Project 63: Semantics of Numeric Computation in Programming Languages.

Argonne 1986:
=============

The project was proposed by Feldman. It will consider questions like "What do the 
language definition and standards demand?", "Do we recommend changes?", "How the 
real compilers match standards?", "Do we recommend changes or fixes?" etc.

Examples of the problem:

Sanctity of the parentheses and functional symbols 

			(a-b)+e = (a+e)-b 
			  (a)+a = a+a

Precision of intermediates

			if (a = b) print (a-b)


Como 1987:
==========

Feldman noted that there was almost no testing of semantics in
programming languages. Dekker pointed out that Brown's model was
available for Ada. Ford said that there were two excellent papers on
scientific Ada and Ada semantic specifications in the "The Ada
Companion Series", B. Ford ed., Cambridge Univ.  Press. Specification
and semantics for elementary functions and arithmetic in C, Fortran 8x
and languages in general were also discussed (Hull, Feldman, Gaffney).

Stanford 1988:
==============

S. Feldman reported on the current state of the project. The
presentation was followed by a discussion (Smith. Gentleman, Ford.
Dekker. Vouk, Reid. Einarsson).

SIAM Mini-symposium on Language Facilities for Scientific Computing.
--------------------------------------------------------------------

T. Hull reported on the forthcoming SIAM mini-symposium on language
facilities for scientific computing. It will take place in San Diego,
CA, July 10-14, 1989. J. Rice offered to organize a section of PSE's.
Discussion followed (Paul. Hull. Cody. Ford, Gentleman).

Beijing 1989:
=============

There was no report, but a general discussion was held. Paul noted
that some numerical floating point aspects had to be relaxed to get
new Fortran standard moving. Arithmetics have been ordered according
to precision. Ordering becomes peculiar if there are two types of
arithmetic (binary and hex). The same is true for different exponent
ranges. However, the general semantics of expression evaluation has
been kept. Mixing of precisions was discussed. Kulisch pointed out
that a version of Pascal-SC offers different precision by block and by
variable, while in Fortran-SC it is possible to adapt precision via an
index which can be changed according to needs (accuracy, condition
number). Precision properties of Numerical Turing were also discussed.
(Paul, Kulisch, Einarsson, Vouk, Dekker, Stetter, Fosdick).

Jerusalem 1990:
===============

Projects [44] and [54] were merged into this project. 

Toronto 1992:
=============

Exception Handling in Fortran 90 

Document:	IFIP/WG 2.5 (Toronto-15) 1915, 5 pages,

J. Reid initiated and led the discussion on exception handling in
Fortran. He presented a particular approach based on definition of
ENABLE and HANDLE statements (see IFIP/WG 2.5, Toronto-15).
Differences between the approach proposed by Reid, and some other
suggested approaches, was noted. The group expressed support for the
effort and Reid, Hull and some other members of the group are expected
to be involved (Einarsson, Sorensen, Ford, Rice, Hanson). Electronic
mailing facility will be established to facilitate communications
related to this topic.

Kyoto 1995:
===========

Document: Kyoto-2207

J. Reid's proposal for exception handling was discussed.
Synchronization with C committee is probably a good idea.
 
IFIP WG 2.5 would like to thank J. Reid for his efforts. WG 2.5
endorses introduction of exception handling in FORTRAN.

Interval Arithmetic 
-------------------

Document: Kyoto-2215

The Kulisch-Walter proposal on interval arithmetic was discussed.
IFIP WG 2.5 endorses the effort and we would like to be informed of
the progress.

Amsterdam 2001:
===============

Interoperability of Numeric Functions 
-------------------------------------

WG11 of ISO/SC22 is the organization that is attempting to develop
standards for interoperability among various programming languages.
They are seeking our advice and/or support on a number of issues. The
current status is that a standard, LIA-1 has been approved, LIA-2
(which deals with the accuracy of numerical functions) is close to
approval, and LIA-3 (which extends the standard to complex-valued
numerical functions) is just being developed. Walter has followed very
closely the activities of this group and has attended their last two
meetings. He feels they need and would welcome any advice we can
provide. A handout was distributed describing in more detail their
recent related work. Members were encouraged to respond directly with
any concerns or suggestions. Boisvert agreed to respond formally on
behalf of our group. Walter will continue to provide liaison with
WG11.