Re: Ada in a C++ Interview

Re: Ada in a C++ Interview

[Jim Showalter]

yow@sweetpea.jsc.nasa.gov (Bill Yow) writes:

>it turns out that the only people that are doing Ada 
>programming of Crays happpen to be oil companies.

Well, this is hardly something to be ashamed of (assuming this
is true)--commercial organizations using Ada by _choice_ are
even better than DoD folks using Ada. This is especially nice
in this particular case because I keep hearing that "only FORTRAN"
is suitable for vectorizing stuff on Crays (obvious malarkey, since
Ada is inherently vectorizable [can you say "tasking"], but what
the heck).

>The DoD guy was saying, "Yeah, we don't do any Ada programming."  
>And I said, "Wait a minute. Isn't it your langauge?" And he said, 
>"Yeah, but all of our people got exceptions, so we don't have to."  
>Nobody wants to use it inside the DoD.

>Any comments?

Yeah: the guy from the DoD is an ignoramus. If anything, the Ada
mandate continues to get stronger every year, and the proportion of
DoD software written in Ada continues to increase at the expense of
other languages. Furthermore, the number of waivers continues to
drop, heading toward zero as the number of lame excuses dries up
("There are no compilers", "There are no tools", "It's not fast
enough", etc).
-- 
*** LIMITLESS SOFTWARE, Inc: Jim Showalter, jls@netcom.com, (408) 243-0630 ****
*Proven solutions to software problems. Consulting and training on all aspects*
*of software development. Management/process/methodology. Architecture/design/*
*reuse. Quality/productivity. Risk reduction. EFFECTIVE OO usage. Ada/C++.    *

Re: Ada in a C++ Interview

[Mike Lijewski]

In article <1991Aug06.220725.4010@netcom.COM> jls@netcom.COM (Jim Showalter) wr
ites:
>yow@sweetpea.jsc.nasa.gov (Bill Yow) writes:
>
>>it turns out that the only people that are doing Ada 
>>programming of Crays happpen to be oil companies.
>
>Well, this is hardly something to be ashamed of (assuming this
>is true)--commercial organizations using Ada by _choice_ are
>even better than DoD folks using Ada. This is especially nice
>*reuse. Quality/productivity. Risk reduction. EFFECTIVE OO usage. Ada/C++.    
*
Vectorizing and tasking are orthogonal.  I've vectorized and parallelized
plenty of code.  The first rule is always to make the code as highly
vectorized as possible.  Only then do you worry about parallelizing the
code.  The rationale being that parallelization always introduces overhead
of its own, so you want the code to run as fast as possible before taking
that step.  My experience is that well vectorized code is both more
efficient and maintainable, even on scalar machines.  So the question is,
are there any Ada compilers which produce highly vectorized code, on say
Crays, IBM 3090s or any of the other vector machines?

-- 
Mike Lijewski  (H)607/272-0238 (W)607/254-8686
Cornell National Supercomputer Facility
ARPA: mjlx@eagle.tc.cornell.edu  BITNET: mjlx@cornellf.bitnet
SMAIL:  25 Renwick Heights Road, Ithaca, NY  14850

Re: Ada in a C++ Interview

[Boris Pelakh]

In article <1991Aug7.121826.20660@batcomputer.tn.cornell.edu> lijewski@theory.T
C.Cornell.EDU (Mike Lijewski) writes:
> ....  My experience is that well vectorized code is both more
>efficient and maintainable, even on scalar machines.  So the question is,
>are there any Ada compilers which produce highly vectorized code, on say
>Crays, IBM 3090s or any of the other vector machines?
>

pragma BEGIN_COMPANY_SPEAK;

The Convex Ada compiler will produce vector code that is alsmost as good
as the one produced by our FORTRAN compiler. The main problem with
vectorizing Ada code is the fact that the exception handling mechanism
produces more branching possibilities within loops, thus inhibiting
optimization. BTW, we also parallelize our programs either by spreading
tasks across processors or parallelizing on the loop level.

pragma END_COMPANY_SPEAK;

I feel like a sales critter. *Cold Chill*

--

Boris Pelakh			"With a tough and smart lawyer, you can get 
Development Software Test	 two to three times the cash and benefits you 
pelakh@convex.com		 deserve !"	- Personal Injury Att. Ad

Re: Ada in a C++ Interview

[Chuck Shotton]

In article <1991Aug7.121826.20660@batcomputer.tn.cornell.edu>, lijewski@theory.
TC.Cornell.EDU (Mike Lijewski) writes:

[Stuff deleted]

> plenty of code.  The first rule is always to make the code as highly
> vectorized as possible.  Only then do you worry about parallelizing the
> code.  The rationale being that parallelization always introduces overhead
> of its own, so you want the code to run as fast as possible before taking
> that step.  My experience is that well vectorized code is both more
> efficient and maintainable, even on scalar machines.  So the question is,
> are there any Ada compilers which produce highly vectorized code, on say
> Crays, IBM 3090s or any of the other vector machines?
> 
I feel that I'm significantly ignorant of parallel processing to ask the
following FAQ:

Exactly what is involved in "vectorizing" code? Particularly, Ada code.

Re: Ada in a C++ Interview

[Greg Titus]

In article <1991Aug06.220725.4010@netcom.COM> jls@netcom.COM (Jim Showalter) wr
ites:
>... I keep hearing that "only FORTRAN"
>is suitable for vectorizing stuff on Crays (obvious malarkey, since
>Ada is inherently vectorizable [can you say "tasking"], but what
>the heck).

Better yet, can you say "loop"?   ;-)   Any language with looping
constructs (quick!  name one without!), whether implicit or explicit
ones, is potentially vectorizable.  You actually may get a break
vectorizing Ada, because the language is such that more information
is available to the compiler regarding the possible values of
variables at runtime than it is in, say, Fortran or C.

[This thread started with someone reporting a DoD person as having
claimed that nobody in DoD was using Ada on Crays.  I won't comment
on customer activities (that's their business), but the CRI PR office
may be able to provide more accurate information on the subject.]

greg
--
--------------------------------------------------------------
Greg Titus (gbt@zia.cray.com)             Compiler Group (Ada)
Cray Research, Inc.                               Santa Fe, NM
Opinions expressed herein (such as they are) are purely my own.