Generics and state machines

[Steve]

Hi, I would like to create a state machine that is of generic size.
It is for accessing offchip memory, having parallel components attempting to
access and delivering access to one at a time. I just don't know how many
different components are trying to access the memory until I use it.

is there a way to create state type of arrays, so that generics can be used
to create the size
of the state machine???

steve

 

[Jim Lewis]

Steve,
Draw a picture and partition your design first.
Start with a fixed number of elements and then
figure out how to apply generics (and where they
are appropriate).

You have N devices. Each must manage its own
address and data. Each must handshake with
the Memory Controller in turn. Each will only
read or write data when it is selected.

You have something to choose one of the N devices:
An arbiter.

You have a multiplexer to route address and data
from the active device to the memory.

You have one memory controller. It probably
does not need to know that there is more than
one device connected to it.

With this view, only the Arbiter
and the Memory multiplexer will have generics.
The arbiter will have request and grant signals.
The memory multiplexer will need to handle
N address and data busses.

This should give you a start.

Cheers,
Jim

Hi, I would like to create a state machine that is of generic size.
It is for accessing offchip memory, having parallel components attempting to
access and delivering access to one at a time. I just don't know how many
different components are trying to access the memory until I use it.

is there a way to create state type of arrays, so that generics can be used
to create the size
of the state machine???

steve

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jim Lewis
Director of Training mailto:[email protected]
SynthWorks Design Inc. http://www.SynthWorks.com
1-503-590-4787

Expert VHDL Training for Hardware Design and Verification
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 

[Tom Hawkins]

Hi, I would like to create a state machine that is of generic size.
It is for accessing offchip memory, having parallel components attempting to
access and delivering access to one at a time. I just don't know how many
different components are trying to access the memory until I use it.

is there a way to create state type of arrays, so that generics can be used
to create the size
of the state machine???

steve

If the machine is farily simple, you might be able to use GENERATE to
extend it's functionality, i.e. increase its size. But I suspect
there may not be a clean solution with VHDL.

This sort of task is a snap with Conflunce. Just write an algorithm
to build the state transition table, then instantiate a state machine
with the given table. At a higher level of abstract, you could simply
pass in a list of component interface, then generate the logic for
each. The length of the interface list acts as a configuration
parameter.

-Tom

 

[Jim Lewis]

If the machine is farily simple, you might be able to use GENERATE to
extend it's functionality, i.e. increase its size. But I suspect
there may not be a clean solution with VHDL.

This sort of task is a snap with Conflunce. Just write an algorithm
to build the state transition table, then instantiate a state machine
with the given table. At a higher level of abstract, you could simply
pass in a list of component interface, then generate the logic for
each. The length of the interface list acts as a configuration
parameter.

Gee Tom, that looks like a shameless marketing plug for your product.
Does confluence allow you to forget about hardware?
That seems to be the main issue here. Once you sort
out what hardware you care to implement, it is
clean and easy with VHDL.

Cheers,
Jim
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jim Lewis
Director of Training mailto:[email protected]
SynthWorks Design Inc. http://www.SynthWorks.com
1-503-590-4787

Expert VHDL Training for Hardware Design and Verification
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 

[Tom Hawkins]

Gee Tom, that looks like a shameless marketing plug for your product.

Yes, it was a shameless plug for a tool that is open-source (GPL).
Here's another one: http://www.launchbird.com/

Does confluence allow you to forget about hardware?

No. On the contrary, Confluence forces you to take an even closer
look at your hardware structures and the algorithms that generate
them.

The key is the fundamental paradigm shift between HDL and Confluence:
An HDL design describes a single structure, where as a Confluence
design is an algorithm to generate an entire class of structures.

That seems to be the main issue here. Once you sort
out what hardware you care to implement, it is
clean and easy with VHDL.

This is true assuming you know the hardware you need to implement.
But here in lies the problem. As Steve clearly states, "I just don't
know how many different components are trying to access the memory
until I use it."

Confluence solves this problem by providing more mechanisms for reuse.
Users can design the generic case up front, then instantiate a
specific configuration once the requirements have solidified.

-Tom

Cheers,
Jim
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jim Lewis
Director of Training mailto:[email protected]
SynthWorks Design Inc. http://www.SynthWorks.com
1-503-590-4787

Expert VHDL Training for Hardware Design and Verification

Jim, your tag-line should read:

Expert Confluence and VHDL Training for Hardware Design and
Verification

;-)

 

[Jim Lewis]

Tom,

No. On the contrary, Confluence forces you to take an even closer
look at your hardware structures and the algorithms that generate
them.

I am glad to see we agree that you can't forget about
the hardware. When one forgets about the hardware, they
get what they deserve.

Once we figure out the nature of the hardware and what to
do for a fixed number of iterations it is fairly straight
forward to genericize it with VHDL.

The problematic places of this problem are the arbiter (select
active device) and the memory multiplexer (select address and
data to memory for the active device).

The arbiter has request and grant signals. Each of these can
use generics an array with the width that matches the number
of devices in the system. The width would be specified by
a generic.

The memory multiplexer is a little more tricky as the objects
that need to be multiplexed are the address and data busses.
We need a couple of types defined to be an unconstrained
array of an array type. One to hold address and one to hold
data.

At this point I think I have done enough of the homework
problem that was posted.

This is not just theoretically do-able in VHDL, I have done it.
It is not difficult, you just need to be organized and
structure the problem right (as you would have to do for any
solution - even confluence).

Expert Confluence and VHDL Training for Hardware Design and
Verification

Hart to tell. I am a strong typing fan. I don't like
debugging in the simulator - I prefer to debug in the
compiler.

Cheers,
Jim
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jim Lewis
Director of Training mailto:[email protected]
SynthWorks Design Inc. http://www.SynthWorks.com
1-503-590-4787

Expert VHDL Training for Hardware Design and Verification
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 

[Tom Hawkins]

[snip]

Hart to tell. I am a strong typing fan. I don't like
debugging in the simulator - I prefer to debug in the
compiler.

I agree completely. Static typing is an incredibly powerful, yet
often undervalued method for uncovering bugs. I wouldn't have been
able to get Confluence to where it is today without the static type
system of Objective Caml.

Confluence is a dynamically typed language from the software
standpoint. However, in light of the "generate" paradigm, running a
Confluence program to produce a netlist is still part of the
compilation process. Since the run-time environment handles type
checking, and the run-time is part of the netlist compilation,
Confluence can be considered a staticly typed language.

In other words, if you can successfully compile a Confluence program
to a netlist, you have strict guarantees of the resulting netlist
quality.


-Tom

 

[Andreas Hinze]

Hi,

I agree completely. Static typing is an incredibly powerful, yet
often undervalued method for uncovering bugs. I wouldn't have been
able to get Confluence to where it is today without the static type
system of Objective Caml.

Yes, but in OCaml typing is very convinient due to type interference
and polymorph functions. Doing the same hack in i.e. C++ will probably
drive you crazy.

And i assume that Confluence would not be where it is without the other
nice features that OCaml brings to you ;-)

However, from my experience bugs from odd typing are not very hard to catch.
Much more ugly are "dynamic" bugs, i.e. a wrong assumption about any system
constraint (typical "This will never happen"). To get rid of these i like to
have a bunch of monitors and condition handlers in my code that tries to fix
it or, at least, halts the program gracefully (i.e. saving the data and write
a stack trace).

Just my $0.01 ;-)

Kind regards
AHz