Re: random variate from power exponential distribution:performances

[Giulio Bottazzi <giulio dot bottazzi at libero dot it>]

On Mon, 27 Sep 2004 11:00:08 -0500
Linas Vepstas <linas@austin.ibm.com> wrote:

> You should test again with optimization turned on. This can make 
> a *huge* difference in relative performance.   Without optimization,
> the compiler does literally what the C code says, with an excess use
> of register-to-register copies.  Thus, and sort of redundant

Hi Linas,
very good point!  What you suggest is longer to say than to do

gcc -O3 -march=i686 -lgsl -lgslcblas testran.c -o testran

results (ignore the value for a, it doesn't matter) :

value for a    value for b   JOHNSON (secs)  TADIKAMALLA (secs)    

1.000000	0.500000	0.540000	3.630000
1.000000	0.600000	1.590000	2.880000
1.000000	0.700000	1.550000	2.500000
1.000000	0.800000	1.490000	2.290000
1.000000	0.900000	1.430000	2.170000
1.000000	1.000000	0.510000	0.220000
1.000000	1.100000	1.330000	2.160000
1.000000	1.200000	1.330000	2.180000
1.000000	1.300000	1.330000	2.160000
1.000000	1.400000	1.320000	2.180000
1.000000	1.500000	1.310000	2.180000
1.000000	1.600000	1.320000	2.170000
1.000000	1.700000	1.300000	2.170000
1.000000	1.800000	1.290000	2.180000
1.000000	1.900000	1.290000	2.180000
1.000000	2.000000	1.290000	0.410000
1.000000	2.100000	1.280000	4.370000
1.000000	2.200000	1.270000	4.370000
1.000000	2.300000	1.270000	4.360000
1.000000	2.400000	1.260000	4.370000
1.000000	2.500000	1.260000	4.370000
1.000000	2.600000	1.250000	4.370000
1.000000	2.700000	1.250000	4.380000
1.000000	2.800000	1.250000	4.380000
1.000000	2.900000	1.250000	4.380000
1.000000	3.000000	1.230000	3.920000
1.000000	3.200000	1.230000	4.390000
1.000000	3.400000	1.230000	4.370000
1.000000	3.600000	1.220000	4.370000
1.000000	3.800000	1.210000	4.370000
1.000000	4.000000	1.210000	3.960000


comment: things did not change, JOHNSON is ALWAYS faster.


BTW I've tested also on a dual opteron compiling with

gcc -O3 -lgsl -lgslcblas testran.c -o testran

guess what? same thing: JOHNSON is ALWAYS faster.


Any other optimization option to suggest?


	G.




-- 
Giulio Bottazzi <bottazzi@sssup.it>
Laboratory of Economics and Management
Sant'Anna School for Advanced Studies, Pisa, Italy
Phone: (+39)-050-883365  Fax: (+39)-050-883344
URL: http://www.sssup.it/~bottazzi/

Attachment: pgp00000.pgp
Description: PGP signature