The Math Package Study Subgroup (Par 1076.2) From: Jose Torres To: Interested Parties Date: June 20, 1994 Subject: 6/10/94 meeting SUMMARY ======== This meeting took place at San Diego (DASC meetings following DAC) from 8:00 a.m. until 2:00 p.m.. The meeting had 3 parts: the first part consisted of a general presentation to DASC about the objectives, status, schedule, and open issues; the second part consisted of a working session to resolve issues and assign action items to the active members of the group; and the third part consisted of a Q&A session for people interested to get additional information on this standard. The first session was attended by active participants of all DASC working groups. The participants for the working session were: Jose Torres, jose@synopsys.com Chuck Swart, Mentor, chuck_swart@mentorg.com Alex Zamfirescu, Intergraph, azamfire@edaca.ingr.com The participants for the third session were: John Hines, Wright-Patterson AFB, hines@el.wpafb.af.mil Dave Ackley, Intel, dackley@ichips.intel.com Kiyoshi Makino, Seiko Instruments, Jose Torres PRESENTATION TO DASC GROUPS =========================== During this session, the objectives, status, proposed schedule, deliverables, and open issues were presented. A summary follows: 1. Objectives - A set of standard VHDL Mathematical Packages that include most oftenly used real and complex elementary functions, and required data types and type conversion functions. 2. Status - A reflector account has been set up (math@vhdl.org). If you wish to be added to the list, please send message to math-req@vhdl.org - Current activities are in the areas of building a test bench, fixing problems reported by early reviewers, and establishing a ballot constituency 3. Schedule - Current schedule calls for establishing a ballot constituency and the package documentation by 9/94, package distribution and validation by 12/94, balloting from 10/94-2/95, ballot review from 12/94-3/95, IEEE approval by 6/95. 4. Deliverables - The deliverables will be two VHDL packages (math_real and math_complex -- to reside in library IEEE) and their documentation, and an appendix with a test bench. The test bench will not be part of the standard. 5. Issues - Man power to finish validation test bench - Accuracy of some algorithms - Man power for documentation - Needs of Analog Group - Shall distribution start w/o test bench? - Mechanism to handle requests for extensions or additional packages The following inter-group issues were raised during this session: - Need to set up a common format with the synthesis package working group for error messages - Need to coordinate with Analog working group on type names for double precision and location for definitions of common data types and functions. WORKING SESSION ================ The main objective of this session was to address the above mentioned issues, review the schedule, and identify action items. Here is a summary of the decisions/action items: 1) Man power to finish validation test bench - We agreed on the following: a) Jose, Chuck, and Alex will check with their companies for support (funds) to hire a contractor to finish the template for the test bench ==> AI for Jose, Chuck, and Alex by end of June b) After the template is done, ask for volunteers (either companies or individuals) to complete the test bench for all the functions in the packages ==> Any volunteers for this task are welcomed. We have 70+ functions for which we need to put together a test bench Jose to send note on the net for volunteers by mid-July c) Jose will convert the VMS ANSI FORTRAN reel to reel tape into a format readable by UNIX machines. ==> AI for Jose by mid-July 2) Accuracy of some algorithms - We agreed on the following: a) Alex will review the cordic algorithms being used and determine if they can be fixed or a different algorithm must be used. ==> AI: Jose to send current version of package to Alex Alex to review algorithms by mid-July b) Chuck to convert the C implementation of logarithmic function into C. ==> AI: Jose to send C version to Chuck Chuck to convert C version to VHDL by mid-July c) As the test benches ar put in place, if additional problems are detected in other functions, then assign an owner and deadline for fixes. 3) Man power for documentation - We agreed on the following: a) We will use as a template the documentation for IEEE 1164. (conventions, contents, index, glossary). ==> AI: John Hines to get electronic version of the documentation (from Jim Balunga?) and send it to Jose by end of July b) Chuck will check with Rita C. and Randy Harr for their availability for this purpose. ==> AI: Chuck to check with Rita and Randy by end of July. c) John Hines mentioned that we need to adhere to new IEEE standards for documentation (sgml.dtd -- mil std 28000) ==> AI: John Hines to get documentation on this standard (from Al Sicker? or Scott Calhoun) and send it to Jose by end of July d) Check with Paul M. for balloting guidelines ==> AI: Jose to check with Paul by mid July 4) Needs of Analog Group - We agreed on the following: a) Check if it is possible to reach and agreement with the Analog working group on the definition of data type names and location of such definitions by mid-August ==> AI: Jose to contact Jean Michel Berge by mid-July b) Check with Analog working group if all the basic functions they need (independent of precision) exist in the current proposal. ==> AI: Jose to check with Analog group by mid-July. c) Create an additional set of packages to cover for the double precision data type that may be required by Analog, when the definition for the new data types is available. These packages would overload the math package functions for the double precision data types. 5) Shall distribution start w/o test bench? - We agreed on the following: a) Start general distribution of the package for validation when the working group feels confident enough on the quality of results even if the test bench is not complete. ==> AI: none 6) Mechanism to handle requests for extensions or additional packages - We agreed on the following: a) Wait for the balloting process and use such feedback to decide if additional functions should be incorporated into these packages or in a different package ==> AI: none 7) Need to set up a common format for error msgs with the synthesis package - We agreed on the following: a) Jose to check with Alex on formats and coordinate activities in the two groups to see if a common format is possible. ==> AI: Jose and Alex to coordinate this issue by end of July.