fail on unusable native index without sources, create assets dir before marker delete

* Support Linux Wii controller mappings

* read kernel remote buttons and pair classic node

.github/workflows/build.yml

@@ -47,6 +47,29 @@ jobs:
       - name: Test
         run: dotnet test translator/Translator.sln -c Release --no-build --verbosity normal
 
+  platform_tests:
+    name: Host platform (${{ matrix.runner }})
+    strategy:
+      fail-fast: false
+      matrix:
+        runner: [ubuntu-24.04, macos-14, macos-15-intel]
+    runs-on: ${{ matrix.runner }}
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          persist-credentials: false
+
+      - name: Configure without graphics dependencies or game assets
+        run: >-
+          cmake -S runtime -B build-platform -G Ninja
+          -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++
+          -DMKW_BUILD_PRODUCTS=OFF -DMKW_PLATFORM_TESTS_ONLY=ON
+
+      - name: Build and run host-platform tests
+        run: |
+          cmake --build build-platform --parallel 3
+          ctest --test-dir build-platform --output-on-failure
+
   macos_substrate:
     name: macOS arm64 (configure + substrate tests)
     runs-on: macos-14

.github/workflows/package.yml

@@ -11,6 +11,11 @@ on:
     tags:
       - '*'
   workflow_dispatch:
+    inputs:
+      version:
+        description: Package version
+        required: true
+        type: string
 
 permissions:
   contents: read
@@ -163,14 +168,17 @@ jobs:
 
       - name: Build Setup.pkg
         env:
+          PACKAGE_VERSION: ${{ inputs.version }}
           TAG_VERSION: ${{ github.ref_name }}
         shell: bash
         run: |
-          package_version=""
-          if [[ "${TAG_VERSION:-}" =~ ^v?[0-9] ]]; then
-            package_version="${TAG_VERSION#v}"
-          elif [[ -f "Launcher/Directory.Build.props" ]]; then
-            package_version=$(grep -m1 '<Version>' Launcher/Directory.Build.props | sed -E 's/.*<Version>([^<]+)<\/Version>.*/\1/')
+          package_version="$PACKAGE_VERSION"
+          if [[ -z "$package_version" ]]; then
+            if [[ "${TAG_VERSION:-}" =~ ^v?[0-9] ]]; then
+              package_version="${TAG_VERSION#v}"
+            elif [[ -f "Launcher/Directory.Build.props" ]]; then
+              package_version=$(grep -m1 '<Version>' Launcher/Directory.Build.props | sed -E 's/.*<Version>([^<]+)<\/Version>.*/\1/')
+            fi
           fi
           mkdir -p Launcher/dist
           Launcher/macos/build-setup-pkg.command \

.github/workflows/recomp-test.yml

@@ -45,6 +45,20 @@ jobs:
         shell: pwsh
         run: ./Launcher/Prepare-PortableTools.ps1
 
+      - name: Build and run Windows host-platform tests
+        shell: pwsh
+        run: |
+          $tools = Join-Path $PWD 'Launcher/artifacts/portable-tools'
+          $env:PATH = "$tools/llvm-mingw/bin;$tools/CMake/bin;$tools/Ninja;$env:PATH"
+          cmake -S runtime -B build-platform -G Ninja `
+            -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ `
+            -DMKW_BUILD_PRODUCTS=OFF -DMKW_PLATFORM_TESTS_ONLY=ON
+          if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
+          cmake --build build-platform --parallel 3
+          if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
+          ctest --test-dir build-platform --output-on-failure
+          if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
+
       - name: Prepare pinned native dependencies
         shell: pwsh
         run: ./Launcher/Prepare-Dependencies.ps1

.gitignore

@@ -1,4 +1,5 @@
 /generated/
+/runtime/native-index.json
 /Assets/
 /assets/
 /PulsarPacks/

Launcher/Build-Installer.ps1

@@ -112,6 +112,20 @@ Assert-Directory $dependencySources 'Pinned offline dependency sources'
 # back to compiling aurora from source, which is what this package exists to
 # avoid - and a mismatched one would be an ABI hazard.
 $portableClangHash = Get-MkwFileSha256 (Join-Path $portableTools 'llvm-mingw\bin\clang-22.exe')
+# The identity the runtime build compares against the package's runtime archive.
+function Get-RuntimeSourceFingerprint([string]$RuntimeDirectory, [string]$DependenciesDirectory) {
+    $arguments = @("-DMKW_RUNTIME_DIR=$(ConvertTo-MkwCMakePath $RuntimeDirectory)")
+    $cppWinRt = Join-Path $DependenciesDirectory 'cppwinrt'
+    if (Test-Path -LiteralPath (Join-Path $cppWinRt 'winrt\base.h') -PathType Leaf) {
+        $arguments += "-DMKW_CPPWINRT_INCLUDE_DIR=$(ConvertTo-MkwCMakePath $cppWinRt)"
+    }
+    $arguments += @('-P', (ConvertTo-MkwCMakePath (Join-Path $RuntimeDirectory 'cmake\RuntimeFingerprint.cmake')))
+    $output = @(& (Join-Path $portableTools 'CMake\bin\cmake.exe') @arguments)
+    if ($LASTEXITCODE -ne 0) { throw "Runtime fingerprinting failed with exit code $LASTEXITCODE." }
+    $line = $output | Where-Object { "$_" -match '^runtime_fingerprint=[0-9a-f]{64}$' } | Select-Object -First 1
+    if (-not $line) { throw "RuntimeFingerprint.cmake printed no fingerprint: $($output -join ' ')" }
+    return "$line".Substring('runtime_fingerprint='.Length)
+}
 function Test-NativePrebuiltCurrent([string]$PackageDirectory) {
     $provenancePath = Join-Path $PackageDirectory 'provenance.json'
     if (-not (Test-Path -LiteralPath $provenancePath -PathType Leaf)) { return $false }
@@ -129,6 +143,9 @@ function Test-NativePrebuiltCurrent([string]$PackageDirectory) {
     # release ships runtime/third_party sources that first-party TUs compile
     # against, so archives harvested before a vendored change are stale.
     $recordedThirdParty = $provenance.PSObject.Properties['ThirdPartySourceFingerprint']
+    # The precompiled runtime archive must be from these runtime sources; a package without one
+    # would make every installation compile the runtime itself.
+    $recordedRuntime = $provenance.PSObject.Properties['RuntimeSourceFingerprint']
     return ($provenance.CompilerSha256 -eq $portableClangHash) -and
            ($provenance.FlagFingerprint -eq (Get-MkwNativeFlagFingerprint)) -and
            ($null -ne $recordedDawn) -and
@@ -136,11 +153,13 @@ function Test-NativePrebuiltCurrent([string]$PackageDirectory) {
            ($null -ne $recordedAurora) -and
            ($recordedAurora.Value -eq (Get-MkwAuroraSourceFingerprint (Join-Path $repoRoot 'aurora-main'))) -and
            ($null -ne $recordedThirdParty) -and
-           ($recordedThirdParty.Value -eq (Get-MkwThirdPartySourceFingerprint (Join-Path $repoRoot 'runtime\third_party')))
+           ($recordedThirdParty.Value -eq (Get-MkwThirdPartySourceFingerprint (Join-Path $repoRoot 'runtime\third_party'))) -and
+           ($null -ne $recordedRuntime) -and
+           ($recordedRuntime.Value -eq (Get-RuntimeSourceFingerprint (Join-Path $repoRoot 'runtime') $dependencySources))
 }
 $nativePrebuiltPackage = Join-Path $dependencySources 'native_prebuilt'
 if (-not (Test-NativePrebuiltCurrent $nativePrebuiltPackage)) {
-    Write-Host '[0/6] Building the precompiled aurora and third-party package...'
+    Write-Host '[0/6] Building the precompiled aurora, third-party and runtime package...'
     & (Join-Path $PSScriptRoot 'Prepare-NativePrebuilt.ps1') `
         -PortableToolsDirectory $portableTools -DependencySourceDirectory $dependencySources
     if ($LASTEXITCODE -ne 0) { throw 'Precompiled aurora package preparation failed.' }
@@ -222,6 +241,21 @@ Copy-Item -LiteralPath (Join-Path $PSScriptRoot 'NativeBuildFlags.ps1') -Destina
 [IO.Directory]::CreateDirectory((Join-Path $workspace 'Dependencies')) | Out-Null
 foreach ($name in $requiredDependencies) { Copy-Directory (Join-Path $dependencySources $name) (Join-Path $workspace "Dependencies\$name") }
 
+# The translation identity covers the runtime's native registrations through this index (see
+# ToolkitFingerprint), so it is extracted from exactly the staged runtime sources.
+$nativeIndex = Join-Path $workspace 'runtime\native-index.json'
+& $translator emit-native-index --project (Join-Path $workspace 'projects\mkwii\recomp.yml') --out $nativeIndex
+if ($LASTEXITCODE -ne 0) { throw "Native index extraction failed with exit code $LASTEXITCODE." }
+Assert-File $nativeIndex 'Runtime native index'
+# A staged runtime that fingerprints differently from the shipped archive would make every
+# installation silently compile the runtime from source.
+$stagedRuntimeFingerprint = Get-RuntimeSourceFingerprint (Join-Path $workspace 'runtime') (Join-Path $workspace 'Dependencies')
+$packagedRuntimeFingerprint = (Get-Content -LiteralPath (Join-Path $workspace 'Dependencies\native_prebuilt\provenance.json') -Raw |
+    ConvertFrom-Json).RuntimeSourceFingerprint
+if ($stagedRuntimeFingerprint -ne $packagedRuntimeFingerprint) {
+    throw "The staged runtime ($stagedRuntimeFingerprint) does not match the precompiled runtime archive ($packagedRuntimeFingerprint)."
+}
+
 [IO.Directory]::CreateDirectory((Join-Path $payloadRoot 'host')) | Out-Null
 Copy-Item $setupHost (Join-Path $payloadRoot 'host\WiiCompiled-Setup.exe')
 

Launcher/LocalBuild.ps1

@@ -429,7 +429,8 @@ try {
             }
             $redist = Join-Path $Toolkit 'Redist'
             if (Test-Path -LiteralPath $redist -PathType Container) {
-                Copy-Item -Path (Join-Path $redist '*.dll') -Destination $Destination -Force
+                Get-ChildItem -LiteralPath $redist -Filter '*.dll' -File |
+                    Copy-Item -Destination $Destination -Force
             }
             foreach ($name in @('config','wii_bootstrap')) {
                 $directory = Join-Path $build $name

Launcher/NativeBuildFlags.ps1

@@ -145,7 +145,8 @@ function Invoke-Checked([string]$FilePath, [string[]]$Arguments, [string]$Descri
     else { Write-Host "${LogPrefix}: $Description" }
     $quotedArguments = @($Arguments | ForEach-Object {
         if ($_.Contains('"')) { throw "A native build argument contains an unsupported quote: $_" }
-        '"' + $_ + '"'
+        # Backslashes directly before the closing quote would otherwise escape it.
+        '"' + ($_ -replace '(\\+)$', '$1$1') + '"'
     })
     if ($WaitForProcessTree) {
         $process = Start-Process -FilePath $FilePath -ArgumentList $quotedArguments `

Launcher/Prepare-NativePrebuilt.ps1

@@ -1,7 +1,8 @@
 # Builds the redistributable precompiled aurora + third-party package: aurora (~43% of local build
 # CPU time) and vendored Crypto++ are identical for every user, so this configures runtime/ like
-# LocalBuild.ps1 (both dot-source NativeBuildFlags.ps1), builds just that closure, and harvests the
-# archives plus a generated CMake description into launcher/artifacts/dependencies/native_prebuilt.
+# LocalBuild.ps1 (both dot-source NativeBuildFlags.ps1), builds just that closure plus the
+# game-independent runtime archive, and harvests the archives and a generated CMake description
+# into launcher/artifacts/dependencies/native_prebuilt.
 [CmdletBinding(PositionalBinding = $false)]
 param(
     [string]$PortableToolsDirectory = 'launcher/artifacts/portable-tools',
@@ -87,24 +88,29 @@ try {
         -AdditionalArguments @("-DMKW_NATIVE_PREBUILT_EXPORT_DIR=$(ConvertTo-MkwCMakePath $exportDirectory)")
     Invoke-Checked $cmake $configure 'Configuring the aurora/third-party staging build' -LogPrefix 'MKWNP'
 
+    $metaPath = Join-Path $exportDirectory 'meta.txt'
+    Assert-File $metaPath 'Native prebuilt export metadata'
+    $meta = @{}
+    foreach ($line in Get-Content -LiteralPath $metaPath) {
+        if ($line -match '^([a-z0-9_]+)=(.*)$') { $meta[$Matches[1]] = $Matches[2] }
+    }
+    # The game-independent runtime ships precompiled too; it is outside the probe closure.
+    if (-not $meta.ContainsKey('runtime_fingerprint')) {
+        throw 'The staging configure exported no runtime_fingerprint; the x86_64 package must carry the runtime archive.'
+    }
+
     # mkw_np_probe links exactly what mkw_runtime_common and the public products
     # link, so building it compiles the whole redistributable closure and nothing
     # else - and a successful link proves the harvested archives are complete.
-    Invoke-Checked $cmake @('--build', $stage, '--target', 'mkw_np_probe', '--parallel', "$Parallel") `
-        'Building the aurora/third-party closure' -LogPrefix 'MKWNP'
+    Invoke-Checked $cmake @('--build', $stage, '--target', 'mkw_np_probe', '--target', 'mkw_runtime',
+        '--parallel', "$Parallel") 'Building the aurora/third-party closure and the runtime' -LogPrefix 'MKWNP'
 } finally {
     $env:PATH = $oldPath
 }
 
 # ---------------------------------------------------------------------------
-# Read the description CMake wrote, plus the resolved command lines Ninja holds.
+# Read the resolved command lines Ninja holds.
 # ---------------------------------------------------------------------------
-$metaPath = Join-Path $exportDirectory 'meta.txt'
-Assert-File $metaPath 'Native prebuilt export metadata'
-$meta = @{}
-foreach ($line in Get-Content -LiteralPath $metaPath) {
-    if ($line -match '^([a-z0-9_]+)=(.*)$') { $meta[$Matches[1]] = $Matches[2] }
-}
 
 $buildNinja = Join-Path $stage 'build.ninja'
 Assert-File $buildNinja 'Generated build.ninja'
@@ -234,6 +240,12 @@ foreach ($name in ($targets.Keys | Sort-Object)) {
 }
 if ($sharedImports.Count -eq 0) { throw 'No shared libraries were harvested; the SDL/zlib/libpng DLLs would be missing.' }
 if ($dawnLinkerFile) { $linkerFileToReference[$dawnLinkerFile] = 'dawn::webgpu_dawn' }
+$runtimeTarget = $targets['mkw_runtime']
+if ($null -eq $runtimeTarget -or $runtimeTarget.Type -ne 'STATIC_LIBRARY' -or
+    -not $linkerFileToReference.ContainsKey($runtimeTarget.LinkerFile)) {
+    throw 'The runtime archive (mkw_runtime) was not built and harvested.'
+}
+$runtimeArchiveReference = $linkerFileToReference[$runtimeTarget.LinkerFile]
 
 # ---------------------------------------------------------------------------
 # Rewrite the compile/link interface into workspace-relative tokens.
@@ -297,6 +309,9 @@ $packageLinkItems = @($linkItems | ForEach-Object {
     if (StartsWithPath $absolute $dependencyRoot) { return "`@DEPS`@/$(RelativeTo $absolute $dependencyRoot)" }
     throw "Link item is neither a system library nor a harvested archive: $item"
 })
+if ($packageLinkItems -contains $runtimeArchiveReference) {
+    throw 'The runtime archive leaked into the aurora link interface.'
+}
 
 # fmt's `-include cstdlib` (and anything like it) is recorded from a C++ compile
 # line, so re-scope it to C++. The runtime also assembles generated .S blobs
@@ -333,6 +348,11 @@ $generated.Add('')
 $generated.Add("set(MKW_NP_DAWN_CONFIG_DIR `"$(ConvertTo-Token $meta['dawn_config_dir'] 'dawn')`")")
 $generated.Add("set(MKW_NP_DAWN_VERSION `"$($meta['aurora_dawn_version'])`")")
 $generated.Add('')
+$generated.Add('# The precompiled game-independent runtime; RuntimeLibrary.cmake decides whether it applies.')
+$generated.Add("set(MKW_NP_RUNTIME_ARCHIVE `"$runtimeArchiveReference`")")
+$generated.Add("set(MKW_NP_RUNTIME_FINGERPRINT `"$($meta['runtime_fingerprint'])`")")
+$generated.Add("set(MKW_NP_RUNTIME_CPU_PROFILE `"$($meta['runtime_cpu_profile'])`")")
+$generated.Add('')
 $generated.Add('# Shared third-party libraries keep imported SHARED targets so that')
 $generated.Add('# $<TARGET_RUNTIME_DLLS> still copies them next to the game executable.')
 foreach ($import in $sharedImports) {
@@ -393,6 +413,9 @@ $provenance = [ordered]@{
     # release build and the local build can refuse a package aurora has outgrown.
     AuroraSourceFingerprint = $auroraSourceFingerprint
     ThirdPartySourceFingerprint = $thirdPartySourceFingerprint
+    # runtime/cmake/RuntimeFingerprint.cmake over the runtime sources the archive was built from.
+    RuntimeSourceFingerprint = $meta['runtime_fingerprint']
+    RuntimeCpuProfile = $meta['runtime_cpu_profile']
     Sdl3Target = $meta['aurora_sdl3_target']
     HarvestedLibraryCount = @($linkerFileToReference.Keys).Count
     Contents = $contents

Launcher/Prepare-NativePrebuilt.sh

@@ -2,8 +2,9 @@
 # Builds the redistributable precompiled aurora + third-party package for native Linux: aurora
 # (~43% of local build CPU time per Prepare-NativePrebuilt.ps1), Crypto++ and mbed TLS are identical
 # for every user under the pinned toolchain prepare-portable-tools.sh bundles, so this configures
-# runtime/ against that toolchain, builds just that closure, and harvests the archives plus a
-# generated CMake description into an output package - the Linux counterpart to
+# runtime/ against that toolchain, builds just that closure (plus, on x86_64, the game-independent
+# runtime archive), and harvests the archives plus a generated CMake description into an output
+# package - the Linux counterpart to
 # Launcher/Prepare-NativePrebuilt.ps1, consumed by the same platform-agnostic
 # runtime/cmake/NativePrebuilt.cmake either script's package works with unmodified.
 #
@@ -191,12 +192,23 @@ if [[ "${cxx_sha256}" != "${compiler_sha256}" ]]; then
     compiler_sha256="${compiler_sha256}:${cxx_sha256}"
 fi
 
+# The game-independent runtime ships precompiled on x86_64 only; aarch64 compiles it locally.
+runtime_fingerprint=""
+if [[ "$arch" == x86_64 ]]; then
+    runtime_fingerprint=$("$cmake_bin" -DMKW_RUNTIME_DIR="$runtime_source" \
+        -P "$runtime_source/cmake/RuntimeFingerprint.cmake" | sed -n 's/^runtime_fingerprint=\([0-9a-f]\{64\}\)$/\1/p')
+    [[ -n "$runtime_fingerprint" ]] || fail "RuntimeFingerprint.cmake printed no runtime fingerprint"
+fi
+
 if [[ "$print_fingerprint_only" -eq 1 ]]; then
     printf 'compiler_sha256=%s\n' "$compiler_sha256"
     printf 'flag_fingerprint=%s\n' "$flag_fingerprint"
     printf 'aurora_fingerprint=%s\n' "$aurora_fingerprint"
     printf 'third_party_fingerprint=%s\n' "$third_party_fingerprint"
     printf 'mbedtls_fingerprint=%s\n' "$mbedtls_fingerprint"
+    if [[ -n "$runtime_fingerprint" ]]; then
+        printf 'runtime_fingerprint=%s\n' "$runtime_fingerprint"
+    fi
     exit 0
 fi
 
@@ -220,20 +232,30 @@ echo "Prepare-NativePrebuilt.sh: configuring the aurora/third-party staging buil
     -DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=lld" \
     -DMKW_NATIVE_PREBUILT_EXPORT_DIR="$export_dir"
 
-# mkw_np_probe links exactly what mkw_runtime_common and the public products link, so building it
-# compiles the whole redistributable closure and nothing else - and a successful link proves the
-# harvested archives are complete.
-echo "Prepare-NativePrebuilt.sh: building the aurora/third-party closure..."
-"$cmake_bin" --build "$stage_dir" --target mkw_np_probe --parallel "$parallel"
-
-# ---------------------------------------------------------------------------
-# Read the description CMake wrote, plus the resolved command lines Ninja holds.
-# ---------------------------------------------------------------------------
-
 meta_path="$export_dir/meta.txt"
 assert_file "$meta_path" "Native prebuilt export metadata"
 get_meta() { awk -F= -v k="$1" '$0 ~ "^" k "=" { sub(/^[^=]*=/, ""); print; exit }' "$meta_path"; }
 
+# mkw_np_probe links exactly what mkw_runtime_common and the public products link, so building it
+# compiles the whole redistributable closure and nothing else - and a successful link proves the
+# harvested archives are complete. The runtime archive is outside that closure and built alongside.
+build_targets=(--target mkw_np_probe)
+meta_runtime_fingerprint=$(get_meta runtime_fingerprint)
+if [[ -n "$meta_runtime_fingerprint" ]]; then
+    if [[ -n "$runtime_fingerprint" && "$meta_runtime_fingerprint" != "$runtime_fingerprint" ]]; then
+        fail "The configure's runtime fingerprint ($meta_runtime_fingerprint) differs from RuntimeFingerprint.cmake's ($runtime_fingerprint)"
+    fi
+    build_targets+=(--target mkw_runtime)
+elif [[ -n "$runtime_fingerprint" ]]; then
+    fail "The staging configure exported no runtime_fingerprint; the $arch package must carry the runtime archive"
+fi
+echo "Prepare-NativePrebuilt.sh: building the aurora/third-party closure..."
+"$cmake_bin" --build "$stage_dir" "${build_targets[@]}" --parallel "$parallel"
+
+# ---------------------------------------------------------------------------
+# Read the resolved command lines Ninja holds.
+# ---------------------------------------------------------------------------
+
 build_ninja="$stage_dir/build.ninja"
 assert_file "$build_ninja" "Generated build.ninja"
 
@@ -397,6 +419,13 @@ while IFS='|' read -r name type file linkerfile; do
         linker_file_to_reference["$linkerfile"]="@PKG@/$relative_linker"
     fi
 done < "$targets_txt"
+runtime_archive_ref=""
+if [[ -n "$meta_runtime_fingerprint" ]]; then
+    runtime_linker=$(awk -F'|' '$1 == "mkw_runtime" && $2 == "STATIC_LIBRARY" { print $4; exit }' "$targets_txt")
+    [[ -n "$runtime_linker" && -f "$runtime_linker" ]] || fail "The runtime archive (mkw_runtime) was not built"
+    runtime_archive_ref=${linker_file_to_reference["$(normalize "$runtime_linker")"]:-}
+    [[ "$runtime_archive_ref" == "@PKG@/lib/"* ]] || fail "The runtime archive (mkw_runtime) was not harvested"
+fi
 mbedtls_refs=()
 mbedtls_txt="$export_dir/mbedtls.txt"
 assert_file "$mbedtls_txt" "Mbed TLS export targets list"
@@ -498,6 +527,11 @@ for item in "${link_items[@]}"; do
     esac
     fail "Link item is neither a system library nor a harvested archive: $item"
 done
+if [[ -n "$runtime_archive_ref" ]]; then
+    for item in "${package_link_items[@]}"; do
+        [[ "$item" != "$runtime_archive_ref" ]] || fail "The runtime archive leaked into the aurora link interface"
+    done
+fi
 
 # ---------------------------------------------------------------------------
 # Emit the consumer-facing CMake description.
@@ -562,6 +596,12 @@ generated_cmake="$output_dir/native_prebuilt.cmake"
     format_cmake_block MKW_NP_MBEDTLS_LIBRARIES "${mbedtls_refs[@]}"
     echo 'set(MKW_NP_MBEDTLS_INCLUDE_DIR "@PKG@/include/mbedtls")'
     printf 'set(MKW_NP_MBEDTLS_FINGERPRINT "%s")\n' "$mbedtls_fingerprint"
+    if [[ -n "$runtime_archive_ref" ]]; then
+        # The precompiled game-independent runtime; RuntimeLibrary.cmake decides whether it applies.
+        printf 'set(MKW_NP_RUNTIME_ARCHIVE "%s")\n' "$runtime_archive_ref"
+        printf 'set(MKW_NP_RUNTIME_FINGERPRINT "%s")\n' "$meta_runtime_fingerprint"
+        printf 'set(MKW_NP_RUNTIME_CPU_PROFILE "%s")\n' "$(get_meta runtime_cpu_profile)"
+    fi
     format_cmake_block MKW_NP_AURORA_TARGETS "aurora::gx" "aurora::pad" "aurora::si" "aurora::vi" "aurora::mtx"
     echo ""
     printf 'set(MKW_NP_DAWN_CONFIG_DIR "%s")\n' "$dawn_config_token"
@@ -597,12 +637,13 @@ harvested_count=${#linker_file_to_reference[@]}
 
 python3 - "$output_dir" "$compiler_sha256" "$compiler_version" "$flag_fingerprint" \
     "$dawn_version" "$dawn_runtime_sha256" "$aurora_fingerprint" "$third_party_fingerprint" \
-    "$sdl3_target" "$harvested_count" "$mbedtls_fingerprint" <<'PY'
+    "$sdl3_target" "$harvested_count" "$mbedtls_fingerprint" \
+    "${runtime_archive_ref:+$meta_runtime_fingerprint}" "${runtime_archive_ref:+$(get_meta runtime_cpu_profile)}" <<'PY'
 import hashlib, json, os, sys, datetime
 
 (output_dir, compiler_sha256, compiler_version, flag_fingerprint, dawn_version,
  dawn_runtime_sha256, aurora_fingerprint, third_party_fingerprint, sdl3_target,
- harvested_count, mbedtls_fingerprint) = sys.argv[1:]
+ harvested_count, mbedtls_fingerprint, runtime_fingerprint, runtime_cpu_profile) = sys.argv[1:]
 
 contents = []
 for root, dirs, files in os.walk(output_dir):
@@ -644,6 +685,9 @@ provenance = {
     "HarvestedLibraryCount": int(harvested_count),
     "Contents": contents,
 }
+if runtime_fingerprint:
+    provenance["RuntimeSourceFingerprint"] = runtime_fingerprint
+    provenance["RuntimeCpuProfile"] = runtime_cpu_profile
 with open(os.path.join(output_dir, "provenance.json"), "w", encoding="utf-8") as fh:
     json.dump(provenance, fh, indent=2)
     fh.write("\n")

Launcher/Test-PinnedFacts.ps1

@@ -127,7 +127,7 @@ foreach ($mandatory in ([regex]::Matches($parameterBlock,
 }
 
 # --- The build steps the installer's progress bar acts on.
-$installProgress = Read-SourceFile (Join-Path $setup 'InstallProgress.cs') 'InstallProgress.cs'
+$installProgress = Read-SourceFile (Join-Path $common 'InstallProgress.cs') 'InstallProgress.cs'
 $knownSteps = Get-QuotedList $installProgress '(?s)class BuildStepIds\s*\{(.*?)\n\}' `
     'const string \w+ = "([^"]+)";' 'BuildStepIds'
 $emittedSteps = @([regex]::Matches($localBuild, "(?:-StepId|Write-MkwBuildStep)\s+'([^']+)'") |

Launcher/WiiCompiled.Setup.Common/InstallProgress.cs

@@ -1,35 +1,19 @@
 using System.Text.Json;
 
-namespace WiiCompiled.Setup.Windows;
-
-/// <summary>
-/// Stable stage identifiers reported by <c>--progress-json</c>. These are part of the public
-/// WheelWizard contract (docs/WHEELWIZARD_CONTRACT.md): the strings may gain new members, but an
-/// existing identifier never changes meaning.
-/// </summary>
-internal static class InstallStages
-{
-    public const string Validate = "validate";
-    public const string ExtractToolkit = "extract-toolkit";
-    public const string ExtractDisc = "extract-disc";
-    public const string PrepareRetro = "prepare-retro";
-    public const string BuildBase = "build-base";
-    public const string BuildRetro = "build-retro";
-    public const string Publish = "publish";
-}
+namespace WiiCompiled.Setup.Common;
 
 /// <summary>
 /// Where an installation reports what it is doing. Progress is coarse and monotonic; raw translator
 /// and compiler output is a diagnostic, never progress, because it is unbounded and machine-hostile.
 /// </summary>
-internal interface IInstallReporter
+public interface IInstallReporter
 {
     void Progress(string stage, string message, int percent);
     void Diagnostic(string line);
 }
 
 /// <summary>Reports to a caller-supplied sink; used by console and maintenance commands.</summary>
-internal sealed class DelegatingInstallReporter : IInstallReporter
+public sealed class DelegatingInstallReporter : IInstallReporter
 {
     private readonly Action<string> _sink;
 
@@ -43,15 +27,20 @@ internal sealed class DelegatingInstallReporter : IInstallReporter
 /// The <c>--progress-json</c> protocol: one JSON object per line on stdout, nothing else on stdout,
 /// diagnostics on stderr. The terminal <c>result</c> line is written exactly once.
 /// </summary>
-internal sealed class NdjsonInstallReporter : IInstallReporter
+public sealed class NdjsonInstallReporter : IInstallReporter
 {
     private static readonly JsonSerializerOptions Options = new() { WriteIndented = false };
+    private readonly string _version;
     private readonly Action<string>? _log;
     private readonly object _gate = new();
     private int _lastPercent;
     private bool _finished;
 
-    public NdjsonInstallReporter(Action<string>? log = null) => _log = log;
+    public NdjsonInstallReporter(string version, Action<string>? log = null)
+    {
+        _version = version;
+        _log = log;
+    }
 
     public void Progress(string stage, string message, int percent)
     {
@@ -78,7 +67,7 @@ internal sealed class NdjsonInstallReporter : IInstallReporter
         {
             if (_finished) return;
             _finished = true;
-            WriteLine(new { type = "result", success = true, version = ProductInfo.Version, installDir = installDirectory });
+            WriteLine(new { type = "result", success = true, version = _version, installDir = installDirectory });
         }
     }
 
@@ -106,8 +95,8 @@ internal sealed class NdjsonInstallReporter : IInstallReporter
     }
 }
 
-/// <summary>Plain-text console reporting for a silent install without <c>--progress-json</c>.</summary>
-internal sealed class ConsoleInstallReporter : IInstallReporter
+/// <summary>Plain-text console reporting for a run without <c>--progress-json</c>.</summary>
+public sealed class ConsoleInstallReporter : IInstallReporter
 {
     private readonly Action<string>? _log;
 
@@ -126,10 +115,10 @@ internal sealed class ConsoleInstallReporter : IInstallReporter
     }
 }
 
-/// <summary>Build step identifiers from the bundled script's <c>MKWCBUILD:STEP:&lt;id&gt;</c> lines. The
+/// <summary>Build step identifiers from the build scripts' <c>MKWCBUILD:STEP:&lt;id&gt;</c> lines. The
 /// id is the contract; LocalBuild.ps1 emits these exactly and Test-PinnedFacts.ps1 fails the release if
 /// the two sets disagree, so progress never depends on matching reworded English prose.</summary>
-internal static class BuildStepIds
+public static class BuildStepIds
 {
     public const string ReuseBaseTranslation = "reuse-base-translation";
     public const string RetranslateBase = "retranslate-base";
@@ -143,16 +132,16 @@ internal static class BuildStepIds
 }
 
 /// <summary>
-/// Maps one local translate-and-compile run onto a slice of the overall percentage. The bundled
-/// build script announces every step it starts with an <c>MKWCBUILD:</c> prefix, so the slice can
-/// advance on real events instead of on a timer.
+/// Maps one local translate-and-compile run onto a slice of the overall percentage. The build
+/// scripts announce every step they start with an <c>MKWCBUILD:</c> prefix, so the slice can advance
+/// on real events instead of on a timer.
 /// </summary>
-internal sealed class BuildProgressWindow
+public sealed class BuildProgressWindow
 {
     private const string Marker = "MKWCBUILD:";
     private const string StepMarker = "STEP:";
 
-    /// <summary>The fraction the compile step reaches; beyond it, compiler output is a heartbeat.</summary>
+    /// <summary>The fraction the compile step reaches; beyond it, Ninja's counter drives progress.</summary>
     private const double CompileFraction = 0.58;
 
     private static readonly (string Id, double Fraction, string Message)[] Steps =
@@ -164,7 +153,7 @@ internal sealed class BuildProgressWindow
         (BuildStepIds.TranslateMod, 0.38, "Translating the Retro Rewind Code.pul"),
         (BuildStepIds.GenerateDataInit, 0.44, "Generating game data initialization"),
         (BuildStepIds.EmitBuildShards, 0.48, "Preparing the native build"),
-        (BuildStepIds.ConfigureNative, 0.52, "Configuring the bundled compiler"),
+        (BuildStepIds.ConfigureNative, 0.52, "Configuring the compiler"),
         (BuildStepIds.Compile, CompileFraction, "Compiling the game. This is the longest step"),
     ];
 
@@ -209,19 +198,30 @@ internal sealed class BuildProgressWindow
             }
         }
 
-        // Anything else - an unknown step identifier from a newer script, a plain MKWCBUILD note, or
-        // raw tool output - stays a diagnostic and only feeds the heartbeat below.
+        // Anything else - an unknown step identifier, a plain MKWCBUILD note, or raw tool output -
+        // stays a diagnostic.
         _reporter.Diagnostic(line);
-        // Compilation announces itself once and then emits thousands of compiler lines. Treat that
-        // output as a heartbeat so the slice keeps creeping forward, but only publish a progress
-        // line when the rounded percentage actually changes.
-        if (_fraction >= CompileFraction)
+        // Ninja prefixes every edge with [done/total]; how many edges there are depends on how much
+        // of the build is already up to date, so follow the counter rather than counting lines.
+        if (_fraction >= CompileFraction && TryParseNinjaProgress(line, out var done, out var total))
         {
-            _fraction = Math.Min(0.97, _fraction + 0.0015);
+            _fraction = Math.Max(_fraction, CompileFraction + (0.97 - CompileFraction) * done / total);
             Emit();
         }
     }
 
+    private static bool TryParseNinjaProgress(string line, out int done, out int total)
+    {
+        done = total = 0;
+        if (!line.StartsWith('[')) return false;
+        var close = line.IndexOf(']');
+        var slash = line.IndexOf('/');
+        return close > 0 && slash > 0 && slash < close
+            && int.TryParse(line.AsSpan(1, slash - 1), out done)
+            && int.TryParse(line.AsSpan(slash + 1, close - slash - 1), out total)
+            && total > 0 && done <= total;
+    }
+
     private void Emit()
     {
         var percent = Interpolate(_fraction);

Launcher/WiiCompiled.Setup.Common/RuntimeConfiguration.cs

@@ -1,4 +1,7 @@
 using System.Text;
+using Tomlyn;
+using Tomlyn.Parsing;
+using Tomlyn.Syntax;
 
 namespace WiiCompiled.Setup.Common;
 
@@ -17,12 +20,6 @@ public static class RuntimeConfiguration
 {
     public const string ConfigFileName = "Config.toml";
 
-    /// <summary>
-    /// Ordinal, matching the runtime parser/writer (<c>runtime_config.h</c>) exactly. Case-insensitive
-    /// matching here would let setup rewrite a <c>[Paths]</c>/<c>Dvd_Root</c> spelling the runtime ignores.
-    /// </summary>
-    private const StringComparison KeyComparison = StringComparison.Ordinal;
-
     /// <summary>The per-user configuration an ordinary (non-portable) installation shares.</summary>
     public static string ApplicationDataConfigPath => Path.Combine(
         Environment.GetFolderPath(Environment.SpecialFolder.LocalApplicationData),
@@ -75,9 +72,11 @@ public static class RuntimeConfiguration
     public static void RemoveDvdRootIfOwned(string configPath, string dvdRoot) =>
         RemovePathIfOwned(configPath, "dvd_root", dvdRoot);
 
-    /// <summary>The raw stored text of a <c>[paths]</c> key, exactly as the file holds it.</summary>
+    /// <summary>The stored string of a <c>[paths]</c> key, exactly as the file holds it.</summary>
     public static string? GetPath(string configPath, string key) =>
-        TryUnquoteToml(GetRawValue(configPath, "paths", key) ?? "", out var value) ? value : null;
+        TryLoad(configPath) is { } document && Find(document, ["paths"], key)?.Item.Value is StringValueSyntax value
+            ? value.Value!
+            : null;
 
     /// <summary>
     /// A <c>[paths]</c> value as the runtime will actually use it. The runtime resolves every relative
@@ -99,9 +98,19 @@ public static class RuntimeConfiguration
 
     private static void SetPath(string configPath, string key, string value)
     {
-        var lines = ReadLinesOrDefault(configPath);
-        SetSectionValue(lines, "paths", key, QuoteToml(FormatPathValue(configPath, value)));
-        WriteLines(configPath, lines);
+        var document = Parse(File.Exists(configPath) ? File.ReadAllText(configPath) : DefaultConfig);
+        var section = new[] { "paths" };
+        var newValue = new StringValueSyntax(FormatPathValue(configPath, value));
+        if (Find(document, section, key) is { } existing)
+        {
+            LastToken(newValue).TrailingTrivia = LastToken(existing.Item.Value!).TrailingTrivia;
+            existing.Item.Value = newValue;
+        }
+        else
+        {
+            Append(document, section, new KeyValueSyntax(key, newValue));
+        }
+        Save(configPath, document);
     }
 
     /// <summary>
@@ -123,176 +132,104 @@ public static class RuntimeConfiguration
 
     private static void RemovePathIfOwned(string configPath, string key, string expectedValue)
     {
-        if (!File.Exists(configPath)) return;
-        var lines = File.ReadAllLines(configPath).ToList();
-        var inPaths = false;
-        var changed = false;
-        for (var index = 0; index < lines.Count; index++)
-        {
-            var trimmed = RemoveComment(lines[index]).Trim();
-            if (IsSection(trimmed))
-            {
-                inPaths = trimmed.Equals("[paths]", KeyComparison);
-                continue;
-            }
-            if (!inPaths || trimmed.Length == 0) continue;
-            var equals = trimmed.IndexOf('=');
-            if (equals < 0 || !trimmed[..equals].Trim().Equals(key, KeyComparison))
-                continue;
-            if (!TryUnquoteToml(trimmed[(equals + 1)..].Trim(), out var configured)) return;
-            // A portable installation stores this relative, so ownership is decided on the path the
-            // runtime would actually resolve, not on the stored text.
-            if (!ResolveAgainstConfig(configPath, configured)
-                    .Equals(Path.GetFullPath(expectedValue), StringComparison.OrdinalIgnoreCase))
-                return;
-            lines.RemoveAt(index);
-            changed = true;
-            break;
-        }
-        if (changed) WriteLines(configPath, lines);
+        if (TryLoad(configPath) is not { } document ||
+            Find(document, ["paths"], key) is not { } entry ||
+            entry.Item.Value is not StringValueSyntax configured)
+            return;
+        // A portable installation stores this relative, so ownership is decided on the path the
+        // runtime would actually resolve, not on the stored text.
+        if (!ResolveAgainstConfig(configPath, configured.Value!)
+                .Equals(Path.GetFullPath(expectedValue), StringComparison.OrdinalIgnoreCase))
+            return;
+        var previous = entry.Previous?.EndOfLineToken ?? entry.Table?.EndOfLineToken;
+        if (previous is not null && entry.Item.EndOfLineToken?.TrailingTrivia is { } trivia)
+            (previous.TrailingTrivia ??= []).AddRange(trivia);
+        entry.Items.RemoveChild(entry.Item);
+        Save(configPath, document);
     }
 
-    /// <summary>The raw TOML literal stored for a key, or null when the section or key is absent.</summary>
-    public static string? GetRawValue(string configPath, string section, string key)
+    private const string DefaultConfig = "# WiiCompiled user configuration\n\n";
+
+    private sealed record Entry(
+        KeyValueSyntax Item, SyntaxList<KeyValueSyntax> Items, TableSyntax? Table, KeyValueSyntax? Previous);
+
+    private static DocumentSyntax Parse(string text) => SyntaxParser.ParseStrict(text);
+
+    private static DocumentSyntax? TryLoad(string configPath)
     {
         if (!File.Exists(configPath)) return null;
-        var header = $"[{section}]";
-        var inSection = false;
-        foreach (var line in File.ReadLines(configPath))
+        try { return Parse(File.ReadAllText(configPath)); }
+        catch (TomlException) { return null; }
+    }
+
+    /// <summary>Re-parses the edited tree before replacing the file, so an invalid edit never reaches disk.</summary>
+    private static void Save(string configPath, DocumentSyntax document)
+    {
+        var text = document.ToString();
+        Parse(text);
+        FileSystemUtilities.WriteAtomic(configPath, new UTF8Encoding(false).GetBytes(text));
+    }
+
+    private static IEnumerable<TableSyntax> Tables(DocumentSyntax document) =>
+        document.Tables.OfType<TableSyntax>();
+
+    private static string[] Parts(KeySyntax key) =>
+        [Part(key.Key!), .. key.DotKeys.Select(dotted => Part(dotted.Key!))];
+
+    private static string Part(BareKeyOrStringValueSyntax part) =>
+        part is StringValueSyntax quoted ? quoted.Value! : ((BareKeySyntax)part).Key!.Text!;
+
+    /// <summary>Matches by TOML meaning, so <c>"dvd_root"</c> and <c>dvd_root</c> are the same key.</summary>
+    private static Entry? Find(DocumentSyntax document, string[] section, string key)
+    {
+        string[] wanted = [.. section, key];
+        var scopes = Tables(document)
+            .Select(table => (Table: (TableSyntax?)table, Prefix: Parts(table.Name!), Items: table.Items))
+            .Prepend((Table: null, Prefix: [], Items: document.KeyValues));
+        foreach (var (table, prefix, items) in scopes)
         {
-            var trimmed = RemoveComment(line).Trim();
-            if (IsSection(trimmed))
+            KeyValueSyntax? previous = null;
+            foreach (var item in items)
             {
-                inSection = trimmed.Equals(header, KeyComparison);
-                continue;
+                if (prefix.Concat(Parts(item.Key!)).SequenceEqual(wanted))
+                    return new Entry(item, items, table, previous);
+                previous = item;
             }
-            if (!inSection || trimmed.Length == 0) continue;
-            var equals = trimmed.IndexOf('=');
-            if (equals < 0 || !trimmed[..equals].Trim().Equals(key, KeyComparison)) continue;
-            return trimmed[(equals + 1)..].Trim();
         }
         return null;
     }
 
-    /// <summary>
-    /// Replaces one key in one section, preserving comments, ordering, and every unrelated setting -
-    /// this file belongs to the user and to the in-game settings bar as much as to setup. A
-    /// commented-out key is deliberately not matched, matching the runtime's own writer.
-    /// </summary>
-    private static void SetSectionValue(List<string> lines, string section, string key, string value)
+    private static void Append(DocumentSyntax document, string[] section, KeyValueSyntax item)
     {
-        var header = $"[{section}]";
-        var sectionStart = -1;
-        var sectionEnd = lines.Count;
-        for (var index = 0; index < lines.Count; index++)
+        var text = document.ToString();
+        var newline = text.Contains("\r\n") ? "\r\n" : "\n";
+        var table = Tables(document).FirstOrDefault(t => Parts(t.Name!).SequenceEqual(section));
+        if (table is null)
         {
-            var trimmed = RemoveComment(lines[index]).Trim();
-            if (!IsSection(trimmed)) continue;
-            if (sectionStart >= 0)
-            {
-                sectionEnd = index;
-                break;
-            }
-            if (trimmed.Equals(header, KeyComparison)) sectionStart = index;
-        }
-        if (sectionStart < 0)
-        {
-            if (lines.Count > 0 && lines[^1].Length != 0) lines.Add("");
-            lines.Add(header);
-            lines.Add($"{key} = {value}");
-            return;
+            var name = new KeySyntax(section[0]);
+            foreach (var part in section.Skip(1)) name.DotKeys.Add(new DottedKeyItemSyntax(part));
+            table = new TableSyntax(name) { EndOfLineToken = NewLine(newline) };
+            var blankLine = text.EndsWith("\n\n") || text.EndsWith("\n\r\n");
+            var gap = text.Length == 0 ? 0 : !text.EndsWith('\n') ? 2 : blankLine ? 0 : 1;
+            for (var i = 0; i < gap; i++) (table.LeadingTrivia ??= []).Add(new SyntaxTrivia(TokenKind.NewLine, newline));
+            document.Tables.Add(table);
         }
 
-        for (var index = sectionStart + 1; index < sectionEnd; index++)
-        {
-            var trimmed = RemoveComment(lines[index]).Trim();
-            if (trimmed.Length == 0) continue;
-            var equals = trimmed.IndexOf('=');
-            if (equals >= 0 && trimmed[..equals].Trim().Equals(key, KeyComparison))
-            {
-                lines[index] = $"{key} = {value}";
-                return;
-            }
-        }
-
-        // Append after the section's last real line, not after the blank line that separates it from
-        // the next header, so a generated file stays readable for the user who also edits it by hand.
-        var insertAt = sectionEnd;
-        while (insertAt > sectionStart + 1 && lines[insertAt - 1].Trim().Length == 0) insertAt--;
-        lines.Insert(insertAt, $"{key} = {value}");
+        // Blank lines and comments after the last entry stay after the new one.
+        var last = table.Items.LastOrDefault();
+        var before = last is null ? table.EndOfLineToken ??= NewLine(newline) : last.EndOfLineToken ??= NewLine(newline);
+        item.EndOfLineToken = NewLine(newline);
+        item.EndOfLineToken.TrailingTrivia = before.TrailingTrivia;
+        before.TrailingTrivia = null;
+        table.Items.Add(item);
     }
 
-    private static List<string> ReadLinesOrDefault(string configPath) =>
-        File.Exists(configPath) ? File.ReadAllLines(configPath).ToList() : DefaultConfigLines();
+    private static SyntaxToken NewLine(string text) => new(TokenKind.NewLine, text);
 
-    private static void WriteLines(string configPath, List<string> lines) =>
-        FileSystemUtilities.WriteAtomic(configPath,
-            Encoding.UTF8.GetBytes(string.Join(Environment.NewLine, lines) + Environment.NewLine));
-
-    private static List<string> DefaultConfigLines() =>
-    [
-        "# WiiCompiled user configuration",
-        "",
-        "[paths]"
-    ];
-
-    private static bool IsSection(string value) =>
-        value.Length >= 2 && value[0] == '[' && value[^1] == ']';
-
-    /// <summary>
-    /// Line-for-line port of <c>RemoveComment</c> in <c>runtime_config.h</c>, including its backslash
-    /// escaping inside basic strings: without it a Windows path ending in <c>\\"</c> reads as
-    /// re-opening the quote, and the two implementations disagree about which <c>#</c> starts a comment.
-    /// </summary>
-    private static string RemoveComment(string line)
+    private static SyntaxToken LastToken(SyntaxNode node)
     {
-        var inSingle = false;
-        var inDouble = false;
-        var escaped = false;
-        for (var index = 0; index < line.Length; index++)
-        {
-            var character = line[index];
-            if (inDouble && character == '\\' && !escaped)
-            {
-                escaped = true;
-                continue;
-            }
-            if (character == '\'' && !inDouble) inSingle = !inSingle;
-            else if (character == '"' && !inSingle && !escaped) inDouble = !inDouble;
-            else if (character == '#' && !inSingle && !inDouble) return line[..index];
-            escaped = false;
-        }
-        return line;
-    }
-
-    private static string QuoteToml(string value) =>
-        "\"" + value.Replace("\\", "\\\\").Replace("\"", "\\\"") + "\"";
-
-    public static bool TryUnquoteToml(string value, out string result)
-    {
-        result = "";
-        if (value.Length < 2) return false;
-        // Literal strings carry no escapes; the runtime's parser treats them the same way.
-        if (value[0] == '\'' && value[^1] == '\'')
-        {
-            result = value[1..^1];
-            return true;
-        }
-        if (value[0] != '"' || value[^1] != '"') return false;
-        var builder = new StringBuilder(value.Length - 2);
-        for (var index = 1; index < value.Length - 1; index++)
-        {
-            if (value[index] == '\\' && index + 1 < value.Length - 1)
-            {
-                var escaped = value[++index];
-                builder.Append(escaped switch { '\\' => '\\', '"' => '"', _ => escaped });
-            }
-            else
-            {
-                builder.Append(value[index]);
-            }
-        }
-        result = builder.ToString();
-        return true;
+        while (node is not SyntaxToken)
+            node = node.GetChild(node.ChildrenCount - 1)!;
+        return (SyntaxToken)node;
     }
 }

Launcher/WiiCompiled.Setup.Common/WiiCompiled.Setup.Common.csproj

@@ -11,4 +11,7 @@
     <DebugType>embedded</DebugType>
     <SatelliteResourceLanguages>en</SatelliteResourceLanguages>
   </PropertyGroup>
+  <ItemGroup>
+    <PackageReference Include="Tomlyn" Version="2.10.1" />
+  </ItemGroup>
 </Project>

Launcher/WiiCompiled.Setup.Linux/BuildRunner.cs

@@ -82,6 +82,7 @@ internal static class BuildRunner
         {
             startInfo.ArgumentList.Add("--sysroot"); startInfo.ArgumentList.Add(sysroot);
         }
+        startInfo.ArgumentList.Add("--setup-version"); startInfo.ArgumentList.Add(ProductInfo.Version);
 
         using var process = new Process { StartInfo = startInfo };
         var window = new BuildProgressWindow(reporter, InstallStages.Build, start: 6, end: 96);

Launcher/WiiCompiled.Setup.Linux/DiscTool.cs

@@ -15,6 +15,12 @@ internal static class DiscTool
         string isoPath, ProjectManifest manifest, string assetsDirectory, string workspace,
         string? nodToolBin, IInstallReporter reporter, CancellationToken cancellationToken)
     {
+        if (IsAlreadyExtracted(manifest, assetsDirectory))
+        {
+            reporter.Progress(InstallStages.ExtractDisc, "Reusing the verified extracted disc", 6);
+            return;
+        }
+
         var nodTool = nodToolBin ?? await NodToolProvider.ResolveAsync(workspace, cancellationToken);
 
         // `nodtool info` only decodes the disc/partition headers (milliseconds); `nodtool extract`
@@ -35,6 +41,8 @@ internal static class DiscTool
         // time, so it has to survive past this install (see Program.cs, which points dvd_root here).
         reporter.Progress(InstallStages.ExtractDisc, "Extracting the disc image", 4);
         var dataDir = Path.Combine(assetsDirectory, "DATA");
+        Directory.CreateDirectory(assetsDirectory);
+        File.Delete(Path.Combine(assetsDirectory, ExtractionMarker));
         if (Directory.Exists(dataDir)) Directory.Delete(dataDir, recursive: true);
         await RunExtractAsync(nodTool, isoPath, dataDir, cancellationToken);
 
@@ -61,6 +69,7 @@ internal static class DiscTool
         Directory.CreateDirectory(assetsDirectory);
         File.Copy(dolPath, Path.Combine(assetsDirectory, "main.dol"), overwrite: true);
         File.Copy(relPath, Path.Combine(assetsDirectory, "StaticR.rel"), overwrite: true);
+        File.WriteAllText(Path.Combine(assetsDirectory, ExtractionMarker), ExtractionRecord(dolSha, dataDir));
         reporter.Progress(InstallStages.ExtractDisc, "Disc validated and extracted", 6);
     }
 
@@ -86,6 +95,36 @@ internal static class DiscTool
         return stdout;
     }
 
+    private const string ExtractionMarker = ".disc-extracted";
+
+    // Only one disc revision is accepted, so a completed, verified extraction is identical to a fresh one.
+    private static bool IsAlreadyExtracted(ProjectManifest manifest, string assetsDirectory)
+    {
+        var marker = Path.Combine(assetsDirectory, ExtractionMarker);
+        var dol = Path.Combine(assetsDirectory, "main.dol");
+        var rel = Path.Combine(assetsDirectory, "StaticR.rel");
+        var dataDir = Path.Combine(assetsDirectory, "DATA");
+        if (!File.Exists(marker) || !File.Exists(dol) || !File.Exists(rel) ||
+            !Directory.Exists(Path.Combine(dataDir, "files")))
+            return false;
+        return string.Equals(File.ReadAllText(marker), ExtractionRecord(manifest.DolSha256, dataDir),
+                   StringComparison.Ordinal) &&
+               string.Equals(Sha256Of(dol), manifest.DolSha256, StringComparison.Ordinal) &&
+               string.Equals(Sha256Of(rel), manifest.RelSha256, StringComparison.Ordinal);
+    }
+
+    // The DOL hash plus every extracted file's path and size, so a later deletion or truncation inside
+    // DATA forces a fresh extraction.
+    private static string ExtractionRecord(string dolSha, string dataDir)
+    {
+        var record = new System.Text.StringBuilder(dolSha).Append('\n');
+        foreach (var path in Directory.EnumerateFiles(dataDir, "*", SearchOption.AllDirectories)
+                     .Select(path => Path.GetRelativePath(dataDir, path).Replace('\\', '/'))
+                     .Order(StringComparer.Ordinal))
+            record.Append(new FileInfo(Path.Combine(dataDir, path)).Length).Append(' ').Append(path).Append('\n');
+        return record.ToString();
+    }
+
     private static string Sha256Of(string path)
     {
         using var stream = File.OpenRead(path);

Launcher/WiiCompiled.Setup.Linux/InstallProgress.cs

@@ -1,209 +0,0 @@
-using System.Text.Json;
-
-namespace WiiCompiled.Setup.Linux;
-
-// Ported near-verbatim from Launcher/WiiCompiled.Setup/InstallProgress.cs: this whole file is
-// platform-neutral (System.Text.Json + Console only), so the NDJSON --progress-json wire protocol
-// stays byte-for-byte the same shape a future GUI already speaks on Windows.
-
-/// <summary>
-/// Stable stage identifiers reported by <c>--progress-json</c>. Kept intentionally small for this
-/// lean Linux installer (no toolkit-extraction/publish-transaction stages, since there is no
-/// bundled toolkit or staged workspace copy here - see the plan's "operate on a git checkout"
-/// scoping decision).
-/// </summary>
-internal static class InstallStages
-{
-    public const string Validate = "validate";
-    public const string ExtractDisc = "extract-disc";
-    public const string Build = "build";
-    public const string Shortcuts = "shortcuts";
-}
-
-/// <summary>
-/// Where an installation reports what it is doing. Progress is coarse and monotonic; raw translator
-/// and compiler output is a diagnostic, never progress, because it is unbounded and machine-hostile.
-/// </summary>
-internal interface IInstallReporter
-{
-    void Progress(string stage, string message, int percent);
-    void Diagnostic(string line);
-}
-
-/// <summary>
-/// The <c>--progress-json</c> protocol: one JSON object per line on stdout, nothing else on stdout,
-/// diagnostics on stderr. The terminal <c>result</c> line is written exactly once.
-/// </summary>
-internal sealed class NdjsonInstallReporter : IInstallReporter
-{
-    private static readonly JsonSerializerOptions Options = new() { WriteIndented = false };
-    private readonly object _gate = new();
-    private int _lastPercent;
-    private bool _finished;
-
-    public void Progress(string stage, string message, int percent)
-    {
-        lock (_gate)
-        {
-            if (_finished) return;
-            // Percentages are clamped monotonic: a caller's progress bar must never walk backwards
-            // because a later stage happened to estimate a lower number.
-            _lastPercent = Math.Clamp(Math.Max(percent, _lastPercent), 0, 99);
-            WriteLine(new { type = "progress", stage, message, percent = _lastPercent });
-        }
-    }
-
-    public void Diagnostic(string line) => Console.Error.WriteLine(line);
-
-    public void Success(string installDirectory)
-    {
-        lock (_gate)
-        {
-            if (_finished) return;
-            _finished = true;
-            WriteLine(new { type = "result", success = true, version = ProductInfo.Version, installDir = installDirectory });
-        }
-    }
-
-    public void Failure(string error)
-    {
-        lock (_gate)
-        {
-            if (_finished) return;
-            _finished = true;
-            WriteLine(new { type = "result", success = false, error });
-        }
-    }
-
-    /// <summary>
-    /// The terminal result line is the caller's only completion signal, so no exit path may skip it.
-    /// Callers invoke this from a finally block; it is a no-op once a result was already written.
-    /// </summary>
-    public void EnsureFinished(string errorIfUnfinished) => Failure(errorIfUnfinished);
-
-    private static void WriteLine(object value)
-    {
-        Console.Out.WriteLine(JsonSerializer.Serialize(value, Options));
-        Console.Out.Flush();
-    }
-}
-
-/// <summary>Plain-text console reporting for a run without <c>--progress-json</c>.</summary>
-internal sealed class ConsoleInstallReporter : IInstallReporter
-{
-    public void Progress(string stage, string message, int percent) =>
-        Console.Out.WriteLine($"[{percent,3}%] {message}");
-
-    public void Diagnostic(string line) => Console.Out.WriteLine(line);
-}
-
-/// <summary>
-/// Build step identifiers from local-build.sh's <c>MKWCBUILD:STEP:&lt;id&gt;</c> lines - the id is
-/// the contract, matched against Launcher/local-build.sh's log_step() call sites.
-/// </summary>
-internal static class BuildStepIds
-{
-    public const string BuildTranslator = "build-translator";
-    public const string ReuseBaseTranslation = "reuse-base-translation";
-    public const string RetranslateBase = "retranslate-base";
-    public const string TranslateBase = "translate-base";
-    public const string EmitBaseManifest = "emit-base-manifest";
-    public const string TranslateMod = "translate-mod";
-    public const string GenerateDataInit = "generate-data-init";
-    public const string EmitBuildShards = "emit-build-shards";
-    public const string ConfigureNative = "configure-native";
-    public const string Compile = "compile";
-}
-
-/// <summary>
-/// Maps one local-build.sh run onto a slice of the overall percentage. local-build.sh announces
-/// every step it starts with an <c>MKWCBUILD:</c> prefix, so the slice can advance on real events
-/// instead of on a timer.
-/// </summary>
-internal sealed class BuildProgressWindow
-{
-    private const string Marker = "MKWCBUILD:";
-    private const string StepMarker = "STEP:";
-
-    /// <summary>The fraction the compile step reaches; beyond it, compiler output is a heartbeat.</summary>
-    private const double CompileFraction = 0.58;
-
-    private static readonly (string Id, double Fraction, string Message)[] Steps =
-    [
-        (BuildStepIds.BuildTranslator, 0.04, "Building the translator"),
-        (BuildStepIds.ReuseBaseTranslation, 0.30, "Reusing the completed base translation"),
-        (BuildStepIds.RetranslateBase, 0.08, "The base translation is stale; retranslating it"),
-        (BuildStepIds.TranslateBase, 0.10, "Translating Mario Kart Wii"),
-        (BuildStepIds.EmitBaseManifest, 0.34, "Creating the translation manifest"),
-        (BuildStepIds.TranslateMod, 0.38, "Translating the Retro Rewind Code.pul"),
-        (BuildStepIds.GenerateDataInit, 0.44, "Generating game data initialization"),
-        (BuildStepIds.EmitBuildShards, 0.48, "Preparing the native build"),
-        (BuildStepIds.ConfigureNative, 0.52, "Configuring the compiler"),
-        (BuildStepIds.Compile, CompileFraction, "Compiling the game. This is the longest step"),
-    ];
-
-    private readonly IInstallReporter _reporter;
-    private readonly string _stage;
-    private readonly int _start;
-    private readonly int _end;
-    private double _fraction;
-    private string _message = "Preparing the local build";
-    private int _reportedPercent = -1;
-
-    public BuildProgressWindow(IInstallReporter reporter, string stage, int start, int end)
-    {
-        _reporter = reporter;
-        _stage = stage;
-        _start = start;
-        _end = end;
-    }
-
-    public void Observe(string line)
-    {
-        var index = line.IndexOf(Marker, StringComparison.Ordinal);
-        if (index >= 0)
-        {
-            var text = line[(index + Marker.Length)..].Trim();
-            if (text.StartsWith(StepMarker, StringComparison.Ordinal))
-            {
-                var identifier = text[StepMarker.Length..];
-                var end = identifier.IndexOf(' ');
-                if (end >= 0) identifier = identifier[..end];
-                foreach (var (id, fraction, message) in Steps)
-                {
-                    if (!id.Equals(identifier, StringComparison.Ordinal)) continue;
-                    if (fraction > _fraction)
-                    {
-                        _fraction = fraction;
-                        _message = message;
-                        Emit();
-                    }
-                    return;
-                }
-            }
-        }
-
-        // Anything else - a plain MKWCBUILD note, or raw tool output - stays a diagnostic and only
-        // feeds the heartbeat below.
-        _reporter.Diagnostic(line);
-        // Compilation announces itself once and then emits thousands of compiler lines. Treat that
-        // output as a heartbeat so the slice keeps creeping forward, but only publish a progress
-        // line when the rounded percentage actually changes.
-        if (_fraction >= CompileFraction)
-        {
-            _fraction = Math.Min(0.97, _fraction + 0.0015);
-            Emit();
-        }
-    }
-
-    private void Emit()
-    {
-        var percent = Interpolate(_fraction);
-        if (percent == _reportedPercent) return;
-        _reportedPercent = percent;
-        _reporter.Progress(_stage, _message, percent);
-    }
-
-    private int Interpolate(double fraction) =>
-        (int)Math.Round(_start + (_end - _start) * Math.Clamp(fraction, 0, 1));
-}

Launcher/WiiCompiled.Setup.Linux/InstallStages.cs

@@ -0,0 +1,10 @@
+namespace WiiCompiled.Setup.Linux;
+
+/// <summary>Stable stage identifiers reported by <c>--progress-json</c>.</summary>
+internal static class InstallStages
+{
+    public const string Validate = "validate";
+    public const string ExtractDisc = "extract-disc";
+    public const string Build = "build";
+    public const string Shortcuts = "shortcuts";
+}

Launcher/WiiCompiled.Setup.Linux/Program.cs

@@ -31,7 +31,7 @@ internal static class Program
         var (command, flags) = ParseArgs(args);
         if (command is null) { PrintUsage(); return 1; }
         var progressJson = flags.ContainsKey("progress-json");
-        IInstallReporter reporter = progressJson ? new NdjsonInstallReporter() : new ConsoleInstallReporter();
+        IInstallReporter reporter = progressJson ? new NdjsonInstallReporter(ProductInfo.Version) : new ConsoleInstallReporter();
 
         try
         {

Launcher/WiiCompiled.Setup.Linux/WiiCompiled.Setup.Linux.csproj

@@ -15,5 +15,6 @@
   </PropertyGroup>
   <ItemGroup>
     <ProjectReference Include="..\WiiCompiled.Setup.Common\WiiCompiled.Setup.Common.csproj" />
+    <Using Include="WiiCompiled.Setup.Common" />
   </ItemGroup>
 </Project>

