Re: Regarding support of wreal

From: Kevin Cameron <kevin_at_.....>
Date: Tue Aug 08 2006 - 00:13:07 PDT
Bresticker, Shalom wrote:

>wreal is fundamentally different from the digital domain.
>In the digital domain, each signal is either 0 or 1. 
>A number is represented by a set of bits (= set of signals) large enough
>to represent the number.
>So the number 5 is represented by 3 bits/signals, the first and last of
>which are 1, the middle one is 0.
>But these are three different wires.
>You may be able to group these wires together into a vector, but they
>remain three wires.
>
>In contrast, wreal (as I understand it) is a single wire with a
>continuous value set (quantized, of course). This is analog SPICE-like
>logic, not digital. The whole idea of digital logic is to abstract
>voltage levels into logical 0 and 1.
>
>  
>
It's only Spice-like if it's piecewise linear (pwl), if it's dicrete 
it's "digital" (or event-driven) simulation.

>The whole idea of resolving two real entities in the digital domain is
>not relevant.
>  
>
I don't think it's unreasonable to want to do something like sum 
currents at a node using a resolution function. There is a lot of analog 
behavioral modeling you can do without using a solver.

VHDL let you have user-defined types like wreal (resolution optional) 
before analog was added.

Kev

>I'm not sure I am explaining this clearly, but I hope the idea comes
>across.
>
>I can access eda.org .
>
>
>Shalom
>
>  
>
>>-----Original Message-----
>>From: owner-verilog-ams@server.eda-stds.org [mailto:owner-verilog-
>>ams@server.eda-stds.org] On Behalf Of Sri Chandra
>>Sent: Tuesday, August 08, 2006 7:52 AM
>>To: Jonathan David
>>Cc: Kevin Cameron; Verilog-A Reflector; Martin O'Leary
>>Subject: Re: Regarding support of wreal
>>
>>Hi all,
>>
>>I would prefer that we keep the discussion on the reflector purely
>>technical - as to what is broken in the language today, what we need
>>    
>>
>to
>  
>
>>do to fix it or if people feel that there is not enough information
>>about a particular feature defined in the language for them to
>>    
>>
>implement
>  
>
>>it correctly. Of course, as part of this we need to understand the
>>original underlying requirement - what was missing pre-2.0 LRM that
>>necessitated the addition of the new syntax, and whether its required,
>>restrictive or better alternatives available?
>>
>>Coming to wreal query, im my opinion i think it needs to be in the
>>digital syntax - given that its a digital/mixed signal feature (and
>>    
>>
>can
>  
>
>>be assigned only in the digital domain), i would think that it should
>>    
>>
>be
>  
>
>>part of the 1364 std. But today 1364 does not allow ports of type
>>    
>>
>real.
>  
>
>>So we need to take this with digital committee in the true spirit of a
>>mixed signal common language for them to support it (if required). If
>>there are good reasons for them not to support it or if it is already
>>supported through existing syntax in digital then we need to
>>    
>>
>reconsider
>  
>
>>the value of additional syntax that we define in AMS.
>>
>>Probably SystemVerilog probably addresses this problem partially by
>>having something called user defined types, but again i am not sure
>>whether they have defined clearly how resolution is performed on this
>>type. I think Graham suggested a while back that wreal should just be
>>treated as another discrete net type with an associated discipline and
>>we can do discipline resolution appropriately.
>>
>>Dave,
>>Well there are only two places for you to look for digital and mixed
>>signal features - AMS LRM2.2 or 1364-2005, if there are other
>>    
>>
>documents
>  
>
>>in which its documented it should be part of these stds :-). Probably
>>you can post an email to the digital std. Some of the contact people
>>    
>>
>for
>  
>
>>1364/SV are sharp@cadence.com (Steven Sharp), Michael Mcnamara
>>(mac@verisity.com) or shalom.bresticker@intel.com (Shalom Bresticker).
>>
>>Is it just me who cant access eda.org, eda-stds.org or verilog.org?
