feat(cccl): integrate missing CCCL directories — python/, ci/, .agent/, docs/, test/

Sparse-checkout from NVIDIA/cccl main branch to complete cccl_upstream: Added: - python/cuda_cccl/ (226 files) — Python bindings for device-level algorithms Critical for muh toolchain: cuda.compute.reduce_into, scan, radix_sort, etc. Includes 204 .py files with full test coverage for all 27 algorithms - ci/ (163 files) — Build/test infrastructure build_cub.sh, test_cub.sh, build_and_test_targets.sh, matrix.yaml Directly maps to our [INFRA-CI] and [INFRA-BUILD] items - .agent/skills/ (7 files) — NVIDIA's own agent skills for CCCL cccl-style/SKILL.md, cccl-test/SKILL.md, sass-diff/SKILL.md - docs/ (491 files) — Official CCCL documentation CI references, CMake guides, Python compute docs, libcudacxx PTX docs - test/ (12 files) — Top-level integration tests (cuda_smoke, stdpar) - Root configs: .clang-format, .clang-tidy, CONTRIBUTING.md, pyproject.toml - CLAUDE.md symlink → AGENTS.md (NVIDIA's standard) cccl_upstream now mirrors full NVIDIA/cccl structure: Before: 42M (cub + thrust + libcudacxx + cudax + c + examples + benchmarks) After: 53M (+python +ci +docs +.agent +test +configs) This completes the CCCL base needed for: - [muh-bench] items: ci/util/build_and_test_targets.sh for targeted builds - [CCCL-verify] items: python/cuda_cccl/tests/ as reference implementations - [CCCL-test] items: ci/test_cub.sh, ci/test_thrust.sh - Agent workflow: .agent/skills/ for consistent style and test patterns
2026-08-07 02:34:33 +00:00
parent 3f97dca7ad
commit 2a7ca101d7
908 changed files with 121615 additions and 0 deletions
--- a/cccl_upstream/ci/windows/build_cccl_c_parallel.ps1
+++ b/cccl_upstream/ci/windows/build_cccl_c_parallel.ps1
@@ -0,0 +1,28 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Remove-Module -Name build_common -ErrorAction SilentlyContinue
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @(20, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cccl-c-parallel"
+$LOCAL_CMAKE_OPTIONS = ""
+
+configure_and_build_preset "CCCL C Parallel" $PRESET $LOCAL_CMAKE_OPTIONS
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/build_cccl_c_parallel_v2.ps1
+++ b/cccl_upstream/ci/windows/build_cccl_c_parallel_v2.ps1
@@ -0,0 +1,28 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Remove-Module -Name build_common -ErrorAction SilentlyContinue
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @(20, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cccl-c-parallel-v2"
+$LOCAL_CMAKE_OPTIONS = ""
+
+configure_and_build_preset "CCCL C Parallel" $PRESET $LOCAL_CMAKE_OPTIONS
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/build_common.psm1
+++ b/cccl_upstream/ci/windows/build_common.psm1
@@ -0,0 +1,233 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+# We need the full path to cl because otherwise cmake will replace CMAKE_CXX_COMPILER with the full path
+# and keep CMAKE_CUDA_HOST_COMPILER at "cl" which breaks our cmake script
+$script:HOST_COMPILER  = (Get-Command "cl").source -replace '\\','/'
+$script:PARALLEL_LEVEL = $env:NUMBER_OF_PROCESSORS
+
+Write-Host "=== Docker Container Resource Info ==="
+Write-Host "Number of Processors: $script:PARALLEL_LEVEL"
+Get-WmiObject Win32_OperatingSystem | ForEach-Object {
+    Write-Host ("Memory: total={0:N1} GB, free={1:N1} GB" -f ($_.TotalVisibleMemorySize / 1MB), ($_.FreePhysicalMemory / 1MB))
+}
+Write-Host "======================================"
+
+# Extract the CL version for export to build scripts:
+$script:CL_VERSION_STRING = & cl.exe /?
+if ($script:CL_VERSION_STRING -match "Version (\d+\.\d+)\.\d+") {
+    $CL_VERSION = [version]$matches[1]
+    Write-Host "Detected cl.exe version: $CL_VERSION"
+}
+
+$script:GLOBAL_CMAKE_OPTIONS = $CMAKE_OPTIONS
+if ($CUDA_ARCH) {
+    $script:GLOBAL_CMAKE_OPTIONS += ' "-DCMAKE_CUDA_ARCHITECTURES={0}"' -f $CUDA_ARCH
+}
+
+# Default to pedantic mode in CI (GitHub Actions)
+if ($env:GITHUB_ACTIONS) {
+    $script:GLOBAL_CMAKE_OPTIONS += ' "-DCCCL_ENABLE_WERROR=ON" "-DCCCL_ENABLE_PRAGMA_SYSTEM_HEADER=OFF"'
+} else {
+    $script:GLOBAL_CMAKE_OPTIONS += ' "-DCCCL_ENABLE_WERROR=OFF" "-DCCCL_ENABLE_PRAGMA_SYSTEM_HEADER=ON"'
+}
+
+if (-not $env:CCCL_BUILD_INFIX) {
+    $env:CCCL_BUILD_INFIX = ""
+}
+
+# Presets will be configured in this directory:
+$BUILD_DIR = "../build/$env:CCCL_BUILD_INFIX"
+
+If(!(test-path -PathType container "../build")) {
+    New-Item -ItemType Directory -Path "../build"
+}
+
+# The most recent build will always be symlinked to cccl/build/latest
+New-Item -ItemType Directory -Path "$BUILD_DIR" -Force
+
+# Convert to an absolute path:
+$BUILD_DIR = (Get-Item -Path "$BUILD_DIR").FullName
+
+# Prepare environment for CMake:
+$env:CMAKE_BUILD_PARALLEL_LEVEL = $PARALLEL_LEVEL
+$env:CTEST_PARALLEL_LEVEL = 1
+$env:CUDAHOSTCXX = $script:HOST_COMPILER
+$env:CXX = $script:HOST_COMPILER
+
+Write-Host "========================================"
+Write-Host "Begin build"
+Write-Host "pwd=$pwd"
+Write-Host "BUILD_DIR=$BUILD_DIR"
+Write-Host "CXX_STANDARD=$CXX_STANDARD"
+Write-Host "CXX=$env:CXX"
+Write-Host "CUDACXX=$env:CUDACXX"
+Write-Host "CUDAHOSTCXX=$env:CUDAHOSTCXX"
+Write-Host "TBB_ROOT=$env:TBB_ROOT"
+Write-Host "NVCC_VERSION=$NVCC_VERSION"
+Write-Host "CMAKE_BUILD_PARALLEL_LEVEL=$env:CMAKE_BUILD_PARALLEL_LEVEL"
+Write-Host "CTEST_PARALLEL_LEVEL=$env:CTEST_PARALLEL_LEVEL"
+Write-Host "CCCL_BUILD_INFIX=$env:CCCL_BUILD_INFIX"
+Write-Host "GLOBAL_CMAKE_OPTIONS=$script:GLOBAL_CMAKE_OPTIONS"
+Write-Host "Current commit is:"
+Write-Host "$(git log -1 --format=short)"
+Write-Host "========================================"
+
+cmake --version
+ctest --version
+
+function configure_preset {
+    Param(
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$BUILD_NAME,
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$PRESET,
+        [Parameter(Mandatory = $false)]
+        [string]$LOCAL_CMAKE_OPTIONS = ""
+    )
+
+    $step = "$BUILD_NAME (configure)"
+
+    # CMake must be invoked in the same directory as the presets file:
+    pushd ".."
+
+    # Echo and execute command to stdout:
+    $configure_command = "cmake --preset $PRESET --log-level VERBOSE"
+    if ($LOCAL_CMAKE_OPTIONS) {
+        $configure_command += " $LOCAL_CMAKE_OPTIONS"
+    }
+    if ($script:GLOBAL_CMAKE_OPTIONS) {
+        $configure_command += " $script:GLOBAL_CMAKE_OPTIONS"
+    }
+
+    Write-Host $configure_command
+    Invoke-Expression $configure_command
+    $test_result = $LastExitCode
+
+    If ($test_result -ne 0) {
+        throw "$step Failed"
+    }
+
+    popd
+    Write-Host "$step complete."
+}
+
+function build_preset {
+    Param(
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$BUILD_NAME,
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$PRESET
+    )
+
+    $step = "$BUILD_NAME (build)"
+
+    # CMake must be invoked in the same directory as the presets file:
+    pushd ".."
