keep steam desktop layout off while focused

* Update README.md

Added NTSC-U and NTSC-J to compatible ISOs. Changed "other regions" to NTSC-K

* Update README.md

Fixed typo and installation.

* Update README.md

Added NTSC-U and NTSC-J to line 130.

* Fixed "and" to "or" and line 130.

.github/workflows/build.yml

@@ -102,6 +102,7 @@ jobs:
             mkw_nand_settings_tests \
             mkw_sc_serial_tests \
             mkw_input_expr_tests \
+            mkw_motion_trick_tests \
             mkw_bluetooth_controller_tests \
             mkw_macos_native_compile \
             mkw_macos_context_abi_tests \

Launcher/Build-Installer.ps1

@@ -278,6 +278,24 @@ $packagedRuntimeFingerprint = (Get-Content -LiteralPath (Join-Path $workspace 'D
 if ($stagedRuntimeFingerprint -ne $packagedRuntimeFingerprint) {
     throw "The staged runtime ($stagedRuntimeFingerprint) does not match the precompiled runtime archive ($packagedRuntimeFingerprint)."
 }
+# Users' builds run LocalBuild.ps1 under Windows PowerShell 5.1, whatever shell built this
+# installer, so the staged package is accepted (or not) exactly the way theirs will be.
+$prebuiltGate = Join-Path $workRoot 'native-prebuilt-gate.ps1'
+@'
+param([string]$Workspace, [string]$Toolkit)
+$ErrorActionPreference = 'Stop'
+Set-StrictMode -Version 3.0
+. (Join-Path $Workspace 'NativeBuildFlags.ps1')
+$package = Resolve-MkwNativePrebuiltDirectory (Join-Path $Workspace 'Dependencies') `
+    (Join-Path $Toolkit 'llvm-mingw\bin') (Join-Path $Workspace 'aurora-main') (Join-Path $Workspace 'runtime\third_party')
+if (-not $package) { exit 3 }
+'@ | Set-Content -LiteralPath $prebuiltGate -Encoding UTF8
+& (Join-Path $env:SystemRoot 'System32\WindowsPowerShell\v1.0\powershell.exe') -NoLogo -NoProfile -NonInteractive `
+    -ExecutionPolicy Bypass -File $prebuiltGate -Workspace $workspace -Toolkit $toolkit
+if ($LASTEXITCODE -ne 0) {
+    throw "Windows PowerShell rejects the staged precompiled aurora package (exit $LASTEXITCODE); every installation would build aurora from source."
+}
+Remove-Item -LiteralPath $prebuiltGate -Force
 
 [IO.Directory]::CreateDirectory((Join-Path $payloadRoot 'host')) | Out-Null
 Copy-Item $setupHost (Join-Path $payloadRoot 'host\WiiCompiled-Setup.exe')

Launcher/Directory.Build.props

@@ -1,7 +1,7 @@
 <Project>
   <PropertyGroup>
     <!-- The one product version. Release tags must match it. -->
-    <Version>0.9.1</Version>
+    <Version>0.9.3</Version>
     <IncludeSourceRevisionInInformationalVersion>false</IncludeSourceRevisionInInformationalVersion>
   </PropertyGroup>
 </Project>

Launcher/LocalBuild.ps1

@@ -34,65 +34,6 @@ Set-StrictMode -Version 3.0
 # See NativeBuildFlags.ps1; Build-Installer.ps1 stages it beside this script.
 . (Join-Path $PSScriptRoot 'NativeBuildFlags.ps1')
 
-function Resolve-NativePrebuiltDirectory([string]$DependenciesDirectory, [string]$ToolchainBin, [string]$AuroraDirectory, [string]$ThirdPartyDirectory) {
-    <#
-    The shipped precompiled aurora/third-party package, verified against its recorded provenance
-    (exact compiler, exact flags) before use. Anything unexpected falls back to building from source.
-    #>
-    $package = Join-Path $DependenciesDirectory 'native_prebuilt'
-    $provenancePath = Join-Path $package 'provenance.json'
-    if (-not (Test-Path -LiteralPath $provenancePath -PathType Leaf)) {
-        Write-Host 'MKWCBUILD: No precompiled aurora package was shipped; building it from source'
-        return ''
-    }
-    try {
-        $provenance = Get-Content -LiteralPath $provenancePath -Raw | ConvertFrom-Json
-        $compiler = Join-Path $ToolchainBin 'clang-22.exe'
-        $compilerHash = Get-MkwFileSha256 $compiler
-        if ($provenance.CompilerSha256 -ne $compilerHash) {
-            Write-Host 'MKWCBUILD: The precompiled aurora package was built by a different compiler; building it from source'
-            return ''
-        }
-        if ($provenance.FlagFingerprint -ne (Get-MkwNativeFlagFingerprint)) {
-            Write-Host 'MKWCBUILD: The precompiled aurora package was built with different flags; building it from source'
-            return ''
-        }
-        # The archives embed Dawn's headers and link its import library, so a
-        # dawn_prebuilt tree that is not the one they were harvested against is
-        # an ABI mismatch that only shows up as a crash inside the GPU backend.
-        $recordedDawn = $provenance.PSObject.Properties['DawnRuntimeSha256']
-        if ($null -eq $recordedDawn -or
-            $recordedDawn.Value -ne (Get-MkwDawnRuntimeSha256 $DependenciesDirectory)) {
-            Write-Host 'MKWCBUILD: The precompiled aurora package was built against a different Dawn; building it from source'
-            return ''
-        }
-        # The archives are aurora; the runtime compiled here calls its headers.
-        # If this workspace's aurora sources are not the ones that were harvested
-        # (a newer toolkit, or a locally modified tree), linking the package would
-        # mix two versions of aurora, so build it from source instead.
-        $recordedAurora = $provenance.PSObject.Properties['AuroraSourceFingerprint']
-        if ($null -eq $recordedAurora -or
-            $recordedAurora.Value -ne (Get-MkwAuroraSourceFingerprint $AuroraDirectory)) {
-            Write-Host 'MKWCBUILD: The precompiled aurora package was built from different aurora sources; building it from source'
-            return ''
-        }
-        # Same reasoning for the vendored Crypto++ the package also carries:
-        # first-party TUs compile against this workspace's third_party headers,
-        # so a package harvested from an older tree must not be linked here.
-        $recordedThirdParty = $provenance.PSObject.Properties['ThirdPartySourceFingerprint']
-        if ($null -eq $recordedThirdParty -or
-            $recordedThirdParty.Value -ne (Get-MkwThirdPartySourceFingerprint $ThirdPartyDirectory)) {
-            Write-Host 'MKWCBUILD: The precompiled aurora package was built from different vendored third-party sources; building it from source'
-            return ''
-        }
-    } catch {
-        Write-Host "MKWCBUILD: The precompiled aurora package could not be validated ($($_.Exception.Message)); building it from source"
-        return ''
-    }
-    Write-Host 'MKWCBUILD: Using the precompiled aurora and third-party libraries'
-    return $package
-}
-
 function Write-MkwBuildStep([string]$StepId, [string]$Message) {
     <#
     Announces a build step. The identifier is the contract (BuildStepIds in
@@ -160,6 +101,8 @@ if (-not [string]::IsNullOrWhiteSpace($RetroRewindPackageDirectory) -and
     $RetroRewindPackageDirectory = $Workspace + $RetroRewindPackageDirectory.Substring($realWorkspace.Length)
 }
 $translator = Join-Path $Toolkit 'Translator\Translator.Cli.exe'
+# A translator step takes 3-19 s on 16 threads; past this it has hung, and failing lets the clean retry run.
+$translatorTimeoutSeconds = 1800
 $toolchain = Get-MkwBuildSafePath $Toolkit 'toolchain' 'CMake\bin\cmake.exe'
 $cmake = Join-Path $toolchain 'CMake\bin\cmake.exe'
 $ninja = Join-Path $toolchain 'Ninja\ninja.exe'
@@ -208,8 +151,9 @@ if ($Parallel -gt 0) {
     $globalJobs = $Parallel
 } else {
     $translatorThreads = [math]::Max(1, [math]::Min([Environment]::ProcessorCount, 16))
+    # A translated TU peaks around 0.26 GB in clang; 0.5 GiB each after a 2 GiB reserve.
     $translatedJobs = [math]::Max(1, [math]::Min([Environment]::ProcessorCount,
-        [math]::Floor($memoryGiB / 2)))
+        [math]::Floor(($memoryGiB - 2) / 0.5)))
     $globalJobs = [math]::Max($translatedJobs, [Environment]::ProcessorCount)
 }
 
@@ -286,11 +230,18 @@ try {
                 # same translated functions produces the same base tree. Only the translator can
                 # decide that - it takes parsing the Kamek patch set against the recorded translated
                 # function ranges - and it fails closed, so anything but exit 0 retranslates.
-                & $translator @(
-                    'check-base-mod-awareness', '--project', $project, '--profile', 'retro-rewind',
-                    '--translation-output-metadata', $baseMetadata, '--code-pul', $retroCodePul
-                ) | Write-Host
-                if ($LASTEXITCODE -ne 0) {
+                $modAware = $true
+                try {
+                    Invoke-Checked $translator @(
+                        'check-base-mod-awareness', '--project', $project, '--profile', 'retro-rewind',
+                        '--translation-output-metadata', $baseMetadata, '--code-pul', $retroCodePul
+                    ) 'Checking whether the base translation covers the new Code.pul' `
+                        -TimeoutSeconds $translatorTimeoutSeconds
+                } catch {
+                    Write-Host "MKWCBUILD: $($_.Exception.Message)"
+                    $modAware = $false
+                }
+                if (-not $modAware) {
                     Write-MkwBuildStep 'retranslate-base' 'The base translation is stale; retranslating the base game for the new Code.pul'
                     $reuseBase = $false
                 }
@@ -313,12 +264,14 @@ try {
                 '--outdir', $functions, '--output-metadata', $baseMetadata,
                 '--production-source-bundle', (Join-Path $generated 'base_translation_sources.bin'),
                 '--no-function-files', '--prune-stale', '--threads', $translatorThreads
-            ) 'Translating the user-owned base game' -StepId 'translate-base'
+            ) 'Translating the user-owned base game' -StepId 'translate-base' `
+                -TimeoutSeconds $translatorTimeoutSeconds
 
             Invoke-Checked $translator @(
                 'emit-base-manifest', '--project', $project, '--out', $baseManifestDir,
                 '--functions-dir', $functions, '--translation-output-metadata', $baseMetadata, '--region', $pins.Region
-            ) 'Creating the local base translation manifest' -StepId 'emit-base-manifest'
+            ) 'Creating the local base translation manifest' -StepId 'emit-base-manifest' `
+                -TimeoutSeconds $translatorTimeoutSeconds
 
             if (-not [string]::IsNullOrWhiteSpace($TranslationFingerprint)) {
                 [ordered]@{ SchemaVersion = 1; TranslationFingerprint = $TranslationFingerprint; DolSha256 = $pins.DolSha256 } |
@@ -348,11 +301,12 @@ try {
                 throw 'Retro Rewind requires either -RetroWfcOfflineDirectory or -SkipRetroWfcPayload.'
             }
             Invoke-Checked $translator $translateModArguments 'Translating the selected Retro Rewind Code.pul' `
-                -StepId 'translate-mod'
+                -StepId 'translate-mod' -TimeoutSeconds $translatorTimeoutSeconds
         }
 
         Invoke-Checked $translator @('generate-data-init', '--project', $project) `
-            'Generating local game data initialization' -StepId 'generate-data-init'
+            'Generating local game data initialization' -StepId 'generate-data-init' `
+            -TimeoutSeconds $translatorTimeoutSeconds
 
         $shardArgs = @(
             'emit-build-shards', '--project', $project, '--base-metadata', $baseMetadata,
@@ -364,7 +318,8 @@ try {
             $shardArgs += @('--resolved-profile', (Join-Path $retroOut 'resolved_dispatch_profile.json'),
                 '--retro-cpp-dir', (Join-Path $retroOut 'cpp'))
         }
-        Invoke-Checked $translator $shardArgs 'Preparing local native build shards' -StepId 'emit-build-shards'
+        Invoke-Checked $translator $shardArgs 'Preparing local native build shards' -StepId 'emit-build-shards' `
+            -TimeoutSeconds $translatorTimeoutSeconds
 
         # The build directory is kept whenever the exact toolchain that filled it is the one
         # installed now and its CMake cache still belongs to this workspace path. Object files are
@@ -396,7 +351,7 @@ try {
             }
         }
         $dependencies = Join-Path $Workspace 'Dependencies'
-        $nativePrebuilt = Resolve-NativePrebuiltDirectory $dependencies $toolchainBin `
+        $nativePrebuilt = Resolve-MkwNativePrebuiltDirectory $dependencies $toolchainBin `
             (Join-Path $Workspace 'aurora-main') (Join-Path $Workspace 'runtime\third_party')
         if ($X86CpuProfile -eq 'v2') {
             Write-Host 'MKWCBUILD: This CPU does not support x86-64-v3; building for x86-64-v2'

Launcher/NativeBuildFlags.ps1

@@ -157,7 +157,8 @@ function Resolve-MkwProjectForAssets([string]$Workspace, [string]$AssetsDirector
 }
 
 function Invoke-Checked([string]$FilePath, [string[]]$Arguments, [string]$Description,
-    [string]$LogPrefix = 'MKWCBUILD', [string]$StepId = '', [bool]$WaitForProcessTree = $true) {
+    [string]$LogPrefix = 'MKWCBUILD', [string]$StepId = '', [bool]$WaitForProcessTree = $true,
+    [int]$TimeoutSeconds = 0) {
     <#
     Runs a build tool and turns a non-zero exit code into a described failure. By default,
     Start-Process -Wait waits for the whole process tree, since a .NET single-file bundle host may
@@ -166,6 +167,7 @@ function Invoke-Checked([string]$FilePath, [string[]]$Arguments, [string]$Descri
     Start-Process doesn't publish $LASTEXITCODE, so this sets it manually for callers that check it.
     -StepId emits the machine-readable form the installer's progress bar consumes (BuildStepIds in
     WiiCompiled.Setup/InstallProgress.cs); the human sentence stays on the same log line.
+    -TimeoutSeconds waits on the process itself and kills its tree when time runs out.
     #>
     if ($StepId) { Write-Host "${LogPrefix}:STEP:${StepId} $Description" }
     else { Write-Host "${LogPrefix}: $Description" }
@@ -174,7 +176,17 @@ function Invoke-Checked([string]$FilePath, [string[]]$Arguments, [string]$Descri
         # Backslashes directly before the closing quote would otherwise escape it.
         '"' + ($_ -replace '(\\+)$', '$1$1') + '"'
     })
-    if ($WaitForProcessTree) {
+    if ($TimeoutSeconds -gt 0) {
+        $process = Start-Process -FilePath $FilePath -ArgumentList $quotedArguments -NoNewWindow -PassThru
+        # Holding the handle keeps ExitCode readable once the process has exited.
+        $null = $process.Handle
+        if (-not $process.WaitForExit($TimeoutSeconds * 1000)) {
+            # taskkill's stderr would end the script under 'Stop' before the message below.
+            try { & taskkill.exe /T /F /PID $process.Id 2>&1 | Out-Null } catch {}
+            throw "$Description did not finish within $TimeoutSeconds seconds and was stopped."
+        }
+        $exitCode = $process.ExitCode
+    } elseif ($WaitForProcessTree) {
         $process = Start-Process -FilePath $FilePath -ArgumentList $quotedArguments `
             -NoNewWindow -Wait -PassThru
         $exitCode = $process.ExitCode
@@ -236,69 +248,121 @@ function Get-MkwDawnRuntimeSha256([string]$DependenciesDirectory) {
     return Get-MkwFileSha256 $runtime
 }
 
-function Get-MkwThirdPartySourceFingerprint([string]$ThirdPartyDirectory) {
+function ConvertTo-MkwHex([byte[]]$Bytes) {
+    return [BitConverter]::ToString($Bytes).Replace('-', '').ToLowerInvariant()
+}
+
+function Get-MkwSourceTreeFingerprint([string]$Directory, [string[]]$ExcludedTopDirectories = @()) {
     <#
-    SHA-256 over runtime/third_party, the vendored sources behind the prebuilt package. The
-    precompiled Crypto++ archive is linked while first-party TUs compile against the same headers,
-    so a workspace whose vendored tree moved on must not consume archives harvested from an older
-    one (same reasoning as the aurora fingerprint below), skipping dot-paths by name the same way.
-    Returns '' when absent.
+    SHA-256 over "relative-path content-sha256" lines of a source tree. It must come out the same
+    under powershell.exe 5.1 and pwsh 7 and for any checkout's line endings, because the package is
+    harvested on a release machine and re-checked on every user's: paths are sorted ordinally (not
+    with Sort-Object, whose culture collation differs between NLS and ICU) and text content is hashed
+    with CRLF folded to LF. Dot-paths are skipped by name, not by the hidden attribute, which the
+    payload zip drops. Returns '' when the tree is absent.
     #>
-    if ([string]::IsNullOrWhiteSpace($ThirdPartyDirectory)) { return '' }
-    if (-not (Test-Path -LiteralPath $ThirdPartyDirectory -PathType Container)) { return '' }
-    $root = [IO.Path]::GetFullPath($ThirdPartyDirectory).TrimEnd('\')
-    $files = Get-ChildItem -LiteralPath $root -Recurse -File -Force |
-        Where-Object { -not ($_.FullName.Substring($root.Length + 1) -split '\\' | Where-Object { $_.StartsWith('.') }) } |
-        Sort-Object { $_.FullName.Substring($root.Length + 1).Replace('\', '/') }
+    if ([string]::IsNullOrWhiteSpace($Directory)) { return '' }
+    if (-not (Test-Path -LiteralPath $Directory -PathType Container)) { return '' }
+    $root = [IO.Path]::GetFullPath($Directory).TrimEnd('\')
+    $relatives = [Collections.Generic.List[string]]::new()
+    foreach ($file in [IO.Directory]::EnumerateFiles($root, '*', [IO.SearchOption]::AllDirectories)) {
+        $relative = $file.Substring($root.Length + 1).Replace('\', '/')
+        $parts = $relative.Split('/')
+        if ($parts.Length -gt 1 -and $ExcludedTopDirectories -contains $parts[0]) { continue }
+        $hidden = $false
+        foreach ($part in $parts) { if ($part.StartsWith('.', [StringComparison]::Ordinal)) { $hidden = $true; break } }
+        if (-not $hidden) { $relatives.Add($relative) }
+    }
+    $sorted = $relatives.ToArray()
+    [Array]::Sort($sorted, [StringComparer]::Ordinal)
+
+    $latin1 = [Text.Encoding]::GetEncoding(28591)
     $sha = [Security.Cryptography.SHA256]::Create()
     try {
         $builder = [Text.StringBuilder]::new()
-        foreach ($file in $files) {
-            $relative = $file.FullName.Substring($root.Length + 1).Replace('\', '/')
-            $content = Get-MkwFileSha256 $file.FullName
-            [void]$builder.Append($relative).Append(' ').Append($content).Append("`n")
+        foreach ($relative in $sorted) {
+            $bytes = [IO.File]::ReadAllBytes($root + '\' + $relative.Replace('/', '\'))
+            # Latin-1 maps bytes 1:1, so the round trip only changes the line endings.
+            $text = $latin1.GetString($bytes)
+            if ($text.IndexOf([char]0) -lt 0 -and $text.Contains("`r`n")) {
+                $bytes = $latin1.GetBytes($text.Replace("`r`n", "`n"))
+            }
+            [void]$builder.Append($relative).Append(' ').Append((ConvertTo-MkwHex $sha.ComputeHash($bytes))).Append("`n")
         }
-        $bytes = $sha.ComputeHash([Text.Encoding]::UTF8.GetBytes($builder.ToString()))
+        return ConvertTo-MkwHex $sha.ComputeHash([Text.Encoding]::UTF8.GetBytes($builder.ToString()))
     } finally { $sha.Dispose() }
-    return (($bytes | ForEach-Object { $_.ToString('x2') }) -join '')
+}
+
+function Get-MkwThirdPartySourceFingerprint([string]$ThirdPartyDirectory) {
+    <#
+    runtime/third_party, the vendored sources behind the prebuilt package. The precompiled Crypto++
+    archive is linked while first-party TUs compile against the same headers, so a workspace whose
+    vendored tree moved on must not consume archives harvested from an older one.
+    #>
+    return Get-MkwSourceTreeFingerprint $ThirdPartyDirectory
 }
 
 function Get-MkwAuroraSourceFingerprint([string]$AuroraDirectory) {
     <#
-    SHA-256 over the aurora sources the precompiled archives were harvested from. The compiler,
-    flag, and Dawn digests only prove the package agrees with its target environment; none notice
-    aurora itself moving on, and since the runtime calls aurora headers directly, a package
-    harvested before an aurora change links old code against new call sites (undefined symbol at
-    best, silently stale renderer at worst). extern/ is excluded because the payload ships those
-    trees separately. Returns '' when the tree is absent.
+    The aurora sources the precompiled archives were harvested from. The compiler, flag, and Dawn
+    digests only prove the package agrees with its target environment; none notice aurora itself
+    moving on, and since the runtime calls aurora headers directly, a package harvested before an
+    aurora change links old code against new call sites. extern/ is excluded because the payload
+    ships those trees separately.
     #>
-    if ([string]::IsNullOrWhiteSpace($AuroraDirectory)) { return '' }
-    if (-not (Test-Path -LiteralPath $AuroraDirectory -PathType Container)) { return '' }
-    $root = [IO.Path]::GetFullPath($AuroraDirectory).TrimEnd('\')
-    $excludedPrefixes = @(
-        (Join-Path $root 'extern') + '\',
-        (Join-Path $root 'build') + '\'
-    )
-    # Dot-paths (the submodule's .git pointer) are skipped by name, not by the hidden attribute:
-    # the payload zip drops that attribute, so the installed side would otherwise hash them.
-    $files = Get-ChildItem -LiteralPath $root -Recurse -File -Force |
-        Where-Object {
-            $full = $_.FullName
-            -not ($excludedPrefixes | Where-Object { $full.StartsWith($_, [StringComparison]::OrdinalIgnoreCase) }) -and
-            -not ($full.Substring($root.Length + 1) -split '\\' | Where-Object { $_.StartsWith('.') })
-        } |
-        Sort-Object { $_.FullName.Substring($root.Length + 1).Replace('\', '/') }
-    $sha = [Security.Cryptography.SHA256]::Create()
+    return Get-MkwSourceTreeFingerprint $AuroraDirectory @('extern', 'build')
+}
+
+function Resolve-MkwNativePrebuiltDirectory([string]$DependenciesDirectory, [string]$ToolchainBin,
+    [string]$AuroraDirectory, [string]$ThirdPartyDirectory) {
+    <#
+    The shipped precompiled aurora/third-party package, verified against its recorded provenance
+    (exact compiler, exact flags, sources) before use. Anything unexpected falls back to building
+    from source. Shared with Build-Installer.ps1, which runs it against the staged payload under
+    powershell.exe so a package users would reject never ships.
+    #>
+    $package = Join-Path $DependenciesDirectory 'native_prebuilt'
+    $provenancePath = Join-Path $package 'provenance.json'
+    if (-not (Test-Path -LiteralPath $provenancePath -PathType Leaf)) {
+        Write-Host 'MKWCBUILD: No precompiled aurora package was shipped; building it from source'
+        return ''
+    }
     try {
-        $builder = [Text.StringBuilder]::new()
-        foreach ($file in $files) {
-            $relative = $file.FullName.Substring($root.Length + 1).Replace('\', '/')
-            $content = Get-MkwFileSha256 $file.FullName
-            [void]$builder.Append($relative).Append(' ').Append($content).Append("`n")
+        $provenance = Get-Content -LiteralPath $provenancePath -Raw | ConvertFrom-Json
+        if ($provenance.CompilerSha256 -ne (Get-MkwFileSha256 (Join-Path $ToolchainBin 'clang-22.exe'))) {
+            Write-Host 'MKWCBUILD: The precompiled aurora package was built by a different compiler; building it from source'
+            return ''
         }
-        $bytes = $sha.ComputeHash([Text.Encoding]::UTF8.GetBytes($builder.ToString()))
-    } finally { $sha.Dispose() }
-    return (($bytes | ForEach-Object { $_.ToString('x2') }) -join '')
+        if ($provenance.FlagFingerprint -ne (Get-MkwNativeFlagFingerprint)) {
+            Write-Host 'MKWCBUILD: The precompiled aurora package was built with different flags; building it from source'
+            return ''
+        }
+        # The archives embed Dawn's headers and link its import library; a different Dawn is an
+        # ABI mismatch that only shows up as a crash inside the GPU backend.
+        $recordedDawn = $provenance.PSObject.Properties['DawnRuntimeSha256']
+        if ($null -eq $recordedDawn -or
+            $recordedDawn.Value -ne (Get-MkwDawnRuntimeSha256 $DependenciesDirectory)) {
+            Write-Host 'MKWCBUILD: The precompiled aurora package was built against a different Dawn; building it from source'
+            return ''
+        }
+        $recordedAurora = $provenance.PSObject.Properties['AuroraSourceFingerprint']
+        if ($null -eq $recordedAurora -or
+            $recordedAurora.Value -ne (Get-MkwAuroraSourceFingerprint $AuroraDirectory)) {
+            Write-Host 'MKWCBUILD: The precompiled aurora package was built from different aurora sources; building it from source'
+            return ''
+        }
+        $recordedThirdParty = $provenance.PSObject.Properties['ThirdPartySourceFingerprint']
+        if ($null -eq $recordedThirdParty -or
+            $recordedThirdParty.Value -ne (Get-MkwThirdPartySourceFingerprint $ThirdPartyDirectory)) {
+            Write-Host 'MKWCBUILD: The precompiled aurora package was built from different vendored third-party sources; building it from source'
+            return ''
+        }
+    } catch {
+        Write-Host "MKWCBUILD: The precompiled aurora package could not be validated ($($_.Exception.Message)); building it from source"
+        return ''
+    }
+    Write-Host 'MKWCBUILD: Using the precompiled aurora and third-party libraries'
+    return $package
 }
 
 function ConvertTo-MkwCMakePath([string]$Path) {
@@ -347,9 +411,9 @@ function Get-MkwNativeConfigureArguments {
         }
     }
 
-    if (-not [string]::IsNullOrWhiteSpace($NativePrebuiltDirectory)) {
-        $arguments += "-DMKW_NATIVE_PREBUILT_DIR=$(ConvertTo-MkwCMakePath $NativePrebuiltDirectory)"
-    }
+    # Always set, so a reused CMake cache can never keep a package this run rejected.
+    $arguments += if ([string]::IsNullOrWhiteSpace($NativePrebuiltDirectory)) { '-DMKW_NATIVE_PREBUILT_DIR=' }
+        else { "-DMKW_NATIVE_PREBUILT_DIR=$(ConvertTo-MkwCMakePath $NativePrebuiltDirectory)" }
 
     $arguments += $AdditionalArguments
     return $arguments

Launcher/local-build-macos.command

@@ -64,6 +64,7 @@ done
 macos_arch=$(uname -m)
 case "$macos_arch" in arm64|x86_64) ;; *) fail "unsupported macOS architecture: $macos_arch" ;; esac
 macos_deployment_target=12.0
+macos_sysroot=$(xcrun --show-sdk-path 2>/dev/null) && [[ -d "$macos_sysroot" ]] || fail 'could not resolve the macOS SDK; install the Xcode Command Line Tools (xcode-select --install)'
 workspace=$(cd "$workspace" && pwd)
 if [[ -z "$setup_version" && -f "$workspace/Launcher/Directory.Build.props" ]]; then
     setup_version=$(sed -nE 's/.*<Version>([^<]+)<\/Version>.*/\1/p' "$workspace/Launcher/Directory.Build.props" | head -n 1)
@@ -186,7 +187,7 @@ fi
 step configure-native 'Configuring the native toolchain'
 # Use Aurora's pinned SDL3 source on macOS.  A system SDL3 can be older than
 # Aurora's required API even when find_package() succeeds.
-"$cmake_bin" -S "$workspace/runtime" -B "$native_build" -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER="$cc_bin" -DCMAKE_CXX_COMPILER="$cxx_bin" -DCMAKE_MAKE_PROGRAM="$ninja_bin" -DCMAKE_OSX_ARCHITECTURES="$macos_arch" -DCMAKE_OSX_DEPLOYMENT_TARGET="$macos_deployment_target" -DMKW_TRANSLATED_COMPILE_JOBS="$translated_jobs" -DMKW_X86_CPU_PROFILE="$x86_cpu_profile" -DAURORA_SDL3_PROVIDER=vendor -DMKW_SETUP_VERSION="$setup_version"
+"$cmake_bin" -S "$workspace/runtime" -B "$native_build" -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER="$cc_bin" -DCMAKE_CXX_COMPILER="$cxx_bin" -DCMAKE_MAKE_PROGRAM="$ninja_bin" -DCMAKE_OSX_ARCHITECTURES="$macos_arch" -DCMAKE_OSX_DEPLOYMENT_TARGET="$macos_deployment_target" -DCMAKE_OSX_SYSROOT="$macos_sysroot" -DMKW_TRANSLATED_COMPILE_JOBS="$translated_jobs" -DMKW_X86_CPU_PROFILE="$x86_cpu_profile" -DAURORA_SDL3_PROVIDER=vendor -DMKW_SETUP_VERSION="$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"

README.md

@@ -22,7 +22,7 @@ anywhere at runtime.
 
