PR 11350

[Jan Hubicka]

> 
> Hi,
> the problem is that the destination of tablejump gets spilled onto stack and
> the tablejump gets elliminated after reload so the load stays in the code.
> Bootstrappping/regtesting i686-pc-gnu-linux in progress, OK?

... forgot to mention.  I realize that this make crossjumping of
tablejumps to leave ugly artefacts, but I really don't see any good way
around.  
Perhaps we can do one crossjumping before register allocation?  It
works, but it is less effecitve as different pseudos used for
temporaries preserve crossjumping.

Honza