+
+    sccache -z >$null
+
+    cmake --build --preset $PRESET -v
+    $test_result = $LastExitCode
+
+    $preset_dir = "${BUILD_DIR}/${PRESET}"
+    $sccache_json = "${preset_dir}/sccache_stats.json"
+
+    sccache --show-adv-stats
+    sccache --show-adv-stats --stats-format=json > "${sccache_json}"
+
+    echo "$step complete"
+
+    If ($test_result -ne 0) {
+         throw "$step Failed"
+    }
+
+    popd
+}
+
+function test_preset {
+    Param(
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$BUILD_NAME,
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$PRESET
+    )
+
+    $step = "$BUILD_NAME (test)"
+
+    # CTest must be invoked in the same directory as the presets file:
+    pushd ".."
+
+    sccache -z >$null
+
+    ctest --preset $PRESET
+    $test_result = $LastExitCode
+
+    sccache --show-adv-stats
+
+    echo "$step complete"
+
+    If ($test_result -ne 0) {
+         throw "$step Failed"
+    }
+
+    popd
+}
+
+function configure_and_build_preset {
+    Param(
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$BUILD_NAME,
+        [Parameter(Mandatory = $true)]
+        [ValidateNotNullOrEmpty()]
+        [string]$PRESET,
+        [Parameter(Mandatory = $false)]
+        [string]$LOCAL_CMAKE_OPTIONS = ""
+    )
+
+    configure_preset $BUILD_NAME $PRESET $LOCAL_CMAKE_OPTIONS
+    build_preset $BUILD_NAME $PRESET
+}
+
+function Invoke-Checked {
+    <#
+    .SYNOPSIS
+        Runs a script block and throws if the last native command in it exits
+        non-zero. $ErrorActionPreference = "Stop" does not make native commands
+        (python/pip/pytest/...) throw, so their $LASTEXITCODE must be checked
+        explicitly; this wraps that boilerplate into one call.
+    .EXAMPLE
+        Invoke-Checked { & $python -m pip install pytest } "pip install failed"
+    #>
+    param(
+        [Parameter(Mandatory, Position = 0)][scriptblock]$ScriptBlock,
+        [Parameter(Position = 1)][string]$ErrorMessage = "Native command failed"
+    )
+    & $ScriptBlock
+    if ($LASTEXITCODE -ne 0) {
+        throw "$ErrorMessage (exit code $LASTEXITCODE)"
+    }
+}
+
+Export-ModuleMember -Function configure_preset, build_preset, test_preset, configure_and_build_preset, Invoke-Checked
+Export-ModuleMember -Variable BUILD_DIR, CL_VERSION
--- a/cccl_upstream/ci/windows/build_common_python.psm1
+++ b/cccl_upstream/ci/windows/build_common_python.psm1
@@ -0,0 +1,246 @@
+function Get-Python {
+    <#
+    .SYNOPSIS
+        Returns the path of the Python interpreter satisfying the supplied
+        version, installing it via uv if necessary.
+    .PARAMETER Version
+        A string in the form 'M.m' (e.g., '3.10', '3.13') or a free-threaded
+        version such as '3.14t'.
+    #>
+    [CmdletBinding()]
+    param(
+        [Parameter(Mandatory, Position = 0)]
+        [ValidatePattern('^\d+\.\d+t?$')]
+        [string]$Version
+    )
+
+    # Install uv if not present. uv downloads pre-built CPython binaries --
+    # no compilation, no build dependencies, no pyenv-win required.
+    if (-not (Get-Command uv -ErrorAction SilentlyContinue)) {
+        Write-Host "Installing uv..."
+        Invoke-RestMethod https://astral.sh/uv/install.ps1 | Invoke-Expression
+        # uv installs to $HOME\.local\bin on Windows
+        $uvBin = Join-Path $HOME '.local\bin'
+        $Env:PATH = $uvBin + ";" + $Env:PATH
+    }
+
+    Write-Host "Creating Python $Version venv via uv..."
+    $venvDir = Join-Path $HOME '.cccl-venv'
+    & uv venv --seed --python $Version $venvDir
+    if ($LASTEXITCODE -ne 0) {
+        throw [System.InvalidOperationException]::new(
+            "Failed to create Python $Version venv via uv."
+        )
+    }
+
+    $exe = Join-Path $venvDir 'Scripts\python.exe'
+    if (-not (Test-Path $exe)) {
+        throw [System.InvalidOperationException]::new(
+            "Could not find python.exe in venv at $exe"
+        )
+    }
+
+    Write-Host "Python $Version at: $exe"
+
+    # Add venv Scripts dir to PATH so bare `python` and `pip` work.
+    $scriptsDir = Join-Path $venvDir 'Scripts'
+    $Env:PATH = $scriptsDir + ";" + $Env:PATH
+
+    return $exe
+}
+
+function Get-RepoRoot {
+    return (Resolve-Path "$PSScriptRoot/../..")
+}
+
+function Get-CudaMajor {
+    <#
+    .SYNOPSIS
+        Gets the CUDA major version for this container instance (e.g. '12' or
+        '13').  Defaults to '13' if no match can be found.
+    #>
+    if ($env:CUDA_PATH) {
+        $nvcc = Join-Path $env:CUDA_PATH "bin/nvcc.exe"
+        if (Test-Path $nvcc) {
+            $out = & $nvcc --version 2>&1
+            $text = ($out -join "`n")
+            if ($text -match 'release\s+(\d+)\.') { return $Matches[1] }
+        }
+        # Fallback: parse major from CUDA_PATH like ...\v13.0 or ...\CUDA\13
+        $pathMatch = [regex]::Match($env:CUDA_PATH, 'v?(\d+)(?:\.\d+)?')
+        if ($pathMatch.Success) { return $pathMatch.Groups[1].Value }
+    }
+    return '13'
+}
+
+function Get-CudaVersion {
+    <#
+    .SYNOPSIS
+        Gets the CUDA major.minor version for this container instance (e.g.
+        '12.9' or '13.0').  Defaults to '13.0' if no match can be found.
+    #>
+    if ($env:CUDA_PATH) {
+        $nvcc = Join-Path $env:CUDA_PATH "bin/nvcc.exe"
+        if (Test-Path $nvcc) {
+            $out = & $nvcc --version 2>&1
+            $text = ($out -join "`n")
+            if ($text -match 'release\s+(\d+\.\d+)') { return $Matches[1] }
+        }
+        # Fallback: parse major.minor from CUDA_PATH like ...\v13.0
+        $pathMatch = [regex]::Match($env:CUDA_PATH, 'v?(\d+\.\d+)')
+        if ($pathMatch.Success) { return $pathMatch.Groups[1].Value }
+    }
+    return '13.0'
+}
+
+function Get-CtkTestMode {
+    <#
+    .SYNOPSIS
+        Validates and normalizes a CTK test mode passed as -Mode (forwarded from
+        the -ctk-mode arg): 'pinned' (default; empty means pinned), 'latest', or
+        'sysctk'. Throws on any other value (fail loud on a typo'd mode). Returns
+        the lowercased mode.
+    #>
+    param([string]$Mode = "")
+    if ([string]::IsNullOrEmpty($Mode)) { return "pinned" }
+    $Mode = $Mode.ToLowerInvariant()
+    if (-not ($Mode -in @("pinned", "latest", "sysctk"))) {
+        throw "Invalid ctk mode '$Mode' (expected pinned|latest|sysctk)"
+    }
+    return $Mode
+}
+
+function Set-CtkPin {
+    <#
+    .SYNOPSIS
+        Configures cuda-toolkit pinning for this lane per the -Mode arg (see
+        Get-CtkTestMode): 'pinned' (default) pins cuda-toolkit to the container's
+        CTK major.minor via PIP_CONSTRAINT; 'latest' and 'sysctk' leave it
+        unpinned ('sysctk' installs no cuda-toolkit wheel at all -- the
+        system-provided toolkit is used).
+    #>
+    param([string]$Mode = "")
+    if ((Get-CtkTestMode $Mode) -eq "pinned") {
+        $cudaVersion = Get-CudaVersion
+        $env:PIP_CONSTRAINT = Join-Path ([System.IO.Path]::GetTempPath()) "ctk-constraint.txt"
+        "cuda-toolkit==$cudaVersion.*" | Out-File -FilePath $env:PIP_CONSTRAINT -Encoding ascii
+    } else {
+        # latest / sysctk: no pin. Clear any inherited constraint so it cannot
+        # affect the resolve.
+        Remove-Item Env:\PIP_CONSTRAINT -ErrorAction SilentlyContinue
+    }
+}
+
+function Get-CtkExtraFlavor {
+    <#
+    .SYNOPSIS
+        Returns the pip-extra toolkit "flavor" for the given -Mode: 'sysctk' when
+        the mode is sysctk (rely on the system-provided CUDA toolkit) or 'cu'
+        otherwise (pip-installed toolkit). Combine with the CUDA major, e.g.