 > [!IMPORTANT]
 > There is no Nintendo code, no assets and no game data anywhere in this project or its releases.
-> You need your own legally dumped copy of the PAL version of the game. Setup only ships the
+> You need your own legally dumped copy of the PAL, NTSC-U, or NTSC-J version of the game. Setup only ships the
 > toolchain, the translation runs on your machine against your disc image, and nothing ever gets
 > uploaded.
 
@@ -103,14 +103,13 @@ Known limitations of the Wii Remote path:
 - About 20 GB of free disk space during installation (Final game size ~5 GB)
 - macOS 12 (Monterey) or later on Apple Silicon (`arm64`) or Intel (`x86_64-v3`); pre-Haswell Intel CPUs are unsupported
 - On macOS, a Metal-capable GPU and Apple Xcode Command Line Tools (Setup opens Apple's installer when they are missing)
-- A clean, unmodified **PAL `RMCP01`** disc image of Mario Kart Wii, dumped by you. ISO, GCM,
+- A clean, unmodified **PAL `RMCP01`**, **NTSC-U `RMCE01`**, or **NTSC-J `RMCJ01`** disc image of Mario Kart Wii, dumped by you. ISO, GCM,
   GCZ, CISO, WBFS, WIA and RVZ are accepted.
 
 > [!NOTE]
 > GPU/CPU minimums are set by driver support and D3D12/Vulkan feature requirements, not by the game's actual demands.
 
-Only the clean PAL revision will work. Anything else (other
-regions, patched executables) is rejected outright.
+Only the clean PAL, NTSC-U, and NTSC-J revisions will work. Anything else (NTSC-K, patched executables) is rejected outright.
 
 > [!NOTE]
 > Nobody here will tell you where to get the game. Dumping your own disc is on you, and links to
@@ -118,7 +117,7 @@ regions, patched executables) is rejected outright.
 
 ## Installing
 
-For an easy experience, use [Wheel Wizard](https://github.com/TeamWheelWizard/WheelWizard). Pick your clean PAL `RMCP01`
+For an easy experience, use [Wheel Wizard](https://github.com/TeamWheelWizard/WheelWizard). Pick your clean PAL `RMCP01`, NTSC-U `RMCE01`, or NTSC-J `RMCJ01`
 image under Settings, turn on **WiiCompiled (beta)**, and hit install from the Home page.
 Wheel Wizard downloads the setup tool from this repo and walks you through install, updates and
 launching. The backend itself is deliberately command-line only, Wheel Wizard is a wrapper around it.
@@ -128,7 +127,7 @@ launching. The backend itself is deliberately command-line only, Wheel Wizard is
 Download `WiiCompiled-Setup.pkg` from this repository's Releases page and open it. The universal
 package selects the appropriate bundled tools for the host architecture, supporting both Apple Silicon (`arm64`)
 and Intel (`x86_64`) Macs. It installs **WiiCompiled Setup** in Applications; open that app, choose
-your clean PAL `RMCP01` disc image, and select either the base game or Retro Rewind. For Retro Rewind, choose the `RetroRewind6` folder
+your clean PAL `RMCP01`, NTSC-U `RMCE01`, or NTSC-J `RMCJ01` disc image, and select either the base game or Retro Rewind. For Retro Rewind, choose the `RetroRewind6` folder
 or its parent folder.
 
 Setup verifies and extracts the image locally, then translates and compiles the native app on your
@@ -196,7 +195,7 @@ don't want to risk it right now, setup has to run a static recompiler over the w
 and then throw a C++ compiler at the result. It's a **one-time cost** on your machine.
 
 **Which game version works?**
-Clean PAL `RMCP01`. Other regions and modified executables are **rejected**. Translating
+Clean PAL `RMCP01`, NTSC-U `RMCE01`, and NTSC-J `RMCJ01`. NTSC-K `RMCK01` and modified executables are **rejected**. Translating
 them against the wrong manifest would give you a subtly broken game that's miserable to debug for us.
 
 **Can I recompile other GameCube/Wii games with it?**

runtime/cmake/PlatformTests.cmake

@@ -63,6 +63,15 @@ target_include_directories(mkw_input_expr_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}
 target_compile_features(mkw_input_expr_tests PRIVATE cxx_std_17)
 add_test(NAME mkw_input_expr_tests COMMAND mkw_input_expr_tests)
 
+add_executable(mkw_motion_trick_tests
+    "${MKW_TEST_RUNTIME_DIR}/tests/motion_trick_tests.cpp"
+    "${MKW_TEST_RUNTIME_DIR}/src/motion_trick_detector.cpp")
+target_include_directories(mkw_motion_trick_tests PRIVATE "${MKW_TEST_RUNTIME_DIR}/include")
+target_compile_features(mkw_motion_trick_tests PRIVATE cxx_std_17)
+# Match the runtime's float model so the bad-data guard is tested as shipped.
+target_compile_options(mkw_motion_trick_tests PRIVATE $<$<NOT:$<CXX_COMPILER_ID:MSVC>>:-ffast-math>)
+add_test(NAME mkw_motion_trick_tests COMMAND mkw_motion_trick_tests)
+
 # The emulated Bluetooth controller is pure host code.
 add_executable(mkw_bluetooth_controller_tests
     "${MKW_TEST_RUNTIME_DIR}/tests/bluetooth_controller_tests.cpp"

runtime/include/abi_bridge.h

@@ -157,7 +157,7 @@ public:
 #define MKW_DETAIL_CAT(a, b) a##b
 #define MKW_DETAIL_MAKE_UNIQUE(a, b) MKW_DETAIL_CAT(a, b)
 
-// Three hand-written registration macros, given the PAL address. Generated code never registers
+// Hand-written registration macros, given the PAL address. Generated code never registers
 // per function; every translated function reaches the registry via the bulk
 // BulkTranslatedFunctionRecord tables in the translator's *_registration TUs.
 #define REGISTER_TRANSLATED_FUNCTION(address, fn) \
@@ -168,3 +168,9 @@ public:
 
 #define REGISTER_NATIVE_FUNCTION_AS(address, fn, pretty_name) \
     static AbiTrampoline<decltype(fn)> MKW_DETAIL_MAKE_UNIQUE(_abi_native_trampoline_named_, __COUNTER__)(::GuestRegion::Translate(address), pretty_name, fn, FunctionKind::Native, false, kPpcAllNonvolatileFprMask, 0, 0, &AbiRawCpuThunk<&fn>::Invoke)
+
+// A native that calls through with MKW_GUEST_NEXT, so a mod overlay at its address still runs.
+#define REGISTER_NATIVE_WRAPPER_AS(address, fn, pretty_name) REGISTER_NATIVE_FUNCTION_AS(address, fn, pretty_name)
+
+// A native that deliberately takes over its address from any mod overlay there.
+#define REGISTER_NATIVE_REPLACEMENT_AS(address, fn, pretty_name) REGISTER_NATIVE_FUNCTION_AS(address, fn, pretty_name)

runtime/include/audio_backend.h

@@ -22,6 +22,8 @@ public:
     // Applied to the final host output, covering both AX and direct AI DMA.
     void SetMasterVolume(float volume);
     void SetMuted(bool muted);
+    // Mutes output without touching the user's persisted mute setting.
+    void SetSilenced(bool silenced);
 
 private:
     AudioBackend() = default;
@@ -43,6 +45,7 @@ private:
     bool m_initialized = false;
     float m_masterVolume = 1.0f;
     bool m_muted = false;
+    bool m_silenced = false;
     bool m_reportedDroppedBlock = false;
     std::vector<int16_t> m_convertBuffer;
 };

runtime/include/aurora_events.h

@@ -2,6 +2,7 @@
 
 #include "friend_code_keyboard.h"
 #include "settings_overlay.h"
+#include "steam_input_layout.h"
 #include "runtime_config.h"
 #include "gx_worker.h"
 #include "perf_stats.h"
@@ -19,9 +20,8 @@
 #endif
 
 extern "C" bool g_dynamicAspectRatioEnabled;
-void ConfigureMkwDynamicAspect(bool widescreen, bool forceAspect169, uint32_t surfaceWidth, uint32_t surfaceHeight);
-void SetMkwForceAspect169(bool enabled);
-bool MkwForceAspect169Requested();
+void ConfigureMkwDynamicAspect(bool widescreen, bool forceAspect, uint32_t surfaceWidth, uint32_t surfaceHeight);
+void SetMkwForceAspect(bool enabled);
 void UpdateMkwDynamicAspectSurface(uint32_t surfaceWidth, uint32_t surfaceHeight);
 // Arms the "keep EGG::Frustum's projection scale" flag on every screen that
 // renders to a fixed-size offscreen target. Cheap and idempotent; called from
@@ -64,6 +64,7 @@ inline void Flush(bool force = false) {
 // process directly and do not run the crash/atexit paths.
 [[noreturn]] inline void ExitForAuroraWindowClose() noexcept {
     settings_overlay::ReleaseControllers();
+    SteamInputLayout::Release();
     WindowPlacementPersistence::Flush(true);
 #if defined(_WIN32)
     ::ExitProcess(0);
@@ -121,6 +122,7 @@ inline void ProcessAuroraEvents(const AuroraEvent* events) {
     }
     settings_overlay::HandleEvents(events);
     FriendCodeKeyboard::HandleEvents(events);
+    SteamInputLayout::HandleEvents(events);
 }
 
 inline void ApplyPendingMkwDynamicAspectSurface() {

runtime/include/boot_race.h

@@ -7,4 +7,5 @@
 namespace BootRace {
 // False when `arg` is not a boot-race option; throws on a malformed one.
 bool ParseArgument(std::string_view arg);
+bool Requested();
 } // namespace BootRace

runtime/include/guest_function_imports.h

@@ -3,6 +3,8 @@
 // Base translations the runtime calls directly, named by PAL identity and resolved through the
 // function registry on first use.
 //   MKW_GUEST_FUNC(801A0620)(ctx);
+// A native registered with REGISTER_NATIVE_WRAPPER_AS calls through with MKW_GUEST_NEXT instead,
+// so a mod overlay at the same address still runs.
 
 #include "guest_abi/cpu_context.h"
 #include "guest_abi/guest_region.h"
@@ -12,6 +14,10 @@ using Entry = void (*)(CpuContext*);
 
 // The base translation at a guest address, also where a native or mod overrides it.
 Entry BaseTranslation(uint32_t address);
+
+// The translation a native wrapper sits on top of: the winning mod overlay, else the base
+// translation.
+Entry NextTranslation(uint32_t address);
 } // namespace GuestFunctions
 
 #define MKW_GUEST_FUNC(pal)                                                                        \
@@ -19,3 +25,9 @@ Entry BaseTranslation(uint32_t address);
         static const ::GuestFunctions::Entry entry = ::GuestFunctions::BaseTranslation(MKW_GADDR(pal)); \
         return entry;                                                                              \
     }())
+
+#define MKW_GUEST_NEXT(pal)                                                                        \
+    (*[] {                                                                                         \
+        static const ::GuestFunctions::Entry entry = ::GuestFunctions::NextTranslation(MKW_GADDR(pal)); \
+        return entry;                                                                              \
+    }())

runtime/include/motion_trick.h

@@ -0,0 +1,36 @@
+#pragma once
+
+// Shake to trick: a flick of a controller's motion sensor briefly presses the
+// matching D-pad direction, which the game treats as a trick (or a wheelie on
+// bikes), like shaking a Wii Remote.
+
+#include <cstdint>
+
+#include <dolphin/pad.h>
+
+#include "motion_trick_detector.h"
+
+namespace MotionTrick {
+
+bool Enabled() noexcept;
+void SetEnabled(bool enabled) noexcept;
+int Sensitivity() noexcept;
+void SetSensitivity(int level) noexcept;
+
+// The controller on this port reports a gyro or accelerometer of its own.
+bool PortHasMotion(uint32_t port) noexcept;
+
+// Call once per guest PADRead, after every other input source has been merged.
+void Apply(PADStatus* statuses) noexcept;
+
+struct PortStatus {
+    float intensity = 0.0f; // decaying peak, 1 being the trigger threshold
+    Direction lastTrick = Direction::None;
+    uint64_t lastTrickUs = 0;
+    bool sensorsUnavailable = false; // the controller refused to turn its sensors on
+};
+PortStatus Status(uint32_t port) noexcept;
+
+uint64_t NowUs() noexcept;
+
+} // namespace MotionTrick

runtime/include/motion_trick_detector.h

@@ -0,0 +1,43 @@
+#pragma once
+
+// Shake detection for controllers with motion sensors. Pure host code, so it is
+// unit-testable without SDL.
+
+#include <cstdint>
+
+namespace MotionTrick {
+
+enum class Direction : uint8_t { None, Up, Down, Left, Right };
+
+inline constexpr int kMinSensitivity = 1;
+inline constexpr int kMaxSensitivity = 5;
+inline constexpr int kDefaultSensitivity = 3;
+
+// SDL gamepad sensor axes: x right, y top, z toward the player.
+struct MotionSample {
+    bool hasGyro = false;
+    bool hasAccel = false;
+    float gyro[3]{};  // rad/s: pitch, yaw, roll
+    float accel[3]{}; // m/s^2, gravity included
+};
+
+// Fed one sample per game frame; reports at most one trick per flick.
+class ShakeDetector {
+public:
+    ShakeDetector() { SetSensitivity(kDefaultSensitivity); }
+
+    void SetSensitivity(int level);
+    Direction Update(const MotionSample& sample, uint64_t nowUs);
+
+    // Strength of the last sample, 1 being the trigger threshold.
+    float Intensity() const { return m_intensity; }
+
+private:
+    float m_gyroThreshold = 0.0f;
+    float m_accelThreshold = 0.0f;
+    float m_intensity = 0.0f;
+    bool m_armed = true;
+    uint64_t m_lastFireUs = 0;
+};
+
+} // namespace MotionTrick

runtime/include/runtime_config.h

@@ -33,7 +33,8 @@
 
 struct RuntimeUserConfig {
     std::optional<bool> widescreen;
-    std::optional<bool> forceAspect169;
+    std::optional<bool> forceAspect;
+    bool legacyForceAspect = false;
     std::optional<int32_t> windowPosX;
     std::optional<int32_t> windowPosY;
     std::optional<uint32_t> windowWidth;
@@ -69,6 +70,10 @@ struct RuntimeUserConfig {
     std::optional<int32_t> wiiIrSensitivity;
     std::optional<int32_t> wiiSpeakerVolume;
     std::optional<bool> wiiRumble;
+    // Outside a Steam launch, ask a running Steam client for a game controller layout while
+    // the window has focus instead of its desktop one (buttons bound to mouse and keyboard).
+    std::optional<bool> steamInputOverride;
+    std::optional<uint32_t> steamInputAppId;
     std::optional<bool> networkEnabled;
     std::optional<bool> discordPresenceEnabled;
     // The application ID of the WiiCompiled Discord application. This is only
@@ -86,6 +91,9 @@ struct RuntimeUserConfig {
     // "dpad_up,left_shoulder") as values; pressing either bound button counts.
     std::array<std::optional<std::string>, 12> controllerButtons;
     std::optional<bool> rumbleEnabled;
+    // Shake a motion controller to press the D-pad, like shaking a Wii Remote.
+    std::optional<bool> motionTrick;
+    std::optional<int32_t> motionTrickSensitivity;
     std::optional<int32_t> muteHotkey;
     std::map<std::string, std::string> controllerExpressions;
 };
@@ -271,7 +279,7 @@ inline void EnsureConfigFile() {
               "# Set paths.dvd_root to an extracted Mario Kart Wii DATA directory.\n\n"
               "[video]\n"
               "widescreen = true\n"
-              "force_16_9 = false\n"
+              "force_aspect = false\n"
               "resolution_multiplier = 1.0\n"
               "frame_interpolation_fps = 0\n"
               "display_mode = \"windowed\"\n"
@@ -304,6 +312,12 @@ inline void EnsureConfigFile() {
               "mix_worker = true\n\n"
               "[network]\n"
               "enabled = true\n\n"
+              "[controller]\n"
+              "# Outside a Steam launch, Steam's desktop layout binds controller buttons\n"
+              "# to mouse and keyboard. While the game has focus, ask a running Steam\n"
+              "# client to use the game layout of steam_input_app_id instead.\n"
+              "steam_input_override = true\n"
+              "# steam_input_app_id = 480\n\n"
               "[discord]\n"
               "# Rich Presence talks only to a locally-running Discord client.\n"
               "# Retro Rewind supplies its official app ID automatically. Set this\n"
@@ -389,6 +403,10 @@ inline RuntimeUserConfig ParseConfigDocument(const toml::value& document) {
     }
 
     config.rumbleEnabled = FindConfigValue<bool>(document, "controller", "rumble");
+    config.motionTrick = FindConfigValue<bool>(document, "controller", "motion_trick");
+    if (const auto value = FindConfigInt(document, "controller", "motion_trick_sensitivity"); value && *value >= 1 && *value <= 5) {
+        config.motionTrickSensitivity = *value;
+    }
     if (auto value = FindConfigInt(document, "audio", "mute_key")) {
         config.muteHotkey = *value;
     }
@@ -403,7 +421,15 @@ inline RuntimeUserConfig ParseConfigDocument(const toml::value& document) {
     }
 
     config.widescreen = FindConfigValue<bool>(document, "video", "widescreen");
-    config.forceAspect169 = FindConfigValue<bool>(document, "video", "force_16_9");
+    config.forceAspect = FindConfigValue<bool>(document, "video", "force_aspect");
+    if (!config.forceAspect) {
+        // Legacy key; it always forced a 16:9 game.
+        config.forceAspect = FindConfigValue<bool>(document, "video", "force_16_9");
+        config.legacyForceAspect = config.forceAspect.has_value();
+        if (config.forceAspect.value_or(false)) {
+            config.widescreen = true;
+        }
+    }
     config.windowPosX = FindConfigInt(document, "video", "window_x");
     config.windowPosY = FindConfigInt(document, "video", "window_y");
     if (auto value = FindConfigUint(document, "video", "window_width"); value && *value != 0) {
@@ -473,6 +499,10 @@ inline RuntimeUserConfig ParseConfigDocument(const toml::value& document) {
         config.wiiSpeakerVolume = *value;
     }
     config.wiiRumble = FindConfigValue<bool>(document, "controller", "wii_rumble");
+    config.steamInputOverride = FindConfigValue<bool>(document, "controller", "steam_input_override");
+    if (const auto value = FindConfigUint(document, "controller", "steam_input_app_id"); value && *value != 0) {
+        config.steamInputAppId = *value;
+    }
     config.networkEnabled = FindConfigValue<bool>(document, "network", "enabled");
     config.discordPresenceEnabled = FindConfigValue<bool>(document, "discord", "enabled");
     config.discordClientId = FindConfigValue<std::string>(document, "discord", "client_id");
@@ -692,6 +722,24 @@ inline bool SetRumbleEnabled(bool value) {
     return WriteSetting("controller", "rumble", value ? "true" : "false");
 }
 
+inline bool MotionTrickEnabled(bool fallback = false) {
+    return Get().motionTrick.value_or(fallback);
+}
+
+inline bool SetMotionTrickEnabled(bool value) {
+    Mutable().motionTrick = value;
+    return WriteSetting("controller", "motion_trick", value ? "true" : "false");
+}
+
+inline int32_t MotionTrickSensitivity(int32_t fallback = 3) {
+    return Get().motionTrickSensitivity.value_or(fallback);
+}
+
+inline bool SetMotionTrickSensitivity(int32_t value) {
+    Mutable().motionTrickSensitivity = value;
+    return WriteSetting("controller", "motion_trick_sensitivity", std::to_string(value));
+}
+
 inline int32_t MuteHotkey(int32_t fallback) {
     return Get().muteHotkey.value_or(fallback);
 }
@@ -760,13 +808,30 @@ inline bool WidescreenEnabled(bool fallback = false) {
     return Get().widescreen.value_or(fallback);
 }
 
-inline bool ForceAspect169Enabled(bool fallback = false) {
-    return Get().forceAspect169.value_or(fallback);
+inline bool ForceAspectEnabled(bool fallback = false) {
+    return Get().forceAspect.value_or(fallback);
 }
 
-inline bool SetForceAspect169(bool value) {
-    Mutable().forceAspect169 = value;
-    return WriteSetting("video", "force_16_9", value ? "true" : "false");
+inline bool SetForceAspect(bool value) {
+    Mutable().forceAspect = value;
+    return WriteSetting("video", "force_aspect", value ? "true" : "false");
+}
+
+// The in-memory value stays as booted; SCGetAspectRatio must not change mid-session.
+inline bool SetWidescreen(bool value) {
+    return WriteSetting("video", "widescreen", value ? "true" : "false");
+}
+
+// Rewrites a legacy force_16_9 as force_aspect plus the widescreen it implied.
+inline void MigrateLegacyForceAspect() {
+    if (!Get().legacyForceAspect) {
+        return;
+    }
+    const bool force = ForceAspectEnabled();
+    if ((!force || WriteSetting("video", "widescreen", "true")) &&
+        WriteSetting("video", "force_aspect", force ? "true" : "false")) {
+        Mutable().legacyForceAspect = false;
+    }
 }
 
 inline bool WindowPosition(int32_t& x, int32_t& y) {
@@ -844,6 +909,15 @@ inline bool SetWiiRemotesEnabled(bool value) {
     return true;
 }
 
+inline bool SteamInputOverrideEnabled(bool fallback = true) {
+    return Get().steamInputOverride.value_or(fallback);
+}
+
+// Spacewar, Valve's Steamworks sample app, available to every Steam account.
+inline uint32_t SteamInputAppId(uint32_t fallback = 480) {
+    return Get().steamInputAppId.value_or(fallback);
+}
+
 inline bool WiiAutoPairEnabled(bool fallback = true) {
     return Get().wiiAutoPair.value_or(fallback);
 }

runtime/include/steam_input_layout.h

@@ -0,0 +1,17 @@
+#pragma once
+
+#include <aurora/event.h>
+
+#include <cstdint>
+
+// Outside a Steam launch, Steam applies its desktop controller layout, which binds
+// controller buttons to mouse and keyboard actions. While the game window has focus,
+// a running Steam client is asked to apply the given app's game layout instead.
+namespace SteamInputLayout {
+
+void Initialize(bool enabled, uint32_t appId);
+void HandleEvents(const AuroraEvent* events);
+// Restores the desktop layout. Idempotent; safe to call from any exit path.
+void Release() noexcept;
+
+} // namespace SteamInputLayout

runtime/src/abi_bridge.cpp

@@ -426,20 +426,10 @@ const TranslatedFunctionInfo* TranslatedFunctionRegistry::FindByAddressPtrSlow(u
     return nullptr;
 }
 
-GuestFunctions::Entry GuestFunctions::BaseTranslation(uint32_t address) {
-    std::unique_lock<std::mutex> lock(RegistryMutex(), std::defer_lock);
-    if (!RegistryFrozen().load(std::memory_order_acquire)) {
-        lock.lock();
-    }
-    for (const auto& entry : Registry()) {
-        if (entry.address == address && entry.kind == FunctionKind::BaseTranslated && entry.rawCpuInvoker) {
-            return entry.rawCpuInvoker;
-        }
-    }
-    if (lock.owns_lock()) {
-        lock.unlock();
-    }
-    RT_LOG(RT_TAG_RUNTIME) << "ERROR: no base translation registered at 0x" << std::hex << address << std::dec
+namespace {
+
+[[noreturn]] void ReportMissingGuestFunction(uint32_t address) {
+    RT_LOG(RT_TAG_RUNTIME) << "ERROR: no translation registered at 0x" << std::hex << address << std::dec
                            << std::endl;
     std::ostringstream message;
     message << "The runtime calls the translated function at guest address 0x" << std::hex << address
@@ -451,6 +441,43 @@ GuestFunctions::Entry GuestFunctions::BaseTranslation(uint32_t address) {
     std::exit(EXIT_FAILURE);
 }
 
+} // namespace
+
+GuestFunctions::Entry GuestFunctions::BaseTranslation(uint32_t address) {
+    {
+        std::unique_lock<std::mutex> lock(RegistryMutex(), std::defer_lock);
+        if (!RegistryFrozen().load(std::memory_order_acquire)) {
+            lock.lock();
+        }
+        for (const auto& entry : Registry()) {
+            if (entry.address == address && entry.kind == FunctionKind::BaseTranslated && entry.rawCpuInvoker) {
+                return entry.rawCpuInvoker;
+            }
+        }
+    }
+    ReportMissingGuestFunction(address);
+}
+
+GuestFunctions::Entry GuestFunctions::NextTranslation(uint32_t address) {
+    {
+        std::unique_lock<std::mutex> lock(RegistryMutex(), std::defer_lock);
+        if (!RegistryFrozen().load(std::memory_order_acquire)) {
+            lock.lock();
+        }
+        const TranslatedFunctionInfo* best = nullptr;
+        for (const auto& entry : Registry()) {
+            if (entry.address == address && entry.kind != FunctionKind::Native && entry.rawCpuInvoker &&
+                (!best || IsBetterCandidate(entry, *best))) {
+                best = &entry;
+            }
+        }
+        if (best) {
+            return best->rawCpuInvoker;
+        }
+    }
+    ReportMissingGuestFunction(address);
+}
+
 const RawDispatchRecord* TranslatedFunctionRegistry::FindRawByAddressPtrSlow(uint32_t address) {
     if (!RegistryFrozen().load(std::memory_order_acquire)) {
         return nullptr;

runtime/src/audio_backend.cpp

@@ -15,7 +15,7 @@ AudioBackend& AudioBackend::Instance() {
 }
 
 float AudioBackend::EffectiveGainLocked() const {
-    return m_muted ? 0.0f : m_masterVolume;
+    return m_muted || m_silenced ? 0.0f : m_masterVolume;
 }
 
 void AudioBackend::ApplyGainLocked() {
@@ -36,6 +36,12 @@ void AudioBackend::SetMuted(bool muted) {
     ApplyGainLocked();
 }
 
