL_GPUCommandBuffer *)barrierCommandBuffer)) {
+            VULKAN_INTERNAL_DestroyTexture(renderer, texture);
+            return NULL;
+        }
    }

    return texture;
@@ -6061,7 +6093,6 @@ static VkRenderPass VULKAN_INTERNAL_CreateRenderPass(
        colorAttachmentReferences[colorAttachmentReferenceCount].layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

        attachmentDescriptionCount += 1;
-        colorAttachmentReferenceCount += 1;

        if (colorTargetInfos[i].store_op == SDL_GPU_STOREOP_RESOLVE || colorTargetInfos[i].store_op == SDL_GPU_STOREOP_RESOLVE_AND_STORE) {
            VulkanTextureContainer *resolveContainer = (VulkanTextureContainer *)colorTargetInfos[i].resolve_texture;
@@ -6076,12 +6107,16 @@ static VkRenderPass VULKAN_INTERNAL_CreateRenderPass(
            attachmentDescriptions[attachmentDescriptionCount].initialLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
            attachmentDescriptions[attachmentDescriptionCount].finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

-            resolveReferences[resolveReferenceCount].attachment = attachmentDescriptionCount;
-            resolveReferences[resolveReferenceCount].layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
+            resolveReferences[colorAttachmentReferenceCount].attachment = attachmentDescriptionCount;
+            resolveReferences[colorAttachmentReferenceCount].layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

            attachmentDescriptionCount += 1;
            resolveReferenceCount += 1;
+        } else {
+            resolveReferences[colorAttachmentReferenceCount].attachment = VK_ATTACHMENT_UNUSED;
        }
+
+        colorAttachmentReferenceCount += 1;
    }

    subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
@@ -6557,6 +6592,16 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
            &nameInfo);
    }

+    // Put this data in the pipeline we can do validation in gpu.c
+    graphicsPipeline->header.num_vertex_samplers = graphicsPipeline->resourceLayout->vertexSamplerCount;
+    graphicsPipeline->header.num_vertex_storage_buffers = graphicsPipeline->resourceLayout->vertexStorageBufferCount;
+    graphicsPipeline->header.num_vertex_storage_textures = graphicsPipeline->resourceLayout->vertexStorageTextureCount;
+    graphicsPipeline->header.num_vertex_uniform_buffers = graphicsPipeline->resourceLayout->vertexUniformBufferCount;
+    graphicsPipeline->header.num_fragment_samplers = graphicsPipeline->resourceLayout->fragmentSamplerCount;
+    graphicsPipeline->header.num_fragment_storage_buffers = graphicsPipeline->resourceLayout->fragmentStorageBufferCount;
+    graphicsPipeline->header.num_fragment_storage_textures = graphicsPipeline->resourceLayout->fragmentStorageTextureCount;
+    graphicsPipeline->header.num_fragment_uniform_buffers = graphicsPipeline->resourceLayout->fragmentUniformBufferCount;
+
    return (SDL_GPUGraphicsPipeline *)graphicsPipeline;
 }

@@ -6651,6 +6696,14 @@ static SDL_GPUComputePipeline *VULKAN_CreateComputePipeline(
            &nameInfo);
    }

+    // Track these here for debug layer
+    vulkanComputePipeline->header.numSamplers = vulkanComputePipeline->resourceLayout->numSamplers;
+    vulkanComputePipeline->header.numReadonlyStorageTextures = vulkanComputePipeline->resourceLayout->numReadonlyStorageTextures;
+    vulkanComputePipeline->header.numReadonlyStorageBuffers = vulkanComputePipeline->resourceLayout->numReadonlyStorageBuffers;
+    vulkanComputePipeline->header.numReadWriteStorageTextures = vulkanComputePipeline->resourceLayout->numReadWriteStorageTextures;
+    vulkanComputePipeline->header.numReadWriteStorageBuffers = vulkanComputePipeline->resourceLayout->numReadWriteStorageBuffers;
+    vulkanComputePipeline->header.numUniformBuffers = vulkanComputePipeline->resourceLayout->numUniformBuffers;
+
    return (SDL_GPUComputePipeline *)vulkanComputePipeline;
 }

@@ -6913,6 +6966,8 @@ static void VULKAN_ReleaseTexture(
        VULKAN_INTERNAL_ReleaseTexture(renderer, vulkanTextureContainer->textures[i]);
    }

+    SDL_DestroyProperties(vulkanTextureContainer->header.info.props);
+
    // Containers are just client handles, so we can destroy immediately
    if (vulkanTextureContainer->debugName != NULL) {
        SDL_free(vulkanTextureContainer->debugName);
@@ -7114,6 +7169,8 @@ static VkRenderPass VULKAN_INTERNAL_FetchRenderPass(
    key.sampleCount = VK_SAMPLE_COUNT_1_BIT;
    if (numColorTargets > 0) {
        key.sampleCount = SDLToVK_SampleCount[((VulkanTextureContainer *)colorTargetInfos[0].texture)->header.info.sample_count];
+    } else if (numColorTargets == 0 && depthStencilTargetInfo != NULL) {
+        key.sampleCount = SDLToVK_SampleCount[((VulkanTextureContainer *)depthStencilTargetInfo->texture)->header.info.sample_count];
    }

    key.numColorTargets = numColorTargets;
@@ -7449,21 +7506,21 @@ static void VULKAN_BindVertexSamplers(
        VulkanTextureContainer *textureContainer = (VulkanTextureContainer *)textureSamplerBindings[i].texture;
        VulkanSampler *sampler = (VulkanSampler *)textureSamplerBindings[i].sampler;

