wouter/d4r
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Execute native Windows K DLSS and validate real transformer weights

Browse Source
This commit is contained in:
Codex committed 2026-09-30 18:23:01 +03:00
1 parent a0c36a6d6f
commit 5592b7a921
20 files changed
+948 -54
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

scripts/windows/build-zluda-windows.ps1
+1
View File
@@ -66,6 +66,7 @@ try {
Copy-Item -LiteralPath (Join-Path $release 'zluda_trace.dll') -Destination (Join-Path $InstallRoot 'trace/nvcuda.dll') -Force
Copy-Item -LiteralPath (Join-Path $release 'examples/d4r_emit.exe') -Destination $InstallRoot -Force
@{base='ee2f25a180099fa42f36b2346732e1f2470a03ad'; target='x86_64-pc-windows-gnu';
sourceCommit=(& git rev-parse HEAD); workingTreeDirty=[bool](& git status --porcelain);
rust='1.98.1'; llvm=(& (Join-Path $LlvmBuildRoot 'bin/llvm-config.exe') --version);
files=@(Get-ChildItem -LiteralPath $InstallRoot -Filter '*.dll' | Get-FileHash -Algorithm SHA256)} |
ConvertTo-Json -Depth 5 | Set-Content -LiteralPath (Join-Path $InstallRoot 'build-info.json') -Encoding UTF8
scripts/windows/stage-native-k.ps1
+24
View File
@@ -0,0 +1,24 @@
[CmdletBinding()]
param([Parameter(Mandatory=$true)][string]$DlssDll, [string]$PackageRoot, [string]$OutputDirectory)
$ErrorActionPreference = 'Stop'
$repo = [IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
if (!$PackageRoot) { $PackageRoot = Join-Path $repo 'dist/windows-rdna4-therock' }
if (!$OutputDirectory) { $OutputDirectory = Join-Path $repo 'build/native-k-gfx1201' }
if (![IO.Path]::IsPathRooted($DlssDll) -or !(Test-Path -LiteralPath $DlssDll)) {
throw 'Supply the absolute path to your local nvngx_dlss.dll.'
}
New-Item -ItemType Directory -Force $OutputDirectory | Out-Null
$layers = @('enc0','enc1','enc2','enc3','enc4','dec5','dec4','dec3','dec2','dec1','dec0')
foreach ($layer in $layers) {
$name = "dltss_pwin_${layer}_layer"
$source = Join-Path $PackageRoot "experimental/k/${name}_gfx1201.hsaco"
if (!(Test-Path -LiteralPath $source)) { throw "Native K module missing: $source" }
Copy-Item -LiteralPath $source -Destination (Join-Path $OutputDirectory "$name.hsaco") -Force
}
& python (Join-Path $repo 'kernels/tools/kernel_manifest.py') $OutputDirectory $DlssDll
if ($LASTEXITCODE) { throw "Native manifest generation failed ($LASTEXITCODE)" }
@{architecture='gfx1201'; family='K'; dlssSha256=(Get-FileHash -LiteralPath $DlssDll -Algorithm SHA256).Hash;
validation='synthetic layers passed; actual DLSS capture/reference validation required';
modules=@(Get-ChildItem -LiteralPath $OutputDirectory -Filter '*.hsaco' | Get-FileHash -Algorithm SHA256)} |
ConvertTo-Json -Depth 5 | Set-Content -LiteralPath (Join-Path $OutputDirectory 'build-info.json') -Encoding UTF8
Write-Host "Experimental native K directory: $OutputDirectory"
scripts/windows/test-k-captures.ps1
+52
View File
@@ -0,0 +1,52 @@
[CmdletBinding()]
param([Parameter(Mandatory=$true)][string]$CaptureDirectory, [string]$HipRoot,
[string]$PackageRoot, [string]$OutputDirectory, [int]$Windows = 12,
[Parameter(Mandatory=$true)][string]$DlssDll, [int]$PtxWindows = 2)
