From: marcov@stack.nl   
      
   On 2005-08-01, Skybuck Flying  wrote:   
   >> > The cpu starts reading data and meta bits until it reaches a meta bit of   
   > 1.   
   >>   
   >> (ignoring the obvious "what is the (B)Pascal angle?" for a moment)   
   >>   
   >> "scanning" is to my best knowledge not easily paralellisable. Hence your   
   >> proposal would already lead to severe performance degradation.   
   >   
   > Assuming each individually bit can be addressed, scanning would not even be   
   > necessary.   
      
   If you have datafields larger than the wordsize it would.   
      
   > 1 Hole plugged.   
      
   Nope:-)   
      
   >>   
   >> Some other reasons:   
   >> - memory use would double.   
   >   
   > True, but ok... a more efficient coding could even plug this hole.   
      
   No. It would lessen it at best. Moreover, if the avg size of your fields   
   is low (e.g. bitpacked), it won't help that much.   
      
   >> - what if I have more metabits than register width? IOW what if I try to   
   > load   
   >>  a 64-bit value on a 16-bit cpu?   
   >   
   > A 1 bit cpu can solve it.   
      
   Only for certain values it is prepared to handle. Not for _all_ values. And   
   the whole idea was being able to handle values that normally cannot be handled.   
      
   > Why not a 16 bit cpu.   
   >   
   > Pretty simply really, a sliding register etc.   
      
   For addition/subtraction that works. For division/multiplication this is   
   costly. This is the reason why most CPUs don't support these operations   
   for more than twice the wordsize (and even that with modifications)   
      
   >> - If I per accident give only zeroes as meta, what would happen?   
   >   
   > Don't do that ;)   
      
   Note acceptable.   
      
   > Add some nice error checking :)   
      
   And then what? If you add a high bound, you are limiting your architecture   
   again, just the thing you want to avoid.   
      
   >> P.s. I'm no electrical engineer. If I can shoot so many obvious holes in   
   > your reasoning,   
   >> you don't even want to know what an EE could do :_)   
   >   
   > So far all holes plugged... bring on them holes ! :)   
      
   IMHO you showed that the performance impact can be less than 100%. That's all,   
   and was not really new to me.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)