Re: "Subtract C, add Ada"

[ronalds60@aol.com (RonaldS60)]

Gary Frankel (gfrankel@rational.com) wrote
>As of late 1989, I was involved in a commerical Ada optimizing 
>compiler... We published results on a wide variety of benchmarks
>(Hennessey,etc) that showed that the Ada compiler was producing
>better code than any commercially available C compiler at that time
>and far better than most C compilers.  This was true for workstations
>and cross compilers.
>
>I cant vouch to the same degree for the current state of the art
>since I havent done the same studies recently, but I believe
>that you will find tha the optimizing Ada compilers can produce
>comparable or better code than C compilers.  I have also noticed
>that a number of C++ compilers produce poorer code - probably due
>to the current level of maturity.

About two years ago a IEEE article (in one of the s/w engineering
journals) compared the percentages of dead code (I assume using some
reachability algorithm) between Fortran, C, and C++.  The numbers
were suprising (to say the least) and indicated that while Fortran
had the least dead code (followed by C) C++ could have upto 60%-75%
dead code.  The exact quantity being undecidable in a static analysis
due to virtual function calls.  This research hasn't had any obvious
effect on the C++ crowd (or any of the OO people) that I have seen.

>Generally, an optimizing Ada compiler is actually at an advantage over
>C compilers due to Ada sources providing more useful type information
>and the aliasing problems of C.  Ada compilers will tend to 
>compile more slowly since they do considerably more checking than
>C compilers.

One of the problems with aliasing is covered in an article titled 
'The Undecidability of Aliasing' by G. Ramalingam in ACM Transactions
on Programming Languages and Systems, vol. 16 no. 5, September 1994.

>So, I think that anyone who does a careful study will simply not
>find that C has any advantages in code efficiency over Ada at all
>(at least on any program more than a few short lines).

Actually there was an article around 1989 to 1991 covering the 
difference between two programs (that did the same thing) written in
C and Ada.  They were both compiled on the best compilers that could
be found.  C won for overall speed, by less than 2% when runtime 
checking was disabled (as it could be when the program was validated.)
A 1% to 2% difference in speed isn't much price to pay when the
overall development cycle is shortened (especially in testing) as
much as Ada allows (when compared to C.)  Just think how nice it
would have been if that company in Redmond had used Ada instead of
C (oooh, software that actually works! and on time! ;{))).

>-- Gary Frankel
>
>
>Mitch Gart (mg@asp.camb.inmet.com) wrote:
>: Jay Martin (jmartin@oahu.cs.ucla.edu) wrote:
>
>: : I read the abstracts to these papers before I posted, there is not
>: : one peep about the relative efficiency between C and Ada which
>: : should be the central topic of discussion in my opinion.
>
>: I wrote a paper on this, which is now pretty dated, but still might
>: possibly be of interest:
>
>: "Benchmarking Ada, C, and Pascal", SigAda conference, July 1986,
>: Pittsburgh.
>
>:  Mitch Gart


Ron
======================================================
Computers now make it possible to make more mistakes
more efficiently than ever before possible.
======================================================