$ErrorActionPreference = 'Stop'
$repo = [IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
if (!$HipRoot) { $HipRoot = Join-Path $repo '.tools/therock-10.2.0a20260929/_rocm_sdk_core' }
if (!$PackageRoot) { $PackageRoot = Join-Path $repo 'dist/windows-rdna4-therock' }
if (!$OutputDirectory) { $OutputDirectory = Join-Path $repo ('test-results/k-real-reference-' + (Get-Date -Format 'yyyyMMdd-HHmmss')) }
New-Item -ItemType Directory -Force $OutputDirectory | Out-Null
$previousPython = $env:PYTHONPATH
$previousDiag = $env:D4R_DIAG_DIR
$env:PYTHONPATH = "$(Join-Path $repo '.tools/python/vendor');$previousPython"
$summary = [ordered]@{utc=[DateTime]::UtcNow.ToString('o'); captureDirectory=$CaptureDirectory; layers=@(); passed=$false}
try {
foreach ($layer in @('enc0','enc1','enc2','enc3','enc4','dec5','dec4','dec3','dec2','dec1','dec0')) {
$name = "dltss_pwin_${layer}_layer"
$captures = @(Get-ChildItem -LiteralPath $CaptureDirectory -Directory | Where-Object Name -Match ("^replay-[0-9]+-" + $name + '$'))
if (!$captures.Count) { throw "Missing actual K capture: $name" }
$capture = $captures | Sort-Object Name | Select-Object -First 1
$folder = Join-Path $OutputDirectory $layer
$output = Join-Path $folder 'output'
New-Item -ItemType Directory -Force $folder | Out-Null
$env:D4R_DIAG_DIR = $folder
& (Join-Path $PackageRoot 'bin/d4r_native_replay_probe.exe') --hip-root $HipRoot `
--module (Join-Path $PackageRoot "experimental/k/${name}_gfx1201.hsaco") `
--fixture-dir $capture.FullName --output-dir $output --iterations 1 `
> (Join-Path $folder 'gpu.stdout.log') 2> (Join-Path $folder 'gpu.stderr.log')
if ($LASTEXITCODE) { throw "K $layer actual GPU replay failed: $folder" }
& python (Join-Path $PackageRoot 'k_replay_validate.py') validate --layer $layer `
--fixture-dir $capture.FullName --output-dir $output --windows $Windows --real-capture `
> (Join-Path $folder 'reference.stdout.log') 2> (Join-Path $folder 'reference.stderr.log')
if ($LASTEXITCODE) { throw "K $layer mismatch; stopping before the next layer: $folder" }
Get-Content -LiteralPath (Join-Path $folder 'reference.stdout.log') | Write-Host
$ptxValidator = Join-Path $PackageRoot 'k_ptx_validate.py'
if (!(Test-Path -LiteralPath $ptxValidator)) { $ptxValidator = Join-Path $repo 'tools/windows/k_ptx_validate.py' }
& python -u $ptxValidator --layer $layer `
--capture-dir $capture.FullName --native-dir $output --dlss-dll $DlssDll `
--output-dir (Join-Path $folder 'private-ptx') --windows $PtxWindows `
> (Join-Path $folder 'ptx-reference.stdout.log') 2> (Join-Path $folder 'ptx-reference.stderr.log')
if ($LASTEXITCODE) { throw "K $layer independent PTX mismatch; stopping: $folder" }
Get-Content -LiteralPath (Join-Path $folder 'ptx-reference.stdout.log') | Write-Host
$summary.layers += @{layer=$layer; passed=$true}
}
$summary.passed = $true
} catch { $summary.error=$_.Exception.Message; Write-Warning $summary.error }
finally {
$env:PYTHONPATH = $previousPython; $env:D4R_DIAG_DIR = $previousDiag
$summary | ConvertTo-Json -Depth 5 | Set-Content -LiteralPath (Join-Path $OutputDirectory 'summary.json') -Encoding UTF8
