I am trying to compile http://www.kevinbeason.com/smallpt/ raytracer using Qt5 and Visual Studio 2012 compiler.
It works Okay but when I try to use openmp by adding
QMAKE_CXXFLAGS += -fopenmp
LIBS += -fopenmp
to project.pro
, Qt says that /fopenmp is not recognized and ignored.
When I compile using VS command
cl /c /O2 /EHsc /openmp main.cpp
it works and I get a program 3 times faster then the one compiled from Qt.
How to make Qt recognize openmp and how to enable the other optimizations in the command line?
Thanks in advance.
I finally got all the required elements to get it compiled with openmp from Qt Creator:
Add #include <omp.h> to the source, which is not required if you compile with
the command line mentioned above.
Add QMAKE_CXXFLAGS += -openmp to the project file. It wont work
with -fopenmp. No need for any openmp lib like I did ( LIBS += -openmp ) neither for QMAKE_LFLAGS += -openmp
Compile as release. With debug it has no impact on performance.
The other optimization options are already configured in mkspecs\win32-msvc2012\
qmake.conf
Hope this helps someone else.
I want to run Qt project with openmp. I have followed the above steps, I could build the project, but execution time is increasing instead of reducing.
Here is the configuration I used
CONFIG += console c++11
CONFIG -= app_bundle
#CONFIG -= qt
QMAKE_CXXFLAGS+= -openmp
Related
I am newbie in openmp. The following is the environment.
OS : Mac OSX Mavericks
Compiler : gcc (MacPorts gcc48 4.8.2_0) 4.8.2
IDE : Eclipse Kepler CDT plugin
I wrote the following openmp program
#include < stdio.h>
#include < omp.h>
int main()
{
#pragma omp parallel
{
int i=omp_get_thread_num();
printf("hello (%d)",i);
printf("world (%d)",i);
}
}
I compiled the above program and got the error that omp.h is not found and lgomp not found. Hence I added in the project properties an include path with /opt/local/lib/gcc48/gcc/x86_64-apple-darwin13/4.8.2/include and a library path /opt/local/lib/gcc48. The include path had the omp.h file and the library path had the file libomp.o.
I include the -fopenmp option in both the linker and the compiler option through project properties. It is compiling with gcc -I/opt/local/lib/gcc48/gcc/x86_64-apple-darwin13/4.8.2/include -O0 -g3 -Wall -c -fmessage-length=0 -fopenmp -MMD -MP -MF"src/OpenMPCourseExamples.d" -MT"src/OpenMPCourseExamples.d" -o "src/OpenMPCourseExamples.o" "../src/OpenMPCourseExamples.c" and linking with the command "gcc -L/opt/local/lib/gcc48 -fopenmp -o "OpenMPCourseExamples" ./src/OpenMPCourseExamples.o".
With the above command it compiles without an error but with a warning - "warning: unknown pragma ignored [-Wunknown-pragmas] #pragma omp parallel".
Also, I set an environment variable in the launch properties with OMP_NUM_THREADS=4. I ran the program that compiled with the above warning. I am getting only "hello (0)world (0)". I was under the impression that I should start four threads and should see the other outputs of "hello(1)world(1)hello(2)world(2)hello(3)world(3)" in some ordering as well. Now, here are my following questions.
Why am I getting the #pragma warning?
Is the compiler really detecting the openmp and building with openmp?
If everything is correct, why am I not seeing four different threads getting started?
The final steps that worked for openmp, macports gcc compiler, eclipse CDT in mac osx mavericks are.
Enable "Make ToolChain(s) Preferred" in Eclipse->Preference->C/C++->New C/C++ Project Wizard.
sudo port select --list gcc and set it sudo port select --set gcc with mp-gcc.
File->New Project->C Project (not C++) and create a hello world project.
In Project->Properties->C/C++ Build->Settings->Tool Settings set the following. (a) GCC C Compiler to /opt/local/bin/gcc-mp-4.8 (b)MAC OSX Linker to /opt/local/bin/gcc-mp-4.8
Build the hello world project and make sure, it compiles and runs successfully.
Include the open mp code. The code asked in the question of this page.
Go to again Project->Properties->C/C++ Build->Settings->Tool Settings set the following. (a) GCC Compiler ->Miscellaneous add -fopenmp (b) MacOSx Linker->Miscellaneous set -fopenmp
Build the code again.
The above steps worked good for me.
MacPorts configures the GCC build process with --program-suffix=-mp-${major} and therefore all compiler executables have the -mp-4.8 suffix. When you call gcc, you end up using Apple's Clang compiler, which does not support OpenMP and therefore does not recognise the -fopenmp option and #pragma omp ....
