Backport gfx1154 support into ZLUDA's Windows LLVM compiler
Adapt official LLVM 7a0829e41228 to the pinned LLVM 22 parser and ELF structures. Apply and audit the complete source delta before building, export its provenance, and package the source patch. Add 33 offline public PTX compilation checks across all eleven GPU targets without hardware execution. Preserve strict arithmetic and document RX9070XT K/M regressions.
This commit is contained in:
xdfnx-dev committed
1 parent
b6b6d91ed9
commit
3d5f42865f
10 files changed
+307
-2
No files matched your search
@@ -128,4 +128,5 @@ uses a previous-frame fallback.
|
||||
|
||||
OptiScaler source modifications are GPL-3.0 and exported under
|
||||
`source-patches/optiscaler`. ZLUDA source modifications are exported under
|
||||
`source-patches/zluda`. Dependency license texts and build hashes are included.
|
||||
`source-patches/zluda`; the LLVM gfx1154 backport is under
|
||||
`source-patches/zluda-llvm`. Dependency license texts and build hashes are included.
|
||||
@@ -99,6 +99,29 @@ Use `-GpuArchitectures gfx1100,gfx1101,gfx1102,gfx1200,gfx1201` for a subset.
|
||||
Per-target logs and `matrix-build.json` distinguish compilation from execution.
|
||||
The supplied DLL generates manifests locally and is never copied into archives.
|
||||
|
||||
Rebuild LLVM/ZLUDA when updating from an older package: the old LLVM 22 backend
|
||||
does not recognize gfx1154 even though Clang 24 compiles native objects for it.
|
||||
`build-zluda-llvm.ps1` now applies the source backport in `patches/zluda-llvm`
|
||||
automatically; relink with `build-zluda-windows.ps1` afterward. Compiler base
|
||||
and patch hashes are recorded in the ZLUDA build metadata and included in game
|
||||
packages. The expected patched LLVM submodule appears modified in its parent
|
||||
checkout; its complete source delta is exported, not a binary target override.
|
||||
|
||||
For a compilation gate requiring neither a GPU nor NVIDIA DLLs, run:
|
||||
|
||||
```powershell
|
||||
.\scripts\windows\test-zluda-target-compilation.ps1 -ZludaRoot "$PWD\dist\zluda-windows-final"
|
||||
```
|
||||
|
||||
All 33 public PTX compilations pass for the eleven targets: Driver API pattern,
|
||||
FP16 MMA strict reference and explicit WMMA modes, with correct ELF targets
|
||||
and wave32. The gfx12 strict F16 reference intentionally retains its existing
|
||||
double-precision helper; the separate WMMA mode checks code generation only.
|
||||
It does not enable relaxed arithmetic in game packages or validate numerical
|
||||
results on other GPUs. The original gfx1154 compiler abort is preserved locally
|
||||
in `build/llvm-gfx1154/before.log`; the regression results are in
|
||||
`test-results/windows-gpu-coverage/zluda-offline`.
|
||||
|
||||
## Fork adaptation
|
||||
|
||||
Reviewed [realdody's branch](https://github.com/realdody/d4r/tree/windows-rdna3)
|
||||
|
||||
@@ -52,6 +52,22 @@ Results: `test-results/windows-gpu-coverage`; build logs and the compile-only
|
||||
matrix report: `build/windows-gpu-coverage`. No physical multi-GPU, APU or
|
||||
other-target validation is claimed. K performance remains open.
|
||||
|
||||
The final translator gate exposed a separate dependency gap: pinned LLVM 22
|
||||
rejects `gfx1154` and aborts (`0xC0000409`) on a public PTX pattern. A minimal
|
||||
source backport of official LLVM `7a0829e41228` adds the processor's real
|
||||
11.5.4 features, parser and ELF identity. It is exported under
|
||||
`patches/zluda-llvm`, applied idempotently by the LLVM builder and included with
|
||||
its base/hash in package metadata. LLVM and ZLUDA are rebuilt; all 33 offline
|
||||
pattern/F16-reference/WMMA compilations pass across the eleven targets without
|
||||
GPU execution. The existing strict gfx12 F16 reference lowering remains intact.