Write-Host "Private real-weight replay results: $OutputDirectory (do not redistribute captures/allocations)"
}
if (!$summary.passed) { exit 1 }
scripts/windows/test-native-overrides.ps1
+50
View File
@@ -0,0 +1,50 @@
[CmdletBinding()]
param([string]$HipRoot, [string]$PackageRoot, [string]$ZludaRoot, [string]$OutputDirectory)
$ErrorActionPreference = 'Stop'
$repo = [IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
if (!$HipRoot) { $HipRoot = Join-Path $repo '.tools/therock-10.2.0a20260929/_rocm_sdk_core' }
if (!$PackageRoot) { $PackageRoot = Join-Path $repo 'dist/windows-rdna4-therock' }
if (!$ZludaRoot) { $ZludaRoot = Join-Path $repo 'dist/zluda-windows-native' }
if (!$OutputDirectory) { $OutputDirectory = Join-Path $repo ('test-results/native-identity-' + (Get-Date -Format 'yyyyMMdd-HHmmss')) }
New-Item -ItemType Directory -Force $OutputDirectory | Out-Null
$previous = @{D4R_ZLUDA_NATIVE_DIR=$env:D4R_ZLUDA_NATIVE_DIR; D4R_ZLUDA_VERBOSE=$env:D4R_ZLUDA_VERBOSE; D4R_DIAG_DIR=$env:D4R_DIAG_DIR}
function Invoke-NativeProbe([string]$Name) {
$info = New-Object Diagnostics.ProcessStartInfo
$info.FileName = Join-Path $PackageRoot 'bin/d4r_cuda_driver_probe.exe'
$info.Arguments = '--hip-root "' + $HipRoot + '" --cuda-dll "' + (Join-Path $ZludaRoot 'nvcuda.dll') + '" --iterations 32'
$info.UseShellExecute=$false; $info.CreateNoWindow=$true; $info.RedirectStandardOutput=$true; $info.RedirectStandardError=$true
$process = New-Object Diagnostics.Process
$process.StartInfo=$info
if (!$process.Start()) { throw 'Cannot start native override probe' }
$stdout=$process.StandardOutput.ReadToEndAsync(); $stderr=$process.StandardError.ReadToEndAsync()
if (!$process.WaitForExit(120000)) { $process.Kill(); $process.WaitForExit(); throw 'Native override test timed out' }
[IO.File]::WriteAllText((Join-Path $OutputDirectory "$Name.stdout.log"),$stdout.GetAwaiter().GetResult())
[IO.File]::WriteAllText((Join-Path $OutputDirectory "$Name.stderr.log"),$stderr.GetAwaiter().GetResult())
$code=$process.ExitCode; $process.Dispose()
return $code
}
try {
$env:D4R_ZLUDA_VERBOSE = '1'; $env:D4R_DIAG_DIR = $OutputDirectory
$bad = Join-Path $OutputDirectory 'wrong-identity'
$good = Join-Path $OutputDirectory 'matching-identity'
New-Item -ItemType Directory -Force $bad,$good | Out-Null
foreach ($directory in @($bad,$good)) {
Copy-Item -LiteralPath (Join-Path $PackageRoot 'bin/probe_gfx1201.hsaco') -Destination (Join-Path $directory 'd4r_ptx_pattern.hsaco') -Force
}
'd4r_ptx_pattern 0000000000000000' | Set-Content -LiteralPath (Join-Path $bad 'd4r-kernels.txt') -Encoding ASCII
$env:D4R_ZLUDA_NATIVE_DIR = $bad
if (Invoke-NativeProbe 'wrong') { throw 'Wrong identity must use the correct translated fallback' }
if (!(Select-String -LiteralPath (Join-Path $OutputDirectory 'wrong.stderr.log') -Pattern 'reason=ptx_identity.*fallback=translated' -Quiet)) {
throw 'Wrong identity was not explicitly rejected'
}
$identity = Select-String -LiteralPath (Join-Path $OutputDirectory 'wrong.stdout.log') -Pattern 'fnv1a64=([0-9a-f]{16})'
if (!$identity) { throw 'PTX probe did not report its identity' }
