Does wgpu insert barrier between draw calls ? What about compute dispatches ?

[newDINO]

For draw calls, it seems reasonable that barriers are not inserted when the pass has a depth attachment, and that barriers are inserted when there is not depth attachment.

For compute dispatches, it seems reasonable that barriers are not inserted when there are no dependencies between dispatches. But wgpu does not have explicit write-only storage buffer, so theoretically, if multiple dispatches writes to a single storage buffer, barriers need to be inserted.
What about accessing buffer using dynamic offset ? What if different dispatches access different regions of the same buffer, will barriers be inserted between them ?

Is the above guesses right ?
My main concern is a bunch of small dispatches or draw calls can only executed sequentially.