News for hackers. Stuff that matters.
Re:Node v9.4.0 by rf15 on Wednesday January 10, 2018 @11:42AM (#16116428)
I'm always impressed by some of the optimisations they find: 0aa403b649 improves readDouble/readFloat in buffers by 200-800% simply by...moving the code to plain javascript binary operations instead of the C++ runtime.
Re:Node v9.4.0 by coolspot on Wednesday January 10, 2018 @12:10PM (#16116686)
My understanding is that transitions between C++ runtime and JS context are relatively expensive.
Re:Node v9.4.0 by Something1234 on Wednesday January 10, 2018 @01:57PM (#16117760)
What generates that improvement percent confidences that was included in that commit?
Re:Node v9.4.0 by dingdingdang on Wednesday January 10, 2018 @02:18PM (#16117981)
Yeah, could be that you see 200% improvement on first call and then proceed to suffer 2000% slowdown unless this is benchmarked properly.
Re:Node v9.4.0 by __s on Wednesday January 10, 2018 @05:07PM (#16119590)
Given the context it's likely the other way around

Nobody really cares if your 1 call is 20x slower though. Much better to make it 2x faster when there's 100 calls

Re:Node v9.4.0 by Something1234 on Wednesday January 10, 2018 @06:58PM (#16120272)
I was talkiing about the tooling that generated the output, but this is a valid point too. Still want to know what the tooling is.
Re:Node v9.4.0 by rf15 on Sunday January 14, 2018 @12:28AM (#16142948)
They probably just timed execution with a few thousand samples and compared results. Nothing too fancy either way, I think.