>>
>>cheers,
>>Sri
>>
>>Jonathan David wrote:
>>    
>>
>>>I'm all for BETTER solutions..
>>>
>>>I don't see the problem about the resolution function..
>>>If I connect wreal to logic by accident.. I don't want the
>>>      
>>>
>simulation
>  
>
>>to run..
>>    
>>
>>>the design is wrong. fix the design,
>>>then there are no more resolution issues..
>>>
>>>Generally I can't afford to have these signals in the Continuous
>>>      
>>>
>Time
>  
>
>>domain..
>>    
>>
>>>any other Suggestions.. ?
>>>
>>>In this case, no solution would leave VHDL as the only available
>>>      
>>>
>>solution.
>>    
>>
>>>Not all nets are something you could swap a logic driver with..
>>>MY complaint about the wreal solution is that CADENCE won't let me
>>>put 10 wreals in a bus (wreal [9:0] signal_p;)
>>>it was nearly the first AMS bug I reported on learning about wreal..
>>>and they still haven't fixed it..
>>>Any better AMS simulators out there?
>>>Jonathan
>>>
>>>----- Original Message ----
>>>From: Kevin Cameron <kevin@sonicsinc.com>
>>>To: Verilog-A Reflector <verilog-ams@eda.org>
>>>Cc: Martin O'Leary <oleary@cadence.com>
>>>Sent: Monday, August 7, 2006 2:12:04 PM
>>>Subject: Re: Regarding support of wreal
>>>
>>>
>>>Martin O'Leary wrote:
>>>
>>>      
>>>
>>>>(another perspective on the wreal feature)
>>>>
>>>>Wreal part of the Verilog-AMS standard not the Verilog standard and
>>>>        
>>>>
>>has
>>    
>>
>>>>been there since LRM2.0,
>>>>released in 2000.
>>>>
>>>>It has proven popular with users and is supported by multiple
>>>>        
>>>>
>>implementors.
>>    
>>
>>>>        
>>>>
>>>- on the principal that a bad solution is better than no solution.
>>>
>>>The reason that it is unnecessary is that wires should be neutral
>>>      
>>>
>and
>  
>
>>>have their type/discipline derived at elaboration be looking at the
>>>drivers/contributions. A wreal is essentially a wire with a driver
>>>      
>>>
>of
>  
>
>>>type real (a non-resloved type). As far as I remember you can't
>>>      
>>>
>>connect
>>    
>>
>>>wreal to anything else so it breaks the plug-and-play approach
>>>      
>>>
>>intended
>>    
>>
>>>in the original design of the language - i.e. you should be able to
>>>      
>>>
>>swap
>>    
>>
>>>a "digital" module with a real driver for an analog module with a
>>>contribution and have it handled automatically.
>>>
>>>Kev.
>>>
>>>      
>>>
>>>>Thanks,
>>>>--Martin
>>>>
>>>>-----Original Message-----
>>>>From: owner-verilog-ams@eda-stds.org
>>>>[mailto:owner-verilog-ams@eda-stds.org] On Behalf Of Kevin Cameron
>>>>Sent: Monday, August 07, 2006 1:14 PM
>>>>To: Dave Miller
>>>>Cc: verilog-ams@eda-stds.org
>>>>Subject: Re: Regarding support of wreal
>>>>
>>>>
>>>>wreal was a (unecessary) Cadence hack that only exists in Verilog-
>>>>        
>>>>
>>A[MS].
>>    
>>
>>>>Hopefully it will go away some day.
>>>>
>>>>Kev.
>>>>
>>>>Dave Miller wrote:
>>>>
>>>>
>>>>        
>>>>
>>>>>Hello all,
>>>>>I am a bit confused about the support of the net type 'wreal'. I
>>>>>          
>>>>>
>see
>  
>
>>>>>that it is included in the AMS syntax that Graham has done but I
>>>>>          
>>>>>
>>can't
>>    
>>
>>>>>          
>>>>>
>>>>        
>>>>
>>>>>find any mention of 'wreal' in 1364-2001, or 1364-2005. Is 'wreal'
>>>>>          
>>>>>
>a
>  
>
>>>>>type that is only valid in the digital domain (i.e. can only
>>>>>          
>>>>>
>assign
>  
>
>>to
>>    
>>
>>>>>          
>>>>>
>>>>        
>>>>
>>>>>it in digital), if so why can't I find it in the digital LRM's or
>>>>>          
>>>>>
>am
>  
>
>>I
>>    
>>
>>>>>          
>>>>>
>>>>        
>>>>
>>>>>looking in the wrong place?
>>>>>Thanks for any help,
>>>>>
>>>>>Regards
>>>>>Dave
>>>>>
>>>>>
>>>>>
>>>>>          
>>>>>
>>>>
>>>>        
>>>>
>>--
>>Srikanth Chandrasekaran
>>DTO Tools Development
>>Freescale Semiconductor Inc.
>>Ph: +91-120-439 7021 F: x5199
>>
>
>
>  
>
Received on Tue Aug 8 14:52:27 2006

This archive was generated by hypermail 2.1.8 : Tue Aug 08 2006 - 14:52:29 PDT