How to make working CUDA 11.6 with CMake 3.23 and MSVC 2019

Question

I cannot find a solution to manage how to use the langage CUDA in a CMake project on Windows with the standard MSVC 2019 compiler.

I am trying to configure and compile this hello-cmake-cuda repository (also described in this blog post).

CMakeLists.txt file contents:

cmake_minimum_required(VERSION 3.8 FATAL_ERROR)
project(hello LANGUAGES CXX CUDA)
enable_language(CUDA)
add_executable(hello hello.cu)

Here is the output to the cmake .. command, run from within the build directory:

PS C:\GitRepo\cuda_hello\build> cmake ..
-- Selecting Windows SDK version 10.0.18362.0 to target Windows 10.0.22000.
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/CMakeDetermineCUDACompiler.cmake:311 (message):
  CMAKE_CUDA_ARCHITECTURES must be valid if set.
Call Stack (most recent call first):
  CMakeLists.txt:5 (project)


-- Configuring incomplete, errors occurred!
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeOutput.log".
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeError.log".

It means that architectures_tested from CMakeDetermineCUDACompiler.cmake:311 is empty...

How can I get CMake to complete its configuration and the simple program to build?

My development environment

• Operating system: Windows 11 Version 10.0.22000 Build 22000
• Compiler: Microsoft Visual Studio Community 2019 Version 16.11.11
• CMake version is 3.23
• CUDA version is 11.6

I have tried different versions of each soft and keep having the same issue. I have decided to stay with these versions at the moment.

My GPU is properly configured: It shows up with nvidia-smi, and I am also able to build and run the deviceQuery CUDA sample:

CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "NVIDIA GeForce GTX 1650"
  CUDA Driver Version / Runtime Version          11.6 / 11.6
  CUDA Capability Major/Minor version number:    7.5
  etc. etc. ...

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 11.6, CUDA Runtime Version = 11.6, NumDevs = 1
Result = PASS

---

My environment PATH variable:

PS C:\GitRepo\hello-cuda-cmake-master> $env:path -split ";"
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\libnvvp
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\libnvvp

C:\Program Files (x86)\Common Files\Oracle\Java\javapath
C:\Python38\Scripts\
C:\Python38\
C:\Windows\system32
C:\Windows
C:\Windows\System32\Wbem
C:\Windows\System32\WindowsPowerShell\v1.0\
C:\Windows\System32\OpenSSH\
C:\Program Files (x86)\NVIDIA Corporation\PhysX\Common
C:\Program Files\NVIDIA Corporation\NVIDIA NvDLISR
C:\Program Files\PuTTY\
C:\Program Files (x86)\PuTTY\
C:\Program Files\Microsoft SQL Server\110\Tools\Binn\
C:\Program Files\TortoiseSVN\bin
C:\Program Files\TortoiseGit\bin
C:\Program Files\Microsoft VS Code\bin
C:\WINDOWS\system32
C:\WINDOWS
C:\WINDOWS\System32\Wbem
C:\WINDOWS\System32\WindowsPowerShell\v1.0\
C:\WINDOWS\System32\OpenSSH\
C:\Program Files\Docker\Docker\resources\bin
C:\ProgramData\DockerDesktop\version-bin
C:\Program Files\Git\cmd
C:\WINDOWS\system32
C:\WINDOWS
C:\WINDOWS\System32\Wbem
C:\WINDOWS\System32\WindowsPowerShell\v1.0\
C:\WINDOWS\System32\OpenSSH\
C:\Program Files\NVIDIA Corporation\Nsight Compute 2022.1.1\
C:\Program Files\CMake\bin
C:\Ruby30-x64\bin
C:\Users\Thibault GEFFROY\.cargo\bin
C:\Users\Thibault GEFFROY\AppData\Local\Microsoft\WindowsApps
C:\Program Files\OpenCppCoverage
C:\intelFPGA\20.1\modelsim_ase\win32aloem

What I've tried and hasn't worked

If I try to insert the wanted CMAKE_CUDA_ARCHITECTURES :

set(CMAKE_CUDA_ARCHITECTURES 75)

I get:

PS C:\GitRepo\cuda_hello\build> cmake ..
-- Selecting Windows SDK version 10.0.18362.0 to target Windows 10.0.22000.
-- The CUDA compiler identification is unknown
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/CMakeDetermineCUDACompiler.cmake:654 (message):
  The CMAKE_CUDA_ARCHITECTURES:

    75

  do not all work with this compiler.  Try:



  instead.
Call Stack (most recent call first):
  CMakeLists.txt:5 (project)


-- Configuring incomplete, errors occurred!
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeOutput.log".
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeError.log".

