[Numpy-discussion] Need help for implementing a fast clip in numpy (was slow clip)

David Cournapeau david at ar.media.kyoto-u.ac.jp
Fri Jan 12 03:32:01 EST 2007

Charles R Harris wrote:
> Well, the common machines are 32 bit and 64 bit, so for instance 
> extended precision (usually 80 bits) ends up as 96 bits (3*32) on the 
> first and 128 (2*64) bits on the second, with the extra bits ignored. 
> The items in c structures will often have empty spaces filling in 
> between them unless specific compiler directives are invoked and the 
> whole will be aligned on the appropriate boundary. For instance, on my 
> machine
> struct bar {
>     char a;
>     int  b;
> };
So the data buffer in numpy is not really a C array, but more like a 
structure ? (I am  talking about the actual binary data)

In C, I think you always have sizeof(*a) bytes between a[0] and a[i]. 
That is sizeof(item) and addresses shift always match: we always have 
a[i] = (char*)a + sizeof(*a) * i. For example, in your example for 
extended precision, the computation may be done in 80 bits, but 
sizeof(long double) is 12 on my machine, not 10. There is a match 
between sizeof and spacing between items; otherwise, I don't see how 
pointer arithmetic would be possible with arrays in C otherwise ?

But this goes quite further from my usual C knowledge, so I may be wrong 
here too.



More information about the NumPy-Discussion mailing list