BRITISH COMPUTER SOCIETY - FORTRAN SPECIALIST GROUP
Minutes of a meeting held on 5th June 1978 at the Polytechnic of Central London,
309 Regent Street, London.
M. R. Lewis (Chairman) Cray Research Inc
R. J. Collins UMRCC
C. Graff RAE Farnborough
G. L. Harding ECMRWF
A. P. Hayes UMRCC
D. Hill CCA
D. J. Holmes Rolls Royce Ltd, Bristol
G. L. Horsnell Imperial College
P. Loftus Marconi-Avionics (GEC)
J. D. Murchland University College, London
T. L. van Raalte AWRE
T. R. H. Sizer RAE Farnborough
F. Smithin RAE Farnborough
A. W. Stewart Honeywell Information Systems NISD
J. D. Wilson University of Leicester
J. M. Roberts-Jones (Secretary) Liverpool City Council
Apologies for Absence:
P. A. Clarke Rothamsted Experimental Station
D. T. Muxworthy Edinburgh RCC
1. Approval of Minutes
The minutes of the meeting of 10th April 1978 were approved.
Mostyn Lewis announced that he was resigning as Chairman because of pressure
Alan Clarke was nominated to succeed as Chairman, and it being reported that he
had given his consent, the nomination was seconded and approved unanimously. A
It was further reported that Dr A. Colin Day was resigning from the steering
committee because of pressure of work.
The Secretary was asked to circulate an up-to-date list of the Officers of the
Group (See Appendix D).
2. Matters Arising from the Minutes
There were no matters arising.
3. Activities of Other Fortran Groups
The ANSI Board of Standards Review approved Fortran 77 as an American National
Standard on 3rd April 1978. The name "Fortran 77" is retained as an informal
designation, although the standard is in fact X3.9-1978. Final copy editing by
ANSI is under way; it is expected that publication will be in July or August.
4. Working Party Reports
No reports were available.
5. Other Recent Fortran Events
The BSI working group (BSI/DPS13/WG6), convened by David Muxworthy, is soliciting
comments on the next round of Fortran revision in preparation for the November ISO
meeting. The working group has requested this Group's co-operation in distribut-
ing a questionnaire (See Appendix B).
6. Retirement of Frank Engel Jr from X3J3
The Group noted with regret that Frank Engel Jr, who was Chairman of X3J3 from
1970 to 1977, had been expelled from the committee because of failure to meet the
stringent attendance requirements. It was resolved to record the Group's appre-
ciation of his contribution to the development of Fortran. John Murchland agreed
to draft the text (See Appendix A).
The Secretary was asked to circulate details (See Appendix C).
8. Other BCS Business
It having been reported that Pauline Halters had now left BCS Headquarters, it was
resolved to record the Group's appreciation of her assistance with our administra-
9. Any Other Business
Mr A. W. Stewart noted with astonishment a comment in the April 1978 issue of
Forward that "no known complete Fortran 77 compilers are available to date"; he
felt that the Honeywell compiler was one such.
Dr J. D. Murchland described various aspects of "pretty-printing" and drew
attention to a number of relevant papers in Sigplan Notices.
Mr D. Hill summarised the progress of the DOD Revised Ironman project, which has
now been refined to two (red and green) of the four Pascal-based candidate
languages. The status of Pebbleman in providing a working environment for Ironman
was also described.
There was an inconclusive discussion of various proposals for statement separators.
Dr J. D. Murchland presented arguments in favour of the adoption of "Fortran
Substandards", referring to Captain Grace Hopper's reported position on Cobol.
Mr G. L. Harding drew attention to a CDC internal document giving guidance on the
transition from current compilers to Fortran 77 compilers.
It was mentioned that various bodies were developing supersets of Fortran 77. The
US Department of Defense is specifying a DO-WHILE construct and various bit-
manipulation facilities. The US Department of Energy is considering asynchronous
transput and NAMELIST. A discussion ensued as to the desirability of strict
enforcement of standards. In this context, it was reported that because of the
stack-oriented nature of the ICL 2900 series architecture, problems could arise in
transferring programs which used non-standard techniques such as reliance on
argument retention which were not detected as illegal by earlier compilers.
10. Next Meeting
The date of the next meeting was agreed as 4th September 1978. However, it has
since been discovered that this conflicts with the BSC/DBAWG conference, and the
date has been changed to Monday 21st August 1978.
11. Talk on the Cray-1 Fortran Compiler
Irene Q. Mallgrave gave a talk describing progress with the Fortran compiler
being developed for the Cray-1 processor.
She first gave an overview of the hardware, describing the use of vector
registers to achieve parallel manipulation of up to sixty-four array elements
and explaining the concept of chaining, whereby an array expression of the form
(A op1 B) op2 C,
where op1 and op2 are binary operators corresponding to distinct functional units,
could be evaluated with overlapping of the two operations as well as of the
elements. The compiler is being developed by a team of six, one of whom is res-
ponsible for debugging aids. The main difference from a compiler for a more
conventional architecture arises from the loop analysis needed to detect loops
that can be replaced by vector operations.
The overall design aim is to maximise hardware efficiency in executing
standard-conforming Fortran programs. A number of immediate goals are achieving
Fortran 77 conformance, scheduling vector operators to exploit chaining, improving
debugging tools (a cross-reference analysis has been implemented and the next step
is provision of a symbol table to the run-time support for use in post-mortem
analysis), and improving detection of vectorisable operations.
The compiler is essentially two passes. The first pass performs parsing and
syntax analysis. It also deletes statement function definitions and expands
references in-line in order to improve vectorisability. The second pass identifies
"blocks" (a "block" appears to be a maximal sequence of graphically contiguous
executable statements the last of which is either an unconditional transfer of
control or a labelled statement), performs a vector analysis of each block (which
appears to correspond to loop detection, motion of loop-invariant code and a
process similar to strength reduction replacing loops of scalar operations by
vector operations) and generates intermediate code which is transformed by
scheduling for chaining and by a register assignment analysis.
A description was given of the way in which source programs could be trans-
formed in order to improve detection of vectorisable loops.
[No appendices are available for digitization. Following on from Appendix A,
Frank Engel Jr was awarded honorary membership of the BCS in a ceremony in London
in December 1978. The questionnaire in Appendix B is part of the BSI submission
to the ISO Fortran Experts meeting which is appended to the February 1979 minutes.
Appendix D would have been similar to the contact details for officers for the
following year which are attached to the April 1979 minutes.]