Launcher/WiiCompiled.Setup.Windows/ConsoleCommands.cs

@@ -57,7 +57,7 @@ internal static class ConsoleCommands
 
     public static int VerifyInputs(CommandLine command)
     {
-        var reporter = command.ProgressJson ? new NdjsonInstallReporter() : null;
+        var reporter = command.ProgressJson ? new NdjsonInstallReporter(ProductInfo.Version) : null;
         var temp = Path.Combine(Path.GetTempPath(), "mkwc-verify-" + Guid.NewGuid().ToString("N"));
         Directory.CreateDirectory(temp);
         try
@@ -107,7 +107,7 @@ internal static class ConsoleCommands
         }
         File.WriteAllText(logPath, $"{ProductInfo.Name} setup {ProductInfo.Version}{Environment.NewLine}");
 
-        var ndjson = command.ProgressJson ? new NdjsonInstallReporter(Log) : null;
+        var ndjson = command.ProgressJson ? new NdjsonInstallReporter(ProductInfo.Version, Log) : null;
         IInstallReporter reporter = ndjson ?? (IInstallReporter)new ConsoleInstallReporter(Log);
         var installDirectory = Path.GetFullPath(command.InstallDirectory!);
         try
@@ -145,7 +145,7 @@ internal static class ConsoleCommands
     public static int CheckProducts(CommandLine command,
         CancellationToken cancellationToken = default)
     {
-        var reporter = command.ProgressJson ? new NdjsonInstallReporter() : null;
+        var reporter = command.ProgressJson ? new NdjsonInstallReporter(ProductInfo.Version) : null;
         var installation = new Installation(string.IsNullOrWhiteSpace(command.InstallDirectory)
             ? AppContext.BaseDirectory
             : Path.GetFullPath(command.InstallDirectory));
@@ -198,7 +198,7 @@ internal static class ConsoleCommands
             }
         }
 
-        var ndjson = command.ProgressJson ? new NdjsonInstallReporter(Log) : null;
+        var ndjson = command.ProgressJson ? new NdjsonInstallReporter(ProductInfo.Version, Log) : null;
         IInstallReporter reporter = ndjson is null ? new ConsoleInstallReporter(Log) : ndjson;
         var installDirectory = Path.GetFullPath(command.InstallDirectory!);
         var installation = new Installation(installDirectory);

Launcher/WiiCompiled.Setup.Windows/InstallStages.cs

@@ -0,0 +1,17 @@
+namespace WiiCompiled.Setup.Windows;
+
+/// <summary>
+/// Stable stage identifiers reported by <c>--progress-json</c>. These are part of the public
+/// WheelWizard contract (docs/WHEELWIZARD_CONTRACT.md): the strings may gain new members, but an
+/// existing identifier never changes meaning.
+/// </summary>
+internal static class InstallStages
+{
+    public const string Validate = "validate";
+    public const string ExtractToolkit = "extract-toolkit";
+    public const string ExtractDisc = "extract-disc";
+    public const string PrepareRetro = "prepare-retro";
+    public const string BuildBase = "build-base";
+    public const string BuildRetro = "build-retro";
+    public const string Publish = "publish";
+}

Launcher/WiiCompiled.Setup.Windows/InstallerEngine.cs

@@ -501,17 +501,31 @@ internal sealed class InstallerEngine
         var dol = Path.Combine(dataRoot, "sys", "main.dol");
         var rel = Path.Combine(dataRoot, "files", "rel", "StaticR.rel");
         var fst = Path.Combine(dataRoot, "sys", "fst.bin");
-        if (!File.Exists(dol) || !File.Exists(rel) || !File.Exists(fst))
-            throw new InvalidDataException("The disc extraction is missing required Mario Kart Wii files.");
-        var dolHash = InputValidation.Sha256File(dol);
-        var relHash = InputValidation.Sha256File(rel);
-        if (!dolHash.Equals(manifest.ExpectedDolSha256, StringComparison.OrdinalIgnoreCase) ||
-            !relHash.Equals(manifest.ExpectedRelSha256, StringComparison.OrdinalIgnoreCase))
-        {
+        var missing = new[] { dol, rel, fst }.Where(path => !File.Exists(path))
+            .Select(path => Path.GetRelativePath(dataRoot, path)).ToList();
+        if (missing.Count > 0)
             throw new InvalidDataException(
-                "The disc is RMCP01 but does not match the supported clean PAL revision. " +
-                "Patched or otherwise modified game code cannot be installed safely.");
+                "The disc extraction is missing required Mario Kart Wii files: " +
+                string.Join(", ", missing) + ".");
+
+        var mismatches = new List<string>();
+        foreach (var (name, path, expected) in new[]
+                 {
+                     ("main.dol", dol, manifest.ExpectedDolSha256),
+                     ("StaticR.rel", rel, manifest.ExpectedRelSha256)
+                 })
+        {
+            var actual = InputValidation.Sha256File(path);
+            if (!actual.Equals(expected, StringComparison.OrdinalIgnoreCase))
+                mismatches.Add($"{name} is {new FileInfo(path).Length:N0} bytes with SHA-256 {actual} " +
+                               $"(expected {expected})");
         }
+
+        if (mismatches.Count > 0)
+            throw new InvalidDataException(
+                "The disc is RMCP01 but its game code does not match the supported clean PAL revision: " +
+                string.Join("; ", mismatches) + ". " +
+                "Patched or otherwise modified game code cannot be installed safely.");
     }
 
     private static void ValidateInstallDirectory(string path) =>

Launcher/WiiCompiled.Setup.Windows/Program.cs

@@ -31,7 +31,7 @@ internal static class Program
             {
                 // A command line was supplied, so a caller is driving this process. Rejecting the
                 // command with a modal dialog would hang that caller forever.
-                if (progressJson) new NdjsonInstallReporter().Failure(ex.Message);
+                if (progressJson) new NdjsonInstallReporter(ProductInfo.Version).Failure(ex.Message);
                 else Console.Error.WriteLine(ex.Message);
                 return 1;
             }
@@ -91,7 +91,7 @@ internal static class Program
             // errors raised outside it.
             if (progressJson)
             {
-                new NdjsonInstallReporter().Failure(ex.Message);
+                new NdjsonInstallReporter(ProductInfo.Version).Failure(ex.Message);
                 Console.Error.WriteLine(ex);
             }
             else

Launcher/WiiCompiled.Setup.Windows/SelfTests.cs

@@ -1,5 +1,6 @@
 using System.Buffers.Binary;
 using System.Security.Cryptography;
+using Tomlyn;
 using WiiCompiled.Setup.Common;
 
 namespace WiiCompiled.Setup.Windows;
@@ -51,6 +52,7 @@ internal static class SelfTests
             RuntimeConfiguration.SetRetroRewindRoot, RuntimeConfiguration.RemoveRetroRewindRootIfOwned),
             failures);
         Test("Runtime configuration key matching", TestRuntimeConfigurationKeyMatching, failures);
+        Test("Runtime configuration quoted keys", TestRuntimeConfigurationQuotedKeys, failures);
         Test("Install location validation", TestInstallLocationValidation, failures);
         Test("Command line contract", TestCommandLineContract, failures);
         Test("Portable root discovery", TestPortableRootDiscovery, failures);
@@ -229,6 +231,70 @@ internal static class SelfTests
         }
     }
 
+    /// <summary>
+    /// Quoted and bare spellings of a key or table are the same TOML key, so rewriting replaces the
+    /// entry in place, keeps everything else byte-identical, and never touches an unparseable file.
+    /// </summary>
+    private static void TestRuntimeConfigurationQuotedKeys()
+    {
+        var root = Path.Combine(Path.GetTempPath(), "mkwc-tomlquoted-" + Guid.NewGuid().ToString("N"));
+        var config = Path.Combine(root, RuntimeConfiguration.ConfigFileName);
+        var owned = Path.Combine(root, "Owned Location");
+        var fullOwned = Path.GetFullPath(owned);
+        try
+        {
+            Directory.CreateDirectory(owned);
+            foreach (var header in new[] { "[\"paths\"]", "[ 'paths' ]", "[paths]" })
+            foreach (var key in new[] { "\"dvd_root\"", "'dvd_root'", "\"dvd\\u005Froot\"", "dvd_root" })
+            {
+                File.WriteAllText(config,
+                    $"# top\r\n[video]\r\nwidescreen = true # keep\r\n\r\n{header}\r\n  {key}=\"D:\\\\Old\" # note\r\nother = 1\r\n");
+                RuntimeConfiguration.SetDvdRoot(config, owned);
+                var text = File.ReadAllText(config);
+                var expected = $"# top\r\n[video]\r\nwidescreen = true # keep\r\n\r\n{header}\r\n  {key}=" +
+                               $"\"{fullOwned.Replace("\\", "\\\\")}\" # note\r\nother = 1\r\n";
+                if (text != expected)
+                    throw new Exception($"{header} {key} was not replaced in place:\r\n{text}");
+                if (RuntimeConfiguration.GetResolvedPath(config, "dvd_root") != fullOwned)
+                    throw new Exception($"The replaced {key} could not be read back.");
+            }
+
+            File.WriteAllText(config, "[paths]\r\nother = 1\r\n\r\n# next\r\n[video]\r\nx = 2");
+            RuntimeConfiguration.SetDvdRoot(config, owned);
+            var appended = File.ReadAllText(config);
+            if (!appended.StartsWith("[paths]\r\nother = 1\r\ndvd_root = ", StringComparison.Ordinal) ||
+                !appended.EndsWith("\r\n\r\n# next\r\n[video]\r\nx = 2", StringComparison.Ordinal))
+                throw new Exception("A missing key was not appended to its section:\r\n" + appended);
+
+            File.WriteAllText(config, "dvd_root = \"D:\\\\Root\"\r\n[video]\r\nx = 2\r\n");
+            RuntimeConfiguration.SetDvdRoot(config, owned);
+            var created = File.ReadAllText(config);
+            if (!created.StartsWith("dvd_root = \"D:\\\\Root\"\r\n[video]\r\nx = 2\r\n\r\n[paths]\r\n", StringComparison.Ordinal) ||
+                RuntimeConfiguration.GetResolvedPath(config, "dvd_root") != fullOwned)
+                throw new Exception("A missing section was not created without touching a same-named root key:\r\n" + created);
+
+            File.Delete(config);
+            RuntimeConfiguration.SetDvdRoot(config, owned);
+            if (RuntimeConfiguration.GetResolvedPath(config, "dvd_root") != fullOwned)
+                throw new Exception("A missing config file was not created.");
+
+            const string broken = "[paths]\r\ndvd_root = 1\r\ndvd_root = 2\r\n";
+            File.WriteAllText(config, broken);
+            try
+            {
+                RuntimeConfiguration.SetDvdRoot(config, owned);
+                throw new Exception("An unparseable config was rewritten.");
+            }
+            catch (TomlException) { }
+            if (File.ReadAllText(config) != broken)
+                throw new Exception("An unparseable config was modified.");
+        }
+        finally
+        {
+            if (Directory.Exists(root)) Directory.Delete(root, recursive: true);
+        }
+    }
+
     /// <summary>
     /// One location rule set covers both an installation directory and a portable root, so the
     /// rejections a portable install performs are the ones an ordinary install performs too.
@@ -1073,10 +1139,10 @@ internal static class SelfTests
             if (topologyChangedRuntimeAssets == changedRuntimeAssets)
                 throw new Exception("An empty copied runtime-asset directory was omitted from its identity.");
 
-            // Cache-reuse subsets: translator or runtime/src changes (the translator regex-scans it
-            // for native overrides and effect contracts) invalidate translation but not toolchain,
-            // since CMake/Ninja rebuild exactly the affected objects; a compiler change invalidates
-            // only toolchain; a native_prebuilt provenance rewrite invalidates neither.
+            // Cache-reuse subsets: translator or native registration changes invalidate translation
+            // but not toolchain, since CMake/Ninja rebuild exactly the affected objects; a compiler
+            // change invalidates only toolchain; a native_prebuilt provenance rewrite invalidates
+            // neither. Without a current native index, runtime/src itself decides translation.
             var components = ToolkitFingerprint.ComputeComponents(root);
             if (components.Compile != afterPatch)
                 throw new Exception("ComputeComponents disagrees with the compile identity.");
@@ -1087,15 +1153,36 @@ internal static class SelfTests
                 throw new Exception("A new translator did not change the translation identity.");
             if (afterNewTranslator.NativeToolchain != components.NativeToolchain)
                 throw new Exception("A new translator needlessly changed the native toolchain identity.");
-            File.WriteAllText(Path.Combine(root, "BuildWorkspace", "runtime", "src", "main.cpp"),
-                "int main3();");
+            var runtimeMain = Path.Combine(root, "BuildWorkspace", "runtime", "src", "main.cpp");
+            File.WriteAllText(runtimeMain, "int main3();");
+            var afterUnindexedRuntime = ToolkitFingerprint.ComputeComponents(root);
+            if (afterUnindexedRuntime.Translation == afterNewTranslator.Translation)
+                throw new Exception("A runtime source change without a native index did not change the translation identity.");
+
+            File.WriteAllText(runtimeMain, "int main();");
+            if (WriteFakeNativeIndex(root, "\"a\"") !=
+                "50b9779591f4b205751361c8899accd362f96ea5dfd32febed1bedca534c523a")
+                throw new Exception("The native index source digest no longer matches the translator's.");
+            var indexed = ToolkitFingerprint.ComputeComponents(root);
+            File.WriteAllText(runtimeMain, "int main3();");
+            WriteFakeNativeIndex(root, "\"a\"");
             var afterNewRuntime = ToolkitFingerprint.ComputeComponents(root);
-            if (afterNewRuntime.Translation == afterNewTranslator.Translation)
-                throw new Exception("A runtime source change did not change the translation identity.");
-            if (afterNewRuntime.NativeToolchain != afterNewTranslator.NativeToolchain)
+            if (afterNewRuntime.Translation != indexed.Translation)
+                throw new Exception("A runtime source change with unchanged registrations changed the translation identity.");
+            if (afterNewRuntime.NativeToolchain != indexed.NativeToolchain)
                 throw new Exception("A runtime source change needlessly changed the native toolchain identity.");
-            if (afterNewRuntime.Compile == afterNewTranslator.Compile)
+            if (afterNewRuntime.Compile == indexed.Compile)
                 throw new Exception("A runtime source change did not change the compile identity.");
+            File.WriteAllText(runtimeMain, "int main4();");
+            if (ToolkitFingerprint.ComputeComponents(root).Translation == afterNewRuntime.Translation)
+                throw new Exception("A stale native index was trusted for the translation identity.");
+            File.WriteAllText(runtimeMain, "int main3();");
+            WriteFakeNativeIndex(root, "\"a\", \"b\"");
+            afterNewRuntime = ToolkitFingerprint.ComputeComponents(root);
+            if (afterNewRuntime.Translation == indexed.Translation)
+                throw new Exception("A native registration change did not change the translation identity.");
+            if (afterNewRuntime.NativeToolchain != indexed.NativeToolchain)
+                throw new Exception("A native registration change needlessly changed the native toolchain identity.");
             File.WriteAllText(Path.Combine(root, "Toolkit", "llvm-mingw", "bin", "clang-22.exe"),
                 "clang-v2");
             var afterNewCompiler = ToolkitFingerprint.ComputeComponents(root);
@@ -1148,6 +1235,26 @@ internal static class SelfTests
         }
     }
 
+    /// <summary>Writes a native index current for the fake runtime/src; returns its source digest.</summary>
+    private static string WriteFakeNativeIndex(string root, string registrations)
+    {
+        var source = Path.Combine(root, "BuildWorkspace", "runtime", "src");
+        var manifest = new System.Text.StringBuilder();
+        foreach (var (relative, path) in Directory.EnumerateFiles(source, "*", SearchOption.AllDirectories)
+                     .Where(path => path.EndsWith(".cpp", StringComparison.OrdinalIgnoreCase) ||
+                                    path.EndsWith(".h", StringComparison.OrdinalIgnoreCase) ||
+                                    path.EndsWith(".hpp", StringComparison.OrdinalIgnoreCase))
+                     .Select(path => (Path.GetRelativePath(source, path).Replace('\\', '/'), path))
+                     .OrderBy(item => item.Item1, StringComparer.Ordinal))
+            manifest.Append(InputValidation.Sha256File(path)).Append("  ").Append(relative).Append('\n');
+        var digest = Convert.ToHexString(SHA256.HashData(System.Text.Encoding.UTF8.GetBytes(manifest.ToString())))
+            .ToLowerInvariant();
+        File.WriteAllText(Path.Combine(root, "BuildWorkspace", "runtime", "native-index.json"),
+            "{\n  \"format\": \"wiicompiled-runtime-native-index\",\n  \"sourceDigest\": \"" + digest +
+            "\",\n  \"registrations\": [" + registrations + "]\n}\n");
+        return digest;
+    }
+
     private static void CreateFakeToolkit(string root)
     {
         Directory.CreateDirectory(Path.Combine(root, "Toolkit", "Translator"));

Launcher/WiiCompiled.Setup.Windows/ToolkitFingerprint.cs

@@ -1,5 +1,7 @@
 using System.Security.Cryptography;
 using System.Text;
+using System.Text.Json;
+using System.Text.Json.Nodes;
 using WiiCompiled.Setup.Common;
 
 namespace WiiCompiled.Setup.Windows;
@@ -12,9 +14,10 @@ namespace WiiCompiled.Setup.Windows;
 internal static class ToolkitFingerprint
 {
     // v6/native-toolchain v2 forces one clean rebuild past a Ninja stale-object bug.
-    // translation v2 added runtime/src, which the translator regex-scans for native overrides.
+    // translation v2 added runtime/src, which the translator regex-scans for native overrides;
+    // v3 replaced it with the native index extracted from it, whenever that index is current.
     private const string Version = "mkwc-toolkit-compile-v6";
-    private const string TranslationVersion = "mkwc-toolkit-translation-v2";
+    private const string TranslationVersion = "mkwc-toolkit-translation-v3";
     private const string NativeToolchainVersion = "mkwc-toolkit-native-toolchain-v2";
     private const string PackageVersion = "mkwc-toolkit-package-v1";
     private const string RuntimeAssetsVersion = "mkwc-product-runtime-assets-v3";
@@ -25,20 +28,25 @@ internal static class ToolkitFingerprint
          ".cmake", ".txt", ".yml", ".yaml", ".toml", ".json", ".ps1", ".in", ".def", ".patch"];
 
     /// <summary>
-    /// Inputs that decide translator output: translator, translation project, runtime sources it
-    /// scans for native override registrations/effect contracts, and the build script's translator
-    /// command lines. Unchanged means the completed base translation can be reused instead of re-run.
+    /// Inputs that decide translator output: translator, translation project, the native override
+    /// registrations/effect contracts it extracts from runtime/src (see
+    /// <see cref="AddNativeRegistrations"/>), and the build script's translator command lines.
+    /// Unchanged means the completed base translation can be reused instead of re-run.
     /// </summary>
     private static readonly string[] TranslationPrefixes =
     [
         InstalledLayout.ToolkitEntryPrefix + "Translator/",
-        InstalledLayout.WorkspaceEntryPrefix + "projects/",
-        InstalledLayout.WorkspaceEntryPrefix + "runtime/src/"
+        InstalledLayout.WorkspaceEntryPrefix + "projects/"
     ];
 
     private static readonly string[] TranslationFiles =
         [InstalledLayout.WorkspaceEntryPrefix + "LocalBuild.ps1"];
 
+    private const string NativeSourcePrefix = InstalledLayout.WorkspaceEntryPrefix + "runtime/src/";
+    private const string NativeIndexEntry = InstalledLayout.WorkspaceEntryPrefix + "runtime/native-index.json";
+    // The file set the translator's native index source digest covers.
+    private static readonly string[] NativeIndexSourceExtensions = [".cpp", ".h", ".hpp"];
+
     /// <summary>
     /// Inputs that make compiled objects unsafe to reuse: the compiler/CMake/Ninja toolchain, the
     /// shipped runtime DLLs, and the scripts owning configure flags. Source/dependency changes are
@@ -108,14 +116,16 @@ internal static class ToolkitFingerprint
         if (entries.Count == 0)
             throw new InvalidDataException($"No toolkit files were found under {root}; the installation is incomplete.");
 
+        var translation = SelectSubset(entries, TranslationPrefixes, TranslationFiles);
+        AddNativeRegistrations(translation, entries, root);
         return new ToolkitFingerprintComponents(
             BuildIdentity(Version, entries),
-            BuildSubsetIdentity(TranslationVersion, entries, TranslationPrefixes, TranslationFiles),
-            BuildSubsetIdentity(NativeToolchainVersion, entries, NativeToolchainPrefixes,
-                NativeToolchainFiles));
+            BuildIdentity(TranslationVersion, translation),
+            BuildIdentity(NativeToolchainVersion,
+                SelectSubset(entries, NativeToolchainPrefixes, NativeToolchainFiles)));
     }
 
-    private static string BuildSubsetIdentity(string version, SortedDictionary<string, string> entries,
+    private static SortedDictionary<string, string> SelectSubset(SortedDictionary<string, string> entries,
         string[] prefixes, string[] files)
     {
         var subset = new SortedDictionary<string, string>(StringComparer.Ordinal);
@@ -125,9 +135,61 @@ internal static class ToolkitFingerprint
                 files.Any(file => relative.Equals(file, StringComparison.Ordinal)))
                 subset[relative] = hash;
         }
-        return BuildIdentity(version, subset);
+        return subset;
     }
 
+    /// <summary>
+    /// A native-index.json extracted from exactly the shipped runtime/src stands in for those
+    /// sources, minus its source digest, so a runtime change that leaves the registrations alone
+    /// keeps the base translation. A missing, unreadable or stale index keeps the sources instead.
+    /// </summary>
+    private static void AddNativeRegistrations(SortedDictionary<string, string> subset,
+        SortedDictionary<string, string> entries, string root)
+    {
+        var sources = entries.Where(entry => entry.Key.StartsWith(NativeSourcePrefix, StringComparison.Ordinal))
+            .ToList();
+        var registrations = TryHashCurrentNativeIndex(
+            Path.Combine(InstalledLayout.Workspace(root), "runtime", "native-index.json"), sources);
+        if (registrations is not null)
+        {
+            subset[NativeIndexEntry] = registrations;
+            return;
+        }
+        foreach (var (relative, hash) in sources) subset[relative] = hash;
+    }
+
+    private static string? TryHashCurrentNativeIndex(string path, List<KeyValuePair<string, string>> sources)
+    {
+        if (!File.Exists(path)) return null;
+        JsonObject? document;
+        try { document = JsonNode.Parse(File.ReadAllText(path)) as JsonObject; }
+        catch (Exception ex) when (ex is IOException or UnauthorizedAccessException or JsonException)
+        {
+            return null;
+        }
+        if (document?["sourceDigest"] is not JsonValue recorded || !recorded.TryGetValue<string>(out var digest))
+            return null;
+
+        // Mirrors the translator's RuntimeNativeIndexFile.ComputeSourceDigest.
+        var manifest = new StringBuilder();
+        foreach (var (relative, hash) in sources)
+        {
+            if (!NativeIndexSourceExtensions.Any(extension =>
+                    relative.EndsWith(extension, StringComparison.OrdinalIgnoreCase)))
+                continue;
+            manifest.Append(hash).Append("  ").Append(relative, NativeSourcePrefix.Length,
+                relative.Length - NativeSourcePrefix.Length).Append('\n');
+        }
+        if (!digest.Equals(Sha256Hex(manifest.ToString()), StringComparison.OrdinalIgnoreCase))
+            return null;
+
+        document.Remove("sourceDigest");
+        return Sha256Hex(document.ToJsonString());
+    }
+
+    private static string Sha256Hex(string text) =>
+        Convert.ToHexString(SHA256.HashData(Encoding.UTF8.GetBytes(text))).ToLowerInvariant();
+
     /// <summary>
     /// Content identity of the source-owned files copied verbatim beside every product. This is
     /// intentionally independent of <see cref="Compute"/>: an asset-only release is published
@@ -273,8 +335,7 @@ internal static class ToolkitFingerprint
         var builder = new StringBuilder(version).Append('\n');
         foreach (var (relative, hash) in entries)
             builder.Append(relative).Append('|').Append(hash).Append('\n');
-        return Convert.ToHexString(SHA256.HashData(Encoding.UTF8.GetBytes(builder.ToString())))
-            .ToLowerInvariant();
+        return Sha256Hex(builder.ToString());
     }
 }
 

Launcher/WiiCompiled.Setup.Windows/WiiCompiled.Setup.Windows.csproj

@@ -15,5 +15,6 @@
   </PropertyGroup>
   <ItemGroup>
     <ProjectReference Include="..\WiiCompiled.Setup.Common\WiiCompiled.Setup.Common.csproj" />
+    <Using Include="WiiCompiled.Setup.Common" />
   </ItemGroup>
 </Project>

Launcher/build-appimage.sh

@@ -127,10 +127,10 @@ mkdir -p "$appdir/usr/toolchain"
 cp -a "$workspace/Launcher/artifacts/portable-tools/toolchain-$appimagetool_arch"/. "$appdir/usr/toolchain/"
 
 # Precompiled aurora + third-party package (see Prepare-NativePrebuilt.sh) so a user's own
-# local-build.sh never has to compile aurora itself (~43% of local build CPU time). Re-harvesting
-# recompiles the aurora/Crypto++/mbed TLS closure with the toolchain above, so this is skipped unless
-# --print-fingerprint-only (a fast, build-free check) says the existing package no longer matches
-# the current compiler, flags, source trees or mbed TLS pin.
+# local-build.sh never has to compile aurora itself (~43% of local build CPU time), nor on x86_64
+# the game-independent runtime. Re-harvesting recompiles that closure with the toolchain above, so
+# this is skipped unless --print-fingerprint-only (a fast, build-free check) says the existing
+# package no longer matches the current compiler, flags, source trees or mbed TLS pin.
 native_prebuilt_dir="$workspace/Launcher/artifacts/native-prebuilt-$appimagetool_arch"
 echo "Checking whether the precompiled aurora + third-party package ($appimagetool_arch) is current..."
 current_fingerprint=$(bash "$script_dir/Prepare-NativePrebuilt.sh" --arch "$appimagetool_arch" --print-fingerprint-only)
@@ -149,6 +149,8 @@ fields = {
     "aurora_fingerprint": "AuroraSourceFingerprint",
     "third_party_fingerprint": "ThirdPartySourceFingerprint",
     "mbedtls_fingerprint": "MbedTlsFingerprint",
+    # Only printed on x86_64, so a package without the runtime archive is stale there.
+    "runtime_fingerprint": "RuntimeSourceFingerprint",
 }
 print(1 if all(provenance.get(v) == current.get(k) for k, v in fields.items()) else 0)
 PY
@@ -174,7 +176,29 @@ done
 # is a plain developer build directory.
 find "$appdir/workspace/aurora-main/extern" -mindepth 1 -maxdepth 1 -type d -exec rm -rf {} +
 rm -rf "$appdir/workspace/runtime/build"
-cp "$workspace/Launcher/local-build.sh" "$appdir/workspace/Launcher/local-build.sh"
+cp "$workspace/Launcher/local-build.sh" "$workspace/Launcher/build-cache.sh" "$appdir/workspace/Launcher/"
+
+# The translation identity (build-cache.sh) covers the runtime's native registrations through this
+# index, so it is extracted from exactly the bundled runtime sources.
+echo "Extracting the runtime native index..."
+"$appdir/usr/bin/translator-cli" emit-native-index \
+    --project "$appdir/workspace/projects/mkwii/recomp.yml" \
+    --out "$appdir/workspace/runtime/native-index.json"
+[[ -f "$appdir/workspace/runtime/native-index.json" ]] || { echo "build-appimage.sh: the native index was not written" >&2; exit 1; }
+
+# A bundled runtime that fingerprints differently from the shipped archive would make every
+# installation silently compile the runtime from source.
+if [[ "$appimagetool_arch" == x86_64 ]]; then
+    staged_runtime_fingerprint=$("$workspace/Launcher/artifacts/portable-tools/toolchain-$appimagetool_arch/bin/cmake" \
+        -DMKW_RUNTIME_DIR="$appdir/workspace/runtime" -P "$appdir/workspace/runtime/cmake/RuntimeFingerprint.cmake" |
+        sed -n 's/^runtime_fingerprint=//p')
+    packaged_runtime_fingerprint=$(python3 -c 'import json, sys; print(json.load(open(sys.argv[1], encoding="utf-8")).get("RuntimeSourceFingerprint", ""))' \
+        "$appdir/native-prebuilt/provenance.json")
+    if [[ -z "$staged_runtime_fingerprint" || "$staged_runtime_fingerprint" != "$packaged_runtime_fingerprint" ]]; then
+        echo "build-appimage.sh: the bundled runtime ($staged_runtime_fingerprint) does not match the precompiled runtime archive ($packaged_runtime_fingerprint)" >&2
+        exit 1
+    fi
+fi
 
 # AppRun re-syncs runtime/aurora-main/projects/local-build.sh into the writable cache only when
 # this changes, so it must change whenever any of those bundled paths actually did - a bare commit
@@ -204,12 +228,9 @@ CACHE="${XDG_DATA_HOME:-$HOME/.local/share}/WiiCompiled/workspace"
 mkdir -p "$CACHE"
 if [ ! -f "$CACHE/.bundle-version" ] || \
    [ "$(cat "$HERE/workspace/.bundle-version")" != "$(cat "$CACHE/.bundle-version")" ]; then
-    mkdir -p "$CACHE/Launcher"
-    for dir in runtime aurora-main projects; do
-        rm -rf "$CACHE/$dir"
-        cp -r "$HERE/workspace/$dir" "$CACHE/$dir"
-    done
-    cp "$HERE/workspace/Launcher/local-build.sh" "$CACHE/Launcher/local-build.sh"
+    # Unchanged files keep their timestamps so the incremental build survives the update.
+    source "$HERE/workspace/Launcher/build-cache.sh"
+    mkw_sync_tree "$HERE/workspace" "$CACHE" runtime aurora-main projects Launcher >&2
     cp "$HERE/workspace/.bundle-version" "$CACHE/.bundle-version"
 fi
 # toolchain/ and native-prebuilt/ are NOT copied into the cache (they're large - ~500 MiB /

Launcher/build-cache.sh

@@ -0,0 +1,148 @@
+# Shared incremental-build logic for the Unix setups (Linux AppImage and macOS), sourced by
+# local-build.sh, local-build-macos.command, the AppImage's AppRun and macos/setup.command.
+# Mirrors the Windows installer's translation/toolchain identities and timestamp preservation.
+# Must stay compatible with macOS's bash 3.2.
+
+if command -v sha256sum >/dev/null 2>&1; then
+    mkw_sha256_cmd=(sha256sum)
+else
+    mkw_sha256_cmd=(shasum -a 256)
+fi
+
+mkw_sha256() {
+    "${mkw_sha256_cmd[@]}" "$@" | awk '{print $1}'
+}
+
+# Prints "<sha>  <relative path>" for every file under each directory, relative to $1.
+mkw_digest_tree() {
+    local root=$1; shift
+    (
+        cd "$root" || exit 1
+        local dir
+        for dir in "$@"; do
+            if [[ -d "$dir" ]]; then
+                printf 'tree %s\n' "$dir"
+                find "$dir" -type f -print0 | LC_ALL=C sort -z | xargs -0 "${mkw_sha256_cmd[@]}" /dev/null
+            else
+                printf 'absent %s\n' "$dir"
+            fi
+        done
+    )
+}
+
+# Content-only digest of a file or directory, independent of where it lives (the AppImage mount
+# path changes every launch).
+mkw_digest_path() {
+    if [[ -d "$1" ]]; then
+        mkw_digest_tree "$1" . | mkw_sha256
+    elif [[ -f "$1" ]]; then
+        mkw_sha256 < "$1"
+    else
+        printf 'absent\n'
+    fi
+}
+
+# Same digest as the translator's RuntimeNativeIndexFile.ComputeSourceDigest. $1 = runtime/src.
+mkw_native_source_digest() {
+    (
+        cd "$1" || exit 1
+        find . -type f \( -iname '*.cpp' -o -iname '*.h' -o -iname '*.hpp' \) -print0 | LC_ALL=C sort -z |
+            xargs -0 "${mkw_sha256_cmd[@]}" /dev/null
+    ) | grep -v '  /dev/null$' | sed 's|  \./|  |' | mkw_sha256
+}
+
+# The native registrations the translator reads from runtime/src. A current native-index.json
+# stands in for the sources (minus its source digest), so runtime edits that leave the
+# registrations alone keep the translation; a missing or stale index falls back to the sources.
+# $1 = workspace.
+mkw_native_index_identity() {
+    local index="$1/runtime/native-index.json" recorded
+    recorded=$(mkw_read_identity "$index" sourceDigest)
+    if [[ -n "$recorded" && -d "$1/runtime/src" && "$recorded" == "$(mkw_native_source_digest "$1/runtime/src")" ]]; then
+        printf 'native-index %s\n' "$(grep -v '^[[:space:]]*"sourceDigest"[[:space:]]*:' "$index" | mkw_sha256)"
+    else
+        mkw_digest_tree "$1" runtime/src
+    fi
+}
+
+# Everything the base translation depends on: game inputs, project, native registrations, the
+# translator itself and the build script that drives it.
+# $1 = workspace, $2 = translator file or directory, $3 = build script.
+mkw_translation_identity() {
+    local workspace=$1 translator=$2 script=$3
+    {
+        printf 'mkw-translation-v2\n'
+        printf 'dol %s\n' "$(mkw_digest_path "$workspace/Assets/main.dol")"
+        printf 'rel %s\n' "$(mkw_digest_path "$workspace/Assets/StaticR.rel")"
+        printf 'translator %s\n' "$(mkw_digest_path "$translator")"
+        printf 'script %s\n' "$(mkw_digest_path "$script")"
+        mkw_digest_tree "$workspace" projects
+        mkw_native_index_identity "$workspace"
+    } | mkw_sha256
+}
+
+# Inputs that make compiled objects unsafe to reuse: compiler/CMake/Ninja versions and the
+# from-source dependency pins. Source changes are excluded; Ninja rebuilds exactly what they touch.
+# $1 = workspace, remaining = tools (C compiler, C++ compiler, cmake, ninja).
+mkw_toolchain_identity() {
+    local workspace=$1; shift
+    local tool
+    {
+        printf 'mkw-toolchain-v2\n'
+        for tool in "$@"; do
+            # First line only: clang appends its InstalledDir, which moves with the AppImage mount.
+            printf '%s\n' "$("$tool" --version 2>/dev/null | head -n 1)"
+        done
+        mkw_digest_tree "$workspace" aurora-main/extern aurora-main/cmake
+        printf 'aurora %s\n' "$(mkw_digest_path "$workspace/aurora-main/CMakeLists.txt")"
+        printf 'mbedtls %s\n' "$(mkw_digest_path "$workspace/runtime/cmake/MbedTLSPin.cmake")"
+    } | mkw_sha256
+}
+
+# $1 = JSON file, $2 = key. Prints the recorded string value, or nothing.
+mkw_read_identity() {
+    [[ -f "$1" ]] || return 0
+    grep -o "\"$2\" *: *\"[^\"]*\"" "$1" 2>/dev/null | sed 's/.*"\([^"]*\)"$/\1/' || true
+}
+
+mkw_write_identity() {
+    printf '{"SchemaVersion":1,"%s":"%s"}\n' "$2" "$3" > "$1"
+}
+
+# Mirrors each named directory from $1 into $2. Unchanged files are left untouched so their
+# timestamps (which clang's PCH validation and Ninja compare) survive; changed and new files are
+# copied fresh and files no longer shipped are removed.
+mkw_sync_tree() {
+    local source=$1 destination=$2; shift 2
+    local dir relative changed=0 unchanged=0 removed=0
+    for dir in "$@"; do
+        if [[ ! -d "$source/$dir" ]]; then
+            rm -rf "$destination/$dir"
+            continue
+        fi
+        if [[ -e "$destination/$dir" && ! -d "$destination/$dir" ]]; then
+            rm -f "$destination/$dir"
+        fi
+        mkdir -p "$destination/$dir"
+        while IFS= read -r -d '' relative; do
+            relative=${relative#./}
+            if [[ -d "$source/$dir/$relative" ]]; then
+                [[ -d "$destination/$dir/$relative" ]] || { rm -f "$destination/$dir/$relative"; mkdir -p "$destination/$dir/$relative"; }
+            elif [[ -f "$destination/$dir/$relative" ]] && cmp -s "$source/$dir/$relative" "$destination/$dir/$relative"; then
+                unchanged=$((unchanged + 1))
+            else
+                rm -rf "$destination/$dir/$relative"
+                cp "$source/$dir/$relative" "$destination/$dir/$relative"
+                changed=$((changed + 1))
+            fi
+        done < <(cd "$source/$dir" && find . -mindepth 1 -print0)
+        while IFS= read -r -d '' relative; do
+            relative=${relative#./}
+            if [[ ! -e "$source/$dir/$relative" && ( -e "$destination/$dir/$relative" || -L "$destination/$dir/$relative" ) ]]; then
+                [[ -d "$destination/$dir/$relative" ]] || removed=$((removed + 1))
+                rm -rf "$destination/$dir/$relative"
+            fi
+        done < <(cd "$destination/$dir" && find . -mindepth 1 -depth -print0)
+    done
+    printf 'Workspace refreshed: %d changed, %d unchanged, %d removed file(s).\n' "$changed" "$unchanged" "$removed"
+}

Launcher/local-build-macos.command

@@ -4,11 +4,12 @@ set -euo pipefail
 
 script_dir=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 default_workspace=$(cd "$script_dir/.." && pwd)
+# shellcheck source=build-cache.sh
+source "$script_dir/build-cache.sh"
 
 fail() { printf 'local-build-macos.command: error: %s\n' "$*" >&2; exit 1; }
 step() { printf 'MKWCBUILD:STEP:%s %s\n' "$1" "$2"; }
 assert_file() { [[ -f "$1" ]] || fail "$2 is missing: $1"; }
-sha256() { shasum -a 256 "$1" | awk '{ print $1 }'; }
 
 usage() {
     cat <<'EOF'
@@ -27,12 +28,15 @@ Usage: local-build-macos.command --output-dir DIR [options]
   --cmake PATH --ninja PATH       Override build tools
   --dotnet PATH                   Override dotnet
   --translator-bin PATH           Use a self-contained Translator.Cli executable
+  --setup-version VERSION         Version recorded in the runtime log (default: <Version> in
+                                  Launcher/Directory.Build.props when present)
 EOF
 }
 
 workspace="$default_workspace"; profile=base; output_dir=""; base_output_dir=""
 game=""; nodtool=""; retro_root=""; retro_wfc=""; skip_retro_wfc=0; force_clean=0
 parallel=0; cmake_bin=cmake; ninja_bin=ninja; dotnet_bin=dotnet; translator_bin=""
+setup_version=""
 while (($#)); do
     case "$1" in
         --workspace) workspace=${2:-}; shift 2 ;;
@@ -50,6 +54,7 @@ while (($#)); do
         --ninja) ninja_bin=${2:-}; shift 2 ;;
         --dotnet) dotnet_bin=${2:-}; shift 2 ;;
         --translator-bin) translator_bin=${2:-}; shift 2 ;;
+        --setup-version) setup_version=${2:-}; shift 2 ;;
         -h|--help) usage; exit 0 ;;
         *) fail "unknown option: $1" ;;
     esac
@@ -60,6 +65,10 @@ macos_arch=$(uname -m)
 case "$macos_arch" in arm64|x86_64) ;; *) fail "unsupported macOS architecture: $macos_arch" ;; esac
 macos_deployment_target=12.0
 workspace=$(cd "$workspace" && pwd)
+if [[ -z "$setup_version" && -f "$workspace/Launcher/Directory.Build.props" ]]; then
+    setup_version=$(sed -nE 's/.*<Version>([^<]+)<\/Version>.*/\1/p' "$workspace/Launcher/Directory.Build.props" | head -n 1)
+fi
+[[ "$setup_version" =~ ^[0-9A-Za-z.+-]*$ ]] || fail '--setup-version contains unsupported characters'
 [[ -n "$output_dir" ]] || fail '--output-dir is required'
 case "$profile" in base|retro-rewind|both) ;; *) fail '--profile must be base, retro-rewind, or both' ;; esac
 builds_retro=0; [[ "$profile" != base ]] && builds_retro=1
@@ -118,11 +127,30 @@ if (( builds_retro )); then
     fi
 fi
 
-step translate-base 'Translating the user-owned base game'
-rm -rf "$functions" "$metadata" "$manifest_dir"; mkdir -p "$functions" "$manifest_dir"
-translator translate-recursive "$entry_point" --project "$project" --outdir "$functions" --output-metadata "$metadata" --production-source-bundle "$generated/base_translation_sources.bin" --no-function-files --prune-stale --threads "$translator_threads"
-step emit-base-manifest 'Creating the local base translation manifest'
-translator emit-base-manifest --project "$project" --out "$manifest_dir" --functions-dir "$functions" --translation-output-metadata "$metadata" --region P
+translation_provenance="$generated/translation-provenance.json"
+translation_fingerprint=$(mkw_translation_identity "$workspace" "${translator_bin:-$(dirname "${translator_dll:-}")}" "$script_dir/local-build-macos.command")
+reuse_base=0
+if [[ "$(mkw_read_identity "$translation_provenance" TranslationFingerprint)" == "$translation_fingerprint" && -f "$metadata" && -f "$manifest" ]]; then
+    reuse_base=1
+fi
+if (( reuse_base && builds_retro )); then
+    # A base tree that never saw this Code.pul would bake vanilla code into the modded product.
+    pul_sha=$(mkw_sha256 "$retro_root/Binaries/Code.pul")
+    if ! grep -q "\"codePulSha256\":\"$pul_sha\"" "$metadata" && \
+       ! translator check-base-mod-awareness --project "$project" --profile retro-rewind --translation-output-metadata "$metadata" --code-pul "$retro_root/Binaries/Code.pul"; then
+        reuse_base=0
+    fi
+fi
+if (( reuse_base )); then
+    step reuse-base-translation 'Reusing the completed base translation'
+else
+    rm -f "$translation_provenance"; mkdir -p "$functions" "$manifest_dir"
+    step translate-base 'Translating the user-owned base game'
+    translator translate-recursive "$entry_point" --project "$project" --outdir "$functions" --output-metadata "$metadata" --production-source-bundle "$generated/base_translation_sources.bin" --no-function-files --prune-stale --threads "$translator_threads"
+    step emit-base-manifest 'Creating the local base translation manifest'
+    translator emit-base-manifest --project "$project" --out "$manifest_dir" --functions-dir "$functions" --translation-output-metadata "$metadata" --region P
+    mkw_write_identity "$translation_provenance" TranslationFingerprint "$translation_fingerprint"
+fi
 if (( builds_retro )); then
     mod_out="$workspace/build/mods/retro_rewind_full_cpp"; args=(translate-mod --project "$project" --profile retro-rewind --base-manifest "$manifest" --base-translation-output-metadata "$metadata" --code-pul "$retro_root/Binaries/Code.pul" --mod-root "$retro_root" --mod-name 'Retro Rewind' --region P --out "$mod_out" --prefer-cached-inputs --emit-cpp --threads "$translator_threads")
     if (( skip_retro_wfc )); then args+=(--skip-retro-wfc); else offline_payload="$retro_wfc/binary/payload.RMCPD00.bin"; assert_file "$offline_payload" 'Offline Retro-WFC payload'; args+=(--retro-wfc-payload "$offline_payload"); fi
@@ -133,10 +161,17 @@ args=(emit-build-shards --project "$project" --base-metadata "$metadata" --base-
 if (( builds_retro )); then args+=(--resolved-profile "$mod_out/resolved_dispatch_profile.json" --retro-cpp-dir "$mod_out/cpp"); fi
 step emit-build-shards 'Preparing native build shards'; translator "${args[@]}"
 
+toolchain_provenance="$native_build/toolchain-provenance.json"
+toolchain_fingerprint=$(mkw_toolchain_identity "$workspace" clang clang++ "$cmake_bin" "$ninja_bin")
+if [[ -d "$native_build" && "$(mkw_read_identity "$toolchain_provenance" ToolchainFingerprint)" != "$toolchain_fingerprint" ]]; then
+    printf 'MKWCBUILD: The toolchain or a pinned dependency changed; rebuilding the native build from scratch\n'
+    rm -rf "$native_build"
+fi
 step configure-native 'Configuring the native toolchain'
 # Use Aurora's pinned SDL3 source on macOS.  A system SDL3 can be older than
 # Aurora's required API even when find_package() succeeds.
-"$cmake_bin" -S "$workspace/runtime" -B "$native_build" -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_MAKE_PROGRAM="$ninja_bin" -DCMAKE_OSX_ARCHITECTURES="$macos_arch" -DCMAKE_OSX_DEPLOYMENT_TARGET="$macos_deployment_target" -DMKW_TRANSLATED_COMPILE_JOBS="$translated_jobs" -DAURORA_SDL3_PROVIDER=vendor
+"$cmake_bin" -S "$workspace/runtime" -B "$native_build" -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_MAKE_PROGRAM="$ninja_bin" -DCMAKE_OSX_ARCHITECTURES="$macos_arch" -DCMAKE_OSX_DEPLOYMENT_TARGET="$macos_deployment_target" -DMKW_TRANSLATED_COMPILE_JOBS="$translated_jobs" -DAURORA_SDL3_PROVIDER=vendor -DMKW_SETUP_VERSION="$setup_version"
+mkw_write_identity "$toolchain_provenance" ToolchainFingerprint "$toolchain_fingerprint"
 targets=(); [[ "$profile" != retro-rewind ]] && targets+=(WiiCompiled); [[ "$profile" != base ]] && targets+=(RetroRewind)
 step compile "Compiling ${targets[*]} locally"; "$cmake_bin" --build "$native_build" --target "${targets[@]}" --parallel "$global_jobs"
 if [[ "$profile" != retro-rewind ]]; then "$script_dir/macos/publish-app.command" --build-dir "$native_build" --product WiiCompiled --output-dir "${base_output_dir:-$output_dir}" --architecture "$macos_arch" --minimum-system-version "$macos_deployment_target"; fi

Launcher/local-build.sh

@@ -39,7 +39,7 @@ require_command() {
 }
 
 sha256_of() {
-    sha256sum "$1" | awk '{print $1}'
+    mkw_sha256 "$1"
 }
 
 # ---------------------------------------------------------------------------
@@ -47,6 +47,8 @@ sha256_of() {
 # ---------------------------------------------------------------------------
 
 script_dir=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+# shellcheck source=build-cache.sh
+source "$script_dir/build-cache.sh"
 workspace=$(cd "$script_dir/.." && pwd)
 profile=base
 output_dir=""
@@ -66,6 +68,7 @@ translator_bin_override=""
 fuse_ld_override=""
 native_prebuilt_dir=""
 sysroot=""
+setup_version=""
 
 usage() {
     cat <<'EOF'
@@ -90,6 +93,8 @@ Usage: local-build.sh --output-dir DIR [options]
                                    skips compiling aurora-main from source entirely
   --sysroot PATH                   Passed to CMake as -DCMAKE_SYSROOT: where the compiler resolves
                                    standard headers/startup files
+  --setup-version VERSION          Version recorded in the runtime log (default: <Version> in
+                                   Launcher/Directory.Build.props when present)
 EOF
 }
 
@@ -114,11 +119,17 @@ while [[ $# -gt 0 ]]; do
         --translator-bin) translator_bin_override=$2; shift 2 ;;
         --native-prebuilt-dir) native_prebuilt_dir=$2; shift 2 ;;
         --sysroot) sysroot=$2; shift 2 ;;
+        --setup-version) setup_version=$2; shift 2 ;;
         -h|--help) usage; exit 0 ;;
         *) fail "unknown argument: $1" ;;
     esac
 done
 
+if [[ -z "$setup_version" && -f "$workspace/Launcher/Directory.Build.props" ]]; then
+    setup_version=$(sed -nE 's/.*<Version>([^<]+)<\/Version>.*/\1/p' "$workspace/Launcher/Directory.Build.props" | head -n 1)
+fi
+[[ "$setup_version" =~ ^[0-9A-Za-z.+-]*$ ]] || fail "--setup-version contains unsupported characters"
+
 [[ -n "$output_dir" ]] || { usage; fail "--output-dir is required"; }
 case "$profile" in
     base|retro-rewind|both) ;;
@@ -266,10 +277,9 @@ else
 fi
 
 # ---------------------------------------------------------------------------
-# Translation cache: this script is the only owner of the reuse decision (unlike LocalBuild.ps1,
-# which is handed caller-computed fingerprints by the Windows installer - there is no Linux
-# installer yet to supply anything). Hash the game inputs the translation actually depends on;
-# a match plus every expected output file present means the previous translation is still good.
+# Translation cache: this script is the only owner of the reuse decision. The identity (see
+# build-cache.sh) covers the game inputs, project, native registrations, translator and this script,
+# matching the Windows installer's translation identity.
 # ---------------------------------------------------------------------------
 
 if (( force_clean_build )); then
@@ -277,13 +287,12 @@ if (( force_clean_build )); then
     rm -rf "$generated" "$base_manifest_dir" "$build"
 fi
 
-translation_fingerprint=$(cat "$assets/main.dol" "$assets/StaticR.rel" "$project" | sha256sum | awk '{print $1}')
+translator_identity_input=${translator_bin:-$(dirname "$translator_dll")}
+translation_fingerprint=$(mkw_translation_identity "$workspace" "$translator_identity_input" "$script_dir/local-build.sh")
 reuse_base=0
-if [[ -f "$translation_provenance" ]]; then
-    recorded=$(grep -o '"TranslationFingerprint" *: *"[^"]*"' "$translation_provenance" 2>/dev/null | sed 's/.*"\([0-9a-f]*\)"$/\1/' || true)
-    if [[ "$recorded" == "$translation_fingerprint" && -f "$base_metadata" && -f "$base_manifest" ]]; then
-        reuse_base=1
-    fi
+if [[ "$(mkw_read_identity "$translation_provenance" TranslationFingerprint)" == "$translation_fingerprint" \
+      && -f "$base_metadata" && -f "$base_manifest" ]]; then
+    reuse_base=1
 fi
 
 if (( builds_retro )); then
@@ -331,8 +340,7 @@ else
     translator emit-base-manifest --project "$project" --out "$base_manifest_dir" \
         --functions-dir "$functions" --translation-output-metadata "$base_metadata" --region P
 
-    printf '{"SchemaVersion":1,"TranslationFingerprint":"%s"}' "$translation_fingerprint" \
-        > "$translation_provenance"
+    mkw_write_identity "$translation_provenance" TranslationFingerprint "$translation_fingerprint"
 fi
 
 if (( builds_retro )); then
@@ -407,6 +415,13 @@ if [[ -f "$build/CMakeCache.txt" ]]; then
         done < <(grep -o ':FILEPATH=.*' "$build/CMakeCache.txt" | sed 's/^:FILEPATH=//')
     fi
 fi
+# The AppImage keeps tool paths stable across updates, so a changed compiler is only visible here.
+toolchain_fingerprint=$(mkw_toolchain_identity "$workspace" "$cc_bin" "$cxx_bin" "$cmake_bin" "$ninja_bin")
+if [[ "$keep_native_build" -eq 1 && \
+      "$(mkw_read_identity "$toolchain_provenance" ToolchainFingerprint)" != "$toolchain_fingerprint" ]]; then
+    keep_native_build=0
+    stale_reason=" (the toolchain or a pinned dependency changed)"
+fi
 if [[ -d "$build" && "$keep_native_build" -eq 0 ]]; then
     echo "MKWCBUILD: The native build cache does not belong to this workspace path or toolchain; rebuilding from scratch$stale_reason"
     rm -rf "$build"
@@ -418,7 +433,8 @@ configure_args=(-S "$workspace/runtime" -B "$build" -G Ninja
     -DCMAKE_BUILD_TYPE=Release
     -DCMAKE_C_COMPILER="$cc_bin" -DCMAKE_CXX_COMPILER="$cxx_bin"
     -DCMAKE_MAKE_PROGRAM="$ninja_bin"
-    -DMKW_TRANSLATED_COMPILE_JOBS="$translated_jobs")
+    -DMKW_TRANSLATED_COMPILE_JOBS="$translated_jobs"
+    -DMKW_SETUP_VERSION="$setup_version")
 # CMAKE_C_COMPILER/CXX_COMPILER stay stable across AppImage runs on their own (they're exactly
 # what's passed via -D above, and AppRun points --cc/--cxx at a symlink it re-targets at the
 # current mount every launch - see build-appimage.sh). CMAKE_AR/RANLIB/LINKER/ASM_COMPILER do NOT
@@ -437,7 +453,9 @@ if [[ "$cc_bin" == */* ]]; then
     configure_args+=(-DCMAKE_ASM_COMPILER="$cc_bin")
 fi
 if [[ -n "$fuse_ld_override" ]]; then
-    configure_args+=(-DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=$fuse_ld_override")
+    # CMAKE_EXE_LINKER_FLAGS is one string that CMake pastes into the shell link command.
+    printf -v fuse_ld_quoted '%q' "$fuse_ld_override"
+    configure_args+=(-DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=$fuse_ld_quoted")
 fi
 if [[ -n "$native_prebuilt_dir" ]]; then
     configure_args+=(-DMKW_NATIVE_PREBUILT_DIR="$native_prebuilt_dir")
@@ -453,6 +471,7 @@ fi
 
 log_step configure-native "Configuring the native toolchain"
 "$cmake_bin" "${configure_args[@]}"
+mkw_write_identity "$toolchain_provenance" ToolchainFingerprint "$toolchain_fingerprint"
 
 case "$profile" in
     base) targets=(WiiCompiled) ;;

Launcher/macos/build-setup-pkg.command

@@ -20,13 +20,14 @@ resulting pkg is unsigned unless
 Developer ID Installer certificate.
 
