regression for 3.1: bad sign extension?

[Richard Henderson]

On Thu, Mar 14, 2002 at 05:11:44PM -0800, Bob Wilson wrote:
> Perhaps I should change the combiner to detect a zero-extend of 
> a value loaded from the constant pool and combine those two ops by 
> putting a new zero-extended value in the pool.  Would that be a good 
> solution?

Yes, because now you are referring to an SImode constant,
rather than an HImode constant.


r~