Nathanael,

Nathanael Schärli wrote:
> 
> I was aware of the fact that the old recognizer is using the pen trail, but
> when I implemented Genie I thought it would be cool if the Genie trail would
> be a real Morph and we could even use it properly on a BouncingAtomsMorph or
> anything else that is moving around. However, since trails are anyway only
> visible for a short time, performance is much more important than fancy
> features like correct updating on top of a BouncingAtomsMorph. If we can
> make it faster using a single Stroke Morph instead of multiple LineMorphs
> that would be cool! Otherwise, I'll probably just switch to using the trail
> pane (as turtles and the old recognizer do it).
> 
> (Historic note: The first pre-alpha-version of Genie was using the trail
> plane instead of Morphs and once more it looks as if the straight-forward
> and simple approach is not so bad as it looks like :)
>
 
> As far as the delay in the lookup is concerned, I'm really thinking of
> writing a primitive for it. 90% of the time is spent in the method
> CRStrokeFeature>>sameClassAbsoluteStrokeDistance:forReference: that
> basically iterates over the vector decompositions of two gestures and
> compares them using integer arithmetic. (I avoided any floating point
> operations because that is even slower on small machines). This method has <
> 100 LOC and it only calls two helper methods (3 LOC each). I don't have much
> experience concerning the performance gain by making a primitive out of such
> a method, but I have a feeling that it could be quite a lot.
> Does anyone know more?

After taking a look onto it I guess a speedup of factor 5 to 10 for this
method, if written as primitive. This is a conservative estimation.

If you or another person haven't written it so far, I'd like to help here.

Greetings,

Stephan

> 
> Nathanael
> 
> > -----Original Message-----
> > From: hmm at heeg.de [mailto:hmm at heeg.de]
> > Sent: Thursday, March 15, 2001 2:32 PM
> > To: squeak at cs.uiuc.edu
> > Subject: Re: Genie on iPaq..comments
> >
> >
> > Kevin Fisher wrote:
> > > I'm not sure what could be the cause of the drawing
> > slowness.  It certainly
> > > isn't system load, as I'm basically only running an
> > X-server and Squeak (no
> > > window managers or anything else).
> > >
> > > Generally Morphic is considerably slower on the iPaq, but
> > as I mentioned, the
> > > CharRecog pen trail isn't laggy at all by comparison.
> > The difference between CharRecog and Genie is that the former
> > writes its
> > trail directly to the screen, while the latter contructs and
> > modifies a
> > Morph for the pen trail, which is of course much slower if Morphic is
> > slow on your machine. I think it should be easy to retrofit
> > Genie with a
> > direct-to-screen trail.
> > I have also noticed that the feedback is accomplished with a number of
> > straight line morphs. Probably a stroke morph which just gets
> > new points
> > appended would make things a bit faster. This kind of morph
> > might either
> > invalidate the rectangle encompassing the added line segment for each
> > new point, or it could draw that segment directly to screen for that
> > special case.
> > I will probably play with it a little bit today, maybe the
> > outcome will
> > be such a faster stroke morph.
> >
> > Cheers,
> > Hans-Martin
> >

-- 
Stephan Rudlof (sr at evolgo.de)
   "Genius doesn't work on an assembly line basis.
    You can't simply say, 'Today I will be brilliant.'"
    -- Kirk, "The Ultimate Computer", stardate 4731.3