[Vm-dev] CogDroid: first public alpha available for testing

Dimitry Golubovsky golubovsky at gmail.com
Sat Sep 3 13:54:15 UTC 2011


Thanks for the interest to the CogDroid alpha. Answering to all questions here.

@Gary: I'm afraid I am almost the one involved with this at the moment
;) I do not have a specific timetable, and my next step  is to create
a meaningful tool to make JNI calls to the Java libraries available on
Android, since Google discourages many things to be done natively. I
think I'll announce beta after that's ready. Regarding etoys, I think
it is possible to try the original image, or whatever is available
from the Scat project [1] (OK, that's Scratch, but IMHO similar
purpose) with the new VM. Monticello should work (or if it does not
work in any way, please let me know). In other words, the more people
help me with testing, the sooner we get a higher grade release.

Regarding older images, one thing to keep in mind is how much polling
they have [2]. Although I (remembering from prior discussions) also
made browserPluginExitIfNeeded (sp?) to exit the interpreter if it is
running too long (current time quant is 200ms although this si very
soft rule), so event serverMode may be not so necessary. The VM writes
a log file located at /sdcard/jni.log. Upon exit (via the Quit/Save
and quit item of the world menu) it writes VM time statistics in that
file: what was average time in interpreter, average time between
interpreter calls, etc. This may also give some idea.

@Igor: There are three things to merge simultanesously. My dmg-blessed
repo at Gitorious [3], and the two packages at
squeaksource.com/EventVM: CMakeVMMaker-golubovsky.138.mcz and
VMMaker-oscog-golubovsky.121.mcz Not sure how you adjust
ConfigurationOfCog for these, but I think they should go to the
bleeding edge configuration. Besides, can Google SDK and NDK be
installed on Hudson to test at least buildability? Or I might install
a build slave at my computer solely to test Android builds, but we
need to discuss this in separate thread.

@Stephane: I don't think of any actions on the image side at the
moment since it works with 1.3, and after involving the timed
executionI was able to even draw a PolygonMorph (although very slowly,
but at least the application did not crash every time, only sometimes

@Mariano: Thanks.

And we have the first benchmark (will also add it to the wiki):

Notion Ink Adam, Android 2.2

1 GHz, Dual-core ARM Cortex -A9, Nvidia Tegra 250

Classic VM - Squeak 3.9 36,930,178 bytecodes/s 1,201,501 sends/s

Stack Cog - Squeak 3.9 41,775,456 bytecodes/s 2,074,373 sends/s

which seems a very good speed improvement.

[1] http://code.google.com/p/scat/
[2] http://lists.squeakfoundation.org/pipermail/vm-dev/2011-August/009209.html
[3] http://gitorious.org/~golubovsky/cogvm/dmg-blessed

Dimitry Golubovsky

Anywhere on the Web

More information about the Vm-dev mailing list