Ok, I've looked into ArchiDroid (he's actually from PL ). I've checked (briefly) those -O flags, but most interesting was the github article about optimizing kernel compile The guy managed to get 0.05% decrease in execution time ;D extraordinary But the read was good because he provided a full analysis that this improvement exist and it's not just measurement error
But in the meantime I've found some docs http://source.android.com/devices/tech/debug/tuning.html and http://source.android.com/devices/tech/ram/low-ram.html. While the latter is already implemented in my kernel, I've made a set of options for hwui:
Defaults given in the docs were for screen 1280x800, which is 2.5x bigger than ours, so I've recalculated those. Now I'm just using the phone to see how it turns out
I'll also dive into source code and see how android uses cgroups - maybe then I'll figure something out about swap.
But in the meantime I've found some docs http://source.android.com/devices/tech/debug/tuning.html and http://source.android.com/devices/tech/ram/low-ram.html. While the latter is already implemented in my kernel, I've made a set of options for hwui:
Code:
root@janice:/ # cat /data/local.prop
ro.hwui.texture_cache_size=8
ro.hwui.layer_cache_size=6
ro.hwui.gradient_cache_size=0.2
ro.hwui.path_cache_size=2
ro.hwui.r_buffer_cache_size=1
ro.hwui.drop_shadow_cache_size=1
I'll also dive into source code and see how android uses cgroups - maybe then I'll figure something out about swap.