runtime: improve timer performance #53953
Description
Needed changes to improve performance (At least in theory):
- Change
timers []*timertotimers []timerBucket - Make new timerBucket struct as
type timerBucket struct {
timer *timer
// Two reason to have timer when here:
// - hot cache to prevent dereference timer to get when field
// - It can be difference with timer when filed in timerModifiedXX status.
when int64
}- Delete
when int64from timer struct in favor oftimerBucketwhen field. - Rename
nextwhen int64towhen int64in timer struct - Make some few other changes to respect above changes in related functions
By this changes, we claim that we improve performance in siftupTimer and siftdownTimer when need to compare timers added when to timing heap, we don't need to get data from many different location in the memory (dereference many timers) and we have timers added when in sequential array.
Additional suggestions to improve code readability:
- make
timer.goandtimers.goin runtime package - make timers struct
- move timers field from p struct in runtime2.go to timers struct and embed this struct instead
- move all timers logic from many files in runtime package to timers.go such as logics in
time.go - almost everthing,proc.go - checkTimers, ...
p.s: If the idea is ok, I can help to make changes by send a PR.