performance problems on sailfish3

asked 2018-11-01 16:43:59 +0200

NeWin gravatar image

updated 2018-11-01 17:13:54 +0200

hi, I just downloaded sailfish3 on my xperia X and dispite having in the log "both the new app launch and in-app page opening paths are considerably quicker now. Things load faster." I have some quite annoying performance problem. It seems the lazy loading of QML engine is creating itches during allocation (espetially image allocation) for example in the Gallery (where I expected it to be faster due to those optim+revamp) just scrolling create hitches. This looks general to all apps that display heavy content. This looks so bad that it begin to feel like android =/.

PS: I would be happy to help with more profiling details but unfortunately the sailfish SDK doesn't provide really great way to profile anything just like we can do on desktop or game consoles.

image description image description

edit retag flag offensive close delete

Comments

yep, scrolling in SF3 in Gallery is horrible =(

virgi26 ( 2018-11-01 19:22:28 +0200 )edit

Now that you mention it I've noticed some stuttering with quick scrolling in a rather heavy list view (WiP twitter client, delegates often with many big pictures and lots of rich text, often in Japanese) that I think might not been there before. Thankfully it seems more like noticeable rather than outright annoying, but I will be monitoring the situation.

MartinK ( 2018-11-02 00:58:21 +0200 )edit

there is some application that delay the display of images to avoid the hitches. In game development (and sometimes in web ?) what could also do is loading a lower resolution version while waiting for a better one (also ask for a transition like the deplayed display). in addition they could also load images in advance on flickable elements it take more ram but it also give more time to load next images. To fix the fact images take a lot of ram I discovered by using QT that images are always loaded on CPU+GPU but as long as you don't do any modification on them (which is 99.9 of the time) there is no need to have a CPU copy, additionally on mobile the ram is shared so QT pay each image twice.

IMO jolla should write a GPU only Image qml element (or a special property of the existing image?) and using the freed ram to load in advance + enable the asyncronous display by default.

NeWin ( 2018-11-03 13:51:26 +0200 )edit