"d4r_ptx_pattern $($identity.Matches[0].Groups[1].Value)" | Set-Content -LiteralPath (Join-Path $good 'd4r-kernels.txt') -Encoding ASCII
$env:D4R_ZLUDA_NATIVE_DIR = $good
if (Invoke-NativeProbe 'matching') { throw 'Matching native override produced incorrect output' }
$hits = @(Select-String -LiteralPath (Join-Path $OutputDirectory 'matching.stderr.log') -Pattern '\[d4r-launch\].*backend=native')
if ($hits.Count -ne 32) { throw "Expected 32 native launches, got $($hits.Count)" }
Write-Host 'PASS NATIVE_IDENTITY wrong_hash=fallback matching_hash=native launches=32'
} finally {
foreach ($name in $previous.Keys) { [Environment]::SetEnvironmentVariable($name,$previous[$name],'Process') }
}
scripts/windows/test-windows-rdna4.ps1
+34 -5
View File
@@ -10,6 +10,8 @@ param(
[ValidateSet('init', 'evaluate')][string]$NgxMode = 'init',
[ValidateSet(5, 11, 13)][int]$Preset = 11,
[switch]$NgxOnly,
[switch]$Trace,
[switch]$RequireNativeNetwork,
[int]$Iterations = 32,
[int]$TimeoutSeconds = 180
)
@@ -48,6 +50,7 @@ $originalLog = $env:ZLUDA_LOG_DIR
$originalCuda = $env:ZLUDA_CUDA_LIB
$originalNvapi = $env:D4R_NVAPI_BACKEND
$originalPythonPath = $env:PYTHONPATH
$originalVerbose = $env:D4R_ZLUDA_VERBOSE
$exitStatus = 1
$ngxRuntimeDirectory = $null
$summary = [ordered]@{utc=[DateTime]::UtcNow.ToString('o'); profile=$RuntimeProfile; hipRoot=$HipRoot; zludaRoot=$ZludaRoot; tests=@()}
@@ -72,13 +75,17 @@ function Invoke-Probe([string]$Name, [string]$Exe, [string[]]$Arguments) {
$process.StartInfo = $info
if (!$process.Start()) { throw "Process start failed: $Exe" }
# Consume both streams concurrently; a verbose JIT trace must not block its child.
$outTask = $process.StandardOutput.ReadToEndAsync()
$errTask = $process.StandardError.ReadToEndAsync()
$outFile = [IO.File]::Open($stdout, [IO.FileMode]::Create, [IO.FileAccess]::Write, [IO.FileShare]::Read)
$errFile = [IO.File]::Open($stderr, [IO.FileMode]::Create, [IO.FileAccess]::Write, [IO.FileShare]::Read)
$outTask = $process.StandardOutput.BaseStream.CopyToAsync($outFile)
$errTask = $process.StandardError.BaseStream.CopyToAsync($errFile)
$finished = $process.WaitForExit($TimeoutSeconds * 1000)
if (!$finished) { $process.Kill(); $process.WaitForExit() }
$code = $process.ExitCode
[IO.File]::WriteAllText($stdout, $outTask.GetAwaiter().GetResult())
[IO.File]::WriteAllText($stderr, $errTask.GetAwaiter().GetResult())
try {
$null = $outTask.GetAwaiter().GetResult()
$null = $errTask.GetAwaiter().GetResult()
} finally { $outFile.Dispose(); $errFile.Dispose() }
$process.Dispose()
$codeHex = if ($null -eq $code) { 'unknown' } else {
'0x{0:x8}' -f [BitConverter]::ToUInt32([BitConverter]::GetBytes([int]$code), 0)
@@ -90,6 +97,12 @@ function Invoke-Probe([string]$Name, [string]$Exe, [string[]]$Arguments) {
}
try {
if ($NgxOnly -and (!$NgxCore -or !$DlssDll)) { throw '-NgxOnly requires both locally supplied NVIDIA DLL paths.' }
if ($RequireNativeNetwork) {
if ($NgxMode -ne 'evaluate' -or $Preset -ne 11 -or !$env:D4R_ZLUDA_NATIVE_DIR) {
throw '-RequireNativeNetwork currently requires K Evaluate and D4R_ZLUDA_NATIVE_DIR.'
