[SOLVED] QT5 on Tegra2, crosscompiling problem

[bricke]

Hi all, I'm Matteo from Italy, I'm finding some problem crosscompiling Qt5 for my Colibri T20 device (Nvidia Tegra 2).

I've the toolchain from the vendor (Toradex) and also the rootfs.
I've downloaded the stable version of Qt5 from git and now I'm compiling with the following command:

@./configure -opengl es2 -device tegra2 -device-option CROSS_COMPILE=~/QT5/compiler/gcc-linaro-arm-linux-gnueabihf-2012.09-20120921_linux/bin/arm-linux-gnueabihf- -sysroot ~/QT5/rootfs/ -opensource -confirm-license -optimized-qmake -release -make libs -prefix /usr/local/qt5pi -v@

But my code stops with the folowing error:
@Note: PKG_CONFIG_SYSROOT_DIR automatically set to /home/matteo/QT5/rootfs/
Determining architecture... ()
/home/matteo/QT5/compiler/gcc-linaro-arm-linux-gnueabihf-2012.09-20120921_linux/bin/arm-linux-gnueabihf-g++ -c -pipe -mtune=cortex-a9 -march=armv7-a -mhard-float -mfloat-abi=softfp -mfpu=vfpv3-d16 -mfloat-abi=softfp --sysroot=/home/matteo/QT5/rootfs/ -g -Wall -W -fPIE -I../../mkspecs/devices/linux-tegra2-g++ -I. -I/home/matteo/QT5/rootfs/usr/include -o arch.o arch.cpp
arch.cpp:45:19: fatal error: stdio.h: No such file or directory
compilation terminated.
make: *** [arch.o] Error 1
Unable to determine architecture!

Could not determine the target architecture!
Turn on verbose messaging (-v) to see the final report.
Determining architecture... ()
g++ -c -pipe -g -Wall -W -fPIE -I../../mkspecs/linux-g++ -I. -o arch.o arch.cpp
g++ -o arch arch.o
{ test -n "" && DESTDIR="" || DESTDIR=.; } && test $(gdb --version | sed -e 's,[^0-9]+([0-9]).([0-9]).*,\1\2,;q') -gt 72 && gdb --nx --batch --quiet -ex 'set confirm off' -ex "save gdb-index $DESTDIR" -ex quit 'arch' && test -f arch.gdb-index && objcopy --add-section '.gdb_index=arch.gdb-index' --set-section-flags '.gdb_index=readonly' 'arch' 'arch' && rm -f arch.gdb-index || true
Found architecture in binary
CFG_HOST_ARCH="i386"
CFG_HOST_CPUFEATURES=""
System architecture: 'unknown'
Host architecture: 'i386'
C++11 auto-detection... ()
...
@

It seems that my system don't found stdio.h, but I've all the relative packet for compiling installed and if I launch a locate stdio.h in the result I've also:
@/usr/include/stdio.h
/usr/include/bits/stdio.h
/usr/include/c++/4.6/tr1/stdio.h
/usr/include/i386-linux-gnu/bits/stdio.h
/usr/lib/perl/5.14.2/CORE/nostdio.h
/usr/lib/syslinux/com32/include/stdio.h@

Do you have some hints? I've searched a lot but it seems that everyone could compile for tegra2 but me :(.
I've also an openembedded env on my machine, so if you have some recipes for qt5 I can use it.

Thanks.

[zhxt]

'stdio.h's in /usr/include are for host, not for target board. the one which you need should be in toolchain directory.

So I think you can check your toolchain first, try another one for test.

[bricke]

So, how can I tell the crosscompiler to use the one in my toolchain dir?

[bricke]

I've managed this error, I've linked in the configure options the right sysroot from my OE environment but now the config command give me other errors:

error: arch uses VFP register arguments, arch.o does not

Is the linux-tegra2-g++/qmake.conf in mkspeks/device folder up to date?

[zhxt]

I'm also not sure if there is a new version qmake.conf for T2.

I guess some of this TEGRA2_CFLAGS flags caused this problem.
but I do not know which one is.

