I side with SiSoft on this, favoring separate DLLs and AMI files.
Also, I'd like to challenge Walter's concession that constructing a unified DLL
is "trivial", by asking the following question:
Using only the existing API, how would the EDA tool indicate to a unified DLL
whether its call of the lone Init() function in that DLL was intended to be
routed to the Tx or Rx model?
-db
From: ibis-macro-bounce@xxxxxxxxxxxxx [mailto:ibis-macro-bounce@xxxxxxxxxxxxx]
On Behalf Of Walter Katz
Sent: Tuesday, March 17, 2015 3:05 AM
To: michael.mirmak@xxxxxxxxx; IBIS-ATM
Subject: [ibis-macro] Re: Questions for tomorrow's meeting
MM,
First, I point out that if a model maker can write separate DLLs for the Tx and
Rx, then it is a trivial exercise to write a third unified DLL that calls
either the Tx or Rx DLL depending on the status of the In parameter
Model_Direction. So there is no valid "programming difficulty" argument that
should prevent the unified DLL approach.
There are much more compelling arguments to require separate AMI files and DLLs
and not allow unified DLLs:
1. We would need to deal with things like Ignore_Bits - can it be
different for Tx and Rx.
2. Which parameters apply when operating as a Tx vs an Rx. When doing a
solution space analysis (either a blind sweep, or an intelligent
co-optimization), which parameters are Tx only, which are Rx only and which are
both?
I also point out that if someone wrote a unified DLL, it would be trivial to
deliver a single DLL and two AMI files, that to an EDA tool looks the same a
separate AMI and DLL files.
In summary, the clear choice for SiSoft is separate AMI and DLL files (allowing
separate AMI files and a single DLL) for the following reasons:
1. For the Model Maker
a. Gives him the most flexibility in developing and maintaining the
DLL(s) as he chooses to.
b. Eliminates the need to identify which parameters control Tx and which
Rx.
2. For the EDA tool
a. Eliminate the need to identify which parameters control Tx and which
Rx.
3. For the IBIS Standards Committee (us)
a. Eliminate the need to identify which parameters control Tx and which
Rx.
b. The only hard part is that in the [Algorithmic Section] there will be
two executable statements with different .ami files and maybe different .dll
files (per OS).
i. The EDA
tool would need to read each .ami file to determine which is Tx and which is Rx.
1. In this case Model_Direction needs to be a Reserved Parameter
ii. We could
make this a bit simpler for the EDA tool by instead of adding Model_Direction
to the .ami file, we created separate Executable keywords for Tx and Rx.
1. In this case Model_Direction is either not required, or would be a
Model Specific "In" parameter of format Value if the two .ami files shared a
common DLL.
4. Finally, this can be supported by EDA vendors today by using Model
Selector, and then using the new Executable or Model Direction keyword when
supported in the spec.
Finally, any of the proposed methods would work, I think the ultimate decision
is what would be most successful in the market - least complexity in the
specification, easiest for the model makers, easiest for the EDA tools, and
mechanisms to get adaptation fastest.
Walter
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Mirmak, Michael
Sent: Monday, March 16, 2015 11:42 PM
To: IBIS-ATM (ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>)
Subject: [ibis-macro] Questions for tomorrow's meeting
For tomorrow's IBIS-ATM meeting, I will be raising two questions and, most
likely, making a motion.
1) To support Directional (I/O) IBIS-AMI model sets, which DLL option
shall we adopt?
* Permit one, unified DLL and, alternately, separate DLLs per direction?
Model authors can create a single DLL with both TX and RX functions.
* Permit only separate DLLs, with individual DLLs for each direction per
[Algorithmic Model] OS and architecture (e.g., 1 is TX and 1 is RX)?
2) To support Directional (I/O) IBIS-AMI model sets, which .ami file
option shall we adopt?
* Permit one, unified .ami file and, alternately, separate .ami files
per direction? Model authors can create a single .ami with both TX and RX
parameters.
* Permit only separate ,ami files, with individual .ami files for each
direction (e.g., 1 for TX and 1 for RX)?
Thanks!
- MM