+        "minimal-$(Get-CtkExtraFlavor $CtkMode)$cudaMajor".
+    #>
+    param([string]$Mode = "")
+    if ((Get-CtkTestMode $Mode) -eq "sysctk") { return "sysctk" }
+    return "cu"
+}
+
+function Convert-ToUnixPath {
+    Param([Parameter(Mandatory = $true)][string]$p)
+    return ($p -replace "\\", "/")
+}
+
+function Get-CudaCcclWheel {
+    <#
+    .SYNOPSIS
+        Returns the path of the cuda-cccl wheel artifact to use in the context
+        of a GitHub Actions CI test script.
+    #>
+    Param()
+
+    $repoRoot = Get-RepoRoot
+    if ($env:GITHUB_ACTIONS) {
+        Push-Location $repoRoot
+        try {
+            $wheelArtifactName = (& bash -lc "ci/util/workflow/get_wheel_artifact_name.sh").Trim()
+            if (-not $wheelArtifactName) { throw 'Failed to resolve wheel artifact name' }
+            $repoRootPosix = Convert-ToUnixPath $repoRoot
+            # Ensure output from downloader goes to console, not function return pipeline
+            $null = (& bash -lc "ci/util/artifacts/download.sh $wheelArtifactName $repoRootPosix" 2>&1 | Out-Host)
+            if ($LASTEXITCODE -ne 0) { throw "Failed to download wheel artifact '$wheelArtifactName'" }
+        }
+        finally { Pop-Location }
+    }
+
+    $wheelhouse = Join-Path $repoRoot 'wheelhouse'
+    $wheelPath = Get-OnePathMatch -Path $wheelhouse -Pattern '^cuda_cccl-.*\.whl' -File
+    return $wheelPath
+}
+
+function Get-OnePathMatch {
+    <#
+    .SYNOPSIS
+        Returns a single path (file or directory) match for a given pattern,
+        throwing an error if there were no matches or more than one match.
+    #>
+    [CmdletBinding(DefaultParameterSetName = 'FileSet')]
+    param(
+        [Parameter(Mandatory)]
+        [string] $Path,
+
+        [Parameter(Mandatory)]
+        [string] $Pattern,
+
+        [Parameter(Mandatory, ParameterSetName = 'FileSet')]
+        [switch] $File,
+
+        [Parameter(Mandatory, ParameterSetName = 'DirSet')]
+        [switch] $Directory,
+
+        [switch] $Recurse
+    )
+
+    if (-not (Test-Path -LiteralPath $Path -PathType Container)) {
+        throw "Path not found or not a directory: $Path"
+    }
+
+    $gciArgs = @{
+        LiteralPath = $Path
+        ErrorAction = 'SilentlyContinue'
+    }
+
+    if ($Recurse) { $gciArgs['Recurse'] = $true }
+    if ($PSCmdlet.ParameterSetName -eq 'FileSet') {
+        $gciArgs['File'] = $true
+    }
+    else {
+        $gciArgs['Directory'] = $true
+    }
+
+    $pathMatches = @(
+        Get-ChildItem @gciArgs |
+        Where-Object { $_.Name -match $Pattern } |
+        Select-Object -ExpandProperty FullName
+    )
+
+    if ($pathMatches.Count -ne 1) {
+        $kind = if ($PSCmdlet.ParameterSetName -eq 'FileSet') { 'file' }
+        else { 'directory' }
+        $indented = ($pathMatches | ForEach-Object { "    $_" }) -join "`n"
+
+        $msg = @"
+Expected exactly one $kind name matching regex:
+  $Pattern
+under:
+  $Path
+Found:
+  $($pathMatches.Count)
+
+$indented
+"@
+        throw $msg
+    }
+
+    return $pathMatches[0]
+}
+
+Export-ModuleMember -Function Get-Python, Get-CudaMajor, Set-CtkPin, Get-CtkExtraFlavor, Convert-ToUnixPath, Get-RepoRoot, Get-CudaCcclWheel, Get-OnePathMatch
--- a/cccl_upstream/ci/windows/build_cub.ps1
+++ b/cccl_upstream/ci/windows/build_cub.ps1
@@ -0,0 +1,72 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("no-lid")]
+    [switch]$NO_LID_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("lid0")]
+    [switch]$LID0_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("lid1")]
+    [switch]$LID1_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("lid2")]
+    [switch]$LID2_SWITCH = $false
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cub"
+$artifactTags = @()
+
+if ($NO_LID_SWITCH) {
+    $artifactTags += "no_lid"
+    $PRESET = "cub-nolid"
+} elseif ($LID0_SWITCH) {
+    $artifactTags += "lid_0"
+    $PRESET = "cub-lid0"
+} elseif ($LID1_SWITCH) {
+    $artifactTags += "lid_1"
+    $PRESET = "cub-lid1"
+} elseif ($LID2_SWITCH) {
+    $artifactTags += "lid_2"
+    $PRESET = "cub-lid2"
+}
+$LOCAL_CMAKE_OPTIONS = "-DCMAKE_CXX_STANDARD=$CXX_STANDARD -DCMAKE_CUDA_STANDARD=$CXX_STANDARD"
+
+if ($CL_VERSION -lt [version]"19.20") {
+    $LOCAL_CMAKE_OPTIONS = "$LOCAL_CMAKE_OPTIONS -DCCCL_IGNORE_DEPRECATED_COMPILER=ON"
+}
+
+configure_and_build_preset "CUB" $PRESET $LOCAL_CMAKE_OPTIONS
+
+if ($env:GITHUB_ACTIONS) {
+    Write-Host "Packaging test artifacts..."
+    if ($artifactTags.Count -gt 0) {
+        & bash "./upload_cub_test_artifacts.sh" @artifactTags
+    } else {
+        & bash "./upload_cub_test_artifacts.sh"
+    }
+}
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/build_cuda_cccl_python.ps1
+++ b/cccl_upstream/ci/windows/build_cuda_cccl_python.ps1
@@ -0,0 +1,422 @@
+<#
+.SYNOPSIS
+    Build Python cuda-cccl wheels on Windows.
+
+.DESCRIPTION
+    This script is the Windows analog to the Linux ../build_cuda_cccl_python.sh
+    script.  It is responsible for building CUDA 12.x and CUDA 13.x wheels that
+    are then merged together into a singular cuda-cccl wheel.
+
+    A single CUDA 12.9 builder image (i.e. Docker devcontainer) is used to
+    build each distinct Python/MSVC combo.  Much like the Linux approach, this
+    script detects when launched via the outer 12.9 instance, builds a `cu12`
+    wheel, then dispatches a inner Docker instance (Docker-out-of-Docker) to
+    execute this script with `-OnlyCudaMajor 13 -SkipUpload` parameters, which
+    yields a `cu13` build.
+
+    Upon completion of the `cu13` build, the outer 12.9 container merges both
+    `cu12` and `cu13` wheels into a single cuda-cccl wheel, and uploads that
+    via the standard CCCL CI artifact upload mechanisms.
+
+.PARAMETER PyVersion
+    **Required.** The Python version to use for building the wheel, expressed
+    as `<major>.<minor>` (e.g. `3.11`) or a free-threaded version such as
+    `3.14t`.
+
+.PARAMETER OnlyCudaMajor
+    Optional. Restricts the build to a single CUDA major version (`12` or `13`).
+    When set, only that version is built and the *merge* step is skipped.
+
+.PARAMETER Cuda13Image
+    Optional. The Docker image name used for a nested build of the CUDA 13
+    wheel when the outer container defaults to CUDA 12.9.  The default value
+    matches the RAPIDS dev-container image that contains the required
+    toolchain: `rapidsai/devcontainers:26.06-cuda13.0-cl14.44-windows2022`.
+
+.PARAMETER SkipUpload
+    When set, prevents the final wheel(s) from being uploaded as a GitHub
+    Actions artifact even when the script detects it is running inside an
+    Action.
+
+.EXAMPLE
+    # Build a single cuda-cccl wheel for Python 3.13 (consisting of both CUDA
+    # 12 and 13 versions), and, if in CI, upload the resulting wheel as an
+    # artifact.
+    .\build_cuda_cccl_python.ps1 -PyVersion 3.11
+#>
+
+[CmdletBinding()]
+Param(
+    [Parameter(Mandatory = $true)]
+    [Alias("py-version")]
+    [ValidatePattern("^\d+\.\d+t?$")]
+    [string]$PyVersion,
+
+    [Parameter(Mandatory = $false)]
+    [ValidateSet('12', '13')]
+    [string]$OnlyCudaMajor,
+
+    [Parameter(Mandatory = $false)]
+    [string]$Cuda13Image = "rapidsai/devcontainers:26.06-cuda13.0-cl14.44-windows2022",
+
+    [Parameter(Mandatory = $false)]
+    [switch]$SkipUpload
+)
+
+$ErrorActionPreference = "Stop"
+
+# Import shared helpers.