-        if (vulkanCommandBuffer->vertexSamplers[firstSlot + i] != sampler) {
+        if (vulkanCommandBuffer->vertexSamplerBindings[firstSlot + i] != sampler->sampler) {
            VULKAN_INTERNAL_TrackSampler(
                vulkanCommandBuffer,
                (VulkanSampler *)textureSamplerBindings[i].sampler);

-            vulkanCommandBuffer->vertexSamplers[firstSlot + i] = (VulkanSampler *)textureSamplerBindings[i].sampler;
+            vulkanCommandBuffer->vertexSamplerBindings[firstSlot + i] = sampler->sampler;
            vulkanCommandBuffer->needNewVertexResourceDescriptorSet = true;
        }

-        if (vulkanCommandBuffer->vertexSamplerTextures[firstSlot + i] != textureContainer->activeTexture) {
+        if (vulkanCommandBuffer->vertexSamplerTextureViewBindings[firstSlot + i] != textureContainer->activeTexture->fullView) {
            VULKAN_INTERNAL_TrackTexture(
                vulkanCommandBuffer,
                textureContainer->activeTexture);

-            vulkanCommandBuffer->vertexSamplerTextures[firstSlot + i] = textureContainer->activeTexture;
+            vulkanCommandBuffer->vertexSamplerTextureViewBindings[firstSlot + i] = textureContainer->activeTexture->fullView;
            vulkanCommandBuffer->needNewVertexResourceDescriptorSet = true;
        }
    }
@@ -7480,12 +7537,12 @@ static void VULKAN_BindVertexStorageTextures(
    for (Uint32 i = 0; i < numBindings; i += 1) {
        VulkanTextureContainer *textureContainer = (VulkanTextureContainer *)storageTextures[i];

-        if (vulkanCommandBuffer->vertexStorageTextures[firstSlot + i] != textureContainer->activeTexture) {
+        if (vulkanCommandBuffer->vertexStorageTextureViewBindings[firstSlot + i] != textureContainer->activeTexture->fullView) {
            VULKAN_INTERNAL_TrackTexture(
                vulkanCommandBuffer,
                textureContainer->activeTexture);

-            vulkanCommandBuffer->vertexStorageTextures[firstSlot + i] = textureContainer->activeTexture;
+            vulkanCommandBuffer->vertexStorageTextureViewBindings[firstSlot + i] = textureContainer->activeTexture->fullView;
            vulkanCommandBuffer->needNewVertexResourceDescriptorSet = true;
        }
    }
@@ -7502,12 +7559,12 @@ static void VULKAN_BindVertexStorageBuffers(
    for (Uint32 i = 0; i < numBindings; i += 1) {
        VulkanBufferContainer *bufferContainer = (VulkanBufferContainer *)storageBuffers[i];

-        if (vulkanCommandBuffer->vertexStorageBuffers[firstSlot + i] != bufferContainer->activeBuffer) {
+        if (vulkanCommandBuffer->vertexStorageBufferBindings[firstSlot + i] != bufferContainer->activeBuffer->buffer) {
            VULKAN_INTERNAL_TrackBuffer(
                vulkanCommandBuffer,
                bufferContainer->activeBuffer);

-            vulkanCommandBuffer->vertexStorageBuffers[firstSlot + i] = bufferContainer->activeBuffer;
+            vulkanCommandBuffer->vertexStorageBufferBindings[firstSlot + i] = bufferContainer->activeBuffer->buffer;
            vulkanCommandBuffer->needNewVertexResourceDescriptorSet = true;
        }
    }
@@ -7525,21 +7582,21 @@ static void VULKAN_BindFragmentSamplers(
        VulkanTextureContainer *textureContainer = (VulkanTextureContainer *)textureSamplerBindings[i].texture;
        VulkanSampler *sampler = (VulkanSampler *)textureSamplerBindings[i].sampler;

-        if (vulkanCommandBuffer->fragmentSamplers[firstSlot + i] != sampler) {
+        if (vulkanCommandBuffer->fragmentSamplerBindings[firstSlot + i] != sampler->sampler) {
            VULKAN_INTERNAL_TrackSampler(
                vulkanCommandBuffer,
                (VulkanSampler *)textureSamplerBindings[i].sampler);

-            vulkanCommandBuffer->fragmentSamplers[firstSlot + i] = (VulkanSampler *)textureSamplerBindings[i].sampler;
+            vulkanCommandBuffer->fragmentSamplerBindings[firstSlot + i] = sampler->sampler;
            vulkanCommandBuffer->needNewFragmentResourceDescriptorSet = true;
        }

-        if (vulkanCommandBuffer->fragmentSamplerTextures[firstSlot + i] != textureContainer->activeTexture) {
+        if (vulkanCommandBuffer->fragmentSamplerTextureViewBindings[firstSlot + i] != textureContainer->activeTexture->fullView) {
            VULKAN_INTERNAL_TrackTexture(
                vulkanCommandBuffer,
                textureContainer->activeTexture);

