--disable-gil & --enable-experimental-jit fails to build on M1 Mac

[smontanaro]

Bug report

Bug description:

On my M1 Mac, if I configure like so: ./configure --disable-gil --enable-experimental-jit I get a compilation error:

gcc -c -fno-strict-overflow -Wsign-compare -Wunreachable-code -DNDEBUG -g -O3 -Wall    -D_Py_TIER2=1 -D_Py_JIT -std=c11 -Wextra -Wno-unused-parameter -Wno-missing-field-initializers -Wstrict-prototypes -Werror=implicit-function-declaration -fvisibility=hidden  -I./Include/internal -I./Include/internal/mimalloc  -I. -I./Include    -DPy_BUILD_CORE -o Python/jit.o Python/jit.c
In file included from Python/jit.c:392:
./jit_stencils.h:13419:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16a(code + 0x198, (uintptr_t)&_PyCriticalSection_Resume);

(followed by many more of the same error).

I don't recall if I've tried this configure combination before or not. I'm using Apple's clang on an up-to-date MacOS (14.4.1):

% gcc --version
Apple clang version 15.0.0 (clang-1500.3.9.4)
Target: arm64-apple-darwin23.4.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

• gh-118844: Fix --disable-gil & --enable-experimental-jit fails to build #118935
• [3.13] GH-118844: Fix build failures when combining --disable-gil with --enable-experimental-jit (GH-118935) #118959

[Eclips4]

Full compilation error:

gcc -c -fno-strict-overflow -Wsign-compare -g -Og -Wall    -D_Py_TIER2=1 -D_Py_JIT -std=c11 -Wextra -Wno-unused-parameter -Wno-missing-field-initializers -Wstrict-prototypes -Werror=implicit-function-declaration -fvisibility=hidden  -I./Include/internal -I./Include/internal/mimalloc  -I. -I./Include    -DPy_BUILD_CORE -o Python/jit.o Python/jit.c
In file included from Python/jit.c:392:
./jit_stencils.h:24576:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16a(code + 0x230, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:24577:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16b(code + 0x234, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:24578:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16c(code + 0x238, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:24579:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16d(code + 0x23c, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:24596:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16a(code + 0x2a8, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:24597:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16b(code + 0x2ac, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:24598:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16c(code + 0x2b0, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:24599:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16d(code + 0x2b4, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:45384:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16a(code + 0x1c0, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:45385:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16b(code + 0x1c4, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:45386:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16c(code + 0x1c8, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:45387:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16d(code + 0x1cc, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:45404:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16a(code + 0x238, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:45405:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16b(code + 0x23c, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:45406:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16c(code + 0x240, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:45407:49: error: use of undeclared identifier '_PyCriticalSection_BeginSlow'
    patch_aarch64_16d(code + 0x244, (uintptr_t)&_PyCriticalSection_BeginSlow);
                                                ^
./jit_stencils.h:48060:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16a(code + 0x358, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:48061:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16b(code + 0x35c, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
./jit_stencils.h:48062:49: error: use of undeclared identifier '_PyCriticalSection_Resume'
    patch_aarch64_16c(code + 0x360, (uintptr_t)&_PyCriticalSection_Resume);
                                                ^
fatal error: too many errors emitted, stopping now [-ferror-limit=]
20 errors generated.
make: *** [Python/jit.o] Error 1

[brandtbucher]

Thanks! I’m not sure if this is a build configuration that’s been given any thought yet... or if it even makes sense right now (I believe our tier 2 optimization pipeline is disabled on free-threaded builds, but I’d need to check). Regardless, we’re headed towards a future with both of these features and should probably make sure that they at least build together.

@savannahostrowski, interested in taking this too? I’m on my phone now, but I think we probably just need to #include the header where this is defined in jit.c.

[sobolevn]

Here' the code sample that is generated:

memcpy(data, data_body, sizeof(data_body));
    patch_64(data + 0x0, instruction->oparg);
    memcpy(code, code_body, sizeof(code_body));
    patch_aarch64_33rx(code + 0x30, (uintptr_t)data);
    patch_aarch64_26r(code + 0x16c, (uintptr_t)code + sizeof(code_body));
    patch_aarch64_33rx(code + 0x190, (uintptr_t)data);
    patch_aarch64_26r(code + 0x1c8, instruction_starts[instruction->error_target]);
    patch_aarch64_16a(code + 0x1d0, (uintptr_t)&_PyEval_FormatExcUnbound);
    patch_aarch64_16b(code + 0x1d4, (uintptr_t)&_PyEval_FormatExcUnbound);
    patch_aarch64_16c(code + 0x1d8, (uintptr_t)&_PyEval_FormatExcUnbound);
    patch_aarch64_16d(code + 0x1dc, (uintptr_t)&_PyEval_FormatExcUnbound);
    patch_aarch64_16a(code + 0x1e8, (uintptr_t)&_PyCriticalSection_Resume);
    patch_aarch64_16b(code + 0x1ec, (uintptr_t)&_PyCriticalSection_Resume);
    patch_aarch64_16c(code + 0x1f0, (uintptr_t)&_PyCriticalSection_Resume);
    patch_aarch64_16d(code + 0x1f4, (uintptr_t)&_PyCriticalSection_Resume);
    patch_aarch64_16a(code + 0x200, (uintptr_t)&_PyThreadState_GetCurrent);
    patch_aarch64_16b(code + 0x204, (uintptr_t)&_PyThreadState_GetCurrent);
    patch_aarch64_16c(code + 0x208, (uintptr_t)&_PyThreadState_GetCurrent);
    patch_aarch64_16d(code + 0x20c, (uintptr_t)&_PyThreadState_GetCurrent);

[savannahostrowski]

Yep - feel free to assign to me @brandtbucher! I'll take a look tonight or tomorrow!

[smontanaro]

FWIW, I'm not sure this is M1-specific. It compiled just fine on my Dell laptop running XUbuntu 22.04. OTOH, on an old MacBook Air I have laying around (dual core Intel i5, MacOS 12.7.4), it failed to compile Python/jit.c in the same fashion as on my M1 MBP. Maybe it's a MacOS issue?