2011/3/14 Shane Bryan <[email protected]>: > On Mon, 14 Mar 2011 16:34:09 +0100 > Carsten Munk <[email protected]> wrote: > >> 2011/3/14 <[email protected]>: >> > I was taking a quick look at the apps on MeeGo images. It seems >> > that none of the current apps utilize launcher (booster) for >> > improving the startup time. Previously, I've seen that using >> > launcher can improve startup time of meego touch based apps 1-2s, I >> > do not know yet how much the improvement would be on N900. There is >> > also no memory penalty in using launcher, it may even save some >> > memory. Prestarting is different thing. >> > > > It was also my understanding that the booster/pre-launch combo was also > really only of benifit to ARM, since the IA platform can efficiently > use PMIC ... though I confess this is not an area of expertise for me. > > My point is, if it helps ARM but hurts IA, then it's a non-starter. > Likewise, if it helps IA and hurts ARM, I suspect the same is true. We > really need to find appropriate mechanisms that work well across > architectures, or clear documentation on how to re-write apps in a > manner that allows them to be conditionally compiled per arch, if it > makes sense to do so. >
Shane actually brings up a relevant point regarding booster here. Let me try to explain.. The "booster" part is that in effect, the booster preloads some libraries/does the initialization phase of Qt/MTF/whatever once and then receives requests to open a application and then dlopen() the application. The application is compiled with -fPIC and linked with -pie -rdynamic to be able to do this. The problem on X86, or Atom specifically is the problem shown in this article: http://software.intel.com/en-us/blogs/2010/10/12/avoid-short-functions-on-atom/ Now of course the discussion would be if the startup performance is more important and if the application hit is in practice that bad. Considering that Qt is built with -fPIC on X86 (and hence have the performance hit already) the hit for rest of application shouldn't be that bad. I think startup performance is more important than runtime, considering it takes like 5s for dialer to start up even on X86. BR Carsten Munk _______________________________________________ MeeGo-handset mailing list [email protected] http://lists.meego.com/listinfo/meego-handset
