Re: combine patch 1/2: compares with sign/zero extensions

[Ian Lance Taylor <ian at airs dot com>]

"Joseph S. Myers" <joseph@codesourcery.com> writes:

> I've tested a revised patch version which only changes i2src to i2src_orig 
> in the COMPARE case (i.e., omits the last patch hunk from my original 
> patch) and verified that it still has the desired effect (of generating 
> all the desired instructions with my new instruction patterns).  Because 
> in the COMPARE case it doesn't matter whether subst's changes are good or 
> bad - any change will lead to the two parts of the PARALLEL not matching 
> and so the combination not being made - I think this cut-down version 
> should be a safe patch.

That seems like a good approach to me.  What I think would work well
would be to test for that case before calling subst at all.  That way
you will avoid a useless call to copy_rtx (in all cases) and to subst
(in a few cases).

Thanks.

Ian