BCrypt is too slow about crypto HOT 10 CLOSED

It's possible to be something eval target specific. On try.haxe ( js target ) is about 0.12 sec.
https://try.haxe.org/#0c7a7 , same on try-mrcdk ( http://try-haxe.mrcdk.com/#e6074 ) with latest haxe dev

from crypto.

Can you factor out some code that executes slowly so I can take a look?

from crypto.

Sorry, maybe is not so clean code, but here is example: https://try.haxe.org/#ed87F
The problem is with function f(x:Int)
On eval I get :
Time: 5.0531008243560791 seconds
For JS :
Time: 0.6010000705718994 seconds

from crypto.

Well, we have to put things in perspective here:

Of course a number cruncher like this is going to be slow on targets that don't have a static compilation or a crazy JIT like JS does.

from crypto.

You can bring this down to 3.7 by making f and sBox static. Eval has trouble optimizing instance calls because it can't be sure if a function is overridden or not. Technically, I can't even rely on final for that because there's this goddamn @:hack thing.

from crypto.

from crypto.

Sure, but then there's no call left for me to optimize.

From Nicolas reaction, it seemed like we don't have to consider @:hack. We might want to restrict it to externs.

from crypto.

Yes, it would be sad not being allowed to optimize anything.

from crypto.

I just checked, @:hack only works on classes, not on fields. So checking cf_final is safe, we just can't check for cl_final until we address HaxeFoundation/haxe#7473.

from crypto.

Using Vector give some performance boost . Nothing else can do here .

from crypto.