From 47da1da414276055102f25177a4dc42a020300f6 Mon Sep 17 00:00:00 2001
From: cm4ker <cm4ker@gmail.com>
Date: Sat, 30 Aug 2025 22:56:28 +0600
Subject: [PATCH 1/3] Windows Arm Support

---
 Directory.Build.props                         | 12 ++---
 Directory.Build.targets                       | 25 ++++++++--
 pkg/TorchAudio/TorchAudio.nupkgproj           |  2 +-
 pkg/TorchVision/TorchVision.nupkgproj         |  2 +-
 .../libtorch-cpu-win-arm64.nupkgproj          | 16 ++++++
 pkg/libtorch-cpu/libtorch-cpu.nupkgproj       |  1 +
 pkg/pack.proj                                 |  6 ++-
 src/Examples.Utils/Examples.Utils.csproj      |  2 +-
 src/Native/build.cmd                          | 25 ++++------
 src/Native/gen-buildsys-win.bat               |  1 +
 src/Redist/libtorch-cpu/libtorch-cpu.proj     | 15 +++++-
 ...arm64-shared-with-deps-2.7.1%2Bcpu.zip.sha |  1 +
 ...shared-with-deps-debug-2.7.1%2Bcpu.zip.sha |  1 +
 src/TorchSharp/Torch.cs                       | 50 +++++++++++++++++--
 14 files changed, 123 insertions(+), 36 deletions(-)
 create mode 100644 pkg/libtorch-cpu-win-arm64/libtorch-cpu-win-arm64.nupkgproj
 create mode 100644 src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-2.7.1%2Bcpu.zip.sha
 create mode 100644 src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-debug-2.7.1%2Bcpu.zip.sha

diff --git a/Directory.Build.props b/Directory.Build.props
index ff94123a3..8ba341083 100644
--- a/Directory.Build.props
+++ b/Directory.Build.props
@@ -52,14 +52,13 @@
     <TargetOS Condition="'$(TargetOS)' == '' AND $([MSBuild]::IsOSPlatform('osx'))">mac</TargetOS>
     <TargetOS Condition="'$(TargetOS)' == '' AND '$(OS)' != 'Windows_NT'">linux</TargetOS>
 
+
     <TargetPlatform Condition="'$(TargetPlatform)' == ''">$(TargetOS)-$(TargetArchitecture)</TargetPlatform>
 
-    <TargetRuntimeID Condition="'$(TargetOS)' == 'windows'">win-x64</TargetRuntimeID>
+    <TargetRuntimeID Condition="'$(TargetOS)' == 'windows' and '$(TargetArchitecture)' == 'arm64'">win-arm64</TargetRuntimeID>
+    <TargetRuntimeID Condition="'$(TargetOS)' == 'windows' and '$(TargetArchitecture)' != 'arm64'">win-x64</TargetRuntimeID>
     <TargetRuntimeID Condition="'$(TargetOS)' == 'linux'">linux-x64</TargetRuntimeID>
     <TargetRuntimeID Condition="'$(TargetPlatform)' == 'mac-arm64'">osx-arm64</TargetRuntimeID>
-
-    <TargetRuntimeID Condition="'$(TargetOS)' == 'windows'">win-x64</TargetRuntimeID>
-    <TargetRuntimeID Condition="'$(TargetOS)' == 'linux'">linux-x64</TargetRuntimeID>
     <TargetRuntimeID Condition="'$(TargetOS)' == 'mac'">osx-$(TargetArchitecture)</TargetRuntimeID>
 
     <!-- on windows separate debug binaries of LibTorch are available-->
@@ -95,8 +94,8 @@
 
   <PropertyGroup>
     <!-- turned on/off manually in separate CI jobs -->
-    <SkipCuda Condition="'$(TargetOS)' == 'mac'">true</SkipCuda>
-    <SkipCuda Condition="'$(TargetOS)' != 'mac'">false</SkipCuda>
+    <SkipCuda Condition="'$(TargetOS)' == 'mac' or ('$(TargetOS)' == 'windows' and '$(TargetArchitecture)' == 'arm64')">true</SkipCuda>
+    <SkipCuda Condition="'$(SkipCuda)' == ''">false</SkipCuda>
     <SkipTests>false</SkipTests>
 
