[Matrix-SIG] RE: Shifted array ops (was: RE: [Matrix-SIG] QUERY: Array indexing)

[Tim Peters]

[Frank Horowitz has an "aha!" experience thanks to Perry Stoll (thanks.
 Perry Stoll!)]

[Tim's toy VectorRotate class in non-NumPy Python]

[Frank]
> If I understand your suggestion correctly, you're doing some trickery
> with indexing *for an individual element* ...

Exactly, and in non-NumPy Python that doesn't generalize to whole-array
operations.

> Does the fact that they're overriding the operator class methods imply
> that the result of the calls to getdata index through the array at C
> speeds (courtesy of the NumPy __getitem__ support code)?  Or is the
> result that the getdata indexes through the array at interpreter speeds?

I *still* haven't installed NumPy <wink>, but my non-NumPy example class
works at interpreter speed.  Presumably combining NumPy slice objects with
NumPy whole-array ops works at C speed, though.  Same idea under the covers:
don't copy the data, apply a transformation to the indices.  Simple things
like that sure do go faster in C!

> Remember, I'm contemplating applying this stuff (over and over :-) to
> arrays with ~12 million elements.  If it runs at interpreter speeds, I'll
> have to look for a different answer.

That part was unclear, since you had mentioned disk thrashing.  A gimmick
that can stop *that* can afford to waste a lot of CPU and still win in the
end.  That's what I use it for (in non-NumPy Python).

As with all things Python, though, you don't have to choose because
everything's perfect <wink>.

although-perfection-is-often-cleverly-hiding-in-plain-view-ly y'rs  - tim