Christian,
I used to work on an LGP-30 over 37 years ago and was fortunate in
having ACT V as the complier for the machine. You will see Graeme
Scott's name in the first part of the ACT V compiler tape on your
website. After joining IBM in 1969 I happened to encounter Graeme
working for Bell Northern Research in Ottawa in 1972 or 1973. He
mentioned that he had written ACT V in a rather short time as he was
frustrated at how long it took to do things in ACT III. He turned out
what I still believe was a masterful piece of software that exploited
the LGP-30 hardware capability to its maximum. You will see that the
compiled code branches to highly optimized run-time subroutines in the
ACT V Run Time portion of the package. The compiled code executed at a
very reasonable rate for such a slow machine as every floating point
operation ran optimized. The compiler itself used HASH coding to
determine if a variable was a reserved word in the language and if not,
was stored away in the hash table of variables. Optimal Assembly
routines were devised for most drum memory machines like the LGP-30, IBM
650 and Bendix G-15. The ACT V compiler made assembly coding on the
LGP-30 unnecessary. It was hard to outperform the compiled ACT V code.
Compile times for ACT V were about 1/3 that of ACT III. And with the
Load & Go compiler, you didn't need to reload the run-time routines [or
the compiler after execution]. The 4096 word machine seemed reasonably
sized for many problems.
Unfortunately I had lost any and all LGP-30 documentation many years
ago but if you put up the ACT V material you have on your web-site, I
can probably offer some additional comments as it will likely jog my
brain cells that haven't thought about these things for many years.
Paul Tardif [former LGP-30 programmer and technician]
--
J.Paul Tardif
17 Lyall Avenue
Toronto, Ontario
CANADA M4E 1V7
(416) 694-5309 FAX: 416 694-2381
Received on Fri Jul 09 2004 - 00:13:00 BST