     <!-- By default only TorchSharp and no libtorch-cpu or libtorch-cuda packages are built.  The CI file controls these via 'BuildLibTorchPackages' -->
@@ -141,6 +140,7 @@
     <LibTorchCpuLocalNameSuffix>cpu</LibTorchCpuLocalNameSuffix>
     <LibTorchCudaLocalNameSuffix>cu$(CudaVersionNoDot)</LibTorchCudaLocalNameSuffix>
     <LibTorchArchiveCoreName Condition="'$(TargetOS)' == 'windows'">libtorch-win-shared-with-deps$(LibTorchDebug)</LibTorchArchiveCoreName>
+    <LibTorchArchiveCoreName Condition="'$(TargetOS)' == 'windows' and '$(TargetArchitecture)' == 'arm64'">libtorch-win-arm64-shared-with-deps$(LibTorchDebug)</LibTorchArchiveCoreName>
     <LibTorchArchiveCoreName Condition="'$(TargetOS)' == 'linux'">libtorch-cxx11-abi-shared-with-deps</LibTorchArchiveCoreName>
     <LibTorchArchiveCoreName Condition="'$(TargetOS)' == 'mac'">libtorch-macos-x86_64</LibTorchArchiveCoreName>
     <LibTorchArchiveCoreName Condition="'$(TargetPlatform)' == 'mac-arm64'">libtorch-macos-arm64</LibTorchArchiveCoreName>
diff --git a/Directory.Build.targets b/Directory.Build.targets
index 70c79f166..dc72e290c 100644
--- a/Directory.Build.targets
+++ b/Directory.Build.targets
@@ -1,7 +1,12 @@
 <Project>
-  
+  <!-- Ensure NativeOutputPath is populated for consumer projects (examples/tests) on Windows ARM64 by staging redist first -->
+  <Target Name="EnsureNativeRedistStaged" BeforeTargets="PrepareForRun" Condition="'$(OS)' == 'Windows_NT' and '$(NativeTargetArchitecture)' == 'arm64' and '$(SkipNative)' != 'true'">
+    <Message Importance="High" Text="Ensuring Windows ARM64 native redist is staged to $(NativeOutputPath)" />
+    <MSBuild Projects="$(RepoRoot)src\Redist\libtorch-cpu\libtorch-cpu.proj" Targets="Build" Properties="Configuration=$(NativeConfiguration);TargetOS=windows;TargetArchitecture=arm64" />
+  </Target>
+
   <!-- We copy over the LibTorch native binaries in order to run tests.  This can probably be automated better. -->
-  <!-- These lists are duplicated in redist/... with file splitting added for package prep --> 
+  <!-- These lists are duplicated in redist/... with file splitting added for package prep -->
 
   <ItemGroup  Condition="'$(TestUsesLibTorch)' == 'true' and '$(SkipNative)' != 'true' ">
     <NativeAssemblyReference Include="LibTorchSharp" />
@@ -20,6 +25,20 @@
     <NativeAssemblyReference Include="uv" />
   </ItemGroup>
 
+  <!-- Windows ARM64 CPU libtorch binary list used for examples and testing -->
+  <ItemGroup Condition="'$(NativeTargetArchitecture)' == 'arm64' and '$(OS)' == 'Windows_NT' and '$(TestUsesLibTorch)' == 'true' and ('$(TestCuda)' != 'true' or '$(SkipCuda)' == 'true') and '$(SkipNative)' != 'true' ">
+    <NativeAssemblyReference Include="aoti_custom_ops" />
+    <NativeAssemblyReference Include="backend_with_compiler" />
+    <NativeAssemblyReference Include="c10" />
+    <NativeAssemblyReference Include="torchbind_test" />
+    <NativeAssemblyReference Include="torch" />
+    <NativeAssemblyReference Include="torch_cpu" />
+    <NativeAssemblyReference Include="torch_global_deps" />
+    <NativeAssemblyReference Include="uv" />
+    <NativeAssemblyReference Include="jitbackend_test" />
+    <NativeAssemblyReference Include="armpl_lp64" />
+  </ItemGroup>
+
   <!-- Windows CUDA 12.1 libtorch binary list used for examples and testing -->
   <ItemGroup Condition="'$(NativeTargetArchitecture)' == 'x64' and '$(OS)' == 'Windows_NT' and '$(TestUsesLibTorch)' == 'true' and ('$(TestCuda)' == 'true' and '$(SkipCuda)' != 'true') and '$(SkipNative)' != 'true' ">
     <NativeAssemblyReference Include="asmjit" Variant="cuda\" />
@@ -116,5 +135,5 @@
     </Copy>
 