   --workspace DIR             Repository root (default: script's grandparent)
-  --version VERSION           Bundle/package version (default: 0.1.0)
+  --version VERSION           Bundle/package version (default: project version, or 0.1.0)
   --installer-identity NAME   Developer ID Installer identity for productbuild
 EOF
 }
 
 script_dir=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 workspace=$(cd "$script_dir/../.." && pwd); nodtool_arm64=""; nodtool_x86_64=""; translator_arm64=""; translator_x86_64=""; cmake_root=""; ninja_arm64=""; ninja_x86_64=""; output=""; version=0.1.0; identity=""
+version_supplied=false
 while (($#)); do
     case "$1" in
         --workspace) workspace=${2:-}; shift 2 ;;
@@ -38,14 +39,14 @@ while (($#)); do
         --ninja-arm64) ninja_arm64=${2:-}; shift 2 ;;
         --ninja-x86_64) ninja_x86_64=${2:-}; shift 2 ;;
         --output) output=${2:-}; shift 2 ;;
-        --version) version=${2:-}; shift 2 ;;
+        --version) version=${2:-}; version_supplied=true; shift 2 ;;
         --installer-identity) identity=${2:-}; shift 2 ;;
         -h|--help) usage; exit 0 ;;
         *) fail "unknown option: $1" ;;
     esac
 done
 version=${version#v}
-if [[ -z "$version" || "$version" == "0.1.0" ]]; then
+if [[ "$version_supplied" == false ]]; then
     local_csproj="$workspace/Launcher/Directory.Build.props"
     if [[ -f "$local_csproj" ]]; then
         detected=$(grep -m1 '<Version>' "$local_csproj" | sed -E 's/.*<Version>([^<]+)<\/Version>.*/\1/' || true)
@@ -149,8 +150,16 @@ for source in aurora-main projects runtime translator; do
     [[ -d "$workspace/$source" ]] || fail "required workspace directory is missing: $source"
     copy_clean "$workspace/$source" "$resources/workspace/$source"
 done
+# The translation identity (build-cache.sh) covers the runtime's native registrations through this
+# index, so it is extracted from exactly the packaged runtime sources.
+if [[ "$host_arch" == arm64 ]]; then host_translator=$translator_arm64; else host_translator=$translator_x86_64; fi
+"$host_translator" emit-native-index --project "$resources/workspace/projects/mkwii/recomp.yml" \
+    --out "$resources/workspace/runtime/native-index.json" || fail 'the runtime native index could not be extracted'
+[[ -f "$resources/workspace/runtime/native-index.json" ]] || fail 'the runtime native index was not written'
 mkdir -p "$resources/workspace/Launcher/macos"
+copy_clean "$workspace/Launcher/Directory.Build.props" "$resources/workspace/Launcher/Directory.Build.props"
 copy_clean "$workspace/Launcher/local-build-macos.command" "$resources/workspace/Launcher/local-build-macos.command"
+copy_clean "$workspace/Launcher/build-cache.sh" "$resources/workspace/Launcher/build-cache.sh"
 copy_clean "$workspace/Launcher/macos/extract-disc.command" "$resources/workspace/Launcher/macos/extract-disc.command"
 copy_clean "$workspace/Launcher/macos/publish-app.command" "$resources/workspace/Launcher/macos/publish-app.command"
 chmod +x "$resources/workspace/Launcher/local-build-macos.command" "$resources/workspace/Launcher/macos/"*.command

Launcher/macos/extract-disc.command

@@ -38,6 +38,16 @@ done
 [[ -x "$nodtool" ]] || fail "nodtool is not executable: $nodtool"
 
 assets_dir=$(mkdir -p "$assets_dir" && cd "$assets_dir" && pwd)
+
+# Sorted "<size> <path>" of every extracted file; recorded at extraction, compared on reuse.
+data_manifest() { (cd "$1" && find . -type f -print0 | LC_ALL=C sort -z | xargs -0 stat -f '%z %N'); }
+
+if [[ -d "$assets_dir/DATA/files" && -f "$assets_dir/.data-manifest" && -f "$assets_dir/main.dol" && -f "$assets_dir/StaticR.rel" ]] && \
+   [[ $(sha256 "$assets_dir/main.dol") == "$EXPECTED_DOL_SHA256" && $(sha256 "$assets_dir/StaticR.rel") == "$EXPECTED_REL_SHA256" ]] && \
+   [[ "$(data_manifest "$assets_dir/DATA")" == "$(cat "$assets_dir/.data-manifest")" ]]; then
+    printf 'MKWCBUILD:STEP:disc-ready Reusing the verified extracted game assets\n'
+    exit 0
+fi
 scratch=$(mktemp -d "${TMPDIR:-/tmp}/wiicompiled-disc.XXXXXX")
 cleanup() { rm -rf "$scratch"; }
 trap cleanup EXIT
@@ -51,8 +61,10 @@ dol=$(find "$scratch/extracted" -type f -path '*/sys/main.dol' -print -quit)
 rel=$(find "$scratch/extracted" -type f -path '*/files/rel/StaticR.rel' -print -quit)
 [[ -n "$dol" ]] || fail 'nodtool extraction did not contain sys/main.dol'
 [[ -n "$rel" ]] || fail 'nodtool extraction did not contain files/rel/StaticR.rel'
-[[ $(sha256 "$dol") == "$EXPECTED_DOL_SHA256" ]] || fail 'disc is not the supported clean PAL RMCP01 Mario Kart Wii image'
-[[ $(sha256 "$rel") == "$EXPECTED_REL_SHA256" ]] || fail 'disc has an unexpected StaticR.rel; use a clean PAL RMCP01 image'
+dol_sha=$(sha256 "$dol")
+rel_sha=$(sha256 "$rel")
+[[ $dol_sha == "$EXPECTED_DOL_SHA256" ]] || fail "main.dol does not match the clean PAL RMCP01 image (expected $EXPECTED_DOL_SHA256, got $dol_sha)"
+[[ $rel_sha == "$EXPECTED_REL_SHA256" ]] || fail "StaticR.rel does not match the clean PAL RMCP01 image (expected $EXPECTED_REL_SHA256, got $rel_sha)"
 
 data_root=$(dirname "$(dirname "$dol")")
 [[ -d "$data_root/files" ]] || fail 'nodtool extraction did not contain the Wii files directory'
@@ -64,15 +76,17 @@ mkdir -p "$stage"
 ditto "$data_root" "$stage/DATA"
 ditto "$dol" "$stage/main.dol"
 ditto "$rel" "$stage/StaticR.rel"
+data_manifest "$stage/DATA" > "$stage/.data-manifest"
 
 backup="$assets_dir/.previous-extraction-$(date +%Y%m%d-%H%M%S)"
-if [[ -e "$assets_dir/DATA" || -e "$assets_dir/main.dol" || -e "$assets_dir/StaticR.rel" ]]; then
+if [[ -e "$assets_dir/DATA" || -e "$assets_dir/.data-manifest" || -e "$assets_dir/main.dol" || -e "$assets_dir/StaticR.rel" ]]; then
     mkdir -p "$backup"
-    for item in DATA main.dol StaticR.rel; do
+    for item in DATA .data-manifest main.dol StaticR.rel; do
         [[ -e "$assets_dir/$item" ]] && mv "$assets_dir/$item" "$backup/$item"
     done
 fi
 mv "$stage/DATA" "$assets_dir/DATA"
+mv "$stage/.data-manifest" "$assets_dir/.data-manifest"
 mv "$stage/main.dol" "$assets_dir/main.dol"
 mv "$stage/StaticR.rel" "$assets_dir/StaticR.rel"
 rmdir "$stage"

Launcher/macos/setup.command

@@ -11,7 +11,13 @@ workspace="$support_root/BuildWorkspace"
 products="$support_root/Products"
 
 fail() { printf 'WiiCompiled Setup: %s\n' "$*" >&2; exit 1; }
-notice() { /usr/bin/osascript -e "display dialog \"${1//\"/\\\"}\" buttons {\"OK\"} default button \"OK\" with icon caution" >/dev/null; }
+notice() {
+    /usr/bin/osascript - "$1" >/dev/null <<'APPLESCRIPT'
+on run argv
+    display dialog (item 1 of argv) buttons {"OK"} default button "OK" with icon caution
+end run
+APPLESCRIPT
+}
 usage() {
     cat <<'EOF'
 Usage: setup.command --game IMAGE [--retro-dir DIR] [--install-location {user|applications}]
@@ -85,15 +91,12 @@ if (( needs_workspace_refresh )); then
         /usr/bin/ditto "$workspace_source" "$workspace"
     else
         # Refresh only packaged source inputs. Assets and the staged Retro
-        # Rewind package belong to the user and stay in place.
-        for source in aurora-main projects runtime translator Launcher; do
-            rm -rf "$workspace/$source"
-            /usr/bin/ditto "$workspace_source/$source" "$workspace/$source"
-        done
+        # Rewind package belong to the user and stay in place; unchanged files
+        # keep their timestamps so the incremental native build survives.
+        # shellcheck source=../build-cache.sh
+        source "$workspace_source/Launcher/build-cache.sh"
+        mkw_sync_tree "$workspace_source" "$workspace" aurora-main projects runtime translator Launcher
         /usr/bin/ditto "$source_bundle_version" "$workspace_bundle_version"
-        # A dependency provider can be cached in this directory, so make the
-        # refreshed sources configure from a clean native build tree.
-        rm -rf "$workspace/native-build-macos-arm64" "$workspace/native-build-macos-x86_64"
     fi
 fi
 
@@ -146,7 +149,7 @@ fi
 "$workspace/Launcher/local-build-macos.command" "${build_args[@]}" --profile "$profile" --cmake "$cmake_bin" --ninja "$ninja_bin" --translator-bin "$translator"
 
 config="$support_root/Config.toml"
-toml_string() { printf '%s' "$1" | sed -e 's/\\\\/\\\\\\\\/g' -e 's/"/\\\\"/g'; }
+toml_string() { printf '%s' "$1" | sed -e 's/\\/\\\\/g' -e 's/"/\\"/g'; }
 set_path() {
     local key=$1 value=$2 encoded line temporary
     encoded=$(toml_string "$value"); line="$key = \"$encoded\""; temporary="$config.tmp"
@@ -172,8 +175,13 @@ install_app() {
     local app=$1
     [[ -d "$products/$app" ]] || return 0
     if [[ "$destination" == /Applications ]]; then
-        command="mkdir -p /Applications && rm -rf '/Applications/$app' && ditto '$products/$app' '/Applications/$app'"
-        /usr/bin/osascript -e "do shell script \"$command\" with administrator privileges"
+        /usr/bin/osascript - "$products/$app" "/Applications/$app" <<'APPLESCRIPT'
+on run argv
+    set sourcePath to quoted form of (item 1 of argv)
+    set installedPath to quoted form of (item 2 of argv)
+    do shell script "mkdir -p /Applications && rm -rf " & installedPath & " && /usr/bin/ditto " & sourcePath & " " & installedPath with administrator privileges
+end run
+APPLESCRIPT
     else
         mkdir -p "$destination"; rm -rf "$destination/$app"; /usr/bin/ditto "$products/$app" "$destination/$app"
     fi

README.md

@@ -11,6 +11,7 @@
   <a href="#building-from-source"><img alt="PowerPC static recompilation" src="https://img.shields.io/badge/PowerPC-static%20recompilation-FF9F0A"></a>
   <a href="#retro-rewind"><img alt="Retro Rewind supported" src="https://img.shields.io/badge/Retro%20Rewind-supported-FF375F"></a>
   <a href="https://github.com/TeamWheelWizard/WheelWizard/releases"><img alt="Install with Wheel Wizard" src="https://img.shields.io/badge/install%20with-Wheel%20Wizard-8B5CF6"></a>
+  <a href="https://github.com/patchzyy/Wiicompiled/releases"><img alt="Total downloads" src="https://img.shields.io/github/downloads/patchzyy/Wiicompiled/total?color=00B8D9&amp;logo=github"></a>
   <a href="LICENSE"><img alt="License: GPLv3" src="https://img.shields.io/badge/license-GPLv3-2EA44F?logo=gnu&amp;logoColor=white"></a>
 </p>
 

aurora-main/cmake/AuroraCopyRuntimeDLLs.cmake

@@ -15,6 +15,7 @@ function(aurora_copy_runtime_dlls target)
     add_custom_command(TARGET ${target} POST_BUILD
         COMMAND ${CMAKE_COMMAND} -E copy_if_different $<TARGET_RUNTIME_DLLS:${_t}> $<TARGET_FILE_DIR:${target}>
         COMMAND_EXPAND_LISTS
+        VERBATIM
     )
   endforeach ()
 
@@ -30,6 +31,7 @@ function(aurora_copy_runtime_dlls target)
         if (EXISTS "${_dawn_bin_dir}/${_dll}")
           add_custom_command(TARGET ${target} POST_BUILD
               COMMAND ${CMAKE_COMMAND} -E copy_if_different "${_dawn_bin_dir}/${_dll}" $<TARGET_FILE_DIR:${target}>
+              VERBATIM
           )
         endif ()
       endforeach ()

aurora-main/cmake/AuroraSDL3Patches.cmake

@@ -32,6 +32,52 @@ function(aurora_sdl3_apply_patches sdl_source_dir)
   if (NOT EXISTS "${_wii}")
     message(FATAL_ERROR "aurora: SDL3 source tree at ${sdl_source_dir} has no SDL_hidapi_wii.c")
   endif ()
+  set(_rumble "${sdl_source_dir}/src/joystick/hidapi/SDL_hidapi_rumble.c")
+  if (NOT EXISTS "${_rumble}")
+    message(FATAL_ERROR "aurora: SDL3 source tree at ${sdl_source_dir} has no SDL_hidapi_rumble.c")
+  endif ()
+  set(_hidapi "${sdl_source_dir}/src/joystick/hidapi/SDL_hidapijoystick.c")
+  if (NOT EXISTS "${_hidapi}")
+    message(FATAL_ERROR "aurora: SDL3 source tree at ${sdl_source_dir} has no SDL_hidapijoystick.c")
+  endif ()
+
+  # Backport of SDL 71f4af732 (drop once SDL includes it): rumble writes must not hold dev_lock.
+  _aurora_sdl3_replace("${_rumble}" "HIDAPI rumble writes no longer hold the device lock"
+[==[        if (request) {
+            SDL_LockMutex(request->device->dev_lock);
+            if (request->device->dev) {
+#ifdef DEBUG_RUMBLE
+                HIDAPI_DumpPacket("Rumble packet: size = %d", request->data, request->size);
+#endif
+                SDL_hid_write(request->device->dev, request->data, request->size);
+            }
+            SDL_UnlockMutex(request->device->dev_lock);
+]==]
+[==[        if (request) {
+            if (request->device->dev) {
+#ifdef DEBUG_RUMBLE
+                HIDAPI_DumpPacket("Rumble packet: size = %d", request->data, request->size);
+#endif
+                SDL_hid_write(request->device->dev, request->data, request->size);
+            }
+]==])
+
+  # Without the lock, closing the handle must wait for in-flight rumble writes.
+  _aurora_sdl3_replace("${_hidapi}" "HIDAPI device close waits for pending rumble"
+[==[    device->driver->FreeDevice(device);
+    device->driver = NULL;
+
+    SDL_LockMutex(device->dev_lock);
+]==]
+[==[    device->driver->FreeDevice(device);
+    device->driver = NULL;
+
+    while (SDL_GetAtomicInt(&device->rumble_pending) > 0) {
+        SDL_Delay(10);
+    }
+
+    SDL_LockMutex(device->dev_lock);
+]==])
 
   # Wii Remote: rebuild the joystick in place after an extension change.
   #
@@ -118,6 +164,219 @@ static void UpdateDeviceIdentity(SDL_HIDAPI_Device *device)
 
     now = SDL_GetTicks();
 ]==])
+
+  # Classic Controller sticks: the stock path stretches each half-range by learned min/max;
+  # this posts WPAD's linear value (raw - first-report centre) * unit, times 64.
+  _aurora_sdl3_replace("${_wii}" "Wii Classic Controller linear sticks (helper)"
+[==[static void HandleGamepadControllerButtonData(SDL_DriverWii_Context *ctx, SDL_Joystick *joystick, const WiiButtonData *data)
+{
+]==]
+[==[/* WiiCompiled: axis = (raw - centre) * unit * 64, y inverted; see AuroraSDL3Patches.cmake */
+static void PostClassicStickLinear(Uint64 timestamp, SDL_Joystick *joystick, StickCalibrationData *calibration, Uint8 axis, Uint16 data, int unit)
+{
+    int value;
+
+    if (!calibration->center) {
+        // Center on first read
+        calibration->center = data;
+        return;
+    }
+    value = ((int)data - (int)calibration->center) * unit * 64;
+    if (axis == SDL_GAMEPAD_AXIS_LEFTY || axis == SDL_GAMEPAD_AXIS_RIGHTY) {
+        value = -value;
+    }
+    SDL_SendJoystickAxis(timestamp, joystick, axis, (Sint16)SDL_clamp(value, SDL_JOYSTICK_AXIS_MIN, SDL_JOYSTICK_AXIS_MAX));
+}
+
+static void HandleGamepadControllerButtonData(SDL_DriverWii_Context *ctx, SDL_Joystick *joystick, const WiiButtonData *data)
+{
+]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Classic Controller linear sticks (axes)"
+[==[    PostStickCalibrated(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[0], SDL_GAMEPAD_AXIS_LEFTX, lx);
+    PostStickCalibrated(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[1], SDL_GAMEPAD_AXIS_LEFTY, ly);
+    PostStickCalibrated(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[2], SDL_GAMEPAD_AXIS_RIGHTX, rx);
+    PostStickCalibrated(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[3], SDL_GAMEPAD_AXIS_RIGHTY, ry);
+]==]
+[==[    PostClassicStickLinear(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[0], SDL_GAMEPAD_AXIS_LEFTX, lx, 16);
+    PostClassicStickLinear(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[1], SDL_GAMEPAD_AXIS_LEFTY, ly, 16);
+    PostClassicStickLinear(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[2], SDL_GAMEPAD_AXIS_RIGHTX, rx, 32);
+    PostClassicStickLinear(ctx->timestamp, joystick, &ctx->m_StickCalibrationData[3], SDL_GAMEPAD_AXIS_RIGHTY, ry, 32);
+]==])
+
+  # Wii Remote: decode the accelerometer with the factory calibration from the EEPROM.
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (context fields)"
+[==[    StickCalibrationData m_StickCalibrationData[6];
+} SDL_DriverWii_Context;
+]==]
+[==[    StickCalibrationData m_StickCalibrationData[6];
+
+    float m_fAccelZeroG[3];
+    float m_fAccelPerG[3];
+} SDL_DriverWii_Context;
+]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (memory space parameter)"
+[==[static bool ReadRegister(SDL_DriverWii_Context *ctx, Uint32 address, int size, bool sync)
+{
+    Uint8 readRequest[7];
+
+    readRequest[0] = k_eWiiOutputReportIDs_ReadMemory;
+    readRequest[1] = (Uint8)(0x04 | (Uint8)ctx->m_bRumbleActive);
+]==]
+[==[static bool ReadMemory(SDL_DriverWii_Context *ctx, Uint8 addressSpace, Uint32 address, int size, bool sync);
+
+static bool ReadRegister(SDL_DriverWii_Context *ctx, Uint32 address, int size, bool sync)
+{
+    return ReadMemory(ctx, 0x04, address, size, sync);
+}
+
+static bool ReadMemory(SDL_DriverWii_Context *ctx, Uint8 addressSpace, Uint32 address, int size, bool sync)
+{
+    Uint8 readRequest[7];
+
+    readRequest[0] = k_eWiiOutputReportIDs_ReadMemory;
+    readRequest[1] = (Uint8)(addressSpace | (Uint8)ctx->m_bRumbleActive);
+]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (EEPROM read)"
+[==[static bool HIDAPI_DriverWii_InitDevice(SDL_HIDAPI_Device *device)
+{
+]==]
+[==[// Nominal decode: zero-g at 0x200, 100 units per g
+#define WII_ACCEL_NOMINAL_ZERO_G  512.0f
+#define WII_ACCEL_NOMINAL_PER_G   100.0f
+
+// EEPROM calibration at 0x16, copy at 0x20: zero-g XYZ, 1 g XYZ, LSBs of both, checksum
+#define WII_ACCEL_CALIBRATION_SIZE 10
+#define WII_ADDRESS_SPACE_EEPROM   0x00
+
+static bool ReadAccelCalibrationBlock(SDL_DriverWii_Context *ctx, Uint8 address, Uint8 *block)
+{
+    if (!ReadMemory(ctx, WII_ADDRESS_SPACE_EEPROM, address, WII_ACCEL_CALIBRATION_SIZE, true)) {
+        return false;
+    }
+    if (ctx->m_rgucReadBuffer[5] != address ||
+        ctx->m_rgucReadBuffer[3] != ((WII_ACCEL_CALIBRATION_SIZE - 1) << 4)) {
+        SDL_SetError("Unexpected EEPROM read response");
+        return false;
+    }
+    SDL_memcpy(block, ctx->m_rgucReadBuffer + 6, WII_ACCEL_CALIBRATION_SIZE);
+    return true;
+}
+
+static bool ParseAccelCalibration(SDL_DriverWii_Context *ctx, const Uint8 *block)
+{
+    int zero[3];
+    int perG[3];
+    Uint8 checksum = 0x55;
+    int i;
+
+    for (i = 0; i < WII_ACCEL_CALIBRATION_SIZE - 1; ++i) {
+        checksum = (Uint8)(checksum + block[i]);
+    }
+    if (checksum != block[WII_ACCEL_CALIBRATION_SIZE - 1]) {
+        SDL_SetError("Bad checksum");
+        return false;
+    }
+
+    for (i = 0; i < 3; ++i) {
+        int shift = 4 - 2 * i;
+        int oneG = (block[4 + i] << 2) | ((block[7] >> shift) & 0x03);
+
+        zero[i] = (block[i] << 2) | ((block[3] >> shift) & 0x03);
+        perG[i] = oneG - zero[i];
+
+        // Sanity limits around the nominal values
+        if (zero[i] < 384 || zero[i] > 640 || perG[i] < 60 || perG[i] > 160) {
+            SDL_SetError("Implausible values");
+            return false;
+        }
+    }
+
+    for (i = 0; i < 3; ++i) {
+        ctx->m_fAccelZeroG[i] = (float)zero[i];
+        ctx->m_fAccelPerG[i] = (float)perG[i];
+    }
+    return true;
+}
+
+static void ReadAccelCalibration(SDL_DriverWii_Context *ctx)
+{
+    static const Uint8 addresses[] = { 0x16, 0x20 };
+    Uint8 block[WII_ACCEL_CALIBRATION_SIZE];
+    int i;
+
+    for (i = 0; i < SDL_arraysize(addresses); ++i) {
+        if (ReadAccelCalibrationBlock(ctx, addresses[i], block) &&
+            ParseAccelCalibration(ctx, block)) {
+            SDL_LogInfo(SDL_LOG_CATEGORY_INPUT, "HIDAPI Wii: Accelerometer calibration from EEPROM 0x%02X: zero %d/%d/%d, %d/%d/%d units per g",
+                        addresses[i],
+                        (int)ctx->m_fAccelZeroG[0], (int)ctx->m_fAccelZeroG[1], (int)ctx->m_fAccelZeroG[2],
+                        (int)ctx->m_fAccelPerG[0], (int)ctx->m_fAccelPerG[1], (int)ctx->m_fAccelPerG[2]);
+            return;
+        }
+        SDL_LogInfo(SDL_LOG_CATEGORY_INPUT, "HIDAPI Wii: Accelerometer calibration at EEPROM 0x%02X unusable: %s", addresses[i], SDL_GetError());
+    }
+    SDL_LogInfo(SDL_LOG_CATEGORY_INPUT, "HIDAPI Wii: Using nominal accelerometer calibration");
+}
+
+static bool HIDAPI_DriverWii_InitDevice(SDL_HIDAPI_Device *device)
+{
+]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (nominal defaults)"
+[==[    ctx->device = device;
+    device->context = ctx;
+]==]
+[==[    ctx->device = device;
+    device->context = ctx;
+    for (int i = 0; i < 3; ++i) {
+        ctx->m_fAccelZeroG[i] = WII_ACCEL_NOMINAL_ZERO_G;
+        ctx->m_fAccelPerG[i] = WII_ACCEL_NOMINAL_PER_G;
+    }
+]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (read at init)"
+[==[        ctx->m_eExtensionControllerType = ReadExtensionControllerType(device);
+]==]
+[==[        ctx->m_eExtensionControllerType = ReadExtensionControllerType(device);
+        ReadAccelCalibration(ctx);
+]==])
+  # Reports without accelerometer bytes (status, memory) decoded as garbage.
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer guard"
+[==[    }
+
+    x = ((data->rgucAccelerometer[0] << 2) | ((data->rgucBaseButtons[0] >> 5) & 0x03))]==]
+[==[    }
+
+    if (!data->hasAccelerometer) {
+        return;
+    }
+
+    x = ((data->rgucAccelerometer[0] << 2) | ((data->rgucBaseButtons[0] >> 5) & 0x03))]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (nominal constant)"
+[==[const WiiButtonData *data)
+{
+    const float ACCEL_RES_PER_G = 100.0f;
+    Sint16 x, y, z;
+]==]
+[==[const WiiButtonData *data)
+{
+    Sint16 x, y, z;
+]==])
+  _aurora_sdl3_replace("${_wii}" "Wii Remote accelerometer calibration (decode)"
+[==[    x = ((data->rgucAccelerometer[0] << 2) | ((data->rgucBaseButtons[0] >> 5) & 0x03)) - 0x200;
+    y = ((data->rgucAccelerometer[1] << 2) | ((data->rgucBaseButtons[1] >> 4) & 0x02)) - 0x200;
+    z = ((data->rgucAccelerometer[2] << 2) | ((data->rgucBaseButtons[1] >> 5) & 0x02)) - 0x200;
+
+    values[0] = -((float)x / ACCEL_RES_PER_G) * SDL_STANDARD_GRAVITY;
+    values[1] = ((float)z / ACCEL_RES_PER_G) * SDL_STANDARD_GRAVITY;
+    values[2] = ((float)y / ACCEL_RES_PER_G) * SDL_STANDARD_GRAVITY;
+]==]
+[==[    x = ((data->rgucAccelerometer[0] << 2) | ((data->rgucBaseButtons[0] >> 5) & 0x03));
+    y = ((data->rgucAccelerometer[1] << 2) | ((data->rgucBaseButtons[1] >> 4) & 0x02));
+    z = ((data->rgucAccelerometer[2] << 2) | ((data->rgucBaseButtons[1] >> 5) & 0x02));
+
+    values[0] = -(((float)x - ctx->m_fAccelZeroG[0]) / ctx->m_fAccelPerG[0]) * SDL_STANDARD_GRAVITY;
+    values[1] = (((float)z - ctx->m_fAccelZeroG[2]) / ctx->m_fAccelPerG[2]) * SDL_STANDARD_GRAVITY;
+    values[2] = (((float)y - ctx->m_fAccelZeroG[1]) / ctx->m_fAccelPerG[1]) * SDL_STANDARD_GRAVITY;
+]==])
 endfunction()
 
 # Script mode (FetchContent PATCH_COMMAND).

aurora-main/lib/dolphin/gx/GXDispList.cpp

@@ -66,7 +66,7 @@ void GXCallDisplayList(const void* data, u32 nbytes) {
 
   // Decode the display list immediately while its borrowed resources are valid.
   aurora::gx::fifo::drain();
-  aurora::gx::fifo::process(static_cast<const u8*>(data), nbytes, true);
+  aurora::gx::fifo::process_all(static_cast<const u8*>(data), nbytes, true);
 }
 
 void GXCallDisplayListLE(const void* data, u32 nbytes) {
@@ -84,6 +84,6 @@ void GXCallDisplayListLE(const void* data, u32 nbytes) {
   aurora::gx::fifo::drain();
 
   // Process the display list through the command processor (little-endian)
-  aurora::gx::fifo::process(static_cast<const u8*>(data), nbytes, false);
+  aurora::gx::fifo::process_all(static_cast<const u8*>(data), nbytes, false);
 }
 }

aurora-main/lib/dolphin/pad/pad.cpp

@@ -195,7 +195,7 @@ std::array<PADButtonMapping, PAD_BUTTON_COUNT> g_defaultButtonsJoyPair{{
 }};
 
 // Wii U Pro Controllers through SDL's HIDAPI Wii driver. No SDL_GamepadType
-// singles them out, so they are picked by the name the driver gives them (see
+// singles them out, so they are picked by name or USB id (see
 // __PADSetDefaultMapping). Wii Remotes, with or without a Nunchuk or Classic
 // Controller, are read by the game through KPAD instead and never get a
 // GameCube mapping (the runtime hides those ports from PADRead).
@@ -451,11 +451,14 @@ static void reset_alt_button_mapping(aurora::input::GameController* controller)
 
 // SDL's hidapi Wii driver names the pad "Nintendo Wii U Pro Controller"; the
 // other names it produces are Wii Remotes, which the game reads through KPAD
-// and which therefore never take a GameCube mapping.
+// and which therefore never take a GameCube mapping. The Linux kernel driver's
+// "Nintendo Wii Remote Pro Controller" is only recognised by its USB id.
 static bool wii_default_mapping(const aurora::input::GameController* controller,
                                 std::array<PADButtonMapping, PAD_BUTTON_COUNT>& out) {
   const char* name = SDL_GetGamepadName(controller->m_controller);
-  if (name == nullptr || SDL_strstr(name, "Wii U Pro Controller") == nullptr) {
+  const bool wiiUProId = SDL_GetGamepadVendor(controller->m_controller) == 0x057e &&
+                         SDL_GetGamepadProduct(controller->m_controller) == 0x0330;
+  if (!wiiUProId && (name == nullptr || SDL_strstr(name, "Wii U Pro Controller") == nullptr)) {
     return false;
   }
   out = g_defaultButtonsWiiUPro;

aurora-main/lib/gfx/texture_replacement.cpp

@@ -553,7 +553,7 @@ void build_index() noexcept {
       continue;
     }
 
-    s_replacementIndex.try_emplace(*parsed, path);
+    s_replacementIndex.try_emplace(*parsed, ReplacementIndexEntry{path});
   }
 
   Log.info("Indexed {} texture replacements", s_replacementIndex.size());

aurora-main/lib/gx/fifo.cpp

@@ -82,19 +82,23 @@ void drain() {
   if (detail::sBufferSize == 0) {
     return;
   }
+  process_all(detail::sBufferData, detail::sBufferSize, true);
+  detail::sBufferSize = 0;
+}
+
+void process_all(const uint8_t* data, uint32_t size, bool bigEndian) {
   uint32_t consumed = 0;
   bool retried = false;
-  while (consumed < detail::sBufferSize) {
-    const auto count = process(detail::sBufferData + consumed, detail::sBufferSize - consumed, true);
+  while (consumed < size) {
+    const auto count = process(data + consumed, size - consumed, bigEndian);
     if (count == 0 && retried)
       throw gfx::StagingCapacityError("FIFO draw does not fit after capacity submission");
     consumed += count;
-    if (consumed == detail::sBufferSize) break;
+    if (consumed == size) break;
     // process returned with its renderer lock released. No recursive drain.
     gfx::split_staging_batch();
     retried = true;
   }
-  detail::sBufferSize = 0;
 }
 
 const uint8_t* get_buffer_data() { return detail::sBufferData; }

aurora-main/lib/gx/fifo.hpp

@@ -81,6 +81,10 @@ bool in_display_list();
 // Drain the internal FIFO buffer through the command processor
 void drain();
 
+// Decode a whole command stream, submitting the staging batch whenever a draw does not fit.
+// process() alone stops at that draw, so a caller that ignores its count drops the rest.
+void process_all(const uint8_t* data, uint32_t size, bool bigEndian);
+
 // Internal buffer inspection
 const uint8_t* get_buffer_data();
 uint32_t get_buffer_size();

aurora-main/lib/gx/shader.cpp

@@ -174,6 +174,12 @@ static std::string color_arg_reg(GXTevColorArg arg, size_t stageIdx, const Shade
   case GX_CC_KONST: {
     switch (stage.kcSel) {
       DEFAULT_FATAL("invalid kcSel {}", underlying(stage.kcSel));
+    // Unsourced encodings evaluate to zero in hardware.
+    case static_cast<GXTevKColorSel>(0x08):
+    case static_cast<GXTevKColorSel>(0x09):
+    case static_cast<GXTevKColorSel>(0x0A):
+    case static_cast<GXTevKColorSel>(0x0B):
+      return "vec3f(0.0)";
     case GX_TEV_KCSEL_8_8:
       return "vec3f(1.0)";
     case GX_TEV_KCSEL_7_8:
@@ -275,6 +281,16 @@ static std::string alpha_arg_reg(GXTevAlphaArg arg, size_t stageIdx, const Shade
   case GX_CA_KONST: {
     switch (stage.kaSel) {
       DEFAULT_FATAL("invalid kaSel {}", underlying(stage.kaSel));
+    // Unsourced encodings evaluate to zero in hardware.
+    case static_cast<GXTevKAlphaSel>(0x08):
+    case static_cast<GXTevKAlphaSel>(0x09):
+    case static_cast<GXTevKAlphaSel>(0x0A):
+    case static_cast<GXTevKAlphaSel>(0x0B):
+    case static_cast<GXTevKAlphaSel>(0x0C):
+    case static_cast<GXTevKAlphaSel>(0x0D):
+    case static_cast<GXTevKAlphaSel>(0x0E):
+    case static_cast<GXTevKAlphaSel>(0x0F):
+      return "0.0";
     case GX_TEV_KASEL_8_8:
       return "1.0";
     case GX_TEV_KASEL_7_8:

aurora-main/lib/input.cpp

@@ -14,6 +14,7 @@
 #include <array>
 #include <string>
 #include <utility>
+#include <vector>
 
 using namespace std::string_view_literals;
 
@@ -23,7 +24,7 @@ absl::flat_hash_map<Uint32, GameController> g_GameControllers;
 
 namespace {
 constexpr uint32_t kPortPreferencesMagic = SBIG('CPRT');
-constexpr uint32_t kPortPreferencesVersion = 2;
+constexpr uint32_t kPortPreferencesVersion = 3;
 constexpr uint32_t kMaxPersistedStringLength = 256;
 
 enum class PortPreferenceState : uint8_t {
@@ -35,6 +36,8 @@ enum class PortPreferenceState : uint8_t {
 struct ControllerIdentity {
   std::string guid;
   std::string serial;
+  std::string path;
+  int32_t slot = -1;
 };
 
 struct PortPreference {
@@ -44,6 +47,9 @@ struct PortPreference {
 
 std::array<PortPreference, PAD_MAX_CONTROLLERS> g_portPreferences;
 bool g_portPreferencesLoaded = false;
+// Startup controllers are added one by one via events; capture their SDL player index at init,
+// before earlier port reassignments can move it.
+absl::flat_hash_map<SDL_JoystickID, int32_t> g_connectSlots;
 
 std::string port_preferences_path() {
   if (g_config.userPath == nullptr) {
@@ -83,6 +89,8 @@ ControllerIdentity controller_identity(const GameController& controller) {
   SDL_GUIDToString(SDL_GetGamepadGUIDForID(SDL_GetGamepadID(controller.m_controller)), guid, sizeof(guid));
   identity.guid = guid;
   identity.serial = normalize_serial(SDL_GetGamepadSerial(controller.m_controller));
+  identity.path = controller.m_path;
+  identity.slot = controller.m_slot;
   return identity;
 }
 
@@ -137,12 +145,16 @@ bool write_string(SDL_IOStream* file, const std::string& value) {
   return write_value(file, size) && (size == 0 || write_exact(file, value.data(), size));
 }
 
-bool read_identity(SDL_IOStream* file, ControllerIdentity& identity) {
-  return read_string(file, identity.guid) && read_string(file, identity.serial);
+bool read_identity(SDL_IOStream* file, uint32_t version, ControllerIdentity& identity) {
+  if (!read_string(file, identity.guid) || !read_string(file, identity.serial)) {
+    return false;
+  }
+  return version < 3 || (read_string(file, identity.path) && read_value(file, identity.slot));
 }
 
 bool write_identity(SDL_IOStream* file, const ControllerIdentity& identity) {
-  return write_string(file, identity.guid) && write_string(file, identity.serial);
+  return write_string(file, identity.guid) && write_string(file, identity.serial) &&
+         write_string(file, identity.path) && write_value(file, identity.slot);
 }
 
 bool read_port_preferences_file(std::array<PortPreference, PAD_MAX_CONTROLLERS>& preferences) {
@@ -159,13 +171,13 @@ bool read_port_preferences_file(std::array<PortPreference, PAD_MAX_CONTROLLERS>&
   uint32_t magic = 0;
   uint32_t version = 0;
   bool ok = read_value(file, magic) && read_value(file, version) && magic == kPortPreferencesMagic &&
-            version == kPortPreferencesVersion;
+            version >= 2 && version <= kPortPreferencesVersion;
 
   for (auto& preference : preferences) {
     uint8_t state = 0;
     ControllerIdentity identity;
     ok = ok && read_value(file, state) && state <= static_cast<uint8_t>(PortPreferenceState::Controller) &&
-         read_identity(file, identity);
+         read_identity(file, version, identity);
     if (ok) {
       preference.state = static_cast<PortPreferenceState>(state);
       preference.identity = std::move(identity);
@@ -225,22 +237,31 @@ void save_port_preferences() {
   }
 }
 
-enum class IdentityMatch {
-  None,
-  Fallback,
-  Exact,
-};
+// Zero means no match. Higher scores claim ports first: path+slot > GUID > VID/PID.
+constexpr int kScoreVidPid = 1;
+constexpr int kScoreGuid = 2;
+constexpr int kScoreSlot = 1;
+constexpr int kScorePath = 2;
 
-IdentityMatch identity_match(const ControllerIdentity& saved, const ControllerIdentity& current) {
+int identity_score(const ControllerIdentity& saved, const ControllerIdentity& current) {
   if (saved.guid.empty()) {
-    return IdentityMatch::None;
+    return 0;
   }
   if (saved.guid == current.guid) {
-    return saved.serial.empty() || serial_matches(saved.serial, current.serial) ? IdentityMatch::Exact
-                                                                                : IdentityMatch::None;
+    if (!saved.serial.empty() && !serial_matches(saved.serial, current.serial)) {
+      return 0;
+    }
+    int score = kScoreGuid;
+    if (!saved.path.empty() && saved.path == current.path) {
+      score += kScorePath;
+    }
+    if (saved.slot >= 0 && saved.slot == current.slot) {
+      score += kScoreSlot;
+    }
+    return score;
   }
   if (!serial_matches(saved.serial, current.serial)) {
-    return IdentityMatch::None;
+    return 0;
   }
 
   // Attempt to match against VID/PID as a fallback if the GUID changes
@@ -250,9 +271,9 @@ IdentityMatch identity_match(const ControllerIdentity& saved, const ControllerId
   uint16_t currentProduct = 0;
   SDL_GetJoystickGUIDInfo(SDL_StringToGUID(saved.guid.c_str()), &savedVendor, &savedProduct, nullptr, nullptr);
   SDL_GetJoystickGUIDInfo(SDL_StringToGUID(current.guid.c_str()), &currentVendor, &currentProduct, nullptr, nullptr);
-  return savedVendor != 0 && savedVendor == currentVendor && savedProduct != 0 && savedProduct == currentProduct
-             ? IdentityMatch::Fallback
-             : IdentityMatch::None;
+  const bool sameDevice =
+      savedVendor != 0 && savedVendor == currentVendor && savedProduct != 0 && savedProduct == currentProduct;
+  return sameDevice ? kScoreVidPid : 0;
 }
 
 void assign_player_index(GameController& controller, int32_t port) {
@@ -267,12 +288,6 @@ int32_t effective_player_index(const GameController& controller) {
   return player >= 0 ? player : controller.m_playerIndex;
 }
 
-bool is_instance_claimed(const std::array<Uint32, PAD_MAX_CONTROLLERS>& claimedControllers, size_t claimedCount,
-                         Uint32 instance) {
-  return std::find(claimedControllers.begin(), claimedControllers.begin() + claimedCount, instance) !=
-         claimedControllers.begin() + claimedCount;
-}
-
 void apply_port_preferences() noexcept {
   ensure_port_preferences_loaded();
   if (!std::any_of(g_portPreferences.begin(), g_portPreferences.end(),
@@ -288,45 +303,63 @@ void apply_port_preferences() noexcept {
     }
   }
 
-  std::array<Uint32, PAD_MAX_CONTROLLERS> claimedControllers{};
+  struct Candidate {
+    int score;
+    uint32_t port;
+    Uint32 instance;
+  };
+  std::vector<Candidate> candidates;
+  for (const auto& [instance, controller] : g_GameControllers) {
+    const auto identity = controller_identity(controller);
+    for (uint32_t port = 0; port < g_portPreferences.size(); ++port) {
+      if (g_portPreferences[port].state != PortPreferenceState::Controller) {
+        continue;
+      }
+      if (const int score = identity_score(g_portPreferences[port].identity, identity); score > 0) {
+        candidates.push_back({score, port, instance});
+      }
+    }
+  }
+  // Best matches claim first so identical controllers each end up on their own port
+  std::sort(candidates.begin(), candidates.end(), [](const Candidate& a, const Candidate& b) {
+    if (a.score != b.score) {
+      return a.score > b.score;
+    }
+    return a.port != b.port ? a.port < b.port : a.instance < b.instance;
+  });
+
+  std::array<bool, PAD_MAX_CONTROLLERS> portClaimed{};
+  std::array<Uint32, PAD_MAX_CONTROLLERS> claimedInstances{};
   size_t claimedCount = 0;
-  for (uint32_t port = 0; port < g_portPreferences.size(); ++port) {
-    const auto& preference = g_portPreferences[port];
-    if (preference.state != PortPreferenceState::Controller) {
+  bool upgraded = false;
+  for (const auto& candidate : candidates) {
+    if (portClaimed[candidate.port] ||
+        std::find(claimedInstances.begin(), claimedInstances.begin() + claimedCount, candidate.instance) !=
+            claimedInstances.begin() + claimedCount) {
       continue;
     }
+    portClaimed[candidate.port] = true;
+    claimedInstances[claimedCount++] = candidate.instance;
+    auto& controller = g_GameControllers[candidate.instance];
+    assign_player_index(controller, static_cast<int32_t>(candidate.port));
 
-    Uint32 fallbackInstance = 0;
-    GameController* fallbackController = nullptr;
-    for (auto& [instance, controller] : g_GameControllers) {
-      if (is_instance_claimed(claimedControllers, claimedCount, instance)) {
-        continue;
+    // Preferences saved before path and slot existed gain them on their first GUID match
+    auto& saved = g_portPreferences[candidate.port].identity;
+    if (candidate.score >= kScoreGuid && (saved.path.empty() || saved.slot < 0)) {
+      const auto current = controller_identity(controller);
+      if (saved.path.empty() && !current.path.empty()) {
+        saved.path = current.path;
+        upgraded = true;
       }
-
-      switch (identity_match(preference.identity, controller_identity(controller))) {
-      case IdentityMatch::Exact:
-        assign_player_index(controller, static_cast<int32_t>(port));
-        claimedControllers[claimedCount++] = instance;
-        fallbackController = nullptr;
-        break;
-      case IdentityMatch::Fallback:
-        // Prefer any later exact match before claiming a fallback candidate
-        if (fallbackController == nullptr) {
-          fallbackInstance = instance;
-          fallbackController = &controller;
-        }
-        continue;
-      case IdentityMatch::None:
-        continue;
+      if (saved.slot < 0 && current.slot >= 0) {
+        saved.slot = current.slot;
+        upgraded = true;
       }
-      break;
-    }
-
-    if (fallbackController != nullptr) {
-      assign_player_index(*fallbackController, static_cast<int32_t>(port));
-      claimedControllers[claimedCount++] = fallbackInstance;
     }
   }
+  if (upgraded) {
+    save_port_preferences();
+  }
 }
 
 // SDL only hands out a player index when the device already had a gamepad mapping
@@ -401,6 +434,15 @@ SDL_JoystickID add_controller(SDL_JoystickID which) noexcept {
       return -1;
     }
     controller.m_isGameCube = controller.m_vid == 0x057E && controller.m_pid == 0x0337;
+    if (const char* path = SDL_GetGamepadPath(ctrl); path != nullptr) {
+      controller.m_path = path;
+    }
+    if (const auto slot = g_connectSlots.find(which); slot != g_connectSlots.end()) {
+      controller.m_slot = slot->second;
+      g_connectSlots.erase(slot);
+    } else {
+      controller.m_slot = SDL_GetGamepadPlayerIndex(ctrl);
+    }
     const char* serial = SDL_GetGamepadSerial(ctrl);
     controller.m_gameCubeUseOrdinaryStop = controller.m_isGameCube && serial && "GCP+"sv == serial;
     if (controller.m_isGameCube ||
@@ -420,8 +462,7 @@ SDL_JoystickID add_controller(SDL_JoystickID which) noexcept {
       bool hasOtherPortPreference = false;
       for (size_t port = 1; port < g_portPreferences.size(); ++port) {
         if (g_portPreferences[port].state == PortPreferenceState::Controller &&
-            identity_match(g_portPreferences[port].identity, controller_identity(g_GameControllers[instance])) !=
-                IdentityMatch::None) {
+            identity_score(g_portPreferences[port].identity, controller_identity(g_GameControllers[instance])) > 0) {
           hasOtherPortPreference = true;
           break;
         }
@@ -457,6 +498,7 @@ bool refresh_controller(SDL_JoystickID instance) noexcept {
 }
 
 void remove_controller(Uint32 instance) noexcept {
+  g_connectSlots.erase(instance);
   if (auto it = g_GameControllers.find(instance); it != g_GameControllers.end()) {
     SDL_CloseGamepad(it->second.m_controller);
     g_GameControllers.erase(it);
@@ -540,6 +582,14 @@ void initialize() noexcept {
    * as expected */
   ASSERT(SDL_Init(SDL_INIT_HAPTIC | SDL_INIT_JOYSTICK | SDL_INIT_GAMEPAD), "Failed to initialize SDL subsystems: {}",
          SDL_GetError());
+
+  int count = 0;
+  if (SDL_JoystickID* ids = SDL_GetGamepads(&count)) {
+    for (int i = 0; i < count; ++i) {
+      g_connectSlots[ids[i]] = SDL_GetGamepadPlayerIndexForID(ids[i]);
+    }
+    SDL_free(ids);
+  }
 }
 
 struct MouseScrollStatus {

aurora-main/lib/input.hpp

@@ -20,6 +20,9 @@ struct GameController {
   bool m_gameCubeUseOrdinaryStop = false;
   Sint32 m_index = -1;
   Sint32 m_playerIndex = -1;
+  // Device path and the SDL player index at connect (the adapter port for GameCube adapters).
+  std::string m_path;
+  Sint32 m_slot = -1;
   bool m_hasRumble = false;
   PADDeadZones m_deadZones{
       .emulateTriggers = true,

aurora-main/lib/webgpu/gpu.cpp

@@ -893,9 +893,14 @@ void release_surface() noexcept {
   }
   g_surface = {};
   if (hadSurface && g_instance && g_device && g_queue) {
+    // Retired swapchain is freed after a later submission completes and the device ticks.
+    submit_staging_commands(g_device.CreateCommandEncoder().Finish());
     const auto future = g_queue.OnSubmittedWorkDone(
         wgpu::CallbackMode::WaitAnyOnly, [](wgpu::QueueWorkDoneStatus, wgpu::StringView) {});
-    g_instance.WaitAny(future, 1000000000);
+    if (const auto status = g_instance.WaitAny(future, 1000000000); status != wgpu::WaitStatus::Success) {
+      Log.warn("Timed out waiting for queue idle on surface release: {}", magic_enum::enum_name(status));
+    }
+    g_device.Tick();
   }
 }
 

aurora-main/tests/gx_test_common.hpp

@@ -50,6 +50,9 @@ void reset_uniform_allocations() noexcept;
 const std::vector<uint8_t>& uniform_allocation(size_t index) noexcept;
 void use_draw_command_tracking(bool enabled) noexcept;
 void use_real_vertex_format_helpers(bool enabled) noexcept;
+// Admit this many draws, then refuse one as if the staging batch were full.
+void refuse_staging_admission_after(uint32_t admissions) noexcept;
+void reset_staging_capacity() noexcept;
 } // namespace aurora::gfx::testing
 
 class GXFifoTest : public ::testing::Test {
@@ -65,6 +68,7 @@ protected:
     aurora::gfx::testing::reset_resolve_pass_records();
     aurora::gfx::testing::reset_vertex_push_record();
     aurora::gfx::testing::use_real_vertex_format_helpers(false);
+    aurora::gfx::testing::reset_staging_capacity();
   }
 
   // Copy the internal FIFO buffer contents and clear it

aurora-main/tests/gx_test_stubs.cpp

@@ -34,6 +34,9 @@ std::optional<aurora::gx::DrawData> s_lastGxDraw;
 bool s_trackDrawCommands = false;
 bool s_useRealVertexFormatHelpers = false;
 std::deque<std::vector<uint8_t>> s_uniformAllocations;
+std::optional<uint32_t> s_stagingAdmissionsBeforeRefusal;
+bool s_stagingRefused = false;
+uint64_t s_stagingSplitCount = 0;
 } // namespace
 
 // --- aurora::g_config ---
@@ -301,8 +304,23 @@ std::pair<ByteBuffer, Range> copy_uniform(Range source) {
 uint32_t align_uniform(uint32_t value) { return (value + 255u) & ~255u; }
 uint64_t staging_uniform_bytes(uint64_t value) { return staging_padded(value, 256); }
 uint64_t staging_storage_bytes(uint64_t value) { return staging_padded(value, 256); }
-bool staging_has_space(const StagingSizes&) { return true; }
-void split_staging_batch() { throw StagingCapacityError("Unexpected split in FIFO unit test"); }
+bool staging_has_space(const StagingSizes&) {
+  if (!s_stagingAdmissionsBeforeRefusal) return true;
+  if (*s_stagingAdmissionsBeforeRefusal > 0) {
+    --*s_stagingAdmissionsBeforeRefusal;
+    return true;
+  }
+  // Refuse once, as a full batch would; the split that follows makes room again.
+  s_stagingAdmissionsBeforeRefusal.reset();
+  s_stagingRefused = true;
+  return false;
+}
+void split_staging_batch() {
+  if (!s_stagingRefused) throw StagingCapacityError("Unexpected split in FIFO unit test");
+  s_stagingRefused = false;
+  ++s_stagingSplitCount;
+}
+uint64_t staging_split_count() noexcept { return s_stagingSplitCount; }
 
 Vec2<uint32_t> get_render_target_size() noexcept { return s_renderTargetSize; }
 Vec2<uint32_t> get_frame_buffer_size() noexcept { return s_renderTargetSize; }
@@ -345,6 +363,15 @@ void use_draw_command_tracking(bool enabled) noexcept {
 void use_real_vertex_format_helpers(bool enabled) noexcept {
   s_useRealVertexFormatHelpers = enabled;
 }
+void refuse_staging_admission_after(uint32_t admissions) noexcept {
+  s_stagingAdmissionsBeforeRefusal = admissions;
+  s_stagingRefused = false;
+}
+void reset_staging_capacity() noexcept {
+  s_stagingAdmissionsBeforeRefusal.reset();
+  s_stagingRefused = false;
+  s_stagingSplitCount = 0;
+}
 } // namespace aurora::gfx::testing
 
 // --- Pipeline/draw command stubs ---
@@ -525,6 +552,7 @@ std::optional<TextureHandle> find_replacement(const GXTexObj_&) noexcept { retur
 namespace aurora::window {
 AuroraWindowSize get_window_size() { return {640, 480, 640, 480, 640, 480, 1.0f}; }
 void set_frame_buffer_aspect_fit(bool) {}
+void set_force_aspect_16_9(bool) {}
 } // namespace aurora::window
 
 // --- WebGPU C API stubs (prevent linker errors from wgpu:: destructors) ---

aurora-main/tests/renderer_regression_test.cpp

@@ -88,6 +88,31 @@ TEST_F(GXFifoTest, SingleExpandedPrimitiveCannotMergeWithTriangles) {
   EXPECT_EQ(aurora::gfx::testing::last_pushed_indices(), (std::vector<u16>{0, 1, 2}));
 }
 
+TEST_F(GXFifoTest, DisplayListSubmitsStagingWhenADrawDoesNotFit) {
+  __GXSetDirtyState();
+  aurora::gx::fifo::clear_buffer();
+  g_gxState.lastVtxFmt = GX_VTXFMT0;
+  g_gxState.lastVtxSize = 1;
+  // admitted=1 refuses the middle draw; admitted=2 refuses the last, so only a replay of the refused draw pushes 0x33.
+  for (const u32 admitted : {1u, 2u}) for (const bool bigEndian : {true, false}) {
+    std::vector<u8> list;
+    for (const u8 fill : {0x11, 0x22, 0x33}) {
+      auto bytes = draw(GX_TRIANGLES, 3);
+      if (!bigEndian) std::swap(bytes[1], bytes[2]);
+      std::fill(bytes.begin() + 3, bytes.end(), fill);
+      list.insert(list.end(), bytes.begin(), bytes.end());
+    }
+    const auto splitsBefore = aurora::gfx::staging_split_count();
+    aurora::gfx::testing::refuse_staging_admission_after(admitted);
+    if (bigEndian) GXCallDisplayList(list.data(), static_cast<u32>(list.size()));
+    else GXCallDisplayListLE(list.data(), static_cast<u32>(list.size()));
+    EXPECT_EQ(aurora::gfx::staging_split_count() - splitsBefore, 1u)
+        << "admitted=" << admitted << " bigEndian=" << bigEndian;
+    EXPECT_EQ(aurora::gfx::testing::last_pushed_vertices(), (std::vector<u8>{0x33, 0x33, 0x33}))
+        << "admitted=" << admitted << " bigEndian=" << bigEndian;
+  }
+}
+
 TEST(StagingMapping, RetiredCallbacksCannotPublishAnotherBuffersReadiness) {
   using namespace aurora::gfx;
   StagingMapState state;

runtime/CMakeLists.txt

@@ -42,12 +42,38 @@ else()
         "WiiCompiled supports 64-bit LLVM-MinGW Clang on Windows, native Linux x86_64/aarch64, or Apple Clang on macOS x86_64/arm64")
 endif()
 
+if(MKW_PLATFORM_WINDOWS)
+    # CMake and Ninja write response files with Windows quoting, but clang tokenizes them with
+    # POSIX rules by default, where a bare apostrophe in a path opens a quoted span.
+    add_link_options(--rsp-quoting=windows)
+endif()
+
 if(NOT CMAKE_BUILD_TYPE STREQUAL "Release")
     message(FATAL_ERROR "WiiCompiled only supports Release builds")
 endif()
 
 option(MKW_BUILD_PRODUCTS "Build translated WiiCompiled product targets" ON)
 option(MKW_BUILD_PSQ_TESTS "Build focused PSQ ISA tests" OFF)
+set(MKW_X86_CPU_PROFILE "v3" CACHE STRING
+    "Minimum x86-64 CPU profile for translated products (v2 or v3)")
+set_property(CACHE MKW_X86_CPU_PROFILE PROPERTY STRINGS v2 v3)
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$" AND
+   NOT MKW_X86_CPU_PROFILE MATCHES "^(v2|v3)$")
+    message(FATAL_ERROR "MKW_X86_CPU_PROFILE must be v2 or v3 (got '${MKW_X86_CPU_PROFILE}')")
+endif()
+
+set(MKW_SETUP_VERSION "" CACHE STRING "WiiCompiled version recorded in the runtime log")
+
+option(MKW_PLATFORM_TESTS_ONLY "Build host-platform tests without graphics dependencies or game assets" OFF)
+include("${CMAKE_CURRENT_LIST_DIR}/cmake/HostLibraries.cmake")
+if(MKW_PLATFORM_TESTS_ONLY)
+    if(MKW_BUILD_PRODUCTS)
+        message(FATAL_ERROR "MKW_PLATFORM_TESTS_ONLY requires MKW_BUILD_PRODUCTS=OFF")
+    endif()
+    add_compile_definitions(NOMINMAX)
+    include("${CMAKE_CURRENT_LIST_DIR}/cmake/PlatformTests.cmake")
+    return()
+endif()
 
 # Preprocessor definitions that belong to this project's own code (the runtime,
 # the translated shards and the product glue) and to nothing else. They are
@@ -68,6 +94,8 @@ set(MKW_NATIVE_PREBUILT_DIR "" CACHE PATH
 # package can be harvested from it.
 set(MKW_NATIVE_PREBUILT_EXPORT_DIR "" CACHE PATH
     "Maintainer-only: write a description of the configured aurora build here")
+option(MKW_USE_PREBUILT_RUNTIME
+    "Link the native prebuilt package's runtime archive when it matches this workspace" ON)
 
 # The two Crypto++ configuration macros are load-bearing for ABI: they shape the
 # structs and inline code the headers generate, so a first-party TU including
@@ -85,21 +113,6 @@ target_include_directories(mkw_pugixml PUBLIC third_party/pugixml)
 target_compile_features(mkw_pugixml PUBLIC cxx_std_17)
 set_target_properties(mkw_pugixml PROPERTIES UNITY_BUILD OFF)
 
-# POSIX x86-64 guest-fiber scheduling (runtime/src/host_context.cpp) needs a symmetric
-# stackful-coroutine primitive to stand in for Win32 Fibers. libco's co_switch() transfers
-# directly to any other created coroutine, matching SwitchToFiber's semantics exactly (unlike
-# asymmetric resume/yield coroutine libraries, which would need every call site restructured).
-# Vendored from upstream (higan-emu/libco @ e18e09d, 2019-10-16, ISC license; valgrind.h is
-# separately BSD-style licensed, see third_party/libco/LICENSE). Windows keeps native Fibers;
-# Apple Silicon uses the project's x18-safe AArch64 assembly backend, while Intel macOS uses
-# libco's existing System V AMD64 backend.
-if(MKW_PLATFORM_LINUX OR MKW_PLATFORM_MACOS_X86_64)
-    add_library(mkw_libco STATIC third_party/libco/libco.c)
-    add_library(mkw::libco ALIAS mkw_libco)
-    target_include_directories(mkw_libco PUBLIC third_party/libco)
-    set_target_properties(mkw_libco PROPERTIES UNITY_BUILD OFF)
-endif()
-
 # Runtime configuration is real TOML, parsed by toml11 rather than a project-
 # specific line parser. Keep it header-only and vendored so disconnected release
 # builds have exactly the same parser as developer builds.
@@ -294,20 +307,16 @@ else()
             TARGET_DIRECTORY aurora_gx
             PROPERTIES COMPILE_OPTIONS "-fno-finite-math-only")
     endif()
-
-    if(MKW_NATIVE_PREBUILT_EXPORT_DIR)
-        include("${CMAKE_CURRENT_LIST_DIR}/cmake/NativePrebuiltExport.cmake")
-    endif()
 endif()
 
 # Everything from here down is this project's own code, so this is where the
 # project definitions enter the directory scope. A directory-scoped definition
 # seeds the COMPILE_DEFINITIONS of targets as they are created, which means:
 #   * aurora-main and its third-party tree were added above -> unaffected;
-#   * mkw_runtime_common, the translated shard libraries and the products in
-#     cmake/PublicProducts.cmake are all created below (that file is include()d,
-#     not add_subdirectory()d, so it shares this directory scope) -> they pick
-#     the definitions up exactly as before.
+#   * the runtime library, the translated shard libraries and the products in
+#     cmake/RuntimeLibrary.cmake and cmake/PublicProducts.cmake are all created
+#     below (those files are include()d, not add_subdirectory()d, so they share
+#     this directory scope) -> they pick the definitions up exactly as before.
 if(MKW_PROJECT_COMPILE_DEFINITIONS)
     add_compile_definitions(${MKW_PROJECT_COMPILE_DEFINITIONS})
 endif()
@@ -336,137 +345,25 @@ set(MKW_BASE_PRODUCT_SOURCE "${CMAKE_CURRENT_LIST_DIR}/src/product/base_product.
 set(MKW_RETRO_REWIND_PRODUCT_SOURCE "${CMAKE_CURRENT_LIST_DIR}/src/product/retro_rewind_product.cpp")
 # The host ISA guard is the one translation unit that must not receive the
 # x86-64-v3 flags, so it gets its own object library instead of riding along in
-# mkw_runtime_common. See cmake/PublicProducts.cmake and the file's own header.
+# the runtime library. See cmake/PublicProducts.cmake and the file's own header.
 set(MKW_CPU_BASELINE_SOURCE "${CMAKE_CURRENT_LIST_DIR}/src/host_cpu_baseline.cpp")
 list(REMOVE_ITEM SOURCES ${MKW_BASE_PRODUCT_SOURCE} ${MKW_RETRO_REWIND_PRODUCT_SOURCE}
      ${MKW_CPU_BASELINE_SOURCE} ${MKW_PLATFORM_SOURCE})
 
-# This deliberately small library contains host services that are safe to
-# validate before guest memory and fiber work makes a full runtime build viable.
-add_library(mkw_platform STATIC "${MKW_PLATFORM_SOURCE}")
-target_include_directories(mkw_platform PUBLIC "${CMAKE_CURRENT_LIST_DIR}/include")
-target_compile_features(mkw_platform PUBLIC cxx_std_17)
-set_target_properties(mkw_platform PROPERTIES UNITY_BUILD OFF)
+include("${CMAKE_CURRENT_LIST_DIR}/cmake/PlatformTests.cmake")
 
-# Keep these independent from Aurora's BUILD_TESTING option: they validate the
-# project's host-platform contracts, not Aurora's third-party test suite.
-enable_testing()
-if(MKW_BUILD_PSQ_TESTS)
-    add_executable(mkw_psq_helpers_tests
-        "${CMAKE_CURRENT_LIST_DIR}/tests/psq_helpers_tests.cpp"
-        "${CMAKE_CURRENT_LIST_DIR}/src/ppc_quantized.cpp")
-    target_include_directories(mkw_psq_helpers_tests PRIVATE
-        "${CMAKE_CURRENT_LIST_DIR}/tests/psq_memory"
-        "${CMAKE_CURRENT_LIST_DIR}/include/isa"
-        "${CMAKE_CURRENT_LIST_DIR}/include")
-    target_compile_features(mkw_psq_helpers_tests PRIVATE cxx_std_17)
-    target_compile_options(mkw_psq_helpers_tests PRIVATE
-        -O2 ${MKW_TRANSLATED_PPC_FP_OPTIONS} -fno-slp-vectorize)
-    if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
-        target_compile_options(mkw_psq_helpers_tests PRIVATE -march=x86-64-v3)
+set(MKW_RUNTIME_SOURCE_DIR "${CMAKE_CURRENT_LIST_DIR}")
+include("${CMAKE_CURRENT_LIST_DIR}/cmake/RuntimeLibrary.cmake")
+
+# The prebuilt export also carries the game-independent runtime as a static
+# archive (x86_64 only; AArch64 compiles it locally with -mcpu=native). It is
+# defined here, after the project definitions, so its flags match the products'.
+if(MKW_NATIVE_PREBUILT_EXPORT_DIR AND NOT MKW_NATIVE_PREBUILT_DIR)
+    if(MKW_HOST_IS_X86_64)
+        mkw_add_runtime_library(mkw_runtime STATIC ${SOURCES})
+        set_target_properties(mkw_runtime PROPERTIES EXCLUDE_FROM_ALL ON)
     endif()
-    set_target_properties(mkw_psq_helpers_tests PROPERTIES UNITY_BUILD OFF)
-    add_test(NAME mkw_psq_helpers_tests COMMAND mkw_psq_helpers_tests)
-    add_test(NAME mkw_psq_reserved_tests COMMAND "${CMAKE_COMMAND}"
-        "-DPSQ_TEST_EXECUTABLE=$<TARGET_FILE:mkw_psq_helpers_tests>"
-        -P "${CMAKE_CURRENT_LIST_DIR}/tests/psq_reserved_tests.cmake")
-endif()
-
-add_executable(mkw_platform_paths_tests "${CMAKE_CURRENT_LIST_DIR}/tests/platform_paths_tests.cpp")
-target_link_libraries(mkw_platform_paths_tests PRIVATE mkw_platform)
-target_compile_features(mkw_platform_paths_tests PRIVATE cxx_std_17)
-add_test(NAME mkw_platform_paths_tests COMMAND mkw_platform_paths_tests)
-
-add_executable(mkw_runtime_config_tests "${CMAKE_CURRENT_LIST_DIR}/tests/runtime_config_tests.cpp")
-target_include_directories(mkw_runtime_config_tests PRIVATE
-    "${CMAKE_CURRENT_LIST_DIR}/include"
-    "${CMAKE_CURRENT_LIST_DIR}/third_party/toml11")
-target_compile_features(mkw_runtime_config_tests PRIVATE cxx_std_20)
-add_test(NAME mkw_runtime_config_tests COMMAND mkw_runtime_config_tests)
-
-add_executable(mkw_nand_save_tests "${CMAKE_CURRENT_LIST_DIR}/tests/nand_save_tests.cpp")
-target_include_directories(mkw_nand_save_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-target_compile_features(mkw_nand_save_tests PRIVATE cxx_std_17)
-add_test(NAME mkw_nand_save_tests COMMAND mkw_nand_save_tests)
-
-add_executable(mkw_nand_settings_tests "${CMAKE_CURRENT_LIST_DIR}/tests/nand_settings_tests.cpp")
-find_package(Threads REQUIRED)
-target_link_libraries(mkw_nand_settings_tests PRIVATE Threads::Threads)
-target_include_directories(mkw_nand_settings_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-target_compile_features(mkw_nand_settings_tests PRIVATE cxx_std_17)
-add_test(NAME mkw_nand_settings_tests COMMAND mkw_nand_settings_tests)
-
-add_executable(mkw_sc_serial_tests "${CMAKE_CURRENT_LIST_DIR}/tests/sc_serial_tests.cpp")
-target_include_directories(mkw_sc_serial_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-target_compile_features(mkw_sc_serial_tests PRIVATE cxx_std_17)
-add_test(NAME mkw_sc_serial_tests COMMAND mkw_sc_serial_tests)
-
-# The input expression engine is self-contained, so it can be exercised without
-# linking the runtime or SDL.
-add_executable(mkw_input_expr_tests
-    "${CMAKE_CURRENT_LIST_DIR}/tests/test_expr.cpp"
-    "${CMAKE_CURRENT_LIST_DIR}/src/input_expr.cpp")
-target_include_directories(mkw_input_expr_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-target_compile_features(mkw_input_expr_tests PRIVATE cxx_std_17)
-add_test(NAME mkw_input_expr_tests COMMAND mkw_input_expr_tests)
-
-# HostContext deliberately keeps the platform-specific context primitive out
-# of fiber_manager.cpp. Exercise the Linux libco handoff directly so future
-# refactors cannot silently remove its headers, implementation, or link edge.
-if(MKW_PLATFORM_LINUX)
-    add_executable(mkw_linux_host_context_tests
-        "${CMAKE_CURRENT_LIST_DIR}/tests/host_context_tests.cpp"
-        "${CMAKE_CURRENT_LIST_DIR}/src/host_context.cpp")
-    target_include_directories(mkw_linux_host_context_tests PRIVATE
-        "${CMAKE_CURRENT_LIST_DIR}/include"
-        "${CMAKE_CURRENT_LIST_DIR}/third_party/libco")
-    target_compile_features(mkw_linux_host_context_tests PRIVATE cxx_std_17)
-    target_link_libraries(mkw_linux_host_context_tests PRIVATE mkw::libco)
-    add_test(NAME mkw_linux_host_context_tests COMMAND mkw_linux_host_context_tests)
-endif()
-
-if(MKW_PLATFORM_MACOS)
-    # Exercise the public host-memory contracts separately from translated products.
-    if(MKW_PLATFORM_MACOS_ARM64)
-        # Apple Silicon's context ABI is implemented by the local assembly backend.
-        enable_language(ASM)
-        add_executable(mkw_macos_context_abi_tests
-            "${CMAKE_CURRENT_LIST_DIR}/tests/macos_context_abi_tests.cpp"
-            "${CMAKE_CURRENT_LIST_DIR}/src/platform/macos/co_switch.S")
-        target_compile_features(mkw_macos_context_abi_tests PRIVATE cxx_std_17)
-        add_test(NAME mkw_macos_context_abi_tests COMMAND mkw_macos_context_abi_tests)
-
-        add_executable(mkw_macos_host_context_tests
-            "${CMAKE_CURRENT_LIST_DIR}/tests/host_context_tests.cpp"
-            "${CMAKE_CURRENT_LIST_DIR}/src/host_context.cpp"
-            "${CMAKE_CURRENT_LIST_DIR}/src/platform/macos/co_switch.S")
-        target_include_directories(mkw_macos_host_context_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-    else()
-        # Intel macOS follows the same System V AMD64 libco path as Linux.
-        add_executable(mkw_macos_host_context_tests
-            "${CMAKE_CURRENT_LIST_DIR}/tests/host_context_tests.cpp"
-            "${CMAKE_CURRENT_LIST_DIR}/src/host_context.cpp")
-        target_include_directories(mkw_macos_host_context_tests PRIVATE
-            "${CMAKE_CURRENT_LIST_DIR}/include"
-            "${CMAKE_CURRENT_LIST_DIR}/third_party/libco")
-        target_link_libraries(mkw_macos_host_context_tests PRIVATE mkw::libco)
-    endif()
-    target_compile_features(mkw_macos_host_context_tests PRIVATE cxx_std_17)
-    add_test(NAME mkw_macos_host_context_tests COMMAND mkw_macos_host_context_tests)
-
-    add_executable(mkw_macos_guest_flat_memory_tests
-        "${CMAKE_CURRENT_LIST_DIR}/tests/macos_guest_flat_memory_tests.cpp"
-        "${CMAKE_CURRENT_LIST_DIR}/src/guest_flat_memory_macos.cpp")
-    target_include_directories(mkw_macos_guest_flat_memory_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-    target_compile_features(mkw_macos_guest_flat_memory_tests PRIVATE cxx_std_17)
-    add_test(NAME mkw_macos_guest_flat_memory_tests COMMAND mkw_macos_guest_flat_memory_tests)
-
-    add_executable(mkw_macos_external_audio_tests
-        "${CMAKE_CURRENT_LIST_DIR}/tests/macos_external_audio_tests.cpp"
-        "${CMAKE_CURRENT_LIST_DIR}/src/external_audio_macos.cpp")
-    target_include_directories(mkw_macos_external_audio_tests PRIVATE "${CMAKE_CURRENT_LIST_DIR}/include")
-    target_compile_features(mkw_macos_external_audio_tests PRIVATE cxx_std_17)
-    add_test(NAME mkw_macos_external_audio_tests COMMAND mkw_macos_external_audio_tests)
+    include("${CMAKE_CURRENT_LIST_DIR}/cmake/NativePrebuiltExport.cmake")
 endif()
 
 # The translator emits the complete, content-addressed source graph. Consuming
@@ -476,8 +373,8 @@ if(MKW_BUILD_PRODUCTS)
     set(MKW_TRANSLATED_SHARD_MANIFEST
         "${CMAKE_CURRENT_LIST_DIR}/../generated/build_shards/shards.cmake"
         CACHE FILEPATH "Translator-owned aggregate shard manifest")
-    # The prebuilt export only needs the aurora/third-party closure configured above, so a
-    # packaging machine without a translation stops here instead of failing.
+    # The prebuilt export only needs the aurora/third-party closure and the runtime
+    # configured above, so a packaging machine without a translation stops here.
     if(MKW_NATIVE_PREBUILT_EXPORT_DIR AND NOT EXISTS "${MKW_TRANSLATED_SHARD_MANIFEST}")
         message(STATUS "No translator shard manifest; configuring the native prebuilt export only")
         return()
@@ -494,20 +391,13 @@ if(MKW_BUILD_PRODUCTS)
         "${MKW_PROFILE_SENSITIVE_CALLER_COUNT} profile-sensitive callers; "
         "${MKW_RETRO_REWIND_FUNCTION_COUNT} Retro Rewind functions")
 
