Re: Migrating to Ada : a mini success story

[Georg Bauhaus <rm-host.bauhaus@maps.futureapps.de>]

On 2/13/10 11:04 PM, Hibou57 (Yannick Duchï¿œne) wrote:
> Hello once again,
>
> Here is a new quote about Ada, in another interesting context (no more
> LP this time) :
> http://gmarceau.qc.ca/blog/2009/05/speed-size-and-dependability-of.html

> The interesting context, is that this comment was one after an article
> containing interesting graphic charts comparing some measurable language
> characteristics. Unfortunately, Ada is not part of any of these charts,
> but there is this comment, and the charts which may be nice for your own
> knowledge anyway.


There is a chart "Ada 2005 GNAT" on the linked page, though.

*

The Shootout GCC compiler appears to have been updated to version 4.4,
recently.
- The good news:
As a consequence, the chameneos-redux Ada program is working
again.
- The bad news:
As a consequence, the chameneos-redux Ada program is working
again.   It runs 16 times slower than the best Java
entry.  Apparently Ada's median has moved from 1.xx to 2.xx.
I believe this is related.
(I'm not criticising, the program has been present for
a long time.  It obviously is performing better than its
compiler :-)

The Ada solution uses full tasking and this seems to be showing
overhead here:
Each task in chameneos-redux does *very* little, but communicates
a *lot* with via POs, delays execution in a selective accept,
the POs have entries, ..., so I think it is Ada's tasking overhead.
I have tried using Ravenscar profile facilities instead (i.e. just
suspension objects entry-less tasks) but rewriting this requires
care (and skills, and time, and looking at different solutions...
and possibly the resulting code is helped by a run-time
that produces faster code if pragma Profile (Ravenscar)
is in effect).