If I try to use the FindCUDA module to set CMAKE_CUDA_ARCHITECTURES - the solution given by @alfC here - I get:

PS C:\GitRepo\cuda_hello\build> cmake ..
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/FindCUDA/select_compute_arch.cmake:120 (file):
  file failed to open for writing (Permission denied):

    /detect_cuda_compute_capabilities.cpp
Call Stack (most recent call first):
  CMakeLists.txt:4 (CUDA_DETECT_INSTALLED_GPUS)


CMake Error: The source directory "CMAKE_FLAGS" does not exist.
Specify --help for usage, or press the help button on the CMake GUI.
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/FindCUDA/select_compute_arch.cmake:141 (try_run):
  Failed to configure test project build system.
Call Stack (most recent call first):
  CMakeLists.txt:4 (CUDA_DETECT_INSTALLED_GPUS)


CMake Error: TRY_COMPILE attempt to remove -rf directory that does not contain CMakeTmp:/detect_cuda_compute_capabilities.cpp
-- Configuring incomplete, errors occurred!
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeOutput.log".
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeError.log".

finally, if I try to invoke find_package(CUDA), I get:

PS C:\GitRepo\cuda_hello\build> cmake ..
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/FindCUDA.cmake:677 (cmake_initialize_per_config_variable):
  Unknown CMake command "cmake_initialize_per_config_variable".
Call Stack (most recent call first):
  CMakeLists.txt:2 (find_package)


-- Configuring incomplete, errors occurred!
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeOutput.log".
See also "C:/GitRepo/cuda_hello/build/CMakeFiles/CMakeError.log".

Edit 1:

Answer to @einpoklum solution this:

Thanks for the proposal but it doesn't work either.

Here is the output of the cmake -B build command in your repository:

PS C:\GitRepo\hello-cuda-cmake-master> cmake -B build
-- Building for: Visual Studio 16 2019
-- Selecting Windows SDK version 10.0.18362.0 to target Windows 10.0.22000.
-- The CUDA compiler identification is unknown
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/CMakeDetermineCUDACompiler.cmake:633 (message):
  Failed to detect a default CUDA architecture.



  Compiler output:

Call Stack (most recent call first):
  CMakeLists.txt:2 (project)


-- Configuring incomplete, errors occurred!
See also "C:/GitRepo/hello-cuda-cmake-master/build/CMakeFiles/CMakeOutput.log".
See also "C:/GitRepo/hello-cuda-cmake-master/build/CMakeFiles/CMakeError.log".

The output is the same using PowerShell or a MSVC command prompt.

---

Here are the cmake variables and their value when using cmake-gui:

---

When using the simple nvcc build command: nvcc hello.cu from MSVC command prompt I get:

nvcc fatal   : Could not set up the environment for Microsoft Visual Studio using 'c:/Program Files (x86)/Microsoft Visual Studio/2019/Community/VC/Tools/MSVC/14.29.30133/bin/HostX86/x86/../../../../../../../VC/Auxiliary/Build/vcvars64.bat'

The PATH is valid though, and the script vcvars64.bat exists at this location.

---

What happens if I add the find_package(CUDAToolkit) to the CMakeLists.txt

The new CMakeLists.txt:

cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
find_package(CUDAToolkit)
project(hello LANGUAGES CUDA)
add_executable(hello hello.cu)

The output :

PS C:\GitRepo\hello-cuda-cmake-master> cmake -B build
-- Building for: Visual Studio 16 2019
-- Found CUDAToolkit: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.6/include (found version "11.6.124")
-- Selecting Windows SDK version 10.0.18362.0 to target Windows 10.0.22000.
-- The CUDA compiler identification is unknown
CMake Error at C:/Program Files/CMake/share/cmake-3.23/Modules/CMakeDetermineCUDACompiler.cmake:633 (message):
  Failed to detect a default CUDA architecture.