|
||||
The rebuilt ZLUDA passes 20/20 RX 9070 XT CTest gates. Its twelve further K
|
||||
4K burst/recreation frames are finite and all six async RGB images are exact;
|
||||
three sync RGB images also exactly match the previous compiler's captures.
|
||||
Four further M 4K frames are finite and match all four previous-compiler RGB
|
||||
images exactly. Both require native transformer hits, zero CPU frame copies
|
||||
and frame age 0. Results: `k-new-zluda`, `m-new-zluda` and compiler comparison
|
||||
logs under `test-results/windows-gpu-coverage`. Other GPUs remain unverified.
|
||||
|
||||
**Current milestone: K performance remains open.** The latest unprofiled 4K
|
||||
Silent Hill 2 scene capture records 65.250 FPS. Functional K/M and the
|
||||
reproducible archive do not close this gate. Prioritize measured K kernel cost
|
||||
@@ -1219,7 +1235,10 @@ Pinned additional build dependencies:
|
||||
rust-std and rust-mingw archives and installs only under `.tools`.
|
||||
* ZLUDA LLVM submodule `ff4dc1f7c9e1c64d4d69e40f4ed30c2280a96dfd`,
|
||||
`llvm-config` reports `22.0.0git`. AMDGPU/LLVM/LLD are built with LLVM-MinGW
|
||||
`20260922`; no Linux build host or runtime is used.
|
||||
`20260922`; no Linux build host or runtime is used. The Windows builder adds
|
||||
`patches/zluda-llvm/0001-backport-gfx1154.patch`, adapted from official LLVM
|
||||
`7a0829e41228513299c5108685b0bc127463c6a1`. Rebuild and relink ZLUDA when
|
||||
updating; native Clang 24 target support alone does not cover translated PTX.
|
||||
* HiGHS submodule `364c83a51e44ba6c27def9c8fc1a49b1daf5ad5c`.
|
||||
* The LLVM IR helper producer is stable HIP 7.2/LLVM 21. The final target is
|
||||
selected by the translator from HIP `gfx1201`; generic helper generation
|
||||
|
||||
@@ -0,0 +1,128 @@
|
||||
diff --git a/llvm/include/llvm/BinaryFormat/ELF.h b/llvm/include/llvm/BinaryFormat/ELF.h
|
||||
index e619b186d..dd3e9d92b 100644
|
||||
--- a/llvm/include/llvm/BinaryFormat/ELF.h
|
||||
+++ b/llvm/include/llvm/BinaryFormat/ELF.h
|
||||
@@ -856,7 +856,7 @@ enum : unsigned {
|
||||
EF_AMDGPU_MACH_AMDGCN_GFX11_GENERIC = 0x054,
|
||||
EF_AMDGPU_MACH_AMDGCN_GFX1152 = 0x055,
|
||||
EF_AMDGPU_MACH_AMDGCN_RESERVED_0X56 = 0x056,
|
||||
- EF_AMDGPU_MACH_AMDGCN_RESERVED_0X57 = 0x057,
|
||||
+ EF_AMDGPU_MACH_AMDGCN_GFX1154 = 0x057,
|
||||
EF_AMDGPU_MACH_AMDGCN_GFX1153 = 0x058,
|
||||
EF_AMDGPU_MACH_AMDGCN_GFX12_GENERIC = 0x059,
|
||||
EF_AMDGPU_MACH_AMDGCN_GFX1251 = 0x05a,
