[libcxx]: std::atomic for cortex-m3 - ARMv7-M

Clang Frontend
#1

Hi there,

I'm trying to use std::atomic<> for a cortex-m3 target (ARMv7-M Thumb-2),
but it seems not to be supported (It compiles OK on my x86_64 host).

error: cannot compile this atomic library call yet
        {return __c11_atomic_fetch_sub(&this->__a_, __op, __m);}
                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I wonder where I can find more information about this subject.
Is anyone working on adding support for this? Or is it even possible
since, from what I know, ARMv7-M provide special instruction that only
works on special memory area (bit-band)?

Thanks in advance.

Regards,
Salvatore

#2

After a bit of investigation I figured that the ARMv7-M also provide the
LDREX/STREX for exclusive memory access.

I've also noticed that by generating the intermediate assembly code
by using as target my host architecture (x86_64), and then compiling
the intermediate
assembly code with llc with -mcpu=cortex-m3 (thumb2) as target, it compiles
fine to assembly code, and it uses the LDREX/STREX instructions,

I'm not that familiar with neither LLVM IR assembly, nor the ARMv7-M instruction
set, so if some expert could jump in I'd really appreciate it. :slight_smile:

Regards,
Salvatore

#3

There's a FIXME in lib/Basic/Targets.cpp "// FIXME: Set
MaxAtomicInlineWidth if we have the feature v6e".

-Eli