-            vulkanCommandBuffer->fragmentSamplerTextures[firstSlot + i] = textureContainer->activeTexture;
+            vulkanCommandBuffer->fragmentSamplerTextureViewBindings[firstSlot + i] = textureContainer->activeTexture->fullView;
            vulkanCommandBuffer->needNewFragmentResourceDescriptorSet = true;
        }
    }
@@ -7556,12 +7613,12 @@ static void VULKAN_BindFragmentStorageTextures(
    for (Uint32 i = 0; i < numBindings; i += 1) {
        VulkanTextureContainer *textureContainer = (VulkanTextureContainer *)storageTextures[i];

-        if (vulkanCommandBuffer->fragmentStorageTextures[firstSlot + i] != textureContainer->activeTexture) {
+        if (vulkanCommandBuffer->fragmentStorageTextureViewBindings[firstSlot + i] != textureContainer->activeTexture->fullView) {
            VULKAN_INTERNAL_TrackTexture(
                vulkanCommandBuffer,
                textureContainer->activeTexture);

-            vulkanCommandBuffer->fragmentStorageTextures[firstSlot + i] = textureContainer->activeTexture;
+            vulkanCommandBuffer->fragmentStorageTextureViewBindings[firstSlot + i] = textureContainer->activeTexture->fullView;
            vulkanCommandBuffer->needNewFragmentResourceDescriptorSet = true;
        }
    }
@@ -7580,12 +7637,12 @@ static void VULKAN_BindFragmentStorageBuffers(
    for (i = 0; i < numBindings; i += 1) {
        bufferContainer = (VulkanBufferContainer *)storageBuffers[i];

-        if (vulkanCommandBuffer->fragmentStorageBuffers[firstSlot + i] != bufferContainer->activeBuffer) {
+        if (vulkanCommandBuffer->fragmentStorageBufferBindings[firstSlot + i] != bufferContainer->activeBuffer->buffer) {
            VULKAN_INTERNAL_TrackBuffer(
                vulkanCommandBuffer,
                bufferContainer->activeBuffer);

-            vulkanCommandBuffer->fragmentStorageBuffers[firstSlot + i] = bufferContainer->activeBuffer;
+            vulkanCommandBuffer->fragmentStorageBufferBindings[firstSlot + i] = bufferContainer->activeBuffer->buffer;
            vulkanCommandBuffer->needNewFragmentResourceDescriptorSet = true;
        }
    }
@@ -7858,15 +7915,17 @@ static void VULKAN_BeginRenderPass(

    clearValues = SDL_stack_alloc(VkClearValue, clearCount);

-    for (i = 0; i < totalColorAttachmentCount; i += 1) {
-        clearValues[i].color.float32[0] = colorTargetInfos[i].clear_color.r;
-        clearValues[i].color.float32[1] = colorTargetInfos[i].clear_color.g;
-        clearValues[i].color.float32[2] = colorTargetInfos[i].clear_color.b;
-        clearValues[i].color.float32[3] = colorTargetInfos[i].clear_color.a;
+    int clearIndex = 0;
+    for (i = 0; i < numColorTargets; i += 1) {
+        clearValues[clearIndex].color.float32[0] = colorTargetInfos[i].clear_color.r;
+        clearValues[clearIndex].color.float32[1] = colorTargetInfos[i].clear_color.g;
+        clearValues[clearIndex].color.float32[2] = colorTargetInfos[i].clear_color.b;
+        clearValues[clearIndex].color.float32[3] = colorTargetInfos[i].clear_color.a;
+        clearIndex += 1;

        if (colorTargetInfos[i].store_op == SDL_GPU_STOREOP_RESOLVE || colorTargetInfos[i].store_op == SDL_GPU_STOREOP_RESOLVE_AND_STORE) {
            // Skip over the resolve texture, we're not clearing it
-            i += 1;
+            clearIndex += 1;
        }
    }

@@ -8099,15 +8158,15 @@ static void VULKAN_EndRenderPass(
    SDL_zeroa(vulkanCommandBuffer->vertexBufferOffsets);
    vulkanCommandBuffer->vertexBufferCount = 0;

-    SDL_zeroa(vulkanCommandBuffer->vertexSamplers);
-    SDL_zeroa(vulkanCommandBuffer->vertexSamplerTextures);
-    SDL_zeroa(vulkanCommandBuffer->vertexStorageTextures);
-    SDL_zeroa(vulkanCommandBuffer->vertexStorageBuffers);
+    SDL_zeroa(vulkanCommandBuffer->vertexSamplerBindings);
+    SDL_zeroa(vulkanCommandBuffer->vertexSamplerTextureViewBindings);
+    SDL_zeroa(vulkanCommandBuffer->vertexStorageTextureViewBindings);
+    SDL_zeroa(vulkanCommandBuffer->vertexStorageBufferBindings);

