Re: [myhdl-list] Floating Point Libraries

[Christopher F. <chr...@gm...>]

<snip>
>
> So why not start with a workaround? We could define an
> interface using a_sign, a_mant, a_exp and b_sign, b_mant, b_exp.
> Ok, the interface is flattened, but in the code the only
> difference would be the underscore instead of dot notation.
> But it would teach us if this type of interface is really
> what we want to work with. And it works today. And if conversion ever
> gets more powerful, an upgrade may be simple.



Are you thinking something like the following?  If I understand, the 
proposal would be to have one Signal for a complex structure vs. a 
structure with multiple Signals.

-----------
from myhdl import *

class TypeStruct(object):

	def __init__(self):
		x = intbv(0)[8:]
		y = intbv(0)[12:]
		z = intbv(0)[23:]
		

if __name__ == '__main__':
     struct = Signal(TypeStruct())
     struct.next.x = 1
     struct.next.y = 2
     struct.next.z = 3
     struct._update()   # usually happens behind the scenes

     print struct.x, struct.y, struct.z

-----------

This isn't specific to the FP interface proposal.  This is the general 
method to handle grouping of types.  And then the conversion proposal 
would be; that a Signal carrying a class type (and dict?) the internal 
intbv types would be expanded to struct_x, struct_y, struct_z, for the 
above example?

I follow, that the conversion work would follow the general modeling 
approach.

Or when you said "code" you are referring to the MyHDL code.  And for 
the near-term it is more important to define a working convertible FP 
module that is support by MyHDL 0.7.  The FP unit would be built using 
the interface you mentioned and in the future it could be replaced by 
something like the above.

Chris Felton