2 ms·
One difference is that before only bursts needed to be super tight, now idle performance have to be in addition. Maybe his rule is still valid, but the two case
by im3w1l 12y ago
One difference is that before only bursts needed to be super tight, now idle performance have to be in addition.
Maybe his rule is still valid, but the two cases do seem a little bit different to me.
Previously it used to be about optimizing that innermost loop that was only a few instructions long so it went twice as fast. A small local change with inline assembly.
But idle behaviour involves architectural optimizations. Like pushing events instead of polling for them.
- ori_b 12y ago> now idle performance have to be in addition. Your idle time should be with all processes blocked in select()/poll()/epoll() and friends, and the CPU fully shut off with no code at all running. If your CPU isn't fully asleep -- if it's running instructions -- you've already lost. There was a bug with my cellphone where a service was holding a wakelock and preventing the CPU from shutting off. This was enough to drain enough power that it would drain even while on the charger.
- im3w1l 12y agoYeah, hopefully nobody busyloops ala for(;;)checkForEvent(); But I could see someone doing for(;;){checkForEvent();sleep(sometime);} or setInterval(checkForEvent, sometime); This wouldn't be terrible but not ideal.