Qt:找不到文件:@ rpath/CUDA.framework/Versions/A/CUDA用于架构x86_64 [英] Qt: File not found: @rpath/CUDA.framework/Versions/A/CUDA for architecture x86_64
问题描述
尝试将Qt与CUDA集成时遇到问题.我在安装了64位CUDA工具包的64位Mac上运行，但是，当我尝试构建代码时，错误 ld:找不到文件:@ rpath/CUDA.framework/Versions/A/CUDA对于体系结构x86_64 .
I'm having problems when trying to integrate Qt with CUDA. I am running on a 64Bit Mac with the 64Bit CUDA toolkit installed, however when I try to build my code the error ld: file not found: @rpath/CUDA.framework/Versions/A/CUDA for architecture x86_64 is thrown.
我已经验证了所有路径，但是始终抛出相同的错误.我的 .pro 配置代码如下:
I have verified all my paths but the same error is consistently thrown. My .pro configuration code is as follows:
QT += core gui
QT += multimedia
QT += multimediawidgets
QT += concurrent
greaterThan(QT_MAJOR_VERSION, 4): QT += widgets
TARGET = WebcamFilter
TEMPLATE = app
SOURCES += main.cpp\
mainwindow.cpp \
camerafeed.cpp \
HEADERS += mainwindow.h \
camerafeed.h
FORMS += mainwindow.ui
# CUDA Resources
CUDA_SOURCES += gaussian.cu
CUDA_DIR = /usr/local/cuda
# Path to header and lib files
INCLUDEPATH += $$CUDA_DIR/include
QMAKE_LIBDIR += $$CUDA_DIR/lib
# Libs used for source code
LIBS += -lcudart -lcuda
# GPU Architecture
CUDA_ARCH = sm_20
# Custom flags for nvcc
NVCCFLAGS = --compiler-options -fno-strict-aliasing -use_fast_math --ptxas-options=-v
# Prepare extra compiler configuration
CUDA_INC = $$join(INCLUDEPATH,' -I','-I',' ')
cuda.commands = $$CUDA_DIR/bin/nvcc -m64 -O3 -arch=$$CUDA_ARCH -c $$NVCCFLAGS \
$$CUDA_INC $$LIBS ${QMAKE_FILE_NAME} -o ${QMAKE_FILE_OUT} \
2>&1 | sed -r \"s/\\(([0-9]+)\\)/:\\1/g\" 1>&2
cuda.dependency_type = TYPE_C
cuda.depend_command = $$CUDA_DIR/bin/nvcc -O3 -M $$CUDA_INC $$NVCCFLAGS ${QMAKE_FILE_NAME}
cuda.input = CUDA_SOURCES
cuda.output = ${OBJECTS_DIR}${QMAKE_FILE_BASE}_cuda.o
# Tell Qt that we want add more stuff to the Makefile
QMAKE_EXTRA_COMPILERS += cuda
推荐答案
几个月前我遇到了这个问题(修复此问题后还有其他一些问题)，所以我想我会发布一个完全正常的QT/CUDA例子，现在我已经弄清楚了.我从较大的项目中提取了大部分 .pro 文件，适用于Linux和Mac(CUDA的东西在gpu文件夹中)，但是这部分代码仅在OS X上经过了测试.
I came across this problem a few months ago (plus some other issues after this was fixed) so I figured I'd just post a fully working QT/CUDA example now that I have it mostly figured out. I pulled most of the .pro file from a larger project for both Linux and Mac (CUDA stuff is in the gpu folder) but this bit of code has only been tested on OS X.
我当前正在使用:
CUDA 7.0驱动程序V7.0.27
OS X Yosemite 10.10.3
QT 5.3.1
CUDA 7.0 driver V7.0.27
OS X Yosemite 10.10.3
QT 5.3.1
如果您最近没有更新，请在尝试此代码之前，确保CUDA deviceQuery 和 bandwidthTest 示例仍在工作.
If you haven't updated recently make sure the CUDA deviceQuery and bandwidthTest samples are still working before trying this code.
下面的 .pro 文件可能是解决问题所需的全部，但是C ++代码也在下面.代码注释完成了大部分解释.
The .pro file below might be all you need to solve your problems but the C++ code is below as well. The code comments do most of the explaining.
