[ibis-macro] Re: Question about clock_times
- From: "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx>
- To: "IBIS-ATM" <ibis-macro@xxxxxxxxxxxxx>
- Date: Thu, 28 Jan 2010 09:21:14 -0800
Mike, Kumar,
Thanks for your replies. Even if we don't change how the AMI
interface works regarding the clock_times vector memory allocation,
I think we should explain it better in the next version of the spec
(in the AMI BIRD Walter is writing).
The problem I see with the current wording is that the spec doesn't
spell out the rules and actions clearly, and because of that we
can run into situations where things can fail or crash...
1) The spec should say that the EDA tool must initialize the
vector with -1 (at least in the first location or the whole vector).
2) Depending on how this was done, he spec should say something
about what the DLL is expected to do. If the EDA tool initialized
the whole vector to -1, the DLL doesn't have to do anything. But if
the EDA tool puts a -1 in the first location only, than the DLL
will have to write a -1 at the end.
3) The spec should spell out that the DLL is not required to fill
the vector if it doesn't have clock recovery.
4) If we don't say that the EDA tool must initialize the vector
than the DLL should be required to write a -1 in the first location
(at least) of this vector to tell the EDA tool that the vector is
empty, even if it is a Tx DLL or an Rx DLL without a clock recovery.
If we don't spell these rules out, we can end up with an uninitialized
vector with no -1 retuned by a DLL. What is the EDA tool going to
do with that? Can't decide whether the content is garbage, or good
data...
Arpad
======================================================================
-----Original Message-----
From: Mike Steinberger [mailto:msteinb@xxxxxxxxxx]
Sent: Thursday, January 28, 2010 10:06 AM
To: Muranyi, Arpad
Cc: IBIS-ATM
Subject: Re: [ibis-macro] Re: Question about clock_times
Arpad-
I think Kumar's done a perfectly adequate job of covering the subject. I
don't see any improvement in going beyond what he's already said.
As far as memory consumption goes, consider:
1. How much memory are we talking about anyway? block_size*8 bytes.
2. How big is a typical block? 1k? 10k? So maybe we're talking about
100kB of memory.
3. How long does this memory have to stay allocated? The duration of the
GetWave call plus a little bit of post processing in the function that
called it.
I don't see the clock_ticks array as a very good candidate for
optimization. Again, Kumar's solution is good enough.
Mike S.
Muranyi, Arpad wrote:
> The point I was trying to make is that if the EDA tool initializes
> the entire vector with -1, then the DLL doesn't have to put a -1
> at the end, because it is there already by default. The DLL can
> just write data and stop, and the next value will be -1 anyway.
> Of course if we don't require for the EDA tool to initialize the
> entire vector with -1, then the DLL should put the -1 at the end.
> Either way, the spec should say something about the EDA tool
> having to initialize this vector one way or another, since it
> appears that there may be situations when the DLL doesn't write
> anything to it, not even a -1...
>
> Regarding #2, I can see your point about simplifying the rules for
> the API, but I think this can waste a lot of memory, especially if
> clock_times length = wave length. If we were able to make other
> arguments optional, like AMI_parameters_out, why can't we figure
> out a way to make this one optional, or at least say something
> about the fact that for Tx, for example, it is not necessary to
> allocate a huge amount of memory, only enough for a single number
> containing a -1...
>
> Arpad
> ====================================================================
>
> -----Original Message-----
> From: ckumar@xxxxxxxxxxx [mailto:ckumar@xxxxxxxxxxx]
> Sent: Thursday, January 28, 2010 8:58 AM
> To: Muranyi, Arpad
> Subject: Re: [ibis-macro] Re: Question about clock_times
>
> 1."putting the value at the end" is required only if he dll sends
clock
> back.
> If it is not c=doing anything with the clock it does not even need to
> look at
> it.
>
> 2. This is mainly done to simplify the API. Rule of thumb is you
shoudl
> keep th
> API to minimum for documentation and programming simplicity. If it is
> optional
> then we have to create another API without clock.
>
> Quoting "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx>:
>
>
>> Kumar,
>>
>> Thanks for the answers.
>>
>> If everyone agrees to what you said, then I would suggest to write
>> this into the specification.
>>
>> #1) The spec doesn't say anywhere that this is supposed to be done
>> by the EDA tool, it only says:
>> "The last clock is indicated by putting a
>>
> value
>
>> | of -1 at the end of clocks for the current wave sample. The
>>
> clock_time
>
>> | vector is allocated by the EDA platform and is guaranteed to be
>> greater
>> | than the number of clocks expected during the AMI_GetWave call."
>>
>> If the EDA tool initializes the vector with a -1, the DLL wouldn't
>>
> have
>
>> to do anything about that, so we wouldn't need to say:
>>
>> "by putting a value of -1 at the end" because it would be there
>>
> already.
>
>> On the other hand, we should add "The clock_time vector is allocated
>>
> by
>
>> the EDA platform AND INITIALIZED WITH -1 at least in the first
>> location..."
>> or something like that...
>>
>> #2 If this vector is not needed for the Tx, shouldn't clock_times be
>> optional (like AMI_parameters_out, for example)? Currently it is
>> required... I find it a little confusing to have something required
>> when it is not used in half the cases. It makes me wonder what am I
>> missing that it is required for...