+void AudioBackend::SetSilenced(bool silenced) {
+    std::lock_guard<std::mutex> lock(m_mutex);
+    m_silenced = silenced;
+    ApplyGainLocked();
+}
+
 bool AudioBackend::EnsureInitializedLocked(uint32_t sampleRate, uint32_t channels) {
     if (m_initialized && m_sampleRate == sampleRate && m_channels == channels) {
         return true;

runtime/src/dynamic_aspect.cpp

@@ -14,10 +14,9 @@ extern "C" int g_gxFrameCount;
 
 namespace {
 
-bool g_widescreenConfigured = false;
-bool g_widescreenSetting = false;
-bool g_forceAspect169 = false;
-std::atomic_bool g_requestedForceAspect169{false};
+bool g_widescreen = true;
+bool g_forceAspect = false;
+std::atomic_bool g_requestedForceAspect{false};
 bool g_policyDirty = false;
 uint32_t g_lastEggWidth43 = 0;
 uint32_t g_lastEggWidth169 = 0;
@@ -130,19 +129,20 @@ void WriteEggScreenRecord(uint32_t recordAddr, uint32_t width) {
 
 void ApplyEggScreenRecords(uint32_t surfaceWidth, uint32_t surfaceHeight) {
     using namespace MkwDynamicAspect;
+    if (g_forceAspect) {
+        surfaceWidth = g_widescreen ? 16u : 4u;
+        surfaceHeight = g_widescreen ? 9u : 3u;
+    }
+    const float referenceAspect = g_widescreen ? kPresentationAspect169 : kPresentationAspect43;
     const uint32_t width43 =
-        g_widescreenConfigured ? EggRecordWidth(kEggRecordWidth43, surfaceWidth, surfaceHeight)
-                               : kEggRecordWidth43;
+        EggRecordWidth(kEggRecordWidth43, surfaceWidth, surfaceHeight, referenceAspect);
     const uint32_t width169 =
-        g_widescreenConfigured ? EggRecordWidth(kEggRecordWidth169, surfaceWidth, surfaceHeight)
-                               : kEggRecordWidth169;
+        EggRecordWidth(kEggRecordWidth169, surfaceWidth, surfaceHeight);
     WriteEggScreenRecord(kEggScreenRecord43, width43);
     WriteEggScreenRecord(kEggScreenRecord169, width169);
     Memory::WriteFloat32(kEggScreenProjScaleX, 1.0f);
     Memory::WriteFloat32(kEggScreenProjScaleY,
-                         g_widescreenConfigured
-                             ? VerticalExpansion(surfaceWidth, surfaceHeight)
-                             : 1.0f);
+                         VerticalExpansion(surfaceWidth, surfaceHeight, referenceAspect));
     AssertOffscreenScreenBypass();
     if (width43 == g_lastEggWidth43 && width169 == g_lastEggWidth169) {
         return;
@@ -172,58 +172,38 @@ void AssertMkwOffscreenScreenBypass() {
 }
 
 void UpdateMkwDynamicAspectSurface(uint32_t surfaceWidth, uint32_t surfaceHeight) {
-    const bool requestedForceAspect169 = g_requestedForceAspect169.load(std::memory_order_acquire);
-    if (g_forceAspect169 != requestedForceAspect169) {
-        g_forceAspect169 = requestedForceAspect169;
-        g_widescreenConfigured = g_widescreenSetting || g_forceAspect169;
-        g_dynamicAspectRatioEnabled = g_widescreenConfigured;
+    const bool requestedForceAspect = g_requestedForceAspect.load(std::memory_order_acquire);
+    if (g_forceAspect != requestedForceAspect) {
+        g_forceAspect = requestedForceAspect;
         g_policyDirty = true;
     }
     if (surfaceWidth == 0 || surfaceHeight == 0) {
         return;
     }
-    if (!g_widescreenConfigured) {
-        if (g_policyDirty) {
+    if (g_policyDirty) {
+        if (g_forceAspect && !g_widescreen) {
             AuroraSetViewportPolicy(AURORA_VIEWPORT_FIT);
             VILockAspectRatio(4, 3);
-            ApplyEggScreenRecords(surfaceWidth, surfaceHeight);
-            g_policyDirty = false;
+        } else {
+            AuroraSetViewportPolicy(g_forceAspect ? AURORA_VIEWPORT_16_9 : AURORA_VIEWPORT_STRETCH);
+            VIUnlockAspectRatio();
         }
-        return;
-    }
-    if (g_policyDirty) {
-        VIUnlockAspectRatio();
         g_policyDirty = false;
     }
-    AuroraSetViewportPolicy(g_forceAspect169 ? AURORA_VIEWPORT_16_9 : AURORA_VIEWPORT_STRETCH);
-    ApplyEggScreenRecords(g_forceAspect169 ? 16u : surfaceWidth,
-                          g_forceAspect169 ? 9u : surfaceHeight);
-}
-
-void SetMkwForceAspect169(bool enabled) {
-    g_requestedForceAspect169.store(enabled, std::memory_order_release);
-}
-
-bool MkwForceAspect169Requested() {
-    return g_requestedForceAspect169.load(std::memory_order_acquire);
-}
-
-void ConfigureMkwDynamicAspect(bool widescreen, bool forceAspect169, uint32_t surfaceWidth, uint32_t surfaceHeight) {
-    g_widescreenSetting = widescreen;
-    g_widescreenConfigured = widescreen || forceAspect169;
-    g_forceAspect169 = forceAspect169;
-    g_requestedForceAspect169.store(forceAspect169, std::memory_order_relaxed);
-    g_dynamicAspectRatioEnabled = g_widescreenConfigured;
-    g_lastEggWidth43 = 0;
-    g_lastEggWidth169 = 0;
-    if (g_widescreenConfigured) {
-        VIUnlockAspectRatio();
-        ApplyEggScreenRecords(forceAspect169 ? 16u : surfaceWidth,
-                              forceAspect169 ? 9u : surfaceHeight);
-        return;
-    }
-
-    AuroraSetViewportPolicy(AURORA_VIEWPORT_FIT);
-    VILockAspectRatio(4, 3);
+    ApplyEggScreenRecords(surfaceWidth, surfaceHeight);
+}
+
+void SetMkwForceAspect(bool enabled) {
+    g_requestedForceAspect.store(enabled, std::memory_order_release);
+}
+
+void ConfigureMkwDynamicAspect(bool widescreen, bool forceAspect, uint32_t surfaceWidth, uint32_t surfaceHeight) {
+    g_widescreen = widescreen;
+    g_forceAspect = forceAspect;
+    g_requestedForceAspect.store(forceAspect, std::memory_order_relaxed);
+    g_policyDirty = true;
+    g_dynamicAspectRatioEnabled = true;
+    g_lastEggWidth43 = 0;
+    g_lastEggWidth169 = 0;
     ApplyEggScreenRecords(surfaceWidth, surfaceHeight);
 }

runtime/src/hle/audio/audio.cpp

@@ -503,19 +503,20 @@ int64_t ConsumeAudioPollDeltaMicros()
     using Clock = std::chrono::steady_clock;
     static thread_local Clock::time_point lastPoll = Clock::now();
 
-    const Clock::time_point now = Clock::now();
-    auto elapsed = std::chrono::duration_cast<std::chrono::microseconds>(now - lastPoll).count();
-    lastPoll = now;
-
-    if (elapsed < 0) {
-        elapsed = 0;
-    }
-
     // Cap the catch-up interval after a debugger pause or host stall; backlog drains via
     // kMaxBlocksPerTick per pass. Never return early on a zero delta: two scheduler passes
     // can land in the same microsecond and the backlog still needs servicing.
-    constexpr int64_t kMaxPollDeltaMicros = 100'000;
-    return std::min(elapsed, kMaxPollDeltaMicros);
+    constexpr std::chrono::microseconds kMaxPollDelta{100'000};
+
+    const Clock::time_point now = Clock::now();
+    const auto elapsed = std::chrono::duration_cast<std::chrono::microseconds>(now - lastPoll);
+    if (elapsed.count() < 0 || elapsed > kMaxPollDelta) {
+        lastPoll = now;
+        return elapsed.count() < 0 ? 0 : kMaxPollDelta.count();
+    }
+    // Keep the sub-microsecond remainder for the next poll.
+    lastPoll += elapsed;
+    return elapsed.count();
 }
 
 } // namespace

runtime/src/hle/boot_race.cpp

@@ -125,6 +125,10 @@ bool ParseArgument(std::string_view arg) {
     return true;
 }
 
+bool Requested() {
+    return g_course.has_value();
+}
+
 } // namespace BootRace
 
 // ArgumentParser::parse, picks the boot section in SectionMgr::Init.
@@ -158,7 +162,7 @@ REGISTER_NATIVE_FUNCTION_AS(0x805B9344, ArgumentParser_Parse_805b9344, "Argument
 // RacedataScenario::InitScreens; local racers keep no real controller channel.
 extern "C" void RacedataScenario_InitScreens_8052f4e8(CpuContext* ctx) {
     if (!g_cpuDriven) {
-        MKW_GUEST_FUNC(8052F4E8)(ctx);
+        MKW_GUEST_NEXT(8052F4E8)(ctx);
         return;
     }
     const uint32_t players = ctx->gpr[3] + kPlayersOffset;
@@ -166,7 +170,7 @@ extern "C" void RacedataScenario_InitScreens_8052f4e8(CpuContext* ctx) {
     for (uint32_t i = 0; i < kPlayerCount; ++i) {
         channels[i] = Memory::Read8(players + i * kPlayerSize + kPlayerControllerChannelOffset);
     }
-    MKW_GUEST_FUNC(8052F4E8)(ctx);
+    MKW_GUEST_NEXT(8052F4E8)(ctx);
     for (uint32_t i = 0; i < kPlayerCount; ++i) {
         const uint32_t player = players + i * kPlayerSize;
         if (Memory::Read32(player + kPlayerTypeOffset) == kPlayerRealLocal) {
@@ -174,7 +178,7 @@ extern "C" void RacedataScenario_InitScreens_8052f4e8(CpuContext* ctx) {
         }
     }
 }
-REGISTER_NATIVE_FUNCTION_AS(0x8052F4E8, RacedataScenario_InitScreens_8052f4e8, "RacedataScenario_InitScreens_8052f4e8");
+REGISTER_NATIVE_WRAPPER_AS(0x8052F4E8, RacedataScenario_InitScreens_8052f4e8, "RacedataScenario_InitScreens_8052f4e8");
 
 // KartAIController::IsCPU
 extern "C" void KartAIController_IsCPU_8072624c(CpuContext* ctx) {

runtime/src/hle/gx/gx_dynamic_aspect.h

@@ -13,6 +13,7 @@
 namespace MkwDynamicAspect {
 
 inline constexpr float kPresentationAspect169 = 16.0f / 9.0f;
+inline constexpr float kPresentationAspect43 = 4.0f / 3.0f;
 inline constexpr float kEggFramebufferWidth = 608.0f;
 inline constexpr uint32_t kEggScreenHeight = 456u;
 inline constexpr uint32_t kEggRecordWidth43 = 608u;
@@ -31,32 +32,35 @@ inline float SurfaceAspect(uint32_t width, uint32_t height) noexcept {
         static_cast<float>(width) / static_cast<float>(height));
 }
 
-// surfaceAspect / (16:9). Above 1 the surface is wider than 16:9, below 1 it
-// is tighter (4:3, portrait).
-inline float AspectRatioFactor(uint32_t width, uint32_t height) noexcept {
-    return SurfaceAspect(width, height) / kPresentationAspect169;
+// surfaceAspect / referenceAspect. Above 1 the surface is wider than the
+// reference, below 1 it is tighter (4:3, portrait).
+inline float AspectRatioFactor(uint32_t width, uint32_t height,
+                               float referenceAspect = kPresentationAspect169) noexcept {
+    return SurfaceAspect(width, height) / referenceAspect;
 }
 
-// Hor+ factor, clamped at 1 so a tighter-than-16:9 surface never narrows the
-// canvas (that would crop away scene the 16:9 view showed).
-inline float HorizontalExpansion(uint32_t width, uint32_t height) noexcept {
-    return std::max(1.0f, AspectRatioFactor(width, height));
+// Hor+ factor, clamped at 1 so a tighter surface never narrows the canvas
+// (that would crop away scene the reference view showed).
+inline float HorizontalExpansion(uint32_t width, uint32_t height,
+                                 float referenceAspect = kPresentationAspect169) noexcept {
+    return std::max(1.0f, AspectRatioFactor(width, height, referenceAspect));
 }
 
-// Vert+ factor, clamped at 1 so a wider-than-16:9 surface never shortens the
-// frustum. Exactly one of the two expansions is ever greater than 1.
-inline float VerticalExpansion(uint32_t width, uint32_t height) noexcept {
-    const float factor = AspectRatioFactor(width, height);
+// Vert+ factor, clamped at 1 so a wider surface never shortens the frustum.
+// Exactly one of the two expansions is ever greater than 1.
+inline float VerticalExpansion(uint32_t width, uint32_t height,
+                               float referenceAspect = kPresentationAspect169) noexcept {
+    const float factor = AspectRatioFactor(width, height, referenceAspect);
     if (!(factor > 0.0f)) {
         return 1.0f;
     }
     return std::clamp(1.0f / factor, 1.0f, kMaxVerticalExpansion);
 }
 
-inline uint32_t EggRecordWidth(uint32_t baseWidth, uint32_t surfaceWidth,
-                               uint32_t surfaceHeight) noexcept {
-    const float scaledWidth =
-        static_cast<float>(baseWidth) * HorizontalExpansion(surfaceWidth, surfaceHeight);
+inline uint32_t EggRecordWidth(uint32_t baseWidth, uint32_t surfaceWidth, uint32_t surfaceHeight,
+                               float referenceAspect = kPresentationAspect169) noexcept {
+    const float scaledWidth = static_cast<float>(baseWidth) *
+                              HorizontalExpansion(surfaceWidth, surfaceHeight, referenceAspect);
     return std::clamp<uint32_t>(
         static_cast<uint32_t>(std::lround(scaledWidth / 4.0f)) * 4u, 4u, 0xFFFCu);
 }

runtime/src/hle/input/pad.cpp

@@ -2,6 +2,7 @@
 #include "memory.h"
 #include "hle/controller_status_contract.h"
 #include "input_bindings.h"
+#include "motion_trick.h"
 
 #include <algorithm>
 #include <atomic>
@@ -116,6 +117,7 @@ extern "C" uint32_t PAD__Read_HLE(uint32_t statusPtr)
 
     FillTriggersHeldByButtons(statuses);
     InputBindings::Apply(statuses);
+    MotionTrick::Apply(statuses);
 
     try {
         for (uint32_t i = 0; i < PAD_CHANMAX; ++i) {

runtime/src/hle/sc.cpp

@@ -53,7 +53,7 @@ PPC_NATIVE_OVERRIDE(801B0220, SCCheckStatus_HLE, uint32_t, (), ());
 // Returns: 0 = 4:3, 1 = 16:9
 extern "C" uint32_t SCGetAspectRatio_HLE()
 {
-    return (RuntimeConfigFile::WidescreenEnabled(true) || MkwForceAspect169Requested()) ? 1u : 0u;
+    return RuntimeConfigFile::WidescreenEnabled(true) ? 1u : 0u;
 }
 
 PPC_NATIVE_OVERRIDE(801B1BE4, SCGetAspectRatio_HLE, uint32_t, (), ());

runtime/src/hle/storage/nand_isfs.cpp

@@ -436,7 +436,7 @@ extern "C" void NAND_IOS_OpenBody_HLE_801938FC(CpuContext* ctx) {
     ctx->gpr[3] = static_cast<uint32_t>(result);
     ctx->gpr[1] = ctx->gpr[1] + 32u;
 }
-REGISTER_NATIVE_FUNCTION_AS(0x801938FC, NAND_IOS_OpenBody_HLE_801938FC, "NAND_IOS_OpenBody_HLE_801938FC");
+REGISTER_NATIVE_REPLACEMENT_AS(0x801938FC, NAND_IOS_OpenBody_HLE_801938FC, "NAND_IOS_OpenBody_HLE_801938FC");
 
 extern "C" int32_t NAND_IOS_Close_HLE(uint32_t fd) {
     if (fd == ISFS_DEV_FD) {

runtime/src/main.cpp

@@ -41,6 +41,7 @@
 #include <mmsystem.h>
 #include <dbghelp.h>
 #else
+#include <fcntl.h>
 #include <signal.h>
 #if defined(__x86_64__)
 // Only the x86 POSIX fault path inspects ucontext_t to recover the page-fault
@@ -57,6 +58,7 @@
 #endif
 
 #include "abi_bridge.h"
+#include "audio_backend.h"
 #include "boot_race.h"
 #include "guest_abi/guest_region.h"
 #include "guest_flat_memory.h"
@@ -67,6 +69,7 @@
 #include "aurora_events.h"
 #include "wiimote_host.h"
 #include "discord_presence.h"
+#include "steam_input_layout.h"
 #include "fiber_manager.h"
 #include "gx_worker.h"
 #include "hle_stubs.h"
@@ -167,6 +170,7 @@ std::atomic_bool g_auroraInitialized{false};
 std::atomic_flag g_abortSignalHandled = ATOMIC_FLAG_INIT;
 std::atomic_bool g_fatalErrorReported{false};
 std::atomic_bool g_fatalPopupShown{false};
+bool g_headless = false;
 std::atomic<int> g_lastExitCode{0};
 std::atomic_bool g_exitCodeSet{false};
 
@@ -335,6 +339,9 @@ bool InstallTranscriptPipe(int& outReadFd, int& outWriteFd, int targetFd) {
     if (::pipe(pipeFds) != 0) {
         return false;
     }
+    // Spawned helpers must not keep the transcript open; its reader waits for EOF on shutdown.
+    ::fcntl(pipeFds[0], F_SETFD, FD_CLOEXEC);
+    ::fcntl(pipeFds[1], F_SETFD, FD_CLOEXEC);
 #endif
 
     outReadFd = pipeFds[0];
@@ -773,7 +780,7 @@ void MarkFatalErrorReported() {
 }
 
 void ShowRuntimeFatalPopup(std::string_view category, std::string_view details) noexcept {
-    if (g_fatalPopupShown.exchange(true, std::memory_order_acq_rel)) {
+    if (g_headless || g_fatalPopupShown.exchange(true, std::memory_order_acq_rel)) {
         return;
     }
 
@@ -1096,6 +1103,7 @@ LONG ReportFatalSehAndExit(EXCEPTION_POINTERS* info) {
                      << " at host address 0x" << record->ExceptionInformation[1];
     }
     popupDetails << ".\n\nThe process transcript and crash log contain the full CPU and stack diagnostics.";
+    SteamInputLayout::Release();
     ShowRuntimeFatalPopup("a native crash occurred", popupDetails.str());
     DumpHostStackTrace();
 
@@ -1196,6 +1204,7 @@ void PosixMemoryFaultHandler(int sig, siginfo_t* info, void* ucontextVoid) {
     }
     popupDetails << ".\n\nThe process transcript and crash log contain the full CPU and stack "
                     "diagnostics.";
+    SteamInputLayout::Release();
     ShowRuntimeFatalPopup("a native crash occurred", popupDetails.str());
     DumpHostStackTrace();
     WriteFatalLogImpl(sig == SIGBUS ? "sigbus" : "sigsegv");
@@ -1229,6 +1238,7 @@ void AbortSignalHandler(int signum) {
     // abort() bypasses atexit, so the guest-memory fault summary has to be
     // emitted here too. It is idempotent, so a later AtExitHandler is a no-op.
     GuestFlat::LogFaultSummary();
+    SteamInputLayout::Release();
 
     ShowRuntimeFatalPopup("a fatal internal error occurred",
                           "The process called abort while running the game or Aurora renderer.\n\n"
@@ -1264,6 +1274,7 @@ void SetRuntimeExitCode(int code) {
 
 // Global handler called via atexit() to flush buffers before any exit
 static void AtExitHandler() {
+    SteamInputLayout::Release();
     // End-of-run guest memory report. This runs before the fatal-report check
     // below because the counters describe the whole session and are just as
     // interesting after a crash as after a clean exit.
@@ -1288,6 +1299,7 @@ static void AtExitHandler() {
 
 // Global terminate handler for uncaught exceptions
 static void TerminateHandler() {
+    SteamInputLayout::Release();
     // Skip detailed dump if already reported
     if (g_fatalErrorReported.load(std::memory_order_acquire)) {
         std::fflush(stderr);
@@ -1345,15 +1357,25 @@ int RuntimeMain(int argc, char** argv) {
     
     std::string currentEntryLabel;
 
+    // `--headless`: with a boot race, run without a window, sound or dialogs.
+    g_headless = std::any_of(argv + 1, argv + argc,
+                             [](const char* arg) { return std::string_view(arg) == "--headless"; });
+
     try {
         for (int i = 1; i < argc; ++i) {
-            if (!BootRace::ParseArgument(argv[i])) {
+            if (std::string_view(argv[i]) != "--headless" && !BootRace::ParseArgument(argv[i])) {
                 throw std::invalid_argument(std::string("Unknown command-line option '") + argv[i] +
-                                            "'; the game runtime only accepts --race[=<course id>] and --race-cpu, use Config.toml through the installed host.");
+                                            "'; the game runtime only accepts --race[=<course id>], --race-cpu and --headless, use Config.toml through the installed host.");
             }
         }
+        if (g_headless && !BootRace::Requested()) {
+            throw std::invalid_argument("--headless needs --race[=<course id>] or --race-cpu");
+        }
+        if (g_headless) {
+            AudioBackend::Instance().SetSilenced(true);
+        }
         RuntimeConfigFile::LogLoadedConfig();
-        if (RuntimeConfigFile::DiscordPresenceEnabled()) {
+        if (!g_headless && RuntimeConfigFile::DiscordPresenceEnabled()) {
             DiscordPresence::Initialize(RuntimeConfigFile::DiscordClientId(), "Mario Kart Wii");
         }
         SystemBridge::Initialize();
@@ -1379,15 +1401,17 @@ int RuntimeMain(int argc, char** argv) {
         auroraConfig.cachePath = auroraCachePath.c_str();
         auroraConfig.logCallback = &RuntimeAuroraLogCallback;
         auroraConfig.logLevel = LOG_DEBUG;
+        RuntimeConfigFile::MigrateLegacyForceAspect();
         const bool configWidescreen = RuntimeConfigFile::WidescreenEnabled(true);
-        const bool forceAspect169 = RuntimeConfigFile::ForceAspect169Enabled();
-        auroraConfig.windowWidth = (configWidescreen || forceAspect169) ? 854 : 640;
+        const bool forceAspect = RuntimeConfigFile::ForceAspectEnabled();
+        auroraConfig.windowWidth = configWidescreen ? 854 : 640;
         auroraConfig.windowHeight = 480;
         auroraConfig.windowWidth = RuntimeConfigFile::WindowWidth(auroraConfig.windowWidth);
         auroraConfig.windowHeight = RuntimeConfigFile::WindowHeight(auroraConfig.windowHeight);
         auroraConfig.hasWindowPosition = RuntimeConfigFile::WindowPosition(
             auroraConfig.windowPosX, auroraConfig.windowPosY);
-        auroraConfig.allowJoystickBackgroundEvents = true;
+        auroraConfig.allowJoystickBackgroundEvents = !g_headless;
+        auroraConfig.startHidden = g_headless;
         auroraConfig.disableCopyFilter = RuntimeConfigFile::DisableCopyFilter(true);
         // Dolphin-style custom textures. Aurora indexes <userPath>/texture_replacements
         // once during aurora_initialize, so both knobs only take effect on the next launch.
@@ -1399,7 +1423,7 @@ int RuntimeMain(int argc, char** argv) {
         auroraConfig.vsync = RuntimeConfigFile::Vsync();
         auroraConfig.desiredBackend = BACKEND_AUTO;
         const float resolutionMultiplier = RuntimeConfigFile::ResolutionMultiplier(1.0f);
-        ConfigureMkwDynamicAspect(configWidescreen, forceAspect169,
+        ConfigureMkwDynamicAspect(configWidescreen, forceAspect,
                                   auroraConfig.windowWidth, auroraConfig.windowHeight);
         VISetFrameBufferScale(resolutionMultiplier);
         // One table for both directions. RuntimeConfigFile::IsSupportedGraphicsApi
@@ -1466,16 +1490,23 @@ int RuntimeMain(int argc, char** argv) {
         UpdateMkwDynamicAspectSurface(auroraInfo.windowSize.native_fb_width,
                                       auroraInfo.windowSize.native_fb_height);
         settings_overlay::InitializeRuntimeSettings();
+        if (!g_headless) {
+            SteamInputLayout::Initialize(RuntimeConfigFile::SteamInputOverrideEnabled(),
+                                         RuntimeConfigFile::SteamInputAppId());
+        }
+        if (BootRace::Requested()) {
+            aurora_set_skip_unready_pipelines(false);
+        }
         RT_LOG(RT_TAG_CONFIG) << "video.widescreen=" << (configWidescreen ? "true" : "false")
-                  << " force_16_9=" << (forceAspect169 ? "true" : "false")
-                  << " SCGetAspectRatio=" << (configWidescreen || forceAspect169 ? 1 : 0)
+                  << " force_aspect=" << (forceAspect ? "true" : "false")
+                  << " SCGetAspectRatio=" << (configWidescreen ? 1 : 0)
                   << " resolutionMultiplier=" << resolutionMultiplier
                   << " window=" << auroraInfo.windowSize.width << "x" << auroraInfo.windowSize.height
                   << " native=" << auroraInfo.windowSize.native_fb_width << "x"
                   << auroraInfo.windowSize.native_fb_height
-                  << " viewportPolicy=" << (forceAspect169 ? "16:9" : (configWidescreen ? "stretch" : "fit"))
+                  << " viewportPolicy=" << (forceAspect ? (configWidescreen ? "16:9" : "fit") : "stretch")
                   << " presentAspect="
-                  << (forceAspect169 ? "16:9" : (configWidescreen ? "surface (dynamic EGG canvas)" : "4:3"))
+                  << (forceAspect ? (configWidescreen ? "16:9" : "4:3") : "surface (dynamic EGG canvas)")
                   << std::endl;
         g_auroraInitialized.store(true, std::memory_order_release);
         GxWorker::SetWaitCallback(ServiceGuestTimingDuringAuroraFrameWait);
@@ -1508,6 +1539,7 @@ int RuntimeMain(int argc, char** argv) {
         Fiber::GuestFiberManager::Shutdown();
         WindowPlacementPersistence::Flush(true);
         Wiimote::Shutdown();
+        SteamInputLayout::Release();
         GxWorker::Stop();
         aurora_shutdown();
         DiscordPresence::Shutdown();
@@ -1522,6 +1554,7 @@ int RuntimeMain(int argc, char** argv) {
                 << " len=0x" << ex.length()
                 << std::dec << std::nouppercase
                 << " reason=" << ex.reason();
+        SteamInputLayout::Release();
         ShowRuntimeFatalPopup("a guest memory access was out of bounds", details.str());
         WriteFatalLogImpl("access_violation", details.str());
         SetRuntimeExitCodeImpl(1);
@@ -1536,6 +1569,7 @@ int RuntimeMain(int argc, char** argv) {
     } catch (const std::exception& ex) {
         std::cerr << "Runtime error: " << ex.what() << std::endl;
         SystemBridge::DumpCpuState(TryGetCpuContext());
+        SteamInputLayout::Release();
         ShowRuntimeFatalPopup("a runtime exception occurred", ex.what());
         WriteFatalLogImpl("exception");
         SetRuntimeExitCodeImpl(1);

