Re: Reload codegen improvement

[Bernd Schmidt <bernds at codesourcery dot com>]

On 01/07/2014 05:22 PM, Bernd Schmidt wrote:
> This fixes a problem identified by Chung-Lin. Once reload is done, all
> equivalencing insns for pseudos that didn't get a hard reg but could be
> eliminated using their REG_EQUIV are deleted. However, we still can
> produce reloads and reload insns for them in certain cases, leading to
> unnecessary spilling. This patch corrects this by making sure we use
> identical tests when deciding whether to ignore an insn while reloading,
> and whether to delete it afterwards.
>
> Bootstrapped and tested on x86_64-linux (with lra_p disabled). Chung-Lin
> says he's tested it as well, I think on arm (probably with something 4.8
> based). Will commit in a few days if no objections.

I decided to wait until after 4.9.  Now committed after testing on 
bfin-elf - the trick of disabling LRA on x86_64 no longer works, the 
target doesn't build with reload enabled.

(I saw one compare-debug randomly failing or passing, which happens with 
or without this change. It seems to be caused by something in IRA).


Bernd