From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me
X-Spam-Level: *
X-Spam-Status: No, score=1.2 required=5.0 tests=BAYES_00,FROM_WORDY,
	INVALID_MSGID autolearn=no autolearn_force=no version=3.4.4
X-Google-Language: ENGLISH,ASCII-7-bit
X-Google-Thread: 103376,fd3a5ba6349a6060
X-Google-Attributes: gid103376,public
From: "Nick Roberts" <Nick.Roberts@dial.pipex.com>
Subject: Re: should I be interested in ada?
Date: 1999/02/23
Message-ID: <7av53r$goc$2@plug.news.pipex.net>#1/1
X-Deja-AN: 447623662
References: <7a72e6$g55$1@probity.mcc.ac.uk> <36C93BB4.1429@ecs.soton.ac.uk>
 <7afc1o$3mi$2@plug.news.pipex.net> <7afttr$7v3$1@nnrp1.dejanews.com>
 <7aganu$qsc$1@plug.news.pipex.net> <36CC3AEA.59E2@lanl.gov>
 <7ai502$6an$1@nnrp1.dejanews.com> <36CD8DBA.237C@lanl.gov>
 <7akvao$j5t$2@plug.news.pipex.net> <36D1D862.41C6@lanl.gov>
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Organization: UUNET WorldCom server (post doesn't reflect views of UUNET
 WorldCom)
Newsgroups: comp.lang.ada
Date: 1999-02-23T00:00:00+00:00
List-Id: <comp.lang.ada>

William Clodius wrote in message <36D1D862.41C6@lanl.gov>...
[...]
|>    procedure Reorder (Object: in out Vector; Index: in Selection) is
|>       Temp: constant Vector := Object;
|>    begin
|>       for i in Index'Range loop
|>          Object(Index(i)) := Temp(i);
|>       end loop;
|>    end;
|> <snip>
|
|Even neglecting parallelization issues, the above code does not catch
|the semantics of the FORALL loop, in particular the FORALL does not
|require that either X or J be dimensioned 1:N, only that:
|
|1. the index range include 1:N;
|2. X's range include all values of J(1:N); and
|3. the values of J(1:N) not have any duplicates.
|
|As a result Temp as declared above is assumed to be larger than the
|requirements might actually be, and Index might range past the desired
|bounds.


Its true that the Ada code doesn't capture the same semantics - essentially
that no component of Object will be assigned more than once (and they can be
assigned in any order) - but I'm not sure that this will make any difference
in practice (for a non-parallel target).

The Ada procedure is inlined, so the optimizer may have the opportunity to
eliminate null operations (e.g. assignment to self). I think Temp being too
long will often not be a problem (for a clever dataflow analysing
optimiser). Anyway, the problem can always be got around 'by force', by
passing one or two more parameters into the procedure.

As for the 'parallelization issues', possibly one or two pragmas might help
for SIMD machines (as suggested by another poster recently, I think).

   pragma Assume(Parallel_Update,Object);
   pragma Assume(Spanning_Bijection,Index);

This is, admittedly, a bit hypothetical, since (to my knowledge) there is no
Ada compiler targetting a SIMD architecture, and no immediate prospect of
there being one.

Parallel processor targets can be catered for, in Ada, by encapsulating a
multiple-task solution inside a procedure. I have constructed such a thing
myself (for matrix inversion), as a 'technology demonstrator'. Not neat, but
effective.

|William B. Clodius Phone: (505)-665-9370
|Los Alamos Nat. Lab., NIS-2     FAX: (505)-667-3815
|PO Box 1663, MS-C323    Group office: (505)-667-5776
|Los Alamos, NM 87545            Email: wclodius@lanl.gov

-------------------------------------
Nick Roberts
-------------------------------------