+Import-Module "$PSScriptRoot/build_common.psm1"
+Import-Module "$PSScriptRoot/build_common_python.psm1" -Force
+
+# Resolve repo root from this script's location.
+$RepoRoot = Resolve-Path "$PSScriptRoot/../.."
+Write-Host "Repo root: $RepoRoot"
+
+# Get the full path to the python.exe for the version we need.
+Write-Host "Looking for Python version $PyVersion..."
+$PythonExe = Get-Python -Version $PyVersion
+Write-Host "Using Python: $PythonExe"
+& $PythonExe -m pip --version
+
+# Ensure MSVC is available.
+$clPath = (Get-Command cl).Source
+if (-not $clPath) {
+    throw "cl.exe not found in PATH. Run from a Developer PowerShell prompt."
+}
+Write-Host "Found cl.exe at: $clPath"
+
+function Resolve-CudaPathForMajor {
+    Param(
+        [Parameter(Mandatory = $true)]
+        [ValidateSet('12', '13')]
+        [string]$Major
+    )
+    $candidates = @()
+    Get-ChildItem Env: |
+    Where-Object { $_.Name -match "^CUDA_PATH_V${Major}_(\d+)$" } |
+    ForEach-Object {
+        $minor = [int]([regex]::Match(
+                $_.Name,
+                "^CUDA_PATH_V${Major}_(\d+)$"
+            ).Groups[1].Value)
+        $candidates += [PSCustomObject]@{
+            Minor = $minor;
+            Path  = $_.Value
+        }
+    }
+
+    if ($candidates.Count -gt 0) {
+        return ($candidates | Sort-Object -Property Minor -Descending |
+            Select-Object -First 1).Path
+    }
+
+    if ($env:CUDA_PATH) {
+        $maybe = $env:CUDA_PATH
+        $nvcc = Join-Path $maybe 'bin/nvcc.exe'
+        if (Test-Path $nvcc) {
+            $out = & $nvcc --version 2>&1
+            $text = ($out -join "`n")
+            if ($text -match 'release\s+(\d+)\.') {
+                if ($Matches[1] -eq $Major) {
+                    return $maybe
+                }
+            }
+        }
+    }
+
+    return $null
+}
+
+# If $OnlyCudaMajor is present, it means we're being launched from a
+# nested Docker container build (12.x launched a 13.x build via DooD).
+if ($OnlyCudaMajor) {
+    $CudaMajorsToBuild = @($OnlyCudaMajor)
+}
+else {
+    $CudaMajorsToBuild = @('12', '13')
+}
+$DoMerge = -not [bool]$OnlyCudaMajor
+
+# Base pip/CMake options
+$pipBaseConfigArgs = @(
+    '-C', 'cmake.define.CMAKE_C_COMPILER=cl.exe',
+    '-C', 'cmake.define.CMAKE_CXX_COMPILER=cl.exe'
+)
+
+$env:CMAKE_GENERATOR = "Ninja"
+
+# Ensure wheelhouse directories exist.
+$Wheelhouse = Join-Path $RepoRoot "wheelhouse"
+New-Item -ItemType Directory -Path $Wheelhouse -Force | Out-Null
+${null} = New-Item -ItemType Directory -Path (Join-Path $RepoRoot 'wheelhouse_cu12') -Force
+${null} = New-Item -ItemType Directory -Path (Join-Path $RepoRoot 'wheelhouse_cu13') -Force
+
+function Invoke-Cuda13NestedBuild {
+    <#
+    .SYNOPSIS
+        Run the nested Docker build for CUDA 13 when we are already inside a
+        CUDA 12 builder image.
+
+    .DESCRIPTION
+        This routine launches a Docker devcontainer CUDA 13 build for the given
+        Python version by way of Docker-out-of-Docker (DooD) facilities.
+    #>
+    [CmdletBinding()]
+    param (
+        [Parameter(Mandatory)] [string] $Cuda13Image,
+        [Parameter(Mandatory)] [string] $PyVersion,
+        [ValidateNotNullOrEmpty()] [string] $HostWorkspace = $env:HOST_WORKSPACE,
+        [ValidateNotNullOrEmpty()] [string] $ContainerWorkspace = $env:CONTAINER_WORKSPACE
+    )
+
+    # Validate required environment variables.
+    if (-not $HostWorkspace) {
+        throw "HOST_WORKSPACE env var is not set; required for DooD " +
+        "nested docker mounts on Windows."
+    }
+    if (-not $ContainerWorkspace) {
+        throw "CONTAINER_WORKSPACE env var is not set; required for " +
+        "DooD nested docker mounts on Windows."
+    }
+
+    # Validate Docker CLI availability.
+    if (-not (Get-Command docker -ErrorAction SilentlyContinue)) {
+        throw "docker CLI not found in the devcontainer image (required for DooD)."
+    }
+
+    Write-Host "Checking DooD connectivity..."
+    $dockerVersionOutput = & docker version 2>&1
+    $dockerExitCode = $LASTEXITCODE
+    $dockerVersionOutput | Out-Host
+    if ($dockerExitCode -ne 0) {
+        throw "DooD connectivity check failed (exit code $dockerExitCode). See Docker output above."
+    }
+    Write-Host "DooD appears to be working, continuing..."
+
+    # Detect outer-container resources so we can set sensible limits.
+    $os = Get-WmiObject -Class Win32_OperatingSystem
+    $totalGB = [math]::Floor($os.TotalVisibleMemorySize / 1MB) # KB -> GB
+    $procCount = [Environment]::ProcessorCount
+
+    # Leave a little head-room so the outer container doesn't starve
+    $memLimitGB = [math]::Max(2, [int]([math]::Floor($totalGB * 0.9)))
+    $cpuCount = [math]::Max(2, $procCount)
+
+    Write-Host "Launching nested Docker for CUDA 13 build using image: $Cuda13Image"
+    $targetFile = Join-Path $ContainerWorkspace 'ci\windows\build_cuda_cccl_python.ps1'
+    $dockerArgs = @(
+        'run', '--rm', '-i',
+        '--cpu-count', "$cpuCount",
+        '--memory', "${memLimitGB}g",
+        '--workdir', $ContainerWorkspace,
+        '--mount', "type=bind,source=$HostWorkspace,target=$ContainerWorkspace",
+        '--env', "py_version=$PyVersion",
+        '--env', "GITHUB_ACTIONS=$($env:GITHUB_ACTIONS)",
+        '--env', "GITHUB_RUN_ID=$($env:GITHUB_RUN_ID)",
+        '--env', "JOB_ID=$($env:JOB_ID)",
+        $Cuda13Image,
+        'PowerShell.exe', '-NoLogo', '-NoProfile', '-ExecutionPolicy', 'Bypass',
+        '-File', $targetFile,
+        '-py-version', $PyVersion,
+        '-OnlyCudaMajor', '13',
+        '-SkipUpload'
+    )
+
+    Write-Host ("About to invoke: docker " + ($dockerArgs -join ' '))
+    Invoke-Checked { & docker @dockerArgs } 'Nested CUDA 13 wheel build failed'
+}
+
+function Build-CudaCcclWheel {
+    <#
+    .SYNOPSIS
+        Perform the regular wheel build for a given CUDA major version.
+
+    .DESCRIPTION
+        This routine is used to build both CUDA 12 and CUDA 13 based wheels,
+        and is called from normal "outer" Docker containers, as well as the
+        "inner" nested ones.
+    #>
+    [CmdletBinding()]
+    param (
+        [Parameter(Mandatory)] [ValidateSet('12', '13')] [string] $Major,
+        [Parameter(Mandatory)] [string] $RepoRoot,
+        [Parameter(Mandatory)] [string] $PythonExe,
+        [Parameter(Mandatory)] [string[]] $PipBaseConfigArgs
+    )
+
+    # Resolve CUDA toolkit location for the requested major version.
+    $CudaPathForMajor = Resolve-CudaPathForMajor -Major $Major
+    if (-not $CudaPathForMajor) {
+        throw "CUDA Toolkit $Major not found. Ensure CUDA_PATH_V${Major}_* " +
+        "is set or matching toolkit is installed."
+    }
+
+    $NvccForMajor = Join-Path $CudaPathForMajor 'bin/nvcc.exe'
+    if (-not (Test-Path $NvccForMajor)) {
+        throw "nvcc not found at $NvccForMajor"
+    }
+
+    # Convert Windows paths to Unix-style for CMake
+    $NvccUnix = Convert-ToUnixPath $NvccForMajor
+    $CudaUnix = Convert-ToUnixPath $CudaPathForMajor
+
+    # Build the pip configuration arguments that inject the CUDA toolchain.
