Huuuge performance improvement & more userfriendly way of digesting changes

schteppe / p2.js

JavaScript 2D physics library

Other

2.63k stars 329 forks source link

Huuuge performance improvement & more userfriendly way of digesting changes #283

Closed issy123 closed 6 years ago

issy123 commented 7 years ago

Changes:

cached calculation of array.length
changed some functions that are heavily used to use a for loop instead of .indexOf()
improved deleting of objects
more userfriendly way of digesting changes. Before it would throw an exception if you try to make changes while the world is busy making a step (affected events beginContact, endContact etc). Now it will digest changes in the next world step

englercj commented 7 years ago

Do you have benchmarks to show the "Huuuge performance improvement"? I'm curious what the actual change is.

issy123 commented 7 years ago

@englercj I have done multiple cpuprofile samples of my game and just modified the ones with the most computation time to be more performant.

I don't really have benchmarks in terms of facts, but in my game (without changing my own game code ofcourse) before i could spawn 100 circles/boxes & sensors with 80-100% cpu usage. After i could spawn 1000 circles/boxes & sensors with 70-90% CPU usage.

But those are not really things that proofs it, just talk of a random guy.

If you can run benchmarks i would love to see them!! :) You can see the changes in the commit.

finscn commented 7 years ago

I like this PR. There are many denseness computations in a physics engine , I think we shouldn't despise any micro-optimization . In particular，when we use JS and run it on mobile device.

issy123 commented 7 years ago

@englercj

also I have changed the code you contributed, it may be fast for FF but not for the v8 engine (webkit browsers & nodejs) the change i made is in experience a more efficient one.

https://github.com/schteppe/p2.js/commit/3879c0194ba91a6cb37bb056f9e6893b9d9688bd

englercj commented 7 years ago

@issy123 I'm sure the browser landscape has changed in the past 3 years but that change was originally to increase performance and not increase GC. I specifically didn't want to allocate a new object each reset.

I also stand by what I said before: benchmarks prove performance increases, otherwise we are just guessing.

issy123 commented 7 years ago

@englercj do you have anything you can suggest to benchmark with?

issy123 commented 7 years ago

Is this something you can work with?

So just like the unit tests in test/ folder i can make a benchmark/ folder and u only have to type grunt benchmark ? just like you type grunt test The results are similar to jsperf.

englercj commented 7 years ago

benchmark.js is the standard, most other js things wrap it (including jsperf).