|
||||
diff --git a/llvm/include/llvm/TargetParser/TargetParser.h b/llvm/include/llvm/TargetParser/TargetParser.h
|
||||
index aad985926..c0aaa465b 100644
|
||||
--- a/llvm/include/llvm/TargetParser/TargetParser.h
|
||||
+++ b/llvm/include/llvm/TargetParser/TargetParser.h
|
||||
@@ -109,6 +109,7 @@ enum GPUKind : uint32_t {
|
||||
GK_GFX1151,
|
||||
GK_GFX1152,
|
||||
GK_GFX1153,
|
||||
+ GK_GFX1154,
|
||||
|
||||
GK_GFX1200,
|
||||
GK_GFX1201,
|
||||
diff --git a/llvm/lib/Object/ELFObjectFile.cpp b/llvm/lib/Object/ELFObjectFile.cpp
|
||||
index f9fda2346..70ddaae22 100644
|
||||
--- a/llvm/lib/Object/ELFObjectFile.cpp
|
||||
+++ b/llvm/lib/Object/ELFObjectFile.cpp
|
||||
@@ -591,6 +591,8 @@ StringRef ELFObjectFileBase::getAMDGPUCPUName() const {
|
||||
return "gfx1152";
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1153:
|
||||
return "gfx1153";
|
||||
+ case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1154:
|
||||
+ return "gfx1154";
|
||||
|
||||
// AMDGCN GFX12.
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1200:
|
||||
diff --git a/llvm/lib/ObjectYAML/ELFYAML.cpp b/llvm/lib/ObjectYAML/ELFYAML.cpp
|
||||
index d9cce1eb7..7936876f4 100644
|
||||
--- a/llvm/lib/ObjectYAML/ELFYAML.cpp
|
||||
+++ b/llvm/lib/ObjectYAML/ELFYAML.cpp
|
||||
@@ -632,6 +632,7 @@ void ScalarBitSetTraits<ELFYAML::ELF_EF>::bitset(IO &IO,
|
||||
BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1151, EF_AMDGPU_MACH);
|
||||
BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1152, EF_AMDGPU_MACH);
|
||||
BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1153, EF_AMDGPU_MACH);
|
||||
+ BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1154, EF_AMDGPU_MACH);
|
||||
BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1200, EF_AMDGPU_MACH);
|
||||
BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1201, EF_AMDGPU_MACH);
|
||||
BCaseMask(EF_AMDGPU_MACH_AMDGCN_GFX1250, EF_AMDGPU_MACH);
|
||||
diff --git a/llvm/lib/Target/AMDGPU/GCNProcessors.td b/llvm/lib/Target/AMDGPU/GCNProcessors.td
|
||||
index c6fb31f9c..5d21894a3 100644
|
||||
--- a/llvm/lib/Target/AMDGPU/GCNProcessors.td
|
||||
+++ b/llvm/lib/Target/AMDGPU/GCNProcessors.td
|
||||
@@ -304,7 +304,12 @@ def : ProcessorModel<"gfx1153", GFX11SpeedModel,
|
||||
FeatureISAVersion11_5_3.Features
|
||||
>;
|
||||
|
||||
-// [gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153]
|
||||
+// Backported from LLVM 7a0829e41228: actual gfx1154 features and identity.
