On Mon, Feb 10, 2025 at 11:34 AM Peter Bindels <dascandy_at_[hidden]> wrote:

> What about the indirect benefits? From how I understand the proposal the
> unstable_* functions do a whole lot less cache line clearing and
> memcpy'ing; this should have a general benefit on software beyond what you
> can measure in a microbenchmark.
>

I agree, but I am unaware of easy way to measure this. Beside cache there
is the benefit of not using branch prediction, meaning other parts of
software get to use it, and their prediction rate should improve. Again
quite hard to prove this will improve performance, but it should.