josyb

One-Process, Two-Process and Three-Process State Machines

"a state machine is worth a thousand equations" - Josy Boelen (yes ...)

and their Resets

About State Machines

On the internet for a dealing with FPGA programming, I often read the advice to write One-Process state machines only.
And I usually beg to differ. (You shouldn't be surprised ...)
The principal arguments given are that a One-Process state machine description is 'free from latches' and at the same time requires less typing. Now being lazy is a hallmark of a good engineer, but ...
In a lot of cases a One-Process description fits exactly. However in my line of work this is rarely so.
To understand the deeper differences between the One- and Two-Process description we have to go back to Moore and Mealy.

josyb

'''
Created on 23 aug. 2018

@author: josy
'''

from __future__ import print_function

from myhdl import Signal, intbv, block, always_comb, always_seq, ResetSignal

josyb

class RingCounter():
    ''' a simple 'Overbeck' ring counter.
        See https://en.wikipedia.org/wiki/Ring_counter
        It has zero overhead cost as no logic is needed to start.
        The following picture is worth a thousand words?
                       A          B          C            D
               ----    |   ----   |   ----   |   ----     |
             -| 0  |   |--| 1  |-----| 2  |-----| 3  |-----
            | |    | o-   |    |     |    |     |    |o-
            |  ----        ----       ----       ----   |

josyb

Following a discussion on Gitter[https://gitter.im/myhdl/myhdl] (November 24th, 14:40 about) I decided to make a small example to test it all. I have two versions of the MyHDL source:

• structuraldesign_block.py: using the @block
• structuraldesign_noblock.py: as in 0.9 using the old toVHDL() function call

You can see the two resulting VHDL files.

The one generated with the @block is a mess as it duplicates a lot of names. Of course the one generated by my local no-block MyHDL packages is ... well ... beautiful

josyb

There is a web-page where Adam Taylor lists 10 alternative FPGA development languages: http://www.eetimes.com/document.asp?doc_id=1329857 On some languages the OO-word was used ... I commented that the only thing I have seen so far from these languages is that they are (truly) class based but that I haven't seen any real example. Yet, as I did not study them to their deepest extent, having not enough time and too much other work ...

I already use class-based design for my MyHDL work, see my gist https://gist.github.com/josyb/afd84c9a06fdec77f2fd, but this is not OO as none of these classes have been subclassed.

In stead of doing some real work today (Sat Oct 22nd 2016), I decided to give OO in MyHDL a try. You can see the results in the two next files.

josyb

'''
Created on 2 Jul 2016

@author: Josy

a question by Merkourios Katsimpris on gitter:myhdl triggered me to look at
    http://paddy3118.blogspot.nl/2008/08/zig-zag.html

    while the solution presented there is (very probably) very pythonic
    you need to read the explanation to understand what or why it is doing

josyb

'''
Created on 29 Dec 2015

@author: Josy
'''
from __future__ import print_function

import random

import myhdl

josyb

def delta(Clk, Reset, D, Wr, Sel, Q):
    r = [Signal( intbv(0)[len(D):]) for _ in range(9)]
    
    @always_seq(Clk.posedge, reset = Reset)
    def rtl():
        if Wr:
            r[0].next = D
            r[1].next = r[0]
            r[2].next = r[1]
            r[3].next = r[2]

josyb

def array_2(clk, reset, D, Q):
    ''' testing a 2-dimensional Array
        just making a simple pipeline
    '''
    mt = myhdl.Array( (4, 3,), myhdl.Signal( myhdl.intbv(0)[len(D):]))
    
    @myhdl.always_comb
    def rtlcomb():
        Q.next = mt[3][2]
      

josyb

def array_3(clk, reset, Sel1, Sel2, Q):
    ''' testing a 2-dimensional Constant Array
    '''
    aoc = myhdl.Array( [[1,2,3], [4,5,6], [7,8,9]],  myhdl.intbv(0)[len(Q):])
      
    @myhdl.always_seq( clk.posedge, reset = reset)
    def rtlreg():
        Q.next = aoc[Sel2][Sel1]
       
    return rtlreg