|
||||
+def : ProcessorModel<"gfx1154", GFX11SpeedModel,
|
||||
+ FeatureISAVersion11_5_Common.Features
|
||||
+>;
|
||||
+
|
||||
+// [gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153, gfx1154]
|
||||
def : ProcessorModel<"gfx11-generic", GFX11SpeedModel,
|
||||
FeatureISAVersion11_Generic.Features
|
||||
>;
|
||||
diff --git a/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp b/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp
|
||||
index 5a0857322..c506fc308 100644
|
||||
--- a/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp
|
||||
+++ b/llvm/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp
|
||||
@@ -114,6 +114,7 @@ StringRef AMDGPUTargetStreamer::getArchNameFromElfMach(unsigned ElfMach) {
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1151: AK = GK_GFX1151; break;
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1152: AK = GK_GFX1152; break;
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1153: AK = GK_GFX1153; break;
|
||||
+ case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1154: AK = GK_GFX1154; break;
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1200: AK = GK_GFX1200; break;
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1201: AK = GK_GFX1201; break;
|
||||
case ELF::EF_AMDGPU_MACH_AMDGCN_GFX1250: AK = GK_GFX1250; break;
|
||||
@@ -201,6 +202,7 @@ unsigned AMDGPUTargetStreamer::getElfMach(StringRef GPU) {
|
||||
case GK_GFX1151: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1151;
|
||||
case GK_GFX1152: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1152;
|
||||
case GK_GFX1153: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1153;
|
||||
+ case GK_GFX1154: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1154;
|
||||
case GK_GFX1200: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1200;
|
||||
case GK_GFX1201: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1201;
|
||||
case GK_GFX1250: return ELF::EF_AMDGPU_MACH_AMDGCN_GFX1250;
|
||||
diff --git a/llvm/lib/TargetParser/TargetParser.cpp b/llvm/lib/TargetParser/TargetParser.cpp
|
||||
index 34b09b14b..39acbb942 100644
|
||||
--- a/llvm/lib/TargetParser/TargetParser.cpp
|
||||
+++ b/llvm/lib/TargetParser/TargetParser.cpp
|
||||
@@ -172,6 +172,7 @@ constexpr GPUInfo AMDGCNGPUs[] = {
|
||||
{{"gfx1151"}, {"gfx1151"}, GK_GFX1151, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
|
||||
{{"gfx1152"}, {"gfx1152"}, GK_GFX1152, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
|
||||
{{"gfx1153"}, {"gfx1153"}, GK_GFX1153, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
|
||||
+ {{"gfx1154"}, {"gfx1154"}, GK_GFX1154, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
|
||||
{{"gfx1200"}, {"gfx1200"}, GK_GFX1200, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
|
||||
{{"gfx1201"}, {"gfx1201"}, GK_GFX1201, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32|FEATURE_WGP},
|
||||
{{"gfx1250"}, {"gfx1250"}, GK_GFX1250, FEATURE_FAST_FMA_F32|FEATURE_FAST_DENORMAL_F32|FEATURE_WAVE32},
|
||||
@@ -328,6 +329,7 @@ AMDGPU::IsaVersion AMDGPU::getIsaVersion(StringRef GPU) {
|
||||
case GK_GFX1151: return {11, 5, 1};
|
||||
case GK_GFX1152: return {11, 5, 2};
|
||||
case GK_GFX1153: return {11, 5, 3};
|
||||
+ case GK_GFX1154: return {11, 5, 4};
|
||||
case GK_GFX1200: return {12, 0, 0};
|
||||
case GK_GFX1201: return {12, 0, 1};
|
||||
case GK_GFX1250: return {12, 5, 0};
|
||||
@@ -475,6 +477,7 @@ static void fillAMDGCNFeatureMap(StringRef GPU, const Triple &T,
|
||||
Features["atomic-fmin-fmax-global-f32"] = true;
|
||||
break;
|
||||
case GK_GFX1153:
|
||||
+ case GK_GFX1154:
|
||||
case GK_GFX1152:
|
||||
case GK_GFX1151:
|
||||
case GK_GFX1150:
|
||||
diff --git a/llvm/tools/llvm-readobj/ELFDumper.cpp b/llvm/tools/llvm-readobj/ELFDumper.cpp
|
||||
index ab9331690..0ade2d412 100644
|
||||
--- a/llvm/tools/llvm-readobj/ELFDumper.cpp
|
||||
+++ b/llvm/tools/llvm-readobj/ELFDumper.cpp
|
||||
@@ -1655,6 +1655,7 @@ const EnumEntry<unsigned> ElfHeaderMipsFlags[] = {
|
||||
ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1151, "gfx1151"), \
|
||||
ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1152, "gfx1152"), \
|
||||
ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1153, "gfx1153"), \
|
||||
+ ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1154, "gfx1154"), \
|
||||
ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1200, "gfx1200"), \
|
||||
ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1201, "gfx1201"), \
|
||||
ENUM_ENT(EF_AMDGPU_MACH_AMDGCN_GFX1250, "gfx1250"), \
|
||||
@@ -0,0 +1,24 @@
|
||||
# LLVM source patches for the Windows compiler
|
||||
|
||||
Base: ZLUDA's LLVM submodule `ff4dc1f7c9e1c64d4d69e40f4ed30c2280a96dfd`
|
||||
(LLVM 22). `build-zluda-llvm.ps1` applies these patches idempotently before
|
||||
building. ZLUDA's parent source patches remain in `patches/zluda`.
