[Qbs] How to pass -march=armv7-a to the g++ linking command line (without -Wl, )

[Christian Gagneraud]

On 16/01/2018 4:23 AM, Wookey wrote:
> On 2018-01-15 15:13 +0100, Ola Røer Thorsen wrote:
>> I've got a product running on Linux arm imx6. It's being cross-compiled on
>> Linux x86_64. I've got it set up with both qmake and now qbs that hopefully
>> should replace qmake at some point.
>>
>> The qmake build ends up linking the application with a command line like this:
>>
>> arm-fslc-linux-gnueabi-g++
>>          -mfloat-abi=hard
> 
>> I'm struggling to find the property or similar in the cpp module that adds the
>> "-mfloat-abi=hard"-option to g++ when used for linking.
> 
> It's not just linking, its an ABI option that applies to the compile
> stage too. ABIs are so fundamental that these are two different
> architectures in Debian-based distros: armel and armhf. Which have
> different toolchains, and cross-toolchains, so if you use the right
> triplet then you get the right output and everything matches up.
> 
> I would question why you are using a linux-gnueabi toolchain when you
> want linux-gnueabihf outputs. Obviously it can be done, but a good way
> to avoid problems with the wrong ABI is to use a toolchain that
> defaults to the right thing. I don't know what your base distro is,
> but if it's debian-based then using a the distro-supplied
> arm-linux-gnueabihf-g++ would avoid this problem completely. If you
> are in redhat-world I presume there are equivalent cross-toolchains.
> 
> Keeping ABI info in triplets is much cleaner than special-casing the
> magic option for getting one ABI whilst building with a toolchain for
> another, especially in upstream projects. (this includes -m32/-m64 in
> x86 world, which is also a special case that doesn't necessarily work
> on other architectures and architecture-pairs).
> 
> I reaslise this is a rather tangential answer to your actual question,
> but it is hopefully helpful.

I found this piece of information very interesting indeed, what's your 
stance about -march then? In my case, I have one toolchain per platform 
(roughly), they all have the same generic triplet (arm-linux-gnueabi), 
but target different ARM arch (armv5te, armv6, cortex A?, ...) and 
actually have sometimes different float ABI.
I'm currently trying to refactor that to have more consistency across 
our toolchains, so your above comments caught my attention.

Chris

PS: Glad to see you on this ML, I remember you from the Emdebian project 
  when I gave it a try, that was 10 years ago! ;)



> 
> Wookey
> 
> 
> 
> _______________________________________________
> Qbs mailing list
> Qbs at qt-project.org
> http://lists.qt-project.org/mailman/listinfo/qbs
>