Dave Harris wrote:

In-Reply-To: <c7bsah$27c$1@sea.gmane.org> ianmcc@lorentz.leidenuniv.nl (Ian McCulloch) wrote (abridged):

...

Nice idea. I was thinking along the lines of how to achieve high-performance, with shared memory & fast network message passing in mind.

I have some experience of that kind of code being quicker than code which naively treats an int as 4 bytes. There was enough compression that the bandwidth saving (eg on disk i/o) compensated for the complexity of the code.

That was some years ago, and there may have been ways to optimise the naive code that I didn't try. Still, I wouldn't write this off on speed grounds without measuring it.

Interesting. I'll try to get some measurements, especially on Alpha, where the memory subsystem is generally pretty good and alignment issues would come into play as well. But as you say, it shouldn't be written off. Cheers, Ian