|
||||
|
||||
`0001-backport-gfx1154.patch` adapts official LLVM
|
||||
[`7a0829e41228`](https://github.com/llvm/llvm-project/commit/7a0829e41228513299c5108685b0bc127463c6a1)
|
||||
to the older parser/ELF structures. It adds actual gfx1154 processor features,
|
||||
ISA 11.5.4, parser identity and ELF machine 0x57. It does not map gfx1154 onto
|
||||
another CPU, patch binary headers or alter existing processors' features.
|
||||
LLVM's gfx1154 uses `FeatureISAVersion11_5_Common` and the GFX11 speed model.
|
||||
LLVM code remains Apache-2.0 WITH LLVM-exception (see the dependency license).
|
||||
|
||||
The prior offline compiler aborts on a public PTX pattern for gfx1154 with
|
||||
unrecognized-processor diagnostics and exit `0xC0000409`. After rebuilding
|
||||
LLVM, rebuild/relink ZLUDA and its offline emitter; changing native Clang
|
||||
objects alone cannot fix the translated PTX path. Run
|
||||
`scripts/windows/test-zluda-target-compilation.ps1` to compile the public
|
||||
Driver API pattern and FP16 MMA fixtures (strict/reference and WMMA modes)
|
||||
for every target without GPU work. All 33 compilations pass with correct ELF
|
||||
targets and wave32. Strict gfx12 F16 reference retains its prior FP64 helper;
|
||||
WMMA mode here is a compilation check, not a default arithmetic change.
|
||||
This checks code generation and ELF identity, not hardware execution/numerics.
|
||||
@@ -12,6 +12,7 @@ if (!(Test-Path (Join-Path $SourceRoot 'ext/llvm-project/llvm/CMakeLists.txt')))
|
||||
throw 'Run git submodule update --init --depth 1 ext/llvm-project in the ZLUDA checkout.'
|
||||
}
|
||||
New-Item -ItemType Directory -Force $BuildRoot | Out-Null
|
||||
& (Join-Path $PSScriptRoot 'prepare-zluda-llvm-source.ps1') -LlvmSourceRoot (Join-Path $SourceRoot 'ext/llvm-project')
|
||||
$oldPath = $env:PATH
|
||||
try {
|
||||
$env:PATH = "$(Join-Path $toolchain 'bin');$oldPath"
|
||||
@@ -36,5 +37,9 @@ try {
|
||||
if ($LASTEXITCODE) { throw "LLVM build failed ($LASTEXITCODE)" }
|
||||
& (Join-Path $BuildRoot 'bin/llvm-config.exe') --version
|
||||
if ($LASTEXITCODE) { throw 'llvm-config verification failed' }
|
||||
@{sourceBase=(& git -C (Join-Path $SourceRoot 'ext/llvm-project') rev-parse HEAD).Trim(); fullSourceDeltaVerified=$true;
|
||||
sourcePatches=@(Get-ChildItem (Join-Path $repo 'patches/zluda-llvm') -Filter *.patch | ForEach-Object {
|
||||
@{name=$_.Name; sha256=(Get-FileHash -LiteralPath $_.FullName).Hash}})} |
|
||||
ConvertTo-Json -Depth 5 | Set-Content (Join-Path $BuildRoot 'd4r-llvm-build-info.json') -Encoding UTF8
|
||||
Write-Host "Native Windows LLVM: $BuildRoot"
|
||||
} finally { $env:PATH = $oldPath }
|
||||
@@ -84,6 +84,9 @@ try {
|
||||