   </Target>
-  
+
 </Project>
\ No newline at end of file
diff --git a/pkg/TorchAudio/TorchAudio.nupkgproj b/pkg/TorchAudio/TorchAudio.nupkgproj
index 7ac14ee82..1ca16a71b 100644
--- a/pkg/TorchAudio/TorchAudio.nupkgproj
+++ b/pkg/TorchAudio/TorchAudio.nupkgproj
@@ -8,7 +8,7 @@
   <ItemGroup>
     <ProjectReference Include="..\TorchSharp\TorchSharp.nupkgproj" />
   </ItemGroup>
-  
+
   <ItemGroup>
     <Content Include="..\common\NormalPackage.props" Pack="true" PackagePath="buildTransitive\net6.0\$(MSBuildProjectName).props" />
     <Content Include="..\common\NormalPackage.targets" Pack="true" PackagePath="buildTransitive\net6.0\$(MSBuildProjectName).targets" />
diff --git a/pkg/TorchVision/TorchVision.nupkgproj b/pkg/TorchVision/TorchVision.nupkgproj
index f389e9b84..813fd1959 100644
--- a/pkg/TorchVision/TorchVision.nupkgproj
+++ b/pkg/TorchVision/TorchVision.nupkgproj
@@ -8,7 +8,7 @@
   <ItemGroup>
     <ProjectReference Include="..\TorchSharp\TorchSharp.nupkgproj" />
   </ItemGroup>
-  
+
   <ItemGroup>
     <Content Include="..\common\NormalPackage.props" Pack="true" PackagePath="buildTransitive\net6.0\$(MSBuildProjectName).props" />
     <Content Include="..\common\NormalPackage.targets" Pack="true" PackagePath="buildTransitive\net6.0\$(MSBuildProjectName).targets" />
diff --git a/pkg/libtorch-cpu-win-arm64/libtorch-cpu-win-arm64.nupkgproj b/pkg/libtorch-cpu-win-arm64/libtorch-cpu-win-arm64.nupkgproj
new file mode 100644
index 000000000..766c20d0d
--- /dev/null
+++ b/pkg/libtorch-cpu-win-arm64/libtorch-cpu-win-arm64.nupkgproj
@@ -0,0 +1,16 @@
+<Project Sdk="Microsoft.NET.Sdk" DefaultTargets="Pack">
+
+  <PropertyGroup>
+    <TargetFramework>netstandard2.0</TargetFramework>
+  </PropertyGroup>
+
+  <ItemGroup Condition="'$(TargetFramework)' == 'netstandard2.0'">
+  </ItemGroup>
+
+  <ItemGroup>
+    <Content Include="..\common\CpuHeadPackage.props" Pack="true" PackagePath="buildTransitive\netstandard2.0\$(MSBuildProjectName).props" />
+    <Content Include="..\common\CpuHeadPackage.targets" Pack="true" PackagePath="buildTransitive\netstandard2.0\$(MSBuildProjectName).targets" />
+    <Content Include="..\empty.txt" Pack="true" PackagePath="lib\netstandard2.0\_._" />
+    <Content Include="$(RepoRoot)\THIRD-PARTY-NOTICES.txt" Pack="true" PackagePath="LICENSE-LIBTORCH.txt" />
+  </ItemGroup>
+</Project>
\ No newline at end of file
diff --git a/pkg/libtorch-cpu/libtorch-cpu.nupkgproj b/pkg/libtorch-cpu/libtorch-cpu.nupkgproj
index 97c3ffe67..97296c6c8 100644
--- a/pkg/libtorch-cpu/libtorch-cpu.nupkgproj
+++ b/pkg/libtorch-cpu/libtorch-cpu.nupkgproj
@@ -8,6 +8,7 @@
      <ProjectReference Include="../libtorch-cpu-linux-x64/libtorch-cpu-linux-x64.nupkgproj" />
      <ProjectReference Include="../libtorch-cpu-osx-arm64/libtorch-cpu-osx-arm64.nupkgproj" />
      <ProjectReference Include="../libtorch-cpu-win-x64/libtorch-cpu-win-x64.nupkgproj" />
+     <ProjectReference Include="../libtorch-cpu-win-arm64/libtorch-cpu-win-arm64.nupkgproj" />
   </ItemGroup>
 