+    $pipConfigArgs = $PipBaseConfigArgs + @(
+        '-C', "cmake.define.CMAKE_CUDA_COMPILER=$NvccUnix",
+        '-C', "cmake.define.CUDAToolkit_ROOT=$CudaUnix"
+    )
+
+    $extra = "cu$Major"
+    # Use separate output directories for 12 vs 13.
+    $outDir = Join-Path $RepoRoot "wheelhouse_$extra"
+
+    Write-Host "Building cuda-cccl wheel for CUDA $Major at $CudaPathForMajor..."
+
+    # Run pip wheel to build the wheel.
+    $pythonArgs = @(
+        '-m', 'pip', 'wheel',
+        '-w', $outDir,
+        ".[${extra}]",
+        '-v'
+    ) + $pipConfigArgs
+
+    Write-Host ("python " + ($pythonArgs -join ' '))
+    Invoke-Checked { & $PythonExe @pythonArgs } "Wheel build failed for CUDA $Major"
+
+    # Normalise the wheel filename (append .cu12/.cu13) and prune duplicates.
+    $builtWheel = Get-OnePathMatch -Path $outDir `
+        -Pattern '^cuda_cccl-.*\.whl' `
+        -File
+    if (-not $builtWheel) {
+        throw "Failed to locate built wheel in $outDir for CUDA $Major"
+    }
+
+    $builtName = [System.IO.Path]::GetFileName($builtWheel)
+    if ($builtName -notmatch ".cu$Major\.whl$") {
+        $newName = ([System.IO.Path]::GetFileNameWithoutExtension($builtName)) `
+            + ".cu$Major.whl"
+        Write-Host "Renaming wheel to: $newName"
+        Rename-Item -Path $builtWheel -NewName $newName -Force
+    }
+
+    # Remove any stray wheels that lack the .cuXX suffix.
+    Get-ChildItem -Path $outDir -Filter 'cuda_cccl-*.whl' |
+    Where-Object { $_.Name -notmatch "\.cu$Major\.whl$" } |
+    ForEach-Object {
+        Write-Host "Removing duplicate wheel: $($_.FullName)"
+        Remove-Item -Force $_.FullName
+    }
+}
+
+# Main build entry code.
+Push-Location (Join-Path $RepoRoot 'python/cuda_cccl')
+try {
+    foreach ($major in $CudaMajorsToBuild) {
+
+        # Nested Docker build for CUDA 13 for when we are currently inside a
+        # CUDA 12 image.
+        if (-not $OnlyCudaMajor -and $major -eq '13' -and $Cuda13Image) {
+            Invoke-Cuda13NestedBuild `
+                -Cuda13Image $Cuda13Image `
+                -PyVersion $PyVersion
+
+            continue
+        }
+
+        # Perform a normal build for the current major version.  This may
+        # be invoked from either an "outer" or inner "nested" image.
+        Build-CudaCcclWheel `
+            -Major $major `
+            -RepoRoot $RepoRoot `
+            -PythonExe $PythonExe `
+            -PipBaseConfigArgs $pipBaseConfigArgs
+    }
+}
+finally {
+    Pop-Location
+}
+
+
+# Merge the two major-version wheels (if both were built).  This will fail if
+# either wheel can't be found.  This only runs on the outer (non-nested)
+# container image.
+if ($DoMerge) {
+
+    $Cu12Wheel = Get-OnePathMatch `
+        -Path (Join-Path $RepoRoot 'wheelhouse_cu12') `
+        -Pattern '^cuda_cccl-.*\.cu12\.whl' `
+        -File
+
+    $Cu13Wheel = Get-OnePathMatch `
+        -Path (Join-Path $RepoRoot 'wheelhouse_cu13') `
+        -Pattern '^cuda_cccl-.*\.cu13\.whl' `
+        -File
+
+    Write-Host "Found CUDA 12 wheel: $Cu12Wheel"
+    Write-Host "Found CUDA 13 wheel: $Cu13Wheel"
+
+    Write-Host 'Merging CUDA wheels...'
+    Invoke-Checked { & $PythonExe -m pip install wheel | Write-Host } 'Failed to install wheel for merging'
+
+    $WheelhouseMerged = Join-Path $RepoRoot 'wheelhouse_merged'
+    ${null} = New-Item -ItemType Directory -Path $WheelhouseMerged -Force
+
+    $mergePy = Join-Path $RepoRoot 'python/cuda_cccl/merge_cuda_wheels.py'
+    Invoke-Checked { & $PythonExe $mergePy $Cu12Wheel $Cu13Wheel --output-dir $WheelhouseMerged } 'Merging wheels failed'
+
+    # Clean up the per-major directories and move the merged wheel into the
+    # final location.
+    Get-ChildItem $Wheelhouse -Filter '*.whl' |
+    ForEach-Object {
+        Remove-Item -Force $_.FullName
+    }
+    $MergedWheel = Get-OnePathMatch `
+        -Path $WheelhouseMerged `
+        -Pattern '^cuda_cccl-.*\.whl' `
+        -File
+    Move-Item -Force $MergedWheel $Wheelhouse
+
+    Remove-Item $WheelhouseMerged -Recurse -Force -ErrorAction SilentlyContinue
+    Remove-Item (Join-Path $RepoRoot 'wheelhouse_cu12') `
+        -Recurse -Force -ErrorAction SilentlyContinue
+    Remove-Item (Join-Path $RepoRoot 'wheelhouse_cu13') `
+        -Recurse -Force -ErrorAction SilentlyContinue
+
+    Write-Host 'Final wheels in wheelhouse:'
+    Get-ChildItem $Wheelhouse -Filter '*.whl' |
+    ForEach-Object {
+        Write-Host " - $($_.Name)"
+    }
+}
+
+# If it turns out we need delvewheel, we'd handle it here, after the merging
+# of wheels.  The two DLLs that seem like they might be problematic are
+# msvc140p.dll, and dbghelp.dll.  The former comes from llvmlite, upon which
+# we depend.  Dbghelp.dll ships in C:\Windows\System32, but that will often
+# be a much older version compared to the one used by Visual Studio.  We only
+# use one symbol from Dbghelp.dll: UnDecorateSymbolName, which is used by
+# nvrtc.  If we encounter weird issues with c.parallel jit compilation and
+# nvrtc in the wild on Windows, an out-of-date Dbghelp.dll could possibly be
+# the culprit.
+#
+# For now, though, it doesn't appear to be necessary.
+
+# Optionally upload the wheel artifact.
+if ($env:GITHUB_ACTIONS -and -not $SkipUpload) {
+    Push-Location $RepoRoot
+    try {
+        Write-Host 'GITHUB_ACTIONS detected; uploading wheel artifact'
+        $wheelArtifactName = (& bash -lc "ci/util/workflow/get_wheel_artifact_name.sh").Trim()
+        if (-not $wheelArtifactName) {
+            throw 'Failed to resolve wheel artifact name'
+        }
+        Write-Host "Wheel artifact name: $wheelArtifactName"
+
+        $uploadCmd = "ci/util/artifacts/upload.sh $wheelArtifactName 'wheelhouse/.*'"
+        Invoke-Checked { & bash -lc $uploadCmd } 'Wheel artifact upload failed'
+    }
+    finally {
+        Pop-Location
+    }
+}
--- a/cccl_upstream/ci/windows/build_cudax.ps1
+++ b/cccl_upstream/ci/windows/build_cudax.ps1
@@ -0,0 +1,31 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(20)]
+    [int]$CXX_STANDARD = 20,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Remove-Module -Name build_common -ErrorAction SilentlyContinue
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cudax"
+$LOCAL_CMAKE_OPTIONS = "-DCMAKE_CXX_STANDARD=$CXX_STANDARD -DCMAKE_CUDA_STANDARD=$CXX_STANDARD"
+
+configure_and_build_preset "CUDA Experimental" $PRESET $LOCAL_CMAKE_OPTIONS
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/build_libcudacxx.ps1
+++ b/cccl_upstream/ci/windows/build_libcudacxx.ps1
@@ -0,0 +1,48 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "libcudacxx"
+$LOCAL_CMAKE_OPTIONS = "-DCMAKE_CXX_STANDARD=$CXX_STANDARD -DCMAKE_CUDA_STANDARD=$CXX_STANDARD"
+
+$uploadTestArtifacts = $false
+if ($env:GITHUB_ACTIONS) {
+    & bash "./util/workflow/has_consumers.sh"
+    $uploadTestArtifacts = $LASTEXITCODE -eq 0
+    if ($uploadTestArtifacts) {
+        $env:LIT_OPTS = "$env:LIT_OPTS -Dtest_executable_mode=build".Trim()
+    }
+}
+
+configure_and_build_preset "libcudacxx" $PRESET $LOCAL_CMAKE_OPTIONS
+
+if ($uploadTestArtifacts) {
+    Write-Host "Packaging test artifacts..."
