Re: Better handling for multi-word values?

[Bernd Schmidt <crux at pool dot informatik dot rwth-aachen dot de>]

> 
>   > My question is now, what could be done about it?  I've played with the comp iler
>   > a bit yesterday and came up with a half-done patch (which I won't post here,
>   > since it doesn't work properly yet; I can send it on request).  I'll describe
>   > what I did so that people can comment on whether they think it's a good
>   > starting point or not.
>   > 
>   > I added a new rtx code, MULTIWORD, which represents a value that consists of
>   > multiple word-sized parts.  The idea is that instead of a DImode pseudo, two
>   > SImode pseudos are allocated and put inside of a MULTIWORD:
> This sounds very similar to the half-finished CONCAT stuff already
> in the compiler.....

Similar, but not identical.  For example, CONCATs go away during tree->rtl
conversion and never show up during later passes.  And CONCATs only work with
exactly two separate objects, which I feel is an unnecessary restriction.

Bernd