@{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 $LlvmBuildRoot 'bin/llvm-config.exe') --version);
|
||||
llvmSource=$(if (Test-Path (Join-Path $LlvmBuildRoot 'd4r-llvm-build-info.json')) {
|
||||
Get-Content (Join-Path $LlvmBuildRoot 'd4r-llvm-build-info.json') -Raw | ConvertFrom-Json
|
||||
} else { $null });
|
||||
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
|
||||
Write-Host "Native Windows d4r ZLUDA runtime: $InstallRoot"
|
||||
|
||||
@@ -94,6 +94,7 @@ Stage (Join-Path $HipRoot 'share/doc/amd_comgr/LICENSE.txt') 'licenses/amd_comgr
|
||||
Stage (Join-Path $repo 'external/clr/LICENSE.md') 'licenses/HIP-CLR.txt'
|
||||
foreach ($patch in Get-ChildItem -LiteralPath (Join-Path $repo 'patches/optiscaler') -Filter '*.patch') { Stage $patch.FullName ('source-patches/optiscaler/' + $patch.Name) }
|
||||
foreach ($patch in Get-ChildItem -LiteralPath (Join-Path $repo 'patches/zluda') -Filter '*.patch') { Stage $patch.FullName ('source-patches/zluda/' + $patch.Name) }
|
||||
foreach ($patch in Get-ChildItem -LiteralPath (Join-Path $repo 'patches/zluda-llvm') -Filter '*.patch') { Stage $patch.FullName ('source-patches/zluda-llvm/' + $patch.Name) }
|
||||
$sourceArchive=Join-Path $repo 'build/d4r-windows-rdna4-source.zip'
|
||||
New-Item -ItemType Directory -Force (Split-Path $sourceArchive) | Out-Null
|
||||
& git -C $repo archive --format=zip --output $sourceArchive $sourceCommit
|
||||
|
||||
@@ -0,0 +1,34 @@
|
||||
[CmdletBinding()]
|
||||
param([Parameter(Mandatory=$true)][string]$LlvmSourceRoot)
|
||||
$ErrorActionPreference='Stop'
|
||||
$repo=[IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
|
||||
$expected='ff4dc1f7c9e1c64d4d69e40f4ed30c2280a96dfd'
|
||||
$actual=(& git -C $LlvmSourceRoot rev-parse HEAD).Trim()
|
||||
if ($LASTEXITCODE -or $actual -ne $expected) { throw "Expected audited ZLUDA LLVM source $expected, found $actual; review the patch against this revision." }
|
||||
$patches=@(Get-ChildItem (Join-Path $repo 'patches/zluda-llvm') -Filter *.patch | Sort-Object Name)
|
||||
foreach ($patch in $patches) {
|
||||
$oldPreference=$ErrorActionPreference
|
||||
try {
|
||||
$ErrorActionPreference='Continue'
|
||||
& git -C $LlvmSourceRoot apply --reverse --check $patch.FullName 2>$null
|
||||
$applied=($LASTEXITCODE -eq 0)
|
||||
} finally { $ErrorActionPreference=$oldPreference }
|
||||
if (!$applied) {
|
||||
& git -C $LlvmSourceRoot apply --check $patch.FullName
|
||||
if ($LASTEXITCODE) { throw "LLVM source patch conflicts: $($patch.Name)" }
|
||||
& git -C $LlvmSourceRoot apply $patch.FullName
|
||||
if ($LASTEXITCODE) { throw "LLVM source patch failed: $($patch.Name)" }
|
||||
}
|
||||
Write-Host "LLVM source patch present: $($patch.Name)"