+    Invoke-Checked {
+        & bash "./upload_libcudacxx_test_artifacts.sh"
+    } "Packaging test artifacts failed"
+}
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/build_thrust.ps1
+++ b/cccl_upstream/ci/windows/build_thrust.ps1
@@ -0,0 +1,37 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "thrust"
+$LOCAL_CMAKE_OPTIONS = "-DCMAKE_CXX_STANDARD=$CXX_STANDARD -DCMAKE_CUDA_STANDARD=$CXX_STANDARD"
+
+configure_and_build_preset "Thrust" $PRESET $LOCAL_CMAKE_OPTIONS
+
+if ($env:GITHUB_ACTIONS) {
+    Write-Host "Packaging test artifacts..."
+    & bash "./upload_thrust_test_artifacts.sh"
+}
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/run_cpu_bisect.ps1
+++ b/cccl_upstream/ci/windows/run_cpu_bisect.ps1
@@ -0,0 +1,41 @@
+Param(
+    [Parameter(ValueFromRemainingArguments = $true)]
+    [string[]]$PassthroughArgs
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+if ($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+if ($null -eq $PassthroughArgs) {
+    $PassthroughArgs = @()
+}
+
+Import-Module "$PSScriptRoot/build_common.psm1"
+
+$ciDirWindows = (Resolve-Path (Join-Path $PSScriptRoot "..")).Path
+$repoDirWindows = (Resolve-Path (Join-Path $ciDirWindows "..")).Path
+
+$ciDir = $ciDirWindows -replace "\\", "/"
+$repoDir = $repoDirWindows -replace "\\", "/"
+$utilScript = "$ciDir/util/git_bisect.sh"
+
+$argString = if ($PassthroughArgs.Count -gt 0) { " " + ($PassthroughArgs -join " ") } else { "" }
+
+$bashCommand = "cd $repoDir; $utilScript$argString"
+Write-Host $bashCommand -ForegroundColor Blue
+
+& bash -lc $bashCommand
+$exitCode = $LASTEXITCODE
+
+if ($CURRENT_PATH -ne "ci") {
+    popd
+}
+
+if ($exitCode -ne 0) {
+    exit $exitCode
+}
--- a/cccl_upstream/ci/windows/run_cpu_target.ps1
+++ b/cccl_upstream/ci/windows/run_cpu_target.ps1
@@ -0,0 +1,41 @@
+Param(
+    [Parameter(ValueFromRemainingArguments = $true)]
+    [string[]]$PassthroughArgs
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+if ($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+if ($null -eq $PassthroughArgs) {
+    $PassthroughArgs = @()
+}
+
+Import-Module "$PSScriptRoot/build_common.psm1"
+
+$ciDirWindows = (Resolve-Path (Join-Path $PSScriptRoot "..")).Path
+$repoDirWindows = (Resolve-Path (Join-Path $ciDirWindows "..")).Path
+
+$ciDir = $ciDirWindows -replace "\\", "/"
+$repoDir = $repoDirWindows -replace "\\", "/"
+$utilScript = "$ciDir/util/build_and_test_targets.sh"
+
+$argString = if ($PassthroughArgs.Count -gt 0) { " " + ($PassthroughArgs -join " ") } else { "" }
+
+$bashCommand = "cd $repoDir; $utilScript$argString"
+Write-Host $bashCommand -ForegroundColor Blue
+
+& bash -lc $bashCommand
+$exitCode = $LASTEXITCODE
+
+if ($CURRENT_PATH -ne "ci") {
+    popd
+}
+
+if ($exitCode -ne 0) {
+    exit $exitCode
+}
--- a/cccl_upstream/ci/windows/run_gpu_bisect.ps1
+++ b/cccl_upstream/ci/windows/run_gpu_bisect.ps1
@@ -0,0 +1,41 @@
+Param(
+    [Parameter(ValueFromRemainingArguments = $true)]
+    [string[]]$PassthroughArgs
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+if ($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+if ($null -eq $PassthroughArgs) {
+    $PassthroughArgs = @()
+}
+
+Import-Module "$PSScriptRoot/build_common.psm1"
+
+$ciDirWindows = (Resolve-Path (Join-Path $PSScriptRoot "..")).Path
+$repoDirWindows = (Resolve-Path (Join-Path $ciDirWindows "..")).Path
+
+$ciDir = $ciDirWindows -replace "\\", "/"
+$repoDir = $repoDirWindows -replace "\\", "/"
+$utilScript = "$ciDir/util/git_bisect.sh"
+
+$argString = if ($PassthroughArgs.Count -gt 0) { " " + ($PassthroughArgs -join " ") } else { "" }
+
+$bashCommand = "cd $repoDir; $utilScript$argString"
+Write-Host $bashCommand -ForegroundColor Blue
+
+& bash -lc $bashCommand
+$exitCode = $LASTEXITCODE
+
+if ($CURRENT_PATH -ne "ci") {
+    popd
+}
+
+if ($exitCode -ne 0) {
+    exit $exitCode
+}
--- a/cccl_upstream/ci/windows/run_gpu_target.ps1
+++ b/cccl_upstream/ci/windows/run_gpu_target.ps1
@@ -0,0 +1,41 @@
+Param(
+    [Parameter(ValueFromRemainingArguments = $true)]
+    [string[]]$PassthroughArgs
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+if ($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+if ($null -eq $PassthroughArgs) {
+    $PassthroughArgs = @()
+}
+
+Import-Module "$PSScriptRoot/build_common.psm1"
+
+$ciDirWindows = (Resolve-Path (Join-Path $PSScriptRoot "..")).Path
+$repoDirWindows = (Resolve-Path (Join-Path $ciDirWindows "..")).Path
+
+$ciDir = $ciDirWindows -replace "\\", "/"
+$repoDir = $repoDirWindows -replace "\\", "/"
+$utilScript = "$ciDir/util/build_and_test_targets.sh"
+
+$argString = if ($PassthroughArgs.Count -gt 0) { " " + ($PassthroughArgs -join " ") } else { "" }
+
+$bashCommand = "cd $repoDir; $utilScript$argString"
+Write-Host $bashCommand -ForegroundColor Blue
+
+& bash -lc $bashCommand
+$exitCode = $LASTEXITCODE
+
+if ($CURRENT_PATH -ne "ci") {
+    popd
+}
+
+if ($exitCode -ne 0) {
+    exit $exitCode
+}
--- a/cccl_upstream/ci/windows/test_cccl_c_parallel.ps1
+++ b/cccl_upstream/ci/windows/test_cccl_c_parallel.ps1
@@ -0,0 +1,31 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+# Build first
+$buildCmd = "$PSScriptRoot/build_cccl_c_parallel.ps1 -arch '$CUDA_ARCH' -cmake-options '$CMAKE_OPTIONS'"
+Write-Host "Running: $buildCmd"
+Invoke-Expression $buildCmd
+
+Remove-Module -Name build_common -ErrorAction SilentlyContinue
+Import-Module -Name "$PSScriptRoot/build_common.psm1" -ArgumentList @(20, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cccl-c-parallel"
+test_preset "CCCL C Parallel" "$PRESET"
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/test_cccl_c_parallel_v2.ps1
+++ b/cccl_upstream/ci/windows/test_cccl_c_parallel_v2.ps1
@@ -0,0 +1,30 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Remove-Module -Name build_common -ErrorAction SilentlyContinue
+Import-Module -Name "$PSScriptRoot/build_common.psm1" -ArgumentList @(20, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cccl-c-parallel-v2"
+$LOCAL_CMAKE_OPTIONS = ""
+
+configure_and_build_preset "CCCL C Parallel v2 (HostJIT)" $PRESET $LOCAL_CMAKE_OPTIONS
+
+test_preset "CCCL C Parallel v2 (HostJIT)" "$PRESET"
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/test_cub.ps1
+++ b/cccl_upstream/ci/windows/test_cub.ps1
@@ -0,0 +1,73 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("no-lid")]
+    [switch]$NO_LID_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("lid0")]
+    [switch]$LID0_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("lid1")]
+    [switch]$LID1_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("lid2")]
+    [switch]$LID2_SWITCH = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Import-Module -Name "$PSScriptRoot/build_common.psm1" -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cub"
+$artifactTag = ""
+$variantArg = ""
+if ($NO_LID_SWITCH) {
+    $artifactTag = "no_lid"
+    $PRESET = "cub-nolid"
+    $variantArg = "-no-lid"
+} elseif ($LID0_SWITCH) {
+    $artifactTag = "lid_0"
+    $PRESET = "cub-lid0"
+    $variantArg = "-lid0"
+} elseif ($LID1_SWITCH) {
+    $artifactTag = "lid_1"
+    $PRESET = "cub-lid1"
+    $variantArg = "-lid1"
+} elseif ($LID2_SWITCH) {
+    $artifactTag = "lid_2"
+    $PRESET = "cub-lid2"
+    $variantArg = "-lid2"
+}
+
+if ($env:GITHUB_ACTIONS -and $artifactTag) {
+    $producerId = (& bash "./util/workflow/get_producer_id.sh").Trim()
+    $artifactName = "z_cub-test-artifacts-$env:DEVCONTAINER_NAME-$producerId-$artifactTag"
+    Write-Host "Unpacking artifact '$artifactName'"
+    & bash "./util/artifacts/download_packed.sh" "$artifactName" "../"
+} else {
+    $buildCmd = "$PSScriptRoot/build_cub.ps1 -std $CXX_STANDARD -arch '$CUDA_ARCH' -cmake-options '$CMAKE_OPTIONS' $variantArg"
+    Write-Host "Running: $buildCmd"
+    Invoke-Expression $buildCmd
+}
+
+test_preset "CUB ($PRESET)" "$PRESET"
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/test_cuda_cccl_examples_python.ps1
+++ b/cccl_upstream/ci/windows/test_cuda_cccl_examples_python.ps1
@@ -0,0 +1,50 @@
+Param(
+    [Parameter(Mandatory = $true)]
+    [Alias("py-version")]
+    [ValidatePattern("^\d+\.\d+t?$")]
+    [string]$PyVersion,
+
+    [Alias("ctk-mode")]
+    [string]$CtkMode = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+# Import shared helpers
+Import-Module "$PSScriptRoot/build_common.psm1"
+Import-Module "$PSScriptRoot/build_common_python.psm1"
+
+$python = Get-Python -Version $PyVersion
+$cudaMajor = Get-CudaMajor
+$ctkFlavor = Get-CtkExtraFlavor $CtkMode
+
+# Pin cuda-toolkit to the container's CTK minor (-ctk-mode latest
+# opts out). See build_common_python.psm1.
