The amount of stack in the L1 cache is essentially that. A shifting fast access working memory area.

Then think of registers as just part of the fetch & store pipelines for staging operations on stack values.

Forth goes all in with this approach.