runtime/src/motion_trick.cpp

@@ -0,0 +1,227 @@
+#include "motion_trick.h"
+
+#include "input_bindings.h"
+#include "runtime_config.h"
+
+#include <algorithm>
+#include <array>
+#include <atomic>
+#include <cstring>
+#include <mutex>
+
+#include <SDL3/SDL_gamepad.h>
+#include <SDL3/SDL_keyboard.h>
+#include <SDL3/SDL_timer.h>
+
+namespace MotionTrick {
+namespace {
+
+// Long enough for the game to see the press, short enough not to hold a trick.
+constexpr int kPulseFrames = 4;
+// Sensors report zeros or stale data for a moment after being switched on.
+constexpr int kWarmupFrames = 6;
+constexpr float kIntensityDecay = 0.85f;
+// Real sensors always jitter; identical readings mean the stream was switched
+// off behind our back, so ask for it again a few times.
+constexpr int kStaleFrames = 30;
+constexpr uint64_t kResyncIntervalUs = 2'000'000;
+constexpr int kMaxResyncs = 3;
+
+struct PortState {
+    SDL_JoystickID id = 0;
+    bool gyro = false;
+    bool accel = false;
+    int warmup = 0;
+    MotionSample previous;
+    int staleFrames = 0;
+    int resyncs = 0;
+    uint64_t lastResyncUs = 0;
+    ShakeDetector detector;
+    float displayIntensity = 0.0f;
+    Direction pulse = Direction::None;
+    int pulseFrames = 0;
+    Direction lastTrick = Direction::None;
+    uint64_t lastTrickUs = 0;
+};
+
+std::atomic<bool> g_enabled{RuntimeConfigFile::MotionTrickEnabled(false)};
+std::atomic<int> g_sensitivity{RuntimeConfigFile::MotionTrickSensitivity(kDefaultSensitivity)};
+
+std::mutex g_mutex;
+std::array<PortState, PAD_CHANMAX> g_ports;
+
+// A keyboard player keeps their controller assigned, but aurora ignores it.
+SDL_Gamepad* GamepadForPort(uint32_t port) {
+    uint32_t keyCount = 0;
+    if (PADGetKeyButtonBindings(port, &keyCount) != nullptr) {
+        return nullptr;
+    }
+    const s32 index = PADGetIndexForPort(port);
+    return index < 0 ? nullptr : PADGetSDLGamepadForIndex(static_cast<u32>(index));
+}
+
+uint16_t ButtonFor(Direction direction) {
+    switch (direction) {
+    case Direction::Up: return PAD_BUTTON_UP;
+    case Direction::Down: return PAD_BUTTON_DOWN;
+    case Direction::Left: return PAD_BUTTON_LEFT;
+    case Direction::Right: return PAD_BUTTON_RIGHT;
+    default: return 0;
+    }
+}
+
+bool HeldByOtherPort(uint32_t port, SDL_JoystickID id) {
+    for (uint32_t other = 0; other < g_ports.size(); ++other) {
+        if (other != port && g_ports[other].id == id) {
+            return true;
+        }
+    }
+    return false;
+}
+
+void SetSensors(SDL_Gamepad* gamepad, const PortState& state, bool enabled) {
+    if (state.gyro) SDL_SetGamepadSensorEnabled(gamepad, SDL_SENSOR_GYRO, enabled);
+    if (state.accel) SDL_SetGamepadSensorEnabled(gamepad, SDL_SENSOR_ACCEL, enabled);
+}
+
+void Attach(uint32_t port, SDL_Gamepad* gamepad) {
+    PortState& state = g_ports[port];
+    const SDL_JoystickID id = gamepad != nullptr ? SDL_GetGamepadID(gamepad) : 0;
+    if (id == state.id) {
+        return;
+    }
+    if (state.id != 0 && !HeldByOtherPort(port, state.id)) {
+        if (SDL_Gamepad* previous = SDL_GetGamepadFromID(state.id)) {
+            SetSensors(previous, state, false);
+        }
+    }
+    state = PortState{};
+    if (gamepad == nullptr) {
+        return;
+    }
+    state.id = id;
+    state.gyro = SDL_GamepadHasSensor(gamepad, SDL_SENSOR_GYRO) &&
+                 SDL_SetGamepadSensorEnabled(gamepad, SDL_SENSOR_GYRO, true);
+    state.accel = SDL_GamepadHasSensor(gamepad, SDL_SENSOR_ACCEL) &&
+                  SDL_SetGamepadSensorEnabled(gamepad, SDL_SENSOR_ACCEL, true);
+    state.warmup = kWarmupFrames;
+}
+
+bool SameReading(const MotionSample& a, const MotionSample& b) {
+    return a.hasGyro == b.hasGyro && a.hasAccel == b.hasAccel &&
+           std::memcmp(a.gyro, b.gyro, sizeof(a.gyro)) == 0 && std::memcmp(a.accel, b.accel, sizeof(a.accel)) == 0;
+}
+
+// Returns true when the sensors were restarted.
+bool ResyncIfStale(SDL_Gamepad* gamepad, PortState& state, const MotionSample& sample, uint64_t now) {
+    const bool stale = SameReading(sample, state.previous) && SDL_GetKeyboardFocus() != nullptr;
+    state.previous = sample;
+    state.staleFrames = stale ? state.staleFrames + 1 : 0;
+    if (!stale) {
+        state.resyncs = 0;
+    }
+    if (state.staleFrames < kStaleFrames || state.resyncs >= kMaxResyncs ||
+        now - state.lastResyncUs < kResyncIntervalUs) {
+        return false;
+    }
+    SetSensors(gamepad, state, false);
+    SetSensors(gamepad, state, true);
+    state.staleFrames = 0;
+    ++state.resyncs;
+    state.lastResyncUs = now;
+    state.warmup = kWarmupFrames;
+    return true;
+}
+
+MotionSample ReadSample(SDL_Gamepad* gamepad, const PortState& state) {
+    MotionSample sample;
+    sample.hasGyro = state.gyro && SDL_GetGamepadSensorData(gamepad, SDL_SENSOR_GYRO, sample.gyro, 3);
+    sample.hasAccel = state.accel && SDL_GetGamepadSensorData(gamepad, SDL_SENSOR_ACCEL, sample.accel, 3) &&
+                      (sample.accel[0] != 0.0f || sample.accel[1] != 0.0f || sample.accel[2] != 0.0f);
+    return sample;
+}
+
+} // namespace
+
+bool Enabled() noexcept {
+    return g_enabled.load(std::memory_order_relaxed);
+}
+
+void SetEnabled(bool enabled) noexcept {
+    g_enabled.store(enabled, std::memory_order_relaxed);
+    RuntimeConfigFile::SetMotionTrickEnabled(enabled);
+}
+
+int Sensitivity() noexcept {
+    return g_sensitivity.load(std::memory_order_relaxed);
+}
+
+void SetSensitivity(int level) noexcept {
+    level = std::clamp(level, kMinSensitivity, kMaxSensitivity);
+    g_sensitivity.store(level, std::memory_order_relaxed);
+    RuntimeConfigFile::SetMotionTrickSensitivity(level);
+}
+
+bool PortHasMotion(uint32_t port) noexcept {
+    SDL_Gamepad* gamepad = GamepadForPort(port);
+    return gamepad != nullptr &&
+           (SDL_GamepadHasSensor(gamepad, SDL_SENSOR_GYRO) || SDL_GamepadHasSensor(gamepad, SDL_SENSOR_ACCEL));
+}
+
+uint64_t NowUs() noexcept {
+    return SDL_GetTicksNS() / 1000;
+}
+
+void Apply(PADStatus* statuses) noexcept {
+    if (statuses == nullptr) {
+        return;
+    }
+    const bool enabled = Enabled();
+    const int sensitivity = Sensitivity();
+    const bool blocked = InputBindings::InputBlocked();
+    const uint64_t now = NowUs();
+
+    std::lock_guard<std::mutex> lock(g_mutex);
+    for (uint32_t port = 0; port < PAD_CHANMAX; ++port) {
+        SDL_Gamepad* gamepad =
+            enabled && statuses[port].err == PAD_ERR_NONE ? GamepadForPort(port) : nullptr;
+        Attach(port, gamepad);
+        PortState& state = g_ports[port];
+        if (gamepad == nullptr || (!state.gyro && !state.accel)) {
+            continue;
+        }
+        if (state.warmup > 0) {
+            --state.warmup;
+        } else if (const MotionSample sample = ReadSample(gamepad, state);
+                   !ResyncIfStale(gamepad, state, sample, now)) {
+            state.detector.SetSensitivity(sensitivity);
+            const Direction fired = state.detector.Update(sample, now);
+            state.displayIntensity =
+                std::max(state.detector.Intensity(), state.displayIntensity * kIntensityDecay);
+            if (fired != Direction::None) {
+                state.lastTrick = fired;
+                state.lastTrickUs = now;
+                state.pulse = fired;
+                state.pulseFrames = kPulseFrames;
+            }
+        }
+        if (blocked) {
+            state.pulseFrames = 0;
+        }
+        if (state.pulseFrames > 0) {
+            statuses[port].button |= ButtonFor(state.pulse);
+            --state.pulseFrames;
+        }
+    }
+}
+
+PortStatus Status(uint32_t port) noexcept {
+    if (port >= PAD_CHANMAX) {
+        return {};
+    }
+    std::lock_guard<std::mutex> lock(g_mutex);
+    const PortState& state = g_ports[port];
+    return {state.displayIntensity, state.lastTrick, state.lastTrickUs, state.id != 0 && !state.gyro && !state.accel};
+}
+
+} // namespace MotionTrick

runtime/src/motion_trick_detector.cpp

@@ -0,0 +1,80 @@
+#include "motion_trick_detector.h"
+
+#include <algorithm>
+#include <array>
+#include <cmath>
+#include <cstdint>
+#include <cstring>
+
+namespace MotionTrick {
+namespace {
+
+constexpr float kGravity = 9.80665f;
+constexpr float kRearmIntensity = 0.35f;
+constexpr uint64_t kCooldownUs = 400'000;
+
+// Indexed by sensitivity level - 1.
+constexpr std::array<float, kMaxSensitivity> kGyroThresholds = {11.0f, 9.0f, 7.0f, 5.5f, 4.0f};
+constexpr std::array<float, kMaxSensitivity> kAccelThresholds = {20.0f, 16.0f, 13.0f, 10.0f, 8.0f};
+
+// The runtime builds with -ffast-math, which may fold std::isfinite away.
+bool Finite(const float (&v)[3]) {
+    for (const float value : v) {
+        uint32_t bits = 0;
+        std::memcpy(&bits, &value, sizeof(bits));
+        if ((bits & 0x7f800000u) == 0x7f800000u) {
+            return false;
+        }
+    }
+    return true;
+}
+
+float Length(const float (&v)[3]) {
+    return std::sqrt(v[0] * v[0] + v[1] * v[1] + v[2] * v[2]);
+}
+
+Direction GyroDirection(const float (&gyro)[3]) {
+    const float pitch = gyro[0];
+    const float yaw = gyro[1];
+    const float roll = gyro[2];
+    if (std::fabs(pitch) >= std::max(std::fabs(yaw), std::fabs(roll))) {
+        return pitch >= 0.0f ? Direction::Up : Direction::Down;
+    }
+    const float lateral = std::fabs(roll) >= std::fabs(yaw) ? roll : yaw;
+    return lateral >= 0.0f ? Direction::Left : Direction::Right;
+}
+
+} // namespace
+
+void ShakeDetector::SetSensitivity(int level) {
+    const size_t index = static_cast<size_t>(std::clamp(level, kMinSensitivity, kMaxSensitivity) - 1);
+    m_gyroThreshold = kGyroThresholds[index];
+    m_accelThreshold = kAccelThresholds[index];
+}
+
+Direction ShakeDetector::Update(const MotionSample& sample, uint64_t nowUs) {
+    // The accelerometer only stands in for a missing gyro: on a gyro pad it
+    // mostly picks up bumps, and it carries no direction.
+    const float(&values)[3] = sample.hasGyro ? sample.gyro : sample.accel;
+    if ((!sample.hasGyro && !sample.hasAccel) || !Finite(values)) {
+        m_intensity = 0.0f;
+        return Direction::None;
+    }
+    m_intensity = sample.hasGyro ? Length(sample.gyro) / m_gyroThreshold
+                                 : std::fabs(Length(sample.accel) - kGravity) / m_accelThreshold;
+
+    if (!m_armed) {
+        if (m_intensity < kRearmIntensity && nowUs - m_lastFireUs >= kCooldownUs) {
+            m_armed = true;
+        }
+        return Direction::None;
+    }
+    if (m_intensity < 1.0f) {
+        return Direction::None;
+    }
+    m_armed = false;
+    m_lastFireUs = nowUs;
+    return sample.hasGyro ? GyroDirection(sample.gyro) : Direction::Up;
+}
+
+} // namespace MotionTrick

runtime/src/settings_overlay.cpp

@@ -5,6 +5,7 @@
 #include "controller_button_names.h"
 #include "controller_mapping_wizard.h"
 #include "input_bindings.h"
+#include "motion_trick.h"
 #include "game_graphics_options.h"
 #include "music_attenuation.h"
 #include "runtime_config.h"
@@ -113,7 +114,8 @@ bool g_skipUnreadyPipelines = RuntimeConfigFile::SkipUnreadyPipelines(true);
 bool g_vsync = RuntimeConfigFile::Vsync();
 bool g_disableCopyFilter = RuntimeConfigFile::DisableCopyFilter(true);
 bool g_showFps = RuntimeConfigFile::ShowFps(true);
-bool g_forceAspect169 = RuntimeConfigFile::ForceAspect169Enabled();
+bool g_forceAspect = RuntimeConfigFile::ForceAspectEnabled();
+bool g_widescreen = RuntimeConfigFile::WidescreenEnabled(true);
 #if defined(__APPLE__)
 bool g_metalFxSpatialUpscaling = RuntimeConfigFile::MetalFxSpatialUpscaling(false);
 #endif
@@ -744,6 +746,69 @@ void DrawRumbleSettings() {
     }
 }
 
+const char* TrickDirectionLabel(MotionTrick::Direction direction) {
+    switch (direction) {
+    case MotionTrick::Direction::Up: return "up";
+    case MotionTrick::Direction::Down: return "down";
+    case MotionTrick::Direction::Left: return "left";
+    case MotionTrick::Direction::Right: return "right";
+    default: return "";
+    }
+}
+
+void DrawMotionTrickSettings(uint32_t port) {
+    const MotionTrick::PortStatus status = MotionTrick::Status(port);
+    const bool hasMotion = MotionTrick::PortHasMotion(port) && !status.sensorsUnavailable;
+    bool enabled = MotionTrick::Enabled();
+    ImGui::BeginDisabled(!hasMotion && !enabled);
+    if (ImGui::Checkbox("Shake to trick", &enabled)) {
+        MotionTrick::SetEnabled(enabled);
+    }
+    ImGui::EndDisabled();
+    if (ImGui::IsItemHovered(ImGuiHoveredFlags_AllowWhenDisabled)) {
+        ImGui::SetTooltip(hasMotion
+            ? "Flick the controller to trick off ramps, or to wheelie on a bike, like shaking a Wii Remote.\n"
+              "The flick direction picks the trick. The D-pad keeps working. Applies to every player."
+            : "Steam Input can hide a controller's motion sensor; turn it off for this game to use it here.");
+    }
+    if (!hasMotion) {
+        ImGui::SameLine();
+        ImGui::TextDisabled("(no motion sensor)");
+        return;
+    }
+    if (!enabled) {
+        return;
+    }
+
+    ImGui::Indent();
+    int sensitivity = MotionTrick::Sensitivity();
+    ImGui::SetNextItemWidth(160.0f);
+    if (ImGui::SliderInt("Sensitivity##motion", &sensitivity, MotionTrick::kMinSensitivity,
+                         MotionTrick::kMaxSensitivity, "%d", ImGuiSliderFlags_AlwaysClamp)) {
+        MotionTrick::SetSensitivity(sensitivity);
+    }
+    if (ImGui::IsItemHovered()) {
+        ImGui::SetTooltip("Higher reacts to smaller flicks. Lower it if you trick or wheelie by accident while steering.");
+    }
+
+    constexpr float kMeterScale = 1.5f;
+    const ImVec2 meterPos = ImGui::GetCursorScreenPos();
+    const ImVec2 meterSize(160.0f, ImGui::GetFrameHeight() * 0.6f);
+    ImGui::ProgressBar(std::min(status.intensity / kMeterScale, 1.0f), meterSize, "");
+    const float thresholdX = meterPos.x + meterSize.x / kMeterScale;
+    ImGui::GetWindowDrawList()->AddLine(ImVec2(thresholdX, meterPos.y), ImVec2(thresholdX, meterPos.y + meterSize.y),
+                                        ImGui::GetColorU32(ImGuiCol_Text), 2.0f);
+    ImGui::SameLine();
+    const bool recent = status.lastTrick != MotionTrick::Direction::None &&
+                        MotionTrick::NowUs() - status.lastTrickUs < 700'000;
+    if (recent) {
+        ImGui::TextColored(ImVec4(0.4f, 0.9f, 0.4f, 1.0f), "Trick! (%s)", TrickDirectionLabel(status.lastTrick));
+    } else {
+        ImGui::TextDisabled("Flick to test");
+    }
+    ImGui::Unindent();
+}
+
 void DrawGamepadMapping(uint32_t port) {
     uint32_t mappingCount = 0;
     PADButtonMapping* mappings = PADGetButtonMappings(port, &mappingCount);
@@ -984,6 +1049,7 @@ void DrawControllerSettings() {
         }
     } else {
         ImGui::TextUnformatted("Ready to play.");
+        DrawMotionTrickSettings(port);
         if (ImGui::CollapsingHeader("Customize buttons")) {
             ImGui::Indent();
             DrawGamepadMapping(port);
@@ -1080,11 +1146,22 @@ void DrawAudioSettings() {
 
 void DrawGraphicsSettings() {
     g_displayMode = static_cast<int>(aurora_get_display_mode());
-    if (ImGui::Checkbox("Force 16:9", &g_forceAspect169)) {
-        SetMkwForceAspect169(g_forceAspect169);
-        RuntimeConfigFile::SetForceAspect169(g_forceAspect169);
+    if (ImGui::BeginCombo("Aspect ratio (restart required)", g_widescreen ? "16:9" : "4:3")) {
+        if (ImGui::Selectable("16:9", g_widescreen)) {
+            g_widescreen = true;
+            RuntimeConfigFile::SetWidescreen(true);
+        }
+        if (ImGui::Selectable("4:3", !g_widescreen)) {
+            g_widescreen = false;
+            RuntimeConfigFile::SetWidescreen(false);
+        }
+        ImGui::EndCombo();
     }
-    ImGui::TextDisabled("Keep a 16:9 image with black bars when the window has another shape.");
+    if (ImGui::Checkbox("Force aspect ratio", &g_forceAspect)) {
+        SetMkwForceAspect(g_forceAspect);
+        RuntimeConfigFile::SetForceAspect(g_forceAspect);
+    }
+    ImGui::TextDisabled("Keep the current aspect ratio with black bars when the window has another shape.");
     ImGui::Separator();
     struct EffectFlag {
         const char* label;
@@ -1470,6 +1547,18 @@ void DrawTopBar() {
     ImGui::EndMainMenuBar();
 }
 
+// Steam's desktop layout maps B to Escape on Valve controllers.
+bool ValveControllerConnected() {
+    int count = 0;
+    SDL_JoystickID* ids = SDL_GetJoysticks(&count);
+    bool found = false;
+    for (int i = 0; ids != nullptr && i < count && !found; ++i) {
+        found = SDL_GetJoystickVendorForID(ids[i]) == 0x28DE && SDL_GetJoystickProductForID(ids[i]) != 0x11FF;
+    }
+    SDL_free(ids);
+    return found;
+}
+
 bool IsToggleKey(const SDL_Event& event, SDL_Scancode code) {
     return event.type == SDL_EVENT_KEY_DOWN && !event.key.repeat && event.key.scancode == code;
 }
@@ -1591,7 +1680,7 @@ void HandleEvents(const AuroraEvent* events) noexcept {
                 g_exitPromptOpen = false;
             } else if (g_topBarVisible) {
                 SetTopBarVisible(false);
-            } else {
+            } else if (!ValveControllerConnected()) {
                 g_exitPromptOpen = true;
             }
             ApplyInputBlockState();