#-------------------------------------------------
#
# Project created by QtCreator 2015-05-02T02:37:39
#
#-------------------------------------------------
QT += core gui
greaterThan(QT_MAJOR_VERSION, 4): QT += widgets
TARGET = qtcuda
TEMPLATE = app
# project build directories (if not using shadow build)
DESTDIR = $$system(pwd)
BUILDDIR = $$DESTDIR/build
MOC_DIR = $$BUILDDIR # moc_... files
UI_DIR = $$BUILDDIR # ui_mainwindow.cpp
OBJECTS_DIR = $$BUILDDIR/bin # .o binary files
SOURCES += main.cpp\
mainwindow.cpp
HEADERS += mainwindow.h
FORMS += mainwindow.ui
# NOTE: C++ flags are needed here for
# the CUDA Thrust library
############### UNIX FLAGS #####################
unix {
QMAKE_CXXFLAGS += -std=c++11
}
############### MAC FLAGS #####################
macx {
# libs that don't get passed to nvcc (we'll remove them from LIBS later)
NON_CUDA_LIBS += -stdlib=libc++
LIBS += $$NON_CUDA_LIBS
QMAKE_CXXFLAGS += -stdlib=libc++ -mmacosx-version-min=10.7
QMAKE_LFLAGS += -mmacosx-version-min=10.7
QMAKE_MACOSX_DEPLOYMENT_TARGET = 10.7
# specific to computers without older sdks
MAC_SDK = /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9. sdk/
if( exists( $$MAC_SDK) ) {
QMAKE_MAC_SDK = macosx10.9 # lowest sdk on my computer :/
}
# don't show warnings for c++11 extentions
QMAKE_CXXFLAGS += -Wno-c++11-extensions
}
################### CUDA ###################### (similar to your setup)
unix:!macx {
SED_STUFF = 2>&1 | sed -r \"s/\\(([0-9]+)\\)/:\\1/g\" 1>&2
}
macx {
SED_STUFF = 2>&1 | sed -E \"s/\\(([0-9]+)\\)/:\\1/g\" 1>&2
}
CUDA_DIR = /usr/local/cuda
# make sure cuda is available on the computer
if ( exists( $$CUDA_DIR/ ) ) {
message( "Configuring for cuda...");
DEFINES += CUDA_7 # # same as putting this in code -> #define CUDA_7
# Cuda sources
CUDA_SOURCES += cuda/wrappers.cu
# show files in working tree
OTHER_FILES += cuda/wrappers.cu \
cuda/wrappers.cuh \
cuda/helper_cuda.h
# Path to cuda install
CUDA_LIB = $$CUDA_DIR/lib
# Pather to header and lib files
INCLUDEPATH += $$CUDA_DIR/include \
cuda # my cuda files
QMAKE_LIBDIR += $$CUDA_LIB
# prevents warnings from code we didn't write
QMAKE_CXXFLAGS += -isystem $$CUDA_DIR/include
LIBS += -lcudart # add other cuda libs here (-lcublas -lcurand, etc.)
# SPECIFY THE R PATH FOR NVCC!!!!! (your problem...previously my problem)
QMAKE_LFLAGS += -Wl,-rpath,$$CUDA_LIB
NVCCFLAGS = -Xlinker -rpath,$$CUDA_LIB
# libs used in the code
CUDA_LIBS = $$LIBS
CUDA_LIBS -= $$NON_CUDA_LIBS # remove libs nvcc won't recognize
# GPU architecture (might be a way to detect this somehow instead of hardcoding)
CUDA_ARCH = sm_20 # <- based on specs from your code. This was tested with sm_30
# Some default NVCC flags
NVCCFLAGS += --compiler-options -fno-strict-aliasing -use_fast_math --ptxas-options=-v --std=c++11
# Prepare the extra compiler configuration (taken from the nvidia forum)
CUDA_INC = $$join(INCLUDEPATH,' -I','-I',' ')
cuda.commands = $$CUDA_DIR/bin/nvcc -m64 -O3 -arch=$$CUDA_ARCH -c $$NVCCFLAGS \
$$CUDA_INC $$CUDA_LIBS ${QMAKE_FILE_NAME} -o ${QMAKE_FILE_OUT} \
$$SED_STUFF
# nvcc error printout format ever so slightly different from gcc
# http://forums.nvidia.com/index.php?showtopic=171651
cuda.dependency_type = TYPE_C
cuda.depend_command = $$CUDA_DIR/bin/nvcc -O3 -M $$CUDA_INC $$NVCCFLAGS ${QMAKE_FILE_NAME}
cuda.input = CUDA_SOURCES
cuda.output = ${OBJECTS_DIR}${QMAKE_FILE_BASE}_cuda.o
# Tell Qt that we want add more stuff to the Makefile
QMAKE_EXTRA_COMPILERS += cuda
} # endif CUDA
以下两个文件由用于执行CUDA代码的外部函数组成. .cu 文件定义包含CUDA代码的函数，并使用NVCC进行编译(如 .pro 文件中所指定). .cuh 文件用作头文件，并且仅声明相同的函数，以便C ++文件可以引用它们.仅 wrappers.cuh 需要包含在C ++代码中.
