From jls@uxb.liv.ac.uk Tue Feb 9 10:28:15 1993 Received: from ib.rl.ac.uk by dkuug.dk with SMTP id AA29010 (5.65c8/IDA-1.4.4j for ); Tue, 9 Feb 1993 11:29:07 +0100 Received: from mail.liv.ac.uk by ib.rl.ac.uk (IBM VM SMTP V2R1) with TCP; Tue, 09 Feb 93 10:28:29 GMT Received: from uxb.liverpool.ac.uk by mailhub.liverpool.ac.uk with SMTP (PP) id <25522-0@mailhub.liverpool.ac.uk>; Tue, 9 Feb 1993 10:28:35 +0000 From: jls Message-Id: <446.9302091028@uxb.liv.ac.uk > Subject: HPF 1.0 comments, chapter 4 To: hpf-comments@cs.rice.edu Date: Tue, 9 Feb 93 10:28:15 GMT Cc: SC22WG5@dkuug.dk (SC22/WG5 members) X-Mailer: ELM [version 2.3 PL11] X-Charset: ASCII X-Char-Esc: 29 Comment on HPF 1.0, Jan'93 Further Comments on Chapter 4, Statements In a previous comment on this chapter I suggested that the PURE and INDEPENDENT directives would be much better as Language syntax and that they be more tightly tied to the entity they qualify. I have not changed my mind on this but I can appreciate why at this stage it is considered desirable for the comment nature of these directives to be preserved. However, I feel strongly that if at some future date HPF functionality is to be included in the official Fortran language standard then the correct linking of these directive to the entity qualified is critical. I would therefore like to suggest a compromise position. Since Fortran 90 allows ! initiated comments to appear inline in both source forms, I would propose that both the PURE directive and the INDEPENDENT directive be recast as inline comments that qualify the proper defining statement for the entity in question. For the PURE directive this would mean that it would always appear as a trailing comment on the qualified FUNCTION or SUBROUTINE header statement. The PURE directive does not then need to specify the procedure name since this is provided by the header that is involved. Subsequent incorporation as a part of the language proper will then be simple and the modification if necessary of any existing programs trivial. Similarly, the INDEPENDENT directive should be attached to the DO loop it qualifies as a trailing comment. Again as it applies to a single loop not to a nest it does not need to specify the index; this is specified elsewhere in the statement. I strongly believe this qualification of a nest by qualifying the individual loops to be essential because the DO statements setting up each loop of a nest may be separated from each other by many lines of code. In fact there is no need for there to be only a single inner loop nested within an outer. The following situation is perfectly allowable Fortran, DO i1=1,n1 ..... DO i2=1,n2 ..... ENDDO ..... DO i2=1,m2 ..... ENDDO ..... ENDDO The independence or otherwise of these loops depends on the code contained within their individual loop bodies. It would be perfectly possible for the loop over i1 to be independent but for one of the nested loops over i2 not to be independent. This also underlines that index variable name is not a suitable name for indentifying a loop. The only name that has this function is the loop construct name which is by definition a unique identifier for a loop. If the intent of the nested loop single independent directive is intended to mean that the distinction between inner and outer loop is also of no significance, then qualifying the DO loop syntax is a highly undesirable vehicle for such an assertion. It would probably be better to look at the FORALL construct as the vehicle in this case or some similar but new syntactic entity. Although, the general aim of the facilities defined in this chapter is reasonable and reasonably clear, the specific details of the syntax and semantics as set out in the current draft are far from adequate. There are too many problems, too many loop-holes caused by a lack of precision in the syntax definitions and the semantic rules. The functionality being proposed would be very unlikely, in my opinion, suitable for inclusion in a future Fortran standard without substantial alteration. It is clearly not desirable for an important extension proposal like HPF to become widespread unless there is a good prospect of it being incorporated without major changes in a future Fortran standard. -- Dr.J.L.Schonfelder Director, Computing Services Dept. University of Liverpool, UK Phone: +44(51)794 3716 FAX : +44(51)794 3759 email: jls@liv.ac.uk