Inline Matrix3 element functions?

[samreid]

While working on phetsims/circuit-construction-kit-common#295 @jonathanolson and I noticed that Safari iPad3 tethering reported a non-negligible amount of time being spent in the Matrix3 element functions like m00, m01, etc.

We tried a CCK-DC test with the usual m00 style functions and saw ?profiler report between 21-22 frames per second. When inlining these functions (and using array accessors instead), the performance reported 22-23 frames per second. We only ran this test once, it would be good to confirm and run on other hardware such as iPad2.

Replacing the named element accesses with indexed array accesses would make Matrix3 harder to read and understand and maintain, but may buy us a bit of performance across all our sims.

[pixelzoom]

... may buy us a bit of performance across all our sims.

Can this be quantified? Will it be significant to the user experience? (The reported improvement for CCK:DC was only 22fps to 23fps.)

[jonathanolson]

If it's the difference between 22 and 23 fps, that's almost a 2ms difference. If that stays somewhat constant from a ~45ms frame to a ~16ms frame (~60fps), it would be a more significant difference. If an almost-60fps sim would instead only gain 0.7ms a frame, it's still potentially relevant.

[samreid]

I'm inclined to investigate things like WebGL phetsims/circuit-construction-kit-common#367 first