-    set(MKW_RUNTIME_SOURCE_DIR "${CMAKE_CURRENT_LIST_DIR}")
     include("${CMAKE_CURRENT_LIST_DIR}/cmake/PublicProducts.cmake")
 else()
     if(MKW_PLATFORM_MACOS)
         # Compile-only audit of native runtime sources. It deliberately avoids
         # translated products until their host dependencies are portable.
-        # These sources depend on generated/RuntimeConfig.h, which is emitted for
-        # a particular game by the translator and is intentionally unavailable
-        # in this platform-only configuration.
-        set(MKW_MACOS_NATIVE_AUDIT_SOURCES ${SOURCES})
-        list(REMOVE_ITEM MKW_MACOS_NATIVE_AUDIT_SOURCES
-            "${CMAKE_CURRENT_LIST_DIR}/src/abi_bridge.cpp"
-            "${CMAKE_CURRENT_LIST_DIR}/src/hle/os/os_alarm.cpp")
-        add_library(mkw_macos_native_compile OBJECT ${MKW_MACOS_NATIVE_AUDIT_SOURCES})
+        # runtime/src is game-independent; per-game glue lives in runtime/game.
+        add_library(mkw_macos_native_compile OBJECT ${SOURCES})
         target_include_directories(mkw_macos_native_compile PRIVATE
             "${CMAKE_CURRENT_LIST_DIR}/include" "${CMAKE_CURRENT_LIST_DIR}/src"
             "${CMAKE_CURRENT_LIST_DIR}/.." "${CMAKE_CURRENT_LIST_DIR}/../aurora-main/include")
@@ -518,10 +408,11 @@ else()
             mkw::pugixml mkw::toml11 mkw::cryptopp mkw::mbedtls)
         if(MKW_PLATFORM_MACOS_X86_64)
             target_link_libraries(mkw_macos_native_compile PRIVATE mkw::libco)
-            # Keep this compile-only audit on the same Haswell-era x86-64-v3
-            # baseline as the translated product. PPC paired FMA helpers use
-            # FMA intrinsics and intentionally cannot compile for plain x86-64.
-            target_compile_options(mkw_macos_native_compile PRIVATE -march=x86-64-v3)
+            target_compile_options(mkw_macos_native_compile PRIVATE
+                -march=x86-64-${MKW_X86_CPU_PROFILE})
+            if(MKW_X86_CPU_PROFILE STREQUAL "v2")
+                target_compile_definitions(mkw_macos_native_compile PRIVATE MKW_X86_CPU_PROFILE_V2=1)
+            endif()
         endif()
         set_target_properties(mkw_macos_native_compile PROPERTIES UNITY_BUILD OFF)
     endif()

runtime/cmake/HostLibraries.cmake

@@ -0,0 +1,29 @@
+# Shared by the product build and the dependency-free platform test configuration.
+get_filename_component(MKW_HOST_RUNTIME_DIR "${CMAKE_CURRENT_LIST_DIR}/.." ABSOLUTE)
+
+# POSIX guest-fiber scheduling (runtime/src/host_context.cpp) needs a symmetric
+# stackful-coroutine primitive to stand in for Win32 Fibers. libco's co_switch() transfers
+# directly to any other created coroutine, matching SwitchToFiber's semantics exactly (unlike
+# asymmetric resume/yield coroutine libraries, which would need every call site restructured).
+# Vendored from upstream (higan-emu/libco @ e18e09d, 2019-10-16, ISC license; valgrind.h is
+# separately BSD-style licensed, see third_party/libco/LICENSE). Windows keeps native Fibers;
+# Apple Silicon uses the project's x18-safe AArch64 assembly backend, while Intel macOS uses
+# libco's existing System V AMD64 backend.
+if(MKW_PLATFORM_LINUX OR MKW_PLATFORM_MACOS_X86_64)
+    add_library(mkw_libco STATIC "${MKW_HOST_RUNTIME_DIR}/third_party/libco/libco.c")
+    add_library(mkw::libco ALIAS mkw_libco)
+    target_include_directories(mkw_libco PUBLIC "${MKW_HOST_RUNTIME_DIR}/third_party/libco")
+    set_target_properties(mkw_libco PROPERTIES UNITY_BUILD OFF)
+endif()
+
+# This deliberately small library contains host services that are safe to
+# validate before guest memory and fiber work makes a full runtime build viable.
+add_library(mkw_platform STATIC "${MKW_HOST_RUNTIME_DIR}/src/platform/host_platform.cpp")
+target_include_directories(mkw_platform PUBLIC "${MKW_HOST_RUNTIME_DIR}/include")
+target_compile_features(mkw_platform PUBLIC cxx_std_17)
+set_target_properties(mkw_platform PROPERTIES UNITY_BUILD OFF)
+
+if(MKW_PLATFORM_WINDOWS)
+    target_compile_definitions(mkw_platform PRIVATE NOMINMAX)
+    target_link_libraries(mkw_platform PUBLIC shell32 ole32 uuid)
+endif()

runtime/cmake/NativePrebuilt.cmake

@@ -3,10 +3,10 @@
 #
 # Produced by launcher/Prepare-NativePrebuilt.ps1 with the same toolchain and the
 # same canonical flag set the consuming build uses, which is what makes a single
-# package valid for every user. Nothing about this project's own code changes:
-# the runtime, the translated shards and the products still compile locally and
-# still link targets named aurora::gx / aurora::pad / aurora::si / aurora::vi /
-# aurora::mtx, so cmake/PublicProducts.cmake needs no prebuilt-specific branch.
+# package valid for every user. The translated shards and the products still
+# compile locally and still link targets named aurora::gx / aurora::pad /
+# aurora::si / aurora::vi / aurora::mtx. The package may also carry the native
+# runtime archive; cmake/RuntimeLibrary.cmake decides whether it can be used.
 #
 # The aurora and dependency source trees are still present in the workspace - the
 # runtime includes aurora headers directly - so the package records include
@@ -42,6 +42,10 @@ endfunction()
 mkw_np_resolve(_mkw_np_includes ${MKW_NP_INCLUDE_DIRECTORIES})
 mkw_np_resolve(_mkw_np_links ${MKW_NP_LINK_LIBRARIES})
 mkw_np_resolve(_mkw_np_dawn_config ${MKW_NP_DAWN_CONFIG_DIR})
+set(MKW_NP_RUNTIME_ARCHIVE_PATH "")
+if(MKW_NP_RUNTIME_ARCHIVE)
+    mkw_np_resolve(MKW_NP_RUNTIME_ARCHIVE_PATH "${MKW_NP_RUNTIME_ARCHIVE}")
+endif()
 if(NOT MKW_PLATFORM_WINDOWS)
     if(NOT MKW_NP_MBEDTLS_LIBRARIES OR NOT MKW_NP_MBEDTLS_INCLUDE_DIR)
         message(FATAL_ERROR

runtime/cmake/NativePrebuiltExport.cmake

@@ -87,6 +87,11 @@ mkw_collect_buildsystem_targets("${_mkw_np_aurora_dir}" _mkw_np_all_targets)
 # the scan above cannot see it; it is appended explicitly. The type and closure
 # checks below still apply to it.
 list(APPEND _mkw_np_all_targets "mkw_cryptopp")
+# The game-independent runtime archive (RuntimeLibrary.cmake). It is outside the
+# probe closure, so it ships only when the harvester builds it explicitly.
+if(TARGET mkw_runtime)
+    list(APPEND _mkw_np_all_targets "mkw_runtime")
+endif()
 if(NOT MKW_PLATFORM_WINDOWS)
     set(_mkw_np_mbedtls_targets mbedtls mbedx509 mbedcrypto)
     # mbedcrypto links these third-party archives when upstream defines them.
@@ -168,5 +173,12 @@ file(WRITE "${MKW_NATIVE_PREBUILT_EXPORT_DIR}/meta.txt"
     "cxx_compiler=${CMAKE_CXX_COMPILER}\n"
     "cxx_compiler_version=${CMAKE_CXX_COMPILER_VERSION}\n"
     "library_targets=${_mkw_np_lib_targets_csv}\n")
+if(TARGET mkw_runtime)
+    mkw_runtime_fingerprint(_mkw_np_runtime_fingerprint "${CMAKE_CURRENT_SOURCE_DIR}"
+        "${MKW_CPPWINRT_INCLUDE_DIR}" "${MKW_X86_CPU_PROFILE}")
+    file(APPEND "${MKW_NATIVE_PREBUILT_EXPORT_DIR}/meta.txt"
+        "runtime_fingerprint=${_mkw_np_runtime_fingerprint}\n"
+        "runtime_cpu_profile=${MKW_X86_CPU_PROFILE}\n")
+endif()
 
 message(STATUS "Native prebuilt export: ${MKW_NATIVE_PREBUILT_EXPORT_DIR}")

runtime/cmake/PlatformTests.cmake

@@ -0,0 +1,188 @@
+get_filename_component(MKW_TEST_RUNTIME_DIR "${CMAKE_CURRENT_LIST_DIR}/.." ABSOLUTE)
+
+# Keep these independent from Aurora's BUILD_TESTING option: they validate the
+# project's host-platform contracts, not Aurora's third-party test suite.
+enable_testing()
+if(MKW_BUILD_PSQ_TESTS)
+    add_executable(mkw_psq_helpers_tests
+        "${MKW_TEST_RUNTIME_DIR}/tests/psq_helpers_tests.cpp"
+        "${MKW_TEST_RUNTIME_DIR}/src/ppc_quantized.cpp")
+    target_include_directories(mkw_psq_helpers_tests PRIVATE
+        "${MKW_TEST_RUNTIME_DIR}/tests/psq_memory"
+        "${MKW_TEST_RUNTIME_DIR}/include/isa"
+        "${MKW_TEST_RUNTIME_DIR}/include")
+    target_compile_features(mkw_psq_helpers_tests PRIVATE cxx_std_17)
+    target_compile_options(mkw_psq_helpers_tests PRIVATE
+        -O2 -fno-fast-math -ffp-contract=off -fno-slp-vectorize)
+    if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
+        target_compile_options(mkw_psq_helpers_tests PRIVATE
+            -march=x86-64-${MKW_X86_CPU_PROFILE})
+    endif()
+    set_target_properties(mkw_psq_helpers_tests PROPERTIES UNITY_BUILD OFF)
+    add_test(NAME mkw_psq_helpers_tests COMMAND mkw_psq_helpers_tests)
+    add_test(NAME mkw_psq_reserved_tests COMMAND "${CMAKE_COMMAND}"
+        "-DPSQ_TEST_EXECUTABLE=$<TARGET_FILE:mkw_psq_helpers_tests>"
+        -P "${MKW_TEST_RUNTIME_DIR}/tests/psq_reserved_tests.cmake")
+endif()
+
+add_executable(mkw_platform_paths_tests "${MKW_TEST_RUNTIME_DIR}/tests/platform_paths_tests.cpp")
+target_link_libraries(mkw_platform_paths_tests PRIVATE mkw_platform)
+target_compile_features(mkw_platform_paths_tests PRIVATE cxx_std_17)
+add_test(NAME mkw_platform_paths_tests COMMAND mkw_platform_paths_tests)
+
+add_executable(mkw_runtime_config_tests "${MKW_TEST_RUNTIME_DIR}/tests/runtime_config_tests.cpp")
+target_include_directories(mkw_runtime_config_tests PRIVATE
+    "${MKW_TEST_RUNTIME_DIR}/include"
+    "${MKW_TEST_RUNTIME_DIR}/third_party/toml11")
+target_compile_features(mkw_runtime_config_tests PRIVATE cxx_std_20)
+add_test(NAME mkw_runtime_config_tests COMMAND mkw_runtime_config_tests)
+
+add_executable(mkw_nand_save_tests "${MKW_TEST_RUNTIME_DIR}/tests/nand_save_tests.cpp")
+target_include_directories(mkw_nand_save_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+target_compile_features(mkw_nand_save_tests PRIVATE cxx_std_17)
+add_test(NAME mkw_nand_save_tests COMMAND mkw_nand_save_tests)
+
+add_executable(mkw_nand_settings_tests "${MKW_TEST_RUNTIME_DIR}/tests/nand_settings_tests.cpp")
+find_package(Threads REQUIRED)
+target_link_libraries(mkw_nand_settings_tests PRIVATE Threads::Threads)
+target_include_directories(mkw_nand_settings_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+target_compile_features(mkw_nand_settings_tests PRIVATE cxx_std_17)
+add_test(NAME mkw_nand_settings_tests COMMAND mkw_nand_settings_tests)
+
+add_executable(mkw_sc_serial_tests "${MKW_TEST_RUNTIME_DIR}/tests/sc_serial_tests.cpp")
+target_include_directories(mkw_sc_serial_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+target_compile_features(mkw_sc_serial_tests PRIVATE cxx_std_17)
+add_test(NAME mkw_sc_serial_tests COMMAND mkw_sc_serial_tests)
+
+# The input expression engine is self-contained, so it can be exercised without
+# linking the runtime or SDL.
+add_executable(mkw_input_expr_tests
+    "${MKW_TEST_RUNTIME_DIR}/tests/test_expr.cpp"
+    "${MKW_TEST_RUNTIME_DIR}/src/input_expr.cpp")
+target_include_directories(mkw_input_expr_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+target_compile_features(mkw_input_expr_tests PRIVATE cxx_std_17)
+add_test(NAME mkw_input_expr_tests COMMAND mkw_input_expr_tests)
+
+# HostContext deliberately keeps the platform-specific context primitive out
+# of fiber_manager.cpp. Exercise the Linux libco handoff directly so future
+# refactors cannot silently remove its headers, implementation, or link edge.
+if(MKW_PLATFORM_LINUX)
+    add_executable(mkw_linux_host_context_tests
+        "${MKW_TEST_RUNTIME_DIR}/tests/host_context_tests.cpp"
+        "${MKW_TEST_RUNTIME_DIR}/src/host_context.cpp")
+    target_include_directories(mkw_linux_host_context_tests PRIVATE
+        "${MKW_TEST_RUNTIME_DIR}/include"
+        "${MKW_TEST_RUNTIME_DIR}/third_party/libco")
+    target_compile_features(mkw_linux_host_context_tests PRIVATE cxx_std_17)
+    target_link_libraries(mkw_linux_host_context_tests PRIVATE mkw::libco)
+    add_test(NAME mkw_linux_host_context_tests COMMAND mkw_linux_host_context_tests)
+endif()
+
+if(MKW_PLATFORM_WINDOWS)
+    add_executable(mkw_windows_host_context_tests
+        "${MKW_TEST_RUNTIME_DIR}/tests/host_context_tests.cpp"
+        "${MKW_TEST_RUNTIME_DIR}/src/host_context.cpp")
+    target_include_directories(mkw_windows_host_context_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+    target_compile_features(mkw_windows_host_context_tests PRIVATE cxx_std_17)
+    add_test(NAME mkw_windows_host_context_tests COMMAND mkw_windows_host_context_tests)
+endif()
+
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
+    # Probe at the base ISA, including OS support for AVX register state. Cross
+    # builds execute target code only through a configured emulator.
+    if(NOT CMAKE_CROSSCOMPILING OR CMAKE_CROSSCOMPILING_EMULATOR)
+        include(CheckCXXSourceRuns)
+        include(CMakePushCheckState)
+        cmake_push_check_state(RESET)
+        set(CMAKE_REQUIRED_FLAGS "-march=x86-64")
+        check_cxx_source_runs([=[
+            #include <cpuid.h>
+            int main() {
+                unsigned a, b, c, d;
+                if (__get_cpuid_max(0, nullptr) < 7 ||
+                    __get_cpuid_max(0x80000000u, nullptr) < 0x80000001u)
+                    return 1;
+                __cpuid_count(1, 0, a, b, c, d);
+                // v2 plus FMA, MOVBE, XSAVE, OSXSAVE, AVX and F16C.
+                const unsigned leaf1 = (1u << 0) | (1u << 9) | (1u << 12) |
+                    (1u << 13) | (1u << 19) | (1u << 20) | (1u << 22) |
+                    (1u << 23) | (1u << 26) | (1u << 27) | (1u << 28) | (1u << 29);
+                if ((c & leaf1) != leaf1) return 1;
+                __cpuid_count(7, 0, a, b, c, d);
+                const unsigned leaf7 = (1u << 3) | (1u << 5) | (1u << 8);
+                if ((b & leaf7) != leaf7) return 1;
+                __cpuid_count(0x80000001u, 0, a, b, c, d);
+                if ((c & 0x21u) != 0x21u) return 1; // LAHF-SAHF and LZCNT.
+                __asm__ __volatile__("xgetbv" : "=a"(a), "=d"(d) : "c"(0));
+                return (a & 0x6u) == 0x6u ? 0 : 1;
+            }
+        ]=] MKW_HOST_SUPPORTS_X86_V3)
+        cmake_pop_check_state()
+    endif()
+    # Compile the same oracle cases for both paths. The v2 binary proves the
+    # scalar fallback stays free of FMA; the v3 binary checks the existing
+    # vector intrinsic path against the same strict result.
+    foreach(profile IN ITEMS v2 v3)
+        add_executable(mkw_ppc_pair_fma_${profile}_tests
+            "${MKW_TEST_RUNTIME_DIR}/tests/ppc_pair_fma_tests.cpp")
+        target_include_directories(mkw_ppc_pair_fma_${profile}_tests PRIVATE
+            "${MKW_TEST_RUNTIME_DIR}/include")
+        target_compile_features(mkw_ppc_pair_fma_${profile}_tests PRIVATE cxx_std_17)
+        target_compile_options(mkw_ppc_pair_fma_${profile}_tests PRIVATE
+            -march=x86-64-${profile} -fno-fast-math -ffp-contract=off)
+        if((NOT CMAKE_CROSSCOMPILING OR CMAKE_CROSSCOMPILING_EMULATOR) AND
+           (profile STREQUAL "v2" OR MKW_HOST_SUPPORTS_X86_V3))
+            add_test(NAME mkw_ppc_pair_fma_${profile}_tests COMMAND mkw_ppc_pair_fma_${profile}_tests)
+        endif()
+    endforeach()
+
+    add_library(mkw_cpu_baseline_v2_compile OBJECT
+        "${MKW_TEST_RUNTIME_DIR}/src/host_cpu_baseline.cpp")
+    target_compile_features(mkw_cpu_baseline_v2_compile PRIVATE cxx_std_17)
+    target_compile_definitions(mkw_cpu_baseline_v2_compile PRIVATE MKW_X86_CPU_PROFILE_V2=1)
+    target_compile_options(mkw_cpu_baseline_v2_compile PRIVATE -march=x86-64-v2 -w)
+endif()
+
+if(MKW_PLATFORM_MACOS)
+    # Exercise the public host-memory contracts separately from translated products.
+    if(MKW_PLATFORM_MACOS_ARM64)
+        # Apple Silicon's context ABI is implemented by the local assembly backend.
+        enable_language(ASM)
+        add_executable(mkw_macos_context_abi_tests
+            "${MKW_TEST_RUNTIME_DIR}/tests/macos_context_abi_tests.cpp"
+            "${MKW_TEST_RUNTIME_DIR}/src/platform/macos/co_switch.S")
+        target_compile_features(mkw_macos_context_abi_tests PRIVATE cxx_std_17)
+        add_test(NAME mkw_macos_context_abi_tests COMMAND mkw_macos_context_abi_tests)
+
+        add_executable(mkw_macos_host_context_tests
+            "${MKW_TEST_RUNTIME_DIR}/tests/host_context_tests.cpp"
+            "${MKW_TEST_RUNTIME_DIR}/src/host_context.cpp"
+            "${MKW_TEST_RUNTIME_DIR}/src/platform/macos/co_switch.S")
+        target_include_directories(mkw_macos_host_context_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+    else()
+        # Intel macOS follows the same System V AMD64 libco path as Linux.
+        add_executable(mkw_macos_host_context_tests
+            "${MKW_TEST_RUNTIME_DIR}/tests/host_context_tests.cpp"
+            "${MKW_TEST_RUNTIME_DIR}/src/host_context.cpp")
+        target_include_directories(mkw_macos_host_context_tests PRIVATE
+            "${MKW_TEST_RUNTIME_DIR}/include"
+            "${MKW_TEST_RUNTIME_DIR}/third_party/libco")
+        target_link_libraries(mkw_macos_host_context_tests PRIVATE mkw::libco)
+    endif()
+    target_compile_features(mkw_macos_host_context_tests PRIVATE cxx_std_17)
+    add_test(NAME mkw_macos_host_context_tests COMMAND mkw_macos_host_context_tests)
+
+    add_executable(mkw_macos_guest_flat_memory_tests
+        "${MKW_TEST_RUNTIME_DIR}/tests/macos_guest_flat_memory_tests.cpp"
+        "${MKW_TEST_RUNTIME_DIR}/src/guest_flat_memory_macos.cpp")
+    target_include_directories(mkw_macos_guest_flat_memory_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+    target_compile_features(mkw_macos_guest_flat_memory_tests PRIVATE cxx_std_17)
+    add_test(NAME mkw_macos_guest_flat_memory_tests COMMAND mkw_macos_guest_flat_memory_tests)
+
+    add_executable(mkw_macos_external_audio_tests
+        "${MKW_TEST_RUNTIME_DIR}/tests/macos_external_audio_tests.cpp"
+        "${MKW_TEST_RUNTIME_DIR}/src/external_audio_macos.cpp")
+    target_include_directories(mkw_macos_external_audio_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+    target_compile_features(mkw_macos_external_audio_tests PRIVATE cxx_std_17)
+    add_test(NAME mkw_macos_external_audio_tests COMMAND mkw_macos_external_audio_tests)
+endif()

runtime/cmake/PublicProducts.cmake

@@ -4,55 +4,42 @@
 # functions are compiled once into mkw_base_shared; only callers whose direct
 # ABI differs between profiles receive small base/RR variants.
 
+# Per-game data the runtime reads but does not own. Kept out of the runtime
+# library so that library stays game-independent (and can be precompiled).
+set(MKW_GAME_DATA_SOURCES "")
 set(DATA_INIT_FILE "${MKW_RUNTIME_SOURCE_DIR}/../generated/data_sections_init.cpp")
 set(DATA_INIT_BLOB_ASM "${MKW_RUNTIME_SOURCE_DIR}/../generated/data_sections_init_blobs.S")
 if(EXISTS "${DATA_INIT_FILE}")
-    list(APPEND SOURCES "${DATA_INIT_FILE}")
+    list(APPEND MKW_GAME_DATA_SOURCES "${DATA_INIT_FILE}")
 endif()
 # Crash-report symbolization table emitted by generate-data-init. The stub
 # (deliberately outside the globbed src/ tree so it is never picked up twice)
 # keeps link succeeding when the generated table has not been produced yet.
 set(GUEST_SYMBOL_TABLE_FILE "${MKW_RUNTIME_SOURCE_DIR}/../generated/guest_symbol_table.cpp")
 if(EXISTS "${GUEST_SYMBOL_TABLE_FILE}")
-    list(APPEND SOURCES "${GUEST_SYMBOL_TABLE_FILE}")
+    list(APPEND MKW_GAME_DATA_SOURCES "${GUEST_SYMBOL_TABLE_FILE}")
 else()
-    list(APPEND SOURCES "${MKW_RUNTIME_SOURCE_DIR}/cmake/guest_symbol_table_stub.cpp")
+    list(APPEND MKW_GAME_DATA_SOURCES "${MKW_RUNTIME_SOURCE_DIR}/cmake/guest_symbol_table_stub.cpp")
 endif()
 if(EXISTS "${DATA_INIT_BLOB_ASM}")
     enable_language(ASM)
     set_source_files_properties("${DATA_INIT_BLOB_ASM}" PROPERTIES LANGUAGE ASM SKIP_UNITY_BUILD_INCLUSION ON)
-    list(APPEND SOURCES "${DATA_INIT_BLOB_ASM}")
+    list(APPEND MKW_GAME_DATA_SOURCES "${DATA_INIT_BLOB_ASM}")
 endif()
-list(REMOVE_DUPLICATES SOURCES)
+# Runtime glue that depends on generated/ headers (e.g. RuntimeConfig.h).
+file(GLOB MKW_GAME_GLUE_SOURCES CONFIGURE_DEPENDS "${MKW_RUNTIME_SOURCE_DIR}/game/*.cpp")
+list(APPEND MKW_GAME_DATA_SOURCES ${MKW_GAME_GLUE_SOURCES})
 
 if(MKW_PLATFORM_MACOS)
     find_library(MKW_IOKIT_FRAMEWORK IOKit REQUIRED)
     find_library(MKW_COREFOUNDATION_FRAMEWORK CoreFoundation REQUIRED)
-    find_library(MKW_COREAUDIO_FRAMEWORK CoreAudio REQUIRED)
 endif()
 
-function(mkw_apply_common_compile_options target)
-    target_compile_options(${target} PRIVATE -O3 -ffast-math -w -pipe)
-endfunction()
-
 function(mkw_apply_translated_compile_options target)
     target_compile_options(${target} PRIVATE
         -O2 ${MKW_TRANSLATED_PPC_FP_OPTIONS} -fno-slp-vectorize -w -pipe)
 endfunction()
 
-function(mkw_configure_object_target target)
-    target_include_directories(${target} PRIVATE
-        "${MKW_RUNTIME_SOURCE_DIR}/include"
-        "${MKW_RUNTIME_SOURCE_DIR}/src"
-        # Workspace root, so translator output is spelled "generated/<x>.h"
-        # instead of a ../ chain whose depth depends on the includer.
-        "${MKW_RUNTIME_SOURCE_DIR}/.."
-        "${MKW_RUNTIME_SOURCE_DIR}/../aurora-main/include")
-    target_compile_definitions(${target} PRIVATE
-        TARGET_PC)
-    set_target_properties(${target} PROPERTIES CXX_STANDARD 17 CXX_STANDARD_REQUIRED ON)
-endfunction()
-
 # Translated shard TUs are the memory-hungry compiles; everything else in the build is
 # comparatively small. A dedicated Ninja job pool caps how many of them run at once so the
 # global parallelism can use every core for the cheap TUs without the memory ceiling being a
@@ -74,85 +61,64 @@ function(mkw_configure_translated_target target)
     mkw_bound_translated_compiles(${target})
 endfunction()
 
-add_library(mkw_runtime_common OBJECT ${SOURCES})
-mkw_configure_object_target(mkw_runtime_common)
-target_compile_features(mkw_runtime_common PRIVATE cxx_std_20)
-target_compile_definitions(mkw_runtime_common PRIVATE
+mkw_select_runtime_provider()
+if(MKW_RUNTIME_USE_PREBUILT)
+    mkw_import_prebuilt_runtime(mkw_runtime_prebuilt "${MKW_NP_RUNTIME_ARCHIVE_PATH}")
+    # A replacement archive can carry an older mtime than the last link, so relink on fingerprint change.
+    set(MKW_RUNTIME_PREBUILT_STAMP "${CMAKE_BINARY_DIR}/mkw_runtime_prebuilt.stamp")
+    file(CONFIGURE OUTPUT "${MKW_RUNTIME_PREBUILT_STAMP}" CONTENT "${MKW_NP_RUNTIME_FINGERPRINT}\n")
+else()
+    mkw_add_runtime_library(mkw_runtime_common OBJECT ${SOURCES})
+endif()
+
+add_library(mkw_game_data OBJECT ${MKW_GAME_DATA_SOURCES})
+mkw_configure_object_target(mkw_game_data)
+target_compile_features(mkw_game_data PRIVATE cxx_std_20)
+target_compile_definitions(mkw_game_data PRIVATE
     SDL_MAIN_HANDLED
     _DISABLE_STRING_ANNOTATION _DISABLE_VECTOR_ANNOTATION)
-target_link_libraries(mkw_runtime_common PRIVATE
-    aurora::gx aurora::pad aurora::si aurora::vi aurora::mtx)
-target_link_libraries(mkw_runtime_common PRIVATE mkw_platform mkw::pugixml mkw::toml11 mkw::cryptopp mkw::mbedtls)
-if(MKW_PLATFORM_WINDOWS)
-    target_link_libraries(mkw_runtime_common PRIVATE shell32 windowsapp)
-elseif(MKW_PLATFORM_LINUX)
-    # ${CMAKE_DL_LIBS} for music_attenuation.cpp's dlopen of libdbus-1 (MPRIS
-    # media monitoring).
-    target_link_libraries(mkw_runtime_common PRIVATE mkw::libco ${CMAKE_DL_LIBS})
+if(MKW_SETUP_VERSION)
+    set_property(SOURCE "${MKW_RUNTIME_SOURCE_DIR}/game/setup_version.cpp" APPEND PROPERTY
+        COMPILE_DEFINITIONS "MKW_SETUP_VERSION=\"${MKW_SETUP_VERSION}\"")
 endif()
+mkw_runtime_link_dependencies(mkw_game_data PRIVATE)
+set_target_properties(mkw_game_data PROPERTIES UNITY_BUILD OFF)
+mkw_apply_common_compile_options(mkw_game_data)
 
-if(MKW_PLATFORM_MACOS)
-    # CoreAudio framework is required for automatic music muting on macOS.
-    target_link_libraries(mkw_runtime_common PRIVATE "${MKW_COREAUDIO_FRAMEWORK}")
-    if(MKW_PLATFORM_MACOS_X86_64)
-        # libco is used by Intel macOS. Apple Silicon uses the local
-        # x18-safe assembly backend and therefore does not define mkw::libco.
-        target_link_libraries(mkw_runtime_common PRIVATE mkw::libco)
-    endif()
-endif()
-
-if(MKW_CPPWINRT_INCLUDE_DIR)
-    if(NOT EXISTS "${MKW_CPPWINRT_INCLUDE_DIR}/winrt/base.h")
-        message(FATAL_ERROR
-            "MKW_CPPWINRT_INCLUDE_DIR does not contain winrt/base.h: ${MKW_CPPWINRT_INCLUDE_DIR}")
-    endif()
-    target_include_directories(mkw_runtime_common PRIVATE "${MKW_CPPWINRT_INCLUDE_DIR}")
-endif()
-
-# Keep runtime unity units small and semantically related. The old generated-TU
-# batch size put all 57 native runtime sources into one memory-heavy compiler job.
-foreach(source IN LISTS SOURCES)
-    get_filename_component(source_name "${source}" NAME_WE)
-    string(REPLACE "\\" "/" source_normalized "${source}")
-    if(source_normalized MATCHES "/hle/gx/")
-        set(runtime_group "gx_bridge")
-    elseif(source_name MATCHES "network|socket|dns|dwc|ios")
-        set(runtime_group "network_ios")
-    elseif(source_name MATCHES "os_|system|memory|fiber|scheduler")
-        set(runtime_group "guest_system")
-    elseif(source_name MATCHES "debug|trace|prof")
-        set(runtime_group "diagnostics")
-    else()
-        string(SHA256 source_hash "${source_name}")
-        string(SUBSTRING "${source_hash}" 0 4 source_hash_prefix)
-        math(EXPR runtime_bucket "0x${source_hash_prefix} % 8")
-        set(runtime_group "runtime_${runtime_bucket}")
-    endif()
-    set_source_files_properties("${source}" PROPERTIES UNITY_GROUP "${runtime_group}")
-endforeach()
-# PPC semantic sources are excluded from unity/PCH and configured in CMakeLists.txt.
-set_target_properties(mkw_runtime_common PROPERTIES UNITY_BUILD ON UNITY_BUILD_MODE GROUP)
-target_precompile_headers(mkw_runtime_common PRIVATE "${MKW_RUNTIME_SOURCE_DIR}/include/mkw_pch.h")
-mkw_apply_common_compile_options(mkw_runtime_common)
-
-# Host ISA guard. Windows and Linux x86_64 product targets use x86-64-v3, so
-# this object deliberately keeps the plain baseline ISA and checks the CPU
-# before any AVX2/FMA code can execute. AArch64 has no equivalent optional ISA
-# floor to probe: NEON/FMA are architectural requirements.
-if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
+# Host ISA guard deliberately keeps the plain baseline ISA and checks the
+# selected x86 profile before optional instructions can execute. AArch64 has
+# no equivalent optional ISA floor to probe: NEON/FMA are architectural
+# requirements.
+if(MKW_HOST_IS_X86_64)
     add_library(mkw_cpu_baseline OBJECT "${MKW_CPU_BASELINE_SOURCE}")
     target_compile_features(mkw_cpu_baseline PRIVATE cxx_std_17)
     set_target_properties(mkw_cpu_baseline PROPERTIES UNITY_BUILD OFF)
     target_compile_options(mkw_cpu_baseline PRIVATE -w)
+    if(MKW_X86_CPU_PROFILE STREQUAL "v2")
+        target_compile_definitions(mkw_cpu_baseline PRIVATE MKW_X86_CPU_PROFILE_V2=1)
+    endif()
 endif()
 
 if(NOT MKW_BASE_COMMON_SHARDS)
     message(FATAL_ERROR "Translator build graph contains no shared base shards")
 endif()
 
+# Translated code compiles against the guest ABI surface only, so runtime
+# implementation edits do not invalidate its PCH.
+set(MKW_TRANSLATED_PCH "${MKW_RUNTIME_SOURCE_DIR}/include/guest_abi/guest_abi_pch.h")
+if(NOT EXISTS "${MKW_TRANSLATED_PCH}")
+    set(MKW_TRANSLATED_PCH "${MKW_RUNTIME_SOURCE_DIR}/include/mkw_pch.h")
+endif()
+
 add_library(mkw_base_shared STATIC ${MKW_BASE_COMMON_SHARDS})
 mkw_configure_translated_target(mkw_base_shared)
-target_precompile_headers(mkw_base_shared PRIVATE "${MKW_RUNTIME_SOURCE_DIR}/include/mkw_pch.h")
+target_precompile_headers(mkw_base_shared PRIVATE "${MKW_TRANSLATED_PCH}")
+
+# Registration and indirect-dispatch tables are translator output too; they
+# compile with the translated configuration rather than inside the products.
+add_library(mkw_base_registration OBJECT ${MKW_BASE_REGISTRATION_SOURCES})
+mkw_configure_translated_target(mkw_base_registration)
+target_precompile_headers(mkw_base_registration REUSE_FROM mkw_base_shared)
 
 if(MKW_BASE_PORTABLE_SENSITIVE_SHARDS)
     add_library(mkw_base_sensitive OBJECT ${MKW_BASE_PORTABLE_SENSITIVE_SHARDS})
@@ -175,16 +141,31 @@ if(MKW_HAVE_RETRO_REWIND)
             set_source_files_properties("${source}" PROPERTIES LANGUAGE ASM SKIP_PRECOMPILE_HEADERS ON)
         endif()
     endforeach()
+    # mod_data_patches (MKW_RETRO_EXTRA_SOURCES) calls runtime APIs directly; it
+    # keeps the full runtime include path this target already has.
     add_library(mkw_retro_rewind_functions OBJECT ${MKW_RETRO_TRANSLATED_SOURCES})
     mkw_configure_translated_target(mkw_retro_rewind_functions)
     target_precompile_headers(mkw_retro_rewind_functions REUSE_FROM mkw_base_shared)
+
+    add_library(mkw_retro_registration OBJECT ${MKW_RETRO_REGISTRATION_SOURCES})
+    mkw_configure_translated_target(mkw_retro_registration)
+    target_precompile_headers(mkw_retro_registration REUSE_FROM mkw_base_shared)
 endif()
 
 function(mkw_configure_product target)
-    target_sources(${target} PRIVATE $<TARGET_OBJECTS:mkw_runtime_common>)
+    if(TARGET mkw_runtime_common)
+        target_sources(${target} PRIVATE $<TARGET_OBJECTS:mkw_runtime_common>)
+    else()
+        # Natives self-register from static constructors, so every archive
+        # member must be kept. Listed before mkw_base_shared so single-pass
+        # linkers resolve the runtime's references to translated functions.
+        target_link_libraries(${target} PRIVATE "$<LINK_LIBRARY:WHOLE_ARCHIVE,mkw_runtime_prebuilt>")
+        set_property(TARGET ${target} APPEND PROPERTY LINK_DEPENDS "${MKW_RUNTIME_PREBUILT_STAMP}")
+    endif()
+    target_sources(${target} PRIVATE $<TARGET_OBJECTS:mkw_game_data>)
     # Startup CPU check. Must stay a separate object library so it keeps the
     # plain baseline ISA while everything around it is built for x86-64-v3.
-    if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
+    if(MKW_HOST_IS_X86_64)
         target_sources(${target} PRIVATE $<TARGET_OBJECTS:mkw_cpu_baseline>)
     endif()
     target_include_directories(${target} PRIVATE
@@ -198,9 +179,6 @@ function(mkw_configure_product target)
         SDL_MAIN_HANDLED _DISABLE_STRING_ANNOTATION _DISABLE_VECTOR_ANNOTATION TARGET_PC)
     target_compile_features(${target} PRIVATE cxx_std_20)
     mkw_apply_common_compile_options(${target})
-    # The dispatch-table and registration shards compile inside the product target itself and
-    # include the same fat translated headers; bound them by the same pool.
-    mkw_bound_translated_compiles(${target})
     target_link_libraries(${target} PRIVATE
         mkw_platform mkw_base_shared mkw::pugixml mkw::toml11 mkw::cryptopp mkw::mbedtls)
 
@@ -223,7 +201,7 @@ function(mkw_configure_product target)
        (MKW_SQLITE_TARGET_TYPE STREQUAL "SHARED_LIBRARY" OR
         MKW_SQLITE_TARGET_TYPE STREQUAL "MODULE_LIBRARY"))
         add_custom_command(TARGET ${target} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different
-            $<TARGET_FILE:sqlite3> $<TARGET_FILE_DIR:${target}>)
+            $<TARGET_FILE:sqlite3> $<TARGET_FILE_DIR:${target}> VERBATIM)
     endif()
 
     if(MKW_PLATFORM_WINDOWS)
@@ -257,7 +235,7 @@ function(mkw_configure_product target)
             endif()
             add_custom_command(TARGET ${target} POST_BUILD
                 COMMAND ${CMAKE_COMMAND} -E copy_if_different
-                    "${runtime_dll_path}" $<TARGET_FILE_DIR:${target}>)
+                    "${runtime_dll_path}" $<TARGET_FILE_DIR:${target}> VERBATIM)
         endforeach()
     endif()
 
@@ -266,7 +244,7 @@ function(mkw_configure_product target)
         message(FATAL_ERROR "Missing Wii first-run bootstrap payload: ${MKW_WII_BOOTSTRAP_SOURCE_DIR}")
     endif()
     add_custom_command(TARGET ${target} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_directory
-        "${MKW_WII_BOOTSTRAP_SOURCE_DIR}" "$<TARGET_FILE_DIR:${target}>/wii_bootstrap")
+        "${MKW_WII_BOOTSTRAP_SOURCE_DIR}" "$<TARGET_FILE_DIR:${target}>/wii_bootstrap" VERBATIM)
 
     set(MKW_DSP_COEFFICIENT_ROM "${MKW_RUNTIME_SOURCE_DIR}/assets/dsp/dsp_coef.bin")
     if(NOT EXISTS "${MKW_DSP_COEFFICIENT_ROM}")
@@ -278,7 +256,7 @@ function(mkw_configure_product target)
         message(FATAL_ERROR "Wii DSP coefficient ROM hash mismatch: ${MKW_DSP_COEFFICIENT_ROM_SHA256}")
     endif()
     add_custom_command(TARGET ${target} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different
-        "${MKW_DSP_COEFFICIENT_ROM}" "$<TARGET_FILE_DIR:${target}>/dsp_coef.bin")
+        "${MKW_DSP_COEFFICIENT_ROM}" "$<TARGET_FILE_DIR:${target}>/dsp_coef.bin" VERBATIM)
 
     # Aurora imports this portable recipe database into each user's writable
     # pipeline cache. Keep the upstream filename so its default resourcesPath
@@ -290,7 +268,7 @@ function(mkw_configure_product target)
     endif()
     add_custom_command(TARGET ${target} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different
         "${MKW_INITIAL_PIPELINE_CACHE}"
-        "$<TARGET_FILE_DIR:${target}>/initial_pipeline_cache.db")
+        "$<TARGET_FILE_DIR:${target}>/initial_pipeline_cache.db" VERBATIM)
 
     # Non-Windows TLS (runtime/src/hle/net/network_ssl.cpp's mbed TLS path) needs a trusted root
     # CA bundle to verify server certificates against - Windows gets this for free from the OS via
@@ -304,12 +282,13 @@ function(mkw_configure_product target)
             message(FATAL_ERROR "Missing TLS root CA bundle: ${MKW_CA_CERTIFICATE_BUNDLE}")
         endif()
         add_custom_command(TARGET ${target} POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different
-            "${MKW_CA_CERTIFICATE_BUNDLE}" "$<TARGET_FILE_DIR:${target}>/cacert.pem")
+            "${MKW_CA_CERTIFICATE_BUNDLE}" "$<TARGET_FILE_DIR:${target}>/cacert.pem" VERBATIM)
     endif()
 endfunction()
 
-add_executable(WiiCompiled "${MKW_BASE_PRODUCT_SOURCE}" ${MKW_BASE_REGISTRATION_SOURCES})
+add_executable(WiiCompiled "${MKW_BASE_PRODUCT_SOURCE}")
 mkw_configure_product(WiiCompiled)
+target_sources(WiiCompiled PRIVATE $<TARGET_OBJECTS:mkw_base_registration>)
 target_precompile_headers(WiiCompiled PRIVATE
     "${MKW_RUNTIME_SOURCE_DIR}/include/mkw_pch.h")
 if(TARGET mkw_base_sensitive)
@@ -317,9 +296,10 @@ if(TARGET mkw_base_sensitive)
 endif()
 
 if(MKW_HAVE_RETRO_REWIND)
-    add_executable(RetroRewind "${MKW_RETRO_REWIND_PRODUCT_SOURCE}" ${MKW_RETRO_REGISTRATION_SOURCES})
+    add_executable(RetroRewind "${MKW_RETRO_REWIND_PRODUCT_SOURCE}")
     mkw_configure_product(RetroRewind)
     target_precompile_headers(RetroRewind REUSE_FROM WiiCompiled)
+    target_sources(RetroRewind PRIVATE $<TARGET_OBJECTS:mkw_retro_registration>)
     if(TARGET mkw_retro_sensitive)
         target_sources(RetroRewind PRIVATE $<TARGET_OBJECTS:mkw_retro_sensitive>)
     endif()
@@ -333,23 +313,12 @@ else()
     message(STATUS "RetroRewind target disabled (run translate-mod and emit-build-shards)")
 endif()
 
-# Windows and Linux x86_64 share the x86-64-v3 floor that the CPU baseline
-# object above checks. AArch64 builds are compiled locally for the host that
-# will run them, so both Linux and Apple Silicon use the compiler's native CPU
-# tuning rather than leaving target-specific performance on the table.
-if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
-    set(MKW_BASELINE_ARCH_FLAG -march=x86-64-v3)
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|ARM64)$")
-    set(MKW_BASELINE_ARCH_FLAG -mcpu=native)
-else()
-    set(MKW_BASELINE_ARCH_FLAG "")
-endif()
-
+# mkw_runtime_common gets the same flags from mkw_add_runtime_library.
 set(MKW_ALL_BUILD_TARGETS
-    mkw_runtime_common mkw_base_shared mkw_base_sensitive mkw_retro_sensitive
-    mkw_retro_rewind_functions WiiCompiled RetroRewind)
+    mkw_game_data mkw_base_shared mkw_base_registration mkw_base_sensitive mkw_retro_sensitive
+    mkw_retro_rewind_functions mkw_retro_registration WiiCompiled RetroRewind)
 foreach(target IN LISTS MKW_ALL_BUILD_TARGETS)
-    if(TARGET ${target} AND MKW_BASELINE_ARCH_FLAG)
-        target_compile_options(${target} PRIVATE ${MKW_BASELINE_ARCH_FLAG})
+    if(TARGET ${target})
+        mkw_apply_baseline_arch(${target})
     endif()
 endforeach()

runtime/cmake/RuntimeFingerprint.cmake

@@ -0,0 +1,98 @@
+# Content identity of the game-independent runtime and the rule deciding whether
+# a precompiled runtime archive may replace compiling it. Script mode prints
+# exactly one stdout line, runtime_fingerprint=<sha256 hex>:
+#   cmake -DMKW_RUNTIME_DIR=<abs runtime dir> [-DMKW_CPPWINRT_INCLUDE_DIR=<dir>]
+#         [-DMKW_X86_CPU_PROFILE=v2|v3] [-DCMAKE_OSX_DEPLOYMENT_TARGET=<ver>]
+#         -P runtime/cmake/RuntimeFingerprint.cmake
+#
+# Inputs: src/, include/, cmake/ and CMakeLists.txt (sorted relative paths and
+# content hashes), winrt/base.h, the CPU profile and the macOS deployment target.
+# third_party/ has its own package fingerprint; game/ is per-user. Dot-files
+# differ between machines.
+
+function(mkw_runtime_fingerprint_inputs out_var runtime_dir)
+    get_filename_component(_root "${runtime_dir}" ABSOLUTE)
+    file(GLOB_RECURSE _files LIST_DIRECTORIES false RELATIVE "${_root}"
+        "${_root}/src/*" "${_root}/include/*" "${_root}/cmake/*")
+    list(FILTER _files EXCLUDE REGEX "(^|/)\\.")
+    list(APPEND _files "CMakeLists.txt")
+    list(SORT _files COMPARE STRING CASE SENSITIVE)
+    set(${out_var} "${_files}" PARENT_SCOPE)
+endfunction()
+
+function(mkw_runtime_fingerprint out_var runtime_dir cppwinrt_dir cpu_profile)
+    get_filename_component(_root "${runtime_dir}" ABSOLUTE)
+    mkw_runtime_fingerprint_inputs(_files "${_root}")
+    set(_manifest "")
+    foreach(_file IN LISTS _files)
+        if(NOT EXISTS "${_root}/${_file}")
+            message(FATAL_ERROR "Runtime fingerprint input is missing: ${_root}/${_file}")
+        endif()
+        file(SHA256 "${_root}/${_file}" _hash)
+        string(APPEND _manifest "${_file} ${_hash}\n")
+    endforeach()
+    if(cppwinrt_dir)
+        if(NOT EXISTS "${cppwinrt_dir}/winrt/base.h")
+            message(FATAL_ERROR "MKW_CPPWINRT_INCLUDE_DIR does not contain winrt/base.h: ${cppwinrt_dir}")
+        endif()
+        file(SHA256 "${cppwinrt_dir}/winrt/base.h" _hash)
+        string(APPEND _manifest "cppwinrt:winrt/base.h ${_hash}\n")
+    endif()
+    string(APPEND _manifest "x86_cpu_profile ${cpu_profile}\n")
+    string(APPEND _manifest "osx_deployment_target ${CMAKE_OSX_DEPLOYMENT_TARGET}\n")
+    string(SHA256 _fingerprint "${_manifest}")
+    set(${out_var} "${_fingerprint}" PARENT_SCOPE)
+endfunction()
+
+# Sets out_use to TRUE/FALSE and out_reason to a one-line explanation. Pure, so
+# it can be exercised in script mode without a configured project.
+cmake_policy(PUSH)
+if(POLICY CMP0174)
+    cmake_policy(SET CMP0174 NEW)
+endif()
+function(mkw_runtime_prebuilt_decision out_use out_reason)
+    cmake_parse_arguments(PARSE_ARGV 2 _d ""
+        "ARCHIVE;ENABLED;X86_64;ARCHIVE_EXISTS;PACKAGE_PROFILE;PROFILE;PACKAGE_FINGERPRINT;FINGERPRINT" "")
+    set(_use FALSE)
+    if(NOT _d_ARCHIVE)
+        set(_reason "the native prebuilt package has no runtime archive")
+    elseif(NOT _d_ENABLED)
+        set(_reason "MKW_USE_PREBUILT_RUNTIME is OFF")
+    elseif(NOT _d_X86_64)
+        set(_reason "the precompiled runtime is x86_64-only")
+    elseif(NOT _d_ARCHIVE_EXISTS)
+        set(_reason "the runtime archive is missing: ${_d_ARCHIVE}")
+    elseif(NOT _d_PACKAGE_PROFILE STREQUAL _d_PROFILE)
+        set(_reason "the package was built for x86-64-${_d_PACKAGE_PROFILE}, this build targets x86-64-${_d_PROFILE}")
+    elseif(NOT _d_PACKAGE_FINGERPRINT)
+        set(_reason "the package records no runtime fingerprint")
+    elseif(NOT _d_FINGERPRINT OR NOT _d_PACKAGE_FINGERPRINT STREQUAL _d_FINGERPRINT)
+        set(_reason "runtime sources differ from the package (package ${_d_PACKAGE_FINGERPRINT}, workspace ${_d_FINGERPRINT})")
+    else()
+        set(_use TRUE)
+        set(_reason "fingerprint ${_d_FINGERPRINT} matches")
+    endif()
+    set(${out_use} ${_use} PARENT_SCOPE)
+    set(${out_reason} "${_reason}" PARENT_SCOPE)
+endfunction()
+cmake_policy(POP)
+
+if(CMAKE_SCRIPT_MODE_FILE)
+    get_filename_component(_mkw_script "${CMAKE_SCRIPT_MODE_FILE}" REALPATH)
+    get_filename_component(_mkw_self "${CMAKE_CURRENT_LIST_FILE}" REALPATH)
+endif()
+if(CMAKE_SCRIPT_MODE_FILE AND _mkw_script STREQUAL _mkw_self)
+    if(NOT MKW_RUNTIME_DIR OR NOT IS_DIRECTORY "${MKW_RUNTIME_DIR}")
+        message(FATAL_ERROR "Pass -DMKW_RUNTIME_DIR=<runtime directory>")
+    endif()
+    if(NOT MKW_X86_CPU_PROFILE)
+        set(MKW_X86_CPU_PROFILE "v3")
+    endif()
+    # Same default runtime/CMakeLists.txt forces on every platform before project().
+    if(NOT CMAKE_OSX_DEPLOYMENT_TARGET)
+        set(CMAKE_OSX_DEPLOYMENT_TARGET "12.0")
+    endif()
+    mkw_runtime_fingerprint(_mkw_fingerprint "${MKW_RUNTIME_DIR}" "${MKW_CPPWINRT_INCLUDE_DIR}" "${MKW_X86_CPU_PROFILE}")
+    # message() writes to stderr; the echo child inherits stdout.
+    execute_process(COMMAND "${CMAKE_COMMAND}" -E echo "runtime_fingerprint=${_mkw_fingerprint}")
+endif()

runtime/cmake/RuntimeLibrary.cmake

@@ -0,0 +1,180 @@
+# The game-independent native runtime (runtime/src). One definition serves
+# three uses: the OBJECT library the products link from source, the STATIC
+# archive the native prebuilt export ships (mkw_runtime), and that archive
+# imported in place of compiling it (mkw_runtime_prebuilt).
+
+include("${CMAKE_CURRENT_LIST_DIR}/RuntimeFingerprint.cmake")
+
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(AMD64|amd64|x86_64|X86_64)$")
+    set(MKW_HOST_IS_X86_64 TRUE)
+else()
+    set(MKW_HOST_IS_X86_64 FALSE)
+endif()
+
+# Windows, Linux and Intel macOS x86_64 use the selected profile. AArch64
+# builds are compiled locally for the host that will run them, so both Linux
+# and Apple Silicon use the compiler's native CPU tuning rather than leaving
+# target-specific performance on the table.
+if(MKW_HOST_IS_X86_64)
+    set(MKW_BASELINE_ARCH_FLAG -march=x86-64-${MKW_X86_CPU_PROFILE})
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|ARM64)$")
+    set(MKW_BASELINE_ARCH_FLAG -mcpu=native)
+else()
+    set(MKW_BASELINE_ARCH_FLAG "")
+endif()
+
+if(MKW_PLATFORM_MACOS)
+    find_library(MKW_COREAUDIO_FRAMEWORK CoreAudio REQUIRED)
+endif()
+
+# CMake decides WHOLE_ARCHIVE's form by probing ld.lld in ELF mode, which
+# accepts --push-state; the MinGW lld driver rejects it at link time.
+if(MKW_PLATFORM_WINDOWS)
+    set(CMAKE_CXX_LINK_LIBRARY_USING_WHOLE_ARCHIVE
+        "LINKER:--whole-archive" "<LINK_ITEM>" "LINKER:--no-whole-archive")
+endif()
+
+function(mkw_apply_common_compile_options target)
+    target_compile_options(${target} PRIVATE -O3 -ffast-math -w -pipe)
+endfunction()
+
+function(mkw_apply_baseline_arch target)
+    if(MKW_BASELINE_ARCH_FLAG)
+        target_compile_options(${target} PRIVATE ${MKW_BASELINE_ARCH_FLAG})
+        if(MKW_X86_CPU_PROFILE STREQUAL "v2")
+            target_compile_definitions(${target} PRIVATE MKW_X86_CPU_PROFILE_V2=1)
+        endif()
+    endif()
+endfunction()
+
+function(mkw_configure_object_target target)
+    target_include_directories(${target} PRIVATE
+        "${MKW_RUNTIME_SOURCE_DIR}/include"
+        "${MKW_RUNTIME_SOURCE_DIR}/src"
+        # Workspace root, so translator output is spelled "generated/<x>.h"
+        # instead of a ../ chain whose depth depends on the includer.
+        "${MKW_RUNTIME_SOURCE_DIR}/.."
+        "${MKW_RUNTIME_SOURCE_DIR}/../aurora-main/include")
+    target_compile_definitions(${target} PRIVATE
+        TARGET_PC)
+    set_target_properties(${target} PROPERTIES CXX_STANDARD 17 CXX_STANDARD_REQUIRED ON)
+endfunction()
+
+# scope is PRIVATE for a compiled target and INTERFACE for the imported archive.
+function(mkw_runtime_link_dependencies target scope)
+    target_link_libraries(${target} ${scope}
+        aurora::gx aurora::pad aurora::si aurora::vi aurora::mtx)
+    target_link_libraries(${target} ${scope} mkw_platform mkw::pugixml mkw::toml11 mkw::cryptopp mkw::mbedtls)
+    if(MKW_PLATFORM_WINDOWS)
+        target_link_libraries(${target} ${scope} shell32 windowsapp)
+    elseif(MKW_PLATFORM_LINUX OR MKW_PLATFORM_MACOS_X86_64)
+        # ${CMAKE_DL_LIBS} for music_attenuation.cpp's dlopen of libdbus-1 (MPRIS
+        # media monitoring). Empty on platforms where dl* is already in libc/libSystem.
+        target_link_libraries(${target} ${scope} mkw::libco ${CMAKE_DL_LIBS})
+    endif()
+    if(MKW_PLATFORM_MACOS)
+        # CoreAudio framework is required for automatic music muting on macOS.
+        target_link_libraries(${target} ${scope} "${MKW_COREAUDIO_FRAMEWORK}")
+        if(MKW_PLATFORM_MACOS_X86_64)
+            # libco is used by Intel macOS. Apple Silicon uses the local
+            # x18-safe assembly backend and therefore does not define mkw::libco.
+            target_link_libraries(${target} ${scope} mkw::libco)
+        endif()
+    endif()
+endfunction()
+
+# Compiles the runtime sources as an OBJECT or STATIC library.
+function(mkw_add_runtime_library target type)
+    set(_sources ${ARGN})
+    add_library(${target} ${type} ${_sources})
+    mkw_configure_object_target(${target})
+    target_compile_features(${target} PRIVATE cxx_std_20)
+    target_compile_definitions(${target} PRIVATE
+        SDL_MAIN_HANDLED
+        _DISABLE_STRING_ANNOTATION _DISABLE_VECTOR_ANNOTATION)
+    mkw_runtime_link_dependencies(${target} PRIVATE)
+
+    if(MKW_CPPWINRT_INCLUDE_DIR)
+        if(NOT EXISTS "${MKW_CPPWINRT_INCLUDE_DIR}/winrt/base.h")
+            message(FATAL_ERROR
+                "MKW_CPPWINRT_INCLUDE_DIR does not contain winrt/base.h: ${MKW_CPPWINRT_INCLUDE_DIR}")
+        endif()
+        target_include_directories(${target} PRIVATE "${MKW_CPPWINRT_INCLUDE_DIR}")
+    endif()
+
+    # Keep runtime unity units small and semantically related. The old generated-TU
+    # batch size put all 57 native runtime sources into one memory-heavy compiler job.
+    foreach(source IN LISTS _sources)
+        get_filename_component(source_name "${source}" NAME_WE)
+        string(REPLACE "\\" "/" source_normalized "${source}")
+        if(source_normalized MATCHES "/hle/gx/")
+            set(runtime_group "gx_bridge")
+        elseif(source_name MATCHES "network|socket|dns|dwc|ios")
+            set(runtime_group "network_ios")
+        elseif(source_name MATCHES "os_|system|memory|fiber|scheduler")
+            set(runtime_group "guest_system")
+        elseif(source_name MATCHES "debug|trace|prof")
+            set(runtime_group "diagnostics")
+        else()
+            string(SHA256 source_hash "${source_name}")
+            string(SUBSTRING "${source_hash}" 0 4 source_hash_prefix)
+            math(EXPR runtime_bucket "0x${source_hash_prefix} % 8")
+            set(runtime_group "runtime_${runtime_bucket}")
+        endif()
+        set_source_files_properties("${source}" PROPERTIES UNITY_GROUP "${runtime_group}")
+    endforeach()
+    # PPC semantic sources are excluded from unity/PCH and configured in CMakeLists.txt.
+    set_target_properties(${target} PROPERTIES UNITY_BUILD ON UNITY_BUILD_MODE GROUP)
+    target_precompile_headers(${target} PRIVATE "${MKW_RUNTIME_SOURCE_DIR}/include/mkw_pch.h")
+    mkw_apply_common_compile_options(${target})
+    mkw_apply_baseline_arch(${target})
+endfunction()
+
+function(mkw_import_prebuilt_runtime target archive)
+    add_library(${target} STATIC IMPORTED GLOBAL)
+    set_target_properties(${target} PROPERTIES IMPORTED_LOCATION "${archive}")
+    mkw_runtime_link_dependencies(${target} INTERFACE)
+endfunction()
+
+# Sets MKW_RUNTIME_USE_PREBUILT in the caller. The archive replaces the
+# runtime's objects only when it was built from exactly this workspace's
+# runtime inputs for the same x86-64 profile.
+function(mkw_select_runtime_provider)
+    set(_archive "${MKW_NP_RUNTIME_ARCHIVE_PATH}")
+    set(_exists FALSE)
+    if(_archive AND EXISTS "${_archive}")
+        set(_exists TRUE)
+    endif()
+    set(_fingerprint "")
+    if(_exists AND MKW_USE_PREBUILT_RUNTIME AND MKW_HOST_IS_X86_64 AND
+       MKW_NP_RUNTIME_CPU_PROFILE STREQUAL MKW_X86_CPU_PROFILE)
+        mkw_runtime_fingerprint(_fingerprint "${MKW_RUNTIME_SOURCE_DIR}"
+            "${MKW_CPPWINRT_INCLUDE_DIR}" "${MKW_X86_CPU_PROFILE}")
+    endif()
+    mkw_runtime_prebuilt_decision(_use _reason
+        ARCHIVE "${_archive}"
+        ENABLED "${MKW_USE_PREBUILT_RUNTIME}"
+        X86_64 "${MKW_HOST_IS_X86_64}"
+        ARCHIVE_EXISTS "${_exists}"
+        PACKAGE_PROFILE "${MKW_NP_RUNTIME_CPU_PROFILE}"
+        PROFILE "${MKW_X86_CPU_PROFILE}"
+        PACKAGE_FINGERPRINT "${MKW_NP_RUNTIME_FINGERPRINT}"
+        FINGERPRINT "${_fingerprint}")
+    if(NOT MKW_NATIVE_PREBUILT_DIR)
+        set(_reason "no native prebuilt package is configured")
+    endif()
+
+    if(_use)
+        message(STATUS "Native runtime: precompiled archive ${_archive} (${_reason})")
+        # Any runtime edit must re-run configure so the mismatch is noticed.
+        mkw_runtime_fingerprint_inputs(_inputs "${MKW_RUNTIME_SOURCE_DIR}")
+        list(TRANSFORM _inputs PREPEND "${MKW_RUNTIME_SOURCE_DIR}/")
+        if(MKW_CPPWINRT_INCLUDE_DIR)
+            list(APPEND _inputs "${MKW_CPPWINRT_INCLUDE_DIR}/winrt/base.h")
+        endif()
+        set_property(DIRECTORY "${CMAKE_SOURCE_DIR}" APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS ${_inputs})
+    else()
+        message(STATUS "Native runtime: compiled from source (${_reason})")
+    endif()
+    set(MKW_RUNTIME_USE_PREBUILT ${_use} PARENT_SCOPE)
+endfunction()

runtime/game/runtime_config_values.cpp

@@ -0,0 +1,8 @@
+#include "guest_small_data.h"
+
+#include "generated/RuntimeConfig.h"
+
+namespace GuestSmallData {
+extern const uint32_t kSda1Base = RuntimeConfig::SDA1_BASE;
+extern const uint32_t kSda2Base = RuntimeConfig::SDA2_BASE;
+} // namespace GuestSmallData

runtime/game/setup_version.cpp

@@ -0,0 +1,7 @@
+// Compiled per user rather than into the shared runtime archive.
+extern const char* const kMkwSetupVersion =
+#if defined(MKW_SETUP_VERSION)
+    MKW_SETUP_VERSION;
+#else
+    "";
+#endif

runtime/include/abi_bridge.h

@@ -18,638 +18,14 @@
 #include <tuple>
 #include <type_traits>
 
-inline void InvokeIndirectCpu(uint32_t target, CpuContext* ctx);
+#include "guest_abi/dispatch.h"
 
-// Some game-facing runtime options alter arguments at well-defined ABI
-// boundaries. Keep this independent of the dispatch mechanism: generated
-// static calls deliberately bypass InvokeDirectCpu for performance.
-// (used for the path mask filtering in ScnRenderer::createPath: depth of
-// field is always removed, bloom when the user disabled it)
-inline void ApplyRuntimeCallOptions(uint32_t target, CpuContext* ctx) {
-    if (target == 0x8023BD38u) {
-        // ScnRenderer::createPath receives the post-processing path mask in r4.
-        ctx->gpr[4] = RuntimeGameGraphicsOptions::FilterScnRendererPathMask(ctx->gpr[4]);
-    }
-}
-
-// Persistent per-thread CPU context used across translated function calls.
-CpuContext& GetPersistentCpuContext();
 void InitializePersistentCpuContext();
 
