Windows下qtCreator+MSVC2017编译器+CUDA混合编程的三种方法

qtCreator+MSVC2017编译器+CUDA混合编程的三种方法

一、准备好工程文件

（一）Qt中创建一个工程：（二）cuda程序内容（三）cuda程序的头文件内容（四）主程序入口内容二、混合编程的三种方法

（一）使用NVCC编译器预编译生成*.obj后在Qt工程文档中调用（二）在pro文件中添加额外编译器nvcc和主程序同时编译（三）将cu文件封装为库文件，而后在工程文件中引用三、总结
尝试在Qt中实现CUDA的混合编程，因为CUDA不支持MinGW编译器，因此只能使用VS编译器，于是选用MSVC2017编译器。
为了实现混合编程，其实实际上是同一种方法，核心在于对CUDA部分的代码使用NVCC编译器，而对别的代码内容使用VS2017编译器。区别在于，对CUDA代码的编译结果在VS2017中调用的方式。

一、准备好工程文件（一）Qt中创建一个工程：

（二）cuda程序内容

bilinear.cu

#include "bilinear.h"

extern "C"
__global__ void hellofromGPU(void)
{
    printf("GPU:hello sunyin");
}

void showhello(void)
{
    hellofromGPU <<<1,10>>>();
    cudaDeviceSynchronize();

}

（三）cuda程序的头文件内容

bilinear.h

#ifndef BILINEAR_H
#define BILINEAR_H
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include 
#include "malloc.h"
#define  WIDTH 11

#define  HEIGHT 10
#define  X_INTER 3
#define  Y_INTER 3
#define  BLOCK_SIZE 8

void showhello(void);
#endif // BINLINEARINTERPOLATIONCUDA_H

（四）主程序入口内容

main.cpp

#include
#include "bilinear.h"
int main(void)
{
   showhello();
   while(1);
}
//nvcc -arch sm_45

二、混合编程的三种方法（一）使用NVCC编译器预编译生成*.obj后在Qt工程文档中调用

将VS2017的include，lib，bin写入环境变量。如果是默认的安装目录，应该是在C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6的目录下。

使用VS2017自带的命令行工具，适用于VS 2017的x64本机工具命令提示（也有可能是英文名，搞不清是哪个，可以将命令工具挨个试一下）。转到bilinear.cu目录下。

输入nvcc -Xcompiler "/MDd" -D_DEBUG -c bilinear.cu，则会生成bilinear.obj。

而后在testCUDA.pro中添加OBJECTS += bilinear.obj，就可以正常编译了。

将bilinear.obj放到编译目录下，debug下。

（二）在pro文件中添加额外编译器nvcc和主程序同时编译

这个不需要事先生成bilinear.obj文件，但在pro文件中添加额外编译器nvcc的目的也是生成bilinear.obj，原理是一样的，只是将生成的命令写入pro文件中，在编译前先生成pro文件。
testCUDA.pro

QT -= gui

ConFIG += c++11 console
ConFIG -= app_bundle

# You can make your code fail to compile if it uses deprecated APIs.
# In order to do so, uncomment the following line.
#DEFINES += QT_DISABLE_DEPRECATED_BEFORE=0x060000    # disables all the APIs deprecated before Qt 6.0.0

SOURCES += 
        main.cpp
OBJECTS += 

# CUDA 设置
CUDA_SOURCES += bilinear.cu
CUDA_DIR = "C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6"
SYSTEM_NAME = x64
SYSTEM_TYPE = 64
CUDA_ARCH  = sm_61
NVCC_OPTIONS = --use_fast_math

# 头文件路径
INCLUDEPATH += "$$CUDA_DIR/include"

# 导入库文件路径
QMAKE_LIBDIR += "$$CUDA_DIR/lib/x64"
CUDA_INC = $$join(INCLUDEPATH,'" -I"','-I"','"')

# 依赖库
CUDA_LIB_NAMES += 
cuda 
cudadevrt 
cudart 

CONFIG(debug, debug|release) {
CUDA_LIB_NAMES += ucrtd 
}
else{
CUDA_LIB_NAMES += ucrt 
}

for(lib, CUDA_LIB_NAMES) {
    CUDA_LIBS += $$lib.lib
}
for(lib, CUDA_LIB_NAMES) {
    NVCC_LIBS += -l$$lib
}
LIBS += $$NVCC_LIBS


