Re: ratioanl number type

[Robert Dewar <robert_dewar@my-deja.com>]

In article <s57e138d53112@corp.supernews.com>,
  "Vladimir Olensky" <vladimir_olensky@yahoo.com> wrote:

> This technique was used for more than 20 years in Russian
> supercomputers "Elbrus". So no wonder that one of the
> leading scientists from "Elbrus" team (Pentkovsky) was invited
> to the Intel and he lead the development of SIMD extensions
> for Intel chips.

Well there is absolutely nothing new conceptually in the SIMD
extensions (or in the Elbrus), these techniques are very old
and very well known and have been for 30 years. Indeed the jury
is still out on whether such extensions are worthwhile. The ia64
should help to answer that question. Remember that the Elbrus
team as well as similar contemporary architectures were very
much under the CISC philosophy (and of course Intel still is,
indeed I would really call the ia64 a CISC design, full of
crufty stuff [e.g. long offset arithmetic available only in
4 of the 128 registers].

So the point is not the general approach, but rather,
specificaly for the ia32 extensions:

a) whether it is useful in the context of an optimizing Ada
compiler. Answer: very dubious, almost certainly there is a
long list of better optimization opportunities for any existing
compiler. Obviously you can find some test cases where pattern
matching will do nice things, but I doubt as a general
optimization it will have a noticable affect on applications
in general.

b) Whether it would help a hand written multi-precision integer
package. Again I think dubious. Vladimir, assuming you have some
experience with multi-precision arithmetic packages, why not
try writing some critical inner loops, and timing them both
ways. Vague conjecture here is not very convincing :-)



Sent via Deja.com http://www.deja.com/
Before you buy.