-mhard-float -mfloat-abi=softfp -mfpu=vfpv3-d16

[bricke]

I've made a step, my configure settings are:
@
./configure -make libs
-xplatform linux-tegra2-g++
-platform linux-g++-32
-prefix /home/matteo/oe-core/build/out-eglibc/sysroots/colibri-t20/opt
-sysroot /home/matteo/oe-core/sysroot
-confirm-license
-opensource -v
@

I've created a linux-tegra2-gxx folder under mkspecs and I've populated it as follow:
@
include(../devices/common/linux_device_pre.conf)

modifications to g++.conf

QMAKE_CC = arm-angstrom-linux-gnueabi-gcc
QMAKE_CXX = arm-angstrom-linux-gnueabi-g++
QMAKE_LINK = arm-angstrom-linux-gnueabi-g++
QMAKE_LINK_SHLIB = arm-angstrom-linux-gnueabi-g++

modifications to linux.conf

QMAKE_AR = arm-angstrom-linux-gnueabi-ar cqs
QMAKE_OBJCOPY = arm-angstrom-linux-gnueabi-objcopy
QMAKE_STRIP = arm-angstrom-linux-gnueabi-strip

QMAKE_INCDIR += $$[QT_SYSROOT]/usr/include

QMAKE_LIBDIR += $$[QT_SYSROOT]/usr/lib
$$[QT_SYSROOT]/lib
$$[QT_SYSROOT]/usr/lib/arm-angstrom-linux-gnueabi

QMAKE_LFLAGS += -Wl,-rpath-link,$$[QT_SYSROOT]/usr/lib
-Wl,-rpath-link,$$[QT_SYSROOT]/usr/lib/arm-angstrom-linux-gnueabi
-Wl,-rpath-link,$$[QT_SYSROOT]/lib

TEGRA2_CFLAGS = -mtune=cortex-a9 -march=armv7-a -mhard-float -mfloat-abi=softfp -mfpu=vfpv3-d16
QMAKE_CFLAGS += $$TEGRA2_CFLAGS
QMAKE_CXXFLAGS += $$TEGRA2_CFLAGS

include(../devices/common/linux_device_post.conf)

load(qt_config)
@

arm-angstrom-linux-gnueabi- is on my path

But when I launch the command it exits with the Error: "Unable to determine the architecture", but the strange things is that the first line in the console after executing the command is:
@cc1plus: warning: include location "/usr/include/i386-linux-gnu" is unsafe for cross-compilation [-Wpoison-system-directories]@

Any ideas? It's seems that my system uses the host machine files instead of the tegra2 for configuring. I think that the final error is due to this wrong configuration, but I don't know how to eliminate it.

[zhxt]

The makespec for Tegra2 is in the same dir as for RaspPi. So I think a easy way is starting from build instructions for pi.
just simply change the device option to linux-tegra2-g++ (gussed), set the proper toolchain path, sysroot path and prefix.

@
$ ./configure -prefix <your prefix> -release
-device linux-tegra2-g++ -make libs
-device-option CROSS_COMPILE=<your toolchain path>/bin/arm-none-linux-gnueabi-
-sysroot <your sysroot path> -v
@

of course you also can append some options like -opensoure -confirm-license which is unconcerned with these ERRORs.

And see what will happed.

[bricke]

If you read my previous post you can note that I've already done some test in that way.

The -device option for tegra is tegra2, as already written, not linux-tegra2-g++, and one week ago I've started this "adventure" from the Rasp guide, but unfortunately it's not that easy.

I'm thinking that my toolchain is broken or my env, when building the errors comes with the host libraries that my env caches in "/usr/include/i386-linux-gnu" and I don't know why.

[zhxt]

1, My guess about linux-tegra2-g++ have no effect with this errors, cause it's the same as tegra2. (the pattern is: "* tegra2*" ).

2, As my first reply, I doubt the toolchain, because mine has these missing headers, So it can found automatically. If it does not have, use -sysroot to override the default searching path should work, if -sysroot is not set, it will goes to host path.

Because when the variable QT_SYSROOT is blank.
@QMAKE_INCDIR += $$[QT_SYSROOT]/usr/include@

equal to

@QMAKE_INCDIR += /usr/include@

I'm not exactly sure, but it make sense.

3, Continue digging my second reply. TRY TO:

add "DISTRO_OPTS += hard-float" and change the value of "-mfloat-abi" to "hard" to the default qmake.conf in mkspec/device/linux-tegra2-g++

@
DISTRO_OPTS += hard-float
TEGRA2_CFLAGS = -mtune=cortex-a9 -march=armv7-a -mhard-float -mfloat-abi=hard -mfpu=vfpv3-d1
@

Then use the first configuration(rpi-like) to build.

BTW, attention to the sentence bellow in that qmake.conf:
CAUTION: The Tegra2 chips don't support neon, make sure it will not be used (-no-neon)

[bricke]

Thanks for the tip, I will try it later.

