IEEE P1076.1 Working Group - Table-Driven Modeling Project - Telco Meeting Minutes of August 9, 2010

Status: approved on September 13, 2010.

Attendees

• Ernst Christen
• Joachim Haase (chair, meeting convener)
• Alain Vachoux (note taker)

Agenda

• Call to order
• Approval for agenda
• Review and approve minutes from July 5 meeting
• Review action items and any content provided from them
• Discussion of TDM Requirements document
• Next meeting
• AOB
• Adjourn

Call to order on August 9, 2010 at 8am PDT (5pm CEST)

Joachim welcomes the participants and starts presenting the meeting slides.

Volunteer for minutes

Alain volunteers to take notes.

Approval for agenda

Agenda unanimously approved.

Minutes from last meeting

Minutes of July 5, 2010 meeting unanimously approved.

Current action items

Ernst:

• Look at the requirements and consequences of supporting multiple dependent variables. - done
• Writeup on the use of Verilog-AMS control strings with different language definition versions and tools. - done

Discussion (see $table_model in Verilog-AMS):

• There are conflicts between the Verilog-AMS LRM and the Cadence implementation. It seems that the LRM has different interpretations.
• Multiple dependent variables may be useful for semi-conductor models. One function call with N outputs may be replaced by N function calls with one output. Recommendation is to first support single-output lookup functions and to evaluate the cost and the gain to support multiple-output functions later.
• Complex dependent variables should be considered as a single dependent variables.
• Verilog-AMS control string should be implemented as enumeration types. Verilog-AMS capabilities are mostly sufficient.
• All combinations of interpolation/extrapolation methods make sense, except non-spline interpolation (linear, nearest) with spline extrapolation. Can we consider linear interpolation as first-order spline interpolation? If yes, only one non-compatible combination would remain.
• Interpolation and extrapolation can be done at elaboration time (fixed preparation/optimization tasks) and at run time (e.g., actual table lookups).
• The same extrapolation method as the one used for interpolation should be the default case. Update requirement document.

Joachim:

• Details on the Matlab 4 storage format. - done

Discussion (see Tables in Matlab):

• The Matlab format defines a way to describe matrices. This format can be interpreted without problem in the TDM package.
• Other waveform formats from Simulink could be also investigated.
• The two-step approach for spline interpolation is an example of splitting tasks between elaboration and run time.

Alain:

• Update requirement document. - version 0.3 published on web site

Discussion:

• We should finalize the document for submission to the WG. Only minor changes may need to be done later.
• Plan to present the requirement document to WG at FDL meeting on September 14. Then proceed to a WG vote.
• Version 0.4 to include a few changes from discussion. Also requirement item must be consistently renumbered.

Next steps and next meeting

The next step is to prepare the material for presentation to the WG and the following vote. The material will be the requirement document and additional details (context, rationales) that are useful to explain the proposed requirements. The form for additional information could be slides or separate text documents.

There was also a discussion on the direction to go:

• Developing a better understanding of interpolation/extrapolation algorithms. This may have impact on the second item.
• Developping a use model (file format, elaboration vs. run time tasks, architecture of the VHDL-AMS package, VHDL-AMS data structure and function prototypes). This is not related to the algorithms, but it is closely related to the VHDL-AMS package(s) to be standardised. Package body(ies) and internal data structures will be only a proof-of-concept implementation. It is likely that vendors will develop different implementations. There is then the question to avoid reinventing the wheel. Joachim proposes to look at existing software packages such as FITPACK.

The next telco is planned on September 13, 2010 at 8:00am PDT or 5:00pm CEST.

AOB

None.

Adjourn

Joachim adjourns the meeting at 6:45 CEST.

Action items

Alain:

• Update requirement document from the meeting discussion.
• Prepare the material for WG presentation and vote.

Joachim:

• Prepare the material for WG presentation and vote.
• Provide references to FITPACK software package.