[PATCH] fix PR middle-end/22480

[Richard Henderson]

On Sat, Sep 10, 2005 at 01:02:20PM +0200, Uros Bizjak wrote:
> A couple of lines above, there is a loop that checks all tree operands 
> with 'vect_is_simple_use (..., &dt)'. At the check quoted above, dt 
> holds the definition of  the last operand, which is just what we need.

That's not what I meant.  At what point do you return false when
the operand is something that sse cannot support?


r~