I am using Math.abs for my 1000 points as below
Math.abs(a - b) < tolerance;
My profiler shows that Math.abs uses 62 ms, I need to optimize it, so tried this
((a-b) < 0 ?-(a-b) : a-b) < tolerance;
I want to know which is better and fast
Asked
Active
Viewed 220 times
-2
James A Mohler
- 11,060
- 15
- 46
- 72
user2918741
- 87
- 1
- 10
-
2Is it **the real bottleneck** in your application? – aga Nov 18 '13 at 06:52
-
1Try testing it! You've already written the code, so why are you asking here? – elixenide Nov 18 '13 at 06:52
-
1@aga It seems that this code runs a 1000 times. So yes, that can be a performance issue. – James A Mohler Nov 18 '13 at 06:54
-
Test both at http://jsperf.com/ – Anto Jurković Nov 18 '13 at 06:54
-
What does your profiler say about the optimization? – JJJ Nov 18 '13 at 06:54
-
1Proof that profilers are useless :D It is physically impossible for Math.abs to take 62ms for 1000 calls. You probably interpreted it wrong. – Esailija Nov 18 '13 at 09:43
1 Answers
0
It highly depends on the browser (and version) used as you can see in this jsperf and is discussed in related question on SO:
Anno 2013 one could state that:
Math.abs()is mostly fastest in IE, FireFox and Safari- Bitwise
(x ^ (x >> 31)) - (x >> 31)is mostly fastest in Chrome and Operah - Ternary
(x < 0 ? -x : x)was only the fastest in FireFox 3.x and 4 era.
You could provide a dynamic function-loader to initiate the optimal function (based on live in-browser speed-profiling or browser-detection), containing the fastest routine for that browser.
Naturally this comes at a couple of costs:
- time wasted on browser-profiling (reliability while loading the page) or browser-detection (well, reliability in general..)
- extra bandwidth for larger javascript
- (usually) larger memory-footprint (because on the fly generated functions are often slower then static defined functions (that are given the proper indentifier)
So in general I'd prefer:
Mah.abs() (since that is often the smallest and clearly indicates intent).
Followed by the ternary (often coupled with the comma-operator by which I can perform more logic with just one test, this can be both faster/shorter and even require less programming-logic).
Finally the bitwise variant reveals the least intent, doesn't offer some of the above ternary-advantages AND only works reliable if you have no more than 32 bits..
Hope this helps!