-enum class FunctionKind : uint8_t {
-    BaseTranslated = 0,
-    ModTranslated = 1,
-    Native = 2,
-};
-
-inline constexpr uint32_t kPpcAllNonvolatileFprMask = 0xffffc000u;
-inline constexpr uint32_t kBaseTranslatedFunctionPriority = 0;
-inline constexpr uint32_t kModTranslatedFunctionPriorityBase = 100;
-inline constexpr uint32_t kNativeFunctionPriority = 10000;
-
-#include "isa/ppc_isa_cr.h"
-
-struct TranslatedFunctionInfo {
-    uint32_t address = 0;
-    const char* name = "";
-    uint64_t moduleId = 0;
-    uint32_t priority = 0;
-    uint32_t nonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
-    void* entryPoint = nullptr; // Raw typed function pointer for indirect-call resolution
-    void (*rawCpuInvoker)(CpuContext*) = nullptr;
-    bool mustRemainDynamicallyDispatchable = true;
-    FunctionKind kind = FunctionKind::BaseTranslated;
-};
-
-struct RawDispatchRecord {
-    uint32_t address = 0;
-    void (*entry)(CpuContext*) = nullptr;
-    uint32_t nonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
-    bool preserveNonvolatileGprs = false;
-};
-
-// Translator-owned indirect dispatch is split by the high address byte and
-// then by 4 KiB guest pages. Each populated segment stores only the contiguous
-// page range that contains verified function entry points. The final search is
-// bounded to at most the 1024 aligned instruction addresses in one guest page.
-struct StaticIndirectDispatchPage {
-    uint32_t firstEntry = 0;
-    uint16_t entryCount = 0;
-    // Explicit tail padding to an 8-byte stride. The emitter writes it as a
-    // literal third initializer, so it is part of the generated-data contract.
-    uint16_t reserved = 0;
-};
-
-struct StaticIndirectDispatchSegment {
-    const StaticIndirectDispatchPage* pages = nullptr;
-    uint16_t firstPage = 0;
-    uint16_t pageCount = 0;
-};
-
-struct StaticIndirectDispatchTable {
-    const char* profileName = nullptr;
-    const StaticIndirectDispatchSegment* segments = nullptr; // Exactly 256 high-byte segments.
-    const RawDispatchRecord* entries = nullptr;
-    size_t entryCount = 0;
-};
-
-inline const RawDispatchRecord* FindStaticIndirectDispatchEntry(
-    const StaticIndirectDispatchTable* table,
-    uint32_t address) noexcept {
-    if (!table || !table->segments || !table->entries || table->entryCount == 0) {
-        return nullptr;
-    }
-
-    const auto& segment = table->segments[address >> 24];
-    const uint32_t pageNumber = (address >> 12) & 0x0FFFu;
-    if (!segment.pages || pageNumber < segment.firstPage) {
-        return nullptr;
-    }
-    const uint32_t relativePage = pageNumber - segment.firstPage;
-    if (relativePage >= segment.pageCount) {
-        return nullptr;
-    }
-
-    const auto& page = segment.pages[relativePage];
-    const size_t first = page.firstEntry;
-    const size_t count = page.entryCount;
-    if (first > table->entryCount || count > table->entryCount - first) {
-        return nullptr;
-    }
-
-    size_t lower = first;
-    size_t upper = first + count;
-    while (lower < upper) {
-        const size_t middle = lower + (upper - lower) / 2;
-        const uint32_t candidate = table->entries[middle].address;
-        if (candidate < address) {
-            lower = middle + 1;
-        } else {
-            upper = middle;
-        }
-    }
-    return lower < first + count && table->entries[lower].address == address
-        ? &table->entries[lower]
-        : nullptr;
-}
-
-void RegisterStaticIndirectDispatchTable(const StaticIndirectDispatchTable* table);
-inline std::atomic<const StaticIndirectDispatchTable*> g_publishedStaticIndirectDispatchTable{nullptr};
-
-class StaticIndirectDispatchTableRegistrar {
-public:
-    explicit StaticIndirectDispatchTableRegistrar(const StaticIndirectDispatchTable* table) {
-        RegisterStaticIndirectDispatchTable(table);
-    }
-};
-
-inline unsigned PpcLowestSetBitIndex(uint32_t value) noexcept {
-    return static_cast<unsigned>(__builtin_ctz(value));
-}
-
-class PpcNonvolatileFprGuard {
-public:
-    explicit PpcNonvolatileFprGuard(CpuContext* cpu, uint32_t mask = kPpcAllNonvolatileFprMask) noexcept
-        : cpu_(cpu), mask_(mask & kPpcAllNonvolatileFprMask) {
-        if (!cpu_ || mask_ == 0) {
-            return;
-        }
-        // Real masks are a single contiguous register run - f14..f31 for the
-        // conservative default and f23..f31 or similar for a measured writer -
-        // so resolve the run once and copy it as one straight-line block instead
-        // of testing 18 mask bits. Any other mask keeps the per-bit loop, which
-        // saves and restores exactly the same registers.
-        const uint32_t lowest = mask_ & (0u - mask_);
-        const uint32_t aboveRun = mask_ + lowest;
-        if ((aboveRun & (aboveRun - 1u)) == 0u) {
-            const unsigned first = PpcLowestSetBitIndex(mask_);
-            const unsigned end = aboveRun == 0u ? 32u : PpcLowestSetBitIndex(aboveRun);
-            firstSlot_ = static_cast<uint8_t>(first - 14u);
-            slotCount_ = static_cast<uint8_t>(end - first);
-            const size_t count = slotCount_;
-            for (size_t i = 0; i < count; ++i) {
-                saved_[firstSlot_ + i] = cpu_->fpr[first + i];
-            }
-            return;
-        }
-        for (size_t i = 0; i < saved_.size(); ++i) {
-            if ((mask_ & (1u << (14 + i))) == 0) {
-                continue;
-            }
-            saved_[i] = cpu_->fpr[14 + i];
-        }
-    }
-
-    ~PpcNonvolatileFprGuard() noexcept {
-        if (!cpu_ || mask_ == 0) {
-            return;
-        }
-        if (slotCount_ != 0) {
-            const unsigned first = 14u + firstSlot_;
-            const size_t count = slotCount_;
-            for (size_t i = 0; i < count; ++i) {
-                cpu_->fpr[first + i] = saved_[firstSlot_ + i];
-            }
-            return;
-        }
-        for (size_t i = 0; i < saved_.size(); ++i) {
-            if ((mask_ & (1u << (14 + i))) == 0) {
-                continue;
-            }
-            cpu_->fpr[14 + i] = saved_[i];
-        }
-    }
-
-    PpcNonvolatileFprGuard(const PpcNonvolatileFprGuard&) = delete;
-    PpcNonvolatileFprGuard& operator=(const PpcNonvolatileFprGuard&) = delete;
-
-private:
-    CpuContext* cpu_ = nullptr;
-    uint32_t mask_ = 0;
-    // Contiguous-run description of mask_. slotCount_ == 0 means the mask is not
-    // a single run, so the destructor mirrors the constructor's per-bit loop.
-    uint8_t firstSlot_ = 0;
-    uint8_t slotCount_ = 0;
-    // Every element selected by mask_ is written before it is read. Leaving the
-    // remaining slots uninitialized avoids zero-filling all 18 values around a
-    // call that may preserve only one or two architectural FPRs.
-    std::array<PPC_FPR, 18> saved_;
-};
-
-class PpcNonvolatileGprGuard {
-public:
-    explicit PpcNonvolatileGprGuard(CpuContext* cpu, bool enabled = true) noexcept
-        : cpu_(enabled ? cpu : nullptr) {
-        if (!cpu_) {
-            return;
-        }
-        for (size_t i = 0; i < saved_.size(); ++i) {
-            saved_[i] = cpu_->gpr[14 + i];
-        }
-    }
-
-    ~PpcNonvolatileGprGuard() noexcept {
-        if (!cpu_) {
-            return;
-        }
-        for (size_t i = 0; i < saved_.size(); ++i) {
-            cpu_->gpr[14 + i] = saved_[i];
-        }
-    }
-
-    PpcNonvolatileGprGuard(const PpcNonvolatileGprGuard&) = delete;
-    PpcNonvolatileGprGuard& operator=(const PpcNonvolatileGprGuard&) = delete;
-
-private:
-    CpuContext* cpu_ = nullptr;
-    // Left uninitialized: enabled writes all 18 slots before reading them, disabled reads none,
-    // so skipping the zero-fill saves 72 bytes of stack work on the common disabled hot path.
-    std::array<uint32_t, 18> saved_;
-};
-
-// Guest address of OSLoadContext, which models rfi-style context restoration
-// and intentionally replaces the full guest register file instead of returning
-// like a normal ABI call. It can never be guarded.
-inline constexpr uint32_t kOSLoadContextAddress = 0x801A1F58u;
-
-inline bool ShouldPreserveNonvolatileGprsForRawCpuCall(const TranslatedFunctionInfo* info) noexcept {
-    return info->kind == FunctionKind::Native && info->address != kOSLoadContextAddress;
-}
-
-inline uint32_t NonvolatileFprGuardMaskFor(const TranslatedFunctionInfo* info) noexcept {
-    if (!info || info->nonvolatileFprWriteMask == 0) {
-        return 0;
-    }
-    return info->nonvolatileFprWriteMask & kPpcAllNonvolatileFprMask;
-}
-
-template <uint32_t Target>
-struct KnownTranslatedCpuCall {
-    static constexpr bool kAvailable = false;
-    static constexpr uint32_t kNonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
-    static constexpr bool kMustRemainDynamicallyDispatchable = true;
-    static constexpr void (*Entry)(CpuContext*) = nullptr;
-};
-
-// Translator-emitted trait specializations, one macro shared by every generated shard.
-// `addr` is the 8-digit hex entry point, `winner` the resolved symbol (mods/Retro Rewind can
-// publish their own name). A specialization only exists for a single statically bound winner,
-// so overridable/dynamically-dispatchable are hardcoded here rather than passed in.
-#define MKW_TRANSLATED_TRAIT(addr, winner, nonvolatile_fpr_write_mask)                    \
-    extern "C" void winner(CpuContext* ctx);                                              \
-    template <>                                                                           \
-    struct KnownTranslatedCpuCall<0x##addr##u> {                                          \
-        static constexpr bool kAvailable = true;                                          \
-        static constexpr uint32_t kNonvolatileFprWriteMask = nonvolatile_fpr_write_mask;  \
-        static constexpr bool kMustRemainDynamicallyDispatchable = false;                 \
-        static constexpr void (*Entry)(CpuContext*) = &winner;                            \
-    }
-
-template <uint32_t Target>
-struct KnownNativeCpuCall {
-    static constexpr bool kAvailable = false;
-    static constexpr uint32_t kNonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
-    static constexpr void (*Entry)(CpuContext*) = nullptr;
-};
-
-template <uint32_t Target>
-struct KnownTypedNativeCpuCall {
-    static constexpr bool kAvailable = false;
-};
-
-#include "native_cpu_calls.inc"
-
-// Direct-mapped thread-local memo in front of the sorted registry lookup every bctrl performs;
-// a miss just re-runs the sorted lookup. 512 entries (12 KiB) covers the per-frame indirect
-// working set while staying L1/L2 resident, unlike 4096 which would thrash L2. Must stay a
-// power of two, the index masks with (size - 1).
-inline constexpr size_t kIndirectDispatchCacheEntries = 512;
-
-// Namespace-scope `inline thread_local`, not function-local `static thread_local`, to avoid a
-// thread-static init epoch check on every bctrl (same as g_currentCpuContext in ppc_runtime.h).
-struct IndirectResolvedDispatchMemoEntry {
-    bool valid;
-    uint32_t address;
-    const TranslatedFunctionInfo* info;
-};
-inline thread_local IndirectResolvedDispatchMemoEntry
-    g_indirectResolvedDispatchMemo[kIndirectDispatchCacheEntries]{};
-
-struct IndirectRawDispatchMemoEntry {
-    bool valid;
-    uint32_t address;
-    const RawDispatchRecord* record;
-};
-inline thread_local IndirectRawDispatchMemoEntry
-    g_indirectRawDispatchMemo[kIndirectDispatchCacheEntries]{};
-
-class TranslatedFunctionRegistry {
-public:
-    static void Register(TranslatedFunctionInfo info);
-    static void Finalize();
-    static inline bool IsLookupPublished() noexcept {
-        return lookupPublished_.load(std::memory_order_acquire);
-    }
-    static std::optional<TranslatedFunctionInfo> FindNearestByAddress(uint32_t address);
-    static inline const TranslatedFunctionInfo* FindByAddressPtr(uint32_t address) {
-        if (!lookupPublished_.load(std::memory_order_acquire)) {
-            return FindByAddressPtrSlow(address);
-        }
-
-        auto& cached = g_indirectResolvedDispatchMemo[
-            ((address >> 2) * 2654435761u) & (kIndirectDispatchCacheEntries - 1)];
-        if (cached.valid && cached.address == address) {
-            return cached.info;
-        }
-
-        const auto* info = FindByAddressPtrSlow(address);
-        cached = {true, address, info};
-        return info;
-    }
-
-    // Hot wrapper: only the direct-mapped memo probe stays inline. Inlining the
-    // generated-table binary search alongside it pushed the whole function past
-    // the inliner's budget and out-of-lined the probe itself, which is the part
-    // every bctrl executes.
-    static MKW_PPC_FORCE_INLINE const RawDispatchRecord* FindRawByAddressPtr(uint32_t address) {
-        if (!lookupPublished_.load(std::memory_order_acquire)) {
-            return nullptr;
-        }
-
-        auto& cached = g_indirectRawDispatchMemo[
-            ((address >> 2) * 2654435761u) & (kIndirectDispatchCacheEntries - 1)];
-        if (cached.valid && cached.address == address) {
-            return cached.record;
-        }
-
-        return FindRawByAddressPtrMiss(address, cached);
-    }
-    // Look up a registered function by host (native) address.
-    // This enables stack trace symbolication for translated functions.
-    static std::optional<TranslatedFunctionInfo> FindByHostAddress(uintptr_t hostAddr);
-
-private:
-    // Memo miss: resolve against the header-inline generated table first; only dynamically
-    // registered records or a not-yet-published table fall through to the slow path.
-    static MKW_PPC_NO_INLINE MKW_PPC_COLD const RawDispatchRecord* FindRawByAddressPtrMiss(
-        uint32_t address, IndirectRawDispatchMemoEntry& cached) {
-        const RawDispatchRecord* record = nullptr;
-        if (const auto* table =
-                g_publishedStaticIndirectDispatchTable.load(std::memory_order_acquire)) {
-            record = FindStaticIndirectDispatchEntry(table, address);
-        }
-        if (record == nullptr) {
-            record = FindRawByAddressPtrSlow(address);
-        }
-        cached = {true, address, record};
-        return record;
-    }
-
-    static const TranslatedFunctionInfo* FindByAddressPtrSlow(uint32_t address);
-    static const RawDispatchRecord* FindRawByAddressPtrSlow(uint32_t address);
-
-    // Finalize publishes immutable registry, generated-table, and dynamic raw
-    // dispatch storage. Header-inlined cache hits are enabled only after that
-    // release publication, so a pre-finalization miss can never poison them.
-    inline static std::atomic_bool lookupPublished_{false};
-};
-
-// Cold, profile-owned registration data is emitted in compact table shards.
-// Keeping it out of generated function bodies avoids one static constructor
-// per translated function.
-struct BulkTranslatedFunctionRecord {
-    uint32_t address;
-    const char* name;
-    void (*entry)(CpuContext*);
-    FunctionKind kind;
-    bool preservesNonvolatileFprs;
-    uint32_t nonvolatileFprWriteMask;
-    uint32_t priority;
-    uint64_t moduleId;
-    bool mustRemainDynamicallyDispatchable;
-};
-
-void RegisterBulkTranslatedFunctions(const BulkTranslatedFunctionRecord* records, size_t count);
-
-class BulkTranslatedFunctionRegistrar {
-public:
-    BulkTranslatedFunctionRegistrar(const BulkTranslatedFunctionRecord* records, size_t count) {
-        RegisterBulkTranslatedFunctions(records, count);
-    }
-};
-
-MKW_PPC_FORCE_INLINE bool TryDispatchResolvedCpuTarget(const TranslatedFunctionInfo* info, CpuContext* cpu) {
-    if (!info || !info->rawCpuInvoker) {
-        return false;
-    }
-
-    RecompMod::ScopedTranslatedExecutionAddress translatedExecution(info->address);
-    PpcNonvolatileFprGuard fprGuard(cpu, NonvolatileFprGuardMaskFor(info));
-    PpcNonvolatileGprGuard gprGuard(cpu, ShouldPreserveNonvolatileGprsForRawCpuCall(info));
-    if (TryGetCpuContext() != cpu) {
-        CpuContextScope scope(cpu);
-        info->rawCpuInvoker(cpu);
-        return true;
-    }
-    info->rawCpuInvoker(cpu);
-    return true;
-}
-
-inline bool TryDispatchRawCpuTarget(const RawDispatchRecord* record, CpuContext* cpu) {
-    if (!record || !record->entry) {
-        return false;
-    }
-    RecompMod::ScopedTranslatedExecutionAddress translatedExecution(record->address);
-    PpcNonvolatileGprGuard gprGuard(cpu, record->preserveNonvolatileGprs);
-    PpcNonvolatileFprGuard fprGuard(cpu, record->nonvolatileFprWriteMask);
-    if (TryGetCpuContext() != cpu) {
-        CpuContextScope scope(cpu);
-        record->entry(cpu);
-    } else {
-        record->entry(cpu);
-    }
-    return true;
-}
-
-template <uint32_t Target>
-inline void DispatchKnownTranslatedCpuTargetStatic(CpuContext* cpu) {
-    const auto invokeKnownTranslated = [&]() {
-        KnownTranslatedCpuCall<Target>::Entry(cpu);
-    };
-
-    // A statically resolved, non-overridable translated call already running
-    // in this CpuContext needs no registry, context, or diagnostic target
-    // transition. This is the normal generated-to-generated hot path.
-    if (TryGetCpuContext() == cpu) {
-        if constexpr (KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask == 0) {
-            invokeKnownTranslated();
-            return;
-        } else {
-            PpcNonvolatileFprGuard fprGuard(
-                cpu, KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask);
-            invokeKnownTranslated();
-            return;
-        }
-    }
-
-    CpuContextScope contextScope(cpu);
-    RecompMod::ScopedTranslatedExecutionAddress translatedExecution(Target);
-    if constexpr (KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask == 0) {
-        invokeKnownTranslated();
-        return;
-    } else {
-        PpcNonvolatileFprGuard fprGuard(cpu, KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask);
-        invokeKnownTranslated();
-    }
-}
-
-[[noreturn]] inline void ReportMissingCpuTarget(uint32_t target, CpuContext* cpu) {
-    // Centralized crash path: stderr diagnostics, heuristics, crash artifacts
-    // in the run's log folder, popup, exit.
-    RuntimeCrash::FatalMissingGuestTarget(target, cpu);
-}
-
-template <uint32_t Target>
-inline const TranslatedFunctionInfo* ResolveDirectCpuTargetInfo() {
-    return TranslatedFunctionRegistry::FindByAddressPtr(Target);
-}
-
-template <uint32_t Target>
-inline const TranslatedFunctionInfo* ResolveDirectCpuTargetInfoCached() {
-    // Only latch a result once the registry is published (immutable for the process lifetime);
-    // a pre-finalization miss or provisional winner must never be cached. 0 = not yet resolved,
-    // 1 = negative-cache sentinel, avoiding a second ready flag on the hot path.
-    static std::atomic<uintptr_t> cachedValue{0};
-    constexpr uintptr_t kNegativeCache = 1;
-    const uintptr_t cached = cachedValue.load(std::memory_order_acquire);
-    if (cached != 0) {
-        return cached == kNegativeCache
-            ? nullptr
-            : reinterpret_cast<const TranslatedFunctionInfo*>(cached);
-    }
-    if (!TranslatedFunctionRegistry::IsLookupPublished()) {
-        return ResolveDirectCpuTargetInfo<Target>();
-    }
-    const auto* info = ResolveDirectCpuTargetInfo<Target>();
-    cachedValue.store(info != nullptr
-                          ? reinterpret_cast<uintptr_t>(info)
-                          : kNegativeCache,
-                      std::memory_order_release);
-    return info;
-}
-
-// Caches the "is the published winner still base translation?" verdict for state-free fast
-// paths guarding tens of thousands of call sites. Same latching rule as above: 0 = unresolved,
-// 1 = negative, 2 = positive, and a verdict computed pre-publication is never stored.
-template <uint32_t Target>
-inline bool IsBaseTranslatedCpuTargetActive() {
-    static std::atomic<uint8_t> cachedVerdict{0};
-    constexpr uint8_t kNegativeVerdict = 1;
-    constexpr uint8_t kPositiveVerdict = 2;
-    const uint8_t cached = cachedVerdict.load(std::memory_order_acquire);
-    if (cached != 0) {
-        return cached == kPositiveVerdict;
-    }
-    const bool publishable = TranslatedFunctionRegistry::IsLookupPublished();
-    const auto* info = ResolveDirectCpuTargetInfo<Target>();
-    const bool active = info != nullptr && info->kind == FunctionKind::BaseTranslated;
-    if (publishable) {
-        cachedVerdict.store(active ? kPositiveVerdict : kNegativeVerdict,
-                            std::memory_order_release);
-    }
-    return active;
-}
-
-template <uint32_t Target>
-inline void InvokeDirectCpu(CpuContext* ctx) {
-    static_assert(Target != 0, "InvokeDirectCpu cannot target address 0");
-    CpuContext* cpu = ctx ? ctx : &GetPersistentCpuContext();
-    ApplyRuntimeCallOptions(Target, cpu);
-    if constexpr (KnownNativeCpuCall<Target>::kAvailable) {
-        const auto invokeKnownNative = [&]() {
-            PpcNonvolatileGprGuard gprGuard(cpu);
-            if (TryGetCpuContext() != cpu) {
-                CpuContextScope scope(cpu);
-                KnownNativeCpuCall<Target>::Entry(cpu);
-                return;
-            }
-            KnownNativeCpuCall<Target>::Entry(cpu);
-        };
-        if constexpr (KnownNativeCpuCall<Target>::kNonvolatileFprWriteMask == 0) {
-            invokeKnownNative();
-        } else {
-            PpcNonvolatileFprGuard fprGuard(cpu, KnownNativeCpuCall<Target>::kNonvolatileFprWriteMask);
-            invokeKnownNative();
-        }
-        return;
-    }
-
-    if constexpr (KnownTypedNativeCpuCall<Target>::kAvailable) {
-        PpcNonvolatileGprGuard gprGuard(cpu);
-        if (TryGetCpuContext() != cpu) {
-            CpuContextScope scope(cpu);
-            KnownTypedNativeCpuCall<Target>::Invoke(cpu);
-        } else {
-            KnownTypedNativeCpuCall<Target>::Invoke(cpu);
-        }
-        return;
-    }
-
-    if constexpr (KnownTranslatedCpuCall<Target>::kAvailable) {
-        DispatchKnownTranslatedCpuTargetStatic<Target>(cpu);
-        return;
-    }
-
-    const auto* registeredInfo = ResolveDirectCpuTargetInfoCached<Target>();
-    if (TryDispatchResolvedCpuTarget(registeredInfo, cpu)) {
-        return;
-    }
-
-    ReportMissingCpuTarget(Target, cpu);
-}
-
-// Indirect jump (bctr without link) - used for switch tables.
-inline void InvokeIndirectJump(uint32_t target, CpuContext* ctx) {
-    // Use the caller's context, not the persistent global context!
-    // This is essential for tail calls via bctr where the 'this' pointer (r3)
-    // must be passed correctly to the target function.
-    CpuContext* cpu = ctx ? ctx : &GetPersistentCpuContext();
-    if (TryDispatchRawCpuTarget(TranslatedFunctionRegistry::FindRawByAddressPtr(target), cpu)) {
-        return;
-    }
-
-    // Cold fallback for dynamically registered or signature-only targets.
-    const auto* info = TranslatedFunctionRegistry::FindByAddressPtr(target);
-    if (info) {
-        if (TryDispatchResolvedCpuTarget(info, cpu)) {
-            return;
-        }
-    }
-    
-    // If not registered, this is likely a switch table jump to an intra-function label.
-    // The translator should have recognized this pattern and generated proper switch code.
-    RT_LOG(RT_TAG_RUNTIME) << "InvokeIndirectJump: target 0x"
-              << std::hex << target << std::dec
-              << " is not a registered function.\n"
-              << "This is likely a switch statement jump table that the translator\n"
-              << "failed to recognize. The function containing this bctr needs\n"
-              << "proper switch pattern recognition." << std::endl;
-    RT_LOG(RT_TAG_RUNTIME) << "Indirect jump context: pc=0x" << std::hex << cpu->pc
-              << " lr=0x" << cpu->lr << " ctr=0x" << cpu->ctr
-              << " r1=0x" << cpu->gpr[1] << " r3=0x" << cpu->gpr[3]
-              << " r29=0x" << cpu->gpr[29] << " r30=0x" << cpu->gpr[30]
-              << " r31=0x" << cpu->gpr[31] << std::dec << std::endl;
-    DumpHostStackTraceForRuntimeHelper();
-    std::fflush(stderr);
-    std::ostringstream message;
-    message << "The game stopped because an indirect jump targeted guest address 0x"
-            << std::hex << target
-            << ", but that address is not a registered translated function.\n\n"
-            << "The translator may have missed a switch/jump-table target.";
-    // Same artifact set as RuntimeCrash::FatalMissingGuestTarget, which is this
-    // path's sibling for indirect *calls*: MarkFatalErrorReported below stops the
-    // atexit reporter, so the crash log has to be written here.
-    RuntimeCrash::WriteCrashArtifacts("missing_jump_target", message.str(), &target);
-    SetRuntimeExitCode(EXIT_FAILURE);
-    ShowRuntimeFatalPopup("Missing indirect jump target", message.str());
-    MarkFatalErrorReported();
-    std::exit(EXIT_FAILURE);
-}
-
-inline void InvokeIndirectCpu(uint32_t target, CpuContext* ctx) {
-    CpuContext* cpu = ctx ? ctx : &GetPersistentCpuContext();
-    if (target == 0) {
-        ReportMissingCpuTarget(target, cpu);
-    }
-    ApplyRuntimeCallOptions(target, cpu);
-    if (TryDispatchRawCpuTarget(TranslatedFunctionRegistry::FindRawByAddressPtr(target), cpu)) {
-        return;
-    }
-    if (TryDispatchResolvedCpuTarget(TranslatedFunctionRegistry::FindByAddressPtr(target), cpu)) {
-        return;
-    }
-    ReportMissingCpuTarget(target, cpu);
-}
+std::optional<TranslatedFunctionInfo> FindNearestTranslatedFunction(uint32_t address);
+// Look up a registered function by host (native) address.
+// This enables stack trace symbolication for translated functions.
+std::optional<TranslatedFunctionInfo> FindTranslatedFunctionByHostAddress(uintptr_t hostAddr);
 
 template <typename Signature>
 class AbiTrampoline;

runtime/include/guest_abi/abi_version.h

@@ -0,0 +1,4 @@
+#pragma once
+
+// Bump whenever guest_abi/, isa/ or ppc_isa_memory.h changes what translated code compiles against.
+#define MKW_GUEST_ABI_VERSION 1

runtime/include/guest_abi/cpu_context.h

@@ -0,0 +1,84 @@
+#pragma once
+
+#include "isa/ppc_isa_fpenv.h"
+
+#include <cstdint>
+
+union PPC_FPR {
+    uint64_t raw;
+    double d;
+    struct {
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+        float ps1; // Low word (Least Significant)
+        float ps0; // High word (Most Significant)
+#else
+        float ps0; // High word
+        float ps1; // Low word
+#endif
+    } paired;
+};
+
+// PowerPC CPU Context
+struct CpuContext {
+    // Standard GPRs
+    uint32_t gpr[32];
+
+    // Special Purpose Registers defined by standard PPC
+    uint32_t cr;        // Condition Register
+    uint32_t lr;        // Link Register
+    uint32_t ctr;       // Count Register
+    uint32_t xer;       // Integer Exception Register
+    uint32_t fpscr;     // Floating-Point Status and Control Register
+
+    // Program State
+    uint32_t pc;        // Program Counter (not used alot tho)
+
+    // Floating Point Registers (Modified for Paired Single support)
+    PPC_FPR  fpr[32];
+
+    // Broadway Specific Extensions
+    uint32_t gqr[8];    // Graphics Quantization Registers
+    uint32_t hid0;      // HID0
+    uint32_t hid1;      // HID1
+    uint32_t hid2;      // HID2
+
+    uint32_t srr0;      // Save/Restore Register 0
+    uint32_t srr1;      // Save/Restore Register 1
+    uint32_t msr;       // Machine State Register
+};
+
+inline thread_local CpuContext* g_currentCpuContext = nullptr;
+
+class CpuContextScope {
+public:
+    explicit CpuContextScope(CpuContext* ctx)
+        : previous_(g_currentCpuContext)
+    {
+        g_currentCpuContext = ctx;
+
+        savedMxcsr_ = MkwGetHostFpControl();
+        if (ctx != nullptr)
+            MkwApplyHostNiMode(ctx->fpscr);
+    }
+
+    ~CpuContextScope()
+    {
+        g_currentCpuContext = previous_;
+        if (previous_ != nullptr)
+            MkwApplyHostNiMode(previous_->fpscr);
+        else
+            MkwRestoreHostMxcsr(savedMxcsr_);
+    }
+
+    CpuContextScope(const CpuContextScope&) = delete;
+    CpuContextScope& operator=(const CpuContextScope&) = delete;
+
+private:
+    CpuContext* previous_ = nullptr;
+    uint32_t savedMxcsr_ = 0;
+};
+
+inline CpuContext* TryGetCpuContext() noexcept
+{
+    return g_currentCpuContext;
+}

runtime/include/guest_abi/dispatch.h

@@ -0,0 +1,618 @@
+#pragma once
+
+#include "guest_abi/cpu_context.h"
+#include "guest_abi/exec_guard.h"
+#include "isa/ppc_isa_config.h"
+#include "isa/ppc_isa_cr.h"
+
+#include <array>
+#include <atomic>
+#include <cstddef>
+#include <cstdint>
+
+namespace RuntimeCrash {
+[[noreturn]] void FatalMissingGuestTarget(uint32_t target, struct CpuContext* cpu) noexcept;
+} // namespace RuntimeCrash
+
+inline void InvokeIndirectCpu(uint32_t target, CpuContext* ctx);
+
+uint32_t FilterScnRendererPathMaskOption(uint32_t pathMask) noexcept;
+
+// Some game-facing runtime options alter arguments at well-defined ABI
+// boundaries. Keep this independent of the dispatch mechanism: generated
+// static calls deliberately bypass InvokeDirectCpu for performance.
+// (used for the path mask filtering in ScnRenderer::createPath: depth of
+// field is always removed, bloom when the user disabled it)
+inline void ApplyRuntimeCallOptions(uint32_t target, CpuContext* ctx) {
+    if (target == 0x8023BD38u) {
+        // ScnRenderer::createPath receives the post-processing path mask in r4.
+        ctx->gpr[4] = FilterScnRendererPathMaskOption(ctx->gpr[4]);
+    }
+}
+
+// Persistent per-thread CPU context used across translated function calls.
+CpuContext& GetPersistentCpuContext();
+
+enum class FunctionKind : uint8_t {
+    BaseTranslated = 0,
+    ModTranslated = 1,
+    Native = 2,
+};
+
+inline constexpr uint32_t kPpcAllNonvolatileFprMask = 0xffffc000u;
+inline constexpr uint32_t kBaseTranslatedFunctionPriority = 0;
+inline constexpr uint32_t kModTranslatedFunctionPriorityBase = 100;
+inline constexpr uint32_t kNativeFunctionPriority = 10000;
+
+struct TranslatedFunctionInfo {
+    uint32_t address = 0;
+    const char* name = "";
+    uint64_t moduleId = 0;
+    uint32_t priority = 0;
+    uint32_t nonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
+    void* entryPoint = nullptr; // Raw typed function pointer for indirect-call resolution
+    void (*rawCpuInvoker)(CpuContext*) = nullptr;
+    bool mustRemainDynamicallyDispatchable = true;
+    FunctionKind kind = FunctionKind::BaseTranslated;
+};
+
+struct RawDispatchRecord {
+    uint32_t address = 0;
+    void (*entry)(CpuContext*) = nullptr;
+    uint32_t nonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
+    bool preserveNonvolatileGprs = false;
+};
+
+// Translator-owned indirect dispatch is split by the high address byte and
+// then by 4 KiB guest pages. Each populated segment stores only the contiguous
+// page range that contains verified function entry points. The final search is
+// bounded to at most the 1024 aligned instruction addresses in one guest page.
+struct StaticIndirectDispatchPage {
+    uint32_t firstEntry = 0;
+    uint16_t entryCount = 0;
+    // Explicit tail padding to an 8-byte stride. The emitter writes it as a
+    // literal third initializer, so it is part of the generated-data contract.
+    uint16_t reserved = 0;
+};
+
+struct StaticIndirectDispatchSegment {
+    const StaticIndirectDispatchPage* pages = nullptr;
+    uint16_t firstPage = 0;
+    uint16_t pageCount = 0;
+};
+
+struct StaticIndirectDispatchTable {
+    const char* profileName = nullptr;
+    const StaticIndirectDispatchSegment* segments = nullptr; // Exactly 256 high-byte segments.
+    const RawDispatchRecord* entries = nullptr;
+    size_t entryCount = 0;
+};
+
+inline const RawDispatchRecord* FindStaticIndirectDispatchEntry(
+    const StaticIndirectDispatchTable* table,
+    uint32_t address) noexcept {
+    if (!table || !table->segments || !table->entries || table->entryCount == 0) {
+        return nullptr;
+    }
+
+    const auto& segment = table->segments[address >> 24];
+    const uint32_t pageNumber = (address >> 12) & 0x0FFFu;
+    if (!segment.pages || pageNumber < segment.firstPage) {
+        return nullptr;
+    }
+    const uint32_t relativePage = pageNumber - segment.firstPage;
+    if (relativePage >= segment.pageCount) {
+        return nullptr;
+    }
+
+    const auto& page = segment.pages[relativePage];
+    const size_t first = page.firstEntry;
+    const size_t count = page.entryCount;
+    if (first > table->entryCount || count > table->entryCount - first) {
+        return nullptr;
+    }
+
+    size_t lower = first;
+    size_t upper = first + count;
+    while (lower < upper) {
+        const size_t middle = lower + (upper - lower) / 2;
+        const uint32_t candidate = table->entries[middle].address;
+        if (candidate < address) {
+            lower = middle + 1;
+        } else {
+            upper = middle;
+        }
+    }
+    return lower < first + count && table->entries[lower].address == address
+        ? &table->entries[lower]
+        : nullptr;
+}
+
+void RegisterStaticIndirectDispatchTable(const StaticIndirectDispatchTable* table);
+inline std::atomic<const StaticIndirectDispatchTable*> g_publishedStaticIndirectDispatchTable{nullptr};
+
+class StaticIndirectDispatchTableRegistrar {
+public:
+    explicit StaticIndirectDispatchTableRegistrar(const StaticIndirectDispatchTable* table) {
+        RegisterStaticIndirectDispatchTable(table);
+    }
+};
+
+inline unsigned PpcLowestSetBitIndex(uint32_t value) noexcept {
+    return static_cast<unsigned>(__builtin_ctz(value));
+}
+
+class PpcNonvolatileFprGuard {
+public:
+    explicit PpcNonvolatileFprGuard(CpuContext* cpu, uint32_t mask = kPpcAllNonvolatileFprMask) noexcept
+        : cpu_(cpu), mask_(mask & kPpcAllNonvolatileFprMask) {
+        if (!cpu_ || mask_ == 0) {
+            return;
+        }
+        // Real masks are a single contiguous register run - f14..f31 for the
+        // conservative default and f23..f31 or similar for a measured writer -
+        // so resolve the run once and copy it as one straight-line block instead
+        // of testing 18 mask bits. Any other mask keeps the per-bit loop, which
+        // saves and restores exactly the same registers.
+        const uint32_t lowest = mask_ & (0u - mask_);
+        const uint32_t aboveRun = mask_ + lowest;
+        if ((aboveRun & (aboveRun - 1u)) == 0u) {
+            const unsigned first = PpcLowestSetBitIndex(mask_);
+            const unsigned end = aboveRun == 0u ? 32u : PpcLowestSetBitIndex(aboveRun);
+            firstSlot_ = static_cast<uint8_t>(first - 14u);
+            slotCount_ = static_cast<uint8_t>(end - first);
+            const size_t count = slotCount_;
+            for (size_t i = 0; i < count; ++i) {
+                saved_[firstSlot_ + i] = cpu_->fpr[first + i];
+            }
+            return;
+        }
+        for (size_t i = 0; i < saved_.size(); ++i) {
+            if ((mask_ & (1u << (14 + i))) == 0) {
+                continue;
+            }
+            saved_[i] = cpu_->fpr[14 + i];
+        }
+    }
+
+    ~PpcNonvolatileFprGuard() noexcept {
+        if (!cpu_ || mask_ == 0) {
+            return;
+        }
+        if (slotCount_ != 0) {
+            const unsigned first = 14u + firstSlot_;
+            const size_t count = slotCount_;
+            for (size_t i = 0; i < count; ++i) {
+                cpu_->fpr[first + i] = saved_[firstSlot_ + i];
+            }
+            return;
+        }
+        for (size_t i = 0; i < saved_.size(); ++i) {
+            if ((mask_ & (1u << (14 + i))) == 0) {
+                continue;
+            }
+            cpu_->fpr[14 + i] = saved_[i];
+        }
+    }
+
+    PpcNonvolatileFprGuard(const PpcNonvolatileFprGuard&) = delete;
+    PpcNonvolatileFprGuard& operator=(const PpcNonvolatileFprGuard&) = delete;
+
+private:
+    CpuContext* cpu_ = nullptr;
+    uint32_t mask_ = 0;
+    // Contiguous-run description of mask_. slotCount_ == 0 means the mask is not
+    // a single run, so the destructor mirrors the constructor's per-bit loop.
+    uint8_t firstSlot_ = 0;
+    uint8_t slotCount_ = 0;
+    // Every element selected by mask_ is written before it is read. Leaving the
+    // remaining slots uninitialized avoids zero-filling all 18 values around a
+    // call that may preserve only one or two architectural FPRs.
+    std::array<PPC_FPR, 18> saved_;
+};
+
+class PpcNonvolatileGprGuard {
+public:
+    explicit PpcNonvolatileGprGuard(CpuContext* cpu, bool enabled = true) noexcept
+        : cpu_(enabled ? cpu : nullptr) {
+        if (!cpu_) {
+            return;
+        }
+        for (size_t i = 0; i < saved_.size(); ++i) {
+            saved_[i] = cpu_->gpr[14 + i];
+        }
+    }
+
+    ~PpcNonvolatileGprGuard() noexcept {
+        if (!cpu_) {
+            return;
+        }
+        for (size_t i = 0; i < saved_.size(); ++i) {
+            cpu_->gpr[14 + i] = saved_[i];
+        }
+    }
+
+    PpcNonvolatileGprGuard(const PpcNonvolatileGprGuard&) = delete;
+    PpcNonvolatileGprGuard& operator=(const PpcNonvolatileGprGuard&) = delete;
+
+private:
+    CpuContext* cpu_ = nullptr;
+    // Left uninitialized: enabled writes all 18 slots before reading them, disabled reads none,
+    // so skipping the zero-fill saves 72 bytes of stack work on the common disabled hot path.
+    std::array<uint32_t, 18> saved_;
+};
+
+// Guest address of OSLoadContext, which models rfi-style context restoration
+// and intentionally replaces the full guest register file instead of returning
+// like a normal ABI call. It can never be guarded.
+inline constexpr uint32_t kOSLoadContextAddress = 0x801A1F58u;
+
+inline bool ShouldPreserveNonvolatileGprsForRawCpuCall(const TranslatedFunctionInfo* info) noexcept {
+    return info->kind == FunctionKind::Native && info->address != kOSLoadContextAddress;
+}
+
+inline uint32_t NonvolatileFprGuardMaskFor(const TranslatedFunctionInfo* info) noexcept {
+    if (!info || info->nonvolatileFprWriteMask == 0) {
+        return 0;
+    }
+    return info->nonvolatileFprWriteMask & kPpcAllNonvolatileFprMask;
+}
+
+template <uint32_t Target>
+struct KnownTranslatedCpuCall {
+    static constexpr bool kAvailable = false;
+    static constexpr uint32_t kNonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
+    static constexpr bool kMustRemainDynamicallyDispatchable = true;
+    static constexpr void (*Entry)(CpuContext*) = nullptr;
+};
+
+// Translator-emitted trait specializations, one macro shared by every generated shard.
+// `addr` is the 8-digit hex entry point, `winner` the resolved symbol (mods/Retro Rewind can
+// publish their own name). A specialization only exists for a single statically bound winner,
+// so overridable/dynamically-dispatchable are hardcoded here rather than passed in.
+#define MKW_TRANSLATED_TRAIT(addr, winner, nonvolatile_fpr_write_mask)                    \
+    extern "C" void winner(CpuContext* ctx);                                              \
+    template <>                                                                           \
+    struct KnownTranslatedCpuCall<0x##addr##u> {                                          \
+        static constexpr bool kAvailable = true;                                          \
+        static constexpr uint32_t kNonvolatileFprWriteMask = nonvolatile_fpr_write_mask;  \
+        static constexpr bool kMustRemainDynamicallyDispatchable = false;                 \
+        static constexpr void (*Entry)(CpuContext*) = &winner;                            \
+    }
+
+template <uint32_t Target>
+struct KnownNativeCpuCall {
+    static constexpr bool kAvailable = false;
+    static constexpr uint32_t kNonvolatileFprWriteMask = kPpcAllNonvolatileFprMask;
+    static constexpr void (*Entry)(CpuContext*) = nullptr;
+};
+
+template <uint32_t Target>
+struct KnownTypedNativeCpuCall {
+    static constexpr bool kAvailable = false;
+};
+
+#include "guest_abi/native_cpu_calls.inc"
+
+// Direct-mapped thread-local memo in front of the sorted registry lookup every bctrl performs;
+// a miss just re-runs the sorted lookup. 512 entries (12 KiB) covers the per-frame indirect
+// working set while staying L1/L2 resident, unlike 4096 which would thrash L2. Must stay a
+// power of two, the index masks with (size - 1).
+inline constexpr size_t kIndirectDispatchCacheEntries = 512;
+
+// Namespace-scope `inline thread_local`, not function-local `static thread_local`, to avoid a
+// thread-static init epoch check on every bctrl (same as g_currentCpuContext in guest_abi/cpu_context.h).
+struct IndirectResolvedDispatchMemoEntry {
+    bool valid;
+    uint32_t address;
+    const TranslatedFunctionInfo* info;
+};
+inline thread_local IndirectResolvedDispatchMemoEntry
+    g_indirectResolvedDispatchMemo[kIndirectDispatchCacheEntries]{};
+
+struct IndirectRawDispatchMemoEntry {
+    bool valid;
+    uint32_t address;
+    const RawDispatchRecord* record;
+};
+inline thread_local IndirectRawDispatchMemoEntry
+    g_indirectRawDispatchMemo[kIndirectDispatchCacheEntries]{};
+
+class TranslatedFunctionRegistry {
+public:
+    static void Register(TranslatedFunctionInfo info);
+    static void Finalize();
+    static inline bool IsLookupPublished() noexcept {
+        return lookupPublished_.load(std::memory_order_acquire);
+    }
+    static inline const TranslatedFunctionInfo* FindByAddressPtr(uint32_t address) {
+        if (!lookupPublished_.load(std::memory_order_acquire)) {
+            return FindByAddressPtrSlow(address);
+        }
+
+        auto& cached = g_indirectResolvedDispatchMemo[
+            ((address >> 2) * 2654435761u) & (kIndirectDispatchCacheEntries - 1)];
+        if (cached.valid && cached.address == address) {
+            return cached.info;
+        }
+
+        const auto* info = FindByAddressPtrSlow(address);
+        cached = {true, address, info};
+        return info;
+    }
+
+    // Hot wrapper: only the direct-mapped memo probe stays inline. Inlining the
+    // generated-table binary search alongside it pushed the whole function past
+    // the inliner's budget and out-of-lined the probe itself, which is the part
+    // every bctrl executes.
+    static MKW_PPC_FORCE_INLINE const RawDispatchRecord* FindRawByAddressPtr(uint32_t address) {
+        if (!lookupPublished_.load(std::memory_order_acquire)) {
+            return nullptr;
+        }
+
+        auto& cached = g_indirectRawDispatchMemo[
+            ((address >> 2) * 2654435761u) & (kIndirectDispatchCacheEntries - 1)];
+        if (cached.valid && cached.address == address) {
+            return cached.record;
+        }
+
+        return FindRawByAddressPtrMiss(address, cached);
+    }
+
+private:
+    // Memo miss: resolve against the header-inline generated table first; only dynamically
+    // registered records or a not-yet-published table fall through to the slow path.
+    static MKW_PPC_NO_INLINE MKW_PPC_COLD const RawDispatchRecord* FindRawByAddressPtrMiss(
+        uint32_t address, IndirectRawDispatchMemoEntry& cached) {
+        const RawDispatchRecord* record = nullptr;
+        if (const auto* table =
+                g_publishedStaticIndirectDispatchTable.load(std::memory_order_acquire)) {
+            record = FindStaticIndirectDispatchEntry(table, address);
+        }
+        if (record == nullptr) {
+            record = FindRawByAddressPtrSlow(address);
+        }
+        cached = {true, address, record};
+        return record;
+    }
+
+    static const TranslatedFunctionInfo* FindByAddressPtrSlow(uint32_t address);
+    static const RawDispatchRecord* FindRawByAddressPtrSlow(uint32_t address);
+
+    // Finalize publishes immutable registry, generated-table, and dynamic raw
+    // dispatch storage. Header-inlined cache hits are enabled only after that
+    // release publication, so a pre-finalization miss can never poison them.
+    inline static std::atomic_bool lookupPublished_{false};
+};
+
+// Cold, profile-owned registration data is emitted in compact table shards.
+// Keeping it out of generated function bodies avoids one static constructor
+// per translated function.
+struct BulkTranslatedFunctionRecord {
+    uint32_t address;
+    const char* name;
+    void (*entry)(CpuContext*);
+    FunctionKind kind;
+    bool preservesNonvolatileFprs;
+    uint32_t nonvolatileFprWriteMask;
+    uint32_t priority;
+    uint64_t moduleId;
+    bool mustRemainDynamicallyDispatchable;
+};
+
+void RegisterBulkTranslatedFunctions(const BulkTranslatedFunctionRecord* records, size_t count);
+
+class BulkTranslatedFunctionRegistrar {
+public:
+    BulkTranslatedFunctionRegistrar(const BulkTranslatedFunctionRecord* records, size_t count) {
+        RegisterBulkTranslatedFunctions(records, count);
+    }
+};
+
+MKW_PPC_FORCE_INLINE bool TryDispatchResolvedCpuTarget(const TranslatedFunctionInfo* info, CpuContext* cpu) {
+    if (!info || !info->rawCpuInvoker) {
+        return false;
+    }
+
+    RecompMod::ScopedTranslatedExecutionAddress translatedExecution(info->address);
+    PpcNonvolatileFprGuard fprGuard(cpu, NonvolatileFprGuardMaskFor(info));
+    PpcNonvolatileGprGuard gprGuard(cpu, ShouldPreserveNonvolatileGprsForRawCpuCall(info));
+    if (TryGetCpuContext() != cpu) {
+        CpuContextScope scope(cpu);
+        info->rawCpuInvoker(cpu);
+        return true;
+    }
+    info->rawCpuInvoker(cpu);
+    return true;
+}
+
+inline bool TryDispatchRawCpuTarget(const RawDispatchRecord* record, CpuContext* cpu) {
+    if (!record || !record->entry) {
+        return false;
+    }
+    RecompMod::ScopedTranslatedExecutionAddress translatedExecution(record->address);
+    PpcNonvolatileGprGuard gprGuard(cpu, record->preserveNonvolatileGprs);
+    PpcNonvolatileFprGuard fprGuard(cpu, record->nonvolatileFprWriteMask);
+    if (TryGetCpuContext() != cpu) {
+        CpuContextScope scope(cpu);
+        record->entry(cpu);
+    } else {
+        record->entry(cpu);
+    }
+    return true;
+}
+
+template <uint32_t Target>
+inline void DispatchKnownTranslatedCpuTargetStatic(CpuContext* cpu) {
+    const auto invokeKnownTranslated = [&]() {
+        KnownTranslatedCpuCall<Target>::Entry(cpu);
+    };
+
+    // A statically resolved, non-overridable translated call already running
+    // in this CpuContext needs no registry, context, or diagnostic target
+    // transition. This is the normal generated-to-generated hot path.
+    if (TryGetCpuContext() == cpu) {
+        if constexpr (KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask == 0) {
+            invokeKnownTranslated();
+            return;
+        } else {
+            PpcNonvolatileFprGuard fprGuard(
+                cpu, KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask);
+            invokeKnownTranslated();
+            return;
+        }
+    }
+
+    CpuContextScope contextScope(cpu);
+    RecompMod::ScopedTranslatedExecutionAddress translatedExecution(Target);
+    if constexpr (KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask == 0) {
+        invokeKnownTranslated();
+        return;
+    } else {
+        PpcNonvolatileFprGuard fprGuard(cpu, KnownTranslatedCpuCall<Target>::kNonvolatileFprWriteMask);
+        invokeKnownTranslated();
+    }
+}
+
+[[noreturn]] inline void ReportMissingCpuTarget(uint32_t target, CpuContext* cpu) {
+    // Centralized crash path: stderr diagnostics, heuristics, crash artifacts
+    // in the run's log folder, popup, exit.
+    RuntimeCrash::FatalMissingGuestTarget(target, cpu);
+}
+
+template <uint32_t Target>
+inline const TranslatedFunctionInfo* ResolveDirectCpuTargetInfo() {
+    return TranslatedFunctionRegistry::FindByAddressPtr(Target);
+}
+
+template <uint32_t Target>
+inline const TranslatedFunctionInfo* ResolveDirectCpuTargetInfoCached() {
+    // Only latch a result once the registry is published (immutable for the process lifetime);
+    // a pre-finalization miss or provisional winner must never be cached. 0 = not yet resolved,
+    // 1 = negative-cache sentinel, avoiding a second ready flag on the hot path.
+    static std::atomic<uintptr_t> cachedValue{0};
+    constexpr uintptr_t kNegativeCache = 1;
+    const uintptr_t cached = cachedValue.load(std::memory_order_acquire);
+    if (cached != 0) {
+        return cached == kNegativeCache
+            ? nullptr
+            : reinterpret_cast<const TranslatedFunctionInfo*>(cached);
+    }
+    if (!TranslatedFunctionRegistry::IsLookupPublished()) {
+        return ResolveDirectCpuTargetInfo<Target>();
+    }
+    const auto* info = ResolveDirectCpuTargetInfo<Target>();
+    cachedValue.store(info != nullptr
+                          ? reinterpret_cast<uintptr_t>(info)
+                          : kNegativeCache,
+                      std::memory_order_release);
+    return info;
+}
+
+// Caches the "is the published winner still base translation?" verdict for state-free fast
+// paths guarding tens of thousands of call sites. Same latching rule as above: 0 = unresolved,
+// 1 = negative, 2 = positive, and a verdict computed pre-publication is never stored.
+template <uint32_t Target>
+inline bool IsBaseTranslatedCpuTargetActive() {
+    static std::atomic<uint8_t> cachedVerdict{0};
+    constexpr uint8_t kNegativeVerdict = 1;
+    constexpr uint8_t kPositiveVerdict = 2;
+    const uint8_t cached = cachedVerdict.load(std::memory_order_acquire);
+    if (cached != 0) {
+        return cached == kPositiveVerdict;
+    }
+    const bool publishable = TranslatedFunctionRegistry::IsLookupPublished();
+    const auto* info = ResolveDirectCpuTargetInfo<Target>();
+    const bool active = info != nullptr && info->kind == FunctionKind::BaseTranslated;
+    if (publishable) {
+        cachedVerdict.store(active ? kPositiveVerdict : kNegativeVerdict,
+                            std::memory_order_release);
+    }
+    return active;
+}
+
+template <uint32_t Target>
+inline void InvokeDirectCpu(CpuContext* ctx) {
+    static_assert(Target != 0, "InvokeDirectCpu cannot target address 0");
+    CpuContext* cpu = ctx ? ctx : &GetPersistentCpuContext();
+    ApplyRuntimeCallOptions(Target, cpu);
+    if constexpr (KnownNativeCpuCall<Target>::kAvailable) {
+        const auto invokeKnownNative = [&]() {
+            PpcNonvolatileGprGuard gprGuard(cpu);
+            if (TryGetCpuContext() != cpu) {
+                CpuContextScope scope(cpu);
+                KnownNativeCpuCall<Target>::Entry(cpu);
+                return;
+            }
+            KnownNativeCpuCall<Target>::Entry(cpu);
+        };
+        if constexpr (KnownNativeCpuCall<Target>::kNonvolatileFprWriteMask == 0) {
+            invokeKnownNative();
+        } else {
+            PpcNonvolatileFprGuard fprGuard(cpu, KnownNativeCpuCall<Target>::kNonvolatileFprWriteMask);
+            invokeKnownNative();
+        }
+        return;
+    }
+
+    if constexpr (KnownTypedNativeCpuCall<Target>::kAvailable) {
+        PpcNonvolatileGprGuard gprGuard(cpu);
+        if (TryGetCpuContext() != cpu) {
+            CpuContextScope scope(cpu);
+            KnownTypedNativeCpuCall<Target>::Invoke(cpu);
+        } else {
+            KnownTypedNativeCpuCall<Target>::Invoke(cpu);
+        }
+        return;
+    }
+
+    if constexpr (KnownTranslatedCpuCall<Target>::kAvailable) {
+        DispatchKnownTranslatedCpuTargetStatic<Target>(cpu);
+        return;
+    }
+
+    const auto* registeredInfo = ResolveDirectCpuTargetInfoCached<Target>();
+    if (TryDispatchResolvedCpuTarget(registeredInfo, cpu)) {
+        return;
+    }
+
+    ReportMissingCpuTarget(Target, cpu);
+}
+
+// Cold tail of InvokeIndirectJump: a bctr target that is not a registered function.
+[[noreturn]] MKW_PPC_NO_INLINE MKW_PPC_COLD void ReportMissingIndirectJumpTarget(uint32_t target, CpuContext* cpu);
+
+// Indirect jump (bctr without link) - used for switch tables.
+inline void InvokeIndirectJump(uint32_t target, CpuContext* ctx) {
+    // Use the caller's context, not the persistent global context!
+    // This is essential for tail calls via bctr where the 'this' pointer (r3)
+    // must be passed correctly to the target function.
+    CpuContext* cpu = ctx ? ctx : &GetPersistentCpuContext();
+    if (TryDispatchRawCpuTarget(TranslatedFunctionRegistry::FindRawByAddressPtr(target), cpu)) {
+        return;
+    }
+
+    // Cold fallback for dynamically registered or signature-only targets.
+    const auto* info = TranslatedFunctionRegistry::FindByAddressPtr(target);
+    if (info) {
+        if (TryDispatchResolvedCpuTarget(info, cpu)) {
+            return;
+        }
+    }
+    ReportMissingIndirectJumpTarget(target, cpu);
+}
+
+inline void InvokeIndirectCpu(uint32_t target, CpuContext* ctx) {
+    CpuContext* cpu = ctx ? ctx : &GetPersistentCpuContext();
+    if (target == 0) {
+        ReportMissingCpuTarget(target, cpu);
+    }
+    ApplyRuntimeCallOptions(target, cpu);
+    if (TryDispatchRawCpuTarget(TranslatedFunctionRegistry::FindRawByAddressPtr(target), cpu)) {
+        return;
+    }
+    if (TryDispatchResolvedCpuTarget(TranslatedFunctionRegistry::FindByAddressPtr(target), cpu)) {
+        return;
+    }
+    ReportMissingCpuTarget(target, cpu);
+}

runtime/include/guest_abi/exec_guard.h

@@ -0,0 +1,48 @@
+#pragma once
+
+#include <atomic>
+#include <cstdint>
+
+namespace RecompMod {
+
+// Diagnostic-only, read by fatal reporters (unmapped access, executable-write, CPU dumps).
+// Defined here as `inline thread_local` with a constant initializer, not `extern thread_local`
+// in the .cpp: every indirect dispatch scopes this, so an out-of-line ctor/dtor would cost two
+// un-inlinable calls plus a register spill each for three instructions of work.
+inline thread_local uint32_t g_currentTranslatedExecutionAddress = 0;
+
+class ScopedTranslatedExecutionAddress {
+public:
+    explicit ScopedTranslatedExecutionAddress(uint32_t address) noexcept
+        : previous_(g_currentTranslatedExecutionAddress) {
+        // Address 0 means "nothing new to report" - the enclosing scope's value
+        // stays visible, and the unconditional restore below keeps nesting exact.
+        if (address != 0) {
+            g_currentTranslatedExecutionAddress = address;
+        }
+    }
+
+    ~ScopedTranslatedExecutionAddress() noexcept {
+        g_currentTranslatedExecutionAddress = previous_;
+    }
+
+    ScopedTranslatedExecutionAddress(const ScopedTranslatedExecutionAddress&) = delete;
+    ScopedTranslatedExecutionAddress& operator=(const ScopedTranslatedExecutionAddress&) = delete;
+
+private:
+    uint32_t previous_ = 0;
+};
+
+inline constexpr uint32_t kExecutableWriteGuardPageShift = 12;
+inline constexpr uint32_t kExecutableWriteGuardPageCount = 1u << (32 - kExecutableWriteGuardPageShift);
+inline constexpr uint32_t kExecutableWriteGuardCoarsePageShift = 20;
+inline constexpr uint32_t kExecutableWriteGuardCoarsePageCount = 1u << (32 - kExecutableWriteGuardCoarsePageShift);
+inline constexpr uint32_t kExecutableWriteGuardMidPageShift = 16;
+inline constexpr uint32_t kExecutableWriteGuardMidPageCount = 1u << (32 - kExecutableWriteGuardMidPageShift);
+
+extern std::atomic<bool> g_executableWriteGuardEnabled;
+extern std::atomic<uint8_t> g_executableWriteGuardPages[kExecutableWriteGuardPageCount];
+extern std::atomic<uint8_t> g_executableWriteGuardCoarsePages[kExecutableWriteGuardCoarsePageCount];
+extern std::atomic<uint8_t> g_executableWriteGuardMidPages[kExecutableWriteGuardMidPageCount];
+
+} // namespace RecompMod

runtime/include/guest_abi/flat_memory.h

@@ -0,0 +1,59 @@
+#pragma once
+
+// Flat 4 GiB guest address space constants used by translated accesses; see guest_flat_memory.h.
+#include <cstddef>
+#include <cstdint>
+
+namespace GuestFlat {
+
+// Fixed base so the emitted access is `[reg + imm64-in-register]` with no load
+// of a global.
+inline constexpr uint64_t kGuestSpaceSize = 0x1'0000'0000ull;
+inline constexpr size_t kGuestPageSize = 0x1000;
+#if defined(__x86_64__)
+// 16 TiB: clear of the Windows ASan shadow (32 TiB) and of the usual image/heap
+// placement.
+inline constexpr uintptr_t kFixedFlatGuestBase = 0x0000'1000'0000'0000ull;
+#elif defined(__aarch64__) && defined(__APPLE__)
+// Keep this well above the low address ranges that Darwin's ASLR may use for
+// a PIE executable and its shared cache. Apple Silicon's user VA is wider
+// than Linux's 39-bit minimum, so this 512 GiB region is available while the
+// Linux AArch64 target retains its 64 GiB placement below.
+inline constexpr uintptr_t kFixedFlatGuestBase = 0x0000'0080'0000'0000ull;
+#elif defined(__aarch64__)
+// 16 TiB (this arch's x86_64 sibling value) is unreachable on any AArch64
+// kernel configured for 39-bit virtual addresses (512 GiB ceiling) - common on
+// older/embedded targets (e.g. this project's own tested Jetson/L4T board, kernel
+// 4.9). There mmap() silently ignores a hint above the ceiling and hands back an
+// address near the top of the real range instead, which this module's caller
+// then rejects as "already occupied". 64 GiB is reachable on every AArch64 VA
+// width in real use (39-bit minimum and up) and was confirmed via direct mmap
+// probing to sit far below where the PIE image, heap, shared libraries and
+// stack actually land (all clustered above ~340 GiB on a 39-bit/512 GiB system).
+inline constexpr uintptr_t kFixedFlatGuestBase = 0x0000'0010'0000'0000ull;
+#else
+#error "guest_abi/flat_memory.h has no fixed flat guest base chosen for this architecture"
+#endif
+
+#define MKW_FLAT_GUEST_BASE (reinterpret_cast<uint8_t*>(GuestFlat::kFixedFlatGuestBase))
+
+// True when a host VM page covers more than one 4 KiB Wii page. In that
+// configuration, guest-view page protection cannot safely represent per-Wii-
+// page MMIO, deferred-read, or executable-write state, so general translated
+// accesses must use the checked Memory::* path.
+// Windows user mode and x86-64 always use a 4 KiB base page, so those builds
+// fold this to a compile-time false: it appears in every flat access and must
+// not become a hot-path load. Only AArch64, where the page size is a kernel
+// configuration (4/16/64 KiB), has to probe it at runtime.
+#if defined(_WIN32) || defined(__x86_64__)
+#define MKW_GUEST_FLAT_FIXED_PAGE_SIZE 1
+#endif
+
+#if defined(MKW_GUEST_FLAT_FIXED_PAGE_SIZE)
+inline constexpr bool RequiresCheckedAccess() noexcept { return false; }
+#else
+extern bool g_requiresCheckedAccess;
+inline bool RequiresCheckedAccess() noexcept { return g_requiresCheckedAccess; }
+#endif
+
+} // namespace GuestFlat

runtime/include/guest_abi/guest_abi.h

@@ -0,0 +1,19 @@
+#pragma once
+
+// The only header translated code includes. Everything reachable from here is the
+// guest ABI; runtime-only headers must never be pulled in.
+#include "guest_abi/abi_version.h"
+
+#include "isa/ppc_isa_config.h"
+#include "isa/big_endian.h"
+#include "isa/ppc_isa_fpenv.h"
+#include "guest_abi/cpu_context.h"
+#include "isa/ppc_isa_int.h"
+#include "isa/ppc_isa_float.h"
+#include "guest_abi/flat_memory.h"
+#include "guest_abi/exec_guard.h"
+#include "guest_abi/memory.h"
+#include "isa/ppc_isa_quantized.h"
+#include "isa/ppc_isa_cr.h"
+#include "guest_abi/guest_cr.h"
+#include "guest_abi/dispatch.h"

runtime/include/guest_abi/guest_abi_pch.h

@@ -0,0 +1,7 @@
+#pragma once
+
+#include <cstdint>
+#include <cstring>
+#include <cmath>
+
+#include "guest_abi/guest_abi.h"