>>
>> Thanks,
>>
>> Arpad
>>
>>
>
========================================================================
>
>> ======
>>
>> -----Original Message-----
>> From: ckumar@xxxxxxxxxxx [mailto:ckumar@xxxxxxxxxxx]
>> Sent: Thursday, January 28, 2010 5:47 AM
>> To: Muranyi, Arpad
>> Subject: Re: [ibis-macro] Re: Question about clock_times
>>
>> Arpad:
>>
>> 1. The EDA tool allocates the clock memory and initializes that with
a
>> -1.
>> If the Rx does not have a CDR, it will not modify this vector
>>
>> 2. For the Tx , whether the clock vector exists or not is irrelevant.
>>
> It
>
>> should
>> not be doing anything to that vector. (like logical don't care state)
>>
>> 3. clock length = wave length is a robust way of memory allocation.
It
>> is much
>> more important to be robust than worry about 'efficiency' of memory
>> allocation
>>
>> Quoting "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx>:
>>
>>
>>> A related question:
>>>
>>> Is an Rx DLL required to fill the clock_times memory
>>> with good numbers, i.e. does it have to have a clock
>>> recovery, or is even an Rx allowed to basically not
>>> return anything useful?
>>>
>>> Related to this, are the Tx and Rx DLL-s required to
>>> put a -1 in the first location of the clock_times
>>> memory space if they do not have anything to return,
>>> or is the EDA tool supposed to initialize the allocated
>>> memory so that at least the first location would have
>>> a -1?
>>>
>>> Thanks,
>>>
>>> Arpad
>>> =======================================================
>>>
>>> ________________________________
>>>
>>> From: ibis-macro-bounce@xxxxxxxxxxxxx
>>> [mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
>>> Sent: Wednesday, January 27, 2010 11:44 PM
>>> To: IBIS-ATM
>>> Subject: [ibis-macro] Question about clock_times
>>>
>>>
>>> Hello AMI experts,
>>>
>>> I just noticed something in the AMI portion of the official IBIS v5
>>> specification.
>>>
>>> Section 3.2.2.3 clock_times is not labeled as optional, yet the text
>>> implies that this is only applicable to Rx models. Note the words
>>> "clock recovery loop" towards the end:
>>>
>>>
>>> | 3.2.2.3 clock_times
>>> |
>>> | Vector to return clock times. The clock times are referenced to
>>>
> the
>
>>> start
>>> | of the simulation (the first AMI_GetWave call). The time is always
>>> | greater or equal to zero. The last clock is indicated by putting a
>>> value
>>> | of -1 at the end of clocks for the current wave sample. The
>>>
>> clock_time
>>
>>> | vector is allocated by the EDA platform and is guaranteed to be
>>> greater
>>> | than the number of clocks expected during the AMI_GetWave call.
>>>
> The
>
>>> clock
>>> | times are the times at which clock signal at the output of the
>>>
> clock
>
>>> | recovery loop crosses the logic threshold. It is to be assumed
>>>
> that
>
>>> the
>>> | input data signal is sampled at exactly one half clock period
>>>
> after
>
>> a
>>
>>> | clock time.
>>>
>>>
>>> A Tx model would not have a clock recovery loop, so I would think
>>>
> that
>
>>> a Tx GetWave function would not return anything in clock_times. Is
>>>
>> the
>>
>>> EDA tool still required to allocate memory for clock_times even for
>>>
>> Tx?
>>
>>> Also note that the spec doesn't spell out exactly how much memory
>>>
> the
>
>>> EDA
>>> tool needs to allocate, only that it should be "greater than the
>>>
>> number
>>
>>> of clocks expected". Since we have seen crashes due to a DLL
>>>
> wanting
>
>> to
>>
>>> write more clock_times than what we "expected" from it, and the
>>>
> safest
>
>>> thing to do would be to allocate as much memory as the number of
>>> simulated
>>> bits, this seems to be quite a waste if we had to allocate the same
>>> amount
>>> of memory to a Tx model's clock_times also...
>>>
>>> Isn't there a better way of doing this?
>>>
>>> Thanks,
>>>
>>> Arpad
>>>
>>>
>
========================================================================
>
>>> =====
>>>
>>>
>>
>> ---------------------------------------------------------------------
>> IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
>> IBIS Macro reflector: //www.freelists.org/list/ibis-macro
>> To unsubscribe send an email:
>> To: ibis-macro-request@xxxxxxxxxxxxx
>> Subject: unsubscribe
>>
>>
>>
>
>
>
> ---------------------------------------------------------------------
> IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
> IBIS Macro reflector: //www.freelists.org/list/ibis-macro
> To unsubscribe send an email:
> To: ibis-macro-request@xxxxxxxxxxxxx
> Subject: unsubscribe
>
>
---------------------------------------------------------------------
IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
IBIS Macro reflector: //www.freelists.org/list/ibis-macro
To unsubscribe send an email:
To: ibis-macro-request@xxxxxxxxxxxxx
Subject: unsubscribe
Other related posts:
