Conferences certainly dominate journals, but most people publish their work on arxiv first, anyway.

I'm in CS (at the intersection of PL/compilers/HPC), and I've never heard of anyone in my field doing that. In fact, the only papers I've read on arxiv have been ones linked on HN.

I'm in a similar intersection (hi!), and same goes for me. I want to change that, though. I have started publishing tech reports (I work in an industry research lab) whenever I submit a paper for review. I'm tired of work being stuck in endless review cycles, not public and not referenceable. Were I still in academia, I would submit to arxiv, and I have even recommended this to grad students.

Virtually every ML paper is posted on arxiv before conferences.