-    SDL_zeroa(vulkanCommandBuffer->fragmentSamplers);
-    SDL_zeroa(vulkanCommandBuffer->fragmentSamplerTextures);
-    SDL_zeroa(vulkanCommandBuffer->fragmentStorageTextures);
-    SDL_zeroa(vulkanCommandBuffer->fragmentStorageBuffers);
+    SDL_zeroa(vulkanCommandBuffer->fragmentSamplerBindings);
+    SDL_zeroa(vulkanCommandBuffer->fragmentSamplerTextureViewBindings);
+    SDL_zeroa(vulkanCommandBuffer->fragmentStorageTextureViewBindings);
+    SDL_zeroa(vulkanCommandBuffer->fragmentStorageBufferBindings);
 }

 static void VULKAN_BeginComputePass(
@@ -8137,6 +8196,7 @@ static void VULKAN_BeginComputePass(
            VULKAN_TEXTURE_USAGE_MODE_COMPUTE_STORAGE_READ_WRITE);

        vulkanCommandBuffer->readWriteComputeStorageTextureSubresources[i] = subresource;
+        vulkanCommandBuffer->readWriteComputeStorageTextureViewBindings[i] = subresource->computeWriteView;

        VULKAN_INTERNAL_TrackTexture(
            vulkanCommandBuffer,
@@ -8150,9 +8210,10 @@ static void VULKAN_BeginComputePass(
            vulkanCommandBuffer,
            bufferContainer,
            storageBufferBindings[i].cycle,
-            VULKAN_BUFFER_USAGE_MODE_COMPUTE_STORAGE_READ);
+            VULKAN_BUFFER_USAGE_MODE_COMPUTE_STORAGE_READ_WRITE);

        vulkanCommandBuffer->readWriteComputeStorageBuffers[i] = buffer;
+        vulkanCommandBuffer->readWriteComputeStorageBufferBindings[i] = buffer->buffer;

        VULKAN_INTERNAL_TrackBuffer(
            vulkanCommandBuffer,
@@ -8204,21 +8265,21 @@ static void VULKAN_BindComputeSamplers(
        VulkanTextureContainer *textureContainer = (VulkanTextureContainer *)textureSamplerBindings[i].texture;
        VulkanSampler *sampler = (VulkanSampler *)textureSamplerBindings[i].sampler;

-        if (vulkanCommandBuffer->computeSamplers[firstSlot + i] != sampler) {
+        if (vulkanCommandBuffer->computeSamplerBindings[firstSlot + i] != sampler->sampler) {
            VULKAN_INTERNAL_TrackSampler(
                vulkanCommandBuffer,
                sampler);

-            vulkanCommandBuffer->computeSamplers[firstSlot + i] = sampler;
+            vulkanCommandBuffer->computeSamplerBindings[firstSlot + i] = sampler->sampler;
            vulkanCommandBuffer->needNewComputeReadOnlyDescriptorSet = true;
        }

-        if (vulkanCommandBuffer->computeSamplerTextures[firstSlot + i] != textureContainer->activeTexture) {
+        if (vulkanCommandBuffer->computeSamplerTextureViewBindings[firstSlot + i] != textureContainer->activeTexture->fullView) {
            VULKAN_INTERNAL_TrackTexture(
                vulkanCommandBuffer,
                textureContainer->activeTexture);

-            vulkanCommandBuffer->computeSamplerTextures[firstSlot + i] = textureContainer->activeTexture;
+            vulkanCommandBuffer->computeSamplerTextureViewBindings[firstSlot + i] = textureContainer->activeTexture->fullView;
            vulkanCommandBuffer->needNewComputeReadOnlyDescriptorSet = true;
        }
    }
@@ -8259,6 +8320,7 @@ static void VULKAN_BindComputeStorageTextures(
                textureContainer->activeTexture);

            vulkanCommandBuffer->readOnlyComputeStorageTextures[firstSlot + i] = textureContainer->activeTexture;
+            vulkanCommandBuffer->readOnlyComputeStorageTextureViewBindings[firstSlot + i] = textureContainer->activeTexture->fullView;
            vulkanCommandBuffer->needNewComputeReadOnlyDescriptorSet = true;
        }
    }
@@ -8298,6 +8360,7 @@ static void VULKAN_BindComputeStorageBuffers(
                bufferContainer->activeBuffer);

            vulkanCommandBuffer->readOnlyComputeStorageBuffers[firstSlot + i] = bufferContainer->activeBuffer;
+            vulkanCommandBuffer->readOnlyComputeStorageBufferBindings[firstSlot + i] = bufferContainer->activeBuffer->buffer;
            vulkanCommandBuffer->needNewComputeReadOnlyDescriptorSet = true;
        }
    }
@@ -8372,8 +8435,8 @@ static void VULKAN_INTERNAL_BindComputeDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pBufferInfo = NULL;

-            imageInfos[imageInfoCount].sampler = commandBuffer->computeSamplers[i]->sampler;
-            imageInfos[imageInfoCount].imageView = commandBuffer->computeSamplerTextures[i]->fullView;
+            imageInfos[imageInfoCount].sampler = commandBuffer->computeSamplerBindings[i];
+            imageInfos[imageInfoCount].imageView = commandBuffer->computeSamplerTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -8396,7 +8459,7 @@ static void VULKAN_INTERNAL_BindComputeDescriptorSets(
            currentWriteDescriptorSet->pBufferInfo = NULL;

            imageInfos[imageInfoCount].sampler = VK_NULL_HANDLE;
-            imageInfos[imageInfoCount].imageView = commandBuffer->readOnlyComputeStorageTextures[i]->fullView;
+            imageInfos[imageInfoCount].imageView = commandBuffer->readOnlyComputeStorageTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_GENERAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -8418,7 +8481,7 @@ static void VULKAN_INTERNAL_BindComputeDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pImageInfo = NULL;