runtime/src/steam_input_layout.cpp

@@ -0,0 +1,603 @@
+#include "steam_input_layout.h"
+
+#include "runtime_log.h"
+
+#include <aurora/aurora.h>
+#include <SDL3/SDL_video.h>
+
+#include <atomic>
+#include <chrono>
+#include <condition_variable>
+#include <cstdlib>
+#include <memory>
+#include <mutex>
+#include <string>
+#include <system_error>
+#include <thread>
+#include <utility>
+#include <vector>
+
+#if defined(_WIN32)
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif
+#include <windows.h>
+
+#include <cwchar>
+#include <filesystem>
+#include <optional>
+#else
+#include <cerrno>
+#include <csignal>
+#include <fcntl.h>
+#include <spawn.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#if defined(__APPLE__)
+#include <libproc.h>
+#include <sys/proc.h>
+#include <sys/proc_info.h>
+#else
+#include <dirent.h>
+#include <fstream>
+#include <sstream>
+#endif
+
+extern char** environ;
+#endif
+
+namespace SteamInputLayout {
+namespace {
+
+using Clock = std::chrono::steady_clock;
+
+enum class Handoff { None, Unconfirmed, Confirmed };
+
+constexpr std::chrono::milliseconds kHandoffTimeout{2000};
+#if defined(__linux__)
+// xdg-open can return before the Steam forwarder it started has delivered the URL.
+constexpr std::chrono::milliseconds kDeliverySpacing{1500};
+#else
+constexpr std::chrono::milliseconds kDeliverySpacing{0};
+#endif
+
+struct State {
+    std::mutex mutex;
+    std::condition_variable wake;
+    // Serializes requests to Steam so they arrive in order.
+    std::timed_mutex delivery;
+    bool enabled = false;
+    bool stopping = false;
+    bool desired = false;
+    bool sent = false;
+    bool maybeForced = false;
+    uint32_t appId = 0;
+    SDL_WindowID windowId = 0;
+    std::atomic<Clock::rep> lastDelivery{0};
+    std::atomic<bool> released{false};
+};
+
+// Never destroyed: the detached worker and late exit paths may still touch it.
+std::atomic<State*> g_state{nullptr};
+thread_local bool t_isWorker = false;
+thread_local bool t_inRelease = false;
+
+// The worker launches and publishes its request helper under this lock, so an exit that
+// can't wait for the worker can cancel it: either the helper is published, or it never starts.
+std::mutex g_helperMutex;
+std::atomic<bool> g_helperCancelled{false};
+#if defined(_WIN32)
+HANDLE g_helper = nullptr;
+#else
+pid_t g_helper = 0;
+#endif
+
+// These locks are only held briefly; failing to get one means a crash inside it.
+bool TryLockBriefly(std::unique_lock<std::mutex>& lock,
+                    std::chrono::milliseconds timeout = std::chrono::milliseconds(250)) {
+    const auto deadline = Clock::now() + timeout;
+    while (!lock.try_lock()) {
+        if (Clock::now() >= deadline) {
+            return false;
+        }
+        std::this_thread::sleep_for(std::chrono::milliseconds(1));
+    }
+    return true;
+}
+
+// Steam applies the launching app's own layout to everything it starts, children included.
+bool LaunchedBySteam() {
+    for (const char* name : {"SteamGameId", "SteamAppId", "SteamClientLaunch"}) {
+        const char* value = std::getenv(name);
+        if (value != nullptr && *value != '\0') {
+            return true;
+        }
+    }
+#if defined(_WIN32)
+    if (GetModuleHandleW(L"GameOverlayRenderer64.dll") != nullptr) {
+        return true;
+    }
+#endif
+    return false;
+}
+
+#if defined(_WIN32)
+
+std::vector<BYTE> TokenUserOf(HANDLE process) {
+    std::vector<BYTE> user;
+    HANDLE token = nullptr;
+    if (!OpenProcessToken(process, TOKEN_QUERY, &token)) {
+        return user;
+    }
+    DWORD size = 0;
+    GetTokenInformation(token, TokenUser, nullptr, 0, &size);
+    user.resize(size);
+    if (size == 0 || !GetTokenInformation(token, TokenUser, user.data(), size, &size)) {
+        user.clear();
+    }
+    CloseHandle(token);
+    return user;
+}
+
+// A stale pid can be reused by another signed-in user's Steam.
+bool OwnedByCurrentUser(HANDLE process) {
+    const auto theirs = TokenUserOf(process);
+    const auto ours = TokenUserOf(GetCurrentProcess());
+    return !theirs.empty() && !ours.empty() &&
+           EqualSid(reinterpret_cast<const TOKEN_USER*>(theirs.data())->User.Sid,
+                    reinterpret_cast<const TOKEN_USER*>(ours.data())->User.Sid);
+}
+
+// ActiveProcess\pid outlives a Steam crash, so the process behind it is verified too.
+std::optional<std::wstring> RunningSteamExecutable() {
+    DWORD pid = 0;
+    DWORD size = sizeof(pid);
+    if (RegGetValueW(HKEY_CURRENT_USER, L"Software\\Valve\\Steam\\ActiveProcess", L"pid",
+                     RRF_RT_REG_DWORD, nullptr, &pid, &size) != ERROR_SUCCESS ||
+        pid == 0) {
+        return std::nullopt;
+    }
+    HANDLE process = OpenProcess(PROCESS_QUERY_LIMITED_INFORMATION, FALSE, pid);
+    if (process == nullptr) {
+        return std::nullopt;
+    }
+    std::wstring path(32768, L'\0');
+    DWORD length = static_cast<DWORD>(path.size());
+    DWORD exitCode = 0;
+    const bool alive = QueryFullProcessImageNameW(process, 0, path.data(), &length) &&
+                       GetExitCodeProcess(process, &exitCode) && exitCode == STILL_ACTIVE &&
+                       OwnedByCurrentUser(process);
+    CloseHandle(process);
+    if (!alive) {
+        return std::nullopt;
+    }
+    path.resize(length);
+    if (_wcsicmp(std::filesystem::path(path).filename().c_str(), L"steam.exe") != 0) {
+        return std::nullopt;
+    }
+    return path;
+}
+
+// Same command line as Steam's registered steam:// handler, sent to the verified client.
+Handoff OpenSteamUrl(const std::string& url, bool bounded) {
+    const auto steam = RunningSteamExecutable();
+    if (!steam) {
+        return Handoff::None;
+    }
+    std::wstring commandLine = L"\"" + *steam + L"\" -- \"" + std::wstring(url.begin(), url.end()) + L"\"";
+    const std::wstring directory = std::filesystem::path(*steam).parent_path().wstring();
+    STARTUPINFOW startup{};
+    startup.cb = sizeof(startup);
+    PROCESS_INFORMATION info{};
+    std::unique_lock launch(g_helperMutex, std::defer_lock);
+    if (!bounded) {
+        launch.lock();
+        if (g_helperCancelled) {
+            return Handoff::None;
+        }
+    }
+    // The worker's helper starts suspended so a cancellation that couldn't take the lock while
+    // CreateProcessW ran still stops it before it can reach Steam.
+    if (!CreateProcessW(steam->c_str(), commandLine.data(), nullptr, nullptr, FALSE,
+                        bounded ? 0 : CREATE_SUSPENDED, nullptr, directory.c_str(), &startup, &info)) {
+        return Handoff::None;
+    }
+    if (!bounded) {
+        g_helper = info.hProcess;
+        launch.unlock();
+        if (g_helperCancelled) {
+            TerminateProcess(info.hProcess, 1);
+        } else {
+            ResumeThread(info.hThread);
+        }
+    }
+    CloseHandle(info.hThread);
+    const DWORD waited = WaitForSingleObject(info.hProcess, bounded ? static_cast<DWORD>(kHandoffTimeout.count()) : INFINITE);
+    if (!bounded) {
+        std::lock_guard lock(g_helperMutex);
+        g_helper = nullptr;
+    }
+    DWORD exitCode = 1;
+    const bool succeeded = waited == WAIT_OBJECT_0 && GetExitCodeProcess(info.hProcess, &exitCode) && exitCode == 0;
+    CloseHandle(info.hProcess);
+    return succeeded ? Handoff::Confirmed : Handoff::Unconfirmed;
+}
+
+void CancelWorkerRequests() {
+    g_helperCancelled = true;
+    std::unique_lock lock(g_helperMutex, std::defer_lock);
+    if (!TryLockBriefly(lock, kHandoffTimeout)) {
+        return;
+    }
+    if (g_helper != nullptr && TerminateProcess(g_helper, 1)) {
+        WaitForSingleObject(g_helper, static_cast<DWORD>(kHandoffTimeout.count()));
+    }
+}
+
+#else
+
+// Only a live client of this user; a URL handler with no client running would start Steam.
+bool SteamRunning() {
+    const uid_t uid = getuid();
+#if defined(__APPLE__)
+    const int count = proc_listallpids(nullptr, 0);
+    if (count <= 0) {
+        return false;
+    }
+    std::vector<pid_t> pids(static_cast<size_t>(count) + 32);
+    const int filled = proc_listallpids(pids.data(), static_cast<int>(pids.size() * sizeof(pid_t)));
+    for (int i = 0; i < filled; ++i) {
+        char name[2 * MAXCOMLEN + 1] = {};
+        if (proc_name(pids[i], name, sizeof(name)) <= 0 || std::string(name) != "steam_osx") {
+            continue;
+        }
+        proc_bsdinfo info{};
+        if (proc_pidinfo(pids[i], PROC_PIDTBSDINFO, 0, &info, sizeof(info)) == sizeof(info) &&
+            info.pbi_uid == uid && info.pbi_status != SZOMB) {
+            return true;
+        }
+    }
+    return false;
+#else
+    DIR* proc = opendir("/proc");
+    if (proc == nullptr) {
+        return false;
+    }
+    bool found = false;
+    while (const dirent* entry = readdir(proc)) {
+        if (entry->d_name[0] < '1' || entry->d_name[0] > '9') {
+            continue;
+        }
+        std::ifstream status(std::string("/proc/") + entry->d_name + "/status");
+        std::string line;
+        bool named = false;
+        bool live = false;
+        bool owned = false;
+        while (std::getline(status, line)) {
+            if (line.rfind("Name:", 0) == 0) {
+                named = line == "Name:\tsteam";
+            } else if (line.rfind("State:", 0) == 0) {
+                live = line.find_first_of("ZX") == std::string::npos;
+            } else if (line.rfind("Uid:", 0) == 0) {
+                std::istringstream fields(line.substr(4));
+                uid_t real = 0;
+                owned = static_cast<bool>(fields >> real) && real == uid;
+                break;
+            }
+        }
+        if (named && live && owned) {
+            found = true;
+            break;
+        }
+    }
+    closedir(proc);
+    return found;
+#endif
+}
+
+Handoff OpenSteamUrl(const std::string& url, bool bounded) {
+    if (!SteamRunning()) {
+        return Handoff::None;
+    }
+    // The worker's helper waits for a go-ahead on fd 3 before it runs the URL handler, so a
+    // cancellation that couldn't take the lock while posix_spawn ran still stops it in time.
+    static char gateScript[] = "read -r go <&3 || exit 1; exec 3<&-; exec \"$0\" \"$@\"";
+#if defined(__APPLE__)
+    const char* program = "/usr/bin/open";
+    char* argv[] = {const_cast<char*>("open"), const_cast<char*>("-g"), const_cast<char*>(url.c_str()), nullptr};
+    char* gated[] = {const_cast<char*>("sh"), const_cast<char*>("-c"), gateScript, const_cast<char*>(program),
+                     const_cast<char*>("-g"), const_cast<char*>(url.c_str()), nullptr};
+    constexpr int kNoSignal = 0;
+#else
+    const char* program = "xdg-open";
+    char* argv[] = {const_cast<char*>("xdg-open"), const_cast<char*>(url.c_str()), nullptr};
+    char* gated[] = {const_cast<char*>("sh"), const_cast<char*>("-c"), gateScript, const_cast<char*>(program),
+                     const_cast<char*>(url.c_str()), nullptr};
+    constexpr int kNoSignal = MSG_NOSIGNAL;
+#endif
+    int gate[2] = {-1, -1};
+    if (!bounded) {
+        if (socketpair(AF_UNIX, SOCK_STREAM, 0, gate) != 0) {
+            return Handoff::None;
+        }
+        // dup2(3, 3) keeps FD_CLOEXEC on macOS, which would close the gate in the child.
+        if (gate[0] == 3) {
+            std::swap(gate[0], gate[1]);
+        }
+        fcntl(gate[0], F_SETFD, FD_CLOEXEC);
+        fcntl(gate[1], F_SETFD, FD_CLOEXEC);
+#if defined(__APPLE__)
+        int noSigPipe = 1;
+        setsockopt(gate[1], SOL_SOCKET, SO_NOSIGPIPE, &noSigPipe, sizeof(noSigPipe));
+#endif
+    }
+    const auto closeGate = [&gate] {
+        for (int& end : gate) {
+            if (end != -1) {
+                close(end);
+                end = -1;
+            }
+        }
+    };
+    posix_spawn_file_actions_t actions;
+    if (posix_spawn_file_actions_init(&actions) != 0) {
+        closeGate();
+        return Handoff::None;
+    }
+    if (!bounded) {
+        posix_spawn_file_actions_adddup2(&actions, gate[0], 3);
+    }
+    posix_spawn_file_actions_addopen(&actions, STDIN_FILENO, "/dev/null", O_RDONLY, 0);
+    posix_spawn_file_actions_addopen(&actions, STDOUT_FILENO, "/dev/null", O_WRONLY, 0);
+    posix_spawn_file_actions_addopen(&actions, STDERR_FILENO, "/dev/null", O_WRONLY, 0);
+    // A crash handler can call this with signals blocked; the child should not inherit that.
+    posix_spawnattr_t attributes;
+    posix_spawnattr_init(&attributes);
+    sigset_t noSignals;
+    sigemptyset(&noSignals);
+    posix_spawnattr_setsigmask(&attributes, &noSignals);
+    posix_spawnattr_setflags(&attributes, POSIX_SPAWN_SETSIGMASK);
+    pid_t child = 0;
+    std::unique_lock launch(g_helperMutex, std::defer_lock);
+    if (!bounded) {
+        launch.lock();
+    }
+    int error = ECANCELED;
+    if (bounded) {
+        error = posix_spawnp(&child, program, &actions, &attributes, argv, environ);
+    } else if (!g_helperCancelled) {
+        error = posix_spawn(&child, "/bin/sh", &actions, &attributes, gated, environ);
+    }
+    posix_spawnattr_destroy(&attributes);
+    posix_spawn_file_actions_destroy(&actions);
+    if (error != 0) {
+        closeGate();
+        return Handoff::None;
+    }
+    int status = 0;
+    if (!bounded) {
+        g_helper = child;
+        if (!g_helperCancelled) {
+            send(gate[1], "1\n", 2, kNoSignal);
+        }
+        closeGate();
+        launch.unlock();
+        // Wait without reaping so the pid can't be reused while it is still published.
+        siginfo_t info{};
+        while (waitid(P_PID, static_cast<id_t>(child), &info, WEXITED | WNOWAIT) == -1 && errno == EINTR) {
+        }
+        {
+            std::lock_guard lock(g_helperMutex);
+            g_helper = 0;
+        }
+    }
+    const auto deadline = Clock::now() + kHandoffTimeout;
+    for (;;) {
+        const pid_t waited = waitpid(child, &status, bounded ? WNOHANG : 0);
+        if (waited == child) {
+            return WIFEXITED(status) && WEXITSTATUS(status) == 0 ? Handoff::Confirmed : Handoff::Unconfirmed;
+        }
+        if (waited == -1 && errno != EINTR) {
+            return Handoff::Unconfirmed;
+        }
+        if (waited == 0) {
+            if (Clock::now() >= deadline) {
+                try {
+                    std::thread([child] { waitpid(child, nullptr, 0); }).detach();
+                } catch (...) {
+                }
+                return Handoff::Unconfirmed;
+            }
+            std::this_thread::sleep_for(std::chrono::milliseconds(10));
+        }
+    }
+}
+
+void CancelWorkerRequests() {
+    g_helperCancelled = true;
+    std::unique_lock lock(g_helperMutex, std::defer_lock);
+    if (!TryLockBriefly(lock, kHandoffTimeout)) {
+        return;
+    }
+    if (g_helper == 0 || kill(g_helper, SIGKILL) != 0) {
+        return;
+    }
+    // The worker only reaps after clearing g_helper under this lock, so the pid stays ours.
+    const auto deadline = Clock::now() + kHandoffTimeout;
+    for (;;) {
+        siginfo_t info{};
+        if (waitid(P_PID, static_cast<id_t>(g_helper), &info, WEXITED | WNOWAIT | WNOHANG) != 0 ||
+            info.si_pid != 0 || Clock::now() >= deadline) {
+            return;
+        }
+        std::this_thread::sleep_for(std::chrono::milliseconds(10));
+    }
+}
+
+#endif
+
+// Caller holds state.delivery. Only exit paths bound the wait; the worker waits for each
+// request to finish so a slow one can never land after the next.
+Handoff Deliver(State& state, bool forced, uint32_t appId, bool bounded) {
+    const auto earliest = Clock::time_point(Clock::duration(state.lastDelivery.load())) + kDeliverySpacing;
+    if (const auto now = Clock::now(); now < earliest) {
+        std::this_thread::sleep_for(earliest - now);
+    }
+    const Handoff delivered = OpenSteamUrl("steam://forceinputappid/" + std::to_string(forced ? appId : 0), bounded);
+    if (delivered != Handoff::None) {
+        state.lastDelivery = Clock::now().time_since_epoch().count();
+        if (forced) {
+            RT_LOG(RT_TAG_RUNTIME) << "Steam Input: applying the controller layout of app " << appId << std::endl;
+        } else {
+            RT_LOG(RT_TAG_RUNTIME) << "Steam Input: restored the desktop controller layout" << std::endl;
+        }
+    }
+    return delivered;
+}
+
+void Worker(State& state) {
+    t_isWorker = true;
+    std::unique_lock lock(state.mutex);
+    for (;;) {
+        state.wake.wait(lock, [&] { return state.stopping || state.desired != state.sent; });
+        if (state.stopping) {
+            return;
+        }
+        const bool target = state.desired;
+        const uint32_t appId = state.appId;
+        // A failed request is retried on the next focus change rather than in a loop.
+        state.sent = target;
+        lock.unlock();
+        {
+            std::lock_guard delivery(state.delivery);
+            {
+                // Release may have run while this thread waited for the delivery lock.
+                std::lock_guard guard(state.mutex);
+                if (state.stopping) {
+                    return;
+                }
+            }
+            const Handoff delivered = Deliver(state, target, appId, false);
+            std::lock_guard guard(state.mutex);
+            state.maybeForced = target ? delivered != Handoff::None || state.maybeForced
+                                       : state.maybeForced && delivered != Handoff::Confirmed;
+        }
+        lock.lock();
+    }
+}
+
+} // namespace
+
+void Initialize(bool enabled, uint32_t appId) {
+    if (!enabled || appId == 0 || g_state.load() != nullptr) {
+        return;
+    }
+    if (LaunchedBySteam()) {
+        RT_LOG(RT_TAG_RUNTIME) << "Steam Input: launched by Steam, leaving its controller layout alone" << std::endl;
+        return;
+    }
+    SDL_Window* window = aurora_get_window();
+    if (window == nullptr) {
+        return;
+    }
+    auto* state = new State();
+    state->enabled = true;
+    state->appId = appId;
+    state->windowId = SDL_GetWindowID(window);
+    state->desired = (SDL_GetWindowFlags(window) & SDL_WINDOW_INPUT_FOCUS) != 0;
+    try {
+        std::thread(Worker, std::ref(*state)).detach();
+    } catch (const std::exception& error) {
+        RT_LOG(RT_TAG_RUNTIME) << "Steam Input: " << error.what() << std::endl;
+        delete state;
+        return;
+    }
+    g_state.store(state);
+}
+
+void HandleEvents(const AuroraEvent* events) {
+    State* state = g_state.load();
+    if (state == nullptr || events == nullptr) {
+        return;
+    }
+    int focus = -1;
+    for (const AuroraEvent* event = events; event->type != AURORA_NONE; ++event) {
+        if (event->type != AURORA_SDL_EVENT) {
+            continue;
+        }
+        const SDL_Event& sdl = event->sdl;
+        const bool gained = sdl.type == SDL_EVENT_WINDOW_FOCUS_GAINED;
+        if ((gained || sdl.type == SDL_EVENT_WINDOW_FOCUS_LOST) && sdl.window.windowID == state->windowId) {
+            focus = gained ? 1 : 0;
+        }
+    }
+    if (focus < 0) {
+        return;
+    }
+    std::lock_guard lock(state->mutex);
+    if (state->enabled) {
+        state->desired = focus == 1;
+        state->wake.notify_one();
+    }
+}
+
+void Release() noexcept {
+    State* state = g_state.load();
+    if (state == nullptr) {
+        return;
+    }
+    const auto releaseWait = kHandoffTimeout + kDeliverySpacing + std::chrono::milliseconds(500);
+    try {
+        {
+            std::unique_lock lock(state->mutex, std::defer_lock);
+            if (!TryLockBriefly(lock)) {
+                return;
+            }
+            if (!state->enabled) {
+                lock.unlock();
+                // Another thread is restoring; exiting under it would kill its request.
+                const auto deadline = Clock::now() + 2 * releaseWait;
+                while (!t_inRelease && !state->released.load() && Clock::now() < deadline) {
+                    std::this_thread::sleep_for(std::chrono::milliseconds(5));
+                }
+                return;
+            }
+            t_inRelease = true;
+            state->enabled = false;
+            state->stopping = true;
+            state->wake.notify_one();
+        }
+        // Let an in-flight request land first so the restore is the last one Steam sees.
+        std::unique_lock delivery(state->delivery, std::defer_lock);
+        const bool ordered = !t_isWorker && delivery.try_lock_for(releaseWait);
+        bool maybeForced = true;
+        if (ordered) {
+            std::unique_lock lock(state->mutex, std::defer_lock);
+            maybeForced = !TryLockBriefly(lock) || state->maybeForced;
+        }
+        // A request still in flight, or about to start, could otherwise land after the restore.
+        CancelWorkerRequests();
+        if (maybeForced) {
+            // Discovery and spawning can block as well; they must not hold up the exit.
+            auto done = std::make_shared<std::atomic<bool>>(false);
+            try {
+                std::thread([state, done] {
+                    Deliver(*state, false, 0, true);
+                    done->store(true);
+                }).detach();
+            } catch (const std::system_error&) {
+                done->store(true);
+            }
+            const auto deadline = Clock::now() + releaseWait;
+            while (!done->load() && Clock::now() < deadline) {
+                std::this_thread::sleep_for(std::chrono::milliseconds(5));
+            }
+        }
+    } catch (...) {
+    }
+    state->released.store(true);
+}
+
+} // namespace SteamInputLayout

runtime/tests/motion_trick_tests.cpp

@@ -0,0 +1,247 @@
+#include "motion_trick_detector.h"
+
+#include <cmath>
+#include <cstdio>
+#include <cstring>
+#include <string>
+#include <vector>
+
+using MotionTrick::Direction;
+using MotionTrick::MotionSample;
+using MotionTrick::ShakeDetector;
+
+namespace {
+
+int g_failures = 0;
+constexpr uint64_t kFrameUs = 16'667;
+
+void Check(bool ok, const std::string& what) {
+    if (!ok) {
+        std::printf("  FAIL: %s\n", what.c_str());
+        ++g_failures;
+    }
+}
+
+// Written as raw bits so no float operation under -ffast-math can drop them.
+void SetBits(float& target, uint32_t bits) {
+    std::memcpy(&target, &bits, sizeof(target));
+}
+
+MotionSample Gyro(float pitch, float yaw, float roll) {
+    MotionSample sample;
+    sample.hasGyro = true;
+    sample.gyro[0] = pitch;
+    sample.gyro[1] = yaw;
+    sample.gyro[2] = roll;
+    sample.hasAccel = true;
+    sample.accel[1] = 9.80665f;
+    return sample;
+}
+
+MotionSample AtRest() {
+    return Gyro(0.0f, 0.0f, 0.0f);
+}
+
+struct Run {
+    ShakeDetector detector;
+    uint64_t now = 0;
+    std::vector<Direction> fired;
+
+    void Feed(const MotionSample& sample, int frames = 1) {
+        for (int i = 0; i < frames; ++i) {
+            const Direction direction = detector.Update(sample, now);
+            if (direction != Direction::None) fired.push_back(direction);
+            now += kFrameUs;
+        }
+    }
+
+    // Ramp up to the peak and back down over a few frames, like a real flick.
+    void Flick(const MotionSample& peak) {
+        for (const float scale : {0.3f, 0.7f, 1.0f, 0.6f, 0.2f}) {
+            MotionSample sample = peak;
+            for (float& value : sample.gyro) value *= scale;
+            Feed(sample);
+        }
+    }
+};
+
+void TestRestNeverFires() {
+    Run run;
+    run.Feed(AtRest(), 600);
+    Check(run.fired.empty(), "controller at rest fired a trick");
+}
+
+void TestSteeringWobbleNeverFires() {
+    Run run;
+    for (int i = 0; i < 600; ++i) {
+        run.Feed(Gyro(1.5f * std::sin(i * 0.2f), 2.0f * std::sin(i * 0.13f), 2.5f * std::cos(i * 0.17f)));
+    }
+    Check(run.fired.empty(), "steering wobble fired a trick");
+}
+
+void TestOneFlickOneTrick() {
+    Run run;
+    run.Flick(Gyro(10.0f, 0.0f, 0.0f));
+    run.Feed(AtRest(), 60);
+    Check(run.fired.size() == 1, "one flick did not give exactly one trick");
+    Check(!run.fired.empty() && run.fired[0] == Direction::Up, "upward flick was not Up");
+}
+
+void TestReturnStrokeIsIgnored() {
+    Run run;
+    run.Flick(Gyro(10.0f, 0.0f, 0.0f));
+    run.Flick(Gyro(-9.0f, 0.0f, 0.0f));
+    run.Feed(AtRest(), 60);
+    Check(run.fired.size() == 1, "bringing the controller back gave a second trick");
+}
+
+void TestSeparateFlicksBothFire() {
+    Run run;
+    run.Flick(Gyro(10.0f, 0.0f, 0.0f));
+    run.Feed(AtRest(), 40);
+    run.Flick(Gyro(10.0f, 0.0f, 0.0f));
+    run.Feed(AtRest(), 40);
+    Check(run.fired.size() == 2, "two separate flicks did not give two tricks");
+}
+
+void TestSustainedShakeFiresOnce() {
+    Run run;
+    for (int i = 0; i < 120; ++i) {
+        run.Feed(Gyro(i % 2 == 0 ? 12.0f : -12.0f, 0.0f, 0.0f));
+    }
+    Check(run.fired.size() == 1, "continuous shaking fired more than once");
+}
+
+void TestDirections() {
+    const struct {
+        MotionSample peak;
+        Direction expected;
+        const char* name;
+    } cases[] = {
+        {Gyro(10.0f, 0.0f, 0.0f), Direction::Up, "pitch up"},
+        {Gyro(-10.0f, 0.0f, 0.0f), Direction::Down, "pitch down"},
+        {Gyro(1.0f, 0.0f, 10.0f), Direction::Left, "roll left"},
+        {Gyro(1.0f, 0.0f, -10.0f), Direction::Right, "roll right"},
+        {Gyro(0.0f, 10.0f, 2.0f), Direction::Left, "yaw left"},
+        {Gyro(0.0f, -10.0f, -2.0f), Direction::Right, "yaw right"},
+    };
+    for (const auto& c : cases) {
+        Run run;
+        run.Flick(c.peak);
+        Check(run.fired.size() == 1 && run.fired[0] == c.expected, std::string("wrong direction for ") + c.name);
+    }
+}
+
+void TestSensitivity() {
+    Run low;
+    low.detector.SetSensitivity(1);
+    low.Flick(Gyro(6.0f, 0.0f, 0.0f));
+    Check(low.fired.empty(), "sensitivity 1 fired on a small flick");
+
+    Run high;
+    high.detector.SetSensitivity(5);
+    high.Flick(Gyro(6.0f, 0.0f, 0.0f));
+    Check(high.fired.size() == 1, "sensitivity 5 missed a small flick");
+
+    Run clamped;
+    clamped.detector.SetSensitivity(99);
+    clamped.Flick(Gyro(6.0f, 0.0f, 0.0f));
+    Check(clamped.fired.size() == 1, "out of range sensitivity was not clamped");
+}
+
+void TestAccelOnlyController() {
+    Run run;
+    MotionSample rest;
+    rest.hasAccel = true;
+    rest.accel[1] = 9.80665f;
+    run.Feed(rest, 120);
+    Check(run.fired.empty(), "gravity alone fired a trick");
+
+    MotionSample jolt = rest;
+    jolt.accel[1] = 35.0f;
+    run.Feed(jolt);
+    run.Feed(rest, 60);
+    Check(run.fired.size() == 1 && run.fired[0] == Direction::Up, "accelerometer jolt did not give Up");
+}
+
+void TestGyroOnlyController() {
+    Run run;
+    MotionSample flick = Gyro(10.0f, 0.0f, 0.0f);
+    flick.hasAccel = false;
+    MotionSample rest = AtRest();
+    rest.hasAccel = false;
+    run.Flick(flick);
+    run.Feed(rest, 60);
+    run.Flick(flick);
+    run.Feed(rest, 60);
+    Check(run.fired.size() == 2 && run.fired[0] == Direction::Up && run.fired[1] == Direction::Up,
+          "repeated gyro-only flicks did not give two Up tricks");
+}
+
+void TestGyroPadIgnoresAccel() {
+    Run bump;
+    MotionSample jolt = AtRest();
+    jolt.accel[1] = 40.0f;
+    bump.Feed(jolt, 3);
+    bump.Feed(AtRest(), 60);
+    Check(bump.fired.empty(), "an accelerometer bump on a gyro pad fired a trick");
+
+    Run flick;
+    MotionSample sideways = Gyro(0.0f, 10.0f, 0.0f);
+    sideways.accel[1] = 40.0f;
+    flick.Feed(sideways);
+    Check(flick.fired.size() == 1 && flick.fired[0] == Direction::Left,
+          "a jolt changed the direction of a sideways flick");
+}
+
+void TestNoSensorsNeverFires() {
+    Run run;
+    MotionSample none;
+    none.gyro[0] = 50.0f;
+    none.accel[1] = 80.0f;
+    run.Feed(none, 60);
+    Check(run.fired.empty(), "a sample without sensors fired a trick");
+}
+
+void TestBadDataIgnored() {
+    Run run;
+    for (const uint32_t bits : {0x7fc00000u, 0x7f800000u}) {
+        for (int axis = 0; axis < 3; ++axis) {
+            MotionSample gyro = AtRest();
+            SetBits(gyro.gyro[axis], bits);
+            run.Feed(gyro, 10);
+
+            MotionSample accel;
+            accel.hasAccel = true;
+            SetBits(accel.accel[axis], bits);
+            run.Feed(accel, 10);
+        }
+    }
+    Check(run.fired.empty(), "NaN or infinite sensor data fired a trick");
+    run.Flick(Gyro(10.0f, 0.0f, 0.0f));
+    Check(run.fired.size() == 1, "detector did not recover after NaN data");
+}
+
+} // namespace
+
+int main() {
+    TestRestNeverFires();
+    TestSteeringWobbleNeverFires();
+    TestOneFlickOneTrick();
+    TestReturnStrokeIsIgnored();
+    TestSeparateFlicksBothFire();
+    TestSustainedShakeFiresOnce();
+    TestDirections();
+    TestSensitivity();
+    TestAccelOnlyController();
+    TestGyroOnlyController();
+    TestGyroPadIgnoresAccel();
+    TestNoSensorsNeverFires();
+    TestBadDataIgnored();
+    if (g_failures != 0) {
+        std::printf("%d failure(s)\n", g_failures);
+        return 1;
+    }
+    std::printf("motion trick tests passed\n");
+    return 0;
+}

runtime/tests/runtime_config_tests.cpp

@@ -51,6 +51,45 @@ int main() {
         const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "invalid.toml");
         if (!Require(!config.metalFxSpatialUpscaling, "invalid MetalFX setting was accepted")) return 1;
     }
+    {
+        std::istringstream input("[video]\nwidescreen = false\nforce_16_9 = true\n");
+        const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "legacy.toml");
+        if (!Require(config.forceAspect == std::optional<bool>{true} &&
+                         config.widescreen == std::optional<bool>{true} && config.legacyForceAspect,
+                     "legacy force_16_9 did not force a 16:9 game")) return 1;
+    }
+    {
+        std::istringstream input("[video]\nwidescreen = false\nforce_aspect = true\nforce_16_9 = true\n");
+        const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "migrated.toml");
+        if (!Require(config.forceAspect == std::optional<bool>{true} &&
+                         config.widescreen == std::optional<bool>{false} && !config.legacyForceAspect,
+                     "legacy force_16_9 overrode force_aspect")) return 1;
+    }
+    {
+        std::istringstream input("[controller]\nmotion_trick = true\nmotion_trick_sensitivity = 5\n");
+        const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "motion.toml");
+        if (!Require(config.motionTrick == std::optional<bool>{true} &&
+                         config.motionTrickSensitivity == std::optional<int32_t>{5},
+                     "shake to trick settings were not loaded")) return 1;
+    }
+    {
+        std::istringstream input("[controller]\nmotion_trick_sensitivity = 9\n");
+        const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "motion-range.toml");
+        if (!Require(!config.motionTrick && !config.motionTrickSensitivity,
+                     "out of range shake to trick sensitivity was accepted")) return 1;
+    }
+    {
+        std::istringstream input("[controller]\nsteam_input_override = false\nsteam_input_app_id = 1234\n");
+        const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "steam.toml");
+        if (!Require(config.steamInputOverride == std::optional<bool>{false} &&
+                         config.steamInputAppId == std::optional<uint32_t>{1234},
+                     "Steam Input settings were not loaded")) return 1;
+    }
+    {
+        std::istringstream input("[controller]\nsteam_input_app_id = 0\n");
+        const RuntimeUserConfig config = RuntimeConfigFile::ParseConfig(input, "steam-zero.toml");
+        if (!Require(!config.steamInputAppId, "zero Steam Input app id 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;

translator/src/Translator.Cli/Program.cs

@@ -139,6 +139,15 @@ TranslationOptions WithProjectFrontEndPolicy(
         LeafInliningBlockedTargets = leafInliningBlockedTargets.Value
     };
 string? translateModInputCacheDirectory = null;
+// translate-mod reads the same (large) base metadata from several places.
+var baseMetadataCache = new Dictionary<string, BaseTranslationOutputMetadata>(StringComparer.OrdinalIgnoreCase);
+BaseTranslationOutputMetadata ReadBaseMetadataCached(string path)
+{
+    var fullPath = Path.GetFullPath(path);
+    if (!baseMetadataCache.TryGetValue(fullPath, out var metadata))
+        baseMetadataCache[fullPath] = metadata = BaseTranslationOutputMetadataFile.Read(fullPath);
+    return metadata;
+}
 string? translateModPublishedOutputDirectory = null;
 List<AddressRange>? _executableRanges = null;
 // Callback discovery now runs on the discovery worker pool, so the range cache
@@ -472,10 +481,6 @@ int RunTranslateRecursive(string[] argsTail)
     var pruneStale = HasFlag(argsTail, "--prune-stale");
     var threadsOption = OptionValue(argsTail, "--threads");
     var maxThreads = threadsOption != null ? Math.Max(1, ParseInt(threadsOption)) : Math.Max(1, Environment.ProcessorCount);
-    // Discovery batches: cheap front-end-only work items, so the batch is sized
-    // generously to keep queue overhead off the hot path. Deliberately larger
-    // than the mod emission wave below, which carries whole C++ emissions.
-    var batchSize = Math.Max(64, maxThreads * 8);
     Console.WriteLine($"[translator] Using up to {maxThreads} worker thread(s).");
 