MSVCRT_link_FLAG_DEBUG   = "/MDd"
MSVCRT_link_FLAG_RELEASE = "/MD"

# 这部分内容，实质上就是在Qt中调用环境中的命令行工具来先编译生成bilinear.obj
CONFIG(debug, debug|release) {
    # Debug 模式
    OBJECTS_DIR = debug/obj
    CUDA_OBJECTS_DIR = debug
    cuda_d.input = CUDA_SOURCES
    cuda_d.output = $$CUDA_OBJECTS_DIR/${QMAKE_FILE_base}.obj
    cuda_d.commands = $$CUDA_DIR/bin/nvcc.exe -D_DEBUG $$NVCC_OPTIONS $$CUDA_INC $$LIBS 
                      --machine $$SYSTEM_TYPE -arch=$$CUDA_ARCH 
                      --compile -cudart static -g -DWIN32 -D_MBCS 
                      -Xcompiler "/wd4819,/EHsc,/W3,/nologo,/Od,/Zi,/RTC1" 
                      -Xcompiler $$MSVCRT_link_FLAG_DEBUG 
                      -c -o ${QMAKE_FILE_OUT} ${QMAKE_FILE_NAME}
    cuda_d.dependency_type = TYPE_C
    QMAKE_EXTRA_COMPILERS += cuda_d
}
else {
    # Release 模式
    OBJECTS_DIR = release/obj
    CUDA_OBJECTS_DIR = release
    cuda.input = CUDA_SOURCES
    cuda.output = $$CUDA_OBJECTS_DIR/${QMAKE_FILE_base}.obj
    cuda.commands = $$CUDA_DIR/bin/nvcc.exe $$NVCC_OPTIONS $$CUDA_INC $$LIBS 
                    --machine $$SYSTEM_TYPE -arch=$$CUDA_ARCH 
                    --compile -cudart static -D_MBCS 
                    -Xcompiler "/wd4819,/EHsc,/W3,/nologo,/O3,/Zi" 
                    -Xcompiler $$MSVCRT_link_FLAG_RELEASE 
                    -c -o ${QMAKE_FILE_OUT} ${QMAKE_FILE_NAME}
    cuda.dependency_type = TYPE_C
    QMAKE_EXTRA_COMPILERS += cuda
}

HEADERS += 
    bilinear.h

DISTFILES += 
    bilinear.cu

注意！这个方法，我在windows sever2019服务器上使用，不知道为什么，一直没有编译生成bilinear.obj文件，试了很久都没成功，但是换了一台windows10，几乎一样的工具，一遍就编译通过了。所以如果你也遇到了类似的问题，换个电脑，换个系统试试。
正常编译输出内容如下，仅供参考：

12:00:27: 为项目testCUDA执行步骤 ...
12:00:27: 配置没有改变, 跳过 qmake 步骤。
12:00:28: 正在启动 "C:QtQt5.12.12ToolsQtCreatorbinjomjom.exe" 

	C:QtQt5.12.12ToolsQtCreatorbinjomjom.exe -f Makefile.Debug
	C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6/bin/nvcc.exe -D_DEBUG --use_fast_math -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6/include" -lcuda -lcudadevrt -lcudart -lucrtd --machine 64 -arch=sm_61 --compile -cudart static -g -DWIN32 -D_MBCS -Xcompiler /wd4819,/EHsc,/W3,/nologo,/Od,/Zi,/RTC1 -Xcompiler /MDd -c -o debugbilinear.obj ..testCUDAbilinear.cu
	cl -c -nologo -Zc:wchar_t -FS -Zc:rvalueCast -Zc:inline -Zc:strictStrings -Zc:throwingNew -Zc:referenceBinding -Zc:__cplusplus -Zi -MDd -W3 -w34100 -w34189 -w44996 -w44456 -w44457 -w44458 -wd4577 -wd4467 -EHsc /FddebugobjtestCUDA.vc.pdb -DUNICODE -D_UNICODE -DWIN32 -D_ENABLE_EXTENDED_ALIGNED_STORAGE -DWIN64 -DQT_QML_DEBUG -DQT_CORE_LIB -I..testCUDA -I. -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6include" -IC:QtQt5.12.125.12.12msvc2017_64include -IC:QtQt5.12.125.12.12msvc2017_64includeQtCore -Idebug -IC:QtQt5.12.125.12.12msvc2017_64mkspecswin32-msvc -Fodebugobj @C:UsersADMINI~1AppDataLocalTempmain.obj.31692.343.jom
main.cpp
bilinear.cu
	link /NOLOGO /DYNAMICbase /NXCOMPAT /DEBUG /SUBSYSTEM:ConSOLE "/MANIFESTDEPENDENCY:type='win32' name='Microsoft.Windows.Common-Controls' version='6.0.0.0' publicKeyToken='6595b64144ccf1df' language='*' processorArchitecture='*'" /MANIFEST:embed /OUT:debugtestCUDA.exe @C:UsersADMINI~1AppDataLocalTemptestCUDA.exe.31692.12781.jom
12:00:42: 进程"C:QtQt5.12.12ToolsQtCreatorbinjomjom.exe"正常退出。
12:00:42: Elapsed time: 00:15.