  Compiler output:

Call Stack (most recent call first):
  CMakeLists.txt:3 (project)


-- Configuring incomplete, errors occurred!
See also "C:/GitRepo/hello-cuda-cmake-master/build/CMakeFiles/CMakeOutput.log".
See also "C:/GitRepo/hello-cuda-cmake-master/build/CMakeFiles/CMakeError.log".

Edit 2:

I am trying to compile the CUDA sample BlackScholes without CMake, with the MSVC 2019 solution provided.

I end up with this error:

Severity        Code        Description        Project        File        Line        Suppression State
Error        MSB3721        The command ""C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin\nvcc.exe" -gencode=arch=compute_35,code=\"sm_35,compute_35\" -gencode=arch=compute_37,code=\"sm_37,compute_37\" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=compute_60,code=\"sm_60,compute_60\" -gencode=arch=compute_61,code=\"sm_61,compute_61\" -gencode=arch=compute_70,code=\"sm_70,compute_70\" -gencode=arch=compute_75,code=\"sm_75,compute_75\" -gencode=arch=compute_80,code=\"sm_80,compute_80\" -gencode=arch=compute_86,code=\"sm_86,compute_86\" --use-local-env -ccbin "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\bin\HostX86\x64" -x cu   -I./ -I../../../Common -I./ -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\/include" -I../../../Common -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\include"  -G   --keep-dir x64\Debug  -maxrregcount=0  --machine 64 --compile -cudart static -Xcompiler "/wd 4819"  --threads 0 -g  -DWIN32 -DWIN32 -D_MBCS -D_MBCS -Xcompiler "/EHsc /W3 /nologo /Od /Fdx64/Debug/vc142.pdb /FS /Zi /RTC1 /MTd " -o "C:\ProgramData\NVIDIA Corporation\CUDA Samples\v11.6\cuda-samples\Samples\5_Domain_Specific\BlackScholes\x64\Debug\BlackScholes.cu.obj" "C:\ProgramData\NVIDIA Corporation\CUDA Samples\v11.6\cuda-samples\Samples\5_Domain_Specific\BlackScholes\BlackScholes.cu"" exited with code 1.        BlackScholes        C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\MSBuild\Microsoft\VC\v160\BuildCustomizations\CUDA 11.6.targets        790

While using WSL 2 Ubuntu 20.4 and the following CUDA installation and these instructions to build the BlackScholes sample I get this output:

$ sudo make BlackScholes
/usr/local/cuda/bin/nvcc -ccbin g++ -I../../../Common  -m64    -maxrregcount=16 --threads 0 --std=c++11 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_86,code=compute_86 -o BlackScholes.o -c BlackScholes.cu
nvcc warning : The 'compute_35', 'compute_37', 'sm_35', and 'sm_37' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
ptxas warning : For profile sm_86 adjusting per thread register count of 16 to lower bound of 24
ptxas warning : For profile sm_80 adjusting per thread register count of 16 to lower bound of 24
ptxas warning : For profile sm_70 adjusting per thread register count of 16 to lower bound of 24
ptxas warning : For profile sm_75 adjusting per thread register count of 16 to lower bound of 24
/usr/local/cuda/bin/nvcc -ccbin g++ -I../../../Common  -m64    -maxrregcount=16 --threads 0 --std=c++11 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_86,code=compute_86 -o BlackScholes_gold.o -c BlackScholes_gold.cpp
nvcc warning : The 'compute_35', 'compute_37', 'sm_35', and 'sm_37' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
/usr/local/cuda/bin/nvcc -ccbin g++   -m64      -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_86,code=compute_86 -o BlackScholes BlackScholes.o BlackScholes_gold.o
nvcc warning : The 'compute_35', 'compute_37', 'sm_35', and 'sm_37' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).
mkdir -p ../../../bin/x86_64/linux/release
cp BlackScholes ../../../bin/x86_64/linux/release


$ ./BlackScholes
[./BlackScholes] - Starting...
GPU Device 0: "Turing" with compute capability 7.5

Initializing data...
...allocating CPU memory for options.
...allocating GPU memory for options.
...generating input data in CPU mem.
...copying input data to GPU mem.
Data init done.

