Inherit QObject in a cuda class

skishore1998

I am c++ beginner working on an app development that acquires data in C++, processes it in GPU, and plots it in GUI using Qcustom plot library. I am using signals and slots to share data pointers between classes. But how to inherit Qobject in Cuda class?

I will give a sample code to represent my problem.

cmake_minimum_required(VERSION 3.5)

project(Trial04 VERSION 0.1 LANGUAGES CXX CUDA)

set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)

set(CMAKE_CUDA_STANDARD 11)
set(CMAKE_CUDA_STANDARD_REQUIRED ON)

find_package(Qt6 REQUIRED COMPONENTS Widgets PrintSupport)

qt_standard_project_setup()

include_directories(${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})

add_executable(t04a 
    mainwindow.ui
    mainwindow.cpp
    mainwindow.h
    main.cpp
    CudaProcessing.cu
    CudaProcessing.h
)

target_link_libraries(t04 PRIVATE Qt6::Widgets)

set_target_properties(t04 PROPERTIES
    WIN32_EXECUTABLE ON
    MACOSX_BUNDLE ON
)
set_target_properties(t04 PROPERTIES CUDA_SEPARABLE_COMPILATION ON)

//mainwindow.cpp
MainWindow::MainWindow(QWidget *parent)
    : QMainWindow(parent)
    , ui(new Ui::MainWindow)
{
    ui->setupUi(this);
    CudaProcessing *c1 = new CudaProcessing(this);
    c1->hello_world();
}

//CudaProcessing.cu
#include "CudaProcessing.h"
#include <stdio.h>

using namespace std;

__global__ void hello(){
    printf("hello from GPU");
}

void CudaProcessing::hello_world(){
    hello<<<1,1>>>();
}

//CudaProcessing.h
#pragma once
class CudaProcessing{
public:
    CudaProcessing() {}
    void hello_world();
    ~CudaProcessing() {}
};

The above code without QObject inheritance works fine.

#include <QObject>
class CudaProcessing:public QObject {
    Q_OBJECT
public:
    CudaProcessing(QObject *parent):QObject(parent){ }
    void hello_world();
    ~CudaProcessing() {}
};

When I inherit QObject in the above CUDA class I am getting an error

error "Qt requires a C++17 compiler, and a suitable value for __cplusplus. On MSVC, you must pass the /Zc:__cplusplus option to the compiler."

If I add the following

if (MSVC)
    add_compile_options(/Zc:__cplusplus)
endif()

to cmake file it shows a new error

error: Cannot specify link libraries for target "t04" which is not built by this project.

Is my approach of signal and slot in CUDA class right? If not, any suggestion would be greatly appreciated. Thanks

SGaist

Hi,

As @Pl45m4 wrote, CUDA kernels are a different beast that you should keep separated from QObject.
You won't be able to use signals and slots from CUDA code anyway.
Make your QObject based classes use CUDA kernels and keep compilation units separate.

Pl45m4

@skishore1998 said in Inherit QObject in a cuda class:

I am using signals and slots to share data pointers between classes

I don't see any Qt signal-slot connection in your code?!

jsulm

@skishore1998 said in Inherit QObject in a cuda class:

add_executable(t04a

in other places you use t04 - why? Fix you CMakeLists.txt file.

skishore1998

@Pl45m4
The purpose of inheriting Qobject is to use signals and slots. But I see the error immediately after inheriting Qobject. I have tried going ahead and implementing signals and slots in my main code but the error remains the same.

skishore1998

@jsulm Thanks, that was a typo. Now the error is

:-1: error: ninja: build stopped: subcommand failed.

jsulm

@skishore1998 Please post the actual error

skishore1998

@jsulm Sorry, This is what I get at the compile output

18:32:43: Running steps for project Trial04...
18:32:43: Starting: "C:\Qt\Tools\CMake_64\bin\cmake.exe" --build D:/desktop_c/Kishore/qwt_testing/10_contour_ADQ_linked/build-Trial04-Desktop_Qt_6_6_1_MSVC2019_64bit-Debug --target all
[1/3 20.0/sec] Building CUDA object CMakeFiles\t04.dir\CudaProcessing.cu.obj
FAILED: CMakeFiles/t04.dir/CudaProcessing.cu.obj 
C:\PROGRA~1\NVIDIA~2\CUDA\v12.3\bin\nvcc.exe -forward-unknown-to-host-compiler -DQT_CORE_LIB -DQT_GUI_LIB -DQT_WIDGETS_LIB -DUNICODE -DWIN32 -DWIN64 -D_ENABLE_EXTENDED_ALIGNED_STORAGE -D_UNICODE -D_WIN64 -ID:\desktop_c\Kishore\qwt_testing\10_contour_ADQ_linked\build-Trial04-Desktop_Qt_6_6_1_MSVC2019_64bit-Debug\t04_autogen\include -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.3\include" -isystem C:\Qt\6.6.1\msvc2019_64\include\QtWidgets -isystem C:\Qt\6.6.1\msvc2019_64\include -isystem C:\Qt\6.6.1\msvc2019_64\include\QtCore -isystem C:\Qt\6.6.1\msvc2019_64\mkspecs\win32-msvc -isystem C:\Qt\6.6.1\msvc2019_64\include\QtGui -D_WINDOWS -Xcompiler="/W3 /GR /EHsc" -Xcompiler="-MDd  -Zi -Ob0 -Od /RTC1" /Zc:__cplusplus -MD -MT CMakeFiles\t04.dir\CudaProcessing.cu.obj -MF CMakeFiles\t04.dir\CudaProcessing.cu.obj.d -x cu -rdc=true -c D:\desktop_c\Kishore\qwt_testing\10_contour_ADQ_linked\Trial04\CudaProcessing.cu -o CMakeFiles\t04.dir\CudaProcessing.cu.obj -Xcompiler=-FdCMakeFiles\t04.dir\,-FS
nvcc fatal   : A single input file is required for a non-link phase when an outputfile is specified
ninja: build stopped: subcommand failed.
18:32:44: The process "C:\Qt\Tools\CMake_64\bin\cmake.exe" exited with code 1.
Error while building/deploying project Trial04 (kit: Desktop Qt 6.6.1 MSVC2019 64bit)
When executing step "Build"
18:32:44: Elapsed time: 00:00.

Pl45m4

@skishore1998 said in Inherit QObject in a cuda class:

The above code without QObject inheritance works fine.

I think it's not a good idea to make your CUDA class (including the *.cu file) a QObject.
You need to compile your CUDA/cu file separately.

Here is an example
(it's somewhat old and it uses QMake, you have to adapt it to CMake)

https://www.qt.io/blog/2015/03/03/qt-weekly-28-qt-and-cuda-on-the-jetson-tk1

jsulm

Something is wrong with the nvcc.exe call, but I'm not a CUDA expert.

SGaist

Hi,

As @Pl45m4 wrote, CUDA kernels are a different beast that you should keep separated from QObject.
You won't be able to use signals and slots from CUDA code anyway.
Make your QObject based classes use CUDA kernels and keep compilation units separate.