   <ItemGroup Condition="'$(TargetFramework)' == 'netstandard2.0'">
diff --git a/pkg/pack.proj b/pkg/pack.proj
index 3c9db2f98..d5f0f86f3 100644
--- a/pkg/pack.proj
+++ b/pkg/pack.proj
@@ -22,8 +22,10 @@
   <Target Name="Pack" DependsOnTargets="Restore">
     <Warning Text="Packages will be incomplete and unusable on linux platforms. To get a complete package you need the LibTorchSharp.so binaries for other platforms and copy them into '$(PackagePreparationPath)' to make complete packages. This is automated by Azure Pipelines."
              Condition="'$(IncludeTorchSharpPackage)' == 'true' AND !Exists('$(PackagePreparationPath)\TorchSharp\runtimes\linux-x64\native\libLibTorchSharp.so')" />
-    <Warning Text="Packages will be incomplete and unusable on win-x64 platform. To get a complete package you need the LibTorchSharp.dll binaries for other platforms and copy them into '$(PackagePreparationPath)' to make complete packages. This is automated by Azure Pipelines."
-             Condition="'$(IncludeTorchSharpPackage)' == 'true' AND !Exists('$(PackagePreparationPath)\TorchSharp\runtimes\win-x64\native\LibTorchSharp.dll')" />
+  <Warning Text="Packages will be incomplete and unusable on win-x64 platform. To get a complete package you need the LibTorchSharp.dll binaries for other platforms and copy them into '$(PackagePreparationPath)' to make complete packages. This is automated by Azure Pipelines."
+       Condition="'$(IncludeTorchSharpPackage)' == 'true' AND !Exists('$(PackagePreparationPath)\TorchSharp\runtimes\win-x64\native\LibTorchSharp.dll')" />
+  <Warning Text="Packages will be incomplete and unusable on win-arm64 platform. To get a complete package you need the LibTorchSharp.dll binaries for other platforms and copy them into '$(PackagePreparationPath)' to make complete packages. This is automated by Azure Pipelines."
+       Condition="'$(IncludeTorchSharpPackage)' == 'true' AND !Exists('$(PackagePreparationPath)\TorchSharp\runtimes\win-arm64\native\LibTorchSharp.dll')" />
 
     <MSBuild Projects="%(PackProject.Identity)" Targets="Pack" />
 
diff --git a/src/Examples.Utils/Examples.Utils.csproj b/src/Examples.Utils/Examples.Utils.csproj
index a542b181d..620d5b487 100644
--- a/src/Examples.Utils/Examples.Utils.csproj
+++ b/src/Examples.Utils/Examples.Utils.csproj
@@ -17,7 +17,7 @@
 
   <ItemGroup>
     <PackageReference Include="SharpZipLib" Version="1.4.0" />
-    <PackageReference Include="SixLabors.ImageSharp" Version="3.1.7" />
+    <PackageReference Include="SixLabors.ImageSharp" Version="3.1.11" />
   </ItemGroup>
 
