N
как всегда короче
М?
Size: a a a
2019 November 05
ҪҸ
да те, кто публикуют это, не в курсе что там на самом деле
ҪҸ
(я постоянно им пишу правки в релиз ноутс safari tp, например)
US
(and sorry for adding you here without asking 😇 btw we have some non-russian speaking people here already)
No worries. The benefit of telling me on dm would be that I would have noticed sooner 😂😂😂
US
(worst case)
Depends. Worst case? Never.
US
It only optimizes what it considers "hot". If you have poorly optimized bytecode somewhere but it's not your bottleneck, it won't be a hot function and thus V8 won't waste time.
EO
It only optimizes what it considers "hot". If you have poorly optimized bytecode somewhere but it's not your bottleneck, it won't be a hot function and thus V8 won't waste time.
Hi Ujjwal,
The question was more about “how many function calls must be done in order to tell the engine, hey, optimize the function, pls”
And, we figured that it depends on the bytecode length. Does it sounds like true?)
The question was more about “how many function calls must be done in order to tell the engine, hey, optimize the function, pls”
And, we figured that it depends on the bytecode length. Does it sounds like true?)
US
Hi Ujjwal,
The question was more about “how many function calls must be done in order to tell the engine, hey, optimize the function, pls”
And, we figured that it depends on the bytecode length. Does it sounds like true?)
The question was more about “how many function calls must be done in order to tell the engine, hey, optimize the function, pls”
And, we figured that it depends on the bytecode length. Does it sounds like true?)
I think it's a rather simplistic conclusion.
EO
US
I mean, sure, it should depend on the length, bit also a bunch of other things.
US
For one, you need the feedback vectors to have warmed up with enough type information.
US
If they haven't had decisive information, it's hard to optimize based on indecisive data.
US
Another thing which I'm most afraid of is that V8 arbitrary decides if it's worth it to optimize.
US
So if your function is hot or not is somewhat weird. Sometimes some other calls are slower than your core logic (maybe you're using a shitty crypto library and that becomes your bottleneck).
US
In cases like these, V8 refuses to optimize sometimes because it doesn't see enough value in optimizing which is very fair, I guess. So you need to make sure your core application logic is "hot". And not some miscellaneous other thing.
US
In cases like these, V8 refuses to optimize sometimes because it doesn't see enough value in optimizing which is very fair, I guess. So you need to make sure your core application logic is "hot". And not some miscellaneous other thing.
Sometimes the hottest part of people's code is Google Analytics 😅😂
EO
If they haven't had decisive information, it's hard to optimize based on indecisive data.
as I know V8 is a master of speculative optimizations :)
US
I did an entire talk about it, it's my favorite optimization pass of all...