     // Log start time
@@ -493,7 +498,6 @@ int RunTranslateRecursive(string[] argsTail)
         // callees are decoded separately and never appear here). Feeds the
         // fall-through interior-entry detection for the base manifest.
         var translatedFlowCoverage = new Dictionary<uint, uint[]>();
-        var gqrUnknownEntryRoots = new HashSet<uint> { startAddr };
         var translatedFunctionEnds = new Dictionary<uint, uint>();
         // Every map entry is a function start, so the whole boundary set is known
         // before the first decode instead of accumulating as discovery proceeds.
@@ -565,13 +569,15 @@ int RunTranslateRecursive(string[] argsTail)
 
         void ProcessQueue(bool speculative = false)
         {
+            // Each round decodes the whole frontier: the output depends only on the final
+            // boundary set (boundary repair re-decodes stale owners), never on batching.
             while (queue.Count > 0)
             {
-                var batch = new List<(uint Address, int Depth, string Name)>(batchSize);
+                var batch = new List<(uint Address, int Depth, string Name)>(queue.Count);
                 var batchAddresses = new HashSet<uint>();
                 var candidates = QueueBatch.Dequeue(
                     queue,
-                    batchSize,
+                    queue.Count,
                     next => ShouldSkip(next) || !batchAddresses.Add(next.addr));
                 foreach (var next in candidates)
                 {
@@ -598,23 +604,19 @@ int RunTranslateRecursive(string[] argsTail)
             }
         }
 
-        // Batches drain one at a time so a callee of a prior speculative entry arrives with a known
-        // caller/GQR state instead of as another unknown root. Everything here is speculative, so a
-        // "call" decoded from an unreached map entry isn't proven code; treat failures as soft, not fatal.
+        // Everything here is speculative, so a "call" decoded from an unreached map entry
+        // isn't proven code; treat failures as soft, not fatal.
         void ProcessSpeculativeSeeds()
         {
             while (speculativeSeeds.Count > 0)
             {
-                var batch = new List<(uint Address, int Depth, string Name)>(batchSize);
+                var batch = new List<(uint Address, int Depth, string Name)>(speculativeSeeds.Count);
                 var candidates = QueueBatch.Dequeue(
                     speculativeSeeds,
-                    batchSize,
+                    speculativeSeeds.Count,
                     address => translated.Contains(address) || !visited.Add(address));
                 foreach (var address in candidates)
                 {
-                    // Nothing calls this from the translated set, so its entry
-                    // GQR state is unconstrained.
-                    gqrUnknownEntryRoots.Add(address);
                     batch.Add((address, 0, ResolveFunctionName(address, preferred: null, $"func_{address:X8}")));
                 }
 
@@ -652,22 +654,42 @@ int RunTranslateRecursive(string[] argsTail)
         // decoded out of one - is dropped with a counted warning.
         void TranslateBatch(List<(uint Address, int Depth, string Name)> batch, bool speculative = false)
         {
-            var results = new FunctionDiscoveryResult?[batch.Count];
-            IndexedParallel.For(batch.Count, parallelOptions, i => results[i] = Attempt(batch[i], speculative));
+            var metrics = new TranslationMetrics?[batch.Count];
+            var entryPoints = new uint[batch.Count];
+            var discoveryTargets = new uint[batch.Count][];
+            var swallowedSiblingEntries = new uint[batch.Count][];
+            var summaries = new GqrFunctionSummary[batch.Count];
+            var functionEnds = new uint[batch.Count];
+            var flowCoverage = new uint[batch.Count][];
+
+            // Everything kept from a decode is derived in the decode's own pass, so the full
+            // result is never retained; the few results a newly revealed boundary invalidates
+            // are simply decoded again. Commits stay serial and in batch order.
+            void Decode(int i, bool findSwallowedSiblings)
+            {
+                var result = Attempt(batch[i], speculative);
+                metrics[i] = result?.Metrics;
+                if (result is null)
+                {
+                    canonicalIrStore.Remove(batch[i].Address);
+                    return;
+                }
+                entryPoints[i] = result.EntryPoint;
+                if (findSwallowedSiblings)
+                    swallowedSiblingEntries[i] = DiscoverSwallowedSiblingEntries(result).ToArray();
+                discoveryTargets[i] = DiscoverDiscoveryTranslationTargets(result).ToArray();
+                summaries[i] = GqrFunctionSummary.Create(result.Ssa.Function);
+                canonicalIrStore.Put(batch[i].Address, result.Ssa.Function);
+                functionEnds[i] = GetFunctionInstructionEnd(result);
+                flowCoverage[i] = CoalesceInstructionRanges(result);
+            }
+
+            IndexedParallel.For(batch.Count, parallelOptions, i => Decode(i, findSwallowedSiblings: true));
 
             // Re-run only translations that swallowed a newly revealed sibling entry, or direct
-            // branches could bypass a mod overlay registered there. Both scans are pure per-result
-            // so they run on the same worker pool; the dedup filter below keeps first sighting.
-            var discoveryTargets = new uint[results.Length][];
-            var swallowedSiblingEntries = new uint[results.Length][];
-            IndexedParallel.For(results.Length, parallelOptions, i =>
-            {
-                if (results[i] is not { } result) return;
-                discoveryTargets[i] = DiscoverDiscoveryTranslationTargets(result).ToArray();
-                swallowedSiblingEntries[i] = DiscoverSwallowedSiblingEntries(result).ToArray();
-            });
-            var newlyKnownEntries = Enumerable.Range(0, results.Length)
-                .Where(i => results[i] is not null)
+            // branches could bypass a mod overlay registered there. The dedup filter keeps first sighting.
+            var newlyKnownEntries = Enumerable.Range(0, batch.Count)
+                .Where(i => metrics[i] is not null)
                 .SelectMany(i => discoveryTargets[i].Concat(swallowedSiblingEntries[i]))
                 .Where(knownBaseFunctionEntryPoints.Add)
                 .ToHashSet();
@@ -675,49 +697,39 @@ int RunTranslateRecursive(string[] argsTail)
             {
                 // The boundary set is final for this batch once the filter above has run,
                 // so the affected re-translations are independent of one another.
-                var affected = new List<int>();
-                for (var i = 0; i < results.Length; i++)
-                {
-                    if (results[i] is not { } result) continue;
-                    var entry = result.EntryPoint;
-                    if (result.Instructions.Any(instruction =>
-                            instruction.Address != entry && newlyKnownEntries.Contains(instruction.Address)))
-                    {
-                        affected.Add(i);
-                    }
-                }
-
-                IndexedParallel.For(affected.Count, parallelOptions, k =>
-                {
-                    var i = affected[k];
-                    results[i] = Attempt(batch[i], speculative);
-                    if (results[i] is { } result)
-                    {
-                        discoveryTargets[i] = DiscoverDiscoveryTranslationTargets(result).ToArray();
-                    }
-                });
+                var sortedNewEntries = newlyKnownEntries.Order().ToArray();
+                var affected = Enumerable.Range(0, batch.Count)
+                    .Where(i => metrics[i] is not null &&
+                        CoversInteriorInstruction(flowCoverage[i], entryPoints[i], sortedNewEntries))
+                    .ToArray();
+                IndexedParallel.For(affected.Length, parallelOptions, k => Decode(affected[k], findSwallowedSiblings: false));
             }
 
-            // The compact GQR projection is a pure walk of the final SSA graph and the
-            // canonical store serializes its own dictionary, so both move off the serial
-            // commit. Every address in a batch is distinct and untranslated, so the store
-            // sees one Put per key regardless of completion order.
-            var summaries = new GqrFunctionSummary[results.Length];
-            IndexedParallel.For(results.Length, parallelOptions, i =>
-            {
-                if (results[i] is not { } result) return;
-                summaries[i] = GqrFunctionSummary.Create(result.Ssa.Function);
-                canonicalIrStore.Put(batch[i].Address, result.Ssa.Function);
-            });
-
-            // Update state sequentially (thread-safe)
             for (var i = 0; i < batch.Count; i++)
             {
-                if (results[i] is not { } result) continue;
-                CommitResult(batch[i], result, discoveryTargets[i], summaries[i]);
+                if (metrics[i] is not { } functionMetrics) continue;
+                CommitResult(batch[i], functionMetrics, discoveryTargets[i], summaries[i], functionEnds[i], flowCoverage[i]);
             }
         }
 
+        // True when an instruction of the coalesced [start, end) coverage, other than the
+        // entry, starts at one of the sorted addresses.
+        static bool CoversInteriorInstruction(uint[] coverage, uint entry, uint[] sortedAddresses)
+        {
+            for (var r = 0; r < coverage.Length; r += 2)
+            {
+                var index = Array.BinarySearch(sortedAddresses, coverage[r]);
+                for (index = index >= 0 ? index : ~index;
+                     index < sortedAddresses.Length && sortedAddresses[index] < coverage[r + 1];
+                     index++)
+                {
+                    var address = sortedAddresses[index];
+                    if (address != entry && (address & 3u) == 0) return true;
+                }
+            }
+            return false;
+        }
+
         FunctionDiscoveryResult? Attempt((uint Address, int Depth, string Name) work, bool speculative)
         {
             if (!speculative)
@@ -773,16 +785,18 @@ int RunTranslateRecursive(string[] argsTail)
 
         void CommitResult(
             (uint Address, int Depth, string Name) work,
-            FunctionDiscoveryResult result,
+            TranslationMetrics functionMetrics,
             IReadOnlyList<uint> discoveryTargets,
-            GqrFunctionSummary summary)
+            GqrFunctionSummary summary,
+            uint functionEnd,
+            uint[] flowCoverage)
         {
             // File writing moved to parallel section above
-            totals.Add(result.Metrics);
+            totals.Add(functionMetrics);
             count++;
             translated.Add(work.Address);
-            translatedFunctionEnds[work.Address] = GetFunctionInstructionEnd(result);
-            translatedFlowCoverage[work.Address] = CoalesceInstructionRanges(result);
+            translatedFunctionEnds[work.Address] = functionEnd;
+            translatedFlowCoverage[work.Address] = flowCoverage;
 
             foreach (var target in discoveryTargets)
             {
@@ -821,6 +835,7 @@ int RunTranslateRecursive(string[] argsTail)
         // list that existed to paper over the addresses they still missed.
         ProcessQueue();
         var walkReachable = count;
+        var walkTranslated = translated.ToHashSet();
         ProcessSpeculativeSeeds();
         if (functionMap is not null)
         {
@@ -864,26 +879,31 @@ int RunTranslateRecursive(string[] argsTail)
             if (boundaryRepairAddresses.Length != 0)
             {
                 Console.WriteLine($"[translator] Reanalyzing {boundaryRepairAddresses.Length:N0} function(s) affected by later-discovered boundaries...");
-                var repaired = new FunctionDiscoveryResult[boundaryRepairAddresses.Length];
+                var repairedSummaries = new GqrFunctionSummary[boundaryRepairAddresses.Length];
+                var repairedEnds = new uint[boundaryRepairAddresses.Length];
+                var repairedCoverage = new uint[boundaryRepairAddresses.Length][];
                 IndexedParallel.For(boundaryRepairAddresses.Length, parallelOptions, i =>
                 {
                     var address = boundaryRepairAddresses[i];
                     var work = baseWork[address];
-                    repaired[i] = translator.Value.Discover(address, WithProjectFrontEndPolicy(new TranslationOptions(
+                    var repaired = translator.Value.Discover(address, WithProjectFrontEndPolicy(new TranslationOptions(
                         work.Name,
                         MaxInstructions: work.MaxInstructions,
                         MaxBytes: work.MaxBytes,
                         AllowUnsupportedInstructions: work.AllowUnsupported,
                         KnownFunctionEntryPoints: knownBaseFunctionEntryPoints),
                         RequireProject().Translation));
+                    repairedSummaries[i] = GqrFunctionSummary.Create(repaired.Ssa.Function);
+                    canonicalIrStore.Put(address, repaired.Ssa.Function);
+                    repairedEnds[i] = GetFunctionInstructionEnd(repaired);
+                    repairedCoverage[i] = CoalesceInstructionRanges(repaired);
                 });
                 for (var i = 0; i < boundaryRepairAddresses.Length; i++)
                 {
                     var address = boundaryRepairAddresses[i];
-                    baseSummaries[address] = GqrFunctionSummary.Create(repaired[i].Ssa.Function);
-                    canonicalIrStore.Put(address, repaired[i].Ssa.Function);
-                    translatedFunctionEnds[address] = GetFunctionInstructionEnd(repaired[i]);
-                    translatedFlowCoverage[address] = CoalesceInstructionRanges(repaired[i]);
+                    baseSummaries[address] = repairedSummaries[i];
+                    translatedFunctionEnds[address] = repairedEnds[i];
+                    translatedFlowCoverage[address] = repairedCoverage[i];
                 }
             }
 
@@ -906,6 +926,12 @@ int RunTranslateRecursive(string[] argsTail)
             }
 
             Console.WriteLine("[translator] Computing interprocedural GQR constants and callee write summaries...");
+            // Map entries the walk never reached have an unconstrained entry GQR state unless some
+            // translated function calls them directly.
+            var gqrUnknownEntryRoots = GqrInterproceduralAnalysis.SelectUnknownEntryRoots(
+                baseSummaries,
+                [startAddr],
+                (functionMap?.Addresses ?? Array.Empty<uint>()).Where(address => !walkTranslated.Contains(address)));
             var gqrAnalysis = GqrInterproceduralAnalysis.Analyze(
                 baseSummaries,
                 gqrUnknownEntryRoots);
@@ -1183,7 +1209,7 @@ int RunTranslateRecursive(string[] argsTail)
             // Populate the recursive ABI inference cache once, in stable global order, before
             // refinement and parallel emission, so cyclic call graphs don't depend on thread count.
             LogPhase("state-free selection + leaf residency");
-            ReportAbiPrewarm(inferredGuestAbi.Value.Prewarm(emissionAddresses));
+            ReportAbiPrewarm(inferredGuestAbi.Value.Prewarm(emissionAddresses, maxThreads));
             LogPhase("ABI prewarm");
 
             // State-free entries use the transformed leaf's exact live-at-entry
@@ -1404,6 +1430,7 @@ int RunTranslateRecursive(string[] argsTail)
             // source visible. `completionPulse` only wakes the writer; it carries
             // no state of its own.
             var emissionDone = new bool[pendingSources.Length];
+            var unsupportedByIndex = new int[emissionAddresses.Length];
             using var completionPulse = new ManualResetEventSlim(false);
             var emissionAborted = false;
             Task? pendingBundleWrite = null;
@@ -1459,6 +1486,8 @@ int RunTranslateRecursive(string[] argsTail)
                         var options = BuildEmissionOptions(address, stateFreeAbiContracts);
                         if (!canonicalIrStore.TryGet(address, out var canonicalSsa))
                             throw new InvalidOperationException($"Missing canonical SSA for final lowering at 0x{address:X8}.");
+                        unsupportedByIndex[i] = canonicalSsa.Blocks.Sum(block =>
+                            block.Instructions.Count(static instruction => instruction is IrUndefined));
                         var emitted = translator.Value.LowerCanonical(address, canonicalSsa, options);
                         var path = work.Path;
                         if (outputMetadata is not null)
@@ -1517,16 +1546,7 @@ int RunTranslateRecursive(string[] argsTail)
                     : Path.GetRelativePath(
                         Path.GetDirectoryName(Path.GetFullPath(outputMetadataPath!))!,
                         Path.GetFullPath(productionSourceBundlePath)).Replace('\\', '/');
-                var metadataUnsupportedInstructionCount = emissionAddresses.Sum(address =>
-                {
-                    if (!canonicalIrStore.TryGet(address, out var canonicalSsa))
-                    {
-                        throw new InvalidOperationException(
-                            $"Missing canonical SSA while computing translation quality for 0x{address:X8}.");
-                    }
-                    return canonicalSsa.Blocks.Sum(block =>
-                        block.Instructions.Count(static instruction => instruction is IrUndefined));
-                });
+                var metadataUnsupportedInstructionCount = unsupportedByIndex.Sum();
                 completedOutputMetadata = BaseTranslationOutputMetadata.Create(
                     outputMetadata,
                     new TranslationQualityMetadata(
@@ -1701,7 +1721,7 @@ int RunTranslateModCore(string[] argsTail, string? outputDirectoryOverride)
     // win (this shipped once as a character-select LinkList.h:573 panic).
     if (File.Exists(baseTranslationOutputMetadataPath) && File.Exists(codePul))
     {
-        var guardedBaseMetadata = BaseTranslationOutputMetadataFile.Read(baseTranslationOutputMetadataPath);
+        var guardedBaseMetadata = ReadBaseMetadataCached(baseTranslationOutputMetadataPath);
         var codePulSha256 = ChecksumUtilities.Sha256HexOfFile(codePul);
         var baseKnowsThisPul = guardedBaseMetadata.ModPatchAwareness is { } awareness &&
             awareness.Any(entry => string.Equals(entry.CodePulSha256, codePulSha256, StringComparison.OrdinalIgnoreCase));
@@ -1780,6 +1800,8 @@ int RunTranslateModCore(string[] argsTail, string? outputDirectoryOverride)
 
     var pul = KamekPulFile.Load(codePul);
     var selected = pul.SelectRegion(region);
+    // The server validates the hash of the unmodified chunk.
+    var sourceKamekCodeSha1 = SHA1.HashData(selected.CodeBlob);
     if (string.Equals(profile?.Name, "retro-rewind", StringComparison.OrdinalIgnoreCase))
     {
         var corrected = pul.IsCombined && !string.Equals(region, "P", StringComparison.OrdinalIgnoreCase)
@@ -1988,7 +2010,7 @@ int RunTranslateModCore(string[] argsTail, string? outputDirectoryOverride)
         baseManifest,
         dvdOverlayRoots,
         relocatedModuleImage,
-        SHA1.HashData(selected.CodeBlob),
+        sourceKamekCodeSha1,
         retroWfcLoweringPlan?.StaticPointers,
         selected.CodeSize,
         selected.BssSize,
@@ -2077,7 +2099,7 @@ void WriteResolvedDispatchProfile(
         var metadataPath = baseTranslationOutputMetadataPath;
         if (File.Exists(metadataPath))
         {
-            var metadata = BaseTranslationOutputMetadataFile.Read(metadataPath);
+            var metadata = ReadBaseMetadataCached(metadataPath);
             if (!string.IsNullOrWhiteSpace(metadata.SourceBundlePath))
             {
                 var bundlePath = Path.IsPathRooted(metadata.SourceBundlePath)
@@ -2152,7 +2174,7 @@ void WriteResolvedDispatchProfile(
         BaseTranslationOutputMetadata? metadata = null;
         if (File.Exists(baseTranslationOutputMetadataPath))
         {
-            metadata = BaseTranslationOutputMetadataFile.Read(baseTranslationOutputMetadataPath);
+            metadata = ReadBaseMetadataCached(baseTranslationOutputMetadataPath);
         }
         if (metadata is null || metadata.Functions.Any(static function => function.Build is null))
         {
@@ -2256,12 +2278,21 @@ void WriteResolvedDispatchProfile(
     // functions the runtime must own (hardware, the strap scene). But when the mod's whole point
     // is replacing that function, e.g. Retro Rewind's DVDOpen override at 0x8015E2BC, the overlay
     // is silently dropped. Name every such address so the tradeoff is a choice, not an accident.
+    // Wrapper natives still run the overlay, and replacement natives already made that choice.
+    var nativesKeepingModOverlayDecision = runtimeNativeIndex.Value.Registrations
+        .Where(static registration => registration.ModOverlay != NativeModOverlay.Unspecified)
+        .Select(static registration => (registration.Address, registration.Symbol))
+        .ToHashSet();
     var nativeOverriddenModPatches = groups
         .Where(group =>
-            group.Any(entry => entry.Kind == "rr") &&
-            group.OrderByDescending(entry => entry.Priority)
-                 .ThenByDescending(entry => KindRank(entry.Kind))
-                 .First().Kind == "native")
+        {
+            var winner = group.OrderByDescending(entry => entry.Priority)
+                .ThenByDescending(entry => KindRank(entry.Kind))
+                .First();
+            return group.Any(entry => entry.Kind == "rr") &&
+                   winner.Kind == "native" &&
+                   !nativesKeepingModOverlayDecision.Contains((winner.Address, winner.Symbol));
+        })
         .Select(group => new
         {
             Address = group.Key,
@@ -2324,12 +2355,9 @@ int EmitModCpp(
         : new CompositeGuestFunctionAbiProvider(inferredAbi);
     var modTranslator = new FunctionTranslator(modImage, abiProvider);
     var parallelOptions = new ParallelOptions { MaxDegreeOfParallelism = maxThreads };
-    // Mod emission waves: each item is a full translate+emit, and each wave is a
-    // barrier that also drains newly discovered continuations, so it is
-    // deliberately smaller than the discovery batch above - a long wave delays
-    // the discoveries the next wave has to translate.
-    var waveSize = Math.Max(16, maxThreads * 4);
-    Console.WriteLine($"  translating mod C++ with up to {maxThreads} worker thread(s) in waves of {waveSize}");
+    // Each wave drains its whole queue: a wave is a barrier that publishes newly discovered
+    // entries, and draining everything makes the output independent of any wave size.
+    Console.WriteLine($"  translating mod C++ with up to {maxThreads} worker thread(s)");
 
     var cppRoot = Path.Combine(outDir, "cpp");
     if (Directory.Exists(cppRoot))
@@ -2408,7 +2436,7 @@ int EmitModCpp(
         // Some dynamically discovered module/continuation waves bypass
         // ProcessBoundedWaves. Prewarm every parallel emission boundary as a
         // final deterministic guard; phase-prewarmed entries are cheap hits.
-        ReportAbiPrewarm(inferredAbi.Prewarm(wave.Select(static work => work.Address)));
+        ReportAbiPrewarm(inferredAbi.Prewarm(wave.Select(static work => work.Address), maxThreads));
         return ModTranslationWaveRunner.Translate(() => modTranslator, wave, parallelOptions);
     }
 
@@ -2445,7 +2473,7 @@ int EmitModCpp(
         // Populate the recursive ABI cache in a stable global order before any worker emits C++,
         // so shared closure reuse and cyclic call graphs never depend on thread count, wave size,
         // or worker order.
-        ReportAbiPrewarm(inferredAbi.Prewarm(ordered.Select(static work => work.Address)));
+        ReportAbiPrewarm(inferredAbi.Prewarm(ordered.Select(static work => work.Address), maxThreads));
         // Commit callbacks here only append to the discovered-continuation queue that a later
         // phase drains, so nothing in this phase reads back its own commits; wave barriers would
         // only cost idle workers, so it runs as one parallel batch and the runner re-sorts by the
@@ -2590,7 +2618,7 @@ int EmitModCpp(
     {
         while (discoveredContinuationQueue.Count > 0)
         {
-            var wave = QueueBatch.Dequeue(discoveredContinuationQueue, waveSize)
+            var wave = QueueBatch.Dequeue(discoveredContinuationQueue, discoveredContinuationQueue.Count)
                 .Select(CreateContinuationWork)
                 .ToArray();
             CommitWave(TranslateWave(wave), (result, work) =>
@@ -2609,10 +2637,10 @@ int EmitModCpp(
     var emittedModuleStarts = new HashSet<uint>();
     while (moduleQueue.Count > 0)
     {
-        var wave = new List<ModTranslationWork>(waveSize);
+        var wave = new List<ModTranslationWork>(moduleQueue.Count);
         foreach (var start in QueueBatch.Dequeue(
                      moduleQueue,
-                     waveSize,
+                     moduleQueue.Count,
                      start => !emittedModuleStarts.Add(start.Address)))
         {
             var name = ModuleFunctionName(start.Address);

translator/src/Translator.Cli/Translator.Cli.csproj

@@ -8,6 +8,7 @@
     <TreatWarningsAsErrors>true</TreatWarningsAsErrors>
     <ServerGarbageCollection>true</ServerGarbageCollection>
     <ConcurrentGarbageCollection>true</ConcurrentGarbageCollection>
+    <TieredPGO>false</TieredPGO>
     <VersionPrefix>0.5.0</VersionPrefix>
     <Company>WiiCompiled</Company>
     <Product>Mario Kart Wii Static Translator</Product>

translator/src/Translator.Core/CodeGen/GuestMemoryRangeLowering.cs

@@ -102,12 +102,7 @@ public static class GuestMemoryRangeLowering
             .Any(group => group.Count() >= minimumAccesses))
             return function;
 