   <ItemGroup>
diff --git a/src/Native/build.cmd b/src/Native/build.cmd
index c805b2608..ee5d19838 100644
--- a/src/Native/build.cmd
+++ b/src/Native/build.cmd
@@ -23,6 +23,7 @@ if /i [%1] == [Debug]       ( set CMAKE_BUILD_TYPE=Debug&&shift&goto Arg_Loop)
 if /i [%1] == [x86]         ( set __BuildArch=x86&&set __VCBuildArch=x86&&shift&goto Arg_Loop)
 if /i [%1] == [x64]         ( set __BuildArch=x64&&set __VCBuildArch=x86_amd64&&shift&goto Arg_Loop)
 if /i [%1] == [amd64]       ( set __BuildArch=x64&&set __VCBuildArch=x86_amd64&&shift&goto Arg_Loop)
+if /i [%1] == [arm64]       ( set __BuildArch=arm64&&set __VCBuildArch=x86_arm64&&shift&goto Arg_Loop)
 
 if /i [%1] == [--libtorchpath] ( set LIBTORCH_PATH=%2&&shift&goto Arg_Loop)
 
@@ -64,40 +65,32 @@ exit /b 1
 :: Setup vars for VS2022
 set __PlatformToolset=v143
 set __VSVersion=17 2022
-if NOT "%__BuildArch%" == "arm64" (
-    :: Set the environment for the native build
-    call "%VS160COMNTOOLS%..\..\VC\Auxiliary\Build\vcvarsall.bat" %__VCBuildArch%
-)
+:: Set the environment for the native build
+call "%VS160COMNTOOLS%..\..\VC\Auxiliary\Build\vcvarsall.bat" %__VCBuildArch%
 goto :SetupDirs
 
 :VS2019
 :: Setup vars for VS2019
 set __PlatformToolset=v142
 set __VSVersion=16 2019
-if NOT "%__BuildArch%" == "arm64" (
-    :: Set the environment for the native build
-    call "%VS160COMNTOOLS%..\..\VC\Auxiliary\Build\vcvarsall.bat" %__VCBuildArch%
-)
+:: Set the environment for the native build
+call "%VS160COMNTOOLS%..\..\VC\Auxiliary\Build\vcvarsall.bat" %__VCBuildArch%
 goto :SetupDirs
 
 :VS2017
 :: Setup vars for VS2017
 set __PlatformToolset=v141
 set __VSVersion=15 2017
-if NOT "%__BuildArch%" == "arm64" (
-    :: Set the environment for the native build
-    call "%VS150COMNTOOLS%..\..\VC\Auxiliary\Build\vcvarsall.bat" %__VCBuildArch%
-)
+:: Set the environment for the native build
+call "%VS150COMNTOOLS%..\..\VC\Auxiliary\Build\vcvarsall.bat" %__VCBuildArch%
 goto :SetupDirs
 
 :VS2015
 :: Setup vars for VS2015build
 set __PlatformToolset=v140
 set __VSVersion=14 2015
-if NOT "%__BuildArch%" == "arm64" (
-    :: Set the environment for the native build
-    call "%VS140COMNTOOLS%..\..\VC\vcvarsall.bat" %__VCBuildArch%
-)
+:: Set the environment for the native build
+call "%VS140COMNTOOLS%..\..\VC\vcvarsall.bat" %__VCBuildArch%
 
 :SetupDirs
 :: Setup to cmake the native components
diff --git a/src/Native/gen-buildsys-win.bat b/src/Native/gen-buildsys-win.bat
index b3870c171..e8ec2fd39 100644
--- a/src/Native/gen-buildsys-win.bat
+++ b/src/Native/gen-buildsys-win.bat
@@ -30,6 +30,7 @@ popd
 :: Set the target architecture to a format cmake understands.
 if /i "%3" == "x64"     (set __ExtraCmakeParams=%__ExtraCmakeParams% -A x64)
 if /i "%3" == "x86"     (set __ExtraCmakeParams=%__ExtraCmakeParams% -A Win32)
+if /i "%3" == "arm64"   (set __ExtraCmakeParams=%__ExtraCmakeParams% -A ARM64)
 
