How to add an NVCC flag

I am attempting to compile a program using CUDA, and am getting the following error in response

C:\Program Files\MATLAB\R2024a\sys\cuda\win64\cuda\include\crt/host_config.h(164): fatal error C1189:

#error: -- unsupported Microsoft Visual Studio version! Only the versions between 2017 and 2022

(inclusive) are supported! The nvcc flag '-allow-unsupported-compiler' can be used to override this

version check; however, using an unsupported host compiler may cause compilation failure or incorrect run

time execution. Use at your own risk.

I looked and it seems the problem is caused by a small glitch in a recent update of visual studio that shouldnt have any effect on the compiler. How do I add the flag to override the version check? My current code is as follows

mex -setup:'C:\Program Files\MATLAB\R2024a\bin\win64\mexopts\msvcpp2022.xml' C++

mexGPUall

[Matt J]

I looked and it seems the problem is caused by a small glitch in a recent update of visual studio that shouldnt have any effect on the compiler.

I am having the same difficulty. How did you verify this?

Answer 1

[Xiao Haonan]

Hi Aiden,

I encountered the same problem, and solved it by setting environmental variable "NVCC_APPEND_FLAGS".

To do this, you can try:

setenv("NVCC_APPEND_FLAGS", '-allow-unsupported-compiler')
mexcuda -v mexGPUExample.cu

Hope it helps!

[Aidan Fishenden]

This worked perfectly, thank you!

[Matt J]

Didn't work for me, unfortunately.

Answer 2

[Akash]

Hi Aidan,

To address the issue, you can ensure that the compiler and its version are supported with the product. You can find the list of supported and compatible compilers at the following link:-

https://www.mathworks.com/support/requirements/supported-compilers.html

The tables on this page outline the compilers that are supported by various MathWorks products.

For configuring parameters for CUDA code generation, you can use "coder.gpuConfig". This allows setting additional flags for the GPU compiler, including the NVCC flag. You can find more information on how to set these properties in the below provided MATLAB documentation link:-

https://www.mathworks.com/help/gpucoder/ref/coder.gpucodeconfig.html?searchHighlight=nvcc%20flag&s_tid=srchtitle_support_results_2_nvcc%20flag#:~:text=default)%20%7C%20character%20vector-,CompilerFlags,-%E2%80%94%20Additional%20flags

[Aidan Fishenden]

Would it be possible for you to show example code? I've gone through the page you reccomended but am still getting the same error despite the changes. It still does not recognize the flag.

mex -setup:'C:\Program Files\MATLAB\R2024a\bin\win64\mexopts\msvcpp2022.xml' C++ 
cfg = coder.gpuConfig( "mex" );
cfg.TargetLang = "C++";
cfg.GpuConfig.CompilerFlags = "--allow-unsupported-compiler";
cfg.GpuConfig.SelectCudaDevice = 0;
mexGPUall

Answer 3

[Matt J]

Downgrade to Visual Studio 2019. See also,

https://www.mathworks.com/matlabcentral/answers/2158705-mexcuda-can-t-find-visual-studio-2022-but-mex-can?s_tid=srchtitle

Answer 4

[.json]

I had a similar problem using gpucoder on linux (ubuntu 24.04 is currently on gcc 13.2, matlab 2025a ships with cuda 12.2 which only supports gcc 12.2). The error message was:

#error -- unsupported GNU version! gcc versions later than 12 are not supported! The nvcc flag '-allow-unsupported-compiler' can be used to override this version check; however, using an unsupported host compiler may cause compilation failure or incorrect run time execution. Use at your own risk.

The "--allow-unsupported-compiler" flag didn't work for me (maybe the gcc version delta was too wide). I installed gcc-12 using anaconda:

conda create --name gcc-12 gcc=12.2 gxx=12.2
conda activate gcc-12

Then pointed matlab to the correct compiler with:

cfg = coder.gpuConfig('mex');
cfg.GpuConfig.CompilerFlags = '--compiler-bindir ${HOME}/miniforge3/envs/gcc-12/bin';

where ${HOME} is my home folder. codegen worked fine from there.