-        var cfg = IrCfg.Build(function);
-        var dominators = ComputeDominators(cfg);
-        var reachability = ComputeReachability(cfg);
-        var cyclicBlocks = cfg.Blocks.Keys.Where(label =>
-                cfg.Successors(label).Any(successor => reachability[successor].Contains(label)))
-            .ToHashSet(StringComparer.OrdinalIgnoreCase);
+        var graph = BlockGraph.Build(function);
 
         var plans = new Dictionary<(int Block, int Instruction), Plan>();
         var resolves = new Dictionary<int, List<PlacedResolve>>();
@@ -129,24 +124,35 @@ public static class GuestMemoryRangeLowering
                 if (definitionSites.Count != 1) continue;
                 definition = definitionSites[0];
                 var defined = definitionSites[0];
-                var definitionBlock = function.Blocks[defined.BlockIndex].Label;
                 if (remaining.Any(candidate =>
-                        !dominators[candidate.AccessBlockLabel(function)].Contains(definitionBlock) ||
+                        !graph.Dominates(defined.BlockIndex, candidate.BlockIndex) ||
                         candidate.BlockIndex == defined.BlockIndex &&
                         candidate.Access.Index <= defined.InstructionIndex))
                     continue;
             }
+            // The pair predicate is invariant across rounds, so memoize it.
+            var groupIndex = new Dictionary<FunctionAccess, int>(ReferenceEqualityComparer.Instance);
+            foreach (var access in remaining) groupIndex.Add(access, groupIndex.Count);
+            var compatible = remaining.Count <= 4096 ? new sbyte[remaining.Count * remaining.Count] : null;
+            bool Compatible(FunctionAccess anchor, FunctionAccess candidate)
+            {
+                if (compatible is null)
+                    return DominatesAccess(graph, anchor, candidate) &&
+                        !HasInvalidatingBoundaryOnPath(graph, invalidators, anchor, candidate);
+                var slot = groupIndex[anchor] * groupIndex.Count + groupIndex[candidate];
+                if (compatible[slot] != 0) return compatible[slot] > 0;
+                var value = DominatesAccess(graph, anchor, candidate) &&
+                    !HasInvalidatingBoundaryOnPath(graph, invalidators, anchor, candidate);
+                compatible[slot] = value ? (sbyte)1 : (sbyte)-1;
+                return value;
+            }
             while (remaining.Count >= minimumAccesses)
             {
                 FunctionAccess? bestAnchor = null;
                 FunctionAccess[] bestMembers = [];
                 foreach (var anchor in remaining)
                 {
-                    var members = remaining.Where(candidate =>
-                            DominatesAccess(function, dominators, anchor, candidate) &&
-                            !HasInvalidatingBoundaryOnPath(
-                                function, reachability, cyclicBlocks,
-                                invalidators, anchor, candidate))
+                    var members = remaining.Where(candidate => Compatible(anchor, candidate))
                         .OrderBy(candidate => candidate.Access.Index)
                         .ToList();
                     // Keep a bounded affine window around the real anchor. A
@@ -248,15 +254,13 @@ public static class GuestMemoryRangeLowering
     }
 
     private static bool DominatesAccess(
-        IrFunction function,
-        IReadOnlyDictionary<string, HashSet<string>> dominators,
+        BlockGraph graph,
         FunctionAccess anchor,
         FunctionAccess consumer)
     {
         if (anchor.BlockIndex == consumer.BlockIndex)
             return anchor.Access.Index <= consumer.Access.Index;
-        return dominators[consumer.AccessBlockLabel(function)].Contains(
-            anchor.AccessBlockLabel(function));
+        return graph.Dominates(anchor.BlockIndex, consumer.BlockIndex);
     }
 
     private static bool HasOverlappingOppositeAccess(
@@ -273,27 +277,25 @@ public static class GuestMemoryRangeLowering
     }
 
     private static bool HasInvalidatingBoundaryOnPath(
-        IrFunction function,
-        IReadOnlyDictionary<string, HashSet<string>> reachability,
-        IReadOnlySet<string> cyclicBlocks,
+        BlockGraph graph,
         IReadOnlyList<InstructionSite> invalidators,
         FunctionAccess anchor,
         FunctionAccess consumer)
     {
-        var anchorLabel = anchor.AccessBlockLabel(function);
-        var consumerLabel = consumer.AccessBlockLabel(function);
+        var anchorBlock = anchor.BlockIndex;
+        var consumerBlock = consumer.BlockIndex;
+        var consumerReentered = anchorBlock != consumerBlock && graph.Cyclic[consumerBlock];
         foreach (var invalidator in invalidators)
         {
-            var boundaryLabel = function.Blocks[invalidator.BlockIndex].Label;
-            var afterAnchor = invalidator.BlockIndex == anchor.BlockIndex
+            var boundaryBlock = invalidator.BlockIndex;
+            var afterAnchor = boundaryBlock == anchorBlock
                 ? invalidator.InstructionIndex > anchor.Access.Index
-                : reachability[anchorLabel].Contains(boundaryLabel);
-            var beforeConsumer = invalidator.BlockIndex == consumer.BlockIndex
-                ? invalidator.InstructionIndex < consumer.Access.Index ||
-                  anchor.BlockIndex != consumer.BlockIndex &&
-                  cyclicBlocks.Contains(consumerLabel)
-                : reachability[boundaryLabel].Contains(consumerLabel);
-            if (afterAnchor && beforeConsumer) return true;
+                : graph.Reaches(anchorBlock, boundaryBlock);
+            if (!afterAnchor) continue;
+            var beforeConsumer = boundaryBlock == consumerBlock
+                ? invalidator.InstructionIndex < consumer.Access.Index || consumerReentered
+                : graph.Reaches(boundaryBlock, consumerBlock);
+            if (beforeConsumer) return true;
         }
         return false;
     }
@@ -315,54 +317,117 @@ public static class GuestMemoryRangeLowering
         return result;
     }
 
-    private static Dictionary<string, HashSet<string>> ComputeDominators(IrCfg cfg)
+    /// <summary>Dominator and reachability sets as bitsets over block ordinals.</summary>
+    private sealed class BlockGraph
     {
-        var comparer = StringComparer.OrdinalIgnoreCase;
-        var labels = cfg.Blocks.Keys.ToArray();
-        var all = labels.ToHashSet(comparer);
-        var result = labels.ToDictionary(label => label,
-            label => label.Equals(cfg.Entry, StringComparison.OrdinalIgnoreCase)
-                ? new HashSet<string>([label], comparer)
-                : new HashSet<string>(all, comparer), comparer);
-        bool changed;
-        do
-        {
-            changed = false;
-            foreach (var label in labels)
-            {
-                if (label.Equals(cfg.Entry, StringComparison.OrdinalIgnoreCase)) continue;
-                var predecessors = cfg.Predecessors(label);
-                var next = predecessors.Count == 0
-                    ? new HashSet<string>(comparer)
-                    : new HashSet<string>(result[predecessors[0]], comparer);
-                foreach (var predecessor in predecessors.Skip(1)) next.IntersectWith(result[predecessor]);
-                next.Add(label);
-                if (result[label].SetEquals(next)) continue;
-                result[label] = next;
-                changed = true;
-            }
-        } while (changed);
-        return result;
-    }
+        private readonly ulong[][] _dominators;
+        private readonly ulong[][] _reach;
 
-    private static Dictionary<string, HashSet<string>> ComputeReachability(IrCfg cfg)
-    {
-        var comparer = StringComparer.OrdinalIgnoreCase;
-        var result = new Dictionary<string, HashSet<string>>(comparer);
-        foreach (var start in cfg.Blocks.Keys)
+        private BlockGraph(ulong[][] dominators, ulong[][] reach, bool[] cyclic)
         {
-            var reachable = new HashSet<string>(comparer);
-            var pending = new Stack<string>();
-            pending.Push(start);
-            while (pending.Count != 0)
-            {
-                var current = pending.Pop();
-                if (!reachable.Add(current)) continue;
-                foreach (var successor in cfg.Successors(current)) pending.Push(successor);
-            }
-            result[start] = reachable;
+            _dominators = dominators;
+            _reach = reach;
+            Cyclic = cyclic;
+        }
+
+        /// <summary>Blocks that can reach themselves again through a successor.</summary>
+        public bool[] Cyclic { get; }
+
+        public bool Dominates(int dominator, int block) =>
+            (_dominators[block][dominator >> 6] & (1UL << (dominator & 63))) != 0;
+
+        /// <summary>Reflexive: every block reaches itself.</summary>
+        public bool Reaches(int from, int to) =>
+            (_reach[from][to >> 6] & (1UL << (to & 63))) != 0;
+
+        public static BlockGraph Build(IrFunction function)
+        {
+            var cfg = IrCfg.Build(function);
+            var count = function.Blocks.Count;
+            var words = (count + 63) >> 6;
+            var index = new Dictionary<string, int>(count, StringComparer.OrdinalIgnoreCase);
+            for (var i = 0; i < count; i++) index.Add(function.Blocks[i].Label, i);
+            var successors = new int[count][];
+            var predecessors = new int[count][];
+            for (var i = 0; i < count; i++)
+            {
+                var label = function.Blocks[i].Label;
+                successors[i] = cfg.Successors(label).Select(successor => index[successor]).ToArray();
+                predecessors[i] = cfg.Predecessors(label).Select(predecessor => index[predecessor]).ToArray();
+            }
+
+            var reach = new ulong[count][];
+            var pending = new Stack<int>();
+            for (var start = 0; start < count; start++)
+            {
+                var row = new ulong[words];
+                pending.Push(start);
+                while (pending.Count != 0)
+                {
+                    var current = pending.Pop();
+                    ref var word = ref row[current >> 6];
+                    var bit = 1UL << (current & 63);
+                    if ((word & bit) != 0) continue;
+                    word |= bit;
+                    foreach (var successor in successors[current]) pending.Push(successor);
+                }
+                reach[start] = row;
+            }
+
+            var cyclic = new bool[count];
+            for (var i = 0; i < count; i++)
+            {
+                foreach (var successor in successors[i])
+                {
+                    if ((reach[successor][i >> 6] & (1UL << (i & 63))) == 0) continue;
+                    cyclic[i] = true;
+                    break;
+                }
+            }
+
+            var entry = index[cfg.Entry];
+            var dominators = new ulong[count][];
+            for (var i = 0; i < count; i++)
+            {
+                dominators[i] = new ulong[words];
+                if (i == entry)
+                {
+                    dominators[i][i >> 6] = 1UL << (i & 63);
+                    continue;
+                }
+                Array.Fill(dominators[i], ulong.MaxValue);
+                if ((count & 63) != 0) dominators[i][words - 1] = (1UL << (count & 63)) - 1;
+            }
+            var next = new ulong[words];
+            bool changed;
+            do
+            {
+                changed = false;
+                for (var i = 0; i < count; i++)
+                {
+                    if (i == entry) continue;
+                    var preds = predecessors[i];
+                    if (preds.Length == 0)
+                    {
+                        Array.Clear(next);
+                    }
+                    else
+                    {
+                        Array.Copy(dominators[preds[0]], next, words);
+                        for (var p = 1; p < preds.Length; p++)
+                        {
+                            var other = dominators[preds[p]];
+                            for (var w = 0; w < words; w++) next[w] &= other[w];
+                        }
+                    }
+                    next[i >> 6] |= 1UL << (i & 63);
+                    if (next.AsSpan().SequenceEqual(dominators[i])) continue;
+                    next.AsSpan().CopyTo(dominators[i]);
+                    changed = true;
+                }
+            } while (changed);
+            return new BlockGraph(dominators, reach, cyclic);
         }
-        return result;
     }
 
     private static IrInstruction ToResolved(IrInstruction instruction, Access access, Plan plan)

translator/src/Translator.Core/CodeGen/InferredGuestFunctionAbiProvider.cs

@@ -3,6 +3,7 @@ using System.Collections.Concurrent;
 using System.Collections.Generic;
 using System.Globalization;
 using System.Linq;
+using System.Numerics;
 using System.Threading;
 using System.Threading.Tasks;
 using Translator.Core.Analysis;
@@ -72,6 +73,9 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
         /// <summary>Reusable backing list for <see cref="FrameTaint"/>; phase A is never reentrant.</summary>
         public readonly List<bool> FrameTaintBuffer = new();
 
+        /// <summary>Address per <see cref="FrameTaint"/> entry, innermost last.</summary>
+        public readonly List<uint> FrameAddresses = new();
+
         /// <summary>Addresses already known tainted, shared by every phase-A worker of the current Prewarm
         /// call and reset each invocation. Without this memoization an unpublishable address gets
         /// re-analyzed, subtree and all, by every caller that reaches it, which compounds combinatorially
@@ -95,14 +99,21 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
             return false;
         }
 
+        var state = _threadState.Value!;
+        var frameTaint = state.FrameTaint;
+        if (frameTaint is not null && state.FrameAddresses.Count > 0 && state.FrameAddresses[^1] == address)
+        {
+            // Direct self-recursion sees what a sequential run would, even once another thread has published it.
+            abi = GuestFunctionAbi.Empty;
+            return false;
+        }
+
         if (_cache.TryGetValue(address, out abi!))
         {
             return true;
         }
 
-        var state = _threadState.Value!;
         var inProgress = state.InProgress;
-        var frameTaint = state.FrameTaint;
 
         if (frameTaint is not null && state.TaintedAddresses!.ContainsKey(address))
         {
@@ -151,6 +162,7 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
 
         // Prewarm phase A: track taint for this frame and publish only if it stayed clean.
         frameTaint.Add(false);
+        state.FrameAddresses.Add(address);
         var frameIndex = frameTaint.Count - 1;
         bool succeeded;
         bool tainted;
@@ -172,6 +184,7 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
         finally
         {
             inProgress.Remove(address);
+            state.FrameAddresses.RemoveAt(frameIndex);
             tainted = frameTaint[frameIndex];
             frameTaint.RemoveAt(frameIndex);
             if (tainted && frameIndex > 0)
@@ -274,6 +287,7 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
         var state = _threadState.Value!;
         var frames = state.FrameTaintBuffer;
         frames.Clear();
+        state.FrameAddresses.Clear();
         state.TaintedAddresses = taintedAddresses;
         state.FrameTaint = frames;
         try
@@ -418,82 +432,102 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
     private HashSet<string> InferArgumentRegisters(IrFunction function, IrCfg cfg)
     {
         var comparer = StringComparer.OrdinalIgnoreCase;
-        var blockUses = new Dictionary<string, HashSet<string>>(comparer);
-        var blockDefs = new Dictionary<string, HashSet<string>>(comparer);
-
-        foreach (var block in function.Blocks)
+        // Interned register ids so the liveness fixpoint runs over bitsets.
+        var ids = new Dictionary<string, int>(comparer);
+        var names = new List<string>();
+        int Id(string name)
         {
-            var uses = new HashSet<string>(comparer);
-            var defs = new HashSet<string>(comparer);
+            if (!ids.TryGetValue(name, out var id))
+            {
+                id = names.Count;
+                ids.Add(name, id);
+                names.Add(name);
+            }
+            return id;
+        }
+
+        var blockCount = function.Blocks.Count;
+        var useLists = new List<int>[blockCount];
+        var defLists = new List<int>[blockCount];
+        var labelIndex = new Dictionary<string, int>(blockCount, comparer);
+        for (var b = 0; b < blockCount; b++)
+        {
+            var block = function.Blocks[b];
+            var uses = new List<int>();
+            var defs = new List<int>();
+            var defined = new HashSet<int>();
             foreach (var instruction in block.Instructions)
             {
                 foreach (var use in InstructionRegisterUsesForArgumentInference(instruction))
                 {
-                    if (!defs.Contains(use))
-                    {
-                        uses.Add(use);
-                    }
+                    var id = Id(use);
+                    if (!defined.Contains(id))
+                        uses.Add(id);
                 }
 
                 foreach (var def in InstructionRegisterDefsForArgumentInference(instruction))
                 {
-                    defs.Add(def);
+                    var id = Id(def);
+                    if (defined.Add(id))
+                        defs.Add(id);
                 }
             }
-
-            blockUses[block.Label] = uses;
-            blockDefs[block.Label] = defs;
+            useLists[b] = uses;
+            defLists[b] = defs;
         }
 
-        var liveIn = function.Blocks.ToDictionary(
-            static block => block.Label,
-            _ => new HashSet<string>(comparer),
-            comparer);
-        var liveOut = function.Blocks.ToDictionary(
-            static block => block.Label,
-            _ => new HashSet<string>(comparer),
-            comparer);
+        for (var b = 0; b < blockCount; b++)
+            labelIndex.Add(function.Blocks[b].Label, b);
 
+        var words = Math.Max(1, (names.Count + 63) >> 6);
+        var useBits = new ulong[blockCount * words];
+        var keepBits = new ulong[blockCount * words];
+        Array.Fill(keepBits, ulong.MaxValue);
+        var successors = new int[blockCount][];
+        for (var b = 0; b < blockCount; b++)
+        {
+            foreach (var id in useLists[b]) useBits[b * words + (id >> 6)] |= 1UL << (id & 63);
+            foreach (var id in defLists[b]) keepBits[b * words + (id >> 6)] &= ~(1UL << (id & 63));
+            var targets = new List<int>();
+            foreach (var successor in cfg.Successors(function.Blocks[b].Label))
+                if (labelIndex.TryGetValue(successor, out var index))
+                    targets.Add(index);
+            successors[b] = targets.ToArray();
+        }
+
+        var liveIn = new ulong[blockCount * words];
         var changed = true;
         while (changed)
         {
             changed = false;
-            for (var blockIndex = function.Blocks.Count - 1; blockIndex >= 0; blockIndex--)
+            for (var b = blockCount - 1; b >= 0; b--)
             {
-                var block = function.Blocks[blockIndex];
-                var newOut = new HashSet<string>(comparer);
-                foreach (var successor in cfg.Successors(block.Label))
+                var row = b * words;
+                for (var w = 0; w < words; w++)
                 {
-                    if (liveIn.TryGetValue(successor, out var successorLiveIn))
+                    ulong liveOut = 0;
+                    foreach (var successor in successors[b])
+                        liveOut |= liveIn[successor * words + w];
+                    var value = useBits[row + w] | (liveOut & keepBits[row + w]);
+                    if (value != liveIn[row + w])
                     {
-                        newOut.UnionWith(successorLiveIn);
+                        liveIn[row + w] = value;
+                        changed = true;
                     }
                 }
-
-                var newIn = new HashSet<string>(blockUses[block.Label], comparer);
-                var outMinusDefs = new HashSet<string>(newOut, comparer);
-                outMinusDefs.ExceptWith(blockDefs[block.Label]);
-                newIn.UnionWith(outMinusDefs);
-
-                if (!newOut.SetEquals(liveOut[block.Label]))
-                {
-                    liveOut[block.Label] = newOut;
-                    changed = true;
-                }
-
-                if (!newIn.SetEquals(liveIn[block.Label]))
-                {
-                    liveIn[block.Label] = newIn;
-                    changed = true;
-                }
             }
         }
 
-        return liveIn.TryGetValue(function.EntryLabel, out var entryLive)
-            ? entryLive
-                .Where(IsGprOrFprRegister)
-                .ToHashSet(comparer)
-            : new HashSet<string>(comparer);
+        var result = new HashSet<string>(comparer);
+        if (labelIndex.TryGetValue(function.EntryLabel, out var entry))
+        {
+            for (var id = 0; id < names.Count; id++)
+            {
+                if ((liveIn[entry * words + (id >> 6)] & (1UL << (id & 63))) != 0 && IsGprOrFprRegister(names[id]))
+                    result.Add(names[id]);
+            }
+        }
+        return result;
     }
 
     private IEnumerable<string> InstructionRegisterUsesForArgumentInference(IrInstruction instruction)
@@ -755,45 +789,74 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
     private HashSet<string> InferScalarFloatArguments(IrFunction function, IrCfg cfg)
     {
         var scalarArgs = new HashSet<string>(StringComparer.OrdinalIgnoreCase);
-        var comparer = StringComparer.OrdinalIgnoreCase;
-        var inMap = function.Blocks.ToDictionary(
-            static b => b.Label,
-            b => b.Label.Equals(function.EntryLabel, StringComparison.OrdinalIgnoreCase)
-                ? CreateInitialArgumentAliasState()
-                : new Dictionary<string, HashSet<string>>(comparer),
-            comparer);
-        var outMap = function.Blocks.ToDictionary(
-            static b => b.Label,
-            b => b.Label.Equals(function.EntryLabel, StringComparison.OrdinalIgnoreCase)
-                ? CreateInitialArgumentAliasState()
-                : new Dictionary<string, HashSet<string>>(comparer),
-            comparer);
+        var keys = new AliasKeys(function);
+        var blockCount = function.Blocks.Count;
+        var blockIndex = new Dictionary<string, int>(blockCount, StringComparer.OrdinalIgnoreCase);
+        for (var b = 0; b < blockCount; b++)
+            blockIndex[function.Blocks[b].Label] = b;
+        var predecessors = new int[blockCount][];
+        var isEntry = new bool[blockCount];
+        for (var b = 0; b < blockCount; b++)
+        {
+            var block = function.Blocks[b];
+            isEntry[b] = block.Label.Equals(function.EntryLabel, StringComparison.OrdinalIgnoreCase);
+            var list = new List<int>();
+            foreach (var predecessor in cfg.Predecessors(block.Label))
+                if (blockIndex.TryGetValue(predecessor, out var index))
+                    list.Add(index);
+            predecessors[b] = list.ToArray();
+        }
+
+        // Per alias key: -1 when absent, otherwise the mask of ABI float argument sources.
+        var initial = keys.InitialState();
+        var inMap = new int[blockCount][];
+        var outMap = new int[blockCount][];
+        for (var b = 0; b < blockCount; b++)
+        {
+            inMap[b] = isEntry[b] ? (int[])initial.Clone() : keys.EmptyState();
+            outMap[b] = isEntry[b] ? (int[])initial.Clone() : keys.EmptyState();
+        }
 
         var changed = true;
         while (changed)
         {
             changed = false;
-            foreach (var block in function.Blocks)
+            for (var b = 0; b < blockCount; b++)
             {
-                var newIn = block.Label.Equals(function.EntryLabel, StringComparison.OrdinalIgnoreCase)
-                    ? CreateInitialArgumentAliasState()
-                    : UnionPredecessorStates(cfg.Predecessors(block.Label), outMap);
-
-                var newOut = CopyAliasState(newIn);
-                foreach (var ins in block.Instructions)
+                int[] newIn;
+                if (isEntry[b])
                 {
-                    ProcessScalarArgumentInstruction(ins, newOut, scalarArgs);
+                    newIn = (int[])initial.Clone();
+                }
+                else
+                {
+                    newIn = keys.EmptyState();
+                    foreach (var predecessor in predecessors[b])
+                    {
+                        var predState = outMap[predecessor];
+                        for (var key = 0; key < newIn.Length; key++)
+                        {
+                            if (predState[key] < 0) continue;
+                            newIn[key] = newIn[key] < 0 ? predState[key] : newIn[key] | predState[key];
+                        }
+                    }
                 }
 
-                if (!AliasStatesEqual(newIn, inMap[block.Label]))
+                var newOut = (int[])newIn.Clone();
+                foreach (var ins in function.Blocks[b].Instructions)
                 {
-                    inMap[block.Label] = newIn;
+                    ProcessScalarArgumentInstruction(ins, keys, newOut, scalarArgs);
+                }
+
+                if (!newIn.AsSpan().SequenceEqual(inMap[b]))
+                {
+                    inMap[b] = newIn;
                     changed = true;
                 }
 
-                if (!AliasStatesEqual(newOut, outMap[block.Label]))
+                if (!newOut.AsSpan().SequenceEqual(outMap[b]))
                 {
-                    outMap[block.Label] = newOut;
+                    outMap[b] = newOut;
                     changed = true;
                 }
             }
@@ -802,63 +865,153 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
         return scalarArgs;
     }
 
+    /// <summary>Interned float register names the scalar-argument alias analysis can track.
+    /// Ids 0..12 are the ABI float argument registers, so a source set is a 13-bit mask.</summary>
+    private sealed class AliasKeys
+    {
+        private readonly Dictionary<string, int> _ids = new(StringComparer.OrdinalIgnoreCase);
+
+        public AliasKeys(IrFunction function)
+        {
+            foreach (var register in AbiFloatArgumentRegisters)
+                _ids.Add(register, _ids.Count);
+            foreach (var block in function.Blocks)
+            {
+                foreach (var instruction in block.Instructions)
+                {
+                    switch (instruction)
+                    {
+                        case IrAssign assign:
+                            AddName(assign.Destination);
+                            AddValue(assign.Value);
+                            break;
+                        case IrBinary binary:
+                            AddValue(binary.Left);
+                            AddValue(binary.Right);
+                            AddName(binary.Destination);
+                            break;
+                        case IrLoad load:
+                            AddName(load.Destination);
+                            break;
+                        case IrStore store:
+                            AddValue(store.Source);
+                            break;
+                        case IrCall call:
+                            foreach (var argument in call.Arguments) AddValue(argument);
+                            AddName(call.Destination);
+                            break;
+                        case IrIndirectCall indirectCall:
+                            foreach (var argument in indirectCall.Arguments) AddValue(argument);
+                            AddName(indirectCall.Destination);
+                            break;
+                        case IrSetCrField setCr:
+                            AddValue(setCr.Left);
+                            AddValue(setCr.Right);
+                            break;
+                        case IrPhi phi:
+                            AddName(phi.Destination);
+                            break;
+                    }
+                }
+            }
+        }
+
+        public int Count => _ids.Count;
+
+        public int[] EmptyState()
+        {
+            var state = new int[_ids.Count];
+            Array.Fill(state, -1);
+            return state;
+        }
+
+        public int[] InitialState()
+        {
+            var state = EmptyState();
+            for (var i = 0; i < AbiFloatArgumentRegisters.Length; i++)
+                state[i] = 1 << i;
+            return state;
+        }
+
+        public bool TryGet(string baseName, out int id) => _ids.TryGetValue(baseName, out id);
+
+        private void AddValue(IrValue? value)
+        {
+            if (value is { Kind: "register", RegisterName: { } name })
+                AddName(name);
+        }
+
+        private void AddName(string? name)
+        {
+            if (string.IsNullOrEmpty(name)) return;
+            var baseName = Base(name);
+            if (IsFloatRegister(baseName))
+                _ids.TryAdd(baseName, _ids.Count);
+        }
+    }
+
     private void ProcessScalarArgumentInstruction(
         IrInstruction ins,
-        Dictionary<string, HashSet<string>> aliases,
+        AliasKeys keys,
+        int[] aliases,
         HashSet<string> scalarArgs)
     {
         switch (ins)
         {
             case IrAssign assign:
-                PropagateFloatAlias(assign.Destination, assign.Value, aliases);
+                PropagateFloatAlias(assign.Destination, assign.Value, keys, aliases);
                 break;
 
             case IrBinary bin:
-                MarkScalarUse(bin.Left, aliases, scalarArgs);
-                MarkScalarUse(bin.Right, aliases, scalarArgs);
-                KillFloatAlias(bin.Destination, aliases);
+                MarkScalarUse(bin.Left, keys, aliases, scalarArgs);
+                MarkScalarUse(bin.Right, keys, aliases, scalarArgs);
+                KillFloatAlias(bin.Destination, keys, aliases);
                 break;
 
             case IrLoad load:
-                KillFloatAlias(load.Destination, aliases);
+                KillFloatAlias(load.Destination, keys, aliases);
                 break;
 
             case IrStore store:
                 if (store.SizeBytes is 4 or 8)
                 {
-                    MarkScalarUse(store.Source, aliases, scalarArgs);
+                    MarkScalarUse(store.Source, keys, aliases, scalarArgs);
                 }
                 break;
 
             case IrCall call:
-                ProcessScalarArgumentCall(call.Target, call.Arguments, aliases, scalarArgs);
-                KillFloatAlias(call.Destination, aliases);
+                ProcessScalarArgumentCall(call.Target, call.Arguments, keys, aliases, scalarArgs);
+                KillFloatAlias(call.Destination, keys, aliases);
                 break;
 
             case IrIndirectCall icall:
                 foreach (var arg in icall.Arguments)
                 {
-                    MarkScalarUse(arg, aliases, scalarArgs);
+                    MarkScalarUse(arg, keys, aliases, scalarArgs);
                 }
-                aliases.Remove("f1");
-                KillFloatAlias(icall.Destination, aliases);
+                aliases[F1AliasKey] = -1;
+                KillFloatAlias(icall.Destination, keys, aliases);
                 break;
 
             case IrSetCrField setCr:
-                MarkScalarUse(setCr.Left, aliases, scalarArgs);
-                MarkScalarUse(setCr.Right, aliases, scalarArgs);
+                MarkScalarUse(setCr.Left, keys, aliases, scalarArgs);
+                MarkScalarUse(setCr.Right, keys, aliases, scalarArgs);
                 break;
 
             case IrPhi phi:
-                KillFloatAlias(phi.Destination, aliases);
+                KillFloatAlias(phi.Destination, keys, aliases);
                 break;
         }
     }
 