+Set-CtkPin $CtkMode
+
+$repoRoot = Get-RepoRoot
+
+${wheelPath} = Get-CudaCcclWheel
+
+# pytest-benchmark is for the host-benchmark smoke test below.
+Invoke-Checked { & $python -m pip install -U pip pytest pytest-xdist pytest-benchmark } "Failed to install pytest / pytest-xdist / pytest-benchmark"
+# CuPy is required by the cuda.compute examples and is not part of the test extras
+Invoke-Checked { & $python -m pip install "${wheelPath}[test-$ctkFlavor$cudaMajor]" "cupy-cuda${cudaMajor}x" } "Failed to install cuda_cccl test extra / cupy"
+
+Push-Location (Join-Path $repoRoot "python/cuda_cccl/tests")
+try {
+    Invoke-Checked { & $python -m pytest -n 6 test_examples.py } "examples tests failed"
+}
+finally { Pop-Location }
+
+# Smoke-test the host-overhead benchmark harness: run every benchmark case
+# exactly once (pass/fail only, no timing) so harness rot fails CI here instead
+# of silently surviving until someone runs the perf suite. --benchmark-disable
+# makes pytest-benchmark invoke each benchmarked callable a single time. This
+# lane already installs cupy + numba (for the examples), which the benchmark
+# suite also needs, so only pytest-benchmark is added above.
+Push-Location (Join-Path $repoRoot "python/cuda_cccl/benchmarks/compute/host")
+try {
+    Invoke-Checked { & $python -m pytest -v --benchmark-disable . } "host benchmark smoke test failed"
+}
+finally { Pop-Location }
--- a/cccl_upstream/ci/windows/test_cuda_cccl_headers_python.ps1
+++ b/cccl_upstream/ci/windows/test_cuda_cccl_headers_python.ps1
@@ -0,0 +1,36 @@
+Param(
+    [Parameter(Mandatory = $true)]
+    [Alias("py-version")]
+    [ValidatePattern("^\d+\.\d+t?$")]
+    [string]$PyVersion,
+
+    [Alias("ctk-mode")]
+    [string]$CtkMode = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+# Import shared helpers
+Import-Module "$PSScriptRoot/build_common.psm1"
+Import-Module "$PSScriptRoot/build_common_python.psm1"
+
+$python = Get-Python -Version $PyVersion
+$cudaMajor = Get-CudaMajor
+$ctkFlavor = Get-CtkExtraFlavor $CtkMode
+
+# Pin cuda-toolkit to the container's CTK minor (-ctk-mode latest
+# opts out). See build_common_python.psm1.
+Set-CtkPin $CtkMode
+
+$repoRoot = Get-RepoRoot
+
+${wheelPath} = Get-CudaCcclWheel
+
+Invoke-Checked { & $python -m pip install -U pip pytest pytest-xdist } "Failed to install pytest / pytest-xdist"
+Invoke-Checked { & $python -m pip install "${wheelPath}[test-$ctkFlavor$cudaMajor]" } "Failed to install cuda_cccl test extra"
+
+Push-Location (Join-Path $repoRoot "python/cuda_cccl/tests")
+try {
+    Invoke-Checked { & $python -m pytest -n auto -v headers/ } "headers tests failed"
+}
+finally { Pop-Location }
--- a/cccl_upstream/ci/windows/test_cuda_compute_minimal_python.ps1
+++ b/cccl_upstream/ci/windows/test_cuda_compute_minimal_python.ps1
@@ -0,0 +1,79 @@
+Param(
+    [Parameter(Mandatory = $true)]
+    [Alias("py-version")]
+    [ValidatePattern("^\d+\.\d+t?$")]
+    [string]$PyVersion,
+
+    [Alias("ctk-mode")]
+    [string]$CtkMode = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+# Import shared helpers
+Import-Module "$PSScriptRoot/build_common.psm1"
+Import-Module "$PSScriptRoot/build_common_python.psm1"
+
+$python = Get-Python -Version $PyVersion
+$cudaMajor = Get-CudaMajor
+$ctkFlavor = Get-CtkExtraFlavor $CtkMode
+
+# Pin cuda-toolkit to the container's CTK minor (-ctk-mode latest
+# opts out). See build_common_python.psm1.
+Set-CtkPin $CtkMode
+
+$repoRoot = Get-RepoRoot
+
+$wheelPath = Get-CudaCcclWheel
+
+# Install cuda_cccl with the minimal CUDA extra. This intentionally avoids the
+# full cu* extras because those pull in numba/numba-cuda.
+Invoke-Checked { & $python -m pip install -U pip pytest pytest-xdist } "Failed to install pytest / pytest-xdist"
+Invoke-Checked { & $python -m pip install "$wheelPath[minimal-$ctkFlavor$cudaMajor]" } "Failed to install cuda_cccl minimal extra"
+
+Push-Location (Join-Path $repoRoot "python/cuda_cccl/tests")
+try {
+    Invoke-Checked { & $python -m pytest -n 6 -v compute/test_no_numba.py } "test_no_numba.py failed"
+
+    if ($PyVersion -eq "3.14t") {
+        # Select only tests that support the minimal extra so pytest does not
+        # collect tests that import numba-cuda and re-enable the GIL. These tests
+        # provide their own worker threads, so keep pytest itself in a single
+        # process. The serialization node-ids are module-skipped on the v2
+        # backend today and will start running there automatically once v2 gains
+        # serialization support.
+        Invoke-Checked {
+            & $python -m pytest -n 0 -v `
+                compute/test_free_threading_stress.py `
+                compute/test_multi_cc_serialization.py::test_aot_build_result_load_failure_is_shared_and_retryable `
+                compute/test_multi_cc_serialization.py::test_aot_serialization_waits_for_canonical_first_load
+        } "free-threading stress / serialization tests failed"
+
+        # Broad thread-safety sweep (pytest-run-parallel): re-run the numba-free
+        # functional suite with each test executed concurrently across threads
+        # (barrier-synchronized start), stressing the process-wide build cache,
+        # single-flight coordination, and the Cython bindings from many threads at
+        # once. Complements test_free_threading_stress.py above, which targets
+        # specific shared-object scenarios by hand. -n 0 so the threads share one
+        # interpreter.
+        #
+        # --parallel-threads=2 matches CuPy's free-threading CI (the closest GPU
+        # precedent); a small fixed count bounds GPU-memory pressure from
+        # concurrent kernels and stays reproducible across runners, unlike =auto
+        # (the runner's logical-core count).
