Ada and multi-core CPUs

Ada and multi-core CPUs

[tmoran]

I finally got around to testing a sort.  I split the data array in half,
sorting one half with the main program and the other with a separate task,
then merged the two half-results.  It's substantially (about 60% as long
for 100K floats) faster than splitting, sort one half, then the other
half, then merge.  And easy to program in Ada.

Re: Ada and multi-core CPUs

[Wilhelm Spickermann]

Hi,

tmoran@acm.org wrote:

> Iï¿œsplitï¿œtheï¿œdataï¿œarrayï¿œinï¿œhalf,
> sorting one half with the main program and the other with a
> separate task, then merged the two half-results.

Instead of merging, you could also insert another parallel sort,
one for both of the upper quarters and the other for both of the
lower quarters. After that, only the two inner quarters need to
be merged or sorted (perhaps recursively). 
(see D. E. Knuth: The Art of Computer Programming, Vol III
Sorting and Searching, 5.3.4 Networks for Sorting)

Wilhelm Spickermann

Re: Ada and multi-core CPUs

[tmoran]

> Instead of merging, you could ...
  There are many ways to make a parallel sort.  My point was that here
is a very standard time consuming operation, very simple Ada code can
make a simple dual-tasking version, and that version on a hyperthreading
CPU really does run a lot faster.

Re: Ada and multi-core CPUs

[Adrian Knoth]

tmoran@acm.org <tmoran@acm.org> wrote:

>> Instead of merging, you could ...
>   There are many ways to make a parallel sort.  My point was that here
> is a very standard time consuming operation, very simple Ada code can
> make a simple dual-tasking version, and that version on a hyperthreading
> CPU really does run a lot faster.

Only Hyperthreading with single-core or a real dual-core/SMP-system?


-- 
mail: adi@thur.de  	http://adi.thur.de	PGP: v2-key via keyserver

Einer spinnt immer. Wenn zwei spinnen, wird's schlimmer

Re: Ada and multi-core CPUs

[tmoran]

> Only Hyperthreading with single-core or a real dual-core/SMP-system?
I ran the test on my single-core hyperthreaded system.  If you have a
true dual-core system I'd be curious to hear your results.
  After running in 60% of the time by splitting in half, sorting,
then merging the two halves, I tried burying the tasking inside
a quicksort.  On the first partitioning, one of the partitions goes
to a separate task.  That ran around 75%, perhaps because the
partitions are likely not the same size.

Re: Ada and multi-core CPUs

[Marc A. Criley]

tmoran@acm.org wrote:
> I finally got around to testing a sort.  I split the data array in half,
> sorting one half with the main program and the other with a separate task,
> then merged the two half-results.  It's substantially (about 60% as long
> for 100K floats) faster than splitting, sort one half, then the other
> half, then merge.  And easy to program in Ada.

Write it up and submit for publication.  Chip Multi-Processing 
(multi-core CPUs) seems to be a hot technical issue; it was the theme of 
the latest (9/05) issue of ACM Queue.

-- Marc A. Criley
-- McKae Technologies
-- www.mckae.com
-- DTraq - XPath In Ada - XML EZ Out

Re: Ada and multi-core CPUs

[Jeffrey Carter]

tmoran@acm.org wrote:
> 
> I ran the test on my single-core hyperthreaded system.

Curious Subject for a report on a single-core CPU. I find the results 
more interesting now that I know they apply to a single-core system.

My work computer runs WinXP on one or more P4HT processors. Control 
Panel => System, General tab, reports 4 CPUs. Task manager allows 
processes to have "affinity" for up to 2 CPUs. How many CPUs are in the 
machine? I'm not sure, but I think it's one.

-- 
Jeffrey Carter
"Now go away or I shall taunt you a second time."
Monty Python and the Holy Grail
E-mail: jeffrey_r_carter-nr [commercial-at]
         raytheon [period | full stop] com

Re: Ada and multi-core CPUs

[tmoran]

> Write it up and submit for publication.  Chip Multi-Processing
> (multi-core CPUs) seems to be a hot technical issue; it was the theme of
> the latest (9/05) issue of ACM Queue.
    I sent them a letter to the editor.  Perhaps someone with better
writing skills than myself could send them an article.