Error with valgrind on a code using boost::python - c++

Question

I try to track a memory leak in a code using boost::python.
I reduced my code to a simple main. This simple main run with valgrind without error/leak :
valgrind --leak-check=full --track-origins=yes /HIDDEN_PATH/MemLeak
==23316== Memcheck, a memory error detector
==23316== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.
==23316== Using Valgrind-3.19.0 and LibVEX; rerun with -h for copyright info
==23316== Command: /HIDDEN_PATH/MemLeak
==23316==
==23316==
==23316== HEAP SUMMARY:
==23316== in use at exit: 0 bytes in 0 blocks
==23316== total heap usage: 192 allocs, 192 frees, 148,668 bytes allocated
==23316==
==23316== All heap blocks were freed -- no leaks are possible
==23316==
==23316== For lists of detected and suppressed errors, rerun with: -s
==23316== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
When I add the first call to something using boost::python, valgrind fails to execute the code (the error is before the main):
> valgrind --leak-check=full --track-origins=yes /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak
==23595== Memcheck, a memory error detector
==23595== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.
==23595== Using Valgrind-3.19.0 and LibVEX; rerun with -h for copyright info
==23595== Command: /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak
==23595==
vex amd64->IR: unhandled instruction bytes: 0x62 0xF1 0xFD 0x8 0x6F 0x4 0x24 0x48 0x89 0x68
vex amd64->IR: REX=0 REX.W=0 REX.R=0 REX.X=0 REX.B=0
vex amd64->IR: VEX=0 VEX.L=0 VEX.nVVVV=0x0 ESC=NONE
vex amd64->IR: PFX.66=0 PFX.F2=0 PFX.F3=0
==23595== valgrind: Unrecognised instruction at address 0x61ac14.
==23595== at 0x61AC14: boost::python::converter::registry::insert(void* (*)(_object*), void (*)(_object*, boost::python::converter::rvalue_from_python_stage1_data*), boost::python::type_info, _typeobject const* (*)()) (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x628753: boost::python::converter::initialize_builtin_converters() (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x61A55B: boost::python::converter::(anonymous namespace)::get(boost::python::type_info, bool) (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x528FCD: registry_lookup2<const volatile std::__cxx11::basic_string<char> > (registered.hpp:100)
==23595== by 0x528FCD: registry_lookup1<const volatile std::__cxx11::basic_string<char>&> (registered.hpp:107)
==23595== by 0x528FCD: __static_initialization_and_destruction_0 (registered.hpp:118)
==23595== by 0x528FCD: _GLOBAL__sub_I__ZN6minamo22PyMod_VirtualEvaluatorC2Ev (PyMod_VirtualEvaluator.cpp:188)
==23595== by 0x8A61AC: __libc_csu_init (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x68E1C2D: (below main) (libc-start.c:270)
==23595== Your program just tried to execute an instruction that Valgrind
==23595== did not recognise. There are two possible reasons for this.
==23595== 1. Your program has a bug and erroneously jumped to a non-code
==23595== location. If you are running Memcheck and you just saw a
==23595== warning about a bad jump, it's probably your program's fault.
==23595== 2. The instruction is legitimate but Valgrind doesn't handle it,
==23595== i.e. it's Valgrind's fault. If you think this is the case or
==23595== you are not sure, please let us know and we'll try to fix it.
==23595== Either way, Valgrind will now raise a SIGILL signal which will
==23595== probably kill your program.
==23595==
==23595== Process terminating with default action of signal 4 (SIGILL)
==23595== Illegal opcode at address 0x61AC14
==23595== at 0x61AC14: boost::python::converter::registry::insert(void* (*)(_object*), void (*)(_object*, boost::python::converter::rvalue_from_python_stage1_data*), boost::python::type_info, _typeobject const* (*)()) (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x628753: boost::python::converter::initialize_builtin_converters() (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x61A55B: boost::python::converter::(anonymous namespace)::get(boost::python::type_info, bool) (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x528FCD: registry_lookup2<const volatile std::__cxx11::basic_string<char> > (registered.hpp:100)
==23595== by 0x528FCD: registry_lookup1<const volatile std::__cxx11::basic_string<char>&> (registered.hpp:107)
==23595== by 0x528FCD: __static_initialization_and_destruction_0 (registered.hpp:118)
==23595== by 0x528FCD: _GLOBAL__sub_I__ZN6minamo22PyMod_VirtualEvaluatorC2Ev (PyMod_VirtualEvaluator.cpp:188)
==23595== by 0x8A61AC: __libc_csu_init (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x68E1C2D: (below main) (libc-start.c:270)
==23595==
==23595== HEAP SUMMARY:
==23595== in use at exit: 86,344 bytes in 24 blocks
==23595== total heap usage: 24 allocs, 0 frees, 86,344 bytes allocated
==23595==
==23595== 32 bytes in 1 blocks are definitely lost in loss record 8 of 24
==23595== at 0x4C378C3: operator new(unsigned long) (vg_replace_malloc.c:422)
==23595== by 0x61AC13: boost::python::converter::registry::insert(void* (*)(_object*), void (*)(_object*, boost::python::converter::rvalue_from_python_stage1_data*), boost::python::type_info, _typeobject const* (*)()) (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x628753: boost::python::converter::initialize_builtin_converters() (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x61A55B: boost::python::converter::(anonymous namespace)::get(boost::python::type_info, bool) (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x528FCD: registry_lookup2<const volatile std::__cxx11::basic_string<char> > (registered.hpp:100)
==23595== by 0x528FCD: registry_lookup1<const volatile std::__cxx11::basic_string<char>&> (registered.hpp:107)
==23595== by 0x528FCD: __static_initialization_and_destruction_0 (registered.hpp:118)
==23595== by 0x528FCD: _GLOBAL__sub_I__ZN6minamo22PyMod_VirtualEvaluatorC2Ev (PyMod_VirtualEvaluator.cpp:188)
==23595== by 0x8A61AC: __libc_csu_init (in /HIDDEN_PATH/BUILD_DEBUG/test/experimentation/MemLeak)
==23595== by 0x68E1C2D: (below main) (libc-start.c:270)
==23595==
==23595== LEAK SUMMARY:
==23595== definitely lost: 32 bytes in 1 blocks
==23595== indirectly lost: 0 bytes in 0 blocks
==23595== possibly lost: 0 bytes in 0 blocks
==23595== still reachable: 86,312 bytes in 23 blocks
==23595== suppressed: 0 bytes in 0 blocks
==23595== Reachable blocks (those to which a pointer was found) are not shown.
==23595== To see them, rerun with: --leak-check=full --show-leak-kinds=all
==23595==
==23595== For lists of detected and suppressed errors, rerun with: -s
==23595== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
Illegal instruction (core dumped)
I don't understand what is wrong.
Configuration: CentOS Stream 8, GCC 8.5.0, valgrind 3.19.0, python 3.9.10 static fPIC, boost 1.71.0 static fPIC
Note: I tried with option --suppressions=valgrind-python.supp but no difference.