ayo
247fe1092d
Implement native NGX feature-13 RR through a LUID-matched D3D12/Vulkan shared-memory and timeline-fence bridge. Add owned parameter snapshots, GPU resource lifetime handling, inherited/enhanced barriers, optional guides, sampled input formats and depth-plane support with stencil preservation. Separate RR initialization from the SR CUDA runtime and expose explicit backend/model selection through player configuration. Add synthetic bridge, snapshot and NGX lifecycle harnesses covering camera/history/reset, parameter mutation, delayed submission, early release, capability gating and unsupported-input rejection. Add exact gfx11 D/E model generation, verified caches, versioned local deployment with configuration backups, and isolated portable model tools with pinned Python, NumPy and glslang dependencies. Extend Windows K benchmarking and private native-cache selection. Fix LLVM 22 emitter/helper compatibility and preserve exact-output validation for local output-kernel optimization. Add native FSR RR provider diagnostics, local development package refresh tooling, installer/checker updates, FidelityFX headers and Windows parity documentation. Keep NVIDIA binaries, derived models and private optimized kernels out of distributable packages. Native RR evidence remains synthetic; game validation, FSR RR dispatch, installer integration and Linux performance parity remain incomplete.
94 lines
5.2 KiB
PowerShell
94 lines
5.2 KiB
PowerShell
[CmdletBinding()]
|
|
param([string]$SourceRoot, [string]$HipRoot = 'C:\Program Files\AMD\ROCm\7.2', [string]$WorkRoot, [string]$LlvmBuildRoot)
|
|
$ErrorActionPreference = 'Stop'
|
|
$repo = [IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
|
|
if (!$SourceRoot) { $SourceRoot = Join-Path $repo 'external/ZLUDA' }
|
|
if (!$WorkRoot) { $WorkRoot = Join-Path $repo 'build/zluda-helpers-windows' }
|
|
$toolchain = Join-Path $repo '.tools/llvm-mingw-20260922-ucrt-x86_64'
|
|
function Find-HipHelperTool([string]$Name) {
|
|
foreach ($folder in @('lib/llvm/bin','bin')) {
|
|
$candidate=Join-Path $HipRoot "$folder/$Name.exe"
|
|
if(Test-Path -LiteralPath $candidate){return $candidate}
|
|
}
|
|
throw "HIP helper tool missing: $Name under $HipRoot"
|
|
}
|
|
$compiler = Find-HipHelperTool 'clang++'
|
|
$dis = Find-HipHelperTool 'llvm-dis'
|
|
$assembler = Find-HipHelperTool 'llvm-as'
|
|
if(!$LlvmBuildRoot){$LlvmBuildRoot=Join-Path $repo 'build/zluda-llvm-windows-gnu'}
|
|
$pinnedAssembler=Join-Path $LlvmBuildRoot 'bin/llvm-as.exe'
|
|
if(!(Test-Path -LiteralPath $pinnedAssembler)){throw 'Build the pinned LLVM llvm-as target before the PTX helpers.'}
|
|
$assembler=$pinnedAssembler
|
|
$ocml = Join-Path $HipRoot 'lib/llvm/amdgcn/bitcode/ocml.bc'
|
|
if(!(Test-Path -LiteralPath $ocml)){$ocml=Join-Path $HipRoot 'amdgcn/bitcode/ocml.bc'}
|
|
foreach ($file in @($compiler,$dis,$assembler,$ocml)) {
|
|
if (!(Test-Path $file)) { throw "HIP helper dependency missing: $file" }
|
|
}
|
|
New-Item -ItemType Directory -Force $WorkRoot | Out-Null
|
|
$configInclude = Join-Path $WorkRoot 'libcxx-config'
|
|
New-Item -ItemType Directory -Force $configInclude | Out-Null
|
|
Copy-Item -LiteralPath (Join-Path $toolchain 'include/c++/v1/__config_site') -Destination $configInclude -Force
|
|
Copy-Item -LiteralPath (Join-Path $SourceRoot 'ext/llvm-project/libcxx/vendor/llvm/default_assertion_handler.in') `