You have to do the following changes to the project settings:
Change the compiler command to gcc-mp-4.8
Change the linker command to gcc-mp-4.8
Remove the explicit specification of the include and library paths since the presence of -fopenmp adds them automatically.
I have installed Qt 5.1.1 with MinGW 4.8 and QtCreator through Qt Online installer.
I launched Qt creator and made new project (Qt application Desktop) with MainWindow class based on QMainWindow (default). It works fine and popped up a mainwindow after compilation.
However, when I used C:\Qt\Qt5.1.1\Tools\QtCreator\bin\jom.exe instead of C:\Qt\Qt5.1.1\Tools\mingw48_32\bin\mingw32-make.exe , I got some output like below.
jom 1.0.13 - empower your cores
C:\Qt\Qt5.1.1\Tools\QtCreator\bin\jom.exe -f Makefile.Release
g++ -c -pipe -fno-keep-inline-dllexport -O2 -frtti -Wall -Wextra -fexceptions -mthreads -DUNICODE -DQT_NO_DEBUG -DQT_WIDGETS_LIB -DQT_GUI_LIB -DQT_CORE_LIB -DQT_NEEDS_QMAIN -I../foobar -I'C:/Qt/Qt5.1.1/5.1.1/mingw48_32/include' -I'C:/Qt/Qt5.1.1/5.1.1/mingw48_32/include/QtWidgets' -I'C:/Qt/Qt5.1.1/5.1.1/mingw48_32/include/QtGui' -I'C:/Qt/Qt5.1.1/5.1.1/mingw48_32/include/QtCore' -I'release' -I'.' -I'.' -I'C:/Qt/Qt5.1.1/5.1.1/mingw48_32/mkspecs/win32-g++' -o release/main.o ../foobar/main.cpp
In file included from ../foobar/main.cpp:1:0:
../foobar/mainwindow.h:4:23: fatal error: QMainWindow: No such file or directory
#include <QMainWindow>
^
compilation terminated.
jom: D:\work\build-foobar-Desktop_Qt_5_1_1_MinGW_32bit-Release\Makefile.Release [release\main.o] Error 1
jom: D:\work\build-foobar-Desktop_Qt_5_1_1_MinGW_32bit-Release\Makefile [release] Error 2
15:17:38: Process "C:\Qt\Qt5.1.1\Tools\QtCreator\bin\jom.exe" stopped with code 2
Error while building/deploying project foobar (kit: Desktop Qt 5.1.1 MinGW 32bit)
During step 'Make'
15:17:38: Elapsed time: 00:00.
My .pro is like this.
#-------------------------------------------------
#
# Project created by QtCreator 2013-09-08T15:16:13
#
#-------------------------------------------------
QT += core gui
greaterThan(QT_MAJOR_VERSION, 4): QT += widgets
TARGET = foobar
TEMPLATE = app
SOURCES += main.cpp\ mainwindow.cpp
HEADERS += mainwindow.h
FORMS += mainwindow.ui
I think the line in .pro greaterThan(QT_MAJOR_VERSION, 4): QT += widgets and in the output during compilation -I'C:/Qt/Qt5.1.1/5.1.1/mingw48_32/include/QtWidgets' looks fine, but failed.
How can I compile with Qt 5.1 and jom on Windows 7 (32 bit)?
Thanks in advance.
Jom is a replacement of nmake and not MinGW's make.
nmake is the Visual Studio command line tool used to compile makefiles. Unfortunately that tool is using only one core (Parallel builds are handled inside Visual Studio either directly or using MSBuild, I'm not really sure). That being said, Jörg Bornemann wrote a tool called jom that sort of adds the -j command to nmake. It will allow you to compile your code using the Visual C++ compiler across different threads.
MinGW on the other end contains (amongst other things) a port of the GNU GCC compiler for Windows. Being a port, not everything that is possible to do with GCC is possible with MinGW and I guess that -jN is one of these. However, MinGW do support the option -j that will spawn as many processes as it can (beware that if you use this option, you may not be able to use your computer whilst compiling).
In your case, you probably want to use MinGW as your Qt version is compiled with it. If you want to use Visual Studio, you will have to install it but also either:
recompile Qt (and all other library dependencies) or
download the binary corresponding to the Visual Studio version you just installed.
Using jom should just work with QtCreator once this is done.
In my case I fixed this error by overriding system environment PATH in project settings by adding only bin path to qt, mingw and qtcreator: "E:\Qt\4.8.1\bin;E:\qt\mingw-4.4.0\bin;E:\Qt\qtcreator-4.0.0\bin" and after Run qmake and Rebuild
I have compiled TBB from source using Mingw following the comment #5 in this post: http://software.intel.com/en-us/forums/topic/291331. That went ok.
