On 07/27/2013 02:20 PM, Sundareson, Prabindh wrote:
The device approach is useful for adding platform specific hooks for
eglfs via EGLFS_PLATFORM_HOOKS_SOURCES, so it is a good thing. So it is
more of a convenience/cleaner way, than anything related to performance.
I do agree with Dennys about the compiler flags, yocto knows what are
the best flags for the machine/device.
What is nice with the device file approach is the specification for
OpenGL includes, libs, QPA and the EGLFS hooks as well.
Basically I like the fact that all the OpenGL "mess" is sorted out in
one place (when it comes to compilation).
Interesting examples are the Rasberry Pi [1] and Freescale i.mx6 [2].
when considering Qt alone (out of Yocto/OE), I think as a SoC vendor, it
is the best way to make sure your users get the best of Qt5/OpenGL, and
very easily according to [3] (not tested).
When it comes to Yocto, I didn't see any easy way for meta-bsp to tell
meta-qt5 how to configure/build qt5 w/ OpenGL support.
It seems to me that the only way is through the .bbappend, in qtbase.inc:
[...]
# if you want to enable egl/gles2 in your .bbappend, don't forget to add
right dependencies, e.g.
# GL_DEPENDS = "virtual/libgles2 virtual/egl"
# QT_GLFLAGS = "-opengl es2 -eglfs"
GL_DEPENDS ?= "virtual/libgl"
QT_GLFLAGS ?= "-opengl"
[...]
Could/should this be used by the meta-ti layer?
I spent some time recently going through meta-qt5 recently (as well as
the qt4 stuff), I even tried an alternative configure/method, so far
unsuccessfully. My feeling is that meta-qt5 tries to adapt qt5 to OE, I
think it should be done the other way around, that's my very personal view.
I really would like to see a Yocto Qt5 SDK one of these days.
Prabindh, could you tell more about the status of Qt5/OpenGL for TI
SoCs, what about this commit [4]? Is it possible to build within Yocto
Qt5 w/ OpenGL support for AM335x platforms?
Regards,
Chris
[1]
http://qt.gitorious.org/qt/qtbase/trees/stable/mkspecs/devices/linux-rasp-pi-g++
[2]
http://qt.gitorious.org/qt/qtbase/trees/stable/mkspecs/devices/linux-imx6-g++
[3]
http://blog.qt.digia.com/blog/2012/04/13/cross-compiling-qt-for-the-masses/
[4]
https://github.com/prabindh/meta-qt5/commit/c8129fb5057f9254c773f82d2fdbd85d9c56410a
regards
Prabindh
________________________________________
From: [email protected] [[email protected]] on
behalf of Dmytriyenko, Denys
Sent: Saturday, July 27, 2013 2:36 AM
To: Christian Gagneraud
Cc: [email protected]
Subject: Re: [meta-ti] [Qt5] mkspecs/devices: beagleboard == am335x ?
On Fri, Jul 26, 2013 at 10:45:02AM +1200, Christian Gagneraud wrote:
Hi there,
In qtbase sources, there are a couple of "device" definitions files,
according to [1], this greatly help cross-compiling qt5.
This device file specify things for CPU and GPU optimisations. So
basically these files are tightly coupled with BSP providers.
One of them is linux-beagleboard-g++ [2], is it appropriate to use
this one for AM335x based platform as well?
The meta-qt5 layer doesn't seem to use this feature, but if it did,
would this device config files help to get an optimised Qt5 build
for TI SoCs? Especially, would that help to get OpenGL stuff sorted
out?
Regards,
Chris
[1] http://blog.qt.digia.com/blog/2012/04/13/cross-compiling-qt-for-the-masses/
[2]
https://qt.gitorious.org/qt/qtbase/trees/stable/mkspecs/devices/linux-beagleboard-g++
Chris,
Looking at those files shows very basic compiler optimization flags being set,
-march, -mtune, -mfpu etc. I don't see any GL or GPU specifics in there. It
also hardcodes things like soft-fp, which is not good. All this stuff is
alredy taken care of by the build framework and I'm sure meta-qt5, like the
old qt4, sets all the right COMPILER_FLAGS to get the most performance from
the target CPU.
--
Denys
_______________________________________________
meta-ti mailing list
[email protected]
https://lists.yoctoproject.org/listinfo/meta-ti
_______________________________________________
meta-ti mailing list
[email protected]
https://lists.yoctoproject.org/listinfo/meta-ti