The following two files are composed of extern functions used to execute CUDA code. The .cu file defines functions that contain CUDA code and gets compiled with NVCC (as specified in the .pro file). The .cuh file is used as a header file and simply declares the same functions so they can be referenced by C++ files. Only wrappers.cuh needs to be included in the C++ code.
注意:引用的 helper_cuda.h 文件可在 NoteNote :此项目假定保留 wrappers.cuh ， wrappers.cu 和 helper_cuda.h 在项目目录中标有 cuda 的文件夹中.
NoteNote: This project assumes wrappers.cuh, wrappers.cu, and helper_cuda.h are kept in a folder labeled cuda within the project directory.
#ifndef WRAPPERS_CUH
#define WRAPPERS_CUH
typedef unsigned int uint;
extern "C"
{
void cudaInit();
void allocateArray(void **devPtr, int size);
void freeArray(void *devPtr);
void copyArrayToDevice(void *device, const void *host, int offset, int size);
void copyArrayFromDevice(void *host, const void *device, int size);
uint sumNumbers(uint *dNumbers, uint n);
// not used here but useful when calling kernel functions
void computeGridSize(uint n, uint blockSize, uint &numBlocks, uint &numThreads);
}
#endif // WRAPPERS_CUH
cuda/wrappers.cu
#include <cuda_runtime.h>
#include <cuda_gl_interop.h>
#include <thrust/device_ptr.h>
#include <thrust/reduce.h>
#include "helper_cuda.h"
typedef unsigned int uint;
extern "C"
{
void cudaInit()
{
int devID;
// use device with highest Gflops/s
devID = findCudaDevice();
if (devID < 0)
{
printf("No CUDA Capable devices found, exiting...\n");
exit(EXIT_SUCCESS);
}
}
void allocateArray(void **devPtr, size_t size)
{
checkCudaErrors(cudaMalloc(devPtr, size));
}
void freeArray(void *devPtr)
{
checkCudaErrors(cudaFree(devPtr));
}
void copyArrayToDevice(void *device, const void *host, int offset, int size)
{
checkCudaErrors(cudaMemcpy((char *) device + offset, host, size, cudaMemcpyHostToDevice));
}
void copyArrayFromDevice(void *host, const void *device, int size)
{
checkCudaErrors(cudaMemcpy(host, device, size, cudaMemcpyDeviceToHost));
}
uint sumNumbers(uint *dNumbers, uint n)
{
// simple reduction from 1 to n
thrust::device_ptr<uint> dp_numbers(dNumbers);
return thrust::reduce(dp_numbers, dp_numbers + n);
}
//Round a / b to nearest higher integer value
uint iDivUp(uint a, uint b)
{
return (a % b != 0) ? (a / b + 1) : (a / b);
}
// compute grid and thread block size for a given number of elements
void computeGridSize(uint n, uint blockSize, uint &numBlocks, uint &numThreads)
{
numThreads = min(blockSize, n);
numBlocks = iDivUp(n, numThreads);
}
}
接下来的三个文件将创建一个简单的QT窗口并检查鼠标事件.每次移动鼠标时，X和Y像素位置都会加在一起以创建 n .然后，使用CUDA函数查找 1 + 2 + ... + n (是的，这很奇怪而且很随机；目的是显示CUDA以快速简便的方式运行).
The next three files create a simple QT window and check for mouse events. Every time the mouse is moved the X and Y pixel positions are added together to create n. Then a CUDA function is used to find 1 + 2 + ... + n (yes this is weird and random; the point was to show CUDA running in a quick and easy way).
因此，如果鼠标位于(23，45)，则:
n =(23 + 45)= 68 和
1 + 2 + ... + n = 2346
So if the mouse is at (23, 45) then:
n = (23 + 45) = 68 and
1 + 2 + ... + n = 2346
然后将其显示在窗口底部.