When I try to use the new TBB library in a QtCreator project, I end with this errors (ignore the warning messages): http://postimage.org/image/yrrecugix/
Here's the sample code I tried (I omit the non-tbb code):
#include "tbb/task_scheduler_init.h"
int main()
{
tbb::task_scheduler_init init;
/// more things.
}
And here's the .pro file:
TEMPLATE = app
CONFIG += console
CONFIG -= qt
SOURCES += main.cpp
#QMAKE_CXXFLAGS += -fopenmp
#QMAKE_LFLAGS += -fopenmp
INCLUDEPATH += "E:\TRABAJO\LIBRERIAS\tbb-4.1_src\include"
LIBS += -L"E:\TRABAJO\LIBRERIAS\tbb-4.1_src\build\windows_intel64_gcc_mingw4.5.4_debug\" \
-ltbb_debug
Any idea?.
Thanks!.
When built with MinGW on Windows, TBB binaries are tbb.dll and tbb_debug.dll. The option -ltbb_debug in your configuration files probably causes the linker to look for libtbb_debug.<something>. It can't find such a binary and reports about unresolved symbols.
I've created a basic app in visual studio with Qt and an openGL widget. I export a .pro file from the addin so that I can compile in Linux.
I get the message:
/usr/bin/ld: cannot find -lopengl32
/usr/bin/ld: cannot find -lglu32
... when running make.
The issue goes away and the program compiles fine if I remove:
LIBS += -lopengl32 \
-lglu32
... from the generated .pro file.
I understand the "export .pro" feature is not flawless, but is this the correct way to fix this problem?
I don't quite remember the exact name of the OpenGL library on Linux, but it's not opengl32. Rather libopengl. So it's perfectly normal.
This will make you link with opengl32.lib on windows only:
win32 : LIBS += -lopengl32 -lglu32
and if needed, this will make you link with libopengl on Linux :
unix : LIBS += -lopengl -lglu
For the macx platform scope, I guess there is additional work to link with the OpenGL Framework.
I have a Qt project and i wanted to use google tests framework GoogleTestFramework in order to do google test. The framework compiled fine on Linux.
I have managed to compile the libraries on windows, using tips provided in the following link:
Installing Gtest without pthreads
The following advice do not let me compile the libraries, because of errors due to pthreads:
Sackoverflow compiling with MSYS
I have :
GNU Make 3.81 for make.exe
GNU Make 3.82 for mingw32-make
CMake 2.8.7
QT 4.7.4
Qt MingW gcc v 4.4.0
When i try to compile my test projet, i get an error saying:
undefined reference to '__chkstk_ms'
File not found: gtest-all.cc
Here is my qmake project code (I have semplified the code).
QT += core gui
DEPENDPATH += . headers src ../FastTrans/headers ../FastTrans/src ../FastTrans/forms C:/MinGW/msys/1.0/home/zakharan/gtest-1.6.0/include
INCLUDEPATH += . headers src ../FastTrans/headers ../FastTrans/src ../FastTrans/forms C:/MinGW/msys/1.0/home/zakharan/gtest-1.6.0/include
TARGET = tstall
#CONFIG += console
#CONFIG -= app_bundle
TEMPLATE = app
HEADERS += ../FastTrans/headers/shortcutsactionsitem.h \
../FastTrans/headers/shortcutsactionsmodel.h \
../FastTrans/headers/isavesettings.h
SOURCES += ../FastTrans/src/shortcutsactionsitem.cpp \
../FastTrans/src/shortcutsactionsmodel.cpp \
../FastTrans/src/isavesettings.cpp \
src/tstshortcutsactionsitem.cpp \
src/tstshortcutsactionmodel.cpp
INCLUDEPATH += $$PWD/
DEPENDPATH += $$PWD/
win32: LIBS += C:/MinGW/msys/1.0/home/zakharan/gtest-1.6.0/mybuid/libgtest.a
win32: LIBS += C:/MinGW/msys/1.0/home/zakharan/gtest-1.6.0/mybuid/libgtest_main.a
Google recommends the following for compiling in Windows:
### Windows Requirements ###
* Microsoft Visual C++ 7.1 or newer
### Cygwin Requirements ###
* Cygwin 1.5.25-14 or newer
I've heard of some people using MingW for compiling this. Supposedly compiling with MingW can be a little faster because there doesn't need to be a comparability layer between the OS and the application.
But, since this is just for dev testing purposes Cygwin should suffice. Just download the version that Google recommends.
See Also:
Compile Cygwin project in Eclipse
How Cygwin compares to MingW for porting Linux applications to Windows