Toon said:
{I've snipped, but history will tell you what.]
>What did happen was that Fortran 90 was such a large change over FORTRAN
>77 that vendors started to write their compilers from scratch (or at
>least their front-ends).
>
>That might have had an effect. I know for sure that my younger brother
>stuck to F77 on the Cray he had access to because it vectorized certain
>constructs that the Fortran 90 compiler didn't.
>
>If I confine myself to the one compiler I actually know something about
>- g77 - and its Fortran 95 cousin, I can say with almost certainty that
>the performance of the generated code on 64-bit targets will be worse
>with the latter than the former. The reason for this is that the former
>contains hacks in the front-end to make up for deficiencies in the
>back-end with regards to the formation of mixed-mode induction variables
>(array addresses are 64-bit quantities, but the indices are 32-bit
>INTEGERs).
>
>Unless this is repaired in the back-end before g95 hits the archives,
>you'll definitely see this effect.
Yes, and F2K or 2002/3/4 will be another large change, and again the vendors
will write from scratch to Fortran++. More benchmarking!!
The initial differences betwixt F77 and F90 were vectorising. The other things
that came in and gave a difference of about 14 years from the advantages that I
wanted became a frustration.
If I wanted an OO type language, I'd want it now. I don't, I want something to
number-crunch with efficiently, otherwise I'd go into my dreaded area of C and
C++. And I want a stable language, not one full of traps and dangerous
constructs. This is what I feel that Fortran will devolve to in its next
upgrade/incarnation from all our vendors.
We have seen enough threads here and in c.l.f deploring the efficiency of our
vendors' compilers. Worse compilation than user F77 DO loops. I exclude my
vendor here, and (from another thread) they do give a qualifier ( or -something
on UNIX) that tells you if a copy has been done.
Regards, Paddy
|