Possible 4x speed up of merge --gvcf ref.fa

[pd3]

This needs to be tested, but it appears that merge --gvcf ref.fa is 4x slower than merge --gvcf -. This may be caused by inefficient use of faidx_fetch_seq; would pre-loading and caching the reference sequence help?

[jkbonfield]

Have you tried doing a small file and using perf record (and perf report) to profile it? It'll soon point out where the differences in CPU lie.