把Makefile.Debug中的部分内容也贴一下，仅供参考：

compiler_cuda_d_make_all: debugbilinear.obj
compiler_cuda_d_clean:
	-$(DEL_FILE) debugbilinear.obj
debugbilinear.obj: ..testCUDAbilinear.cu 
		..testCUDAbilinear.h 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecuda_runtime.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrthost_config.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includebuiltin_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includedevice_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrthost_defines.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includedriver_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includevector_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesurface_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includetexture_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includelibrary_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includechannel_descriptor.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecuda_runtime_api.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecuda_device_runtime_api.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includedriver_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includevector_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includevector_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtdevice_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtdevice_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includedevice_atomic_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includedevice_atomic_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtdevice_double_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtdevice_double_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_20_atomic_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_20_atomic_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_32_atomic_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_32_atomic_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_35_atomic_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_60_atomic_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_60_atomic_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_20_intrinsics.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_20_intrinsics.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_30_intrinsics.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_30_intrinsics.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_32_intrinsics.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_32_intrinsics.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_35_intrinsics.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_61_intrinsics.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesm_61_intrinsics.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtsm_70_rt.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtsm_70_rt.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtsm_80_rt.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtsm_80_rt.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesurface_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecuda_surface_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includetexture_fetch_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecuda_texture_types.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includetexture_indirect_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includesurface_indirect_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtcommon_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtmath_functions.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtfunc_macro.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includecrtmath_functions.hpp" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includemath_constants.h" 
		"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6includedevice_launch_parameters.h"
	C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6/bin/nvcc.exe -D_DEBUG --use_fast_math -I"C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6/include" -lcuda -lcudadevrt -lcudart -lucrtd --machine 64 -arch=sm_61 --compile -cudart static -g -DWIN32 -D_MBCS -Xcompiler /wd4819,/EHsc,/W3,/nologo,/Od,/Zi,/RTC1 -Xcompiler /MDd -c -o debugbilinear.obj ..testCUDAbilinear.cu

（三）将cu文件封装为库文件，而后在工程文件中引用

这个方法我具体没有测试，但是理论上跟方法一应该是同个道理，只是方法一是将bilinear.cu文件编译成了bilinear.obj文件，在这个方法中，是将bilinear.cu封装成bilinear.lib或者bilinear.dll文件，而后调用。是一个道理。
在这个方法中，在编译完成后，应该至少生成两个文件。
bilinear.h和bilinear.lib。或者还有一个bilinear.dll。
具体的我就不试了，但是想想应该是简单的。
包括在openCV中调用到CUDA的内容，应该实质上也是类似于此的方法，因此调用openCV的库文件，并不需要自己解决CUDA的配置问题。

三、总结

千万记住CUDA的编译暂不支持MinGW，所以别在这方面浪费时间了。如果生成obj文件失败，可以换个系统试试。根据我的观察，我在之前那个系统中一直生成失败，是因为在之前那个系统中，压根就没启用cmd命令行来调用额外编译器nvcc，具体原因暂时不知道。关于不同显卡的sm值是多少的填写，在英伟达官网可以查到，比如我的GTX1060 3G写的是6.1，则sm=61。查询GPU算力的官网

Windows下qtCreator+MSVC2017编译器+CUDA混合编程的三种方法

C/C++/C#相关栏目本月热门文章