[R] crossprod(x) vs crossprod(x,x)

[Prof Brian Ripley]

On Tue, 21 Nov 2006, Duncan Murdoch wrote:

> On 11/21/2006 6:39 AM, Robin Hankin wrote:
>> Professor Ripley
>> 
>> thanks for this.
>> 
>> I see what you say about gc().  I'll try your looped test on my system:
>> 
>> 
>>
>>  > x <- matrix(rnorm(3000000),ncol=3)
>>  >
>>  > system.time(for(i in 1:100){crossprod(x)})
>> [1]  7.931 20.420 28.619  0.000  0.000
>>  > system.time(for(i in 1:100){crossprod(x)})
>> [1]  7.975 20.590 29.512  0.000  0.000
>>  > system.time(for(i in 1:100){crossprod(x)})
>> [1]  8.003 20.641 30.160  0.000  0.000
>>  > system.time(for(i in 1:100){crossprod(x,x)})
>> [1] 2.350 0.032 2.552 0.000 0.000
>>  > system.time(for(i in 1:100){crossprod(x,x)})
>> [1] 2.330 0.015 2.333 0.000 0.000
>>  > system.time(for(i in 1:100){crossprod(x,x)})
>> [1] 2.333 0.034 2.447 0.000 0.000
>>  >
>> 
>> 
>> 
>> How come my findings are qualitatively different from yours?
>
> Have you said what version of R you're using?  I tried your code in an Intel 
> Mac with R 2.4.0, and a Windows machine with R-devel, and saw what Brian saw. 
> Maybe the Power Mac changes this, but that doesn't make a lot of sense.  On 
> the other hand, it could be that an older R release is making unnecessary 
> copies in one case but not the other.

I think this indicates a problem with the BLAS on the PPC MacOS, which is 
very different from the Intel one, AFAIK.  Nothing in R itself could be 
directly responsible for that system-time usage, since R is a user 
process.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595