Executing Black-Scholes GPU kernel (512 iterations)...
Options count             : 8000000
BlackScholesGPU() time    : 0.722482 msec
Effective memory bandwidth: 110.729334 GB/s
Gigaoptions per second    : 11.072933

BlackScholes, Throughput = 11.0729 GOptions/s, Time = 0.00072 s, Size = 8000000 options, NumDevsUsed = 1, Workgroup = 128

Reading back GPU results...
Checking the results...
...running CPU calculations.

Comparing the results...
L1 norm: 1.741792E-07
Max absolute error: 1.192093E-05

Shutting down...
...releasing GPU memory.
...releasing CPU memory.
Shutdown done.

[BlackScholes] - Test Summary

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

Test passed

Answer 1

I had the same issue and the main issue was, that on CMake 3.23.2 it was just not working.

My steps to solve this problem were:

1. Remove all CUDA versions that were installed on the machine
2. Remove all CUDA-related environmental variables
3. Install CUDA v12.2
4. Install newest CMAKE 3.27.4 (GUI version)
5. Copy CUDA Visual Studio Integration files into Visual Studio 2022 link

Answer 2

I had the same problem and I solved it by installing older version of CMake. More precisely: a version before 3.18.

Apparently CMake added first party language support for CUDA in 3.18 and that is where these nonsensical problems ("Try: indead") were coming from.

Answer 3

os env :

1. window 10 (visual studio 2022 Community)
2. cuda: cuda 11.6 , nvcc
3. cpp standard: 17
4. ide: vscode
5. cmake.

this project name is :hellogpu

cmake file:

cmake_minimum_required(VERSION 3.0.0)
project(hellogpu CUDA)

include(CTest)
enable_testing()

add_executable(${PROJECT_NAME} main.cu)

set_target_properties(${PROJECT_NAME} PROPERTIES
        CUDA_SEPARABLE_COMPILATION ON)
set(CPACK_PROJECT_NAME ${PROJECT_NAME})
set(CPACK_PROJECT_VERSION ${PROJECT_VERSION})
include(CPack)

Answer 4

Try to add:

set(CMAKE_CUDA_ARCHITECTURES 60 61 62 70 72 75 86)
set(CMAKE_CUDA_COMPILER /usr/local/cuda-11.6/bin/nvcc)

check your CUDA arch in https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/ and change the parameter of CMAKE_CUDA_ARCHITECTURES.

And link the CMAKE_CUDA_COMPILER to nvcc.

this is my full CMakeLists.txt:

cmake_minimum_required(VERSION 3.20 FATAL_ERROR)

set(CMAKE_CUDA_ARCHITECTURES 60 61 62 70 72 75 86)
set(CMAKE_CUDA_COMPILER /usr/local/cuda-11.6/bin/nvcc)

project(cudatest CUDA)
find_package(CUDAToolkit)

set(CMAKE_CUDA_STANDARD 14)

add_executable(cudatest main.cu)

set_target_properties(cudatest PROPERTIES
    CUDA_SEPARABLE_COMPILATION ON)

My GPU is GeForce GTX 1660, CMake version 3.23, CUDA Version 11.6.

And this is a Docker image I made for developmenting some projects: https://github.com/GuangchenJ/cuda-dev, you can try to use it.

Answer 5

Beginning with CMake 3.18, we no longer use the FindCUDA.cmake module - neither directly nor via find_package(CUDA). This has been replaced with find_package(CUDAToolkit) (which used the FindCUDAToolkit.cmake module).

But actually, for your simple hello-world project - you don't even need to do that, since starting with CMake 3.8, CUDA is a "first-class citizen" language for CMake. Well, kind of. So, here's a CMakeLists.txt file you can use:

cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
PROJECT(cuda_hello LANGUAGES CUDA)
add_executable(hello hello.cu)

I've tested this on a Windows 10 (Enterprise Evaluation) VM, using CUDA 11.6 and Visual Studio 16 (a.k.a. VS 2019).

Note: The version number in the cmake_minimum_required() line may be critical! With the version number at the cuda_hello repository - it doesn't work for me, since a CMAKE_CUDA_ARCHITECTURES value is demanded to be present.

Now, after you configure using CMake, you can run ccmake, where you'll see the CMAKE_CUDA_ARCHITECTURES value. Change it to what you want to use. Again, I'm offering you the simplest and most basic way to do things, not necessarily the fanciest and most robust.

---

I've set all of this up for you in a fork of the hello-cuda-cmake repository.