runtime/include/guest_abi/guest_cr.h

@@ -0,0 +1,66 @@
+#pragma once
+#include "guest_abi/cpu_context.h"
+
+#include <cmath>
+#include <cstdint>
+
+// Condition Register Fields
+#define CR_LT 0
+#define CR_GT 1
+#define CR_EQ 2
+#define CR_SO 3
+
+[[noreturn]] void PPC_Undefined(uint32_t pc, uint32_t rawInstruction, const char* details);
+
+// Used by generated code when the translator encounters privileged/unmodeled PPC instructions
+// (e.g. rfi). This is intentionally loud so missing HLE hooks are easy to find.
+#define UNDEFINED(pc, rawInstruction, details) PPC_Undefined((pc), (rawInstruction), (details))
+
+// Helper to set CR bits (Signed)
+inline void SetCR(CpuContext* cpu, int field, int32_t a, int32_t b) {
+    uint32_t crField = 0;
+    if (a < b) crField |= 0x8; // LT
+    if (a > b) crField |= 0x4; // GT
+    if (a == b) crField |= 0x2; // EQ
+    crField |= (cpu->xer >> 31) & 1u; // SO
+    
+    int shift = (7 - field) * 4;
+    uint32_t mask = 0xF << shift;
+    cpu->cr = (cpu->cr & ~mask) | (crField << shift);
+}
+
+// Helper to set CR bits (Unsigned)
+inline void SetCR(CpuContext* cpu, int field, uint32_t a, uint32_t b) {
+    uint32_t crField = 0;
+    if (a < b) crField |= 0x8; // LT
+    if (a > b) crField |= 0x4; // GT
+    if (a == b) crField |= 0x2; // EQ
+    crField |= (cpu->xer >> 31) & 1u; // SO
+    
+    int shift = (7 - field) * 4;
+    uint32_t mask = 0xF << shift;
+    cpu->cr = (cpu->cr & ~mask) | (crField << shift);
+}
+
+// Helper to set CR bits (Floating-point)
+inline void SetCRFloat(CpuContext* cpu, int field, double a, double b) {
+    uint32_t crField = 0;
+    if (std::isnan(a) || std::isnan(b)) {
+        // Unordered: LT/GT/EQ clear, SO set.
+        crField = 0x1;
+    } else {
+        if (a < b) crField |= 0x8; // LT
+        if (a > b) crField |= 0x4; // GT
+        if (a == b) crField |= 0x2; // EQ
+    }
+
+    int shift = (7 - field) * 4;
+    uint32_t mask = 0xF << shift;
+    cpu->cr = (cpu->cr & ~mask) | (crField << shift);
+}
+
+// Helper to get CR bit
+inline bool GetCRBit(CpuContext* cpu, int field, int bit) {
+    int shift = (7 - field) * 4 + (3 - bit);
+    return (cpu->cr >> shift) & 1;
+}

runtime/include/guest_abi/memory.h

@@ -0,0 +1,114 @@
+#pragma once
+
+#include <cstddef>
+#include <cstdint>
+#include <stdexcept>
+#include <string>
+#include <string_view>
+#include <vector>
+
+class Memory {
+public:
+    using DeferredReadCallback = bool (*)(void* user);
+
+    static constexpr size_t kMem1Size = 24u * 1024u * 1024u;
+    static constexpr size_t kMem2Size = 128u * 1024u * 1024u;
+    static constexpr uint32_t kMem1PhysicalBase = 0x00000000u;
+    static constexpr uint32_t kMem1CachedBase = 0x80000000u;
+    static constexpr uint32_t kMem1UncachedBase = 0xC0000000u;
+    static constexpr uint32_t kMem2PhysicalBase = 0x10000000u;
+    static constexpr uint32_t kMem2CachedBase = 0x90000000u;
+    static constexpr uint32_t kMem2UncachedBase = 0xD0000000u;
+    static constexpr uint32_t kMem2PhysicalEnd =
+        kMem2PhysicalBase + static_cast<uint32_t>(kMem2Size);
+    static constexpr uint32_t kMem2CachedEnd =
+        kMem2CachedBase + static_cast<uint32_t>(kMem2Size);
+    static constexpr uint32_t kMem2UncachedEnd =
+        kMem2UncachedBase + static_cast<uint32_t>(kMem2Size);
+
+    struct RegionConfig {
+        std::string name;
+        uint32_t baseAddress = 0;
+        size_t sizeBytes = 0;
+    };
+
+    struct Config {
+        std::vector<RegionConfig> regions;
+        static Config WiiDefaults();
+    };
+
+    class AccessViolation : public std::runtime_error {
+    public:
+        AccessViolation(uint32_t address, size_t length, std::string_view reason);
+
+        uint32_t address() const noexcept { return address_; }
+        size_t length() const noexcept { return length_; }
+        std::string_view reason() const noexcept { return reason_; }
+
+    private:
+        uint32_t address_ = 0;
+        size_t length_ = 0;
+        std::string reason_;
+    };
+
+    static void Init(const Config& config);
+    // Single flat MEM1 region. Not used by the shipped runtime, but the
+    // translator integration-test harnesses emit calls to it.
+    static void Init(size_t mem1Size);
+    static void Reset();
+    // Executable ranges are normally registered during startup. Rebuild the
+    // writable fast-path classification after each registration so a page
+    // previously classified as ordinary data cannot retain a stale direct
+    // write bias.
+    static void RefreshWritableFastPathsForExecutableRanges();
+
+    static uint8_t Read8(uint32_t addr);
+    static uint16_t Read16(uint32_t addr);
+    static uint32_t Read32(uint32_t addr);
+    static uint64_t Read64(uint32_t addr);
+    static float ReadFloat32(uint32_t addr);
+    static double ReadFloat64(uint32_t addr);
+    static void Write8(uint32_t addr, uint8_t val);
+    static void Write16(uint32_t addr, uint16_t val);
+    static void Write32(uint32_t addr, uint32_t val);
+    static void Write64(uint32_t addr, uint64_t val);
+    static void WriteFloat32(uint32_t addr, double val);
+    static void WriteFloat64(uint32_t addr, double val);
+
+    // Exception-safe scalar access for HLE code. These keep Read32/Write32's
+    // full mapping behavior and only convert an unmapped address into a failure
+    // result; they are deliberately not MemoryInline::Try*GuestScalar, which is
+    // the translated-code fast path over the page table.
+    static bool TryRead32(uint32_t addr, uint32_t& value) noexcept {
+        try {
+            value = Read32(addr);
+            return true;
+        } catch (const AccessViolation&) {
+            return false;
+        }
+    }
+
+    static bool TryWrite32(uint32_t addr, uint32_t value) noexcept {
+        try {
+            Write32(addr, value);
+            return true;
+        } catch (const AccessViolation&) {
+            return false;
+        }
+    }
+
+    static uint8_t* GetPointer(uint32_t addr);
+    static uint8_t* GetPointer(uint32_t addr, size_t length);
+    static bool Contains(uint32_t addr, size_t length = 1);
+
+    static uint64_t RegisterDeferredRead(uint32_t addr, size_t length,
+                                         DeferredReadCallback callback, void* user);
+    static void ClearDeferredReads();
+
+    // sizeBytes reports the storage actually allocated, which is not always the
+    // configured size (aliased MEM1/MEM2 windows are clamped to what is behind
+    // them). Used by the crash dump.
+    static std::vector<RegionConfig> DescribeRegions();
+};
+
+#include "guest_abi/memory_access.h"

runtime/include/guest_abi/memory_access.h

@@ -0,0 +1,732 @@
+#pragma once
+
+#include "guest_abi/exec_guard.h"
+#include "guest_abi/flat_memory.h"
+#include "guest_abi/memory.h"
+
+#include <cstddef>
+#include <cstdint>
+#include <cstring>
+
+// Hooks for GX HLE FIFO handling
+extern "C" {
+    void GX_HLE_FIFO_WriteFloat(float val);
+    void GX_HLE_FIFO_Write32(uint32_t val);
+    void GX_HLE_FIFO_Write16(uint16_t val);
+    void GX_HLE_FIFO_Write8(uint8_t val);
+    void GX_HLE_FIFO_WriteBurst(const uint8_t* data, uint32_t sizeBytes);
+}
+
+namespace MemoryInline {
+#if defined(_WIN32)
+#define MKW_MEMORY_FORCE_INLINE __forceinline
+#define MKW_MEMORY_NO_INLINE __declspec(noinline)
+#else
+// See runtime/include/isa/ppc_isa_config.h for why non-Windows Clang needs the GNU-attribute
+// spellings instead of the MS-extension keywords.
+#define MKW_MEMORY_FORCE_INLINE __attribute__((always_inline)) inline
+#define MKW_MEMORY_NO_INLINE __attribute__((noinline))
+#endif
+#define MKW_MEMORY_COLD __attribute__((cold))
+inline constexpr uint32_t kPageShift = 20;
+inline constexpr uint32_t kPageSize = 1u << kPageShift;
+inline constexpr uint32_t kPageMask = kPageSize - 1u;
+inline constexpr uint32_t kPageCount = 1u << (32 - kPageShift);
+inline constexpr uint32_t kMaxFastScalarSize = 8;
+inline constexpr uint32_t kWritableSubPageShift = RecompMod::kExecutableWriteGuardPageShift;
+inline constexpr uint32_t kWritableSubPageSize = 1u << kWritableSubPageShift;
+inline constexpr uint32_t kWritableSubPagesPerPage = kPageSize / kWritableSubPageSize;
+
+struct PageEntry {
+    uint8_t* base = nullptr;
+    uint32_t limit = 0;
+};
+
+extern PageEntry g_pageTable[kPageCount];
+// Encoded (host page base - guest page base) + 1 for full pages whose next
+// page is contiguous. This permits any native access up to 8 bytes without a
+// per-access mask/limit check. Zero retains the general PageEntry fallback.
+extern uintptr_t g_fullPageBias[kPageCount];
+// Runtime-active readable biases. Deferred-read pages clear these entries once
+// instead of paying a mode branch on every translated read.
+extern uintptr_t g_fullReadablePageBias[kPageCount];
+// Same encoding, but only for pages proven not to contain executable bytes.
+// Executable-range registration invalidates entries before guest execution.
+extern uintptr_t g_fullWritablePageBias[kPageCount];
+
+// Allocated only for mapped 1 MiB pages that contain both executable and data
+// 4 KiB pages. Entries use the same encoded host bias as the coarse table.
+// Exact executable bits remain authoritative and are checked at lookup time,
+// including both sides of a cross-4-KiB access.
+struct SparseWritablePageTable {
+    uintptr_t encodedBias[kWritableSubPagesPerPage]{};
+};
+extern const SparseWritablePageTable* g_sparseWritablePageTables[kPageCount];
+
+// Nonzero while any registered deferred read overlaps the page. Small
+// mappings (e.g. the 16 KiB locked cache) have no coarse bias entry, so a
+// zero readable bias alone cannot distinguish "deferred content pending"
+// from "small but plain memory"; range resolution needs the explicit flag.
+extern uint8_t g_deferredReadCoveredPages[kPageCount];
+
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint8_t Read8Slow(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint16_t Read16Slow(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint32_t Read32Slow(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint64_t Read64Slow(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD float ReadFloat32Slow(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD double ReadFloat64Slow(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write8Slow(uint32_t addr, uint8_t val);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write16Slow(uint32_t addr, uint16_t val);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write32Slow(uint32_t addr, uint32_t val);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write64Slow(uint32_t addr, uint64_t val);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteFloat32Slow(uint32_t addr, double val);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteFloat64Slow(uint32_t addr, double val);
+template <typename T>
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD T ReadResolvedFallback(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD float ReadResolvedFallbackFloat32(uint32_t addr);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD double ReadResolvedFallbackFloat64(uint32_t addr);
+template <typename T>
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteResolvedFallback(uint32_t addr, T value);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteResolvedFallbackFloat32(uint32_t addr, double val);
+MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteResolvedFallbackFloat64(uint32_t addr, double val);
+bool ResolveDeferredReads(uint32_t addr, size_t length);
+
+constexpr bool IsMmioAddress(uint32_t addr) {
+    return addr >= 0xCC000000u && addr < 0xCE000000u;
+}
+
+constexpr bool IsGpuFifoAddress(uint32_t addr) {
+    return addr >= 0xCC008000u && addr < 0xCC008100u;
+}
+
+// Page protections can't cover this: an MMIO write must reach GX HLE with its value or be
+// reported, and a fault record can't carry the value, so this mask/compare sits in front of
+// every flat store instead.
+MKW_MEMORY_FORCE_INLINE bool FlatWriteNeedsPolicy(uint32_t address) {
+    return (address & 0xFE000000u) == 0xCC000000u;  // 0xCC000000..0xCDFFFFFF
+}
+
+// Gekko stfs conversion is a bit-level narrowing operation.  In particular it
+// does not behave like a host double-to-float cast for values which were left
+// in double precision, and it has hardware-tested handling for tiny values.
+MKW_MEMORY_FORCE_INLINE uint32_t ConvertPpcDoubleToSingleBits(double value) {
+    uint64_t bits = 0;
+    std::memcpy(&bits, &value, sizeof(bits));
+    const uint32_t exponent = static_cast<uint32_t>((bits >> 52) & 0x7FFu);
+    // The subnormal-single arm applies to exactly the exponents 874..896; every
+    // other exponent takes the plain sign/exponent/fraction narrowing below.
+    // Inside that window the exponent field is nonzero, so the magnitude cannot
+    // be zero and needs no separate test - the zero case (exponent 0) reaches
+    // the narrowing exactly as it did when the two arms shared that test.
+    if (exponent - 874u <= 22u) [[unlikely]]
+    {
+        uint32_t narrowed = static_cast<uint32_t>(
+            0x80000000ULL | ((bits & 0x000FFFFFFFFFFFFFULL) >> 21));
+        narrowed >>= (905u - exponent);
+        narrowed |= static_cast<uint32_t>((bits >> 32) & 0x80000000ULL);
+        return narrowed;
+    }
+
+    // Results below the documented conversion range are architecturally
+    // undefined; this is the behavior measured on Gekko/Broadway hardware.
+    return static_cast<uint32_t>(
+        ((bits >> 32) & 0xC0000000ULL) | ((bits >> 29) & 0x3FFFFFFFULL));
+}
+
+MKW_MEMORY_FORCE_INLINE float PpcSingleBitsToFloat(uint32_t bits) {
+    float value = 0.0f;
+    std::memcpy(&value, &bits, sizeof(value));
+    return value;
+}
+
+MKW_MEMORY_FORCE_INLINE bool TryGetPointerFast(uint32_t address, size_t length, uint8_t*& pointer) {
+    const uint32_t page = address >> kPageShift;
+    if (length <= 8) {
+        const uintptr_t encodedBias = g_fullPageBias[page];
+        if (encodedBias != 0) {
+            pointer = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
+            return true;
+        }
+    }
+    const uint32_t offset = address & kPageMask;
+    const auto& entry = g_pageTable[page];
+    if (!entry.base || offset + length > entry.limit) {
+        pointer = nullptr;
+        return false;
+    }
+    pointer = entry.base + offset;
+    return true;
+}
+
+inline uint8_t* GetPointerFast(uint32_t address, size_t length) {
+    uint8_t* pointer = nullptr;
+    return TryGetPointerFast(address, length, pointer) ? pointer : nullptr;
+}
+
+MKW_MEMORY_FORCE_INLINE bool TryGetWritablePointerFast(
+    uint32_t address, size_t length, uint8_t*& pointer) {
+    // The sparse-table hit below distinguishes itself from the small-mapping
+    // hit through `pointer`, so the out-parameter must start null regardless
+    // of what the caller passed in. (A caller handing in an uninitialized
+    // pointer used to turn every sparse hit into a write through stack
+    // garbage - random host memory corruption.)
+    pointer = nullptr;
+    if (length == 0 || length > 8 || address > UINT32_MAX - (length - 1))
+        return false;
+    const uint32_t coarsePage = address >> kPageShift;
+    uintptr_t encodedBias = g_fullWritablePageBias[coarsePage];
+    const uint32_t endAddress = address + static_cast<uint32_t>(length - 1);
+    const uint32_t firstExactPage = address >> kWritableSubPageShift;
+    const uint32_t lastExactPage = endAddress >> kWritableSubPageShift;
+    if (encodedBias != 0 && (endAddress >> kPageShift) != coarsePage &&
+        RecompMod::g_executableWriteGuardPages[lastExactPage].load(
+            std::memory_order_relaxed) != 0) {
+        return false;
+    }
+    if (encodedBias == 0) {
+        const auto* subTable = g_sparseWritablePageTables[coarsePage];
+        if (subTable != nullptr) {
+            encodedBias = subTable->encodedBias[
+                (address & kPageMask) >> kWritableSubPageShift];
+            if (encodedBias == 0)
+                return false;
+        } else {
+            // Small mappings such as Broadway's 16 KiB locked cache cannot
+            // populate the full-1-MiB bias table. Keep them native by proving
+            // the exact access against the ordinary page entry, then applying
+            // the same executable-write policy as a sparse-table hit.
+            const uint32_t offset = address & kPageMask;
+            const auto& entry = g_pageTable[coarsePage];
+            if (!entry.base || offset + length > entry.limit)
+                return false;
+            pointer = entry.base + offset;
+        }
+
+        // The exact 4 KiB guard is the final authority. Checking it on every
+        // checked/sparse hit also makes later executable-range registration
+        // safe when a prebuilt table still contains the old mapped bias.
+        if (RecompMod::g_executableWriteGuardPages[firstExactPage].load(
+                std::memory_order_relaxed) != 0 ||
+            (lastExactPage != firstExactPage &&
+             RecompMod::g_executableWriteGuardPages[lastExactPage].load(
+                 std::memory_order_relaxed) != 0)) {
+            return false;
+        }
+
+        if (pointer != nullptr)
+            return true;
+    }
+    if (encodedBias == 0)
+        return false;
+    pointer = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
+    return true;
+}
+
+// Flat form: guest_flat_memory.h's page protections already answer mapped/non-deferred/non-executable, so resolving is pure address arithmetic.
+// Two checks stay inline: a wrapped guest address can't survive 64-bit `host + rangeOffset`, and an MMIO write's value isn't recoverable from a
+// fault record, so a write touching that window must resolve null and fall back to Memory::Write*. Checking both range endpoints is a complete
+// proof since length <= kPageSize (1 MiB) can't straddle the 32 MiB MMIO window.
+MKW_MEMORY_FORCE_INLINE uint8_t* ResolveRangeHost(uint32_t base, int32_t minOffset, uint32_t length,
+                                                 bool needsRead, bool needsWrite) {
+    (void)needsRead;
+    const uint32_t guestStart = base + static_cast<uint32_t>(minOffset);
+    if (length == 0 || length > kPageSize || guestStart > UINT32_MAX - (length - 1)) return nullptr;
+    if (GuestFlat::RequiresCheckedAccess()) {
+        // A host page can cover multiple independently-special Wii pages.
+        // Returning null keeps resolved accesses on the checked Memory::*
+        // path, which materializes deferred reads and applies write policy.
+        (void)needsWrite;
+        return nullptr;
+    }
+    if (needsWrite &&
+        (FlatWriteNeedsPolicy(guestStart) || FlatWriteNeedsPolicy(guestStart + (length - 1))))
+        [[unlikely]] return nullptr;
+    return MKW_FLAT_GUEST_BASE + guestStart;
+}
+
+// Guest-address byte order. Distinct from isa/big_endian.h, which is the
+// host-pointer codec; do not "unify" them.
+inline uint16_t ByteSwap16(uint16_t value) {
+    return __builtin_bswap16(value);
+}
+
+inline uint32_t ByteSwap32(uint32_t value) {
+    return __builtin_bswap32(value);
+}
+
+inline uint64_t ByteSwap64(uint64_t value) {
+    return __builtin_bswap64(value);
+}
+
+template <typename T>
+inline T MaybeByteSwap(T value) {
+    if constexpr (sizeof(T) == 1) {
+        return value;
+    } else if constexpr (sizeof(T) == 2) {
+        return static_cast<T>(ByteSwap16(static_cast<uint16_t>(value)));
+    } else if constexpr (sizeof(T) == 4) {
+        return static_cast<T>(ByteSwap32(static_cast<uint32_t>(value)));
+    } else if constexpr (sizeof(T) == 8) {
+        return static_cast<T>(ByteSwap64(static_cast<uint64_t>(value)));
+    } else {
+        return value;
+    }
+}
+
+template <typename T>
+MKW_MEMORY_FORCE_INLINE bool ReadResolvedScalar(uint8_t* host, uint32_t rangeOffset, T& outValue) {
+    if (!host) return false;
+    if constexpr (sizeof(T) == 1) {
+        outValue = host[rangeOffset];
+    } else {
+        T value = 0;
+        std::memcpy(&value, host + rangeOffset, sizeof(T));
+        outValue = MaybeByteSwap(value);
+    }
+    return true;
+}
+
+struct ResolvedLoadPair {
+    uint32_t first = 0;
+    uint32_t second = 0;
+    bool valid = false;
+};
+
+MKW_MEMORY_FORCE_INLINE ResolvedLoadPair ReadResolvedPair16(
+    uint8_t* host, uint32_t rangeOffset) {
+    uint32_t packed = 0;
+    if (!ReadResolvedScalar(host, rangeOffset, packed)) return {};
+    return {packed >> 16, packed & 0xFFFFu, true};
+}
+
+MKW_MEMORY_FORCE_INLINE ResolvedLoadPair ReadResolvedPair32(uint8_t* host, uint32_t rangeOffset) {
+    uint64_t packed = 0;
+    if (!ReadResolvedScalar(host, rangeOffset, packed)) return {};
+    return {static_cast<uint32_t>(packed >> 32), static_cast<uint32_t>(packed), true};
+}
+
+template <typename Packed>
+MKW_MEMORY_FORCE_INLINE bool WriteResolvedPairFast(
+    uint8_t* host, uint32_t rangeOffset, Packed packed) {
+    if (!host) return false;
+    const Packed swapped = MaybeByteSwap(packed);
+    std::memcpy(host + rangeOffset, &swapped, sizeof(swapped));
+    return true;
+}
+
+MKW_MEMORY_FORCE_INLINE bool WriteResolvedPair16(
+    uint8_t* host, uint32_t rangeOffset, uint32_t packed) {
+    return WriteResolvedPairFast(host, rangeOffset, packed);
+}
+
+MKW_MEMORY_FORCE_INLINE bool WriteResolvedPair32(
+    uint8_t* host, uint32_t rangeOffset, uint64_t packed) {
+    return WriteResolvedPairFast(host, rangeOffset, packed);
+}
+
+template <typename T>
+MKW_MEMORY_FORCE_INLINE bool WriteResolvedScalar(uint8_t* host, uint32_t rangeOffset, T value) {
+    if (!host) return false;
+    if constexpr (sizeof(T) == 1) {
+        host[rangeOffset] = static_cast<uint8_t>(value);
+    } else {
+        const T swapped = MaybeByteSwap(value);
+        std::memcpy(host + rangeOffset, &swapped, sizeof(T));
+    }
+    return true;
+}
+
+template <typename T>
+inline bool TryReadMappedScalar(uint32_t address, T& outValue) {
+    uint8_t* ptr = nullptr;
+    if (TryGetPointerFast(address, sizeof(T), ptr)) {
+        if constexpr (sizeof(T) == 1) {
+            outValue = *ptr;
+        } else {
+            T value = 0;
+            std::memcpy(&value, ptr, sizeof(T));
+            outValue = MaybeByteSwap(value);
+        }
+        return true;
+    }
+    return false;
+}
+
+template <typename T>
+MKW_MEMORY_FORCE_INLINE bool TryReadGuestScalar(uint32_t address, T& outValue) {
+    const uintptr_t encodedBias = g_fullReadablePageBias[address >> kPageShift];
+    if (encodedBias == 0) [[unlikely]]
+        return false;
+    auto* ptr = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
+    if constexpr (sizeof(T) == 1) {
+        outValue = *ptr;
+    } else {
+        T value = 0;
+        std::memcpy(&value, ptr, sizeof(T));
+        outValue = MaybeByteSwap(value);
+    }
+    return true;
+}
+
+template <typename T>
+MKW_MEMORY_FORCE_INLINE bool TryWriteGuestScalar(uint32_t address, T value) {
+    static_assert(sizeof(T) >= 1 && sizeof(T) <= kMaxFastScalarSize);
+    // A mixed executable/data 1 MiB page zeroes the coarse writable bias even though most of
+    // its 4 KiB sub-pages are plain data; MKW's THP buffers share such a page with .text, which
+    // used to force ~15% of total CPU through the cold path. The sparse sub-page tier below
+    // keeps those stores native while the exact 4 KiB executable guards stay authoritative.
+    if (address > UINT32_MAX - static_cast<uint32_t>(sizeof(T) - 1u)) [[unlikely]]
+        return false;
+    const uint32_t coarsePage = address >> kPageShift;
+    const uint32_t endAddress = address + static_cast<uint32_t>(sizeof(T) - 1u);
+    if ((endAddress >> kPageShift) != coarsePage) [[unlikely]]
+        return false;
+    const uintptr_t encodedBias = g_fullWritablePageBias[coarsePage];
+    uint8_t* ptr = nullptr;
+    if (encodedBias != 0) {
+        ptr = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
+    } else if (!TryGetWritablePointerFast(address, sizeof(T), ptr)) [[unlikely]] {
+        return false;
+    }
+    if constexpr (sizeof(T) == 1) {
+        *ptr = static_cast<uint8_t>(value);
+    } else {
+        const T swapped = MaybeByteSwap(value);
+        std::memcpy(ptr, &swapped, sizeof(T));
+    }
+    return true;
+}
+
+template <typename T>
+inline bool WriteStackScalarFast(uint32_t address, T value) {
+    uint8_t* ptr = nullptr;
+    if (TryGetPointerFast(address, sizeof(T), ptr)) {
+        if constexpr (sizeof(T) == 1) {
+            *ptr = static_cast<uint8_t>(value);
+        } else {
+            const T swapped = MaybeByteSwap(value);
+            std::memcpy(ptr, &swapped, sizeof(T));
+        }
+        return true;
+    }
+    return false;
+}
+
+inline uint8_t ReadStack8(uint32_t address) {
+    uint8_t value = 0;
+    return TryReadMappedScalar(address, value) ? value : Memory::Read8(address);
+}
+
+inline uint16_t ReadStack16(uint32_t address) {
+    uint16_t value = 0;
+    return TryReadMappedScalar(address, value) ? value : Memory::Read16(address);
+}
+
+inline uint32_t ReadStack32(uint32_t address) {
+    uint32_t value = 0;
+    return TryReadMappedScalar(address, value) ? value : Memory::Read32(address);
+}
+
+inline uint64_t ReadStack64(uint32_t address) {
+    uint64_t value = 0;
+    return TryReadMappedScalar(address, value) ? value : Memory::Read64(address);
+}
+
+
+
+inline void WriteStack8(uint32_t address, uint8_t value) {
+    if (!WriteStackScalarFast(address, value)) {
+        Memory::Write8(address, value);
+    }
+}
+
+inline void WriteStack16(uint32_t address, uint16_t value) {
+    if (!WriteStackScalarFast(address, value)) {
+        Memory::Write16(address, value);
+    }
+}
+
+inline void WriteStack32(uint32_t address, uint32_t value) {
+    if (!WriteStackScalarFast(address, value)) {
+        Memory::Write32(address, value);
+    }
+}
+
+inline void WriteStack64(uint32_t address, uint64_t value) {
+    if (!WriteStackScalarFast(address, value)) {
+        Memory::Write64(address, value);
+    }
+}
+
+inline void WriteStackFloat32(uint32_t address, double value) {
+    const uint32_t bits = ConvertPpcDoubleToSingleBits(value);
+    if (!WriteStackScalarFast(address, bits)) {
+        Memory::WriteFloat32(address, value);
+    }
+}
+
+inline void WriteStackFloat64(uint32_t address, double value) {
+    uint64_t bits = 0;
+    std::memcpy(&bits, &value, sizeof(bits));
+    if (!WriteStackScalarFast(address, bits)) {
+        Memory::WriteFloat64(address, value);
+    }
+}
+
+template <typename T, typename SlowRead>
+MKW_MEMORY_FORCE_INLINE T ReadResolved(uint8_t* host, uint32_t rangeOffset, uint32_t address,
+                      SlowRead slow) {
+    T value = 0;
+    if (ReadResolvedScalar(host, rangeOffset, value)) {
+        return value;
+    }
+    [[unlikely]] return slow(address);
+}
+
+MKW_MEMORY_FORCE_INLINE uint8_t ReadResolved8(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint8_t>(r, o, a, ReadResolvedFallback<uint8_t>); }
+MKW_MEMORY_FORCE_INLINE uint16_t ReadResolved16(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint16_t>(r, o, a, ReadResolvedFallback<uint16_t>); }
+MKW_MEMORY_FORCE_INLINE uint32_t ReadResolved32(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint32_t>(r, o, a, ReadResolvedFallback<uint32_t>); }
+// Live via isa/ppc_isa_quantized.h (the psq resolved tier packs two lanes into
+// one 64-bit access); generated code never names it directly.
+MKW_MEMORY_FORCE_INLINE uint64_t ReadResolved64(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint64_t>(r, o, a, ReadResolvedFallback<uint64_t>); }
+MKW_MEMORY_FORCE_INLINE float ReadResolvedFloat32(uint8_t* r, uint32_t o, uint32_t a) {
+    uint32_t bits = 0;
+    if (!ReadResolvedScalar(r, o, bits)) [[unlikely]] return ReadResolvedFallbackFloat32(a);
+    float value; std::memcpy(&value, &bits, sizeof(value)); return value;
+}
+MKW_MEMORY_FORCE_INLINE double ReadResolvedFloat64(uint8_t* r, uint32_t o, uint32_t a) {
+    uint64_t bits = 0;
+    if (!ReadResolvedScalar(r, o, bits)) [[unlikely]] return ReadResolvedFallbackFloat64(a);
+    double value; std::memcpy(&value, &bits, sizeof(value)); return value;
+}
+
+template <typename T, typename SlowWrite>
+MKW_MEMORY_FORCE_INLINE void WriteResolved(uint8_t* host, uint32_t rangeOffset, uint32_t address, T value,
+                          SlowWrite slow) {
+    if (WriteResolvedScalar(host, rangeOffset, value)) {
+        return;
+    }
+    [[unlikely]] slow(address, value);
+}
+
+MKW_MEMORY_FORCE_INLINE void WriteResolved8(uint8_t* r, uint32_t o, uint32_t a, uint8_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint8_t>); }
+MKW_MEMORY_FORCE_INLINE void WriteResolved16(uint8_t* r, uint32_t o, uint32_t a, uint16_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint16_t>); }
+MKW_MEMORY_FORCE_INLINE void WriteResolved32(uint8_t* r, uint32_t o, uint32_t a, uint32_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint32_t>); }
+// Live via isa/ppc_isa_quantized.h, as ReadResolved64 above.
+MKW_MEMORY_FORCE_INLINE void WriteResolved64(uint8_t* r, uint32_t o, uint32_t a, uint64_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint64_t>); }
+MKW_MEMORY_FORCE_INLINE void WriteResolvedFloat32(uint8_t* r, uint32_t o, uint32_t a, double v) {
+    const uint32_t bits = ConvertPpcDoubleToSingleBits(v);
+    if (WriteResolvedScalar(r, o, bits)) return;
+    [[unlikely]] WriteResolvedFallbackFloat32(a, v);
+}
+MKW_MEMORY_FORCE_INLINE void WriteResolvedFloat64(uint8_t* r, uint32_t o, uint32_t a, double v) {
+    uint64_t bits; std::memcpy(&bits, &v, sizeof(bits));
+    if (WriteResolvedScalar(r, o, bits)) return;
+    [[unlikely]] WriteResolvedFallbackFloat64(a, v);
+}
+
+// Flat guest memory (audit item T-MEM): the 4 GiB reservation makes a guest access a byte swap
+// around `*(T*)(base + addr)`, no page-table load or limit check (interception model documented
+// in guest_flat_memory.h). The one exception kept inline is the MMIO write policy, since the
+// written value can't be recovered from a fault record.
+//
+// When a host VM page is larger than a 4 KiB Wii page, guest-view protections
+// cannot distinguish adjacent special Wii pages. The general FlatRead*/
+// FlatWrite* helpers then use the checked page-table path, which materializes
+// deferred reads and applies executable-write/MMIO policy before touching RAM.
+// FlatWriteRam* remains direct because the translator emits it only for
+// addresses it has proven are ordinary RAM.
+
+template <typename T>
+MKW_MEMORY_FORCE_INLINE T FlatLoad(uint32_t address) {
+    T value{};
+    std::memcpy(&value, MKW_FLAT_GUEST_BASE + address, sizeof(T));
+    return MaybeByteSwap(value);
+}
+
+template <typename T>
+MKW_MEMORY_FORCE_INLINE void FlatStore(uint32_t address, T value) {
+    const T swapped = MaybeByteSwap(value);
+    std::memcpy(MKW_FLAT_GUEST_BASE + address, &swapped, sizeof(T));
+}
+
+MKW_MEMORY_FORCE_INLINE uint8_t FlatRead8(uint32_t address) {
+    if (GuestFlat::RequiresCheckedAccess()) return Memory::Read8(address);
+    return FlatLoad<uint8_t>(address);
+}
+MKW_MEMORY_FORCE_INLINE uint16_t FlatRead16(uint32_t address) {
+    if (GuestFlat::RequiresCheckedAccess()) return Memory::Read16(address);
+    return FlatLoad<uint16_t>(address);
+}
+MKW_MEMORY_FORCE_INLINE uint32_t FlatRead32(uint32_t address) {
+    if (GuestFlat::RequiresCheckedAccess()) return Memory::Read32(address);
+    return FlatLoad<uint32_t>(address);
+}
+
+MKW_MEMORY_FORCE_INLINE float FlatReadFloat32(uint32_t address) {
+    const uint32_t bits = FlatRead32(address);
+    float value = 0.0f;
+    std::memcpy(&value, &bits, sizeof(value));
+    return value;
+}
+
+MKW_MEMORY_FORCE_INLINE double FlatReadFloat64(uint32_t address) {
+    const uint64_t bits = GuestFlat::RequiresCheckedAccess()
+        ? Memory::Read64(address) : FlatLoad<uint64_t>(address);
+    double value = 0.0;
+    std::memcpy(&value, &bits, sizeof(value));
+    return value;
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWrite8(uint32_t address, uint8_t value) {
+    if (GuestFlat::RequiresCheckedAccess()) { Memory::Write8(address, value); return; }
+    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { Write8Slow(address, value); return; }
+    FlatStore<uint8_t>(address, value);
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWrite16(uint32_t address, uint16_t value) {
+    if (GuestFlat::RequiresCheckedAccess()) { Memory::Write16(address, value); return; }
+    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { Write16Slow(address, value); return; }
+    FlatStore<uint16_t>(address, value);
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWrite32(uint32_t address, uint32_t value) {
+    if (GuestFlat::RequiresCheckedAccess()) { Memory::Write32(address, value); return; }
+    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { Write32Slow(address, value); return; }
+    FlatStore<uint32_t>(address, value);
+}
+
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteFloat32(uint32_t address, double value) {
+    if (GuestFlat::RequiresCheckedAccess()) { Memory::WriteFloat32(address, value); return; }
+    const uint32_t bits = ConvertPpcDoubleToSingleBits(value);
+    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { WriteFloat32Slow(address, value); return; }
+    FlatStore<uint32_t>(address, bits);
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteFloat64(uint32_t address, double value) {
+    if (GuestFlat::RequiresCheckedAccess()) { Memory::WriteFloat64(address, value); return; }
+    uint64_t bits = 0;
+    std::memcpy(&bits, &value, sizeof(bits));
+    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { WriteFloat64Slow(address, value); return; }
+    FlatStore<uint64_t>(address, bits);
+}
+
+// Check-free stores: emitted ONLY for addresses the translator proved at translate time are ordinary guest RAM (r1-relative stack slots, ~45%
+// of flat stores), skipping the MMIO mask/compare that's pure overhead there. Still safe if that proof were ever wrong: the flat view maps
+// 0xCC000000..0xCDFFFFFF PAGE_NOACCESS, so a stray MMIO store faults into the same handler and diagnostic as the checked path, just reported
+// instead of dispatched inline. Never use these for an address the translator hasn't proven.
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteRam8(uint32_t address, uint8_t value) {
+    FlatStore<uint8_t>(address, value);
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteRam16(uint32_t address, uint16_t value) {
+    FlatStore<uint16_t>(address, value);
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteRam32(uint32_t address, uint32_t value) {
+    FlatStore<uint32_t>(address, value);
+}
+
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteRamFloat32(uint32_t address, double value) {
+    FlatStore<uint32_t>(address, ConvertPpcDoubleToSingleBits(value));
+}
+
+MKW_MEMORY_FORCE_INLINE void FlatWriteRamFloat64(uint32_t address, double value) {
+    uint64_t bits = 0;
+    std::memcpy(&bits, &value, sizeof(bits));
+    FlatStore<uint64_t>(address, bits);
+}
+
+} // namespace MemoryInline
+
+#undef MKW_MEMORY_FORCE_INLINE
+#undef MKW_MEMORY_NO_INLINE
+#undef MKW_MEMORY_COLD
+
+inline uint8_t Memory::Read8(uint32_t addr) {
+    uint8_t value = 0;
+    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
+        return MemoryInline::Read8Slow(addr);
+    return value;
+}
+
+inline uint16_t Memory::Read16(uint32_t addr) {
+    uint16_t value = 0;
+    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
+        return MemoryInline::Read16Slow(addr);
+    return value;
+}
+
+inline uint32_t Memory::Read32(uint32_t addr) {
+    uint32_t value = 0;
+    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
+        return MemoryInline::Read32Slow(addr);
+    return value;
+}
+
+inline uint64_t Memory::Read64(uint32_t addr) {
+    uint64_t value = 0;
+    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
+        return MemoryInline::Read64Slow(addr);
+    return value;
+}
+
+inline float Memory::ReadFloat32(uint32_t addr) {
+    uint32_t bits = 0;
+    if (!MemoryInline::TryReadGuestScalar(addr, bits)) [[unlikely]]
+        return MemoryInline::ReadFloat32Slow(addr);
+    float value = 0.0f;
+    std::memcpy(&value, &bits, sizeof(value));
+    return value;
+}
+
+inline double Memory::ReadFloat64(uint32_t addr) {
+    uint64_t bits = 0;
+    if (!MemoryInline::TryReadGuestScalar(addr, bits)) [[unlikely]]
+        return MemoryInline::ReadFloat64Slow(addr);
+    double value = 0.0;
+    std::memcpy(&value, &bits, sizeof(value));
+    return value;
+}
+
+inline void Memory::Write8(uint32_t addr, uint8_t val) {
+    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
+        MemoryInline::Write8Slow(addr, val);
+}
+
+inline void Memory::Write16(uint32_t addr, uint16_t val) {
+    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
+        MemoryInline::Write16Slow(addr, val);
+}
+
+inline void Memory::Write32(uint32_t addr, uint32_t val) {
+    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
+        MemoryInline::Write32Slow(addr, val);
+}
+
+inline void Memory::Write64(uint32_t addr, uint64_t val) {
+    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
+        MemoryInline::Write64Slow(addr, val);
+}
+
+inline void Memory::WriteFloat32(uint32_t addr, double val) {
+    {
+        const uint32_t bits = MemoryInline::ConvertPpcDoubleToSingleBits(val);
+        if (MemoryInline::TryWriteGuestScalar(addr, bits))
+            return;
+    }
+    [[unlikely]] MemoryInline::WriteFloat32Slow(addr, val);
+}
+
+inline void Memory::WriteFloat64(uint32_t addr, double val) {
+    {
+        uint64_t bits = 0;
+        std::memcpy(&bits, &val, sizeof(bits));
+        if (MemoryInline::TryWriteGuestScalar(addr, bits))
+            return;
+    }
+    [[unlikely]] MemoryInline::WriteFloat64Slow(addr, val);
+}

runtime/include/guest_abi/native_cpu_calls.inc

@@ -1,9 +1,10 @@
 // Native call catalog used by InvokeDirectCpu. Keep project-specific HLE
-// address mappings here; abi_bridge.h contains only the generic ABI machinery.
+// address mappings here; dispatch.h contains only the generic ABI machinery.
+// Part of the guest ABI: translated shards bind these calls at compile time.
 //
 // This is a .inc, not a header: it has no includes of its own and references
 // CpuContext, KnownTypedNativeCpuCall<> and KnownNativeCpuCall<>, all of which
-// are declared above its single mid-file include site in abi_bridge.h. It
+// are declared above its single mid-file include site in dispatch.h. It
 // cannot compile standalone and must never be included anywhere else.
 
 extern "C" void nw4r__lyt__detail__DrawQuad_800847c0(CpuContext* ctx);

runtime/include/guest_flat_memory.h

@@ -9,39 +9,10 @@
 #include <cstdint>
 #include <vector>
 
+#include "guest_abi/flat_memory.h"
+
 namespace GuestFlat {
 
-// Fixed base so the emitted access is `[reg + imm64-in-register]` with no load
-// of a global.
-inline constexpr uint64_t kGuestSpaceSize = 0x1'0000'0000ull;
-inline constexpr size_t kGuestPageSize = 0x1000;
-#if defined(__x86_64__)
-// 16 TiB: clear of the Windows ASan shadow (32 TiB) and of the usual image/heap
-// placement.
-inline constexpr uintptr_t kFixedFlatGuestBase = 0x0000'1000'0000'0000ull;
-#elif defined(__aarch64__) && defined(__APPLE__)
-// Keep this well above the low address ranges that Darwin's ASLR may use for
-// a PIE executable and its shared cache. Apple Silicon's user VA is wider
-// than Linux's 39-bit minimum, so this 512 GiB region is available while the
-// Linux AArch64 target retains its 64 GiB placement below.
-inline constexpr uintptr_t kFixedFlatGuestBase = 0x0000'0080'0000'0000ull;
-#elif defined(__aarch64__)
-// 16 TiB (this arch's x86_64 sibling value) is unreachable on any AArch64
-// kernel configured for 39-bit virtual addresses (512 GiB ceiling) - common on
-// older/embedded targets (e.g. this project's own tested Jetson/L4T board, kernel
-// 4.9). There mmap() silently ignores a hint above the ceiling and hands back an
-// address near the top of the real range instead, which this module's caller
-// then rejects as "already occupied". 64 GiB is reachable on every AArch64 VA
-// width in real use (39-bit minimum and up) and was confirmed via direct mmap
-// probing to sit far below where the PIE image, heap, shared libraries and
-// stack actually land (all clustered above ~340 GiB on a 39-bit/512 GiB system).
-inline constexpr uintptr_t kFixedFlatGuestBase = 0x0000'0010'0000'0000ull;
-#else
-#error "guest_flat_memory.h has no fixed flat guest base chosen for this architecture"
-#endif
-
-#define MKW_FLAT_GUEST_BASE (reinterpret_cast<uint8_t*>(GuestFlat::kFixedFlatGuestBase))
-
 enum class Backing {
     Owned,
     Mem1,
@@ -65,25 +36,6 @@ struct FaultCounters {
 // True once the reservation exists and translated code may use the flat path.
 bool IsActive();
 
-// True when a host VM page covers more than one 4 KiB Wii page. In that
-// configuration, guest-view page protection cannot safely represent per-Wii-
-// page MMIO, deferred-read, or executable-write state, so general translated
-// accesses must use the checked Memory::* path.
-// Windows user mode and x86-64 always use a 4 KiB base page, so those builds
-// fold this to a compile-time false: it appears in every flat access and must
-// not become a hot-path load. Only AArch64, where the page size is a kernel
-// configuration (4/16/64 KiB), has to probe it at runtime.
-#if defined(_WIN32) || defined(__x86_64__)
-#define MKW_GUEST_FLAT_FIXED_PAGE_SIZE 1
-#endif
-
-#if defined(MKW_GUEST_FLAT_FIXED_PAGE_SIZE)
-inline constexpr bool RequiresCheckedAccess() noexcept { return false; }
-#else
-extern bool g_requiresCheckedAccess;
-inline bool RequiresCheckedAccess() noexcept { return g_requiresCheckedAccess; }
-#endif
-
 // Reserves the 4 GiB space (once per process) and maps every requested region
 // into both views. Throws std::runtime_error with a precise diagnosis when the
 // reservation, the section objects or a view cannot be created - a silent

runtime/include/guest_small_data.h

@@ -0,0 +1,10 @@
+#pragma once
+
+#include <cstdint>
+
+// PowerPC EABI small-data base pointers of the translated game. Defined per game
+// in runtime/game/runtime_config_values.cpp so the runtime itself stays game-independent.
+namespace GuestSmallData {
+extern const uint32_t kSda1Base; // _SDA_BASE_, loaded into r13
+extern const uint32_t kSda2Base; // _SDA2_BASE_, loaded into r2
+} // namespace GuestSmallData

runtime/include/host_context.h

@@ -4,7 +4,7 @@
 
 // HostContext is the deliberately small boundary between the guest scheduler
 // and the host's cooperative-context facility. Windows uses native Fibers;
-// Linux and Intel macOS use libco's System V x86-64 backend. macOS AArch64 uses
+// Linux uses libco's host backend; Intel macOS uses its System V x86-64 backend. macOS AArch64 uses
 // the local assembly backend because it must preserve Darwin's platform-reserved
 // x18 register, which libco's AArch64 backend does not save. Its handles are
 // only valid on the thread that initialized the scheduler.

runtime/include/isa/ppc_isa_context.h

@@ -1,5 +1,6 @@
 #pragma once
 #include "ppc_isa_fpenv.h"
+#include "guest_abi/cpu_context.h"
 
 #include <cmath>
 #include <cstdint>
@@ -10,85 +11,6 @@
 
 void ShowRuntimeFatalPopup(std::string_view category, std::string_view details) noexcept;
 
-union PPC_FPR {
-    uint64_t raw;
-    double d;
-    struct {
-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
-        float ps1; // Low word (Least Significant)
-        float ps0; // High word (Most Significant)
-#else
-        float ps0; // High word
-        float ps1; // Low word
-#endif
-    } paired;
-};
-
-// PowerPC CPU Context
-struct CpuContext {
-    // Standard GPRs
-    uint32_t gpr[32];
-
-    // Special Purpose Registers defined by standard PPC
-    uint32_t cr;        // Condition Register
-    uint32_t lr;        // Link Register
-    uint32_t ctr;       // Count Register
-    uint32_t xer;       // Integer Exception Register
-    uint32_t fpscr;     // Floating-Point Status and Control Register
-
-    // Program State
-    uint32_t pc;        // Program Counter (not used alot tho)
-
-    // Floating Point Registers (Modified for Paired Single support)
-    PPC_FPR  fpr[32];
-
-    // Broadway Specific Extensions
-    uint32_t gqr[8];    // Graphics Quantization Registers
-    uint32_t hid0;      // HID0
-    uint32_t hid1;      // HID1
-    uint32_t hid2;      // HID2
-
-    uint32_t srr0;      // Save/Restore Register 0
-    uint32_t srr1;      // Save/Restore Register 1
-    uint32_t msr;       // Machine State Register
-};
-
-inline thread_local CpuContext* g_currentCpuContext = nullptr;
-
-class CpuContextScope {
-public:
-    explicit CpuContextScope(CpuContext* ctx)
-        : previous_(g_currentCpuContext)
-    {
-        g_currentCpuContext = ctx;
-
-        savedMxcsr_ = MkwGetHostFpControl();
-        if (ctx != nullptr)
-            MkwApplyHostNiMode(ctx->fpscr);
-    }
-
-    ~CpuContextScope()
-    {
-        g_currentCpuContext = previous_;
-        if (previous_ != nullptr)
-            MkwApplyHostNiMode(previous_->fpscr);
-        else
-            MkwRestoreHostMxcsr(savedMxcsr_);
-    }
-
-    CpuContextScope(const CpuContextScope&) = delete;
-    CpuContextScope& operator=(const CpuContextScope&) = delete;
-
-private:
-    CpuContext* previous_ = nullptr;
-    uint32_t savedMxcsr_ = 0;
-};
-
-inline CpuContext* TryGetCpuContext() noexcept
-{
-    return g_currentCpuContext;
-}
-
 inline CpuContext* CurrentCpuContext()
 {
     CpuContext* cpu = TryGetCpuContext();
@@ -110,79 +32,7 @@ inline CpuContext* CurrentCpuContext()
     return cpu;
 }
 
-// Condition Register Fields
-#define CR_LT 0
-#define CR_GT 1
-#define CR_EQ 2
-#define CR_SO 3
+#include "guest_abi/guest_cr.h"
 
 extern "C" void DumpHostStackTraceForRuntimeHelper();
 void MarkFatalErrorReported();
-
-[[noreturn]] inline void PPC_Undefined(uint32_t pc, uint32_t rawInstruction, const char* details)
-{
-    std::fprintf(stderr,
-                 "[runtime] UNDEFINED guest instruction: pc=0x%08X raw=0x%08X %s\n",
-                 pc,
-                 rawInstruction,
-                 details ? details : "");
-    char message[256]{};
-    std::snprintf(message, sizeof(message),
-                  "The game stopped because it reached an unsupported guest instruction at PC 0x%08X (instruction 0x%08X).\n\n%s",
-                  pc, rawInstruction, details ? details : "No additional details were provided.");
-    ShowRuntimeFatalPopup("Unsupported guest instruction", message);
-    std::abort();
-}
-
-// Used by generated code when the translator encounters privileged/unmodeled PPC instructions
-// (e.g. rfi). This is intentionally loud so missing HLE hooks are easy to find.
-#define UNDEFINED(pc, rawInstruction, details) PPC_Undefined((pc), (rawInstruction), (details))
-
-// Helper to set CR bits (Signed)
-inline void SetCR(CpuContext* cpu, int field, int32_t a, int32_t b) {
-    uint32_t crField = 0;
-    if (a < b) crField |= 0x8; // LT
-    if (a > b) crField |= 0x4; // GT
-    if (a == b) crField |= 0x2; // EQ
-    crField |= (cpu->xer >> 31) & 1u; // SO
-    
-    int shift = (7 - field) * 4;
-    uint32_t mask = 0xF << shift;
-    cpu->cr = (cpu->cr & ~mask) | (crField << shift);
-}
-
-// Helper to set CR bits (Unsigned)
-inline void SetCR(CpuContext* cpu, int field, uint32_t a, uint32_t b) {
-    uint32_t crField = 0;
-    if (a < b) crField |= 0x8; // LT
-    if (a > b) crField |= 0x4; // GT
-    if (a == b) crField |= 0x2; // EQ
-    crField |= (cpu->xer >> 31) & 1u; // SO
-    
-    int shift = (7 - field) * 4;
-    uint32_t mask = 0xF << shift;
-    cpu->cr = (cpu->cr & ~mask) | (crField << shift);
-}
-
-// Helper to set CR bits (Floating-point)
-inline void SetCRFloat(CpuContext* cpu, int field, double a, double b) {
-    uint32_t crField = 0;
-    if (std::isnan(a) || std::isnan(b)) {
-        // Unordered: LT/GT/EQ clear, SO set.
-        crField = 0x1;
-    } else {
-        if (a < b) crField |= 0x8; // LT
-        if (a > b) crField |= 0x4; // GT
-        if (a == b) crField |= 0x2; // EQ
-    }
-
-    int shift = (7 - field) * 4;
-    uint32_t mask = 0xF << shift;
-    cpu->cr = (cpu->cr & ~mask) | (crField << shift);
-}
-
-// Helper to get CR bit
-inline bool GetCRBit(CpuContext* cpu, int field, int bit) {
-    int shift = (7 - field) * 4 + (3 - bit);
-    return (cpu->cr >> shift) & 1;
-}

runtime/include/isa/ppc_isa_float.h

@@ -22,7 +22,7 @@
 // the psq_l/psq_st tier that does lives in ppc_isa_quantized.h.
 
 #include "ppc_isa_config.h"
-#include "ppc_isa_context.h"
+#include "guest_abi/cpu_context.h"
 
 #include <array>
 #include <cmath>
@@ -607,7 +607,18 @@ inline double PPC_PsMulNoNiInline(double lhs, double rhs)
 inline PpcPairVec PpcFmaddPairInline(PpcPairVec multiplicand, PpcPairVec multiplier, PpcPairVec addend)
 {
 #if defined(__x86_64__)
+#if defined(__FMA__)
     return _mm_fmadd_ps(multiplicand, multiplier, addend);
+#else
+    // x86-64-v2 has SSE4.2 but no FMA. Preserve the one rounding point per
+    // lane through the scalar C++ FMA rather than decomposing into mul+add.
+    const double a = PpcM128ToPsInline(multiplicand);
+    const double c = PpcM128ToPsInline(multiplier);
+    const double b = PpcM128ToPsInline(addend);
+    return PpcPsToM128Inline(PpcPackPairedInline(
+        PpcAccuratePsMaddLaneNoNiInline<false>(PpcGetPs0Inline(a), PpcGetPs0Inline(c), PpcGetPs0Inline(b)),
+        PpcAccuratePsMaddLaneNoNiInline<false>(PpcGetPs1Inline(a), PpcGetPs1Inline(c), PpcGetPs1Inline(b))));
+#endif
 #elif defined(__aarch64__)
     const PpcPairVec result = vfma_f32(addend, multiplicand, multiplier);
     if (PpcPairNanLaneBitsInline(result) != 0) [[unlikely]]
@@ -619,7 +630,16 @@ inline PpcPairVec PpcFmaddPairInline(PpcPairVec multiplicand, PpcPairVec multipl
 inline PpcPairVec PpcFmsubPairInline(PpcPairVec multiplicand, PpcPairVec multiplier, PpcPairVec subtractor)
 {
 #if defined(__x86_64__)
+#if defined(__FMA__)
     return _mm_fmsub_ps(multiplicand, multiplier, subtractor);
+#else
+    const double a = PpcM128ToPsInline(multiplicand);
+    const double c = PpcM128ToPsInline(multiplier);
+    const double b = PpcM128ToPsInline(subtractor);
+    return PpcPsToM128Inline(PpcPackPairedInline(
+        PpcAccuratePsMaddLaneNoNiInline<true>(PpcGetPs0Inline(a), PpcGetPs0Inline(c), PpcGetPs0Inline(b)),
+        PpcAccuratePsMaddLaneNoNiInline<true>(PpcGetPs1Inline(a), PpcGetPs1Inline(c), PpcGetPs1Inline(b))));
+#endif
 #elif defined(__aarch64__)
     const PpcPairVec result = vfma_f32(vneg_f32(subtractor), multiplicand, multiplier);
     if (PpcPairNanLaneBitsInline(result) != 0) [[unlikely]]

runtime/include/isa/ppc_isa_fpenv.h

@@ -34,7 +34,7 @@ inline constexpr double kMkwNiFlushThreshold = 0x1p-126;  // 0x3810000000000000
 inline thread_local double g_mkwNiFlushThreshold = 0.0;
 
 // Host FP control register access (MXCSR on x86_64, FPCR on AArch64), abstracted so
-// MkwApplyHostNiMode/MkwRestoreHostMxcsr and CpuContextScope (ppc_isa_context.h) don't each need
+// MkwApplyHostNiMode/MkwRestoreHostMxcsr and CpuContextScope (guest_abi/cpu_context.h) don't each need
 // their own per-arch branch.
 inline uint32_t MkwGetHostFpControl() noexcept
 {

runtime/include/isa/ppc_isa_quantized.h

@@ -2,7 +2,7 @@
 
 
 #include "ppc_isa_config.h"
-#include "ppc_isa_context.h"
+#include "guest_abi/cpu_context.h"
 #include "ppc_isa_float.h"
 #include "big_endian.h"
 

runtime/include/memory.h

@@ -8,109 +8,7 @@
 #include <string_view>
 #include <vector>
 
-class Memory {
-public:
-    using DeferredReadCallback = bool (*)(void* user);
-
-    static constexpr size_t kMem1Size = 24u * 1024u * 1024u;
-    static constexpr size_t kMem2Size = 128u * 1024u * 1024u;
-    static constexpr uint32_t kMem1PhysicalBase = 0x00000000u;
-    static constexpr uint32_t kMem1CachedBase = 0x80000000u;
-    static constexpr uint32_t kMem1UncachedBase = 0xC0000000u;
-    static constexpr uint32_t kMem2PhysicalBase = 0x10000000u;
-    static constexpr uint32_t kMem2CachedBase = 0x90000000u;
-    static constexpr uint32_t kMem2UncachedBase = 0xD0000000u;
-    static constexpr uint32_t kMem2PhysicalEnd =
-        kMem2PhysicalBase + static_cast<uint32_t>(kMem2Size);
-    static constexpr uint32_t kMem2CachedEnd =
-        kMem2CachedBase + static_cast<uint32_t>(kMem2Size);
-    static constexpr uint32_t kMem2UncachedEnd =
-        kMem2UncachedBase + static_cast<uint32_t>(kMem2Size);
-
-    struct RegionConfig {
-        std::string name;
-        uint32_t baseAddress = 0;
-        size_t sizeBytes = 0;
-    };
-
-    struct Config {
-        std::vector<RegionConfig> regions;
-        static Config WiiDefaults();
-    };
-
-    class AccessViolation : public std::runtime_error {
-    public:
-        AccessViolation(uint32_t address, size_t length, std::string_view reason);
-
-        uint32_t address() const noexcept { return address_; }
-        size_t length() const noexcept { return length_; }
-        std::string_view reason() const noexcept { return reason_; }
-
-    private:
-        uint32_t address_ = 0;
-        size_t length_ = 0;
-        std::string reason_;
-    };
-
-    static void Init(const Config& config);
-    // Single flat MEM1 region. Not used by the shipped runtime, but the
-    // translator integration-test harnesses emit calls to it.
-    static void Init(size_t mem1Size);
-    static void Reset();
-    // Executable ranges are normally registered during startup. Rebuild the
-    // writable fast-path classification after each registration so a page
-    // previously classified as ordinary data cannot retain a stale direct
-    // write bias.
-    static void RefreshWritableFastPathsForExecutableRanges();
-
-    static uint8_t Read8(uint32_t addr);
-    static uint16_t Read16(uint32_t addr);
-    static uint32_t Read32(uint32_t addr);
-    static uint64_t Read64(uint32_t addr);
-    static float ReadFloat32(uint32_t addr);
-    static double ReadFloat64(uint32_t addr);
-    static void Write8(uint32_t addr, uint8_t val);
-    static void Write16(uint32_t addr, uint16_t val);
-    static void Write32(uint32_t addr, uint32_t val);
-    static void Write64(uint32_t addr, uint64_t val);
-    static void WriteFloat32(uint32_t addr, double val);
-    static void WriteFloat64(uint32_t addr, double val);
-
-    // Exception-safe scalar access for HLE code. These keep Read32/Write32's
-    // full mapping behavior and only convert an unmapped address into a failure
-    // result; they are deliberately not MemoryInline::Try*GuestScalar, which is
-    // the translated-code fast path over the page table.
-    static bool TryRead32(uint32_t addr, uint32_t& value) noexcept {
-        try {
-            value = Read32(addr);
-            return true;
-        } catch (const AccessViolation&) {
-            return false;
-        }
-    }
-
-    static bool TryWrite32(uint32_t addr, uint32_t value) noexcept {
-        try {
-            Write32(addr, value);
-            return true;
-        } catch (const AccessViolation&) {
-            return false;
-        }
-    }
-
-    static uint8_t* GetPointer(uint32_t addr);
-    static uint8_t* GetPointer(uint32_t addr, size_t length);
-    static bool Contains(uint32_t addr, size_t length = 1);
-
-    static uint64_t RegisterDeferredRead(uint32_t addr, size_t length,
-                                         DeferredReadCallback callback, void* user);
-    static void ClearDeferredReads();
-
-    // sizeBytes reports the storage actually allocated, which is not always the
-    // configured size (aliased MEM1/MEM2 windows are clamped to what is behind
-    // them). Used by the crash dump.
-    static std::vector<RegionConfig> DescribeRegions();
-};
+#include "guest_abi/memory.h"
 
 // The translated-code access layer is kept separately from the public memory API.
 #include "memory_access.h"

runtime/include/memory_access.h

@@ -4,729 +4,4 @@
 #include "memory.h"
 #include "recomp_mod_loader.h"
 
