SOME EARLY UK FORTRAN COMPILERS

I attended the Fifty Years of FORTRAN meeting and was pleased to see how the early prominence of the UK in FORTRAN use, development and standardisation has continued to the present day. This note provides some further detail on some of the pioneering work only hinted at the 1984 Annals of Computing published by AFIPS.

One speaker, Ron Bell of AWE described what was probably the first major UK FORTRAN compiler, designed by Alick Glennie for the IBM 7030 (Stretch). At about the same time AERE Harwell was preparing to take delivery of a Ferranti Atlas, which was the only computer at that time comparable in power to Stretch. AERE decided to write their own compiler under the leadership of Dr. Ian Pyle and I, as a member of the Ferranti Atlas team joined them in 1961 to write the code generation and optimisation phase. Ian maintained steady contact with Alick Glennie during development, although the design differed, as did the aims.

Atlas featured a 'one-level store' (a concept blurring the physical distinction between RAM and drum storage) which provided a very large address space. The AERE FORTRAN was basically FORTRAN II with some features from the emerging FORTRAN IV, but also added arrays whose size could be defined at run time rather than at compilation, utilising that address space. This was the first ever availability of dynamic arrays.

The Atlas Operating System provided only basic batch processing capability, albeit with multitasking of jobs, and no-one seemed aware of the magnitude of prevailing FORTRAN programs and the need to compile subroutines independently. Consequently apart from writing the compiler we also had to write a subroutine linking system, as well as the run-time routines. Nevertheless there were only 5 people in the team, although a sixth joined later.

Another interesting feature of the Harwell compiler is that it was written almost entirely in FORTRAN. This enabled it to be largely tested before Atlas itself was available. The testing took place on an IBM 7094 at Risley, which meant a 2 day turn around on every test! A great advantage was that a loadable and tested version of the compiler could be generated on the 7094 and taken en masse to Atlas.

Testing eventually transferred to the Atlas prototype in Manchester University but the only time available was overnight. The testing of our compiler and loading system could only really get under way after I had tested the card reader and punch which no-one else was interested in. Inputting some 3000 binary punched cards certainly provided a very thorough test of reliability! Anyway, testing on Atlas progressed well and all was largely ready when the AERE Atlas was delivered (to adjacent Chilton). I then transferred the completed Harwell FORTRAN system to the Atlas Computers at the Universities of Manchester and London.

Later, AWRE took delivery of a Ferranti Atlas 2, which although using the same processor as Atlas had a different operating system oriented towards time sharing. Alick Glennie provided an S3 compiler, which was essentially language compatible with the S2 Stretch compiler. I supported that by porting the Harwell loading system. I then transferred the complete system to Titan, the Atlas 2 prototype, at Cambridge University, in early 1966.

Yet another FORTRAN compiler was written nearby for the Ferranti Orion computer at the Rutherford Laboratory at Chilton by Dr. Bob Taylor and team from the staff there, again in the early 60s.

Brian Chapman MA, CEng, FBCS, CITP

Comments on this or any other of the Group's pages should be sent by e-mail to the FSG Web Editor.