I’m using Visual Studio 2019 and I’m trying to build a DLL that includes PyTorch c++ CPU only. I’d like to do this because the DLL will later be used in two different applications one in c++ and the other in C#.
When I’m compiling I mainly get this error: Macro clash with min and max -- define NOMINMAX when compiling your program on Windows
out of +1700 errors.
What I’m doing is the following:
I created a DLL project from the List that Visual Studio offers.
I made a super easy code, that didn’t use PyTorch to make sure I’m able to run it from C# and c++.
#include "pch.h"
BOOL APIENTRY DllMain( HMODULE hModule,
DWORD ul_reason_for_call,
LPVOID lpReserved
)
{
switch (ul_reason_for_call)
{
case DLL_PROCESS_ATTACH:
case DLL_THREAD_ATTACH:
case DLL_THREAD_DETACH:
case DLL_PROCESS_DETACH:
break;
}
return TRUE;
}
extern "C"
{
__declspec(dllexport) int sum(int a, int b)
{
return (a + b);
}
}
Code in point 3 works perfectly, so now its time to include the torch libraries and test the code again.
#include "pch.h"
#include <torch/torch.h> // <---------- Only modification in previous code
#include <torch/script.h> // <---------- Only modification in previous code
BOOL APIENTRY DllMain( HMODULE hModule,
DWORD ul_reason_for_call,
LPVOID lpReserved
)
{
switch (ul_reason_for_call)
{
case DLL_PROCESS_ATTACH:
case DLL_THREAD_ATTACH:
case DLL_THREAD_DETACH:
case DLL_PROCESS_DETACH:
break;
}
return TRUE;
}
extern "C"
{
__declspec(dllexport) int sum(int a, int b)
{
return (a + b);
}
}
When I try to compile this code, all the errors appear (attached image).
I’ve tried many things, including adding NOMINMAX to preprocessor definitions. The DLL compiles, but calling it from c# or c++ makes the application crash.
Also tried to:
#undef max
#undef min
in C++17.h, it compiles but happens the same as before.
I’ve also searched the forums and could’nt find anything similar.
I use the NOMINMAX macro and call my dll from C#, everything works fine. If it compiles, are you sure the crash is related to preprocessor definitions ? Is there error message when you try to call it from C# ?
Hello, thanks for replying.
I’m not sure the crash is related to the preprocessor definitions. I’m not sure where it’s coming from.
In C# I get a DllNotFoundException. I make the same process that I did in the step 2 that worked fine.
I call it with the following code:
public const string CppFunctionsDLL = @"DycareAPIBLE\Dependencies\Dll1.dll";
[DllImport(CppFunctionsDLL, CallingConvention = CallingConvention.Cdecl)]
extern static int sum(int a, int b);
var a = sum(2, 7);
I’ve also used a dependency walker. This one: Dependencies and nothing weird shows up.
From your experience, if the code in step 2 works and I add NOMINMAX to preprocessor and add the includes to the code. This should work fine?
Oh! I saw that topic the other day. That’s the reason I used a Dependency Walker. Firstly I used the one you mentioned there and then the other I linked in my previous reply.
I don’t see any red folder, what I see are a few yellow question marks. And there are some errors I don’t know how to solve in the middle box. Is this normal?
I’ve also followed your recommendation there and manually copied ALL the files from libtorch-win-shared-with-deps-1.10.0+cpu\libtorch\lib to my output folder. In my case there are 13 files, since I’m not using the CUDA version
But the error is still there.
Both my DLL and C# application are compiled in Debug x64. What I did is to create a folder in my solution, in which I copied my DLL and all necessary DLLs, I set every file to:
public const string CppFunctionsDLL = @"DycareAPIBLE\Dependencies\Dll1.dll";
Does it mean your DLL is in ‘DycareAPIBLE\Dependencies’ ?
You could try using just [DllImport(“Dll1.dll”)] and place the dll in the root folder along the pytorch DLLs