|
|
-Destination (Join-Path $configInclude '__assertion_handler') -Force
|
|
# HIP 7.2's long shuffle overloads assume every non-MSVC compiler has LP64.
|
|
# Windows GNU also has 32-bit long. Correct that public header in a private
|
|
# mirror; the installed SDK remains untouched and the GPU target is unchanged.
|
|
$hipInclude = Join-Path $WorkRoot 'hip-include'
|
|
New-Item -ItemType Directory -Force $hipInclude | Out-Null
|
|
Copy-Item -LiteralPath (Join-Path $HipRoot 'include/hip') -Destination $hipInclude -Recurse -Force
|
|
$warpHeader = Join-Path $hipInclude 'hip/amd_detail/amd_warp_functions.h'
|
|
$warpSource = [IO.File]::ReadAllText($warpHeader)
|
|
$warpSource = $warpSource.Replace('#ifndef _MSC_VER', '#if __SIZEOF_LONG__ == 8')
|
|
[IO.File]::WriteAllText($warpHeader,$warpSource,[Text.UTF8Encoding]::new($false))
|
|
$variants = @(
|
|
@{Suffix=''; Flags=@()},
|
|
@{Suffix='_constrained'; Flags=@('-ffp-model=strict','-ffp-exception-behavior=ignore')},
|
|
@{Suffix='_w64'; Flags=@('-mwavefrontsize64','-DD4R_WAVE64')},
|
|
@{Suffix='_constrained_w64'; Flags=@('-ffp-model=strict','-ffp-exception-behavior=ignore','-mwavefrontsize64','-DD4R_WAVE64')}
|
|
)
|
|
foreach ($variant in $variants) {
|
|
$name = 'zluda_ptx_impl' + $variant.Suffix
|
|
$raw = Join-Path $WorkRoot "$name.raw.bc"
|
|
$ir = Join-Path $WorkRoot "$name.ll"
|
|
$output = Join-Path $SourceRoot "ptx/lib/$name.bc"
|
|
$arguments = @('-x','hip','-std=c++20','--target=x86_64-w64-windows-gnu','-stdlib=libc++',
|
|
'--offload-arch=gfx1030','--offload-device-only','-emit-llvm','-c','-nogpulib','-O3',
|
|
'-mno-wavefrontsize64','-DHIP_ENABLE_WARP_SYNC_BUILTINS',
|
|
'-Xclang','-fdenormal-fp-math=dynamic',
|
|
'-isystem',(Join-Path $SourceRoot 'ext/llvm-project/libcxx/include'),
|
|
'-isystem',$configInclude,
|
|
'-isystem',(Join-Path $toolchain 'include'),
|
|
'-I',$hipInclude,
|
|
'-Xclang','-mlink-bitcode-file','-Xclang',$ocml)
|
|
$arguments += $variant.Flags
|
|
$arguments += @('-o',$raw,(Join-Path $SourceRoot 'ptx/lib/zluda_ptx_impl.cpp'))
|
|
$ErrorActionPreference = 'Continue'
|
|
& $compiler @arguments 2>&1 | Tee-Object (Join-Path $WorkRoot "$name.compile.log")
|
|
$ErrorActionPreference = 'Stop'
|
|
if ($LASTEXITCODE) { throw "PTX helper compile failed: $name ($LASTEXITCODE)" }
|
|
& $dis $raw -o $ir
|
|
if ($LASTEXITCODE) { throw "Disassemble failed: $raw" }
|
|
$clean = [Collections.Generic.List[string]]::new()
|
|
foreach ($line in [IO.File]::ReadLines($ir)) {
|
|
# Directly linked helpers do not need a ThinLTO index, which may retain
|
|
# references to the stripped HIP/LLVM globals.
|
|
if ($line -match '^\^\d+\s*=') { continue }
|
|
if ($line -match '@llvm.used|wchar_size|llvm.module.flags|__hip_cuid') { continue }
|
|
$line = $line -replace 'optnone','' -replace 'define hidden','define linkonce_odr'
|
|
$line = $line -replace '"target-cpu"="gfx1030"','' -replace '"target-features"="[^"]+"',''
|
|
$clean.Add($line)
|
|
}
|
|
[IO.File]::WriteAllLines($ir,$clean,[Text.UTF8Encoding]::new($false))
|
|
if($assembler -eq $pinnedAssembler){
|
|
& python (Join-Path $repo 'tools/windows/llvm22_ir.py') $ir $ir
|
|
if($LASTEXITCODE){throw 'LLVM 22 helper IR compatibility conversion failed.'}
|
|
}
|
|
$previousPath=$env:PATH
|
|
$env:PATH="$(Join-Path $toolchain 'bin');$previousPath"
|
|
try {
|
|
& $assembler $ir -o $output
|
|
if ($LASTEXITCODE) { throw "Assemble failed: $ir" }
|
|
} finally {$env:PATH=$previousPath}
|
|
Write-Host "Built $output ($((Get-Item $output).Length) bytes)"
|
|
}
|