}
$env:D4R_ZLUDA_VERBOSE = '1'
}
$summary.foundationDiagnosticsPerformed = !$NgxOnly
if ($Iterations -lt 1 -or $Iterations -gt 10000) { throw 'Iterations must be 1..10000' }
if ($TimeoutSeconds -lt 1) { throw 'TimeoutSeconds must be positive' }
@@ -199,12 +212,27 @@ try {
Copy-Item -LiteralPath (Join-Path $bin 'd4r_ngx_cuda_init_probe.exe') -Destination $ngxExe
Copy-Item -LiteralPath $NgxCore -Destination $localCore
Copy-Item -LiteralPath $DlssDll -Destination $localDlss
$ngxArguments = @('--hip-root', $HipRoot, '--cuda-dll', (Join-Path $ZludaRoot 'nvcuda.dll'),
$selectedCuda = if ($Trace) { Join-Path $ZludaRoot 'trace/nvcuda.dll' } else { Join-Path $ZludaRoot 'nvcuda.dll' }
if (!(Test-Path -LiteralPath $selectedCuda)) { throw "CUDA runtime missing: $selectedCuda" }
$summary.traceRequested = [bool]$Trace
$ngxArguments = @('--hip-root', $HipRoot, '--cuda-dll', $selectedCuda,
'--ngx-core', $localCore, '--dlss-dll', $localDlss,
'--nvapi-dll', (Join-Path $PackageRoot 'nvapi-compat/nvapi64.dll'),
'--ngx-mode', $NgxMode, '--preset', "$Preset", '--iterations', "$Iterations")
$ngxName = if ($NgxMode -eq 'evaluate') { "ngx-evaluate-preset-$Preset" } else { 'ngx-init' }
$ngxOk = Invoke-Probe $ngxName $ngxExe $ngxArguments
if ($ngxOk -and $RequireNativeNetwork) {
$hits = @()
foreach ($layer in @('enc0','enc1','enc2','enc3','enc4','dec5','dec4','dec3','dec2','dec1','dec0')) {
$pattern = '\[d4r-launch\] kernel="dltss_pwin_' + $layer + '_layer" backend=native'
$count = @(Select-String -LiteralPath (Join-Path $OutputDirectory "$ngxName.stderr.log") -Pattern $pattern).Count
$hits += @{layer=$layer; nativeLaunches=$count; expectedFrames=$Iterations}
if ($count -lt $Iterations) { $ngxOk=$false }
}
$summary.nativeTransformer = $hits
$summary.nativeTransformerPassed = $ngxOk
Write-Host "Native K transformer launches validated=$ngxOk"
}
if (!$ngxOk) {
$exitStatus = 1
if (Test-Path (Join-Path $ZludaRoot 'trace/nvcuda.dll')) {
@@ -240,6 +268,7 @@ try {
$env:ZLUDA_CUDA_LIB = $originalCuda
$env:D4R_NVAPI_BACKEND = $originalNvapi
$env:PYTHONPATH = $originalPythonPath
$env:D4R_ZLUDA_VERBOSE = $originalVerbose
$archive = "$OutputDirectory.zip"
Compress-Archive -LiteralPath $OutputDirectory -DestinationPath $archive -Force
Write-Host "Diagnostic bundle: $archive"