[cfe-users] Cross compiling a bare metal Cortex M4 Application with LTO

[Dan Čermák via cfe-users]

Hello everyone,

I am developing bare metal applications on a Cortex M4 CPU mostly with
gcc but wanted to also use clang, as I do for all my programs targeting
an OS-environment.

After a lot of fiddling around I finally managed to get clang to
successfully compile & link my projects. I have achieved that by using
my ordinary cross compiler (arm-none-eabi-gcc):
- provide clang with all the include paths of arm-none-eabi-gcc
- setting --sysroot to the installation root of arm-none-eabi-gcc
- set the gcc name using the option '-ccc-gcc-name' to
  arm-none-eabi-gcc, otherwise clang would try to link the application
  using the ordinary gcc (which obviously won't work)

This works pretty well, but has one big downside: link time optimization
does not work, as gcc is linking the object files and not clang. (And it
has a "cosmetic" downside of relying on the gcc cross compiler.)

Does anyone know how I could setup a "pure" clang build environment,
that would not have to rely on the gcc cross compiler? I guess I would
have to also build the libc for arm-none-eabi for that and tell clang
about the ARM headers, but I have little experience with that (and I did
not find a lot of useful information as most embedded Devs seem to use
the vendor supplied IDEs with the bundled compilers on Windows instead).


Thanks in advance,

Dan