By now I've successfully terminated the configure script with some hack (I've bypassed the /usr/include/i386-linux-gnu folder by linking the arm folder in the same place)

@
Debug .................. no
C++11 support .......... yes
pkg-config ............. yes
Qt D-Bus module ........ yes (run-time)
Qt Concurrent code ..... yes
Qt GUI module .......... yes
Qt Widgets module ...... yes
JavaScriptCore JIT ..... To be decided by JavaScriptCore
QML debugging .......... yes
PCH support ............ no
iWMMXt support ......... no
NEON support ........... no
IPv6 ifname support .... yes
getaddrinfo support .... yes
getifaddrs support ..... yes
Accessibility .......... yes
NIS support ............ yes
CUPS support ........... no
Iconv support .......... yes
Glib support ........... yes
GStreamer support ...... yes
PulseAudio support ..... yes
Large File support ..... yes
GIF support ............ plugin
JPEG support ........... plugin (system)
PNG support ............ yes (system)
zlib support ........... system
Session management ..... auto
libudev support ........ yes
Use system proxies ..... no
OpenGL support ......... yes (OpenGL ES 2.x)
OpenVG support ......... no
XShape support ......... auto
XVideo support ......... auto
XSync support .......... auto
Xinerama support ....... runtime
Xcursor support ........ runtime
Xfixes support ......... runtime
Xrandr support ......... runtime
Xi support ............. no
Xi2 support ............ yes
MIT-SHM support ........ auto
FontConfig support ..... yes
XKB Support ............ auto
GTK theme support ...... yes
MySQL support .......... plugin
PostgreSQL support ..... plugin
SQLite 2 support ....... plugin
SQLite support ......... plugin (qt)
OpenSSL support ........ yes (run-time)
Alsa support ........... yes
libICU support ......... no
PCRE support ........... qt
Xcb support ............ qt
Xrender support ........ yes
EGLFS support .......... yes
DirectFB support ....... no
LinuxFB support ........ yes
KMS support ............ no
@

but I've used the -platform and -xplatform options instead of the -device one.

However, I cannod understand if I will be able to run GUI application with hardware acceleration, I need to run gui app on FB, but I can only see LinuxFB support...

[zhxt]

LinuxFB is not FB?

[bricke]

I don't know, I read:
DirectFB support no
LinuxFB support yes

So I'm having some doubt...

[zhxt]

I think yes, Linux FB is FB, but DirectFB is on top of FB. http://qt-project.org/doc/qt-4.8/qt-embeddedlinux-accel.html

If you want to use hardware acceleration, May be you also need DirectFB or OpenGL ES | Open VG. learnd from the above link.

[bricke]

I have the CPLUS_INCLUDE_PATH and C_INCLUDE_PATH variable setted to /usr/include/i386-linux-gnu that gives me a lot of trouble!

Unsetting the CPLUS one let me terminate my configure correctly, and then my make command blocked on the same dir, so I've insetted the C_INCLUDE_PATH and now it's compiling...

Now I'm able to terminate the configure script with:
@./configure -make libs
-prefix /opt/qt5_teo
-sysroot /home/matteo/oe-core/sysroot
-device tegra2
-device-option CROSS_COMPILE=/home/matteo/matteo_test/gcc-linaro/bin/arm-linux-gnueabihf-
-no-neon
-opengl es2
-qt-xcb
-confirm-license
-nomake examples -nomake tests
-optimized-qmake
-reduce-relocations
-release
-opensource -v
@

and the tegra2 device qmake as follow:
@
include(../common/linux_device_pre.conf)

QMAKE_INCDIR += $$[QT_SYSROOT]/usr/include

QMAKE_LIBDIR += $$[QT_SYSROOT]/usr/lib
$$[QT_SYSROOT]/lib
$$[QT_SYSROOT]/usr/lib/arm-angstrom-linux-gnueabi/4.7.2

QMAKE_LFLAGS += -Wl,-rpath-link,$$[QT_SYSROOT]/usr/lib
-Wl,-rpath-link,$$[QT_SYSROOT]/usr/lib/arm-angstrom-linux-gnueabi/4.7.2
-Wl,-rpath-link,$$[QT_SYSROOT]/lib

DISTRO_OPTS += hard-float
TEGRA2_CFLAGS = -mtune=cortex-a9 -march=armv7-a -mhard-float -mfloat-abi=hard -mfpu=vfpv3-d16
QMAKE_CFLAGS += $$TEGRA2_CFLAGS
QMAKE_CXXFLAGS += $$TEGRA2_CFLAGS

include(../common/linux_device_post.conf)

load(qt_config)
@

[RhnBlk]

I'm using qt 4.8.4 for toradex colibri_t20. Configuring it was not a problem, but I'm facing issues in executing programs. U can find the post here "://http://qt-project.org/forums/viewthread/29274/"