This is then displayed at the bottom of the window.
#include "mainwindow.h"
#include <QApplication>
int main(int argc, char *argv[])
{
QApplication a(argc, argv);
MainWindow w;
w.show();
return a.exec();
}
mainwindow.h
#ifndef MAINWINDOW_H
#define MAINWINDOW_H
#include <QMainWindow>
namespace Ui {
class MainWindow;
}
class MainWindow : public QMainWindow
{
Q_OBJECT
public:
explicit MainWindow(QWidget *parent = 0);
~MainWindow();
// events are passed here
virtual bool eventFilter(QObject *obj, QEvent *event);
private:
Ui::MainWindow *ui;
uint *m_dNumbers; // device array
};
#endif // MAINWINDOW_H
mainwindow.cpp
#include "mainwindow.h"
#include "ui_mainwindow.h"
#include <QEvent>
#include <QMouseEvent>
#include <assert.h>
#include "wrappers.cuh"
const uint MAX_NUMBERS = 5000;
MainWindow::MainWindow(QWidget *parent) :
QMainWindow(parent),
ui(new Ui::MainWindow)
{
// basic ui setup and event filter for mouse movements
ui->setupUi(this);
qApp->installEventFilter(this);
// create a host array and initialize it to {1, 2, 3, ..., MAX_NUMBERS}
uint hNumbers[MAX_NUMBERS];
for (uint i = 0; i < MAX_NUMBERS; i++)
{
hNumbers[i] = i + 1;
}
// CUDA FUNCTIONS:
cudaInit(); // initialiaze the cuda device
allocateArray((void**)&m_dNumbers, MAX_NUMBERS*sizeof(int)); // allocate device array
copyArrayToDevice(m_dNumbers, hNumbers, 0, MAX_NUMBERS*sizeof(int)); // copy host array to device array
}
MainWindow::~MainWindow()
{
// CUDA FUNCTION: free device memory
freeArray(m_dNumbers);
delete ui;
}
// used to detect mouse movement events
bool MainWindow::eventFilter(QObject *, QEvent *event)
{
if (event->type() == QEvent::MouseMove)
{
// find mouseX + mouseY
QMouseEvent *mouseEvent = static_cast<QMouseEvent*>(event);
QPoint p = mouseEvent->pos();
uint n = std::min((uint)(p.x() + p.y()), MAX_NUMBERS);
// CUDA FUNCTION:
// compute the sum of 1 + 2 + 3 + ... + n
uint sum = sumNumbers(m_dNumbers, n);
// check that the sum is correct
assert(sum == ( (n * (n+1) ) / 2 ) );
// show the sum at the bottom of the window
statusBar()->showMessage(QString("Mouse pos: (%1, %2) Sum from 0 to %3 = %4").arg(p.x()).arg(p.y()). arg(n).arg(sum));
}
return false;
}
最后一个但并非最不重要的 .ui 文件，如果您想实际构建并运行项目:
And last but not least the .ui file if you want to actually build and run the project:
<?xml version="1.0" encoding="UTF-8"?>
<ui version="4.0">
<class>MainWindow</class>
<widget class="QMainWindow" name="MainWindow">
<property name="geometry">
<rect>
<x>0</x>
<y>0</y>
<width>400</width>
<height>300</height>
</rect>
</property>
<property name="windowTitle">
<string>MainWindow</string>
</property>
<widget class="QWidget" name="centralWidget"/>
<widget class="QMenuBar" name="menuBar">
<property name="geometry">
<rect>
<x>0</x>
<y>0</y>
<width>400</width>
<height>22</height>
</rect>
</property>
</widget>
<widget class="QToolBar" name="mainToolBar">
<attribute name="toolBarArea">
<enum>TopToolBarArea</enum>
</attribute>
<attribute name="toolBarBreak">
<bool>false</bool>
</attribute>
</widget>
<widget class="QStatusBar" name="statusBar"/>
</widget>
<layoutdefault spacing="6" margin="11"/>
<resources/>
<connections/>
</ui>
我知道QT/CUDA流程可能很烦人，而且自您提出问题以来已经沉默了半年，但希望这会有所帮助.
I know the QT/CUDA process can be annoying and it's been half a year of silence since you asked the question but hopefully this helps.
这篇关于Qt:找不到文件:@ rpath/CUDA.framework/Versions/A/CUDA用于架构x86_64的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
