Kivy performance...

[Ryan]

I'm curious to know how I can improve Kivy's performance ...

I'm interested in getting several hundred (thousand?) images moving around at the same time. Now, I realize this will be expensive to compute, but I was hoping that someone here could enlighten me in to options at improving Kivy's performance... perhaps it's just not possible to do on current hardware?

Still, here's what I'm using for hardware:

- Intel Xeon W3670 (6 cores with hyperthreading) @ 3.2Ghz

- 12GB RAM

- nVidia GTX580

- Windows 7 Professional

I've modified the Pictures demo example to load 100 copies of each image (currently 3 images x 100), added an FPS counter, and made it such that an 'on_touch_move' will move *ALL* images.

Here's what I came up with:

import kivy

kivy.require('1.0.6')

from glob import glob

from random import randint

from os.path import join, dirname

from kivy.app import App

from kivy.logger import Logger

from kivy.uix.scatter import Scatter

from kivy.properties import StringProperty

# FIXME this shouldn't be necessary

from kivy.core.window import Window

from kivy.clock import Clock

from kivy.uix.label import Label

class Picture(Scatter):

    source = StringProperty(None)

    

class PicturesApp(App):

    def build(self):

        # the root is created in pictures.kv

        root = self.root

        self.root.on_touch_move = self.on_touch_move

        

        self.fps_label = Label(halign='left')

        self.root.add_widget(self.fps_label)

        Clock.schedule_interval(self.set_fps, 1.0/60)

        # get any files into images directory

        curdir = dirname(__file__)

        for filename in glob(join(curdir, 'images', '*')):

            try:

                # load the image

                for x in range(100):

                    picture = Picture(source=filename, rotation=randint(-30,30), pos=(randint(0,400), randint(0,400)))

                    # add to the main field

                    root.add_widget(picture)

            except Exception, e:

                Logger.exception('Pictures: Unable to load <%s>' % filename)

    def on_pause(self):

        return True

    

    def on_touch_move(self, touch):

        for p in self.root.children:

            if (p.__class__.__name__ == "Picture" and touch.grab_current != p):

                p.pos = (p.x + touch.dx/2, p.y + touch.dy/2)

                

    def set_fps(self, dt):

        self.fps_label.text = 'Current FPS: ' + str(Clock.get_fps())

        self.fps_label.pos = (-Window.width/2 + 120, Window.height/2 - 40) 

if __name__ in ('__main__', '__android__'):

    PicturesApp().run()

Based on my current hardware, when I do nothing (or if I change the code to move only one image at a time) performance is great!

But if I use the above code, Kivy chugs down to around 10 fps. Also, it seems to be independent of the window size, so it doesn't seem to be a rendering issue, but more of a raw computation issue.

Would I be better off spawning multiple threads to handle the movement?

Ryan

[Mathieu Virbel]

Interesting.

I would like to know if it's the computation from the on_touch_move, or the graphics computation than came just after the on_touch_move (like all graphics instructions are put in dirty state, and reuploaded or something. or if all of them are recompiled.)

100 doesn't seem big, but still, you can have  (if FPS is limited to 60) 60 touch move = 6000 collision test that involve Matrix computation.

A deeper test would be great to see where it's slowing down exactly.

Also, the 8 cores have no meaning, Kivy is using only one, and Python have GIL so...

Mathieu