|
||||
}
|
||||
# The pinned backport is the entire intended LLVM delta.
|
||||
if ($patches.Count -ne 1) { throw 'Review combined LLVM source provenance before adding more patches.' }
|
||||
$audit=Join-Path $repo ('build/llvm-source-audit-'+[Guid]::NewGuid().ToString('N')+'.patch')
|
||||
New-Item -ItemType Directory -Force (Split-Path $audit) | Out-Null
|
||||
try {
|
||||
& git -C $LlvmSourceRoot diff --binary --output=$audit
|
||||
if ($LASTEXITCODE) { throw 'Cannot audit LLVM source diff.' }
|
||||
$actualDiff=[IO.File]::ReadAllText($audit).Replace("`r`n","`n").TrimEnd()
|
||||
$expectedDiff=[IO.File]::ReadAllText($patches[0].FullName).Replace("`r`n","`n").TrimEnd()
|
||||
if ($actualDiff -cne $expectedDiff) { throw 'LLVM source contains changes outside the exported audited patch.' }
|
||||
} finally { if (Test-Path -LiteralPath $audit) { Remove-Item -LiteralPath $audit } }
|
||||
@@ -0,0 +1,67 @@
|
||||
[CmdletBinding()]
|
||||
param([string]$ZludaRoot, [string[]]$GpuArchitectures, [string]$OutputDirectory)
|
||||
$ErrorActionPreference='Stop'
|
||||
. (Join-Path $PSScriptRoot 'gpu-target.ps1')
|
||||
$repo=[IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
|
||||
if (!$ZludaRoot) { $ZludaRoot=Join-Path $repo 'dist/zluda-windows-final' }
|
||||
if (!$OutputDirectory) { $OutputDirectory=Join-Path $repo 'test-results/zluda-target-compilation' }
|
||||
if (!$GpuArchitectures) { $GpuArchitectures=@(Get-D4RGpuTargets | ForEach-Object { $_.architecture }) }
|
||||
foreach ($arch in $GpuArchitectures) { [void](Get-D4RGpuTargetInfo $arch) }
|
||||
$emitter=Join-Path $ZludaRoot 'd4r_emit.exe'
|
||||
if (!(Test-Path -LiteralPath $emitter)) { throw "Rebuild ZLUDA and its offline emitter: $emitter" }
|
||||
New-Item -ItemType Directory -Force $OutputDirectory | Out-Null
|
||||
$settings=@{PATH="$ZludaRoot;$env:PATH"; D4R_ZLUDA_WMMA='1'; D4R_ZLUDA_WMMA_FP8='1';
|
||||
D4R_ZLUDA_WMMA_F16_REFERENCE='1'; D4R_ZLUDA_WMMA_FP8_NATIVE='0'; D4R_ZLUDA_WMMA_LAYOUT=$null;
|
||||
D4R_ZLUDA_WAVE64=$null; D4R_ZLUDA_FAST_MATH=$null; D4R_ZLUDA_IGNORE_DENORMAL=$null;
|
||||
D4R_ZLUDA_WMMA_F32ACC=$null; D4R_ZLUDA_EXTRA_BC=$null; D4R_ZLUDA_WGP=$null}
|
||||
$old=@{}; $rows=[Collections.Generic.List[object]]::new()
|
||||
try {
|
||||
foreach ($key in $settings.Keys) {
|
||||
$old[$key]=[Environment]::GetEnvironmentVariable($key,'Process')
|
||||
if ($null -eq $settings[$key]) { Remove-Item -LiteralPath "Env:$key" -ErrorAction SilentlyContinue }
|
||||
else { [Environment]::SetEnvironmentVariable($key,$settings[$key],'Process') }
|
||||
}
|
||||
foreach ($fixture in @(@{name='pattern'; source='cuda_driver_probe.cpp'; reference='1'},
