Not related to this PR: we could also lift the 1 byte restriction as was done in the CreateSpan PR. Somehow I didn't realize until now.
One second thought, it's less beneficial here. We'd still need to copy the data, so maybe the savings aren't as meaningful.

#61414 doesn't do anything on stackalloc, unlike #57123 but I think that's not merged in.

Right, I mean we can do something like #61414 but for stackalloc. I didn't realize we'd already done a demo branch with exactly that. Thanks for digging up #57123. I'll ping there to see whether we should make a real PR with that change.

@jcouv, what became of this? Even just emitting it as a copy from the data section (which the JIT should be able to unroll in many cases) would likely be useful.

@stephentoub It looks like the hackathon project didn't transfer into a committed project/PR. The last comment from Jared there is that this just needs keyboard time, but we support it in principle.
We're out of dev bandwidth for .NET 8 (ie. at least for one month) unless urgent. Let's discuss around that time. We could file an 17.9 issue (refresh, resolve conflicts and bring #57123 into main) for tracking purpose.

I opened #69325. Thanks.