-            bufferInfos[bufferInfoCount].buffer = commandBuffer->readOnlyComputeStorageBuffers[i]->buffer;
+            bufferInfos[bufferInfoCount].buffer = commandBuffer->readOnlyComputeStorageBufferBindings[i];
            bufferInfos[bufferInfoCount].offset = 0;
            bufferInfos[bufferInfoCount].range = VK_WHOLE_SIZE;

@@ -8453,7 +8516,7 @@ static void VULKAN_INTERNAL_BindComputeDescriptorSets(
            currentWriteDescriptorSet->pBufferInfo = NULL;

            imageInfos[imageInfoCount].sampler = VK_NULL_HANDLE;
-            imageInfos[imageInfoCount].imageView = commandBuffer->readWriteComputeStorageTextureSubresources[i]->computeWriteView;
+            imageInfos[imageInfoCount].imageView = commandBuffer->readWriteComputeStorageTextureViewBindings[i];
            imageInfos[imageInfoCount].imageLayout = VK_IMAGE_LAYOUT_GENERAL;

            currentWriteDescriptorSet->pImageInfo = &imageInfos[imageInfoCount];
@@ -8475,7 +8538,7 @@ static void VULKAN_INTERNAL_BindComputeDescriptorSets(
            currentWriteDescriptorSet->pTexelBufferView = NULL;
            currentWriteDescriptorSet->pImageInfo = NULL;

-            bufferInfos[bufferInfoCount].buffer = commandBuffer->readWriteComputeStorageBuffers[i]->buffer;
+            bufferInfos[bufferInfoCount].buffer = commandBuffer->readWriteComputeStorageBufferBindings[i];
            bufferInfos[bufferInfoCount].offset = 0;
            bufferInfos[bufferInfoCount].range = VK_WHOLE_SIZE;

@@ -8550,7 +8613,7 @@ static void VULKAN_INTERNAL_BindComputeDescriptorSets(
        dynamicOffsetCount,
        dynamicOffsets);

-    commandBuffer->needNewVertexUniformOffsets = false;
+    commandBuffer->needNewComputeUniformOffsets = false;
 }

 static void VULKAN_DispatchCompute(
@@ -8642,9 +8705,12 @@ static void VULKAN_EndComputePass(
        }
    }

-    // we don't need a barrier because sampler state is always the default if sampler bit is set
-    SDL_zeroa(vulkanCommandBuffer->computeSamplerTextures);
-    SDL_zeroa(vulkanCommandBuffer->computeSamplers);
+    // we don't need a barrier for sampler resources because sampler state is always the default if sampler bit is set
+    SDL_zeroa(vulkanCommandBuffer->computeSamplerTextureViewBindings);
+    SDL_zeroa(vulkanCommandBuffer->computeSamplerBindings);
+
+    SDL_zeroa(vulkanCommandBuffer->readWriteComputeStorageTextureViewBindings);
+    SDL_zeroa(vulkanCommandBuffer->readWriteComputeStorageBufferBindings);

    vulkanCommandBuffer->currentComputePipeline = NULL;

@@ -9350,6 +9416,8 @@ static bool VULKAN_INTERNAL_AllocateCommandBuffer(
    commandBuffer->usedUniformBuffers = SDL_malloc(
        commandBuffer->usedUniformBufferCapacity * sizeof(VulkanUniformBuffer *));

+    commandBuffer->swapchainRequested = false;
+
    // Pool it!

    vulkanCommandPool->inactiveCommandBuffers[vulkanCommandPool->inactiveCommandBufferCount] = commandBuffer;
@@ -9460,7 +9528,8 @@ static SDL_GPUCommandBuffer *VULKAN_AcquireCommandBuffer(
    VulkanCommandBuffer *commandBuffer =
        VULKAN_INTERNAL_GetInactiveCommandBufferFromPool(renderer, threadID);

-    commandBuffer->descriptorSetCache = VULKAN_INTERNAL_AcquireDescriptorSetCache(renderer);
+    DescriptorSetCache *descriptorSetCache =
+        VULKAN_INTERNAL_AcquireDescriptorSetCache(renderer);

    SDL_UnlockMutex(renderer->acquireCommandBufferLock);

@@ -9468,6 +9537,8 @@ static SDL_GPUCommandBuffer *VULKAN_AcquireCommandBuffer(
        return NULL;
    }

+    commandBuffer->descriptorSetCache = descriptorSetCache;
+
    // Reset state

    commandBuffer->currentComputePipeline = NULL;
@@ -9510,26 +9581,29 @@ static SDL_GPUCommandBuffer *VULKAN_AcquireCommandBuffer(
    SDL_zeroa(commandBuffer->vertexBufferOffsets);
    commandBuffer->vertexBufferCount = 0;

