Re: [Boost-users] R: R: [multi-array] Performance? Difference from GIL or uBLAS?

From: John C. Femiani (john.femiani_at_[hidden])
Date: 2008-08-08 04:03:14

Andrea Denzler wrote:
>
> Working with indexes is defintively easier and more confortable. And 
> probably in many situation the overhead is not worth to be optimized 
> at all.
>
> I only was surprised when you said there is a performance gain using 
> indexes.
>
>  
>
> But I hardly would write something like this (from your array.cpp example)
>
>  
>
>     for (int i = 0 ; i < h ; ++i)
>
>         for (int j = 0 ; j < w ; ++j) {
>
>             array[i][j] = uni() ;
>
>         }
>
>  
>
> You are loosing performance unless the compiler is so intelligent to 
> avoid the multiplication at each step. When performance is really an 
> issue then I first re-think the algorithm, after I check the produced 
> assembler listing to see what the compiler did. In the example above 
> at each step you have to calculate this address:
>
>     array + i*rowsize + j
>
>  
>
> If you are lucky the compiler can produce code that does this in one 
> instruction set, if not you will get an overhead. Again if h and w are 
> low values you will not notice it.
>
>  
>
> This example is of course much faster, and yes, it is not elegant nor 
> clear.
>
>      int *p=array*,**pend=&array[h][w];
>
>      while (p < pend)
>
>             *p++ = uni() ;
>

The standard algorithms use iterators; An array implementation that 
translates iterators to indexes and expects the indexes to be converted 
back into pointers is questionable to me. Especially when my trivial 
test code showed the multi-array implementation was 10-20 times slower.  
This slow down is way more than I expected.

The multi_array implementation is extremely scary to me because it does 
not seem to be low overhead at all. 

-- John