Re: [oc] Beyond Transmeta...

[Suboner@aol.com]

> Since you are SW guy I should warn you: in CMOS there is practicaly no
> redundant power consumption, since gates consume power only when
> changed. No extra circuits are neccessary...

Hmm didn't know that but if at one time I add 2 variables and then at another 
I redo those 2 variables, there is redunt work there. Of course this is that 
if you have setup the network as being persistent (which makes me realize 
that the reason it is like this is because it consumer more memory when its 
in persistent).

> There is no need for 10 bit processors. Your network have to build '32b
> instructions',
> since compiler doesn't really know whether you will use only 10b in 'int'...

Hmm, I'm not sure you got that one. What I meant was that there may be 
changes to bits that require 10 relationships to be recalulated in one pass, 
at the end result of those 10 relationships being recalculated may effect 
another 32 relationships to be recalculated in the next cycle. Assuming you 
have at least 32 1bit processors, 10 1bit processors would be used in the 
first pass and 32 1bit processors would be in the second pass. If you have 
only 5 1bit processors, then you would have to spend 2 clock cycles doing the 
first pass (2 * 5 = 10), and then 7 clocks on the second pass (7 * 5 = 35 > 
32). I think that may clear it up. In reality the compiler could do diffrent 
things, it could either convert the program to code for a high level 
processor, or it could convert it to a persistent network, the persistent 
network would consume a lot of memory (because the processing will be 
incorperated into the program).

I think maybe calling them instructions is kind of throwing you off. I should 
probably call them a relationship that would connect 3 bits (2 in, 1 out), 
and the out bit would be the result of a 4bit instruction that the 
relationship is based upon. Or something like that.

> I don't think that there is any problem of making it in HW (except routing),
> but there are still problems in arhitecture itself.

Ahh, that is very true, but I still have a number of ideas of diffrent ways 
it can be done, and knowing the hardware limitations would probably make it 
easier, but probably also knowing architecture limitations would probably 
help know what kind of hardware problems would arise. I guess its a matter of 
what came first the chicken or the egg.. :).. well, I'll probably start 
working on something, if anything I could potentialy convert it to other 
formats if need be, I'll probably use GTK+...

Leyland Needham