-    SDL_zeroa(commandBuffer->vertexSamplerTextures);
-    SDL_zeroa(commandBuffer->vertexSamplers);
-    SDL_zeroa(commandBuffer->vertexStorageTextures);
-    SDL_zeroa(commandBuffer->vertexStorageBuffers);
+    SDL_zeroa(commandBuffer->vertexSamplerTextureViewBindings);
+    SDL_zeroa(commandBuffer->vertexSamplerBindings);
+    SDL_zeroa(commandBuffer->vertexStorageTextureViewBindings);
+    SDL_zeroa(commandBuffer->vertexStorageBufferBindings);

-    SDL_zeroa(commandBuffer->fragmentSamplerTextures);
-    SDL_zeroa(commandBuffer->fragmentSamplers);
-    SDL_zeroa(commandBuffer->fragmentStorageTextures);
-    SDL_zeroa(commandBuffer->fragmentStorageBuffers);
+    SDL_zeroa(commandBuffer->fragmentSamplerTextureViewBindings);
+    SDL_zeroa(commandBuffer->fragmentSamplerBindings);
+    SDL_zeroa(commandBuffer->fragmentStorageTextureViewBindings);
+    SDL_zeroa(commandBuffer->fragmentStorageBufferBindings);

    SDL_zeroa(commandBuffer->readWriteComputeStorageTextureSubresources);
    commandBuffer->readWriteComputeStorageTextureSubresourceCount = 0;
    SDL_zeroa(commandBuffer->readWriteComputeStorageBuffers);
-    SDL_zeroa(commandBuffer->computeSamplerTextures);
-    SDL_zeroa(commandBuffer->computeSamplers);
+    SDL_zeroa(commandBuffer->computeSamplerTextureViewBindings);
+    SDL_zeroa(commandBuffer->computeSamplerBindings);
+    SDL_zeroa(commandBuffer->readOnlyComputeStorageTextureViewBindings);
+    SDL_zeroa(commandBuffer->readOnlyComputeStorageBufferBindings);
    SDL_zeroa(commandBuffer->readOnlyComputeStorageTextures);
    SDL_zeroa(commandBuffer->readOnlyComputeStorageBuffers);

    commandBuffer->autoReleaseFence = true;

+    commandBuffer->swapchainRequested = false;
    commandBuffer->isDefrag = 0;

    /* Reset the command buffer here to avoid resets being called
@@ -9616,6 +9690,13 @@ static bool VULKAN_INTERNAL_OnWindowResize(void *userdata, SDL_Event *e)
        data->swapchainCreateHeight = e->window.data2;
    }

+#ifdef SDL_PLATFORM_ANDROID
+    if (e->type == SDL_EVENT_DID_ENTER_BACKGROUND) {
+        data = VULKAN_INTERNAL_FetchWindowData(w);
+        data->needsSwapchainRecreate = true;
+    }
+#endif
+
    return true;
 }

@@ -9878,6 +9959,14 @@ static bool VULKAN_INTERNAL_AcquireSwapchainTexture(
        SET_STRING_ERROR_AND_RETURN("Cannot acquire a swapchain texture from an unclaimed window!", false);
    }

+    // The command buffer is flagged for cleanup when the swapchain is requested as a cleanup timing mechanism
+    vulkanCommandBuffer->swapchainRequested = true;
+
+    if (window->flags & SDL_WINDOW_HIDDEN) {
+        // Edge case, texture is filled in with NULL but not an error
+        return true;
+    }
+
    // If window data marked as needing swapchain recreate, try to recreate
    if (windowData->needsSwapchainRecreate) {
        Uint32 recreateSwapchainResult = VULKAN_INTERNAL_RecreateSwapchain(renderer, windowData);
@@ -9895,13 +9984,6 @@ static bool VULKAN_INTERNAL_AcquireSwapchainTexture(
        }
    }

-    if (swapchainTextureWidth) {
-        *swapchainTextureWidth = windowData->width;
-    }
-    if (swapchainTextureHeight) {
-        *swapchainTextureHeight = windowData->height;
-    }
-
    if (windowData->inFlightFences[windowData->frameCounter] != NULL) {
        if (block) {
            // If we are blocking, just wait for the fence!
@@ -9953,6 +10035,14 @@ static bool VULKAN_INTERNAL_AcquireSwapchainTexture(
        }
    }

+    if (swapchainTextureWidth) {
+        *swapchainTextureWidth = windowData->width;
+    }
+
+    if (swapchainTextureHeight) {
+        *swapchainTextureHeight = windowData->height;
+    }
+
    swapchainTextureContainer = &windowData->textureContainers[swapchainImageIndex];

    // We need a special execution dependency with pWaitDstStageMask or image transition can start before acquire finishes
@@ -10021,7 +10111,7 @@ static bool VULKAN_INTERNAL_AcquireSwapchainTexture(
    }

    vulkanCommandBuffer->signalSemaphores[vulkanCommandBuffer->signalSemaphoreCount] =
-        windowData->renderFinishedSemaphore[windowData->frameCounter];
+        windowData->renderFinishedSemaphore[swapchainImageIndex];
    vulkanCommandBuffer->signalSemaphoreCount += 1;

    *swapchainTexture = (SDL_GPUTexture *)swapchainTextureContainer;
@@ -10330,6 +10420,7 @@ static void VULKAN_INTERNAL_CleanCommandBuffer(
    commandBuffer->presentDataCount = 0;
    commandBuffer->waitSemaphoreCount = 0;
    commandBuffer->signalSemaphoreCount = 0;
+    commandBuffer->swapchainRequested = false;