+    /// <summary>Alias key of f1, the first ABI float argument register.</summary>
+    private const int F1AliasKey = 0;
+
     private void ProcessScalarArgumentCall(
         string target,
         IReadOnlyList<IrValue> arguments,
-        Dictionary<string, HashSet<string>> aliases,
+        AliasKeys keys,
+        int[] aliases,
         HashSet<string> scalarArgs)
     {
         if (IsGuestCallTarget(target))
@@ -873,14 +1026,14 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
                         arg.RegisterName != null &&
                         abi.HasScalarFloatArgument(Base(arg.RegisterName)))
                     {
-                        MarkScalarUse(arg, aliases, scalarArgs);
+                        MarkScalarUse(arg, keys, aliases, scalarArgs);
                     }
                 }
             }
 
             if (writesFloatReturnRegister)
             {
-                aliases.Remove("f1");
+                aliases[F1AliasKey] = -1;
             }
             return;
         }
@@ -892,7 +1045,7 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
 
         foreach (var arg in arguments)
         {
-            MarkScalarUse(arg, aliases, scalarArgs);
+            MarkScalarUse(arg, keys, aliases, scalarArgs);
         }
     }
 
@@ -1083,7 +1236,8 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
 
     private static void MarkScalarUse(
         IrValue value,
-        IReadOnlyDictionary<string, HashSet<string>> aliases,
+        AliasKeys keys,
+        int[] aliases,
         HashSet<string> scalarArgs)
     {
         if (value.Kind != "register" || value.RegisterName == null)
@@ -1092,24 +1246,22 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
         }
 
         var baseName = Base(value.RegisterName);
-        if (!IsFloatRegister(baseName) || !aliases.TryGetValue(baseName, out var sources))
+        if (!IsFloatRegister(baseName) || !keys.TryGet(baseName, out var key) || aliases[key] < 0)
         {
             return;
         }
 
-        foreach (var source in sources)
+        for (var sources = aliases[key]; sources != 0; sources &= sources - 1)
         {
-            if (AbiFloatArgumentSet.Contains(source))
-            {
-                scalarArgs.Add(source);
-            }
+            scalarArgs.Add(AbiFloatArgumentRegisters[BitOperations.TrailingZeroCount(sources)]);
         }
     }
 
     private static void PropagateFloatAlias(
         string destination,
         IrValue value,
-        Dictionary<string, HashSet<string>> aliases)
+        AliasKeys keys,
+        int[] aliases)
     {
         var destBase = Base(destination);
         if (!IsFloatRegister(destBase))
@@ -1117,24 +1269,27 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
             return;
         }
 
+        keys.TryGet(destBase, out var destKey);
         if (value.Kind == "register" &&
             value.RegisterName != null &&
             IsFloatRegister(Base(value.RegisterName)) &&
-            aliases.TryGetValue(Base(value.RegisterName), out var srcAliases))
+            keys.TryGet(Base(value.RegisterName), out var sourceKey) &&
+            aliases[sourceKey] >= 0)
         {
-            aliases[destBase] = new HashSet<string>(srcAliases, StringComparer.OrdinalIgnoreCase);
+            aliases[destKey] = aliases[sourceKey];
         }
         else
         {
-            aliases.Remove(destBase);
+            aliases[destKey] = -1;
         }
     }
 
-    private static void KillFloatAlias(string destination, Dictionary<string, HashSet<string>> aliases)
+    private static void KillFloatAlias(string destination, AliasKeys keys, int[] aliases)
     {
-        if (!string.IsNullOrWhiteSpace(destination) && IsFloatRegister(Base(destination)))
+        if (!string.IsNullOrWhiteSpace(destination) && IsFloatRegister(Base(destination)) &&
+            keys.TryGet(Base(destination), out var key))
         {
-            aliases.Remove(Base(destination));
+            aliases[key] = -1;
         }
     }
 
@@ -1146,70 +1301,6 @@ public sealed class InferredGuestFunctionAbiProvider : IGuestFunctionAbiProvider
         }
     }
 
-    private static Dictionary<string, HashSet<string>> CreateInitialArgumentAliasState()
-    {
-        return AbiFloatArgumentRegisters.ToDictionary(
-            static r => r,
-            static r => new HashSet<string>(StringComparer.OrdinalIgnoreCase) { r },
-            StringComparer.OrdinalIgnoreCase);
-    }
-
-    private static Dictionary<string, HashSet<string>> UnionPredecessorStates(
-        IReadOnlyList<string> predecessors,
-        IReadOnlyDictionary<string, Dictionary<string, HashSet<string>>> outMap)
-    {
-        var result = new Dictionary<string, HashSet<string>>(StringComparer.OrdinalIgnoreCase);
-        foreach (var pred in predecessors)
-        {
-            if (!outMap.TryGetValue(pred, out var predState))
-            {
-                continue;
-            }
-
-            foreach (var (register, sources) in predState)
-            {
-                if (!result.TryGetValue(register, out var destSources))
-                {
-                    destSources = new HashSet<string>(StringComparer.OrdinalIgnoreCase);
-                    result[register] = destSources;
-                }
-
-                destSources.UnionWith(sources);
-            }
-        }
-
-        return result;
-    }
-
-    private static Dictionary<string, HashSet<string>> CopyAliasState(
-        IReadOnlyDictionary<string, HashSet<string>> state)
-    {
-        return state.ToDictionary(
-            static kv => kv.Key,
-            static kv => new HashSet<string>(kv.Value, StringComparer.OrdinalIgnoreCase),
-            StringComparer.OrdinalIgnoreCase);
-    }
-
-    private static bool AliasStatesEqual(
-        IReadOnlyDictionary<string, HashSet<string>> left,
-        IReadOnlyDictionary<string, HashSet<string>> right)
-    {
-        if (left.Count != right.Count)
-        {
-            return false;
-        }
-
-        foreach (var (key, leftSet) in left)
-        {
-            if (!right.TryGetValue(key, out var rightSet) || !leftSet.SetEquals(rightSet))
-            {
-                return false;
-            }
-        }
-
-        return true;
-    }
-
     private static bool IsGuestCallTarget(string target)
     {
         if (string.IsNullOrWhiteSpace(target))

translator/src/Translator.Core/GeneratedMarkers.cs

@@ -83,9 +83,11 @@ public static partial class GeneratedMarkers
 
     /// <summary>
     /// <c>REGISTER_NATIVE_FUNCTION(0x…, symbol)</c> and its <c>_AS</c> alias form;
-    /// the <c>as</c> group tells the two apart.
+    /// the <c>as</c> group tells the two apart. <c>REGISTER_NATIVE_WRAPPER_AS</c> and
+    /// <c>REGISTER_NATIVE_REPLACEMENT_AS</c> are alias forms too, named by the
+    /// <c>modOverlay</c> group.
     /// </summary>
-    [GeneratedRegex(@"REGISTER_NATIVE_FUNCTION(?<as>_AS)?\s*\(\s*0x(?<address>[0-9A-Fa-f]{8})\s*,\s*(?<symbol>[A-Za-z_][A-Za-z0-9_]*)", RegexOptions.CultureInvariant)]
+    [GeneratedRegex(@"REGISTER_NATIVE_(?:FUNCTION(?<as>_AS)?|(?<modOverlay>WRAPPER|REPLACEMENT)_AS)\s*\(\s*0x(?<address>[0-9A-Fa-f]{8})\s*,\s*(?<symbol>[A-Za-z_][A-Za-z0-9_]*)", RegexOptions.CultureInvariant)]
     public static partial Regex NativeFunctionRegistrationPattern();
 
     [GeneratedRegex(@"REGISTER_TRANSLATED_FUNCTION\s*\(\s*0x(?<address>[0-9A-Fa-f]{8})\s*,\s*(?<symbol>[A-Za-z_][A-Za-z0-9_]*)", RegexOptions.CultureInvariant)]

translator/src/Translator.Core/RuntimeNativeIndex.cs

@@ -11,7 +11,19 @@ public sealed record RuntimeNativeRegistration(
     string Symbol,
     string SourceFile,
     bool IsTranslatedOverride,
-    bool ExcludesBaseTranslation);
+    bool ExcludesBaseTranslation,
+    NativeModOverlay ModOverlay = NativeModOverlay.Unspecified);
+
+/// <summary>What happens to a mod overlay at a native's address.</summary>
+public enum NativeModOverlay
+{
+    /// <summary>Plain registration: the overlay is lost, and reported as such.</summary>
+    Unspecified,
+    /// <summary>The native calls through to the overlay (<c>MKW_GUEST_NEXT</c>).</summary>
+    Wrapper,
+    /// <summary>The native deliberately takes over the overlay's address.</summary>
+    Replacement
+}
 
 public sealed record RuntimeNativeAbiEntry(
     uint Address,
@@ -119,7 +131,17 @@ public static class RuntimeNativeIndexBuilder
         {
             var source = sourceFile.Content;
             foreach (Match match in GeneratedMarkers.NativeFunctionRegistrationPattern().Matches(source))
-                Add(match, match.Groups["symbol"].Value, false, !match.Groups["as"].Success);
+            {
+                var modOverlay = match.Groups["modOverlay"];
+                Add(match, match.Groups["symbol"].Value, false,
+                    !match.Groups["as"].Success && !modOverlay.Success,
+                    modOverlay.Value switch
+                    {
+                        "WRAPPER" => NativeModOverlay.Wrapper,
+                        "REPLACEMENT" => NativeModOverlay.Replacement,
+                        _ => NativeModOverlay.Unspecified
+                    });
+            }
             foreach (Match match in GeneratedMarkers.TranslatedFunctionRegistrationPattern().Matches(source))
                 Add(match, match.Groups["symbol"].Value, true, true);
             foreach (Match match in GeneratedMarkers.NativeOverridePattern().Matches(source))
@@ -127,10 +149,11 @@ public static class RuntimeNativeIndexBuilder
             foreach (Match match in GeneratedMarkers.FatalStubPattern().Matches(source))
                 Add(match, $"GX_FATAL_STUB_{match.Groups["address"].Value}", false, true);
 
-            void Add(Match match, string symbol, bool translated, bool excludesBase) =>
+            void Add(Match match, string symbol, bool translated, bool excludesBase,
+                NativeModOverlay modOverlay = NativeModOverlay.Unspecified) =>
                 registrations.Add(new RuntimeNativeRegistration(
                     ParseAddress(match.Groups["address"].Value), symbol, sourceFile.RelativePath,
-                    translated, excludesBase));
+                    translated, excludesBase, modOverlay));
         }
 
         return registrations
@@ -139,7 +162,8 @@ public static class RuntimeNativeIndexBuilder
             .ThenBy(static registration => registration.Symbol, StringComparer.Ordinal)
             .ThenBy(static registration => registration.SourceFile, StringComparer.Ordinal)
             .ThenBy(static registration => registration.IsTranslatedOverride)
-            .ThenBy(static registration => registration.ExcludesBaseTranslation);
+            .ThenBy(static registration => registration.ExcludesBaseTranslation)
+            .ThenBy(static registration => registration.ModOverlay);
     }
 
     private static uint ParseAddress(string value) => GuestTargetParser.ParseHexAddress(value);

translator/src/Translator.Core/RuntimeNativeIndexFile.cs

@@ -14,9 +14,9 @@ namespace Translator.Core;
 public static class RuntimeNativeIndexFile
 {
     public const string Format = "wiicompiled-runtime-native-index";
-    public const int FormatVersion = 1;
+    public const int FormatVersion = 2;
     // Bump whenever the scanner or the ABI/effect analyzers can produce different output for the same sources.
-    public const int ExtractorVersion = 1;
+    public const int ExtractorVersion = 2;
     public const string DefaultFileName = "native-index.json";
 
     private static readonly JsonSerializerOptions Options = new()
@@ -59,13 +59,15 @@ public static class RuntimeNativeIndexFile
                 .ThenBy(static registration => registration.SourceFile, StringComparer.Ordinal)
                 .ThenBy(static registration => registration.IsTranslatedOverride)
                 .ThenBy(static registration => registration.ExcludesBaseTranslation)
+                .ThenBy(static registration => registration.ModOverlay)
                 .Select(static registration => new RegistrationDto
                 {
                     Address = FormatAddress(registration.Address),
                     Symbol = registration.Symbol,
                     SourceFile = registration.SourceFile,
                     IsTranslatedOverride = registration.IsTranslatedOverride,
-                    ExcludesBaseTranslation = registration.ExcludesBaseTranslation
+                    ExcludesBaseTranslation = registration.ExcludesBaseTranslation,
+                    ModOverlay = registration.ModOverlay.ToString()
                 })
                 .ToList(),
             VoidStubAbis = index.VoidStubAbis
@@ -160,7 +162,8 @@ public static class RuntimeNativeIndexFile
                         Present(registration.Symbol, "symbol"),
                         Present(registration.SourceFile, "sourceFile"),
                         registration.IsTranslatedOverride,
-                        registration.ExcludesBaseTranslation))
+                        registration.ExcludesBaseTranslation,
+                        ParseModOverlay(Present(registration.ModOverlay, "modOverlay"))))
                     .ToArray(),
                 Present(document.VoidStubAbis, "voidStubAbis")
                     .Select(static entry => new RuntimeNativeAbiEntry(
@@ -185,6 +188,11 @@ public static class RuntimeNativeIndexFile
 
     private static string FormatAddress(uint address) => $"0x{address:X8}";
 
+    private static NativeModOverlay ParseModOverlay(string text) =>
+        Enum.TryParse<NativeModOverlay>(text, ignoreCase: false, out var value) && Enum.IsDefined(value)
+            ? value
+            : throw new FormatException($"Unknown modOverlay '{text}'.");
+
     private static uint ParseAddress(string text) => GuestTargetParser.ParseHexAddress(text);
 
     private static T Present<T>(T? value, string name) where T : class =>
@@ -244,6 +252,7 @@ public static class RuntimeNativeIndexFile
         public required string SourceFile { get; init; }
         public required bool IsTranslatedOverride { get; init; }
         public required bool ExcludesBaseTranslation { get; init; }
+        public required string ModOverlay { get; init; }
     }
 
     private sealed class AbiDto

translator/src/Translator.Core/Translation/CanonicalIrStore.cs

@@ -30,6 +30,14 @@ public sealed class CanonicalIrStore : ICanonicalIrProvider
         }
     }
 
+    public void Remove(uint entryPoint)
+    {
+        lock (_gate)
+        {
+            _entries.Remove(entryPoint);
+        }
+    }
+
     public bool TryGet(uint entryPoint, out IrFunction function)
     {
         byte[]? encoded;
@@ -55,12 +63,16 @@ internal static class CanonicalIrBinaryCodec
         IndirectJump, JumpTable, Return, Comment, TracePpc, Undefined
     }
 
+    // Reused per thread so encoding doesn't regrow (and re-fault) a fresh buffer per function.
+    [ThreadStatic] private static MemoryStream? _encodeStream;
+
     public static byte[] Encode(IrFunction function)
     {
         var strings = BuildStringTable(function);
         var ids = strings.Select((value, index) => (value, index))
             .ToDictionary(static pair => pair.value, static pair => pair.index, StringComparer.Ordinal);
-        using var stream = new MemoryStream();
+        var stream = _encodeStream ??= new MemoryStream();
+        stream.SetLength(0);
         using var writer = new BinaryWriter(stream, Encoding.UTF8, leaveOpen: true);
         writer.Write(Magic);
         WriteVarUInt(writer, (uint)CanonicalIrStore.FormatVersion);
@@ -85,9 +97,11 @@ internal static class CanonicalIrBinaryCodec
         return stream.ToArray();
     }
 
-    public static IrFunction Decode(ReadOnlySpan<byte> encoded)
+    public static IrFunction Decode(ReadOnlySpan<byte> encoded) => Decode(encoded.ToArray());
+
+    public static IrFunction Decode(byte[] encoded)
     {
-        using var stream = new MemoryStream(encoded.ToArray(), writable: false);
+        using var stream = new MemoryStream(encoded, writable: false);
         using var reader = new BinaryReader(stream, Encoding.UTF8, leaveOpen: true);
         if (!reader.ReadBytes(Magic.Length).SequenceEqual(Magic) ||
             ReadVarUInt(reader) != CanonicalIrStore.FormatVersion)

translator/src/Translator.Core/Translation/GqrInterproceduralAnalysis.cs

@@ -497,12 +497,52 @@ public static class GqrInterproceduralAnalysis
         return masks;
     }
 
+    /// <summary>
+    /// Unknown-entry roots that depend only on the final call graph, never on discovery order:
+    /// <paramref name="fixedRoots"/>, every candidate without a direct caller, then (ascending) any
+    /// candidate the roots still don't reach, so no candidate is left without an entry fact source.
+    /// </summary>
+    public static HashSet<uint> SelectUnknownEntryRoots(
+        IReadOnlyDictionary<uint, GqrFunctionSummary> functions,
+        IEnumerable<uint> fixedRoots,
+        IEnumerable<uint> candidates)
+    {
+        var called = new HashSet<uint>();
+        foreach (var summary in functions.Values)
+            foreach (var target in summary.DirectGuestTargets)
+                called.Add(target);
+        var ordered = candidates.Where(functions.ContainsKey).Distinct().OrderBy(static a => a).ToArray();
+        var roots = fixedRoots.Where(functions.ContainsKey).ToHashSet();
+        foreach (var address in ordered)
+            if (!called.Contains(address))
+                roots.Add(address);
+        var reachable = ComputeCompactReachable(functions, roots);
+        foreach (var address in ordered)
+        {
+            if (reachable.Contains(address)) continue;
+            roots.Add(address);
+            ExtendCompactReachable(functions, reachable, address);
+        }
+        return roots;
+    }
+
     private static HashSet<uint> ComputeCompactReachable(
         IReadOnlyDictionary<uint, GqrFunctionSummary> functions,
         IReadOnlySet<uint> roots)
     {
         var reachable = new HashSet<uint>();
-        var queue = new Queue<uint>(roots);
+        foreach (var root in roots)
+            ExtendCompactReachable(functions, reachable, root);
+        return reachable;
+    }
+
+    private static void ExtendCompactReachable(
+        IReadOnlyDictionary<uint, GqrFunctionSummary> functions,
+        HashSet<uint> reachable,
+        uint root)
+    {
+        var queue = new Queue<uint>();
+        queue.Enqueue(root);
         while (queue.TryDequeue(out var address))
         {
             if (!reachable.Add(address)) continue;
@@ -510,7 +550,6 @@ public static class GqrInterproceduralAnalysis
                 if (functions.ContainsKey(target) && !reachable.Contains(target))
                     queue.Enqueue(target);
         }
-        return reachable;
     }
 
     private static void TransferCompact(

translator/tests/Translator.Tests/GqrConstantPropagationTests.cs

@@ -167,6 +167,31 @@ public sealed class GqrConstantPropagationTests
             instruction => instruction is IrCall { Target: "PPC_PsqLKnown_00070007" });
     }
 
+    [Fact]
+    public void UnknownEntryRootsDependOnlyOnTheCallGraph()
+    {
+        IrFunction Calls(params uint[] targets) => Function(targets
+            .Select(target => (IrInstruction)new IrCall(string.Empty, $"0x{target:X8}", Array.Empty<IrValue>()))
+            .Append(new IrReturn(null))
+            .ToArray());
+        var functions = new Dictionary<uint, IrFunction>
+        {
+            [0x80001000] = Calls(),
+            [0x80002000] = Calls(0x80003000),
+            [0x80003000] = Calls(),
+            [0x80004000] = Calls(0x80005000),
+            [0x80005000] = Calls(0x80004000)
+        };
+        var summaries = functions.ToDictionary(pair => pair.Key, pair => GqrFunctionSummary.Create(pair.Value));
+        uint[] candidates = [0x80002000, 0x80003000, 0x80004000, 0x80005000];
+
+        var roots = GqrInterproceduralAnalysis.SelectUnknownEntryRoots(summaries, [0x80001000u], candidates);
+        var reversed = GqrInterproceduralAnalysis.SelectUnknownEntryRoots(summaries, [0x80001000u], candidates.Reverse());
+
+        Assert.Equal(new uint[] { 0x80001000, 0x80002000, 0x80004000 }, roots.Order());
+        Assert.Equal(roots.Order(), reversed.Order());
+    }
+
     [Fact]
     public void InterproceduralAnalysisIntersectsConflictingCallers()
     {

translator/tests/Translator.Tests/GuestAddressTableTests.cs

@@ -129,7 +129,7 @@ public sealed class GuestAddressTableTests
         var calls = Directory.EnumerateFiles(Path.Combine(root, "runtime"), "*.*", SearchOption.AllDirectories)
             .Where(path => (path.EndsWith(".cpp", StringComparison.Ordinal) || path.EndsWith(".h", StringComparison.Ordinal)) &&
                            !path.Contains(thirdParty, StringComparison.Ordinal))
-            .SelectMany(static path => Regex.Matches(File.ReadAllText(path), @"MKW_GUEST_FUNC\(\s*([0-9A-Fa-f]{8})\s*\)")
+            .SelectMany(static path => Regex.Matches(File.ReadAllText(path), @"MKW_GUEST_(?:FUNC|NEXT)\(\s*([0-9A-Fa-f]{8})\s*\)")
                 .Select(static match => Convert.ToUInt32(match.Groups[1].Value, 16)))
             .Distinct()
             .ToArray();
@@ -145,7 +145,7 @@ public sealed class GuestAddressTableTests
             foreach (var (region, starts) in entries)
             {
                 var address = table.Map(pal, region);
-                Assert.True(starts.Contains(address), $"MKW_GUEST_FUNC({pal:X8}) is not a {region} function entry (0x{address:X8})");
+                Assert.True(starts.Contains(address), $"MKW_GUEST_FUNC/NEXT({pal:X8}) is not a {region} function entry (0x{address:X8})");
             }
         }
     }

translator/tests/Translator.Tests/RuntimeNativeIndexTests.cs

@@ -1,5 +1,6 @@
 using System.Security.Cryptography;
 using System.Text;
+using System.Text.RegularExpressions;
 using Translator.Core;
 using Translator.Core.CodeGen;
 using Xunit;
@@ -40,14 +41,24 @@ public sealed class RuntimeNativeIndexTests
                 PPC_NATIVE_OVERRIDE_VOID(80000040, Stub, (void), ());
                 GX_FATAL_STUB(80000050, "Fatal")
                 // REGISTER_NATIVE_FUNCTION(0x80000060, CommentedOut);
+                REGISTER_NATIVE_WRAPPER_AS(0x80000070, Wrapper, "wrapper");
+                REGISTER_NATIVE_REPLACEMENT_AS(0x80000080, Replacement, "replacement");
                 """);
 
             var registrations = RuntimeNativeIndexBuilder.Build(directory).Registrations;
-            Assert.Equal(5, registrations.Length);
+            Assert.Equal(7, registrations.Length);
             Assert.Contains(registrations, static entry =>
-                entry.Address == 0x80000010u && entry.ExcludesBaseTranslation);
+                entry.Address == 0x80000010u && entry.ExcludesBaseTranslation &&
+                entry.ModOverlay == NativeModOverlay.Unspecified);
             Assert.Contains(registrations, static entry =>
-                entry.Address == 0x80000020u && !entry.ExcludesBaseTranslation);
+                entry.Address == 0x80000020u && !entry.ExcludesBaseTranslation &&
+                entry.ModOverlay == NativeModOverlay.Unspecified);
+            Assert.Contains(registrations, static entry =>
+                entry.Address == 0x80000070u && entry.Symbol == "Wrapper" && !entry.ExcludesBaseTranslation &&
+                entry.ModOverlay == NativeModOverlay.Wrapper);
+            Assert.Contains(registrations, static entry =>
+                entry.Address == 0x80000080u && entry.Symbol == "Replacement" && !entry.ExcludesBaseTranslation &&
+                entry.ModOverlay == NativeModOverlay.Replacement);
             Assert.Contains(registrations, static entry =>
                 entry.Address == 0x80000030u && entry.IsTranslatedOverride);
             Assert.Contains(registrations, static entry =>
@@ -108,6 +119,8 @@ public sealed class RuntimeNativeIndexTests
         var sources = Path.Combine(ProjectPathsForTests.FindRepositoryRoot(), "runtime", "src");
         var built = RuntimeNativeIndexBuilder.Build(sources);
         Assert.NotEmpty(built.Registrations);
+        Assert.Contains(built.Registrations, static entry => entry.ModOverlay == NativeModOverlay.Wrapper);
+        Assert.Contains(built.Registrations, static entry => entry.ModOverlay == NativeModOverlay.Replacement);
         Assert.NotEmpty(built.VoidStubAbis);
         Assert.NotEmpty(built.Effects);
 
@@ -126,6 +139,26 @@ public sealed class RuntimeNativeIndexTests
         }
     }
 
+    // A wrapper that calls through with MKW_GUEST_FUNC would silently drop the mod overlay it promises to run.
+    [Fact]
+    public void RepositoryWrapperNativesCallThroughToTheNextTranslation()
+    {
+        var sources = Path.Combine(ProjectPathsForTests.FindRepositoryRoot(), "runtime", "src");
+        var wrappers = RuntimeNativeIndexBuilder.Build(sources).Registrations
+            .Where(static entry => entry.ModOverlay == NativeModOverlay.Wrapper)
+            .ToArray();
+        Assert.NotEmpty(wrappers);
+        foreach (var wrapper in wrappers)
+        {
+            var source = File.ReadAllText(Path.Combine(sources, wrapper.SourceFile));
+            var pal = $"{wrapper.Address:X8}";
+            Assert.True(Regex.IsMatch(source, $@"MKW_GUEST_NEXT\(\s*{pal}\s*\)", RegexOptions.IgnoreCase),
+                $"{wrapper.Symbol} is a wrapper but never calls MKW_GUEST_NEXT({pal})");
+            Assert.False(Regex.IsMatch(source, $@"MKW_GUEST_FUNC\(\s*{pal}\s*\)", RegexOptions.IgnoreCase),
+                $"{wrapper.Symbol} is a wrapper but calls MKW_GUEST_FUNC({pal})");
+        }
+    }
+
     [Fact]
     public void ResolveUsesAMatchingIndexAndKeepsWorkingWithoutSources()
     {