libtorch build failure

[stalek71]

Operating system

Windows

Compiler

MSVC

Steps to reproduce the behavior

I created a new project in Visual Studio 2022 for C++ based on CMake.
I added configuration for vcpkg (integration) and set the triplet to x64-windows.
In the vcpkg.json file, I specified a dependency for libtorch.
During compilation, an error appears as shown in the attached log file (it cannot find NVCC).
I have CUDA SDK (12.5) installed, and the nvcc file is available in the path.

Failure logs

vcpkg-manifest-install.log

Additional context

I have nvcc available in the path of course (info below).

I use the latest version of VS 2022 and vcpkg.

[stalek71]

I switched env vars to CUDA 12.4 (as I I overlooked it), but it still doesn't compile LibTorch from vcpkg defs.
I decided to manually download LibTorch from Pytorch website and it worked like a charm.
I had to add only this operation below to force linker to link to cuda libs (for WIN32) and it reported
that cuda libs are in place and available for calc operations.

set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /INCLUDE:?ignore_this_library_placeholder@@YAHXZ")

This is contents of vcpkgs file (so libtorch is without cuda feature):

{
  "dependencies": [
    "fmt",
    "nlohmann-json",
    "spdlog",
    "gtest",
    "libtorch"
  ]
}