bwa-mem2.avx2 crashes (SIGSEGV) on Intel Core Ultra (Arrow Lake) under WSL2 — workaround: force SSE4.2 binary

[pn-pierre-joubert]

Description of the bug

BWAMEM2_MEM consistently crashes mid-run with exit code 1. The last line of .command.err is always:

samtools sort: truncated file. Aborting

bwa-mem2 itself exits silently without any error message. dmesg reveals the true cause — repeated SIGSEGV in bwa-mem2.avx2 at a consistent code offset:

bwa-mem2.avx2[...]: segfault at ... ip 00005eb590639d46 sp ... error 4 in bwa-mem2.avx2[5eb59061d000+2d000] likely on CPU 5 (core 5, socket 0)
bwa-mem2.avx2: bwa-mem2.avx2: potentially unexpected fatal signal 11.
WSL (... - CaptureCrash): Capturing crash for pid: ..., executable: !opt!conda!bin!bwa-mem2.avx2, signal: 11

This crash was reproduced across multiple runs, always at the same binary offset (~0x1cd46), on different CPU cores, ruling out transient hardware faults.

Root cause: The bwa-mem2 launcher detects AVX2 support and exec's bwa-mem2.avx2. On Intel Arrow Lake (Core Ultra 200S series), the P-cores (Lion Cove) have a significantly redesigned AVX2 microarchitecture versus previous generations. The bioconda-built bwa-mem2.avx2 binary appears to trigger an illegal memory access on this architecture, particularly under WSL2's Hyper-V virtualization. This is not an OOM issue — dmesg shows no OOM events at this CPU count, and memory usage remained well within limits.

---

Workaround

Build a custom Docker image that bypasses the auto-detecting launcher and calls bwa-mem2.sse42 directly:

FROM community.wave.seqera.io/library/bwa-mem2_htslib_samtools:e1f420694f8e42bd

RUN mv /opt/conda/bin/bwa-mem2 /opt/conda/bin/bwa-mem2.launcher \
    && printf '#!/bin/sh\nexec /opt/conda/bin/bwa-mem2.sse42 "$@"\n' > /opt/conda/bin/bwa-mem2 \
    && chmod +x /opt/conda/bin/bwa-mem2

Then override the container and tag it so Nextflow resolves it locally (given docker.registry = 'quay.io' in the pipeline config):

// custom.config
process {
    withName: 'BWAMEM2_MEM' {
        container = 'bwa-mem2_htslib_samtools_sse42'
    }
}

docker build -t bwa-mem2_htslib_samtools_sse42 .
docker tag bwa-mem2_htslib_samtools_sse42 quay.io/bwa-mem2_htslib_samtools_sse42
nextflow run nf-core/sarek ... -c custom.config

---

Suggested improvements

1. Document this as a known incompatibility for Arrow Lake CPUs and/or WSL2 environments in the troubleshooting guide.
2. Consider exposing a parameter (e.g. params.bwa_mem2_simd) or a config example that allows users to substitute a custom bwa-mem2 container without patching the pipeline — the current container override via withName works but requires knowing to tag under quay.io/ to avoid the registry prepend.
3. The issue likely also affects other recent Intel microarchitectures. It may be worth opening a corresponding issue in the bwa-mem2 or bioconda-recipes repositories to get the AVX2 binary recompiled/tested against Arrow Lake.

Command used and terminal output

nextflow run nf-core/sarek \
    --input samplesheet.csv \
    --outdir results \
    --tools strelka,vep \
    --genome GATK.GRCh38 \
    -profile docker \
    --aligner bwa-mem2

Relevant files

No response

System information

nf-core/sarek	3.9.0
bwa-mem2	2.2.1 (community.wave.seqera.io/library/bwa-mem2_htslib_samtools:e1f420694f8e42bd)
Executor	Docker
OS	WSL2 — Linux 6.6.87.2-microsoft-standard-WSL2
CPU	Intel Core Ultra 9 285K (Arrow Lake / Lion Cove)
Genome	GRCh38 (--genome GATK.GRCh38)
Aligner	--aligner bwa-mem2