 echo "%CMakePath%" "-DCMAKE_BUILD_TYPE=%CMAKE_BUILD_TYPE%" "-DCMAKE_INSTALL_PREFIX=%__CMakeBinDir%" "-DLIBTORCH_PATH=%LIBTORCH_PATH%" -G "Visual Studio %__VSString%" %__ExtraCmakeParams% -B. -H%1
 "%CMakePath%" "-DCMAKE_BUILD_TYPE=%CMAKE_BUILD_TYPE%" "-DCMAKE_INSTALL_PREFIX=%__CMakeBinDir%" "-DLIBTORCH_PATH=%LIBTORCH_PATH%" -G "Visual Studio %__VSString%" %__ExtraCmakeParams% -B. -H%1
diff --git a/src/Redist/libtorch-cpu/libtorch-cpu.proj b/src/Redist/libtorch-cpu/libtorch-cpu.proj
index 9c538dbe8..16705c006 100644
--- a/src/Redist/libtorch-cpu/libtorch-cpu.proj
+++ b/src/Redist/libtorch-cpu/libtorch-cpu.proj
@@ -30,7 +30,7 @@
     <CopyToPackageSemaphore>$(MainPackageFolder)\.copied.SkipTests.$(SkipTests).IncludeLibTorchCpuPackages.$(IncludeLibTorchCpuPackages)</CopyToPackageSemaphore>
   </PropertyGroup>
 
-  <ItemGroup Condition="'$(TargetOS)' == 'windows'">
+  <ItemGroup Condition="'$(TargetOS)' == 'windows' and '$(NativeTargetArchitecture)' != 'arm64'">
     <File Include="libtorch\lib\asmjit.dll" />
     <File Include="libtorch\lib\c10.dll" />
     <File Include="libtorch\lib\fbgemm.dll" />
@@ -41,6 +41,19 @@
     <File Include="libtorch\lib\torch_global_deps.dll" />
     <File Include="libtorch\lib\uv.dll" />
   </ItemGroup>
+  <!-- Windows ARM64 has a smaller set of DLLs and different deps names -->
+  <ItemGroup Condition="'$(TargetOS)' == 'windows' and '$(NativeTargetArchitecture)' == 'arm64'">
+    <File Include="lib\aoti_custom_ops.dll"/>
+    <File Include="lib\backend_with_compiler.dll"/>
+    <File Include="lib\c10.dll"/>
+    <File Include="lib\jitbackend_test.dll"/>
+    <File Include="lib\torchbind_test.dll"/>
+    <File Include="lib\torch.dll"/>
+    <File Include="lib\torch_cpu.dll"/>
+    <File Include="lib\torch_global_deps.dll"/>
+    <File Include="lib\uv.dll"/>
+    <File Include="lib\armpl_lp64.dll"/>
+  </ItemGroup>
   <ItemGroup Condition="'$(NativeTargetArchitecture)' == 'arm64' and '$(TargetOS)' == 'mac'">
     <File Include="libtorch\lib\libc10.dylib" />
     <File Include="libtorch\lib\libshm.dylib" />
diff --git a/src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-2.7.1%2Bcpu.zip.sha b/src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-2.7.1%2Bcpu.zip.sha
new file mode 100644
index 000000000..2a96235c2
--- /dev/null
+++ b/src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-2.7.1%2Bcpu.zip.sha
@@ -0,0 +1 @@
+F9DD47C792C900601F08265DDC0186036E01503ADA7895F0C7C90F8AF64FBC4B
diff --git a/src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-debug-2.7.1%2Bcpu.zip.sha b/src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-debug-2.7.1%2Bcpu.zip.sha
new file mode 100644
index 000000000..f35825d9a
--- /dev/null
+++ b/src/Redist/libtorch-cpu/libtorch-win-arm64-shared-with-deps-debug-2.7.1%2Bcpu.zip.sha
@@ -0,0 +1 @@
+DFDD2BECA32B1B3D4894EFA801DC60570849F0ED47A318734BA798A5C4D6513B
diff --git a/src/TorchSharp/Torch.cs b/src/TorchSharp/Torch.cs
index c59196bfe..de68b2a91 100644
--- a/src/TorchSharp/Torch.cs
+++ b/src/TorchSharp/Torch.cs
@@ -36,13 +36,13 @@ public static partial class torch
             RuntimeInformation.OSArchitecture == Architecture.Arm64;
 