    // Reset defrag state

@@ -10427,11 +10518,18 @@ static bool VULKAN_Wait(
    VkResult result;
    Sint32 i;

+    SDL_LockMutex(renderer->submitLock);
+
    result = renderer->vkDeviceWaitIdle(renderer->logicalDevice);

-    CHECK_VULKAN_ERROR_AND_RETURN(result, vkDeviceWaitIdle, false);
-
-    SDL_LockMutex(renderer->submitLock);
+    if (result != VK_SUCCESS) {
+        if (renderer->debugMode) {
+            SDL_LogError(SDL_LOG_CATEGORY_GPU, "%s %s", "vkDeviceWaitIdle", VkErrorMessages(result));
+        }
+        SDL_SetError("%s %s", "vkDeviceWaitIdle", VkErrorMessages(result));
+        SDL_UnlockMutex(renderer->submitLock);
+        return false;
+    }

    for (i = renderer->submittedCommandBufferCount - 1; i >= 0; i -= 1) {
        commandBuffer = renderer->submittedCommandBuffers[i];
@@ -10486,7 +10584,7 @@ static bool VULKAN_Submit(
    VulkanTextureSubresource *swapchainTextureSubresource;
    VulkanMemorySubAllocator *allocator;
    bool performCleanups =
-        (renderer->claimedWindowCount > 0 && vulkanCommandBuffer->presentDataCount > 0) ||
+        (renderer->claimedWindowCount > 0 && vulkanCommandBuffer->swapchainRequested) ||
        renderer->claimedWindowCount == 0;

    SDL_LockMutex(renderer->submitLock);
@@ -10562,7 +10660,7 @@ static bool VULKAN_Submit(
        presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
        presentInfo.pNext = NULL;
        presentInfo.pWaitSemaphores =
-            &presentData->windowData->renderFinishedSemaphore[presentData->windowData->frameCounter];
+            &presentData->windowData->renderFinishedSemaphore[presentData->swapchainImageIndex];
        presentInfo.waitSemaphoreCount = 1;
        presentInfo.pSwapchains = &presentData->windowData->swapchain;
        presentInfo.swapchainCount = 1;
@@ -10578,7 +10676,13 @@ static bool VULKAN_Submit(
            presentData->windowData->inFlightFences[presentData->windowData->frameCounter] = (SDL_GPUFence*)vulkanCommandBuffer->inFlightFence;
            (void)SDL_AtomicIncRef(&vulkanCommandBuffer->inFlightFence->referenceCount);

-            if (presentResult == VK_SUBOPTIMAL_KHR || presentResult == VK_ERROR_OUT_OF_DATE_KHR) {
+// On the Android platform, VK_SUBOPTIMAL_KHR is returned whenever the device is rotated. We'll just ignore this for now.
+#ifndef SDL_PLATFORM_ANDROID
+            if (presentResult == VK_SUBOPTIMAL_KHR) {
+                presentData->windowData->needsSwapchainRecreate = true;
+            }
+#endif
+            if (presentResult == VK_ERROR_OUT_OF_DATE_KHR) {
                presentData->windowData->needsSwapchainRecreate = true;
            }
        } else {
@@ -10956,7 +11060,8 @@ static Uint8 VULKAN_INTERNAL_CheckInstanceExtensions(
    const char **requiredExtensions,
    Uint32 requiredExtensionsLength,
    bool *supportsDebugUtils,
-    bool *supportsColorspace)
+    bool *supportsColorspace,
+    bool *supportsPhysicalDeviceProperties2)
 {
    Uint32 extensionCount, i;
    VkExtensionProperties *availableExtensions;
@@ -10995,6 +11100,12 @@ static Uint8 VULKAN_INTERNAL_CheckInstanceExtensions(
        availableExtensions,
        extensionCount);

+    // Only needed for KHR_driver_properties!
+    *supportsPhysicalDeviceProperties2 = SupportsInstanceExtension(
+        VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME,
+        availableExtensions,
+        extensionCount);
+
    SDL_free(availableExtensions);
    return allExtensionsSupported;
 }
@@ -11105,10 +11216,6 @@ static Uint8 VULKAN_INTERNAL_CreateInstance(VulkanRenderer *renderer)
        instanceExtensionCount + 4);
    SDL_memcpy((void *)instanceExtensionNames, originalInstanceExtensionNames, instanceExtensionCount * sizeof(const char *));

-    // Core since 1.1
-    instanceExtensionNames[instanceExtensionCount++] =
-        VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME;
-
 #ifdef SDL_PLATFORM_APPLE
    instanceExtensionNames[instanceExtensionCount++] =
        VK_KHR_PORTABILITY_ENUMERATION_EXTENSION_NAME;
@@ -11119,7 +11226,8 @@ static Uint8 VULKAN_INTERNAL_CreateInstance(VulkanRenderer *renderer)
            instanceExtensionNames,
            instanceExtensionCount,
            &renderer->supportsDebugUtils,
-            &renderer->supportsColorspace)) {
+            &renderer->supportsColorspace,
+            &renderer->supportsPhysicalDeviceProperties2)) {
        SDL_stack_free((char *)instanceExtensionNames);
        SET_STRING_ERROR_AND_RETURN("Required Vulkan instance extensions not supported", false);
    }
@@ -11141,6 +11249,12 @@ static Uint8 VULKAN_INTERNAL_CreateInstance(VulkanRenderer *renderer)
            VK_EXT_SWAPCHAIN_COLOR_SPACE_EXTENSION_NAME;
    }

