Re: [boost] Re: Probing interest in fixed dimension matrix class

From: christopher diggins (cdiggins_at_[hidden])
Date: 2005-06-03 07:51:03

----- Original Message ----- 
From: "Kresimir Fresl" <fresl_at_[hidden]>
To: <boost_at_[hidden]>
Sent: Friday, June 03, 2005 2:10 AM
Subject: Re: [boost] Re: Probing interest in fixed dimension matrix class

>
> christopher diggins wrote:
>
>> Done! The url is the same. The new benchmarks improve speed for ublas by 
>> using noalias(A) = prod(B, C) but the kmatrix speed is still far superior 
>> on my machine. (celeron 1.6, yadda yadda yadda)
>
> Did you try to compare kmatrix with
>
>   ublas::matrix<T, row_major, ublas::bounded_array<T, M*N> >;
>
> fres

Just did, basically the same results:

comparing integer matricies
2,2 X 2,2
kmatrix 63 msec elapsed
ublas 156 msec elapsed
3,3 X 3,3
kmatrix 219 msec elapsed
ublas 438 msec elapsed
100,100 X 100,100
kmatrix 406 msec elapsed
ublas 1187 msec elapsed
100,1 X 1,100
kmatrix 125 msec elapsed
ublas 203 msec elapsed
1,100 X 100, 1
kmatrix 406 msec elapsed
ublas 1172 msec elapsed
comparing double matricies
2,2 X 2,2
kmatrix 78 msec elapsed
ublas 172 msec elapsed
3,3 X 3,3
kmatrix 219 msec elapsed
ublas 469 msec elapsed
100,100 X 100,100
kmatrix 547 msec elapsed
ublas 1390 msec elapsed
100,1 X 1,100
kmatrix 156 msec elapsed
ublas 219 msec elapsed
1,100 X 100, 1
kmatrix 484 msec elapsed
ublas 1375 msec elapsed

- Christopher Diggins