Building project using libtorch results in "Failed to find nvToolsExt"

[willjallen]

🐛 Describe the bug

• Installed CUDA 12.1 from https://developer.nvidia.com/cuda-12-1-0-download-archive with all options enabled
• Follow instructions from https://pytorch.org/cppdocs/installing.html using libtorch version 2.1.2+cu121

Environment Variables:
CUDA_PATH = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1
CudaToolkitDir = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1

CMakeLists.txt:

cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
project(example-app)

find_package(Torch REQUIRED)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS}")

add_executable(example-app example-app.cpp)
target_link_libraries(example-app "${TORCH_LIBRARIES}")
set_property(TARGET example-app PROPERTY CXX_STANDARD 17)

# The following code block is suggested to be used on Windows.
# According to https://github.com/pytorch/pytorch/issues/25457,
# the DLLs need to be copied to avoid memory errors.
if (MSVC)
  file(GLOB TORCH_DLLS "${TORCH_INSTALL_PREFIX}/lib/*.dll")
  add_custom_command(TARGET example-app
                     POST_BUILD
                     COMMAND ${CMAKE_COMMAND} -E copy_if_different
                     ${TORCH_DLLS}
                     $<TARGET_FILE_DIR:example-app>)
endif (MSVC)

Execute commands

mkdir build
cd build
cmake -DCMAKE_PREFIX_PATH=libtorch/ ..

Get the output:

-- Building for: Visual Studio 17 2022
-- Selecting Windows SDK version 10.0.19041.0 to target Windows 10.0.19045.
-- The C compiler identification is MSVC 19.31.31104.0
-- The CXX compiler identification is MSVC 19.31.31104.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files (x86)/Microsoft Visual Studio/2022/BuildTools/VC/Tools/MSVC/14.31.31103/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files (x86)/Microsoft Visual Studio/2022/BuildTools/VC/Tools/MSVC/14.31.31103/bin/Hostx64/x64/cl.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.1 (found version "12.1")
-- The CUDA compiler identification is NVIDIA 12.1.66
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - done
-- Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.1/bin/nvcc.exe - skipped
-- Detecting CUDA compile features
-- Detecting CUDA compile features - done
-- Found CUDAToolkit: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.1/include (found version "12.1.66")
CMake Error at libtorch/share/cmake/Caffe2/public/cuda.cmake:71 (message):
  Failed to find nvToolsExt
Call Stack (most recent call first):
  libtorch/share/cmake/Caffe2/Caffe2Config.cmake:87 (include)
  libtorch/share/cmake/Torch/TorchConfig.cmake:68 (find_package)
  CMakeLists.txt:4 (find_package)


-- Configuring incomplete, errors occurred!

Versions

Collecting environment information...
PyTorch version: N/A
Is debug build: N/A
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: N/A

OS: Microsoft Windows 10 Home
GCC version: (GCC) 11.3.0
Clang version: Could not collect
CMake version: version 3.27.0
Libc version: N/A

Python version: 3.12.0 (tags/v3.12.0:0fb18b0, Oct 2 2023, 13:03:39) [MSC v.1935 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-10-10.0.19045-SP0
Is CUDA available: N/A
CUDA runtime version: 12.1.66
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: GPU 0: NVIDIA GeForce GTX 1080 Ti
Nvidia driver version: 546.12
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: N/A

CPU:
Architecture=9
CurrentClockSpeed=3696
DeviceID=CPU0
Family=198
L2CacheSize=1536
L2CacheSpeed=
Manufacturer=GenuineIntel
MaxClockSpeed=3696
Name=Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz
ProcessorType=3
Revision=

Versions of relevant libraries:
[pip3] numpy==1.26.1
[conda] Could not collect

cc @ptrblck

[janeyx99]

Do the answers in https://discuss.pytorch.org/t/failed-to-find-nvtoolsext/179635 help?

Generally, these type of build questions would get faster and more responses through our forum (linked above)!

[willjallen]

Unfortunately no. The Nsight NVTX option is missing from the CUDA installer and the nvToolsExt64_1.lib is nowhere to be found. Thanks for the suggestion, i'll try asking on the forums.

Edit: Upon closer inspection I see the 11.8 CUDA installer is specifically required, I was using 12.1. The Nsight NVTX option is present and after installing it the build works as expected.

It seems strange that I need to use two different CUDA installers to satisfy the build requirements, nonetheless this resolves the issue.

[BradXiao]

Identical problem encountered.
Solved by installing CUDA 11.8 (with Nsight NVTX) and CUDA 12.1 (missing Nsight NVTX option).
Using pytorch 2.1.2 (CUDA 12.1).

[janeyx99]

cc’ing @ptrblck

[patrikhuber]

See my comments here #116926 (comment) - NVTX is still present in CUDA 12, it's just header-only now.