A call to a no-op function is taking an abnormally large amount of time. Something like 75ms for 1000 of them where in a desktop browser it’s taking around 1ms. We had a bunch of other tests centered around mootools and the canvas tag that all came out fairly equally slow and the guess is that there is just some global overhead to function calls that’s accounting for the general slowness rather than problems with particular API’s (whose core functionality seems quite fast for a small device when you subtract out the appropriate function overhead).

You can run the tests here: http://lowtidegames.com/test/shapesonly.html (note: it does send the results to our server for easily getting the results off the iPhone).

A sample iPhone run:

no-op took 0 milliseconds.
get context took 1 milliseconds.
canvas bunch took 11 milliseconds.
100 times empty loop took 2 milliseconds.
1000 times empty loop with nothing() took 74 milliseconds.
set fillStyle 100 times took 88 milliseconds.
set strokeStyle 100 times took 93 milliseconds.
100 strokes of 20px took 63 milliseconds.
100 10×10 fillRects took 38 milliseconds.
100 25 radius triangles stroked took 179 milliseconds.
100 25 radius triangles filled took 180 milliseconds.
100 25 radius triangles stroked with style each time took 277 milliseconds.
100 25 radius triangles filled with style each time took 271 milliseconds.
100 strokes of 10px with style each time took 78 milliseconds.
100 5×5 fillRects with style each time took 124 milliseconds.
line math 10000 times took 1456 milliseconds.

Try/Catch with errors 12
Layer movement 79
Random number engine 128
Math engine 174
DOM speed 45
Array functions 483
String functions 48
Ajax declaration NOT RUN
Total Duration 569 ms

So I’d say Opera Mini (v 3.1) doesn’t do too badly either. Well, except it didn’t run the Ajax declration.

Opera Mini 4 (beta) gave the following results:
Try/Catch with errors 0
Layer movement 0
Random number engine 155
Math engine 74
DOM speed 37
Array functions 120
String functions 12
Ajax declaration 68
Total Duration 466 ms

Interestingly though, Microsoft’s Pocket IE running on my SP5 will only give results for the first test (Try/Catch with errors) giving that test a result of 2000!

But also note that these results were obtained with the phone connected via USB to a PC with internet.

What does that mean? Does iPhone has a Random number engine

Here’s to Adobe porting Flash (and hopefully AIR) to iPhone soon.

I’ve definitely noticed the speed difference anecdotally. I tried to put it through its paces with the Dojo Chart benchmark, but the iPhone is not Safari 3 and therefore doesn’t have SVG support.