GPU Coder: compilation error

Question

Daigo 2022-3-20

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1676104-gpu-coder-compilation-error

回答： Kumar Pallav 2022-3-23

I am trying to create a MEX function from the following simple function:

function G = ft2(g, delta)
    % Input
    % - g:      gpuArray, double, complex, size: N x N
    % - delta:  double, size: 1 x 1
    G = fftshift(fft2(fftshift(g))) * delta^2;
end

Although the MEX generation was successful when I specify the input g as a double array (non-gpuArray), it was not successful when I specify it as a double gpuArray. The build was successful but the compilation seems to have failed. The buildLog is like below:

----

C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.29.30133\include\vcruntime.h(197): error #256: invalid redeclaration of type name "size_t"

C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.29.30133\include\vcruntime_new.h(48): error #351: first parameter of allocation function must be of type "size_t"

C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.29.30133\include\type_traits(407): error #247: class template "std::_Is_memfunptr" has already been defined

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\include\crt/common_functions.h(117): error #351: first parameter of allocation function must be of type "size_t"

....

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\include\sm_32_intrinsics.hpp(132): error #3041: asm operand type size(8) does not match type/size implied by constraint 'r'

Error limit reached.

100 errors detected in the compilation of "C:/Users/Daigo/AppData/Local/Temp/tmpxft_00002b68_00000000-10_ft2_initialize.cpp1.ii".

Compilation terminated.

ft2_initialize.cu

ninja: build stopped: subcommand failed.

----

There are more than 100 error messages. It seems like they have something to do with C complier and CUDA toolkit. I cannot imagine how much effort I will need to fix all the issues in the log. Does anyone have advice on the use of GPU Coder?