[Bug]: SIGILL (Illegal Instruction) during import on x86_64 (AVX-512BW `vmovdqu8` requirement)

[gavmor]

Description

The zvec Python package (v0.1.0 - v0.2.0) crashes with an Illegal instruction (core dumped) error immediately upon import on certain x86_64 systems. Debugging reveals that the _zvec shared object contains instructions from the AVX-512BW subset (specifically vmovdqu8) which are not supported by many CPUs that otherwise support base AVX-512 or AVX2.

System Information:

• OS: Linux (Ubuntu 24.04)
• Architecture: x86_64
• Python Version: 3.11.10
• Zvec Version: 0.2.0 (also reproduced on 0.1.1 and 0.1.0)
• CPU Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es ibpb_exit_to_user
• Missing Instructions: Notably lacks avx512bw.

Suggested Fix:
The current wheels seem to be compiled with -march=skylake-avx512 or similar aggressive optimizations. Providing a generic x86-64-v3 (AVX2) wheel or ensuring the build pipeline targets a broader set of x86_64 features would resolve this for users on older or mid-range hardware.

Steps to Reproduce

1. Install `zvec` via pip: `pip install zvec`
2. Attempt to import: `python -c "import zvec"`
3. Result: `Illegal instruction (core dumped)`

Logs / Stack Trace

Program received signal SIGILL, Illegal instruction.
   0x00007fffc65e33a7 in ?? () from .venv/lib/python3.11/site-packages/_zvec.cpython-311-x86_64-linux-gnu.so

   (gdb) x/i $rip
   => 0x7fffc65e33a7:      vmovdqu8 %xmm0,0x28(%r15)

   (gdb) bt
   #0  0x00007fffc65e33a7 in ?? () from .venv/lib/python3.11/site-packages/_zvec.cpython-311-x86_64-linux-gnu.so
   #1  0x00007fffc65e380f in ?? () from .venv/lib/python3.11/site-packages/_zvec.cpython-311-x86_64-linux-gnu.so
   #2  0x00007fffc65e3a5a in ?? () from .venv/lib/python3.11/site-packages/_zvec.cpython-311-x86_64-linux-gnu.so
   #3  0x00007fffc655804d in ?? () from .venv/lib/python3.11/site-packages/_zvec.cpython-311-x86_64-linux-gnu.so
   #4  0x00007ffff7fca71f in call_init (...) at ./elf/dl-init.c:74
   ...
   #11 0x00007ffff7fd2164 in _dl_open (file="/.../site-packages/_zvec.cpython-311-x86_64-linux-gnu.so", ...)

Operating System

Ubuntu 24.04.4 LTS (Noble Numbat) Kernel: Linux 6.8.0-52-generic #53-Ubuntu SMP PREEMPT_DYNAMIC Architecture: x86_64

Build & Runtime Environment

Runtime: Python 3.11.10    Package Version: zvec==0.2.0 (reproducible on 0.1.1 and 0.1.0)    Processor: x86_64    CPU Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni    pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext    perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero    irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es    ibpb_exit_to_user    Note: The CPU supports AVX and AVX2 but lacks the AVX-512BW extension required for the 'vmovdqu8' instruction present in the wheel.

Additional Context

• I've checked git status — no uncommitted submodule changes
• I built with CMAKE_BUILD_TYPE=Debug
• This occurs with or without COVERAGE=ON
• The issue involves Python ↔ C++ integration (pybind11)

[richyreachy]

Thanks @gavmor. It occurs in zvec class factory registration when loading class according to specific name. The standard library, such as map, will inject SIMD codes when doing string comparison.

Since the resolution is in evaluation (and a traditional vacation in the mean time), a build-from-source way could be short-term workaround described as: https://zvec.org/en/docs/build/