From J.Reid@directory.rl.ac.uk Thu Dec 24 12:40:38 1992 Received: from danpost4.uni-c.dk by dkuug.dk with SMTP id AA27021 (5.65c8/IDA-1.4.4j for ); Thu, 24 Dec 1992 11:39:09 +0100 X400-Received: by mta danpost4.uni-c.dk in /PRMD=minerva/ADMD=dk400/C=dk/; Relayed; Thu, 24 Dec 1992 11:38:33 +0100 X400-Received: by /PRMD=uk.ac/ADMD= /C=gb/; Relayed; Thu, 24 Dec 1992 11:38:57 +0100 X400-Received: by /PRMD=UK.AC/ADMD= /C=GB/; Relayed; Thu, 24 Dec 1992 11:25:33 +0100 X400-Received: by /PRMD=UK.AC/ADMD= /C=GB/; Relayed; Thu, 24 Dec 1992 11:40:38 +0100 X400-Received: by /PRMD=UK.AC/ADMD= /C=GB/; Relayed; Thu, 24 Dec 1992 11:40:38 +0100 Date: Thu, 24 Dec 1992 11:40:38 +0100 X400-Originator: J.Reid@directory.rutherford.ac.uk X400-Recipients: SC22WG5@dkuug.dk X400-Mts-Identifier: [/PRMD=UK.AC/ADMD= /C=GB/;<9212241040.AA01031@numerical.cc] X400-Content-Type: P2-1984 (2) Content-Identifier: Note on LIA-1... From: " (John Reid)" Sender: J.Reid@directory.rl.ac.uk Message-Id: <9212241040.AA01031@numerical.cc.rl.ac.uk> To: JLS@liverpool.ac.uk, MEllis@vax.ox.ac.uk, SC22WG5@dkuug.dk, baw@seg.npl.co.uk, d.t.muxworthy@edinburgh.ac.uk Subject: Note on LIA-1 to appear in IMA Numerical Analysis Newsletter X-Charset: ASCII X-Char-Esc: 29 LANGUAGE INDEPENDENT ARITHMETIC (LIA) - A DRAFT INTERNATIONAL STANDARD (DIS) John Reid, Rutherford Appleton Laboratory The UK vote has been decided on the draft international standard ISO/IEC CD 10967-1:1992: Language independent arithmetic - part 1: Integer and floating point arithmetic. The stated goals of the draft are: 1. To enhance the portability of numeric programs across a wide range of numeric architectures. 2. To help programming languages express the semantics of their numerical types. It defines integer and (normalized and denormalized) floating point types. It specifies the primitive computational operations with source and destination operands of the same type, comparison operations on two operands of the same type, and conversions from any arithmetic type to any other arithmetic type. The values of parameters define the exact arithmetic behaviour of any particular platform. At its meeting on December 8, BSI IST/5 decided to vote YES with some comments. The resolution is attached as an appendix to this note and was passed nem-con. I was invited to address the meeting and criticized the draft for not fully supporting the IEEE standard for binary floating point arithmetic. The increasing popularity of the IEEE standard is a very encouraging trend of recent years and is a very real help to those of us writing portable software. LIA is intended to provide "a bridge between the language and the underlying arithmetic hardware" . It appears to me that the current draft cripples IEEE arithmetic by not supporting NaNs ("Not-a-Number"s), infinities, signed zeros, invalid exceptions, divide-by-zero exceptions, inexact exceptions, rounding modes, and trapping modes. The suggestion in last paragraph of the UK comments of an informative annex containing a specific binding to the IEEE standard was added to meet this objection. I said at the time that I felt that such an annex would be helpful, but on reflection have decided otherwise. Such an annex would either fail to provide some important IEEE facilities or would represent an extension of the new standard (and those of us writing software to standards have learnt that extensions should be avoided). The draft standard already contains a parameter that specifies whether denormalized values are supported, so I can see no reason for not including other parameters for other aspects of the IEEE standard or for an IEEE parameter. This would allow us to write clean codes that do not port beyond the IEEE implementations, and less elegant code of wider applicability. It would positively encourage better support of the IEEE facilities on IEEE implementations, which is certainly needed. It would also encourage the further proliferation of the use of the IEEE standard, which, after all, is also an ISO standard. If you wish to read the draft, it is available as postscript or compressed postscript by anonymous ftp: Node: crl.dec.com Directory: pub/misc File: lia-1-v40-cb.ps (569K bytes) File: lia-1-v40-cb.ps.Z (216K bytes) IFIP WG 2.5's collection of commentary is available by anonymous ftp from Node: osl.csc.ncsu.edu (152.1.58.11) Directory: pub/wg25 The ANSI public review period ends on January 12, 1993, which is probably before you receive this newsletter. If you wish to comment, send hardcopy to: X3 Secretariat Attn: Monica Vago Suite 200 - 1250 Eye Street, NW Washington, DC 20005 My experience is that late comments, though not of the same force, are passed on. If you wish to comment on the UK position, write to Dr Andrew Walker, IST/5 Chairman, Standards Manager, X/Open Company Ltd., Apex Plaza, Forbury Road, Reading RG1 1AX email: andrew@xopen.co.uk Appendix. The IST/5 resolution ------------------------------ The UK, noting that there are no known significant technical objections to the content of this document, and believing that it is valuable to users of programs and important to the development of Programming Language standards votes YES to DIS registration of this document, with the following comments. Comments The following changes to the Introduction, under Goals, should be made: 1. In the first goal replace the two occurrences of "portability" with "predictability". Rationale: Portability, in a bit-wise sense, is not achievable. The proposed change better describes this aim of the standard, and is achievable. 2. Replace the second sentence of the first goal with: "The focus is on assuring users of single programs in a specific programming language of a documented level of accuracy of arithmetic operations on which they can rely, and the portability of such programs across platforms." Rationale: This changes gives needed further emphasis to the value of the standard to users of programs, whether or not they are ported across platforms. 3. In the second paragraph replace "Our second goal" with "The second goal of this international standard". For the benefit of many numerical users, LIA-1 should be accompanied, in an informative annex, by a specific binding to IEC 559 (IEEE 754). In addition, definitions should be adapted so that they are consistent with IEC 559 (IEE 754).