|
||||
@{name='f16-mma-reference'; source='cuda_mma_probe.cpp'; reference='1'},
|
||||
@{name='f16-mma-wmma'; source='cuda_mma_probe.cpp'; reference='0'})) {
|
||||
$env:D4R_ZLUDA_WMMA_F16_REFERENCE=$fixture.reference
|
||||
# Reuse the public synthetic PTX exercised by the hardware diagnostics.
|
||||
# Reading source does not execute a probe or load HIP/NVIDIA libraries.
|
||||
$match=[regex]::Match((Get-Content (Join-Path $repo ('tools/windows/'+$fixture.source)) -Raw),'(?s)R"PTX\((.*?)\)PTX"')
|
||||
if (!$match.Success) { throw 'Public PTX fixture not found.' }
|
||||
$ptx=Join-Path $OutputDirectory ($fixture.name+'.ptx')
|
||||
$match.Groups[1].Value | Set-Content -LiteralPath $ptx -Encoding ASCII
|
||||
foreach ($arch in $GpuArchitectures) {
|
||||
$output=Join-Path $OutputDirectory ($arch+'/'+$fixture.name)
|
||||
New-Item -ItemType Directory -Force $output | Out-Null
|
||||
$log=Join-Path $output 'compile.log'
|
||||
$oldPreference=$ErrorActionPreference
|
||||
try { $ErrorActionPreference='Continue'; & $emitter $ptx $output $arch *> $log; $code=$LASTEXITCODE }
|
||||
finally { $ErrorActionPreference=$oldPreference }
|
||||
if ($code) { throw "Offline ZLUDA compilation failed: $arch $($fixture.name), exit=$code; see $log" }
|
||||
Assert-D4RCodeObjectTarget (Join-Path $output 'module.hsaco') $arch
|
||||
$text=Get-Content -LiteralPath $log -Raw
|
||||
if ($text -match 'not a recognized processor|ignoring processor|unsupported processor') { throw "Target rejected by compiler: $log" }
|
||||
$asm=Get-Content (Join-Path $output 'asm.s') -Raw
|
||||
if ($asm -notmatch '\.amdhsa_wavefront_size32\s+1') { throw "Missing wave32 code generation: $output" }
|
||||
$hasWmma=($asm -match 'v_wmma_')
|
||||
if (($fixture.name -eq 'f16-mma-wmma' -or ($fixture.name -eq 'f16-mma-reference' -and $arch.StartsWith('gfx11'))) -and !$hasWmma) {
|
||||
throw "WMMA lowering missing: $output"
|
||||
}
|
||||
# gfx12 strict F16 accumulation intentionally uses the existing
|
||||
# double-precision reference helper. Do not label it native WMMA.
|
||||
if ($fixture.name -eq 'f16-mma-reference' -and $arch.StartsWith('gfx12') -and
|
||||
($hasWmma -or $asm -notmatch 'v_fma_f64')) { throw "Strict gfx12 reference lowering changed: $output" }
|
||||
$rows.Add(@{architecture=$arch; fixture=$fixture.name; compiled=$true; elfTargetCorrect=$true;
|
||||
nativeWmma=$hasWmma; hardwareExecuted=$false})
|
||||
Write-Host "PASS ZLUDA_OFFLINE target=$arch fixture=$($fixture.name) wave=32 hardware_executed=0"
|
||||
}
|
||||
}
|
||||
@{passed=$true; fixtures=@($rows); emitterSha256=(Get-FileHash -LiteralPath $emitter).Hash; gpuWorkExecuted=$false} |
|
||||
ConvertTo-Json -Depth 6 | Set-Content (Join-Path $OutputDirectory 'summary.json') -Encoding UTF8
|
||||
} finally {
|
||||
foreach ($key in $old.Keys) {
|
||||
if ($null -eq $old[$key]) { Remove-Item -LiteralPath "Env:$key" -ErrorAction SilentlyContinue }
|
||||
else { [Environment]::SetEnvironmentVariable($key,$old[$key],'Process') }
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user