         static string nativeRid =>
-            RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? $"win-x64" :
+            RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? ($"win-{(RuntimeInformation.OSArchitecture == Architecture.Arm64 ? "arm64" : "x64")}") :
             RuntimeInformation.IsOSPlatform(OSPlatform.Linux) ? $"linux-x64" :
             isAppleSilicon ? "osx-arm64" :
             "any";
 
         static string nativeGlob =>
-            RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? @".*\.dll" :
+            RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? @".*\.dll(\.dll)?" :
             RuntimeInformation.IsOSPlatform(OSPlatform.OSX) ? @".*\.dylib\.*" :
             // must match
             //   lib.so
@@ -127,7 +127,7 @@ private static void LoadNativeBackend(bool useCudaBackend, out StringBuilder? tr
 
                 if (useCudaBackend) {
                     var isWindows = RuntimeInformation.IsOSPlatform(OSPlatform.Windows);
-                    if (isWindows) {
+                    if (isWindows && RuntimeInformation.OSArchitecture != Architecture.Arm64) {
                         trace.AppendLine($"    Try loading Windows cuda native components");
                         // Preloading these DLLs on windows seems to iron out problems where one native DLL
                         // requests a load of another through dynamic linking techniques.
@@ -151,9 +151,49 @@ private static void LoadNativeBackend(bool useCudaBackend, out StringBuilder? tr
 
                     ok = TryLoadNativeLibraryByName("torch_cuda", typeof(torch).Assembly, trace);
                     ok = TryLoadNativeLibraryByName("LibTorchSharp", typeof(torch).Assembly, trace);
+
+                    // On Windows, also attempt direct absolute-path loads from the assembly directory
+                    if (!ok && isWindows) {
+                        var asmDir = Path.GetDirectoryName(typeof(torch).Assembly.Location)!;
+                        var torchCudaPath = Path.Combine(asmDir, "torch_cuda.dll");
+                        var libTorchSharpPath = Path.Combine(asmDir, "LibTorchSharp.dll");
+                        trace.AppendLine($"    Attempting absolute-path load of native components from {asmDir}...");
+                        ok = TryLoadNativeLibraryFromFile(torchCudaPath, trace);
+                        if (ok) ok = TryLoadNativeLibraryFromFile(libTorchSharpPath, trace);
+                    }
                 } else {
-                    ok = TryLoadNativeLibraryByName("torch_cpu", typeof(torch).Assembly, trace);
-                    ok = TryLoadNativeLibraryByName("LibTorchSharp", typeof(torch).Assembly, trace);
+                    var isWindows = RuntimeInformation.IsOSPlatform(OSPlatform.Windows);
+                    if (isWindows) {
+                        // Preload dependency chain explicitly on Windows: torch_global_deps -> c10 -> uv -> torch -> torch_cpu -> LibTorchSharp
+                        trace.AppendLine($"    Try loading Windows CPU native dependency chain: torch_global_deps -> c10 -> uv -> torch -> torch_cpu -> LibTorchSharp");
+                        ok = TryLoadNativeLibraryByName("torch_global_deps", typeof(torch).Assembly, trace);
+                        if (ok) ok = TryLoadNativeLibraryByName("c10", typeof(torch).Assembly, trace);
+                        if (ok) ok = TryLoadNativeLibraryByName("uv", typeof(torch).Assembly, trace);
+                        if (ok) ok = TryLoadNativeLibraryByName("torch", typeof(torch).Assembly, trace);
+                        if (ok) ok = TryLoadNativeLibraryByName("torch_cpu", typeof(torch).Assembly, trace);
+                        if (ok) ok = TryLoadNativeLibraryByName("LibTorchSharp", typeof(torch).Assembly, trace);
+                    } else {
+                        ok = TryLoadNativeLibraryByName("torch_cpu", typeof(torch).Assembly, trace);
+                        if (ok) ok = TryLoadNativeLibraryByName("LibTorchSharp", typeof(torch).Assembly, trace);
+                    }
+
+                    // On Windows, also attempt direct absolute-path loads from the assembly directory
+                    if (!ok && isWindows) {
+                        var asmDir = Path.GetDirectoryName(typeof(torch).Assembly.Location)!;
+                        var torchGlobalDepsPath = Path.Combine(asmDir, "torch_global_deps.dll");
+                        var c10Path = Path.Combine(asmDir, "c10.dll");
+                        var uvPath = Path.Combine(asmDir, "uv.dll");
+                        var torchPath = Path.Combine(asmDir, "torch.dll");
+                        var torchCpuPath = Path.Combine(asmDir, "torch_cpu.dll");
+                        var libTorchSharpPath = Path.Combine(asmDir, "LibTorchSharp.dll");
+                        trace.AppendLine($"    Attempting absolute-path load of native components from {asmDir}...");
+                        ok = TryLoadNativeLibraryFromFile(torchGlobalDepsPath, trace);
+                        if (ok) ok = TryLoadNativeLibraryFromFile(c10Path, trace);
+                        if (ok) ok = TryLoadNativeLibraryFromFile(uvPath, trace);
+                        if (ok) ok = TryLoadNativeLibraryFromFile(torchPath, trace);
+                        if (ok) ok = TryLoadNativeLibraryFromFile(torchCpuPath, trace);
+                        if (ok) ok = TryLoadNativeLibraryFromFile(libTorchSharpPath, trace);
+                    }
                 }
 