-#include <cstddef>
-#include <cstdint>
-#include <cstring>
-
-// Hooks for GX HLE FIFO handling
-extern "C" {
-    void GX_HLE_FIFO_WriteFloat(float val);
-    void GX_HLE_FIFO_Write32(uint32_t val);
-    void GX_HLE_FIFO_Write16(uint16_t val);
-    void GX_HLE_FIFO_Write8(uint8_t val);
-    void GX_HLE_FIFO_WriteBurst(const uint8_t* data, uint32_t sizeBytes);
-}
-
-namespace MemoryInline {
-#if defined(_WIN32)
-#define MKW_MEMORY_FORCE_INLINE __forceinline
-#define MKW_MEMORY_NO_INLINE __declspec(noinline)
-#else
-// See runtime/include/isa/ppc_isa_config.h for why non-Windows Clang needs the GNU-attribute
-// spellings instead of the MS-extension keywords.
-#define MKW_MEMORY_FORCE_INLINE __attribute__((always_inline)) inline
-#define MKW_MEMORY_NO_INLINE __attribute__((noinline))
-#endif
-#define MKW_MEMORY_COLD __attribute__((cold))
-inline constexpr uint32_t kPageShift = 20;
-inline constexpr uint32_t kPageSize = 1u << kPageShift;
-inline constexpr uint32_t kPageMask = kPageSize - 1u;
-inline constexpr uint32_t kPageCount = 1u << (32 - kPageShift);
-inline constexpr uint32_t kMaxFastScalarSize = 8;
-inline constexpr uint32_t kWritableSubPageShift = RecompMod::kExecutableWriteGuardPageShift;
-inline constexpr uint32_t kWritableSubPageSize = 1u << kWritableSubPageShift;
-inline constexpr uint32_t kWritableSubPagesPerPage = kPageSize / kWritableSubPageSize;
-
-struct PageEntry {
-    uint8_t* base = nullptr;
-    uint32_t limit = 0;
-};
-
-extern PageEntry g_pageTable[kPageCount];
-// Encoded (host page base - guest page base) + 1 for full pages whose next
-// page is contiguous. This permits any native access up to 8 bytes without a
-// per-access mask/limit check. Zero retains the general PageEntry fallback.
-extern uintptr_t g_fullPageBias[kPageCount];
-// Runtime-active readable biases. Deferred-read pages clear these entries once
-// instead of paying a mode branch on every translated read.
-extern uintptr_t g_fullReadablePageBias[kPageCount];
-// Same encoding, but only for pages proven not to contain executable bytes.
-// Executable-range registration invalidates entries before guest execution.
-extern uintptr_t g_fullWritablePageBias[kPageCount];
-
-// Allocated only for mapped 1 MiB pages that contain both executable and data
-// 4 KiB pages. Entries use the same encoded host bias as the coarse table.
-// Exact executable bits remain authoritative and are checked at lookup time,
-// including both sides of a cross-4-KiB access.
-struct SparseWritablePageTable {
-    uintptr_t encodedBias[kWritableSubPagesPerPage]{};
-};
-extern const SparseWritablePageTable* g_sparseWritablePageTables[kPageCount];
-
-// Nonzero while any registered deferred read overlaps the page. Small
-// mappings (e.g. the 16 KiB locked cache) have no coarse bias entry, so a
-// zero readable bias alone cannot distinguish "deferred content pending"
-// from "small but plain memory"; range resolution needs the explicit flag.
-extern uint8_t g_deferredReadCoveredPages[kPageCount];
-
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint8_t Read8Slow(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint16_t Read16Slow(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint32_t Read32Slow(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD uint64_t Read64Slow(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD float ReadFloat32Slow(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD double ReadFloat64Slow(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write8Slow(uint32_t addr, uint8_t val);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write16Slow(uint32_t addr, uint16_t val);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write32Slow(uint32_t addr, uint32_t val);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void Write64Slow(uint32_t addr, uint64_t val);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteFloat32Slow(uint32_t addr, double val);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteFloat64Slow(uint32_t addr, double val);
-template <typename T>
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD T ReadResolvedFallback(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD float ReadResolvedFallbackFloat32(uint32_t addr);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD double ReadResolvedFallbackFloat64(uint32_t addr);
-template <typename T>
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteResolvedFallback(uint32_t addr, T value);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteResolvedFallbackFloat32(uint32_t addr, double val);
-MKW_MEMORY_NO_INLINE MKW_MEMORY_COLD void WriteResolvedFallbackFloat64(uint32_t addr, double val);
-bool ResolveDeferredReads(uint32_t addr, size_t length);
-
-constexpr bool IsMmioAddress(uint32_t addr) {
-    return addr >= 0xCC000000u && addr < 0xCE000000u;
-}
-
-constexpr bool IsGpuFifoAddress(uint32_t addr) {
-    return addr >= 0xCC008000u && addr < 0xCC008100u;
-}
-
-// Page protections can't cover this: an MMIO write must reach GX HLE with its value or be
-// reported, and a fault record can't carry the value, so this mask/compare sits in front of
-// every flat store instead.
-MKW_MEMORY_FORCE_INLINE bool FlatWriteNeedsPolicy(uint32_t address) {
-    return (address & 0xFE000000u) == 0xCC000000u;  // 0xCC000000..0xCDFFFFFF
-}
-
-// Gekko stfs conversion is a bit-level narrowing operation.  In particular it
-// does not behave like a host double-to-float cast for values which were left
-// in double precision, and it has hardware-tested handling for tiny values.
-MKW_MEMORY_FORCE_INLINE uint32_t ConvertPpcDoubleToSingleBits(double value) {
-    uint64_t bits = 0;
-    std::memcpy(&bits, &value, sizeof(bits));
-    const uint32_t exponent = static_cast<uint32_t>((bits >> 52) & 0x7FFu);
-    // The subnormal-single arm applies to exactly the exponents 874..896; every
-    // other exponent takes the plain sign/exponent/fraction narrowing below.
-    // Inside that window the exponent field is nonzero, so the magnitude cannot
-    // be zero and needs no separate test - the zero case (exponent 0) reaches
-    // the narrowing exactly as it did when the two arms shared that test.
-    if (exponent - 874u <= 22u) [[unlikely]]
-    {
-        uint32_t narrowed = static_cast<uint32_t>(
-            0x80000000ULL | ((bits & 0x000FFFFFFFFFFFFFULL) >> 21));
-        narrowed >>= (905u - exponent);
-        narrowed |= static_cast<uint32_t>((bits >> 32) & 0x80000000ULL);
-        return narrowed;
-    }
-
-    // Results below the documented conversion range are architecturally
-    // undefined; this is the behavior measured on Gekko/Broadway hardware.
-    return static_cast<uint32_t>(
-        ((bits >> 32) & 0xC0000000ULL) | ((bits >> 29) & 0x3FFFFFFFULL));
-}
-
-MKW_MEMORY_FORCE_INLINE float PpcSingleBitsToFloat(uint32_t bits) {
-    float value = 0.0f;
-    std::memcpy(&value, &bits, sizeof(value));
-    return value;
-}
-
-MKW_MEMORY_FORCE_INLINE bool TryGetPointerFast(uint32_t address, size_t length, uint8_t*& pointer) {
-    const uint32_t page = address >> kPageShift;
-    if (length <= 8) {
-        const uintptr_t encodedBias = g_fullPageBias[page];
-        if (encodedBias != 0) {
-            pointer = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
-            return true;
-        }
-    }
-    const uint32_t offset = address & kPageMask;
-    const auto& entry = g_pageTable[page];
-    if (!entry.base || offset + length > entry.limit) {
-        pointer = nullptr;
-        return false;
-    }
-    pointer = entry.base + offset;
-    return true;
-}
-
-inline uint8_t* GetPointerFast(uint32_t address, size_t length) {
-    uint8_t* pointer = nullptr;
-    return TryGetPointerFast(address, length, pointer) ? pointer : nullptr;
-}
-
-MKW_MEMORY_FORCE_INLINE bool TryGetWritablePointerFast(
-    uint32_t address, size_t length, uint8_t*& pointer) {
-    // The sparse-table hit below distinguishes itself from the small-mapping
-    // hit through `pointer`, so the out-parameter must start null regardless
-    // of what the caller passed in. (A caller handing in an uninitialized
-    // pointer used to turn every sparse hit into a write through stack
-    // garbage - random host memory corruption.)
-    pointer = nullptr;
-    if (length == 0 || length > 8 || address > UINT32_MAX - (length - 1))
-        return false;
-    const uint32_t coarsePage = address >> kPageShift;
-    uintptr_t encodedBias = g_fullWritablePageBias[coarsePage];
-    const uint32_t endAddress = address + static_cast<uint32_t>(length - 1);
-    const uint32_t firstExactPage = address >> kWritableSubPageShift;
-    const uint32_t lastExactPage = endAddress >> kWritableSubPageShift;
-    if (encodedBias != 0 && (endAddress >> kPageShift) != coarsePage &&
-        RecompMod::g_executableWriteGuardPages[lastExactPage].load(
-            std::memory_order_relaxed) != 0) {
-        return false;
-    }
-    if (encodedBias == 0) {
-        const auto* subTable = g_sparseWritablePageTables[coarsePage];
-        if (subTable != nullptr) {
-            encodedBias = subTable->encodedBias[
-                (address & kPageMask) >> kWritableSubPageShift];
-            if (encodedBias == 0)
-                return false;
-        } else {
-            // Small mappings such as Broadway's 16 KiB locked cache cannot
-            // populate the full-1-MiB bias table. Keep them native by proving
-            // the exact access against the ordinary page entry, then applying
-            // the same executable-write policy as a sparse-table hit.
-            const uint32_t offset = address & kPageMask;
-            const auto& entry = g_pageTable[coarsePage];
-            if (!entry.base || offset + length > entry.limit)
-                return false;
-            pointer = entry.base + offset;
-        }
-
-        // The exact 4 KiB guard is the final authority. Checking it on every
-        // checked/sparse hit also makes later executable-range registration
-        // safe when a prebuilt table still contains the old mapped bias.
-        if (RecompMod::g_executableWriteGuardPages[firstExactPage].load(
-                std::memory_order_relaxed) != 0 ||
-            (lastExactPage != firstExactPage &&
-             RecompMod::g_executableWriteGuardPages[lastExactPage].load(
-                 std::memory_order_relaxed) != 0)) {
-            return false;
-        }
-
-        if (pointer != nullptr)
-            return true;
-    }
-    if (encodedBias == 0)
-        return false;
-    pointer = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
-    return true;
-}
-
-// Flat form: guest_flat_memory.h's page protections already answer mapped/non-deferred/non-executable, so resolving is pure address arithmetic.
-// Two checks stay inline: a wrapped guest address can't survive 64-bit `host + rangeOffset`, and an MMIO write's value isn't recoverable from a
-// fault record, so a write touching that window must resolve null and fall back to Memory::Write*. Checking both range endpoints is a complete
-// proof since length <= kPageSize (1 MiB) can't straddle the 32 MiB MMIO window.
-MKW_MEMORY_FORCE_INLINE uint8_t* ResolveRangeHost(uint32_t base, int32_t minOffset, uint32_t length,
-                                                 bool needsRead, bool needsWrite) {
-    (void)needsRead;
-    const uint32_t guestStart = base + static_cast<uint32_t>(minOffset);
-    if (length == 0 || length > kPageSize || guestStart > UINT32_MAX - (length - 1)) return nullptr;
-    if (GuestFlat::RequiresCheckedAccess()) {
-        // A host page can cover multiple independently-special Wii pages.
-        // Returning null keeps resolved accesses on the checked Memory::*
-        // path, which materializes deferred reads and applies write policy.
-        (void)needsWrite;
-        return nullptr;
-    }
-    if (needsWrite &&
-        (FlatWriteNeedsPolicy(guestStart) || FlatWriteNeedsPolicy(guestStart + (length - 1))))
-        [[unlikely]] return nullptr;
-    return MKW_FLAT_GUEST_BASE + guestStart;
-}
-
-// Guest-address byte order. Distinct from isa/big_endian.h, which is the
-// host-pointer codec; do not "unify" them.
-inline uint16_t ByteSwap16(uint16_t value) {
-    return __builtin_bswap16(value);
-}
-
-inline uint32_t ByteSwap32(uint32_t value) {
-    return __builtin_bswap32(value);
-}
-
-inline uint64_t ByteSwap64(uint64_t value) {
-    return __builtin_bswap64(value);
-}
-
-template <typename T>
-inline T MaybeByteSwap(T value) {
-    if constexpr (sizeof(T) == 1) {
-        return value;
-    } else if constexpr (sizeof(T) == 2) {
-        return static_cast<T>(ByteSwap16(static_cast<uint16_t>(value)));
-    } else if constexpr (sizeof(T) == 4) {
-        return static_cast<T>(ByteSwap32(static_cast<uint32_t>(value)));
-    } else if constexpr (sizeof(T) == 8) {
-        return static_cast<T>(ByteSwap64(static_cast<uint64_t>(value)));
-    } else {
-        return value;
-    }
-}
-
-template <typename T>
-MKW_MEMORY_FORCE_INLINE bool ReadResolvedScalar(uint8_t* host, uint32_t rangeOffset, T& outValue) {
-    if (!host) return false;
-    if constexpr (sizeof(T) == 1) {
-        outValue = host[rangeOffset];
-    } else {
-        T value = 0;
-        std::memcpy(&value, host + rangeOffset, sizeof(T));
-        outValue = MaybeByteSwap(value);
-    }
-    return true;
-}
-
-struct ResolvedLoadPair {
-    uint32_t first = 0;
-    uint32_t second = 0;
-    bool valid = false;
-};
-
-MKW_MEMORY_FORCE_INLINE ResolvedLoadPair ReadResolvedPair16(
-    uint8_t* host, uint32_t rangeOffset) {
-    uint32_t packed = 0;
-    if (!ReadResolvedScalar(host, rangeOffset, packed)) return {};
-    return {packed >> 16, packed & 0xFFFFu, true};
-}
-
-MKW_MEMORY_FORCE_INLINE ResolvedLoadPair ReadResolvedPair32(uint8_t* host, uint32_t rangeOffset) {
-    uint64_t packed = 0;
-    if (!ReadResolvedScalar(host, rangeOffset, packed)) return {};
-    return {static_cast<uint32_t>(packed >> 32), static_cast<uint32_t>(packed), true};
-}
-
-template <typename Packed>
-MKW_MEMORY_FORCE_INLINE bool WriteResolvedPairFast(
-    uint8_t* host, uint32_t rangeOffset, Packed packed) {
-    if (!host) return false;
-    const Packed swapped = MaybeByteSwap(packed);
-    std::memcpy(host + rangeOffset, &swapped, sizeof(swapped));
-    return true;
-}
-
-MKW_MEMORY_FORCE_INLINE bool WriteResolvedPair16(
-    uint8_t* host, uint32_t rangeOffset, uint32_t packed) {
-    return WriteResolvedPairFast(host, rangeOffset, packed);
-}
-
-MKW_MEMORY_FORCE_INLINE bool WriteResolvedPair32(
-    uint8_t* host, uint32_t rangeOffset, uint64_t packed) {
-    return WriteResolvedPairFast(host, rangeOffset, packed);
-}
-
-template <typename T>
-MKW_MEMORY_FORCE_INLINE bool WriteResolvedScalar(uint8_t* host, uint32_t rangeOffset, T value) {
-    if (!host) return false;
-    if constexpr (sizeof(T) == 1) {
-        host[rangeOffset] = static_cast<uint8_t>(value);
-    } else {
-        const T swapped = MaybeByteSwap(value);
-        std::memcpy(host + rangeOffset, &swapped, sizeof(T));
-    }
-    return true;
-}
-
-template <typename T>
-inline bool TryReadMappedScalar(uint32_t address, T& outValue) {
-    uint8_t* ptr = nullptr;
-    if (TryGetPointerFast(address, sizeof(T), ptr)) {
-        if constexpr (sizeof(T) == 1) {
-            outValue = *ptr;
-        } else {
-            T value = 0;
-            std::memcpy(&value, ptr, sizeof(T));
-            outValue = MaybeByteSwap(value);
-        }
-        return true;
-    }
-    return false;
-}
-
-template <typename T>
-MKW_MEMORY_FORCE_INLINE bool TryReadGuestScalar(uint32_t address, T& outValue) {
-    const uintptr_t encodedBias = g_fullReadablePageBias[address >> kPageShift];
-    if (encodedBias == 0) [[unlikely]]
-        return false;
-    auto* ptr = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
-    if constexpr (sizeof(T) == 1) {
-        outValue = *ptr;
-    } else {
-        T value = 0;
-        std::memcpy(&value, ptr, sizeof(T));
-        outValue = MaybeByteSwap(value);
-    }
-    return true;
-}
-
-template <typename T>
-MKW_MEMORY_FORCE_INLINE bool TryWriteGuestScalar(uint32_t address, T value) {
-    static_assert(sizeof(T) >= 1 && sizeof(T) <= kMaxFastScalarSize);
-    // A mixed executable/data 1 MiB page zeroes the coarse writable bias even though most of
-    // its 4 KiB sub-pages are plain data; MKW's THP buffers share such a page with .text, which
-    // used to force ~15% of total CPU through the cold path. The sparse sub-page tier below
-    // keeps those stores native while the exact 4 KiB executable guards stay authoritative.
-    if (address > UINT32_MAX - static_cast<uint32_t>(sizeof(T) - 1u)) [[unlikely]]
-        return false;
-    const uint32_t coarsePage = address >> kPageShift;
-    const uint32_t endAddress = address + static_cast<uint32_t>(sizeof(T) - 1u);
-    if ((endAddress >> kPageShift) != coarsePage) [[unlikely]]
-        return false;
-    const uintptr_t encodedBias = g_fullWritablePageBias[coarsePage];
-    uint8_t* ptr = nullptr;
-    if (encodedBias != 0) {
-        ptr = reinterpret_cast<uint8_t*>((encodedBias - 1u) + address);
-    } else if (!TryGetWritablePointerFast(address, sizeof(T), ptr)) [[unlikely]] {
-        return false;
-    }
-    if constexpr (sizeof(T) == 1) {
-        *ptr = static_cast<uint8_t>(value);
-    } else {
-        const T swapped = MaybeByteSwap(value);
-        std::memcpy(ptr, &swapped, sizeof(T));
-    }
-    return true;
-}
-
-template <typename T>
-inline bool WriteStackScalarFast(uint32_t address, T value) {
-    uint8_t* ptr = nullptr;
-    if (TryGetPointerFast(address, sizeof(T), ptr)) {
-        if constexpr (sizeof(T) == 1) {
-            *ptr = static_cast<uint8_t>(value);
-        } else {
-            const T swapped = MaybeByteSwap(value);
-            std::memcpy(ptr, &swapped, sizeof(T));
-        }
-        return true;
-    }
-    return false;
-}
-
-inline uint8_t ReadStack8(uint32_t address) {
-    uint8_t value = 0;
-    return TryReadMappedScalar(address, value) ? value : Memory::Read8(address);
-}
-
-inline uint16_t ReadStack16(uint32_t address) {
-    uint16_t value = 0;
-    return TryReadMappedScalar(address, value) ? value : Memory::Read16(address);
-}
-
-inline uint32_t ReadStack32(uint32_t address) {
-    uint32_t value = 0;
-    return TryReadMappedScalar(address, value) ? value : Memory::Read32(address);
-}
-
-inline uint64_t ReadStack64(uint32_t address) {
-    uint64_t value = 0;
-    return TryReadMappedScalar(address, value) ? value : Memory::Read64(address);
-}
-
-
-
-inline void WriteStack8(uint32_t address, uint8_t value) {
-    if (!WriteStackScalarFast(address, value)) {
-        Memory::Write8(address, value);
-    }
-}
-
-inline void WriteStack16(uint32_t address, uint16_t value) {
-    if (!WriteStackScalarFast(address, value)) {
-        Memory::Write16(address, value);
-    }
-}
-
-inline void WriteStack32(uint32_t address, uint32_t value) {
-    if (!WriteStackScalarFast(address, value)) {
-        Memory::Write32(address, value);
-    }
-}
-
-inline void WriteStack64(uint32_t address, uint64_t value) {
-    if (!WriteStackScalarFast(address, value)) {
-        Memory::Write64(address, value);
-    }
-}
-
-inline void WriteStackFloat32(uint32_t address, double value) {
-    const uint32_t bits = ConvertPpcDoubleToSingleBits(value);
-    if (!WriteStackScalarFast(address, bits)) {
-        Memory::WriteFloat32(address, value);
-    }
-}
-
-inline void WriteStackFloat64(uint32_t address, double value) {
-    uint64_t bits = 0;
-    std::memcpy(&bits, &value, sizeof(bits));
-    if (!WriteStackScalarFast(address, bits)) {
-        Memory::WriteFloat64(address, value);
-    }
-}
-
-template <typename T, typename SlowRead>
-MKW_MEMORY_FORCE_INLINE T ReadResolved(uint8_t* host, uint32_t rangeOffset, uint32_t address,
-                      SlowRead slow) {
-    T value = 0;
-    if (ReadResolvedScalar(host, rangeOffset, value)) {
-        return value;
-    }
-    [[unlikely]] return slow(address);
-}
-
-MKW_MEMORY_FORCE_INLINE uint8_t ReadResolved8(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint8_t>(r, o, a, ReadResolvedFallback<uint8_t>); }
-MKW_MEMORY_FORCE_INLINE uint16_t ReadResolved16(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint16_t>(r, o, a, ReadResolvedFallback<uint16_t>); }
-MKW_MEMORY_FORCE_INLINE uint32_t ReadResolved32(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint32_t>(r, o, a, ReadResolvedFallback<uint32_t>); }
-// Live via isa/ppc_isa_quantized.h (the psq resolved tier packs two lanes into
-// one 64-bit access); generated code never names it directly.
-MKW_MEMORY_FORCE_INLINE uint64_t ReadResolved64(uint8_t* r, uint32_t o, uint32_t a) { return ReadResolved<uint64_t>(r, o, a, ReadResolvedFallback<uint64_t>); }
-MKW_MEMORY_FORCE_INLINE float ReadResolvedFloat32(uint8_t* r, uint32_t o, uint32_t a) {
-    uint32_t bits = 0;
-    if (!ReadResolvedScalar(r, o, bits)) [[unlikely]] return ReadResolvedFallbackFloat32(a);
-    float value; std::memcpy(&value, &bits, sizeof(value)); return value;
-}
-MKW_MEMORY_FORCE_INLINE double ReadResolvedFloat64(uint8_t* r, uint32_t o, uint32_t a) {
-    uint64_t bits = 0;
-    if (!ReadResolvedScalar(r, o, bits)) [[unlikely]] return ReadResolvedFallbackFloat64(a);
-    double value; std::memcpy(&value, &bits, sizeof(value)); return value;
-}
-
-template <typename T, typename SlowWrite>
-MKW_MEMORY_FORCE_INLINE void WriteResolved(uint8_t* host, uint32_t rangeOffset, uint32_t address, T value,
-                          SlowWrite slow) {
-    if (WriteResolvedScalar(host, rangeOffset, value)) {
-        return;
-    }
-    [[unlikely]] slow(address, value);
-}
-
-MKW_MEMORY_FORCE_INLINE void WriteResolved8(uint8_t* r, uint32_t o, uint32_t a, uint8_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint8_t>); }
-MKW_MEMORY_FORCE_INLINE void WriteResolved16(uint8_t* r, uint32_t o, uint32_t a, uint16_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint16_t>); }
-MKW_MEMORY_FORCE_INLINE void WriteResolved32(uint8_t* r, uint32_t o, uint32_t a, uint32_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint32_t>); }
-// Live via isa/ppc_isa_quantized.h, as ReadResolved64 above.
-MKW_MEMORY_FORCE_INLINE void WriteResolved64(uint8_t* r, uint32_t o, uint32_t a, uint64_t v) { WriteResolved(r, o, a, v, WriteResolvedFallback<uint64_t>); }
-MKW_MEMORY_FORCE_INLINE void WriteResolvedFloat32(uint8_t* r, uint32_t o, uint32_t a, double v) {
-    const uint32_t bits = ConvertPpcDoubleToSingleBits(v);
-    if (WriteResolvedScalar(r, o, bits)) return;
-    [[unlikely]] WriteResolvedFallbackFloat32(a, v);
-}
-MKW_MEMORY_FORCE_INLINE void WriteResolvedFloat64(uint8_t* r, uint32_t o, uint32_t a, double v) {
-    uint64_t bits; std::memcpy(&bits, &v, sizeof(bits));
-    if (WriteResolvedScalar(r, o, bits)) return;
-    [[unlikely]] WriteResolvedFallbackFloat64(a, v);
-}
-
-// Flat guest memory (audit item T-MEM): the 4 GiB reservation makes a guest access a byte swap
-// around `*(T*)(base + addr)`, no page-table load or limit check (interception model documented
-// in guest_flat_memory.h). The one exception kept inline is the MMIO write policy, since the
-// written value can't be recovered from a fault record.
-//
-// When a host VM page is larger than a 4 KiB Wii page, guest-view protections
-// cannot distinguish adjacent special Wii pages. The general FlatRead*/
-// FlatWrite* helpers then use the checked page-table path, which materializes
-// deferred reads and applies executable-write/MMIO policy before touching RAM.
-// FlatWriteRam* remains direct because the translator emits it only for
-// addresses it has proven are ordinary RAM.
-
-template <typename T>
-MKW_MEMORY_FORCE_INLINE T FlatLoad(uint32_t address) {
-    T value{};
-    std::memcpy(&value, MKW_FLAT_GUEST_BASE + address, sizeof(T));
-    return MaybeByteSwap(value);
-}
-
-template <typename T>
-MKW_MEMORY_FORCE_INLINE void FlatStore(uint32_t address, T value) {
-    const T swapped = MaybeByteSwap(value);
-    std::memcpy(MKW_FLAT_GUEST_BASE + address, &swapped, sizeof(T));
-}
-
-MKW_MEMORY_FORCE_INLINE uint8_t FlatRead8(uint32_t address) {
-    if (GuestFlat::RequiresCheckedAccess()) return Memory::Read8(address);
-    return FlatLoad<uint8_t>(address);
-}
-MKW_MEMORY_FORCE_INLINE uint16_t FlatRead16(uint32_t address) {
-    if (GuestFlat::RequiresCheckedAccess()) return Memory::Read16(address);
-    return FlatLoad<uint16_t>(address);
-}
-MKW_MEMORY_FORCE_INLINE uint32_t FlatRead32(uint32_t address) {
-    if (GuestFlat::RequiresCheckedAccess()) return Memory::Read32(address);
-    return FlatLoad<uint32_t>(address);
-}
-
-MKW_MEMORY_FORCE_INLINE float FlatReadFloat32(uint32_t address) {
-    const uint32_t bits = FlatRead32(address);
-    float value = 0.0f;
-    std::memcpy(&value, &bits, sizeof(value));
-    return value;
-}
-
-MKW_MEMORY_FORCE_INLINE double FlatReadFloat64(uint32_t address) {
-    const uint64_t bits = GuestFlat::RequiresCheckedAccess()
-        ? Memory::Read64(address) : FlatLoad<uint64_t>(address);
-    double value = 0.0;
-    std::memcpy(&value, &bits, sizeof(value));
-    return value;
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWrite8(uint32_t address, uint8_t value) {
-    if (GuestFlat::RequiresCheckedAccess()) { Memory::Write8(address, value); return; }
-    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { Write8Slow(address, value); return; }
-    FlatStore<uint8_t>(address, value);
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWrite16(uint32_t address, uint16_t value) {
-    if (GuestFlat::RequiresCheckedAccess()) { Memory::Write16(address, value); return; }
-    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { Write16Slow(address, value); return; }
-    FlatStore<uint16_t>(address, value);
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWrite32(uint32_t address, uint32_t value) {
-    if (GuestFlat::RequiresCheckedAccess()) { Memory::Write32(address, value); return; }
-    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { Write32Slow(address, value); return; }
-    FlatStore<uint32_t>(address, value);
-}
-
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteFloat32(uint32_t address, double value) {
-    if (GuestFlat::RequiresCheckedAccess()) { Memory::WriteFloat32(address, value); return; }
-    const uint32_t bits = ConvertPpcDoubleToSingleBits(value);
-    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { WriteFloat32Slow(address, value); return; }
-    FlatStore<uint32_t>(address, bits);
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteFloat64(uint32_t address, double value) {
-    if (GuestFlat::RequiresCheckedAccess()) { Memory::WriteFloat64(address, value); return; }
-    uint64_t bits = 0;
-    std::memcpy(&bits, &value, sizeof(bits));
-    if (FlatWriteNeedsPolicy(address)) [[unlikely]] { WriteFloat64Slow(address, value); return; }
-    FlatStore<uint64_t>(address, bits);
-}
-
-// Check-free stores: emitted ONLY for addresses the translator proved at translate time are ordinary guest RAM (r1-relative stack slots, ~45%
-// of flat stores), skipping the MMIO mask/compare that's pure overhead there. Still safe if that proof were ever wrong: the flat view maps
-// 0xCC000000..0xCDFFFFFF PAGE_NOACCESS, so a stray MMIO store faults into the same handler and diagnostic as the checked path, just reported
-// instead of dispatched inline. Never use these for an address the translator hasn't proven.
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteRam8(uint32_t address, uint8_t value) {
-    FlatStore<uint8_t>(address, value);
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteRam16(uint32_t address, uint16_t value) {
-    FlatStore<uint16_t>(address, value);
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteRam32(uint32_t address, uint32_t value) {
-    FlatStore<uint32_t>(address, value);
-}
-
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteRamFloat32(uint32_t address, double value) {
-    FlatStore<uint32_t>(address, ConvertPpcDoubleToSingleBits(value));
-}
-
-MKW_MEMORY_FORCE_INLINE void FlatWriteRamFloat64(uint32_t address, double value) {
-    uint64_t bits = 0;
-    std::memcpy(&bits, &value, sizeof(bits));
-    FlatStore<uint64_t>(address, bits);
-}
-
-} // namespace MemoryInline
-
-#undef MKW_MEMORY_FORCE_INLINE
-#undef MKW_MEMORY_NO_INLINE
-#undef MKW_MEMORY_COLD
-
-inline uint8_t Memory::Read8(uint32_t addr) {
-    uint8_t value = 0;
-    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
-        return MemoryInline::Read8Slow(addr);
-    return value;
-}
-
-inline uint16_t Memory::Read16(uint32_t addr) {
-    uint16_t value = 0;
-    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
-        return MemoryInline::Read16Slow(addr);
-    return value;
-}
-
-inline uint32_t Memory::Read32(uint32_t addr) {
-    uint32_t value = 0;
-    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
-        return MemoryInline::Read32Slow(addr);
-    return value;
-}
-
-inline uint64_t Memory::Read64(uint32_t addr) {
-    uint64_t value = 0;
-    if (!MemoryInline::TryReadGuestScalar(addr, value)) [[unlikely]]
-        return MemoryInline::Read64Slow(addr);
-    return value;
-}
-
-inline float Memory::ReadFloat32(uint32_t addr) {
-    uint32_t bits = 0;
-    if (!MemoryInline::TryReadGuestScalar(addr, bits)) [[unlikely]]
-        return MemoryInline::ReadFloat32Slow(addr);
-    float value = 0.0f;
-    std::memcpy(&value, &bits, sizeof(value));
-    return value;
-}
-
-inline double Memory::ReadFloat64(uint32_t addr) {
-    uint64_t bits = 0;
-    if (!MemoryInline::TryReadGuestScalar(addr, bits)) [[unlikely]]
-        return MemoryInline::ReadFloat64Slow(addr);
-    double value = 0.0;
-    std::memcpy(&value, &bits, sizeof(value));
-    return value;
-}
-
-inline void Memory::Write8(uint32_t addr, uint8_t val) {
-    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
-        MemoryInline::Write8Slow(addr, val);
-}
-
-inline void Memory::Write16(uint32_t addr, uint16_t val) {
-    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
-        MemoryInline::Write16Slow(addr, val);
-}
-
-inline void Memory::Write32(uint32_t addr, uint32_t val) {
-    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
-        MemoryInline::Write32Slow(addr, val);
-}
-
-inline void Memory::Write64(uint32_t addr, uint64_t val) {
-    if (!MemoryInline::TryWriteGuestScalar(addr, val)) [[unlikely]]
-        MemoryInline::Write64Slow(addr, val);
-}
-
-inline void Memory::WriteFloat32(uint32_t addr, double val) {
-    {
-        const uint32_t bits = MemoryInline::ConvertPpcDoubleToSingleBits(val);
-        if (MemoryInline::TryWriteGuestScalar(addr, bits))
-            return;
-    }
-    [[unlikely]] MemoryInline::WriteFloat32Slow(addr, val);
-}
-
-inline void Memory::WriteFloat64(uint32_t addr, double val) {
-    {
-        uint64_t bits = 0;
-        std::memcpy(&bits, &val, sizeof(bits));
-        if (MemoryInline::TryWriteGuestScalar(addr, bits))
-            return;
-    }
-    [[unlikely]] MemoryInline::WriteFloat64Slow(addr, val);
-}
+#include "guest_abi/memory_access.h"

runtime/include/ppc_isa_memory.h

@@ -3,8 +3,6 @@
 // Host memory seam for isa/ppc_isa_quantized.h. That header includes
 // "ppc_isa_memory.h" by plain quoted name and expects whatever it resolves to
 // to supply the Memory / MemoryInline APIs and the flat-guest-base macros its
-// psq fast paths use. This file is this project's implementation of that
-// contract; memory.h chains memory_access.h, which defines MemoryInline.
+// psq fast paths use. Part of the guest ABI, so it resolves to guest_abi/ only.
 
-#include "memory.h" // chains memory_access.h, which supplies MemoryInline and
-                    // includes guest_flat_memory.h for the flat-base macros
+#include "guest_abi/memory.h"

runtime/include/ppc_runtime.h

@@ -12,3 +12,4 @@
 #include "isa/ppc_isa_int.h"
 #include "isa/ppc_isa_float.h"
 #include "isa/ppc_isa_quantized.h"
+#include "memory.h"

runtime/include/recomp_mod_loader.h

@@ -8,6 +8,8 @@
 #include <string_view>
 #include <vector>
 
+#include "guest_abi/exec_guard.h"
+
 struct CpuContext;
 
 namespace RecompMod {
@@ -20,46 +22,6 @@ struct MemoryReservation {
     std::string name;
 };
 
-// Diagnostic-only, read by fatal reporters (unmapped access, executable-write, CPU dumps).
-// Defined here as `inline thread_local` with a constant initializer, not `extern thread_local`
-// in the .cpp: every indirect dispatch scopes this, so an out-of-line ctor/dtor would cost two
-// un-inlinable calls plus a register spill each for three instructions of work.
-inline thread_local uint32_t g_currentTranslatedExecutionAddress = 0;
-
-class ScopedTranslatedExecutionAddress {
-public:
-    explicit ScopedTranslatedExecutionAddress(uint32_t address) noexcept
-        : previous_(g_currentTranslatedExecutionAddress) {
-        // Address 0 means "nothing new to report" - the enclosing scope's value
-        // stays visible, and the unconditional restore below keeps nesting exact.
-        if (address != 0) {
-            g_currentTranslatedExecutionAddress = address;
-        }
-    }
-
-    ~ScopedTranslatedExecutionAddress() noexcept {
-        g_currentTranslatedExecutionAddress = previous_;
-    }
-
-    ScopedTranslatedExecutionAddress(const ScopedTranslatedExecutionAddress&) = delete;
-    ScopedTranslatedExecutionAddress& operator=(const ScopedTranslatedExecutionAddress&) = delete;
-
-private:
-    uint32_t previous_ = 0;
-};
-
-inline constexpr uint32_t kExecutableWriteGuardPageShift = 12;
-inline constexpr uint32_t kExecutableWriteGuardPageCount = 1u << (32 - kExecutableWriteGuardPageShift);
-inline constexpr uint32_t kExecutableWriteGuardCoarsePageShift = 20;
-inline constexpr uint32_t kExecutableWriteGuardCoarsePageCount = 1u << (32 - kExecutableWriteGuardCoarsePageShift);
-inline constexpr uint32_t kExecutableWriteGuardMidPageShift = 16;
-inline constexpr uint32_t kExecutableWriteGuardMidPageCount = 1u << (32 - kExecutableWriteGuardMidPageShift);
-
-extern std::atomic<bool> g_executableWriteGuardEnabled;
-extern std::atomic<uint8_t> g_executableWriteGuardPages[kExecutableWriteGuardPageCount];
-extern std::atomic<uint8_t> g_executableWriteGuardCoarsePages[kExecutableWriteGuardCoarsePageCount];
-extern std::atomic<uint8_t> g_executableWriteGuardMidPages[kExecutableWriteGuardMidPageCount];
-
 // Two initializer phases, both emitted by the translator's mod data-patch
 // writer. (A third, plain RegisterInitializer/RunInitializers pair existed with
 // no registrant on either side and was removed.)

runtime/include/runtime_config.h

@@ -69,10 +69,11 @@ struct RuntimeUserConfig {
     // SDL's sensor frame (x right, y out of the button face, z towards the user).
     // SDL's Wii driver falls back to a nominal zero point when its read of the
     // remote's calibration block times out (common over Bluetooth), so this is
-    // measured in the overlay with the remote at rest.
-    std::optional<double> wiiAccelOffsetX;
-    std::optional<double> wiiAccelOffsetY;
-    std::optional<double> wiiAccelOffsetZ;
+    // measured in the overlay with the remote at rest. Stored per remote under
+    // [controller.wii_accel_offsets]; the old shared
+    // wii_accel_offset_* keys are only detected, to report that they are ignored.
+    std::map<std::string, std::array<double, 3>> wiiAccelOffsets;
+    bool hasLegacyWiiAccelOffset = false;
     // Debugging aid: append every KPAD sample of the Bluetooth remote (raw and
     // corrected accelerometer, buttons) to wii_accel_trace.csv next to Config.toml.
     std::optional<bool> wiiAccelTrace;
@@ -138,28 +139,6 @@ inline std::string Trim(std::string_view text) {
     return std::string(text.substr(begin, end - begin));
 }
 
-inline std::string RemoveComment(std::string_view line) {
-    bool inSingle = false;
-    bool inDouble = false;
-    bool escaped = false;
-    for (size_t i = 0; i < line.size(); ++i) {
-        const char ch = line[i];
-        if (inDouble && ch == '\\' && !escaped) {
-            escaped = true;
-            continue;
-        }
-        if (ch == '\'' && !inDouble) {
-            inSingle = !inSingle;
-        } else if (ch == '"' && !inSingle && !escaped) {
-            inDouble = !inDouble;
-        } else if (ch == '#' && !inSingle && !inDouble) {
-            return std::string(line.substr(0, i));
-        }
-        escaped = false;
-    }
-    return std::string(line);
-}
-
 inline bool IsSupportedResolutionMultiplier(float value) {
     static constexpr std::array values{0.0f, 1.0f, 1.5f, 2.0f, 3.0f, 4.0f, 6.0f, 8.0f};
     return std::find(values.begin(), values.end(), value) != values.end();
@@ -403,6 +382,28 @@ inline void AppendOverlayRoots(RuntimeUserConfig& config, const std::string& roo
 }
 
 // Reads every supported setting out of a parsed Config.toml document.
+// [x, y, z] of finite numbers (integers allowed), else nullopt.
+inline std::optional<std::array<double, 3>> ParseAccelOffset(const toml::value& value) {
+    if (!value.is_array() || value.as_array().size() != 3) {
+        return std::nullopt;
+    }
+    std::array<double, 3> offset{};
+    for (size_t i = 0; i < 3; ++i) {
+        const toml::value& element = value.as_array()[i];
+        if (element.is_floating()) {
+            offset[i] = element.as_floating();
+        } else if (element.is_integer()) {
+            offset[i] = static_cast<double>(element.as_integer());
+        } else {
+            return std::nullopt;
+        }
+        if (!std::isfinite(offset[i])) {
+            return std::nullopt;
+        }
+    }
+    return offset;
+}
+
 inline RuntimeUserConfig ParseConfigDocument(const toml::value& document) {
     RuntimeUserConfig config;
 
@@ -426,6 +427,13 @@ inline RuntimeUserConfig ParseConfigDocument(const toml::value& document) {
                 config.controllerExpressions[key] = value.as_string();
             }
         }
+        if (section->contains("wii_accel_offsets") && section->at("wii_accel_offsets").is_table()) {
+            for (const auto& [key, value] : section->at("wii_accel_offsets").as_table()) {
+                if (const auto offset = ParseAccelOffset(value)) {
+                    config.wiiAccelOffsets[key] = *offset;
+                }
+            }
+        }
     }
 
     config.widescreen = FindConfigValue<bool>(document, "video", "widescreen");
@@ -487,9 +495,9 @@ inline RuntimeUserConfig ParseConfigDocument(const toml::value& document) {
         FindConfigValue<bool>(document, "audio", "attenuate_music_when_media_plays");
     config.wiiRemotes = FindConfigValue<bool>(document, "controller", "wii_remotes");
     config.wiiContinuousScan = FindConfigValue<bool>(document, "controller", "wii_continuous_scan");
-    config.wiiAccelOffsetX = FindConfigValue<double>(document, "controller", "wii_accel_offset_x");
-    config.wiiAccelOffsetY = FindConfigValue<double>(document, "controller", "wii_accel_offset_y");
-    config.wiiAccelOffsetZ = FindConfigValue<double>(document, "controller", "wii_accel_offset_z");
+    for (const char* key : {"wii_accel_offset_x", "wii_accel_offset_y", "wii_accel_offset_z"}) {
+        config.hasLegacyWiiAccelOffset |= FindConfigValue<double>(document, "controller", key).value_or(0.0) != 0.0;
+    }
     config.wiiAccelTrace = FindConfigValue<bool>(document, "controller", "wii_accel_trace");
     config.networkEnabled = FindConfigValue<bool>(document, "network", "enabled");
     config.discordPresenceEnabled = FindConfigValue<bool>(document, "discord", "enabled");
@@ -546,68 +554,50 @@ inline const std::optional<std::string>& ControllerButton(size_t index) {
     return index < Get().controllerButtons.size() ? Get().controllerButtons[index] : empty;
 }
 
-// Update one TOML value without discarding comments, unrelated settings, or
-// user-specific paths. This is used by the in-game F10 settings bar.
+// Sets section.key (section may be a dotted table path) to a TOML literal, keeping the rest of the
+// document's keys, order and comments. Returns nullopt when the document or value does not parse.
+inline std::optional<std::string> UpdateSettingText(const std::string& text, std::string_view section,
+                                                    std::string_view key, std::string_view value) {
+    try {
+        auto root = toml::parse_str<toml::ordered_type_config>(text);
+        auto literal = toml::parse_str<toml::ordered_type_config>("v = " + std::string(value));
+        auto* table = &root;
+        std::istringstream path{std::string(section)};
+        for (std::string part; std::getline(path, part, '.');) {
+            auto& parent = table->as_table();
+            if (!parent.contains(part)) {
+                parent[part] = toml::ordered_value(toml::ordered_table{});
+            }
+            table = &parent.at(part);
+            if (!table->is_table()) {
+                return std::nullopt;
+            }
+        }
+        if (table->as_table_fmt().fmt == toml::table_format::implicit) {
+            table->as_table_fmt().fmt = toml::table_format::multiline;
+        }
+        auto& slot = table->as_table()[std::string(key)];
+        auto& updated = literal.as_table().at("v");
+        updated.comments() = slot.comments();
+        slot = std::move(updated);
+        return toml::format(root);
+    } catch (const std::exception&) {
+        return std::nullopt;
+    }
+}
+
+// Update one TOML value without discarding unrelated settings or user-specific paths. This is
+// used by the in-game F10 settings bar. An unparseable file is left untouched.
 inline bool WriteSetting(std::string_view section, std::string_view key, std::string_view value) {
     const auto path = ResolveConfigPath();
-    std::ifstream input(path);
-    std::vector<std::string> lines;
-    std::string line;
-    while (std::getline(input, line)) {
-        if (!line.empty() && line.back() == '\r') {
-            line.pop_back();
-        }
-        lines.push_back(std::move(line));
+    std::ostringstream existing;
+    if (std::ifstream input(path, std::ios::binary); input) {
+        existing << input.rdbuf();
     }
-
-    const std::string normalizedSection = Trim(section);
-    const std::string normalizedKey = Trim(key);
-    size_t sectionStart = lines.size();
-    size_t sectionEnd = lines.size();
-    for (size_t i = 0; i < lines.size(); ++i) {
-        const std::string trimmed = Trim(RemoveComment(lines[i]));
-        if (trimmed.size() >= 2 && trimmed.front() == '[' && trimmed.back() == ']') {
-            const std::string found = Trim(std::string_view(trimmed).substr(1, trimmed.size() - 2));
-            if (sectionStart != lines.size()) {
-                sectionEnd = i;
-                break;
-            }
-            if (found == normalizedSection) {
-                sectionStart = i;
-            }
-        }
-    }
-
-    const std::string replacement = normalizedKey + " = " + std::string(value);
-    if (sectionStart == lines.size()) {
-        if (!lines.empty() && !lines.back().empty()) {
-            lines.emplace_back();
-        }
-        lines.emplace_back("[" + normalizedSection + "]");
-        lines.push_back(replacement);
-    } else {
-        bool replaced = false;
-        for (size_t i = sectionStart + 1; i < sectionEnd; ++i) {
-            const std::string uncommented = Trim(RemoveComment(lines[i]));
-            const size_t equals = uncommented.find('=');
-            if (equals != std::string::npos && Trim(std::string_view(uncommented).substr(0, equals)) == normalizedKey) {
-                lines[i] = replacement;
-                replaced = true;
-                break;
-            }
-        }
-        if (!replaced) {
-            // Append after the section's last real line rather than after the blank line that
-            // separates it from the next header: this file is edited by hand and by the host
-            // installer as well, and a key parked below the separator reads as if it belonged to
-            // the next section. The host writer (Launcher/WiiCompiled.Setup/RuntimeConfiguration.cs)
-            // applies exactly this rule.
-            size_t insertAt = sectionEnd;
-            while (insertAt > sectionStart + 1 && Trim(lines[insertAt - 1]).empty()) {
-                --insertAt;
-            }
-            lines.insert(lines.begin() + static_cast<std::ptrdiff_t>(insertAt), replacement);
-        }
+    const auto updated = UpdateSettingText(existing.str(), section, key, value);
+    if (!updated) {
+        std::cerr << "[runtime-config] Not updating invalid " << PathToUtf8(path) << std::endl;
+        return false;
     }
 
     std::error_code ec;
@@ -619,9 +609,7 @@ inline bool WriteSetting(std::string_view section, std::string_view key, std::st
         std::cerr << "[runtime-config] Unable to write " << PathToUtf8(path) << std::endl;
         return false;
     }
-    for (const auto& outputLine : lines) {
-        output << outputLine << '\n';
-    }
+    output << *updated;
     return static_cast<bool>(output);
 }
 
@@ -884,12 +872,13 @@ inline bool SetWiiContinuousScanEnabled(bool value) {
     return WriteSetting("controller", "wii_continuous_scan", value ? "true" : "false");
 }
 
-// Wii Remote accelerometer zero-point correction (g, SDL sensor frame); all zero
-// when the remote has not been calibrated.
-inline std::array<double, 3> WiiAccelOffset() {
-    const RuntimeUserConfig& config = Get();
-    return {config.wiiAccelOffsetX.value_or(0.0), config.wiiAccelOffsetY.value_or(0.0),
-            config.wiiAccelOffsetZ.value_or(0.0)};
+// Stored correction of the remote identified by `remoteKey`, if it has one.
+inline std::optional<std::array<double, 3>> WiiAccelOffsetFor(const std::string& remoteKey) {
+    const auto it = Get().wiiAccelOffsets.find(remoteKey);
+    if (it == Get().wiiAccelOffsets.end()) {
+        return std::nullopt;
+    }
+    return it->second;
 }
 
 // Whether to write the per-frame accelerometer trace (off unless asked for).
@@ -897,26 +886,14 @@ inline bool WiiAccelTraceEnabled(bool fallback = false) {
     return Get().wiiAccelTrace.value_or(fallback);
 }
 
-// True while a non-zero correction is stored ("Clear calibration" writes zeros).
-inline bool HasWiiAccelOffset() {
-    const std::array<double, 3> offset = WiiAccelOffset();
-    return offset[0] != 0.0 || offset[1] != 0.0 || offset[2] != 0.0;
-}
-
-// Persists the accelerometer correction measured by the overlay's calibration.
-inline bool SetWiiAccelOffset(const std::array<double, 3>& offset) {
-    Mutable().wiiAccelOffsetX = offset[0];
-    Mutable().wiiAccelOffsetY = offset[1];
-    Mutable().wiiAccelOffsetZ = offset[2];
-    bool ok = true;
-    const char* keys[3] = {"wii_accel_offset_x", "wii_accel_offset_y", "wii_accel_offset_z"};
-    for (size_t i = 0; i < 3; ++i) {
-        // Always a float literal, so a whole-number offset does not come back as a TOML integer.
-        std::ostringstream formatted;
-        formatted << std::fixed << std::setprecision(4) << offset[i];
-        ok = WriteSetting("controller", keys[i], formatted.str()) && ok;
-    }
-    return ok;
+// Persists the accelerometer correction of one remote. `remoteKey` must be a bare TOML key.
+inline bool SetWiiAccelOffsetFor(const std::string& remoteKey, const std::array<double, 3>& offset) {
+    Mutable().wiiAccelOffsets[remoteKey] = offset;
+    // Always float literals, so a whole-number offset does not come back as a TOML integer.
+    std::ostringstream formatted;
+    formatted << std::fixed << std::setprecision(4) << '[' << offset[0] << ", " << offset[1] << ", " << offset[2]
+              << ']';
+    return WriteSetting("controller.wii_accel_offsets", remoteKey, formatted.str());
 }
 
 // Target frame rate for frame interpolation, or 0 to disable it.

runtime/include/wii_remote_input.h

@@ -1,6 +1,8 @@
 #pragma once
 
+#include <array>
 #include <cstdint>
+#include <optional>
 
 struct PADStatus;
 
@@ -73,23 +75,20 @@ bool ReadKpadSample(uint32_t chan, KpadSample& sample);
 // no remote or SDL has not delivered a sample yet.
 bool ReadAccelDebug(uint32_t chan, float sdlG[3], float kpadAcc[3]);
 
-// Accelerometer zero-point calibration. SDL's Wii driver tries to read the
-// remote's factory calibration block, but over Bluetooth that read often times
-// out ("Using fallback accelerometer calibration" in console.log) and it falls
-// back to a nominal 0x200 zero point; a real remote then carries a per-axis
-// bias of up to ~0.2 g, which held sideways as a wheel is a permanent steering
-// offset.
-// StartAccelCalibration expects the remote lying still with the buttons up;
-// Poll() then collects samples for about a second and stores the difference to
-// the ideal (0, 1, 0) g in Config.toml.
+// Residual zero-point trim on top of SDL's factory-calibrated decode, stored per
+// physical remote under [controller.wii_accel_offsets]. Expects the remote lying
+// still, buttons up; Poll() averages about a second and stores it for that remote.
 void StartAccelCalibration(uint32_t chan);
-// Forgets the stored correction.
-void ClearAccelCalibration();
-bool IsAccelCalibrating();
+// Forgets the stored correction of the remote on `chan`.
+void ClearAccelCalibration(uint32_t chan);
+bool IsAccelCalibrating(uint32_t chan);
 // 0..1 while a calibration is collecting samples.
 float AccelCalibrationProgress();
-// Outcome of the last calibration attempt for the overlay, or nullptr.
-const char* AccelCalibrationMessage();
+// Correction in effect for the remote on `chan` (zeros when uncalibrated), or
+// nullopt when the port has no remote.
+std::optional<std::array<double, 3>> ReadAccelOffset(uint32_t chan);
+// Outcome of the last calibration attempt on `chan` for the overlay, or nullptr.
+const char* AccelCalibrationMessage(uint32_t chan);
 
 // Reports "no controller" on the GameCube side for every port served through
 // KPAD, so the game never sees the same remote twice. Runs after aurora's PADRead.

runtime/src/abi_bridge.cpp

@@ -4,13 +4,15 @@
 #include <array>
 #include <atomic>
 #include <cstddef>
+#include <cstdio>
+#include <cstdlib>
 #include <iomanip>
 #include <iostream>
 #include <mutex>
 #include <sstream>
 #include <unordered_map>
 #include <vector>
-#include "generated/RuntimeConfig.h"
+#include "guest_small_data.h"
 #include "runtime_log.h"
 
 // Nothing reads CpuContext by offset; these asserts just canary accidental field reordering
@@ -253,6 +255,41 @@ std::string ValidateGeneratedIndirectDispatchLocked() {
 }
 }
 
+uint32_t FilterScnRendererPathMaskOption(uint32_t pathMask) noexcept {
+    return RuntimeGameGraphicsOptions::FilterScnRendererPathMask(pathMask);
+}
+
+void ReportMissingIndirectJumpTarget(uint32_t target, CpuContext* cpu) {
+    // If not registered, this is likely a switch table jump to an intra-function label.
+    // The translator should have recognized this pattern and generated proper switch code.
+    RT_LOG(RT_TAG_RUNTIME) << "InvokeIndirectJump: target 0x"
+              << std::hex << target << std::dec
+              << " is not a registered function.\n"
+              << "This is likely a switch statement jump table that the translator\n"
+              << "failed to recognize. The function containing this bctr needs\n"
+              << "proper switch pattern recognition." << std::endl;
+    RT_LOG(RT_TAG_RUNTIME) << "Indirect jump context: pc=0x" << std::hex << cpu->pc
+              << " lr=0x" << cpu->lr << " ctr=0x" << cpu->ctr
+              << " r1=0x" << cpu->gpr[1] << " r3=0x" << cpu->gpr[3]
+              << " r29=0x" << cpu->gpr[29] << " r30=0x" << cpu->gpr[30]
+              << " r31=0x" << cpu->gpr[31] << std::dec << std::endl;
+    DumpHostStackTraceForRuntimeHelper();
+    std::fflush(stderr);
+    std::ostringstream message;
+    message << "The game stopped because an indirect jump targeted guest address 0x"
+            << std::hex << target
+            << ", but that address is not a registered translated function.\n\n"
+            << "The translator may have missed a switch/jump-table target.";
+    // Same artifact set as RuntimeCrash::FatalMissingGuestTarget, which is this
+    // path's sibling for indirect *calls*: MarkFatalErrorReported below stops the
+    // atexit reporter, so the crash log has to be written here.
+    RuntimeCrash::WriteCrashArtifacts("missing_jump_target", message.str(), &target);
+    SetRuntimeExitCode(EXIT_FAILURE);
+    ShowRuntimeFatalPopup("Missing indirect jump target", message.str());
+    MarkFatalErrorReported();
+    std::exit(EXIT_FAILURE);
+}
+
 void RegisterStaticIndirectDispatchTable(const StaticIndirectDispatchTable* table) {
     std::lock_guard<std::mutex> lock(RegistryMutex());
     if (RegistryFrozen().load(std::memory_order_acquire)) {
@@ -403,7 +440,7 @@ const RawDispatchRecord* TranslatedFunctionRegistry::FindRawByAddressPtrSlow(uin
     return it != entries.end() && it->address == address ? &*it : nullptr;
 }
 
-std::optional<TranslatedFunctionInfo> TranslatedFunctionRegistry::FindByHostAddress(uintptr_t hostAddr) {
+std::optional<TranslatedFunctionInfo> FindTranslatedFunctionByHostAddress(uintptr_t hostAddr) {
     const bool frozen = RegistryFrozen().load(std::memory_order_acquire);
     if (!frozen) {
         std::lock_guard<std::mutex> lock(RegistryMutex());
@@ -439,7 +476,7 @@ std::optional<TranslatedFunctionInfo> TranslatedFunctionRegistry::FindByHostAddr
     return std::nullopt;
 }
 
-std::optional<TranslatedFunctionInfo> TranslatedFunctionRegistry::FindNearestByAddress(uint32_t address) {
+std::optional<TranslatedFunctionInfo> FindNearestTranslatedFunction(uint32_t address) {
     std::optional<TranslatedFunctionInfo> best;
     uint32_t bestDelta = UINT32_MAX;
     const auto scan = [&] {
@@ -478,9 +515,24 @@ void InitializePersistentCpuContext() {
     // seed the PowerPC ABI environmental registers if no one has yet:
     // r2 = _SDA2_BASE_ (read-only small data), r13 = _SDA_BASE_.
     if (g_persistentCpu.gpr[2] == 0) {
-        g_persistentCpu.gpr[2] = RuntimeConfig::SDA2_BASE;
+        g_persistentCpu.gpr[2] = GuestSmallData::kSda2Base;
     }
     if (g_persistentCpu.gpr[13] == 0) {
-        g_persistentCpu.gpr[13] = RuntimeConfig::SDA1_BASE;
+        g_persistentCpu.gpr[13] = GuestSmallData::kSda1Base;
     }
 }
+
+[[noreturn]] void PPC_Undefined(uint32_t pc, uint32_t rawInstruction, const char* details)
+{
+    std::fprintf(stderr,
+                 "[runtime] UNDEFINED guest instruction: pc=0x%08X raw=0x%08X %s\n",
+                 pc,
+                 rawInstruction,
+                 details ? details : "");
+    char message[256]{};
+    std::snprintf(message, sizeof(message),
+                  "The game stopped because it reached an unsupported guest instruction at PC 0x%08X (instruction 0x%08X).\n\n%s",
+                  pc, rawInstruction, details ? details : "No additional details were provided.");
+    ShowRuntimeFatalPopup("Unsupported guest instruction", message);
+    std::abort();
+}

runtime/src/hle/input/pad.cpp

@@ -89,6 +89,11 @@ void WritePadStatus(uint32_t base, const PADStatus& status) {
 
 } // namespace
 
+extern "C" bool PAD_HLE_RumbleEnabled()
+{
+    return g_rumbleEnabled.load(std::memory_order_relaxed);
+}
+
 extern "C" void PAD_HLE_SetRumbleEnabled(bool enabled)
 {
     g_rumbleEnabled.store(enabled, std::memory_order_relaxed);

runtime/src/hle/input/wpad.cpp

@@ -5,6 +5,9 @@
 
 #include <cstdint>
 
+#include <dolphin/pad.h>
+
+extern "C" bool PAD_HLE_RumbleEnabled();
 void NandQueueIosCallback(uint32_t callbackPtr, int32_t result, uint32_t callbackArg);
 
 namespace {
@@ -143,8 +146,8 @@ PPC_NATIVE_OVERRIDE(801C0990, WPADProbe_HLE, int32_t, (uint32_t chan, uint32_t t
 
 extern "C" void WPADControlMotor_HLE(uint32_t chan, uint32_t command)
 {
-    (void)chan;
-    (void)command;
+    const bool on = command != 0 && PAD_HLE_RumbleEnabled();
+    PADControlMotor(chan, on ? PAD_MOTOR_RUMBLE : PAD_MOTOR_STOP_HARD);
 }
 PPC_NATIVE_OVERRIDE(801C0EC4, WPADControlMotor_HLE, void, (uint32_t chan, uint32_t command), (chan, command));
 

runtime/src/hle/net/network_internal.h

@@ -242,7 +242,6 @@ void WritePollResults(uint32_t outAddress,
                       const std::vector<NetworkPollContract::CopiedDescriptor>& descriptors);
 
 // network_socket.cpp
-int32_t DeleteWiiSocket(uint32_t fd);
 void CleanupAllWiiSockets();
 sockaddr_in ReadWiiSockAddr(uint32_t addr);
 int32_t HandleIpTopIoctl(uint32_t cmd, uint32_t inBuf, uint32_t inLen, uint32_t outBuf,

runtime/src/hle/net/network_ssl.cpp

@@ -54,6 +54,9 @@ constexpr int kMaxSslSessions = 4;
 struct SslSession {
     bool active = false;
     bool handshaked = false;
+    // A failed POSIX TLS write cannot be resumed with a new guest buffer.
+    // Keep the slot and socket ownership intact until explicit teardown.
+    bool failed = false;
     bool plaintextWfc = false;
     uint32_t socketFd = UINT32_MAX;
     NativeSocket native = kInvalidSocket;
@@ -761,6 +764,14 @@ static int32_t SslHandshakeImpl(SslSession& ssl) {
     return SSL_OK;
 }
 
+static int32_t FailSslWrite(SslSession& ssl) {
+    ssl.failed = true;
+    // Stop transport I/O without deleting the guest descriptor or freeing a
+    // session still referenced by the IOCTLV_NET_SSL_WRITE caller.
+    ::shutdown(ssl.native, SHUT_RDWR);
+    return SSL_ERR_FAILED;
+}
+
 static int32_t SslWrite(SslSession& ssl, const uint8_t* data, uint32_t size) {
     if (!data || size == 0) {
         return SSL_ERR_ZERO;
@@ -795,12 +806,11 @@ static int32_t SslWrite(SslSession& ssl, const uint8_t* data, uint32_t size) {
         }
         if (ret == MBEDTLS_ERR_SSL_WANT_READ || ret == MBEDTLS_ERR_SSL_WANT_WRITE) {
             if (std::chrono::steady_clock::now() >= writeDeadline) {
-                DeleteWiiSocket(ssl.socketFd);
-                return SSL_ERR_FAILED;
+                return FailSslWrite(ssl);
             }
             continue;
         }
-        return SSL_ERR_FAILED;
+        return FailSslWrite(ssl);
     }
     return static_cast<int32_t>(totalWritten);
 }
@@ -842,6 +852,9 @@ static int32_t SslRead(SslSession& ssl, uint8_t* out, uint32_t size) {
 // The handshake runs on every SSL read/write, so a failure repeats for as long
 // as the session lives; report only the first one.
 static int32_t SslHandshake(SslSession& ssl) {
+    if (ssl.failed) {
+        return SSL_ERR_FAILED;
+    }
     const int32_t result = SslHandshakeImpl(ssl);
     if (result != SSL_OK && !ssl.loggedHandshakeFail) {
         ssl.loggedHandshakeFail = true;
@@ -947,6 +960,11 @@ int32_t HandleSslIoctlv(uint32_t cmd, const std::vector<IoVector>& in, const std
             WriteSslReturn(in, SSL_ERR_ID);
             return 0;
         }
+        // Reject before NAS buffering can acknowledge data on a failed session.
+        if (g_sslSessions[sslId].failed) {
+            WriteSslReturn(in, SSL_ERR_FAILED);
+            return 0;
+        }
         if (out.size() < 2 || !out[1].address) {
             WriteSslReturn(in, SSL_ERR_FAILED);
             return 0;

runtime/src/hle/os/os_alarm.cpp

@@ -11,7 +11,7 @@
 #include "ppc_runtime.h"
 #include "runtime_log.h"
 #include "hle/net/network.h"
-#include "generated/RuntimeConfig.h"
+#include "guest_small_data.h"
 #include "os_internal.h"
 
 extern "C" void func_801AADE0(CpuContext* ctx);
@@ -41,7 +41,7 @@ struct AlarmProcessScope {
 // the other call sites have never logged.
 void EnsureSda1Base(CpuContext* cpu, bool logOnce = false)
 {
-    if (cpu->gpr[13] == RuntimeConfig::SDA1_BASE) {
+    if (cpu->gpr[13] == GuestSmallData::kSda1Base) {
         return;
     }
 
@@ -51,7 +51,7 @@ void EnsureSda1Base(CpuContext* cpu, bool logOnce = false)
             RT_LOG(RT_TAG_OS) << "OS_HLE_ProcessAlarms: fixing r13 to SDA1 base" << std::endl;
         });
     }
-    cpu->gpr[13] = RuntimeConfig::SDA1_BASE;
+    cpu->gpr[13] = GuestSmallData::kSda1Base;
 }
 
 void IncrementSchedulerDisableCount()

runtime/src/host_cpu_baseline.cpp

@@ -1,14 +1,12 @@
-// Host ISA guard. Every other product target builds with -march=x86-64-v3, so a pre-Haswell
-// Intel or pre-Excavator AMD machine would otherwise die on an illegal-instruction fault with no
+// Host ISA guard. Product targets build with an explicit x86-64-v2 or x86-64-v3 profile, so an
+// unsupported machine would otherwise die on an illegal-instruction fault with no
 // explanation. This TU alone skips that flag (own CMake object library, excluded from unity
 // build/PCH) and runs from a priority-101 C initializer, ahead of every C++ dynamic initializer
 // and thus the first AVX2 code that could execute. Keep it free of anything that could pull in
 // vectorized code: no iostreams, no std::string, no runtime-wide headers.
 //
-// x86-64-v3 is an x86-specific optional-feature baseline (AVX2/BMI2/FMA and friends are not
-// guaranteed present on every x86_64 chip); nothing here applies on AArch64, where ASIMD/NEON is
-// mandatory in the base architecture and PublicProducts.cmake never applies an -march=x86-64-v3
-// equivalent flag to begin with. That branch below is a no-op stub, not a port of this check.
+// x86-64-v2/v3 are x86-specific optional-feature baselines; nothing here applies on AArch64,
+// where ASIMD/NEON is mandatory in the base architecture.
 
