Fall back to the CUDA_PATH toolkit when the preferred CUDA version is not installed (#6360)

* Fall back to the CUDA_PATH toolkit when the preferred CUDA version is not installed

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* Document CUDA version fallback in the Windows setup guide

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: Fedr

doxygen/general_pages/CppSetupGuide.dox

@@ -41,10 +41,12 @@ xcopy C:\meshlib-built\install\app\$(Configuration)\*.dll $(TargetDir)
    - Include support for the C++ programming language.
    - Ensure the English language pack is installed (required for `vcpkg`).
  3. **CUDA Toolkit**
-  - For **Visual Studio 2019**: Install [CUDA v11.4](https://developer.nvidia.com/cuda-11-4-0-download-archive)
-  - For **Visual Studio 2022**: Install [CUDA v12.0](https://developer.nvidia.com/cuda-12-0-0-download-archive)
-  - For **Visual Studio 2026**: Install [CUDA v13.2](https://developer.nvidia.com/cuda-13-2-0-download-archive)
- \n Choose the appropriate version based on your Visual Studio installation.
+  - Recommended CUDA versions (the ones MeshLib CI builds with):
+   - For **Visual Studio 2019**: Install [CUDA v11.4](https://developer.nvidia.com/cuda-11-4-0-download-archive)
+   - For **Visual Studio 2022**: Install [CUDA v12.0](https://developer.nvidia.com/cuda-12-0-0-download-archive)
+   - For **Visual Studio 2026**: Install [CUDA v13.2](https://developer.nvidia.com/cuda-13-2-0-download-archive)
+  - Other CUDA versions work as well: if the recommended one is not installed, the build automatically falls back to the toolkit pointed to by the `CUDA_PATH` environment variable, which the NVIDIA installer keeps at the latest installed version.
+  - In the CUDA installer, keep the **Visual Studio Integration** component enabled: the Visual Studio build needs it.
  4. **vcpkg**
   - To install `vcpkg`, follow these steps:
    1. **Open a command prompt (CMD) or PowerShell window**.

source/platform.props

@@ -22,6 +22,13 @@
     <PreferredToolArchitecture>x86</PreferredToolArchitecture> <!-- Using 32-bit build tools reduces memory consumption and fixes error C3859: Failed to create virtual memory for PCH -->
   </PropertyGroup>
   <Import Project="$(CustomPlatfotmPropertyDir)\CustomMRPlatform.props" Condition="'$(CustomPlatfotmPropertyDir)' != ''"/>
+  <!-- If the CUDA version selected above has no MSBuild integration in this Visual Studio, fall back
+       to the toolkit pointed to by the CUDA_PATH environment variable (the NVIDIA installer keeps it
+       at the latest installed version), when that version's integration is present. -->
+  <PropertyGroup Condition="('$(MRCudaVersion)' == '' OR !Exists('$(VCTargetsPath)\BuildCustomizations\CUDA $(MRCudaVersion).props')) AND '$(CUDA_PATH)' != ''">
+    <_MRCudaVersionFromCudaPath>$([System.IO.Path]::GetFileName('$(CUDA_PATH)').TrimStart('v'))</_MRCudaVersionFromCudaPath>
+    <MRCudaVersion Condition="Exists('$(VCTargetsPath)\BuildCustomizations\CUDA $(_MRCudaVersionFromCudaPath).props')">$(_MRCudaVersionFromCudaPath)</MRCudaVersion>
+  </PropertyGroup>
   <!-- Remaining GPU architectures to match the CMake build (cmake/Modules/ConfigureCuda.cmake): the CUDA
        props' unconditional default CodeGeneration (compute_52,sm_52; compute_75,sm_75 on CUDA 13+) already
        supplies the oldest SASS+PTX; these add the newer SASS targets plus PTX for the newest one. -->