                 trace.AppendLine($"    Result from regular native load of LibTorchSharp is {ok}");

From de6d26f5824ed460b1cb9457123b6dbab97981ba Mon Sep 17 00:00:00 2001
From: cm4ker <cm4ker@gmail.com>
Date: Sun, 31 Aug 2025 02:47:26 +0600
Subject: [PATCH 2/3] Update src/TorchSharp/Torch.cs

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---
 src/TorchSharp/Torch.cs | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/TorchSharp/Torch.cs b/src/TorchSharp/Torch.cs
index de68b2a91..61b43a763 100644
--- a/src/TorchSharp/Torch.cs
+++ b/src/TorchSharp/Torch.cs
@@ -42,7 +42,7 @@ public static partial class torch
             "any";
 
         static string nativeGlob =>
-            RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? @".*\.dll(\.dll)?" :
+            RuntimeInformation.IsOSPlatform(OSPlatform.Windows) ? @".*\.dll$" :
             RuntimeInformation.IsOSPlatform(OSPlatform.OSX) ? @".*\.dylib\.*" :
             // must match
             //   lib.so

From 68b3eb7d1b78155e7a07a062c222e4399395b658 Mon Sep 17 00:00:00 2001
From: cm4ker <cm4ker@gmail.com>
Date: Sun, 31 Aug 2025 02:48:03 +0600
Subject: [PATCH 3/3] Update src/TorchSharp/Torch.cs

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---
 src/TorchSharp/Torch.cs | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/TorchSharp/Torch.cs b/src/TorchSharp/Torch.cs
index 61b43a763..cfb4770d8 100644
--- a/src/TorchSharp/Torch.cs
+++ b/src/TorchSharp/Torch.cs
@@ -127,7 +127,8 @@ private static void LoadNativeBackend(bool useCudaBackend, out StringBuilder? tr
 
                 if (useCudaBackend) {
                     var isWindows = RuntimeInformation.IsOSPlatform(OSPlatform.Windows);
-                    if (isWindows && RuntimeInformation.OSArchitecture != Architecture.Arm64) {
+                    // CUDA is not supported on Windows ARM64. Only attempt to load CUDA components on Windows x64.
+                    if (isWindows && RuntimeInformation.OSArchitecture == Architecture.X64) {
                         trace.AppendLine($"    Try loading Windows cuda native components");
                         // Preloading these DLLs on windows seems to iron out problems where one native DLL
                         // requests a load of another through dynamic linking techniques.