error: -mcpu=cortex-a7 conflicts with -march=armv7-a

[james-bay]

ct-ng version: 1.23.0
OS: Kbuntu 18.04 x64

I ran into a error in the build proces:

[INFO ]  Installing final gcc compiler
[EXTRA]    Configuring final gcc compiler
[EXTRA]    Building final gcc compiler
[ERROR]    cc1: error: switch -mcpu=cortex-a7 conflicts with -march=armv7-a switch [-Werror]
[ERROR]    make[6]: *** [load_1_1_.lo] Error 1
[ERROR]    make[5]: *** [all-recursive] Error 1
[ERROR]    make[4]: *** [all] Error 2
[ERROR]    make[3]: *** [all-target-libatomic] Error 2
[ERROR]    make[2]: *** [all] Error 2

I did not change the .config after ct-ng armv7-rpi2-linux-gnueabihf, I just did the following:

~/prms/crosstool-ng$ ./bootstrap
~/prms/crosstool-ng$ ./configure --enable-local
~/prms/crosstool-ng$ ./ct-ng armv7-rpi2-linux-gnueabihf
~/prms/crosstool-ng$ ./ct-ng build.1

• build.log

[bhesmans]

Hey,

I ran into a similar issue, I edited the config file ... I removed CT_ARCH_CPU="cortex-a7" from the config file and add instead CT_ARCH_ARCH="armv7ve" . Now it works. I'm new to crosstool-ng and cross compile in general so it may no be a good solution at all :-P

[stilor]

This seems to be an issue with GCC8: build of libstdc++ now throws in -march=armv7-a unconditionally for ARM/Linux configurations - which conflicts with the --with-cpu=... configuration options. I asked about this on GCC mailing list, haven't got a response so far.

With setting architecture, you still need to verify how libstdc++ has been built - whether your -march=armv7ve came before or after -march=armv7-a. If your option came first, it is overridden by -march=armv7-a.

Until I have a better solution from GCC list, I am going to disallow GCC8 for ARM if any of these options (ARCH_ARCH, ARCH_CPU, ARCH_TUNE) is selected.

[james-bay]

@bhesmans
@stilor
many thanks! Большое спасибо!

[cirosantilli]

Linking to stilor's GCC thread for convenience: https://gcc.gnu.org/ml/gcc-help/2018-05/msg00120.html

Same problem with ./ct-ng arm-cortex_a15-linux-gnueabihf.

[johan13]

This issue appears to have been fixed by #978.

[stilor]

Indeed. Forgot to mention this issue in the commit message.