Re: GNAT compiler switches and optimization

["Jeffrey R. Carter" <spam.not.jrcarter@acm.not.spam.org>]

tmoran@acm.org wrote:
>    Each 800x800 Float matrix is about 2.5 megabytes so I would expect two
> threads to be fighting each other over the cache.  What are the single vs
> dual threaded times for, say, 250x250 arrays and a 1 MB cache on your
> machine, and on the dual Xeon machine?
>    Curiously, on a 3 GHz Pentium 4 with 1 MB cache and compiling
> with -O2 -gnatp using the venerable gnat 3.15p, I get
> Time:           8.511855190       800 1 thread
> Time:           5.682686332       800 2 threads
> for a speedup of 33%
> Time:           0.092724434       200 1 thread
> Time:           0.059209520       200 2 threads
> for a speedup of 36%

On my 3.2 GHz P4 HT (best of 3 runs), I get

Time:           0.170563050       250 1 thread
Time:           0.134637140       250 2 threads

for a speedup of 24%.

Time:           0.113739732       200 1 thread
Time:           0.070599932       200 2 threads

for a speedup of 36%.

I'm still only getting about 10% speedup for 800. Is it possible my 
cache is smaller?

-- 
Jeff Carter
"I unclog my nose towards you."
Monty Python & the Holy Grail
11