+    if (renderer->supportsPhysicalDeviceProperties2) {
+        // Append KHR_physical_device_properties2 extension
+        instanceExtensionNames[instanceExtensionCount++] =
+            VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME;
+    }
+
    createInfo.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO;
    createInfo.pNext = NULL;
    createInfo.flags = createFlags;
@@ -11601,7 +11715,7 @@ static bool VULKAN_PrepareDriver(SDL_VideoDevice *_this)
 {
    // Set up dummy VulkanRenderer
    VulkanRenderer *renderer;
-    Uint8 result;
+    bool result = false;

    if (_this->Vulkan_CreateSurface == NULL) {
        return false;
@@ -11611,16 +11725,16 @@ static bool VULKAN_PrepareDriver(SDL_VideoDevice *_this)
        return false;
    }

-    renderer = (VulkanRenderer *)SDL_malloc(sizeof(VulkanRenderer));
-    SDL_memset(renderer, '\0', sizeof(VulkanRenderer));
-
-    result = VULKAN_INTERNAL_PrepareVulkan(renderer);
-
-    if (result) {
-        renderer->vkDestroyInstance(renderer->instance, NULL);
+    renderer = (VulkanRenderer *)SDL_calloc(1, sizeof(*renderer));
+    if (renderer) {
+        result = VULKAN_INTERNAL_PrepareVulkan(renderer);
+        if (result) {
+            renderer->vkDestroyInstance(renderer->instance, NULL);
+        }
+        SDL_free(renderer);
    }
-    SDL_free(renderer);
    SDL_Vulkan_UnloadLibrary();
+
    return result;
 }

@@ -11636,16 +11750,21 @@ static SDL_GPUDevice *VULKAN_CreateDevice(bool debugMode, bool preferLowPower, S
        return NULL;
    }

-    renderer = (VulkanRenderer *)SDL_malloc(sizeof(VulkanRenderer));
-    SDL_memset(renderer, '\0', sizeof(VulkanRenderer));
+    renderer = (VulkanRenderer *)SDL_calloc(1, sizeof(*renderer));
+    if (!renderer) {
+        SDL_Vulkan_UnloadLibrary();
+        return NULL;
+    }
+
    renderer->debugMode = debugMode;
    renderer->preferLowPower = preferLowPower;
    renderer->allowedFramesInFlight = 2;

    if (!VULKAN_INTERNAL_PrepareVulkan(renderer)) {
+        SET_STRING_ERROR("Failed to initialize Vulkan!");
        SDL_free(renderer);
        SDL_Vulkan_UnloadLibrary();
-        SET_STRING_ERROR_AND_RETURN("Failed to initialize Vulkan!", NULL);
+        return NULL;
    }

    SDL_LogInfo(SDL_LOG_CATEGORY_GPU, "SDL_GPU Driver: Vulkan");
@@ -11671,16 +11790,18 @@ static SDL_GPUDevice *VULKAN_CreateDevice(bool debugMode, bool preferLowPower, S

    if (!VULKAN_INTERNAL_CreateLogicalDevice(
            renderer)) {
+        SET_STRING_ERROR("Failed to create logical device!");
        SDL_free(renderer);
        SDL_Vulkan_UnloadLibrary();
-        SET_STRING_ERROR_AND_RETURN("Failed to create logical device!", NULL);
+        return NULL;
    }

    // FIXME: just move this into this function
-    result = (SDL_GPUDevice *)SDL_malloc(sizeof(SDL_GPUDevice));
+    result = (SDL_GPUDevice *)SDL_calloc(1, sizeof(SDL_GPUDevice));
    ASSIGN_DRIVER(VULKAN)

    result->driverData = (SDL_GPURenderer *)renderer;
+    result->shader_formats = SDL_GPU_SHADERFORMAT_SPIRV;

    /*
     * Create initial swapchain array
@@ -1117,13 +1117,12 @@ bool SDL_SYS_HapticResume(SDL_Haptic *haptic)
 */
 bool SDL_SYS_HapticStopAll(SDL_Haptic *haptic)
 {
-    int i, ret;
+    int i;

    // Linux does not support this natively so we have to loop.
    for (i = 0; i < haptic->neffects; i++) {
        if (haptic->effects[i].hweffect != NULL) {
-            ret = SDL_SYS_HapticStopEffect(haptic, &haptic->effects[i]);
-            if (ret < 0) {
+            if (!SDL_SYS_HapticStopEffect(haptic, &haptic->effects[i])) {
                return SDL_SetError("Haptic: Error while trying to stop all playing effects.");
            }
        }
@@ -31,11 +31,7 @@
 /*
 * External stuff.
 */
-#ifdef SDL_VIDEO_DRIVER_WINDOWS
 extern HWND SDL_HelperWindow;
-#else
-static const HWND SDL_HelperWindow = NULL;
-#endif

 /*
 * Internal stuff.