 #if defined(__x86_64__)
 
@@ -58,27 +56,27 @@ struct CpuFeature {
     bool isOsXsave;
 };
 
-// Everything x86-64-v3 implies, which includes all of x86-64-v2. Spelled out so
-// the error message can name the exact instruction sets the machine lacks
-// rather than only "AVX2", which is merely the best known member of the set.
+// x86-64-v2 requirements. The v3-only extension below retains a useful
+// feature-by-feature diagnostic rather than reducing a failed v3 check to AVX2.
 constexpr CpuFeature kRequiredFeatures[] = {
     {"SSE3", 1, 0, 2, 0, false},
     {"SSSE3", 1, 0, 2, 9, false},
-    {"FMA", 1, 0, 2, 12, false},
     {"CMPXCHG16B", 1, 0, 2, 13, false},
     {"SSE4.1", 1, 0, 2, 19, false},
     {"SSE4.2", 1, 0, 2, 20, false},
-    {"MOVBE", 1, 0, 2, 22, false},
     {"POPCNT", 1, 0, 2, 23, false},
-    {"OSXSAVE", 1, 0, 2, 27, true},
-    {"AVX", 1, 0, 2, 28, false},
-    {"F16C", 1, 0, 2, 29, false},
-    {"BMI1", 7, 0, 1, 3, false},
-    {"AVX2", 7, 0, 1, 5, false},
-    {"BMI2", 7, 0, 1, 8, false},
     {"LAHF-SAHF", 0x80000001u, 0, 2, 0, false},
+};
+
+#if !defined(MKW_X86_CPU_PROFILE_V2)
+constexpr CpuFeature kV3RequiredFeatures[] = {
+    {"FMA", 1, 0, 2, 12, false}, {"MOVBE", 1, 0, 2, 22, false},
+    {"OSXSAVE", 1, 0, 2, 27, true}, {"AVX", 1, 0, 2, 28, false},
+    {"F16C", 1, 0, 2, 29, false}, {"BMI1", 7, 0, 1, 3, false},
+    {"AVX2", 7, 0, 1, 5, false}, {"BMI2", 7, 0, 1, 8, false},
     {"LZCNT", 0x80000001u, 0, 2, 5, false},
 };
+#endif
 
 // Fixed-capacity text accumulation: no allocation, no exceptions, nothing that
 // could route through code this file is trying to stay ahead of.
@@ -130,6 +128,27 @@ bool CollectMissingBaselineFeatures(TextBuffer& missing) {
         ok = false;
     }
 
+#if !defined(MKW_X86_CPU_PROFILE_V2)
+    for (const CpuFeature& feature : kV3RequiredFeatures) {
+        const bool leafAvailable = (feature.leaf & 0x80000000u) != 0
+                                       ? feature.leaf <= maxExtended
+                                       : feature.leaf <= maxBasic;
+        bool present = false;
+        if (leafAvailable) {
+            unsigned regs[4] = {0, 0, 0, 0};
+            HostCpuId(feature.leaf, feature.subleaf, regs);
+            present = (regs[feature.reg] & (1u << feature.bit)) != 0;
+        }
+        if (present) {
+            haveOsXsave = haveOsXsave || feature.isOsXsave;
+            continue;
+        }
+        if (!ok) missing.Append(", ");
+        missing.Append(feature.name);
+        ok = false;
+    }
+#endif
+
     // CPUID reporting AVX is not sufficient: the OS also has to have enabled
     // XMM and YMM state saving or every VEX-encoded instruction faults. This is
     // the same guard a compiler's own runtime feature dispatch applies.
@@ -175,13 +194,17 @@ void WriteStdErrEarly(const char* text) {
 
 [[noreturn]] void ReportUnsupportedCpu(const char* missing) {
     TextBuffer message;
-    message.Append(
-        "This build needs a processor that supports AVX2 and the rest of the "
-        "x86-64-v3 instruction set.\n\nMissing on this machine: ");
+#if defined(MKW_X86_CPU_PROFILE_V2)
+    message.Append("This build needs a processor that supports the x86-64-v2 instruction set.\n\nMissing on this machine: ");
+#else
+    message.Append("This build needs a processor that supports the x86-64-v3 instruction set.\n\nMissing on this machine: ");
+#endif
     message.Append(missing);
-    message.Append(
-        "\n\nx86-64-v3 covers Intel Core processors from Haswell (4th "
-        "generation, 2013) onward and AMD processors from Excavator (2015) onward.");
+#if defined(MKW_X86_CPU_PROFILE_V2)
+    message.Append("\n\nx86-64-v2 covers Intel Core processors from Nehalem (2008) onward and AMD processors from Jaguar (2013) onward.");
+#else
+    message.Append("\n\nx86-64-v3 covers Intel Core processors from Haswell (4th generation, 2013) onward and AMD processors from Excavator (2015) onward.");
+#endif
 
     // The tag matches RT_TAG_RUNTIME in runtime_log.h. It is spelled out here
     // because this translation unit must not include runtime-wide headers (see

runtime/src/main.cpp

@@ -152,6 +152,8 @@ private:
 #endif
 } // namespace
 
+extern const char* const kMkwSetupVersion;
+
 namespace {
 
 std::string g_lastEntryLabel;
@@ -378,10 +380,13 @@ void AttachParentConsoleForDiagnostics() {
 void AttachParentConsoleForDiagnostics() {}
 #endif
 
-// The setup writes build-fingerprint.json beside every product executable; its SetupVersion is
-// the only version identity the runtime has (products are compiled locally, so nothing is baked
-// into the binary). Surface it at the top of the transcript so every attached log self-identifies.
+// The macOS and Linux local builds bake the setup version in; the Windows setup instead writes it
+// to build-fingerprint.json beside the executable. Surfaced at the top of the transcript so every
+// attached log self-identifies.
 std::string ReadInstalledSetupVersion() {
+    if (const std::string_view builtIn = kMkwSetupVersion; !builtIn.empty()) {
+        return std::string(builtIn);
+    }
     const auto directory = RuntimeConfigFile::ExecutableDirectory();
     if (!directory) {
         return {};
@@ -625,7 +630,7 @@ std::string FormatHostStackTrace(unsigned framesToSkip) {
         symbol->MaxNameLen = MAX_SYM_NAME;
         if (symbolsReady && SymFromAddr(process, addr, &symbolDisp, symbol)) {
             symbolName = symbol->Name;
-        } else if (auto info = TranslatedFunctionRegistry::FindByHostAddress(static_cast<uintptr_t>(addr))) {
+        } else if (auto info = FindTranslatedFunctionByHostAddress(static_cast<uintptr_t>(addr))) {
             // DbgHelp could not name it; the translated-function registry can.
             translatedFuncName = info->name;
             ppcAddress = info->address;

runtime/src/platform/host_platform.cpp

@@ -1,6 +1,7 @@
 #include "platform/host_platform.h"
 
 #include <cstdlib>
+#include <vector>
 
 #if defined(_WIN32)
 #ifndef WIN32_LEAN_AND_MEAN
@@ -46,6 +47,18 @@ std::optional<std::filesystem::path> ExecutableDirectory() noexcept {
     std::error_code ec;
     const auto resolved = std::filesystem::weakly_canonical(path, ec);
     return (ec ? std::filesystem::path(path) : resolved).parent_path();
+#elif defined(__linux__)
+    std::vector<char> buffer(256);
+    for (;;) {
+        const auto length = ::readlink("/proc/self/exe", buffer.data(), buffer.size());
+        if (length < 0) {
+            return std::nullopt;
+        }
+        if (static_cast<std::size_t>(length) < buffer.size()) {
+            return std::filesystem::path(std::string(buffer.data(), length)).parent_path();
+        }
+        buffer.resize(buffer.size() * 2);
+    }
 #else
     return std::nullopt;
 #endif

runtime/src/ppc_helpers.cpp

@@ -12,6 +12,7 @@
 #include <cmath>
 #include <cstdint>
 #include <cstdlib>
+#include <ctime>
 #include <stdexcept>
 #include <limits>
 
@@ -44,11 +45,48 @@ uint32_t g_reservationAddr = 0;
 bool g_hasReservation = false;
 const auto g_timeBaseStart = std::chrono::steady_clock::now();
 
+// Days from 1970-01-01 to the given civil date (Howard Hinnant's algorithm).
+constexpr int64_t DaysFromCivil(int64_t y, unsigned m, unsigned d)
+{
+    y -= m <= 2;
+    const int64_t era = (y >= 0 ? y : y - 399) / 400;
+    const unsigned yoe = static_cast<unsigned>(y - era * 400);
+    const unsigned doy = (153 * (m > 2 ? m - 3 : m + 9) + 2) / 5 + d - 1;
+    const unsigned doe = yoe * 365 + yoe / 4 - yoe / 100 + doy;
+    return era * 146097 + static_cast<int64_t>(doe) - 719468;
+}
+
+// The Wii RTC holds local time; the OS counts ticks from 2000-01-01.
+uint64_t InitialTimeBase()
+{
+    const std::time_t now = std::time(nullptr);
+    std::tm local{};
+#if defined(_WIN32)
+    if (localtime_s(&local, &now) != 0) {
+        return 0;
+    }
+#else
+    if (localtime_r(&now, &local) == nullptr) {
+        return 0;
+    }
+#endif
+    const int64_t days = DaysFromCivil(local.tm_year + 1900, static_cast<unsigned>(local.tm_mon + 1),
+                                       static_cast<unsigned>(local.tm_mday)) -
+                         DaysFromCivil(2000, 1, 1);
+    const int64_t seconds = days * 86400 + local.tm_hour * 3600 + local.tm_min * 60 + local.tm_sec;
+    if (seconds <= 0) {
+        return 0;
+    }
+    return static_cast<uint64_t>(seconds) * TimeBaseContract::kTicksPerSecond;
+}
+
+const uint64_t g_timeBaseOrigin = InitialTimeBase();
+
 uint64_t GetTimeBase() {
     const auto elapsed = std::chrono::steady_clock::now() - g_timeBaseStart;
     const auto nanoseconds =
         std::chrono::duration_cast<std::chrono::nanoseconds>(elapsed).count();
-    return TimeBaseContract::NanosecondsToTicks(static_cast<uint64_t>(nanoseconds));
+    return g_timeBaseOrigin + TimeBaseContract::NanosecondsToTicks(static_cast<uint64_t>(nanoseconds));
 }
 
 constexpr uint32_t kFpscrFx = 1u << 31;

runtime/src/settings_overlay.cpp

@@ -309,10 +309,9 @@ void DrawWiiRemoteAccelerometer(uint32_t port) {
     } else {
         ImGui::TextDisabled("No accelerometer data yet.");
     }
-    // SDL's read of the remote's calibration block often times out over Bluetooth
-    // and it falls back to a nominal zero point, leaving a small per-axis bias;
-    // measured here with the remote at rest.
-    if (WiiRemoteInput::IsAccelCalibrating()) {
+    // Residual per-axis bias left after SDL's factory-calibrated decode, measured
+    // here with the remote at rest.
+    if (WiiRemoteInput::IsAccelCalibrating(port)) {
         ImGui::ProgressBar(WiiRemoteInput::AccelCalibrationProgress(), ImVec2(220.0f, 0.0f), "Hold still...");
     } else if (ImGui::Button("Calibrate (remote lying flat, buttons up)")) {
         WiiRemoteInput::StartAccelCalibration(port);
@@ -322,18 +321,20 @@ void DrawWiiRemoteAccelerometer(uint32_t port) {
                           "for about two seconds. Corrects the steering offset of a remote held sideways.");
     }
     ImGui::SameLine();
-    ImGui::BeginDisabled(!RuntimeConfigFile::HasWiiAccelOffset() || WiiRemoteInput::IsAccelCalibrating());
+    const auto stored = WiiRemoteInput::ReadAccelOffset(port);
+    const std::array<double, 3> offset = stored.value_or(std::array<double, 3>{});
+    const bool calibrated = offset[0] != 0.0 || offset[1] != 0.0 || offset[2] != 0.0;
+    ImGui::BeginDisabled(!calibrated || WiiRemoteInput::IsAccelCalibrating(port));
     if (ImGui::Button("Clear calibration")) {
-        WiiRemoteInput::ClearAccelCalibration();
+        WiiRemoteInput::ClearAccelCalibration(port);
     }
     ImGui::EndDisabled();
-    if (const char* message = WiiRemoteInput::AccelCalibrationMessage()) {
+    if (const char* message = WiiRemoteInput::AccelCalibrationMessage(port)) {
         ImGui::TextWrapped("%s", message);
-    } else if (RuntimeConfigFile::HasWiiAccelOffset()) {
-        const std::array<double, 3> offset = RuntimeConfigFile::WiiAccelOffset();
+    } else if (calibrated) {
         ImGui::TextDisabled("Stored offset: x %+.3f  y %+.3f  z %+.3f g", offset[0], offset[1], offset[2]);
     } else {
-        ImGui::TextDisabled("Not calibrated (using SDL's zero point; see console.log for \"fallback accelerometer calibration\").");
+        ImGui::TextDisabled("No trim stored.");
     }
 }
 
@@ -779,8 +780,15 @@ void DrawControllerSettings() {
     if (ImGui::BeginMenu("Assign connected controller")) {
         for (uint32_t index = 0; index < controllerCount; ++index) {
             const char* name = PADGetNameForControllerIndex(index);
+            std::string label = name != nullptr ? name : "Unknown controller";
+            for (int port = 0; port < PAD_MAX_CONTROLLERS; ++port) {
+                if (PADGetIndexForPort(port) == static_cast<s32>(index)) {
+                    label += " (Port " + std::to_string(port + 1) + ")";
+                    break;
+                }
+            }
             ImGui::PushID(static_cast<int>(index));
-            if (ImGui::MenuItem(name != nullptr ? name : "Unknown controller")) {
+            if (ImGui::MenuItem(label.c_str())) {
                 PADSetPortForIndex(index, selectedGamePort);
                 g_configuredControllerIndices.fill(std::numeric_limits<int32_t>::min());
                 ApplyConfiguredMappings();
@@ -790,6 +798,12 @@ void DrawControllerSettings() {
         ImGui::EndMenu();
     }
 
+    if (WiiRemoteInput::IsRemoteChannel(selectedGamePort)) {
+        ImGui::TextDisabled("The game reads this controller as a Wii Remote, so the button mapping does not apply.");
+        DrawRumbleSettings();
+        return;
+    }
+
     uint32_t mappingCount = 0;
     PADButtonMapping* mappings = PADGetButtonMappings(static_cast<uint32_t>(g_controllerPort), &mappingCount);
     if (mappings == nullptr || mappingCount != PAD_BUTTON_COUNT) {

runtime/src/system_bridge.cpp

@@ -57,8 +57,8 @@ std::optional<std::string> PrintableU32(uint32_t value) {
     return out;
 }
 
-std::optional<TranslatedFunctionInfo> FindNearestTranslatedFunction(uint32_t address) {
-    auto best = TranslatedFunctionRegistry::FindNearestByAddress(address);
+std::optional<TranslatedFunctionInfo> FindNearbyTranslatedFunction(uint32_t address) {
+    auto best = FindNearestTranslatedFunction(address);
 
     // Crash-only heuristic: large gaps are usually data/FourCC values, not code offsets.
     if (best && address - best->address < 0x10000u) {
@@ -161,7 +161,7 @@ void PrintGuestValue(std::ostream& os, std::string_view label, uint32_t value) {
         } else {
             os << "<" << exact->name << ">";
         }
-    } else if (auto nearest = FindNearestTranslatedFunction(value)) {
+    } else if (auto nearest = FindNearbyTranslatedFunction(value)) {
         if (mapName && mapStart >= nearest->address) {
             os << "<" << mapName;
             if (value != mapStart) {
@@ -255,7 +255,7 @@ void DumpGuestStackHints(std::ostream& os, const CpuContext* cpu) {
         if (!value) {
             continue;
         }
-        if (!TranslatedFunctionRegistry::FindByAddressPtr(*value) && !FindNearestTranslatedFunction(*value)) {
+        if (!TranslatedFunctionRegistry::FindByAddressPtr(*value) && !FindNearbyTranslatedFunction(*value)) {
             continue;
         }
         if (printed == 0) {

runtime/src/wii_remote_input.cpp

@@ -5,6 +5,7 @@
 
 #include <dolphin/pad.h>
 #include <SDL3/SDL_gamepad.h>
+#include <SDL3/SDL_hidapi.h>
 #include <SDL3/SDL_hints.h>
 #include <SDL3/SDL_log.h>
 #include <SDL3/SDL_sensor.h>
@@ -12,11 +13,16 @@
 
 #include <algorithm>
 #include <array>
+#include <cctype>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
+#include <filesystem>
 #include <fstream>
+#include <optional>
 #include <string>
+#include <system_error>
+#include <unordered_map>
 #include <unordered_set>
 
 namespace WiiRemoteInput {
@@ -63,6 +69,8 @@ constexpr uint32_t kClUp = 0x0001, kClLeft = 0x0002, kClZR = 0x0004, kClX = 0x00
                    kClB = 0x0040, kClZL = 0x0080, kClR = 0x0200, kClPlus = 0x0400, kClHome = 0x0800,
                    kClMinus = 0x1000, kClL = 0x2000, kClDown = 0x4000, kClRight = 0x8000;
 
+constexpr Uint16 kNintendoVendor = 0x057e;
+
 // How long a vanished remote keeps its channel alive with neutral input. SDL's
 // in-place reconnect after an extension change takes well under a second; a
 // remote that is really gone shows up as disconnected after this.
@@ -86,36 +94,46 @@ bool g_scanning = false;
 uint64_t g_driverOffSinceMs = 0;
 // When the current scan started (last Wii controller seen).
 uint64_t g_lostAtMs = 0;
+bool g_hidrawWarned = false;
 
 // Instance ids whose accelerometers have been switched on. SDL keeps sensors
 // off until asked and forgets that when the gamepad is closed, so a re-paired
 // remote gets a fresh id and is enabled again.
 std::array<SDL_JoystickID, PAD_MAX_CONTROLLERS> g_sensorsEnabledFor{};
 
-// Zero-point correction subtracted from the remote's accelerometer, in g and in
-// SDL's sensor frame; loaded from Config.toml on first use, replaced by a
-// calibration run. The Nunchuk accelerometer is left uncorrected.
-std::array<float, 3> g_accelOffset{};
-bool g_accelOffsetLoaded = false;
+// Zero-point correction subtracted from a remote's accelerometer, in g and in
+// SDL's sensor frame, with the config key it is stored under. Resolved once per
+// gamepad instance (a re-created joystick of the same remote resolves to the same
+// key) and replaced by a calibration run. The Nunchuk accelerometer is left
+// uncorrected.
+struct RemoteCalibration {
+    SDL_JoystickID instance = 0;
+    std::string key;
+    std::array<float, 3> offset{};
+};
+std::array<RemoteCalibration, PAD_MAX_CONTROLLERS> g_remoteCalibration{};
 
 // Frames sampled by a calibration run (about 1.5 s at 60 Hz) and how far a
 // sample may stray from the first one before the run is declared "moved".
 constexpr int kCalibrationSamples = 90;
 constexpr float kCalibrationMaxDeviationG = 0.15f;
-// |mean| outside this range means the remote was not at rest or the
-// accelerometer is far off its nominal scale; either way the offset is useless.
-constexpr float kCalibrationMinGravityG = 0.8f;
-constexpr float kCalibrationMaxGravityG = 1.2f;
+// A resting remote reads about 1 g and lies within 20 degrees of face up.
+constexpr double kCalibrationMinGravityG = 0.8;
+constexpr double kCalibrationMaxGravityG = 1.2;
+constexpr double kCalibrationMinTiltCosine = 0.9396926;
 
 struct AccelCalibration {
     bool active = false;
     uint32_t chan = 0;
+    SDL_JoystickID instance = 0;
     int count = 0;
     double sum[3] = {};
     float first[3] = {};
 };
 AccelCalibration g_calibration;
 char g_calibrationMessage[160] = {};
+uint32_t g_calibrationMessageChan = 0;
+SDL_JoystickID g_calibrationMessageRemote = 0;
 
 // Last accepted KPAD acc per port, for the remote and for the Nunchuk. Reused on
 // frames that bring no sample or a glitched one, so the game never sees a jump.
@@ -155,14 +173,73 @@ void EnsureSensors(SDL_Gamepad* gamepad, uint32_t port) {
     }
 }
 
-// Loads the stored zero-point correction once.
-const std::array<float, 3>& AccelOffset() {
-    if (!g_accelOffsetLoaded) {
-        const std::array<double, 3> stored = RuntimeConfigFile::WiiAccelOffset();
-        for (size_t i = 0; i < 3; ++i) g_accelOffset[i] = static_cast<float>(stored[i]);
-        g_accelOffsetLoaded = true;
+// USB interface number of the HID device at `path`, or -1 when it is not a USB
+// interface or cannot be found. A DolphinBar shows each remote slot as one.
+int UsbInterfaceForPath(const char* path, Uint16 vendor, Uint16 product) {
+    if (path == nullptr) {
+        return -1;
     }
-    return g_accelOffset;
+    int interfaceNumber = -1;
+    SDL_hid_device_info* list = SDL_hid_enumerate(vendor, product);
+    for (const SDL_hid_device_info* info = list; info != nullptr; info = info->next) {
+        if (info->path != nullptr && std::strcmp(info->path, path) == 0) {
+            interfaceNumber = info->interface_number;
+            break;
+        }
+    }
+    SDL_hid_free_enumeration(list);
+    return interfaceNumber;
+}
+
+// Config key identifying the physical remote behind a gamepad: its HID serial
+// (the Bluetooth address when paired directly), else its USB interface number
+// (a DolphinBar has no serial and one interface per remote slot) and the game
+// port, else the game port. Only characters valid in a bare TOML key.
+std::string RemoteKey(SDL_Gamepad* gamepad, uint32_t port) {
+    std::string serial;
+    if (const char* raw = SDL_GetGamepadSerial(gamepad)) {
+        for (const char* c = raw; *c != '\0'; ++c) {
+            if (std::isalnum(static_cast<unsigned char>(*c))) {
+                serial += static_cast<char>(std::tolower(static_cast<unsigned char>(*c)));
+            }
+        }
+    }
+    if (serial.find_first_not_of('0') != std::string::npos) {
+        return "serial_" + serial;
+    }
+    const Uint16 vendor = SDL_GetGamepadVendor(gamepad);
+    const Uint16 product = SDL_GetGamepadProduct(gamepad);
+    const int interfaceNumber = UsbInterfaceForPath(SDL_GetGamepadPath(gamepad), vendor, product);
+    char key[48];
+    if (interfaceNumber >= 0) {
+        std::snprintf(key, sizeof(key), "usb_%04x_%04x_if%d_port%u", vendor, product, interfaceNumber,
+                      port + 1);
+    } else {
+        std::snprintf(key, sizeof(key), "port_%u", port + 1);
+    }
+    return key;
+}
+
+// Calibration of the remote currently on `port`, resolved on first use per gamepad
+// instance. A remote without a stored entry has a zero offset.
+RemoteCalibration& CalibrationFor(SDL_Gamepad* gamepad, uint32_t port) {
+    RemoteCalibration& calibration = g_remoteCalibration[port];
+    const SDL_JoystickID id = SDL_GetGamepadID(gamepad);
+    if (calibration.instance != id) {
+        static bool s_legacyReported = false;
+        if (!s_legacyReported && RuntimeConfigFile::Get().hasLegacyWiiAccelOffset) {
+            s_legacyReported = true;
+            RT_LOG(RT_TAG_CONFIG) << "Old shared Wii Remote calibration ignored; recalibrate each remote in F10"
+                                  << std::endl;
+        }
+        calibration.instance = id;
+        calibration.key = RemoteKey(gamepad, port);
+        const std::optional<std::array<double, 3>> stored = RuntimeConfigFile::WiiAccelOffsetFor(calibration.key);
+        for (size_t i = 0; i < 3; ++i) calibration.offset[i] = stored ? static_cast<float>((*stored)[i]) : 0.0f;
+        RT_LOG(RT_TAG_CONFIG) << "Wii Remote on port " << (port + 1) << ": calibration key " << calibration.key
+                              << (stored ? "" : " (no stored offset)") << std::endl;
+    }
+    return calibration;
 }
 
 // Raw SDL sample in m/s^2, rejecting anything SDL has not delivered yet.
@@ -200,9 +277,10 @@ bool IsGlitchedSample(SDL_SensorType sensor, const float* g) {
 // +x left, +y towards the user, +z out of the button face (wiibrew, Dolphin);
 // SDL_hidapi_wii.c posts (-wiiX, wiiZ, wiiY): x right across the face, y out of
 // the face (+1 at rest, buttons up), z towards the user, i.e. away from the tip.
-// The remote's sample gets the zero-point correction; the Nunchuk's does not.
-// False when there is no sample yet or the sample is a glitch (see above).
-bool ReadAccelG(SDL_Gamepad* gamepad, SDL_SensorType sensor, float* g) {
+// The remote's sample gets the zero-point correction of the remote on `port`; the
+// Nunchuk's does not. False when there is no sample yet or the sample is a glitch
+// (see above).
+bool ReadAccelG(SDL_Gamepad* gamepad, uint32_t port, SDL_SensorType sensor, float* g) {
     float sdl[3] = {};
     if (!ReadSdlAccel(gamepad, sensor, sdl)) {
         return false;
@@ -212,7 +290,7 @@ bool ReadAccelG(SDL_Gamepad* gamepad, SDL_SensorType sensor, float* g) {
         return false;
     }
     if (sensor == SDL_SENSOR_ACCEL) {
-        const std::array<float, 3>& offset = AccelOffset();
+        const std::array<float, 3>& offset = CalibrationFor(gamepad, port).offset;
         for (int i = 0; i < 3; ++i) g[i] -= offset[i];
     }
     return true;
@@ -231,9 +309,9 @@ void AccelGToKpad(const float* g, float* kpad) {
 }
 
 // Sensor -> KPAD acc for ReadKpadSample.
-bool ReadAccelAsKpad(SDL_Gamepad* gamepad, SDL_SensorType sensor, float* kpad) {
+bool ReadAccelAsKpad(SDL_Gamepad* gamepad, uint32_t port, SDL_SensorType sensor, float* kpad) {
     float g[3] = {};
-    if (!ReadAccelG(gamepad, sensor, g)) {
+    if (!ReadAccelG(gamepad, port, sensor, g)) {
         return false;
     }
     AccelGToKpad(g, kpad);
@@ -275,10 +353,25 @@ void TraceSample(uint32_t chan, const KpadSample& sample, const float* rawG, boo
     trace.flush();
 }
 
-// Ends a calibration run with a message for the overlay.
-void FinishAccelCalibration(const char* message) {
-    g_calibration.active = false;
+// Gamepad of the Wii Remote on `chan`, or nullptr.
+SDL_Gamepad* RemoteGamepad(uint32_t chan) {
+    if (chan >= PAD_MAX_CONTROLLERS || !IsRemoteChannel(chan)) {
+        return nullptr;
+    }
+    return SDL_GetGamepadFromPlayerIndex(static_cast<int>(chan));
+}
+
+void SetCalibrationMessage(uint32_t chan, const char* message) {
+    SDL_Gamepad* gamepad = RemoteGamepad(chan);
+    g_calibrationMessageChan = chan;
+    g_calibrationMessageRemote = gamepad != nullptr ? SDL_GetGamepadID(gamepad) : 0;
     std::snprintf(g_calibrationMessage, sizeof(g_calibrationMessage), "%s", message);
+}
+
+// Ends a calibration run with a message for the overlay.
+void FinishAccelCalibration(uint32_t chan, const char* message) {
+    g_calibration.active = false;
+    SetCalibrationMessage(chan, message);
     RT_LOG(RT_TAG_CONFIG) << "Wii Remote accelerometer calibration: " << message << std::endl;
 }
 
@@ -288,9 +381,10 @@ void StepAccelCalibration() {
     if (!g_calibration.active) {
         return;
     }
-    SDL_Gamepad* gamepad = SDL_GetGamepadFromPlayerIndex(static_cast<int>(g_calibration.chan));
-    if (gamepad == nullptr || !IsRemoteChannel(g_calibration.chan)) {
-        FinishAccelCalibration("Cancelled: the Wii Remote went away.");
+    SDL_Gamepad* gamepad = RemoteGamepad(g_calibration.chan);
+    if (gamepad == nullptr ||
+        SDL_GetGamepadID(gamepad) != g_calibration.instance) {
+        FinishAccelCalibration(g_calibration.chan, "Cancelled: the Wii Remote went away.");
         return;
     }
     EnsureSensors(gamepad, g_calibration.chan);
@@ -308,7 +402,7 @@ void StepAccelCalibration() {
     } else {
         for (int i = 0; i < 3; ++i) {
             if (std::fabs(g[i] - g_calibration.first[i]) > kCalibrationMaxDeviationG) {
-                FinishAccelCalibration("Failed: the remote moved. Put it down, buttons up, and try again.");
+                FinishAccelCalibration(g_calibration.chan, "Failed: the remote moved. Put it down, buttons up, and try again.");
                 return;
             }
         }
@@ -320,19 +414,20 @@ void StepAccelCalibration() {
     std::array<double, 3> mean{};
     for (int i = 0; i < 3; ++i) mean[i] = g_calibration.sum[i] / g_calibration.count;
     const double length = std::sqrt(mean[0] * mean[0] + mean[1] * mean[1] + mean[2] * mean[2]);
-    if (length < kCalibrationMinGravityG || length > kCalibrationMaxGravityG || mean[1] < 0.5) {
-        FinishAccelCalibration("Failed: the remote was not resting flat with the buttons up.");
+    if (length < kCalibrationMinGravityG || length > kCalibrationMaxGravityG ||
+        mean[1] / length < kCalibrationMinTiltCosine) {
+        FinishAccelCalibration(g_calibration.chan, "Failed: the remote was not resting flat with the buttons up.");
         return;
     }
     const std::array<double, 3> offset = {mean[0], mean[1] - 1.0, mean[2]};
-    for (int i = 0; i < 3; ++i) g_accelOffset[i] = static_cast<float>(offset[i]);
-    g_accelOffsetLoaded = true;
-    const bool saved = RuntimeConfigFile::SetWiiAccelOffset(offset);
+    RemoteCalibration& calibration = CalibrationFor(gamepad, g_calibration.chan);
+    for (int i = 0; i < 3; ++i) calibration.offset[i] = static_cast<float>(offset[i]);
+    const bool saved = RuntimeConfigFile::SetWiiAccelOffsetFor(calibration.key, offset);
     char message[160];
     std::snprintf(message, sizeof(message), "%s offset x %+.3f  y %+.3f  z %+.3f g",
                   saved ? "Calibrated:" : "Calibrated (could not write Config.toml):", offset[0], offset[1],
                   offset[2]);
-    FinishAccelCalibration(message);
+    FinishAccelCalibration(g_calibration.chan, message);
 }
 
 // True when any controller aurora knows about is a Wii device.
@@ -346,6 +441,110 @@ bool AnyWiiControllerConnected() {
     return false;
 }
 
+// True for an evdev node served by the kernel hid-wiimote driver.
+bool IsKernelWiiNode(SDL_JoystickID id) {
+    const char* path = SDL_GetGamepadPathForID(id);
+    return SDL_GetGamepadVendorForID(id) == kNintendoVendor && path != nullptr &&
+           std::strncmp(path, "/dev/input/", 11) == 0;
+}
+
+bool IsClassicNodeName(const char* name) {
+    return NameContains(name, "Classic Controller");
+}
+
+// HID device behind a kernel evdev node; the remote and its extension nodes share it.
+std::string KernelParent(SDL_JoystickID id) {
+    static std::unordered_map<SDL_JoystickID, std::string> s_parents;
+    if (const auto it = s_parents.find(id); it != s_parents.end()) {
+        return it->second;
+    }
+    const char* path = SDL_GetGamepadPathForID(id);
+    if (path == nullptr) {
+        return {};
+    }
+    std::error_code ec;
+    const auto parent = std::filesystem::canonical(
+        std::filesystem::path("/sys/class/input") / std::filesystem::path(path).filename() / "device" / "device", ec);
+    if (ec) {
+        return {};
+    }
+    return s_parents.emplace(id, parent.string()).first->second;
+}
+
+// The kernel node on the same remote as `gamepad`: its Classic Controller node
+// when `wantClassic`, else its remote node.
+SDL_Gamepad* KernelSibling(SDL_Gamepad* gamepad, bool wantClassic) {
+    const SDL_JoystickID self = SDL_GetGamepadID(gamepad);
+    const std::string parent = KernelParent(self);
+    if (parent.empty()) {
+        return nullptr;
+    }
+    int count = 0;
+    SDL_JoystickID* ids = SDL_GetGamepads(&count);
+    if (ids == nullptr) {
+        return nullptr;
+    }
+    SDL_Gamepad* sibling = nullptr;
+    for (int i = 0; i < count && sibling == nullptr; ++i) {
+        if (ids[i] != self && IsKernelWiiNode(ids[i]) &&
+            IsClassicNodeName(SDL_GetGamepadNameForID(ids[i])) == wantClassic && KernelParent(ids[i]) == parent) {
+            sibling = SDL_GetGamepadFromID(ids[i]);
+        }
+    }
+    SDL_free(ids);
+    return sibling;
+}
+
+// The kernel exposes a remote and its Classic Controller as separate gamepads.
+// The remote's port reads both, as on a Wii; the Classic node's own port stays empty.
+Kind KindForGamepad(SDL_Gamepad* gamepad) {
+    if (gamepad == nullptr) {
+        return Kind::NotWii;
+    }
+    const char* name = SDL_GetGamepadName(gamepad);
+    const Kind kind = KindForName(name);
+    if ((kind != Kind::Remote && kind != Kind::RemoteWithClassic) || !IsKernelWiiNode(SDL_GetGamepadID(gamepad))) {
+        return kind;
+    }
+    if (IsClassicNodeName(name)) {
+        return KernelSibling(gamepad, false) != nullptr ? Kind::NotWii : kind;
+    }
+    return KernelSibling(gamepad, true) != nullptr ? Kind::RemoteWithClassic : kind;
+}
+
+// A kernel Classic Controller node whose remote node carries it.
+bool IsCarriedKernelNode(SDL_Gamepad* gamepad) {
+    return gamepad != nullptr && IsKernelWiiNode(SDL_GetGamepadID(gamepad)) &&
+           IsClassicNodeName(SDL_GetGamepadName(gamepad)) && KernelSibling(gamepad, false) != nullptr;
+}
+
+// Logs once when the kernel driver serves a Wii controller SDL's driver could have had.
+void WarnIfKernelDriverServing() {
+    if (g_hidrawWarned || !g_wiiDriverEnabled) {
+        return;
+    }
+    const uint64_t now = SDL_GetTicks();
+    static uint64_t s_nextCheckMs = 0;
+    if (now < s_nextCheckMs) {
+        return;
+    }
+    s_nextCheckMs = now + 2000;
+    int count = 0;
+    SDL_JoystickID* ids = SDL_GetGamepads(&count);
+    if (ids == nullptr) {
+        return;
+    }
+    for (int i = 0; i < count && !g_hidrawWarned; ++i) {
+        const Uint16 product = SDL_GetGamepadProductForID(ids[i]);
+        if ((product == 0x0306 || product == 0x0330) && IsKernelWiiNode(ids[i])) {
+            g_hidrawWarned = true;
+            RT_LOG(RT_TAG_CONFIG) << "A Wii controller is served by the kernel hid-wiimote driver. For full Wii "
+                                  << "Remote support see the Linux udev rule in the README." << std::endl;
+        }
+    }
+    SDL_free(ids);
+}
+
 // Route SDL's input diagnostics (HIDAPI open failures, the Wii driver's
 // extension/status messages) into console.log, minus the periodic chatter.
 // A sub-warning message is written once: SDL repeats the same line on every
@@ -464,6 +663,7 @@ void Poll() {
     for (uint32_t port = 0; port < PAD_MAX_CONTROLLERS; ++port) {
         (void)EffectiveKind(port);
     }
+    WarnIfKernelDriverServing();
     if (!g_wiiDriverEnabled) {
         return;
     }
@@ -518,11 +718,13 @@ uint32_t ScanCount() {
     return g_scanCount;
 }
 
-// Maps the gamepad name SDL's Wii driver reports to a Kind.
+// Maps the gamepad name SDL reports to a Kind: "Nintendo Wii Remote",
+// "... with Nunchuk", "... with Classic Controller" and "Nintendo Wii U Pro
+// Controller" from SDL's Wii driver; the Linux kernel's "Nintendo Wii Remote
+// Classic Controller" is read like the Classic Controller.
 Kind KindForName(const char* name) {
-    // Names come from SDL's hidapi Wii driver: "Nintendo Wii Remote",
-    // "Nintendo Wii Remote with Nunchuk", "Nintendo Wii Remote with Classic
-    // Controller" and "Nintendo Wii U Pro Controller".
+    // The kernel's Pro Controller is a plain gamepad with a built-in SDL mapping.
+    if (NameContains(name, "Wii Remote Pro Controller")) return Kind::NotWii;
     if (NameContains(name, "Wii U Pro Controller")) return Kind::WiiUPro;
     if (!NameContains(name, "Wii Remote")) return Kind::NotWii;
     if (NameContains(name, "Nunchuk")) return Kind::RemoteWithNunchuk;
@@ -534,8 +736,7 @@ Kind KindForName(const char* name) {
 Kind KindForPort(uint32_t port) {
     if (port >= PAD_MAX_CONTROLLERS) return Kind::NotWii;
     SDL_Gamepad* gamepad = SDL_GetGamepadFromPlayerIndex(static_cast<int>(port));
-    if (gamepad == nullptr) return Kind::NotWii;
-    return KindForName(SDL_GetGamepadName(gamepad));
+    return KindForGamepad(gamepad);
 }
 
 // Human-readable name of a Kind for the settings overlay.
@@ -561,7 +762,7 @@ Kind EffectiveKind(uint32_t chan) {
     if (chan >= PAD_MAX_CONTROLLERS) return Kind::NotWii;
     PortMemory& memory = g_ports[chan];
     SDL_Gamepad* gamepad = SDL_GetGamepadFromPlayerIndex(static_cast<int>(chan));
-    const Kind live = gamepad != nullptr ? KindForName(SDL_GetGamepadName(gamepad)) : Kind::NotWii;
+    const Kind live = KindForGamepad(gamepad);
     const uint64_t now = SDL_GetTicks();
     if (live != Kind::NotWii) {
         memory.lastKind = live;
@@ -587,7 +788,8 @@ bool IsRemoteChannel(uint32_t chan) {
 // Marks KPAD-served ports as "no controller" in the GameCube pad statuses.
 void HideRemotesFromPad(PADStatus* statuses, uint32_t count) {
     for (uint32_t port = 0; port < count && port < PAD_MAX_CONTROLLERS; ++port) {
-        if (IsRemoteChannel(port)) {
+        if (IsRemoteChannel(port) ||
+            IsCarriedKernelNode(SDL_GetGamepadFromPlayerIndex(static_cast<int>(port)))) {
             statuses[port] = {};
             statuses[port].err = PAD_ERR_NO_CONTROLLER;
         }
@@ -602,24 +804,47 @@ void FillGraceSample(uint32_t chan, Kind kind, KpadSample& sample) {
     sample.hasClassic = kind == Kind::RemoteWithClassic;
 }
 
-// SDL's stick axis (-32767..32767, y down) as WPADCLStatus carries it: WPAD
-// normalises every Classic Controller stick, whatever the report's resolution,
-// to a signed 10-bit value, -512..511 with 0 at the centre and +y up (RVL SDK
-// WPAD.h; wut's WPADStatusClassic documents the same range).
+// The patched SDL Wii driver posts Classic Controller sticks as WPAD's own
+// value times 64 (aurora-main/cmake/AuroraSDL3Patches.cmake), so WPADCLStatus
+// is the axis divided back: signed, 0 at the centre, -512..511, +y up.
 int16_t ClassicStickRaw(Sint16 axis, bool invert) {
+    const long value = invert ? -static_cast<long>(axis) : axis;
+    return static_cast<int16_t>(std::clamp(std::lround(value / 64.0), -512L, 511L));
+}
+
+// The kernel driver's evdev axes span SDL's full range; scaled to WPAD's.
+int16_t KernelClassicStickRaw(Sint16 axis, bool invert) {
     float value = static_cast<float>(axis) / 32767.0f;
     if (invert) value = -value;
     value = std::clamp(value, -1.0f, 1.0f);
     return static_cast<int16_t>(std::clamp(std::lround(value * 512.0f), -512L, 511L));
 }
 
+// KPAD's default Classic Controller stick limits; the game uses the cross clamp.
+constexpr float kKpadStickMin = 60.0f;
+constexpr float kKpadStickMax = 308.0f;
+
+// KPAD cross clamp: per-axis dead zone and scale, then the unit circle.
+void ClassicStickFloat(const int16_t raw[2], float out[2]) {
+    for (int i = 0; i < 2; ++i) {
+        const float magnitude = std::fabs(static_cast<float>(raw[i]));
+        const float scaled = std::clamp((magnitude - kKpadStickMin) / (kKpadStickMax - kKpadStickMin), 0.0f, 1.0f);
+        out[i] = raw[i] < 0 ? -scaled : scaled;
+    }
+    const float length = std::sqrt(out[0] * out[0] + out[1] * out[1]);
+    if (length > 1.0f) {
+        out[0] /= length;
+        out[1] /= length;
+    }
+}
+
 // Samples buttons, accelerometers and the extension of the remote on a port.
 bool ReadKpadSample(uint32_t chan, KpadSample& sample) {
     if (chan >= PAD_MAX_CONTROLLERS) {
         return false;
     }
     SDL_Gamepad* gamepad = SDL_GetGamepadFromPlayerIndex(static_cast<int>(chan));
-    const Kind kind = gamepad != nullptr ? KindForName(SDL_GetGamepadName(gamepad)) : Kind::NotWii;
+    const Kind kind = KindForGamepad(gamepad);
     if (!IsKpadKind(kind)) {
         const Kind remembered = EffectiveKind(chan);
         if (!IsKpadKind(remembered)) {
@@ -635,26 +860,48 @@ bool ReadKpadSample(uint32_t chan, KpadSample& sample) {
     EnsureSensors(gamepad, chan);
 
     sample = {};
-    const auto raw = [&](int index, uint32_t bit) {
-        if (SDL_GetJoystickButton(joystick, index)) sample.hold |= bit;
-    };
-    raw(kRawA, kWpadA);
-    raw(kRawB, kWpadB);
-    raw(kRawOne, kWpadOne);
-    raw(kRawTwo, kWpadTwo);
-    raw(kRawPlus, kWpadPlus);
-    raw(kRawMinus, kWpadMinus);
-    raw(kRawHome, kWpadHome);
-    raw(kRawDpadUp, kWpadUp);
-    raw(kRawDpadDown, kWpadDown);
-    raw(kRawDpadLeft, kWpadLeft);
-    raw(kRawDpadRight, kWpadRight);
+    // The kernel's evdev nodes have regular SDL mappings, so they are read as gamepads.
+    const bool kernel = IsKernelWiiNode(SDL_GetGamepadID(gamepad));
+    const bool classicNode = kernel && IsClassicNodeName(SDL_GetGamepadName(gamepad));
+    if (kernel) {
+        const auto mapped = [&](SDL_GamepadButton button, uint32_t bit) {
+            if (SDL_GetGamepadButton(gamepad, button)) sample.hold |= bit;
+        };
+        if (!classicNode) {
+            mapped(SDL_GAMEPAD_BUTTON_EAST, kWpadA);
+            mapped(SDL_GAMEPAD_BUTTON_SOUTH, kWpadB);
+            mapped(SDL_GAMEPAD_BUTTON_WEST, kWpadOne);
+            mapped(SDL_GAMEPAD_BUTTON_NORTH, kWpadTwo);
+            mapped(SDL_GAMEPAD_BUTTON_START, kWpadPlus);
+            mapped(SDL_GAMEPAD_BUTTON_BACK, kWpadMinus);
+            mapped(SDL_GAMEPAD_BUTTON_GUIDE, kWpadHome);
+            mapped(SDL_GAMEPAD_BUTTON_DPAD_UP, kWpadUp);
+            mapped(SDL_GAMEPAD_BUTTON_DPAD_DOWN, kWpadDown);
+            mapped(SDL_GAMEPAD_BUTTON_DPAD_LEFT, kWpadLeft);
+            mapped(SDL_GAMEPAD_BUTTON_DPAD_RIGHT, kWpadRight);
+        }
+    } else {
+        const auto raw = [&](int index, uint32_t bit) {
+            if (SDL_GetJoystickButton(joystick, index)) sample.hold |= bit;
+        };
+        raw(kRawA, kWpadA);
+        raw(kRawB, kWpadB);
+        raw(kRawOne, kWpadOne);
+        raw(kRawTwo, kWpadTwo);
+        raw(kRawPlus, kWpadPlus);
+        raw(kRawMinus, kWpadMinus);
+        raw(kRawHome, kWpadHome);
+        raw(kRawDpadUp, kWpadUp);
+        raw(kRawDpadDown, kWpadDown);
+        raw(kRawDpadLeft, kWpadLeft);
+        raw(kRawDpadRight, kWpadRight);
+    }
 
     float rawG[3] = {};
     const bool haveRaw = ReadSdlAccel(gamepad, SDL_SENSOR_ACCEL, rawG);
     for (float& v : rawG) v /= kStandardGravity;
     LastAcc& last = g_lastAcc[chan];
-    const bool accepted = ReadAccelAsKpad(gamepad, SDL_SENSOR_ACCEL, sample.acc);
+    const bool accepted = ReadAccelAsKpad(gamepad, chan, SDL_SENSOR_ACCEL, sample.acc);
     if (accepted) {
         last.valid = true;
         for (int i = 0; i < 3; ++i) last.acc[i] = sample.acc[i];
@@ -674,8 +921,17 @@ bool ReadKpadSample(uint32_t chan, KpadSample& sample) {
         // Home, the L/R clicks as shoulders, the D-pad as buttons 11-14 (never
         // through SDL's gamepad mapping, which expects a hat), ZL/ZR as the
         // trigger axes.
-        const auto cl = [&](int index, uint32_t bit) {
-            if (SDL_GetJoystickButton(joystick, index)) sample.clHold |= bit;
+        SDL_Gamepad* classic = gamepad;
+        if (kernel && !classicNode) {
+            classic = KernelSibling(gamepad, true);
+            if (classic == nullptr) {
+                classic = gamepad;
+            }
+        }
+        const auto cl = [&](SDL_GamepadButton button, uint32_t bit) {
+            if (kernel ? SDL_GetGamepadButton(classic, button) : SDL_GetJoystickButton(joystick, button)) {
+                sample.clHold |= bit;
+            }
         };
         cl(SDL_GAMEPAD_BUTTON_EAST, kClA);
         cl(SDL_GAMEPAD_BUTTON_SOUTH, kClB);
@@ -690,20 +946,29 @@ bool ReadKpadSample(uint32_t chan, KpadSample& sample) {
         cl(SDL_GAMEPAD_BUTTON_DPAD_DOWN, kClDown);
         cl(SDL_GAMEPAD_BUTTON_DPAD_LEFT, kClLeft);
         cl(SDL_GAMEPAD_BUTTON_DPAD_RIGHT, kClRight);
-        if (SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_LEFT_TRIGGER) > 0) sample.clHold |= kClZL;
-        if (SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_RIGHT_TRIGGER) > 0) sample.clHold |= kClZR;
-        const Sint16 lx = SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_LEFTX);
-        const Sint16 ly = SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_LEFTY);
-        const Sint16 rx = SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_RIGHTX);
-        const Sint16 ry = SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_RIGHTY);
-        sample.clLStick[0] = std::clamp(static_cast<float>(lx) / 32767.0f, -1.0f, 1.0f);
-        sample.clLStick[1] = std::clamp(-static_cast<float>(ly) / 32767.0f, -1.0f, 1.0f);
-        sample.clRStick[0] = std::clamp(static_cast<float>(rx) / 32767.0f, -1.0f, 1.0f);
-        sample.clRStick[1] = std::clamp(-static_cast<float>(ry) / 32767.0f, -1.0f, 1.0f);
-        sample.clLStickRaw[0] = ClassicStickRaw(lx, false);
-        sample.clLStickRaw[1] = ClassicStickRaw(ly, true);
-        sample.clRStickRaw[0] = ClassicStickRaw(rx, false);
-        sample.clRStickRaw[1] = ClassicStickRaw(ry, true);
+        if (SDL_GetGamepadAxis(classic, SDL_GAMEPAD_AXIS_LEFT_TRIGGER) > 0) sample.clHold |= kClZL;
+        if (SDL_GetGamepadAxis(classic, SDL_GAMEPAD_AXIS_RIGHT_TRIGGER) > 0) sample.clHold |= kClZR;
+        const Sint16 lx = SDL_GetGamepadAxis(classic, SDL_GAMEPAD_AXIS_LEFTX);
+        const Sint16 ly = SDL_GetGamepadAxis(classic, SDL_GAMEPAD_AXIS_LEFTY);
+        const Sint16 rx = SDL_GetGamepadAxis(classic, SDL_GAMEPAD_AXIS_RIGHTX);
+        const Sint16 ry = SDL_GetGamepadAxis(classic, SDL_GAMEPAD_AXIS_RIGHTY);
+        if (kernel) {
+            sample.clLStickRaw[0] = KernelClassicStickRaw(lx, false);
+            sample.clLStickRaw[1] = KernelClassicStickRaw(ly, true);
+            sample.clRStickRaw[0] = KernelClassicStickRaw(rx, false);
+            sample.clRStickRaw[1] = KernelClassicStickRaw(ry, true);
+            sample.clLStick[0] = std::clamp(static_cast<float>(lx) / 32767.0f, -1.0f, 1.0f);
+            sample.clLStick[1] = std::clamp(-static_cast<float>(ly) / 32767.0f, -1.0f, 1.0f);
+            sample.clRStick[0] = std::clamp(static_cast<float>(rx) / 32767.0f, -1.0f, 1.0f);
+            sample.clRStick[1] = std::clamp(-static_cast<float>(ry) / 32767.0f, -1.0f, 1.0f);
+        } else {
+            sample.clLStickRaw[0] = ClassicStickRaw(lx, false);
+            sample.clLStickRaw[1] = ClassicStickRaw(ly, true);
+            sample.clRStickRaw[0] = ClassicStickRaw(rx, false);
+            sample.clRStickRaw[1] = ClassicStickRaw(ry, true);
+            ClassicStickFloat(sample.clLStickRaw, sample.clLStick);
+            ClassicStickFloat(sample.clRStickRaw, sample.clRStick);
+        }
         // Only the full-press click of L/R reaches SDL; report it as a full pull.
         sample.clTriggerL = (sample.clHold & kClL) ? 255 : 0;
         sample.clTriggerR = (sample.clHold & kClR) ? 255 : 0;
@@ -718,7 +983,7 @@ bool ReadKpadSample(uint32_t chan, KpadSample& sample) {
         sample.stick[1] = -static_cast<float>(SDL_GetGamepadAxis(gamepad, SDL_GAMEPAD_AXIS_LEFTY)) / 32767.0f;
         for (float& v : sample.stick) v = std::clamp(v, -1.0f, 1.0f);
         LastAcc& lastNunchuk = g_lastNunchukAcc[chan];
-        if (ReadAccelAsKpad(gamepad, SDL_SENSOR_ACCEL_L, sample.nunchukAcc)) {
+        if (ReadAccelAsKpad(gamepad, chan, SDL_SENSOR_ACCEL_L, sample.nunchukAcc)) {
             lastNunchuk.valid = true;
             for (int i = 0; i < 3; ++i) lastNunchuk.acc[i] = sample.nunchukAcc[i];
         } else {
@@ -738,7 +1003,7 @@ bool ReadAccelDebug(uint32_t chan, float sdlG[3], float kpadAcc[3]) {
         return false;
     }
     EnsureSensors(gamepad, chan);
-    if (!ReadAccelG(gamepad, SDL_SENSOR_ACCEL, sdlG)) {
+    if (!ReadAccelG(gamepad, chan, SDL_SENSOR_ACCEL, sdlG)) {
         return false;
     }
     AccelGToKpad(sdlG, kpadAcc);
@@ -747,28 +1012,47 @@ bool ReadAccelDebug(uint32_t chan, float sdlG[3], float kpadAcc[3]) {
 
 // Begins collecting rest samples from the remote on `chan`.
 void StartAccelCalibration(uint32_t chan) {
-    if (!IsRemoteChannel(chan)) {
-        FinishAccelCalibration("No Wii Remote on this port.");
+    SDL_Gamepad* gamepad = RemoteGamepad(chan);
+    if (gamepad == nullptr) {
+        FinishAccelCalibration(chan, "No Wii Remote on this port.");
         return;
     }
     g_calibration = {};
     g_calibration.active = true;
     g_calibration.chan = chan;
+    g_calibration.instance = SDL_GetGamepadID(gamepad);
     g_calibrationMessage[0] = '\0';
 }
 
-// Drops the stored correction and goes back to SDL's raw reading.
-void ClearAccelCalibration() {
-    g_calibration.active = false;
-    g_accelOffset = {};
-    g_accelOffsetLoaded = true;
-    RuntimeConfigFile::SetWiiAccelOffset({0.0, 0.0, 0.0});
-    std::snprintf(g_calibrationMessage, sizeof(g_calibrationMessage), "Calibration cleared.");
+// Zeroes the stored correction of the remote on `chan`.
+void ClearAccelCalibration(uint32_t chan) {
+    if (g_calibration.chan == chan) {
+        g_calibration.active = false;
+    }
+    SDL_Gamepad* gamepad = RemoteGamepad(chan);
+    if (gamepad == nullptr) {
+        SetCalibrationMessage(chan, "No Wii Remote on this port.");
+        return;
+    }
+    RemoteCalibration& calibration = CalibrationFor(gamepad, chan);
+    calibration.offset = {};
+    RuntimeConfigFile::SetWiiAccelOffsetFor(calibration.key, {0.0, 0.0, 0.0});
+    SetCalibrationMessage(chan, "Calibration cleared.");
 }
 
-// True while a calibration run is collecting samples.
-bool IsAccelCalibrating() {
-    return g_calibration.active;
+// Correction in effect for the remote on `chan`, for the overlay.
+std::optional<std::array<double, 3>> ReadAccelOffset(uint32_t chan) {
+    SDL_Gamepad* gamepad = RemoteGamepad(chan);
+    if (gamepad == nullptr) {
+        return std::nullopt;
+    }
+    const RemoteCalibration& calibration = CalibrationFor(gamepad, chan);
+    return std::array<double, 3>{calibration.offset[0], calibration.offset[1], calibration.offset[2]};
+}
+
+// True while a calibration run is collecting samples from `chan`.
+bool IsAccelCalibrating(uint32_t chan) {
+    return g_calibration.active && g_calibration.chan == chan;
 }
 
 // Share of the calibration samples collected so far, 0..1; 0 when idle.
@@ -776,9 +1060,14 @@ float AccelCalibrationProgress() {
     return g_calibration.active ? static_cast<float>(g_calibration.count) / kCalibrationSamples : 0.0f;
 }
 
-// Outcome of the last calibration run for the overlay, or nullptr before any.
-const char* AccelCalibrationMessage() {
-    return g_calibrationMessage[0] != '\0' ? g_calibrationMessage : nullptr;
+// Outcome of the last calibration run on `chan` for the overlay, or nullptr.
+const char* AccelCalibrationMessage(uint32_t chan) {
+    if (g_calibrationMessage[0] == '\0' || g_calibrationMessageChan != chan) {
+        return nullptr;
+    }
+    SDL_Gamepad* gamepad = RemoteGamepad(chan);
+    const SDL_JoystickID remote = gamepad != nullptr ? SDL_GetGamepadID(gamepad) : 0;
+    return remote == g_calibrationMessageRemote ? g_calibrationMessage : nullptr;
 }
 
 } // namespace WiiRemoteInput

runtime/tests/ppc_pair_fma_tests.cpp

@@ -0,0 +1,58 @@
+#include "isa/ppc_isa_float.h"
+
+#include <cmath>
+#include <cstdint>
+#include <cstdio>
+
+namespace {
+
+bool SameBits(double lhs, double rhs)
+{
+    return PpcBitCastToU64Inline(lhs) == PpcBitCastToU64Inline(rhs);
+}
+
+bool CheckPair(double actual, float a0, float a1, float c0, float c1, float b0, float b1,
+               bool subtract, const char* operation)
+{
+    const float expected0 = subtract ? std::fma(a0, c0, -b0) : std::fma(a0, c0, b0);
+    const float expected1 = subtract ? std::fma(a1, c1, -b1) : std::fma(a1, c1, b1);
+    const double expected = PpcPackPairedInline(expected0, expected1);
+    if (SameBits(actual, expected)) {
+        return true;
+    }
+    std::fprintf(stderr, "%s produced the wrong paired lanes\n", operation);
+    return false;
+}
+
+bool CheckNegatedPair(double actual, float a0, float a1, float c0, float c1, float b0, float b1,
+                      bool subtract, const char* operation)
+{
+    const float expected0 = -std::fma(a0, c0, subtract ? -b0 : b0);
+    const float expected1 = -std::fma(a1, c1, subtract ? -b1 : b1);
+    const double expected = PpcPackPairedInline(expected0, expected1);
+    if (SameBits(actual, expected)) {
+        return true;
+    }
+    std::fprintf(stderr, "%s produced the wrong paired lanes\n", operation);
+    return false;
+}
+
+}  // namespace
+
+int main()
+{
+    const double a = PpcPackPairedInline(1.000000119f, -123.25f);
+    const double c = PpcPackPairedInline(33554431.0f, 0.0625f);
+    const double b = PpcPackPairedInline(-33554430.0f, 7.75f);
+
+    bool ok = true;
+    ok &= CheckPair(PPC_PsMaddNoNiInline(a, c, b), 1.000000119f, -123.25f,
+                    33554431.0f, 0.0625f, -33554430.0f, 7.75f, false, "ps_madd");
+    ok &= CheckPair(PPC_PsMsubNoNiInline(a, c, b), 1.000000119f, -123.25f,
+                    33554431.0f, 0.0625f, -33554430.0f, 7.75f, true, "ps_msub");
+    ok &= CheckNegatedPair(PPC_PsNmaddInline(a, c, b), 1.000000119f, -123.25f,
+                           33554431.0f, 0.0625f, -33554430.0f, 7.75f, false, "ps_nmadd");
+    ok &= CheckNegatedPair(PPC_PsNmsubNoNiInline(a, c, b), 1.000000119f, -123.25f,
+                           33554431.0f, 0.0625f, -33554430.0f, 7.75f, true, "ps_nmsub");
+    return ok ? 0 : 1;
+}

runtime/tests/runtime_config_tests.cpp

@@ -1,7 +1,9 @@
 #include "runtime_config.h"
 
 #include <iostream>
+#include <optional>
 #include <sstream>
+#include <string>
 
 namespace {
 bool Require(bool value, const char* message) {
@@ -11,6 +13,24 @@ bool Require(bool value, const char* message) {
     }
     return true;
 }
+std::optional<toml::ordered_value> Update(const std::string& original, const char* section, const char* key,
+                                          const char* value, std::string* text = nullptr) {
+    const auto updated = RuntimeConfigFile::UpdateSettingText(original, section, key, value);
+    if (!updated) return std::nullopt;
+    if (text) *text = *updated;
+    return toml::parse_str<toml::ordered_type_config>(*updated);
+}
+bool SetsFps(const std::string& original, const char* message) {
+    const auto document = Update(original, "video", "show_fps", "true");
+    return Require(document && document->at("video").at("show_fps").as_boolean() &&
+                       document->at("video").as_table().size() == 1,
+                   message);
+}
+bool SetsOffset(const std::string& original, const char* message) {
+    const auto document = Update(original, "controller.wii_accel_offsets", "x", "0.5");
+    return Require(document && document->at("controller").at("wii_accel_offsets").at("x").as_floating() == 0.5,
+                   message);
+}
 } // namespace
 
 int main() {
@@ -31,5 +51,49 @@ int main() {
         const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "invalid.toml");
         if (!Require(!config.metalFxSpatialUpscaling, "invalid MetalFX setting was accepted")) return 1;
     }
+    if (!SetsFps("[video]\nshow_fps = false\n", "bare key was not replaced")) return 1;
+    if (!SetsFps("[video]\n\"show_fps\" = false\n", "basic-quoted key was not replaced")) return 1;
+    if (!SetsFps("[video]\n'show_fps'=false\n", "literal-quoted key was not replaced")) return 1;
+    if (!SetsFps("[video]\n\"show\\u005Ffps\" = false\n", "escaped key was not replaced")) return 1;
+    if (!SetsFps("[\"video\"]\n  show_fps = false\n", "quoted table header was not matched")) return 1;
+    if (!SetsFps("[video]\nshow_fps = [\n  1,\n]\n", "multi-line value was not replaced")) return 1;
+    if (!SetsFps("[video]\n", "missing key was not created")) return 1;
+    if (!SetsFps("", "missing table was not created")) return 1;
+    {
+        std::string text;
+        const auto document = Update("# top\nfoo = 1\n\n[video]\nzzz = 2 # keep\nshow_fps = false # note\naaa = 3\n\n"
+                                     "[audio]\nshow_fps = 1\n",
+                                     "video", "show_fps", "true", &text);
+        if (!Require(document && document->at("foo").as_integer() == 1 &&
+                         document->at("audio").at("show_fps").as_integer() == 1 &&
+                         document->at("video").at("zzz").as_integer() == 2 &&
+                         document->at("video").at("aaa").as_integer() == 3 &&
+                         document->at("video").at("show_fps").as_boolean(),
+                     "unrelated settings were not preserved")) return 1;
+        const size_t zzz = text.find("zzz"), fps = text.find("show_fps"), aaa = text.find("aaa");
+        if (!Require(text.find("# top") != std::string::npos && text.find("# keep") != std::string::npos &&
+                         text.find("# note") != std::string::npos && zzz < fps && fps < aaa,
+                     "comments or key order were not preserved")) return 1;
+    }
+    if (!Require(!RuntimeConfigFile::UpdateSettingText("[video]\nshow_fps = 1\nshow_fps = 2\n", "video", "show_fps", "true"),
+                 "a document with a duplicate key was rewritten")) return 1;
+    if (!Require(!RuntimeConfigFile::UpdateSettingText("[video\nshow_fps = 1\n", "video", "show_fps", "true"),
+                 "an unparseable document was rewritten")) return 1;
+    if (!SetsOffset("[controller]\nrumble = true\n\n[controller.wii_accel_offsets]\nx = 0.1\ny = 2\n",
+                    "key in an existing nested table was not replaced")) return 1;
+    if (!SetsOffset("[controller.wii_accel_offsets]\ny = 2\n", "key was not added to a nested table")) return 1;
+    if (!SetsOffset("[controller]\nrumble = true\n", "missing nested table was not created")) return 1;
+    if (!SetsOffset("[controller.\"wii_accel_offsets\"]\nx = 0.1\n", "quoted nested header was not matched")) return 1;
+    if (!SetsOffset("[controller]\nwii_accel_offsets.y = 2\n", "dotted-key table was not extended")) return 1;
+    if (!SetsOffset("controller.wii_accel_offsets.y = 2\n", "root-level dotted table was not extended")) return 1;
+    if (!Require(!RuntimeConfigFile::UpdateSettingText("controller = \"custom\"\n", "controller.wii_accel_offsets",
+                                                       "x", "0.5"),
+                 "a scalar was replaced by a table")) return 1;
+    {
+        const auto document = Update("[video.advanced]\nfoo = 1\n", "video", "show_fps", "true");
+        if (!Require(document && document->at("video").at("show_fps").as_boolean() &&
+                         document->at("video").at("advanced").at("foo").as_integer() == 1,
+                     "key was not added to an implicit table")) return 1;
+    }
     return 0;
 }

translator/README.md

@@ -66,6 +66,7 @@ Relative paths resolve from `workspace_root`, which itself resolves from the man
 -  `generate-data-init --project path`
 -  `emit-base-manifest --project path`
 -  `emit-build-shards --project path`
+-  `emit-native-index --project path --out path [--native-source-dir path]` - writes the serialized runtime native index
 -  `translate-mod --project path [--profile name] ...` - static Kamek/Pulsar module translation
 
 Any command prints its own option list with `--help`.