error in building 3.13.2 for arm free-threaded with jit

[eyalk11]

Bug report

Bug description:

CONFIGURE_OPTS="--disable-gil --enable-experimental-jit" pyenv install 3.13.2

Results logged to /tmp/python-build.20250304140853.1135365.log

Last 10 log lines:
./jit_stencils.h: In function ‘emit__UNPACK_SEQUENCE_TUPLE’:
./jit_stencils.h:36885:49: error: ‘__aarch64_ldadd8_acq_rel’ undeclared (first use in this function)
36885 |     patch_aarch64_16a(code + 0x148, (uintptr_t)&__aarch64_ldadd8_acq_rel);
      |                                                 ^~~~~~~~~~~~~~~~~~~~~~~~
./jit_stencils.h: In function ‘emit__UNPACK_SEQUENCE_TWO_TUPLE’:
./jit_stencils.h:37151:49: error: ‘__aarch64_ldadd8_acq_rel’ undeclared (first use in this function)
37151 |     patch_aarch64_16a(code + 0x1d8, (uintptr_t)&__aarch64_ldadd8_acq_rel);
      |                                                 ^~~~~~~~~~~~~~~~~~~~~~~~
make: *** [Makefile:3009: Python/jit.o] Error 1

CPython versions tested on:

3.13

Operating systems tested on:

Linux

Linked PRs

• [3.13] gh-130834: Fix free-threaded build with JIT for aarch64 linux #132368

[corona10]

@eyalk11 I want to check that passing -mno-outline-atomics to CFLAGS resolve your issue.
IIRC, you can pass the custom CFLAGS to pyenv.
cc @diegorusso

[diegorusso]

@corona10 is right. I've just checked and in 3.13.2 the flag is not set. https://github.com/python/cpython/blob/v3.13.2/Tools/jit/_targets.py#L514

You can try installing the libatomic library.

sudo apt-get install libatomic1

CFLAGS are not passed down to the compilation of the JIT stencils. This because we don't use Makefile to compile the JIT stencils. Also because the compiler needed for compiling CPython might be different from clang (which is mandatory for the JIT stencils).
So if you want to pass that flag you need to modify the _target.py file and pass the flag.

Anyway we do this in 3.14: https://github.com/python/cpython/blob/main/Tools/jit/_targets.py#L524 and I'm wondering if it is worth to back port this change. @brandtbucher opinions?

[corona10]

Hmm, I think that there is no reason not to backport it since we triage it as a bug.
Any ideas? @savannahostrowski @brandtbucher ?

[brandtbucher]

Yeah, I'm fine with backporting it. On my phone now, is anyone else able to do it?

[corona10]

Patch sent: #132368

[corona10]

@eyalk11 I think that we can close the issue, please let us know if this is still an issue at the latest 3.13 branch or the next release.