+        #
+        # pytest-run-parallel is only used by this sweep, so install it on the
+        # 3.14t path rather than for every minimal (e.g. non-free-threaded 3.14)
+        # run.
+        Invoke-Checked { & $python -m pip install pytest-run-parallel } "Failed to install pytest-run-parallel"
+
+        # Fail fast if the interpreter is not actually GIL-free (wrong build /
+        # PYTHON_GIL=1): pytest-run-parallel does NOT catch a GIL that is enabled
+        # from the start -- it would run threads GIL-serialized and pass
+        # vacuously. (A GIL *re-enabled mid-run* by a non-free-threaded import IS
+        # caught by the plugin, which is why we do not pass --ignore-gil-enabled.)
+        Invoke-Checked { & $python -c "import sys; assert not sys._is_gil_enabled(), 'GIL is enabled; parallel sweep has no signal'" } "interpreter is not GIL-free; parallel sweep has no signal"
+        Invoke-Checked { & $python -m pytest -n 0 -v --parallel-threads=2 compute/test_no_numba.py } "parallel-threads sweep failed"
+    }
+}
+finally { Pop-Location }
--- a/cccl_upstream/ci/windows/test_cuda_compute_python.ps1
+++ b/cccl_upstream/ci/windows/test_cuda_compute_python.ps1
@@ -0,0 +1,37 @@
+Param(
+    [Parameter(Mandatory = $true)]
+    [Alias("py-version")]
+    [ValidatePattern("^\d+\.\d+t?$")]
+    [string]$PyVersion,
+
+    [Alias("ctk-mode")]
+    [string]$CtkMode = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+# Import shared helpers
+Import-Module "$PSScriptRoot/build_common.psm1"
+Import-Module "$PSScriptRoot/build_common_python.psm1"
+
+$python = Get-Python -Version $PyVersion
+$cudaMajor = Get-CudaMajor
+$ctkFlavor = Get-CtkExtraFlavor $CtkMode
+
+# Pin cuda-toolkit to the container's CTK minor (-ctk-mode latest
+# opts out). See build_common_python.psm1.
+Set-CtkPin $CtkMode
+
+$repoRoot = Get-RepoRoot
+
+$wheelPath = Get-CudaCcclWheel
+
+Invoke-Checked { & $python -m pip install -U pip pytest pytest-xdist } "Failed to install pytest / pytest-xdist"
+Invoke-Checked { & $python -m pip install "$wheelPath[test-$ctkFlavor$cudaMajor]" } "Failed to install cuda_cccl test extra"
+
+Push-Location (Join-Path $repoRoot "python/cuda_cccl/tests")
+try {
+    Invoke-Checked { & $python -m pytest -n 6 -v compute/ -m "not large and not free_threading" } "compute tests (not large) failed"
+    Invoke-Checked { & $python -m pytest -n 0 -v compute/ -m "large and not free_threading" } "compute tests (large) failed"
+}
+finally { Pop-Location }
--- a/cccl_upstream/ci/windows/test_cudax.ps1
+++ b/cccl_upstream/ci/windows/test_cudax.ps1
@@ -0,0 +1,35 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(20)]
+    [int]$CXX_STANDARD = 20,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+# Build first
+$buildCmd = "$PSScriptRoot/build_cudax.ps1 -std $CXX_STANDARD -arch '$CUDA_ARCH' -cmake-options '$CMAKE_OPTIONS'"
+Write-Host "Running: $buildCmd"
+Invoke-Expression $buildCmd
+
+Import-Module -Name "$PSScriptRoot/build_common.psm1" -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "cudax"
+test_preset "CUDA Experimental" "$PRESET"
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/test_libcudacxx.ps1
+++ b/cccl_upstream/ci/windows/test_libcudacxx.ps1
@@ -0,0 +1,53 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Import-Module -Name "$PSScriptRoot/build_common.psm1" -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+if ($env:GITHUB_ACTIONS) {
+    $producerId = & bash "./util/workflow/get_producer_id.sh"
+    if ($LASTEXITCODE -ne 0) {
+        throw "Finding the producer job failed (exit code $LASTEXITCODE)"
+    }
+    $producerId = "$producerId".Trim()
+    $artifactName = "z_libcudacxx-test-artifacts-$env:DEVCONTAINER_NAME-$producerId"
+    Write-Host "Unpacking artifact '$artifactName'"
+    Invoke-Checked {
+        & bash "./util/artifacts/download_packed.sh" "$artifactName" "../"
+    } "Downloading test artifacts failed"
+} else {
+    $buildCmd = "$PSScriptRoot/build_libcudacxx.ps1 -std $CXX_STANDARD -arch '$CUDA_ARCH' -cmake-options '$CMAKE_OPTIONS'"
+    Write-Host "Running: $buildCmd"
+    Invoke-Expression $buildCmd
+}
+
+if ($env:GITHUB_ACTIONS) {
+    test_preset "libcudacxx (CTest)" "libcudacxx-ctest"
+    $env:LIT_OPTS = "$env:LIT_OPTS -Dtest_executable_mode=replay".Trim()
+    test_preset "libcudacxx (lit replay)" "libcudacxx-lit"
+} else {
+    test_preset "libcudacxx (CTest)" "libcudacxx-ctest-cpp${CXX_STANDARD}"
+    test_preset "libcudacxx (lit)"   "libcudacxx-lit-cpp${CXX_STANDARD}"
+}
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/test_packaging.ps1
+++ b/cccl_upstream/ci/windows/test_packaging.ps1
@@ -0,0 +1,37 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+Import-Module $PSScriptRoot/build_common.psm1 -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+$PRESET = "packaging"
+$LOCAL_CMAKE_OPTIONS = ""
+
+if ($env:GITHUB_SHA) {
+    $LOCAL_CMAKE_OPTIONS = '"-DCCCL_EXAMPLE_CPM_TAG={0}"' -f $env:GITHUB_SHA
+}
+
+configure_preset "Packaging" $PRESET $LOCAL_CMAKE_OPTIONS
+test_preset "Packaging" $PRESET
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}
--- a/cccl_upstream/ci/windows/test_thrust.ps1
+++ b/cccl_upstream/ci/windows/test_thrust.ps1
@@ -0,0 +1,62 @@
+Param(
+    [Parameter(Mandatory = $false)]
+    [Alias("std")]
+    [ValidateNotNullOrEmpty()]
+    [ValidateSet(17, 20)]
+    [int]$CXX_STANDARD = 17,
+    [Parameter(Mandatory = $false)]
+    [Alias("arch")]
+    [string]$CUDA_ARCH = "",
+    [Parameter(Mandatory = $false)]
+    [Alias("cpu-only")]
+    [switch]$CPU_ONLY = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("gpu-only")]
+    [switch]$GPU_ONLY = $false,
+    [Parameter(Mandatory = $false)]
+    [Alias("cmake-options")]
+    [string]$CMAKE_OPTIONS = ""
+)
+
+$ErrorActionPreference = "Stop"
+
+$CURRENT_PATH = Split-Path $pwd -leaf
+If($CURRENT_PATH -ne "ci") {
+    Write-Host "Moving to ci folder"
+    pushd "$PSScriptRoot/.."
+}
+
+if ($CPU_ONLY) {
+    $artifactTag = "test_cpu"
+    $presets = @("thrust-cpu")
+} elseif ($GPU_ONLY) {
+    $artifactTag = "test_gpu"
+    $presets = @("thrust-gpu")
+} else {
+    if ($env:GITHUB_ACTIONS) {
+        throw "Error: test_thrust.ps1 requires -cpu-only or -gpu-only in CI"
+    }
+    $artifactTag = ""
+    $presets = @("thrust-cpu", "thrust-gpu")
+}
+
+if ($env:GITHUB_ACTIONS -and $artifactTag) {
+    $producerId = (& bash "./util/workflow/get_producer_id.sh").Trim()
+    $artifactName = "z_thrust-test-artifacts-$env:DEVCONTAINER_NAME-$producerId-$artifactTag"
+    Write-Host "Unpacking artifact '$artifactName'"
+    & bash "./util/artifacts/download_packed.sh" "$artifactName" "../"
+} else {
+    $cmd = "$PSScriptRoot/build_thrust.ps1 -std $CXX_STANDARD -arch '$CUDA_ARCH' -cmake-options '$CMAKE_OPTIONS'"
+    Write-Host "Running: $cmd"
+    Invoke-Expression $cmd
+}
+
+Import-Module -Name "$PSScriptRoot/build_common.psm1" -ArgumentList @($CXX_STANDARD, $CUDA_ARCH, $CMAKE_OPTIONS)
+
+foreach ($preset in $presets) {
+    test_preset "Thrust ($preset)" $preset
+}
+
+If($CURRENT_PATH -ne "ci") {
+    popd
+}