Boost logo

Ublas :

Subject: Re: [ublas] [uBLAS] What's inside the uBLAS matrix multiplication algorithm?
From: Rui Maciel (rui.maciel_at_[hidden])
Date: 2010-06-14 06:48:36


On 14 June 2010 10:47, Anders Kabell Kristensen <dalko_at_[hidden]> wrote:
> Thanks for your answer Gunter, I will try some other libraries the, but
>>
>> It is O(n^2) for full dense matrices. It is O(nnz) for sparse matrices
>> and it is O(n*k) for band matrices. You can check matrix_expression.hpp
>> for the class matrix_:matrix_product (or similar name) and
>> operations.hpp for axpy_prod().
>>
>
> don't you mean O(n^3) for dense matrices? If not, can you explain why?
> Surely, naive mat. mult. should be O(nm^2) or (n^2m), right?

It does appear that it was a mistake, as the fastest matrix
multiplication algorithm is said to be O(2.38) [1]

Rui Maciel

[1] http://www.cs.berkeley.edu/~oholtz/Talks/mit.pdf