QML optimization #4339
QML optimization #4339
Conversation
Be-ing
force-pushed
the
qml_optimization
branch
3 times, most recently
from
1c2706e
to
01465f5
Compare
res/qml/Deck.qml
Outdated
| @@ -185,18 +185,15 @@ Item { | |||
|
|
|||
| let minutes = Math.floor(positionSeconds / 60); | |||
| let seconds = positionSeconds - (minutes * 60); | |||
| let centiseconds = Math.trunc((seconds - Math.trunc(seconds)) * 100); | |||
| let deciseconds = Math.trunc((seconds - Math.trunc(seconds)) * 10); | |||
There was a problem hiding this comment.
| let deciseconds = Math.trunc((seconds - Math.trunc(seconds)) * 10); | |
| const deciseconds = Math.trunc((seconds - Math.trunc(seconds)) * 10); |
| onValueChanged: marker.x = parent.width * value | ||
| onValueChanged: { | ||
| // Math.round saves tons of CPU by avoiding redrawing for fractional pixel positions. | ||
| marker.x = Math.round(parent.width * value * Screen.devicePixelRatio) / Screen.devicePixelRatio; |
There was a problem hiding this comment.
won't the divide make the position fractional again?
There was a problem hiding this comment.
Yes, but only for fractional logical pixels that correspond to real display pixels. So for a scale factor of 2, updates are done for every 0.5 logical pixels, or every real display pixel, but not extremely tiny fractions in between.
There was a problem hiding this comment.
Ah, so I guess the speedup comes from the fact that Qt doesn't have to do anti aliasing?
There was a problem hiding this comment.
That doesn't make sense either though.
This feels like it should not be something programming QML should have to concern themselves with. So I feel like this is bug in Qt?
There was a problem hiding this comment.
I don't know, but I know that this little change saves a lot of CPU 🤷
There was a problem hiding this comment.
Mhmm the only other idea I have is that the COs value changed multiple times per frame so the callback is also executed quite often. Since the position is always slightly different, the return value is also different. When rounding we always return the same value in the same pixel, so we often return the same value which enables Qt to do some caching optimization (so only doing work when the value actually changes).
This reduces CPU usage by about 10% on my Intel Core i7 8550U.
Be-ing
force-pushed
the
qml_optimization
branch
from
01465f5
to
95d7af3
Compare
based on #4327