optimization was: python was: transputers again was: Commodore

I lucked out and got a copy of Wulf via comp.arch gratis from a guy who had 2 copies. Optimization went underground in the 80s only to resurface later..

I would not call any of those compilers "highly". I think FTN V was poo. CDC FTN (and I knew one of the compiler writers who pbutted away recently pretty well) wasn't particularly optimizing. And even Cray Fortran (cft, less cft77) was antiquated by 1982.

One big dirty secret is that unlike compiler validation optimization doesn't have test suites: only small samples of user programs figured to be useful. It is conceivable, and I know from personal experience, that small scale features are merely buttumed to work (and I've found cases where they didn't).

The Japanese did their homework, and they really started to push optimization, then the companies like Convex and Alliant (Kuck certainly gets credit with that). I am not certain whether Elxsi (the Indians) maybe have also done their homework. Also the first US compilers got top heavy being written in buttembler. This is one reason why Cray was so smart to rewrite cft77 in Pascal (which initially had slower code than cft) it was the maintenance of that compiler which improved it (another reasons why CDC never caught up. Then IBM got around to catching up in the later 80s 88-89. Another problem was end user reluctance to shift out of F66 which is much harder to optimize.

Later I found a great quote from Bill Wulf about problems done in the name of optimization (blind desire for speed).

--